Considering that the strain radiation yielding the present-day chlamydial serovars likely occurred over millions of years 26, the use of reference strains is an accurate strategy as they were isolated only a few decades ago. Thus, in this evolutionary survey, we used the traditional reference strains that represent all 15 C. trachomatis serovars. We selected 51 polymorphic loci (approximately 51,000 bp) dispersed throughout the chromosome (Figure 1; Additional data file 1) that represent the following loci categories: 16 intergenomic regions (IGRs); 16 genes encoding cell envelope proteins (CEPs); 13 housekeeping genes (HKs); and 6 genes encoding hypothetical or unclassified proteins (HPs) (Additional data file 2). In order to evaluate the degree of polymorphism of these loci in comparison with the whole chromosome, we used the data generated from two of the five fully sequenced genomes, A/Har13 (ocular) 23 and D/UW3 (epithelial-genital) 21. We observed in the studied 51 loci a global mutation rate 14.3-fold higher than in the remaining chromosome regions (Fisher's exact test, P < 0.001). Moreover, we found 1,099 SNPs in these 51 loci between A/Har13 and D/UW3, which is greater than 200-fold more than what has been studied to date through concatenation 27, and comprises about 33% of the whole chromosome SNPs, indicating that our results could be scaled up to the full-chromosome level.

Additionally, a global overview of GC content revealed a mean value for all loci categories (data not shown) that is similar to the total mean GC content of approximately 41% observed for the fully sequenced genomes 21232425 with a standard deviation of 2.9%, which is not indicative of any putative HGT event.

We used phylogenomics to correlate each individual locus with tissue-specific strain radiation. Only four (25.0%) CEPs (incD, incE, pmpF and pmpH) and one (6.3%) IGR (incD/incE) comprehensively grouped the strains according to their cell-type/organ appetence (that is, revealed a larger evolutionary distance between strains with different niche appetencies than between strains infecting the same niche; Figure 2a). This clustering seems to be associated with loci revealing a higher p-distance-based polymorphism (Mann-Whitney P = 0.025). A full segregation by cell-type/organ appetence was not seen for most of the remaining CEPs due to the heterogeneity among the genital strains, where serovars E and F frequently form a separate cluster for 62.5% of CEPs (Figure 2a). Globally, 77.6% of loci belonging to different functional categories grouped strains that invade the lymph nodes as an individual cluster (LGV cluster), and the clustering of strains infecting the ocular tissue (ocular cluster) was also frequent. As above, we identified a significant association between a higher absolute number of SNPs and both the occurrence of a LGV cluster and an ocular cluster for each locus (Mann-Whitney P = 0.037 and P = 0.045, respectively). Interestingly, from the loci that better illustrate adaptation to lymph nodes, 80% of HPs and 53% of CEPs, compared with only 29% of HKs, show >50% non-synonymous SNPs (Figure 2b). Considering the DNA replication process, all SNPs on one strand that may imply strain segregation will also have the same impact on the other DNA strand. However, from the 51 loci that we used, only 4 pairs of loci overlap and the overlapping region never exceeds 10 bp (data not shown), which makes this effect negligible. Overall, these results suggest that the distinct genetic variability of strains infecting a specific cell-type/organ likely reflects an evolutionary adaptation process.

By performing intra-locus analysis, we observed that three HPs (CT049, CT144 and CT622) and two IGRs (rs2/ompA and ompA/pbpB) revealed distinct domains in which SNPs are concentrated, instead of being randomly distributed, and are associated with strains that infect a specific cell-type/organ (Figure 3). For these HPs, the SNP domains correspond to clusters of amino acid changes in the protein sequence (data not shown), mirroring the previous findings for some polymorphic membrane protein genes 28. Unfortunately, there is no assigned role for these open reading frames, which rules out any speculation about the functional implications of these specific clustered amino acid alterations. Nevertheless, this tissue-specific amino acid clustering points to a targeted fixation of mutations that may reflect the host-pathogen specific interaction within each organ.

