In prokaryotic species for which genomes of several different strains have been completely sequenced, orthologous genes of the species core can usually be identified easily; they are conserved in chromosomal position as well as in sequence. For deeper and more inclusive taxa, cores become progressively smaller and more elusive because of weak phylogenetic signals, genomic rearrangements, and problems in recognizing paralogy. Nevertheless, there is much interest in such deep cores, especially the universal (Bacteria + Archaea + Eukarya) core. Several investigators argue that this core's composition might reflect that of the genome of the Last Universal Common Ancestor (LUCA), while phylogenies of its genes, if congruent, could delineate the earliest branchings of the Tree of Life (Brown et al. 2001; Woese 2002; Koonin 2003).

Recent attempts to define the universal core have concluded that it contains very few genes. Harris et al. (2003), in a study of 34 genomes report 80 core genes, Koonin (2003) using about 100 genomes finds something like 60 core genes, Brown et al. (2001) with 45 genomes (and greater requirements for stringency) 23. Although the operational and largely arbitrary nature of the definition of this minimal set has been appreciated, several authors have attributed biological significance to the fact that the number of ubiquitous genes is so small. For instance, for Koonin (2003), “the important realization that comes from this type of analysis is the remarkable evolutionary plasticity of even the central, essential biological functions. Only a tiny group of genes (nearly all of them associated with translation and transcription) is truly ubiquitous among living things”. Woese (1987) asks, in the phylogenetic context, “What does it mean, then, to speak of an organismal genealogy when nearly all of the genes in the cell, genes that give it its general character, do not share a common history?”

The PBC approach we describe will address the problem of errors and the problem of disproportionate weighting of popular phyla when ubiquity is not required. It also favors universality (within a phylogenetic context), but it still does not tell us where to draw the line. We do find it encouraging that the core defined in this way increases less than 20% in size, when prevalence required within taxa is dropped from 90% to 80%. If, indeed, there is a conserved set of genes that might be considered the basic heritage for all prokaryotes, but which defies precise definition because of occasional loss or orthologous replacement in scattered lineages, the PBC might provide a good approximation.

Although a nonarbitrary delimitation of core size may be impossible, core composition is not random. Cores are generally dominated by genes of the translational apparatus. Why should this be? The widely accepted explanation is that genes of this important informational process are intrinsically less exchangeable than genes of operational processes, like metabolism (Woese 1987, 2002; Jain et al. 2002). There are so many complex coevolved interactions with other cellular informational constituents, this argument holds, that any replacement of one of these genes by a distant homolog or by an analog would be disadvantageous. Although one might object that there are individual instances in which such components (not only ribosomal proteins and translation factors, but ribosomal RNAs themselves) have been transferred, this complexity hypothesis (or annealing hypothesis) remains appealing and popular.

But there is an alternative explanation, which we will sketch out here, and have represented by a cartoon in Figure 4. All cells need genes of each of several functional categories, such as (put very crudely) translation, metabolism, and cell envelope formation. Lake et al. (1999) and Jain et al. (2002) describe genes for replication, transcription, and translation as informational, and other classes as operational, but the distinction is not hard and fast. Imagine that evolution has assembled cells in a mix-and-match fashion, choosing to satisfy each of the several functional categories with genes available from a global repertoire of determinants, whether homologous, analogous, or in the case of a function superfluous in a current niche, null. Evolution no longer makes cells from scratch, so in general, what happens is that one gene or suite of cofunctional genes (a pathway) replaces another, a general process that includes both replacement by close or distant homologs and by analogs, which Koonin and his collaborators call nonorthologous displacement (Koonin et al. 1996). Where there are many analogous (nonhomologous) types of genes (or suites of genes) that can perform the same general function (e.g., energy production or cell envelope formation), the living world will collectively exhibit much variability, and there will be no ubiquitous sets of genes that appear as part of any universal core. Where choices are more limited, most genes performing the needed function (some step in translation for instance) being homologous, there will appear to be little variability. Our analysis and all similar studies of the core will not generally detect orthologous replacements, in which a resident gene is replaced by an ortholog from another species. Such genes will appear as part of a universal core, but this need not mean that they are exchanged any less frequently than genes in other categories (as implied in the complexity hypothesis). Against the objection that there would be no selective advantage to such cryptic orthologous replacements of generally essential genes, we note that antimicrobials are often targeted against the products of such genes, and here the advantages would be obvious and large.

