In Vitro Recombinants of Antibiotic-Resistant Chlamydia trachomatis Strains Have Statistically More Breakpoints than Clinical Recombinants for the Same Sequenced Loci and Exhibit Selection at Unexpected Loci

Abstract

Lateral gene transfer (LGT) is essential for generating between-strain genomic recombinants of Chlamydia trachomatis to facilitate the organism's evolution. Because there is no reliable laboratory-based gene transfer system for C. trachomatis, in vitro generation of recombinants from antibiotic-resistant strains is being used to study LGT. However, selection pressures imposed on in vitro recombinants likely affect statistical properties of recombination relative to naturally occurring clinical recombinants, including prevalence at particular loci. We examined multiple loci for 16 in vitro-derived recombinants of ofloxacin- and rifampin-resistant L₁ and D strains, respectively, grown with both antibiotics, and compared these with the same sequenced loci among 11 clinical recombinants. Breakpoints and recombination frequency were examined using phylogenetics, bioinformatics, and statistics. In vitro and clinical isolates clustered perfectly into two groups, without misclassification, using Ward's minimum variance based on breakpoint data. As expected, gyrA (confers ofloxacin resistance) and rpoB (confers rifampin resistance) had significantly more breakpoints among in vitro recombinants than among clinical recombinants (P < 0.0001 and P = 0.02, respectively, using the Wilcoxon rank sum test). Unexpectedly, trpA also had significantly more breakpoints for in vitro recombinants (P < 0.0001). There was also significant selection at other loci. The strongest bias was for ompA in strain D (P = 3.3 × 10⁻⁸). Our results indicate that the in vitro model differs statistically from natural recombination events. Additional genomic studies are needed to determine the factors responsible for the observed selection biases at unexpected loci and whether these are important for LGT to inform approaches for genetically manipulating C. trachomatis.

INTRODUCTION

Chlamydia trachomatis is the most prevalent sexually transmitted bacterium and the leading cause of preventable blindness in the world. More than 92 million cases are reported annually (42). Infection leads to diseases such as urethritis, cervicitis, pelvic inflammatory disease, infertility, proctitis, reactive arthritis, lymphogranuloma venereum, and trachoma. Over 70% of women and 50% of men who are infected with C. trachomatis are asymptomatic, providing a huge opportunity for unchecked transmission (23, 42). For symptomatic individuals who do seek medical care, diagnostic tests for C. trachomatis, either for initial diagnosis or test of cure, are not always performed, largely due to cost (8). Thus, empirical treatment with antibiotics is common, which may hasten the development of drug resistance. This has already occurred for Chlamydia suis, a species that is closely related to C. trachomatis in which a tetracycline transposon was acquired from other pathogens residing in the guts of pigs (13). C. trachomatis may not be far behind; a number of studies have shown increased MICs of various antibiotics for clinical C. trachomatis isolates or reference strains propagated in tissue culture under selective antimicrobial pressure (2, 3, 26, 33–35).

In addition to the example of lateral gene transfer (LGT) in C. suis, a growing body of literature points toward recombination as a major mechanism by which C. trachomatis has advanced its evolutionary reproductive fitness and ability to evade the host immune response (14, 16–18, 20, 21, 25, 37). It is well accepted that, among asexual populations evolving in a changing environment, deleterious mutations increase without bound until they reach an intolerable level that would presumably cause population extinction, referred to as Muller's ratchet, unless rescued by recombination (27, 28). Agent-based simulations of finite populations have shown good agreement with analytical theory, where an increased mutation rate is strongly correlated with decreased fitness to the point of population collapse in the absence of recombination (15). These considerations explain why chlamydiae must include recombination, despite their intracellular lifestyle, in order to survive.

The study of the biology of chlamydiae is complicated by their obligate intracellular development and the lack of a routine laboratory system for directed mutagenesis or reliable gene transfer. However, knowledge of the mechanism(s) that chlamydiae utilize to undergo recombination would provide the necessary insight to presumably effectively genetically manipulate the organism. Toward this goal, investigators have developed an in vitro system to model in vivo LGT in chlamydiae. DeMars et al. were the first to culture different antibiotic-resistant D (rifampin) and L₁ (ofloxacin) strains of C. trachomatis to produce recombinant progeny that were doubly resistant when grown in media containing both antibiotics (11). Suchland et al. subsequently produced recombinants of various antibiotic-resistant strains of C. trachomatis, C. suis, and Chlamydia muridarum, including cross-species recombinant events (38). It is not clear whether recombinants created by antimicrobial pressure have the same breakpoint location and frequency of LGT events, or even necessarily the same mechanisms of recombination, compared with those circulating among human populations. Therefore, we examined multiple genomic loci for 16 in vitro-derived recombinants of ofloxacin- and rifampin-resistant strains and compared these with the same loci among 11 naturally occurring clinical recombinants to determine to what extent the properties of the in vitro model reflect natural recombination events.

