Plasmid Constructions. A derivative of pBRR1MCS mob^– (Cm^r) (14), which was suitable for reporting interbacterial protein translocation (called pZL184; Fig. 1), was constructed in a cloning process involving multiple steps. Sequences of oligonucleotides used, sources of gene cassettes, and details of cloning are found in Supporting Methods, which is published as supporting information on the PNAS web site. To express the Cre fusions, we first constructed pZL180 (Amp^r and thyA⁺), which contains cre on the transfer deficient RSF1010 plasmid pJB908 (J. Vogel, Washington University School of Medicine, St. Louis), and fusions were constructed as follows: Whole ORFs were amplified by PCR and were fused to the 3′ end of cre on pZL180 for genes smaller than 2 kb, whereas for genes larger than 2 kb, the 3′ regions encoding 500 or 700 amino acids of each ORF was fused to cre.

Bacterial Two-Hybrid Screening. First, ralF was translationally fused to the 3′ end of fragment T18 of the Bordetella pertussis cya gene, encoding adenylate cyclase, on pUT18 (15). Sau3AI-generated genomic DNA fragments from L. pneumophila strain Lp02 (11), ranging from 800 base pairs to 5 kb, were inserted into pKT25 (15), resulting in a library of L. pneumophila proteins fused to the C terminus of B. pertussis adenylate cyclase T25 fragment. Escherichia coli strain BTH101 (15), expressing cya::ralF(C), was used to identify RalF-interacting proteins by screening either on LB medium containing 40 μg/ml X-Gal or on the synthetic M63 medium (15) with lactose as the sole carbon source. Strains with functional adenylate cyclase proteins were identified based on the presence of detectable Lac⁺ phenotypes on these media. Proteins that interact with DotF were identified in a similar manner by using either DotF or DotF (28–123) as the bait.

Bacterial Matings and Intercellular Protein Translocation. For RP4 Trb-mediated translocation of the Cre::MobA hybrid, ≈2.5 × 10⁸ cells from saturated cultures of S17–1 (Trb⁺) (16) or DH5α (Trb^–) expressing the fusion grown in LB broth were mixed with a 15-fold excess of recipient strain XL1Blue carrying pZL184. The mixtures were spotted onto 0.45-μm nitrocellulose filters, placed on LB plates, and incubated at 37°C for 3 h. The excisants were selected on LB medium containing 5% sucrose and 30 μg/ml kanamycin.

For L. pneumophila to L. pneumophila translocation, Lp02 (11) or Lp03dotA^– (12) containing plasmids expressing the appropriate Cre fusions were grown to postexponential phase in AYE broth. For matings, 0.45-μm nitrocellulose membranes were first placed onto CYET medium containing 100 nM isopropyl β-d-thiogalactoside (IPTG) and the plates were incubated at 37°C for 1 h before spotting with a mixture of ≈3.5 × 10⁸ donor and a 15-fold excess of recipient Lp03(pZL184). The mating plates were incubated at 37°C for 14 h and excisants were selected on CYET –5% sucrose, 30 μg/ml kanamycin, which kills both parents. Each mating was performed in triplicate and repeated at least three times. Transfer frequencies were expressed as number of excisants (resistant to kanamycin and sucrose but sensitive to gentamicin) per donor bacterium and the number of donor bacteria was determined by counting colonyforming units derived from appropriately diluted donor cells plated onto CYET medium before mating.

For plasmid transfer, transconjugants were detected by plating the protein translocation mating mixture onto LB containing 100 μg/ml ampicillin and 30 μg/ml chroramphenicol (for E. coli) or onto charcoal yeast extract containing 5 μg/ml chloramphenicol (selecting Thy⁺ and Cm^R) for transfer in L. pneumophila). As a positive control, the wild-type RSF1010 derivative pKB5 (11) was used in both E. coli and L. pneumophila matings (6).

Construction of In-Frame Deletions and Complementation of the Deletion Mutants. In-frame deletions of L. pneumophila genes were performed by a two-step allelic exchange strategy as described (17). In each case, the deletion construct was designed such that the intact gene was replaced by an ORF predicted to express a 20-aa polypeptide consisting of the first 10 amino acids and the last 10 amino acids of the gene. To perform complementation studies on deletion mutants, the gene of interest was amplified by PCR and inserted into pJB908 or pBBRMCS2 (14).

Protein Purification and Antibody Preparation. The predicted ORF of SidC was inserted into pQE30 and the resulting His₆-SidC fusion protein was purified from E. coli by using Ninitrilotriacetic acid resin (Qiagen, Valencia, CA). Rabbit polyclonal serum was generated by the Pocono Rabbit Farm (Canadensis, PA; ref. 10). Antibodies were affinity-purified by using a matrix containing purified (His)₆-SidC covalently coupled to Affigel-10 beads (Bio-Rad) (18).

