We developed GenComp for the visualization of gene order structures and pseudogene relationships across multiple closely related genomes. The program was applied to a comparative analysis of eight Rickettsia genomes and a closely related outgroup, O. tsutsugamushi, plus the more distantly related outgroup W. pipientis. We first predicted open reading frames (ORFs) with the aid of Glimmer 22 using similar settings for all genomes. Homolog identification across the seven most closely related Rickettsia genomes (excluding R. bellii) was accomplished by basic local alignment search tool (BLAST) searches 23 followed by clustering with the aid of Tribe-MCL 24. A total of 9,450 Glimmer-predicted ORFs were clustered into 2,940 homologous gene families when applying the length ratio criteria 0.80 for homologous groups, with up to 59 genes in each family, including 359 single-gene families.

Information about gene location was considered with the aid of the visualization component of GenComp to identify the final set of positional homologs. This resolved many of the clusters with large numbers of homologs into groups of true orthologs that are conserved in two to six species or, in the full set, seven species. The conserved orthologs were fused into 86 metaclusters (segments with conserved gene order structures). These represent from 84% (R. felis) to 93% (R. prowazekii) of the Rickettsia genomes. Present in the metaclusters were 665 gene families with a cluster size of seven, representing single-copy genes that are conserved in sequence across all seven taxa, using 80% as the length cut-off value. We refer to the positional orthologs present in all seven species as the 'R7 core genes'. In total, this set comprises 688 genes, which accounts for 62% of the TG genomes and 47% to 56% of the SFG genomes. The mean size of the R7 core genes in Rickettsia was 1,006 to 1,010 bp (median size 850 to 860 bp) per species (typical of bacterial genes; Table 1).

We identified a combined total of 658 Rickettsia intergenic segments, defined as sequences flanked by orthologs in the R7 core gene set. These segments represent from 30% of the TG genomes to 38% of the R. felis genome. Glimmer-predicted ORFs located inside Rickettsia variable segments (RVSs) and present in one to six species, or present in all seven species but differing by more than 20% in size, are here referred to as the 'strain-variable ORFs', irrespectively of their origin, size, and functional status. An ORF-cluster was defined as a set of positional homologs conserved across two or more species. Paralogs located in different genomic regions were manually sorted into separate ORF-clusters. In total, 1,160 ORF-clusters were predicted in 304 of the 658 RVSs; these were used as the starting point for all subsequent analyses, as schematically outlined in Figure 1.

The ORF-containing segments ranged in size from a mean of 510 bp to 854 bp per species (median sizes: 143 bp to 186 bp) and contained on average 3.82 ORF-clusters per RVS over the seven species. The longest RVS (reg_id 546) was 19,665 kilobases and contained 29 ORF-clusters in R. felis. The sizes of the strain-variable ORFs were found to be roughly proportional to the prevalence of the ORF across the various species. Thus, strain-variable ORFs identified in six members exhibited a mean gene size of 976 bp (median 744 bp), whereas the mean size of strain-variable ORFs solely present in two species was only 332 bp (median 219 bp), which is only 30% of the R7 core gene size (Table 1).

The nonsynonymous substitution frequency (dN) ranged from 0.5 to 6.3 × 10^-2 substitutions per site for the R7 core genes (Figure 2a). The synonymous substitution frequency (dS) values were more than tenfold higher than the dN values in all pair-wise comparisons, which is indicative of purifying selection. In comparison, the dN values for the strain-variable ORFs were much more variable, ranging from 0.5 to 18 × 10^-2 nonsynymous substitutions per site (Figure 2a). Median dN values were inversely related to the prevalence of the ORF across the seven species and approached the dS values in some pair-wise comparisons that included only two or three species. A difference between the R7 core genes and the strain-variable ORFs in the seven-ortholog clusters was observed even if only ORFs that are more than 1 kilobase in size were included (Figure 2b). The smaller size and higher substitution frequency suggests that many strain-variable ORFs, particularly those present in a limited set of species, have evolved as pseudogenes.

With the aid of the visualization component of GenComp, we produced graphical images of the positions and relative sizes of all strain-variable ORFs in the 304 ORF-containing RVSs (Figure 3; see Additional date file 1 for graphical images of all RVSs). Note that many of the individual strain-variable ORFs represent fragments of the same pseudogene, broken up by indels and stop codons into multiple short ORFs. To follow the gene deterioration process in detail, we implemented a digital code to track sequence similarity across the strain-variable ORFs, with the first digits being identical for homologs within and across genomes and the latter two numbers representing a size index such that ORFs that are more than 80% similar in length are given the same number. A total of 482 of the 1,160 ORF-clusters produced significant hits (E < e^-10) to genes in the National Center for Biotechnology Information (NCBI) database other than of the seven Rickettsia genomes used to identify the variable segments. We selected all Rickettsia ORFs for which multiple gene alignments with homologs in other species could be created for an in-depth analysis; this amounted to 469 of the 482 strain-variable ORF-clusters and 681 of the 688 R7 core genes.

