Appendix AContributed Manuscripts



,2,3 ,4 ,5 ,4 ,2 ,5,6,7 and 4,8,9.


2 Department of Surgery, University of Chicago Pritzker School of Medicine, Chicago, IL 60637.
3 Present address: Department of Surgery, University of Pittsburgh Medical Center, Pittsburgh, PA 15213.
4 Department of Earth and Planetary Science, University of California, Berkeley, CA 94720.
5 Department of Microbiology and Immunology, Stanford University School of Medicine, Stanford, CA 94305.
6 Department of Medicine, Stanford University School of Medicine, Stanford, CA 94305.
7 Veteran’s Affairs Palo Alto Heath Care System, Palo Alto, CA 94304.
8 Department of Environmental Science, Policy, and Management, University of California, Berkeley, CA 94720.
9 To whom correspondence should be addressed. E-mail: ude.yelekreb@dleifnabj.

The intestinal microbiome is a critical determinant of human health. Alterations in its composition have been correlated with chronic disorders, such as obesity and inflammatory bowel disease in adults, and may be associated with neonatal necrotizing enterocolitis in premature infants. Increasing evidence suggests that strain-level genomic variation may underpin distinct ecological trajectories within mixed populations, yet there have been few strain-resolved analyses of genotype–phenotype connections in the context of the human ecosystem. Here, we document strainlevel genomic divergence during the first 3 wk of life within the fecal microbiota of an infant born at 28-wk gestation. We observed three compositional phases during colonization, and reconstructed and intensively curated population genomic datasets from the third phase. The relative abundance of two Citrobacter strains sharing ~99% nucleotide identity changed significantly over time within a community dominated by a nearly clonal Serratia population and harboring a lower abundance Enterococcus population and multiple plasmids and bacteriophage. Modeling of Citrobacter strain abundance suggests differences in growth rates and host colonization patterns. We identified genotypic variation potentially responsible for divergent strain ecologies, including hotspots of sequence variation in regulatory genes and intergenic regions, and in genes involved in transport, flagellar biosynthesis, substrate metabolism, and host colonization, as well as differences in the complements of these genes. Our results demonstrate that a community genomic approach can elucidate gut microbial colonization at the resolution required to discern medically relevant strain and species population dynamics, and hence improve our ability to diagnose and treat microbial community-mediated disorders.

Intestinal microbes influence human health through harvesting of energy from dietary substrates, production of essential nutrients, and protection against colonization by pathogens (Dethlefsen et al., 2007; Hooper et al., 2002). Although the adult gut microbiota is highly variable between individuals, it displays limited diversity at the phylum level: only two bacterial phyla (Bacteroidetes and Firmicutes) contribute ~90% of all microbes (Eckburg et al., 2005). In infants, early assembly of the gut microbiota has been linked to development of innate immune responses and terminal differentiation of intestinal structures (Hooper et al., 2001). The dynamic process of colonization has been well studied at high taxonomic levels (Palmer et al., 2007) and seems predictable based on competitive interactions between and within the dominant phyla (Trosvik et al., 2010). Yet at lower taxonomic levels, and at early stages of development, our knowledge of this process is incomplete.

Strain-level analyses of clinical isolates using multilocus sequence typing (MLST) and comparative genomics have been used to differentiate closely related organisms (Hanage et al., 2009; Palmer et al., 2010). However, important contextual information may be lost when interpreting genomic variation between strains isolated from different communities. Microbial population dynamics can be strongly influenced by synergism and competition with coexisting microorganisms and through phage predation (Sandaa et al., 2009). The mobile element pool, which is generally excluded when analyzing isolates, can rapidly give rise to the genomic variation that underpins strain differentiation (Oliver et al., 2009).

Cultivation-independent genomic analyses of time-series samples provide a way to link shifts in population abundance to genetic characteristics that underlie physiological traits, such as virulence. Here, we analyzed human intestinal colonization during the neonatal period. We conducted a 16S rRNA gene-based survey of fecal samples collected daily during the first 3 wk of life of a premature infant and reconstructed and manually curated population genomic datasets for the dominant gut microorganisms in the third of three colonization phases. We chose to focus on the premature infant microbiome because, in addition to its medical relevance, the limited number of dominant bacterial species in the community allows for deep sequence coverage of multiple subpopulations.

Results and Discussion

Study Subject

We studied fecal samples from a female infant delivered by caesarean section at 28-wk gestation due to premature rupture of membranes. She was treated empirically with broad-spectrum antibiotics (ampicillin/gentamicin) for the first 7 d of life but did not receive antibiotics during the remainder of the study period. She received enteral feedings with maternal breast milk between the fourth and ninth days of life. Feedings were withheld between days 9 and 13 because of abdominal distension. On day 13, feedings were slowly resumed with artificial infant formula (Similac Special Care 20 cal/fl oz; Abbott Nutrition). She also received parenteral nutrition until caloric intake from enteral nutrition was adequate (day 28). She had no major illnesses during her hospitalization and was discharged to home at 64 d of life. Fecal samples were collected daily as available between days 5 and 21.

Day-to-Day Dynamics of Community Composition

Sequencing of amplified bacterial 16S rRNA genes (SI Materials and Methods and Table S1 A and B) from 15 fecal samples collected on different days during the first 3 wk revealed three distinct community configurations demarcated by rapid transitions. This finding is consistent with previously reported colonization patterns in term infants: relative stability over days to months punctuated by rapid compositional change (Koenig et al., 2010; Palmer et al., 2007). Marked shifts in abundant lineages around days 9 and 15 seemed to follow dietary adjustments. On days 5 through 9, communities were largely composed of Leuconostoc, Weissella, and Lactococcus (Fig. A1-1A). The genera Pseudomonas and Staphylococcus, which were relatively scarce on days 8 and 9, became abundant by day 10. On days 10 through 13, species richness and evenness were relatively low (Table S1) and Pseudomonadaceae predominated (Fig. A1-1A). After resuming feedings on day 13, taxa characteristic of the next phase appeared (Fig. A1-1A). On days 16 through 21, species richness and evenness recovered (Table S1) and the family Enterobacteriaceae and its constituent genera Citrobacter and Serratia came into the majority. Sample clustering based on community-wide similarity in membership and structure (Fig. A1-1B and Fig. S1 C–F) further delineated three microbiome configurations. Bacterial community membership and structure were significantly more similar within, than between these colonization phases (P < 0.001; PERMANOVA with Monte Carlo). A crossstudy comparison suggests that the infant studied here harbored similar bacteria to those found in other premature infants surveyed using equivalent methods, especially during the first and third colonization phases (Fig. A1-1B) (de la Cochetiere et al., 2004; Gewolb et al., 1999; Palmer et al., 2007; Mackie et al., 1999; Magne et al., 2006; Millar et al., 1996; Mshvildadze et al., 2010; Schwiertz et al., 2003; Wang et al., 2009).

Two graphs showing multiple stable compositional states in the developing gut microbiota of the premature infant


Multiple stable compositional states in the developing gut microbiota of the premature infant. (A) Relative abundance of the 20 most dominant bacterial taxa in 15 fecal samples collected between days 5 and 21. Sequences were classified to the highest (more...)

Metagenomic Data Processing

Genome-wide sequencing of DNA from fecal samples collected on days 10, 16, 18, and 21 yielded 245 Mbp of metagenomic sequence data. These data were coassembled using Newbler, keeping track of each read’s sample of origin for quantification. Quantification of community composition based on read abundance can be confounded by DNA extraction and sequencing biases (Morgan et al., 2010). However, we could analyze relative abundance shifts across the third colonization phase because the same biases were expected in all samples (Fig. A1-2). We identified three major sequence “bins” for Serratia, Citrobacter, and Enterococcus, which dominated the third phase of colonization (Figs. A1-1A and A1-2). Projecting the smaller contig data (500–1,500 bp) onto an emergent self-organizing map generated based on tetranucleotide frequencies of contigs >1,500 bp and reference genomes allowed us to assign additional fragments to Enterococcus and provide partial coverage for one or more Pseudomonas populations from the day 10 sample (SI Materials and Methods and Fig. S2). Most fragments from other minor populations were assigned to higher taxonomic levels (mostly Enterobacteriaceae) (Table S3 in Dataset S1). We also identified multiple plasmid and phage populations, some of which were completely sequenced (Table S4 in Dataset S1).

A graphical representation of population dynamics based on metagenomic profiling


Population dynamics based on metagenomic profiling. Distribution of the reads over the curated sequence bins across each library (as percentage of all reads in the libraries from day 10, 16, 18, and 21, respectively).

Manual curation resulted in a Serratia genome (strain UC1SER) with nine gaps, seven of which involve rRNA operons. Based on the sequence coverage of Serratia (~17×) compared with other bacterial contigs (Table S2), UC1SER dominated the community genomic datasets from the formula fed (third) phase. We detected remarkably low levels of nucleotide polymorphisms in the UC1SER sequences (close to the expected sequencing substitution error rate), and only very few regions in which gene content varied.

Serratia, a genus comprising motile, facultative anaerobes from the family Enterobacteriaceae, is found in many environments. The UC1SER genome assembled de novo from metagenomic data was compared with the publicly available genomes of Serratia proteamaculans (21) and Serratia marcescens (Sanger Institute, United Kingdom). S. marcescens is an important opportunistic pathogen and a known cause of nosocomial disease in neonatal intensive care units (22). S. proteamaculans is an endophytic bacterium rarely identified in human specimens. All curated UC1SER genome fragments (up to 2.36 Mb in length) share a syntenous backbone with the previously reported genomes, although numerous genomic differences were noted relative to the previously sequenced species (Table S5 in Dataset S1). For syntenous orthologs, UC1SER predicted proteins share 97.3% average amino acid identity (AAI) over 4,089 genes and 88.6% AAI over 3,672 genes with S. marcescens and S. proteamaculans, respectively. Given the overall synteny with S. marcescens and S. proteamaculans across reconstructed genome fragments, we ordered the nine UC1SER genome fragments according to the reference genomes (Table S5 in Dataset S1).

Assembly of a Near-Clonal Serratia Genome and Comparative Genomics

Within syntenous regions in UC1SER, there are small clusters of genes that occur elsewhere in S. marcescens and S. proteamaculans. These clusters encode proteins involved in protocatechuate utilization, fimbrial biosynthesis and export, nitrate reduction, general secretion, siderophore (enterobactin) synthesis and transport, tetrathionate reduction and regulation, osmoprotectant transport, and general metabolism, including amino acid biosynthesis. These rearranged or “indel” regions show elevated sequence divergence relative to syntenous orthologs (AAI of 77 and 58% relative to S. marcescens and S. proteamaculans, respectively). Thus, these regions may contribute to metabolic variation that differentiates these species.

Regions of the UC1SER genome that are absent in one or both of the other Serratia species encode factors involved in transport (most notably iron uptake) and regulation, outer membrane and exopolysaccharide biosynthesis, adhesion, antibiotic biosynthesis, virulence, quorum sensing, biosynthesis of the redox cofactor pyrroloquinoline quinone, arsenate resistance, and propanoate metabolism (Table S5 in Dataset S1). Only UC1SER contains pga operon genes involved in polysaccharide synthesis for biofilm adhesion and a regulon for allantoin utilization, which may be associated with virulence (Chou et al., 2004). It is also the only genome with yjf-sga operon genes (phosphotransferase system components sgaH, U, E), which enable some strains of gut bacteria to use vitamin C as an energy source (Campos et al., 2008). UC1SERalso has a large nonribosomal peptide biosynthesis protein not found in the other genomes. In contrast to the other reconstructed genomes in this study, UC1SER contains few mobile element-derived sequences.

Analyses of Two Ecologically Distinct Citrobacter Subpopulations

Based on 16S rRNA gene sequences on assembled contigs, Citrobacter in the third colonization phase is closely related to Citrobacter freundii. Despite average coverage of ~13× on larger Citrobacter fragments, automated assembly resulted in a highly fragmented genome. Citrobacter contigs displayed many diallelic sites among their reads that were almost always linked (i.e., no evidence for homologous recombination), indicating the presence of two coassembled strain populations. Examination of most contig ends revealed path bifurcation (Fig. A1-3A) because of local strain sequence divergence, differences in gene content, and intergenic region length (see below).

Six graphs showing analyses of two ecologically divergent Citrobacter UC1CIT subpopulations


Analyses of two ecologically divergent Citrobacter UC1CIT subpopulations. (A) Schematic representation of part of the fragmented UC1CIT assembly. At the ends of many Citrobacter contigs (e.g., contig number 699), reads that partially coassembled led to (more...)

Manual curation resolved these bifurcations and reduced the number of Citrobacter contigs from ~1,400 to 10 (the largest curated contig is 2.55 Mb) (Fig. A1-3B). The final contigs are generally syntenous with the Citrobacter 30_2 strain draft genome (Broad Institute, Cambridge, MA) and the complete Citrobacter koseri ATCC BAA-895 genome (Washington University, St. Louis, MO). Consequently, the fragments were oriented and ordered by reference to the C. koseri genome to generate a final genome representation for the dominant strain, UC1CIT-i (Table S6 in Dataset S2). Of the ten genome gaps, eight are the rRNA-encoding regions that could not be resolved, one is within a prophage, and one is in the intergenic region between genes on contig ends that are adjacent in both isolate genomes.

Citrobacter species are facultative anaerobes from the family Enterobacteriaceae and are commonly found as commensals within the mammalian intestinal tract. Like Serratia, they have been frequently documented as pathogens in premature newborns (Doran, 1999) (e.g., in cases of neonatal meningitis). Citrobacter 30_2 was isolated from a patient with Crohn disease, whereas C. koseri was isolated from an infant with meningitis. UC1CIT strains lack a “supercontig” of 402 genes reported as part of Citrobacter 30_2; based on our assembly and the functional annotation, we suspect this supercontig derives from a megaplasmid.

As expected based upon the known physiology of human-associated Citrobacter strains (Doran, 1999), the UC1CIT strains have numerous genes for uptake and utilization of a wide variety of substrates. Similar to C. koseri and Citrobacter 30_2, the UC1CIT strains are predicted to express curli and fimbriae that mediate biofilm formation and binding to host epithelial cells (Barnhart and Chapman, 2006) (Table S6 in Dataset S2). Interestingly, the UC1CIT strains and C. koseri have dual flagellar systems but Citrobacter sp. 30_2 lacks a lateral flagellar apparatus (Table S7 in Dataset S2). Lateral flagella confer swarming motility in viscous fluids (e.g., mucus) and have been associated with virulence, adhesion, and biofilm formation (Gavín et al., 2002; Merino et al., 2006).

UC1CIT sequence variation occurs genome-wide, but one sequence type dominates at most loci (Table S6 in Dataset S2). Given evidence for clonal rather than recombined strains, we defined the minor strain type (UC1CIT-ii) by separating reads primarily using polymorphism patterns in Strainer (Eppley et al., 2007) (Fig. A1-3C), which allowed for direct comparison of the two aligned strains. UC1CIT-ii sequence blocks (up to a few kilobases in length) share 98.5% average nucleotide identity with UC1CIT-i. In regions of shared gene content, ~90% of the UC1CIT-ii genome was reconstructed. When the UC1CIT-ii strain blocks were linked and intervening gaps filled by UC1CIT-i sequence, the strains shared 99.1 ± 0.3% average nucleotide identity across their genomes (Table S8 in Dataset S2). The true level of similarity for orthologous sequences likely lies between these values.

Based on the relative frequency of strain-associated reads in the combined dataset for days 10, 16, 18, and 21, UC1CIT-i comprised 77% of the Citrobacter population (SI Materials and Methods and Table S8 in Dataset S2). However, the relative abundance of the strains changed dramatically during the third colonization phase (Fig. A1-3D and Table S8 in Dataset S2). Possible explanations for the strain abundance shifts include: (i) a bloom of a strain-specific phage that decimated the UC1CIT-ii population around day 18; (ii) a reduced growth rate of UC1CIT-ii when it was outcompeted for resources by UC1CIT-i, Serratia or Enterococcus populations; and (iii) a higher potential of UC1CIT-ii for intestinal wall colonization, leading to an observed decrease in the luminal (fecal) population.

A phage bloom is unlikely because we did not observe an increase in the abundance of Citrobacter phage sequences across the time series. To evaluate the other hypotheses, we constructed two models of bacterial growth in the colon (SI Materials and Methods and Fig. S3). First, using a simplified colon chemostat model, we calculated the differences in growth rates needed to fit the strain population abundance shifts from days 16 to 18 and days 18 to 21 (Fig. A1-3E). Assuming approximately equal numbers of cells per milliliter luminal content, the model predicts nearly constant generation times for UC1CIT-i. The UC1CIT-ii generation time estimates equaled those for UC1CIT-i between days 18 and 21, but increased above the colon transit time (CTT) between days 16 and 18, resulting in washout between days 16 and 18. Based on CTT in children (12– 84 h) (Wagener et al., 2004) and estimates for Escherichia coli generation times in animal models (~2 h) (Rang et al., 1999), results from this model guided us to select parameters for a second model (SI Materials and Methods). The second model incorporated intestinal wall-associated growth and enabled fitting of the empirical data by assuming three orders of magnitude higher intestinal-wall affinity for UC1CIT-ii compared with UC1CIT-i (Fig. A1-3F and Fig. S3). In addition, to avoid rapid washout of UC1CIT-i, its maximum growth rate had to be doubled relative to UC1CIT-ii and the maximum growth rate of wall-adherent cells had to be lowered by an order of magnitude relative to luminal cells. Because these models were built upon a small amount of data, they are inherently limited in their ability to explain the Citrobacter strain behavior. However, they do strongly suggest that the strain shifts are not the result of random fluctuations. Regardless of whether the growth rates and intestinal niches differ, these Citrobacter strains are distinct in their ability to persist in, and interact with, the human host. The availability of genomic data for both strains provides the opportunity to identify possible metabolic characteristics upon which their physiological and ecological divergence is founded.

A prominent form of variation that differentiated the two UC1CIT strains involved insertions and deletions in intergenic regions (Fig. A1-4 and Table S9 in Dataset S2). In most of the 31 observed cases, intergenic regions differed in length between the strains by >10% and in most cases differed by ≥ 30%. Most variable intergenic segments were flanked by gene sequences that were nearly identical in both strains. Transcriptional regulators [25% of cases; e.g., the LexA repressor, and the NanR regulator of fimbrial adhesins previously shown to be affected by sequence variation (Sohanpal et al., 2004)] and transporters (30% of cases) were common among the flanking genes. We identified strong predicted secondary structure for many divergent intergenic regions and shared sequence similarity with known E. coli sRNAs (Fig. S4).

An illustration presenting a Citrobacter UC1CIT genomic overview


(A) Citrobacter UC1CIT genomic overview. A larger version of this figure is included as Fig. S9. (a) Outside circle represents the ten contigs of the UC1CIT-i genome. Coloring indicates read temporal distribution clusters of the contigs condensed during (more...)

Hotspots of sequence variation that differentiated the UC1CIT strains (mostly substitutions rather than sequence insertions/deletions) also occurred within genes involved in transport, regulation, motility, cell-surface composition, carbohydrate metabolism, virulence, and stress response (Tables S6 and S10 in Dataset S2). Sequence polymorphisms that could potentially affect pathogenicity included the misL-like gene (autotransporter), fimbrial proteins, and a polysaccharide antigen-chain regulator. Interestingly, a large gene encoding RatA, believed to promote intestinal colonization, was a hotspot for microdiversity and was found to be absent in both Citrobacter sp. 30_2 and C. koseri. In Salmonella Typhimurium, RatB (and ShdA, see below) are associated blarwith cecal colonization and fecal shedding, and the gene encoding this protein exhibits strain-associated sequence variation in the form of variable-number direct repeats (Kingsley et al., 2003). If RatA were associated with similar phenotypes, then sequence variation between the two strains could explain differences in niche-partitioning and fecal abundance. We also observed unusually high amino acid sequence divergence in lateral flagellar genes between the UC1CIT strains, which could impact interactions with host cell surfaces (Table S7 in Dataset S2) (Lüneberg et al., 1998). High divergence between the UC1CIT strains in both copies of the gene encoding carbonic anhydrase is also notable because this gene is involved in pH homeostasis and has been identified as a colonization factor in some pathogens (Bury-Moné et al., 2008).

Finally, gene content differentiated the UC1CIT strains (Tables S6, S11, and S12 in Dataset S2). Although many strain-specific genes were clearly associated with phage, several may confer specific metabolic traits. Potentially important genes that were found in both C1CIT-i and Citrobacter sp. 30_2 but not in UC1CIT-ii encoded (i) ShdA, a large virulence protein that is part of a pathogenicity island in Salmonella Typhimurium and essential for successful intestinal colonization (Kingsley et al., 2003); (ii) the inner membrane protein YjfL; (iii) a permease specific for transport of products of pectinolysis (KdgT); (iv) a cluster of four proteins involved in cyclic nucleotide metabolism; (v) fimbrial proteins; (vi) a cluster of 13 proteins involved in phenylacetate degradation; and (vii) genes involved in lipopolysaccharide and polysaccharide/O antigen biosynthesis (abequose). Genes unique to UC1CIT-ii included many fimbrial genes, and genes enabling fructose and other sugar import, streptomycin 3 biosynthesis, and acetoacetate metabolism.

In summary, comparative genomic analyses of the UC1CIT strains highlight metabolic and host interaction traits with the potential to influence strain ecology (Fig. A1-4B). The observation that both regulatory genes and large intergenic regions are hotspots for sequence divergence indicates that one basis for physiological differentiation involves gene regulation, consistent with prior studies implicating regulation as an evolutionary mechanism underlying early ecological differentiation (Denef et al., 2010; Konstantinidis et al., 2009).


The Enterococcus population increased in abundance during the third phase of colonization (Figs. A1-1 and A1-2). The 16S rRNA gene sequence of strain UC1ENC (from our data) is identical to those of several E. faecalis isolates. UC1ENC shares 98.7% AAI with E. faecalis V583 (Paulsen et al., 2003). We mapped the UC1ENC contigs and reads to the V583 genome and recovered ~81% of the latter (Fig. S5 and Table S14 in Dataset S1). The genome size is similar to that of E. faecalis T3 and T11 [available in high-quality draft (Palmer et al., 2010)]. Absence of multiple UC1ENC contigs covering the same genomic region and low SNP frequency indicated that only one strain was present (Fig. S5).

We compared the sequences of seven UC1ENC genes to sequences of genes used in MLST analyses of clinical isolates (, and found that UC1ENC was identical at all seven MLST loci to a sequence type 179, the profile of an isolate recovered from a hospitalized patient’s blood sample in The Netherlands. Furthermore, six out of seven loci were identical to sequence type 16 from an isolate found in a Norwegian infant’s fecal sample (Solheim et al., 2009). Consistent with physiological characterization of the latter isolate, we found genes linked to antibiotic transport or modification and genes encoding virulence factors including collagen-binding adhesin, aggregation substance, enterococcal surface protein, gelatinase (gelE), and cytolysin (Solheim et al., 2009). Additional predicted virulence factors included an exfoliative toxin A and a serine protease known to be transcribed with gelE (Fisher and Phillips, 2009). Comparison with the V583 genome revealed the absence in UC1ENC of the mobile element containing the vancomycin resistance genes (except for vanZ), as well as small sections of the pathogenicity island and most of the plasmid regions and prophages (Fig. S5).

Mobile Elements and Minor Bacterial Populations

Manual curation allowed for genomic reconstruction of a Citrobacter plasmid distinct from the above-mentioned megaplasmid of Citrobacter sp. 30_2, except for two shared regions encoding arsenate and Cu/Ag resistance (~85% AAI). Unlike the UC1CIT plasmid, the putative Citrobacter sp. 30_2 megaplasmid encodes tellurite resistance genes, which have been speculated to confer protection against mammalian host defenses (e.g., by counteracting toxic substances produced by macrophages) (Taylor, 1999). The UC1CIT plasmid (~1.4 plasmid copies per cell) has two variants that differ slightly in gene content and have read distributions across the libraries matching the UC1CIT-i and UC1CIT-ii strains, suggesting that they are strain-specific (Table S4 in Dataset S1). Several phage-like contigs were also recovered, and some displayed boom-and-bust dynamics, indicative of a lytic phage. We also reconstructed two plasmids and two phage of Enterococcus with fluctuating copy numbers (Fig. S6 and Table S4 in Dataset S1). No plasmids or phages were linked to the Serratia population. Low-abundance bacterial populations were genomically sampled as well. As predicted by the daily 16S rRNA screening (Fig. A1-1), genomic sequence-abundance data suggest that Pseudomonas peaked around day 10, whereas Enterobacter peaked on day 16, and the Klebsiella population fluctuated over time (Fig. A1-2 and Fig. S6). Several mobile elements have dynamics corresponding to the minor Klebsiella and Enterobacter populations and may derive from them (Fig. S6).

We performed a community-level analysis of functional potential using genomic information from all populations (Fig. S7). This analysis involved comparison of the microbiome of the preterm infant studied here to the core human microbiome (Qin et al., 2010). Most of the core adult orthologous groups missing from the UC1 infant communities have poorly characterized and unknown functions. There is also a depletion of functions related to carbohydrate metabolism in the infant studied, perhaps because of differences in diet and species composition, with a notable absence of lineages typical of adults from the phyla Firmicutes, Bacteroidetes, and Actinobacteria.


Attempts to correlate gut microbial community structure with onset of disease in premature infants have yielded conflicting results. For example, in some studies, infants with and without necrotizing enterocolitis (NEC) harbored similar species, whereas in other studies samples from infants with this disease were enriched for a particular species (e.g., Clostridium perfringens) or a particular phylum (e.g., Proteobacteria) (de la Cochetiere et al., 2004; Wang et al., 2009). In a recent study, Citrobacter was detected in fecal samples from three of four infants with NEC, but in none of the 17 control samples (Mshvildadze et al., 2010). Although it remains possible that Citrobacter is a causative agent for NEC, its presence in samples from the unaffected infant in this study highlights the difficulty in connecting a specific bacterium to disease.

We infer from the results of this study that substantial shifts in Citrobacter strain abundances arise as a result of strain-specific physiology, despite a level of sequence similarity that would typically result in classification of these species as functionally comparable. Given the differences in genetic, especially pathogenic, potential among the otherwise closely related Citrobacter strains reported here, it is perhaps not surprising that medical comparisons at the species or higher level are often inconclusive. The intriguing differences between the UC1CIT strains in size and sequence of a subset of intergenic regions with similarity to small regulatory RNAs, as well as sequence divergence in regulatory genes emphasize the understudied importance of the evolution of gene expression in strain ecology (Denef et al., 2010).

Application of our approach to more complex microbial communities is feasible if organisms of interest within those communities can be deeply sampled, an objective that can be achieved with current platforms for high throughput sequencing. In fact, a recent study of adult gut communities that used ~10 times more sequencing than did our study succeeded in deeply sampling several populations (Turnbaugh et al., 2010). Thus, ultimately, strain-resolved community genomic approaches can provide the resolution needed for appropriate diagnosis and treatment of a range of microbial community associated conditions.

Materials and Methods

Sample Collection

The protocol for sample collection and processing was approved by the Institutional Review Board of The University of Chicago (IRB #15895A). The sampling method involved manual perineal stimulation with a lubricated cotton swab, which induced prompt defecation. Samples were placed at −80 °C within 10 min.

Sequence Analysis of 16S rRNA Genes

Bacterial 16S rRNA genes were amplified using the broad-range bacterial primers 8–27F and 788–806R. Sequences were processed using the QIIME software package (Caporaso et al., 2010) (SI Materials and Methods, Fig. S1, and Table S1). Fecal 16S rRNA gene sequences from previous studies were obtained directly from GenBank or provided by the authors. Pairwise UniFrac distances were calculated and subjected to principal coordinates analysis (SI Materials and Methods).

Metagenomic Data Analyses

Sequencing reads from the four libraries were coassembled using Newbler (GSassembler v. 2.0.01; Roche) after removal of replicated reads (SI Materials and Methods). We annotated contigs larger than 1,500 bp with an in-house annotation pipeline. Sequence bin assignments were based on a combination of manual assembly curation, blastn, blastp, GC%, sequencing depth, SNP density, and emergent self-organizing maps (eSOM) based on tetranucleotide frequency in combination with a K-means clustering of the temporal profiles of the reads of each contig (SI Materials and Methods). In cases of ambiguity, contigs were assigned to a higher phylogenetic category. Contigs of virus and plasmid origin were primarily identified based on boom-and-bust dynamics deduced from read temporal profiles, colocalization with plasmid/phage reference genome fragments on the eSOM map, and functional annotation information. Contigs between 500 and 1,500 bp were assigned to genomic bins based on an approach similar to that used for the large contigs, except for the use of eSOM projection. Contigs smaller than 500 nt that were not incorporated during manual assembly curation were not further analyzed.

Assemblies for the dominant bacterial, viral, and plasmid populations were manually curated in Consed (Gordon et al., 1998). Sequences that matched the human genome (blastn e-value cutoff of 1e−35) were removed from the dataset. For each Citrobacter contig, sequence types were identified based on SNP patterns and separated for downstream analyses in Strainer (Eppley et al., 2007). Details on the straining process and identification of variation hotspots is described in SI Materials and Methods. Modeling of Citrobacter strain dynamics relied on a simplified model of interstrain competition within the colon, assuming chemostat dynamics (Ballyk et al., 2001) (SI Materials and Methods and Fig. S6). The ORFs predicted on all contigs >500 bp were contrasted to the 4,055 core adult microbiome orthologous groups by blastp analysis using the same parameters and database used by Qin et al. (Qin et al., 2010).


The authors thank the Sanger Institute for S. marcescens genome data access; Dr. J. Raes for details on the adult core metagenome dataset; Dr. V. Mai for sharing 16S rRNA sequence data; Dr. C. Fischer for help with MatLab simulations; and C. Sun, N. Justice and Dr. C. Miller for comments on the manuscript. This work was supported in part by the Surgical Infection Society and the March of Dimes Foundation research Grant 5-FY10-103 (to M.J.M.), Department of Energy Genomic Science program Grant DE-FG02-05ER64134 (to J.F.B.), a Walter V. and Idun Berry Postdoctoral Fellowhip (to E.K.C.), National Institutes of Health Pioneer Award DP1OD000964 (to D.A.R.), and by the National Institute of Allergy and Infectious Diseases, National Institutes of Health, and Department of Health and Human Services (Contracts HHSN27220090018C and HHSN266200400001C; Broad Institute Citrobacter sp. 30_2 sequencing). D.A.R. is supported by the Thomas C. and Joan M. Merigan Endowment at Stanford University.


  • Ballyk MM, Jones DA, Smith HL. Microbial competition in reactors with wall attachment. Microb Ecol. 2001;41:210–221. [PubMed: 11391459]
  • Barnhart MM, Chapman MR. Curli biogenesis and function. Annu Rev Microbiol. 2006;60:131–147. [PMC free article: PMC2838481] [PubMed: 16704339]
  • Bury-Moné S, et al. Roles of alpha and beta carbonic anhydrases of Helicobacter pylori in the urease-dependent response to acidity and in colonization of the murine gastric mucosa. Infect Immun. 2008;76:497–509. [PMC free article: PMC2223474] [PubMed: 18025096]
  • Campos E, et al. The yiaKLX1X2PQRS and ulaABCDEFG gene systems are required for the aerobic utilization of L-ascorbate in Klebsiella pneumoniae strain 13882 with L-ascorbate-6-phosphate as the inducer. J Bacteriol. 2008;190:6615–6624. [PMC free article: PMC2566198] [PubMed: 18708499]
  • Caporaso JG, et al. QIIME allows analysis of high-throughput community sequencing data. Nat Methods. 2010;7:335–336. [PMC free article: PMC3156573] [PubMed: 20383131]
  • Chou HC, et al. Isolation of a chromosomal region of Klebsiella pneumonia associated with allantoin metabolism and liver infection. Infect Immun. 2004;72:3783–3792. [PMC free article: PMC427404] [PubMed: 15213119]
  • Costello EK, et al. Bacterial community variation in human body habitats across space and time. Science. 2009;326:1694–1697. [PMC free article: PMC3602444] [PubMed: 19892944]
  • de la Cochetiere MF, et al. Early intestinal bacterial colonization and necrotizing enterocolitis in premature infants: The putative role of Clostridium. Pediatr Res. 2004;56:366–370. [PubMed: 15201403]
  • Denef VJ, et al. Proteogenomic basis for ecological divergence of closely related bacteria in natural acidophilic microbial communities. Proc Natl Acad Sci USA. 2010;107:2383–2390. [PMC free article: PMC2823883] [PubMed: 20133593]
  • Dethlefsen L, McFall-Ngai M, Relman DA. An ecological and evolutionary perspective on human-microbe mutualism and disease. Nature. 2007;449:811–818. [PubMed: 17943117]
  • Doran TI. The role of Citrobacter in clinical disease of children: Review. Clin Infect Dis. 1999;28:384–394. [PubMed: 10064257]
  • Eckburg PB, et al. Diversity of the human intestinal microbial flora. Science. 2005;308:1635–1638. [PMC free article: PMC1395357] [PubMed: 15831718]
  • Eppley JM, Tyson GW, Getz WM, Banfield JF. Strainer: Software for analysis of population variation in community genomic datasets. BMC Bioinformatics. 2007;8:398. [PMC free article: PMC2110895] [PubMed: 17941997]
  • Fisher K, Phillips C. The ecology, epidemiology and virulence of Enterococcus. Microbiology. 2009;155:1749–1757. [PubMed: 19383684]
  • Gavín R, et al. Lateral flagella of Aeromonas species are essential for epithelial cell adherence and biofilm formation. Mol Microbiol. 2002;43:383–397. [PubMed: 11985716]
  • Gewolb IH, Schwalbe RS, Taciak VL, Harrison TS, Panigrahi P. Stool microflora in extremely low birthweight infants. Arch Dis Child Fetal Neonatal Ed. 1999;80:F167–F173. [PMC free article: PMC1720923] [PubMed: 10212075]
  • Gordon D, Abajian C, Green P. Consed: A graphical tool for sequence finishing. Genome Res. 1998;8:195–0202. [PubMed: 9521923]
  • Hanage WP, Fraser C, Tang J, Connor TR, Corander J. Hyper-recombination, diversity, and antibiotic resistance in pneumococcus. Science. 2009;324:1454–1457. [PubMed: 19520963]
  • Hooper LV, et al. Molecular analysis of commensal host-microbial relationships in the intestine. Science. 2001;291:881–884. [PubMed: 11157169]
  • Hooper LV, Midtvedt T, Gordon JI. How host-microbial interactions shape the nutrient environment of the mammalian intestine. Annu Rev Nutr. 2002;22:283–307. [PubMed: 12055347]
  • Kingsley RA, et al. Molecular and phenotypic analysis of the CS54 island of Salmonella enterica serotype typhimurium: Identification of intestinal colonization and persistence determinants. Infect Immun. 2003;71:629–640. [PMC free article: PMC145368] [PubMed: 12540539]
  • Koenig JE, et al. Succession of microbial consortia in the developing infant gut microbiome. Proc Natl Acad Sci USA. 2010 [PMC free article: PMC3063592] [PubMed: 20668239] [CrossRef]
  • Konstantinidis KT, et al. Comparative systems biology across an evolutionary gradient within the Shewanella genus. Proc Natl Acad Sci USA. 2009;106:15909–15914. [PMC free article: PMC2747217] [PubMed: 19805231]
  • Lüneberg E, Glenn-Calvo E, Hartmann M, Bär W, Frosch M. The central, surface-exposed region of the flagellar hook protein FlgE of Campylobacter jejuni shows hypervariability among strains. J Bacteriol. 1998;180:3711–3714. [PMC free article: PMC107344] [PubMed: 9658019]
  • Mackie RI, Sghir A, Gaskins HR. Developmental microbial ecology of the neonatal gastrointestinal tract. Am J Clin Nutr. 1999;69:1035S–1045S. [PubMed: 10232646]
  • Magne F, et al. Low species diversity and high interindividual variability in faeces of preterm infants as revealed by sequences of 16S rRNA genes and PCR-temporal temperature gradient gel electrophoresis profiles. FEMS Microbiol Ecol. 2006;57:128–138. [PubMed: 16819956]
  • Merino S, Shaw JG, Tomás JM. Bacterial lateral flagella: An inducible flagella system. FEMS Microbiol Lett. 2006;263:127–135. [PubMed: 16978346]
  • Millar MR, et al. Application of 16S rRNA gene PCR to study bowel flora of preterm infants with and without necrotizing enterocolitis. J Clin Microbiol. 1996;34:2506–2510. [PMC free article: PMC229306] [PubMed: 8880510]
  • Morgan JL, Darling AE, Eisen JA. Metagenomic sequencing of an in vitrosimulated microbial community. PLoS ONE. 2010;5:e10209. [PMC free article: PMC2855710] [PubMed: 20419134]
  • Mshvildadze M, et al. Intestinal microbial ecology in premature infants assessed with nonculture-based techniques. J Pediatr. 2010;156:20–25. [PMC free article: PMC3628625] [PubMed: 19783002]
  • Oliver KM, Degnan PH, Hunter MS, Moran NA. Bacteriophages encode factors required for protection in a symbiotic mutualism. Science. 2009;325:992–994. [PMC free article: PMC5473335] [PubMed: 19696350]
  • Palmer C, Bik EM, DiGiulio DB, Relman DA, Brown PO. Development of the human infant intestinal microbiota. PLoS Biol. 2007;5(7):e177. [PMC free article: PMC1896187] [PubMed: 17594176]
  • Palmer KL, et al. High-quality draft genome sequences of 28 Enterococcus sp. isolates. J Bacteriol. 2010;192:2469–2470. [PMC free article: PMC2863476] [PubMed: 20207762]
  • Paulsen IT, et al. Role of mobile DNA in the evolution of vancomycin-resistant Enterococcus faecalis. Science. 2003;299:2071–2074. [PubMed: 12663927]
  • Qin J, et al. A human gut microbial gene catalogue established by metagenomic sequencing. Nature. 2010;464:59–65. [PMC free article: PMC3779803] [PubMed: 20203603]
  • Rang CU, et al. Estimation of growth rates of Escherichia coli BJ4 in streptomycin-treated and previously germfree mice by in situ rRNA hybridization. Clin Diagn Lab Immunol. 1999;6:434–436. [PMC free article: PMC103738] [PubMed: 10225851]
  • Sandaa R-A, et al. Viral control of bacterial biodiversity—Evidence from a nutrient-enriched marine mesocosm experiment. Environ Microbiol. 2009;11:2585–2597. [PubMed: 19558511]
  • Schwiertz A, et al. Development of the intestinal bacterial composition in hospitalized preterm infants in comparison with breast-fed, full-term infants. Pediatr Res. 2003;54:393–399. [PubMed: 12788986]
  • Sohanpal BK, El-Labany S, Lahooti M, Plumbridge JA, Blomfield IC. Integrated regulatory responses of fimB to N-acetylneuraminic (sialic) acid and GlcNAc in Escherichia coli K-12. Proc Natl Acad Sci USA. 2004;101:16322–16327. [PMC free article: PMC526197] [PubMed: 15534208]
  • Solheim M, Aakra A, Snipen LG, Brede DA, Nes IF. Comparative genomics of Enterococcus faecalis from healthy Norwegian infants. BMC Genomics. 2009;10:194. [PMC free article: PMC2680900] [PubMed: 19393078]
  • Taghavi S, et al. Genome survey and characterization of endophytic bacteria exhibiting a beneficial effect on growth and development of poplar. Appl Environ Microbiol. 2008;75:748–757. [PMC free article: PMC2632133] [PubMed: 19060168]
  • Taylor DE. Bacterial tellurite resistance. Trends Microbiol. 1999;7:111–115. [PubMed: 10203839]
  • Trosvik P, Stenseth NC, Rudi K. Convergent temporal dynamics of the human infant gut microbiota. ISME J. 2010;4:151–158. [PubMed: 19710708]
  • Turnbaugh PJ, et al. Organismal, genetic, and transcriptional variation in the deeply sequenced gut microbiomes of identical twins. Proc Natl Acad Sci USA. 2010;107:7503–7508. [PMC free article: PMC2867707] [PubMed: 20363958]
  • Voelz A, et al. Outbreaks of Serratia marcescens in neonatal and pediatric intensive care units: Clinical aspects, risk factors and management. Int J Hyg Environ Health. 2010;213:79–87. [PubMed: 19783209]
  • Wagener S, Shankar KR, Turnock RR, Lamont GL, Baillie CT. Colonic transit time—What is normal? J Pediatr Surg. 2004;39:166–169. [PubMed: 14966733]
  • Wang Y, et al. 16S rRNA gene-based analysis of fecal microbiota from preterm infants with and without necrotizing enterocolitis. ISME J. 2009;3:944–954. [PMC free article: PMC2713796] [PubMed: 19369970]



, , and .


11 Centre for Immunity, Infection, and Evolution, School of Biological Sciences, University of Edinburgh, West Mains Road, Edinburgh EH9 3JT, UK.

Standard virulence evolution theory assumes that virulence factors are maintained because they aid parasitic exploitation, increasing growth within and/or transmission between hosts. An increasing number of studies now demonstrate that many opportunistic pathogens (OPs) do not conform to these assumptions, with virulence factors maintained instead because of advantages in non-parasitic contexts. Here we review virulence evolution theory in the context of OPs and highlight the importance of incorporating environments outside a focal virulence site. We illustrate that virulence selection is constrained by correlations between these external and focal settings and pinpoint drivers of key environmental correlations, with a focus on generalist strategies and phenotypic plasticity. We end with a summary of key theoretical and empirical challenges to be met for a fuller understanding of OPs.

Opportunistic Pathogens and a Challenge to Virulence Evolution Theory

The study of infectious diseases has become a major focus within evolutionary biology; however, remarkably little attention has been paid to an extremely broad class of pathogens, the opportunists. This oversight stems from the theoretical convenience of treating hostparasite interactions as closed systems in which a single, obligate pathogen specialises on a single host (Alizon et al., 2009; Anderson and May, 1982). Most pathogens actually fail to meet these assumptions, with many coexisting relatively peacefully with their human host (i.e., they are not obligately pathogenic) or even exploiting an entirely different environment outside of human hosts (Woolhouse et al., 2001).

Opportunistic pathogens (OPs) are typically characterised in the medical literature as organisms that can become pathogenic following a perturbation to their host (e.g., disease, wound, medication, prior infection, immunodeficiency, and ageing). These opportunists can emerge from among the ranks of normally commensal symbionts (e.g., Streptococcus pneumoniae and Staphylococcus aureus) or from environmentally acquired microbes (e.g., Pseudomonas aeruginosa and Burkholderia cepacia). Many more pathogens are recognised as opportunists in the sense that although they regularly cause disease in health humans, they are also zoonotic and exploit numerous other hosts (e.g. Bacillus anthracis and rabies virus).

We propose a broad and simple definition of OPs: non-obligate and/or non-specialist parasites of a focal host. Thus, if the classic assumptions (obligate parasite and specialist on one host) of virulence evolution theory fail, we have an OP. In Table A2-1 we outline, with examples, how the combinations of these two conditions give us a classification of OPs into commensal opportunists, environmental opportunists and parasite opportunists (or zoonoses).

TABLE A2-1. An Ecological Classification of Pathogens with Representative Examples.


An Ecological Classification of Pathogens with Representative Examples.

Given the failure of the two key assumptions of classical virulence evolution theory (Table A2-1), what can this body of theory tell us about OPs? For some parasites, the strict failure to meet these assumptions might not be important if the approximation is reasonable in practice, for instance if humans are the major source of parasite transmission (to any host) and the parasite does not routinely enter a commensal stage (e.g., Salmonella enterica among humans living in dense and unsanitary conditions). In these cases, standard predictions from virulence evolution theory may still apply, such as a trade-off between transmission and virulence (Fraser et al., 2007; Mackinnon and Read, 1999). Yet as the biological reality moves further away from these assumptions, we are left only with the prediction that multi-environment opportunists are likely to experience nonoptimal virulence in a given host (Bull and Ebert, 2008; Gandon, 2004). However, ecological and evolutionary theory offers an increasing number of insights into other key features of many opportunists, in particular plasticity and generalism. All human OPs are generalists in the sense that they are able to grow in more than one environment. In addition, many OPs display remarkable phenotypic plasticity, being able to modify phenotypic expression as a function of their changing environmental context.

If we can understand microbial plasticity and generalism, we can understand why opportunistic bugs become pathogens, when they are likely to do this and how we can interfere with their plastic responses to control their virulence in a sustainable manner. In this review, we aim to develop a general and integrative framework for the understanding and management (on ecological and evolutionary time scales) of opportunistic pathogens.

What is Virulence, and Why Damage Your Host?

For population biologists, virulence is typically the increase in host mortality resulting from parasite infection (Anderson and May, 1982). Although this is an explicit and measurable quantity, it ignores many aspects of parasite biology that cause harm without death (Bull, 1994; Casadevall and Pirofski, 1999; Gandon, 2004; Read, 1994). For medical microbiologists, virulence is understood as harm or morbidity to the host, but the focus is on the mechanistic basis of harm, such as identifying virulence determinants or factors (VFs). VFs are typically defined as pathogen components whose loss specifically impairs virulence but not viability (in rich media); classic bacterial examples include toxins, exoenzymes, adhesins, and secretion systems (Brogden et al., 2007).

VFs can be mechanistically complex and therefore are presumably products of natural selection. However, the nature of selection for maintaining and strengthening VFs remains controversial. Levin and Svanborg Edén made an important distinction between direct and coincidental selection for VFs (Levin and Svanborg Edén, 1990). Under direct selection (by far the most influential model), expression of the VF is correlated with the ability of the parasite to exploit and/or be transmitted from the host; in other words, parasitic VF expression (and consequent costs in terms of host mortality) is rewarded by some benefit. These benefits can either be gained through transmission (Alizon et al., 2009; Anderson and May, 1982) or through within-host growth Levin and Bull, 1994). This dichotomy, highlighting the importance of multiple scales in disease processes (Mideo et al., 2008), forms the basis of the standard evolutionary view of virulence.

By contrast, purely coincidental selection argues that VF expression is not positively correlated with any measure of parasitic success within the focal host; in other words, there are no benefits in the parasitic context. VFs are then fascinating spandrels (Gould and Lewontin, 1979; Levin, 1996), byproducts of selection for adaptations not related to disease. The mystery of why a VF exists must then be answered elsewhere in the parasite life history, with VF maintenance caused by some benefit in an extra-host habitat or a within-host habitat in which the organism does not cause disease. In this case, can we still make general statements about the dynamics (ecological and evolutionary) of virulence, or are we relegated to case-by-case considerations (Ebert and Bull, 2003; Levin and Svanborg Edén, 1990)? To begin to answer this challenge, we develop a descriptive model framework to outline how four key selective pressures (coincidental, colonisation, export, and within-host) can combine to shape the evolutionary dynamics of VFs.

Virulence Factor Dynamics Across Multiple Environments

A characteristic of all opportunistic pathogens is compartmentalisation into environments where they cause disease (e.g., burn wounds for P. aeurignosa and the circulatory system for S. pneumoniae) and environments where they do not (soil and nasopharynx, respectively). This compartmentalisation can be within a focal host (in particular, either side of mucosal barriers) or between a focal host and another environment (e.g., animal reservoirs vs human hosts). To conceptualise this split, we divide the world of a microbe into two compartments: the virulence compartment V, the sensitive parts of a focal host where microbial VFs result in disease symptoms; and the asymptomatic compartment A, which is everywhere else it can grow ( Figure A2-1a, schema inspired by [Margolis and Levin, 2007]). In Figure A2-1b–d, we illustrate how four selective pressures (coincidental, colonisation, export and within-host; the four arrows in Figure A2-1a) can combine to recover existing theories on the evolution of virulence (Box A2-1).

A diagram showing ecological and evolutionary dynamics of virulence factors across two growth environments


Ecological and evolutionary dynamics of virulence factors across two growth environments. NA and NV represent bacterial densities in the asymptomatic and virulence sites, respectively. Arrows represent demographic processes of growth (g, r) and transmission (more...)

Box Icon

BOX A2-1

Ecological and Evolutionary Dynamics in Structured Environments. In Figure A2-1a, we introduce a simple diagrammatic sketch of bacterial population growth in two compartments, one incurring virulence (site V) and the other asymptomatic (site A). To track (more...)

The analysis in Box A2-1 illustrates that if the within-host and transmission (colonisation and export) pressures select against VFs, then there are no counter-vailing benefits of VFs during host exploitation or transmission (and VFs will be less common in virulence compartments, variation permitting): we are left with purely coincidental virulence ( Figure A2-1c). Pure coincidental virulence implies a positive association with damage but not growth or transmission from V. A classic example is found in the soil bacterium Clostridium botulinum: botulinum toxin is an extremely potent virulence factor when introduced into humans, but C. botulinum itself cannot grow in, let alone be transmitted from, humans (Levin and Svanborg Edén, 1990) (thus, humans are an ecological sink [Sokurenko et al., 2006]). The simple formulation in Box A2-1 therefore clarifies how and why some empirical studies may fail to find a selective advantage to VFs in infections (site V). Even without any benefit in site V, selection could favour VF expression, depending on the frequency at which bacteria encounter sites A and V and the relative magnitude of benefits (replication in site A) and costs (growth in site V or movement between A and V). We now use the framework outlined in Figure A2-1 and Box A2-1 to discuss the importance of positive and negative correlations between bacterial environments A and V.


Pure coincidental selection as exemplified by C. botulinum virulence is, however, an extreme situation: coincidental selection can also coexist with positive within-host and/or transmission selection for VFs. In these cases, the VFs are multi-functional. If we assume that environment A is the primary site of adaptation, then we can conclude that selection in site A generates pre-adaptations for the virulent exploitation of site V (i.e., once in site V, the VF confers some advantage in terms of growth or transmission, but evolution of the VF was driven by selection in site A). If, for example, both coincidental and within-host selection favour VF expression, then there is a positive environmental correlation between growth in environment A and that in environment V (the site of virulence, Figure A2-2).

A diagram showing how adaptation to a benign environment can pre-adapt an opportunistic pathogen for virulent growth


Adaptation to a benign environment A can pre-adapt an opportunistic pathogen for virulent growth in V if there is a significant positive association between the properties of environments A and V (i.e., if fast growth in A, g, is correlated with fast (more...)

An emerging paradigm of VF pre-adaptation driven by environmental correlation is the ability of bacteria to generalise mechanisms for resisting protists for use in other situations. Protists are an important class of bacterial predators across diverse environments (including within host-associated microbiotas), and increasing evidence points to the evolution of resistance to protist predation pre-adapting certain environmental microbes for survival and even proliferation within human macrophages (Brüssow, 2007; Cirillo et al., 1999; Harb et al., 2000; Lainhart et al., 2009; Matz and Kjelleberg, 2005; Rasmussen et al., 2005; Steinberg and Levin, 2007). For example, Steinberg and Levin demonstrated that a Shiga toxin VF of Escherichia coli O157:H7 increases survival in the presence of grazing protozoa (Steinberg and Levin, 2007). This result suggests that protozoan predation within ruminants or in the soil may have selected for the VFs that drive pathogenicity and, in particular, export and transmission through Shiga toxin-induced diarrhoea in humans.

Other potential examples of pre-adaptation include selection for capsule carriage (a VF increasing the risk of invasive disease) among pneumococcal strains in the nasopharynx (Lysenko et al., 2010). The most common disease states caused by S. pneumoniae are pneumonia, otitis media and sepsis, and these are not contagious conditions and therefore represent a dead end, especially when the result is rapid demise of the host (Musher, 2003). Rather, transmission occurs from the reservoir of pneumococci residing asymptomatically in the nasopharynx during the organism’s commensal state (Austrian, 1986). However, among the more than 92 types of pneumococci expressing structurally distinct capsular polysaccharides, only a few are potentially virulent (Hausdorff et al., 2000, 2005). So why has the pneumococcus evolved or maintained the capacity for virulent, invasive behaviour through the expression of certain thick capsular polysaccharide coats? The results of Lysenko et al. demonstrate that capsule selection is driven in the nasopharynx by competitive interactions with another commensal, Haemophillus influenzae (Lysenko et al., 2010). While pneumococcal growth is suppressed by H. influenzae, the capsule offers a survival advantage by reducing susceptibility to this suppression. These results also present an important reminder that OPs will often face many distinct non-virulent environments (various A, A′, etc.), such as environments with or without a key predator or competitor (here, H. influenzae).

Lysenko et al. illustrate that growth in a crowded, immunogenic nasopharynx selects for serum resistance, which then pre-adapts S. pneumoniae for growth in blood (Lysenko et al., 2010). Similarly, survival against protists (in soil, say) selects for survival against macrophages (in human hosts) (Brüssow, 2007; Cirillo et al., 1999; Harb et al., 2000; Lainhart et al., 2009; Matz and Kjelleberg, 2005; Rasmussen et al., 2005; Steinberg and Levin, 2007). Both of these examples highlight that shared or similar environmental challenges can shape the potential for new outbreaks, by building positive correlations between environments (Figure A2-2).

Environmental Tradeoffs: Specialism and Plasticity

The examples above describe cases in which selection for VFs may have occurred in a setting outside of infection, but incidentally provides some benefit in terms of transmission or within-host growth (Figure A2-2). Alternatively, the association between growth in A and V can be negative (Figure A2-3). How do OPs deal with such environmental trade-offs? A first possibility is that they do not: the focal lineage continues to adapt to its primary environment A, and in certain V environments, bacteria will be unsuccessful. This would be a reasonable strategy if V environments were infrequently encountered and/or unproductive (C. botulinum is a candidate here). However, if bacteria frequently encounter environments across which the costs of the trade-off are felt (and if sufficient genetic variation exists), then something is likely to give: evolution in the face of an important trade-off can lead to a loss of the trade-off (if the underlying constraint is weak), specialisation or plasticity.

A diagram showing how adaptation to a benign environment can reduce the capacity for virulent growth


Adaptation to a benign environment A can reduce the capacity for virulent growth in V (and vice versa) if there is a significant negative association (trade-off) between growth rates g and r. (A negative selective impact of virulence factor expression (more...)

A common motor of bacterial specialisation is horizontal gene transfer and loss; plasmids and phages shuttle an array of genes conferring local adaptations to heterogeneous environments (Frost et al., 2005; Rankin et al., 2010), including a strikingly large number of VFs (Levin and Svanborg Edén, 1990; Nogueira et al., 2009; Smith, 2001). The acquisition of VFs via horizontal transfer can render harmless bugs more pathogenic, switching (specializing) or even extending (generalizing) their environmental repertoire. Turner et al. posed the question as to whether generalists or specialists would be better able to exploit an entirely novel host type, previously unseen by either line (Turner et al., 2010). In other words, which would make the better OP? They illustrated that specialist RNA viruses (evolved under a single host condition) were able to outperform generalists in specific novel host challenges, highlighting the importance of coincidental (or indirect) selection. However, generalists tended to be more consistent across a range of novel challenges, suggesting that consistency is characteristic of generalists. Generalist phenotypes, whether selected directly or indirectly, result from either increased phenotypic constancy across environmental variation or plasticity (phenotypic switching) (de Visser et al., 2003). For OPs there are many examples of remarkable plasticity that we now discuss.

Plasticity is the ability of an organism to change its phenotype without corresponding changes in genotype. Mechanisms such as altering gene expression can allow an organism to display different phenotypes in different environments (Schlichting and Pigliucci, 1998), and when these responses match the changing environmental requirements (i.e., improve the organism’s fitness in that environment) this is called adaptive phenotypic plasticity. Standard theory for virulence evolution has only recently started to incorporate phenotypic plasticity (Choisy and de Roode, 2010; Taylor et al., 2006), but for OPs this phenomenon is of clear importance.

Bacterial VFs are by definition “optional extras” and are often under regulatory control and are not always on, with expression responsive to both physical (e.g., pH and temperature) and social (e.g., density and diversity) environmental dimensions (Bielecki et al., 2011; Duan and Surette, 2007; Kümmerli et al., 2009; Kümmerli and Brown, 2010). The underlying regulatory machinery is highly complex and variable in extent, with the number of global regulatory sigma factors varying from three in the specialist Helicobacter pylori to 24 in the generalist P. aeruginosa (Dale and Park, 2010). This variation in regulatory investment makes sense in the light of plasticity theory: it is only the challenge of frequent exposure to distinct environments that selects for adaptive phenotypic plasticity, in which case the benefits of adaptive plasticity outweigh the likely costs of the machinery necessary to generate such plasticity (DeWitt et al., 1998).

Although the direct costs and benefits of a complex regulatory machinery are readily appreciated, there is also potential for indirect costs of making “bad decisions” (Figure A2-3), as hinted by recent findings for P. aeruginosa. On initial colonisation of a mammalian host, P. aeruginosa upregulates an array of VFs (Bielecki et al., 2011; Duan and Surette, 2007). However, during subsequent evolution in chronically infected cystic fibrosis patients, many of these VFs are subsequently lost, leading to a reduction in the ability to cause acute disease and mortality (Bragonzi et al., 2009; Smith et al., 2006; Woolhouse et al., 2001). It has been argued that the loss of secreted VFs may be caused by social interactions favouring “cheater” strains that do not pay the costs of collectively useful VF production (Griffin et al., 2004; Jiricny et al., 2010). However, the continued ability of these strains to persist chronically (Bragonzi et al., 2009) suggests the possibility that VFs are redundant in the cystic fibrosis lung, and their initial upregulation was a “bad decision” (alternatively, the benefits of VF expression may change through the course of infection as the infection environment develops).

The causes of some aspects of this decision-making have been brought into closer focus for the P. aeruginosa toxin pyocyanin, expression of which is driven by exposure to N-acetylglucosamine and its polymer peptidoglycan, commonly shed by Gram-positive bacteria (Korgaonkar and Whiteley, 2011). In addition to damaging eukaryotic cells, pyocyanin is a potent antimicrobial, suggesting that N-acetylglucosamine-dependent pyocyanin expression is an antimicrobial mechanism in environments rich in competitors (Korgaonkar and Whiteley, 2011); this may then be triggered inappropriately in the cystic fibrosis lung due to human-derived N-acetylglucosamine. There are also many well-studied examples of global regulation in quorum-sensing and stress responses (like RpoS in many proteobacteria) that strongly impact virulence (Antunes et al., 2010; Dong and Schellhorn, 2010). The impressive and expanding mechanistic understanding of bacterial plasticity (regulatory control) provides a particularly rich arena for evolutionary investigation, with clear importance for applied questions of bacterial control.

Managing Antibiotic Resistance and Virulence

Finally, we turn to the implications of opportunism for parasite control. If most human pathogens are largely shaped by selective pressures outside of disease-causing compartments, then why is antibiotic resistance such a clear and growing problem? A major part of the answer is that for many VFs discussed above, antibiotic resistance genes can confer advantages outside of the context of human medical interventions via resistance to bacterially derived antimicrobial compounds. Consistent with this broader functionality, resistance to a range of antibiotics have been found in ancient DNA from 30,000-year-old permafrost sediments (D’Costa et al., 2011). Nevertheless, antibiotic resistance has spread rapidly in many bacteria since the introduction of antibiotics into medical and farming practice (Palumbi, 2001), indicating that selective pressures are stronger in patients than in nature.

For commensal opportunists, exposure to antibiotics is routine because of their specialisation on human hosts, and therefore the emergence of antibiotic resistance in these species poses little puzzle. By contrast, non-specialists may encounter humans merely as a dynamical “sink” (Sokurenko et al., 2006), and thus human interactions are unlikely to drive the evolution of antibiotic resistance genes among these populations. However, resistance may pose a significant problem in these lineages because of to a mix of innate resistance properties (Poole, 2001) and shuttling of resistance genes by horizontal gene transfer, particularly during chains of human–human transmission (Winstanley et al., 2009).

Interest is now growing in the use of antivirulence drugs that directly target the expression of virulence factors (André and Godelle, 2005; Clatworthy et al., 2007; Defoirdt et al., 2010; Maeda et al., 2011; Mellbye and Schuster, 2011; Rasko and Sperandio, 2010). It has been argued that these drugs will limit the evolution of resistance, because they do not kill or halt the growth of their targets (Clatworthy et al., 2007; Rasko and Sperandio, 2010). How does this claim stand up in the context of OPs? If bacteria only see the drug in V (the virulence site) and the VF is only selected for in A (purely “coincidental selection”), then the drugs have potential: the antivirulence drug in this context will enhance a natural tendency towards virulence attenuation within hosts. However, if bacteria see the drug in their “non-virulent” compartment (A) and/or the VF is positively correlated with transmission, then the risks are far greater. A simple implication is that these drugs will hold more long-term promise for the treatment of environmental opportunists because of greater isolation between compartments A and V.

Concluding Remarks

Although there is a broad range of conceptual models for the evolution of virulence (Figure A2-1, Box A2-1), formal mathematical treatments have focused overwhelmingly on the most tractable subset, the specialist, obligate parasite (Alizon et al., 2009; Anderson and May, 1982; Frank, 1996). Here we argue that this bias has hindered effective evolutionary studies of opportunistic pathogens. The admission of multiple growth environments inevitably makes the mathematics more complex (Box A2-1) (Gandon, 2004; Regoes et al., 2000). More importantly, it also highlights the extent to which biological details matter, with selection on VFs dependent on a complex web of environmental correlations that are only beginning to be picked apart via careful study of bacterial population biology inside and outside of the sites where bacteria cause disease (Brüssow, 2007; Cirillo et al., 1999; Harb et al., 2000; Lainhart et al., 2009; Lysenko et al., 2010; Matz and Kjelleberg, 2005; Rasmussen et al., 2005; Steinberg and Levin, 2007).

Our formal treatment was presented in the context of distinct physical environments (e.g., blood versus mucosa); however, the control of bacterial VF expression in response to contrasting social conditions highlights an even greater complexity and a key theoretical challenge. For instance, many bacteria can discriminate between low- and high-density environments, and even clonal versus polymicrobial conditions, via quorum-sensing mechanisms (Bassler, 1999; Fuqua et al., 1994; Williams et al., 2007) and cues (Korgaonkar and Whitele, 2011). The development and testing of a novel theory integrating the molecular, ecological and evolutionary dynamics of VFs across complex social and physical environments hold real promise for accelerating our understanding of VFs and their potential as targets for evolutionarily robust antivirulence drugs.


We thank Nick Colegrave, Rolf Kümmerli, Stuart West, and Marvin Whiteley’s laboratory for helpful comments, and the Wellcome Trust for funding (Grant No. WT082273).


