Interstrain Gene Transfer in Chlamydia trachomatis In Vitro: Mechanism and Significance

Abstract

The high frequency of between-strain genetic recombinants of Chlamydia trachomatis among isolates obtained from human sexually transmitted infections suggests that lateral gene transfer (LGT) is an important means by which C. trachomatis generates variants that have enhanced relative fitness. A mechanism for LGT in C. trachomatis has not been described, and investigation of this phenomenon by experimentation has been hampered by the obligate intracellular development of this pathogen. We describe here experiments that readily detected LGT between strains of C. trachomatis in vitro. Host cells were simultaneously infected with an ofloxacin-resistant (Ofx^r) mutant of a serovar L1 strain (L1:Ofx^r-1) and a rifampin-resistant (Rif^r) mutant of a serovar D strain (D:Rif^r-1). Development occurred in the absence of antibiotics, and the progeny were subjected to selection for Ofx^r Rif^r recombinants. The parental strains differed at many polymorphic nucleotide sites, and DNA sequencing was used to map genetic crossovers and to determine the parental sources of DNA segments in 14 recombinants. Depending on the assumed DNA donor, the estimated minimal length of the transferred DNA was ≥123 kb in one recombinant but was ≥336 to ≥790 kb in all other recombinants. Such trans-DNA lengths have been associated only with conjugation in known microbial LGT systems, but natural DNA transformation remains a conceivable mechanism. LGT studies can now be performed with diverse combinations of C. trachomatis strains, and they could have evolutionary interest and yield useful recombinants for functional analysis of allelic differences between strains.

We describe here a substantial advance in genetic studies of Chlamydia trachomatis, which is the leading worldwide cause of sexually transmitted bacterial infections (38) and, mostly in the developing world, the leading cause of preventable blindness (4). The obligate intracellular development of C. trachomatis has impeded the creation of in vitro methodologies for manipulating its genetics, thereby hampering many kinds of investigation that might improve our understanding of and ability to control this widely prevalent pathogen. We show here how this obstacle to genetic analysis can be partially overcome; recently described in vitro lateral gene transfer (LGT) (8) now allows genetic crosses that create diverse genetic recombinants between C. trachomatis strains that differ with respect to many genetic markers. Analysis of the genetic makeups of recombinants should provide new opportunities for studying questions that were hardly accessible to experimental analysis until now. We illustrate below how our in vitro LGT system can be used to investigate several such questions, including the following. (i) What are the mechanisms of LGT in C. trachomatis? (ii) Are there especially notable features of the genetic recombination that occurs after DNA has been transferred between chlamydiae? (iii) How might the in vitro processes discovered in investigations of these two questions contribute to the origins of numerous LGT recombinants that are detected among clinical isolates of C. trachomatis (see below)?

Pathogenic microbes employ several genetic mechanisms for producing variants that counter host defenses. A role for spontaneous mutation followed by in vivo selection in the escape of C. trachomatis from human immune defenses is suggested by the high proportion of urogenital tract isolates that have amino acid substitutions in the polymorphic ompA gene, which encodes the major outer membrane protein (MOMP) (3, 5, 7, 11, 17, 21, 27). Humans have diverse B-cell-mediated (33) and T-cell-mediated (16, 17, 27, 28) responses to MOMP. The possibility that at least some of these responses are protective is suggested by the occurrence of mutations in the same ompA segments of multiple clinical isolates (5); some of the mutations alter human T-cell epitopes (27). This implication that there is in vivo selection for certain ompA mutants (21) is supported by the fact that the frequencies of spontaneous mutants involving several non-ompA loci in C. trachomatis are <10⁻⁷ (2, 8) but the frequency of ompA mutants among DNA-sequenced clinical isolates is about 10⁻² to 10⁻¹ (22, 23); while the unmeasured ompA mutation rate may prove to be above average for C. trachomatis, many MOMP mutants probably have been selected for in vivo.