Two proteins known to be transferred from bacterial cells to mammalian cells could also be translocated interbacterially by using this strategy. Full-length ralF and the 3′ half of lidA [called lidA(C)] were each translationally fused to cre and the resulting hybrids were examined for bacterium-bacterium translocation. We found that wild type L. pneumophila, but not a Dot/Icmdeficient strain, could transfer Cre::RalF or Cre::LidA(C), based on the ability of recipient strain to excise the floxed DNA fragment and to express kanamycin resistance [Fig. 1D; Cre::RalF and Cre::LidA(C)]. However, Cre itself could not be translocated by wild-type L. pneumophila (Fig. 1D). The translocation proficiency of Cre::LidA(C) indicates that targeting information resides in the C terminus of the protein, at least for interbacterial transfer. We examined whether this finding was also true for RalF by testing Dot/Icm-mediated translocation of Cre::RalF(C) and Cre::RalF(N) respectively. Only Cre:RalF(C) was translocated at a detectable frequency (Fig. 1D). These observations are similar to findings regarding A. tumefaciens Vir proteins, in which secretion signals are localized to the C termini of the substrates (20, 23).

Identification of Substrates of the Dot/Icm Transporter. Because the translocated proteins LidA and RalF could be transferred interbacterially, we used the bacterial translocation assay to identify proteins transferred by this transporter. To simplify the screening of a randomly generated fusion library, we chose to screen through a preselected pool of candidate substrates. We hypothesized that some translocated substrates may interact with specific components of the Dot/Icm complex. Therefore, a bacterial two-hybrid screen (15) was used to identify L. pneumophila proteins that specifically interacted with the carboxyl portion of RalF. From 41 positive clones sequenced, DotF, predicted to be an inner membrane component of the Dot/Icm complex, was identified nine times independently. Interestingly, in all cases, only the portion of DotF that spans amino acids 28–123 was obtained. The full-length DotF also gave a positive two-hybrid readout, although at a somewhat lower level (Fig. 2). A bacterial two-hybrid study was then performed by using Cya fusions to the 3′ ends of L. pneumophila genes and baits of DotF or its derivative, DotF (28–123). From ≈150,000 candidates, we isolated 148 clones that showed some level of interaction higher than background, and these were sequenced and analyzed. Although it was clear that some strains gave barely detectable β-galactosidase activity and were of questionable significance, this strategy greatly reduced the number of strains to be analyzed. Complete ORFs of these genes were retrieved from the L. pneumophila genome database (http://genome3.cpmc.columbia.edu/~legion) and 68 genes were identified. Of these genes, 20 candidates were eliminated based primarily on their high hydrophobicity, which may have caused nonspecific interactions with DotF. Forty-eight candidate genes were subsequently tested for interbacterial transfer (see Table 2, which is published as supporting information on the PNAS web site, for data on 17 such candidates from a single experiment). Table 1 shows eight clones that gave a positive signal by using the Cre/loxP reporter system. We designated these proteins substrate of Icm/Dot transporter (Sid) (Table 1). The efficiency of interbacterial transfer varied, ranging from 10^–6 to 10^–5 excisants per input donor cell. Moreover, transfer of all of these proteins depended on the Dot/Icm apparatus, because mating with dotA^– strains expressing these fusions failed to cause the excision of the floxed DNA fragment and the subsequent expression of the kanamycin resistance marker (Table 1).

Characteristics of sid Genes. Sequence analyses revealed that with the exception of sidB, the sid genes have no significant orthologs present in the GenBank NR database (Table 1). SidB contains a putative active site found in some lipases and shows a region similar to a portion of the Rtx toxin from Vibrio cholerae (24). In contrast, many of these proteins have one or more paralogs present in the L. pneumophila genome (Table 1). The similarities among the paralogs range from almost identical predicted proteins (E = 0) to rather loose similarity (E = 5e^–2 to 2e^–8; Table 1). Interestingly, in some cases, subsets of paralogs are organized into contiguous ORFs. For instance, sidC and its homolog sdcA are separated by only 150 base pairs, whereas sdeA, sdeB, and sdeC are closely clustered (Fig. 3). In addition, a paralog of sidE (sdeD) is located directly upstream from what appears to be the operon encoding sidC (Fig. 3B). Interestingly, two putative genes, orf1 and orf3, located near the sidA and sdeC regions are highly similar (E = 0; Fig. 3 A and B), although these genes were not identified in the original two-hybrid assay.