  • Alizon S, et al. Virulence evolution and the trade-off hypothesis: history, current state of affairs and the future. J Evol Biol. 2009;22:245–259. [PubMed: 19196383]
  • Anderson R, May R. Coevolution of hosts and parasites. Parasitology. 1982;85:411–426. [PubMed: 6755367]
  • Antunes LCM, et al. Quorum sensing in bacterial virulence. Microbiology. 2010;156:2271–2282. [PubMed: 20488878]
  • Austrian R. Some aspects of the pneumococcal carrier state. J. Antimicrob Chemother. 1986;18(Suppl A):35–45. [PubMed: 3745031]
  • Bielecki P, et al. In-vivo expression profiling of Pseudomonas aeruginosa infections reveals niche-specific and strain-independent transcriptional programs. PLoS ONE. 2011;6:e24235. [PMC free article: PMC3171414] [PubMed: 21931663]
  • Bragonzi A, et al. Pseudomonas aeruginosa microevolution during cystic fibrosis lung infection establishes clones with adapted virulence. Am J Respir Crit Care Med. 2009;180:138–145. [PubMed: 19423715]
  • Brogden KA, et al. Virulence Mechanisms of Bacterial Pathogens. 1st edn. ASM Press; 2007.
  • Brüssow H. Bacteria between protists and phages: from antipredation strategies to the evolution of pathogenicity. Mol Microbiol. 2007;65:583–589. [PubMed: 17608793]
  • Bull JJ. Perspective: virulence. Evolution. 1994;48:1423–1437. [PubMed: 28568406]
  • Bull JJ, Ebert D. Invasion thresholds and the evolution of nonequilibrium virulence. Evol Appl. 2008;1:172–182. [PMC free article: PMC3352400] [PubMed: 25567500]
  • Casadevall A, Pirofski L. Host–pathogen interactions: redefining the basic concepts of virulence and pathogenicity. Infect Immun. 1999;67:3703–3713. [PMC free article: PMC96643] [PubMed: 10417127]
  • Choisy M, de Roode JC. Mixed infections and the evolution of virulence: effects of resource competition, parasite plasticity, and impaired host immunity. Am Nat. 2010;175:E105–E118. [PubMed: 20297955]
  • Cirillo JD, et al. Intracellular growth in Acanthamoeba castellanii affects monocyte entry mechanisms and enhances virulence of Legionella pneumophila. Infect Immun. 1999;67:4427–4434. [PMC free article: PMC96761] [PubMed: 10456883]
  • Dale JW, Park SF. Molecular Genetics of Bacteria. 5th edn. Wiley; 2010.
  • D’Costa VM, et al. Antibiotic resistance is ancient. Nature. 2011;477:457–461. [PubMed: 21881561]
  • de Visser JAGM, et al. Perspective: evolution and detection of genetic robustness. Evolution. 2003;57:1959–1972. [PubMed: 14575319]
  • DeWitt TJ, et al. Costs and limits of phenotypic plasticity. Trends Ecol Evol. 1998;13:77–81. [PubMed: 21238209]
  • Dong T, Schellhorn HE. Role of RpoS in virulence of pathogens. Infect Immun. 2010;78:887–897. [PMC free article: PMC2825926] [PubMed: 19948835]
  • Duan K, Surette MG. Environmental regulation of Pseudomonas aeruginosa PAO1 Las and Rhl quorum-sensing systems. J Bacteriol. 2007;189:4827–4836. [PMC free article: PMC1913434] [PubMed: 17449617]
  • Ebert D, Bull JJ. Challenging the trade-off model for the evolution of virulence: is virulence management feasible? Trends Microbiol. 2003;11:15–20. [PubMed: 12526850]
  • Fraser C, et al. Variation in HIV-1 set-point viral load: epidemiological analysis and an evolutionary hypothesis. Proc Natl Acad Sci USA. 2007;104:17441–17446. [PMC free article: PMC2077275] [PubMed: 17954909]
  • Frost LS, et al. Mobile genetic elements: the agents of open source evolution. Nat Rev Microbiol. 2005;3:722–732. [PubMed: 16138100]
  • Gandon S. Evolution of multihost parasites. Evolution. 2004;58:455–469. [PubMed: 15119430]
  • Gould SJ, Lewontin RC. The spandrels of San Marco and the Panglossian paradigm: a critique of the adaptationist programme. Proc R Soc Lond Ser B Biol Sci. 1979;205:581–598. [PubMed: 42062]
  • Griffin AS, et al. Cooperation and competition in pathogenic bacteria. Nature. 2004;430:1024–1027. [PubMed: 15329720]
  • Harb OS, et al. From protozoa to mammalian cells: a new paradigm in the life cycle of intracellular bacterial pathogens. Environ Microbiol. 2000;2:251–265. [PubMed: 11200426]
  • Hausdorff WP, et al. Which pneumococcal serogroups cause the most invasive disease: implications for conjugate vaccine formulation and use, part I. Clin Infect Dis. 2000;30:100–121. [PubMed: 10619740]
  • Hausdorff WP, et al. Epidemiological differences among pneumococcal serotypes. Lancet Infect Dis. 2005;5:83–93. [PubMed: 15680778]
  • Jiricny N, et al. Fitness correlates with the extent of cheating in a bacterium. J Evol Biol. 2010;23:738–747. [PubMed: 20210835]
  • Korgaonkar AK, Whiteley M. Pseudomonas aeruginosa enhances production of an antimicrobial in response to Nacetylglucosamine and peptidoglycan. J Bacteriol. 2011;193:909–917. [PMC free article: PMC3028681] [PubMed: 21169497]
  • Kümmerli R, Brown SP. Molecular and regulatory properties of a public good shape the evolution of cooperation. Proc Natl Acad Sci USA. 2010;107:18921–18926. [PMC free article: PMC2973908] [PubMed: 20944065]
  • Kümmerli R, et al. Phenotypic plasticity of a cooperative behaviour in bacteria. J Evol Biol. 2009;22:589–598. [PubMed: 19170825]
  • Lainhart W, et al. Shiga toxin as a bacterial defense against a eukaryotic predator, Tetrahymena thermophila. J. Bacteriol. 2009;191:5116. [PMC free article: PMC2725575] [PubMed: 19502393]
  • Levin BR. The evolution and maintenance of virulence in microparasites. Emerg Infect Dis. 1996;2:93–102. [PMC free article: PMC2639826] [PubMed: 8903208]
  • Levin BR, Bull JJ. Short-sighted evolution and the virulence of pathogenic microorganisms. Trends Microbiol. 1994;2:76–81. [PubMed: 8156275]
  • Levin BR, Svanborg Edén C. Selection and evolution of virulence in bacteria: an ecumenical excursion and modest suggestion. Parasitology. 1990;100(Suppl):S103–S115. [PubMed: 2235060]
  • Lysenko ES, et al. Within-host competition drives selection for the capsule virulence determinant of Streptococcus pneumoniae. Curr Biol. 2010;20:1222–1226. [PMC free article: PMC2913241] [PubMed: 20619820]
  • Mackinnon MJ, Read AF. Genetic relationships between parasite virulence and transmission in the rodent malaria Plasmodium chabaudi. Evolution. 1999;53:689–70. [PubMed: 28565637]
  • Margolis E, Levin BR. The evolution of bacteria–host interactions: virulence and the immune over-response. In: Evolutionary Biology of Bacterial and Fungal Pathogens. ASM Press; 2007. pp. 3–12.
  • Matz C, Kjelleberg S. Off the hook – how bacteria survive protozoan grazing. Trends Microbiol. 2005;13:302–307. [PubMed: 15935676]
  • Mideo N, et al. Linking within- and between-host dynamics in the evolutionary epidemiology of infectious diseases. Trends Ecol Evol. 2008;23:511–517. [PubMed: 18657880]
  • Musher DM. How contagious are common respiratory tract infections? N Engl J Med. 2003;348:1256–1266. [PubMed: 12660390]
  • Nogueira T, et al. Horizontal gene transfer of the secretome drives the evolution of bacterial cooperation and virulence. Curr Biol. 2009;19:1683–1691. [PMC free article: PMC2773837] [PubMed: 19800234]
  • Palumbi SR. Humans as the world’s greatest evolutionary force. Science. 2001;293:1786. [PubMed: 11546863]
  • Poole K. Multidrug efflux pumps and antimicrobial resistance in Pseudomonas aeruginosa and related organisms. J Mol Microbiol Biotechnol. 2001;3:255–264. [PubMed: 11321581]
  • Rankin D, et al. What traits are carried on mobile genetic elements, and why? Heredity. 2010;106:1–10. [PMC free article: PMC3183850] [PubMed: 20332804]
  • Rasmussen MA, et al. Exposure to rumen protozoa leads to enhancement of pathogenicity of and invasion by multiple-antibiotic-resistant Salmonella enterica bearing SGI1. Infect Immun. 2005;73:4668–4675. [PMC free article: PMC1201270] [PubMed: 16040979]
  • Read AF. The evolution of virulence. Trends Microbiol. 1994;2:73–76. [PubMed: 8156274]
  • Schlichting C, Pigliucci M. Phenotypic Evolution: A Reaction Norm Perspective. Sinauer; 1998.
  • Smith EE, et al. Genetic adaptation by Pseudomonas aeruginosa to the airways of cystic fibrosis patients. Proc Natl Acad Sci USA. 2006;103:8487–8492. [PMC free article: PMC1482519] [PubMed: 16687478]
  • Smith J. The social evolution of bacterial pathogenesis. Proc R Soc Ser B Biol Sci. 2001;268:61–69. [PMC free article: PMC1087601] [PubMed: 12123299]
  • Sokurenko EV, et al. Source–sink dynamics of virulence evolution. Nat Rev Microbiol. 2006;4:548–555. [PubMed: 16778839]
  • Steinberg KM, Levin BR. Grazing protozoa and the evolution of the Escherichia coli O157: H7 Shiga toxin-encoding prophage. Proc R Soc B Biol Sci. 2007;274:1921–1929. [PMC free article: PMC2211389] [PubMed: 17535798]
  • Taylor PD, et al. The evolutionary consequences of plasticity in host–pathogen interactions. Theor Popul Biol. 2006;69:323–331. [PubMed: 16469343]
  • Turner PE, et al. Role of evolved host breadth in the initial emergence of an RNA virus. Evolution. 2010;64:3273–3286. [PubMed: 20633045]
  • Woolhouse ME, et al. Population biology of multihost pathogens. Science. 2001;292:1109–1112. [PubMed: 11352066]
  • Woolhouse MEJ, et al. Population biology of multihost pathogens. Science. 2001;292:1109. [PubMed: 11352066]



,12,‡ ,12 ,13 ,13 ,12 ,14 and 12.


12 Organismic and Evolutionary Biology, Harvard University, 16 Divinity Avenue, Cambridge, Massachusetts 02138.
13 Marine Biology, University of Vienna, Althanstrasse 14, A-1090 Vienna, Austria.
14 Biology Department MS 9160, Western Washington University, 516 High Street, Bellingham, Washington 98225.
Present address: Division of Natural Sciences and Mathematics, The Richard Stockton College of New Jersey, P.O. Box 195, Pomona, NJ 08240.

Recent evidence suggests that deep-sea vestimentiferan tube worms acquire their endosymbiotic bacteria from the environment each generation; thus, free-living symbionts should exist. Here, free-living tube worm symbiont phylotypes were detected in vent seawater and in biofilms at multiple deep-sea vent habitats by PCR amplification, DNA sequence analysis, and fluorescence in situ hybridization. These findings support environmental transmission as a means of symbiont acquisition for deep-sea tube worms.

The mode by which symbionts are passed between successive host generations is a primary question in symbiosis research. Symbiont transmission typically occurs vertically via transfer from parent to offspring, horizontally between co-occurring host individuals, or environmentally via uptake from a free-living population (Buchner, 1965). Determining which of these mechanisms operates within a symbiosis is critical, as the transmission mode impacts fundamental ecological and evolutionary processes, including genome evolution, symbiont-host specificity, and coevolution (for examples, see references Dale and Moran, 2006; Moya et al., 2008; and Stewart et al., 2005). Deep-sea vestimentiferan tube worms, which dominate the fauna at hydrothermal vents and cold seeps, are hypothesized to acquire their bacterial symbionts environmentally from a free-living population. Attempts to detect tube worm symbionts in host eggs and larvae by the use of microscopy and PCR have been unsuccessful (Cary et al., 1989; Cary et al., 1993; Cavanaugh et al., 1981; and Jones and Gardiner, 1988), suggesting that transmission does not occur vertically. Furthermore, most vent vestimentiferan species host symbionts that share identical 16S rRNA sequences, which is also consistent with the hypothesis of environmental transmission (Laue and Nelson, 1997; Nelson and Fisher, 2000). Unlike adults, the larvae and small juveniles of vestimentiferan tube worms have a mouth and gut, suggesting environmental acquisition via the ingestion of symbionts during larval development (Jones and Gardiner, 1988; Southward, 1988). However, Nussbaumer et al. (2006) recently demonstrated that bacterial symbionts are found on the developing tubes of settled larvae, entering the host worm through the epidermis and body wall of both larvae and young juveniles (Nussbaumer et al., 2006). These studies strongly suggest that tube worms acquire their symbionts from the surrounding environment and, therefore, that these endosymbionts should be detectable in a free-living form.

Sample Collection

A systematic search for the free-living counterpart to the gammaproteobacterial endosymbiont phylotype shared by three species of vestimentiferan tube worms, Riftia pachyptila, Oasisia alvinae, and Tevnia jerichonana, was conducted at the Tica hydrothermal vent site (~2,600-m depth) on the East Pacific Rise (EPR) (9°50.447′N, 104°17.493′W) during December 2002 and December 2003. Symbiont-containing tissue was dissected from all three vestimentiferan tube worm hosts (from the trophosome) and from Calyptogena magnifica clams (from the gills) at the Tica vent site for future use as positive and negative controls, respectively. Environmental samples were collected from two distinct habitats: surface-attached biofilms and seawater.

Symbionts in surface-attached biofilms were collected on bacterial settlement devices deployed in four hydrothermal vent environments at increasing distances from tube worm clusters: (i) among tube worms, (ii) adjacent to tube worms, (iii) away from tube worms (~10 m), and (iv) off-axis (~100 m) outside the axial summit of the caldera) (see Fig. S1 in the supplemental material). Settlement devices were constructed of polyvinyl chloride holders containing three to five basalt pieces (8 by 1 by 1 cm) and 4 to 12 glass microscope slides that were washed, autoclaved, and kept sterile until deployment. Devices were collected within 1 month or after 1 year. Upon collection, the basalt pieces were examined under a dissecting microscope to detect any settled tube worm larvae or juveniles and then immediately stored at ~80°C. Pieces with observable tube worms were excluded to eliminate the risk of detecting symbionts living within host tissue. Microscope slides were fixed for fluorescence in situ hybridization (FISH) analysis in 4% paraformaldehyde and stored in 70% ethanol at 4°C.

Seawater samples were collected 1 m away from an R. pachyptila tube worm cluster using a McLane large-volume water transfer system water pump attached to the deep submergence vehicle Alvin. Samples (200 liters each) were filtered in situ through a 1-μm Petex prefilter (Sefar) and then through a 0.45-μm mixed-cellulose ester filter (Millipore). Control seawater samples (80 liters each) were collected from the ocean surface above the EPR and from the Atlantic Ocean in Nahant, MA. All filters were stored at −80°C until DNA extraction.

16S rRNA Gene Sequence Analyses

PCR amplification and DNA sequence analyses were used to test for the presence of the vestimentiferan symbiont in biofilm and seawater samples. DNA was extracted by standard methods (Sambrook and Russel, 2001). The vestimentiferan symbiont 16S rRNA gene (a 401-bp fragment) was PCR amplified using primers specific for the shared 16S phylotype: RifTO44 (5′-GGCCTAGATT-GACGCTGCGGTA-3′) (this study) and RifTO445 (23). To detect contamination by host tissue, primers specific for the genes encoding the vestimentiferan host exoskeleton protein RP43 (GenBank accession no. AF233595), RifTOExoF (5′-CTAAAGGCAGTGTCAAGAGCGGGAC-3′) and RifTOExoR (5′-TTCCTC-GAAGTTGCCGTATGCCG-3′), were used. PCR products were cloned into a pCR2 cloning vector (Invitrogen) and sequenced by standard methods using BigDye Terminator cycle sequencing reaction kits (PE Biosystems) with M13 forward and reverse primers. Symbiont- and host-specific primers amplified their target genes in the control symbiont-containing tissue samples from R. pachyptila, T. jerichonana, and O. alvinae worms, while vestimentiferan symbionts were not amplified from C. magnifica gill tissue, the negative control.

The free-living vestimentiferan symbiont 16S rRNA phylotype was detected in both biofilm and seawater samples collected at the Tica vent site. The symbiont phylotype (GenBank accession no. U77478) (Feldman et al., 1997) was amplified from all basalt pieces retrieved after 1 month and after 1 year, including those from the off-axis site, away from active venting, and those from vent seawater samples on both 0.45- and 1-μm-pore-size water filters (Table A3-1). Host tissue was detected only on a single prefilter (1 μm) water sample. PCR amplifications from surface seawater control samples yielded positive PCR results with universal Bacteria primers (27f and 1492r) (Lane, 1991) but yielded negative results when either the vestimentiferan symbiont- or host-specific primers were used, suggesting that symbiont phylotypes were present only in deep-seawater samples. Little is yet known about the metabolic state or energy source for symbionts outside of their tube worm hosts, but it is possible that free-living symbionts may be cystic or quiescent while awaiting the inoculation of larval or juvenile tube worms.

TABLE A3-1. Detection of Free-Living Symbiont Phylotype of Vent Vestimentiferan Tube Worms Via PCR and Sequence Analyses of Biofilms.


Detection of Free-Living Symbiont Phylotype of Vent Vestimentiferan Tube Worms Via PCR and Sequence Analyses of Biofilms.


FISH was used to provide direct visual evidence of the tube worm symbiont on glass slides recovered from bacterial settlement devices. For each slide, a universal Bacteria probe,

Eub338 (1), either 5′end labeled with fluorescein or stained with the DNA-binding fluorescent dye 4′, 6′-diamidino-2-phenylindole (DAPI), was used as a positive control along with the symbiont-specific probe RifTO147, RifTO445, or RifTO830 that was 5′end labeled with Cy3 (Nussbaumer et al., 2006). The images from the control and symbiont-specific probes were then overlaid. The probe specificity was tested on R. pachyptila trophosome tissue, and the formamide concentration was increased until no probe remained hybridized (probe dependent, 20% [for Fig. S2 in the supplemental material] or 35% [for Fig. A3-1]). On each slide, either a nonsense probe, NON338 (Nussbaumer et al., 2006), or a 1-base-mismatch probe was used as a negative control. Hybridized slides were viewed and digitally photographed using a Leica model DMRB fluorescence microscope.

A four-panel illustration of FISH detection of free-living vestimentiferan bacterial symbionts


FISH detection of free-living vestimentiferan bacterial symbionts. Representative slides deployed at the Tica hydrothermal vent site on the EPR for ~1 month among Riftia pachyptila tube worms (A), near tube worms (B and C), and 10 m away from tube worms (more...)

The tube worm symbiont phylotype was detected using FISH on all slides tested (Fig. A3-1; see Fig. S2 in the supplemental material) with the exception of the off-axis samples that were collected from devices deployed for less than 1 month. Although not directly quantified, the overall bacterial abundance appeared to be greatest on slides deployed for 1 year among, adjacent to, or away from the tube worms. The direct detection of the tube worm symbiont in biofilms supports the hypothesis that these bacteria exist in the free-living vent environment.

Indeed, in a coastal marine endosymbiosis, the 16S phylotype of bacterial symbionts of Codakia orbicularis clams is readily found in the sea grass sediment surrounding their hosts (Gros et al., 2003).

Endosymbiont ITS Diversity

If vestimentiferan tube worms acquire their symbionts from a diverse environmental source population, it can be hypothesized that the symbiont population within a host may consist of multiple closely related phylotypes (DeChaine et al., 2006; Won et al., 2003). The symbiont internal transcribed spacer (ITS), which is under relaxed selection relative to the 16S and has been used extensively to assess strain-level variation in bacteria (Stewart et al., 2007), was cloned and sequenced to test for the presence of multiple symbiont phylotypes within individual tube worms. The ITS, located between the 16S and 23S rRNA genes in the bacterial rRNA operon, occurs as a single copy in the vestimentiferan symbiont genome (Markert et al., 2007; Robidart et al., 2008). By using symbionts-pecific primers embedded in the 16S and 23S rRNA genes (Sym-ITS-1322F and Sym-ITS-23SR) (31), the ITS was PCR amplified (30 cycles with Taq polymerase) from DNA extracted from the trophosomes of three adult R. pachyptila worms. PCR products were cloned and sequenced (96 clones per specimen; 288 in total).

Analysis of the ITS sequences from the three R. pachyptila symbiont clone libraries revealed high levels of genetic homogeneity in intracellular symbiont populations. Sequence analysis revealed one dominant symbiont phylotype within each of the three host specimens (accounting for 65, 77, and 41% of the sequences, respectively), and the third specimen hosted a second phylotype (27%), which consistently differed by the same two nucleotides. The majority of the remaining ITS sequences were singletons that cannot be distinguished from errors resulting from PCR or Taq analyses. The detection of diverse ITS sequences in R. pachyptila worms further supports the acquisition of bacteria from the environment, but the diversity of free-living symbionts has not yet been investigated.

Evidence for Environmental Symbiont Acquisition

Detection of the free-living tube worm symbiont phylotype supports the hypothesis that newly settled tube worms obtain their bacteria from the vent environment. Along a spatial gradient, free-living symbionts were present among, adjacent to, and away from (within 10 m) tube worms and were also detected 100 m outside the areas of hydrothermal activity. The presence of free-living symbiotic bacteria at multiple spatial scales within a vent site suggests a potentially large environmental pool of symbionts. During host larval development and the colonization of new vents (Marsh et al., 2001; Mullineaux et al., 2000, 2005), an abundant free-living bacterial population would facilitate the initiation of the symbiosis. The environmental transmission of symbionts seems to be a risky strategy for obligate tube worm symbioses, as the survival of the mouthless and gutless adult host requires that developing larvae or juveniles successfully acquire their symbionts from a potentially unstable free-living source population. However, this developmental mode might be beneficial if it provides the host with opportunities to acquire specific, locally adapted symbiont genotypes.

Influence of Symbiotic Bacteria on Free-Living Microbial Diversity

Symbioses, notably those that are facultative, clearly have an impact on and may be a driving force of local microbial diversity in varied ecosystems (Baker, 2003; Finlay, 2005). Indeed, the bacterial symbionts of the shrimp Rimicaris exoculata make up a major component of the surrounding microbial community at hydrothermal vents in the Atlantic Ocean (Polz and Cavanaugh, 1995). Likewise, a free-living counterpart to the bioluminescent symbiotic bacterium Vibrio fischeri of squid has been identified in coastal environments, revealing a connection between the symbiotic relationship and microbial abundance and distribution (Lee and Ruby, 1992). The same situation appears to be true in legume-rhizobium symbioses; the host species is thought to be a major factor in determining the characteristics of the soil microbial community (Miethling et al., 2000). Endosymbiont and free-living populations may affect each other via positive feedback cycles, whereby the host inoculates the free-living population, and the free-living population inoculates the host (Polz et al., 2000). This study serves as the basis for future investigations of the biodiversity and biogeography of free-living marine symbionts at multiple spatial scales.

We thank Chief Scientists Charles Fisher and Craig Cary; George Silva, Gary Chiljean, and the crew of the research vessel Atlantis; and the deep submergence vehicle Alvin group for collaborative expeditions, expert sample deployment, and collection. We thank Alan Fleer for instruction and use of the McLane pumps, Ansel Payne for technical assistance, and Thomas Auchtung, Stephanie Huff, and Irene Garcia Newton for assistance both on land and at sea. The paper was greatly improved thanks to Frank Stewart and four anonymous reviewers. This work was supported by grants from the Austria Science Foundation (FWF H00087 and P13762) and the Austrian Academy of Science to M.B., the National Science Foundation (NSF DBI-0400591) to E.G.D, and the NOAA National Undersea Research Center for the West Coast and Polar Regions (UAF 03-0092) and the NSF (OCE-0453901) to C.M.C., which we gratefully acknowledge.


  • Amann RI, Binder BJ, Olson RJ, Chisholm SW, Devereux R, Stahl DA. Combination of 16S rRNA-targeted oligonucleotide probes with flow cytometry for analyzing mixed microbial populations. Appl Environ Microbiol. 1990;56:1919–1925. [PMC free article: PMC184531] [PubMed: 2200342]
  • Baker AC. Flexibility and specificity in coral-algal symbiosis: diversity, ecology, and biogeography of Symbiodinium. Annu Rev Ecol Evol Syst. 2003;34:661–689.
  • Buchner P. Endosymbiosis of animals with plant microorganisms. Interscience Publishers, Inc; New York, NY: 1965.
  • Cary SC, Felbeck H, Holland ND. Observations on the reproductive biology of the hydrothermal vent tubeworm Riftia pachyptila. Mar Ecol Prog Ser. 1989;52:89–94.
  • Cary SC, Warren W, Anderson E, Giovannoni SJ. Identification and localization of bacterial endosymbionts in hydrothermal vent taxa with symbiont-specific polymerase chain reaction amplification and in situ hybridization techniques. Mol Mar Biol Biotechnol. 1993;2:51–62. [PubMed: 8364689]
  • Cavanaugh CM, Gardiner SL, Jones ML, Jannasch HW, Waterbury JB. Prokaryotic cells in the hydrothermal vent tube worm Riftia pachyptila Jones: possible chemoautotrophic symbionts. Science. 1981;213:340–342. [PubMed: 17819907]
  • Dale C, Moran NA. Molecular interactions between bacterial symbionts and their hosts. Cell. 2006;126:453–465. [PubMed: 16901780]
  • DeChaine EG, Bates AE, Shank TM, Cavanaugh CM. Off-axis symbiosis found: characterization and biogeography of bacterial symbionts of Bathymodiolus mussels from Lost City hydrothermal vents. Environ Microbiol. 2006;8:1902–1912. [PubMed: 17014490]
  • Feldman R, Black M, Cary C, Lutz R, Vrijenhoek R. Molecular phylogenetics of bacterial endosymbionts and their vestimentiferan hosts. Mol Mar Biol Biotechnol. 1997;6:268–277. [PubMed: 9284565]
  • Finlay RD. Mycorrhizal symbiosis: myths, misconceptions, new perspectives and future research priorities. Mycologist. 2005;19:90–95.
  • Gros O, Liberge M, Heddi A, Khatchadourian C, Felbeck H. Detection of the free-living forms of sulfide-oxidizing gill endosymbionts in the lucinid habitat (Thalassia testudinum environment). Appl Environ Microbiol. 2003;69:6264–6267. [PMC free article: PMC201202] [PubMed: 14532089]
  • Jones ML, Gardiner SL. Evidence for a transient digestive tract in Vestimentifera. Proc Biol Soc Wash. 1988;101:423–433.
  • Lane DJ. 16S/23S rRNA sequencing. In: Stackebrandt E, Goodfellow M, editors. Nucleic acid techniques in bacterial systematics. John Wiley & Sons; New York, NY: 1991. pp. 115–175.
  • Laue BE, Nelson DC. Sulfur-oxidizing symbionts have not co-evolved with their hydrothermal vent tubeworm hosts: RFLP analysis. Mol Mar Biol Biotechnol. 1997;6:180–188. [PubMed: 9284558]
  • Lee K-H, Ruby EG. Detection of the light organ symbiont, Vibrio fischeri, in Hawaiian seawater by using lux gene probes. Appl Environ Microbiol. 1992;58:942–947. [PMC free article: PMC195360] [PubMed: 16348678]
  • Markert S, Arndt C, Felbeck H, Becher D, Sievert SM, Hugler M, Albrecht D, Robidart J, Bench S, Feldman RA, Hecker M, Schweder T. Physiological proteomics of the uncultured symbiont of Riftia pachyptila. Science. 2007;315:247–250. [PubMed: 17218528]
  • Marsh AG, Mullineaux LS, Young CM, Manahan DT. Larval dispersal potential of the tubeworm Riftia pachyptila at deep-sea hydrothermal vents. Nature. 2001;411:77–80. [PubMed: 11333980]
  • Miethling R, Wieland G, Backhaus H, Tebbe GC. Variation of microbial rhizosphere communities in response to crop species, soil origin, and inoculation with Sinorhizobium meliloti L33. Microb Ecol. 2000;40:43–56. [PubMed: 10977876]
  • Moya A, Pereto J, Gil R, Latorre A. Learning how to live together: genomic insights into prokaryote-animal symbioses. Nat Rev Genet. 2008;9:218–229. [PubMed: 18268509]
  • Mullineaux LS, Fisher CR, Peterson CH, Schaeffer SW. Tubeworm succession at hydrothermal vents: use of biogenic cues to reduce habitat selection error? Oecologia. 2000;123:275–284. [PubMed: 28308732]
  • Mullineaux LS, Mills SW, Sweetman AK, Beaudreau AH, Metaxas A, Hunt HL. Vertical, lateral and temporal structure in larval distributions at hydrothermal vents. Mar Ecol Prog Ser. 2005;293:1–16.
  • Nelson K, Fisher CR. Absence of cospeciation in deep-sea vestimentiferan tube worms and their bacterial endosymbionts. Symbiosis. 2000;28:1–15.
  • Nussbaumer AD, Fisher CR, Bright M. Horizontal endosymbiont transmission in hydrothermal vent tubeworms. Nature. 2006;441:345–348. [PubMed: 16710420]
  • Polz MF, Cavanaugh CM. Dominance of one bacterial phylotype at a Mid-Atlantic Ridge hydrothermal vent site. Proc. Natl. Acad. Sci USA. 1995;92:7232–7236. [PMC free article: PMC41313] [PubMed: 7543678]
  • Polz MF, Ott JA, Bright M, Cavanaugh CM. When bacteria hitch a ride: associations between sulfur-oxidizing bacteria and eukaryotes represent spectacular adaptations to environmental gradients. ASM News. 2000;66:531–539.
  • Robidart JC, Bench SR, Feldman RA, Novoradovsky A, Podell SB, Gaasterland T, Allen EE, Felbeck H. Metabolic versatility of the Riftia pachyptila endosymbiont revealed through metagenomics. Environ Microbiol. 2008;10:727–737. [PubMed: 18237306]
  • Sambrook J, Russell DW. Molecular cloning: a laboratory manual. 3rd ed. Cold Spring Harbor Laboratory; Cold Spring Harbor, NY: 2001.
  • Southward EC. Development of the gut and segmentation of newly settled stages of Ridgeia (Vestimentifera): implications for relationships between Vestimentifera and Pogonophora. J Mar Biol Assoc UK. 1988;68:465–487.
  • Stewart FJ, Cavanaugh CM. Intragenomic variation and evolution of the internal transcribed spacer of the rRNA operon in bacteria. J Mol Evol. 2007;65:44–67. [PubMed: 17568983]
  • Stewart FJ, Newton IL, Cavanaugh CM. Chemosynthetic endosymbioses: adaptations to oxic-anoxic interfaces. Trends Microbiol. 2005;13:439–448. [PubMed: 16054816]
  • Won Y-J, Hallam SJ, O’Mullan GD, Pan IL, Buck KR, Vrijenhoek RC. Environmental acquisition of thiotrophic endosymbionts by deep-sea mussels of the genus Bathymodiolus. Appl Environ Microbiol. 2003;69:6785–6792. [PMC free article: PMC262266] [PubMed: 14602641]



16,17,18 and 16,17,18,*.


16 Department of Bacteriology, University of Wisconsin-Madison, Madison, WI 53706, USA.
17 Smithsonian Tropical Research Institute, Apartado Box 2072, Balboa, Ancon, Panama.
18 Department of Ecology and Evolutionary Biology, University of Kansas, Lawrence, KS 66045, USA.



The persistence of cooperative relationships is an evolutionary paradox; selection should favor those individuals that exploit their partners (cheating), resulting in the breakdown of cooperation over evolutionary time. Our current understanding of the evolutionary stability of mutualisms (cooperation between species) is strongly shaped by the view that they are often maintained by partners having mechanisms to avoid or retaliate against exploitation by cheaters. In contrast, we empirically and theoretically examine how additional symbionts, specifically specialized parasites, potentially influence the stability of bipartite mutualistic associations. In our empirical work we focus on the obligate mutualism between fungus-growing ants and the fungi they cultivate for food. This mutualism is exploited by specialized microfungal parasites (genus Escovopsis) that infect the ant’s fungal gardens. Using sub-colonies of fungus-growing ants, we investigate the interactions between the fungus garden parasite and cooperative and experimentally-enforced uncooperative (“cheating”) pairs of ants and fungi. To further examine if parasites have the potential to help stabilize some mutualisms we conduct Iterative Prisoner’s Dilemma (IPD) simulations, a common framework for predicting the outcomes of cooperative/non-cooperative interactions, which incorporate parasitism as an additional factor.


In our empirical work employing sub-colonies of fungus-growing ants, we found that Escovopsis-infected sub-colonies composed of cheating populations of ants or fungi lost significantly more garden biomass than sub-colonies subjected to infection or cheating (ants or fungi) alone. Since the loss of fungus garden compromises the fitness of both mutualists, our findings suggest that the potential benefit received by the ants or fungi for cheating is outweighed by the increased concomitant cost of parasitism engendered by non-cooperation (cheating). IPD simulations support our empirical results by confirming that a purely cooperative strategy, which is unsuccessful in the classic IPD model, becomes stable when parasites are included.


Here we suggest, and provide evidence for, parasitism being an external force that has the potential to help stabilize cooperation by aligning the selfish interests of cooperative partners in opposition to a common enemy. Specifically, our empirical results and IPD simulations suggest that when two mutualists share a common enemy selection can favor cooperation over cheating, which may help explain the evolutionary stability of some mutualisms.


The stability of cooperation is an evolutionary paradox—partners should be selected to cheat, pursuing their own selfish interests by obtaining benefits without providing a reward in return. Despite the inherent selfishness of individuals, cooperation within and between species is common in nature (Aanen et al., 2002; Chapela et al., 1994; Pellmyr and Huth, 1994). Hamilton’s Kin Selection Theory (Hamilton, 1964) helps explain cooperation among closely related individuals: organisms increase their fitness through altruism with close relatives due to their shared genes. The main theories used to help explain cooperation among unrelated individuals have been categorized as either directed reciprocation or by-product benefits (Sachs et al., 2004). Many of the models that fit the former category, including host sanction and partner fidelity, have developed out of, and are supported by, years of game theory modelling and focus on how individuals avoid being exploited by their partners (Bergstrom et al., 2003; Bull and Rice, 1991; Doebeli and Hauert, 2005; Dugatkin, 1997; Nowak, 2004). Specifically, within directed reciprocation cooperation is thought to be maintained when partners prevent one another from pursuing their own selfish interests (“cheating”), such as retaliation against cheating. However, empirical support for mechanisms of directed reciprocation that stabilize interspecific mutualisms is mostly lacking (Bergstrom et al., 2003; Bull and Rice, 1991; Doebeli and Hauert, 2005; Dugatkin, 1997; Nowak, 2004; Sachs et al., 2004).

Studies on the stability of mutualisms have generated extensive and valuable information about cooperation between unrelated individuals. However, these studies have primarily been framed within the traditional view of pair-wise partner associations occurring in isolation (Bergstrom et al., 2003; Bronstein and Barbosa, 2002; Stanton, 2003), while it is becoming increasingly clear that mutualisms are usually embedded within complex ecological communities (Althoff et al., 2004; Bronstein and Barbosa, 2002; Bronstein, 2001; Currie et al., 1999; Curry et al., 2006; Currie et al., 1999; Irwin, 2006; Little and Currie, 2008; Morris et al., 2003; Segrages et al., 2005; Strauss et al., 2004), and that these additional symbionts or interactants (tertiary, quaternary etc.) play important roles in mutualism dynamics. Indeed, third parties have been shown to alter the intensity, outcomes, and ultimately even the symbiotic state (mutualistic or parasitic) of an association (see [Bronstein and Barbosa, 2002] for review). For example, some mutualisms are known to exist only in the presence of other species, such as protective mutualisms where the presence of natural enemies is required for benefits to be conferred (Agrawal and Fordyce, 2000; Fischer and Shingleton, 2001; Currie et al., 1999; Morales, 2000; Palmer et al., 2008). In addition, recent work on bird-dispersed pine trees has revealed that the presence/absence of a competitor, pine squirrels, alters selection on a trait specifically associated with the bird-pine mutualism (Siepielski and Benkman, 2007). This illustrates that selection imposed on a mutualism by a third party can disrupt the success and/or stability of the association. Despite mounting evidence to support the importance of additional symbionts and community members in the biology of mutualisms, the concept has not yet been extensively explored with respect to the evolutionary stability of mutualisms.

Parasites of mutualisms may be particularly important in altering the dynamics of cooperative relationships in ways that influence their stability. Parasites not only drive host evolution, they also shape community dynamics by indirectly influencing the organisms their hosts interact with. The indirect influence of parasites should be especially pronounced within mutualistic relationships, as increased morbidity and mortality caused by a parasite of one partner will significantly influence the success of the other partner (Currie, 2001; Oliver et al., 2005). Since cooperative partners frequently face a “common enemy” in the form of parasites, we hypothesize that the presence of an abundant and virulent parasite of one member of a mutualism could provide selective pressure such that cooperation between partners is favored over exploitation. Our hypothesis is similar to triadic models developed by social scientists to investigate the role of third parties in cooperative dynamics among humans (Lee et al., 1994), however as mentioned above, the impact of additional players on the evolutionary stability of mutualisms has not been examined empirically.

The mutualism between fungus-growing ants and the fungi they cultivate for food is an example of a cooperative relationship that has persisted over evolutionary time despite continual impact from a virulent parasite (Chapela et al., 1994); Currie, 2001; Currie et al., 2003; Little et al., 2006; Mueller et al., 2001). To test our hypothesis that parasites may stabilize cooperative relationships, we experimentally manipulated sub-colonies of fungus-growing ants to determine the impact of parasites on (i) ants with fungal partners who provide decreased benefit to ants (cheater fungi), and (ii) fungi being tended by ants who provide limited benefits and increased costs to the fungus garden (cheater ants). We explore our empirical results further by utilizing the classic Iterative Prisoner’s Dilemma model (Axelrod and Hamilton, 1981) to confirm how the addition of a virulent parasite influences the traditional victors of the model: Always Defect and Tit for Tat.


Fungus-Growing Ant Symbiosis

Ants in the tribe Attini engage in an obligate mutualism with basidiomycetous fungi (Lepiotaceae and Pterulaceae) (Chapela et al., 1994; Munkacsi et al., 2004). The fungus is maintained in specialized gardens, often subterranean, and ant workers forage for substrate to support the growth of their fungal mutualist and help protect it from potential competitors or parasites. The fungus is vertically transmitted between generations, with new queens carrying a fungal pellet, collected from their natal garden, on the nuptial flight (Weber, 1972). In exchange for these benefits, the fungus serves as the primary food source for the ant colony. Fungal cultivation in ants has a single origin, ~45 million years ago (Schultz and Brady, 2008). The subsequent evolutionary history has generated a diverse collection of ants (more than 230 species) and fungi.

The fungus-growing ant symbiosis is a good model system to investigate the ecological and evolutionary effects parasites have on mutualists for several reasons. First, the symbionts are widely distributed in the new world tropics, and are conspicuous and populous enough to allow for adequate collection. Second, symbionts are amenable to laboratory maintenance, being readily cultivable, thus allowing researchers to study and manipulate each symbiont separately and in combination. Third, an entire tribe of ants culture fungi for food. Each lineage in the tribe tends specific fungal cultivars, which each host specialized mycoparasites in the genus Escovopsis (Ascomycota: Hypocreales). Escovopsis exploits the ant-fungal mutualism by extracting nutrients from the fungal mycelium at a significant cost to both the ants (indirectly) and fungi (directly) (Currie, 2001; Reynolds and Currie, 2004). The high prevalence of Escovopsis and early origin in the ant-fungal symbiosis (Currie, 2001; Currie et al., 2003) suggests that it could help stabilize the ant-fungal mutualism over evolutionary time by aligning the selfish interests of the partners against the parasite.

Experimental Design and Overview

To empirically test the potential role Escovopsis plays in stabilizing cooperation between fungus-growing ants and their cultivated fungi, we investigated the interaction between the fungus garden parasite and cooperative and uncooperative (“cheating”) pairs of ants and fungi. The benefits gained by cheating could be diminished if there is a severe parasitic infection which results in increased costs to ants or fungi. In this instance, Escovopsis could help stabilize the ant-fungus mutualism, by selecting against cheaters. More specifically, if cheating by either mutualist results in increased morbidity due to the garden parasite, then the selective advantage of cheating would be reduced or potentially nullified. We examine this possibility by using a two-by-two factorial design, crossing the presence/absence of parasitism with the presence/absence of a cheating partner. Sub-colonies were randomly assigned to one of four treatments: i) no infection and no cheating, ii) no infection and cheating, iii) infection and no cheating or, iv) infection and cheating. “Ant-cheating” was simulated by altering the male to female ratio (worker castes are always female, while the only function of males is reproduction) in Trachymyrmex cf. zeteki sub-colonies. The ant-cheating treatment mimics ant colonies investing more energy into colony reproduction and less into workers that tend the garden. Consequently, there is less investment in colony/garden maintenance (by worker ants), and additional costs imposed on the fungus garden while males inhabit the nest. In separate Atta colombica sub-colonies, “fungus-cheating” was simulated by removing the specialized nutrient-rich hyphal swellings (gongylidia) produced by the cultivated fungus for the ants to feed on. Gongylidia benefit ants but they are not necessary for cultivar growth or survival (Bass and Cherrett, 1995), thus removing gongylidia simulates fungal cheating by decreasing the nutrient benefit the fungus provides the ants. Colony fitness following treatment was assessed by measuring fungal garden biomass fluctuations (note: number of ant workers within colonies and colony production of reproductives is highly correlated to fungus garden biomass) (Currie, 2001). The parasitism treatment involved infecting the ants’ fungus garden with the specialized parasite Escovopsis. Each treatment is described in detail below.

Sub-Colony Setup

Ten six-month-old queenright A. colombica colonies and 10 queenright T. cf. zeteki colonies with a single fungus chamber were collected in Gamboa, Panama. Sub-colonies of A. colombica were composed of 1.0 g of fungus garden and ~115 ants (consistent ratio of worker size and age castes, and brood), and were maintained in plastic dual chambers (one housing the garden, and one for feeding, foraging, and dumping of refuse) connected by plastic tubes. Trachymyrmex cf. zeteki sub-colonies were composed of 0.1 g of fungal cultivar and four ants, and housed in plastic Petri dishes (60 mm diameter). Colonies were placed on mineral oil islands to prevent potential transfer of microbes between sub-colonies via vectors (e.g., mites), were given unrestricted access to foliage (A. colombica) or a mixture of dried oats and oak catkins (T. cf. zeteki), and watered three times a week. All sub-colonies used in the experiment were healthy, stable, incorporating new substrate into the fungus garden, and free of detectable Escovopsis infection (Currie et al., 1999).

Simulation of Cheating by Ants

To simulate cheating by the ants the sex ratio of T. cf. zeteki sub-colonies was altered; in cheater sub-colonies two females and two males were present, while four females and zero males were present in control sub-colonies. We use colonies of T. cf. zeteki to simulate cheating by the ants because this species regularly produces males in the laboratory (A. colombica do not). All worker ants are female, and female reproductives (gynes) are responsible for maternal vertical transmission of the fungal mutualist. Males are reared on the nutrients of the fungal mutualist and stay within the fungus garden prior to the nuptial flight however they do not contribute towards tending the fungus garden. Thus, male ants are a direct cost to the fungus garden; they provide no known benefit to the ant’s fungal mutualist, neither dispersing nor contributing to fungus garden maintenance (Mueller et al., 2001).

Simulation of Cheating Fungi

To simulate cheating by the fungal mutualist, 10% of A. colombica garden biomass (containing ~276 gongylidia clusters) and all gongylidia clusters on the top surface of the garden (~265 clusters per nest) were removed. Gongylidia removal was done by hand, using a dissecting scope (Accu-Scope, Sea Cliff, NY) and jewel-tip forceps (Bioquip, Rancho Dominguez, CA). Atta colombica was used to simulate cheating by the fungal mutualist because the cultivated fungus of this species produces large, tightly clustered, nutrient-rich hyphal swellings, called gongylidia. Worker ants preferentially feed on gongylidia and harvest them as nutrients to support the growth of larvae (Mueller et al., 2001; Quinlan and Cherrett, 1978). The production of gongylidia by the cultivated fungi provides no apparent benefit to the fungus, but instead serves as a food source that is more beneficial to the ants than the regular hyphae of the fungus (Bass and Cherrett, 1995; Quinlan nd Cherrett, 1978). Gongylidia do not directly benefit the fungus garden and they do not help defend the garden from Escovopsis (see below). While T. cf. zeteki fungi also produce gongylidia, they are smaller, less abundant and fewer per cluster than those of A. colombica (A. Little pers. obs.). To achieve sufficient gongylidia removal and limit fungus garden destruction during sub-colony treatment preparation, it was necessary to use A. colombica, rather than T. cf. zeteki sub-colonies to mimic fungal cheating.

Infection of Sub-Colonies with Escovopsis

Escovopsis strains used in experiments were isolated in Gamboa, Panama from T. cf. zeteki and A. colombica colonies. Isolates were grown on potato dextrose agar (PDA) (Difco, Sparks, MD) with 1000 iu/ml of penicillin-streptomycin (MP Biomedicals Inc., Aurora, OH). Spores were added to ddH2O with Tween 20 [5 × 10−5] (Fisher Scientific, Pittsburgh, PA) to evenly disperse spores in solution. Trachymyrmex cf. zeteki and A. colombica colonies received 0.05 and 0.5 ml of solution, respectively (ca 6000 spores/T. cf. zeteki sub-colony, ca. 20000 spores/A. colombica sub-colony) via mist inoculation. Sub-colony biomass was measured prior to, and 72 hours after treatment. The relative changes in biomass/sub-colony/treatment were subjected to 2-way ANOVA in Minitab (2003). The success of the ants is directly dependent upon the health and biomass of the fungus gardens, therefore, as in other studies (Bot et al., 2001; Currie and Stuart, 2001), we use garden biomass as an indirect fitness indicator for the ant in the ant-cheating experiment.

Prisoner’s Dilemma Computer Simulation

Using the classic Prisoner’s Dilemma model (PD) (Axelrod and Hamilton, 1981), we further explore our empirical results that indicate parasites can play a role in stabilizing cooperation. In the PD two players interact, each has the ability to cooperate or cheat. Cooperation provides the opponent a benefit (b), while incurring a cost (c) to the player (b > c > 0). The highest payoff is received when a player cheats while its partner cooperates: the cheater benefits without paying the cost of cooperation (Temptation to cheat T = b). If both players cooperate each receives a net benefit (Reward) of R = bc, while mutual cheating results in a Punishment payoff of P = 0. The lowest payoff is received by a player that cooperates while its opponent cheats (Suckers payoff S = -c). In single interactions, cheating is the best strategy (T > R > P > S). Our results from ant-fungal manipulations suggest that the presence of a parasite would alter the PD payoff matrix such that pure cooperation is the best strategy. More specifically, cheating by either ants or fungi results in an increased parasite impact, reducing the benefit of “temptation” to cheat (R > T = P > S). In addition, because cooperation by one player (i.e. ants) provides some degree of defense against parasitism (i.e. Escovopsis) further alters the payoff matrix to favor cooperation (R > T = P = S).

The Iterative Prisoner’s Dilemma Model (IPD), where players engage in multiple interactions, is much more relevant to natural system. Based on the alteration of the payoff matrix of the single interaction PD model (see above), it is clear that if parasites impact every interaction in the IPD model they will help favor cooperation over cheating. However, it is very unlikely that parasites are so ubiquitous in natural populations that they influence every interaction. Thus, we utilized a computer program called DILEMMA to determine what level of parasite prevalence is required to potentially help stabilize cooperation within the IPD model [see Additional file 1].

Using DILEMMA, we explored the role of parasites in altering the dynamics within the IPD model by conducting simulations involving various combinations of strategies in the presence and absence of parasites. Simulations involved populations of 10,000 individuals, each individual engaged in 25 interactions per generation. Simulations were run for 500 generations, which we previously determined to be sufficient to obtain a stable proportion of strategies across generations. The average of 100 independent runs for each different simulation is presented. In the first simulation a 50:50 ratio of individuals playing “always defect” (uncooperative strategy), and “always cooperate,” (cooperation in every interaction) was used. Subsequently, simulations using the same 50:50 ratio with parasites present were run. Ten independent runs (500 generations each) were run with proportions of parasitism increasing by increments of 10%, and the means of the final frequencies of each strategy in the population were plotted by proportion of population infected with parasites. A second set of simulations was conducted, with three additional strategies (“tit for tat,” “sneaker,” and “random” [see Additional file 1]). All strategies started with a 20% frequency in the population. As above, this simulation was run in the absence of parasites for 500 generations, and then parasites were added with varying prevalence up to infection rates of 100%.


Empirical Test of Hypothesis

In our experiments we found that infected sub-colonies with cheating populations of ants or fungi each lost significantly more garden biomass than sub-colonies subjected to infection or cheating (ants or fungi) alone (2-way ANOVA p < 0.001 df = 3, for both treatments) (Fig. A4-1). When a cheater is present, the cost of Escovopsis infection is substantially greater than it is in sub-colonies with only cooperative partners. This suggests that the negative consequences Escovopsis has on ant and fungal health could result in parasite-induced selection eliminating cheating by either mutualist. There are several reasons colonies with cheaters are likely to be less successful at fighting garden infection, than those with cooperative ants and fungi. Increased virulence of Escovopsis in the “cheating ant” treatment is likely because the enforced shift to a 50% male:worker ratio results in fewer worker ants present to defend the garden from infection (Currie and Stuart, 2001). In the “cheating-fungi” treatment, the mechanism(s) causing a greater impact of infection is less clear. The high concentration of nutrients found in gongylidia may be a necessary energy source for worker ants that remove parasitic spores. The ants may also retaliate against fungal cheaters by allocating less effort into garden maintenance, which would result in greater garden biomass loss during infection.

A two-way interaction graph showing the impact of experimental infection and cheating ants or mutualistic fungi on the garden biomass of two types of ant colonies


Cheating/infection experiments. Two-way interaction graph illustrating the impact of experimental infection and cheating ants or mutualistic fungi on the garden biomass in two types of fungus-growing ant colonies. A) Trachymyrmex cf.zeteki sub-colonies (more...)

Prisoner’s Dilemma Simulations

In the classic model, when one partner always cooperates and the other always cheats, the cheater population quickly out-competes the cooperator population (Fig. A4-2a). In contrast, parasite infection rates of 51% or higher results in the strategy “always cooperate” being successful and stable (Fig. A4-2b). In the classic model cooperation is successful and stable if cooperative strategies can retaliate against cheating, such as the well-known IPD strategy “tit for tat” (TFT, Fig. A4-2c). In our DILEMMA simulations, when cooperative strategies capable of retaliation against cheaters (i.e., TFT) are included, the strategy “always cooperate” forms a stable population at infection levels of 10%, and out-competes TFT when infection levels are greater than 35% (Fig. A4-2d). These simulations support our empirical results by indicating that when parasites are common, cooperation is stable as the benefits gained by cheating are outweighed by the increased cost of infection.

Graphical output from DILEMMA simulating the Iterated Prisoner™s Dilemma


Prisoner’s Dilemma simulations. Graphical output from DILEMMA, a computer program simulating the Iterated Prisoner’s Dilemma (IPD) with the ability to incorporate parasites into the classic model to determine how prevalent parasite would (more...)


Despite the important role mutually beneficial associations play in shaping all levels of biological organization, how these relationships establish and maintain stability is not well understood. The challenge is elucidating the factor(s) that prevent selection from favoring partners who pursue their own selfish interests, cheaters who obtain benefits without providing rewards in return. Most theories proposed to help explain the evolutionary stability of mutualism argue that cooperation is stabilized by individuals employing mechanisms to avoid being exploited by their partners (e.g., host sanctions, partner choice) (Bull and Rice, 1991; Kiers et al., 2003). In contrast to this typical view that partners enforce reciprocity within beneficial exchanges, here we suggest, and provide empirical and theoretic evidence for, the possibility that an external force, such as parasitism, can help stabilize cooperation by aligning the selfish interests of partners.

One way parasites may help stabilize mutualisms is if “cheating” by one partner results in greater parasite-induced morbidity or mortality in one or both partners, resulting in a net loss to the “cheater” (i.e., the benefits obtained from “cheating” are diminished by the increased costs from more severe infection by the parasite). Indeed, here we found, using the fungus-growing ant mutualism as a model system, that cheaters can suffer disproportionately more in the presence of a parasite than their non-cheater counterparts. More specifically, enforced cheating by either the ants or their fungal partner had little to no negative impact on the health of the fungus garden, which both mutualists obligately depend on. In the presence of the garden parasite, cheating by either mutualist resulted in significantly higher parasite induced garden morbidity, as compared to controls involving garden infections with cooperative mutualist partners. Thus, our empirical results indicate that the increased impact of parasitism in the presence of cheating can reduce the inherent conflict (Herre et al., 1999; Mueller, 2002; Poulsen and Boomsma, 2005) between mutualists (Fig. A4-3), potentially contributing to the stability of the beneficial association.

Four panel diagram showing cooperation and conflict within the among fungus-growing ants


Cooperation and conflict within the fungus-growing ant microbe symbiosis. A) Fungus-growing ants forage for substrate to nourish their cultivated fungus, which they also groom to help remove garden parasites. B) In return, the fungus serves as the primary (more...)

Escovopsis can be extremely prevalent, infecting more than 75% of colonies of fungus-growing ant nests in some populations, and is known to be virulent (Currie et al, 1999; Currie, 2001). Thus, we believe our results indicate that cheating by either ants or fungi could be rapidly eliminated within natural populations by previously established infections or by new infections of the horizontally transmitted parasite. The alignment of interests between the ants and their cultivated fungi, in opposition to the garden parasite, is further illustrated by the contribution the ants make to cultivar defense. Specifically, the ants employ specialized behaviors to physically remove parasitic inoculum from the fungus garden (Currie and Stuart, 2001). Without ant behavioral defenses, the garden is rapidly overgrown by the parasite (Currie et al., 1999), indicating that defense against Escovopsis requires cooperation between ants and their fungal mutualists. The early origin of Escovopsis within the symbiosis and its coevolutionary history with the ants and their fungal cultivar (Currie et al., 2003), suggests that the parasite may have been a stabilizing force within the ant-fungal mutualism for millions of years.

Our view of the stability of cooperation has largely developed out of game theory, especially the PD model (Axelrod and Hamilton, 1981). In the classic single interaction model, cheating is always favored over cooperation (T > R > P > S, see methods above and Fig. A4-2a). However, when the model involves multiple interactions among players (IPD), strategies that are cooperative but capable of retaliating against cheating can out-compete cheating strategies (e.g., the well-known TFT). As outlined above, our empirical results indicate that a parasite has the potential to alter the payoff matrix so that cooperation is favored over cheating in the single interaction PD model. This illustrates the potential for third parties to alter the dynamics of cooperation in ways that shape mutualism stability. Our simulations revealed that even at relatively low prevalence parasitism can select for stability of a cooperative strategy that is incapable of retaliating against cheating. Specifically, “always cooperative” out-competes “always defect” when 51% of interacting partners in a population are infected, which is well within the known infection rates in the fungus-growing ant mutualism. When TFT was integrated into the simulation, surprisingly, we found that “always cooperative” forms a stable population at infection levels of 10%, and out-competes TFT when parasite prevalence is greater than 35%. These findings provide theoretical support to our empirical results from the fungus-growing ant mutualism, further suggesting that parasites can provide an external sanction against one partner’s cheating, or simply alter the costs and benefits received from cooperation versus cheating in such a way that natural selection favors cooperation.

We believe our findings are applicable beyond the fungus-growing ant microbe-symbiosis. Mutualisms in which survival and reproduction are tightly linked to cooperation are especially likely to be stabilized by antagonists, as morbidity and mortality in one partner is expected to have a significant cost to the other; this is complimentary to partner fidelity feedback (Bull and Rice, 1991). Protective mutualisms, in which one partner defends the other from a natural enemy, are common in nature. Just as fungus-growing ants protect their mutualistic fungi from parasites, there are ants that protect plants from herbivores (Huxley et al., 1991), bacteria that protect their insect hosts from disease (Oliver et al., 2005), and endophytic fungi that protect their plant hosts from herbivores via secondary metabolite production (Tanaka et al., 2005). Our results support the prediction that in these interactions it is likely that when the threat imposed by a tertiary symbiont (i.e. predator, parasite) is absent, the protective mutualism may break down. Indeed, a recent paper by Palmer et al. (2008) revealed the breakdown of an ant-plant mutualism in the absence of large-herbivores. Furthermore, the contribution symbionts make to protect their hosts, which appear to be widespread (see Haine [2008]) may be evidence of parasites aligning the interests of mutualists.


Cooperative relationships that occur in natural systems persist in complex ecological communities where interspecific interactions are continuous. In some instances one, or a combination, of the models included in the Sachs et al. framework of directed reciprocation, shared genes, and by-product benefits, adequately explains stable cooperation among organisms. However, our results suggest that a third species eliciting selective pressure on one member of a mutualism can limit cheating by a mechanism that does not neatly fit the current framework. Cooperative dynamics in which two partners have their selfish interests aligned in opposition to a third (parasitic) party, can provide a stabilizing force that helps maintain cooperation between species, that is neither a by-product (e.g. coincident of a selfish action), nor directed reciprocation. Additionally, it is important to be clear that parasitism need not be a mutually exclusive factor stabilizing cooperation. It is possible, and perhaps likely, that parasite pressure works in concert with other well-defined mechanisms that promote cooperation. It would be interesting to empirically test how the addition of a third parasitic species influences cooperative interactions that are believed to be governed by reciprocation, by-product benefits or shared genes.

Author’s Contributions

AL and CC conceived and designed the experiments. AL performed the experiments and analyzed the data. AL and CC wrote the paper and approved the final manuscript.


We are grateful to STRI and ANAM of the Republic of Panama for facilitating the research and granting collecting permits. For valuable logistical support we thank: M. Bergsbaken, M. Cafaro, S. Covington, P. Foley, S. Ingram, M. Leone, H. Reynolds, and M. Tourtellot. We are thankful to K. Boomsma, J. Bull, E. Caldera, N. Gerardo, K. Grubbs, H. Goodrich-Blair, S. Hoover, J. Kelly, A. Pinto, M. Poulsen, G. Roberts, R. Steffensen, and S. West for invaluable comments on this study and/or paper. This work was supported by NSF IRCEB grant DEB-0110073 to CRC, NSF CAREER grant DEB-747002 to CRC, and NSF Doctoral Dissertation Improvement Grant DEB-0608078 to CRC and AEFL.

Additional Material

Additional file 1

Dilemma information

Available at:


  • Aanen DK, Eggleton P, Rouland-Lefevre C, Guldberg-Froslev T, Rosendahl S, Boomsma JJ. The evolution of fungus-growing termites and their mutualistic fungal symbionts. Proc Natl Acad Sci USA. 2002;99(23):14887–14892. [PMC free article: PMC137514] [PubMed: 12386341]
  • Agrawal AA, Fordyce JA. Induced indirect defence in a lycaenidant association: the regulation of a resource in a mutualism. P Roy Soc Lond B Bio. 2000;267(1455):1857–1861. [PMC free article: PMC1690751] [PubMed: 11052536]
  • Althoff DM, Segraves KA, Sparks JP. Characterizing the interaction between the bogus yucca moth and yuccas: do bogus yucca moths impact yucca reproductive success? Oecologia. 2004;140(2):321–327. [PubMed: 15170561]
  • Axelrod R, Hamilton WD. The Evolution of Cooperation. Science. 1981;211(4489):1390–1396. [PubMed: 7466396]
  • Bass M, Cherrett JM. Fungal Hyphae as a Source of Nutrients for the Leaf-Cutting Ant Atta Sexdens. Physiological Entomology. 1995;20(1):1–6.
  • Bergstrom CT, Bronstein JL, Bshary R, Connor RC, Daly M, Frank SA, Gintis H, Keller L, Leimar O, Noe R, et al. Interspecific mutualism. In: Hammerstein P, editor. Genetic and cultural evolution of cooperation vol Workshop Report 90. Cambridge: MIT Press; 2003. pp. 242–256.
  • Bot ANM, Rehner SA, Boomsma JJ. Partial incompatibility between ants and symbiotic fungi in two sympatric species of Acromyrmex leaf-cutting ants. Evolution. 2001;55(10):1980–1991. [PubMed: 11761059]
  • Bronstein JL. The exploitation of mutualisms. Ecol Lett. 2001;4(3):277–287.
  • Bronstein JL, Barbosa P. Multitrophic/multispecies mutualistic interactions: the role of non-mutualists in shaping and mediating mutualisms. In: Tscharntke T, Hawkins BA, editors. Multitrophic level interactions. Cambridge; New York: Cambridge University Press; 2002. pp. 44–66.
  • Bull JJ, Rice WR. Distinguishing mechanisms for the evolution of cooperation. J Theo Biol. 1991;149(1):63–74. [PubMed: 1881147]
  • Chapela IH, Rehner SA, Schultz TR, Mueller UG. Evolutionary history of the symbiosis between fungus-growing ants and their fungi. Science. 1994;266(5191):1691–1694. [PubMed: 17775630]
  • Currie CR. Prevalence and impact of a virulent parasite on a tripartite mutualism. Oecologia. 2001;128(1):99–106. [PubMed: 28547096]
  • Currie CR, Mueller UG, Malloch D. The agricultural pathology of ant fungus gardens. Proc Natl Acad Sci USA. 1999;96(14):7998–8002. [PMC free article: PMC22176] [PubMed: 10393936]
  • Currie CR, Poulsen M, Mendenhall J, Boomsma JJ, Billen J. Coevolved crypts and exocrine glands support mutualistic bacteria in fungus-growing ants. Science. 2006;311(5757):81–83. [PubMed: 16400148]
  • Currie CR, Scott JA, Summerbell RC, Malloch D. Fungus-growing ants use antibiotic-producing bacteria to control garden parasites. Nature. 1999;398(6729):701–704.
  • Currie CR, Stuart AE. Weeding and grooming of pathogens in agriculture by ants. Proc Roy Soc Lond B Bio. 2001;268(1471):1033–1039. [PMC free article: PMC1088705] [PubMed: 11375087]
  • Currie CR, Wong B, Stuart AE, Schultz TR, Rehner SA, Mueller UG, Sung GH, Spatafora JW, Straus NA. Ancient tripartite coevolution in the attine ant-microbe symbiosis. Science. 2003;299(5605):386–388. [PubMed: 12532015]
  • Doebeli M, Hauert C. Models of cooperation based on the Prisoner’s Dilemma and the Snowdrift game. Ecol Lett. 2005;8(7):748–766.
  • Dugatkin LA. Cooperation among animals: an evolutionary perspective. New York: Oxford University Press; 1997.
  • Fischer MK, Shingleton AW. Host plant and ants influence the honeydew sugar composition of aphids. Funct Ecol. 2001;15(4):544–550.
  • Haine ER. Symbiont mediated protection. Proc Roy Soc Lond B Bio. 2008;275:353–361. [PMC free article: PMC2213712] [PubMed: 18055391]
  • Hamilton WD. The genetical theory of social behaviour I, II. J Theo Biol. 1964;7:1–52. [PubMed: 5875341]
  • Herre EA, Knowlton N, Mueller UG, Rehner SA. The evolution of mutualisms: exploring the paths between conflict and cooperation. Trends Ecol Evol. 1999;14(2):49–53. [PubMed: 10234251]
  • Huxley CR, Cutler DF.Linnean Society of London., University of Oxford. Dept of Zoology, University of Oxford Dept of Plant Sciences. Ant-plant interactions. Oxford; New York: Oxford University Press; 1991.
  • Irwin RE. The consequences of direct versus indirect species interactions to selection on traits: Pollination and nectar robbing in Ipomopsis aggregata. Am Nat. 2006;167(3):315–328. [PubMed: 16673341]
  • Kiers ET, Rousseau RA, West SA, Denison RF. Host sanctions and the legume-rhizobium mutualism. Nature. 2003;425(6953):78–81. [PubMed: 12955144]
  • Lee SC, Muncaster RG, Zinnes DA. The friend of my enemy is my enemy: modeling triadic interaction relationships. Synthese. 1994;100:333–358.
  • Little AEF, Currie CR. Black yeast symbionts compromise the efficiency of antibiotic defenses in fungus-growing ants. Ecology. 2008;89(5):1216–1222. [PubMed: 18543616]
  • Little AEF, Murakami T, Mueller UG, Currie CR. Defending against parasites: fungus-growing ants combine specialized behaviours and microbial symbionts to protect their fungus gardens. Biol Lett. 2006;2:12–16. [PMC free article: PMC1617182] [PubMed: 17148313]
  • Minitab statistical software. 14th edition. USA: Minitab Inc; 2003.
  • Morales Ma. Survivorship of an ant-tended membracid as a function of ant recruitment. Oikos. 2000;90(3):469–476.
  • Morris WF, Bronstein JL, Wilson WG. Three-way coexistence in obligate mutualist-exploiter interactions: The potential role of competition. Am Nat. 2003;161(6):860–875. [PubMed: 12858272]
  • Mueller UG. Ant versus fungus versus mutualism: Ant-cultivar conflict and the deconstruction of the attine ant-fungus symbiosis. Am Nat. 2002;160:S67–S98. 47. [PubMed: 18707454]
  • Poulsen M, Boomsma JJ. Mutualistic fungi control crop diversity in fungus-growing ants. Science. 2005;307(5710):741–744. [PubMed: 15692054]
  • Mueller UG, Schultz TR, Currie CR, Adams RMM, Malloch D. The origin of the attine ant-fungus mutualism. Quart Rev Biol. 2001;76(2):169–197. [PubMed: 11409051]
  • Munkacsi AB, Pan JJ, Villesen P, Mueller UG, Blackwell M, McLaughlin DJ. Convergent coevolution in the domestication of coral mushrooms by fungus-growing ants. Proc Roy Soc Lond B Bio. 2004;271(1550):1777–1782. [PMC free article: PMC1691797] [PubMed: 15315892]
  • Nowak MA, Sigmund K. Evolutionary dynamics of biological games. Science. 2004;303(5659):793–799. [PubMed: 14764867]
  • Oliver KM, Moran NA, Hunter MS. Variation in resistance to parasitism in aphids is due to symbionts not host genotype. Proc Natl Acad Sci USA. 2005;102(36):12795–12800. [PMC free article: PMC1200300] [PubMed: 16120675]
  • Palmer TM, Stanton ML, Young TP, Goheen JR, Pringle RM, Karban R. Breakdown of an ant-plant mutualism follows the loss of large herbivores from an African savanna. Science. 2008;319(192–195):192–195. [PubMed: 18187652]
  • Pellmyr O, Huth CJ. Evolutionary stability of mutualism between yuccas and yucca moths. Nature. 1994;372(6503):257–260.
  • Quinlan RJ, Cherrett JM. Aspects of Symbiosis of Leaf-Cutting Ant Acromyrmex-Octospinosus (Reich) and Its Food Fungus. Ecological Entomology. 1978;3(3):221–230.
  • Quinlan RJ, Cherrett JM. Role of Fungus in the Diet of the Leaf-Cutting Ant Atta-Cephalotes (L). Ecological Entomology. 1979;4(2):151–160.
  • Reynolds HT, Currie CR. Pathogenicity of Escovopsis weberi: The parasite of the attine ant-microbe symbiosis directly consumes the ant-cultivated fungus. Mycologia. 2004;96(5):955–959. [PubMed: 21148916]
  • Sachs JL, Mueller UG, Wilcox TP, Bull JJ. The evolution of cooperation. Quart Rev Biol. 2004;79(2):135–160. [PubMed: 15232949]
  • Schultz TR, Brady SG. Major evolutionary transitions in ant agriculture. Proc Natl Acad Sci USA. 2008;105:5435–5440. [PMC free article: PMC2291119] [PubMed: 18362345]
  • Segraves KA, Althoff DM, Pellmyr O. Limiting cheaters in mutualism: evidence from hybridization between mutualist and cheater yucca moths. P Roy Soc B-Biol Sci. 2005;272(1577):2195–2201. [PMC free article: PMC1559949] [PubMed: 16191630]
  • Siepielski AM, Benkman CW. Selection by a predispersal seed predator constrains the evolution of avian seed dispersal in pines. Funct Ecol. 2007;21(3):611–618.
  • Stanton ML. Interacting guilds: moving beyond the pairwise perspective on mutualisms. Am Nat. 2003;162:S10–S23. [PubMed: 14583854]
  • Strauss SY, Irwin RE, Lambrix VM. Optimal defence theory and flower petal colour predict variation in the secondary chemistry of wild radish. J Ecol. 2004;92(1):132–141.
  • Tanaka A, Tapper BA, Popay A, Parker EJ, Scott B. A symbiosis expressed non-ribosomal peptide synthetase from a mutualistic fungal endophyte of perennial ryegrass confers protection to the symbiotum from insect herbivory. Mol Microbiol. 2005;57(4):1036–1050. [PubMed: 16091042]
  • Weber NA. Gardening Ants: The Attines. Philadelphia: American Philosophical Society; 1972.