Exchanges of this nature need not be cryptic in phylogenetic analyses. Informational genes of the universal prokaryotic core should not produce congruent phylogenetic trees if cells have exchanged them for foreign (but still orthologous) versions as often as they seem to have traded nonorthologous (indeed, nonhomologous) operational genes. Whether or not genes of the universal core do show congruent phylogenies is, however, still a matter of legitimate debate; at this (phylum) depth few individual genes have reliable phylogenetic signal. Several years ago, Teichmann and Mitchison (1999) observed that only three of 32 protein families shared between selected bacterial, archaeal, and eukaryotic genomes showed significant phylogenetic signal, and concluded that this signal was actually due to recent LGT events. Several recent studies that have used concatenated sequences of core genes to construct universal trees have argued that the robustness of these trees reflects an underlying phylogenetic coherence (Brown et al. 2001; Brochier et al. 2002; Matte-Tailliez et al. 2002). But, it was the failure of the genes individually to produce resolved trees that motivated concatenation in the first place, and in any event, all such reports conclude that there is significant divergent signal among core translational genes. Harris et al. (2003) note, from preliminary phylogenetic analyses, that 30 of their 80 universal core genes do not maintain domain monophyly, showing clades in which Bacteria, Archaea, and Eukarya are mixed. Brown et al. (2001) were reduced to a core of only 14 genes they considered congruent phylogenetically. Overall, it might be safe to say that informational genes support the Bacterial/Archaeal dichotomy more often than do operational genes— and perhaps, more often than not. But, a consistent and extensive pattern of congruence among informational genes in branching patterns at the level of bacterial and archaeal phyla has not been established.

Using BLASTP scores, we can distinguish between ordinary matches (including many which may be paralogous) and reciprocal best matches (RBMs), which are more often orthologous. Additionally, we can make use of information present in genomic annotations, permitting genes with standardized gene names to find their orthologs despite BLASTP matches that may fall slightly below threshold. Where annotations are reasonably consistent, this permits the union of overlapping sets of genes where BLASTP matches alone might miss members outside of the sets' intersection. Matches are still based on BLASTP, but some-what disconnected outliers can then link through the bridge of a common consensus gene name (CGN). These are computed as follows: For each ORF in a genome, its RBM, if any, is found in each of the other genomes, and the RBM's annotated name is appended to a list. The dominant name in this list (e.g., ftsA) becomes the query ORF's CGN. Lists of prevalent names, found in most or all members of a set of genomes, are generated by the “List named ubiquitous genes” query within NGIBWS (http://www.neurogadgets.com/bws.php) (Charlebois et al. 2003).

A strict definition of a core of genes has those genes present in every member of the set of genomes under consideration. We performed such an analysis on clades of genomes (defined according to a genomic phylogeny [Gophna et al. 2004]), using the “Find ubiquitous genes” query within NGIBWS (Charlebois et al. 2003). Using each of the genomes from the clade in turn, ORFs are found that have an RBM (with allowance for near ties) better than the specified BLASTP threshold, in each of the other genomes in the clade.

Both RBM and CGN approaches present some limitations. False negatives can arise in the RBM approach when BLASTP matches fall below threshold; false positives can arise when match thresholds are set so low as to permit spurious matches, and when paralogs match for want of lost orthologs. The use of CGNs as bridges for weak matches overcomes some of the problems inherent in the pure RBM approach, but introduces new problems relating to inconsistent annotation. Where alternative names for a gene are popular, an orthologous cluster may break up into name cliques, where a gene is truly ubiquitous, but fails to appear as a ubiquitous CGN. False positives are theoretically possible with the CGN approach, but only if homonymous names are used in annotation, which should be rare among core or otherwise prevalent genes, lysyl-tRNA synthetase (lysS) not-withstanding (Ibba et al. 1999).