MATERIALS AND METHODS

Reference strains and clinical isolates of C. trachomatis.

The following 16 reference strains of C. trachomatis were used in this study: A/HAR-13, B/TW-5, D/UW-3, Da/TW-448, E/Bour, F/IC-Cal3, G/UW-57, H/UW-4, I/UW-12, Ia/UW-20, J/UW-36, Ja/UW-92, K/UW-31, L₁/440, L₂/434, and L₂b/UCH-1/proctitis. In addition, 11 urogenital samples were obtained from patients visiting sexually transmitted disease clinics. These samples represent ompA genotypes D, F, J, Ja, L₂, and L₂b and were identified as recombinants either by multilocus sequence typing (MLST) (9) or whole-genome sequencing (37). Briefly, both the reference and clinical strains were propagated in HeLa 229 cells, and the elementary bodies were harvested and purified from host cells via discontinuous density centrifugation in Renografin as we described previously (9, 37). The high-pure PCR template preparation kit (Roche Diagnostics) was used to extract genomic DNA following instructions provided by the manufacturer. The ompA genotype of the 16 reference strains and 11 clinical strains was confirmed by comparison with sequences in GenBank using BLAST (www.ncbi.nlm.nih.gov/BLAST/) as we described previously (36).

Selection and sequencing of specific genomic regions.

Ten previously sequenced genomic regions of 16 clones of in vitro-derived recombinants of rifampin-resistant reference strain D (strain D:Rif^r-1, which has a mutation in the rpoB gene) and ofloxacin-resistant reference strain L₁ (strain L₁:Ofx^r-1, which has a mutation in the gyrA gene) (10) deposited in GenBank were analyzed (n = 160 sequences; 132,096 bp [8,256 per clone]). Similar regions including the recF, incA, trpA, trpB, gyrA, rpoB, murA, rs2, and ompA genes and intergenic regions (IGRs) between rs2 and ompA and between ompA and pbpB were selected for sequencing the 11 clinical isolates (n = 121 sequences; 159,149 bp [14,454 to 14,481 per isolate]). PCR and sequencing primers (see Table S1 in the supplemental material) were designed based on available GenBank sequences of the respective genes for reference strain D/UW-3 (GenBank accession no. AE001273). Due to the large size of rpoB (3,759 bp) and gyrA (2,510 bp), sets of PCR primers were designed to generate overlapping amplicons for the entire gene. Four thermocycling profiles were used as shown in Table S2. The sequences of the clinical strains were subjected to BLAST and aligned to the corresponding gene or IGR of the 16 reference strains using MegAlign software (DNASTAR, Madison, WI) to identify single-nucleotide polymorphisms. The GenBank sequences for each in vitro clone were similarly analyzed in comparison to reference strains D/UW-3 and L₁/440. The least number of nucleotide mutations was used to determine the closest match to a strain for that gene or IGR.

Molecular evolution and phylogenetic analyses.

For each of the in vitro and clinical recombinants, the percent G+C content of the genetic loci was calculated using EditSeq software (DNASTAR, Madison, WI). The molecular evolution of the genes of interest were each evaluated using the Nei-Gojobori method (29) to estimate the ratio of nonsynonymous to synonymous substitutions (d_N/d_S) as we have described previously (37). Using the p-distance model, values were normalized against the number of potential d_N and d_S sites; 95% confidence intervals were used to determine significant differences in the mean d_N and d_S. One thousand bootstrap replicates were performed to calculate the standard error.

The evolutionary history of the genes of interest was reconstructed using a bootstrap test of neighbor-joining tree topologies generated by MEGA (39) where trees were created using the Kimura 2-parameter model, and reconstructions at the protein level were created utilizing the gamma rate distribution model as we described previously (37).

Phylogenetic networks were constructed using SplitsTree4 (19) (http://www.splitstree.org/) as we described previously (9). The network estimates the phylogenetic relationship between all 16 reference strains and all 11 clinical strains and was drawn according to the default settings. The network was created based on 1,000 bootstrap replicates.

Determination of genomic crossover regions and breakpoints.

Mosaic structures of the clinical isolates were proposed based on SimPlot analyses (http://sray.med.som.jhmi.edu/SCRoftware/simplot/) as we described previously (17). Briefly, the query sequence was compared with two potential parental strains and an outgroup strain using a window size of 100 bp with a 10-bp sliding step size to locate and determine breakpoints. Significant informative sites were determined using the maximum chi-squared test. Each informative site supports one phylogenetic tree based on the Kimura 2-parameter model with confidence levels for each branch in the tree calculated from 1,000 bootstrap replicates (17).