We evaluated the nucleotide sequence variation in each concatenated loci category (Table 1). We highlight the multi-loci concatenation approach as a powerful tool to generate robust phylogenomic inferences, even when individual loci have evolved with different substitution patterns 293031. Overall, the HPs exhibit the highest number of variable sites (10.3%), whereas the HKs are the least variable (3.3%), which is supported by the mean p-distance values. Curiously, the IGRs show polymorphism similar to the CEPs. Globally, concatenation of all 51 loci yielded a 'super' sequence of up to 51,074 bp for each of the 15 reference strains, showing a mean of 1,032.1 (standard error (SE) 17.2) nucleotide differences.

Due to the speed and efficiency of the neighbor joining (NJ) method in inferring large phylogenies 3233, we used this approach on concatenated data. The NJ phylogenies inferred from the four concatenated loci categories (Additional data file 3) are consistent with most of the respective individual loci trees. Although only the CEP category clearly segregates strains by the disease they cause, the other categories show a notable segregation of at least one disease group, suggesting that heterogeneous loci categories are involved in the arms race process. The global phylogenetic tree presented in Figure 4 (where each taxon is represented by about 50,000 bp) reveals the putative final picture of C. trachomatis's evolution, showing strain grouping according to the cell-type (epithelial and lymph cells) and organ (eyes and genitalia) that they infect. These distinct segregations are supported by maximum bootstrap values (99-100%) in the nodes that separate disease groups, reinforcing that the targeted and distinct fixation of nucleotide changes on strains infecting a specific cell-type/organ are likely adaptive and barely the consequence of genetic drift. In fact, the genetic distance matrix (Table 2) shows that all strains that preferentially infect the eyes revealed only 0.27% (SE 0.02%) differences among them, but shows a mean genetic distance 7.4- and 11.2-fold higher (corresponding to 983 (SE 20) and 1,484 (SE 42) nucleotides) to strains infecting the epithelial-genital and lymph node tissues, respectively. Also, the LGV strains differ by only 69 (SE 8) nucleotides, whereas their distance to the epithelial-genital strains is 1,226 (SE 34) nucleotides. A separate main branch involving all epithelial-genital strains was not comprehensively seen for any individual loci (except for the CEPs pmpF and pmpH; data not shown) due to the separation of E and F strains. Indeed, the latter has a mean genetic distance of 673 (SE 16) nucleotides to the other epithelial-genital strains (Table 2). Similar NJ tree topologies were obtained for the three models used to estimate evolutionary distances (Kimura 2-parameter (K2P), Jukes-Cantor or Tamura-Nei) as well as for the maximum parsimony method (data not shown), with only slight variations in the bootstrap values, which supports the robustness of these distinct arms race scenarios.

We also highlight the loci that most contribute to the final tree topology (Figure 4), as they may be relevant for the evolutionary adaptation to each specific niche. Among these loci, we have found either highly conserved or polymorphic loci for strains infecting the same cell-type/organ. The former may represent a step forward in the evolutionary process by revealing the final stages 1 of this tissue-specific adaptive evolution, while the latter may also be involved in pathogenic differences between strains infecting the same tissue 25. The most extreme case is given by the CEP pmpF, where all the strains that infect the lymph nodes are 100% similar but show a mean distance of 312 and 421 SNPs to strains infecting the epithelial-genital and ocular tissues, respectively. In contrast, the epithelial-genital strains reveal up to 129 SNPs among them (data not shown). Although less markedly, CT049 is polymorphic among the LGV strains but near 100% identical among the ocular strains.

Additionally, we identified loci that do not seem to have influenced adaptation to each niche, since they generate an incongruent strain-radiation (Table 3), and whose polymorphism may thus be a consequence of genetic drift. However, previous results have demonstrated the involvement of some of these loci (CT622, tsf, rs2 and pbpB) in the pathogenesis of trachoma 25. As expected because of the serovar multiplicity, the epithelial-genital group revealed a higher number of polymorphic loci, and, overall, these loci belong to different categories. In contrast, strains infecting the lymph nodes constitute the most homogeneous group.