Accumulating evidence indicates that LGT may also substantially contribute to the origin of enhanced-fitness variants in C. trachomatis. DNA sequencing of long-established strains (23) and clinical isolates (3, 9, 10, 12) has detected numerous, diverse genetic recombinants possessing alleles of polymorphic loci that were derived from at least two different strains. Traditionally, C. trachomatis strains called “serovars” have been serologically defined by polymorphic variations in the MOMP. These differences in B-cell epitopes have been shown to be localized in four variable MOMP domains (1) that are encoded by polymorphic nucleotide sequences in four variable segments of the ompA gene (15). While various studies have suggested associations between serovar specificity, tissue tropism, and virulence (20), continuing DNA sequencing of ompA and other polymorphic loci has shown that (i) strains can no longer be defined simply by their serologically defined ompA alleles because the sequences of these alleles often vary nonsynonymously within a serovar (6, 22, 32); (ii) some ompA alleles appear to have resulted from intragenic recombination after in vivo LGT between strains; and (iii) various combinations of alleles of polymorphic loci other than ompA can be associated with the same serovar specificity. Such reassortment of alleles of diverse loci must have resulted from interstrain LGT, and there is not an invariant association between the ompA allele and tissue tropism or invasiveness (6, 22).

Recent studies reported that at least 7 of 12 (9) and all 10 (10) sequenced clinical isolates had recombinant chromosomes, and most of the chromosomes in the latter group had crossovers in the same short DNA segments (“hot spots”). Such large proportions of recombinants among numerous clinical isolates bring two questions in particular into sharp focus. (i) By what mechanism(s) did the in vivo LGT recombinants of C. trachomatis originate, and is the LGT ongoing? Below we present numerous estimates of the lengths of transferred DNA segments that are suggestive of possible DNA mechanisms. (ii) To what extent do the frequency of LGT events, on the one hand, and selection for LGT recombinants, on the other hand, contribute to the high frequency of in vivo recombinants? With respect to the latter possibility, we show below (see Discussion) how the in vitro LGT system could be used to distinguish between (i) short, recombination-prone DNA segments in which nonrandomly high frequencies of recombination among both in vitro and clinical recombinant isolates are observed and (ii) short DNA segments in which between-strain recombination occurring at perhaps unremarkable rates generates variants that have enhanced fitness and that accumulate under in vivo selection conditions so that they comprise a large proportion of clinical isolates.

The loss by C. trachomatis of many genes that are necessary for extracellular proliferation has prevented investigation of the questions described above by direct experimental methods. Intracellular development of C. trachomatis is initiated by attachment of a metabolically inert “elementary body” (EB) to a host cell. The EB is internalized in a vacuole and enlarges to become a noninfectious “reticulate body” (RB) in which the DNA decondenses, transcription begins, and metabolism ensues. RBs multiply by means of binary fission, and, as the number of RBs increases, the initial vacuole forms an “inclusion” that gradually enlarges to virtually fill the cell. About midway through development, conversion of some RBs into EBs begins, while other RBs continue to multiply. At about 48 to 72 h postinfection (p.i.), depending on the strain and conditions, the infected cells liberate several hundred to about 1,000 EBs that can infect other cells and that can be assayed as “inclusion-forming units” (IFU) on monolayers of host cells; when initiated by a single IFU, an inclusion corresponds to a colony of bacteria or initiation of a plaque by a virus particle.

The intracellular development of C. trachomatis and the noninfectivity of RBs have limited investigation of LGT in C. trachomatis to retrospective DNA-sequencing studies of recombinants among established reference strains and recent clinical isolates. The reported recombination data indicate that LGT can involve widely separated loci distributed over much or all of the 1.04-Mb C. trachomatis chromosome (10). Sequences resembling insertion sequences (9) and other sequences that were interpreted as recombination “hot spots” (10) have been reported, but LGT mechanisms were not definitely identified. The life cycle impediment to direct experimental investigation of LGT mechanisms was partially removed by our recent use of in vitro methodology to discover LGT within an established C. trachomatis strain that has a serovar L1 ompA allele (8). In various genetic crosses, host cells were simultaneously infected with an ofloxacin (Ofx)-resistant (Ofx^r) mutant of a serovar L1 strain (L1:Ofx^r-1) and second serovar L1 mutants that were either lincomycin resistant (Lin^r), trimethoprim resistant, or rifampin (Rif) resistant (Rif^r). Selection for doubly resistant C. trachomatis isolates among the progeny of mixed infections produced in the absence of antibiotics resulted in recombinant frequencies of 10⁻⁴ to 10⁻³. These frequencies were ∼10³ to 10⁴ times higher than those of doubly resistant spontaneous mutants in progeny from one-parent control infections. Doubly resistant C. trachomatis strains isolated from mixed infections were cloned in the absence of antibiotics and proved to have both of the specific resistance-conferring mutant nucleotides present in the parental strains used in each cross; the absence of the corresponding normal nucleotides indicated that they had been replaced by homologous recombination. These results eliminated spontaneous mutation, between-strain complementation, and heterotypic resistance (14, 35) as origins of the doubly resistant clones that were isolated from mixed infections and demonstrated their genetic stability.