For each putative recombinant locus, a P value was calculated using Fisher's exact test for the best supported breakpoint region (i.e., between informative sites) based on the number of informative sites. A Bonferroni multiple test factor was conservatively applied for the number of ways to choose the observed number of breakpoints among the observed number of informative sites for the locus (this is a standard binomial coefficient). The resulting P value is the probability of observing the given pattern of informative sites supporting the two different ancestors in the absence of recombination.

Analysis of in vitro and clinical breakpoints.

To compare the pattern of recombination among the clinical and in vitro sequences, a table of breakpoints was generated from the pooled clinical and in vitro data. First, data for seven genes (recF, incA, trpA, gyrA, rpoB, murA, and ompA) and the intervening regions (int) between each gene were arranged based on gene order in the genome. This analysis was performed using complete (clinical samples) and partial (in vitro) sequence data for the genes. The intervening regions do not represent sequence data; they represent only breakpoints between genes. Second, for each of the 27 samples, a value of 1 was entered if a recombination breakpoint was present within the gene and a value of 0 was entered if no breakpoint was observed. In each intervening region, if both genes before and after the region have the same strain, then a value of 0 was entered. Otherwise, a value of 1 was entered, indicating that a recombination breakpoint resides somewhere within the intervening region.

To determine whether the clinical and in vitro samples have different patterns of recombination, cluster analysis was performed. The patterns are different if the 27 samples can be grouped into distinct clinical and in vitro clusters based on the presence or absence of recombination breakpoints within each region. Before doing the cluster analysis, the binary data described above were converted to a dissimilarity matrix of Jaccard distances.

For each pair of samples A and B, the Jaccard dissimilarity is given by the equation J′ = (M₀₁ + M₁₀)/(M₀₁ + M₁₀ + M₁₁) where M₁₁ is the total number of regions where A and B both have a value of 1, M₀₁ is the total number of regions where the value of A is 0 and the value of B is 1, and M₁₀ is the total number of regions where the value of A is 1 and the value of B is 0.

The Jaccard dissimilarity matrix is calculated by the DISTANCE procedure in SAS. The output is used as the input to the CLUSTER procedure using Ward's minimum variance method (33a) where the distance between two clusters is the analysis of variance (ANOVA) sum of squares between the two clusters added up over all the samples. The output is then displayed by using the TREE procedure. To determine the regions where the clusters have significantly different recombination patterns, a Wilcoxon score (rank sum) test from the NPAR1WAY procedure is then performed for each region from the output of the TREE procedure.

Nucleotide sequence accession numbers.

The sequences from this study have been deposited in GenBank under the following accession numbers: JN795254 to JN795264 for recF, JN795281 to JN795291 for incA, JN795308 to JN795318 for trpA, JN795335 to JN795345 for trpB, JN795362 to JN795372 for gyrA, JN795389 to JN795399 for rpoB, JN795416 to JN795426 for murA, JN795470 to JN795480 for rs2, JN795443 to JN795453 for ompA, JN795497 to JN795507 for the IGR between rs2 and ompA, and JN795524 to JN795534 for the IGR between ompA and pbpB.

RESULTS

Molecular evolution, phylogenetics, and recombination for the clinical recombinants.

The initial analyses were based on the 11 genomic regions sequenced for 11 clinical recombinants in comparison with 16 reference strains. The d_N and d_S mutations and d_N/d_S were analyzed for each gene as shown in Fig. S1 in the supplemental material. All values except for IncA were <1, which would indicate an overall conservation of the protein structure with purifying selection for those loci. For IncA, however, a value of >1 suggests that it is under evolutionary pressure. There was a large variation for the mean genetic distance for the genes; the least variant gene was trpB (9.0 nucleotides [nt], ±0.1% nt, 6.0 amino acids [aa], ±0.3% aa) while the most variant was ompA (336 nt, ±11.2% nt, 94.0 aa, ±8.9% aa) (Fig. S2). ompA codes for the immunodominant major outer membrane protein of C. trachomatis. There was no significant variation in the G·C content for the genomic regions compared across in vitro and clinical recombinants (data not shown).

Phylogenetic reconstructions for the nine genes, excluding IGRs, were constructed based on the sequences for the 16 reference strains and 11 clinical strains (see Fig. S3 in the supplemental material). The trees were incongruent with each other, which supports prior evidence from the MLST data (9) that the clinical strains are recombinants.

The concatenated phylogenetic analysis of the genetic loci as shown in Fig. 1 revealed distinct clusters by invasive (lymphogranuloma venereum [LGV] strains, lavender), prevalent noninvasive (non-LGV strains, green), and less prevalent noninvasive (non-LGV strains, blue) urogenital disease grouping in which clinical Ja strains formed a distinct branch from the reference Ja strain (Fig. 1). The bootstrap SplitsTree decomposition result was congruent with the clustering in the MEGA phylogenetic tree and also displayed wide evolutionary paths, as indicated by the network structure between strains (Fig. 2), further supporting recombinant events in the evolution of the clinical strains.