In order to have a more complete picture of the evolution of the serovars, we studied the chromosomal occurrence of small insertion/deletion (indel) events, which are non-phylogenetic parameters. We observed 84 small indel events (from 1-43 bp) inside the global concatenated loci for all strains, which mainly occurred within the IGR and CEP categories (Additional data file 4). None of these events was found to disrupt the coding sequence of the respective loci, indicating the absence of gene decay in the studied regions.

For the global concatenated data, we estimated the evolutionary distances using the indel-based parameter γ 34, which computes the number of gap nucleotides per nucleotide site between those sequences, while SNPs are not considered. The γ-distances (Figure 5a) are highly concordant with phylogenomic analyses, showing high heterogeneity within the epithelial-genital strains, and remarkable homogeneity among the LGV strains. Also, they revealed a segregation of strains by their cell-type/organ appetencies, which supports the tissue-specific arms race scenario.

Analysis of the global phylogenetic tree (Figure 4) also shows that the two most prevalent genital serovars worldwide, E and F, are closely related and separated from the other epithelial-genital strains. This segregation is observed for the majority of loci, with the exception of the HPs (Figure 2a). From all these loci, 70% of CEPs show an amino acid replacement for >50% of SNPs, compared to only 20% of HKs (Figure 2b). Curiously, the most remarkable segregation of E and F was seen for two IGRs (rs2/ompA and yfh0_1/parB) and three HKs (karG, tsf and rs2) (Figure 4). Furthermore, for the still unclassified protein gene CT622 and for the IGR rs2/ompA, we observed a non-random distribution of SNPs that are present in serovars E and F but not in the other epithelial-genital strains (Figure 3c,d). Finally, the mean γ-distance from any epithelial-genital strain to serovar E or F was from 3.4-fold (between G and E/F) to 4.7-fold (between I and E/F) higher than the distance between E and F (Figure 5b), which supports this close relationship between the two most ecologically successful serovars.

We used the most common reference strains representing the 15 C. trachomatis serovars: A/Har13, B/TW5, Ba/Apache2, C/TW3, D/UW3, E/Bour, F/IC-Cal3, G/UW57, H/UW4, I/UW12, J/UW36, K/UW31, L1/440, L2/434 and L3/404. McCoy cell culture of all strains plated in T-25 cm² flasks was performed as previously described 50. At 48-72 h post-infection, elementary bodies were harvested, and DNA was extracted using QIAamp^® DNA Mini Kit (Qiagen, Valencia, CA, USA) according to the manufacturer's instructions. Serovar confirmation of each reference strain was performed using ompA genotyping with BLAST comparison of the available GenBank sequences.

A GenBank search was performed to look for genomic regions that had been sequenced for at least one C. trachomatis reference strain from each of the three disease groups. Up to 93 loci were found, comprising about 84,000 bp of the chromosome, and involving IGRs, HKs, HPs and CEPs. Only non-constant loci were selected (51 of the 93; Figure 1; Additional data file 2) for sequencing the other reference strains if their sequences were not available yet. Automated sequencing was performed as previously described 28. The DNA sequence data have been deposited in a public database ([GenBank: EU239694-EU239702], [GenBank:EU239705-EU239712], and [GenBank:EU247618-EU247753]). Primer sequences are given in Additional data file 5. For all strains, five types of concatenated sequences were created in a head-to-tail fashion: one for each loci category (IGRs, HKs, HPs and CEPs) and a global concatenated sequence involving all loci (approximately 50,000 bp for each taxon).

We used data from the fully sequenced genomes A/Har13 and D/UW3 for this evaluation. Thus, considering the 3,354 SNPs identified between these two genomes 23, we evaluated whether 1,099 SNPs restricted to the 51,074 bp analyzed in this study are overrepresented relative to the 2,255 SNPs found in the rest of the chromosome. We framed this as a contingency table (Table 4) with a restricted sequence of 1,519,042 bp for each strain (corresponding to the length of the D/UW3 chromosome), and we estimated P-values using the Fisher's exact test as well as the odds ratios with a 95% confidence interval.