,20,21 ,22 ,20,21,23 ,20,21 ,24 ,20,21 ,24 ,22 ,22 ,22 ,20,21 ,24 ,22 ,22 and 20,21,23.


20 Department of Bacteriology, University of Wisconsin-Madison, Madison, WI, USA.
21 Department of Energy Great Lakes Bioenergy Research Center, University of Wisconsin-Madison, Madison, WI, USA.
22 Biological Sciences Division, Pacific Northwest National Laboratory, Richland, WA, USA.
23 Smithsonian Tropical Research Institute, Balboa, Ancon, Panama.
24 Department of Energy Joint Genome Institute, Walnut Creek, CA, USA.

Herbivores gain access to nutrients stored in plant biomass largely by harnessing the metabolic activities of microbes. Leaf-cutter ants of the genusAttaare a hallmark example; these dominant neotropical herbivores cultivate symbiotic fungus gardens on large quantities of fresh plant forage. As the external digestive system of the ants, fungus gardens facilitate the production and sustenance of millions of workers. Using metagenomic and metaproteomic techniques, we characterize the bacterial diversity and physiological potential of fungus gardens from two species ofAtta. Our analysis of over 1.2Gbp of community metagenomic sequence and three 16S pyrotag libraries reveals that in addition to harboring the dominant fungal crop, these ecosystems contain abundant populations ofEnterobacteriaceae, including the generaEnterobacter,Pantoea,Klebsiella,CitrobacterandEscherichia. We show that these bacterial communities possess genes associated with lignocellulose degradation and diverse biosynthetic pathways, suggesting that they play a role in nutrient cycling by converting the nitrogen-poor forage of the ants into B-vitamins, amino acids and other cellular components. Our metaproteomic analysis confirms that bacterial glycosyl hydrolases and proteins with putative biosynthetic functions are produced in both field-collected and laboratory-reared colonies. These results are consistent with the hypothesis that fungus gardens are specialized fungus–bacteria communities that convert plant material into energy for their ant hosts. Together with recent investigations into the microbial symbionts of vertebrates, our work underscores the importance of microbial communities in the ecology and evolution of herbivorous metazoans.


Ants are critical components of terrestrial ecosystems around the world (Hölldobler and Wilson, 1990). Among ants, leaf-cutters in the genus Atta (Figure A5-1a) are particularly dominant, with mature colonies achieving immense sizes and housing millions of workers (Hölldobler and Wilson, 2008, 2010). Ranging from the southern United States to Argentina, species of leaf-cutter ants can construct elaborate subterranean nests containing hundreds of chambers and displacing up to 40 000 kg of soil (Hölldobler and Wilson, 2010). The ant societies housed within these nests are equally impressive, with an intricate division of labor observed between different castes of workers (Hölldobler and Wilson, 2010). Associated with this division of labor is substantial worker-size polymorphism: the dry weight of individual workers in the same colony can differ by 200-fold (Hölldobler and Wilson, 2010). The success of leaf-cutter ants is largely attributed to their obligate mutualism with a basidiomycetous fungus (Leucoagaricus gongylophorus) that they culture for food in specialized gardens (Figure A5-1b) (Weber, 1966; Hölldobler and Wilson, 2008, 2010). Fresh plant forage collected by the ants serves to nourish the fungus, which in turn is thought to have originated 8–12 million years ago, and numerous adaptations in both the ants and the fungus have occurred over this long history of agriculture (Weber, 1966; Chapela et al., 1994; Schultz and Brady, 2008).

Two photos of leaf cutter ants and two bar graphs showing microbial community composition within fungus gardens


Leaf cutter ants forage on plant material. (a) that they use as manure for specialized fungus gardens (b). Pyrosequencing of bacterial 16S genes from fungus gardens of the leaf-cutter ants A. colombica and A. cephalotes recovered 8000–12000 sequences (more...)

The fresh-foliar biomass leaf-cutter ants integrate into their fungus gardens is composed largely of recalcitrant lignocellulosic polymers. The ants presumably gain indirect access to the carbon stored in plant cell walls through the metabolic activities of their fungus gardens, which act as an ancillary digestive system (Pinto-Tomas et al., 2009). Despite being a critical aspect of leaf-cutter ant biology, the process through which fungus gardens degrade plant forage has only recently been intensely investigated (De Fine Licht et al., 2010; Schiott et al., 2010; Suen et al., 2010; Semenova et al., 2011). Originally it was thought that the fungal cultivar primarily degraded cellulose, and that this was the main polymer converted into nutrients for the ants (Martin and Weber, 1969). However, the cellulolytic capacity of this fungus has come into question, as it has been shown that pure cultures cannot grow on cellulose as a sole carbon source (Abril and Bucher, 2002). This has led to the suggestion that cellulose is not deconstructed in leaf-cutter ant fungus gardens, but rather that the fungal cultivar uses a variety of hemicellulases to deconstruct primarily starch, xylan and other plant polymers (Gomes De Siqueira et al., 1998; Silva et al., 2006a,b; Schiott et al., 2008).

Another model posits that plant cell wall degradation in fungus gardens is partially mediated by lignocellulolytic bacteria. There is some support for this model. Importantly, recent work has found evidence for substantial cellulose deconstruction in the fungus gardens of Atta colombica and the presence of lignocellulolytic bacteria in these ecosystems (Suen et al., 2010). Another study, employing the culture-independent analysis of membrane-lipid markers, has supported the hypothesis that a distinct community of predominantly Gram-negative bacteria resides in fungus gardens (Scott et al., 2010), and the presence of symbiotic nitrogen-fixing bacteria in the genera Pantoea and Klebsiella has also been shown (Pinto-Tomas et al., 2009). Together with culture-dependent investigations recovering microbial groups with a broad array of metabolic activities (Bacci et al., 1995; Santos et al., 2004), these experiments have led to the suggestion that fungus gardens represent specialized fungus–bacteria consortia selected for by the ants, and that the bacteria have essential roles, including plant biomass degradation, nutrient biosynthesis, and competitive or antibiotic-mediated exclusion of pathogens (Mueller et al., 2005; Haeder et al., 2009; Pinto-Tomas et al., 2009; Suen et al., 2010).

Using a combination of metagenomics and metaproteomics, we provide insights into the microbial activities in leaf-cutter ant fungus gardens. Culture independent investigations have previously been performed on leaf-cutter ant fungus gardens (Scott et al., 2010; Suen et al., 2010), but to date only a small quantity of bacterial sequences (~6 Mb) from the fungus gardens of a single ant species have been characterized. Here, by expanding on previous work, we sought to document the non-eukaryotic component of fungus gardens, describe the similarity of communities from different ant species and examine potential microbial activities in situ. To this end, we generated three 16S pyrotag libraries of over 8,000 sequences each and over 1.2 Gbp of raw 454 Titanium community metagenomic data from the bacterial component of A. cephalotes and A. colombica fungus gardens. To account for potential differences in microbial communities due to the extent of plant biomass degradation, we individually examined the top and bottom strata of A. colombica fungus gardens, which correspond to where the ants integrate fresh forage and remove partially degraded plant substrate, respectively. We then conducted metaproteomic analyses on whole fungus gardens to identify proteins produced in these ecosystems and examine the physiology of resident bacteria in more detail. We found that similar bacterial communities inhabit all fungus garden samples analyzed, and that the metabolic potential of resident bacteria includes nutrient biosynthesis, hemicellulose and oligosaccharide degradation, and other functions that potentially enhance plant biomass processing in these ecosystems. Below we discuss a novel framework for understanding the complex interplay between leafcutter ants and the symbiotic communities residing in their fungus gardens.

Materials and Methods

Sample Processing for Community Metagenomes and 16S Pyrotag Libraries

Fungus gardens from healthy A. cephalotes and A. colombica colonies were collected from nests near Gamboa, Panama, in April 2009. Whole A. cephalotes gardens were combined for subsequent analyses, whereas fungus gardens of A. colombica were laterally bisected to separate the top and bottom strata. Immediately after collection, the bacterial fraction of the samples was isolated and DNA was extracted as previously described (Suen et al., 2010). Briefly, plant, ant and fungal material were removed from all samples through a series of washing or centrifugation steps using 1×PBS (137mM NaCl, 2.7mM KCl, 10mM Na2HPO4, 2mM KH2PO4). DNA was subsequently extracted from the remaining bacterial fraction using a Qiagen DNeasy Plant Maxi Kit (Qiagen Sciences, Germantown, MD, USA). One community metagenome and one 16S library were generated from each of the three samples using 454 Titanium-pyrosequencing technology (Margulies et al., 2005). Draft genomes of three bacteria isolated from Atta fungus gardens were also generated to supplement the reference databases used for the phylogenetic binning of metagenomic data. Technical details for the sequencing, assembly, and annotation of all data can be found in the Supplementary Information.


Metaproteomic analysis was conducted on fungus garden material collected in Gamboa, Panama, from a nest of A. colombica distinct from that used for metagenomic analyses. Moreover, we also conducted metaproteomic analyses on a lab-reared colony of A. sexdens for comparison. Detailed methods can be found in Supplementary Information. Briefly, proteins were extracted from whole fungus-garden material, and the resulting protein solution was digested into peptides and subsequently analyzed by liquid chromatography tandem mass spectrometry (LC-MS/MS). The resulting peptide tandem mass spectra were compared with predicted protein datasets of the three community metagenomes individually. Peptide matches were filtered using Sequest (Eng et al., 1994) scores, MSGF spectral probabilities (Kim et al., 2008), and false discovery rates. We restricted our functional analyses to peptides mapped to proteins phylogenetically binned as bacterial, and the IMG-ER and KEGG annotations of these proteins were inspected to identify those potentially involved in biomass degradation or nutrient cycling (Table A5-5). Peptides mapping to these select proteins were inspected manually (Figure A5-5, Supplementary Dataset 6).

TABLE A5-5. A Subset of Bacterial Proteins Identified in Leaf-Cutter Ant Fungus Gardens Using Liquid Chromatography-Tandem Mass Spectrometry.


A Subset of Bacterial Proteins Identified in Leaf-Cutter Ant Fungus Gardens Using Liquid Chromatography-Tandem Mass Spectrometry.

Two graphs show an example of overlap between field-collected and laboratory-reared fungus-garden samples


Example of overlap between field-collected and laboratory-reared fungus-garden samples for the glycoside hydrolase family 3 peptide N.AIADLLFGDVNPSGK.L. (a) Theoretical b and y ions, identified m/z values are highlighted in red. (b) Field sample MS/MS (more...)


Community metagenomes and 16S pyrotag libraries Pyrosequencing of the V6–V8 variable region of the bacterial 16S rRNA gene for the same three samples yielded between 8000–12,000 reads (termed “pyrotags”) each (Table A5-2). Previous attempts to recover Archaeal 16S sequences from fungus gardens were unsuccessful (Suen et al., 2010), and amplification of these genes was not attempted here. Pyrosequencing of community DNA from three samples, representing both the individual top and bottom strata of A. colombica fungus gardens as well as the combined strata of A. cephalotes gardens, each yielded 382–441Mb of raw sequence data (Table A5-1). Reads from each library were assembled into community metagenomes comprising 40–100 Mbp of sequence data.

TABLE A5-2. Family-Level Classification of Partial-Length 16S Sequences Recovered from Atta Colombica and Atta Cephalotes Fungus Gardens.


Family-Level Classification of Partial-Length 16S Sequences Recovered from Atta Colombica and Atta Cephalotes Fungus Gardens.

TABLE A5-1. Sequencing Statistics of the Community Metagenomes.


Sequencing Statistics of the Community Metagenomes.

Microbial Diversity in Fungus Gardens

Clustering of sequences in the 16S pyrotag libraries from the A. colombica top, A. colombica bottom and A. cephalotes fungus-garden samples recovered 204, 274 and 25 operational taxonomic units (OTUs, 97% identity cutoff), respectively. The majority of the OTUs were most similar to sequences of Gammaproteobacteria and Firmicutes (22–217 OTUs, 72–89% of OTUs, and 2–35 OTUs, 5–12% of OTUs, respectively), and only OTUs corresponding to those phyla were represented in all three samples (Figure A5-1c). Phyla represented in lower abundance and more sporadically included the Betaproteobacteria (≤10 OTUs, ≤4.9% of OTUs), Alphaproteobacteria (≤7 OTUs, ≤3.4% of OTUs), Bacteriodetes (≤4 OTUs, ≤2% of OTUs), Acidobacteria (≤1 OTU, ≤1% of OTUs) and Actinobacteria (≤1 OTU, ≤4% of OTUs) (Figure A5-1c). Most pyrotags corresponded to the Gammaproteobacterial family Enterobacteriaceae (79–99% of individual pyrotags, Table A5-2). Although taxonomic profiles were similar in all three pyrotag libraries, the bacterial diversity of each of the A. colombica samples was greater than that recovered from the A. cephalotes fungus garden sample.

Community metagenomic analyses recovered primarily bacterial sequences (71–80% of total assembled bp) (Table A5-3). Consistent with the 16S pyrotag libraries, the majority of sequences in all three data sets matched most closely to Gammaproteobacteria (69–72%), especially Enterobacteriaceae (53–70%). To refine taxonomic resolution and infer the relative abundance of microbial groups, raw reads were phylogenetically classified using the Genome relative Abundance and Average Size (GAAS) tool (Angly et al., 2009). Estimates based on GAAS analyses indicate that the Gammaproteobacteria were particularly abundant, with the genus Enterobacter comprising over 50% of the bacterial population in all the three metagenomes (Figure A5-1d). The community metagenomes also contained representative sequences from the genera Klebsiella (3.8–4.9%), Pantoea (1.8–15.6%), Escherichia (5.3–6.3%), Citrobacter (3–5.8%), Pseudomonas (0.04–4.2%) and Lactococcus (0.01–2.2%). BLAST-based classification of the assembly indicated that B1% of the sequences corresponded to bacteriophage in each of the community metagenomes (Table A5-3), whereas the GAAS tool estimated that 15.1%, 16.8% and 6.8% of the A. colombica top, A. colombica bottom and A. cephalotes metagenomes could be comprised of bacteriophage, respectively.

TABLE A5-3. Phylogenetic Classification of All Assembled Contigs and Singletons in the Leaf-Cutter Ant Fungus Garden Metagenomes.


Phylogenetic Classification of All Assembled Contigs and Singletons in the Leaf-Cutter Ant Fungus Garden Metagenomes.

Consistent with the GAAS- and BLAST-based analyses, the largest phylogenetic bins created by phymmBL were assigned to the genera Enterobacter, Pantoea, Klebsiella, Escherichia, Citrobacter and Pseudomonas. The Enterobacter bins were by far the largest, containing 15.3–29.5Mb of sequence. The majority of these sequences were most similar to the draft genome of Enterobacter FGI 35, a strain isolated in this study from an A. colombica fungus garden. The Pantoea bins were the next largest, containing between 5–7.2Mb of sequence each.

Metabolic Potential of Bacterial Lineages

To compare the coding potential of different bacterial groups in fungus gardens, we analyzed genus-level phylogenetic bins of sequences constructed from the community metagenomes. Comparison of the coding potential in the bins with the KEGG database (Kanehisa et al., 2008) recovered well-represented sugar metabolism pathways in most of the Enterobacteriaceae bins (Figure A5-2). Moreover, pathways involved in B-vitamin and amino-acid metabolism were found to be highly represented in both the Pseudomonas and Enterobacteriaceae bins. The Lactococcus bins showed relatively low representation in most of these pathways. Clustering of phylogenetic bins from each of the metagenomes by their KEGG pathway representation indicated that bacterial members corresponding to the same genus, with the exception of Citrobacter, had similar metabolic profiles.

Graphical reconstruction of KEGG pathways generated from the leaf-cutter ant fungus-garden metagenomes


Reconstruction of KEGG pathways recovered from phylogenetic bins generated from the leaf-cutter ant fungus-garden metagenomes. KEGG profiles normalized by the number of predicted proteins in each phylogenetic bin were used for the clustering analysis. (more...)

To examine how leaf-cutter ant fungus garden microbial communities differed from other environments, we predicted Clusters of Orthologous Groups (COGs) (Tatusov et al., 2001) from all contigs and reads from the three fungus-garden metagenomes and compared these with COG profiles from all other metagenomes available on the Integrated Microbial Genomes/Microbiomes (IMG/M) database (Markowitz et al., 2008) (Figure A5-3). COG profiles for the three fungus-garden metagenomes were found to be highly similar. Compared with all other metagenomes on IMG, many COG categories were overrepresented in fungus gardens (Fisher’s exact test, P < 0.01), including amino-acid transport and metabolism, carbohydrate transport and metabolism, and inorganic ion transport and metabolism (Figure A5-3). Specific COGs involved in carbohydrate transport and metabolism were analyzed in more detail to investigate possible bacterial roles in polysaccharide degradation, and sugar transporters and phosphotransferase system components in particular were found to be significantly overrepresented in the fungus-garden metagenomes (Fisher’s exact test, P < 0.01) (Supplementary Dataset S2).

A bar graph giving a comparison of the COG category distributions of the three combined fungus-garden metagenomes


Comparison of the COG category distributions of the three combined fungus-garden metagenomes (A. colombica top, A. colombica bottom, and A. cephalotes combined) and all other metagenomes available in IMG. The average COG values are shown ±s.d. (more...)

To further investigate potential bacterial roles in plant-polymer deconstruction, we compared all predicted proteins in the three community metagenomes with the carbohydrate active enzymes (CAZy) database (Cantarel et al., 2009) and identified numerous enzymes potentially involved in this process (Table A5-4). The largest proportion of the identified proteins were most similar to oligosaccharidedegrading enzymes (176–566 CAZymes, 28–30%), and relatively few were found to be predicted cellulases (4–5 CAZymes, 0.2–0.6%). Compared with other well-known lignocellulose-degrading communities such as the Tammar wallaby foregut (Pope et al., 2010) and termite hindgut (Warnecke et al., 2007), fungus gardens contained relatively fewer cellulases and hemicellulases, but similar numbers of oligosaccharide-degrading enzymes.

TABLE A5-4. Partial List of CAZymes Identified in the Leaf-Cutter Ant Fungus-Garden Metagenomes, as Compared with Those Found in the Termite Hindgut and Wallaby Foregut.


Partial List of CAZymes Identified in the Leaf-Cutter Ant Fungus-Garden Metagenomes, as Compared with Those Found in the Termite Hindgut and Wallaby Foregut.

Comparison of Enterobacter Populations

To identify the similarities between the Enterobacter populations across different metagenomes, we performed a fragment recruitment analysis comparing all predicted genes from the Enterobacter FGI 35 phylogenetic bins from each metagenome with the draft Enterobacter FGI 35 genome (Figure A5-4). The fragment-recruitment analysis identified near-uniform coverage of >95% nucleic acid identity BLAST hits across the 33 Enterobacter FGI 35 contigs, with the exception of four regions between 18–66 kb large that we termed variable regions I–IV. Moreover, we found that there was also near-uniform coverage of 70–85% identity BLASTN hits across the draft genome. Investigation of the coding potential in these conserved regions identified genes required for the synthesis of thiamine, pyridoxine, nicotinate, nicotinamide, pantothenate, folate and 19 amino acids. Only the later stages of the histidine biosynthetic pathway could be identified, although the full pathway is present in other Enterobacter contigs. These regions also encoded ABC transporters and phosphotransferase system components predicted to uptake cellobiose, xylose, glucose, sucrose, β-glucosides, arbutin or salicin, N-acetyl-muramic acid, mannitol, mannose, sorbitol, galactitol, L-ascorbate, fructose, ribose, L-arabinose, methylgalactoside, sulfate, sulfonate, spermidine/putrescine, 2-aminoethylphosphonate, iron and other nutrients. The variable regions were found to contain primarily hypothetical genes and genes of unknown function, although some phage integrases were also identified.

A graphical representation of a fragment recruitment analysis of genes phylogenetically binned to Enterobacter FGI 35 against the draft Enterobacter FGI 35 genome


Fragment recruitment analysis of genes phylogenetically binned to Enterobacter FGI 35 against the draft Enterobacter FGI 35 genome. Each point indicates the best BLASTN match of a gene. Tick marks on the bottom indicate contig boundaries of the FGI 35 (more...)


Individual searches of the metaproteomic data against the predicted protein databases of each community metagenome recovered a total of 1186 redundant and 869 non-redundant peptides. Of all the distinct peptides recovered, 129 were found in both laboratory and field samples, while 351 were unique to the laboratory sample and 389 were unique to the field sample. A total of 747, 238 and 201 peptides were recovered for the searches against the A. cephalotes, A. colombica top and A. colombica bottom datasets, respectively. These peptides were mapped onto a total of 653 proteins, of which 354 were predicted from contigs or singletons that were phylogenetically binned as bacterial (see Supplementary Information for details on the phylogenetic binning procedure). The majority of bacterial proteins identified were predicted to belong to the Enterobacteriaceae, and functions predicted from these proteins included a variety of metabolic processes (Table A5-5, Supplementary Dataset 5). Figure A5-5 highlights the overlap observed between laboratory-reared samples and field-collected samples for one peptide mapped to a predicted glycosyl hydrolase. Details for all mass spectra and the annotations for the bacterial proteins they mapped to can be found in Supplementary Datasets 3 and 5, respectively.


Leaf-cutter ants are dominant New World herbivores, foraging on up to 17% of the foliar biomass in some ecosystems (Costa et al., 2009). In 1874 Thomas Belt established that leaf-cutters do not consume leaf material directly, as had been previously assumed, but instead use it as manure to cultivate a fungus for food in specialized gardens (Belt, 1874). For over a hundred years after Belt’s pioneering discovery it was believed that the fungus gardens of leaf-cutter ants represented a monoculture of the fungal cultivar that degraded plant cellwall material and converted it into nutrients for the ants (Weber, 1966; Martin and Weber, 1969). However, both the lignocellulolytic capacity of the cultivar and the view that fungus gardens are composed solely of the fungal mutualist have been recently challenged (Gomes De Siqueira et al., 1998; Abril and Bucher, 2002; Scott et al., 2010; Suen et al., 2010). In this study, we explored the hypothesis that bacteria are common constituents of fungus gardens that could be participating in plant biomass degradation and nutrient cycling.

Our work demonstrates that a distinct community of bacteria resides in the fungus gardens of A. colombica and A. cephalotes leaf cutter ants. Our identification of similar bacterial groups in fungus-garden samples taken from different ant species and garden strata supports this conclusion. Moreover, this is consistent with our finding that relatively few bacterial genera comprise the majority of the metagenomic sequences recovered in this study (see below). This, combined with the previous work on nitrogen fixation, plant biomass degradation and membrane-lipid profiles in these ecosystems, indicates that bacteria are long-term residents of fungus gardens and not merely allochthonous organisms introduced from leaf material or the surrounding soil (Bacci et al., 1995; Pinto-Tomas et al., 2009; Scott et al., 2010; Suen et al., 2010). Thus, the term “fungus garden” may be misleading, as these environments are composed of a fungus–bacteria community.

The bacterial component of the microbial ecosystem in fungus gardens appears to be dominated by only a few groups. Specifically, the genera Enterobacter, Klebsiella, Citrobacter, Escherichia and Pantoea represent over two-thirds of the bacterial component in each of the community metagenomes (Figure A5-1d). This narrow genus-level diversity is likely the result of both the nutrient composition of the plant–fungal matrix and the meticulous hygienic practices of the ants. For example, leaf-cutters continuously weed their gardens to remove areas infected with microbial pathogens (Currie and Stuart, 2001), and also apply antimicrobials derived from both glandular secretions and symbiotic actinobacteria (Currie et al., 1999; Fernández-Marín et al., 2006). The extent of plant biomass degradation could also affect microbial diversity, but if this was a critical factor we would expect to find distinct communities between top and bottom garden strata, which contain fresh leaf material and largely degraded biomass, respectively. The similarity between different strata observed here, consistent with previous work reporting little difference between 16S libraries constructed from these two regions (Suen et al., 2010), indicates that the extent of plant biomass degradation is not a major contributor to community structuring. The consistent presence of bacterial groups within the Enterobacteriaceae throughout different garden strata and leaf-cutter ant species implicates them as having a consistent role in fungus gardens, and suggests that these environments represent highly structured communities rather than a random collection of opportunistic microbes. Although it remains a possibility that while removing the fungal matrix and plant debris from fungus gardens our analysis excluded microbial groups adhering to fungal or plant biomass, thereby skewing the composition of the metagenomes, our results are generally consistent with previous culture-independent investigations that either analyzed whole fungus gardens or utilized different methods to isolate bacterial cells (Scott et al., 2010; Suen et al., 2010). Moreover, our processing of fresh rather than frozen fungusgarden material may be partially responsible for our success in removing fungal or plant debris from our samples.

Bacteria of the genus Enterobacter appear to be particularly prevalent in fungus gardens. In contrast to the narrow genus-level diversity observed in these environments, multiple species of Enterobacter appear to be present in all the gardens analyzed. Our fragment-recruitment analysis demonstrates that populations of bacteria with >95% and 70–85% nucleic acid identity to the reference Enterobacter FGI 35 genome exist in these environments (Figure A5-4). The four large gaps identified in the recruitment plot likely represent prophage or other variable elements in the reference genome. Because Enterobacter FGI 35 was isolated from an A. colombica fungus garden, the near-uniform coverage of genes at >95% identity across all metagenomes indicates that highly similar strains of Enterobacter are present in all of the samples analyzed. The near-uniform coverage of genes at 70–85% identity likely represents multiple distinct species, as it is improbable that genes from a single population of bacteria would have such a large range of nucleotide identity to a single reference genome. Genes 70–85% identical to the Enterobacter FGI 35 genome may represent divergent Enterobacter species or even novel Enterobacteriaceae for which an appropriate reference for phylogenetic binning does not exist. That different species of leaf-cutter ant harbor abundant Enterobacter populations indicates that this group may be an important constituent of the fungus garden community.

The overall functional potential of the metagenomes includes a diversity of bacterial genes associated with plant biomass degradation, supporting previous work that has suggested a role for bacteria in this process. The vast majority of CAZymes identified in the metagenomes are associated with oligosaccharide degradation or simple sugar metabolism, suggesting that bacteria are processing partially degraded plant material. We also found KEGG pathways involved in hexose and pentose sugar metabolism to be highly represented in the Enterobacteriaceae, indicating that sugar monomers can be readily metabolized by many of these bacteria. Moreover, our KEGG, COG and metaproteomic analyses recovered numerous sugar transporters (Figure A5-2, Table A5-5, Supplementary Dataset 2), including a large number of cellobiose-specific phosphotransferase system components that are known to be involved in the uptake of the byproducts of cellulose hydrolysis (Figure A5-1, Table A5-5, Supplementary Dataset 2). Together, these data suggest that bacterial community members are metabolizing predominantly partially degraded plant material, although it remains a possibility that unidentified bacterial lignocellulases also have a role in the degradation of more recalcitrant biomass.

Bacterial lineages in fungus gardens were also found to possess diverse biosynthetic pathways. Pathways involved in amino-acid and B-vitamin metabolism were particularly well-represented in the detected Enterobacteriaceae and Pseudomonas sequences, and biosynthetic pathways for thiamin, pyridoxine, nicotinate, nicotinamide, pantothenate, folate, and all 20 amino acids could be reconstructed from the Enterobacter bins. As mentioned above, enzymes involved in the metabolism of oligosaccharides and simple sugars were also identified in many of these groups, indicating that they may convert carbon-rich plant biomass into amino acids, B-vitamins, proteins or other nutrients. Previous work has indicated that bacteria have a role in the introduction and cycling of nitrogen in fungus gardens (Pinto-Tomas et al., 2009). Together with our work, this suggests that the combined metabolism of resident bacteria may enrich the nutrient composition of fungus gardens through the conversion of carbohydrate-rich oligosaccharides into a variety of other nutrients that could promote the growth of the fungal cultivar or even nourish the ants themselves.

Our metaproteomic analysis recovered peptides mapping to bacterial proteins predicted to participate in biomass degradation and nutrient biosynthesis, supporting the results of our metagenomic characterization and further indicating that bacteria are involved in these processes (Table A5-5, Supplementary Datasets 3, 5, and 6). Our manual inspection of the metaproteomic data identified multiple peptides belonging to glycoside hydrolases, sugar transporters and amino acid and B-vitamin biosynthetic pathways. That multiple peptides could be assigned to proteins with similar predicted functions indicates that these processes may be prevalent in fungus gardens. Moreover, many of the mapped peptides originated from both laboratoryreared and field-collected samples, including one that belonged to a family 3 glycosyl hydrolase (Figure A5-5). Although these data should be interpreted cautiously due to the few bacterial proteins identified overall, this may indicate physiological similarities between bacteria in laboratoryreared versus field-collected colonies.

Not all bacteria in fungus gardens were found to have substantial biosynthetic capacity, and in particular the Lactococcus groups appeared to have limited coding potential in the majority of pathways analyzed. This may be a result of lower sequencing coverage, as only a relatively small fraction of the metagenomes was predicted to belong to these groups. Alternatively, these groups may not be contributing substantially to nutrient cycling and are able to subsist on free sugars and other nutrients available in fungus gardens. Importantly, the byproducts of Lactococci metabolism may acidify fungus gardens and contribute to the maintenance of the lower pH in these ecosystems, which has previously been observed at 4.4–5.0 (Powell and Stradling, 1986). Regulation of the pH of fungus gardens to this narrow range has been hypothesized to be critical to the growth of fungal cultivar, but the mechanism through which this occurs has remained unknown (Powell and Stradling, 1986). Few peptides from our metaproteomic data sets were recovered from this group, indicating that they may be present in low abundance.

In addition to bacteria, we also found that fungus gardens contain substantial populations of bacteriophage (Figure A5-1). These organisms could play key roles by limiting bacterial abundance or decreasing ecosystem productivity. Moreover, because fungus gardens contain numerous closely related genera in the Enterobacteriaceae, bacteriophage could provide a common mechanism for gene transfer between lineages. The presence of bacteriophage in fungus gardens adds to the number of organisms that are shaping these ecosystems and introduces a new layer of complexity into the ecology of fungus gardens.

Metagenomics and metaproteomics have previously been shown to be invaluable tools for analyzing microbial communities (Ram et al., 2005; Gill et al., 2006; Woyke et al., 2006; Kalyuzhnaya et al., 2008; Wilmes et al., 2008; Allgaier et al., 2009; Verberkmoes et al., 2009; Burnum et al., 2011), including those associated with herbivores (Warnecke et al., 2007; Brulc et al., 2009; Pope et al., 2010; Burnum et al., 2011). Here we use these techniques to provide insight into the fungus gardens of leaf-cutter ants. Our work shows that relatively few genera dominate the bacterial fraction of these communities, and that the genus Enterobacter appears to be particularly prevalent. We show that bacteria have diverse metabolic potential associated with the degradation of plant biomass, and we confirm the production of two bacterial glycoside hydrolases in situ. Moreover, we show that bacteria in fungus gardens likely participate in the biosynthesis of amino acids, B-vitamins and other nutrients that potentially enhance the growth or biomass-processing efficiency of the fungal cultivar. This is consistent with a model of synergistic biomass degradation by a fungus–bacteria consortium. Our work enhances our knowledge of how leaf-cutter ants process massive quantities of plant biomass in their ancillary digestive systems, and underscores the importance of symbiotic communities on the evolution and ecology of herbivores.


We thank the staff of the Joint Genome Institute, Pacific Northwest National Laboratories, and the Smithsonian Tropical Research Institute for their expertise and support in the collection and processing of all samples, in particular S Malfatti, L Seid, Y Clemons, R Urriola, M Paz and O Arosemena. We thank all members of the Currie lab for their comments on the manuscript. We also thank three anonymous reviewers for their comments on the manuscript. The US Department of Energy Joint Genome Institute effort was supported by the Office of Science of the US Department of Energy under Contract No. DE-AC02-05CH11231. Proteomic work was performed in the Environmental Molecular Sciences Laboratory, a US Department of Energy (DOE) Office of Biological and Environmental Research national scientific user facility on the Pacific Northwest National Laboratory (PNNL) campus. Portions of this research were supported by the US Department of Energy’s (DOE) Office of Biological and Environmental Research (OBER) Panomics program. PNNL is a multiprogram national laboratory operated by Battelle for the DOE under Contract DE-AC05-76RL01830. This work is also supported by the National Science Foundation (grants DEB-0747002, MCB-0702025, and MCB-0731822 to CRC) and the DOE Great Lakes Bioenergy Research Center (DOE Office of Science BER DE-FC02-07ER64494).

Supplementary Information accompanies the paper on The ISME Journal website (


  • Abril AB, Bucher EH. Evidence that the fungus cultured by leaf-cutting ants does not metabolize cellulose. Ecol Lett. 2002;5:325–328.
  • Allgaier M, Reddy A, Park JI, Ivanova N, D’Haeseleer P, Lowry S, et al. Targeted discovery of glycoside hydrolases from a switchgrass-adapted compost community. PLoS One. 2009;5:e8812. [PMC free article: PMC2809096] [PubMed: 20098679]
  • Angly FE, Willner D, Prieto-Davo A, Edwards RA, Schmieder R, Vega-Thurber R, et al. The GAAS metagenomic tool and its estimations of viral and microbial average genome size in four major biomes. PLoS Comput Biol. 2009;5:e1000593. [PMC free article: PMC2781106] [PubMed: 20011103]
  • Bacci M, Ribeiro SB, Casarotto MEF, Pagnocca FC. Biopolymer-degrading bacteria from nests of the leafcutting ant Atta Sexdens Rubropilosa. Braz J Med Biol Res. 1995;28:79–82.
  • Belt T. The Naturalist in Nicaragua. E. Bumpus; London: 1874.
  • Brulc JM, Antonopoulos DA, Miller ME, Wilson MK, Yannarell AC, Dinsdale EA, et al. Gene-centric metagenomics of the fiber-adherent bovine rumen microbiome reveals forage specific glycoside hydrolases. Proc Natl Acad Sci USA. 2009;106:1948–1953. [PMC free article: PMC2633212] [PubMed: 19181843]
  • Burnum KE, Callister SJ, Nicora CD, Purvine SO, Hugenholtz P, Warnecke F, et al. Proteome insights into the symbiotic relationship between a captive colony of Nasutitermes corniger and its hindgut microbiome. ISME J. 2011;5:161–164. [PMC free article: PMC3105668] [PubMed: 20613792]
  • Cantarel BL, Coutinho PM, Rancurel C, Bernard T, Lombard V, Henrissat B. The Carbohydrate-Active EnZymes database (CAZy): an expert resource for glycogenomics. Nucleic Acids Res. 2009;37(Database issue):D233–D238. [PMC free article: PMC2686590] [PubMed: 18838391]
  • Chapela IH, Rehner SA, Schultz TR, Mueller UG. Evolutionary history of the symbiosis between fungus-growing ants and their fungi. Science. 1994;266:1691–1694. [PubMed: 17775630]
  • Costa AN, Vasconcelos HL, Vieira-Neto EHM, Bruna EM. Do herbivores exert top-down effects in Neotropical savannas? Estimates of biomass consumption by leaf-cutter ants. J Veg Sci. 2009;19:849–854.
  • Currie CR, Scott JA, Summerbell RC, Malloch D. Fungus-growing ants use antibiotic-producing bacteria to control garden parasites. Nature. 1999;398:701–704.
  • Currie CR, Stuart AE. Weeding and grooming of pathogens in agriculture by ants. Proc R Soc Lond B Biol Sci. 2001;268:1033–1039. [PMC free article: PMC1088705] [PubMed: 11375087]
  • De Fine Licht HH, Schiott M, Mueller UG, Boomsma JJ. Evolutionary transitions in enzyme activity of ant fungus gardens. Evolution. 2010;64:2055–2069. [PubMed: 20067517]
  • Eng J, McCormack A, Yates J. An approach to correlate tandem mass spectral data of peptides with amino acid sequences in a protein database. J Am Soc Mass Spectrom. 1994;5:976–989. [PubMed: 24226387]
  • Fernández-Marín H, Zimmerman JK, Rehner SA, Wcislo WT. Active use of the metaplural glands by ants in controlling fungal infection. Proc R Soc London B Biol Sci. 2006;273:1689–1695. [PMC free article: PMC1634922] [PubMed: 16769642]
  • Gill SR, Pop M, Deboy RT, Eckburg PB, Turnbaugh PJ, Samuel BS, et al. Metagenomic analysis of the human distal gut microbiome. Science. 2006;312:1355–1359. [PMC free article: PMC3027896] [PubMed: 16741115]
  • Gomes De Siqueira C, Bacci M Jr, Pagnocca FC, Bueno OC, Hebling MJA. Metabolism of plant polysaccharides by Leucoagaricus gongylophorus, the symbiotic fungus of the leaf-cutting ant Atta sexdens. Appl Environ Microbiol. 1998;64:4820–4822. [PMC free article: PMC90928] [PubMed: 9835568]
  • Haeder S, Wirth R, Herz H, Spiteller D. Candicidinproducing Streptomyces support leaf-cutting ants to protect their fungus garden against the pathogenic fungus Escovopsis. Proc Natl Acad Sci USA. 2009;106:4742–4746. [PMC free article: PMC2660719] [PubMed: 19270078]
  • Hölldobler B, Wilson EO. The Ants. Harvard University Press; Cambridge, MA: 1990.
  • Hölldobler B, Wilson EO. The Superorganism: The Beauty, Elegance, and Strangeness of Insect Societies. W. W. Norton & Company; New York, NY: 2008.
  • Hölldobler B, Wilson EO. The Leafcutter Ants: Civilization by Instinct. W. W. Norton & Company; New York, NY: 2010.
  • Kalyuzhnaya MG, Lapidus A, Ivanova N, Copeland AC, McHardy AC, Szeto E, et al. High-resolution metagenomics targets specific functional types in complex microbial communities. Nat Biotechnol. 2008;26:1029–1034. [PubMed: 18711340]
  • Kanehisa M, Araki M, Goto S, Hattori M, Hirakawa M, Itoh M, et al. KEGG for linking genomes to life and the environment. Nucleic Acids Res. 2008;36(Database issue):D480–D484. [PMC free article: PMC2238879] [PubMed: 18077471]
  • Kim S, Gupta N, Pevzner PA. Spectral probabilities and generating functions of tandem mass spectra: a strike against decoy databases. J Proteome Res. 2008;7:3354–3363. [PMC free article: PMC2689316] [PubMed: 18597511]
  • Margulies M, Egholm M, Altman WE, Attiya S, Bader JS, Bemben LA, et al. Genome sequencing in microfabricated high-density picolitre reactors. Nature. 2005;437:376–380. [PMC free article: PMC1464427] [PubMed: 16056220]
  • Markowitz VM, Ivanova NN, Szeto E, Palaniappan K, Chu K, Dalevi D, et al. IMG/M: a data management and analysis system for metagenomes. Nucleic Acids Res. 2008;36(Database issue):D534–D538. [PMC free article: PMC2238950] [PubMed: 17932063]
  • Martin MM, Weber NA. The cellulose-utilizing capability of the fungus cultured by the attine ant Atta colombica tonsipes. Ann Entomol Soc Am. 1969;62:1386–1387. [PubMed: 5374175]
  • Mueller U, Gerardo N, Aanen D, Six D, Schultz T. The evolution of agriculture in insects. Annu Rev Ecol Evol Syst. 2005;36:563–595.
  • Pinto-Tomas AA, Anderson MA, Suen G, Stevenson DM, Chu FS, Cleland WW, et al. Symbiotic nitrogen fixation in the fungus gardens of leaf-cutter ants. Science. 2009;326:1120–1123. [PubMed: 19965433]
  • Pope PB, Denman SE, Jones M, Tringe SG, Barry K, Malfatti SA, et al. Adaptation to herbivory by the Tammar wallaby includes bacterial and glycoside hydrolase profiles different from other herbivores. Proc Natl Acad Sci USA. 2010;107:14793–14798. [PMC free article: PMC2930436] [PubMed: 20668243]
  • Powell RJ, Stradling DJ. Factors influencing the growth of Attamyces bromatificus, a symbiont of the attine ants. Trans Br Mycol Soc. 1986;87:205–213.
  • Ram RJ, Verberkmoes NC, Thelen MP, Tyson GW, Baker BJ, Blake RC II, et al. Community proteomics of a natural microbial biofilm. Science. 2005;308:1915–1920. [PubMed: 15879173]
  • Santos AV, Dillon RJ, Dillon VM, Reynolds SE, Samuels RI. Ocurrence of the antibiotic producing bacterium Burkholderia sp. in colonies of the leafcutting ant Atta sexdens rubropilosa. FEMS Microbiol Lett. 2004;239:319–323. [PubMed: 15476982]
  • Schiott M, De Fine Licht HH, Lange L, Boomsma JJ. Towards a molecular understanding of symbiont function: identification of a fungal gene for the degradation of xylan in the fungus gardens of leafcutting ants. BMC Microbiol. 2008;8:40. [PMC free article: PMC2291056] [PubMed: 18307762]
  • Schiott M, Rogowska-Wrzesinska A, Roepstorff P, Boomsma JJ. Leaf-cutting ant fungi produce cell wall degrading pectinase complexes reminiscent of phytopathogenic fungi. BMC Biol. 2010;8:156. [PMC free article: PMC3022778] [PubMed: 21194476]
  • Schultz TR, Brady SG. Major evolutionary transitions in ant agriculture. Proc Natl Acad Sci USA. 2008;105:5435–5440. [PMC free article: PMC2291119] [PubMed: 18362345]
  • Scott JJ, Budsberg KJ, Suen G, Wixon DL, Balser TC, Currie CR. Microbial community structure of leafcutter ant fungus gardens and refuse dumps. PLoS One. 2010;5:e9922. [PMC free article: PMC2847949] [PubMed: 20360970]
  • Semenova TA, Hughes DP, Boomsma JJ, Schiott M. Evolutionary patterns of proteinase activity in attine ant fungus gardens. BMC Microbiol. 2011;11:15. [PMC free article: PMC3033787] [PubMed: 21247468]
  • Silva A, Bacci M Jr, Pagnocca FC, Bueno OC, Hebling MJ. Production of polysaccharidases in different carbon sources by Leucoagaricus gongylophorus Moller (Singer), the symbiotic fungus of the leaf-cutting ant Atta sexdens Linnaeus. Curr Microbiol. 2006;53:68–71. [PubMed: 16775790]
  • Silva A, Bacci M Jr, Pagnocca FC, Bueno OC, Hebling MJ. Starch metabolism in Leucoagaricus gongylophorus, the symbiotic fungus of leaf-cutting ants. Microbiol Res. 2006;161:299–303. [PubMed: 16380244]
  • Suen G, Scott JJ, Aylward FO, Adams SM, Tringe SG, Pinto-Tomas AA, et al. An insect herbivore microbiome with high plant biomass-degrading capacity. PLoS Genet. 2010;6:e1001129. [PMC free article: PMC2944797] [PubMed: 20885794]
  • Tatusov RL, Natale DA, Garkavtsev IV, Tatusova TA, Shankavaram UT, Rao BS, et al. The COG database: new developments in phylogenetic classification of proteins from complete genomes. Nucleic Acids Res. 2001;29:22–28. [PMC free article: PMC29819] [PubMed: 11125040]
  • Verberkmoes NC, Russell AL, Shah M, Godzik A, Rosenquist M, Halfvarson J, et al. Shotgun metaproteomics of the human distal gut microbiota. ISME J. 2009;3:179–189. [PubMed: 18971961]
  • Warnecke F, Luginbuhl P, Ivanova N, Ghassemian M, Richardson TH, Stege JT, et al. Metagenomic and functional analysis of hindgut microbiota of a wood-feeding higher termite. Nature. 2007;450:560–565. [PubMed: 18033299]
  • Weber NA. Fungus-growing ants. Science. 1966;153:587–604. [PubMed: 17757227]
  • Wilmes P, Wexler M, Bond PL. Metaproteomics provides functional insight into activated sludge wastewater treatment. PLoS One. 2008;3:e1778. [PMC free article: PMC2289847] [PubMed: 18392150]
  • Woyke T, Teeling H, Ivanova NN, Huntemann M, Richter M, Gloeckner FO, et al. Symbiosis insights through metagenomic analysis of a microbial consortium. Nature. 2006;443:950–955. [PubMed: 16980956]





25 University of California, Davis, California.

Note: this paper is based on a transcript of a talk given at the IOM Forum on Microbial Threats in March, 2012. Only minor modifications have been made (e.g., additional of section headers, addition of references, removal of side comments) in order to as accurately as possible reflect the presentation. A recording of the talk with slides is available on YouTube at Because my presentation was in essence a review of my work in the area, this should not be viewed as a review of the field but rather of my work in this area.

Acknowledgements and Introduction

Thank you. I guess I have the awkward after-lunch talk here, so I will try not to use the most complicated slide I used, although I am not so sure about that. Since I frequently don’t get to my last slide, I just want to do like some other people have been doing, acknowledgements at the beginning. And what I am going to talk about is work funded by a lot of different agencies that has gone on in my lab for about 10, 15 years, including in particular work funded by the Department of Energy, the National Science Foundation, the Gordon and Betty Moore Foundation, and recently Homeland Security, all related to phylogenetic analysis of genomes and metagenomes. And there are a lot of people I will mention, many of the people involved in this. But this is the trans-disciplinary type of work. It hurts my head a lot of time to think of all the people involved in some of these projects, but I will try to acknowledge as many of them as possible.

So what I am going to do is give just a quick introduction to phylogeny and then talk about three examples of the uses of phylogeny in studying microbial communities via DNA sequencing—phylotyping, functional prediction (just a tiny bit, because I want to raise the point as [the topic] has come up a few times [at this meeting], and then selection of organisms for study. And then I will end with just a couple of things about future directions.

What Is Phylogeny?

I assume most people here know what phylogeny is, but just [a quick reminder]: phylogeny is a representation of the history of entities, and that could be the history of genes, the history of genomes, the history of species. And in many cases, people have represented this history by a bifurcating tree-like structure. Phylogeny doesn’t have to be represented as a bifurcating tree-like structure. We can have reticulation events, like recombination and lateral gene transfer. I include all those complexities within the concepts of phylogeny, so I am not trying to discriminate between vertical evolution versus lateral evolution, but really this sort of representation of the history of organisms. I am also not going to get into the debates about what that exact history is. People are still trying to resolve the evolutionary history of microbes as well as other organisms, and it is a constant area of research.

Whatever your belief of the latest model is, in my opinion if you incorporate phylogenetic approaches in your analysis of genome and metagenome and other data, it can improve what you are doing relative to not trying to incorporate phylogenetic approaches. And what I am going to do is try and walk you through a couple of examples of this.

Example I: Phylotyping

The first one I want to talk about is phylotyping, which we have heard either directly or indirectly a lot about at this meeting. Phylotyping, I was exposed to as a young, budding scientist in the lab of Colleen Cavanaugh. I was an undergraduate at Harvard and ended up in Colleen’s lab, and I spent a year and a half sequencing one 16s ribosomal RNA gene. But I got a paper out of that one 16s ribosomal RNA gene (Eisen et al., 1992). And the point of sequencing that 16s ribosomal RNA gene, as well as the point, even today in many cases, of ribosomal RNA sequencing, is to try and figure out what the organism is related to where that 16s came from.

And the way phylotyping works, this is basically developed by Norm Pace and colleagues (e.g., Hugenholtz et al., 1988). You collect DNA from your sample, you clone out some sequence like ribosomal RNA, you build an evolutionary tree of that sequence. So this is where the phylo part comes in phylotyping, a phylogenetic tree of that sequence. And you compare your unknowns to known things that are out there. And this is the tree from our Solemya velum chemosynthetic symbiont 16s, which by the way was accepted 20 years ago tomorrow, I think, my first scientific paper (see Figure A6-1).

An unrooted phylogenetic tree showing the position of the S. velum symbionts in relation to that of other Proteobacteria species


Unrooted phylogenetic tree showing the position of the S. velum symbionts in relation to that of other Proteobacteria species on the basis of 16s rRNA gene sequences. The tree was constructed from evolutionary distances in Table 1. Members of the alpha (more...)

Ribosomal RNA phylotyping has been amazing at revolutionizing our understanding of microbes in the world. I assume most people here appreciate the vast diversity of things that have been discovered by using phylogenetic trees of ribosomal RNAs from the environment to understand what the organisms are that those ribosomal RNAs came from. I am not going to go into the whole history of this. What I really want to talk about is three challenges that now come up with phylogenetic typing that largely relate to this issue of the cost of sequencing dropping and dropping and dropping and getting easier and easier and easier, accelerating at a rate faster than Moore’s Law.

And with new sequencing machines being announced every 2 or 3 days, not that all of them work, but there are all sorts of cool things coming out there. And so this affects things like PCR amplification of ribosomal RNA sequences. We now have literally trillions of ribosomal RNA sequences to analyze, as opposed to that one that I got a paper out of. It also is really important in terms of revolutionizing metagenomic approaches. And I appreciate what Jo [Handelsman] was talking about, with metagenomics is not everything about a community, but the cheaper sequencing is, the more data we are going to have for metagenomics. Even though it doesn’t tell us everything we need to analyze that data. And a third challenge is that most of the DNA sequencing technologies that people have been using generate short sequence reads, as opposed to long contigs that are easier to analyze.

And so it is sort of obvious that when metagenomic data was generated, you could scan through the metagenomic data to build evolutionary trees of sequences that were in that data. And this is what I did with Craig Venter in the analysis of the Sargasso Sea data (Venter et al., 2004). You can scan through the metagenomes, find ribosomal RNA sequences, and build evolutionary trees of those ribosomal RNAs just like we did with PCR amplified data (Figure A6-2).

An illustration of an rRNA tree


rRNA tree. Phylogenetic tree of 16s rRNA. Phylogenetic trees are shown for this gene, with sequences from this study colored red, and with major phylogenetic groups outlined (clades of sequences that could not be assigned to any group are labeled as “Unknown”). (more...)

The great thing about metagenomic data is, we can build phylogenetic trees of other genes that are good phylogenetic markers that we never could really get a good sample of most of these because PCR amplification of protein coding genes across broad diversity does not work very well. So now with metagenomic data we can look at protein coding genes and compare and contrast the results with those to the results with ribosomal RNA. I have been obsessed with the RecA gene for a long time (e.g., Eisen, 199526), so I always end up working on RecA (Figure A6-3).

An illustration of a RecA tree


RecA tree. Phylogenetic tree of 16s rRNA. Phylogenetic trees are shown for this gene, with sequences from this study colored Red, and with major phylogenetic groups outlined (clades of sequences that could not be assigned to any group are labeled as “Unknown”). (more...)

But there are lots of others genes that you can analyze, and we did this in the Sargasso Sea analysis. And if you compare and contrast the results that you get with phylogenetic typing of protein coding genes and assigning those types into phyla, into bins that correspond to the phyla of organisms, you see some interesting patterns (Figure A6-4).

A graph showing phylogenetic diversity of Sargasso Sea sequences using multiple phylogenetic markers


Phylogenetic diversity of Sargasso Sea sequences using multiple phylogenetic markers. The relative contribution of organisms from different major phylogenetic groups (phylotypes) was measured using multiple phylogenetic markers that have been used previously (more...)

There are some differences between what you get with ribosomal RNA and what you get with protein coding genes. I think a lot of this is due to the differences in copy number. So if you estimate relative abundance of organisms from ribosomal RNA, the copy number of ribosomal RNA varies a lot between taxa, but the copy number of many protein coding genes does not vary a lot between taxa. So the protein coding genes, even though they are not as richly sampled, are probably better markers for estimating relative abundance than ribosomal RNA sequences. People have been doing this now with metagenomic data in many different contexts.

Need for Automation

I am not going to cover all of the phylogenetic approaches to metagenomic data. But what I want to talk about is this issue of automation. I think as we get more and more sequence data, we can’t look at trees any more. We can’t look at sequence alignments. We can’t even handle all the data at all. But we certainly need to automate everything.

There are multiple strategies to trying to automate phylogenetic typing of ribosomal RNA or metagenomic data. And one of them has been to use the BLAST program (Altschul et al., 1990), or analogs of the BLAST program, which basically looks at sequence similarity of your sequence to sequences in databases. This is not the best approach to analyzing data. Percent similarity or other measures of similarity are not a good indicator of evolutionary relatedness and can produce misleading patterns about the taxonomy and other parts of information that you want to analyze (Eisen, 1998). There are also approaches that look at compositional and word frequencies. Now both of these approaches are very fast, so you can generate a lot of results very rapidly, and that can be an advantage in many cases. But phylogenetic analysis is generally better than most of these approaches, and the challenge is, how do you implement phylogenetic analysis on a massive scale?

And so what I am going to do is just give you sort of four examples of some of the issues related to implementing automated phylogenetic analysis on a large scale.

Method 1: Each Sequence Is an Island

You can scan through the data and say I am going to take each individual sequence, each individual new thing that I get, and build an evolutionary tree of it relative to known things. So in essence, each sequence is an island in and of itself. So we have done this with a variety of tools. We built our automated ribosomal RNA tool called STAP (Wu et al., 2008), which goes through and takes a reference alignment of known ribosomal RNAs and then for each new sequence aligns your new sequence to that and builds an evolutionary tree in a completely automated manner, and then can scan through the tree to look at the taxonomy results from the tree (Figure A6-5).

A flow chart of the STAP pipeline


A flow chart of the STAP pipeline. SOURCE: Wu et al. (2008).

We also built a tool that will do this with protein coating genes, called AMPHORA (Wu and Eisen, 2008). So it can automatically scan through metagenomic data, find homologues of particular protein families, build an alignment of them, build an evolutionary tree of them (Figure A6-6). And if you have a good reference alignment from known organisms, you can identify a candidate—sort of identify what those protein coating genes come from in your environmental sample (Figure A6-7).

A flow chart illustrating the major components of AMPHORA


A flow chart illustrating the major components of AMPHORA. The marker protein sequences from representative genomes are retrieved, aligned, and masked. Profile hidden Markov models (HMMs) are then built from those “seed” alignments. New (more...)

An illustration of an unrooted maximum likelihood bacterial genome tree


An unrooted maximum likelihood bacterial genome tree. The tree was constructed from concatenated protein sequence alignments derived from 31 housekeeping genes. All major phyla are separated into their monophyletic groups and are highlighted by color. (more...)

Now having a good reference database is challenging for protein coating sequences, whereas we have now trillions of ribosomal RNA sequences and tens of thousands of complete ribosomal RNA sequences. We don’t have good databases of protein coating sequences. All of the good data are now coming from genome sequencing projects. So whatever has been sequenced in terms of genomes is basically our source of protein coating genes for building these evolutionary trees. And so you build a reference tree from the genomes, you take your new data, stream them against the reference tree, build a new tree with that, and assign your new sequence to somewhere compared to the reference data. And we have done this with this AMPHORA. It can allow you to stream through massive metagenomic data sets, and do taxonomic assignments for a suite of protein coating genes, just again like you would do with ribosomal RNA sequences (Figure A6-8).

A graph showing major phylotypes identified in Sargasso Sea metagenomic data


Major phylotypes identified in Sargasso Sea metagenomic data. The metagenomic data previously obtained from the Sargasso Sea was reanalyzed using AMPHORA and the 31 protein phylogenetic markers. The microbial diversity profiles obtained from individual (more...)

And again, I think this is very advantageous in particular because of the copy number variation with ribosomal RNA. We have shown that it is better than similarity-based approaches (Figure A6-9).

Two graphs show a comparison of the phylotyping performance by AMPHORA and MEGAN


Comparison of the phylotyping performance by AMPHORA and MEGAN. The sensitivity and specificity of the phylotyping methods were measured across taxonomic ranks using simulated Sanger shotgun sequences of 31 genes from 100 representative bacterial genomes. (more...)

Method 2: Most in the Family

This [approach involves] analyzing each individual sequence on its own. But of course, when you sequence from a new environment, you also want to compare the new sequences to each other. You don’t want to compare each one individually to the reference data. And so there are a lot of methods that people have been trying to develop to build evolutionary trees of all the new sequences compared to each other.

One of the challenges with this is when you have metagenomic data in particular, the new sequences that you get might not correspond to the entire length of the reference sequences that you are analyzing. So you might have an alignment that looks like this (Figure A6-10).

An illustration of a hypothetical multiple sequence alignment including full length fireferencefl sequences as well as fragmentary sequences from metagenomic data


Hypothetical multiple sequence alignment including full length “reference” sequences as well as fragmentary sequences from metagenomic data. Xs represent areas where a sequence lines up with other sequences. Dashes represent gaps in the (more...)

One solution to this is to just trim the alignment and only pick out regions from the metagenomic data that overlap with everything in your reference database. We and other people have built methods to do this, to go through, take all the new sequences, align them to the reference data, and chop out a core region that everything has, and build an evolutionary tree of that region (Figure A6-11).

An illustration of a hypothetical multiple sequence alignment showing one approach to carrying out phylogenetic analysis of metagenomic data


Hypothetical multiple sequence alignment showing one approach to carrying out phylogenetic analysis of metagenomic data—to extract a “core” region of the alignment and only analyze sequences that contain most of this core. Xs represent (more...)

I did this by hand, to analyze the Sargasso data with ribosomal RNAs and a variety of other sequences including RecA, et cetera. All of that was done by hand. It is much better to automate this. So we have added a step in this for this ribosomal RNA pipeline. There are many other tools to do this with ribosomal RNA. Qiime (Caporaso et al., 2010), mother (Schloss et al., 2009), a lot of tools out there will build alignments for you with ribosomal RNA and help you build trees of everything. Usually these work best when you have all the sequences overlap with each other. So the challenge again is, what do you do in cases where the sequences don’t overlap with each other completely, as you would get with metagenomic data.

Again, I did this by hand but we have developed methods that can allow you to do this for protein coating sequences and compare them all to each other. So in the Sargasso data, in red were sequences from the Sargasso Sea and in black were sequences that were from genomes. So you can see how those new sequences relate to each other, in addition to how they relate to the reference data. (See Figures A6-2 and A6-3.)

Method 3: All in the Family

Method 2 is limited in that it involves constraining yourself to this core region of the sequence alignment. But there are methods available that people have used, primarily in analysis of morphological data or of express sequence tag (EST) data, where you can build an evolutionary tree of sequences that don’t overlap with each other at all (Figure A6-12).

An illustration of a hypothetical multiple sequence alignment showing an alternative strategy for phylogenetic analysis of metagenomic data


Hypothetical multiple sequence alignment showing an alternative strategy for phylogenetic analysis of metagenomic data—to analyze everything even if some sequences do not overlap with each other. Xs represent areas where a sequence lines up with (more...)

So if you have good reference data, and you have a sequence that matches the left hand of the reference data and a sequence that matches the right hand of the reference data, that is sort of like if you went to an archaeological or paleontological dig and you had a femur bone from one organism and maybe some teeth from another. And you can figure out in essence whether or not they might have come from the same organism by comparing them to references. You can do the same thing with sequences. And the latest in the phylogenetic analysis of metagenomic data has been to try and build methods that will build evolutionary trees even when sequences don’t overlap with each other at all, by using the reference sequences as your anchor.

We have developed a few tools in my lab, in collaboration with Katie Pollard and Jessica Green that take this “all in the family” approach. One is called PhylOTU (Sharpton et al., 2011), which identifies operational taxonomic units (OTUs) using this approach (Figure A6-13).

A flow chart showing the generalized workflow of PhylOTU


Computational processes are represented as squares and databases are represented as cylinders in this generalized workflow of PhylOTU. SOURCE: Sharpton et al. (2011).

We have another one called PhyloSift, which is like the new version of AMPHORA, that will do this for protein coating genes (see

There is a great method called pplacer (Matsen et al., 2010) that we have integrated in PhyloSift from Erick Matsen. That has been developed to do this exact type of thing. Again, you can build trees for sequences even if they don’t overlap with each other.

Method 4: All in the Genome

So the final frontier in this is to try and build trees, even with different genes, when they do not overlap with each other. We did a little test case of this in collaboration with Stephen Kembel and Jessica Green in Oregon, where we basically took all of the genes that we had been analyzing in that AMPHORA package, found homologues of those, and now built a reference tree of a concatenation of all of those sequences (Kembel et al., 2011). For each sequence that matched any of those individual sequences, different protein families in the environmental data, we can build an evolutionary tree that fits them relative to this anchor of the concatenated alignment of all sequences (Figures A6-14 and A6-15).

A flow chart presenting a conceptual overview of approach to infer phylogenetic relationships among sequences from metagenomic data sets


Conceptual overview of approach to infer phylogenetic relationships among sequences from metagenomic data sets. SOURCE: Kembel et al. (2011).

An illustration of a phylogenetic tree linking metagenomic sequences from 31 gene families along an oceanic depth gradient


Phylogenetic tree linking metagenomic sequences from 31 gene families along an oceanic depth gradient at the HOT ALOHA site (DeLong et al., 2006). The depth from which sequences were collected is indicated by bar color (green = photic zone (< (more...)

So in the long run, I think this is what we are going to want to do with environmental data, is for all genomes, build up a reference tree of all the gene families in those genomes, and then anchor environmental data to that reference tree. And you can figure out much more precisely where those sequences came from, even if they are not part of a traditional sort of evolutionary marker gene family.

Steve Kembel did this analysis, not to do phylotyping, but because he wanted an evolutionary tree to do what is called phylogenetic ecology. Many people are probably familiar with UniFrac analysis to compare the diversity of communities by their overlap in the amount of phylogenetic tree that they cover from the two communities, the unique fraction of the evolutionary tree (Figure A6-16).

Two graphs showing the taxonomic diversity and standardized phylogenetic diversity versus depth in environmental samples along an oceanic depth gradient


Taxonomic diversity and standardized phylogenetic diversity versus depth in environmental samples along an oceanic depth gradient at the HOT ALOHA site. Taxonomic diversity is calculated as OTU richness (number of OTUs) based on binning of SSU-rRNA gene (more...)

That is an approach that could generally be called phylogenetic ecology. And Steve Kembel was really interested in comparing phylogenetic diversity between communities with metagenomic data. And the reason he wanted to concatenate all of these different genes was, we didn’t have enough sequences from individual genes to have enough signal. But if you have 100 genes that you can analyze at once from across the genomes, you can build up enough signal to ask questions about beta diversity, et cetera, in ecological communities.

Method 5: Novel Lineages and Decluttering

So another thing that I am very interested in, and have been interested in for awhile, is to look for novel lineages in metagenomic data. We wanted to do this a long time ago, and we ran into some bioinformatic roadblocks. So what I really wanted to do was scan through metagenomic data to find whether or not there was evidence for a fourth branch in the tree, or something to that effect, things that were really phylogenetically novel compared to other sequences (Figure A6-17).

An illustration of the search for novel phylogenetic lineages


Searching for novel phylogenetic lineages. One key question we have been trying to answer in my lab is “Are there sequences out there that fall in between the current main groups on the tree of life?” Figure by J. A. Eisen.

BOX A6-1Questions During Talk

PARTICIPANT: (off microphone) question about long branches in Figure A6-15.

Answer: The question is what is the meaning of the especially long branches? In this case I think it is taxa that actually evolve rapidly. So in the reference data here we have some organisms like endosymbionts, mycoplasmas, et cetera, that every one of their genes evolves on a long branch. You can get artifacts in some of these cases where you have too small of a fragment, and the phylogenetic methods just get confused by that and give you a really long branch length. I don’t think that’s the case here. So I think most of the cases here are where taxa are known to evolve more rapidly. And the branch length is in essence a representation of evolutionary rate.

PARTICIPANT: (off microphone) question about meaning of colors in Figure A6-15.

Answer: The colors were different. Sorry, I didn’t want to go into the ecological detail here. What Steve analyzes was Ed DeLong’s Hawaii Ocean Time Series Data, re-analyzed that, and the colors correspond to different samples from Ed’s data. What he was asking basically was primarily whether or not phylogenetic approaches to calculating beta diversity gave different answers than taxonomic approaches to calculating beta diversity, where you just count organisms as opposed to comparing the phylogenetic relatedness of organisms. So again, it is analogous to UniFrac, but now you can do it with metagenomic data, not just with ribosomal RNA data.

We had this problem, which was, if we did this for any type of data set, like RecA, at the time, we had something like 10,000 RecA sequences from bacteria, 200 from Archaea, and 200 from Eukaryotes. And at the time, building evolutionary trees of 10,000 sequences or more were challenging. So we wanted to sort of remove many of the bacterial sequences and only analyze a couple of them, as opposed to analyzing the 10,000 bacterial sequences. And what we did was develop a method that is an analog of something called Lek clustering, where you take sequences, you group them together, in essence into subfamilies, and you can identify sets of sequences that are related to each other really rapidly. And that is what we use to find, to flag different subgroups in these massive data sets of 10,000, 15,000, 100,000 sequences (Wu et al., 2011).

And when we do this for RecA or RNA polymerase or other protein family sequences, and scan through metagenomic data, you find lineages that don’t group into any known current lineages of organisms (Figures A6-18 and A6-19).

A phylogenetic tree of the RecA superfamily


Phylogenetic tree of the RecA superfamily. All RecA sequences were grouped into clusters using the Lek algorithm. Representatives of each cluster that contained >2 members were then selected and aligned using MUSCLE. A phylogenetic tree was built (more...)

A phylogenetic tree of the RpoB superfamily


Phylogenetic tree of the RpoB superfamily. All RpoB sequences were grouped into clusters using the Lek algorithm. Representatives of each cluster that contained >2 members were then selected and aligned using MUSCLE. A phylogenetic tree was built (more...)