Fig 1.

Phylogenetic tree for C. trachomatis strains. The 16 reference strains and 11 clinical strains are represented in the phylogeny. The tree was constructed with the concatenated sequences of the nine genes and two IGRs for the above strains using the neighbor-joining method in MEGA with 1,000 bootstrap replicates (see Materials and Methods). The numbers at tree branches indicate the percent bootstrap value as the level of confidence for the node. The colors refer to strains by disease groupings: lavender, invasive urogenital disease (LGV); blue, noninvasive urogenital disease; green, prevalent strains associated with noninvasive urogenital disease; and yellow, trachoma. The scale bar at the bottom of the figure represents the number of substitutions per site.

Fig 2.

Phylogeny network interaction network for C. trachomatis strains. The tree was constructed using the concatenated sequences of the nine genes and two IGRs for the 16 reference strains and 11 clinical strains using SplitsTree (see Materials and Methods). Bootstrap replicates of 1,000 were performed to determine support for the tree. The colors refer to disease groupings: lavender, invasive urogenital disease; blue, noninvasive urogenital disease; green, prevalent strains causing urogenital disease; and yellow, trachoma. The scale bar represents the number of substitutions per site.

Genomic crossover regions and breakpoints for in vitro and clinical recombinants.

To determine putative crossover regions and breakpoints, the phylogenetic and BLAST analyses for each gene for the clinical samples were used to reveal putative mosaic structures. The right-hand column of Table S3 in the supplemental material displays the similarity for the closest hit to a reference strain for each clinical isolate. A graphical representation of the mosaics of the clinical recombinants is shown along with the in vitro recombinants in Fig. 3 based on sequence alignments (see Table S4 in the supplemental material for the genetic distance between clinical isolate and parental reference strains). Each colored arrow corresponds to a particular locus and the reference strain from which it is likely derived (see legend for Fig. 3). Ten of the 11 clinical isolates contain at least two breakpoint regions, as indicated by the black vertical arrows, which were confirmed by phylogenetic, SimPlot, and SAS analyses (see below). Open arrows indicate regions that could not be confirmed other than by comparative genetics. Interestingly, the clinical D43nl strain was identical to the reference strain D/UW-3 for the genomic regions sequenced, although it had previously been identified as a recombinant based on MLST where the sequences differed substantially in two housekeeping genes (9). Strains D2s, D83s, and D84s are recombinants of strains D, Da, and F with the exception of D2s where the parental strain is either H or Ia for the rs2 gene. Strain F38nl is derived from strains E, F, and Da. Strain J27s is a recombinant of strains J and Ia, while the clinical Ja strains are mosaics of strains Da, E, F, and Ja.

Fig 3.

Proposed chromosomal mosaic structure for the 11 clinical recombinants (top half) and the 16 in vitro recombinants (bottom half) of C. trachomatis. The genomic loci are represented by horizontal colored arrows, where the arrow direction signifies the coding strand direction of the gene. The color of the arrow denotes the reference strain from which the mosaic is most likely derived as follows: gray, strain D; green, strain Da; dark blue, strain E; orange, strain F; beige, strain I; yellow, strain Ia; purple, strain J; red, strain Ja; white, strain L₁; light blue, strain L₂; and lavender, strain L₂b. The black horizontal arrows for strains D2s and L₂b48nl denote ambiguity, as the parental strains may be H/UW-4 or Ia/UW-202 for rs2 or D/UW-3, Ia/UW-20, or K/UW-31 for murA, respectively. The boxes for the clinical recombinants represent the two noncoding IGRs, between rs2 and ompA, and upstream of ompA. The proposed crossover regions are indicated by vertical arrows determined by sequence alignments and statistical analyses (see Materials and Methods). The open arrows were confirmed by sequence alignments only. The roman numeral notation for the in vitro recombinant strains corresponds to the identification tags of the DeMars and Weinfurter (10) recombinants. Labels indicated in bold type and parentheses [e.g., (c10–4)] are the corresponding names of the in vitro clones in GenBank.