The within-strain LGT that we discovered in C. trachomatis did not directly provide information about involvement of known LGT mechanisms or their possible generation of recombinants in vivo. We proposed (8) that information about the sizes of LGT DNA segments should help in distinguishing between the three main known LGT mechanisms (see Discussion) and that such information could be obtained by producing recombinants between different strains of C. trachomatis that had numerous polymorphic nucleotide differences distributed throughout the chromosome. The sizes of LGT DNA segments could then be estimated by sequencing for the polymorphic nucleotides at numerous sites in recombinants and then using the genetic map of C. trachomatis (36) to localize crossovers. New research reported here revealed in vitro LGT between different C. trachomatis serovars and fulfills these aims. This sets the stage for using in vitro experiments to investigate LGT mechanisms, possible origins of in vivo LGT recombinants, and functional effects of allelic substitutions created in recombinants by means of in vitro LGT. When it becomes feasible to make planned genetic modifications in candidate LGT genes in C. trachomatis by means of artificial DNA transfer, the system that we described previously (8) and that we describe here should be useful for assessing the effects of the modifications.

MATERIALS AND METHODS

C. trachomatis strains.

We isolated mutant L1:Ofx^r-1 (8) from a wild-type stock that was provided by G. Byrne and had the ompA allele of serovar L1. Mutant D:Rif^r-1 was isolated from the D/UW-3/CX strain (American Type Culture Collection, Gaithersburg, MD) using the mutant isolation procedures described previously (8).

C. trachomatis was propagated essentially as described previously (8), except that SPG buffer (10 mM sodium phosphate [pH 7.2], 0.25 M sucrose, 5 mM l-glutamic acid) was substituted for infection medium. Inocula in SPG buffer were removed after the EB attachment period and before addition of growth medium to the newly infected cultures. EBs were harvested and IFU were assayed as described previously (8).

A D:Rif^r-1 × L1:Ofx^r-1 cross was performed by infecting 10-flask sets of primary (P-0) 12.5-cm² flasks (F12s) containing 1.5 × 10⁶ HeLa 229 cells with either D:Rif^r-1 at a multiplicity of infection (MOI) of 0.82 (set A), L1:Ofx^r-1 at an MOI of 0.70 (set B), or both strains at the indicated MOI (set C). C. trachomatis development occurred in the absence of antibiotics and was terminated at 48 h p.i. IFU were harvested (8) in 1 ml of SPG buffer per F12 and were frozen at −80°C. Ofx^r Rif^r C. trachomatis cells that were present in the P-0 flasks as a result of new spontaneous mutations (sets A and B) or of LGT (set C) were selected by using 200 μl of each P-0 IFU harvest to inoculate a first-passage HeLa cell F12 in which C. trachomatis development occurred in medium containing Ofx (1.0 μg per ml) and Rif (0.015 μg per ml). IFU assays of sample P-0 IFU harvests indicated that set C first-passage F12s were infected with an average of 2.08 × 10⁷ IFU. Three additional passages in the presence of Ofx plus Rif were performed at ∼48-h intervals by infecting HeLa cell F12s at each passage with one-half of the IFU harvested from the preceding passage. The resulting P-4 flasks were scanned with a phase-contrast microscope to determine the presence of Ofx^r Rif^r inclusions at ∼48 h p.i. The reasons for using passaging instead of plaque formation to select for recombinants have been explained previously (8).