These novel sequences easily could be coming from viruses that are out there in the environment, they could be new paralogs of RecA that are previously uncharacterized. Or they could be coming from phylogenetically very novel lineages that are out there in the environment. And the way to find phylogenetically novel lineages is to build evolutionary trees of all the sequences that you can get from environmental samples. I don’t know if Jill [Banfield] is going to talk about this. I know Jill has done this and found novel archaeal lineages, for example, in metagenomic data, within the archaea. What we were looking for here was basically is there anything that can show up between bacteria, archaea, and eukaryotes. My guess is these are not cellular organisms. They are just DNA sequences, and they are probably from viruses or something to that effect. But again, phylogenetic approaches are the way to scan through this type of data to find novel lineages.

I am not going to talk about this, but phylotyping is also very useful for binning metagenomic data, for trying to pull things together into one group that corresponds to a particular organism. We have done this previously with endosymbionts, for example (Wu et al., 2006). But you can use it with any type of data.

Example II: Function

So I want to shift gears. I think phylotyping is a really important area, and we are doing a lot of research on it, as are many others. And I want to shift gears just quickly to talk about a few other uses of phylogenetic analysis for studying microbial communities or microbes. And one relates to functional diversity and functional prediction, and I have spent a lot of time working on this over the years, and I have been very interested in basically when you have new sequence data, how do you make a prediction of the function of that gene.

We have talked about this a little bit at this meeting with the examples of the cytochrome, oxidases, and a few other examples here. How do you make a robust prediction for a sequence of a gene? And I think just like analyzing ribosome RNA sequences to understand an organism by building an evolutionary tree of those sequences, you should build an evolutionary tree of protein family sequences in order to understand the functional diversity in a family.

I developed an approach that I originally called phylogenomics (Figure A6-20), to do this many years ago. And [the approach] is basically: you take a sequence, you compare it to its homologues, you build an evolutionary tree of that sequence and its homologues, you overlay experimentally determined functions onto the tree. And then you use character state reconstruction methods to predict the functions of unknowns.

An outline of a phylogenomic methodology


Outline of a phylogenomic methodology. In this method, information about the evolutionary relationships among genes is used to predict the functions of uncharacterized genes (see text for details). Two hypothetical scenarios are presented and the path (more...)

Sound familiar, anybody? It’s phylotyping. I in essence co-opted this from the ribosome RNA world (Figure A6-21). But you can apply it to functions as opposed to organisms. And it is a very powerful tool in predicting functions for uncharacterized genes.

Phylogenomic functional prediction is based on concept of phylotyping with an overlay of an unrooted phylogenetic tree


Phylogenomic functional prediction is based on the concept of phylotyping. This figure is a merging of Figure 1 and Figure 20. By J. A. Eisen.

Again, placing them in their phylogenetic context is incredibly powerful. I think in the interest of time I won’t go into the multiple examples that I have of this. I would be happy to talk to people about this. This is routinely used now in many genome analysis projects to build evolutionary trees of various sequences. You could do it with whatever sequences you want from environmental data as well as from sequence genomes.

The latest thing in functional prediction, which I think is really interesting, is to use non-homology functional prediction methods, which look at things like distribution of patterns of genes across organisms (e.g., see our use of phylogenetic profiling [Wu et al., 2005]). You can also use distribution patterns of genes across environments to try and help you make predictions of functions of genes. This has been done in a variety of metagenomic projects. Exactly how you group genes and analyze the correlation between the distribution pattern of a gene and the taxa present in a sample, and the metadata of the sample, is still sort of a work in progress.

We have now been collaborating with Simon Levin and others as part of a project that Simon Levin was in charge of. In this “DARPA fundamental laws of biology” project we have been working to apply non-homology methods to metagenomic data (Jiang et al., in press).

Example III: Selecting Organisms to Study

The last thing I want to talk about is this issue of selecting organisms for study. A different use of phylogenetics is to try and understand what we have known about in the diversity of life. And so if you go through the ribosomal RNA tree of life there are many different lineages of bacteria (Figure A6-22). In 2000, when we first sort of noted this ourselves, most of the genomes came from three phyla. That is still basically true. There were some genome sequences available from other lineages, but most lineages were poorly sampled. The same trend is true for eukaryotes, true for archaea, and true for viruses (e.g., see Eisen [2000]).

A three panel illustration of phylogenetically biased genome sequencing


Phylogenetically biased genome sequencing. Phylogenetic tree is based on one from Hugenholtz 2002. SOURCE: Adapted from Hugenholtz (2002).

We had a project when I was at TIGR to sequence eight representatives of novel phyla for which genomes were not available that was funded by the National Science Foundation. More recently I have been coordinating a project called the genomic encyclopedia of bacteria and archaea at the Department of Energy (DOE) Joint Genome Institute where we have been really filling in the tree of bacteria and archaea, of cultured organisms with representative genome sequences.

This is one of these massive projects with hundreds of people involved, and it has been this amazing collaboration with the DOE JGI and the DSMZ culture collection. What we did is basically go through the tree of life, select the genomes to sequence by their phylogenetic novelty, and then ask questions about whether or not phylogeny ended up being a useful guide in selecting genome sequences. And we have shown now about five or six areas where phylogenetic sampling has improved our analysis of genome or metagenome data. So one is in functional predictions of genes, another is in discovery of genetic diversity. So a phylogenetically novel organism, if all else is equal, is more likely to have new protein families than a phylogenetically not novel organism (Figure A6-23).

A graph showing a comparison of proteins from four groupings (species, family, phylum, and domain), used to identify protein families


For each of four groupings (species, different strains of Streptococcus agalactiae; family, Enterobacteriaceae; phylum, Actinobacteria; domain, GEBA bacteria), all proteins from that group were compared to each other to identify protein families. Then (more...)

And we also showed that these phylogenetically novel organisms could help you analyze metagenomic data, by providing more reference data across the tree in essence. But when we did this, there was very little benefit to analyzing the metagenomic data, from the first 50 or even 100 genomes that we sequenced from this genomic encyclopedia project (Wu et al., 2009). And the reason for this is we need to adapt many of the methods that we are doing to improve our ability to make use of this phylogenetically diverse data. So we need to design new phylogenetic methods, new metagenomic methods, to take into account this environmental data.

I have been involved in this great collaboration with Jessica Green and Katie Pollard and Martin Wu to try and develop methods to take advantage of this. It just ended. We called it ISEEM (see

This is one of the products of iSEEM, which is a tree of 2,500 genomes, that Jenna Morgan and Aaron Darling in my lab generated. We have been building new protein family markers from all of these genomes, so we can improve that AMPHORA pipeline by having hundreds to thousands of new phylogenetic markers to use to scan through metagenomic data.

And the last thing I want to leave you with is the reason why I think this project did not really help analyze metagenomic data, which is we haven’t even begun to scratch the surface of microbial diversity in terms of reference genome data. So if you go through the tree of life, and you count the branch length in the tree, it is something called PD, or phylogenetic diversity (Figure A6-24). If you sum up the total length of the branches, for all the genomes sequenced before our project, it came to 25 units. Each genome that we added, added a lot of new units of phylogenetic diversity. It better have, because that is how we picked them. If you go through cultured organisms that are known and described—so the 8,000 or so described bacteria in archaea—we would need about 1,000 genomes to capture half of that diversity. This will be done in the next year or two. However, if you go through all the environmental data, we would need about 10,000 genomes to capture half of the diversity known 5 years ago in full-length ribosomal RNA sequences. So the vast majority of genomic diversity out there is uncaptured in most studies of cultured organisms.

A graph showing measurements of phylogenetic diversity for four subsets of the SSU rRNA phylogenetic tree


Using a phylogenetic tree of unique SSU rRNA gene sequences, phylogenetic diversity was measured for four subsets of this tree: organisms with sequenced genomes pre-GEBA (blue), the GEBA organisms (red), all cultured organisms (dark grey), and all available (more...)

I think the solution to this is to do genomes of uncultured organisms, either via single cell capture or, as I think [meeting participants] will hear from Jill [Banfield] and other people, metagenomic sequencing and assembly of those metagenomes can generate genome data from uncultured organisms. [Note added after talk: see Narasingarao et al. (2012) for an example of this.] And by doing that, we will really fill in the tree. And that will enable all sorts of different uses of phylogeny in analysis of environmental data.

And I will leave it at that and say that of course we need experiments from across the tree, too. Sequencing is great. I love sequencing, but it doesn’t tell us everything. And what we need is an organized effort like this genomic encyclopedia to target functional diversity from across the tree of life, too. And I will leave it there.

[Update. I realize in retrospect that my “conclusion” for my talk was pretty minimal so I am adding a few sentences below to wrap up this paper.]


All biological entities have a history. Making sense out of biological data is best done in the context of that history. What I have tried to show in this paper are examples of how phylogenetic approaches can be useful in studies of microbial diversity. I gave three main kinds of examples—phylotyping, functional prediction, and identifying gaps in our genomic reference data. There are hundreds more, some developed by myself, most developed by others. To best understand the present, and even predict the future, we need to understand the past and how things changed over time.


  • Altschul SF, Gish W, Miller W, Myers EW, Lipman DJ. Basic local alignment search tool. Journal of Molecular Biology. 1990;215:403–410. [PubMed: 2231712]
  • Caporaso JG, Kuczynski J, Stombaugh J, Bittinger K, Bushman FD, Costello EK, Fierer N, Peña AG, Goodrich JK, Gordon JI, Huttley GA, Kelley ST, Knights D, Koenig JE, Ley RE, Lozupone CA, McDonald D, Muegge BD, Pirrung M, Reeder J, Sevinsky JR, Turnbaugh PJ, Walters WA, Widmann J, Yatsunenko T, Zaneveld J, Knight R. QIIME allows analysis of high-throughput community sequencing data. Nature Methods. 2010;7(5):335–336. [PMC free article: PMC3156573] [PubMed: 20383131]
  • DeLong EF, Preston CM, Mincer T, Rich V, Hallam SJ, Frigaard N-U, Martinez A, Sullivan MB, Edwards R, Rodriguez Brito B, Chisholm SW, Karl DM. Community genomics among stratified microbial assemblages in the ocean’s interior. Science. 2006;311:496–503. [PubMed: 16439655]
  • Eisen JA. Phylogenomics: Improving functional predictions for uncharacterized genes by evolutionary analysis. Genome Research. 1998;8(3):163–167. [PubMed: 9521918]
  • Eisen JA. Assessing evolutionary relationships among microbes from whole genome analysis. Current Opinion in Microbiology. 2000;3:475–480. [PubMed: 11050445]
  • Eisen JA, Smith SW, Cavanaugh CM. Phylogenetic relationships of chemoautotrophic bacterial symbionts of Solemya velum say (Mollusca: Bivalvia) determined by 16S rRNA gene sequence analysis. Journal of Bacteriology. 1992;174(10):3416–3421. [PMC free article: PMC206016] [PubMed: 1577710]
  • Hugenholtz P. Exploring prokaryotic diversity in the genomic era. Genome Biology. 2002;3(2):REVIEWS0003. [PMC free article: PMC139013] [PubMed: 11864374]
  • Hugenholtz P, Pitulle C, Hershberger KL, Pace NR. Novel division level bacterial diversity in a Yellowstone hot spring. Journal of Bacteriology. 1998;180(2):366–376. [PMC free article: PMC106892] [PubMed: 9440526]
  • Jiang X, Langille MGI, Neches RY, Elliot MA, Levin SA, Eisen JA, Weitz JS, Dushoff J. Functional biogeography of ocean microbes: Dimension reduction of metagenomic data identifies biological patterns across scales. PLoS One. In press.
  • Kembel SW, Eisen JA, Pollard KS, Green JL. The phylogenetic diversity of metagenomes. PLoS ONE. 2011;6(8):e23214. [PMC free article: PMC3166145] [PubMed: 21912589] [CrossRef]
  • Matsen FA, Kodner RB, Armbrust EV. pplacer: Linear time maximum-likelihood and Bayesian phylogenetic placement of sequences onto a fixed reference tree. BMC Bioinformatics. 2010;11:538. [PMC free article: PMC3098090] [PubMed: 21034504]
  • Narasingarao P, Podell S, Ugalde JA, Brochier-Armanet C, Emerson JB, Brocks JJ, Heidelberg KB, Banfield JF, Allen EE. De novo metagenomic assembly reveals abundant novel major lineage of Archaea in hypersaline microbial communities. ISME Journal. 2012;6(1):81–93. [PMC free article: PMC3246234] [PubMed: 21716304] [CrossRef]
  • Schloss PD, Westcott SL, Ryabin T, Hall JR, Hartmann M, Hollister EB, Lesniewski RA, Oakley BB, Parks DH, Robinson CJ, Sahl JW, Stres B, Thallinger GG, Van Horn DJ, Weber CF. Introducing mothur: Open-source, platform-independent, community-supported software for describing and comparing microbial communities. Applied Environmental Microbiology. 2009;75(23):7537–7541. [PMC free article: PMC2786419] [PubMed: 19801464]
  • Sharpton TJ, Riesenfeld SJ, Kembel SW, Ladau J, O’Dwyer JP, Green JL, Eisen JA, Pollard KS. PhylOTU: A high-throughput procedure quantifies microbial community diversity and resolves novel taxa from metagenomic data. PLoS Computation Biology. 2011;7(1):e1001061. [PMC free article: PMC3024254] [PubMed: 21283775]
  • Stamatakis A. RAxML-VI-HPC: Maximum likelihood-based phylogenetic analyses with thousands of taxa and mixed models. Bioinformatics. 2006;22:2688–2690. [PubMed: 16928733]
  • Venter JC, Remington K, Heidelberg JF, Halpern AL, Rusch D, Eisen JA, Wu D, Paulsen I, Nelson KE, Nelson W, Fouts DE, Levy S, Knap AH, Lomas MW, Nealson K, White O, Peterson J, Hoffman J, Parsons R, Baden-Tillson H, Pfannkoch C, Rogers YH, Smith HO. Environmental genome shotgun sequencing of the Sargasso Sea. Science. 2004;304(5667):66–74. [PubMed: 15001713]
  • Wu D, Daugherty SC, Van Aken SE, Pai GH, Watkins KL, Khouri H, Tallon LJ, Zaborsky JM, Dunbar HE, Tran PL, Moran NA, Eisen JA. Metabolic complementarity and genomics of the dual bacterial symbiosis of sharpshooters. PLoS Biology. 2006;4(6):e188. [PMC free article: PMC1472245] [PubMed: 16729848] [CrossRef]
  • Wu M, Eisen JA. A simple, fast, and accurate method of phylogenomic inference. Genome Biology. 2008;9:R151. [PMC free article: PMC2760878] [PubMed: 18851752] [CrossRef]
  • Wu D, Hartman A, Ward N, Eisen JA. An automated phylogenetic tree-based small subunit rRNA taxonomy and alignment pipeline (STAP). PLoS ONE. 2008;3(7):e2566. [PMC free article: PMC2432038] [PubMed: 18596968] [CrossRef]
  • Wu D, Wu M, Halpern A, Rusch DB, Yooseph S, Frazier M, Venter JC, Eisen JA. Stalking the fourth domain in metagenomic data: Searching for, discovering, and interpreting novel, deep branches in marker gene phylogenetic trees. PLoS ONE. 2011;6(3):e18011. [PMC free article: PMC3060911] [PubMed: 21437252] [CrossRef]
  • Wu D, Hugenholtz P, Mavromatis K, Pukall R, Dalin E, Ivanova NN, Kunin V, Goodwin L, Wu M, Tindall BJ, Hooper SD, Pati A, Lykidis A, Spring S, Anderson IJ, D’haeseleer P, Zemla A, Singer M, Lapidus A, Nolan M, Copeland A, Han C, Chen F, Cheng JF, Lucas S, Kerfeld C, Lang E, Gronow S, Chain P, Bruce D, Rubin EM, Kyrpides NC, Klenk HP, Eisen JA. A phylogeny-driven genomic encyclopaedia of Bacteria and Archaea. Nature. 2009;462(7276):1056–1060. [PMC free article: PMC3073058] [PubMed: 20033048]
  • Wu M, Ren Q, Durkin AS, Daugherty SC, Brinkac LM, Dodson RJ, Madupu R, Sullivan SA, Kolonay JF, Haft DH, Nelson WC, Tallon LJ, Jones KM, Ulrich LE, Gonzalez JM, Zhulin IB, Robb FT, Eisen JA. Life in hot carbon monoxide: The complete genome sequence of Carboxydothermus hydrogenoformans Z-2901. PLoS Genetics. 2005;1(5):e65. [PMC free article: PMC1287953] [PubMed: 16311624]



and .


27 Department of Microbiology, University of Washington.

Cell-Cell Communication and Group Activities

Although bacteria were long thought to be individuals that act alone, it is now accepted that they are social creatures that can act together to exhibit a range of extraordinary group activities (Costerton et al., 1999; West et al., 2006). Many of these activities are involved in virulence and for this reason have been studied in the context of pathogenesis, but microbial social behaviors are important in a variety of other contexts (Diggle et al., 2007a). One type of social trait that has been studied extensively at the molecular level is the ability of bacteria to communicate with one another by using chemical signals. Bacterial communication can coordinate a wide range of activities in different bacterial species (Whitehead et al., 2001). Bacterial communication often, but not always, coordinates transcription as a function of population density. This type of communication is called quorum sensing (Fuqua et al., 1994). At least in some cases it is clear that quorum sensing controls cooperative activities (Diggle et al., 2007a). The following text is an adaptation of Greenberg (2011).

If one strives to really understand bacteria, social aspects of their biology cannot be ignored. Although the field is very young, we see two reasons that it is critical to study quorum sensing and control of social activities in bacteria: (1) Now that we understand that bacteria are social and we understand sociality at an unprecedented level of molecular detail in several model organisms, bacteria have become wonderful tools to study fundamental questions about the costs and benefits of cooperation, the selective pressures that lead to cooperative behaviors, and the advantages of controlling cooperative behaviors by cell-to-cell communication systems like quorum sensing. (2) There is an idea that we might be able to develop novel antivirulence therapeutics that target quorum sensing in bacterial species that control virulence gene expression by cell-to-cell signaling. Although many investigators are working to identify potent quorum sensing and biofilm inhibitors, including our own group (Banin et al., 2008; Muh et al., 2006a,b), we add a cautionary note—we currently understand very little about how, when, or where quorum sensing inhibition might be of therapeutic value.

Quorum Sensing in Proteobacteria

In the late 1960s and early 1970s there was a very modest literature describing pheromone production and activity in bacteria (Eberhard, 1972; Tomasz, 1965). It was not until the 1980s that work on quorum sensing in marine luminescent bacteria led to the idea that these sorts of gene regulatory activities function as intercellular communication systems that coordinate group activities. Not until the 1990s did we begin to understand the prevalence of quorum sensing in bacteria. Now we know that there are many different types of bacterial cell-to-cell signaling systems considered as quorum-sensing systems. Our work focuses on acyl-homoserine lactone (acyl-HSL) quorum-sensing systems prevalent but not universal among the Proteobacteria. Acyl-HSLs are synthesized by LuxI-family enzymes and detected by LuxR-family signal receptors that are transcriptional regulators. There are related but different signals that are specific to each system (Figure A7-1). Our original model system was a marine bacterium, Vibrio fischeri, which controls a small set of about 25 or fewer genes, including genes for light production, by LuxR and 3-oxo-hexanoyl-homoserine lactone (3OC6-HSL), which is produced by the luxI gene product (Antunes et al., 2007; Engebrecht et al., 1983) (Figure A7-2A). V. fischeri exists in the light organs of specific marine animals where it produces light, which serves the mutualistic symbiosis. Of note, 3OC6-HSL moves in and out of cells by passive diffusion (Kaplan and Greenberg, 1985; Pearson et al., 1994). In this way the environmental signal concentration is a reflection of cell density. This quorum-sensing system allows V. fischeri to discriminate between its free-living low-population-density lifestyle and its high-density host-associated lifestyle. The luxI gene itself is controlled by quorum sensing—it is activated by 3OC6-HSL-bound LuxR. In terms of biology this provides hysteresis to the system. The population density required to activate quorum-controlled genes is much higher than the density required to shut down an activated system.

An illustration of acyl-HSL quorum-sensing signals


Some examples of acyl-HSL quorum-sensing signals. The structures and corresponding names are shown. The P. aeruginosa signal synthase RhlI produces C4-HSL and LasI produces 3OC12-HSL. The V. fischeri signal synthase LuxI produces 3OC6-HSL and the Rhodopseudomonas (more...)

A two-panel illustration of acyl-HSL signaling in V. fischeri and P. aeruginosa


Acyl-HSL signaling in V. fischeri (A) and P. aeruginosa (B). Acyl-HSL signals (see Figure A7-1) are made by members of the LuxI family of signal synthases and specifically interact with LuxR family transcription factors. At high cell density, acyl-HSLs (more...)

Later we turned our attention to the opportunistic pathogen P. aeruginosa. We learned that there were two acyl-HSL circuits, the C4-HSL-RhlI-RhlR circuit and the 3OC12-HSL-LasI-LasR circuits, and the 3OC12-HSL-responsive orphan receptor QscR, which together are required for activation of about 325 genes (Figure A7-2B) (Lee et al., 2006; Pearson et al., 1994, 1995; Schuster and Greenberg, 2006). Other investigators showed that at least under certain experimental conditions quorum-sensing mutants were impaired in virulence (Pearson et al., 2000; Tang et al., 1996), and thus the belief was that, as for V. fischeri quorum sensing, P. aeruginosa quorum sensing allowed discrimination between host and free-living states. However, it is not clear from the evidence that this is in fact the case. P. aeruginosa can use quorum sensing–controlled processes to compete with other bacteria (An et al., 2006), and this may provide an advantage in mixed microbial communities. In addition, some quorum-controlled genes are important for general metabolic processes (Schuster et al., 2003) suggesting quorum sensing may be important for regulating broad physiological changes that prepare the population for high-cell-density stress. Of note, among the 300-plus P. aeruginosa quorum-controlled genes, those coding for extracellular products like exoenzymes or production of extracellular products like phenazine pigments are grossly overrepresented in the quorum-controlled regulon (Schuster and Greenberg, 2006). In a social context these sorts of extracellular products can be considered public goods or resources produced by individuals and shared by all members of the group. Thus quorum sensing may be important to coordinate cooperative group behaviors.

Although quorum sensing promotes virulence in multiple acute infection models, many P. aeruginosa isolates from chronic human infections are lasR mutants (D’Argenio et al., 2007; Fothergill et al., 2007; Heurlier et al., 2006; Hoffman et al., 2009; Smith et al., 2006; Tingpej et al., 2007; Wilder et al., 2009). This suggests that lasR mutants have an advantage during the chronic disease state, and in support of this, lasR mutants do exhibit specific conditional growth advantages (D’Argenio et al., 2007; Heurlier et al., 2005). Alternatively, lasR mutants may arise because they can exploit quorum sensing–intact individuals with which they coexist. This would allow them the advantage of utilizing shared quorum sensing–controlled products without the metabolic burden of producing them. For this reason, these naturally occurring quorum-sensing mutants might be social cheaters (Diggle et al., 2007b; Rumbaugh et al., 2009; Sandoz et al., 2007). It is unclear if social cheaters occur in the environment; however, a small survey of P. aeruginosa environmental isolates showed that many retained their quorum-sensing systems (Cabrol et al., 2003). P. aeruginosa is one example where the study of the social aspects of quorum sensing is particularly important to understand the possible utility of novel anti–quorum sensing therapeutics (Kohler et al., 2010; Mellbye and Schuster, 2011).

Common Themes in Quorum Sensing

Although acyl-HSL systems are found in diverse bacterial species that occupy disparate ecological niches, many of the controlled factors fall into several general groups that are conserved. Some of the most common types are toxins (e.g., virulence factors and antimicrobials), exoenzymes (e.g., proteases), and biofilm components (e.g., extracellular polysaccharides [EPSs]) (for a review, see Majerczyk et al., 2012). The overlap of acyl-HSL-regulated targets across diverse species and environments suggests that these factors provide a general benefit to bacteria in a community structure.

Many quorum sensing–controlled processes may be most valuable when carried out in populations of a sufficient cell density. For example, biofilms consist of heterogeneous bacterial groups that organize on surfaces. Bacteria in biofilms secrete EPSs and other biofilm matrix components that surround and protect the group, but this may be a wasteful process for an isolated individual. Another idea that has been proposed is that, during interactions with a host or competitor, delayed production of immunogenic or toxic exoproducts may allow the population to avoid detection until the cell density is sufficient for the group to mount an effective attack. Thus quorum sensing is thought to either provide an efficient method to regulate metabolically expensive products, or allow the population to remain undetected in a hostile environment, or perhaps some combination of each of these. The cost savings of quorum-sensing regulation may also protect cooperating populations from invasion by noncooperating cheaters that do not produce the expensive products but can take advantage of their public availability. This idea was recently proposed as a mechanism for swarming populations to resist invasion by nonswarming cheaters (Xavier et al., 2011). It has also been proposed that acyl-HSLs could serve as a proxy for environmental diffusion potential and prevent production of secreted goods in conditions of high diffusion (Redfield, 2002). According to this hypothesis, quorum sensing also acts to efficiently regulate the production of expensive products but in this case it acts as an environmental sensor, and benefits individuals acting alone as well as individuals that are part of a group. This has been put forth as an alternative to quorum sensing, providing a nonsocial explanation of acyl-HSL signaling. In our view these two hypotheses are not exclusive of each other.

New Models to Study Quorum Sensing

Despite a large volume of research on the molecular mechanisms of quorum sensing we still do not have a clear understanding of the potential social benefits of quorum sensing. Acyl-HSL quorum-sensing systems are thought to have been established in an ancient species of Proteobacteria (Lerat and Moran, 2004) and are encoded in the genomes of many organisms with disparate lifestyles. This supports the view that quorum sensing has likely evolved to benefit bacteria in many different environments. Because many of the well-characterized quorum-sensing systems are in species that alternate between free-living and host-associated lifestyles, our current view is somewhat limited. To better understand how quorum sensing may be important in other contexts, we have recently turned our attention to a group of three closely related Burkholderia species that have adopted distinctly different lifestyles: Burkholderia pseudomallei, an opportunistic pathogen, B. thailandensis, a nonvirulent close relative of B. pseudomallei, and B. mallei, a host-restricted pathogen that evolved from B. pseudomallei. All three species have almost identical quorum-sensing systems (Figure A7-3); QS-1, QS-3, and two orphan receptors (R4 and R5) (Ulrich et al., 2004a,b). By studying quorum sensing in this group we hope to gain new perspective on the benefits of quorum sensing in different pathogenic and nonpathogenic environments.

An illustration of quorum-sensing circuits of B. thailandensis, B. pseudomallei, and B. mallei


The quorum-sensing circuits of B. thailandensis, B. pseudomallei, and B. mallei. The figure shows the genetic context of the homologous quorum-sensing circuits (QS-1, QS-2, and QS-3) in B. thailandensis (Bt), B. pseudomallei (Bp), and B. mallei (Bm). (more...)

B. thailandensis and B. pseudomallei, but not B. mallei, contain another quorum-sensing system, QS-2, and flanking DNA encoding the biosynthetic genes for a potent, broad-spectrum antibiotic called bactobolin (Carr et al., 2011; Duerkop et al., 2009; Seyedsayamdost et al., 2010) (Figure A7-3). The absence of this genetic content in B. mallei supports the idea that QS-2 and bactobolin are important for saprophytic growth outside of the host. Many other saprophytic Proteobacteria also use quorum sensing to control the production of antimicrobials, including species of Pseudomonas (Bainton et al., 1992; McGowan et al., 1995; Park et al., 2001; Pierson et al., 1994), suggesting that quorum sensing–controlled antibiotic production may be commonly used to compete in mixed-species environments. Our group and others have become interested in understanding why quorum sensing is often used to control antibiotics in saprophytes (Hibbing et al., 2010). To address this we developed a dual-species experimental competition model and a related in silico model (Chandler et al., 2012). The results suggested that quorum sensing may be important for competition and that the ability to sense and respond to acyl-HSLs from another species, which we have called eavesdropping, may provide an advantage in some situations. The in silico modeling also suggested that quorum sensing may serve to delay the cost of producing antibiotics until the population can produce a sufficient concentration to kill a competitor (Chandler et al., 2012). This idea is difficult to probe experimentally, but we can now do this with the dual-species model we have developed. This model may fall under the category of “synthetic ecology” and is one example of the new types of systems being developed to probe unanswered questions about the social benefits of quorum sensing.

There has been a recent effort to probe the social benefits of quorum sensing in single-species culture. Investigators have devised experiments where growth of P. aeruginosa requires quorum sensing. This can be accomplished by providing protein as the sole source of carbon and energy. Growth requires quorum sensing–induced production of the exoproteases. In this setting LasR quorum-sensing mutants will emerge and become a stable, significant minority of the overall population. These cheaters or free-loaders do not bear the cost of contributing to the public good (elastase in this case) but they benefit from use of the public goods (Sandoz et al., 2007). This system has been used to study the ability of lasR, rhlR, and qscR mutants to invade a quorum sensing–intact population (Wilder et al., 2011) and to show that secretion of elastase is more beneficial to populations at high density than to populations at low density (Darch et al., 2012). These types of studies provide support for the long-held view that quorum sensing can be used to control and coordinate cooperative activities that benefit individuals within groups of bacteria.

Concluding Comments

This article seeks to highlight the importance of studying cell-to-cell communication in the context of the emerging field of sociomicrobiology. The article emphasizes quorum sensing in Proteobacteria and provides examples where quorum sensing serves to allow individuals of a species to communicate. This rudimentary form of chemical communication serves to coordinate certain group behaviors, such as the production of antibiotics, exoenzymes such as proteases, and biofilm matrix components. New models are being developed to understand the connection between quorum sensing and cooperation. Because this is an emerging field we currently know very little about how communication is important for cooperation, and how this may influence the evolution of quorum-sensing systems in different pathogenic and nonpathogenic environments. These types of studies are critical to understand fundamental principles of cooperation in bacteria as well as to acquire information needed to develop novel therapeutics to treat microbial diseases.


  • An D, Danhorn T, Fuqua C, Parsek MR. Quorum sensing and motility mediate interactions between Pseudomonas aeruginosa and Agrobacterium tumefaciens in biofilm cocultures. Proceedings of the National Academy of Sciences of the United States of America. 2006;103:3828–3833. [PMC free article: PMC1533783] [PubMed: 16537456]
  • Antunes LC, Schaefer AL, Ferreira RB, Qin N, Stevens AM, Ruby EG, Greenberg EP. Transcriptome analysis of the Vibrio fischeri LuxR-LuxI regulon. Journal of Bacteriology. 2007;189:8387–8391. [PMC free article: PMC2168698] [PubMed: 17827287]
  • Bainton NJ, Stead P, Chhabra SR, Bycroft BW, Salmond GP, Stewart GW, Williams P. N-(3-oxohexanoyl)-L-homoserine lactone regulates carbapenem antibiotic production in Erwinia carotovora. Biochemical Journal. 1992;288(Pt 3):997–1004. [PMC free article: PMC1131986] [PubMed: 1335238]
  • Banin E, Lozinski A, Brady KM, Berenshtein E, Butterfield PW, Moshe M, Chevion M, Greenberg EP, Banin E. The potential of desferrioxamine-gallium as an anti-Pseudomonas therapeutic agent. Proceedings of the National Academy of Sciences of the United States of America. 2008;105:16761–16766. [PMC free article: PMC2575493] [PubMed: 18931304]
  • Cabrol S, Olliver A, Pier GB, Andremont A, Ruimy R. Transcription of quorum-sensing system genes in clinical and environmental isolates of Pseudomonas aeruginosa. Journal of Bacteriology. 2003;185:7222–7230. [PMC free article: PMC296264] [PubMed: 14645283]
  • Carr G, Seyedsayamdost MR, Chandler JR, Greenberg EP, Clardy J. Sources of diversity in bactobolin biosynthesis by Burkholderia thailandensis E264. Organic Letters. 2011;13:3048–3051. [PMC free article: PMC3111747] [PubMed: 21615115]
  • Chandler JR, Heilmann S, Mittler JE, Greenberg EP. Acyl-homoserine lactone-dependent eavesdropping promotes competition in a laboratory co-culture model. ISME Journal. 2012 [PMC free article: PMC3504959] [PubMed: 22763647]
  • Costerton JW, Stewart PS, Greenberg EP. Bacterial biofilms: A common cause of persistent infections. Science. 1999;284:1318–1322. [PubMed: 10334980]
  • Darch SE, West SA, Winzer K, Diggle SP. Density-dependent fitness benefits in quorum-sensing bacterial populations. Proceedings of the National Academy of Sciences of the United States of America. 2012;109:8259–8263. [PMC free article: PMC3361460] [PubMed: 22566647]
  • D’Argenio DA, Wu M, Hoffman LR, Kulasekara HD, Deziel E, Smith EE, Nguyen H, Ernst RK, Freeman Larson TJ, Spencer DH, Brittnacher M, Hayden HS, Selgrade S, Klausen M, Goodlett DR, Burns JL, Ramsey BW, Miller SI. Growth phenotypes of Pseudomonas aeruginosa lasR mutants adapted to the airways of cystic fibrosis patients. Molecular Microbiology. 2007;64:512–533. [PMC free article: PMC2742308] [PubMed: 17493132]
  • Diggle SP, Gardner A, West SA, Griffin AS. Evolutionary theory of bacterial quorum sensing: When is a signal not a signal? Philosophical Transactions of the Royal Society B: Biological Sciences. 2007;362:1241–1249. [PMC free article: PMC2435587] [PubMed: 17360270]
  • Diggle SP, Griffin AS, Campbell GS, West SA. Cooperation and conflict in quorum-sensing bacterial populations. Nature. 2007;450:411–414. [PubMed: 18004383]
  • Duerkop BA, Varga J, Chandler JR, Peterson SB, Herman JP, Churchill ME, Parsek MR, Nierman WC, Greenberg EP. Quorum-sensing control of antibiotic synthesis in Burkholderia thailandensis. Journal of Bacteriology. 2009;191:3909–3918. [PMC free article: PMC2698390] [PubMed: 19376863]
  • Eberhard A. Inhibition and activation of bacterial luciferase synthesis. Journal of Bacteriology. 1972;109:1101–1105. [PMC free article: PMC247330] [PubMed: 5011244]
  • Engebrecht J, Nealson K, Silverman M. Bacterial bioluminescence: Isolation and genetic analysis of functions from Vibrio fischeri. Cell. 1983;32:773–781. [PubMed: 6831560]
  • Fothergill JL, Panagea S, Hart CA, Walshaw MJ, Pitt TL, Winstanley C. Wide-spread pyocyanin over-production among isolates of a cystic fibrosis epidemic strain. BMC Microbiology. 2007;7:45. [PMC free article: PMC1890549] [PubMed: 17521417]
  • Fuqua WC, Winans SC, Greenberg EP. Quorum sensing in bacteria: The LuxR-LuxI family of cell density-responsive transcriptional regulators. Journal of Bacteriology. 1994;176:269–275. [PMC free article: PMC205046] [PubMed: 8288518]
  • Greenberg EP. The science and applications of synthetic and systems biology: Workshop summary. Institute of Medicine; Washington DC: The National Academies Press; 2011. The new science of sociomicrobiology and the realm of synthetic and systems ecology; pp. 213–221.
  • Heurlier K, Denervaud V, Haenni M, Guy L, Krishnapillai V, Haas D. Quorum-sensing-negative (lasR) mutants of Pseudomonas aeruginosa avoid cell lysis and death. Journal of Bacteriology. 2005;187:4875–4883. [PMC free article: PMC1169536] [PubMed: 15995202]
  • Heurlier K, Denervaud V, Haas D. Impact of quorum sensing on fitness of Pseudomonas aeruginosa. International Journal of Medical Microbiology. 2006;296:93–102. [PubMed: 16503417]
  • Hibbing ME, Fuqua C, Parsek MR, Peterson SB. Bacterial competition: Surviving and thriving in the microbial jungle. Nature Reviews Microbiology. 2010;8:15–25. [PMC free article: PMC2879262] [PubMed: 19946288]
  • Hoffman LR, Kulasekara HD, Emerson J, Houston LS, Burns JL, Ramsey BW, Miller SI. Pseudomonas aeruginosa lasR mutants are associated with cystic fibrosis lung disease progression. Journal of Cystic Fibrosis. 2009;8:66–70. [PMC free article: PMC2631641] [PubMed: 18974024]
  • Kaplan HB, Greenberg EP. Diffusion of autoinducer is involved in regulation of the Vibrio fischeri luminescence system. Journal of Bacteriology. 1985;163:1210–1214. [PMC free article: PMC219261] [PubMed: 3897188]
  • Kohler T, Perron GG, Buckling A, van Delden C. Quorum sensing inhibition selects for virulence and cooperation in Pseudomonas aeruginosa. PLoS Pathogens. 2010;6:e1000883. [PMC free article: PMC2865528] [PubMed: 20463812]
  • Lee JH, Lequette Y, Greenberg EP. Activity of purified QscR, a Pseudomonas aeruginosa orphan quorum-sensing transcription factor. Molecular Microbiology. 2006;59:602–609. [PubMed: 16390453]
  • Lerat E, Moran NA. The evolutionary history of quorum-sensing systems in bacteria. Molecular Biology and Evolution. 2004;21:903–913. [PubMed: 15014168]
  • Majerczyk CD, Greenberg EP, Chandler JR. Quorum sensing in Burkholderia. In: Vasil ML, Darwin AJ, editors. Regulation of bacterial virulence. Washington, DC: ASM Press; 2012.
  • McGowan S, Sebaihia M, Jones S, Yu B, Bainton N, Chan PF, Bycroft B, Stewart GS, Williams P, Salmond GP. Carbapenem antibiotic production in Erwinia carotovora is regulated by CarR, a homologue of the LuxR transcriptional activator. Microbiology. 1995;141(Pt 3):541–550. [PubMed: 7711893]
  • Mellbye B, Schuster M. The sociomicrobiology of antivirulence drug resistance: A proof of concept. mBio. 2011;2 [PMC free article: PMC3190357] [PubMed: 21990612]
  • Muh U, Hare BJ, Duerkop BA, Schuster M, Hanzelka BL, Heim R, Olson ER, Greenberg EP. A structurally unrelated mimic of a Pseudomonas aeruginosa acyl-homoserine lactone quorum-sensing signal. Proceedings of the National Academy of Sciences of the United States of America. 2006;103(45):16948–16952. [PMC free article: PMC1636559] [PubMed: 17075036]
  • Muh U, Schuster M, Heim R, Singh A, Olson ER, Greenberg EP. Novel Pseudomonas aeruginosa quorum-sensing inhibitors identified in an ultra-high-throughput screen. Antimicrobial Agents and Chemotherapy. 2006;50:3674–3679. [PMC free article: PMC1635174] [PubMed: 16966394]
  • Park JH, Hwang J, Kim JW, Lee SO, Conway BA, Greenberg EP, Lee K. Characterization of quorum-sensing signaling molecules produced by Burkholderia cepacia G4. Journal of Microbiology and Biotechnology. 2001;11:804–811.
  • Pearson JP, Gray KM, Passador L, Tucker KD, Eberhard A, Iglewski BH, Greenberg EP. Structure of the autoinducer required for expression of Pseudomonas aeruginosa virulence genes. Proceedings of the National Academy of Sciences of the United States of America. 1994;91:197–201. [PMC free article: PMC42913] [PubMed: 8278364]
  • Pearson JP, Passador L, Iglewski BH, Greenberg EP. A second N-acylhomoserine lactone signal produced by Pseudomonas aeruginosa. Proceedings of the National Academy of Sciences of the United States of America. 1995;92:1490–1494. [PMC free article: PMC42545] [PubMed: 7878006]
  • Pearson JP, Feldman M, Iglewski BH, Prince A. Pseudomonas aeruginosa cell-to-cell signaling is required for virulence in a model of acute pulmonary infection. Infection and Immunity. 2000;68:4331–4334. [PMC free article: PMC101761] [PubMed: 10858254]
  • Pierson LS 3rd, Keppenne VD, Wood DW. Phenazine antibiotic biosynthesis in Pseudomonas aureofaciens 30–84 is regulated by PhzR in response to cell density. Journal of Bacteriology. 1994;176:3966–3974. [PMC free article: PMC205594] [PubMed: 8021179]
  • Redfield RJ. Is quorum sensing a side effect of diffusion sensing? Trends in Microbiology. 2002;10:365–370. [PubMed: 12160634]
  • Rumbaugh KP, Diggle SP, Watters CM, Ross-Gillespie A, Griffin AS, West SA. Quorum sensing and the social evolution of bacterial virulence. Current Biology. 2009;19:341–345. [PubMed: 19230668]
  • Sandoz KM, Mitzimberg SM, Schuster M. Social cheating in Pseudomonas aeruginosa quorum sensing. Proceedings of the National Academy of Sciences of the United States of America. 2007;104:15876–15881. [PMC free article: PMC2000394] [PubMed: 17898171]
  • Schuster M, Greenberg EP. A network of networks: Quorum-sensing gene regulation in Pseudomonas aeruginosa. International Journal of Medical Microbiology. 2006;296:73–81. [PubMed: 16476569]
  • Schuster M, Lostroh CP, Ogi T, Greenberg EP. Identification, timing, and signal specificity of Pseudomonas aeruginosa quorum-controlled genes: A transcriptome analysis. Journal of Bacteriology. 2003;185:2066–2079. [PMC free article: PMC151497] [PubMed: 12644476]
  • Seyedsayamdost MR, Chandler JR, Blodgett JA, Lima PS, Duerkop BA, Oinuma K, Greenberg EP, Clardy J. Quorum-sensing-regulated bactobolin production by Burkholderia thailandensis E264. Organic Letters. 2010;12:716–719. [PMC free article: PMC2821070] [PubMed: 20095633]
  • Smith EE, Buckley DG, Wu Z, Saenphimmachak C, Hoffman LR, D’Argenio DA, Miller SI, Ramsey BW, Speert DP, Moskowitz SM, Burns JL, Kaul R, Olson MV. Genetic adaptation by Pseudomonas aeruginosa to the airways of cystic fibrosis patients. Proceedings of the National Academy of Sciences of the United States of America. 2006;103:8487–8492. [PMC free article: PMC1482519] [PubMed: 16687478]
  • Tang HB, DiMango E, Bryan R, Gambello M, Iglewski BH, Goldberg JB, Prince A. Contribution of specific Pseudomonas aeruginosa virulence factors to pathogenesis of pneumonia in a neonatal mouse model of infection. Infection and Immunitiy. 1996;64:37–43. [PMC free article: PMC173724] [PubMed: 8557368]
  • Tingpej P, Smith L, Rose B, Zhu H, Conibear T, Al Nassafi K, Manos J, Elkins M, Bye P, Wilcox M, Bell S, Wainwright C, Harbour C. Phenotypic characterization of clonal and nonclonal Pseudomonas aeruginosa strains isolated from lungs of adults with cystic fibrosis. Journal of Clinical Microbiology. 2007;45:1697–1704. [PMC free article: PMC1933084] [PubMed: 17392437]
  • Tomasz A. Control of the competent state in Pneumococcus by a hormone-like cell product: An example for a new type of regulatory mechanism in bacteria. Nature. 1965;208:155–159. [PubMed: 5884251]
  • Ulrich RL, Deshazer D, Brueggemann EE, Hines HB, Oyston PC, Jeddeloh JA. Role of quorum sensing in the pathogenicity of Burkholderia pseudomallei. Journal of Medical Microbiology. 2004;53:1053–1064. [PubMed: 15496380]
  • Ulrich RL, Hines HB, Parthasarathy N, Jeddeloh JA. Mutational analysis and biochemical characterization of the Burkholderia thailandensis DW503 quorum-sensing network. Journal of Bacteriology. 2004;186:4350–4360. [PMC free article: PMC421622] [PubMed: 15205437]
  • West SA, Griffin AS, Gardner A, Diggle SP. Social evolution theory for microorganisms. Nature Reviews Microbiology. 2006;4:597–607. [PubMed: 16845430]
  • Whitehead NA, Barnard AM, Slater H, Simpson NJ, Salmond GP. Quorum-sensing in Gram-negative bacteria. FEMS Microbiology Reviews. 2001;25:365–404. [PubMed: 11524130]
  • Wilder CN, Allada G, Schuster M. Instantaneous within-patient diversity of Pseudomonas aeruginosa quorum-sensing populations from cystic fibrosis lung infections. Infection and Immunitiy. 2009;77:5631–5639. [PMC free article: PMC2786440] [PubMed: 19805523]
  • Wilder CN, Diggle SP, Schuster M. Cooperation and cheating in Pseudomonas aeruginosa: The roles of the las, rhl and pqs quorum-sensing systems. ISME Journal. 2011;5:1332–1343. [PMC free article: PMC3146268] [PubMed: 21368905]
  • Xavier JB, Kim W, Foster KR. A molecular mechanism that stabilizes cooperative secretions in Pseudomonas aeruginosa. Molecular Microbiology. 2011;79:166–179. [PMC free article: PMC3038674] [PubMed: 21166901]



,29 ,30 ,29 and 29.


29 Department of Microbiology, University of Washington School of Medicine, Seattle, WA, USA.
30 Center for Models of Life, Niels Bohr Institute, University of Copenhagen, Copenhagen, Denmark.

Many Proteobacteria use acyl-homoserine lactone (AHL)-mediated quorum sensing to activate the production of antibiotics at high cell density. Extracellular factors like antibiotics can be considered public goods shared by individuals within a group. Quorum-sensing control of antibiotic production may be important for protecting a niche or competing for limited resources in mixed bacterial communities. To begin to investigate the role of quorum sensing in interspecies competition, we developed a dual-species co-culture model using the soil saprophytes Burkholderia thailandensis (Bt) and Chromobacterium violaceum (Cv). These bacteria require quorum sensing to activate the production of antimicrobial factors that inhibit growth of the other species. We demonstrate that quorum-sensing-dependent antimicrobials can provide a competitive advantage to either Bt or Cv by inhibiting growth of the other species in co-culture. Although the quorum-sensing signals differ for each species, we show that the promiscuous signal receptor encoded by Cv can sense signals produced by Bt, and that this ability to eavesdrop on Bt can provide Cv an advantage in certain situations. We use an in silico approach to investigate the effect of eavesdropping in competition, and show conditions where early activation of antibiotic production resulting from eavesdropping can promote competitiveness. Our work supports the idea that quorum sensing is important for interspecies competition and that promiscuous signal receptors allow eavesdropping on competitors in mixed microbial habitats.


Quorum sensing affords bacteria the ability to control the expression of specific genes in a cell density-dependent manner (Fuqua et al., 1994, 2001; Bassler, 2002; Waters and Bassler, 2005). Many species of Proteobacteria use small molecules, acylated homoserine lactones (AHLs), as quorumsensing signals. AHLs are produced by LuxI family synthases, and specifically interact with cytoplasmic LuxR family transcription factors to influence gene expression. AHL specificity is defined by the nature of the acyl side group. AHLs can diffuse through lipid bilayers and thus can move out of and into cells by diffusion. Because of the signal diffusibility, AHLs must reach a critical environmental concentration before they cause changes in gene expression. It is common that the AHL synthase gene is among the genes activated, creating a positive feedback loop that results in increased production of signal (Engebrecht et al., 1983; Seed et al., 1995; Latifi et al., 1996; Duerkop et al., 2009; Stauff and Bassler, 2011). Thus, AHL signaling can coordinate population-wide changes in a cell density-dependent manner.

Quorum-sensing-regulated genes are predominated by those required for the production of shared “public goods,” such as secreted or excreted factors. One commonly occurring example is antimicrobials. Quorum-controlled antimicrobials have been described in many saprophytic Proteobacteria including Erwinia carotovora (Bainton et al., 1992), Pseudomonas aeruginosa (Kownatzki et al., 1987; Bainton et al., 1992; Gallagher and Manoil, 2001; Ran et al., 2003; Schuster and Greenberg, 2006), Burkholderia thailandensis (Bt) (Duerkop et al., 2009) and Chromobacterium violaceum (Cv) (Latifi et al., 1995; McClean et al., 1997). Although some groups have proposed that antimicrobial activity of secondary metabolites is a side effect and the primary function of these compounds is as signals (Davies et al., 2006; Yim et al., 2007), the classic view is that they are used for competition with other strains or species in multi-species environments. This classic view suggests that quorum sensing may be important for interspecies competition. Quorum sensing is best understood in the context of virulence, and few studies have addressed its importance in competition (Mazzola et al., 1992; Moons et al., 2005, 2006; An et al., 2006). The advantage of using quorum sensing to control the production of antimicrobials is unknown, but it may allow a population to coordinate delivery of a sudden killing dose that deprives competitors of the ability to adapt during exposure to subinhibitory antimicrobial concentrations (Hibbing et al., 2010, D An and M Parsek, unpublished). Quorum sensing may also defer production of an antimicrobial to minimize the metabolic cost of production.

We are interested in the connection between quorum sensing and production of antibiotics, and specifically whether quorum-sensing-controlled antibiotics are important for interspecies competition. Thus, we developed a dual-bacterial species model with two soil saprophytes, Bt and Cv. Although it is not unlikely that these species coexist in nature, we selected this pair of bacteria because we have a base of knowledge about their quorumsensing systems, about quorum-sensing control of antibiotic synthesis and because these species exhibit similar laboratory growth characteristics. The Bt genome encodes three LuxR–LuxI pairs. The BtaI1–R1 pair produces and responds to octanoyl-HSL (C8-HSL). Little is known about the genes controlled by this system, but it facilitates clumping under some conditions (Chandler et al., 2009). BtaI3 is a 3-hydroxy-octanoyl-HSL synthase, but little is known about BtaI3–R3 (Chandler et al., 2009). Finally, BtaR2–I2 senses and produces 3-hydroxy-octanoyl-HSL and 3-hydroxy-decanoyl-HSL (Duerkop et al., 2009). The BtaR2–I2 system activates btaI2 and a set of genes responsible for the production of a family of hydrophilic antibiotics, the bactobolins, that have activity against a broad range of bacterial species (Duerkop et al., 2009; Seyedsayamdost et al., 2010; Carr et al., 2011) including Cv (see below). The most potent of these is bactobolin A (Carr et al., 2011).

Cv has a single AHL circuit, the CviR–CviI quorum-sensing system. This circuit activates genes required for the production of a purple pigment called violacein and related compounds that have broad-spectrum antimicrobial activity (McClean et al., 1997). We found that Bt is resistant to purified violacein, but shows sensitivity to other quorum-sensing-dependent factors produced by Cv. The CviI-produced AHL signal is hexanoyl-HSL (C6-HSL), and although CviR is a C6-HSL-responsive transcription factor, it is promiscuous and also responds to a number of different AHL signals (McClean et al., 1997; Swem et al., 2009). This promiscuity may allow Cv to eavesdrop on other AHL-producing species. There are now a number of examples of Proteobacteria with promiscuous LuxR homologs (Pierson et al., 1998; Riedel et al., 2001; Steidle et al., 2001; Venturi et al., 2004; Dulla and Lindow, 2009; Ahlgren et al., 2011; Hosni et al., 2011). It is not known if AHL receptor promiscuity provides any advantage over more signal-specific receptors.

We report here that quorum-sensing-dependent production of antimicrobials can provide a competitive advantage to either Bt or Cv by inhibiting growth of the other species in co-culture. We also present evidence that although Bt and Cv produce different AHLs, the promiscuous signal receptor of Cv can sense Bt signals, and that this ability to eavesdrop on Bt can provide a competitive advantage to Cv. We describe a mathematical model of our dual species system and use this model to show that eavesdropping can promote fitness during competition as long as the population can produce sufficient antibiotic to kill the competitor. Our results support the idea that quorum sensing is important for interspecies competition and that promiscuous signal receptors promote fitness in some situations by enabling eavesdropping on AHLs produced by competitors.

Materials and Methods

Bacterial Strains and Growth

Strains and plasmids are described in the Supplementary Text and Supplementary Table S1. All bacteria were grown in Luria–Bertani (LB) broth containing morpholinepropanesulfonic acid (50mM; pH 7). Bactobolin Awas generously supplied by Jon Clardy (Seyedsayamdost et al., 2010) and dissolved in filter-sterilized water. Synthetic C6-HSL and purified violacein were purchased from Sigma-Aldrich (St Louis, MO, USA) and dissolved in acidified ethyl acetate (0.1 ml l−1 glacial acetic acid) or in dimethylformamide, respectively. AHLs were prepared from the Bt bactobolin strain BD20 by extracting stationary-phase (OD600 8–10) culture fluid with two equal volumes of acidified ethyl acetate and drying to completion under a stream of nitrogen gas. The dried extracts were dissolved in volumes of media equivalent to the volumes from which they were extracted. The extracts did not affect growth of Bt or Cv. Extracts similarly prepared from cultures of an AHL, bactobolin double mutant had no effect on the outcome of co-culture experiments. Co-cultures and cultures for AHL preparation were grown at 30 °C. All other growth was at 30 °C for Cv and 37 °C for Bt. Pure cultures and co-cultures containing visibly aggregated cells of Cv were dispersed by homogenization or waterbath sonication before plating for viable counts. Gentamicin was used at 10 μgml−1 (Cv and Escherichia coli) or 100 μgml−1 (Bt) and trimethoprim was used at 100 μgml−1. For selection of Bt and Cv transconjugants, gentamicin was at 10 μgml−1 and trimethoprim was at 100 μgml−1.

Antimicrobial Susceptibility Testing

We determined the minimum inhibitory concentration of bactobolin or violacein using a protocol modified from the 2003 guidelines of the Clinical and Laboratory Standards Institute (CLSI, formerly NCCLS). Inocula were prepared from logarithmicphase cultures and suspended to 5×106 cells in 1ml morpholinepropanesulfonic acid-buffered LB containing dilutions of antibiotic compounds. The minimum inhibitory concentration was defined as the lowest concentration (μgml−1) that prevented visible growth of bacteria after 24 h. To assess susceptibility to cell culture fluid, bacteria were similarly suspended in a broth with 10% (Bt) or 75% (Cv) (vol vol−1) filtered fluid from stationaryphase cultures grown for 24 or 16 h, respectively. Culture fluid was filtered through a 0.22-μm pore-size membrane and tested immediately. Fluid from cultures of Cv was diluted into 4 × concentrated LB to a 1× final LB concentration. Cv and Bt were treated for 24 and 10 h, respectively, before plating for viability. All antimicrobial susceptibility testing was at 30 °C with shaking.

Co-Culture Experiments

To inoculate co-cultures, pure cultures were grown to mid-logarithmic phase, subcultured to fresh medium at an optical density at 600nm (OD600) 0.05 and grown an additional 3 h before combining at the appropriate ratios in 10 ml (Figures A8-1 and A8-2) or 20 ml (Figures A8-3 and A8-4) of medium in 125-ml culture flasks. The initial OD600 of the co-culture was 0.05 (2–4×107 cells per ml) for Bt and 0.005 (2–4×106 cells per ml) for Cv. Co-cultures were incubated with shaking at 250 r.p.m. Colony-forming units (CFUs) of each species were determined by using differential antibiotic selection on LB agar plates. Bt was selected with gentamicin and Cv was selected with trimethoprim.

A graph showing B. thailandensisŒC. violaceum competition


B. thailandensisC. violaceum competition. Initial cell densities were 2–4×107B. thailandensis (Bt) cells per ml and 2–4×106C. violaceum (Cv) cells per ml. The initial and final cell densities of Bt and Cv were (more...)

A graph showing competition in co-cultures of wild-type Cv and Bt strains


Competition in co-cultures of wild-type C. violaceum (Cv) and wild-type or mutant B. thailandensis (Bt) strains. The dashed line indicates the starting 10:1 ratio of Bt to Cv. The ratio of Bt to Cv after 24 h was determined by selective plating and colony (more...)

A graph showing the ratio of Bt to Cv in a co-culture


Co-cultures of the C. violaceum (Cv) wild-type Cv017 or the AHL mutant Cv026 and the B. thailandensis (Bt) competition-impaired AHL, bactobolin double mutant JBT125. The dashed line shows the initial ratio of Bt to Cv. After 24 h, the ratio of Bt to (more...)

A two panel photograph of Cv quorum sensing activated by Bt


C. violaceum (Cv) quorum sensing is activated by B. thailandensis (Bt) AHLs. Quorum-sensing activation is indicated by the Cv quorum-sensing-dependent purple pigment, violacein, in stationary-phase cultures. (a) Cv wild-type (Cv017) and the AHL mutant (more...)


Antibiotic Sensitivities

As a first step in developing our binary culture model, we needed to test the sensitivity of Cv to bactobolin and the sensitivity of Bt to violacein. Thus, we used purified antibiotics to determine the minimum inhibitory concentrations. The minimum inhibitory concentration of bactobolin A for Cv was 8 μgml−1, and at concentrations exceeding 8 μgml−1, Cv was killed during treatment (data not shown). This bactobolin was estimated to be at 5.3 μgml−1 in pure Bt culture fluid in growth conditions similar to those we use (Seyedsayamdost et al., 2010). Bt produces at least seven other bactobolin compounds (Seyedsayamdost et al., 2010; Carr et al., 2011). To test if Bt-produced bactobolins in cell culture fluid are sufficient to kill Cv, we assessed Cv viability after treatment with filtered fluid from a stationary-phase (OD600 8–10) Bt culture. After treatment with 10% (vol vol−1) culture fluid from a wild-type Bt culture diluted into fresh broth, we were unable to recover viable Cv. After similar treatment with 10% (vol vol−1) culture fluid from a Bt bactobolin-defective mutant (btaK) or in broth alone, Cv grew to 2–3×109 CFU per ml (Table A8-1). Our results show that stationary-phase Bt cultures produce sufficient bactobolins to kill Cv.

TABLE A8-1. Sensitivity of C. violaceum (Cv) Strains to B. thailandensis (Bt) culture fluid.


Sensitivity of C. violaceum (Cv) Strains to B. thailandensis (Bt) culture fluid.

Bt was resistant to violacein at the highest concentration tested, 125 μgml−1 (data not shown), which is in excess of amounts produced by Cv (Tobie, 1935; Strong, 1944). Cv codes for other putative antimicrobial factors, including phenazines and hydrogen cyanide (Brazilian National Genome Project Consortium, 2003). To test whether Cv produces quorum-sensing-dependent antimicrobials with activity against Bt, we incubated Bt with filtered fluid from Cv wild-type or mutant stationary-phase cultures (OD600 4–5). After 10 h, Bt grew modestly to 3×108 in the presence of wild-type Cv culture fluid, but grew to 2×109 in the presence of fluid from the AHL synthesis mutant (Table A8-2). This indicates that Cv quorum sensing regulates production of extracellular factors that inhibit growth of Bt, and that this inhibition is not due to violacein alone.

TABLE A8-2. Sensitivity of B. thailandensis (Bt) Strains to C. violaceum (Cv) Culture Fluid.


Sensitivity of B. thailandensis (Bt) Strains to C. violaceum (Cv) Culture Fluid.

The BtCv Co-Culture Model

In pure culture, the doubling times of all Bt strains were 60 min±5% and Cv strains were 48 min±5% (see Supplementary Table S2), and both species reached densities of about 3×109 cells per ml in early stationary phase. Because of the modest growth-rate discrepancy, we used an inoculum of 2–4×107Bt per ml and 2–4×106Cv per ml in our co-culture experiments. Wild-type Bt outcompetes wild-type Cv, increasing in relative abundance by about 100-fold in 24 h (Figure A8-1). To study the competition further, we enumerated bacteria during logarithmic, early stationary and late stationary growth phases. In logarithmic and early stationary phase, both species reached densities in co-culture that were identical to the densities in pure culture (2–5×109 cells per ml). However, the final densities of both species in late stationary phase (24 h) was lower in co-culture than in pure culture (Table A8-3). The final cell density of Cv decreased over three logs from 5×109 cells per ml in early stationary phase to 1×106 cells per ml at 24 h. There was no significant decrease in Cv density in pure culture (Table A8-3). The final density of Bt was 10-fold lower in co-culture than in pure culture (Table A8-2). Our results are consistent with the hypothesis that both species produce quorum-sensing-controlled antimicrobials during stationary phase that inhibit growth of or kill the other species.

TABLE A8-3. Final Yields of B. thailandensis (Bt) and C. Violaceum (Cv) in a Pure Culture and Co-culture.


Final Yields of B. thailandensis (Bt) and C. Violaceum (Cv) in a Pure Culture and Co-culture.

Quorum-Sensing-Controlled Bactobolin Synthesis Promotes Bt Competitiveness in Binary Culture

To test the hypothesis that quorum sensing promotes Bt competitiveness in co-culture, we assessed competition with a Bt AHL mutant and wild-type Cv. We also assessed the competitiveness of a Bt bactobolin mutant. In co-culture conditions where wild-type Bt had a robust competitive advantage, either the Bt AHL or bactobolin mutant were outcompeted by Cv (Figure A8-2). We could rescue competitiveness of the AHL mutant by supplementing our co-cultures with Bt AHLs that were obtained by ethyl acetate extraction of culture fluid from a stationary-phase (OD600 8–10) Bt bactobolin mutant (Materials and methods). These results demonstrate that quorum sensing and quorum sensing-dependent bactobolin production are critical for the competitive success of Bt in our co-culture model.

Bactobolin production is controlled by the BtaI2–R2 quorum-sensing system (Duerkop et al., 2009). Next, we assessed the importance of BtaI2–R2 and each of the other two Bt quorum-sensing systems, BtaI1–R1 and BtaI3–R3, to the competitiveness of Bt in our co-culture model. For this, we used Bt strains harboring individual deletions in each of the AHL receptor genes btaR1, btaR2 or btaR3 (Figure A8-2). Not surprisingly, the btaR2 mutant competed poorly with Cv. Results were similar to those with the bactobolin mutant and the AHL synthesis mutant. The outcome with the btaR3 mutant was identical to wild type, indicating that BtaR3 is not important for competition in our model. The btaR1 mutant showed an intermediate ability to compete with Cv, suggesting that this regulator may be important for the production of bactobolin or production of other factors that enhance competition or bactobolin activity. In support of the former, we found that expression of a bactobolin btaK-lacZ transcriptional fusion is delayed in a btaR1 mutant (data not shown), suggesting that BtaR1 may advance the production of bactobolin. We also tested the competitiveness of strains with individual mutations in each of the AHL synthase genes. All three individual AHL synthase mutants outcompeted Cv with results similar to competitions with wild-type Bt (data not shown). These findings suggest that the AHL synthases have overlapping abilities to induce expression of bactobolin. This is supported by our previous finding that BtaR2 can respond to both 3-hydroxy-octanoyl-HSL and 3-hydroxy-decanoyl-HSL, which are produced by the BtaI3 and BtaI2 synthases, respectively (Duerkop et al., 2009).

Quorum Sensing Can Promote Competitiveness of Cv

Our results indicate that Cv also produces quorum-sensing-dependent antimicrobial factors that inhibit growth of Bt (Table A8-1). Thus, we hypothesized that quorum sensing promotes competitiveness of Cv as it does for Bt. To address this, we compared the competitiveness of the Cv wild-type and AHL mutant strains in co-culture with Bt. We modified our experiment to give wild-type Cv a competitive advantage by using a competition-defective Bt AHL, bactobolin double mutant, and we increased the co-culture volume to 20 ml because we observed that this further improves Cv competitiveness for reasons that are unknown (data not shown). In these conditions, wild-type Cv strongly outcompeted the Bt mutant, whereas the Cv AHL mutant barely outcompeted the Bt mutant (Figure A8-3). Competitiveness could be restored to the Cv AHL mutant by the addition of C6-HSL (the AHL produced by Cv) (Figure A8-3). These results show that quorum sensing can promote the competitiveness of Cv. Because violacein does not have any antimicrobial activity against Bt, we note that this is not due to violacein, but must be caused by as-yet undefined quorum-sensing-dependent factors.

Cv Can Sense and Respond to Bt AHLs

The Cv AHL receptor CviR can be activated by a range of AHLs including at least one of the AHLs produced by Bt, C8-HSL (McClean et al., 1997; Swem et al., 2009). We hypothesized that Bt AHLs can activate the Cv quorum-sensing receptor CviR and that this promotes competitiveness of Cv in co-culture with Bt. We first tested whether a pure culture of Cv can sense and respond to Bt AHLs; these AHLs were ethyl acetate extracted and concentrated from stationary-phase (OD600 8–10) culture fluid and added to Cv cultures to match concentrations in the culture from which they were extracted. As a read-out for quorum-sensing activation, we followed the purple pigment violacein. The Cv AHL mutant is not pigmented, but pigmentation can be restored by supplementing the culture medium with Bt AHL extracts (Figure A8-4a). This result shows that Cv can sense and respond to physiological levels of Bt AHLs.

Next, we tested whether the Cv AHL mutant can respond to Bt AHLs during co-culture growth. Because Cv is killed by Bt-produced bactobolin in co-culture (Table A8-1), we used the Bt bactobolin mutant BD20 for these experiments (Figure A8-4b). When in co-culture with a Bt AHL, bactobolin double mutant, the Cv AHL mutant did not turn purple. However, in co-culture with the AHL-producing Bt bactobolin mutant BD20, or with exogenously supplied Bt AHLs, the co-culture turned purple. This finding indicates that the Cv CviR responds to Bt AHLs. We conclude that Bt AHLs are cues that alter the behavior of Cv, although they did not evolve for that purpose (Keller and Surette, 2006). In our experiment, the Cv AHL synthase mutant can eavesdrop on Bt.