SidC Is Translocated by Dot/Icm Into the Cytosol of Host Cells. To verify that proteins identified in our assay are translocated into mammalian cells, affinity-purified antiserum against His₆-SidC was used to analyze expression and translocation of SidC by L. pneumophila. As predicted, a protein of ≈110 kDa was detected in both a dot/icm⁺ strain and a dotA^– mutant of L. pneumophila, but not in a mutant missing sidC and its paralog sdcA [Δ(sdcA-sidC)] (Fig. 3C). Moreover, expression of SidC was induced ≈7-fold in cells grown to post exponential phase (Fig. 3C). The translocated substrates RalF and LidA also show induced expression in post exponential phase (ref. 9 and G. M. Conover and R.R.I., unpublished observations), and a large body of evidence indicates that factors critical for intracellular survival of L. pneumophila are similarly regulated (25, 26). When infected macrophages were probed for SidC by indirect immunofluorescence microscopy, we found that 1 h after uptake, ≈86% of L. pneumophila phagosomes stained positively for SidC (Fig. 4A), with the protein localizing about the phagosomal membrane (Fig. 4 C and D). In some cases, there was asymmetric diffusion of the protein emanating from the phagosome surface, with most infected cells showing protein only in the region near the phagosome (Fig. 4D).

Because the above strategy only detects secretion of SidC and does not demonstrate translocation across the phagosomal membrane, a second approach was pursued (10). To demonstrate SidC translocation across the phagosomal membrane by the Dot/Icm transporter, we prepared postnuclear supernatants (PNS) from macrophages incubated with L. pneumophila and probed intact phagosomes for SidC in the absence of permeabilization reagents. Approximately 85% of these phagosomes stained positively for SidC (Fig. 4 B and E), whereas <5% stained positively with anti-L. pneumophila serum, demonstrating that the phagosomal membranes were intact. Less than 0.3% of macrophages containing a L. pneumophila dotA^– strain stained positively for SidC (Fig. 4 A and C), which was consistent with our data showing that interbacterial translocation of this protein requires the Dot/Icm apparatus. Similarly, in PNS, no isolated phagosome containing a dotA^– strain stained positively for SidC (Fig. 4B). The antibody reactivity was specific, because no macrophages harboring a mutant lacking sidC and its upstream paralog stained positively for this protein [Fig. 4A; Δ(sdcA-sidC)], nor did macrophages infected with a simple deletion of sidC (data not shown). These results demonstrate that during infection, SidC is translocated into mammalian cells by the Dot/Icm system, and that translocation occurs across the phagosomal membrane. Because the transfer frequencies of other proteins in our Cre/loxP system were comparable to that of SidC, we postulate that these proteins are also targeted to the host cell by the Dot/Icm transporter.

The lack of intracellular growth defects observed in L. pneumophila mutants lacking translocated substrates is reminiscent of a L. pneumophila ralF mutant, which is indistinguishable from the parental wild-type strain in regards to intracellular growth (9). Nevertheless, a ralF mutant is unable to recruit Arf1 to the surface of the replicative phagosome, a property that may play some unknown role in the lifestyle of the organism (27). It is possible that elimination of some of the genes here identified similarly results in the formation of phagosomes with alterations in either morphology or host protein content that have little consequence with respect to growth in cultured cells. Alternatively, the roles played by these proteins may be substituted by other substrates of Dot/Icm, or some of these proteins may be important for growth in some untested host cell.

It is common that a bacterial pathogen codes for numerous effectors but the presence of multiple paralogs of a specific translocated effector in the same organism is only occasionally found (28–30). The close similarity among proteins in a family points to functional redundancy, which may provide an explanation for the failure to identify these proteins in previous genetic screens for bacterial mutants defective in intracellular growth. In the case of the sidB family, however, functional redundancy clearly extends beyond the identified paralogs, and proteins of little similarity in sequence or function may be redundant. It is plausible that each paralog may be adapted to promote growth only in specific host cells. Because L. pneumophila is a versatile pathogen that interacts with very diverse hosts in the environment, the establishment of a successful intracellular replication niche may require only a subset of translocated substrates, with a single subset adapted for a particular host cell type.

Sorting out the minimal complement of translocated proteins necessary for intracellular replication will be a challenge for the future, because the number of such proteins may be quite large, based on the data displayed in Table 1. In addition to the products of the genes identified in the original Cre/loxP assay and sdeC, we have tested four paralogs of Sid proteins for interbacterial transfer, and all can promote transfer of Cre (data not shown). Based on this finding, we speculate that it is likely that all of the paralogs are capable of translocation. Furthermore, in another study we found an additional family of four proteins that can be transferred (S. M. VanRheenen, Z.-Q.L., and R.R.I., unpublished results). Altogether, this brings the minimum number of translocated proteins encoded by L. pneumophila to 24. In fact, there are probably many more proteins translocated by Dot/Icm, as the DotF interaction screen was not carried to saturation. The gene bank required in-frame fusions to a Sau3AI restrictions site, which may not be present in all genes encoding translocated substrates. Furthermore, the screen described here focused on only the subset of DotF interactors, and it appears that not all translocated substrates bind DotF at detectable levels in this assay.

The other salient feature of the Dot/Icm translocation system that was uncovered here is the large size of the proteins that were translocated. Many of the proteins identified in this study are >90 kDa, with one predicted to be >240 kDa. It appears likely that TFSS have evolved to transport large molecules, such as DNA–protein complexes, and these transfer systems may be the preferred translocators for high molecule weight substrates.