Clones were isolated in the absence of antibiotics from set C P-4 or P-5 F12s by means of limiting dilution essentially as described previously (8). IFU were harvested from the cloning plates on days 7 to 9 p.i. by draining each clone-containing well, rinsing it with 500 μl of phosphate-buffered saline, and freezing at −80°C. After it was thawed at room temperature, each well received 200 μl of ice-cold SPG buffer and ∼20 sterile 1-mm glass beads (Fisher catalog number 50212144). The plates were shaken horizontally 30 times to disrupt the infected cells, and the resultant EBs harvested were used to infect wells in 24-well plates containing 1.5 × 10⁵ HeLa 229 cells per well. EBs harvested from the 24-well plates were used to infect HeLa cell F12s from which EBs were harvested for DNA isolation.

DNA sequence analysis.

DNA was isolated as previously described (8) (omitting cetyltrimethylammonium bromide and NaCl) from EBs that were harvested from one F12 and was collected by centrifugation at 12,000 × g for 10 min. EBs from one F12 usually yielded ∼10 μg of DNA.

The relevant parts of the genes of interest (see Table S1 in the supplemental material) were sequenced after PCR amplification in 50-μl reaction mixtures containing 10 μl of 5× Phusion HF buffer, 1 μl of 10 mM deoxynucleoside triphosphates, 0.5 μl of Phusion polymerase (New England Biolabs), 25 pmol of sense and antisense primers (see Table S2 in the supplemental material), 50 ng of EB template DNA, and water. Each PCR included an initial denaturation step at 98°C for 30 s and a final extension step at 72°C for 7 min in addition to 35 cycles that included a denaturation step at 98°C for 15 s. Table S2 in the supplemental material shows the annealing temperature, the extension time, and the amplicon sizes. PCR amplification primers that allowed detection of the polymorphic nucleotides listed in Table S1 in the supplemental material were used. PCR amplicons were purified using a QIAquick PCR purification kit (Qiagen) as instructed by the manufacturer. Purified amplicons were eluted in 50 μl of EB buffer, and sequencing reactions were performed in 20-μl (final volume) mixtures containing 2 μl of ET Terminator (GE Healthcare), 6 μl of sequencing buffer (20 mM Tris [pH 9.0], 5 mM MgCl₂), 1 μl of 3.2 mM oligonucleotide (see Table S3 in the supplemental material), approximately 50 ng of amplified product, and water. The following cycling conditions were used: 30 cycles of denaturation at 96°C for 15 s, annealing at 50°C for 5 s, and extension at 60°C for 2 min. Completed sequencing reactions were run on an ABI 3730 in house.

Nucleotide sequence accession numbers.

Clone sequences have been deposited in the GenBank database under the following accession numbers: for gyrA, EU104987 to EU105002; for gyrB, EU105003 to EU105018; for incA, EU105019 to EU105034; for murA, EU105035 to EU105050; for ompA, EU105051 to EU105066; for pmpC, EU105067 to EU105082; for recF, EU105083 to EU105098; for ribF, EU105099 to EU105114; for rpoB, EU105115 to EU105130; for trpA, EU105131 to EU105146; and for pCT, EU180710 to EU180743.

RESULTS

LGT between C. trachomatis serovars D and L1 occurs in vitro.

In order to detect LGT between serovar D and serovar L1 strains of C. trachomatis, we simultaneously infected HeLa 229 cells with mutant L1:Ofx^r-1 and mutant D:Rif^r-1 (“mixed infection”) (see Methods and Materials). L1:Ofx^r-1 has a T249→G mutation in the gyrA gene (CT189) (8), and D:Rif^r-1 has a C1400→T mutation in the rpoB gene (CT315). Development occurred in the absence of antibiotics in the originally infected (P-0) F12s. We assumed that LGT could occur during development in the absence of antibiotics because inclusions initiated by at least certain combinations of strains in the same cell fuse (30). Part of each P-0 IFU harvest was used to initiate a sequence of four passages in the presence of Ofx and Rif to select for Ofx^r Rif^r recombinants. P-4 F12s were microscopically scanned for Ofx^r Rif^r inclusions.