Eavesdropping Promotes Competitiveness of Cv

To determine whether eavesdropping can influence competitiveness of Cv, we enumerated Bt and Cv in co-cultures (Figure A8-5). As in our previous experiments, we grew the Cv AHL mutant with the Bt bactobolin mutant or an AHL, bactobolin double mutant. The Cv AHL mutant was more competitive with the Bt bactobolin mutant than it was with the double mutant. As a control, we added Bt AHLs to the co-culture with the Bt double mutant and observed that this improved the competitiveness of Cv. These results suggest that eavesdropping on Bt AHLs promotes Cv competitiveness. As an additional control, we tested whether the Cv AHL receptor CviR is required for eavesdropping. To address this, we constructed a Cv AHL synthase, receptor double mutant. We found that CviR is required for the competitive advantage provided to Cv by eavesdropping on Bt AHLs (Figure A8-5).

A graph showing how eavesdropping promotes competitiveness of Cv in co-cultures with a Bt bactobolin mutant


Eavesdropping promotes competitiveness of Cv in co-cultures with a B. Bt bactobolin mutant. After 24 h of co-culture, the ratio of Bt to Cv was determined. Co-cultures of the Cv AHL mutant (Cv026), or the Cv AHL synthase, receptor double mutant (Cv026R) (more...)

An in silico Eavesdropping Model

Our experimental approach has limitations and with the conditions we used, we could not observe an effect of eavesdropping with wild-type strains (data not shown). However, we suspect there may be conditions where eavesdropping provides an advantage to wild-type Cv. This may be as the population nears the critical density required for quorum-sensing activation. At this density, AHLs produced by a nearby competitor may cause early activation of quorum-sensing-dependent antibiotics and would improve competitiveness of the eavesdropping microbe.

To explore this hypothesis further, we developed a mathematical model of our binary culture system (see Supplementary Text and Supplementary Table S3). The model accounts for two wild-type species that produce antibiotics in response to AHL signals in a well-mixed environment, similar to species Bt and Cv in our experimental system. In silico, the antibiotic produced by each species has equal killing efficiency towards the competing species, but no influence on the producing species. The two species in our in silico model also have identical growth rates, rates of antibiotic and AHL production, and antibiotic-production costs. However, as we observed experimentally, in some conditions one species (which we refer to here as species C) can eavesdrop on the other (species B). In the in silico model, we assume that antibiotic production accelerates once the inducer reaches a critical threshold concentration. However, antibiotic-production rates eventually level off as AHL concentrations exceed the quorum-sensing threshold. We use several different activation thresholds in our analysis.

Our in silico model has a bistable dynamic where one species completely dominates under most conditions. In the absence of eavesdropping, the outcome favors the species that is numerically dominant at the beginning (Figure A8-6). When we vary the activation thresholds for antibiotic production (by varying KB and KC of B and C, respectively, see Supplementary Text and Supplementary Table S3), there is an optimal value (Koptimal) where one species can dominate the other; if we fix KB at this value, B can dominate C at any value of KC (other than when KC was equal to Koptimal), and the same is true for C if KC is set at Koptimal (see Supplementary Figure S1). For every set of parameter values we explored, we find that Koptimal is greater than zero. Thus, waiting until a population reaches a quorum provides a fitness benefit for antibiotic-producing bacteria.

A four panel illustration of in silico modeling


In silico modeling. Our model accounts for two species with quorum-sensing-controlled antibiotics, similar to our experimental model of Bt and Cv. As in our experimental model, our in silico model accounts for two species (B and C) that produce antibiotics (more...)

We then investigated eavesdropping in our in silico model when species B and C had identical thresholds above (high), equal to (optimal) and below (low) the optimal threshold. At a relatively high threshold, eavesdropping provided a distinct advantage to C by allowing it to invade B from lower starting frequencies (Figure A8-6a), supporting our initial hypothesis. However, with an optimal or low threshold, eavesdropping was disadvantageous (Figures A8-6b and A8-6c). We posit that in the latter two cases, the eavesdropping population activates production of antibiotic too early to accumulate a sufficient killing dose and antibiotic production is an ineffective metabolic burden. To test this hypothesis, we kept the same conditions as in Figure A8-6c and increased the toxicity of the antibiotic of both species. In these conditions, eavesdropping provides an advantage (Figure A8-6d), supporting our hypothesis. Furthermore, eavesdropping is also advantageous if the antibiotic cost is decreased (Supplementary Figures S2 and S3). However, these changes in toxicity and cost alter the optimal threshold (Supplementary Figure S1B and data not shown), effectively resetting the system so that antibiotic production is induced after the optimal threshold is achieved. Thus, eavesdropping-dependent early production of antibiotics promotes competition in a population that has already reached a sufficient density to produce a killing dose.


We have developed a dual-species competition model with two soil saprophytes, Bt and Cv, which both use quorum sensing to control production of antimicrobial factors. We show that both of these species can gain a competitive advantage over the other with success dependent on quorum sensing. The advantage of quorum-sensing control of antimicrobials has also been shown in other laboratory co-culture models (Moons et al., 2005, 2006; An et al., 2006). The previous reports, together with the results reported here, support the idea that quorum-sensing regulation is important in multi-species competition. Our results indicate that competitiveness of Bt relies on the btaI2-R2-controlled antibiotic bactobolin and Cv uses as-yet unidentified quorum-sensing-dependent factors for competition. The bactobolin biosynthetic genes and btaI2-R2 are encoded within a large (120-kb) DNA element that is absent from a close relative, the host-adapted pathogen Burkholderia mallei. That this element is retained in Bt supports the view that btaI2-R2 and bactobolin are important for competition during saprophytic growth.

Why do bacteria use quorum sensing to regulate antibiotic production? Our in silico model provides some possible clues. The results indicate that when antibiotic production is costly, early production slows population growth without effectively killing the competitor. Thus, quorum sensing defers the cost of antibiotic production until a sufficient killing dose can be delivered. We do not include in our model the additional possibility that sublethal concentrations of antibiotics may induce in the competitor an adaptation to higher concentrations of antibiotic. Both of these possibilities can be further explored with our experimental co-culture model. An alternative hypothesis is that deferred production may also protect the producing population against the emergence of non-producing cheaters. Cheaters can exploit public goods producers by utilizing the available goods without incurring the cost of their production. In a recent study by Xavier et al. (2011), delayed production of an exploitable public good, surfactant, protected the producing population against the emergence of cheaters. This strategy maximized growth of the producing population, thereby increasing its ability to compete with cheaters. Quorum-sensing regulation may similarly promote competitiveness with non-producing cheaters.

Our experimental model also showed that cross species AHL activation of the Cv broad-specificity AHL receptor can promote the competitiveness of Cv (Figure A8-5). In addition to Cv, there are several other species with broad-specificity AHL receptors and these are also saprophytes: E. carotovora (ExpR2) (Sjoblom et al., 2006); P. aeruginosa (QscR) (Lee et al., 2006); and receptors encoded by two species of Bradyrhizobium (BraR and BjaR) (Ahlgren et al., 2011; Lindemann et al., 2011). ExpR2 and QscR are both orphan receptors without a cognate AHL synthase gene (Cui et al., 2006; Fuqua, 2006; Sjoblom et al., 2006). The potential role of each of these receptors in competition has not been determined. AHL receptor specificity can be easily altered by single amino-acid changes (Collins et al., 2005; Hawkins et al., 2007; Chen et al., 2011; Lintz et al., 2011), suggesting that AHL recognition may be very adaptable in nature. In contrast to these broad-specificity AHL receptors, the receptor of the squid symbiont Vibrio fischeri is quite specific for its cognate AHL (Visick and Ruby, 1999). V. fischeri activates quorum-sensing-dependent functions when it is at high cell densities in its squid host; in this environment it rarely encounters other bacterial species (Visick and McFall-Ngai, 2000). Thus, AHL receptors may evolve broad signal specificity in specific environments where eavesdropping might be of use, although the role of these receptors in inter-species competition and eavesdropping requires further study.

In the conditions of our experimental model, eavesdropping did not provide an observable fitness advantage to wild-type strains during competition. However in another study, AHLs produced by epiphytic bacteria on plant leaves altered the quorum-sensing-regulated virulence phenotype of a wild-type Pseudomonas syringae strain (Dulla and Lindow, 2009), suggesting that wild-type strains can be responsive to AHLs from other species in natural environments. Our co-culture model may provide a limited view of the possible interactions between species in nature, for example, Dulla and Lindow 2009) identified several epiphytic species that produce 10-fold more AHL than their laboratory P. syringae strain. High-level signal producers may play a significant role in cross-species induction.

Our mathematical model allowed a simple assessment of the costs and benefits of eavesdropping between competing wild-type strains. For the model, we made the basic assumption that detection of exogenous AHLs can cause early quorum-sensing dependent activation of antibiotic genes. We have observed this experimentally in Bt with a transcriptional fusion to the bactobolin biosynthetic gene btaK (data not shown), but it is more difficult to address with Cv because we do not yet know what quorum-controlled genes are involved in competition, and during early logarithmic phase the activity of the antimicrobials is too low for our methods of detection. The in silico model indicates that eavesdropping can promote competition in certain conditions where production of antibiotic occurs relatively late during growth. However, eavesdropping can also be detrimental if the activation threshold is relatively low. We observed similar results in other variations of this model (data not shown). Our results suggest that receptors would evolve broad specificity only in particular circumstances where eavesdropping is beneficial. Our bias is that specificity is the more evolved trait and that highly specific receptors likely arose from receptors with less specificity.


This work was funded by a National Institute of Allergy and Infectious Disease (NIAID) award to the Northwest Regional Center of Excellence for Biodefense and Emerging Infectious Diseases (U54AI057141) to EPG. JRC was funded by the NIAID (NRSA 1 F32 AI073027-01A2), SH was funded by the Danish National Research Foundation and JEM was funded by the UW Royalty Research Grant. We thank Drs Ajai Dandekar, Snow Brook Peterson, Ben Kerr and Matt Parsek for helpful discussions regarding this manuscript.


  • Ahlgren NA, Harwood CS, Schaefer AL, Giraud E, Greenberg EP. Aryl-homoserine lactone quorum sensing in stem-nodulating photosynthetic bradyrhizobia. Proc Natl Acad Sci USA. 2011;108:7183–7188. [PMC free article: PMC3084126] [PubMed: 21471459]
  • An D, Danhorn T, Fuqua C, Parsek MR. Quorum sensing and motility mediate interactions between Pseudomonas aeruginosa and Agrobacterium tumefaciens in biofilm cocultures. Proc Natl Acad Sci USA. 2006;103:3828–3833. [PMC free article: PMC1533783] [PubMed: 16537456]
  • Bainton NJ, Bycroft BW, Chhabra SR, Stead P, Gledhill L, Hill PJ, et al. A general role for the lux autoinducer in bacterial cell signalling: control of antibiotic biosynthesis in Erwinia. Gene. 1992;116:87–91. [PubMed: 1628848]
  • Bassler BL. Small talk. Cell-to-cell communication in bacteria. Cell. 2002;109:421–424. [PubMed: 12086599]
  • Brazilian National Genome Project Consortium. The complete genome sequence of Chromobacterium violaceum reveals remarkable and exploitable bacterial adaptability. Proc Natl Acad Sci USA. 2003;100:11660–11665. [PMC free article: PMC208814] [PubMed: 14500782]
  • Carr G, Seyedsayamdost MR, Chandler JR, Greenberg EP, Clardy J. Sources of diversity in bactobolin biosynthesis by Burkholderia thailandensis E264. Org Lett. 2011;13:3048–3051. [PMC free article: PMC3111747] [PubMed: 21615115]
  • Chandler JR, Duerkop BA, Hinz A, West TE, Herman JP, Churchill ME, et al. Mutational analysis of Burkholderia thailandensis quorum sensing and self-aggregation. J Bacteriol. 2009;191:5901–5909. [PMC free article: PMC2747893] [PubMed: 19648250]
  • Chen G, Swem LR, Swem DL, Stauff DL, O’Loughlin CT, Jeffrey PD, et al. A strategy for antagonizing quorum sensing. Mol Cell. 2011;42:199–209. [PMC free article: PMC3092643] [PubMed: 21504831]
  • Collins CH, Arnold FH, Leadbetter JR. Directed evolution of Vibrio fischeri LuxR for increased sensitivity to a broad spectrum of acyl-homoserine lactones. Mol Microbiol. 2005;55:712–723. [PubMed: 15660998]
  • Cui Y, Chatterjee A, Hasegawa H, Chatterjee AK. Erwinia carotovora subspecies produce duplicate variants of ExpR, LuxR homologs that activate rsmA transcription but differ in their interactions with N-acylhomoserine lactone signals. J Bacteriol. 2006;188:4715–4726. [PMC free article: PMC1483022] [PubMed: 16788181]
  • Davies J, Spiegelman GB, Yim G. The world of subinhibitory antibiotic concentrations. Curr Opin Microbiol. 2006;9:445–453. [PubMed: 16942902]
  • Duerkop BA, Varga J, Chandler JR, Peterson SB, Herman JP, Churchill ME, et al. Quorum-sensing control of antibiotic synthesis in Burkholderia thailandensis. J Bacteriol. 2009;191:3909–3918. [PMC free article: PMC2698390] [PubMed: 19376863]
  • Dulla GF, Lindow SE. Acyl-homoserine lactonemediated cross talk among epiphytic bacteria modulates behavior of Pseudomonas syringae on leaves. ISME J. 2009;3:825–834. [PubMed: 19340082]
  • Engebrecht J, Nealson K, Silverman M. Bacterial bioluminescence: isolation and genetic analysis of functions from Vibrio fischeri. Cell. 1983;32:773–781. [PubMed: 6831560]
  • Fuqua C. The QscR quorum-sensing regulon of Pseudomonas aeruginosa: an orphan claims its identity. J Bacteriol. 2006;188:3169–3171. [PMC free article: PMC1447470] [PubMed: 16621807]
  • Fuqua C, Parsek MR, Greenberg EP. Regulation of gene expression by cell-to-cell communication: acyl-homoserine lactone quorum sensing. Annu Rev Genet. 2001;35:439–468. [PubMed: 11700290]
  • Fuqua WC, Winans SC, Greenberg EP. Quorum sensing in bacteria: the LuxR–LuxI family of cell density-responsive transcriptional regulators. J Bacteriol. 1994;176:269–275. [PMC free article: PMC205046] [PubMed: 8288518]
  • Gallagher LA, Manoil C. Pseudomonas aeruginosa PAO1 kills Caenorhabditis elegans by cyanide poisoning. J Bacteriol. 2001;183:6207–6214. [PMC free article: PMC100099] [PubMed: 11591663]
  • Hawkins AC, Arnold FH, Stuermer R, Hauer B, Leadbetter JR. Directed evolution of Vibrio fischeri LuxR for improved response to butanoyl-homoserine lactone. Appl Environ Microbiol. 2007;73:5775–5781. [PMC free article: PMC2074898] [PubMed: 17675429]
  • Hibbing ME, Fuqua C, Parsek MR, Peterson SB. Bacterial competition: surviving and thriving in the microbial jungle. Nat Rev Microbiol. 2010;8:15–25. [PMC free article: PMC2879262] [PubMed: 19946288]
  • Hosni T, Moretti C, Devescovi G, Suarez-Moreno ZR, Fatmi MB, Guarnaccia C, et al. Sharing of quorum-sensing signals and role of interspecies communities in a bacterial plant disease. ISME J. 2011;5:1857–1870. [PMC free article: PMC3223305] [PubMed: 21677694]
  • Keller L, Surette MG. Communication in bacteria: an ecological and evolutionary perspective. Nat Rev Microbiol. 2006;4:249–258. [PubMed: 16501584]
  • Kownatzki R, Tummler B, Doring G. Rhamnolipid of Pseudomonas aeruginosa in sputum of cystic fibrosis patients. Lancet. 1987;1:1026–1027. [PubMed: 2883357]
  • Latifi A, Foglino M, Tanaka K, Williams P, Lazdunski A. A hierarchical quorum-sensing cascade in Pseudomonas aeruginosa links the transcriptional activators LasR and RhIR (VsmR) to expression of the stationary-phase sigma factor RpoS. Mol Microbiol. 1996;21:1137–1146. [PubMed: 8898383]
  • Latifi A, Winson MK, Foglino M, Bycroft BW, Stewart GS, Lazdunski A, et al. Multiple homologues of LuxR and LuxI control expression of virulence determinants and secondary metabolites through quorum sensing in Pseudomonas aeruginosa PAO1. Mol Microbiol. 1995;17:333–343. [PubMed: 7494482]
  • Lee JH, Lequette Y, Greenberg EP. Activity of purified QscR, a Pseudomonas aeruginosa orphan quorum-sensing transcription factor. Mol Microbiol. 2006;59:602–609. [PubMed: 16390453]
  • Lindemann A, Pessi G, Schaefer AL, Mattmann ME, Christensen QH, Kessler A, et al. Isovalerylhomoserine lactone, an unusual branched-chain quorum-sensing signal from the soybean symbiont Bradyrhizobium japonicum. Proc Natl Acad Sci USA. 2011;108:16765–16770. [PMC free article: PMC3189028] [PubMed: 21949379]
  • Lintz MJ, Oinuma K, Wysoczynski CL, Greenberg EP, Churchill ME. Crystal structure of QscR, a Pseudomonas aeruginosa quorum sensing signal receptor. Proc Natl Acad Sci USA. 2011;108:15763–15768. [PMC free article: PMC3179110] [PubMed: 21911405]
  • Mazzola M, Cook RJ, Thomashow LS, Weller DM, Pierson LS III. Contribution of phenazine antibiotic biosynthesis to the ecological competence of fluorescent pseudomonads in soil habitats. Appl Environ Microbiol. 1992;58:2616–2624. [PMC free article: PMC195829] [PubMed: 1514808]
  • McClean KH, Winson MK, Fish L, Taylor A, Chhabra SR, Camara M, et al. Quorum sensing and Chromobacterium violaceum: exploitation of violacein production and inhibition for the detection of N-acylhomoserine lactones. Microbiology. 1997;143(Part 12):3703–11. [PubMed: 9421896]
  • Moons P, Van Houdt R, Aertsen A, Vanoirbeek K, Engelborghs Y, Michiels CW. Role of quorum sensing and antimicrobial component production by Serratia plymuthica in formation of biofilms, including mixed biofilms with Escherichia coli. Appl Environ Microbiol. 2006;72:7294–7300. [PMC free article: PMC1636202] [PubMed: 16997989]
  • Moons P, Van Houdt R, Aertsen A, Vanoirbeek K, Michiels CW. Quorum sensing dependent production of antimicrobial component influences establishment of E. coli in dual species biofilms with Serratia plymuthica. Commun Agric Appl Biol Sci. 2005;70:195–198. [PubMed: 16366307]
  • Pierson EA, Wood DW, Cannon JA, Blachere FM, Pierson LS 3rd. Interpopulation signaling via N-acyl-homoserine lactones among bacteria in the wheat rhizosphere. Mol Plant Microbe Interact. 1998;11:1078–1084.
  • Ran H, Hassett DJ, Lau GW. Human targets of Pseudomonas aeruginosa pyocyanin. Proc Natl Acad Sci USA. 2003;100:14315–14320. [PMC free article: PMC283589] [PubMed: 14605211]
  • Riedel K, Hentzer M, Geisenberger O, Huber B, Steidle A, Wu H, et al. N-acylhomoserine-lactonemediated communication between Pseudomonas aeruginosa and Burkholderia cepacia in mixed biofilms. Microbiology. 2001;147:3249–3262. [PubMed: 11739757]
  • Schuster M, Greenberg EP. A network of networks: quorum-sensing gene regulation in Pseudomonas aeruginosa. Int J Med Microbiol. 2006;296:273–81. [PubMed: 16476569]
  • Seed PC, Passador L, Iglewski BH. Activation of the Pseudomonas aeruginosa lasI gene by LasR and the Pseudomonas autoinducer PAI: an autoinduction regulatory hierarchy. J Bacteriol. 1995;177:654–659. [PMC free article: PMC176640] [PubMed: 7836299]
  • Seyedsayamdost MR, Chandler JR, Blodgett JA, Lima PS, Duerkop BA, Oinuma K, et al. Quorum-sensingregulated bactobolin production by Burkholderia thailandensis E264. Org Lett. 2010;12:716–719. [PMC free article: PMC2821070] [PubMed: 20095633]
  • Sjoblom S, Brader G, Koch G, Palva ET. Cooperation of two distinct ExpR regulators controls quorum sensing specificity and virulence in the plant pathogen Erwinia carotovora. Mol Microbiol. 2006;60:1474–1489. [PubMed: 16796682]
  • Stauff DL, Bassler BL. Quorum sensing in Chromobacterium violaceum: DNA recognition and gene regulation by the CviR receptor. J Bacteriol. 2011;193:3871–3878. [PMC free article: PMC3147534] [PubMed: 21622734]
  • Steidle A, Sigl K, Schuhegger R, Ihring A, Schmid M, Gantner S, et al. Visualization of N-acylhomoserine lactone-mediated cell–cell communication between bacteria colonizing the tomato rhizosphere. Appl Environ Microbiol. 2001;67:5761–5770. [PMC free article: PMC93370] [PubMed: 11722933]
  • Strong FM. Isolation of violacein. Science. 1944;100:287. [PubMed: 17753079]
  • Swem LR, Swem DL, O’Loughlin CT, Gatmaitan R, Zhao B, Ulrich SM, et al. A quorum-sensing antagonist targets both membrane-bound and cytoplasmic receptors and controls bacterial pathogenicity. Mol Cell. 2009;35:143–153. [PMC free article: PMC2741501] [PubMed: 19647512]
  • Tobie WC. The pigment of Bacillus violaceus: I. The production, extraction, and purification of violacein. J Bacteriol. 1935;29:223–227. [PMC free article: PMC543590] [PubMed: 16559782]
  • Venturi V, Friscina A, Bertani I, Devescovi G, Aguilar C. Quorum sensing in the Burkholderia cepacia complex. Res Microbiol. 2004;155:238–244. [PubMed: 15142620]
  • Visick KL, Ruby EG. The emergent properties of quorum sensing: consequences to bacteria of autoinducer signaling in their natural environment. In: Winans GMDaSC., editor. Cell–Cell Signaling in Bacteria. ASM Press; Washington, DC: 1999. pp. 333–352.
  • Visick KL, McFall-Ngai MJ. An exclusive contract: specificity in the Vibrio fischeri–Euprymna scolopes partnership. J Bacteriol. 2000;182:1779–1787. [PMC free article: PMC101858] [PubMed: 10714980]
  • Waters CM, Bassler BL. Quorum sensing: cell-to-cell communication in bacteria. Annu Rev Cell Dev Biol. 2005;21:319–346. [PubMed: 16212498]
  • Wu H, Huang Y, Dykes C, Liu D, Ma J, Perelson AS, et al. Modeling and estimation of replication fitness of human immunodeficiency virus type 1 in vitro experiments by using a growth competition assay. J Virol. 2006;80:2380–2389. [PMC free article: PMC1395363] [PubMed: 16474144]
  • Xavier JB, Kim W, Foster KR. A molecular mechanism that stabilizes cooperative secretions in Pseudomonas aeruginosa. Mol Microbiol. 2011;79:166–179. [PMC free article: PMC3038674] [PubMed: 21166901]
  • Yim G, Wang HH, Davies J. Antibiotics as signalling molecules. Philos Trans R Soc Lond Ser B. 2007;362:1195–1200. [PMC free article: PMC2435582] [PubMed: 17360275]



,32 ,32 ,32 ,33 and 32,34.


32 Department of Bacteriology, University of Wisconsin, Madison, Wisconsin, 53706.
33 Department of Entomology, University of Wisconsin, Madison, Wisconsin, 53706.
34 Department of Plant Pathology, University of Wisconsin, Madison, Wisconsin, 53706.


Microbial communities comprise an interwoven matrix of biological diversity modified by physical and chemical variation over space and time. Although these communities are the major drivers of biosphere processes, relatively little is known about their structure and function, and predictive modeling is limited by a dearth of comprehensive ecological principles that describe microbial community processes. Here we discuss working definitions of central ecological terms that have been used in various fashions in microbial ecology, provide a framework by focusing on different types of interactions within communities, review the status of the interface between evolutionary and ecological study, and highlight important similarities and differences between macro- and microbial ecology. We describe current approaches to study microbial ecology and progress toward predictive modeling.


Biology in the twentieth century was dominated by simplification and order. This was driven by the desire to improve experimental controls and resulted in a landscape of intellectual frameworks unified by reductionism. Study of systems was replaced by study of parts, organism with cells, cells with genes and proteins, and genes and proteins with their atoms. The scientific triumphs were many and the practical outcomes—vaccines, antibiotics, and high yielding crops—transformed human health and food security. But the cost was a reduction of emphasis, training, and vision in systems-level biology, and with that a reduced ability to address some of the most important current environmental and health challenges.

As the twentieth century drew to a close, we were confronted by new challenges that rekindled widespread interest and identified the need to understand systems-level biology. Certain human diseases emerged whose origins were understood from landscape-level events that did not fit neatly into a reductionist scheme. Similarly, global climate change, and its underlying human causation, was recognized as a reality, and any realistic solutions required study of interconnecting spheres of society and the biosphere. These events and others like them arrived just as powerful new methods in microbiology emerged to open the way for a renaissance of ecology in general and microbial community ecology in particular. Although the need for systems biology has always been apparent to ecologists, who can offer many examples of ecosystems in which studying a binary interaction led to an erroneous conclusion that was corrected only by introducing more complexity into the model, the change in perspective was a surprise to much of the microbiology community (Raffa, 2004).

Over the past century of microbiology, the emphasis on the study of microbes in pure culture has isolated microorganisms from their communities and focused on their behavior in the biologically simple environments of the petri dish and test tube. Although simple model systems have driven an explosion of knowledge in cellular processes and host-microbe interactions over the past two decades, the reality of natural communities demands that we direct attention to complex assemblages as well. Global microbial diversity is enormous, likely representing 107 species, of which only 0.01% to 0.1% are known (Curtis and Sloan, 2002, 2004; Dykhuizen, 1998; Gans et al., 2005; Pace, 1997). Microbial communities can be complex, with high species richness and unevenness, and their structures are continually influenced by changing biological, chemical, and physical factors. Most microorganisms do not submit readily to growth in the laboratory, leaving microbiologists to either concentrate on the subsets that do perform well under artificial conditions or grope for other methods to describe the species that compose natural communities. Therefore, the structure of most microbial communities has been difficult to illuminate.

Describing community structure is often a prelude to understanding community function, which has been similarly difficult to elucidate (Handelsman, 2004; Pace, 1995; Pace et al., 1985). One of the barriers confronting microbial ecologists is the lack of ecological principles that provide the foundation for predictive models. Broadly based, validated principles derived from systems that can be manipulated experimentally would allow for predictions regarding behavior of communities that are less tractable for study. Many of these principles can be borrowed from macroecology, although some need to be reformatted to fit the microbial lifestyle. In this review, we explore the internal processes of microbial communities in an effort to begin to define the principles that underpin ecological and evolutionary patterns of microbial communities. Defining these principles is necessary to enable predictive modeling of ecological dynamics of microbial communities. In addition to their importance to fundamental understanding of the biosphere, predictive models have numerous practical applications. For example, they can provide guidance to strategies for manipulating communities on plant or animal surfaces to suppress pathogens, maintaining community integrity when applying chemicals such as antibiotics or pesticides that could destabilize communities, or successfully introducing a beneficial microorganism such as a biocontrol agent in agriculture or a probiotic in veterinary and human medicine.

Ecological Properties of Microbial Communities

The properties of microbial communities can be divided into two categories: structural and functional. Structural properties describe how the community varies and what it looks like—the types and numbers of members across a range of environments. Functional properties define the community’s behavior—how the community processes substrates, interacts with forces in its environment, and responds to perturbations such as invasion.

Structural Properties

One of the simplest ways to characterize a community is to list its members (composition) and to tabulate the total number of members (richness). To answer such questions as How many different species are there in a given community? and What are they? seems easy and straightforward, but answering them is challenging when they are applied to microbial communities (Schloss and Handelsman, 2005). The challenges derive from both biological and statistical issues. Enumeration by culturing limits the description to those members that can be cultured, which constitute the minority (often less than 1%) in most communities. Molecular methods present culture-independent alternatives, which capture far more richness than does culturing (Curtis et al., 2006). Sequence analysis of the 16S rRNA gene is the dominant method of determining identity and phylogenetic relatedness of microorganisms (Curtis et al., 2006), although other genes, such as rpoB, may provide greater resolution in phylogenetic associations at the species and subspecies levels (Case et al., 2007). To avoid the inherent difficulty in sampling every species in every community, macroand microbial ecologists often use estimates of richness based on samples of the communities; however, these estimates can vary depending on the estimator chosen and the type of data analyzed (Begon et al., 1990; Scholoss and Handelsman, 2007; see Curtis et al., 2006 for a review of the difficulties of quantifying properties in microbial communities, the statistics used in the analyses, as well as recent accomplishments in the area). Estimates of richness based on gene sequence relatedness are often calculated using software such as EstimateS and DOTUR (Colwell, 2005; Schloss and Handelsman, 2005). DOTUR, for example, has been used for a variety of genes and environments to assign sequences to operational taxonomic units based on phylogenetic distances and to calculate richness estimates at different degrees of phylogenetic resolution (Brodie et al., 2007; Cox and Gilmore, 2007; Diaz et al., 2006; Francis et al., 2005; Katayama et al., 2007; Schloss and Handelsman, 2006a, Vasanthakumar et al., 2006).

Diversity indices take into account both species richness and evenness of distribution of species (i.e., abundance of individuals) (Begon et al., 1990). The diversity of a community is difficult to interpret on its own but can be valuable when used to compare communities. Community structure, like diversity, is an attribute that is most useful when analyzed for comparative purposes. Community structure incorporates both the composition of the community and the abundance of individual members. Diversity and structure measure different aspects of community characteristics, so they can vary independently of each other (Hartmann and Widmer, 2006). In soil communities, the presence of a plant community, the introduction of various transgenes into a tree community, and rhizomediation influence composition and structure, but not diversity (de Carcer et al., 2007; LeBlanc et al., 2007; Zul et al., 2007). However, in other situations structural changes influence diversity as well.

In addition to the structure of the entire community, understanding the structure of assemblages within a community can also be important. For example, Perez & Sommaruga (Perez and Sommuaruga, 2007) monitored the response of the Betaproteobacteria and Actinobacteria populations within a lake water community to solar radiation and dissolved organic matter from lake, algal, or soil sources. Functionally defined assemblages, such as guilds, are also of interest because of the activities of community members. In another recent study, the structure of a methanotroph assemblage within a rice field community was monitored using terminal restriction-length polymorphism (TRFLP) analysis, pmoA gene analysis, and stable isotope probing (Shrestha et al., 2008). The results indicated that the activity and structure of the methanotroph assemblage (composed of type I and II methanotrophs) fluctuated over time and with CH4 availability and that type I and type II methanotrophs occupied two different niches within the rice field ecosystem. Other guilds such as ammonium-oxidizers, methanogens, and iron-reducers have been analyzed for structural properties (Chu et al., 2007; Hansel et al., 2007; Smith et al., 2007). Similarity of community structure can be calculated using a number of computer programs such as SLIBSHUFF, SONS, TreeClimber, and Unifrac and techniques such as analysis of molecular variance (AMOVA) and homogeneity of molecular variance (HOMOVA) (Lozupone and Knight, 2005; Lozupone et al., 2007; Martin, 2002; Schloss, 2008; Schloss et al., 2006a, 2006b; Schloss et al., 2004). The differences between the hypotheses tested by each of these tools are discussed elsewhere (Schloss, 2008).


Community robustness is the ability of the community to maintain its functional and structural integrity in the face of potential perturbations (Begon et al., 1990). This is consistent with other uses of robustness in engineering and statistics that pertain to the heartiness of a system and its ability to function under various, often adverse, conditions (Geraci, 1991). Just like complex engineered systems, biological systems, such as cells, tissues, organs, and ecological webs, are composed of diverse and often multifunctional components (Kitano, 2002). Systems that maintain their function, characteristic behavior, or some other property despite internal and external perturbations and adapt to their environments are robust (Kitano, 2002, 2003; Stelling et al., 2004). Although robustness is a characteristic of all biological systems, it is a relative property that depends on the perturbation and the behavior monitored (Stelling et al., 2004). For example, cancer cells that establish in the human body are particularly robust against the host’s defenses, but perturbations against which they are weakly robust offer promising therapies (Kitano et al., 2003).

We use structural robustness, similar to ecosystem stability (although ecosystem stability does not always refer to structure), to describe the constancy in community structure over time (temporal stability), the ability to resist change following a perturbation (resistance), and the return to its native structure following a change to structure (resilience). Components of robustness are often studied individually. Temporal stability, though not always referred to as such, is studied far more often than the other components of robustness. For example, Kikuchi & Graf (Kikuchi and Graf, 2007) recently reported that populations in the microbial community of leech crops, comprising Aeromonas veronii bv. sobria and a Rikenella-like organism, fluctuated within 6 h to 14 days after blood feeding. The population size of both members initially increased following feeding, but the abundance of A. veronii decreased 4 days after feeding while the abundance of the Rikenella-like species remained constant over the timescale studied (Kikuchi and Graf, 2007). Temporal stability has also been examined in rice field soil, cabbage white butterfly midguts, and a number of aquatic communities (Alonso-Saez et al., 2007; Carrino-Kyker and Swanson, 2008; Chenier et al., 2006; Kan et al., 2007; Kent et al., 2006; Lepere et al., 2006; Moss et al., 2006; Murase et al., 2006; C. Robinson, P. Schloss, K. Raffa & J. Handelsman, manuscript submitted).

Functional robustness refers to the ability of a community to maintain a particular activity despite perturbation, which, unlike structural robustness, is not necessarily linked to composition, although links between structure and function have been established many times (Adamczyk et al., 2003; Chandler et al., 2006; Chu et al., 2007; Cottrell and Kirchman, 2000; Gentile et al., 2007; Shrestha et al., 2008; Smith et al., 2007). Saison et al. (2006) showed that soil communities exhibited functional and structural resilience to low, but not high, levels of winery compost. In another study, Yannarell et al. (2007) found that nitrogen fixation returned to normal levels in a Bahamian microbial mat following a Category Four hurricane, despite a shift in community structure. At the intersection of functional and structural robustness is the years-old diversitystability debate. Briefly, the debate questions whether increased community diversity increases or decreases community stability (see Ives and Carpenter, 2007, and McCann, 2000, for reviews of the diversity-stability debate). Macro- and, to a lesser extent, microbial ecology have long sought a common rule that governs the relationship between diversity of a given community and its ecosystem stability, with stability often measured by a specific activity or function. Several years ago, McGrady-Steed et al. (1997) found that as diversity increased, decomposition increased, but carbon dioxide uptake decreased in an aquatic microcosm that contained bacteria, protists, and metazoans. They also noted that resistance to invasion increased as the abundance of certain members increased.

More recently, Girvan et al. (2005) examined the temporal structural stability and the functional resilience of two soil communities of different diversities that had been perturbed with benzene or copper using denaturing gradient gel electrophoresis and monitoring of broadscale (mineralization of 14C-labeled wheat shoot) and narrow-scale (mineralization of 14C-labeled 2,4-dichlorophenol) functions for 9 weeks. Temporal shifts in structure were observed in all soils, although copper-treated and control soils were consistently more similar to each other than either was to the benzenetreated soil (Girvan et al., 2005). This indicates that in some systems the source of the perturbation may be more important than community diversity in structural temporal stability. Benzene perturbation reduced the ability of both communities to perform the narrow niche function; however, the more diverse community reacquired the function by week 9 of the experiment, thereby exhibiting functional resilience (Girvan et al., 2005). Copper treatment increased broadscale function but initially reduced narrow niche function for both communities before they both recovered (Girvan et al., 2005). These findings indicate that diversity and source of perturbation may be important for functional robustness. The results also suggest that soil exhibits functional resilience owing in part to functional redundancy, but that greater diversity in soil communities may also lead to greater resistance and functional stability. The positive association between community diversity and stability is consistent with what has been observed in some macroecological systems such as grassplots (Tilman et al., 1996). More study of microbial community robustness is needed to establish the rules of engagement, i.e., governing principles and predictive models.

Interactions Within Microbial Communities

A first step toward understanding the nature of ecological interactions in natural microbial communities is cataloging mechanisms by which microorganisms interact. Symbiotic interactions can be divided into three overlapping categories, which exist in a continuum from parasitic to mutualistic (Figure A9-1). Parasites (Greek, para, “near,” and sitos, “food”) are organisms that live on or in another organism and obtain all or part of their necessary nutrients at the expense of their host. Commensalism (Latin, com, “together,” and mensa, “table”) includes relationships in which one partner derives benefit from the other and the other partner neither is harmed nor benefits from the association. Mutualism (Latin, mutualis, “reciprocal”) is an association in which both organisms derive benefit from one another. In addition to specific symbiotic interactions, microorganisms can interact antagonistically with other microorganisms via competition for a common resource or via predation of one organism upon another. The mechanisms that dictate interactions among microorganisms are largely responsible for the properties of the community as a whole. Dissecting the binary and tertiary interactions among community members is one essential component to understanding the properties of the community.

An illustration of the continuum of interspecific interactions that occur in microbial communities


Continuum of interspecific interactions that occur in microbial communities. (Left) The beneficial relationships that occur between mutualists such as the phototrophic consortia (shown below) comprising central rod-shaped bacteria in the family Comamonadaceae (more...)

Exploitative Competition

As in interactions between macroorganisms, exploitative competition, or competition for nutrients and space, plays an important role in shaping microbial interactions. In eukaryote-associated microbial communities, such as in the human gut, competition for nutrients and space from resident microflora is thought to present one barrier to infection by pathogens, a phenomenon called the barrier effect (Guarner and Malagelada, 2003). In cabbage white butterfly larvae, changes in the resident microflora community structure resulting from treatment with antibiotics facilitates invasion by a nonresident, in support of the role of competition from residents in normally preventing invasion (C. Robinson, Y. Ramos, K. Raffa & J. Handelsman, unpublished observations). Probiotic bacteria, such as some Bifidobacterium and Lactobacillus species, are thought to exert positive effects on host health in part via competitive interactions with pathogenic bacteria for space and nutrients, and also through interference competition by production of toxic compounds (Rastall et al., 2005). In vitro, binding of some probiotic bacteria to cultured epithelial cells can prevent binding of pathogens, in support of the hypothesis that bacteria successfully compete for space (Gopal et al.; 2001, Lee and Puong, 2002). Competition for nutrients among bacterial species from functional groups having different nutritional requirements can be important in structuring microbial communities in nutritionally heterogeneous environments. For example, in microcosms containing picoplanktonic cyanobacteria and heterotrophic bacteria along crossed gradients of glucose and phosphate, the cyanobacteria positively responded to increased phosphate only when glucose was low, presumably because of increased competition from the heterotrophs when organic carbon was supplied (Drakare, 2002).

Exploitative and interference competition. The aspect of microbial interactions that has arguably received the most attention is the ability of some microorganisms to produce compounds that, at least in laboratory studies, directly antagonize other microorganisms. It has been assumed that organisms produce these compounds as a means of chemical warfare, providing a competitive edge to the producers by directly inhibiting growth or killing off potential competitors, a form of interference competition. One example in which evidence supports this proposition is in the rhizosphere, where antibiotic production by a number of bacteria contributes to their ability to protect plants from particular pathogens (Compant et al., 2005; Raaijmakers et al., 2002). The antibiotics these bacteria produce in vitro have been detected in the rhizosphere (Bonsall et al., 1997), and mutants deficient in antibiotic production often exhibit a reduced ability to protect the plant from the pathogen (Keel et al., 1992; Silo-Suh et al., 1994; Thomashow and Weller 1998) or a reduced fitness in the rhizosphere (Mazzola et al., 1992; Pierson and Pierson, 1996). Additionally, production of the peptide antibiotic trifolitoxin by some strains of Rhizobium etli strains contributes to their competitive ability in the rhizosphere, leading to increased occupation of root nodules by producing strains (Robleto et al., 1997; Robleto et al., 1998). In addition to the rhizosphere systems, evidence from a number of invertebrate-microbe interactions supports the role of antibiotics in antagonistic relationships within communities. For example, antibiotics produced by Actinomycetes associated with leafcutter ants protect the ants’ fungal gardens from parasitism by another fungus (Currie et al., 2006; Little et al., 2006). Similarly, larvae of some crustaceans, beewolves, and bark beetles rely on production of antifungal compounds by bacterial symbionts to avoid infection by fungal pathogens (Cardoza et al., 2006; Gil-Turnes and Fenical, 1992; Gil-Turnes et al., 1989; Kaltenpoth et al., 2005).

In many cases, however, antibiotic production in vitro has not been demonstrated to result in antagonism in situ, leading to speculation that antibiotics play roles other than as growth inhibitors (Davies et al., 2006). At subinhibitory concentrations, structurally diverse antibiotics affect transcription of many bacterial genes not necessarily associated with stress responses, suggesting that antibiotics may function as signaling molecules in the environment when produced at low concentrations (Davies et al., 2006).

Interference competition via signal disruption. Some mechanisms of interference competition between microorganisms are independent of antibiotic production, such as disruption of signaling cascades. Diverse bacteria degrade acyl-homoserine lactone signal molecules (Leadbetter and Greenberg, 2000; Lin et al., 2003; Uroz et al., 2003; Uroz et al., 2005; Uroz et al., 2007; Yang et al., 2005), and the rapid turnover rate of acyl-homoserines in nonsterile soil suggests this is a common bacterial trait at least in that environment (Wang and Leadbetter, 2005). Signaling by small peptides can also be disrupted; for example, siamycin, a secondary metabolite produced by a soil Streptomyces strain, inhibits signaling by the gelatinase biosynthesis-activating pheromone of Enterococcus faecalis (Nakayama et al., 2007). The ability to interfere with signaling would provide a competitive advantage if competitive determinants, such as antimicrobial toxin production, were regulated by signaling, as has been suggested (An et al., 2006). However, a clear link between the disruption of signaling and competitive advantage has yet to be established.


In macroecological systems, predation, or the consumption of one organism by another, is frequently a key stabilizer of community structure (Estes, 1996; Estes et al., 2001; Henke and Bryant, 1999). The top predator often regulates abundance of other species that in turn regulate other species, providing a cascade of effects that have a sweeping influence. Often, this effect on the community far exceeds the numerical representation of the predator and can transform entire landscapes, which defines the predator as a keystone species (Paine, 1969; Ripple and Beschtab, 2003).

Predation of bacteria by microbial eukaryotes and bacteriophages provides a key link between microbial and macroorganismal food webs (Clarholm, 2002; Pernthaler, 2005) and has global effects on bacterial community structure and composition in many environments. In both freshwater and marine habitats, predation is a leading cause of bacterial mortality (Pernthaler, 2005; Thingstad, 2000). In some aquatic environments, top-down control by predation also appears to regulate bacterial population sizes (Pernthaler, 2005). Additionally, predation has been suggested as an influence on bacterial species richness and evenness (Pernthaler, 2005; Zhang et al., 2007), selectively limiting population sizes of some readily culturable aquatic bacteria that rapidly proliferate when grazing pressure is experimentally reduced (Beardsley et al., 2003). In soil, predation by protozoa similarly limits bacterial population sizes and can influence bacterial community composition and structure (Clarholm, 2002; Murase et al., 2006). Difficulties associated with quantification of bacteriophage in soil have hampered efforts to ascertain the role of predation by phage in these ecosystems (Ashelford et al., 2003; Williamson et al., 2005). In engineered microbial communities, predation pressure can influence bacterial population structure by selecting for strains adapted to defend against predator attack (Kunin et al., 2008). Manipulation of microbial communities via phage that prey specifically on lineages of bacteria could be used as a tool to study community dynamics, given the strength of phage selective pressures in some communities (Breitbart et al., 2004).

Bacteria or archaea that prey on other bacteria or archaea appear to be relatively rare compared with their eukaryotic counterparts, but examples of each have been described. Bdellovibrio-like organisms are obligate bacterial predators, which now appear to be more diverse and widespread than previously recognized (Davidov et al., 2006a; Pinero et al., 2007). Most of these organisms penetrate the outer membrane and wall of their prey and replicate in the cytoplasm, thereby killing the host. However, some, including the Micavibrio sp. (Alphaproteobacteria), attach to the outside of prey cells and replicate epibiotically (Davidov et al., 2006b). The range of prey organisms targeted by different Bdellovibrio-like predators varies but typically includes only a limited number of species (Jurkevitch et al., 2000; Pineiro et al., 2004; Rogosky et al., 2006). A different predatory strategy is employed by a second group of bacterial predators, the myxobacteria (Deltaproteobacteria). Populations of myxobacteria exhibit cooperative, surface-associated motility and collectively subsume prey organisms they encounter by producing secreted and cell-associated degradative enzymes (Reichenbach, 1999). However, unlike Bdellovibrio-like organisms, myxobacteria can also obtain nutrients by degrading macromolecules instead of live prey; genome sequencing of Myxococcus xanthus suggests that prey bacteria serve as a source of branched-chain amino acids, which the predator does not have the capacity to synthesize (Goldman et al., 2006). In addition to many proteases and cell wall–degrading enzymes, myxobacteria also produce diverse secondary metabolites, which may also play a role in predation or may mediate competition with other species or other myxobacterial strains (Feigna and Velicer, 2005; Goldman et al., 2006). Finally, a number of gram-negative bacteria release membrane vesicles containing hydrolytic enzymes that can fuse with and lyse other bacteria (91, 117). Membrane-vesicle-mediated lysis may help to extract nutrients from target cells, although this has not been established empirically.


The only prokaryote thought to parasitize another prokaryote is Nanoarchaeum equitans, isolated from hydrothermal vents. N. equitans is small in physical stature and has a tiny, compact genome, less than 0.5 Mb, predicted to contain a 95% coding sequence (Huber et al., 2002; McCliment et al., 2006). This species is completely dependent on its host, the larger archaeon, Ignicoccus hospitalis (Huber er al., 2002; Paper et al., 2007). The genome of N. equitans lacks many key metabolic functions, including genes encoding glycolysis and trichloroacetic acid cycle enzymes, as well as most amino acid and lipid biosynthesis pathways, indicating that it must obtain many nutrients and metabolites from its host (Waters et al., 2003). The host species, however, can be found in a free living state, and association with N. equitans appears to take a toll on its fitness, thus suggesting that the smaller associate is a parasite (Huber er al., 2002; Paper et al., 2007).

Mutualistic and Commensal Interactions

At least as prevalent among microorganisms as the antagonistic interactions described above are interactions in which both partners benefit (mutualism) or in which one partner benefits with no apparent effect on the other (commensalism). Purely commensal relationships may not exist; perhaps we simply have not discovered the benefit to the second partner. More likely, those organisms are either beneficial or harmful to their hosts, depending on the community dynamics in the niche, but researchers have yet to delimit and quantify the costs and benefits exchanged between the host and symbiont. For example, many microbes in the humangut historically termed commensal are now recognized as critical factors in gut and immunity development, nutrient uptake, and homeostasis of the system (Hooper, 2004; Hooper et al., 2002, Rakoff-Nahoum et al., 2004). Furthermore, relationships may be context dependent, that is, an organism could be beneficial under certain conditions and commensal under others.

Obligate associations. In obligate mutualisms each partner depends on the other for survival and reproduction. One particularly elegant obligate microbial association is the phototrophic consortium detected in many freshwater habitats (Glaeser and Overmann, 2003a). In these assemblages, a central motile, nonphotosynthetic Betaproteobacterium from the family Comamonadaceae is surrounded by green sulfur bacteria in an organized structure (Frostl and Overmann, 1998; Kanzler et al., 2005). The epibiotic sulfur bacteria are thought to benefit from the motility provided by the central bacterium, enabling the consortia to chemotax toward sulfide (Glaeser and Overmann, 2003a). The central Comamonadaceae may benefit from carbon secreted by the sulfur bacterium during photoautotrophic growth (Glaeser and Overmann, 2003a, 2003b). Additionally, the partners appear to coordinate behaviors via as yet unidentified signal exchange. For example, the consortia chemotax toward sulfide and the organic compound 2-oxoglutarate only in light, which the motile central organism itself cannot detect (Glaeser and Overmann, 2003b). The consortia also move preferentially to the optimal light wavelength absorbed by the nonmobile photosynthetic bacteria (Frostl and Overmann, 1998, 2000).

Several obligate associations between microbial species occur within the context of a eukaryotic host. For example, in the glassywinged sharpshooter, Homalodisca coagulata, metagenomic analysis of its microbial symbionts revealed the presence of complementary amino acid, vitamin, and cofactor biosynthetic pathways in two microbial symbionts, Baumannia cicadellinicola and Sulcia muelleri. Both symbionts are required to sustain the sharpshooter, which feeds on the amino acid–poor diet of plant xylem (Wu et al., 2006). Several species from another group of plant sap-feeding insects, the mealybugs, harbor not only multiple symbionts, but one symbiont, a Gammaproteobacterium, is housed inside the second symbiont, a Betaproteobacterium (von Dohlen et al., 2001). The functions provided to the host by each symbiont have not yet been identified, but the associations appear to be stably maintained and vertically transmitted, as reflected by cospeciation in symbiont phylogenies (Thao et al., 2002).

Facultative associations. Conditions under which one or both species of a mutualism or commensalism survive and maintain populations in the absence of the other partner are called facultative. Many instances of facultative mutualism in microbe-microbe relationships involve the exchange or sharing of nutritional resources. For example, metabolic cooperation can result from complementary degradative capabilities or from the ability of one organism to make use of by-product generated by another. In the human oral cavity, metabolic cooperation plays a key role in structuring the complex, multispecies biofilm formed on tooth surfaces. The late successional stage colonizer Porphymonas gingivalis benefits the earlier colonizer Fusobacterium nucleatum by activating a host protease, plasmin, which F. nucleatum subsequently captures and uses to obtain nutrients (Darenfed et al., 1999). Another facultative commensalism is in plant root exudate, where peptidoglycan from the cell wall of Bacillus cereus rhizosphere isolates provides carbon to sustain the growth of Flavobacterium and Chryseobacterium species, which is otherwise carbon limited (Peterson, 2008; Peterson et al., 2006) without impacting the growth of B. cereus. In some cases, metabolic cooperation results from the ability of one organism to alleviate the effects of a toxin on another organism. For example, in a model system to evaluate effects of mixed organic waste on organisms important for detoxification, the p-cresol-degrading organism Pseudomonas putida DMP1 protected the p-cresol-sensitive strain Pseudomonas sp. strain GJ1, which could then degrade a second common waste compound, 2-chloroethanol (Cowan et al., 2000).

Syntrophy. The hallmark of syntrophic interactions, which can be obligate or facultative, is the coupling of metabolic processes in two organisms, typically by transfer of electrons between the organisms by hydrogen or other carriers, which facilitates metabolisms that would otherwise be thermodynamically unfavorable. Under methanogenic conditions, syntrophy appears to facilitate a number of the intermediate transformations between primary fermentation of complex organic matter and eventual production of methane (Schink, 2002). In methane-rich marine sediments, syntrophy between archaea thought to perform reverse methanogenesis and sulfate-reducing bacteria plays a role in mediating methane oxidation, an important control of the flux of this potent greenhouse gas (Boetius et al., 2000, Hallam et al., 2004). Degradation of some xenobiotic compounds also relies on syntrophy. For example, interspecies hydrogen transfer from a sulfate-reducing organism facilitates tetrachloroethene dehalorespiration by another organism (Drzyzga and Gottschal, 2002). Similarly, vinyl chloride dechlorination by Methanosarcina spp. also requires interspecies hydrogen transfer by a syntrophic partner organism (Heimann et al., 2006).

Coaggregation and multispecies biofilm formation. Many beneficial interactions between microorganisms require the partners to be maintained in close proximity, which is often achieved by the formation of multispecies biofilms or aggregates. In some cases, most notably in the oral cavity of vertebrates, development of complex communities results from specific, receptor-mediated interactions between pairs of organisms, known as coaggregation (Lamont et al., 2002). Early colonizers to tooth surfaces, such as Streptococcus gordonii and other viridans streptococci, can bind a variety of host molecules and subsequently facilitate colonization by the second-stage species through coaggregation with specific partners (Kolenbrander et al., 1990). F. nucleatum, the most abundant gram-negative bacterium in mouths of healthy people, is thought to serve as a bridge between these early and subsequent late colonizers because of its ability to coaggregate with many species from both classes (Kolenbrander et al., 2002).

Evolution in Microbial Communities

The intersection of ecology and evolution is important to our understanding of communities but has not been sufficiently studied to produce a cohesive framework. Antagonistic and mutualistic behaviors have evolved as adaptations to life in a community. Organisms exploit or compete with each other for resources, leading to the grand diversity of ecological mechanisms in the biological world. Individual species evolve in the context of a community, resulting in coevolution, and the community evolves as a composite of many species. Identifying the selection pressures that favor certain interactions is the key to deriving an evolutionary understanding of microbial communities. And perhaps there is a larger conceptual framework to be developed that will describe microbial community evolution, with the entire community as the unit upon which selection acts.

In many ways, the evolution of prokaryotes and eukaryotes is similar. Natural selection and genetic drift operate on populationlevel genetic variation caused by mutation and gene flow. Together these processes alter the genetic composition of populations and directly and indirectly affect the species interactions that dictate community ecology. However, some evolutionary processes play different roles in prokaryotic and eukaryotic populations, and these contrasts are particularly important to consider in the context of microbial community dynamics.

Genetic variation is the target on which selection acts, whereas ecological forces, including biotic factors such as species interactions (competition, parasitism, mutualism), are the agents of selection. Thus, the two fields of study, population genetics and community ecology, seem inevitably coupled through evolution.

Most ecological and evolutionary theory has been developed on the basis of observations made in eukaryotic organisms. Natural selection, developed with plants and animals, and Mendelian genetics, originating from studies of plants, were integrated to form the modern synthesis, which is the basis of current evolutionary theory, but none of the major leaders in the development of the modern synthesis (Fisher, Dobzhansky, Haldane, Wright, Huxley, Mayr, Rensch, Simpson, and Stebbins) focused on prokaryotes. Consequently, models for understanding adaptation, evolution, and speciation in prokaryotic biology were not developed until the early 1980s (Levin, 1981). Advances in molecular biology have propelled the expansion of prokaryotic models for evolution over the past 20 years, from which two major differences between prokaryotic and eukaryotic evolution have emerged: the frequency of recombination and the phylogenetic breadth among which genetic materials can be exchanged. Intragenomic processes such as recombination are likely to have the greatest influence in short-term changes in a community, leading to population adaptation to changing conditions or new metabolic opportunities. Intergenomic processes such as horizontal gene transfer have profound effects on the long-term evolution of communities, possibly leading to the formation of new species (de la Cruz and Davies, 2000; Hoffmeister and Martin, 2003).

Although a superficial examination of evolutionary processes in prokaryotes and eukaryotes reveals stark differences, deeper examination might unite them. Clonal eukaryotes, for example, may be governed by similar principles as prokaryotes. More significantly, the same forces may regulate hybridization between plant species and interspecies gene transfer in prokaryotes and the resultant afront to the integrity of the species (de la Cruz and Davies, 2000; Hoffmeister and Martin, 2003).

Intragenomic alterations. Sequential point mutations and gene rearrangements can lead to adjustments in the genotypic and, consequently, phenotypic content of population members. Mutations or alterations that are selected typically improve the fitness of an organism in its current ecological niche, which is necessary to maintain interspecific interactions such as competition, predation, and mutualism. Several hypotheses regarding microbial fitness are readily testable. For example, improvements in fitness over thousands of generations under glucose limited conditions have been measured (Papadopoulos et al., 1999). An interesting and intensely studied pattern resulting from closely evolving interactions is the coevolutionary process, in which genetically based adaptations in one species invoke reciprocal genetic changes in populations of its partner species (e.g., competitor, parasite) or guilds of species (Thompson, 1994, 2005).

Horizontal gene transfer. The transfer of genetic information between species is a central mechanism of generating genetic variation in microbial communities. For example, multilocus sequence typing data and proteomic and comparative genomics data indicate that bacterial species in acid mine biofilm communities exchange large (up to hundreds of kilobases) regions of DNA as well as smaller sections that may play a role in resistance to phage (Lo et al., 2007; Tyson et al., 2004). Events of horizontal gene transfer can be detected through phylogenetics, by seeking atypical distributions of genes across organisms, or through phylogeny-independent methods that examine genes that appear aberrant in their current genomic context. Complete genome sequencing has arguably been the most important factor in unveiling instances of horizontal gene transfer, illustrating the impact of horizontal gene transfer on bacterial evolution (Bassler, 1999; Boetius et al., 2000; Borneman et al., 1996). Perhaps one of the most dramatic impacts of horizontal transfer of genetic information by accessory genetic elements and vectors of genes (e.g., viruses) is ecological. Horizontal gene transfer can enable a microbe to rapidly expand and/or alter its ecological niche, making this genetic process important when considering evolution in microbial communities through deep time (i.e. an evolutionary timescale rather than an ecological timescale). Horizontal gene transfer has also been proposed to contribute to speciation, which is a critical aspect of community function and evolution (de la Cruz and Davies, 2000; Hoffmeister and Martin, 2003).

Ecological processes, in turn, affect evolution by providing opportunities for interspecies gene transfer and providing selection pressure. The architecture of communities, which dictates proximity of cells of different species, affects the probability of gene exchange across wide phylogenetic distances. The physical and biological features of the community will affect the susceptibility of cells to transformation or transduction, thereby affecting the frequency of gene transfer. The ecological processes and characteristics of the community create the selection pressures that determine the direction of change in frequencies of certain genotypes.

When entire genes or groups of genes are transferred between individuals, especially distantly related individuals, a trait can rapidly sweep through a population under appropriate selection pressure. A contemporary example of this is the rapid spread of antibiotic resistance in bacterial populations. In other instances, the changes can lead to rapid lineage diversification (Riley et al., 2001). In some cases these changes, especially those that involve the metabolic repertoire, enable the recipient of horizontal gene transfer to invade and adapt in a new ecological niche. Classic examples of niche-altering gene acquisitions include the lac operon by Escherichia coli and pathogenicity islands by Salmonella sp. Changes that enable an organism to invade a new niche(s) have strong implications at the community level: They have the potential to alter interactions between species and the structure, diversity, and robustness of communities. Ultimately, the ecological selection pressures that drive microbial evolution are major contributors to the emergent structure and function of the community.

Approaches to the Study of Microbial Communities

Community ecology, as it pertains to microbes, remains in its infancy. Most studies of microbemicrobe and host-microbe interactions have extracted the organisms from their native community and studied them as binary interactions; indeed, this is simpler to understand and a necessary step toward understanding interspecific interactions in a community context (Figure A9-2). However, the information gained from such studies can be inadequate or misleading, as has been shown repeatedly in macroecological research (Bronstein and Barbosa, 2002; Little and Currie, 2008; Silo-Suh et al., 1994; Stanton, 2003), and as such, results obtained from those studies must be interpreted with caution. To this end, we have parsed microbial ecology studies of communities into four groups on the basis of the questions being asked and presented them chronologically in terms of the order in which questions must be addressed and answered to generate further information on microbial community characteristics and processes. Each question is illustrated with classic experiments and recent technological advances that have enabled their investigation (Figure A9-3).

A chart showing the progression from studies on the individual scale to studies on the community scale


Progression from studies on the individual scale to studies on the community scale.

A flow chart showing four groups of questions in microbial ecology and some techniques to address them


Four groups of questions in microbial ecology and some techniques to address them.

Who Is Present in the Community?

The first question a community ecologist asks is, Who makes up this community?, and this is indeed a good starting point. But with microorganisms, this is not a trivial question, nor has it been easy to identify the diversity of species found in various communities. Two central techniques are used to identify microbial phylotypes within community samples: One technique relies on culturing, and the other is culture independent.

Culture-based methods. Koch’s discovery that bacteria could be isolated and grown in pure culture on solid artificial medium enabled the discipline bacteriology to develop. Growing bacteria in pure culture provided morphological and physiological data, which together provided the basis to identify bacterial species. More advanced culture techniques incorporated various nutrients and abiotic conditions that closely mimicked the environments from which the samples were isolated. However, recent estimates indicate that less than 1% of the membership of many communities is culturable, making it necessary to assess the identity of the as yet uncultured organisms to generate a complete list of community members.

Culture-independent methods. One molecular approach that provides a powerful complement to culture-dependent techniques is the amplification of 16S rRNA gene sequences directly from environmental samples, such as soil, using PCR and universal or domain-specific primers, which is usually followed by clone library construction. Clones are then screened to analyze sequence differences (Bond et al., 1995), or restriction fragment length polymorphisms (Moyer et al., 1994), which are then used to identify species. To date, 16S rRNA gene amplification and identification remain the most reliable tool to describe prokaryotic species. Because the gene is universal and can therefore be used as an identifier for any bacterial or archaeal species, it accounts for both culturable and nonculturable prokaryotic organisms, and it has phylogenetic meaning. There are, however, limitations to the 16S rRNA gene approach to phylogeny. Interspecies gene transfer muddies phylogenetic assessments derived from 16S rRNA. If organisms are hybrids with fragments of DNA of different organismal origins, then what is a species? Should they be defined by the 16S rRNA gene affiliation or by a census of functional genes? Prokaryote phylogeny is an emergent field and the species concept will be one to grapple with in the future.

The universal primers used to assess entire communities may not detect all species. Recent work suggests significant differences in the groups whose 16S rRNA genes are amplified when the universal primers are replaced by miniprimers with very broad specificity (Isenbarger et al., 2008). All methods impose bias and it is likely that further development of phylogenetic tools will reveal greater diversity and perhaps groups of organisms that are not suspected from current surveys.

What Are the Functions of Individual Organisms?

After identifying which organism(s) is present within a community, the next challenge for microbial ecologists is to identify which organism demonstrates each of the various metabolic processes in its native community.

Classic in vitro microbial physiology. Up until the last 15 years of the twentieth century, pure-culture experimental setup, as described above, was the central method of associating organisms with metabolic processes. In an effort to understand metabolic processes within the context of a microbe’s native community, it was typical to inoculate enrichment cultures with samples from the natural environment of interest, determine which bacteria grow, and then make inferences on the basis of substrate use about the microbe’s activity in its native community.

Metagenomics. The diversity of the as yet unculturable members of microbial communities is vast compared with that of the culturable members (Bintrim et al., 1997; Borneman et al., 1996, 1997; Kuske et al., 2002; Liles et al., 2003; McCaig et al., 2001; Quaiser et al., 2002). To capture and study the functional diversity of these organisms, a new field designated metagenomics has sprung to life. Metagenomics is the culture-independent analysis of genomes from an assemblage of microorganisms. A metagenomic analysis entails extracting DNA directly from soil, cloning it into a culturable host bacterium, and analyzing it (Rondon et al., 2000; Stein et al., 1996). This method has recently been used for massive capture and sequencing of DNA from the Sargasso Sea (Venter, 2004), acid mine drainage (Tyson et al., 2004), a Minnesota soil sample (Tringe et al., 2005), and a global ocean survey (Rusch et al., 2007). DeLong et al. (2006) applied metagenomics to planktonic microbial communities in the North Pacific Subtropical Gyre, in which they identified stratified microbial communities through comparative genomics delineated by taxonomic zonation and functional and metabolic potential. Through detailed analysis of the genes in each stratum, they inferred the photosynthetic activity at various depths in the ocean. They also found a surprisingly high frequency of cyanophage-infected cells (up to 12%), which likely structure the planktonic community via predation.

Functional metagenomics. Entire phyla in soil are known only by their 16Sr-RNAgene signatures, with nothing known about their physiology, genetics, or role in the soil community. Most work in metagenomics is driven by sequence analysis, but this work is limited by the ability to recognize gene function on the basis of sequence alone. Because many of the genes isolated from the environment have no significant similarity to genes of known function, an alternative approach is to search for genes of a particular function by functional metagenomics. In functional metagenomics, genes are sought that confer a function of interest on a host bacterium. This method requires that the genes be expressed, but it does not require that their functions be recognizable by sequence (Healy et al., 1995; Henne et al., 1999; Knietsch et al., 2003a, 2003b; Majernik et al., 2001; Streit et al., 2004; Winogradsky, 1895). Metagenomics: culture-independent analysis of DNA extracted directly from communities in environmental samples

Stable isotope probing. Stable isotope probing (SIP) involves introducing a stable isotopelabeled substrate into the community and tracking the movement of the substrate by extracting diagnostic molecules (e.g., lipids and nucleic acids) to determine which molecules have incorporated the substrate. Stable isotope ratios have been used by ecologists to follow resource use through trophic levels for decades (Deniro and Epstein, 1978, 1981; Schoeninger et al., 1983). More recently, stable isotopes have become a tool used by microbial ecologists to track the movement of substrates through microbial communities and identify which community members utilize which substrates. The main advantages of SIP are that it does not rely on culturability, and it allows direct observation of substrate movement with minimal disruption of the natural environment and community. SIP in microbial communities has been recently reviewed elsewhere (Kreuzer-Martin, 2007).

Single-cell analyses. Fully comprehending community function necessitates understanding the function and activities at all levels, including that of individual members. Individuals within a population vary in levels of expression of certain genes and growth rates (Kaern et al., 2005). For example, using flow cells and laser scanning, Strovas et al. (Srovas et al., 2007) revealed that individuals within Methylobacterium extorquens grown continuously vary in cell size at division, division time, and growth rate, and they respond differently to a substrate shift. In addition to explaining more about individual members of a community, single-cell analyses should also prove useful for studying rare and uncultured organisms. Although used to amplify low-quantity DNA for metagenomic studies, multiple displacement amplification, a technique that involves random primers and φ29 DNA polymerase), can also be used to amplify whole genomes of single cells (Gonzalex et al., 2005; Hellani et al., 2004; Neufeld et al., 2008). Multiple displacement amplification combined with technologies for capturing individual cells, such as microfluids, may lead to situations in which the genomes of rare members could be analyzed to determine the potential function of the member within its community.