Using SimPlot with phylogenetic, bootscan, and maximum chi-squared test analyses to identify the location of and confidence in breakpoint regions, each recombinant genomic region was confirmed. We used a 2 × 2 matrix for each putative breakpoint locus populated by the number of informative sites supporting the two ancestors on either side of the breakpoint. From this matrix, Fisher's exact test was calculated (Table 1). The total number of informative sites (t) and the number of breakpoint regions (c) used in the analysis are shown. We applied Bonferroni's correction as the number of ways to distribute c in the intervals between t sites (t − 1 chooses c). Figure 4 illustrates the results for strains D83s, F38nl, and Ja41nl. The P values for the informative sites for the two breakpoints within the rs2 region, the IGR between rs2 and ompA, and the ompA gene for strain D83s were determined as shown in Table 1. Breakpoints or breakpoint regions between arrows could not be determined because there was no contiguous sequence between the genes. The first crossover region likely occurs within rs2 (P = 6.2 × 10⁻¹⁰), while the second is within the end of the IGR (flanked by rs2 and ompA) and the beginning of rs2 with a P value of 1.1 × 10⁻⁴⁶. For strain F38nl, the first breakpoint is within the end of the IGR (flanked by rs2 and ompA) and the beginning of rs2 (P = 1.0), while the second is within ompA with a P value of 1.1 × 10⁻⁶⁰. For strain Ja41nl, the crossover regions are the IGR flanked by rs2 and ompA (P = 1.0) and the IGR flanked by ompA and pbpB (P = 0.632). The observed pattern for Ja41nl may have occurred by chance, based on our overly conservative P values, because there are not very many informative sites, but our findings are still consistent with the breakpoints we annotated.

Table 1.

P value calculations for clinical recombinants based on SimPlot analyses

Clinical isolatea	n11b	n12	n21	n22	No. of informative sites (outgroup)	Crossovers (location)c	Fisher's exact test P value	No. of testsd	Bonferroni-corrected P value
D2s (L)	2	0	0	1	64 (B)	2 (trpB, positions 270 to 320)	0.33	2,016	1.0
D2s (R)	0	1	60	0	64 (B)	2 (trpB, position 565)	0.02	2,016	1.0
D83s (L)	2	0	0	37	91 (A)	2 (trpB, position 324)	0.00135	4,005	0.995
D83s (R)	0	37	52	0	91 (A)	2 (trpB, positions 848 to 886)	6.73 × 10−26	4,005	2.69 × 10−22
D83s (L)	20	0	0	33	500 (L2)	2 (rs2, position 525)	4.94 × 10−15	124,251	6.2 × 10−10
D83s (R)	0	33	447	0	500 (L2)	2 (IGR [between rs2 and ompA], positions 365 to 372, to rs2, positions 1 to 11)	8.84 × 10−52	124,251	1.1 × 10−46
D84s (L)	2	0	0	37	91 (A)	2 (trpB, position 324)	0.00135	4,005	0.995
D84s (R)	0	37	52	0	91 (A)	2 (trpB, positions 848 to 886)	6.73 × 10−26	4,005	2.69 × 10−22
D84s (L)	20	0	0	33	122 (L2)	2 (rs2, position 525)	4.94 × 10−15	7,260	3.6 × 10−11
D84s (R)	0	33	69	0	122 (L2)	2 (IGR [between rs2 and ompA], positions 343 to 372, to rs2, positions 1 to 2)	1.55 × 10−27	7,260	1.1 × 10−23
F38nl (L)	2	0	0	55	350 (I)	2 (IGR [between rs2 and ompA], positions 346 to 377, to rs2, positions 1 to 89)	0.00062657	60,726	1.0
F38nl (R)	0	55	293	0	350 (I)	2 (ompA, positions 926 to 1082)	1.88 × 10−65	60,726	1.1 × 10−60
J27s					10 (A)	1 (trpB or trpA, no breakpoint detected)
Ja26s					None (L2)	1 (incA, no breakpoint detected)
Ja26s (L)	38	0	0	149	190 (L2)	2 (ompA, positions 1016 to 1186)	1.39 × 10−40	17,766	2.584−38
Ja26s (R)	0	149	3	0	190 (L2)	2 (IGR [between ompA and pbpB], positions 447 to 608, to ompA, positions 1 to 51)	0.00000174	17,766	0.030
Ja41nl					None (L2)	1 (incA, no breakpoint detected)
Ja41nl (L)	1	0	0	105	108e	2 (ompA, positions 1071 to 1117)	0.00943	5,671	1.0
Ja41nl (R)	0	105	2	0	108e	2 (IGR [between ompA and pbpB], positions 518 to 551)	1.76 × 10−4	5,671	0.632
Ja47nl (L)	1	0	0	104	107e	2 (ompA, positions 1093 to 1191, to IGR [between rs2 and ompA], positions 1 to 377)	0.00952	5,565	1.0
Ja47nl (R)	0	104	3	0	107e	2 (IGR [between ompA and pbpB], positions 548 to 586, to ompA, positions 1 to 123)	5.04 × 10−06	5,565	0.038
L2b48nl	1	0	0	1	2 (K)	1 (IGR [between rs2 and ompA], positions 48 to 151)	1	1	1

^aThe letter in parentheses after the clinical isolate designation indicates the location of the crossover: L for crossover on the left and R for crossover on the right (for strains D83s, F38nl, and Ja41nl; see Fig. 4).

^bn11 through n22 refer to entries in a 2 × 2 contingency table; for example, there are xx informative sites in the first part of the region that favor one tree, and none that favor the other tree, so the first column of the 2 × 2 contingency table has a xx and a 0.