For all individual loci and concatenated sequences, alignments of all strains were generated using LaserGene (DNASTAR, Madison, WI, USA) and MEGA 3.1 51. MEGA 3.1 was also used to create matrices of pairwise comparisons and to estimate the number of variable sites, the number of parsimony informative sites and overall mean genetic distances. The pairwise-deletion option was chosen to remove all sites containing missing data or alignment gaps from all distance estimations, only when the need arose and not prior to the analysis.

In order to search for distinct regions that may be associated with strains belonging to a specific disease group, SimPlot 3.5.1 52 was used on all 51 loci. For each similarity plot, serovars were grouped according to the cell-type/organ that they infect, and nucleotide pairwise distances were calculated using the K2P method (gaps excluded; ts/tv of 2.0) in a sliding window size of 160 bp moved across the alignment in a step size of 10 bp. Additionally, for all loci where serovars E and F clustered apart from the other epithelial-genital serovars was observed, a SimPlot analysis was also performed to evaluate if the E/F nucleotide differences compared to the other genital serovars were clustered in specific domains of each locus.

To evaluate the existence of foreign genetic material 53, SWAAP 1.0.2 54 was used to calculate the percentage of GC content for all sequences of each taxon.

Prior to the phylogenetic reconstructions, and in order to select the appropriate evolutionary models, we evaluated the homogeneity of substitution patterns between sequences by calculating the Monte Carlo test-based Disparity Index per site 55. This gives the probability of rejecting the null hypothesis that sequences have evolved with the same pattern of substitution. The NJ method 56 was used with K2P 57, Jukes-Cantor 58 and Tamura-Nei 59 models to generate phylogenies. For the concatenated sequences, in order to examine the accuracy of the major conclusions reached from the NJ analysis, trees were also constructed under the maximum parsimony criterion 60, using the max-min branch-and-bound algorithm.

Considering that recombination disturbs a phylogenetic signal since the two parts of the recombined region may have different evolutionary histories 6162, one locus (ompA) was excluded from the phylogenetic concatenated analysis, as its highly recombinant nature has already been demonstrated 4863. The use of outgroup sequences was discarded in the present study because no rooted trees were needed to achieve the objectives defined above. Also, the most suitable strain for use as an outgroup, C. muridarum (MoPn strain), has several loci that vary greatly in size and diverge from those in the C. trachomatis strains, which would entail the removal of a huge portion of the sequences being analyzed.

We used a non-phylogenetic method for estimating the evolutionary distance between each pair of homologous DNA sequences, which is given by the parameter γ 34:

γ = -2log_eP

where

P = n_xy/√n_xn_y

where n_xy is the number of nucleotides shared by the two sequences, and n_x and n_y are the number of nucleotides of each sequence. For comparative purposes, we used the same set of loci as for the phylogenetic concatenated analyses, that is, we excluded the recombinant ompA gene. The γ variability was estimated by Monte Carlo using the alignments of each individual locus through the statistical platform R 2.5.1 64. Each time, 20 loci were randomly selected with replacement and γ-distances were calculated by repeating this procedure 50 times.

The statistical association between genetic and phylogenetic variables was performed using the ANOVA test by comparing groups' population means. We considered as genetic variables the overall mean values of percent GC content, p-distance and absolute SNPs obtained for each of the selected loci. The phylogenetic variables were: clustering of strains according to tropism properties; co-segregation of E/F strains; segregation of a LGV cluster or an ocular cluster; and the 'weight' of each locus in the final concatenated tree. The homogeneity of variances was tested using the Levene's test. Whenever the hypothesis of homogeneity of variances was rejected, the non-parametric Mann-Whitney test was used to compare distributions among groups. A P-value of 0.05 or less was considered significant.