How Do Organisms Interact?

Interactions are the fulcrum of communities. Studying them is essential to understanding community function, but that study is challenging. Lessons from study of cultured organisms can guide methodological choices. Culturing has been, by far, the most impactful method introduced into microbiology since the microscope. The study of organisms in pure culture has produced the staggering depth and breadth of current knowledge about cellular microbiology. Likewise, coculture can be used to study interactions in a controlled environment in which variables can be manipulated. Bacterial genetics has been similarly influential, providing experimental precision and rigor that elevate associations to causal relationships. The adaptation of cellular genetics to study interactions among organisms is likely to yield surprises and provide a foundation for community ecology principles.

Coculture experiments. Although the trend in the twentieth century was toward reduction, which meant studying microbes in isolation or sometimes pairs, studying organismal interactions in their native communities, or in assemblages that more closely resemble native communities, is not a recent idea. In 1895, Winogradsky isolated Clostridium pasteurianum, an organism that fixes free nitrogen from the air (Rhee et al., 2004; Tyson et al., 2004). But through a series of pure-culture and coculture experiments, it became clear to Winogradsky that C. pasteurianum, a strict anaerobe, could only fix nitrogen under aerobic conditions. If C. pasteurianum was grown in close association with an extreme aerobe that essentially creates an anaerobic environment for C. pasteurianum, nitrogen fixation was restored. Studying microbial assemblages in culture remains an important method to investigate organismal interactions. In 1993 Gilbert and colleagues (Gilbert et al., 1993) reported that the application of Bacillus cereus UW85, an antibiotic-producing strain used for biological control, to the soybean rhizosphere increased the abundance of bacteria from the Cytophaga-Flavobacterium (CF) group. More recently, Peterson et al. (Peterson et al., 2006) used coculturing to determine that the commensal relationship between B. cereus UW85 and the bacteria from the CF group is mediated by a B. cereus peptidoglycan. Although coculture experiments are invaluable to our understanding of interspecies interactions, a main restriction is that they are limited to a small number of species interacting together in liquid or on a plate.

Bacterial genetics. Just as bacterial genetics has provided profound insight into the function of organisms in pure culture, it can lead to an understanding of species interactions in communities. The foundation of genetics is construction of random mutations followed by mutant hunts. The randomness coupled with assessment of phenotype produces a minimally biased approach in which the genes required for a certain function or process are identified. In this age of gene arrays and genomics, mutant hunts have been replaced by more-directed methods, but broad searches remain critical to expanding our knowledge beyond the human imagination. Targeted approaches typically require the researcher to make educated predictions about the nature or type of genes involved in a process. Broad mutant hunts enable the bacteria to answer that question. Historically, the answers have been surprising, and many of the genes identified would not have been predicted to be involved in the process on the basis of sequence alone.

The community context presents new challenges for bacterial genetics. Identifying genes involved in community function requires complex screens that will be difficult to apply to large collections of mutants, but the search will be worthwhile. In addition, studies of targeted mutants in communities can be revealing. For example, the study of genes involved in quorum sensing in pure culture and in a community provides very different insights. The ubiquitous presence of genes involved in quorumsensing indicates that density-dependent cellcell communication is a common mode of bacterial communication (6, Manefieled and Turner, 2002; Miller and Bassler, 2001). Demonstration of signal exchange between strains in a community in a caterpillar gut provided surprising evidence of stability of quorum-sensing signal in a high pH environment (B. Borlee, G. Geske, C. Robinson, H. Blackwell & J. Handelsman, manuscript in revision). Genes that code for antibacterial compounds are found in several environmental isolates (Brinkhoff et al., 2004; Derzelle et al., 2002; Emmert et al., 2004; Li et al., 2007). The role of antimicrobials in nature is controversial (Brinkhoff et al., 2004; Davies, 2006; Derzelle et al., 2002; Emmert et al., 2004; Li et al., 2007; Yim et al., 2007). Some studies, however, have shown that producing these compounds yields a competitive advantage for bacteria (Franks et al., 2006; Giddens et al., 2003). An antibacterial protein produced by a marine bacterium is important for its competition with other marine bacteria for the formation of biofilms as well as dispersal of cells from biofilms (Mai-Prochnow et al., 2004; Rao et al., 2005).

Can We Make Predictions at the Community Scale?

Perhaps the most important implication of information on ecological community principles and dynamics from both an application and a theoretical standpoint is the potential to build predictive models. By defining fundamental principles of microbial community ecology derived from phylogenetically and ecologically disparate systems amenable to experimental manipulation, we can generate models to predict information about other communities that are less tractable for study. Much of the baseline research needed to generate such models falls into the categories described above, namely, who is present in a community, what is their function, and how do they interact with one another. Advances in molecular biology, and computational biology in particular, have allowed development in predictive ecological modeling that is poised for application to microbial communities.

Integrative modeling. In parallel with the recurring theme of moving from reductionist science to systems-based biology, many current approaches toward understanding properties of microbial communities include integrating biochemistry, thermodynamics, metabolite transport and utilization, metagenomic sequencing, regulatory and metabolic network analysis, and comparative and evolutionary genomics. On the mechanistic scale, several groups aim to predict microbially mediated metabolic activities in specific environments using microarrays and to predict a microbe’s behavior and lifestyle directly from its genomic sequence (Antonovics, 1992). On the ecological scale, ecological niche modeling packages can output predictions of geographic ranges for species on the basis of current species records and layers of environmental data (e.g., GARP, Genetic Algorithm for Rule-set Production).

Community genetics. An integrative field of study becoming increasingly important in generating predictions about how microbial community members interact is community genetics, which involves coupling changes in genetic distributions with species interactions and community structure (Dungey et al., 2000; Whitham et al., 2003). A few studies of macroecological systems have generated predictions. For example, Whitham and colleagues found that genetic variation in plants affects diverse communities of insect herbivores, birds, and fungi (Whitham et al., 2003). Other studies have shown that population dynamics and trophic interactions affect the rate at which pests evolve resistance to genetically engineered crops. Experimental results from a predator-prey system (rotifers that consume algae) provided the basis for building models for ecological and evolutionary hypotheses about the predator-prey cycles. These studies in situations involving rapid evolution model successfully predicted ecological consequences (Shertzer et al., 2002).

There has been little application of community genetics to make predictions of ecological phenomena in predominantly microbial communities, but it will be a fruitful avenue of research. Community genetics in microbial communities has the special power of using constructed, defined mutants that can be introduced into a community. The behavior of the community in the presence of the mutant and wild type can be compared, providing insight into the role of a single gene in a population of a species and in the community context. Just as bacterial genetics brought power and precision to the dissection of bacterial cell processes in the twentieth century, the same approaches will transform our understanding of community processes in the twentyfirst century.

Summary Points

  1. An important feature of microbial communities is robustness, which has structural and functional components.
  2. Understanding microbial community ecology necessitates study of evolutionary mechanisms that underlie community structure.
  3. Recent advances in molecular biology provide a means to address questions about microbial communities, including both culturable and as yet uncultured members.
  4. Microbial communities offer a unique opportunity to apply community genetics in a manner that is difficult in most macroorganism communities: Introduction of defined mutants into communities will advance our understanding of the interplay between community genetic composition and community structure and function under various selection pressures.
  5. The approaches described here will contribute to the inputs needed to build and test predictive models that will elucidate principles that govern community interactions, providing a set of rules of engagement among community members that dictate community structure and function.

Disclosure Statement

The authors are not aware of any biases that might be perceived as affecting the objectivity of this review.


We are grateful for support from an Interdisciplinary Hatch Project from the University of Wisconsin College of Agricultural and Life Sciences and the Howard Hughes Medical Institute.

We thank Karen Cloud-Hansen for helpful discussions of an earlier draft of the manuscript.


  • Adamczyk J, Hesselsoe M, Iversen N, Horn M, Lehner A, et al. The isotope array, a new tool that employs substrate-mediated labeling of rRNA for determination of microbial community structure and function. Appl Environ Microbiol. 2003;69:6875–87. [PMC free article: PMC262286] [PubMed: 14602652]
  • Alonso-Saez L, Balague V, Sa EL, Sanchez O, Gonzalez JM, et al. Seasonality in bacterial diversity in north-west Mediterranean coastal waters: assessment through clone libraries, fingerprinting and FISH. FEMS Microbiol Ecol. 2007;60:98–112. [PubMed: 17250750]
  • An D, Danhorn T, Fuqua C, Parsek MR. Quorum sensing and motility mediate interactions between Pseudomonas aeruginosa and Agrobacterium tumefaciens in biofilm cocultures. Proc. Natl. Acad. Sci USA. 2006;103:3828–33. [PMC free article: PMC1533783] [PubMed: 16537456]
  • Antonovics J. Towards community genetics. In: Fritz RS, Simms EL, editors. Plant Resistance to Herbivores and Pathogens: Ecology, Evolution, and Genetics. Chicago: Univ. Chicago Press; 1992. pp. 426–49.
  • Ashelford KE, Day MJ, Fry JC. Elevated abundance of bacteriophage infecting bacteria in soil. Appl Environ Microbiol. 2003;69:285–89. [PMC free article: PMC152412] [PubMed: 12514006]
  • Bassler BL. How bacteria talk to each other: regulation of gene expression by quorum sensing. Curr Opin Microbiol. 1999;2:582–87. [PubMed: 10607620]
  • Beardsley C, Pernthaler J, Wosniok W, Amann R. Are readily culturable bacteria in coastal North Sea waters suppressed by selective grazing mortality? Appl Environ Microbiol. 2003;69:2624–30. [PMC free article: PMC154555] [PubMed: 12732530]
  • Begon M, Harper J, Townsend C. Ecology: Individuals, Populations, and Communities. Cambridge, MA: Blackwell; 1990. 945 pp.
  • Bintrim SB, Donohue TJ, Handelsman J, Roberts GP, Goodman RM. Molecular phylogeny of Archaea from soil. Proc. Natl. Acad. Sci USA. 1997;94:277–82. [PMC free article: PMC19314] [PubMed: 8990199]
  • Boetius A, Ravenschlag K, Schubert CJ, Rickert D, Widdel F, et al. A marine microbial consortium apparently mediating anaerobic oxidation of methane. Nature. 2000;407:623–26. [PubMed: 11034209]
  • Bond PL, Hugenholtz P, Keller J, Blackall LL. Bacterial community structures of phosphatere-moving and nonphosphate-removing activated sludges from sequencing batch reactors. Appl Environ Microbiol. 1995;61:1910–16. [PMC free article: PMC167453] [PubMed: 7544094]
  • Bonsall RF, Weller DM, Thomashow LS. Quantification of 2,4-diacetylphloroglucinol produced by fluorescent Pseudomonas spp. in vitro and in the rhizosphere of wheat. Appl Environ Microbiol. 1997;63:951–55. [PMC free article: PMC1389124] [PubMed: 16535559]
  • Borneman J, Skroch PW, O’Sullivan KM, Palus JA, Rumjanek NG, et al. Molecular microbial diversity of an agricultural soil in Wisconsin. Appl Environ Microbiol. 1996;62:1935–43. [PMC free article: PMC167971] [PubMed: 8787391]
  • Borneman J, Triplett EW. Molecular microbial diversity in soils from eastern Amazonia: evidence for unusual microorganisms and microbial population shifts associated with deforestation. Appl Environ Microbiol. 1997;63:2647–53. [PMC free article: PMC168563] [PubMed: 9212415]
  • Breitbart M, Wegley L, Leeds S, Schoenfeld T, Rohwer F. Phage community dynamics in hot springs. Appl Environ Microbiol. 2004;70:1633–40. [PMC free article: PMC368299] [PubMed: 15006788]
  • Brinkhoff T, Bach G, Heidorn T, Liang L, Schlingloff A, Simon M. Antibiotic production by a Roseobacter clade-affiliated species from the German Wadden Sea and its antagonistic effects on indigenous isolates. Appl Environ Microbiol. 2004;70:2560–65. [PMC free article: PMC383154] [PubMed: 15066861]
  • Brodie EL, DeSantis TZ, Parker JPM, Zubietta IX, Piceno YM, Andersen GL. Urban aerosols harbor diverse and dynamic bacterial populations. Proc. Natl. Acad. Sci USA. 2007;104:299–304. [PMC free article: PMC1713168] [PubMed: 17182744]
  • Bronstein J, Barbosa P. Multi-trophic/multi-species interactions: the role of nonmutualists in shaping and mediating mutualisms. In: Tscharntke T, Hawkins BA, editors. Multitrophic Interactions. Cambridge, UK: Cambridge Univ. Press; 2002. pp. 44–66.
  • Cardoza YJ, Klepzig KD, Raffa KF. Bacteria in oral secretions of an endophytic insect inhibit antagonistic fungi. Ecol Entomol. 2006;31:636–45.
  • Carrino-Kyker SR, Swanson AK. Temporal and spatial patterns of eukaryotic and bacterial communities found in vernal pools. Appl Environ Microbiol. 2008;74:2554–57. [PMC free article: PMC2293170] [PubMed: 18310431]
  • Case RJ, Boucher Y, Dahllof I, Holmstrom C, Doolittle WF, Kjelleberg S. Use of 16S rRNA and rpoB genes as molecular markers for microbial ecology studies. Appl Environ Microbiol. 2007;73:278–88. [PMC free article: PMC1797146] [PubMed: 17071787]
  • Chandler DP, Jarrell AE, Roden ER, Golova J, Chernov B, et al. Suspension array analysis of 16S rRNA from Fe- and SO42-reducing bacteria in uranium-contaminated sediments undergoing bioremediation. Appl Environ Microbiol. 2006;72:4672–87. [PMC free article: PMC1489301] [PubMed: 16820459]
  • Chenier MR, Beaumier D, Fortin N, Roy R, Driscoll BT, et al. Influence of nutrient inputs, hexadecane, and temporal variations on denitrification and community composition of river biofilms. Appl Environ Microbiol. 2006;72:575–84. [PMC free article: PMC1352192] [PubMed: 16391094]
  • Chu H, Fujii T, Morimoto S, Lin X, Yagi K, et al. Community structure of ammonia-oxidizing bacteria under long-term application of mineral fertilizer and organic manure in a sandy loam soil. Appl Environ Microbiol. 2007;73:485–91. [PMC free article: PMC1796994] [PubMed: 17098920]
  • Clarholm M. Bacteria and protozoa as integral components of the forest ecosystem—their role in creating a naturally varied soil fertility. Antonie van Leeuwenhoek. 2002;81:309–18. [PubMed: 12448729]
  • Colwell RK. EstimateS Ver. 7.5: statistical estimation of species richness and shared species from samples. 2005. (Software and User’s Guide) http://viceroy​.eeb.uconn​.edu/EstimateSPages​/EstimateSSupport.htm.
  • Compant S, Duffy B, Nowak J, Clement C, Barka EA. Use of plant growth-promoting bacteria for biocontrol of plant diseases: principles, mechanisms of action, and future prospects. Appl Environ Microbiol. 2005;71:4951–59. [PMC free article: PMC1214602] [PubMed: 16151072]
  • Cottrell MT, Kirchman DL. Natural assemblages of marine Proteobacteria and members of the Cytophaga-Flavobacter cluster consuming low- and high-molecular-weight dissolved organic matter. Appl Environ Microbiol. 2000;66:1692–97. [PMC free article: PMC92043] [PubMed: 10742262]
  • Cowan SE, Gilbert E, Liepmann D, Keasling JD. Commensal interactions in a dual-species biofilm exposed to mixed organic compounds. Appl Environ Microbiol. 2000;66:4481–85. [PMC free article: PMC92328] [PubMed: 11010902]
  • Cox CR, Gilmore MS. Native microbial colonization of Drosophila melanogaster and its use as a model of Enterococcus faecalis pathogenesis. Infect Immun. 2007;75:1565–76. [PMC free article: PMC1865669] [PubMed: 17220307]
  • Currie CR, Poulsen M, Mendenhall J, Boomsma JJ, Billen J. Coevolved crypts and exocrine glands support mutualistic bacteria in fungus-growing ants. Science. 2006;311:81–83. [PubMed: 16400148]
  • Curtis TP, Head IM, Lunn M, Woodcock S, Schloss PD, Sloan WT. What is the extent of prokaryotic diversity? Philos. Trans. R. Soc. London Ser B. 2006;361:2023–37. [PMC free article: PMC1764926] [PubMed: 17028084]
  • Curtis TP, Sloan WT. Prokaryotic diversity and its limits: microbial community structure in nature and implications for microbial ecology. Curr Opin Microbiol. 2004;7:221–26. [PubMed: 15196488]
  • Curtis TP, Sloan WT, Scannell JW. Estimating prokaryotic diversity and its limits. Proc. Natl. Acad. Sci USA. 2002;99:10494–99. [PMC free article: PMC124953] [PubMed: 12097644]
  • Darenfed H, Grenier D, Mayrand D. Acquisition of plasmin activity by Fusobacterium nucleatum subsp. nucleatum and potential contribution to tissue destruction during periodontitis. Infect Immun. 1999;67:6439–44. [PMC free article: PMC97053] [PubMed: 10569761]
  • Davidov Y, Friedjung A, Jurkevitch E. Structure analysis of a soil community of predatory bacteria using culture-dependent and culture-independent methods reveals a hitherto undetected diversity of Bdellovibrio-and-like organisms. Environ Microbiol. 2006;8:1667–73. [PubMed: 16913926]
  • Davidov Y, Huchon D, Koval SF, Jurkevitch E. A new alpha-proteobacterial clade of Bdellovibriolike predators: implications for the mitochondrial endosymbiotic theory. Environ Microbiol. 2006;8:2179–88. [PubMed: 17107559]
  • Davies J. Are antibiotics naturally antibiotics? J Ind Microbiol Biotechnol. 2006;33:496–99. [PubMed: 16552582]
  • Davies J, Spiegelman GB, Yim G. The world of subinhibitory antibiotic concentrations. Antimicrobials/Genomics. 2006;9:445–53. [PubMed: 16942902]
  • de Carcer DA, Martin M, Karlson U, Rivilla R. Changes in bacterial populations and in biphenyl dioxygenase gene diversity in a polychlorinated biphenyl-polluted soil after introduction of willow trees for rhizoremediation. Appl Environ Microbiol. 2007;73:6224–32. [PMC free article: PMC2075012] [PubMed: 17693557]
  • de la Cruz F, Davies J. Horizontal gene transfer and the origin of species: lessons from bacteria. Trends Microbiol. 2000;8:128–33. [PubMed: 10707066]
  • DeLong EF, Preston CM, Mincer T, Rich V, Hallam SJ, et al. Community genomics among stratified microbial assemblages in the ocean’s interior. Science. 2006;311:496–503. [PubMed: 16439655]
  • Deniro MJ, Epstein S. Influence of diet on distribution of carbon isotopes in animals. Geochim. Cosmochim Acta. 1978;42:495–506.
  • Deniro MJ, Epstein S. Influence of diet on the distribution of nitrogen isotopes in animals. Geochim. Cosmochim Acta. 1981;45:341–51.
  • Derzelle S, Duchaud E, Kunst F, Danchin A, Bertin P. Identification, characterization, and regulation of a cluster of genes involved in carbapenem biosynthesis in Photorhabdus luminescens. Appl Environ Microbiol. 2002;68:3780–89. [PMC free article: PMC124005] [PubMed: 12147472]
  • Diaz PI, Chalmers NI, Rickard AH, Kong C, Milburn CL, et al. Molecular characterization of subject-specific oral microflora during initial colonization of enamel. Appl Environ Microbiol. 2006;72:2837–48. [PMC free article: PMC1449052] [PubMed: 16597990]
  • Drakare S. Competition between picoplanktonic cyanobacteria and heterotrophic bacteria along crossed gradients of glucose and phosphate. Microb Ecol. 2002;44:327–35. [PubMed: 12399896]
  • Drzyzga O, Gottschal JC. Tetrachloroethene dehalorespiration and growth of Desulfitobacterium frappieri TCE1 in strict dependence on the activity of Desulfovibrio fructosivorans. Appl Environ Microbiol. 2002;68:642–49. [PMC free article: PMC126694] [PubMed: 11823202]
  • Dungey HS, Potts BM, Whitham TG, Li HF. Plant genetics affects arthropod community richness and composition: evidence from a synthetic eucalypt hybrid population. Evol Int J Org Evol. 2000;54:1938–46. [PubMed: 11209771]
  • Dykhuizen DE. Santa Rosalia revisited: Why are there so many species of bacteria? Antonie van Leeuwenhoek. 1998;73:25–33. [PubMed: 9602276]
  • Emmert EAB, Klimowicz AK, Thomas MG, Handelsman J. Genetics of zwittermicin A production by Bacillus cereus. Appl Environ Microbiol. 2004;70:104–13. [PMC free article: PMC321298] [PubMed: 14711631]
  • Estes JA. Predators and ecosystem management. Wildl Soc Bull. 1996;24:390–96.
  • Estes JA, Crooks K, Holt R. Predators, ecological role of. In: Levin SA, editor. Encyclopedia of Biodiversity. Vol. 4. San Diego, CA: Academic Press; 2001. pp. 857–78.
  • Feil EJ, Holmes EC, Bessen DE, Chan MS, Day NP, et al. Recombination within natural populations of pathogenic bacteria: short-term empirical estimates and long-term phylogenetic consequences. Proc. Natl. Acad. Sci USA. 2001;98:182–87. [PMC free article: PMC14565] [PubMed: 11136255]
  • Fiegna F, Velicer GJ. Exploitative and hierarchical antagonism in a cooperative bacterium. PLoS Biol. 2005;3:e370. [PMC free article: PMC1275521] [PubMed: 16248676]
  • Francis CA, Roberts KJ, Beman JM, Santoro AE, Oakley BB. Ubiquity and diversity of ammoniaoxidizing archaea in water columns and sediments of the ocean. Proc. Natl. Acad. Sci USA. 2005;102:14683–88. [PMC free article: PMC1253578] [PubMed: 16186488]
  • Franks A, Egan S, Holmstrom C, James S, Lappin-Scott H, Kjelleberg S. Inhibition of fungal colonization by Pseudoalteromonas tunicata provides a competitive advantage during surface colonization. Appl Environ Microbiol. 2006;72:6079–87. [PMC free article: PMC1563610] [PubMed: 16957232]
  • Frostl JM, Overmann J. Physiology and tactic response of the phototrophic consortium “Chlorochromatium aggregatum” Arch Microbiol. 1998;169:129–35. [PubMed: 9446684]
  • Frostl JM, Overmann J. Phylogenetic affiliation of the bacteria that constitute phototrophic consortia. Arch Microbiol. 2000;174:50–58. [PubMed: 10985742]
  • Gans J, Wolinsky M, Dunbar J. Computational improvements reveal great bacterial diversity and high metal toxicity in soil. Science. 2005;309:1387–90. [PubMed: 16123304]
  • Gentile ME, Lynn Nyman J, Criddle CS. Correlation of patterns of denitrification instability in replicated bioreactor communities with shifts in the relative abundance and the denitrification patterns of specific populations. ISME J. 2007;1:714–28. [PubMed: 18059495]
  • Geraci A. IEEE Standard Computer Dictionary: Compilation of IEEE Standard Computer Glossaries. New York: Inst. Elect Electron Eng Inc; 1991.
  • Giddens SR, Houliston GJ, Mahanty HK. The influence of antibiotic production and preemptive colonization on the population dynamics of Pantoea agglomerans (Erwinia herbicola) Eh1087 and Erwinia amylovora in planta. Environ Microbiol. 2003;5:1016–21. [PubMed: 14510856]
  • Gilbert G, Parke JL, Clayton MK, Handelsman J. Effects of an introduced bacterium on bacterial communities on roots. Ecology. 1993;74:840–54.
  • Gil-Turnes MS, Fenical W. Embryos of Homarus americanus are protected by epibiotic bacteria. Biol Bull. 1992;182:105–8. [PubMed: 29304709]
  • Gil-Turnes MS, Hay ME, Fenical W. Symbiotic marine bacteria chemically defend crustacean embryos from a pathogenic fungus. Science. 1989;246:116–18. [PubMed: 2781297]
  • Girvan MS, Campbell CD, Killham K, Prosser JI, Glover LA. Bacterial diversity promotes community stability and functional resilience after perturbation. Environ Microbiol. 2005;7:301–13. [PubMed: 15683391]
  • Glaeser J, Overmann J. Characterization and in situ carbon metabolism of phototrophic consortia. Appl Environ Microbiol. 2003;69:3739–50. [PMC free article: PMC165192] [PubMed: 12839739]
  • Glaeser J, Overmann J. The significance of organic carbon compounds for in situ metabolism and chemotaxis of phototrophic consortia. Environ Microbiol. 2003;5:1053–63. [PubMed: 14641585]
  • Gogarten JP, Doolittle WF, Lawrence JG. Prokaryotic evolution in light of gene transfer. Mol Biol Evol. 2002;19:2226–38. [PubMed: 12446813]
  • Goldman BS, Nierman WC, Kaiser D, Slater SC, Durkin AS, et al. Evolution of sensory complexity recorded in a myxobacterial genome. Proc. Natl. Acad. Sci USA. 2006;103:15200–5. [PMC free article: PMC1622800] [PubMed: 17015832]
  • Gonzalez JM, Portillo MC, Saiz-Jimenez C. Multiple displacement amplification as a prepolymerase chain reaction (pre-PCR) to process difficult to amplify samples and low copy number sequences from natural environments. Environ Microbiol. 2005;7:1024–28. [PubMed: 15946299]
  • Gopal PK, Prasad J, Smart J, Gill HS. In vitro adherence properties of Lactobacillus rhamnosus DR20 and Bifidobacterium lactis DR10 strains and their antagonistic activity against an enterotoxigenic Escherichia coli. Int J Food Microbiol. 2001;67:207–16. [PubMed: 11518430]
  • Guarner F, Malagelada JR. Gut flora in health and disease. Lancet. 2003;361:512–19. [PubMed: 12583961]
  • Hallam SJ, Putnam N, Preston CM, Detter JC, Rokhsar D, et al. Reverse methanogenesis: testing the hypothesis with environmental genomics. Science. 2004;305:1457–62. [PubMed: 15353801]
  • Handelsman J. Metagenomics: application of genomics to uncultured microorganisms. Microbiol Mol Biol Rev. 2004;68:669–85. [PMC free article: PMC539003] [PubMed: 15590779]
  • Hansel CM, Fendorf S, Jardine PM, Francis CA. Changes in bacterial and archaeal community structure and functional diversity along a geochemically variable soil profile. Appl Environ Microbiol. 2008;74:1620–33. [PMC free article: PMC2258623] [PubMed: 18192411]
  • Hartmann M, Widmer F. Community structure analyses are more sensitive to differences in soil bacterial communities than anonymous diversity indices. Appl Environ Microbiol. 2006;72:7804–12. [PMC free article: PMC1694274] [PubMed: 17041161]
  • Healy FG, Ray RM, Aldrich HC, Wilkie AC, Ingram LO, Shanmugam KT. Direct isolation of functional genes encoding cellulases from the microbial consortia in a thermophilic, anaerobic digester maintained on lignocellulose. Appl Microbiol Biotechnol. 1995;43:667–74. [PubMed: 7546604]
  • Heimann AC, Batstone DJ, Jakobsen R. Methanosarcina spp. drive vinyl chloride dechlorination via interspecies hydrogen transfer. Appl Environ Microbiol. 2006;72:2942–49. [PMC free article: PMC1449020] [PubMed: 16598001]
  • Hellani A, Coskun S, Sakati N, Benkhalifa M, Al-Odaib A, Ozand P. Multiple displacement amplification on single cell and possible preimplantation genetic diagnosis applications. Fertil Steril. 2004;82:S28.
  • Henke SE, Bryant FC. Effects of coyote removal on the faunal community in western Texas. J Wildl Manag. 1999;63:1066–81.
  • Henne ADR, Schmitz RA, Gottschalk G. Construction of environmental DNA libraries in Escherichia coli and screening for the presence of genes conferring utilization of 4-hydroxybutyrate. Appl Environ Microbiol. 1999;65:3901–7. [PMC free article: PMC99718] [PubMed: 10473393]
  • Hoffmeister M, Martin W. Interspecific evolution: microbial symbiosis, endosymbiosis and gene transfer. Environ Microbiol. 2003;5:641–49. [PubMed: 12871231]
  • Hooper LV. Bacterial contributions to mammalian gut development. Trends Microbiol. 2004;12:129–34. [PubMed: 15001189]
  • Hooper LV, Midtvedt T, Gordon JI. How host-microbial interactions shape the nutrient environment of the mammalian intestine. Annu Rev Nutr. 2002;22:283–307. [PubMed: 12055347]
  • Huber H, Hohn MJ, Rachel R, Fuchs T, Wimmer VC, Stetter KO. A new phylum of Archaea represented by a nanosized hyperthermophilic symbiont. Nature. 2002;417:63–67. [PubMed: 11986665]
  • Isenbarger TA, Finney M, Ríos-Velázquez C, Handelsman J, Ruvkun G. Miniprimer PCR, a new lens for viewing the microbial world. Appl Environ Microbiol. 2008;74:840–49. [PMC free article: PMC2227730] [PubMed: 18083877]
  • Ives AR, Carpenter SR. Stability and diversity of ecosystems. Science. 2007;317:58–62. [PubMed: 17615333]
  • Jurkevitch E, Minz D, Ramati B, Barel G. Prey range characterization, ribotyping, and diversity of soil and rhizosphere Bdellovibrio spp. isolated on phytopathogenic bacteria. Appl Environ Microbiol. 2000;66:2365–71. [PMC free article: PMC110534] [PubMed: 10831412]
  • Kadurugamuwa JL, Beveridge TJ. Bacteriolytic effect of membrane vesicles from Pseudomonas aeruginosa on other bacteria including pathogens: conceptually new antibiotics. J Bacteriol. 1996;178:2767–74. [PMC free article: PMC178010] [PubMed: 8631663]
  • Kaern M, Elston TC, Blake WJ, Collins JJ. Stochasticity in gene expression: from theories to phenotypes. Nat Rev Genet. 2005;6:451–64. [PubMed: 15883588]
  • Kaltenpoth M, Gottler W, Herzner G, Strohm E. Symbiotic bacteria protect wasp larvae from fungal infestation. Curr Biol. 2005;15:475–79. [PubMed: 15753044]
  • Kan J, Suzuki MT, Wang K, Evans SE, Chen F. High temporal but low spatial heterogeneity of bacterioplankton in the Chesapeake Bay. Appl Environ Microbiol. 2007;73:6776–89. [PMC free article: PMC2074944] [PubMed: 17827310]
  • Kanzler BEM, Pfannes KR, Vogl K, Overmann J. Molecular characterization of the nonphotosynthetic partner bacterium in the consortium “Chlorochromatium aggregatum” Appl Environ Microbiol. 2005;71:7434–41. [PMC free article: PMC1287705] [PubMed: 16269785]
  • Katayama T, Tanaka M, Moriizumi J, Nakamura T, Brouchkov A, et al. Phylogenetic analysis of bacteria preserved in a permafrost ice wedge for 25,000 years. Appl Environ Microbiol. 2007;73:2360–63. [PMC free article: PMC1855676] [PubMed: 17293514]
  • Keel C, Schnider U, Maurhofer M, Voisard C, Laville J, et al. Suppression of root diseases by Pseudomonas fluorescens CHA 0: importance of the bacterial secondary metabolite 2,4-diacetylphloroglucinol. Mol. Plant-Microbe Interact. 1992;5:4–13.
  • Kent AD, Jones SE, Lauster GH, Graham JM, Newton RJ, McMahon KD. Experimental manipulations of microbial food web interactions in a humic lake: shifting biological drivers of bacterial community structure. Environ Microbiol. 2006;8:1448–59. [PubMed: 16872407]
  • Kikuchi Y, Graf J. Spatial and temporal population dynamics of a naturally occurring two-species microbial community inside the digestive tract of the medicinal leech. Appl Environ Microbiol. 2007;73:1984–91. [PMC free article: PMC1828818] [PubMed: 17277211]
  • Kitano H. Computational systems biology. Nature. 2002;420:206–10. [PubMed: 12432404]
  • Kitano H. Cancer robustness: tumour tactics. Nature. 2003;426:125. [PubMed: 14614483]
  • Knietsch A, Bowien S, Whited G, Gottschalk G, Daniel R. Identification and characterization of coenzyme B-12-dependent glycerol dehydratase- and diol dehydratase-encoding genes from metagenomic DNA libraries derived from enrichment cultures. Appl Environ Microbiol. 2003;69:3048–60. [PMC free article: PMC161467] [PubMed: 12788698]
  • Knietsch A, Waschkowitz T, Bowien S, Henne A, Daniel R. Metagenomes of complex microbial consortia derived from different soils as sources for novel genes conferring formation of carbonyls from short-chain polyols on Escherichia coli. J Mol Microbiol Biotechnol. 2003;5:46–56. [PubMed: 12673061]
  • Kolenbrander PE, Andersen RN, Blehert DS, Egland PG, Foster JS, Palmer RJ Jr. Communication among oral bacteria. Microbiol Mol Biol Rev. 2002;66:486–505. [PMC free article: PMC120797] [PubMed: 12209001]
  • Kolenbrander PE, Andersen RN, Moore LV. Intrageneric coaggregation among strains of human oral bacteria: potential role in primary colonization of the tooth surface. Appl Environ Microbiol. 1990;56:3890–94. [PMC free article: PMC185088] [PubMed: 2082831]
  • Kreuzer-Martin HW. Stable isotope probing: linking functional activity to specific members of microbial communities. Soil Sci Soc Am J. 2007;71:611–19.
  • Kunin V, He S, Warnecke F, Peterson SB, Garcia Martin H, et al. A bacterial metapopulation adapts locally to phage predation despite global dispersal. Genome Res. 2008;18:293–97. [PMC free article: PMC2203627] [PubMed: 18077539]
  • Kuske CR, Ticknor LO, Miller ME, Dunbar JM, Davis JA, et al. Comparison of soil bacterial communities in rhizospheres of three plant species and the interspaces in an arid grassland. Appl Environ Microbiol. 2002;68:1854–63. [PMC free article: PMC123825] [PubMed: 11916705]
  • Lamont RJ, El-Sabaeny A, Park Y, Cook GS, Costerton JW, Demuth DR. Role of the Streptococcus gordonii SspB protein in the development of Porphyromonas gingivalis biofilms on streptococcal substrates. Microbiology. 2002;148:1627–36. [PubMed: 12055284]
  • Leadbetter JR, Greenberg EP. Metabolism of acyl-homoserine lactone quorum-sensing signals by Variovorax paradoxus. J Bacteriol. 2000;182:6921–26. [PMC free article: PMC94816] [PubMed: 11092851]
  • LeBlanc PM, Hamelin RC, Filion M. Alteration of soil rhizosphere communities following genetic transformation of white spruce. Appl Environ Microbiol. 2007;73:4128–34. [PMC free article: PMC1932765] [PubMed: 17468272]
  • Lee YK, Puong KY. Competition for adhesion between probiotics and human gastrointestinal pathogens in the presence of carbohydrate. Br J. Nutr. 2002;88(Suppl 1):S101–8. [PubMed: 12215184]
  • Lepere C, Boucher D, Jardillier L, Domaizon I, Debroas D. Succession and regulation factors of small eukaryote community composition in a lacustrine ecosystem (Lake Pavin). Appl Environ Microbiol. 2006;72:2971–81. [PMC free article: PMC1449018] [PubMed: 16598004]
  • Levin BR. Periodic selection, infectious gene exchange, and the genetic structure of E. coli populations. Genetics. 1981;99:1–23. [PMC free article: PMC1214481] [PubMed: 7042453]
  • Levin BR, Bergstrom CT. Bacteria are different: observations, interpretations, speculations, and opinions about the mechanisms of adaptive evolution in prokaryotes. Proc. Natl. Acad. Sci USA. 2000;97:6981–85. [PMC free article: PMC34373] [PubMed: 10860960]
  • Li J, Beatty PK, Shah S, Jensen SE. Use of PCR-targeted mutagenesis to disrupt production of fusaricidin-type antifungal antibiotics in Paenibacillus polymyxa. Appl Environ Microbiol. 2007;73:3480–89. [PMC free article: PMC1932668] [PubMed: 17400768]
  • Li Z, Clarke AJ, Beveridge TJ. Gram-negative bacteria produce membrane vesicles which are capable of killing other bacteria. J Bacteriol. 1998;180:5478–83. [PMC free article: PMC107602] [PubMed: 9765585]
  • Liles MR, Manske BF, Bintrim SB, Handelsman J, Goodman RM. A census of rRNA genes and linked genomic sequences within a soil metagenomic library. Appl Environ Microbiol. 2003;69:2684–91. [PMC free article: PMC154537] [PubMed: 12732537]
  • Lin Y-H, Xu J-L, Hu J, Wang L-H, Ong SL, et al. Acyl-homoserine lactone acylase from Ralstonia strain XJ12B represents a novel and potent class of quorum-quenching enzymes. Mol Microbiol. 2003;47:849–60. [PubMed: 12535081]
  • Little AEF, Currie CR. Black yeast symbionts compromise the efficiency of antibiotic defenses in fungus-growing ants. Ecology. 2008;89:1216–22. [PubMed: 18543616]
  • Little AEF, Murakami T, Mueller UG, Currie CR. Defending against parasites: Fungus-growing ants combine specialized behaviours and microbial symbionts to protect their fungus gardens. Biol Lett. 2006;2:12–16. [PMC free article: PMC1617182] [PubMed: 17148313]
  • Lo I, Denef VJ, Verberkmoes NC, Shah MB, Goltsman D, et al. Strain-resolved community proteomics reveals recombining genomes of acidophilic bacteria. Nature. 2007;446:537–41. [PubMed: 17344860]
  • Lozupone C, Knight R. UniFrac: a new phylogenetic method for comparing microbial communities. Appl Environ Microbiol. 2005;71:8228–35. [PMC free article: PMC1317376] [PubMed: 16332807]
  • Lozupone CA, Hamady M, Kelley ST, Knight R. Quantitative and qualitative beta diversity measures lead to different insights into factors that structure microbial communities. Appl Environ Microbiol. 2007;73:1576–85. [PMC free article: PMC1828774] [PubMed: 17220268]
  • Mai-Prochnow A, Evans F, Dalisay-Saludes D, Stelzer S, Egan S, et al. Biofilm development and cell death in the marine bacterium Pseudoalteromonas tunicata. Appl Environ Microbiol. 2004;70:3232–38. [PMC free article: PMC427794] [PubMed: 15184116]
  • Majernik A, Gottschalk G, Daniel R. Screening of environmental DNA libraries for the presence of genes conferring Na(+)(Li(+))/H(+) antiporter activity on Escherichia coli: characterization of the recovered genes and the corresponding gene products. J Bacteriol. 2001;183:6645–53. [PMC free article: PMC95496] [PubMed: 11673435]
  • Manefield M, Turner SL. Quorum sensing in context: out of molecular biology and into microbial ecology. Microbiol. 2002;148:3762–64. [PubMed: 12480879]
  • Martin AP. Phylogenetic approaches for describing and comparing the diversity of microbial communities. Appl Environ Microbiol. 2002;68:3673–82. [PMC free article: PMC124012] [PubMed: 12147459]
  • Mazzola M, Cook RJ, Thomashow LS, Weller DM, Pierson LS 3rd. Contribution of phenazine antibiotic biosynthesis to the ecological competence of fluorescent pseudomonads in soil habitats. Appl Environ Microbiol. 1992;58:2616–24. [PMC free article: PMC195829] [PubMed: 1514808]
  • McCaig AE, Grayston SJ, Prosser JI, Glover LA. Impact of cultivation on characterisation of species composition of soil bacterial communities. FEMS Microbiol Ecol. 2001;35:37–48. [PubMed: 11248388]
  • McCann KS. The diversity-stability debate. Nature. 2000;405:228–33. [PubMed: 10821283]
  • McCliment EA, Voglesonger KM, O’Day PA, Dunn EE, Holloway JR, Cary SC. Colonization of nascent, deep-sea hydrothermal vents by a novel archaeal and nanoarchaeal assemblage. Environ Microbiol. 2006;8:114–25. [PubMed: 16343327]
  • McGrady-Steed J, Harris PM, Morin PJ. Biodiversity regulates ecosystem predictability. Nature. 1997;390:162–65.
  • Miller MB, Bassler BL. Quorum sensing in bacteria. Annu Rev Microbiol. 2001;55:165–99. [PubMed: 11544353]
  • Moss JA, Nocker A, Lepo JE, Snyder RA. Stability and change in estuarine biofilm bacterial community diversity. Appl Environ Microbiol. 2006;72:5679–88. [PMC free article: PMC1563641] [PubMed: 16957182]
  • Moyer CL, Dobbs FC, Karl DM. Estimation of diversity and community structure through restriction fragment length polymorphism distribution analysis of bacterial 16S rRNA genes from a microbial mat at an active, hydrothermal vent system, Loihi Seamount, Hawaii. Appl Environ Microbiol. 1994;60:871–79. [PMC free article: PMC201404] [PubMed: 7512808]
  • Murase J, Noll M, Frenzel P. Impact of protists on the activity and structure of the bacterial community in a rice field soil. Appl Environ Microbiol. 2006;72:5436–44. [PMC free article: PMC1538762] [PubMed: 16885296]
  • Nakayama J, Tanaka E, Kariyama R, Nagata K, Nishiguchi K, et al. Siamycin attenuates fsr quorum sensing mediated by a gelatinase biosynthesis-activating pheromone in Enterococcus faecalis. J Bacteriol. 2007;189:1358–65. [PMC free article: PMC1797363] [PubMed: 17071762]
  • Neufeld JD, Chen Y, Dumont MG, Murrell JC. Marine methylotrophs revealed by stable-isotope probing, multiple displacement amplification and metagenomics. Environ Microbiol. 2008;10:1526–35. [PubMed: 18294205]
  • Neuhauser C, Andow DA, Heimpel GE, May G, Shaw RG, Wagenius S. Community genetics: expanding the synthesis of ecology and genetics. Ecology. 2003;84:545–58.
  • Ochman H, Lawrence JG, Groisman EA. Lateral gene transfer and the nature of bacterial innovation. Nature. 2000;405:299–304. [PubMed: 10830951]
  • Pace NR. Opening the door onto the natural microbial world: molecular microbial ecology. Harvey Lect. 1995;91:59–78. [PubMed: 9127986]
  • Pace NR. A molecular view of microbial diversity and the biosphere. Science. 1997;276:734–40. [PubMed: 9115194]
  • Pace NR, Stahl DA, Lane DJ, Olsen GJ. Analyzing natural microbial populations by rRNA sequences. ASM News. 1985;51:4–12.
  • Paine RT. A note on trophic complexity and community stability. Am Nat. 1969;103:91–93.
  • Papadopoulos D, Schneider D, Meier-Eiss J, Arber W, Lenski RE, Blot M. Genomic evolution during a 10,000-generation experiment with bacteria. Proc. Natl. Acad. Sci USA. 1999;96:3807–12. [PMC free article: PMC22376] [PubMed: 10097119]
  • Paper W, Jahn U, Hohn MJ, Kronner M, Nather DJ, et al. Ignicoccus hospitalis sp. nov., the host of ‘Nanoarchaeum equitans.’ Int J Syst Evol Microbiol. 2007;57:803–8. [PubMed: 17392210]
  • Perez MT, Sommaruga R. Interactive effects of solar radiation and dissolved organic matter on bacterial activity and community structure. Environ Microbiol. 2007;9:2200–10. [PMC free article: PMC2040261] [PubMed: 17686018]
  • Perna NT, Plunkett G 3rd, Burland V, Mau B, Glasner JD, et al. Genome sequence of enterohaemorrhagic Escherichia coli O157:H7. Nature. 2001;409:529–33. [PubMed: 11206551]
  • Pernthaler J. Predation on prokaryotes in the water column and its ecological implications. Nat Rev Microbiol. 2005;3:537–46. [PubMed: 15953930]
  • Peterson SB. PhD thesis. Univ. Wis.-Madison; 2008. Interactions between rhizosphere bacteria.
  • Peterson SB, Dunn AK, Klimowicz AK, Handelsman J. Peptidoglycan from Bacillus cereus mediates commensalism with rhizosphere bacteria from the Cytophaga-Flavobacterium group. Appl Environ Microbiol. 2006;72:5421–27. [PMC free article: PMC1538759] [PubMed: 16885294]
  • Pierson LS, Pierson EA. Phenazine antibiotic production in Pseudomonas aureofaciens: role in rhizosphere ecology and pathogen suppression. FEMS Microbiol Lett. 1996;136:101–8.
  • Pineiro SA, Sahaniuk GE, Romberg E, Williams HN. Predation pattern and phylogenetic analysis of Bdellovibrionaceae from the Great Salt Lake, Utah. Curr Microbiol. 2004;48:113–17. [PubMed: 15057478]
  • Pineiro SA, Stine OC, Chauhan A, Steyert SR, Smith R, Williams HN. Global survey of diversity among environmental saltwater Bacteriovoracaceae. Environ Microbiol. 2007;9:2441–50. [PubMed: 17803770]
  • Quaiser A, Ochsenreiter T, Klenk HP, Kletzin A, Treusch AH, et al. First insight into the genome of an uncultivated crenarchaeote from soil. Environ Microbiol. 2002;4:603–11. [PubMed: 12366755]
  • Raaijmakers JM, Vlami M, de Souza JT. Antibiotic production by bacterial biocontrol agents. Antonie van Leeuwenhoek. 2002;81:537–47. [PubMed: 12448749]
  • Raffa KF. Transgenic resistance in short-rotation plantation trees. In: Strauss SH, Bradshaw HD, editors. The Bioengineered Forest: Challenges for Science and Society. Washington, DC: RFF Press; 2004. pp. 208–27.
  • Rakoff-Nahoum S, Paglino J, Eslami-Varzaneh F, Edberg S, Medzhitov R. Recognition of commensal microflora by Toll-like receptors is required for intestinal homeostasis. Cell. 2004;118:229–41. [PubMed: 15260992]
  • Rao D, Webb JS, Kjelleberg S. Competitive interactions in mixed-species biofilms containing the marine bacterium Pseudoalteromonas tunicata. Appl Environ Microbiol. 2005;71:1729–36. [PMC free article: PMC1082554] [PubMed: 15811995]
  • Rastall RA, Gibson GR, Gill HS, Guarner F, Klaenhammer TR, et al. Modulation of the microbial ecology of the human colon by probiotics, prebiotics and synbiotics to enhance human health: an overview of enabling science and potential applications. FEMS Microbiol Ecol. 2005;52:145–52. [PubMed: 16329901]
  • Reichenbach H. The ecology of the myxobacteria. Environ Microbiol. 1999;1:15–21. [PubMed: 11207714]
  • Rhee SK, Liu XD, Wu LY, Chong SC, Wan XF, Zhou JZ. Detection of genes involved in biodegradation and biotransformation in microbial communities by using 50-mer oligonucleotide microarrays. Appl Environ Microbiol. 2004;70:4303–17. [PMC free article: PMC444823] [PubMed: 15240314]
  • Riley MS, Cooper VS, Lenski RE, Forney LJ, Marsh TL. Rapid phenotypic change and diversification of a soil bacterium during 1000 generations of experimental evolution. Microbiology. 2001;147:995–1006. [PubMed: 11283295]
  • Ripple WJ, Beschtab RL. Wolf reintroduction, predation risk, and cottonwood recovery in Yellowstone National Park. For Ecol Man. 2003;184:299–313.
  • Robleto EA, Borneman J, Triplett EW. Effects of bacterial antibiotic production on rhizosphere microbial communities from a culture-independent perspective. Appl Environ Microbiol. 1998;64:5020–22. [PMC free article: PMC90960] [PubMed: 9835600]
  • Robleto EA, Scupham AJ, Triplett EW. Trifolitoxin production in Rhizobium etli strain CE3 increases competitiveness for rhizosphere. Mol. Plant-Microbe Interact. 1997;10:228–33.
  • Rogosky AM, Moak PL, Emmert EA. Differential predation by Bdellovibrio bacteriovorus 109J. Curr Microbiol. 2006;52:81–85. [PubMed: 16450066]
  • Rondon MR, August PR, Bettermann AD, Brady SF, Grossman TH, et al. Cloning the soil metagenome: a strategy for accessing the genetic and functional diversity of uncultured microorganisms. Appl Environ Microbiol. 2000;66:2541–47. [PMC free article: PMC110579] [PubMed: 10831436]
  • Rusch DB, Halpern AL, Sutton G, Heidelberg KB, Williamson S, et al. The Sorcerer II global ocean sampling expedition: northwest Atlantic through eastern tropical Pacific. PLoS Biol. 2007;5:e77. [PMC free article: PMC1821060] [PubMed: 17355176]
  • Saison C, Degrange V, Oliver R, Millard P, Commeaux C, et al. Alteration and resilience of the soil microbial community following compost amendment: effects of compost level and compostborne microbial community. Environ Microbiol. 2006;8:247–57. [PubMed: 16423013]
  • Schink B. Synergistic interactions in the microbial world. Antonie van Leeuwenhoek. 2002;81:257–61. [PubMed: 12448724]
  • Schloss PD. Evaluating different approaches that test whether microbial communities have the same structure. ISME J. 2008;2:265–75. [PubMed: 18239608]
  • Schloss PD, Handelsman J. Introducing DOTUR, a computer program for defining operational taxonomic units and estimating species richness. Appl Environ Microbiol. 2005;71:1501–6. [PMC free article: PMC1065144] [PubMed: 15746353]
  • Schloss PD, Handelsman J. Introducing SONS, a tool for operational taxonomic unit-based comparisons of microbial community memberships and structures. Appl Environ Microbiol. 2006;72:6773–79. [PMC free article: PMC1610290] [PubMed: 17021230]
  • Schloss PD, Handelsman J. Introducing TreeClimber, a test to compare microbial community structures. Appl Environ Microbiol. 2006;72:2379–84. [PMC free article: PMC1449046] [PubMed: 16597933]
  • Schloss PD, Handelsman J. The last word: books as a statistical metaphor for microbial communities. Annu Rev Microbiol. 2007;61:23–34. [PubMed: 17439361]
  • Schloss PD, Larget BR, Handelsman J. Integration of microbial ecology and statistics: a test to compare gene libraries. Appl Environ Microbiol. 2004;70:5485–92. [PMC free article: PMC520927] [PubMed: 15345436]
  • Schoeninger MJ, Deniro MJ, Tauber H. Stable nitrogen isotope ratios of bone collagen reflect marine and terrestrial components of prehistoric human diet. Science. 1983;220:1381–83. [PubMed: 6344217]
  • Shertzer KW, Ellner SP, Fussmann GF, Hairston NG. Predator-prey cycles in an aquatic microcosm: testing hypotheses of mechanism. J Anim Ecol. 2002;71:802–15.
  • Shrestha M, Abraham W-R, Shrestha PM, Noll M, Conrad R. Activity and composition of methanotrophic bacterial communities in planted rice soil studied by flux measurements, analyses of pmoA gene and stable isotope probing of phospholipid fatty acids. Environ Microbiol. 2008;10:400–12. [PubMed: 18177369]
  • Sih A, Crowley P, McPeek M, Petranka J, Strohmeier K. Predation, competition, and prey communities—a review of field experiments. Annu Rev Ecol Syst. 1985;16:269–311.
  • Silo-Suh LA, Lethbridge BJ, Raffel SJ, He H, Clardy J, Handelsman J. Biological activities of two fungistatic antibiotics produced by Bacillus cereus UW85. Appl Environ Microbiol. 1994;60:2023–30. [PMC free article: PMC201597] [PubMed: 8031096]
  • Smith JM, Castro H, Ogram A. Structure and function of methanogens along a short-term restoration chronosequence in the Florida Everglades. Appl Environ Microbiol. 2007;73:4135–41. [PMC free article: PMC1932786] [PubMed: 17449688]
  • Stanton ML. Interacting guilds: moving beyond the pairwise perspective on mutualisms. Am Nat. 2003;162:S10–23. [PubMed: 14583854]
  • Stein JL, Marsh TL, Wu KY, Shizuya H, DeLong EF. Characterization of uncultivated prokaryotes: isolation and analysis of a 40-kilobase-pair genome fragment from a planktonic marine archaeon. J Bacteriol. 1996;178:591–99. [PMC free article: PMC177699] [PubMed: 8550487]
  • Stelling J, Sauer U, Szallasi Z, Doyle FJ III, Doyle J. Robustness of cellular functions. Cell. 2004;118:675–85. [PubMed: 15369668]
  • Streit WR, Daniel R, K-E Jaeger. Prospecting for biocatalysts and drugs in the genomes of noncultured microorganisms. Curr Opin Biotechnol. 2004;15:285–90. [PubMed: 15296926]
  • Strovas TJ, Sauter LM, Guo X, Lidstrom ME. Cell-to-cell heterogeneity in growth rate and gene expression in Methylobacterium extorquens AM1. J Bacteriol. 2007;189:7127–33. [PMC free article: PMC2045205] [PubMed: 17644598]
  • Thao ML, Gullan PJ, Baumann P. Secondary (gamma-Proteobacteria) endosymbionts infect the primary (beta-Proteobacteria) endosymbionts of mealybugs multiple times and coevolve with their hosts. Appl Environ Microbiol. 2002;68:3190–97. [PMC free article: PMC126777] [PubMed: 12088994]
  • Thingstad TF. Elements of a theory for the mechanisms controlling abundance, diversity, and biogeochemical role of lytic bacterial viruses in aquatic systems. Limnol Oceanogr. 2000;45:1320–28.
  • Thomashow LS, Weller DM. Role of a phenazine antibiotic from Pseudomonas fluorescens in biological control of Gaeumannomyces graminis var. tritici. J Bacteriol. 1988;170:3499–508. [PMC free article: PMC211320] [PubMed: 2841289]
  • Thompson JN. The Coevolutionary Process. Chicago: Univ. Chicago Press; 1994. 376 pp.
  • Thompson JN. Coevolution: the geographic mosaic of coevolutionary arms races. Curr Biol. 2005;15:R992–94. [PubMed: 16360677]
  • Tilman D, Wedin D, Knops J. Productivity and sustainability influenced by biodiversity in grassland ecosystems. Nature. 1996;379:718–20.
  • Tringe SG, von Mering C, Kobayashi A, Salamov AA, Chen K, et al. Comparative metagenomics of microbial communities. Science. 2005;308:554–57. [PubMed: 15845853]
  • Tyson GW, Chapman J, Hugenholtz P, Allen EE, Ram RJ, et al. Community structure and metabolism through reconstruction of microbial genomes from the environment. Nature. 2004;428:37–43. [PubMed: 14961025]
  • Uroz S, Chhabra SR, Camara M, Williams P, Oger P, Dessaux Y. N-acylhomoserine lactone quorum-sensing molecules are modified and degraded by Rhodococcus erythropolis W2 by both amidolytic and novel oxidoreductase activities. Microbiology. 2005;151:3313–22. [PubMed: 16207914]
  • Uroz S, D’Angelo-Picard C, Carlier A, Elasri M, Sicot C, et al. Novel bacteria degrading Nacylhomoserine lactones and their use as quenchers of quorum-sensing-regulated functions of plantpathogenic bacteria. Microbiology. 2003;149:1981–99. [PubMed: 12904538]
  • Uroz S, Oger P, Chhabra SR, Camara M, Williams P, Dessaux Y. N-acylhomoserine lactones are degraded via an amidolytic activity in Comamonas sp. strain D1. Arch Microbiol. 2007;187:249–56. [PubMed: 17136382]
  • Vasanthakumar A, Delalibera I, Handelsman J, Klepzig KD, Schloss PD, Raffa KF. Characterization of gut-associated bacteria in larvae and adults of the southern pine beetle, Dendroctonus frontalis Zimmermann. Environ Entomol. 2006;35:1710–17.
  • Venter JC, Remington K, Heidelberg JF, Halpern AL, Rusch D, et al. Environmental genome shotgun sequencing of the Sargasso Sea. Science. 2004;304:66–74. [PubMed: 15001713]
  • von Dohlen CD, Kohler S, Alsop ST, McManus WR. Mealybug [beta]-proteobacterial endosymbionts contain [gamma]-proteobacterial symbionts. Nature. 2001;412:433–36. [PubMed: 11473316]
  • Wang Y-J, Leadbetter JR. Rapid acyl-homoserine lactone quorum signal biodegradation in diverse soils. Appl Environ Microbiol. 2005;71:1291–99. [PMC free article: PMC1065188] [PubMed: 15746331]
  • Waters E, Hohn MJ, Ahel I, Graham DE, Adams MD, et al. The genome of Nanoarchaeum equitans: insights into early archaeal evolution and derived parasitism. Proc. Natl. Acad. Sci USA. 2003;100:12984–88. [PMC free article: PMC240731] [PubMed: 14566062]
  • Whitham TG, Young WP, Martinsen GD, Gehring CA, Schweitzer JA, et al. Community and ecosystem genetics: a consequence of the extended phenotype. Ecology. 2003;84:559–73.
  • Williamson KE, Radosevich M, Wommack KE. Abundance and diversity of viruses in six Delaware soils. Appl Environ Microbiol. 2005;71:3119–25. [PMC free article: PMC1151856] [PubMed: 15933010]
  • Winogradsky S. Researches sur l’assimilation de l’azote libre de l’atmosphere par les microbes. Arch Sci Biol. 1895;3:297–352.
  • Wu D, Daugherty SC, Van Aken SE, Pai GH, Watkins KL, et al. Metabolic complementarity and genomics of the dual bacterial symbiosis of sharpshooters. PLoS Biol. 2006;4:e188. [PMC free article: PMC1472245] [PubMed: 16729848]
  • Yang F, Wang L-H, Wang J, Y-H Dong, Hu JY, L-H Zhang. Quorum quenching enzyme activity is widely conserved in the sera of mammalian species. FEBS Lett. 2005;579:3713–17. [PubMed: 15963993]
  • Yannarell AC, Steppe TF, Paerl HW. Disturbance and recovery of microbial community structure and function following Hurricane Frances. Environ Microbiol. 2007;9:576–83. [PubMed: 17298358]
  • Yim G, Wang HH, Davies J. Antibiotics as signalling molecules. Philos. Trans. R. Soc. London Ser B. 2007;362:1195–200. [PMC free article: PMC2435582] [PubMed: 17360275]
  • Zhang R, Weinbauer MG, P-Y Qian. Viruses and flagellates sustain apparent richness and reduce biomass accumulation of bacterioplankton in coastal marine waters. Environ Microbiol. 2007;9:3008–18. [PubMed: 17991029]
  • Zul D, Denzel S, Kotz A, Overmann J. Effects of plant biomass, plant diversity, and water content on bacterial communities in soil lysimeters: implications for the determinants of bacterial diversity. Appl Environ Microbiol. 2007;73:6916–29. [PMC free article: PMC2074975] [PubMed: 17873072]



35,36 and 35.


35 Stanford University.

Where Are We Today?

As an introduction, we describe some of the challenges that microbiologists face today in their data analyses. We then make constructive recommendations involving currently available methods and implementations.


In the 1980s the problems in statistical analyses of biological data centered around the difficulty of having insufficient data from which to draw scientific conclusions effectively. A dripping tap comes to mind. Most measurements were still being collected by hand. We then started using techniques such as the bootstrap to provide confidence statements about small samples (to about the low hundreds) measured on a small number of variables (around 10 to 20 was quite standard).

In the late 1990s many machines became available for generating biologically relevant data, not only sequences, but also microarrays, providing expression measurements on 20,000 or 30,000 genes for 100 samples, which started pouring into our data centers. It was the fire hydrant era. Today, petabytes of input give us a data tsunami in terms of pure bulk. However, we also have to cater to all the different flavors of modern data: images, texts, streaming video, sequences, high-throughput reads, and mass spectrometry features, all coming into the same labs and measured on the same individuals or in the same locations, as well as originating from different labs across the world. Today’s data show levels of heterogeneity and complexity we have not encountered to date (Figure A10-1).

An illustration showing the increasing heterogeneity and complexity of today™s data


1985: A dripping tap. 1998: A fire hydrant. 2012: A perfect storm.

All Homogeneous Data Are the Same, But Heterogeneous Data Are All Heterogeneous in Their Own Way

Here is a sample of the many levels of heterogeneity encountered in today’s microbial ecology and community data.

Date types There are images, sequences, matrices, continuous variables, networks, graphs, maps, rankings, trees, or simple binary outcomes such as healthy/ diseased.

Status Heterogeneity also occurs in terms of status of the variables; some are responses we want to predict (perturbed/normal or survival times), and other variables can serve as useful explanatory variables (such as blood pressure in patients or pH in soil samples).

Explicit or latent factors Some variables are measured and present in the data; others are latent and have to be estimated. For instance, one may actually be measuring a mixture of three separate populations without knowing it, or a hidden gradient may be present in the data. One of the goals of a good exploratory analysis can be to detect the presence of these hidden or latent factors.

Dependence We have differing levels of dependence between observations; spatial or temporal data can be very dependent, as are data collected on a network or along the branches of a river. Spatial location can be important in understanding underlying gradients. Time series allow us to follow the dynamics of biological processes.

Complexity We also want to produce summary outputs that are not just simple numerical summaries; we would like to provide interaction graphs, decision trees, and confidence regions.

Losing Touch

Contemporary biologists feel drowned in their data, often even abandoning the idea of visualizing it at all. There is a feeling that because the data come directly through a pipeline into the computers from the proprietary software developed by the measurement machine manufacturers,37 there is no need for humans to be involved. This is sometimes argued as being more efficient or objective. However, the best neural network available is still the human brain, and many breakthroughs are due to insights gleaned by biologists painstakingly recording and perusing their numbers.

Data still need to be monitored on a regular basis in the same way film directors watch their “dailies.” Data should be visualized early and often. First, we can detect flaws in the experimental design or systematic biases that were not predicted. Overall patterns and a complex picture of the data will appear as the data are tamed in an interactive way. Batch analyses at the end of a long collection process, for instance a few months or weeks before a thesis is due, are always a mistake and result in suboptimal use of resources.

Several transformations are made to the raw data before the user is even involved in the process; this can introduce undocumented biases. A majority of statisticians would recommend obtaining and archiving the raw data and using an open script to make the relevant transformations, thus documenting intermediary steps that may lead to problems later in the analyses. This means that the complete pipeline from the raw data to the final figures will be reproducible. This entails using one of the systems for performing reproducible research (Gentleman and Temple Lang, 2007).

Too Many Choices

At each stage of the analysis of a complex data set, choices have to be made. Extreme value detection and thresholding are done at the start to identify outlier or saturation effects. Variables often differ in scale, even within the same measurement protocol. For instance, gene expression transformations have proved essential in microarray analyses (Allison et al., 2006). Variance stabilizing normalization (vsn) is use to combat heteroscedasticity and LOESS and LOWESS (locally weighted scatterplot smoothing [Cleveland, 1981]) are used to smooth out biases. Such transformations in the case of operational taxonomic unit (OTU) reads can include the replacement of the original data by robustified values, such as ranks or median polish replacements (Holmes et al., 2011).

Filtering out species that are very rare is crucial because these species may appear with inflated influence under some re-weighting schemes and it can be beneficial to delineate true presence from simple noise effects; thus, a relevant presence threshold has to be chosen. Sometimes data are divided into core and variable groups using a threshold for the standard deviations. After transformations and groupings, projection methods such as principal coordinate analysis (PCoA) or multidimensional scaling are popular. Again, an important choice has to be made: which is the appropriate distance to use? There are dozens of available choices. The Bray-Curtis is an distance that is a favorite among some ecologists. A recent study used Jensen-Shannon distances, resulting enterotypes in polarizing the data into clusters (Arumugam et al., 2011). Chi-square distance is a weighted Euclidean distance often preferred by statisticians, whereas others have proposed the Hellinger distance. In the case of simple presence/absence coding of the OTUs, it is better to use a distance based on the Jaccard index because it prioritizes co-occurrence of species as opposed to distances based on a correlation coefficient that put presence and absence on a similar standing. UniFrac distances, both weighted (Lozupone and Knight, 2005) and unweighted (Lozupone et al., 2007), have become popular because they integrate the known phylogenetic tree information into the distance computations. Double Principal Coordinate Analysis (DPCoA) (Eckburg et al., 2005) has the advantage of both integrating the patristic distances using the tree and providing biplot representations of species and samples.

However, each study has its own particularities owing to the specific data at hand. Users need to explore various choices; as we can see from the partial list above, a combinatorial explosion of potential analyses can occur. This is one of the motivations of keeping a complete reproducible record of the scripts involved in the analyses. This facilitates a quick revision of thresholds or distances, and resubmission of the script allows a complete reanalysis of the data in minutes. One can see from recent publications that current studies in microbiology (Human Microbiome Consortium et al., 2012a,b) are very far from being reproducible and are composed of a jigsaw of different steps performed by different players, each choosing their tuning parameters separately. Fortunately, statistical software technology has evolved at a similar pace as the measurement prowess and we have machines powerful enough to summarize and visualize masses of heterogeneous data. In statistics these software advances have come from an interesting self-organizing ecology of open-source projects that bridge many communities of data scientists.

We show in the next few sections how new robust statistical techniques are available as open-source software to address problems such as combining phylogenetic trees and abundance tables or networks. We also show how to combine different data tables, whether they were collected across locations or time. We can now represent evolving microbial communities or detect high-dimensional batch effects. As suggested above, the complexity of the data that now combine sequences, measurements, and phylogenetic trees requires a new approach to data management; we can no longer restrict ourselves to simple matricial formats and analyses.