To illustrate the fragmentation patterns, we sorted the strain-variable ORFs into four sets depending on the different species distribution profiles (Table 2). One set of 75 ORF-clusters contained homologs across all seven species, as exemplified in Figure 3a. Many strain-variable ORFs in this set were only weakly mutated (slightly shorter with only a few indels or internal termination codons compared with their full-length homologs in other species), suggesting that they may encode functional or semifunctional gene products in some species. Another set encompassed 31 strain-variable ORFs in 18 segments (Figure 3b), including 27 ORF-clusters present in the TG plus some but not all members of the SFG, as well as four ORF-clusters uniquely present in the TG. A third set of clusters included strain-variable ORFs that are present in members of the SFG but not in the TG (Figure 3c). This was the largest set, including 215 ORF-clusters, 76 of which have homologs in all members of the SFG. Another 28 strain-variable ORFs were solely present in R. felis and R. akari, and 26 ORFs only in R. conori, R. sibirica, and R. rickettsii. Visual inspection of the erosion patterns in this set provides many examples of how a long ORF in one species, typically R. felis, has been disrupted into numerous short sequence fragments in the other species. The final set included 148 strain-variable ORFs identified in only a single Rickettsia sp., 115 of which were solely present in R. felis; many of these encoded transposons and other mobile elements (Figure 3d).

The strain-variable ORFs were then mapped onto the species phylogenetic tree to estimate the node at which it had been vertically inherited (Figure 4). To this end, core genes and strain-variable ORFs were separately grouped into different phylogroups. The R7 class contained ORFs without homologs in R. bellii, O. tsutsugamushi, or W. pipientis; the R8 class had homologs in R. bellii only; the RO class also in O. tsutsugamushi; and the ROW* class in O. tsutsugamushi and/or W. pipientis (E < e^-10; see Additional data file 2). The results of this categorization revealed a dramatic difference between strain-variable ORFs and core genes, as summarized in Figure 4. Thus, 72% of the strain-variable ORFs were placed in the R7 and R8 classes, 9% in the RO class, and only 19% belonged to the ROW* class. The converse pattern was observed for the core genes; only 21% belonged to the R7 and R8 classes, and 62% traced back to the ROW* ancestor.

The extent of degradation was different for strain-variable ORFs placed at different nodes in the tree (Table 2). In brief, strain-variable ORFs in the ROW* class were often weakly mutated and normally present in all or most Rickettsia spp., whereas strain-variable ORFs in the R7 and R8 classes tended to be heavily degraded or only present in a single Rickettsia sp. For example, 32% of the 89 strain-variable ORFs in the ROW* class were present in all seven Rickettsia spp., whereas only one strain-variable ORF in the R7 phylogroup had homologs in all species. The RO class also contained a high proportion, 34%, of strain-variable ORFs with homologs in all Rickettsia spp., whereas strain-variable ORFs placed in the R8 phylogroup exhibited a more scattered species distribution pattern. Taken together, 71% of the strain-variable ORFs solely present in the SFG belonged to the R8 class, although they represent no more than 46% of the ORFs overall. The R7 class was even more biased in that 52% of the 48 strain-variable ORFs in the R7 phylogroup were members of a single species and another 40% were present solely in the SFG.

With the aid of a classification scheme based on clusters of orthologous groups of proteins (COGs)-based classification scheme, we analyzed the distribution of functional categories for genes and pseudogenes belonging to different phylogroups (Tables 3 and 4). The 89 strain-variable ORFs in the ROW* class exhibited a broad functional distribution profile according to COG classification (Table 3), much like the core genes inherited from the ROW* ancestor (Table 4). Both strain-variable ORFs and core genes in the ROW* phylogroup exhibited a relatively high abundance of genes in categories such as translation, replication, and energy production, as observed previously 17.

The R7 and R8 phylogroups showed strong functional bias. Interestingly, the large majority of the strain-variable ORFs (69%) and the core genes (54%) in the R8 phylogroup represented unknown or poorly characterized genes. Although these hypothetical ORFs may include gene prediction errors, a similar fraction of unknowns was observed in the core genes, suggesting that they represent current or ancestral genes of unknown function. Three functional categories - 'cell wall biosynthesis', 'replication', and 'defense mechanisms' - dominated among the rest of ORFs in these phylogroups. For example, 32 of the 146 core genes in the R8 class represent components of the cell wall that were possibly shed from Orientia and Wolbachia 17. A total of 20 strain-variable ORFs belonged to this category, ten of which were placed in the R8 and three in the R7 phylogroup. These include genes for lipopolysaccharide biosynthesis, which may have been acquired at the base of the Rickettsia lineage.