^cThe numbers in parentheses refer to the location in the gene based on the start codon of the reference strain for that clinical strain.

^dNumber of tests used to determine the Bonferroni-corrected P value. Bonferroni's correction (Sidak formula) was applied because the regions of sequence used to define the contingency tables were chosen based on the data, not in advance. Because of correlations in the multiple test hypotheses, this correction is overly conservative.

^eBecause the parental strains were Ja, E, and F, no outgroup could be used given the limitations of SimPlot.

Fig 4.

SimPlot representation of the proposed crossover regions for clinical isolates D83s (A), F38nl (B), and Ja41nl (C). The vertical lines in panels 2 and 3 correspond to the approximate breakpoint, while the nucleotide positions at the bottom of each figure correspond to alignment positions of the region analyzed. The genomic regions shown are from rs2 to ompA for isolate D83s, from rs2 to ompA for isolate F38nl, and from rs2 to the IGR upstream of ompA for isolate Ja41nl (panels 1). The number of informative sites shared by the recombinant sequence (black) and the parental sequence from which the mosaic is likely derived (red, green, or blue) are shown. Informative sites and supporting four-member trees were determined for each region bounded by a breakpoint (panels 2). SimPlot analysis depicting the similarity of the recombinant sequences (query) to their respective parental reference strains in a 100-bp window with a step size of 10 bp (panels 3). BootScanning analysis showing the phylogenetic relatedness between these sequences (panels 4). Neighbor-joining (NJ) trees constructed for each region bounded by a breakpoint. NJ trees based on the Kimura 2-parameter model and BootScan confidence levels were calculated for 1,000 replicates.

These data further support the same two hot spots for clinical recombination identified in our previous work (17). Clinical strains D83s, D84s, and F38nl had a similar hot spot for recombination in the rs2 and IGR (between rs2 and ompA) genomic regions. In addition, clinical strains F38nl, Ja26s, Ja41nl, and Ja47nl had the same hot spot in the IGR (between rs2 and ompA) and ompA regions.

Similar calculations were performed for the in vitro-derived sequences (data not shown). The nucleotide ranges within which the breakpoints occurred are shown in Table 2, and a graphical representation of the mosaics of the in vitro recombinants is shown in Fig. 3 based on sequence alignments.

Table 2.

Genomic crossover regions and specific breakpoints for in vitro and clinical recombinants^f

^aSixteen recombinant clones resulting from in vitro LGT (10).

^bClone sequences were retrieved from the GenBank database under the following accession numbers for the indicated genes: EU104987 to EU105002 for gyrA, EU105003 to EU105018 for gyrB, EU105019 to EU105034 for incA, EU105035 to EU105050 for murA, EU105051 to EU105066 for ompA, EU105067 to EU105082 for pmpC, EU105083 to EU105098 for recF, EU105099 to EU105114 for ribF, EU105115 to EU105130 for rpoB, and EU105131 to EU105146 for trpA. For the clinical strains, the identification number (ID no) is shown.

^cNucleotide positions of crossover regions.

^dSpontaneous mutation occurring at nucleotide 1400.

^eSpontaneous mutation occurring at nucleotide 249.

^fThe source of the genes is indicated by color as follows: gray, strain D; olive green, strain Da; dark blue, strain E; pumpkin-brown, strain F; beige, strain I; yellow, strain Ia; purple, strain J; red-brown, strain Ja; white, strain L₁; light blue, strain L₂; and lavender, strain L₂b.

We also calculated the P values for strain selection at each gene, using the same test each time (two-tailed Fisher's exact test), comparing the strain of the gene with the combined nearest neighbors on each side. We found that trpA (start [P = 0.000001014] and end [P = 2.9 × 10⁻⁹]), gyrA (start [P = 4.3 × 10⁻¹⁰]), gyrB (start [P = 0.0000805]), rpoB (start [P = 4.3 × 10⁻¹⁰]), and ompA (P = 3.3 × 10⁻⁸) were statistically significant but not recF, ribF, incA, gyrA (nucleotide position 237), pmpC, or murA. In analyzing a gene in this case, start refers to the beginning of the gene; the end (i.e., end of the gene) would be used as the neighbor unless the end and start agreed, in which case the next gene with data is used.

Comparison of in vitro and clinical breakpoints.

The sequences of the 16 recombinant clones resulting from in vitro LGT of parental strains D:Rif^r-1 and L₁:Ofx^r-1 were analyzed for mutations using BLAST. Table 2 shows homology to the parental strain for each of the 10 genes sequenced, although only partial sequences were available in GenBank for each gene. The sequences of the 11 clinical samples were similarly analyzed for nine genes and two IGRs and are shown below the in vitro recombinants in Table 2. Precise breakpoints and the range (denoted by number in parentheses that are the base pair boundaries of the breakpoint regions) within which the breakpoint occurred could be identified in trpA, gyrA, and rpoB for most of the in vitro clones (Table 2) and in incA, trpB or trpA, rs2, the rs2 IGR, ompA, and the ompA IGR for the clinical recombinants (Table 1).