Presentation of R

Specificity of R as a Community-Based Project

The ecology of statistical research has been changed by the interaction of several collaborative communities of data scientists. The R platform was developed by R. Gentleman and R. Ihaka (Ihaka and Gentleman, 1996) as an open-source program based on J. M. Chambers’s S language (Chambers, 1999, 2008). The project has developed organically to incorporate tools from different scientific communities incorporating several desirable features, of which we present only a few here. Each particular set of methods developed by an individual or small team is combined into a documented package containing examples, data, and functions, thus providing a high level of modularity and robustness. Stable versions are provided biannually and standard versioning control enables good reproducibility of studies done with R, even many years after publication of the methods. The object-oriented philosophy of the language allows for the easy management of complex data structures (such as those specifically defined for microbial studies shown in Figure A10-4).

A plot of the values of gPCA variables with a tree using R


Plotting the values of gPCA variables with a tree using R. SOURCE: Purdom (2010).

High-Quality Visualization Tools

Exploratory data analyses of complex data sets presuppose the availability of large numbers of graphical representations of the data.

The package ggplot2 (Wickham, 2009) allows the aesthetic construction of layered plots with multiple levels of information following the principles laid out in Wilkinson’s Grammar of Graphics (Wilkinson and Willis, 2005).38 Features such as α transparency allow the user to evaluate data distributions. Heat maps such as that in Figure A10-2 are useful alternatives to tables of numbers and can be complemented with tree information easily.

A graphical representation of a heatmap of abundance reads and a Volcano plot for expressions with transparency


(A) Heat map of abundance reads. (B) Volcano plot for expressions with transparency. SOURCE: Adapted from Caporaso et al. (2011).

Plotting trees together with supplementary information is possible in several different R packages; here we present some examples.

We created the tree graphic represented in Figure A10-3 using phyloseq, whose graphics are based on ggplot2. Enrichment of the tree was provided by using coloring for the sample-type variable, and also labeling using a shape for the different Families. The symbols are enlarged as the number of individuals increases.

An OTU reference tree


Global patterns global data mapped onto the OTU reference tree. SOURCE: Adapted from Caporaso et al. (2011).

Figure A10-4 is drawn from the paper by Purdom (2010) and shows generalized principal component scores together with the phylogenetic tree information for material phylotypes drawn from Sanger sequences (Eckburg et al., 2005).

Computational Features

As well as providing flexible high-quality graphics, R provides a computational platform that solves many of the challenges we listed in the introduction.

Reproducible research Today there is a movement to ensure statistical analyses follow the same standards of reproducibility as the laboratory sciences. Biologists publish the primers they used to enable other teams to reproduce their results, and statisticians are building platforms to publish all the code used to analyze and visualize the data, so that another researcher could take the same raw data through all the preprocessing steps and produce the same final figures and tables as the original analysts. This is useful for sharing work both within and between laboratories. From within the R environment, there are several ways to track and reproduce analyses. The standard built-in program history keeps an analytic diary of commands. More sophisticated are the Sweave and knitr functions, which allow the analyst to reproduce a whole technical report of the statistical analyses with the graphics and tables without having to handle individual files. A new platform, Rpubs,39 now enables easy web diffusion of all data analytic scripts. Such approaches are essential as we move forward to extremely large data sets so complex that multiple teams have to collaborate together to reach results in a timely manner. Data are often accumulated using different protocols, normalization procedures, outlier filtering functions, or threshold choices. Currently there are more than 40 distances to choose between when summarizing similarities between samples or OTUs. A perfect example of such a patchwork appears in the results of the Human Microbiome Project’s recent publications (Human Microbiome Consortium et al., 2012a,b) where the multiplication of platforms and protocols shows many levels of heterogeneity. The problem with these sequential pipelines is that if the first preprocessing team chooses a noise model and threshold that look optimal in its dimensions of interest and a next team then chooses optimal thresholds for its purpose, then the overall choice may not be a global optimum, in the same way that when trying to find two tuning parameters as an optimum on a two-dimensional surface it is not necessarily sufficient to optimize in one dimension and then in the second dimension. At least one should be able to iterate in such a procedure, preferably using both dimensions at the same time.

High-power computing As the data have become increasingly memory intensive, new parallelization tools for computer-intensive steps are necessary. For instance, foreach is used in the implementation of the Fast UniFrac distance in phyloseq in a way that is transparent from the user’s point of view.

Both standard and cutting-edge solutions All standard statistical methods are available in R, and most recent statistical or machine learning methods are published using R. For instance, packages such as RandomForests allow users to try out ensemble supervised learning methods based on learning trees, and the package caret helps with interpretation of the output. There are a large number of robust and sparse methods that make interpretation much easier than standard projection methods as they decrease the number of variables incorporated in the final models, resulting in parsimonious models. The availability and documentation of the packages make it possible to compare 20 methods on a new type of data in less than a week.

Data input and cleaning R has a high-quality subset of packages known as Bioconductor tailored to the multiple formats of modern bioinformatics (Sequences, Short Reads, RNA-seq, QIIME, HTS, MG-RAST, microarrays, mass spectrometry data, whole genomes, methylation, flow cytometry, microRNAi). A large number of alternative methods for denoising, normalizing, and standardization are available.

Simulation Many different random number generators are available for nonuniform distributions, bootstrapping, and Gibbs sample generation of posterior distributions. These simulations prove to be essential when validating models and doing nonparametric tests.

To illustrate these features, we explore a standard pipeline and mention some of the more useful tools. Teams working in biotechnology have realized that using R can circumvent long waits to access the data due to machine-specific output formats handled easily by Bioconductor. Researchers are encouraged to provide an example data set as soon as one is available so that the appropriate format readers can be found.

The package phyloseq directly takes the output from clustering and annotating programs such as qiime, mothur, RDP, PANGEA, or MG-RAST and combines it with the reference phylogenetic tree and OTU read abundances. The information about the samples is then combined into an object of special heterogeneous class (Figure A10-5).

A diagram showing a multicomponent class typical to R


Multicomponent class typical to R.

Usually the first step will be to normalize or transform the data using robust methods such as winterization, rank transformation, and elimination of very sparse data through filtering. An effective data checking mechanism is to do a simple principal component analysis (PCA) and project batch information as a supplementary variable, as can be seen in Figure A10-6. These are data from an experiment to detect differences between irritable bowel syndrome (IBS) and normal rat microbiota using the phylochip technology.40 Some exploratory verifications of homoscedasticity can reveal problems with the data. In this study, the first two components of a PCA done on the first two batches of data suggested a problem.

A graphical representation of a principal component analysis with batch information projected


Initial analysis with only two batches; on the right we see the addition of a third set of data.

Each color corresponds to a different batch of arrays. The ellipses are computed using the means, variances, and covariance of each group of points on both axes and are drawn with the center of the ellipse defined by the mean point. Its width and height are given by the within-group variance-covariance matrix, where the covariance sets the slope of the main axis of the ellipse. The first two batches (in black and red), although balanced with regard to IBS and healthy rats, show very different levels of variability and overall multivariate location. Batches were done on different days with different sets of arrays. This intermediary study of the data prompted us to make a third batch generated with the same arrays as batch 2 but the same experimental protocol as batch 1. The third group faithfully overlaps with batch 1, thus showing that the batch effect was due not to a difference in arrays but to the experimental protocol. This shows the utility of PCA in quality control.

After outlier detection has been performed, the data need to be normalized to make the columns comparable. Whereas in standard multivariate methods such as PCA, on diverse measurements the columns are all normalized to have the same variance. We recognize that there are often strong disparities in the quality of the various samples. It is common to see 10-fold, or even 100-fold, differences in total numbers of reads across samples.

On the Problems of Standardization and Rarefaction

Species occur in varying numbers in a community. Probabilists call this the multinomial distribution, and the question as to how many individuals one needs to obtain a full set of exemplars (i.e., one instance of every species) is known as the “coupon collectors” problem (Feller, 1957). It is well known that when some of the species are rare, the number of samples needs to be very large to detect all of them. So the exhaustive question is then relaxed somewhat to answer, for instance: “How many samples do we need to collect to obtain 75 percent of the species with probability 0.90?” Although there are formulas for computing the probability distribution of the number of species given the underlying multinomial, usual practice involves simulating the number of species as a function of the number of samples by subsampling without replacement from the data and drawing the rarefaction curve. In fact, the expected number of species in a sample of size is the mean of the probability distribution:


where the ρ’s are the probabilities of the multinomial classes (species relative frequencies).

However justified in the context of drawing rarefaction curves, subsampling for the purpose of “normalizing” sample size is never practiced in statistics, where unjustified elimination of some of the data is considered a sin. The standard practice when comparing counts is to use the chi-square distance, and for testing one can use the parametric chi-square test if the cell counts are mostly larger than 5, or, in the case of smaller counts as for large tables of OTUs, most of which are absent, one can use a nonparametric test using a Monte Carlo approximation (Diaconis and Sturmfels, 1998). It is to be recalled here that the practice of dividing each of the cell counts by the total of each sample, thus making each column sum to 1, will also result in data omission because the total number of reads in the column provides a measure of accuracy that is then lost in later computations. Thus, when using ratios, it is important to retain the denominator information in an extra row in the table. Depending on the context, other transformations, such as replacing the values by their ranks or taking logs, can be more appropriate. For further discussion and comparisons of normalization methods, see McMurdie and Holmes (2012).

Projection Methods

The standard technique after preparation of the normalized multivariate table of OTU abundances in each of the samples is to project the data in the two spaces, one defined by the linear combination of the samples and the other in the linear combination of the OTUs. This involves the eigenanalysis of a specific inner product matrix. Sometimes this can be done just by looking at distances between samples as in PCoA, for instance, on UniFrac distances (Costello et al., 2009).

On the other hand, if one does either a PCA or a simple correspondence analysis, corresponding to a generalized PCA (Holmes, 2006), choices have to be made and justified. One chooses whether and how to normalize the data and how many components to retain. The cutoff for the number of eigenvalues to retain separates the dimensions explained by true signal in the data as opposed to those due to random noise. These methods are all based on the search for new coordinate axes such that the projection of the weighted points onto those axes produces the largest possible inertia in the sense of physics (i.e., the largest weighted sum of squares Σpidi2). In ordinary PCA this is simply the variance that we maximize along the principal directions, which are chosen orthogonal. In correspondence analysis the inertia is proportional to the amount of the chi-square distance from independence explained by each axis (Figure A10-7).

A PCA flowchart, with choice levels highlighted


A PCA flowchart, with choice levels highlighted.

Recycling Methods Between Fields

One of the big advantages of using the R platform is the broad array of methods available from the R and Bioconductor libraries, which avoids having to reinvent the wheel for every new type of data that becomes available.

Hypergeometric test We take a concrete example of methodology transfer. There is a consensus among statisticians and data scientists involved in the analysis of microarrays that differential expression tests should be followed by gene set analysis (Allison et al., 2006) done using data such as the Gene Ontology or Gene Expression Network Analysis (Nacu et al., 2007). These analyses provide ways to make biological sense out of the laundry lists of differentially expressed genes that are provided by the multiple tests done between sample groups.

We can capitalize on this insight in the case of microbial studies. Instead of Gene Ontology, we have the taxonomic organization of the OTUs present in the study. After performing multiple tests we obtain a list of OTUs that come up as differentially represented in the two different samples as shown in Table A10-1.

TABLE A10-1. Differentially Represented OTUs.


Differentially Represented OTUs.

We can thus proceed to use a hypergeometric test to evaluate whether some phyla or families are overrepresented among the “significantly different” list of OTUs. A complete example of such a procedure is provided in a study of IBS and normal rats (Holmes et al., 2011), where we showed using this hypergeometric testing method that (1) taxa with a higher presence in the IBS group had significantly more Bacteriodetes; (2) there was an overrepresentation of Firmicutes in the healthy controls; (3) at the family level, the results showed that the families Oxalobacteraceae, Prevotellaceae, Burkholderiaceae, and Sphingobacteriaceae were significantly overrepresented in IBS; and, (4) conversely, the most significantly enriched family in control rats were Lachnospiraceae, including Ruminococcus sp., followed by Erysipelotrichaeceae and Clostridiaceae.

Once a list of significant families is available though this type of testing, it is useful to illustrate the ways the relevant components of a community assemble. Networks are a wonderful conceptual and visualization tool of which we show a few illustrations. Many other examples of transfer of statistical methods are now available, including uses of clustering, normalization, and supervised learning techniques. Another example of particular interest to researchers aiming for interpretable results is the use of networks to represent and model communities of co-occurring species or to show clumping of samples.

Ease in Network Manipulation and Plotting

We define an ecological community as a set of OTUs that occur often in common across our available samples. The data set in this example came from Arumugam et al. (2011) and contains samples measured with three different technologies on a number of human microbiomes for participants scattered across the world. We can identify samples with common microbial components as in Figure A10-8. Here each sample is a node and an edge appears between two samples if they have more than 70 percent shared OTUs. This is what we call a co-occurrence graph network. Other scientists have preferred to make such networks using the correlation between the vectors of presence/absence of the OTUs (Arumugam et al., 2011); however, that choice is often suboptimal for large numbers of OTUs because co-occurrence is a strong indicator of similarity in communities whereas co-absence is not. Thus, in this context the Jaccard index is preferable to a correlation coefficient. As an example of the ease with which R can handle creating and plotting such a graph, we have appended the code that was used to create Figure A10-8. We used the enterotype data to show the results of the makenetwork() function of the igraph package to create a graphical display of the “connectedness” of samples according to the profile of taxa that they share.

An illustration of a network of enterotype groups among samples as defined by patients having 70% shared OTUs


Network of enterotype groups among samples as defined by patients having 70 percent shared OTUs.

ig<-make_network(enterotype, max.dist=0.3)
plot_network(ig, enterotype, color=”SeqTech”,

Figures A10-8 and A10-9 show features both of the sample network, putting edges connecting samples that have similar OTU compositions, and of the dual network, showing the OTUs that seem to occur together. Often it is useful to represent data with a graph of relationships between metabolites or genes that we would like to use along with our simple numerical measurements.

An illustration of a network of enterotype taxa co-occurring in at least 85% of the samples


Network of enterotype taxa co-occurring in at least 85 percent of the samples.

All the Data

A systems approach leads to structured high-dimensional data We need a system that provides good data integration. We start by understanding what happens when we have multiple variables on the same samples. We call this the multivariate context. It has become standard to use multivariate analyses of many variables to ensure that the associations between the variables are accounted for. In some sense this is the statistician’s version of a “systems-based” approach. The simplest measurement of such associations between the variables taken two by two is called the covariance of the two variables. If we standardize this, we obtain the correlation. Multivariate analyses consider simultaneously all the variables measured in a set of samples. Multitable analyses involve the generalization of this idea to the simultaneous analyses of many tables of data that share either row dimensions or column dimensions. Again, it allows the data scientist to pursue a systems-based approach rather than slicing the data into simple arrays of homogeneous variables; we use all the data together. Let us illustrate these principles with a real example. In an ongoing study with David Relman, Les Dethfelsen, Angela Marcobal, and Justin Sonnenburg, we have accumulated measurements made on 4 patients and a total of 40 samples taken over time.

A first set of OTU abundance columns was collected using 454 technology passed through the qiime pipeline (Dethlefsen and Relman, 2011). Using the mass spectroscopy metabolic feature for the positive- and negative-ion states (Marcobal et al., 2012) has provided us two sets of metabolite tables for the same samples. Finally, we also have short-read data, giving us DNA measurements for 20,000 genes, which were then grouped into KEGG categories.

Thus, we have three different tables of variables measured on one set of samples. The tables share their row dimensions and differ in their column dimensions. In that context, we can compute the co-inertia of the tables (i.e., the amount of variability associated between the tables). This is done by computing the RV coefficient between tables A and B, given in the case of two tables with the same row dimensions by the following formula (Robert and Escoufier, 1976):


where Tr denotes the sum of the diagonal elements.

ig<-make_network(enterotype, type=”species”, max.dist=0.3)
plot_network(ig, enterotype)

Taxa Read counts (4 patients taking cipro: two time courses) : .

Mass-Spec Positive and Negative ion Mass Spec features and their intensities: .

RNA-seq Metagenomic data on genes : .

Here is the resulting RV table of the four array types:

                                     Taxa      Kegg      MassSpec+ MassSpec−
Taxa                             1            0.565      0.561                            0.670
Kegg                             0.565    1              0.686                            0.644
MassSpec+   0.561     0.686      1                                   0.568
MassSpec−   0.670     0.644      0.568                            1

We can ask the same questions about the significance of these “table correlations” as one does with two variable correlations. It can be tested using simple permutation tests where the sample rows are dissociated between tables to create a null distribution to which we compare the values observed. Such a test is implemented in the ade4 package in R. Here, we have a perfect example of the modern approach to testing which forgoes distributional assumptions in favor of computer-intensive randomization procedures. To test the correlation between two variables, we could compute the original value ρ0 on the observed data and then proceed to shuffle the order of one of the vectors, recomputing the correlations between these new shuffled vectors. This would provide us with a complete null distribution of the correlation statistic. Just such an approach is also valid for table correlations. We can think of this as just making each of the tables (each with n rows and p columns) into a vector by taking the first column and adding the second column underneath it until we obtain a vector of length np. We then compute the correlation of the two vectorized tables, call this RV0, and start a shuffling iteration that we repeat 1,000 times. For each shuffle, we compute the correlation between the shuffled vectors. These 1,000 values provide us with a null distribution showing what the sampling distribution would look like if the tables were unrelated. Comparing our original RV0 value to these new shuffled values provides the p value. All this is implemented in several R packages; below we show the output from ade4. This is a procedure similar to the well-known test invented by H. Daniels, commonly known as the Mantel test.

For instance, comparing the taxa table and negative mass spectrometry features between the samples and using the permutation test of RV correlation with 999 Monte Carlo simulations gives a significant p value.

rv12 = RV.rtestpcat$tab, pcamsn$tab, nrepet = 999)
[1] 0.001

Symmetric Two-Table Analysis

When simply trying to visually compare two tables of the same sites we can study the transformation necessary to make one table similar to another; this is often called a Procrustes rotation. In matrix notation, it means we need to find the rotation of the Xn×p points that makes it the most similar to the configuration of the Yn×q points. This occurs when we maximize

RV(Y, XA).

This is often called the Procrustes rotation, and it is found by finding the singular value decomposition of X′Y = UV and the A = UV′. For a detailed review of Procustean co-inertia analysis see Dray et al. (2003) (Figure A10-10).

A large scatterplot of Procrustes and Co-Inertia Analysis of Cipro patients seen in their compromise-common projection


Procrustes and co-inertia analysis of Cipro patients seen in their compromise common projection (the large scatterplot) and the taxa and mass spec feature projections on the left of the plot.

Assymetric Two-Table Analysis

Redundancy analysis and instrumental variables We saw how we can use the RV coefficient to find the level of correlation between two sets of data; however, it is often the case as in regression that the status of one of the tables is considered a response, whereas the other table may be considered explanatory. Explaining one variable by others is well known as multivariate linear regression. We try to find the linear combination of variables that maximizes the following:


In that framework, we find the projection of Y onto the space spanned by X by computing

Px(y) = X(XtX)−1Xty.

Principal Component Analysis (PCA) with respect to instrumental variables (PCAIV) was a technique developed by Rao (1964) to find the best set of coefficients in the multivariate regression setting where the response is multivariate, given by a matrix Y.

Taking the example above, the metabolic features may be taken as a multivariate response and the gene expression measurements the explanatory table. In this context the RV coefficient generalizes ordinary multivariate regression. We can look for the few principal components, known as instrumental variables (Rao, 1964), that explain the response table the best. These can be calculated by first taking the projection of Y onto the X matrix:

Px(Y) = X(XtX)−1XY.

Then, we do the PCA of PX(Y) to find a low-dimensional combination of them. This method is often used in ecology and psychology, where it is known as redundant data analysis (Van Den Wollenberg, 1977).

Things can get a little more complicated when, instead of wanting to predict or explain a table of continuous variables by an X composed also of continuous variables, we want to explain a contingency table of read counts of OTUs, present in a set of samples, by an explanatory matrix of continuous measurements X made on the same set of samples. To extend the method, the notion of variance is generalized to that of inertia of a set of weighted points:


In the case of contingency tables where the row sums and column sums are different, the overall inertia becomes proportional to the chi-square statistic.

Replacing the variance-covariance in the PCAIV computations above by the weighted chi-square gives a favorite method in ecology known as canonical correspondence analysis (CCA) (ter Braak, 1985).

Testing Factors in Canonical Correspondence Analysis

We can extend analysis of variance (ANOVA)-type methods to the decomposition of variability into parts explained by a factor, again using inertia or the chi-square. This is the strength of methods such as CCA and the analysis of distances (Anderson, 2001) provided by functions such as adonis in the vegan (Oksanen et al., 2012) package in R. The standard test statistic for ANOVA is

Test statistic:average sum of squares explained by factoraverage of residual sum of squares,

which is extended to the proportion of weighted sums of squared distances explained by a factor of interest. We can even separate out batch effects in exactly the same way they are removed in standard ANOVA.

Here is a small example on a series of measurements made on a set of mice for which we wanted to remove a cage effect. The actual factor of interest in the study was a shedding effect that we wanted to test after having adjusted for the cage effect.

res= cca(t(tcmall)  ~
                                                        log(salmv + 1) + Condition(as.factor(cage)))
> anova(res)
Permutation test for cca under reduced model
Model: cca(formula= t(tcmall)  ~
                                                                    log(salmv + 1) + Condition(as.factor(cage)))
                         Df             Chisq            F             N.Perm         Pr(>F)
Model               1               0.2017          8.65        199                0.005**
Residual           177           4.1289
Signif. Codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ‘ 1

Best Practices in Data Analyses in Microbiology

To summarize some of the points made in this chapter, we give a short list of best practices to guide the practitioner through the statistical maze of available choices.

Careful Records

Always keep the raw data as well as a complete transcript or diary of the transformations and analyses performed. This is possible through packages such as knitr (Xie, 2012) and Sweave (Leisch, 2002) in R. This ensures that the analysis will be reproducible later if further data become available, but it will also enable a consistent record for future collaborators. Most importantly, reviewers may be interested in knowing how robust the obtained results are to certain choices (thresholds for filters, functions chosen to rescale the data such as log or square roots, down-weighting of outliers, etc). If a script is available, the user can easily check the results for robustness. Exemplary demonstration of the usefulness of such an approach can be found in the illustration of two articles on multitable analyses (see Dray et al., 2004; Thioulouse, 2011).

Keep All the Data Together

The metadata describing the covariates on the samples collected—dates, batches, and procedures—should be made available in a repository for later perusal. This can be essential because at the start of a study some biases or differences may not seem important but show up later as more data give more sensitive results.

Raw data and the transformation scripts are preferable for publication to transformed data as they also provide a way for the reader to reproduce the results in the paper as described. For instance, researchers sometimes take ratios of OTU reads to the total for each sample, thus providing a table of percentages. However, this results in a loss of information that easily jeopardizes the results because the precision with which each ratio is known is dependent on that ratio’s original denominator.

Noise and Outlier Elimination, Data Imputation

In the presence of noise and outliers it can be important to eliminate certain samples or OTUs. This can be done as long as it is documented as above in a diary or script. On the contrary, it may be necessary to impute a few missing points by taking local averages of the k nearest neighbors, for instance, or through other standard data imputation methods (Little and Rubin, 1987), of which there are many choices. If the script used is documented, the reader may replicate the analysis with a slightly different method and see the consequences on the final result. If the data seem overly noisy, it can be beneficial to use robust methods such as nonmetric multidimensional scaling (NMDS) or to replace the data by their ranks.

When Using Ordination Methods, Always Plot the Eigenvalues

It is essential in multivariate projection methods such as PCA, PCoA, discriminant analysis (LDA), CCA, canonical correlation analysis (CCorA), or redundancy analysis (PCAIV or RDA) to show a plot of the successive eigenvalues. Reporting the percentage of inertia or variance explained by the first two components is not sufficient, because instabilities occur when two similar eigenvalues are split. This can easily occur, for instance, if the researcher has decided to retain enough of the eigenvalues to cover 80 percent of the variation of the data and the first five eigenvalues are proportional to 69, 12, 11, 5, and 3. Having a hard threshold of 80 percent will result in the separation of the two eigenvalues 12 and 11, which are probably not statistically different. In that case, the components are not stable but the whole plane defined by the second and third components is. Packages such as ade4 (Chessel et al., 2004) systematically provide the scree plot in all output display functions for projection methods based on eigendecompositions.

Enhance Interpretation Using Redundant and Contingent Information as Well as Regularization

Sanity checks include the plotting of extra information available about factors such as batches, plots, or laboratories. This should be plotted using techniques such as the projections made in Figure A10-10. Known gradients or groupings should be checked on the plots to verify consistency.

Many researchers find the output from simple least-squares projection methods arduous to interpret; with linear combinations of large numbers of genes or OTUs, model selection can be difficult. Recent progress in regularization techniques that incorporate penalties for more complex models can be useful in trying to select more parsimonious models. Such methods include sparse PCA as implemented in the elasticnet package (Zou and Hastie, 2008) or regularized canonical correlation analysis available in the package cca (González et al., 2008) and a sparse version of PCA and CCorA (pma; see Witten et al., 2009).

Always Look at Your Data, All the Data

A complete analysis always separates data into pieces eventually. However, the process has to be completely documented and the effects of outliers and specific groups documented before the data are decomposed. It is only safe to separate out parts of the data by slicing into homogeneous tables after having documented the dependency data using quantitative coefficients. Contiguous information, such as graphs linking related or interacting species or genes, and spatial maps of sampling locations, should always be incorporated into the data. This is important because samples with a high degree of dependency provide less information and standard statistical coefficients should be reweighed to account for this.

Separate Out a Subset of the Data for Confirmatory Purposes

As detailed above, the number of choices made throughout an analysis rapidly multiply. Thus, researchers have the opportunity to thoroughly explore their data to generate a meaningful narrative. However, this often results in overfitting and multiple testing problems. One solution to this problem is to use a separate data set for confirmation of the analysis. The ideal situation is that when a large data set is available, a random subsample is set aside for separate diagnostic purposes. An alternative is to design a separate confirmatory experiment of the hypothesis generated by the analyses. Controlling for false discovery rates after many data transformations and choices of endpoints is a difficult task. Some procedures have been developed and are available in R (Pollard et al., 2010) and mutoss for instance. However, there is usually a high level of dependency in the variables and tables, making standard procedures based on independence assumptions inadequate. For hierarchical dependencies, as in the case of taxonomic ranks, strucSSI provides an implementation of procedures recommended by Benjamini and Yekutieli (2001).

State of the Art in Modern Statistics

Computational power has enabled statisticians to develop many new robust, nonparametric, and sparse methods. Robustness (Huber, 1981) is a statistical quality that ensures that the results of an analysis are not changed if a few of the observations are removed. This is the case for methods with a high breakdown point like the median, whereas the mean of n observations has a low breakdown point at 1n. 41 Robustness to outliers is important for several reasons: analyses based on least squares can be easily perturbed by one or two outliers and multivariate data rarely follow multivariate normal distributions. Many implementations of robust methods are available (robustbase or FRB in R). Rank-based transformations generally ensure results that are robust in the statistical sense. A simple example of a robust transformation is a Winsorization, which changes the top 1 percent of values and the lower 1 percent and makes them equal to the 99th and 1st percentile of the distribution. This is a one-line transformation in R:

  • library(psych)
  • wm=winsor(tab1,trim=0.05)

A consensus has been reached in the study of microarrays (Allison et al., 2006) that rank-based transformations such as rma can help stabilize the variances of the variables. The same is true in ecology where robust PCoA such as NMDS has become popular.

Smoothers for multivariate data such as LOWESS have increased in popularity since their success in reducing bias in the analysis of microarrays; in fact, nonparametric methods in regression and supervised learning have become ubiquitous. Statisticians group together all methods that enable fitting data without prior specification of a small finite number of parameters under the general “nonparametric” umbrella. Methods that prioritize sparseness were developed both to improve interpretability of the results and to counter the difficulties of having a much larger number of potential variables than the number of samples (the large-p, smaller-n paradigm). Sparse PCA (Zou et al., 2006) and elasticnet or glmnet (Zou and Hastie, 2008) provide useful solutions to today’s challenging data containing as many measurements as 30,000 genes or 15,000 OTU frequencies measured on many fewer samples (generally in the hundreds). These sparse methods provide principal components constrained to have very few nonzero coefficients, thus ensuring a high degree of interpretability.


We have made the case that, for science to move forward in step with current levels of data and software availability, communities of users should be able to transfer technologies easily through open-source platforms and reproducible research. A collaborative platform such as R has the advantage of an object-oriented structure that allows good levels of data integration, such as those shown in packages such as edgeR or phyloseq. High-quality graphics make it easy for the user to keep in touch with the data and make publication-level plots through layering functions such as those available in ggplot2. The level of complexity in threshold and tuning parameter choices requires simulation and Monte Carlo tools for creating realistic noise models and sensitivity tests, all of which are currently available.

New standards of publication in statistical work are percolating across the disciplines and many articles are published with all the source code and data in such a way that any user could redo the analyses in the article from beginning to end. A wonderful example of such a case study showing how to do multiple table analyses on paired ecological tables can be found in the web supplement to Thioulouse (2011).42 The goal of these collaborative approaches is to empower a new generation of biologists/data analysts to keep in touch with their data, whether there be megabytes or terabytes of it. Although the new statistical environments such as R can be intimidating, the Complete R Archive Network site43 contains more than 100 books and documentation sources. There are also many tutorials available online so the interested user could get started with the help of the large user community. We hope this review will encourage users to adventure further into the exploration of their data. We have added a short list of R packages in the appendix as a guide to some interesting methods.


We would like to thank our collaborators David Relman, Alfred Spormann, and Les Dethfelsen, as well as our students and postdoctoral fellows Alden Timme, Katie Shelef, Yana Hoy, Miling Yan, Diana Proctor, Audrie Lin, Julia Fukuyama, Kris Sankaran, Sam Pimentel, and Angela Marcobal.

None of this would have been possible without the R community, in particular the participants in the Bioconductor project, the developers of the ade4 and vegan packages, and Emmanuel Paradis for all his work on the ape package.

Funding from NIH/NIGMS R01, NSF-VIGRE, and NSF-DMS.


We give here some of the most useful packages for the analysis of microbial communities from different areas of statistics. Statistical analysis of the microbiome data uses methods from ecology, clinical biostatistics, spatial statistics, bioinformatics, and phylogenetics. Here is an incomplete list of some packages we have found invaluable for this enterprise:

  • High throughput sequencing, RNA-Seq DEseq, shortread, phyloseq, genefilter, edgeR.
  • Microarray analyses Biobase, genefilter, vsn, marray, limma, multtest, made4.
  • Annotation biomaRt, AnnotationDbi, GOstats, KEGG, PFAM, xcms.
  • Decompositions aov, anova, lm, medianpolish, permanova.
  • Ecology and phylogenetic analyses vegan, ade4, ape, phangorn, distory, picante.
  • Plotting ggplot2, lattice, phyloseq, heatmap2, neatmap, rgl.
  • Multivariate analyses ade4, cluster, knn, som, kohonen, sparcl, flexclust.
  • Supervised learning tree, e1071, randomForest, ipred, lasso2,lars, elasticnet, kernlab.
  • Spatial data spatstat, spdep, spBayes.
  • Sparse and robust methods elasticnet.
  • Nonparametric testing bootstrap, coin, Hmisc, asypow, npmc.
  • Bayesian methods arm, bayesclust, lda, DPpackage, topicmodels, coda, rjags, MasterBayes, lmm.


  • Allison D, Cui X, Page G, Sabripour M. Microarray data analysis: From disarray to consolidation and consensus. Nature Reviews Genetics. 2006;7(1):55–65. [PubMed: 16369572]
  • Anderson M. A new method for non-parametric multivariate analysis of variance. Austral Ecology. 2001;26(1):32–46.
  • Arumugam M, Raes J, Pelletier E, Le Paslier D, Yamada T, Mende D, Fernandes G, Tap J, Bruls T, Batto J, Bertalan M, Borruel N, Casellas F, Fernandez L, Gautier L, Hansen T, Hattori M, Hayashi T, Kleerebezem M, Kurokawa K, Leclerc M, Levenez F, Manichanh C, Nielsen HB, Nielsen T, Pons N, Poulain J, Qin J, Sicheritz-Ponten T, Tims S, Torrents D, Ugarte E, Zoetendal EG, Wang J, Guarner F, Pedersen O, deVos WM, Brunak S, Doré J, Antolin M, Artiquenave F, Blottiere HM, Almeida M, Brechot C, Cara C, Chervaux C, Cultrone A, Delorme C, Denariaz G, Dervyn R, Foestner KU, Friss C, van de Guchte M, Guedon E, Haimet F, Huber W, van Hylckama-Vlieg J, Jamet A, Juste C, Kaci G, Knol J, Lakhdari O, Lavec S, LeRoux K, Maquin E, Mérieux A, Melo Minardi R, M’rini C, Muller J, Oozeer R, Parkhill J, Renault P, Rescigno M, Sanchez N, Sunagawa S, Torrjon A, Turner K, Vandemuelebrouck G, Varela E, Winogradsky Y, Zeller G, Weissenbach J, Ehrlich SD, Bork P. MetaHIT Consortium. Enterotypes of the human gut microbiome. Nature. 2011;473(7346):174–180. [PMC free article: PMC3728647] [PubMed: 21508958]
  • Benjamini Y, Yekutieli D. The control of the false discovery rate in multiple testing under dependency. Annals of Statistics. 2001;29(4):1165–1188.
  • Caporaso J, Lauber C, Walters W, Berg-Lyons D, Lozupone C, Turnbaugh P, Fierer N, Knight R. Global patterns of 16s rRNA diversity at a depth of millions of sequences per sample. Proceedings of the National Academy of Sciences of the United States of America. 2011;108(Suppl 1):4516. [PMC free article: PMC3063599] [PubMed: 20534432]
  • Chambers J. Computing with data: Concepts and challenges. American Statistician. 1999;53(1):73–84.
  • Chessel D, Dufour A, Thioulouse J. The ade4 package—I: One-table methods. R News. 2004;4(1):5–10.
  • Cleveland W. LOWESS: A program for smoothing scatterplots by robust locally weighted regression. The American Statistician. 1981;35(1):54.
  • Costello E, Lauber C, Hamady M, Fierer N, Gordon J, Knight R. Bacterial community variation in human body habitats across space and time. Science. 2009;326(5960):1694–1697. [PMC free article: PMC3602444] [PubMed: 19892944]
  • Dethlefsen L, Relman D. Incomplete recovery and individualized responses of the human distal gut microbiota to repeated antibiotic perturbation. Proceedings of the National Academy of Sciences of the United States of America. 2011;108(Suppl 1):4554. [PMC free article: PMC3063582] [PubMed: 20847294]
  • Diaconis P, Sturmfels B. Algebraic algorithms for sampling from conditional distributions. Annals of Statistics. 1998;26(1):363–397.
  • Dray S, Chessel D, Thioulouse J. Co-inertia analysis and the linking of ecological data tables. Ecology. 2003;84(11):3078–3089.
  • Dray S, Dufour A, Chessel D. The ade4 package—II: Two-table and k-table methods. 2004. [accessed September 27, 2012]. Supplementary website. http://pbil​.univ-lyon1​.fr/JTHome/ref/ade4-Rnews.pdf.
  • Eckburg PB, Bik EM, Bernstein CN, Purdom E, Dethlefsen L, Sargent M, Gill SR, Nelson KE, Relman DA. Diversity of the human intestinal microbial flora. Science. 2005;308(5728):1635–1638. [PMC free article: PMC1395357] [PubMed: 15831718]
  • Feller W. Introduction to probability theory and its applications. I. New York: Wiley; 1957.
  • Gentleman R, Temple Lang D. Statistical analyses and reproducible research. Journal of Computational and Graphical Statistics. 2007;16(1):1–23.
  • González I, Déjean S, Martin P, Baccini A. CCA: An R package to extend canonical correlation analysis. Journal of Statistical Software. 2008;23:1–14.
  • Holmes S. Multivariate analysis: The French way. In: Nolan D, Speed TP, editors. Probability and statistics: Essays in honor of David A Freedman. Vol. 56. Beachwood, OH: IMS; 2006. p. 142. (IMS Lecture Notes–Monograph Series). Pacific Symposium on Biocomputing.
  • Holmes S, Alekseyenko A, Timme A, Nelson T, Pasricha P, Spormann A. Visualization and statistical comparisons of microbial communities using R packages on Phylochip data. Pacific Symposium on Biocomputing. 2011;2011:142–153. [PMC free article: PMC3347925] [PubMed: 21121042]
  • Huber PJ. Robust statistic. New York: Wiley; 1981.
  • Human Microbiome Consortium et al. A framework for human microbiome research. Nature. 2012;486:215–221. [PMC free article: PMC3377744] [PubMed: 22699610]
  • Human Microbiome Consortium et al. Structure, function and diversity of human microbiome in an adult reference population. Nature. 2012;486:207–214. [PMC free article: PMC3564958] [PubMed: 22699609]
  • Ihaka R, Gentleman R. R: A language for data analysis and graphics. Journal of Computational and Graphical Statistics. 1996;5(3):299–314.
  • Leisch F. Comstat 2002—proceedings in computational statistics. Heidelberg: Springer Verlag; 2002. Sweave. Dynamic generation of statistical reports using literate data analysis; pp. 575–580.
  • Little R, Rubin D. Statistical analysis with missing data. Vol. 4. New York: Wiley; 1987.
  • Lozupone C, Knight R. UniFrac: A new phylogenetic method for comparing microbial communities. Applied and Environmental Microbiology. 2005;71(12):8228. [PMC free article: PMC1317376] [PubMed: 16332807]
  • Lozupone C, Hamady M, Kelley S, Knight R. Quantitative and qualitative {beta} diversity measures lead to different insights into factors that structure microbial communities. Applied and Environmental Microbiology. 2007;73(5):1576. [PMC free article: PMC1828774] [PubMed: 17220268]
  • Marcobal A, Holmes S, Dethlefsen L, Sonnenburg J, Relman D. Combined analyses of metabolic profiles, metagenomic analyses for patients perturbed by antibiotic. 2012. Unpublished.
  • McMurdie P, Holmes S. Discussion of relative merits of normalization methods for OUT tables. 2012. (in preparation)
  • Nacu S, Critchley-Thorne R, Lee P, Holmes S. Gene expression network analysis and applications to immunology. Bioinformatics. 2007;23(7):850–858. [PubMed: 17267429]
  • Oksanen J, Blanchet FG, Kindt R, Legendre P, Minchin PR, O’Hara RB, Simpson GL, Solymos P, Stevens MHH, Wagner H. vegan: Community ecology package. 2012. R package version 2.0.3.
  • Pollard KS, Gilbert HN, Ge Y, Taylor S, Dudoit S. multtest: Resampling-based multiple hypothesis testing. 2010. R package version 2.4.0.
  • Purdom E. Analysis of a data matrix and a graph: Metagenomic data and the phylogenetic tree. Annals of Applied Statistics. 2010;5(4):2326–2358.
  • Rao CR. The use and interpretation of principal component analysis in applied research. Sankhya A. 1964;26:329–359.
  • ter Braak C. Correspondence analysis of incidence and abundance data. Biometrics. 1985;41:859–873.
  • Thioulouse J. Simultaneous analysis of a sequence of paired ecological tables: A comparison of several methods. Annals of Applied Statistics. 2011;5(4):2300–2325.
  • Van Den Wollenberg A. Redundancy analysis an alternative for canonical correlation analysis. Psychometrika. 1977;42(2):207–219.
  • Wickham H. ggplot2: Elegant graphics for data analysis. New York: Springer; 2009.
  • Wilkinson L, Wills G. The grammar of graphics. New York: Springer-Verlag; 2005.
  • Wilson KH, Wilson WJ, Radosevich JL, DeSantis TZ, Viswanathan VS, Kuczmarski TA, Andersen GL. High-density microarray of small-subunit ribosomal DNA probes. Applied and Environmental Microbiology. 2002;68(5):2535–2541. [PMC free article: PMC127575] [PubMed: 11976131]
  • Witten D, Tibshirani R, Hastie T. A penalized matrix decomposition, with applications to sparse principal components and canonical correlation analysis. Biostatistics. 2009;10(3):515–534. [PMC free article: PMC2697346] [PubMed: 19377034]
  • Xie Y. knitr: A general-purpose package for dynamic report generation in R. 2012. R package version 0.4.1.
  • Zou H, Hastie T. elasticnet: Elastic-net for sparse estimation and sparse PCA. 2008. R package version, 1-0.
  • Zou H, Hastie T, Tibshirani R. Sparse principal component analysis. Journal of Computational and Graphical Statistics. 2006;15(2):265–286.





44 Microdish BV, Utrecht, the Netherlands.

Introduction to Swarming Bacteria

Microorganisms have many ways to move around. Dispersal may be by passive means or by using other organisms as vectors. Additionally, many species of bacteria can migrate; this implies both an active means of propulsion and sufficient sensitivity to external conditions to trigger and/or guide motility. A major form of microbial existence is on surfaces, and under favorable conditions it is perfectly feasible for microbes to travel many centimeters over a hydrated surface. This movement can occur by a number of means, including twitching and gliding motility and other methods, such as spreading, i.e., the elongation of chains of nonmotile cells. Flagellar rotation can also drive collectives of bacteria to migrate over a surface; this is referred to as swarming, by analogy with multicellular organisms such as insects (Harshey, 2003; Henrichsen, 1972; Kearns, 2010). Swarming may be aided by differentiation into specialized swarmer cells; these are often both elongated (compared to vegetative cells of the same species) and hyperflagellated. Some species (e.g., Vibrio spp.) may produce flagella specifically adapted to travel within a thin film of liquid on a surface rather than planktonic motility (swimming in larger volumes of fluid). Swarming bacteria can sense both the presence of a suitable surface and whether they have achieved a suitable cell density for collective action to be effective (often involving “quorum sensing”). Swarming may be viewed as an alternative lifestyle choice compared to sessile colonization of a surface, i.e., a biofilm. Swarming can be associated with the secretion of molecules that facilitate progress. Such extracellular agents include surfactants, lubricants, polymers, antimicrobials, and/or agents of communication (including the molecules involved in quorum sensing). Some secreted compounds (but certainly not all) may be regarded as “public” or “common” goods, resources shared by a microbial community, facilitating the development of powerful models for the study of cooperation and competition within microbial communities. Pseudomonas aeruginosa is one such system: rhamnolipids are critical common goods required for swarming. In the case of P. aeruginosa the metabolic control of rhamnolipid secretion is important in limiting the emergence of noncontributory “cheater” mutants (Xavier et al., 2011). Swarming by the hyperflagellated, elongated cells of Proteus spp. has long been recognized by medical microbiologists as both a fascinating object of study and a confounding factor in clinical assays, the latter by contaminating agar plates during diagnostics assays (Dienes, 1946, 1947). Under permissive conditions, Proteus mirabilis and Proteus vulgaris swarm outward from the inoculation point on an agar plate, alternating zones of swarming with zones of consolidation (the latter as vegetative cells). This produces a characteristic pattern of concentric rings resembling the cross section of a tree. When two identical strains of P. mirabilis meet, the swarming masses of cells interpenetrate without conflict. When non-identical strains meet, there is an incompatibility reaction—a border region containing apparently damaged cells—and the strains fail to mix (Figure A11-1). This effect implies a mechanism of distinguishing self from non-self. The border region between different strains has been known for more than 70 years as a Dienes line (Dienes, 1946, 1947) and is used clinically in strain typing. However, it is only recently that the genetic basis of this form of territoriality and individuality has started to emerge (Gibbs et al., 2008, 2011).

A photograph of swarming by Proteus mirabilis


Swarming by Proteus mirabilis. Swarming on an agar plate, showing a border region between two different strains, one labeled with green fluorescent protein (GFP) and one with dsRED fluorescent protein. The strain expressing the GFP is forming rounded, (more...)

The examples provided by Proteus and Pseudomonas and other swarming bacteria have shown that powerful model systems illuminating fundamental issues in biology may be derived from their study (and indeed by other collective systems of surface motility such as those used in predation and cooperation/defection in the Gram-positive bacterium Myxococcus xanthus) (Velicer et al., 1998). Swarming bacteria have also attracted attention from disciplines outside biology. In particular, mathematicians and related investigators have been attracted by the complex patterns and the self-organizing nature of bacterial swarms (Ben Jacob et al., 1994, 2000; Komoto et al., 2003; Matsuyama and Matsushita, 1992; Ohgiwari et al., 1992; Thar and Kühl, 2005). Issues such as collective decision making, including “swarm intelligence,” where a mass of independent but interacting entities without a fixed hierarchy may make superior decisions to individuals, can be studied (Halloy et al., 2007). In addition, the plastic nature of pattern-forming organisms is a form of temporary multicellularity (Shapiro, 1998), one that appears to be adaptive both to internal constraints (including cell-to-cell contacts and communication) and to external factors, i.e., responding to the environment. This, in turn, raises the question as to what swarming bacteria reveal about the origins and development of more permanent forms of multicellularity. Further questions triggered by swarming bacteria relate to the ecological value and costs of their behavior. For example, what happens when a swarming mass of bacteria meets other microorganisms? Are there issues that relate to the interactions of rapidly moving masses of microorganisms within a short time frame that differ from more static, more slowly developing interactions? How are new niches colonized when it is likely they already contain indigenous microflora? What is the contribution of swarming to dispersal of microorganisms in the soil or infections? This article focuses on the swarming bacterium Paenibacillus vortex and attempts to indicate what this swarming bacterium may contribute to the field.

Paenibacillus vortex: A Pattern-Forming Swarmer

P. vortex is a Gram-positive, sporulating bacterium; its cells are distinctly curved rods, generally 3 microns or longer. P. vortex is a highly effective swarmer and can migrate across an agar plate under a wide range of conditions (Figures A11-2 and A11-3). The genome has been sequenced and contains a set of genes commonly found in bacteria from the rhizosphere, i.e., a very high proportion of two-component systems,45 attack- and defense-related genes (including putative antibiotic synthesis pathways and antibiotic-resistance mechanisms), and many membrane-associated transporters and other (primarily nutrient uptake) related functions (Galperin, 2005; Sirota-Madi et al., 2010). The closest relative (based on 16S rRNA sequence) is Paenibacillus glucanolyticus. The only mechanism of motility apparent from the genome, with this conclusion supported by scanning electron microscopy of swarming cells, uses peritrichous flagella (Ingham and Ben-Jacob, 2008). When swarming, P. vortex has very strong tendencies to chase its own tail, or at least swarm in curved trajectories, and is able to form elegantly rotating colonies on higher percentage agars (> 1.5% w/v). Imaging of swarming colonies reveals lubricant trails (e.g., Figure A11-2A), whereas electron microscopy of elongated, stressed swarmer cells also shows that quite extreme curvature can be compatible with these rotating microcolonies (Figure A10-2B). P. vortex is a very versatile swarmer; for example, it can swarm on very high-percentage agars (up to 2.2% w/v) and quite low-percentage agars (0.4% w/v), which is a wider range than many other bacteria including Gram-negative rods. P. vortex swarms but it does not always colonize the entire plate, instead forming patterns (Figure A10-3) with regions of agar lacking any bacteria. The mechanisms by which this occurs are largely unknown but there appear to be repulsive forces, possibly mediated by signalling molecules or lubricants (Ben-Jacob et al., 2000; Ingham and Ben-Jacob, 2008). These complicated colonies are highly dependent on the environmental conditions (e.g., humidity, nutrients, stress).

A six-panel microscopy of P. vortex cells and microcolonies


Microscopy of P. vortex cells and microcolonies. (A) Light microscopy of a swarming colony moving clockwise in circles (v) on agar (a). (B) Scanning electron microscopy of a colony responding to the antibiotic mitomycin C and showing coiling of the elongated (more...)

A three-panel microscopy showing pattern formation by swarming P. vortex


Pattern formation by swarming P. vortex. Three examples showing regions of P. vortex colonies (7.4 × 5.5 cm area, scale bar indicates 1 cm, growing from a central inoculation point), the bacteria which leave considerable areas of the agar plate (more...)

Transport of Aspergillus fumigatus Conidia by Paenibacillus vortex

Many Paenibacillus spp. interact with fungi. Examples of cooperation of Paeinibacilli spp. with fungi and also competition or antagonism are known (Budi et al., 2000; Dijksterhuis et al., 1999; Hildebrandt et al., 2006). Given this, and the relative ease of distinguishing fungal spores (conidia) from bacterial cells in terms of imaging and growth, we have explored the involvement of P. vortex with fungi, particularly using Aspergillus fumigatus. A. fumigatus is a well-known and widespread filamentous fungus with both sexual and asexual sporulation cycles, the latter producing conidia 3 to 4 μm across. A. fumigatus is an opportunistic pathogen, and an allergen, with a complex genome of 29.4 Mb (Nierman et al., 2005). However, this fungus is not motile (not by mycelia or by spores) and dispersal is largely passive. If the conidia are inoculated in the center of a large agar plate the vegetative mycelia can grow outward over a period of days. Eventually, under permissive conditions, such a colony sporulates (Campbell, 1971). The fungus can reach the edge of the Petri dish either by outgrowth or, in some situations, the spores can become airborne and can inoculate the agar closer to the edge than the parental mycelia. On many growth media, including reduced-strength Mueller–Hinton medium, the fungus takes from 9 to 12 days to reach the edge of a 14 cm–diameter agar plate from an inoculation point in the center. The question posed was whether swarming bacteria can accelerate this process. A very simple experiment was done to test this proposition (Figure A11-4 and Ingham et al., 2011). Actively swarming P. vortex were coinoculated with purified conidia of A. fumigatus at the center of a 14 cm plate containing reduced-strength Mueller–Hinton medium. The bacteria became distributed across the plate by swarming. After a few days, fungal colonies spread out from the central point across the plate, scattered and with fewer colonies toward the edge of the plate. If such plates were examined at an earlier stage, when the fungus had hardly grown and the microcolonies were only visible by microscopy, the fungus could be seen growing inside the mass of the P. vortex. Growth from the P. vortex swarm out into the unoccupied agar was sometimes observed, but the fungus was never seen germinating free of the bacterium. It was concluded that the swarming bacteria were transporting the spores (conidia).

A photo showing transport of A. fumigatus conidia by P. vortex


Transport of A. fumigatus conidia by P. vortex. Swarming P. vortex was co-inoculated with conidia of A. fumigatus in the center of a 14 cm nutrient agar plate. The picture was taken after 72 h; only the fungal colonies are visible, although the plate (more...)

Analysis of movies of the motion of the conidia and of the bacteria surrounding the conidia has proved revealing. Imaging is typically performed using a total magnification of from ×40 to ×100 over a period of a few minutes. Time-lapse movies (Movies 1 to 3) show conidial transport. 46 These movies can be found online at the National Academy of Sciences.47 This analysis has been digitized and compared to a series of models that superimpose “virtual conidia” and predicts their motion using different scenarios based upon assumptions about how the conidia and the bacteria are associating with each other. The model that fits is that the conidia are moving at the same rate as the surrounding microorganisms, and the structure of the microorganisms around the conidia is relatively stable, at least over a period of a few minutes. There appears to be a physical connection, or at least stable linkage, between the two microorganisms. Electron microscopy confirms this; the conidia appear to be trapped or entangled within the flagella of the bacteria.

A photo showing transport of conidia by P. vortex


Transport of conidia by P. vortex. Conidia transport can be seen to be rapid when viewed by microscopy. Left to right: images taken over a period of 2 min showing a bifurcating swarming mass of P. vortex (PV) transporting aggregates of A fumigatus conidia. (more...)

What Is the Significance of Microbial Cargo Transport Outside the Laboratory?

We know that microorganisms can generate mechanical force to drive their own motion and move objects (Figure A11-6 and Darnton et al., 2004). Microengineers tells us that there are a number of ingenious devices that use microorganisms to move wheels (e.g., Hiratsuka et al., 2006). It is possible to conjugate cargo beads to motile algae, then get them to swim and deliver the beads somewhere else within a microfluidic system (Weibel et al., 2005). In this case the movement of the algae was directed by light. Clearly, eukaryotic microorganisms can move prokaryotes (Brock et al., 2011; Rowbotham, 1980). But motile bacteria cooperatively carrying larger eukaryotic microorganisms appear to be a novel observation. A 30 cm translocation of A. fumigatus spores has been achieved using P. vortex (Ingham et al., 2011). This suggests that microbial cargo transport could happen in a way that does not require highly controlled (e.g., microengineered) environment but possibly could work in nature.

An illustration showing examples of motile microorganisms moving other objects within microengineered environments


Examples of motile microorganisms moving other objects within microengineered environments. (A) A cell of the flagellated alga Chlamydomonas reinhardtii with a covalently bound bead (smaller sphere) is shown. Light-directed transport is possible within (more...)

When the bacteria move fungal spores, the spores are largely viable. During the germination process of A. fumigatus without bacteria the conidia swell over a period of a few hours then mycelial outgrowth occurs after 6–7 h on reduced-strength Mueller–Hinton agar. This all happens also within the P. vortex swarm and at the same rate. Therefore, the conidia are active and able to sense their environment inside the swarm, as they are if they are just inoculated in the same medium on an agar plate without bacteria, inasmuch as certain environmental conditions (moisture, nutrients, temperature) triggered outgrowth. Conidia can move about 1 cm per hour in P. vortex swarms. It is known that fungal spores are adapted for some forms of dispersal, including via the wind, water, or insects (Nagarajan and Singh, 1990; Raguso and Roy, 1998; Van Leeuwen et al., 2010). Not all conidia of equal size are transported with equal efficiency (Ingham et al., 2011), but we currently do not understand the relevance of this in nature.

One possible use for conidial transport, from the fungal perspective, could be to escape adverse environments. In an experiment designed to test this proposition, P. vortex and A. fumigatus conidia were co-inoculated in the center of an agar plate in the presence of an antifungal drug, voriconazole (Chryssanthou et al., 2008; Verweij et al., 2007). Some conidia were “rescued” by the bacterium and deposited in a region of the plate lacking the antifungal agent. P. vortex can also move across conidia from another location, pick them up, and deposit them farther away, although not with the same efficiency as if the two organisms are co-inoculated (Ingham et al., 2011).

Many swarming bacteria are refractory to antibiotics, a phenomenon that appears linked to both high cell density and motility (Butler et al., 2010). P. vortex can swarm across regions containing high concentrations of antibiotics (for example using a row of antibiotic-containing tabs) that are sufficient to inhibit the growth of static bacteria. The bacteria that can do this are a subpopulation of the P. vortex population, referred to as explorers or pioneers. It is possible to track the paths of individual swarms as they cross this relatively hostile region. When they get to the other side, the bacteria can start to grow normally. The antibiotic refractory motile colonies are not mutants. They respond in the same way if given the same challenge again. This phenomenon is some form of phenotypic resistance to antibiotics; the relationship between these and antibiotic-refractory subpopulations (potentially including persisters) is unclear. If the P. vortex is transporting cargo conidia some of the swarms will approach the antibiotic tab more effectively than others but the region will be colonized less well than if the antibiotic is absent. A few days later, the fungal spores, which have been hiding out inside the Paenibacillus mass, germinate, and, in this niche, they can compete very effectively with the bacterium, so they take over.

Cargo transport may take place in soil. A simple way to think about the soil is as a series of microniches, many separated by gaps (Figure A11-7). The gaps, microniches in their own right, may have water or air in them. Filamentous fungi can cross air gaps quite effectively by extending growing mycelia. Many bacteria swim well or are otherwise capable of passively crossing a water gap but have problems crossing air gaps (Nazir et al., 2010). The air gap can be simulated on an agar plate (Figure A11-8). One experimental approach is to inoculate the conidia and the bacterium very close to this gap. At least in some cases, if you wait long enough, the growth of the mycelia will cross the gap and inoculate the agar on the other side with the bacteria. The mechanism is not fully understood. Swarming in P. vortex is inhibited by p-nitrophenyl glycerol (Ingham et al., 2008), which is also an antiswarming agent for Proteus spp. (Liaw et al., 2000), and this agent reduces (but does not eliminate) the mycelial-mediated transfer of P. vortex to the far side. This suggests a role for swarming and/or flagella in the process. In a second approach, if the two organisms are co-inoculated farther away from the air gap, then the only way of getting to the target zone is for the P. vortex and A. fumigatus to cooperate. P. vortex move the fungal spores to the gap. Then the mycelia extend across the air gap and this allows the bacteria to reach the far side (Figure A11-8). So in this specific case, the two organisms mutually facilitate their dispersal. There is precedent for fungi assisting bacteria to spread: coswarming bacteria (different motile species that nevertheless aid each other’s spread) are known to be aided in spreading in a model system by fungal mycelia. This idea has been termed “fungal highways” by one group (Warmink and van Elsas, 2009). Fungal highways may have practical applications as well as ecological importance, for example in giving bacteria involved in bioremediation greater access to contaminated soils (Furuno et al., 2009). In the latter case, a hydrocarbon-degrading strain, Pseudomonas putida NAH7, has been shown to travel along pseudomycelia of the filamentous fungus Pythium ultimum (Figure A11-9). Chemotaxis is implicated in this process (Furuno et al., 2009); a non-chemotactic derivative of P. putida was dispersed less effectively, suggesting that directed migration in response to chemical cues is a significant factor. Some of these data, including the current work, can be explained in several ways, including flagella-mediated attachment of cells to growing mycelia (Sen et al., 1996) and/or the mycelia allowing motility over terrain that is otherwise hard to cross. There is another interesting hint from the literature that connects swarming with fungi: the fungal toxin citrinin induces motility in Paenibacillus polymyxa (Park et al., 2008), but the significance of this observation is not clear.

A simple diagram of soil composed of manifold microniches separated by barriers that include air gaps and aqueous environments


Simple diagram of soil composed of manifold microniches separated by barriers that include air gaps and aqueous environments. In many cases mycelial microorganisms, such as fungi, or airborne spores can traverse air gaps that are likely to be difficult (more...)

A five-panel illustration showing dispersal of motile bacteria facilitated by fungal mycelia


Dispersal of motile bacteria facilitated by fungal mycelia. (A) Simulating a 0.5-mm air gap using blocks of agar, posing a significant barrier for P. vortex to cross. (B) One possible solution. Air gap (0.5 mm across) created in a nutrient agar plate (more...)

A micrograph showing an example of dispersal of motile bacteria facilitated by fungal pseudomycelia mycelia


Example of dispersal of motile bacteria facilitated by fungal pseudomycelia mycelia. Confocal laser scanning microscopy of Pythium ultimum hyphae with associated Pseudomonas putida PpG7 bacteria on glass. The image series is shown as maximum intensity (more...)

Transporting cargo can be detrimental to the swarming bacteria. Figure A11-10 shows an experiment in which P. vortex is transporting spores of the actinomycete bacterium Streptomyces coelicolor, an antibiotic transporter. The antibiotics produced by the S. coelicolor inhibit the growth of the P. vortex. So it’s not always a good thing to let other microorganisms into your swarm.

An image showing the dispersal of Streptomyces coelicolor spores facilitated by swarming P. vortex on a 14cm agar plate


Dispersal of Streptomyces coelicolor spores facilitated by swarming P. vortex. Image of a 14 cm–diameter plate containing reduced-strength Mueller–Hinton agar co-inoculated with swarming P. vortex and spores of S. coelicolor at position (more...)

Is there any specificity in the transporting process? To a limited extent, the answer is yes. Different conidia of about the same size can be moved to different degrees. If the surface of the conidia is stripped off, with either a strong protease or SDS, then the transport properties of the treated conidia differ from those of intact ones, suggesting that proteins displayed on the conidial surface matter. However, it is also clear that non-living objects such as microspheres can be transported. There is only limited information on the converse question: do all swarming bacteria transport objects? To date, Proteus mirabilis appears to be a poor transporter of micron-scale objects: attempts to show transport even over a few millimeters have failed. Cautiously, we may speculate that the territorial Proteus swarms tend toward monoculture in other ways than the Dienes line exclusion phenomenon, and they tend not to be very welcoming of other species.

P. vortex itself is a sporulating organism (Sirota-Madi et al., 2010), although spores are not formed during the course of the experiments described in this chapter. However, if exogenous spores purified from a starvation medium are loaded into a swarm these particles are also transported. Given the propensity of P. vortex to swarm into dangerous situations and get itself into trouble, it might make some sense, rather than producing the spores when things get tough, to have them already present in the swarming group.

To summarize the work so far, P. vortex is surprisingly good at moving micron-scale objects around. We have shown that it can move A. fumigatus conidia. It may be able to move conidia either to niches where airborne dispersal of this organism cannot reach, or possibly there are some biases where the Paenibacillus tends to end up in a particular environment, and if it brings along other organisms, then perhaps they benefit. There is also the possibility for rescue from an unfavorable environment, with spores taken to a more favorable location (the converse may also be true although may not be as important). Within the soil a moving colony (swarming) of P. vortex may well be a mixed ecosystem, with the potential to drop things off and have things jump on and thereby redistribute organisms over a centimeter scale in the soil. As noted, P. vortex can produce extended and intricately patterned colonies on agar. In some cases these colonies can be kept motile for several days with a continual traffic system of motile and continuously redistributing bacteria. If a rifampicin-resistant mutant of P. vortex is fed into one part of such a colony of a rifampicin-sensitive strain, rifampicin-resistant progeny can be recovered in other parts of the colony. The degree of connectivity, implied by a P. vortex colony acting as an extended transport network, suggests the possibility that a bacterial logistics system can be created on a 10 cm scale. It will be interesting to explore this and ask what happens when a nitrogen source is in a different location to the carbon source—i.e., is there any logistical benefit to this form of colonial life? Microbial logistics systems, the capability of extended colonies to optimise nutrients over a significant area, are known to be advantageous in fungi and slime moulds (Physarum polycephalum). Quite complicated problems in resource utilization can be solved by slime moulds forming dendritic networks (Nakagaki et al., 2000). There is also interest in using microbial networks in the design of miniaturized devices (Kaehr and Shear, 2009), and adaptive pattern-forming bacteria with the ability to transport micron-scale components may make a contribution.

Future Prospects

One line of future research is to create positive selection systems so that the cargo organism pretty much has no choice but to be carried and to associate with the P. vortex. There is also a strong need for more realistic soil models and some understanding of how swarming and transport works in the soil. Given the amazing recent developments in low-cost and robust sensors and tracking systems, this should be achievable, and maybe we can then understand more about the spread of microorganisms in natural situations. Screening for cargo organisms from the soil—pulling organisms out of the soil and seeing how they partner with P. vortex—may also be fruitful. A spore display library with a wide range of polypeptides expressed on the surfaces of spores could be used to define preferred cargoes. Because P. vortex is currently not amenable to genetic analysis, it may be necessary to exploit the genetics of a partner organism to try to understand transport at a deeper level.


Motile microorganisms can exert physical force and move objects on the micron scale. Specifically, the swarming bacterium Paenibacillus vortex can collectively transport conidia of the non-motile fungus Aspergillus fumigatus. The significance and ecological impact of microbial transport is unknown although there are scenarios in which the benefit appears reciprocal. However, this suggests a new dispersal mechanism for non-motile microorganisms with possible benefits for both transporting and cargo organisms. More broadly, the role and importance of swarming and other mass migrations undertaken by bacteria within the environment are poorly understood. One lesson from recent years is that while swarming bacteria have usually been studied in monoculture, such organisms reveal their most interesting characteristics when in conflict or cooperation with other microorganisms. Swarmers, at least in some cases, may be best thought of as a traveling ecosystem.


Thanks to Eshel Ben Jacob and his group for collaborating over many years. Thanks to Joëlle Dupont, Rolf Geisen, Peter Schneeberger, and Amir Sharon for strains of fungi and advice on fungal culture. Dennis Claessen and Lubbert Dijkhuizen contributed Streptomyces coelicolor strains and Adriaan van Aelst assisted with electron microscopy. The financial support of HEALTH-F3-2011-282004 (EVOTAR) from the European Union is acknowledged.