The large majority of strain-variable ORFs in the R7 and R8 classes represents mobile genetic elements and their associated genes, many of which were classified into the COG categories 'replication and repair' or 'defence mechanisms'. Altogether, these two categories contained 44 strain-variable ORFs, which account for 30% to 50% of strain-variable ORFs with a COG-based functional assignment in the R7 and R8 classes, respectively. These genes include transposases, phage genes, plasmid genes and genes for DNA helicases, RNA helicases, and different types of DNA restriction-modification enzymes. Many more mobile genes were identified in the Rickettsia genomes, but were either unclassified or represented poorly characterized genes, or were categorized in the RO or ROW* classes because of the presence of distant homologs in O. tsutsugamushi and W. pipientis. The latter include numerous strain-variable ORFs for ankyrin repeat and TPR repeat proteins that are normally associated with conjugative transfer elements in O. tsutsugamushi 8 and phage genes in W. pipientis 14.

To search systematically for remnants of mobile genetic elements, we extracted already identified phage and conjugative transfer genes in the individual genomes and BLASTed these against all of the other genomes. Genes for conjugative transfer encoded by the tra operon have been identified on the R. felis plasmid as well as on the chromosomes of R. felis, R. bellii, and O. tsutsugamushi. Using the tra operon as the query, we analyzed each genome for remnants of such genes. No evidence of conjugative transfer genes or remnants of such genes was observed in any of the other Rickettsia genomes nor in W. pipientis. The explanation may be that the tra gene cluster has been horizontally transmitted into or across R. felis, R. belli, and O. tsutsugamushi. Indeed, a phylogenetic analysis of the tra genes present in R. bellii, R. felis, and O. tsutsugamushi showed that the order of divergence was different from the species divergence pattern (data not shown), as was also observed in a recent survey of the Rickettsia massiliae genome 25.

We also observed phage-related genes of the HK97 family in a few RVS with multiple strain-variable ORFs. For example, flanking a gene encoding a HK97 phage portal protein in RVS-308 were five duplicated genes putatively encoding cell surface antigens in R. felis. In RVS-552 we identified a long stretch of 13 ORFs in R. felis, all of which have homologs in R. bellii and six of which (including a copy of the gene for the HK97 phage major capsid protein) have homologs also in O. tsutsugamushi and W. pipientis. Interestingly, not only the sequence but also the order of genes was preserved. None of these six genes, or remnants thereof, was present in any of the other Rickettsia spp. Although no intact prophage was found, the identification of phage genes in the variable Rickettsia segments suggests that bacteriophages are circulating in the Rickettsia population.

Finally, we examined in greater detail the 48 strain-variable ORFs identified in the R7 class, which represents putative horizontal gene transfers into individual Rickettsia spp. and clades, although it cannot be excluded that they were acquired at the base of the Rickettsia lineage followed by species-specific losses. These include a long stretch of genes in RVS-422 that were solely identified in R. prowazekii and encode resolvase-like proteins, transposases, and ankyrin-repeat proteins. Also present in the TG were multiple genes encoding proteins with glycosyltransferase domains in RVS-626, one of which represents the 5' half of a longer gene with two such domains in the other Rickettsia spp. None of these have a close homolog in the Rickettsiales, but were related to glycosyltransferases in distantly related species such as Geobacter spp. and Vibrio cholerae. The genes RP336/RT0325 and RP337/RT0326 have adjacent homologs in V. cholerae.

The SFG and transitional group clades shared gene remnants for aminoglucoside phosphotransferase and acyltransferases. In addition, these clades had unique strain-variable ORFs associated with mobile elements, which encoded products such as lyase, DNA-damage inducible proteins, type I restriction-modification enzyme, transposase, and plasmid stabilization proteins. Twenty strain-variable ORFs in the R7 group were only present in R. felis; these included an exochitinase, a biotin synthase gene bioB, and mobile elements such as a mutator-type transposase (similarity to Psychrobacter), which was located in six different RVS fragments. The only strain-variable ORF with remnants in all species was the metK gene, which encodes S-adenosylmethionine synthetase.

To investigate how the differences in functionality relate to the putative source of the sequences in the R7 and R8 phylogroups, we inferred the closest relatives from the taxonomic classification of the most similar sequences, excluding Rickettsia, Orientia, and Wolbachia (Table 5). The R7 class was dominated by proteobacteria-like sequences, with 17 out of 35 ORFs showing similarity to γ-proteobacteria. Overall, the proteobacteria-like sequences accounted for 73% of strain-variable ORFs in the R7 class, and 11% to 12% of strain-variable ORFs and core genes in the R8 phylogroups. The lower proportion of proteobacteria-like sequences in the R8 classes is due to the high numbers of strain-variable ORFs (60%) and core genes (38%) that are specific to Rickettsia with no identifiable homolog outside the genus. A total of 80 strain-variable ORFs and 80 core genes in the R8 class exhibited sequence similarities to proteobacteria. However, the ratio of ORFs with the highest sequence similarity of α-proteobacteria versus γ-proteobacteria was strikingly different: only 0.6 for the strain-variable ORFs versus 3.2 for the core genes.