The Ward's minimum variance clustering procedure separated the samples into clinical and in vitro clusters without any misclassification (Fig. 5). For the two clusters, the squared multiple correlation R² (the proportion of variance accounted for by the two clusters) was 0.32. As shown in the tree (Fig. 5), the clinical cluster can be further subdivided into two branches. These two clusters both contain samples of clinical D and L strains.

Fig 5.

Phylogenetic tree showing the hierarchical relationship between the in vitro and clinical recombinants calculated using the Jaccard coefficient as a distance metric (see Materials and Methods). The length of each branch represents the similarity in the pattern of breakpoint locations.

The Wilcoxon rank sum test (see Table S5 in the supplemental material) resulted in a P value of < 0.0001 for trpA, intragene region 4 (int4), gyrA, and int5, while the P value was 0.02 for rpoB. All other comparisons were insignificant, showing that this genomic region is where the recombination pattern is different between clinical and in vitro recombinants.

DISCUSSION

There is a growing body of research that supports naturally occurring intra- and interspecies recombination for numerous Chlamydia spp. (13, 14, 17, 18, 25, 30, 31). Despite an abundance of gene sequences and a number of whole-genome sequences, the mechanism(s) of LGT remains elusive. Without this knowledge, the development of a stable gene transfer system has been greatly hampered. In addition, C. trachomatis has been largely refractory to directed mutagenesis, although a recent study by Wang et al. (40) showed that plasmid-free C. trachomatis can be transfected with a chlamydial plasmid to restore glycogen synthesis. Thus, the current lack of reliable methods for genetic manipulation of C. trachomatis profoundly limits our ability to understand the function of many genes involved in virulence and host specificity.

In vitro studies of antibiotic-resistant strains have indeed yielded recombinant organisms that may be useful for evaluating recombination mechanisms (10, 11, 38). However, the results remain difficult to interpret, because while some clones have identical breakpoint regions (e.g., in gyrA), there are no common sites of crossover among recombinants produced by the same or different research groups using the same resistance markers (10, 11, 38). For example, the L₁:Ofx^r-1 and L₁:Rif^r-1 strains that were crossed in one study (11) resulted in recombinant regions in gyrA and rpoB that differed from recombinant regions of the D:Rif^r-1 and L₁:Ofx^r-1 crosses in a second study by the same group (10).

In the present study, we examined multiple genomic loci for 16 in vitro-derived recombinants from strains D:Rif^r-1 and L₁:Ofx^r-1 and compared these with the same loci from 11 clinical recombinants. We initially determined the phylogenies and then the precise regions within which breakpoints occur for nine genes and two IGRs among the 11 clinical strains that were previously identified as recombinants based on MLST (9). Up to five statistically significant breakpoint regions were identified in areas of contiguous gene or IGR sequences among nine clinical recombinants (Fig. 3 and Tables 1 and 2). Three regions, including two hot spots for recombination involving rs2, the IGR (between rs2 and ompA), and ompA, had previously been identified among C. trachomatis strains from a patient population from Lisbon, Portugal, that we had studied (17). The samples in the present study were from San Francisco, California, and Amsterdam, The Netherlands, suggesting that not only are these regions common sites of recombination in the genome, but they are unaffected by geographic perturbations. We and others have also shown the ompA region to be involved in homologous recombination (5, 16, 17, 20–22, 25) in addition to the trpB or trpA region (20). Only one strain in the current study, D43nl, showed no evidence of recombination despite previous identification of LGT in our MLST study (9). This is not a contradiction, given that the MLST scheme produces sequences of seven housekeeping genes not represented among the genomic regions sequenced in this study. In addition, the D43nl strain was from The Netherlands, while the other clinical D strains originated in the San Francisco Bay area in California, suggesting that there are some geographic differences within the strains. This is not caused by a lack of genome diversity among strains in The Netherlands because we previously found strain diversity in that country (9). The exact recombinant nature of strains such as D43nl would be revealed by whole-genome sequencing, which should become standard protocol once costs reach a reasonable level for research laboratories.

We also evaluated the 16 in vitro recombinants for statistically significant breakpoint regions (Table 2). The majority of breakpoints occurred within gyrA and rpoB; 14 contained identical breakpoint regions in gyrA with one recombinant showing an additional breakpoint region in gyrA, while rpoB contained variable breakpoint regions for six recombinants. The majority of the gyrA sequence was L₁ for the 14 crosses and D for the six rpoB crosses.