  • Ben-Jacob E, Schochet O, Tenenbaum A, Cohen I, Czirók A, Vicsek T. Generic modelling of cooperative growth patterns in bacterial colonies. Nature. 1994;368:46–49. [PubMed: 8107881]
  • Ben-Jacob E, Cohen I, Golding I, Gutnick DL, Tcherpakov M, Helbing D, Ron IG. Bacterial cooperative organization under antibiotic stress. Physica A. 2000;282:247–282.
  • Brock DA, Douglas TE, Queller DC, Strassmann JE. Primitive agriculture in a social amoeba. Nature. 2011;469:393–398. [PubMed: 21248849]
  • Budding AE, Ingham CJ, Bitter W, Vandenbroucke-Grauls CM, Schneeberger PM. The Dienes phenomenon: Competition and territoriality in swarming Proteus mirabilis. Journal of Bacteriology. 2009;191:3892–3900. [PMC free article: PMC2698385] [PubMed: 19251852]
  • Budi SW, van Tuinen D, Martinotti G, Gianinazzi S. Isolation from the Sorghum bicolor mycorrhizosphere of a bacterium compatible with arbuscular mycorrhiza development and antagonistic towards soil-borne fungal pathogens. Applied and Environmental Microbiology. 2000;65:5148–5150. [PMC free article: PMC91693] [PubMed: 10543835]
  • Butler MT, Wang Q, Harshey RM. Cell density and mobility protect swarming bacteria against antibiotics. Proceedings of the National Academy of Sciences of the United States of America. 2010;107:3776–3781. [PMC free article: PMC2840483] [PubMed: 20133590]
  • Campbell CK. Fine structure and physiology of conidial germination in Aspergillus fumigatus. Transactions of the British Mycological Society. 1971;57:393–402.
  • Chryssanthou E, Loebig A, Sjölin J. Post-antifungal effect of amphotericin B and voriconazole against germinated Aspergillus fumigatus conidia. Journal of Antimicrobial Chemotherapy. 2008;61:1309–1311. [PubMed: 18367461]
  • Darnton N, Turner L, Breuer K, Berg HC. Moving fluid with bacterial carpet. Biophysical Journal. 2004;86:1863–1870. [PMC free article: PMC1304020] [PubMed: 14990512]
  • Di Leonardo R, Angelani L, Dell’arciprete D, Ruocco G, Lebba V, Schippa S, Conte MP, Mecarini F, De Angelis F, Di Fabrizio E. Proceedings of the National Academy of Sciences of the United States of America. 2010;107:9541–9545. [PMC free article: PMC2906854] [PubMed: 20457936]
  • Dienes L. Reproductive processes in Proteus cultures. Proceedings of the Society for Experimental Biology and Medicine. 1946;63:265–270. [PubMed: 20277719]
  • Dienes L. Further observations on the reproduction of bacilli from large bodies in Proteus cultures. Proceedings of the Society for Experimental Biology and Medicine. 1947;66:97–98. [PubMed: 20270685]
  • Dijksterhuis J, Sanders M, Gorris LG, Smid EJ. Antibiosis plays a role in the context of direct interaction during antagonism of Paenibacillus polymyxa towards Fusarium oxysporum. Journal of Applied Microbiology. 1999;86:13–21. [PubMed: 10030009]
  • Furuno S, Pazolt K, Rabe C, Neu TR, Harms H, Wick LY. Fungal mycelia allow chemotactic dispersal of polycyclic aromatic hydrocarbon-degrading bacteria in water-unsaturated systems. Environmental Microbiology. 2009;12:1391–1398. [PubMed: 19691501]
  • Galperin MY. A census of membrane-bound and intracellular signal transduction proteins in bacteria: Bacterial IQ, extroverts and introverts. BMC Microbiology. 2005;5:35–54. [PMC free article: PMC1183210] [PubMed: 15955239]
  • Gibbs KA, Urbanowski ML, Greenberg EP. Genetic determinants of self-identity and social recognition in bacteria. Science. 2008;321:256–259. [PMC free article: PMC2567286] [PubMed: 18621670]
  • Gibbs KA, Wenren LM, Greenberg EP. Identity gene expression in Proteus mirabilis. Journal of Bacteriology. 2011;193(13):3286–3292. [PMC free article: PMC3133265] [PubMed: 21551301]
  • Halloy J, Sempo G, Caprari G, Rivault C, Asadpour M, Tâche F, Saïd I, Durie V, Canonge S, Amé JM, Detrain C, Correll N, Martinoli A, Mondada F, Siegwart R, Deneubourg JL. Social integration of robots into groups of cockroaches to control self-organized choices. Science. 2007;318:1155–1158. [PubMed: 18006751]
  • Harshey RM. Bacterial motility on surfaces: Many ways to a common goal. Annual Review of Microbiology. 2003;57:249–273. [PubMed: 14527279]
  • Henrichsen J. Bacterial surface translocation: A survey and classification. Bacteriological Reviews. 1972;36:478–505. [PMC free article: PMC408329] [PubMed: 4631369]
  • Hildebrandt U, Ouziad F, Marner FJ, Bothe H. The bacterium Paenibacillus validus stimulates growth of the arbuscular mycorrhizal fungus Glomus intraradices up to the formation of fertile spores. FEMS Microbiol Letters. 2006;254:258–267. [PubMed: 16445754]
  • Hiratsuka Y, Miyata M, Tada T, Uyeda TQP. A microrotary motor powered by bacteria. Proceedings of the National Academy of Sciences of the United States of America. 2006;103:13618–13623. [PMC free article: PMC1564248] [PubMed: 16950878]
  • Ingham CJ, Ben-Jacob E. Swarming and complex pattern formation in Paenibacillus vortex studied by imaging and tracking cells. BMC Microbiology. 2008;8:36. [PMC free article: PMC2268691] [PubMed: 18298829]
  • Ingham CJ, Kalisman O, Finkelstein A, Ben Jacob E. Mutually facilitated dispersal between the nonmotile fungus Aspergillus fumigatus and the swarming bacterium Paenibacillus vortex. Proceedings of the National Academy of Sciences of the United States of America. 2011;108:19731–19736. [PMC free article: PMC3241745] [PubMed: 22106274]
  • Kaehr B, Shear JB. High-throughput design of microfluidics based on directed bacterial motility. Lab on a Chip. 2009;9:2632–2637. [PubMed: 19704977]
  • Kearns DB. A field guide to bacterial swarming motility. Nature Reviews Microbiology. 2010;8(9):634–644. [PMC free article: PMC3135019] [PubMed: 20694026]
  • Komoto A, Hanaki K, Maenosono S, Wakano JY, Yamaguchi Y, Yamamoto K. Growth dynamics of Bacillus circulans colony. Journal of Theoretical Biology. 2003;225:91–97. [PubMed: 14559062]
  • Liaw SJ, Lai HC, Ho SW, Luh KT, Wang WB. Inhibition of virulence factor expression and swarming differentiation in Proteus mirabilis by p-nitrophenylglycerol. Journal of Medical Microbiology. 2000;49:725–731. [PubMed: 10933258]
  • Matsuyama T, Matsushita M. Self-similar colony morphogenesis by Gram-negative rods as the experimental model of fractal growth by a cell population. Applied and Environmental Microbiology. 1992;58:1227–1232. [PMC free article: PMC195579] [PubMed: 1599243]
  • Nagarajan S, Singh DV. Long-distance dispersion of rust pathogens. Annual Review of Phytopathology. 1990;28:139–153. [PubMed: 20540608]
  • Nakagaki T, Yamada H, Toth A. Maze-solving by an amoeboid organism. Nature. 2000;407:470. [PubMed: 11028990]
  • Nazir R, Warmink JA, Boersma H, van Elsas JD. Mechanisms that promote bacterial fitness in fungal-affected soil microhabitats. FEMS Microbiology Ecology. 2010;71:169–185. [PubMed: 20002182]
  • Nierman WC, Pain A, Anderson MJ, Wortman JR, Kim HS, Arroyo J, Berriman M, Abe K, Archer DB, Bermejo C, et al. Genomic sequence of the pathogenic and allergenic filamentous fungus Aspergillus fumigatus. Nature. 2005;438:1151–1156. [PubMed: 16372009]
  • Ohgiwari M, Mitsuge M, Tohey M. Morphological changes in growth of bacterial colony patterns. Journal of the Physical Society of Japan. 1992;61:816–822.
  • Park S-Y, Kim R, Ryu CM, Choi SK, Lee CH, Kim JG, Park SH. Citrinin, a mycotoxin from Penicillium citrinum, plays a role in inducing motility of Paenibacillus polymyxa. FEMS Microbiology Ecology. 2008;65:229–237. [PubMed: 18459968]
  • Raguso RA, Roy BA. ‘Floral’ scent production by Puccinia rust fungi that mimic flowers. Molecular Ecology. 1998;7:1127–1136. [PubMed: 9734071]
  • Rowbotham TJ. Preliminary report on the pathogenicity of Legionella pneumophila for freshwater and soil amoeba. Journal of Clinical Pathology. 1980;33:1179–1183. [PMC free article: PMC1146371] [PubMed: 7451664]
  • Sen R, Nurmiaho-Lassila EL, Haahtela K, Korhonen T. Specificity and mode of primary attachment of Pseudomonas fluorescens strains to the cell walls of ectomycorrhizal fungi. In: Azcon-Aguilar C, Barea JM, editors. 4th European Symposium on Mycorrhizae. Mycorrhizas in Integrated Systems from Genes to Plant Development. Granada, Spain: European Commission; 1996. pp. 661–664.
  • Shapiro J. Thinking about bacterial populations as multicellular organisms. Annual Review of Microbiology. 1998;52:81–104. [PubMed: 9891794]
  • Sirota-Madi A, Olender T, Helman Y, Ingham C, Brainis I, Roth D, Hagi E, Brodsky L, Leshkowitz D, Galatenko V, Nikolaev V, Mugasimangalam RC, Bransburg-Zabary S, Gutnick DL, Lancet D, Ben-Jacob E. Complete genome of the pattern forming Paenibacillus vortex reveals potential for thriving in complex environments. BMC Genomics. 2010;11:710. [PMC free article: PMC3012674] [PubMed: 21167037]
  • Thar R, Kühl M. Complex pattern formation of marine gradient bacteria explained by a simple computer model. FEMS Microbiology Letters. 2005;246:75–79. [PubMed: 15869964]
  • Van Leeuwen MR, Van Doorn TM, Golovina EA, Stark J, Dijksterhuis J. Water-and air-distributed conidia differ in sterol content and cytoplasmic microviscosity. Applied and Environmental Microbiology. 2010;76:366–369. [PMC free article: PMC2798642] [PubMed: 19897752]
  • Velicer GJ, Kroos L, Lenski RE. Loss of social behaviors by Myxococcus xanthus during evolution in an unstructured habitat. Proceedings of the National Academy of Sciences of the United States of America. 1998;95:12376–12380. [PMC free article: PMC22839] [PubMed: 9770494]
  • Verweij PE, Mellado E, Melchers WJ. Multiple-triazole-resistant aspergillosis. New England Journal of Medicine. 2007;356:1481–1483. [PubMed: 17409336]
  • Warmink JA, van Elsas JD. Migratory response of soil bacteria to Lyophyllum sp. strain Karsten in soil microcosms. Applied and Environmental Microbiology. 2009;75:2820–2830. [PMC free article: PMC2681705] [PubMed: 19286795]
  • Weibel DB, Garstecki P, Ryan D, DiLuzio WR, Mayer M, Seto JE, Whitesides GM. Microorganisms to move microscale loads. Proceedings of the National Academy of Sciences of the United States of America. 2005;102:11963–11967. [PMC free article: PMC1189341] [PubMed: 16103369]
  • Xavier JB, Kim W, Foster KR. A molecular mechanism that stabilizes cooperative secretions in Pseudomonas aeruginosa. Molecular Microbiology. 2011;79:166–179. [PMC free article: PMC3038674] [PubMed: 21166901]



48,49 and 48,50.


48 Environmental Science & Engineering, Mail code 100-23, California Institute of Technology, Pasadena, CA 91125.
49 Present address: Department of Geosciences, Guyot Hall, Princeton University, Princeton, NJ 08544.
50 Corresponding author: J. R. Leadbetter, Phone: (626) 395-4182, Fax: (626) 395-2940, Email: ude.hcetlac@rettebdaelj.


Termites and their gut microbes engage in fascinating dietary mutualisms. Less is known about how these complex symbioses have evolved after first emerging in an insect ancestor over 120 Mya. Here, we examined a key bacterial gene of the mutualism, formate dehydrogenase (fdhF), in 8 species of “higher” termite (Termitidae; the youngest, most biomass-abundant, and species-rich termite family). Patterns of fdhF diversity in the gut communities of higher termites contrasted strongly with patterns in less-derived insect relatives (wood-feeding “lower” termites and roaches). We observed phylogenetic evidence for (1) the sweeping loss of several clades of fdhF that may reflect extinctions of symbiotic protozoa and, importantly, bacteria dependent on them, in the last common ancestor of all higher termites and (2) a radiation of genes from the (possibly) single allele that survived. Sweeping gene loss also resulted in (3) the elimination of an entire clade of genes encoding selenium-independent enzymes from higher termite gut communities, perhaps reflecting behavioral or morphological innovations in higher termites that relaxed pre-existing environmental limitations of selenium (Se), a dietary trace element. Curiously, several higher termite gut communities may have subsequently re-encountered Se-limitation, (4) reinventing genes for Se-independent proteins via convergent evolution. Lastly, (5) the presence of a novel fdhF lineage within litter-feeding and subterranean higher (but not other) termites may indicate recent gene “invasion” events. These results imply that cascades of perturbation and adaptation by distinct evolutionary mechanisms have impacted the evolution of complex microbial communities in a highly successful lineage of insects.

Author Summary

Since patterns of relatedness between termite hosts are broadly mirrored by the relatedness of their symbiotic gut microbiota, co-evolution between hosts and gut symbionts is rightly considered as an important force shaping dietary mutualism since its inception over 120 Mya. Apart from lateral gene or symbiont transfer between termite gut communities (for which no evidence yet exists), there has been little discussion of alternative mechanisms impacting the evolution of mutualism. Here, we provide strong gene-based evidence for past environmental perturbations creating significant upheavals that continue to reverberate throughout the gut communities of species comprising a single termite lineage. We suggest that symbiont extinction events, sweeping gene losses, evolutionary radiations, relaxation and re-emergence of key nutritional pressures, convergent evolution of similar traits, and recent gene invasions have all shaped gene composition in the symbiotic gut microbial communities of higher termites, currently the most dominant and successful termite family on Earth.


Identifying factors associated with changing genetic diversity in natural microbial populations is crucial for understanding past and present ecology. Host-associated microbial populations have garnered much interest, as principles of evolution uncovered in the context of host-microbe interactions have wide-ranging applications (e.g. human health, animal development, agriculture) (Dethlefsen et al., 2007; McFall-Ngai, 2002; Van Wees et al., 2008). In particular, studies of animal-microbe mutualism have revealed that microbial symbionts exert an important selective force for evolution in their eukaryotic hosts (Van Wees et al., 2008). Equally intriguing are the pressures that symbiosis imparts on the evolution of the hosts’ microbial counterparts.

Many evolutionary studies on microbial symbiosis have explored the highly intimate mutualisms existing between insects and the microbial endosymbionts that live inside host insect cells (e.g., aphid–Buchnera, tsetse fly–Wigglesworthia) (Chen et al., 1999; Moran and Baumann, 2000; Moran and Telang, 1998). These nutritional symbioses are obligate and characterized by low species complexity (typically 1 symbiont species is present) (Bauman et al., 2006; Buchner, 1965; Wu et al., 2006). As such they have been useful model systems for identifying the major evolutionary consequences of symbiosis in the microbial partners of mutualism: cospeciation with the host, genome reduction, low genomic GC content, and accelerated sequence evolution (Mira et al., 2001; Moran, 2002; Moran et al., 2009; Shigenobu et al., 200; Wernegreen, 2002). However, such studies have limited potential for revealing the consequences of symbiosis between animal hosts and their extracellular symbionts, a category including gut tract microbes. As these symbioses can involve multiple microbial partners functioning as a “community bioreactor” to effect health or disease in their host, determining how interactions between different symbiont species have evolved is necessary to understanding host-symbiont synergisms.

The obligate nutritional mutualism occurring between wood-feeding termites and their specific and complex hindgut microbiota offers an enticing subject for studies of evolutionary themes in polymicrobial, extracellular symbiont communities. Remarkably, mutualism, which underpins the insect’s ability to access lignocellulose as a food source, predates the evolution of termites from their wood-feeding roach ancestors over 120 mya (Grimaldi and Engel, 2005). Similar to endosymbioses, this long-term association has been facilitated by symbiont transfer between host generations (Nalepa and Bandi, 2000), allowing co-evolution between hosts and the symbiont community. Co-speciation as an evolutionary theme has indeed emerged over the past decade (e.g. Berlanga et al., 2007; Brune and Stingl, 2006; Desai et al., 2010; Hongoh et al., 2005; Lo and Eggleton, 2011; Noda et al., 2007; Ohkuma et al., 2009) however factors like diet, gut anatomy, and geography are also potential determinants of gene diversity in gut communities. Compared to endosymbionts, gut symbionts are exposed to considerably more environmental influence as food and environmental microbes continually pass through the gut tract. Under certain conditions, these microbes may establish as new symbionts or may horizontally transfer their genes to pre-existing symbionts. Thus termites and their rich diversity of gut microbial partners, which can number over 300 bacterial phylotypes (Hongo et al., 2003; Yang et al., 2005), provide exceptional opportunities for studying factors affecting evolution in symbiont communities.

Comparing symbiont genes in a wide range of termite host species is one way to learn about factors influencing symbiont evolution. Previously, we applied this comparative approach to symbiont metabolic genes encoding hydrogenase-linked formate dehydrogenase (fdhF), important for CO2 reductive acetogenesis, a key mutualistic process in lignocellulose degradation (Breznak and Switzer, 1986; Odelson and Breznak, 1983). In all wood-feeding termites and roaches, CO2 reductive acetogenesis is the major means of recycling the energy rich H2 derived from wood-polysaccharide fermentations into acetate, the insect’s primary fuel source (Brauman et al., 1992; Pester and Brune, 2007). Study of fdhF in evolutionary less derived wood-feeding insects (the so-called “lower” termites and their wood-feeding roach relatives) implied that the trace element selenium shaped the gene content of gut symbionts since termites diversified from roaches ~150 Mya (Grimaldi and Engel, 2005). More broadly, results suggested that within related insects hosts that feature the same type of gut tract architecture, similar gut communities and diets, acetogenic symbionts as a metabolic subgroup of the greater gut community have maintained remarkably similar strategies to deal with changes in a key nutrient. But how have these symbionts (more precisely, how have their genes) responded to drastic changes in the termite gut, for example to major alterations in gut tract architecture and to restructuring of the gut community itself?

To address this question, a comparison of fdhF genes present in extant insects possessing gut communities representing the “before” and “after” snapshots of drastic change is needed. The gut communities of lower termites represent the “before” condition. Gut communities in higher termites, which form the most recently evolved termite lineage (Termitidae), represent the “after” snapshot (Eggleton, 2006). Higher termites are known for highly segmented gut tracts and their lack of symbiotic gut protozoa (Brugerolle and Radek, 2006), the primary sources of H2 within the guts of earlier evolved wood-feeding insects. Gut segmentation and the extinction of lignocellulose-fermenting protozoa are events that are thought to have occurred early during the emergence of this lineage, possibly in the late cretaceous (80 Mya) to early cenozoic (50 Mya) (Grimaldi and Engel, 2005). Since then, higher termites with their gut symbionts have found ways to access polysaccharides bound in forms other than wood (e.g., dry grass, leaf litter, organic compounds in soil) and have become the most abundant and diverse termite group on Earth.

Here we investigate whether the dramatic shifts in termite biology and microbiology marking the diversification of higher termites into their extant forms left impacts on the H2-consuming symbiont community. To accomplish this we analyzed the phylogeny of fdhF from the symbiotic gut communities of 8 higher termite species. The specimens belonged to Nasutiterminae and Termitinae subfamilies, the two most numerically abundant and species-rich subfamilies within Termitidae (generally recognized as comprising four subfamilies (Grimaldi and Engel, 2005; Eggleton et al., 1996; Engel et al., 2009; Kambhampati and Eggleton, 2000) and were sampled from rainforest, beach, and desert environments in Central America and the Southwestern United States. We compared higher termite fdhF sequences to each other and to previous data from less-derived wood-feeding insects. Our results indicate the H2-consuming bacterial community in higher termites experienced early evolutionary extinctions, possibly due to extinction of H2-producing protozoa, followed by evolutionary radiation, convergent evolution, and even invasion. The latter three outcomes may have been influenced by trace element bioavailability. Together, the data provide a clear example of an extinction propagating through a chain of microbial mutualism, emphasizing the connectivity of symbiosis involving complex extracellular microbial communities such as those in the termite gut.

Results and Discussion

To profile fdhF in the whole gut microbial communities of higher termites with different species affiliations (Figure A12-1), habitats and lifestyles, we constructed fdhF gene inventories (30-107 clones per species, total 684) from the whole gut tracts of six higher termite species from Costa Rica and two higher termite species from California (Table A12-1). The broader taxonomic affiliations for higher termite specimens are indicated in a schematic termite cladogram (Figure A12-2). Multiple genotypes (8 to 59) and phylotypes (4 to 12 operational taxonomic units, or OTUs, at 97% protein sequence similarity cutoff) were recovered from the guts of each termite sample (Table A12-2). This represents significantly more diversity than that discovered by metagenomic analysis of a wood-feeding higher termite’s gut microbiota (Warnecke et al., 2007). In total, 62 novel FDHH OTUs were documented in higher termites. To stringently estimate sampling completeness we compared the number of observed OTUs to that predicted by the 95% higher confidence interval for mean Chao1 (Table A12-2), calculated using EstimateS (Colwell, 2009). On average, 4.9 ± 4.5 (1 stdev) more OTUs were missing per inventory. However, by comparing the observed number of OTUs with mean Chao1 values, on average less than 1 OTU remained undiscovered per inventory.

A phylogeny of termites and related roaches


Mitochondrial cytochrome oxidase II (COII) phylogeny of termites and related roaches. Family names and other descriptions are located on the right side of the tree. Only two of four subfamilies (Macrotermitinae, Apicotermitinae, Nasutitermitinae, and (more...)

TABLE A12-1. Characteristics of Higher Termites Examined in this Study.


Characteristics of Higher Termites Examined in this Study.

A schematic cladogram of major termite families and higher termite subfamilies showing major events in gut habitat evolution


Schematic cladogram of major termite families and higher termite subfamilies showing major events in gut habitat evolution. Key events are A: hindgut fermentation of lignocellulose; B: loss of Blattabacterium fat body endosymbionts; C: loss of mutualistic (more...)

TABLE A12-2. fdhF Inventories Constructed in this Study.


fdhF Inventories Constructed in this Study.

Broad Scale Diversity of Higher Termite fdhF

Our previous studies of fdhF (Matson et al., 2010; Zhang et al., 2011) have shown that hydrogenase-linked formate dehydrogenase enzymes (FDHH, EC1.2.1.2, encoded by enteric γ-Proteobacteria, Spirochaetes, Firmicutes) are widespread in the guts of lower termites and wood-roaches. Almost all fdhF recovered from these insects grouped most closely with genes from the termite gut acetogenic spirochete isolate, Treponema primitia, earning that phylogenetic clade the name “Gut Spirochete Group.” To determine the relationship between higher termite fdhF and previously published sequences (and to guide detailed phylogenetic analyses), we constructed a phylogenetic “guide” tree (Figure A12-3) based on 542 aligned amino acids in FDHH, the catalytic subunit responsible for CO2 reduction/formate oxidation in formate hydrogen lyase complexes (Zinoni et al., 1986).

A protein phylogeny of hydrogenase-linked formate dehydrogenases


Protein phylogeny of hydrogenase-linked formate dehydrogenases (FDHH). Sequences from the gut microbial communities of higher termites, lower termites, wood-feeding roaches, and pure microbial cultures form four major clades (A, B, C, D). Numbers in parentheses (more...)

Based on 3 tree construction methods (maximum likelihood, parsimony, and distance), higher termite sequences consistently clustered into 4 major clades (labeled A–D in Figure A12-3). Sequences from higher termites with diets likely consisting of lignocellulosic substrates such as wood, palm, and dried-grass predominantly (74–100%, Table A12-3) grouped into the Gut Spirochete Group (clade A, Figure A12-3). Similar to other environmental sequences in the Gut Spirochete Group, higher termite sequences (37 phylotypes) most probably belong to uncultured acetogenic spirochetes. The presence of a diagnostic amino acid character uniquely shared amongst Gut spirochete group sequences supported this phylogenetic inference (Zhang et al., 2011). Interestingly, sequences from litter-feeding and subterranean higher termites (8–51%, Table A12-3) formed a novel cluster, designated as the “AGR group” for Amitermes-Gnathamitermes-Rhynchotermes (clade B, 17 phylotypes, Figure A12-3). Relatively few sequences (0–18%, Table A12-3) affiliated with clade C in Figure A12-3 (5 phylotypes), a group defined by proteobacterial sequences, and accordingly named the “Proteobacteria Group.” Clade D in Figure A12-3 (3 phylotypes) was the least represented in inventories (0–7%, Table A12-3) and (like the AGR group) lacked pure culture representatives. This clade was named the “UnHT Group” for unclassified higher termite.

TABLE A12-3. Distribution of Higher Termite Inventory Sequences amongst Four Major FDHH Clades (Figure A12-3).


Distribution of Higher Termite Inventory Sequences amongst Four Major FDHH Clades (Figure A12-3).

Genetic Extinction and Evolutionary Radiation Within Higher Termite Gut Communities

As higher termite sequences predominantly clustered into the Gut Spirochete Group, we performed a phylogenetic analysis using 3 different treeing methods that focused on Gut Spirochete Group sequences (Figure A12-4). Unexpectedly, every higher termite sequence in this major clade group fell into a single subclade, the “Higher Termite Spirochete” clade. This stood in striking contrast to our prior observations that fdhF from lower termites and wood-roaches tend to form multiple, deeply branching but interspersed clades throughout the Gut Spirochete Group. These results indicate that a large swath of fdhF diversity previously present in the guts of lower termite-like ancestors was lost from gut communities during higher termite evolution, consistent with a genetic bottleneck in the fdhF gene pool.

A protein phylogeny of Sec and Cys clade sequences within the fiGut spirochete groupfl of formate dehydrogenases


Protein phylogeny of Sec and Cys clade sequences within the “Gut spirochete group” of FDHH (light grey, previously Clade A, Figure A12-3). Higher termite sequences group with a metagenomic sequence from Nasutitermes (IMG gene object identity (more...)

To substantiate phylogenetic observations of genetic bottlenecking and the accompanying hypothesis of gene extinction during higher termite gut community evolution, we assayed higher termite guts for the presence of “Cys clade” fdhF, a major phylogenetic group which comprised roughly half of all fdhF variants in the guts of lower termites and wood-roaches (Zhang et al., 2011). Using Cys clade specific primers (Cys499F1b, 1045R), we screened the gut DNA of all higher termite samples, 3 lower termite species from Southern California representing 3 termite families, and a wood-feeding roach for Cys clade genes by PCR. No product (or correct sized product) was detected from any higher termite templates after 30 cycles of PCR amplification (Figure A12-5). In contrast, all amplifications from lower termite and roach gut templates yielded robust products. Bearing in mind the inherent limitations of primer-based assays, the results independently corroborate inventory data, which pointed to unique sweeping losses of fdhF diversity having occurred in higher termite gut communities.

A photo of targeted PCR assays on termite and roach gut DNA using Cys clade specific fdhF primers


Targeted PCR assays on termite and roach gut DNA using Cys clade specific fdhF primers (Cys499F1b, 1045R), which yield a ca. 600 bp product (ladder left lane: NEB 2-log). Templates are: ZAS-2, T. primitia str. ZAS-2 genomic DNA; Zn, Z. nevadensis; Rh, (more...)

To gain insight on when such drastic culling of fdhF gene diversity might have occurred, we analyzed the phylogeny of Higher Termite Spirochete clade sequences (Figure A12-6). Sequences from the most closely related higher termites clustered into shallow clades, suggesting recent coevolution between host and fdhF encoded by gut symbionts. However, sequences from higher termite species representing different subfamilies formed deeply branching clades whose branching order could not be resolved. This type of “spoke” topology in phylogenetic trees, wherein clades radiate with no resolvable order like spokes from a wheel, is typical of adaptive (or ‘ecological’) radiations (Sudhaus, 2004). Based on uncertainties in branching order at the sub-family level, the radiation may have occurred sometime during the emergence of higher termite sub-families. Since adaptive radiations commonly occur after massive extinctions, the major loss in fdhF diversity observed in both inventory and PCR data may have taken place earlier than sub-family diversification, possibly during the evolution of the last common ancestor (LCA) of all higher termites over 50 Mya (Grimaldi and Engel, 2005). The earliest evolved higher termites belong to the fungus feeding subfamily Macrotermitinae (Engel et al., 2009; Grimaldi and Engel, 2005). Examining fdhF in a fungus-feeding termite should help clarify the timing of fdhF extinction.

A protein phylogeny of Higher Termite Spirochete group sequences


Protein phylogeny of Higher Termite Spirochete group sequences (grey box, previously dark grey clade in Figure A12-4) within the Sec clade of the Gut spirochete group (light grey clade, Figure A12-4). “HT,” “LT” and “WR” (more...)

“Chain of Extinction” Hypothesis: Disappearance of H2 Producing Protists and H2 Consuming Dependents

Given that the fdhF gene extinction event (or events) occurred during time periods relevant to the LCA of higher termites, the most plausible cause of fdhF extinction would seem to relate to another extinction that transpired in the same time period: the extinction of H2-producing gut protozoa. While these links are circumstantial, the link between fdhF and protozoa also makes functional sense. A dramatic extinction of primary H2-producers (leading to a shift in niche occupancy) in the lower termite-like LCA would undoubtedly propagate down the microbial “food chain” to H2-consuming symbionts, such as H2-consuming acetogens that possess fdhF. The result of this propagation could manifest itself in extant higher termites in one of two ways: (1) a dramatic shift in the abundance of H2-consuming symbionts (and their fdhF genes) relative to that found in less derived termites or (2) a dramatic loss of diversity. Our results support the latter scenario, implying that the consequences of protozoa extinction on symbionts (and their genes) lying “downstream” in H2 metabolism were more wave-like than ripple-like.

Not all genes, however, went extinct. Those that survived waves of extinction underwent an explosive radiation to fill out previously occupied niches. We posit that their genetic descendants form the Higher Termite Spirochete group.

The data also provide circumstantial support for a previous hypothesis on the nature of association between certain protozoa and ectosymbiotic spirochetes. Leadbetter et al. (1999) proposed an H2-based symbiosis to explain the presence of spirochetes attached to protozoa surfaces. The results described here strengthen their implication that some ectosymbiotic spirochetes may be acetogenic and also draw attention to unexplored metabolic dependencies between protists and free-living spirochetes that may not require physical proximity.

Selenium Dynamics: Sweeping Gene Extinction Followed Later by Occasional Convergent Evolution

Our previous study of Gut Spirochete Group FDHH from the guts of less derived wood-feeding insects revealed two functional enzyme variants differing in a key catalytic residue (Figure A12-4) (Zhang et al., 2011). “Sec clade” sequences have been predicted to encode enzymes that contain the trace element selenium in the form selenocysteine at the active site. In contrast, Cys clade sequences generally encode selenium-independent variants containing cysteine, instead of selenocysteine, at the active site. Study of T. primitia (Matson et al., 2010) and other organisms (Jones et al., 1981; Valente et al., 2006; Vorholt et al., 1997) that possess dual enzyme variants indicate that organisms will switch to using their selenium-independent enzymes under selenium limitation.

In higher termites, the striking absence of Cys clade gene variants implied that some selective pressure related to selenium limitation was relaxed in higher termite gut communities, such that genes for selenium-independent enzymes were lost by genome reduction from symbiont genomes. An alternative explanation was that the characteristic absence of Cys clade genes in higher termites related to sampling differences between studies. To address this concern, we collected 2 lower termite species from the same habitats as a subset of higher termites (Costa Rican lower termite Coptotermes sp. Cost009 collected near Microcerotermes sp. Cost006 and Cost008; desert-adapted lower termite R. tibialis JT1 collected near higher termite species Amitermes sp. JT1 and Gnathamitermes sp. JT5) and performed PCR screening on whole gut community DNA with Cys clade specific primers. Correct-sized PCR amplicons were observed for all lower termites, regardless of where they were collected (Figure A12-5) to independently support inventory data. Thus, the absence of Cys clade genes is a characteristic feature of higher termite gut microbial communities, rather than a result of sampling differences.

To further explore the dependence of higher termite FDHH on selenium (a trace nutrient whose bioavailability varies with redox state [Fishbein, 1983; Masscheleyn et al., 1990]), we inspected every higher termite Sec clade FDHH sequence for the selenocysteine amino acid. We discovered that several sequences from Costa Rican higher termites actually encode selenium-independent FDHH (clones cs6_31cys, cs6_F3cys, cs8Bcys, cs8Dcys, 3D6cys, cs7E6cys, 7H1cys). Since these cysteine-containing FDHH variants were nested within the Sec clade, they must originate from the duplication of a selenium-dependent FDHH gene followed by mutational modification of the active site selenocysteine into cysteine. Also of note is the clustering of Microcerotermes cysteine-containing FDHH with each other, to the exclusion of cysteine FDHH from Nasutitermes. This result points to two independent gene duplication events, each of which has resulted in the “reinvention” (or convergent evolution) of a selenium-independent FDHH gene from Sec clade FDHH gene stock in termites. To our knowledge these data provide the first examples of convergent evolution by symbiotic gut microbes in the termite gut.

The forces that have selected for convergent evolution are intriguing. One possibility is that the selenium content of the termite’s diet may vary enough to affect selenium bioavailability in the gut tract and thus select for one or the other gene variant. This hypothesis stems from the observation that the majority of reinvented cysteine variants (Figure A11-6) were identified in termites collected from palm trees in beach areas where plants are regularly submerged in seawater. Estimates of total selenium concentrations in the surface mixed layers of the ocean are 4-orders of magnitude lower than in surface soils (Nriagu, 1989). Thus seawater may flush selenium out from beach soil, reducing selenium levels in plants, and consequently the diet of termites. But even if dietary selenium levels drove convergent evolution, the larger question of why selenium-dependent FDHH genes are favored over selenium-independent variants in higher termites remains unanswered. Perhaps a structural aspect of the gut tract in higher termites makes the same amount of selenium more easily bioavailable in higher termite than in lower termite guts. Perhaps behavioral innovations, e.g. relating to increased and more effective grazing for nutrients, are at play. The continued study of selenium biology and chemistry in termite guts, and the termites in which they reside, should provide further insight into such possibilities.

Recent Gene Invasion into Subterranean and Litter-Feeding Higher Termite Gut Communities

Gut preparations from subterranean (Amitermes sp. Cost010, Amitermes sp. JT2, and Gnathamitermes sp. JT5) and litter-feeding higher termites (Rhynchotermes sp. Cost004) featured a novel clade of FDHH absent in other termites (clade B in Figure A12-3). Figure A12-7 shows the detailed phylogeny of AGR group sequences. Since we could not infer the identity of uncultured organisms encoding these sequences from phylogeny, we inspected the sequences for possible indel signatures. Indeed, AGR sequences contained an amino acid indel (Figure A12-7, right panel) similar to that previously observed in Gut Spirochete Group sequences, weakly suggesting a spirochetal origin for AGR group genes, or for that indel.

A two-panel protein phylogeny and amino acid character analysis of AGR group sequences


Protein phylogeny (left panel) and amino acid character analysis (right panel) of AGR group sequences (Clade B, Figure A12-3). In the left panel, sequences from T. primitia represent the Gut Spirochete Group (Clade A, Figure A12-3). Diamonds located next (more...)

We hypothesized that AGR group genes might be diagnostic markers for subterranean and litter-feeding higher termite diets and behaviors. To test this hypothesis, we designed AGR clade-specific primers (AGR193F, 1045R). We screened lower and higher termite gut DNA templates using nested PCR methods (Figure A12-8). Robust amplicons were consistently detected in every subterranean and litter-feeding higher termite, but not in arboreal higher termites, lower termites, or C. punctulatus to support the conjecture that AGR group alleles are characteristic of subterranean and litter-feeding higher termite gut communities.

A photo of the products from nested PCR reactions using universal fdhF


Products from nested PCR reactions using universal fdhF primers followed by Amitermes-Gnathamitermes-Rhychotermes clade specific primers on gut templates. NEB 2 log ladder was used. See Figure A12-5 legend for template designations. Slight band in lane (more...)

To understand why AGR group sequences would be present in only a subset of higher termites, we compared the relative abundance of AGR sequences (Table A12-3) with the host’s predicted diet (Table A12-1). AGR sequences were the most abundant phylotype in leaf litter-feeding termites (51%, Rhynchotermes sp. Cost004), but appeared at lower frequencies (8–13%) in subterranean termites with diets containing monocots (such as sugarcane roots, grass, and Yucca), and were not recovered from termites feeding on wood. Based on tannin levels reported for bark, leaves, and wood (Hernes and Hedges, 2004), the presence of AGR group sequences may positively relate to dietary tannin levels and represent a novel marker for an as-of-yet unappreciated group of uncultured acetogens, perhaps ones that exhibit greater tolerance to phenolic compounds like tannin. Alternatively, they may represent a group of non-acetogenic, tannin-tolerant, heterotrophic bacteria that ferment residual sugars in decaying leaves, and employ the enzyme in direction of formate oxidation, perhaps in concert with non-acetogenic formyl-tetrahydrofolate synthetase genes inventoried from the guts of litter and subterranean higher termites (Ottesen et al., 2011). In any case, the phylogenetic remoteness of the AGR group from other major FDH clades suggests that a niche that was previously small (or absent) in wood-feeding termites gained importance in higher termites that feed on decaying plant materials that have substantial contact with soil.

The basal location of subterranean termite sequences (Figure A12-7) hints that the influx of AGR type gene stock into gut communities occurred in a termite belonging to the Termitinae subfamily and that such genes may have been laterally transferred into the Nasutitermitinae. It remains to be determined whether the initial influx manifested itself as the lateral transfer of AGR genes from an organism passing through the gut to an established gut symbiont, or the invasive establishment of a novel group of gut symbionts. Indeed complex phylogenetic relationships between spirochete rRNA genes and host termites (Lilburn et all, 1999; Ohkuma et all, 1999) which are not strictly co-cladogenic imply that acquisition of gut symbionts has been ongoing during termite evolution, a concept outlined previously (Breznak et all, 2006; Ohkunma et all, 2006). However, these events are most likely in the very distant past, as there is also strong evidence in support of broad levels of spirochete coevolution with lower termites (Berlanga et al., 2007). The select presence of AGR type genes in litter and subterranean higher termite guts suggests a more recent acquisition in the evolutionary history of this successful lineage of termite hosts.

PCA Analysis: The Past Shapes Most of the Present

To quantify the importance of different factors associated with FDHH phylogeny, we performed a principal component analysis (PCA, Figure A12-9) using the phylogeny statistics software UniFrac (Lozupone and Knight, 2005). The first principal component (P1, 27.64% of total variance, Figure A12-9a, A12-9b) clearly separates lower termite and wood-roach inventories from higher termite inventories, a result consistent with P1 tracking the presence (lower termite, wood-roach) or absence (higher termite) of flagellate protozoa in gut communities. This result supports our hypothesis that fdhF gene extinction results from protozoal extinctions. The second (Figure A12-9a, 9c) and third principal components (Figure A12-9b, A12-9c) accounted for similar levels of variance (15.46%, 14.25%). P2 clustered inventories containing Protobacteria clade sequences together whereas P3 grouped those containing AGR clade sequences. The latter grouping is supports the idea that dietary variables (amount of soil and form of lignocellulose in diet) also play roles in shaping gut communities (Brauman et al., 2001; Miyata et al., 2007; Yamada et all, 2007). PCA analysis does not appear to cluster data based on geography, nest-type, or habitat—the diversity of which were considerable in the sampled species. Based on these data, the transition between lower termite body and gut community plans to the higher termite forms seems to far outweigh in importance other variables for shaping fdhF diversity in higher termites. This is consistent with the notion that the long ago signal imprinted in fdhF sequence diversity in higher termites is the mass extinction of protozoa during their transition.

A three-panel scatterplot showing UniFrac principal component analysis of formate dehydrogenases phylogeny


UniFrac principal component analysis of FDHH phylogeny associated with the gut microbial communities of termites and related insects. Wood-roach and lower termites: Cp, C. punctulatus; Zn, Z. nevadensis; Rh, R. hesperus; Im, I. minor. Higher termites: (more...)

Model for fdhF Evolution in Wood-feeding Dictyopteran Insects

Based on our findings, we constructed a schematic modeling the evolutionary trajectory of fdhF in the guts of wood-feeding insects, beginning with fdhF in the LCA of termites and wood-feeding cockroaches to the present day (Figure A12-10). The evolutionary sequence highlights the importance of past extinction events as key determinants of present diversity. Previous data (Zhang et al., 2011) imply that a spirochete member of the gut community within the LCA of termites and roaches possessed an ancestral fdhF gene, which underwent duplication and mutational modification into selenocysteine and cysteine encoding forms. These two functional variants of fdhF then co-radiated with gut communities and the host insects as wood-feeding insects were diversifying into termite and roach forms to create “Sec” and “Cys clades.” Here we have documented a severe trimming of Sec clade diversity and complete loss of all Cys clade genes. We estimate that this occurred during the emergence of the higher termite line when guts became segmented, foraging behaviors diversified, and cellulolytic protozoa went extinct from the gut community. It is unclear whether losses of Sec and Cys clade genes represent single or multiple extinction events. In any case, genes surviving extinction radiated to fill the newly emptied (or created) niches within higher termite guts. In particular, the convergent evolution of selenium independent fdhF suggests an adaptive radiation into selenium-limited niches that have recently become available in a subset of higher termites. It also appears that some aspect of litter-feeding and subterranean lifestyles has allowed the more recent establishment of a novel clade of fdhF, possibly reflecting an invasion by non–gut adapted species.

A diagram showing the inferred evolutionary history for fdhF in the symbiotic gut microbial communities of lignocellulose-feeding insects


Inferred evolutionary history for fdhF in the symbiotic gut microbial communities of lignocellulose-feeding insects.


The overarching goal of this study was to understand how symbiont communities and their genes have been impacted by drastic change and other perturbations over evolutionary timescales. We accomplished this goal using the obligate nutritional mutualism between termites and their hindgut microbial communities as a “backdrop” for an evolutionary case study of symbiosis. Comparative analysis of a symbiont metabolic gene unveiled a striking implication for evolutionary biology in complex microbial communities wherein the metabolisms of community members form a network of dependent interactions: collapse of a functional population (or network node) within a symbiont community can have dramatic and long lasting effects on the genes encoded by symbionts occupying niches downstream in the chain of community metabolism.

Connectivity and adaptation are themes that have emerged from this study of symbiont communities. The challenge now is to understand the specific interactions on which connectivity was based in the distant past. Studying the genes and organisms involved in present day interactions between specific microbes within termite gut communities should give us clues on how the past has shaped and continues to shape the present.

Materials and Methods

Insect Collection and Classification

Details on insect collection can be found in Ottesen et al. (Masscheleyn et al., 1990). Briefly, termite obtained by permit in Costa Rica were Nasutitermes sp. Cost003, collected from the National Biodiversity Institute of Costa Rica (INBio) forest; Rhynchotermes sp. Cost004, from leaf litter within INBio; Amitermes sp. Cost010, from soil-encrusted decayed sugar cane at a sugar cane plantation in Grecia; Nasutitermes corniger Cost007, Microcerotermes spp. Cost006 and Cost008, from unidentified species of palm growing at a beach in Cahuita National Park (CNP); Coptotermes sp. Cost009 (lower termite, family Rhinotermitidae), from wood near CNP’s Kelly Creek Ranger Station. Termites obtained under US Park Service Research Permit from Joshua Tree National Park, CA were Amitermes sp. JT2 and Gnathamitermes sp. JT5, collected from subterranean nests; Reticulitermes tibialis JT1 (lower termite, family Rhinotermitidae), collected from a decayed log in a dry stream bed.

Termites were identified based on mitochondrial cytochrome oxidase 2 gene sequence (COII, Fig. S1) and morphology. In general, inadequate COII sequence data prevented taxonomic assignments past genus level. Genus names for Rhynchotermes sp. Cost004 and Gnathamitermes sp. JT5 specimens were assigned based on head and mandible morphology, and collection location. COII analyses confirmed that the 8 termite species examined in this study represent distinct lineages in the subfamilies Nasutitermitinae and Termitinae. Classification of termite habitats was based on the Holdridge life zone classification of land areas (Holdridge et al., 1971) and life zone maps in references (Enquist, 2002; Lugo et al., 1999).

DNA Extraction

For each termite species, the entire hindguts of 20 worker termites were extracted within 48 hours of collection, pooled into 500 μl 1X Tris-EDTA buffer (10 mM Tris-HCl, 1 mM EDTA, pH 8), and stored at −20°C until DNA extraction. Whole gut community DNA was obtained using the method described in (Matson et al., 2007).

fdhF Amplification and Cloning

PCR reactions containing universal fdhF primers (1 μM of each degenerate form) (Zhang et al., 2011) were assembled with 1X FAILSAFE Premix D (EPICENTRE Biotechnologies, Madison, WI). Polymerase (0.07 – 0.14 U × μl−1, EXPAND High Fidelity polymerase, Roche Applied Science, Indianapolis, IN) and gut DNA template concentrations (0.05 − 1 ng × μl−1) were adjusted so that reactions would yield similar amounts of PCR product. Thermocycling conditions for PCR on a Mastercycler Model 5331 thermocycler (Eppendorf, Westbury, NY) were: 2 min at 94°C, 25 cycles of (94°C for 30 sec; 51°C, 53.6°C or 55°C for 1 min; 68°C for 2 min 30 sec), and 10 min at 68°C. Details on PCR are presented in Table A12-4. Amplifications at 51°C annealing temperature yielded multiple sized products upon gel electrophoresis with 1.5% w/v agarose (Invitrogen, Carlsbad, CA). The correct-sized bands were excised and gel purified with a QIAquick Gel Extraction Kit (QIAGEN, Valencia, CA). To ensure product specificity, PCR was performed at higher annealing temperatures (53.6°C for Microcerotermes sp. Cost008, Amitermes sp. Cost010; 55°C for Nasutitermes sp. Cost003, Rhynchotermes sp. Cost004). These reactions yielded single product bands upon electrophoresis. All PCR products were cloned using a TOPO-TA cloning kit (Invitrogen). Clones (30-107 per termite species) were screened for the presence of the correct sized insert by PCR and gel electrophoresis. PCR reactions contained T3 (1 μM) and T7 (1 μM) primers, 1X FAILSAFE Premix D (EPICENTRE), 0.05 U ⊕ μl−1 Taq polymerase (New England Biolabs, Beverly, MA) and 1 μL of cells in 1X TE as template. Thermocycling conditions were 2 min at 95°C, 30 cycles of (95°C for 30 sec, 55°C for 1 min, 72°C for 2 min 30 sec), and 10 min at 72°C.

TABLE A12-4. PCR Conditions for Clone Library Construction.


PCR Conditions for Clone Library Construction. Shaded Grey Rows Highlight Templates For Which Multiple Libraries Were Created. Thermocycling Conditions for Each PCR Reaction Were 94°C for 2 Min, 25 Cycles of (94°C for 30 s, Annealing for (more...)

RFLP Analysis, Sequencing, Diversity Assessment

Most inventories were subject to RLFP typing, wherein correct-sized products generated by screening PCRs were digested with the restriction enzyme RsaI (New England Biolabs) and electrophoresed on a 2.5% (w/v) agarose gel (Invitrogen). Plasmids from clones with unique RFLP patterns were purified using a QIAprep Spin Miniprep Kit (QIAGEN). For Nasutitermes sp. Cost003, Rhynchotermes sp. Cost004, and Amitermes sp. Cost010 inventories, generated at an annealing temperature of 51°C, plasmids from clones having the correct-sized products were purified for sequencing without RLFP typing. Plasmids were sequenced with T3 and T7 primers at Laragen, Inc. (Los Angeles, CA) using an Applied Biosystems Incorporated ABI3730 automated sequencer. Lasergene software (DNASTAR, Inc., Madison, WI) was used to assemble and edit sequences. Sequences were aligned with ClustalW (Larkin et al., 2007), manually adjusted, and grouped into operational taxonomic units at a 97% protein similarity level based on distance calculations (Phylip Distance Matrix, Jones-Thorton-Taylor correction) and DOTUR (Schloss and Handelsman, 2005). The program EstimateS v8.2.0 (Colwell, 2009) was used to assess inventory diversity and completeness.

COII Amplification for Termite Identification

Mitochondrial cytochrome oxidase subunit II (COII) gene fragments from Costa Rican termites were amplified from DNA containing both insect and gut community material using primers A-tLEU and B-tLYS and the protocol described by Miura et al. (1998, 2000). COII gene fragments from Californian termites were amplified using the supernatant of a mixture containing an individual termite head crushed in 1X TE as template. Primers and PCR conditions were identical to those employed for Costa Rican termite COII. PCR products were purified using a QIAquick PCR purification kit (QIAGEN), sequenced, and analyzed to verify the identity of termite specimens.

Primer Design and PCR for Cys Clade fdhF Alleles

Degenerate primers (Cys499F1b, 1045R) for a major clade of selenium independent (Cys) fdhF alleles present in lower termites and the wood roach C. punctulatus were designed manually using all sequences recovered from these insects in Zhang et al., 2011 Forward primer Cys499F1b (5′–ATG TCS CTK TCS ATI CCG GAA A–3′) specificity is as follows: 38.9% of the sequences are perfectly matched, 22.2% have 1 mismatch, 27.8% have 2 mismatches, and 8.3% have 3 mismatches. No mismatches are in located in the terminal 3′ position. The reverse primer 1045R (5′–CIC CCA TRT CGC AGG YIC CCT G – 3′) was designed based on 154 sequences from higher termites, lower termites and C. punctulatus. The 1045R primer targets both Sec and Cys fdhF variants; 60.3% of the sequences have 0 primer mismatches, 32.4% have 1, 5.8% have 2, and 1.3% have 3 mismatches. All sequences are perfectly matched at the terminal 3′ position. PCR reactions contained 0.4 ng ⊕ ⎧λ-1 of DNA template, 200 nM Cys4991F1b, 200 nM 1045R, 1X FAILSAFE Premix D (EPICENTRE), and 0.05 U ⊕ ⎧λ−1 Taq polymerase (New England Biolabs). Thermocycling conditions were 2 min at 95°C, 30 cycles of (95°C for 30 sec, 60°C for 30 sec, 72°C for 45 sec), followed by 10 min at 72°C.

TABLE A12-5. Detailed Composition of Higher Termite fdhF Inventories.


Detailed Composition of Higher Termite fdhF Inventories. Clade Affiliations Indicated by GS = Gut Spirochete Group, PRT = Proteobacteria Group, AGR = AGR Group, UNHT = Unclassified Higher Termite Group

Primer Design and PCR for AGR Group fdhF Alleles

AGR clade fdhF sequences were amplified using a nested PCR approach in which the amplicon from the first PCR reaction, generated with universal fdhF primers (Zhang et al., 2011), was used as the template for the second PCR reaction, containing the clade-specific primer set AGR193F and 1045R. Clade specificity was imparted by the forward primer, AGR193F (5′–AGG CTT ACC AAG CCG CCT ATC AGA – 3′), which targets 55.6% of the sequences in the AGR clade with 4 or fewer mismatches, none of them at the terminal 3′ end. PCR amplification of all fdhF types was achieved using the PCR reaction compositions and thermocycling conditions (51°C annealing temperature) previously specified for inventories. Clade specific PCR reactions contained 1 μl of diluted product from the first reaction (1:1000 in water), 250 nM AGR193F, 250 nM 1045R, 1X FAILSAFE Premix D (EPICENTRE), and 0.07 U · μl−1 of EXPAND High Fidelity polymerase (Roche). Thermocycling conditions were 2 min at 95°C, 25 cycles of (95°C for 15 sec, 60°C for 30 sec, 72°C for 1 min), followed by 10 min at 72°C.

Phylogenetic and Principle Component Analysis

Phylogenetic analyses of protein and nucleotide sequences were performed with ARB version 09.08.29 (Ludwig et al., 2004). Genbank accession numbers are listed in Table A12-6. Details of tree construction can be found in figure legends. In general, trees show results from Phylip PROML analysis, node robustness was analyzed with PROTPARS and Fitch distance methods as well (Felsenstein, 1998). The same filter and alignments were employed when additional tree algorithms were used to infer node robustness. All phylogenetic inference models were run assuming a uniform rate of change for each nucleotide or amino acid position. Principal component analysis of FDHH phylogeny and environment data was performed using the software Unifrac (Lozupone and Knight, 2005).

TABLE A12-6. Sequences Used in Phylogenetic Analyses.


Sequences Used in Phylogenetic Analyses. FDH-H, Hydrogenase-Linked Formate Dehydrogenase. FDH-NAD, NAD-Linked Formate Dehydrogenase. FDH-F420, F420-Linked Formate Dehydrogenase. Sup., Supplementary. Multiple Accession Numbers For a FDH Sequence that Appears (more...)


This work was supported by National Science Foundation grant EF-0523267 and Department of Energy grant DE-FG02-07ER64484 to JRL and a National Science Foundation graduate research fellowship to XZ. We thank members of the Leadbetter laboratory for their helpful discussions and comments. We are grateful for the aid of Myriam Hernández, Luis G. Acosta, Giselle Tamayo, and Catalina Murillo of the Instituto Nacional de Biodiversidad (Santo Domingo de Heredia, Costa Rica) and Brian Green, Cathy Chang, and Eric J. Mathur, formerly of Verenium, Inc., in termite collection and site access. Termites from Joshua Tree National Park were collected under permit JOTR-2008-SCI-0002.

Author Contributions

All authors participated in insect collection. XZ and JRL conceived of the experiments. XZ performed the experiments and analyzed the data. XZ and JRL wrote the manuscript.

Competing Interests

The authors have no competing interests.


  • Baumann P, Moran NA, Baumann L. Bacteriocyte-associated endosymbionts of insects. In: Dworkin M, Falkow S, Rosenberg E, Schleifer K-H, Stackebrandt E, editors. The Prokaryotes. Vol. 1. New York: Springer; 2006. pp. 403–438.
  • Berlanga M, Paster BJ, Guerrero R. Coevolution of symbiotic spirochete diversity in lower termites. International Microbiology. 2007;10:133–139. [PubMed: 17661292]
  • Brauman A, Dore J, Eggleton P, Bignell D, Breznak JA, Kane MD. Molecular phylogenetic profiling of prokaryotic communities in guts of termites with different feeding habits. FEMS Microbiology Ecology. 2001;35:27–36. [PubMed: 11248387]
  • Brauman A, Kane MD, Labat M, Breznak JA. Genesis of acetate and methane by gut bacteria of nutritionally diverse termites. Science. 1992;257:1384–1387. [PubMed: 17738281]
  • Breznak JA, Leadbetter JR. Termite gut spirochetes. In: Dworkin M, Falkow S, Rosenber E, Schleifer KH, Stackebrandt E, editors. The Prokaryotes. Vol. 7. New York: Springer; 2006. pp. 318–329.
  • Breznak JA, Switzer JM. Acetate synthesis from H2 plus CO2 by termite gut microbes. Applied and Environmental Microbiology. 1986;52:623–630. [PMC free article: PMC239087] [PubMed: 16347157]
  • Brugerolle G, Radek R. Symbiotic protozoa of termites. In: König H, Varma A, editors. Intestinal microorganisms of termites and other invertebrates. Springer; Heidelberg, Germany: 2006. pp. 244–269.
  • Brune A, Stingl U. Prokaryotic symbionts of termite gut flagellates: phylogenetic and metabolic implications of a tripartite symbiosis. Progress in Molecular and Subcellular Biology. 2006;41:39–60. [PubMed: 16623388]
  • Buchner P. Endosymbiosis of animals with plant microorganisms. Interscience; New York: 1965.
  • Chen X, Li S, Aksoy S. Concordant evolution of a symbiont with its host insect species: molecular phylogeny of genus Glossina and its bacteriome-associated endosymbiont, Wiggles-worthia glossinidia. Journal of Molecular Evolution. 1999;48:49–58. [PubMed: 9873076]
  • Colwell RK. EstimateS: Statistical estimation of species richness and shared species from samples. 2009. posting date. version 8.2.0. [Online.]
  • Desai MS, Strassert JFH, Meuser K, Hertel H, Ikeda-Ohtsubo W, Radek R, Brune A. Strict cospeciation of devescovinid flagellates and Bacteroidales ectosymbionts in the gut of dry-wood termites (Kalotermitidae). Environmental Microbiology. 2010;12:2120–2132. [PubMed: 21966907]
  • Dethlefsen L, Mcfall-Ngai M, Relman D. An ecological and evolutionary perspective on human–microbe mutualism and disease. Nature. 2007;449:811–818. [PubMed: 17943117]
  • Eggleton P. The termite gut habitat: Its evolution and co-evolution. In: König H, Varma A, editors. Intestinal microorganisms of termites and other invertebrates. Springer; Heidelberg, Germany: 2006. pp. 373–404.
  • Eggleton P, Bignell DE, Sands WA, Mawdsley NA, Lawton JH, Wood TG, Bignell NC. The diversity, abundance and biomass of termites under differing levels of disturbance in the Mbalmayo Forest Reserve, southern Cameroon. Philosophical Transactions of the Royal Society B: Biological Sciences. 1996;351:51–68.
  • Engel MS, Grimaldi DA, Krishna K. Termites (Isoptera): Their phylogeny, classification, and rise to ecological dominance. American Museum Novitates. 2009;3650:1–27.
  • Enquist CAF. Predicted regional impacts of climate change on the geographical distribution and diversity of tropical forests in Costa Rica. Journal of Biogeography. 2002;29:519–534.
  • Felsenstein J. PHYLIP - Phylogeny inference package (version 3.2). Cladistics. 1989;5:164–166.
  • Fishbein L. Environmental selenium and its significance. Fundamental and Applied Toxicology. 1983;3:411–419. [PubMed: 6357926]
  • Grimaldi D, Engel MS. Evolution of the insects. New York: Cambridge University Press; 2005.
  • Hernes PJ, Hedges JI. Tannin signatures of barks, needles, leaves, cones, and wood at the molecular level. Geochimica et Cosmochimica Acta. 2004;68:1293–1307.
  • Holdridge LR, Grenke WC, Hatheway WH, Tosi JA. Forest environments in tropical life zones A pilot study. New York: Pergamon Press; 1971.
  • Hongoh Y, Ohkuma M, Kudo T. Molecular analysis of bacterial microbiota in the gut of the termite Reticulitermes speratus (Isoptera; Rhinotermitidae). FEMS Microbiology Ecology. 2003;44:231–242. [PubMed: 19719640]
  • Hongoh Y, Deevong P, Inoue T, Moriya S, Trakulnaleamsai S, Ohkuma M, Vongkaluang C, Noparatnaraporn N, Kudo T. Intra-and interspecific comparisons of bacterial diversity and community structure support coevolution of gut microbiota and termite host. Applied and Environmental Microbiology. 2005;71:6590–6599. [PMC free article: PMC1287746] [PubMed: 16269686]
  • Jones JB, Stadtman TC. Selenium-dependent and selenium-independent formate dehydrogenases of Methanococcus vannielii. Separation of the two forms and characterization of the purified selenium-independent form. The Journal of Biological Chemistry. 1981;256:656–663. [PubMed: 7451465]
  • Kambhampati S, Eggleton P. Taxonomy and phylogeny of termites. In: Abe T, Bignell DE, Higashi M, editors. Termites: Evolution, Sociality, Symbioses, Ecology. Kluwer Academic Publishers; Dordrecht, The Netherlands: 2000. pp. 1–24.
  • Larkin MA, Blackshields G, Brown NP, Chenna R, McGettigan PA, McWilliam H, Valentin F, Wallace IM, Wilm A, Lopez R, Thompson JD, Gibson TJ, Higgins DG. Clustal W and Clustal X version 2.0. Bioinformatics. 2007;23:2947–2948. [PubMed: 17846036]
  • Leadbetter JR, Schmidt TM, Graber JR, Breznak JA. Acetogenesis from H2 plus CO2 by spirochetes from termite guts. Science. 1999;283:686–689. [PubMed: 9924028]
  • Lilburn TG, Schmidt TM, Breznak JA. Phylogenetic diversity of termite gut spirochaetes. Environmental Microbiology. 1999;1:331–345. [PubMed: 11207751]
  • Lo N, Eggleton P. Termite phylogenetics and co-cladogenesis with symbionts. In: Bignell DE, Roisin Y, Lo N, editors. Biology of Termites: A Modern Synthesis. Springer; Dordrecht, Netherlands: 2011. pp. 27–50. [PMC free article: PMC4212872] [PubMed: 25368037]
  • Lozupone C, Knight R. UniFrac: a new phylogenetic method for comparing microbial communities. Applied and Environmental Microbiology. 2005;71:8228–8235. [PMC free article: PMC1317376] [PubMed: 16332807]
  • Ludwig W, Strunk O, Westram R, Richter L, Meier H, Kumar Y, Buchner A, Lai T, Steppi S, Jobb G, Förster W, Brettske I, Gerber S, Ginhart AW, Gross O, Grumann S, Hermann S, Jost R, König A, Liss T, Lüßmann R, May M, Nonhoff B, Reichel B, Strehlow R, Stamatakis A, Stuckmann N, Vilbig A, Lenke M, Ludwigh T, Bode A, Schleifer KH. ARB: a software environment for sequence data. Nucleic Acids Research. 2004;32:1363–1371. [PMC free article: PMC390282] [PubMed: 14985472]
  • Lugo AE, Brown SL, Dodson R, Smith TS, Shugart HH. The Holdridge life zones of the conterminous United States in relation to ecosystem mapping. Journal of Biogeography. 1999;26:1025–1038.
  • Masscheleyn PH, Delaune RD, Patrick WH. Transformations of selenium as affected by sediment oxidation-reduction potential and pH. Environmental Science & Technology. 1990;24:91–96.
  • Matson EG, Ottesen E, Leadbetter JR. Extracting DNA from the gut microbes of the termite Zootermopsis nevadensis. Journal of Visualized Experiments. 2007;4:195. [PMC free article: PMC2556161] [PubMed: 18979000]
  • Matson EG, Zhang X, Leadbetter JR. Selenium controls transcription of paralogous formate dehydrogenase genes in the termite gut acetogen, Treponema primitia. Environmental Microbiology. 2010;12:2245–2258. [PubMed: 21966917]
  • McFall-Ngai M. Unseen forces: the influence of bacteria on animal development. Developmental Biology. 2002;242:1–14. [PubMed: 11795936]
  • Mira A, Ochman H, Moran NA. Deletional bias and the evolution of bacterial genomes. Trends in Genetics. 2001;17:589–596. [PubMed: 11585665]
  • Miura T, Maekawa K, Kitade O, Abe T, Matsumoto T. Phylogenetic relationships among subfamilies in higher termites (Isoptera: Termitidae) based on mitochondrial COII gene sequences. Annals of the Entomological Society of America. 1998;91:515–521.
  • Miura T, Roisin Y, Matsumoto T. Molecular phylogeny and biogeography of the Nasute termite genus Nasutitermes (Isoptera: Termitidae) in the Pacific Tropics. Molecular Phylogenetics and Evolution. 2000;17:1–10. [PubMed: 11020299]
  • Miyata R, Noda N, Tamaki H, Kinjyo K, Aoyagi H, Uchiyama H, Tanaka J. Influence of feed components on symbiotic bacterial community structure in the gut of the wood-feeding higher termite Nasutitermes takasagoensis. Bioscience, Biotechnology, and Biochemistry. 2007;71:1244–1251. [PubMed: 17485852]
  • Moran NA. Microbial minimalism: genome reduction in bacterial pathogens. Cell. 2002;108:583–6. [PubMed: 11893328]
  • Moran NA, Baumann P. Bacterial endosymbionts in animals. Current Opinion in Microbiology. 2000;3:270–275. [PubMed: 10851160]
  • Moran NA, Telang A. Bacteriocyte-associated symbionts of insects. Bioscience. 1998;48:295–304.
  • Moran NA, McLaughlin HJ, Sorek R. The dynamics and time scale of ongoing genomic erosion in symbiotic bacteria. Science. 2009;323:379–382. [PubMed: 19150844]
  • Nalepa C, Bandi C. Characterizing the ancestors: Paedomorphosis and termite evolution. In: Abe T, Bignell DE, Higashi M, editors. Termites: Evolution, Sociality, Symbioses, Ecology. Kluwer Academic Publishers; Dordrecht, Netherlands: 2000. pp. 53–77.
  • Noda S, Kitade O, Inoue T, Kawai M, Kanuka M. Cospeciation in the triplex symbiosis of termite gut protists (Pseudotrichonympha spp.), their hosts, and their bacterial endosymbionts. Molecular Ecology. 2007;16:1257–1266. [PubMed: 17391411]
  • Nriagu JO. Global cycling of selenium. In: Ihnat M, editor. Occurrence and Distribution of Selenium. CRC Press; Boca Baton, Florida: 1989. pp. 327–340.
  • Odelson DA, Breznak JA. Volatile fatty acid production by the hindgut microbiota of xylophagous termites. Applied and Environmental Microbiology. 1983;45:1602–1613. [PMC free article: PMC242507] [PubMed: 16346296]
  • Ohkuma M, Iida T, Kudo T. Phylogenetic relationships of symbiotic spirochetes in the gut of diverse termites. FEMS Microbiology Letters. 1999;181:123–129. [PubMed: 10564797]
  • Ohkuma M, Hongoh Y, Kudo T. Diversity and molecular analyses of yet-uncultivated microorganisms. In: König H, Varma A, editors. Intestinal microorganisms of termites and other invertebrates. Springer; Heidelberg, Germany: 2006. pp. 303–317.
  • Ohkuma M, Noda S, Hongoh Y, Nalepa CA, Inoue T. Inheritance and diversification of symbiotic trichonymphid flagellates from a common ancestor of termites and the cockroach Cryptocercus. Proceedings of the Royal Society B: Biological Sciences. 2009;276:239–245. [PMC free article: PMC2674353] [PubMed: 18812290]
  • Ottesen EA, Leadbetter JR. Formyltetrahydrofolate synthetase gene diversity in the guts of higher termites with different diets and lifestyles. Applied and Environmental Microbiology. 2011;77:3461–3467. [PMC free article: PMC3126463] [PubMed: 21441328]
  • Pester M, Brune A. Hydrogen is the central free intermediate during lignocellulose degradation by termite gut symbionts. ISME Journal. 2007;1:551–565. [PubMed: 18043656]
  • Schloss PD, Handelsman J. Introducing DOTUR, a computer program for defining operational taxonomic units and estimating species richness. Applied and Environmental Microbiology. 2005;71:1501–1506. [PMC free article: PMC1065144] [PubMed: 15746353]
  • Shigenobu S, Watanabe H, Hattori M, Sakaki Y, Ishikawa H. Genome sequence of the endocellular bacterial symbiont of aphids Buchnera sp. APS. Nature. 2000;407:81–86. [PubMed: 10993077]
  • Stamatakis AP, Ludwig T, Meier H. The AxML program family for maximum likelihood-based phylogenetic tree inference. Concurrency and Computation: Practice and Experience. <