Surprisingly, the in vitro recombinant VIII had an entirely D sequence for gyrA, which should have been selected against, since strains D:Rif^r-1 and L₁:Ofx^r-1 were used in the cross. Ofloxacin is one of many fluoroquinolone antibiotics, and previous studies of fluoroquinolone-resistant isolates revealed single-nucleotide polymorphisms conferring resistance within the established quinolone resistance-determining region of gyrA but no mutations in the gyrB, parC, or parE quinolone resistance-determining regions (12, 32). However, one study found that isolates with elevated MICs against fluoroquinolones did not have gyrA or parC mutations in positions known to be critical for fluoroquinolone resistance (43), suggesting that other genes may be involved in resistance. In our study, we identified a mutation in gyrB that was the breakpoint crossover from D to L₁ where the majority of the gyrB sequence was L₁. Consequently, gyrB may be able to confer fluoroquinolone resistance, as this is the most plausible explanation for the occurrence of the VIII clone.

In comparing all 27 recombinants, two distinct clusters were identified, and the two clusters reflected the clinical and in vitro groups (Fig. 5), indicating a significant difference in occurrence of recombination breakpoints within the genes we sequenced. Because the clinical cluster contained D and L strains in two different branches, neither D nor L have patterns of recombination that are distinct from other strains. Therefore, the clustering is not due to the subtypes used in the in vitro experiments (D and L) being different from other strains. The in vitro recombinants had significantly more breakpoints in gyrA, which confers ofloxacin resistance, and in rpoB, which confers rifampin resistance, than the clinical recombinants. This was not surprising due to the introduced antibiotic pressure that selects for recombination in genes that confer a specific resistance, although the crossover breakpoints could have occurred outside the gene boundaries.

The trpA gene also showed significantly more breakpoints with an identical breakpoint region for 13 (81.25%) of the 16 in vitro recombinants that contained a crossover in this gene (Table 2). This was unexpected because rifampin and ofloxacin have not been known to induce mutations in trpA, although there may be some selection pressure within, upstream, or downstream of this gene. trpA encodes TrpA, an alpha chain that interacts with another alpha chain and two TrpB beta chains to form the tetrameric tryptophan synthase. Tryptophan is an essential amino acid for replication during differentiation of the metabolically active reticulate body into the infectious elementary body (4) that, upon release from the cell, sets up another round of infection in adjacent tissue or infects a new host following mucosal transmission. TrpR, TrpB, and TrpA are products of genes that make up the partial tryptophan operon, which is functional in the sexually transmitted strains of C. trachomatis but truncated and rendered nonfunctional among ocular trachoma strains (1, 6). In in vitro cell culture systems, gamma interferon (IFN-γ), which has been observed to be endogenously produced by C. trachomatis infection of HeLa cells (7), induces expression of indoleamine 2,3-dioxygenase that in turn degrades tryptophan. The inhibitory effect of IFN-γ varies according to strain. It is less effective against the LGV strains, which include L₁, compared with other urogenital strains; this effect is likely linked to two amino acid polymorphisms at residues 177 and 178 in the C terminus of TrpA (6). Each of the 13 in vitro TrpA recombinants were derived from L₁ in the C-terminal half of the protein, a critical region of loop 6 involved in alpha-beta subunit interactions and binding of important substrates such as indole-3-glycerol phosphate (24, 41). Consequently, while IFN-γ was not added to the growth medium, it is probable that in vitro induced recombination that places L₁ in the C terminus of TrpA contributes to activity of the protein and counters the effects of endogenously produced IFN-γ (7) in a way that is advantageous to the progeny.

There was also significant selection among the in vitro recombinants for ompA (P = 3.3 × 10⁻⁸), which, like trpA, is not known to be involved in antibiotic resistance. The strongest bias was for strain D (Table 2). Our previous work on clinical recombinants revealed two breakpoint regions, one of which was identified as a hot spot that flanks the ompA gene (17). Our findings support these data. The similarities in breakpoint regions flanking ompA among the in vitro and clinical recombinants in this study suggest a common mechanism of LGT, but additional studies will be needed to identify the precise mechanisms involved.

Recombination is prevalent in C. trachomatis and plays a major role in its genetic diversification and evolution. Our findings suggest that in vitro-derived recombinants of antibiotic resistance strains do not reflect the natural statistical distribution of recombination events. While frequent LGT with multiple breakpoints may occur in vivo among various strains, these events may be selected against given the mucosal environment or may produce mutants that are less fit. In any case, the current in vitro model should be used to evaluate mechanism(s) of recombination given the observed selection biases at the unexpected loci of trpA and ompA. Additional studies are needed to determine the rate and diversity of recombination for C. trachomatis clinical samples. In this process, comparative and functional genomics of many clinical and in vitro isolates that do not rely on antibiotic selection will be key to determining the precise mechanism of LGT in C. trachomatis and, ultimately, to develop a robust gene transfer system for the organism.