Three of the 10 control set A P-4 F12s that resulted from infection of P-0 F12s with just L1:Ofx^r-1 contained spontaneous Rif^r mutants that were initially present in P-0 harvests of this parental strain. DNA sequencing revealed C1411→A, G1399→A, and C1400→T mutations in the rpoB gene. None of the 10 set B P-4 F12s that resulted from infection of P-0 F12s with just D:Rif^r-1 contained spontaneous Ofx^r mutants. Thus, a total of 20 control F12s in sets A and B yielded just one spontaneous mutant having a specific nucleotide substitution that was present in a parental strain. In contrast, the presence of abundant Ofx^r Rif^r inclusions in all P-4 F12s derived from mixed-infection P-0 cultures indicated that each set C P-0 F12 initially contained multiple Ofx^r Rif^r clones. Recombinants were not precisely enumerated in this cross as they were previously (8), but the variety of recombinants subsequently isolated from four individual set C flask lineages (see below) suggests that the average recombinant frequency was much less than 1/10 that observed for within-strain crosses (8).

The specific resistance-conferring mutant nucleotides that were present in the rpoB and gyrA genes of the parental strains were detected in all 10 set C bulk isolates. A total of 30 clones were isolated from four set C P-4 F12s in the absence of antibiotics; cloned recombinants derived from different flasks must have had independent origins. DNAs isolated from the clones were sequenced in order to localize crossovers and estimate minimum trans-DNA lengths. All of the clones that were isolated from mixed-infection progeny had the gyrA T249→G and rpoB C1400→T mutant nucleotides that conferred resistance to Ofx and Rif, respectively, in each of the parental mutant strains. The absence of the corresponding normal nucleotides indicated that they had been replaced by homologous recombination. Therefore, the Ofx^r Rif^r isolates were genetically stable LGT recombinants that had not originated by means of complementation, spontaneous mutation, or heterotypic resistance (14, 35).

Genetic analysis of the serovar D × serovar L1 recombinants was performed by sequencing DNA segments that contained polymorphic nucleotides in the following loci: ompA, murA, pmpC, rpoB, gyrA, trpA, incA, ribF, and recF. Table S1 in the supplemental material shows nucleotide differences that allowed us to determine the parental origins of each of the indicated nucleotide sites in recombinant chromosomes. The collection of nucleotide alleles in each recombinant allowed assignment of allelic signatures that were used to construct chromosome maps showing the parental origins of DNA segments that were present in the cloned recombinants (Fig. 1). The following two assumptions were made as a basis for analyzing the data.

FIG. 1.

Distribution of genetic crossovers in 14 independent recombinant clones resulting from in vitro LGT. The locations of the loci on the 1,043-kb circular C. trachomatis chromosome are indicated by their midpoints (in kilobases of DNA) from the chromosomal Ori. Sequencing of DNA segments within each of the loci shown (see Tables S1, S2, and S3 in the supplemental material) that included polymorphic nucleotide differences between the serovar D and L1 parental strains used in the cross was used to determine the allelic signatures of LGT recombinants (Table 1) and the parental sources of chromosomal segments shown; DNA between these polymorphism-containing segments was not sequenced. Multiple occurrences (n > 1) of the same recombinant type indicate independent origins of the same general type of recombinant in more than one mixed-infection culture. The placement of crossovers in all recombinants indicates that a crossover could have occurred anywhere in the interval between the nearest polymorphic nucleotides on both sides of the crossover. Asterisks identify recombinants that had an exchange within a short, completely sequenced segment of either the rpoB gene (types III, V, V-A, and V-B) or the gyrA gene (types VII and VIII) (Table 1). In the type V-B recombinant, a ≤512-bp segment of serovar L1-derived DNA was inserted into the rpoB gene of a serovar D strain, while in the type VIII recombinant, a ≤257-bp segment of serovar L1-derived DNA was inserted into the gyrA gene of a serovar D strain. Gray bars above the chromosome diagrams indicate the minimum lengths of trans-DNA that could account for the allelic signatures of the recombinant chromosomes. The derivation of these lengths is based on assumed DNA donors and is explained in the text. (A) The assumed DNA donor is L1:Ofx^r-1. (B) The assumed donor is D:Rif^r-1. Recombinant type VI is used to illustrate the two ways in which lengths are estimated for trans-DNA segments that extend either clockwise from ompA (type VI-C) or counterclockwise from ompA (type VI-CC). The estimated minimum serovar D-derived trans-DNA lengths for all recombinants were 386 to 794 kb.

First, homologous recombination occurred between a linear molecule of trans-DNA and a circular chromosome in DNA recipients. Any even number of crossovers would result in replacement of chromosomal DNA segments with homologous segments of trans-DNA. The resultant circular recombinant chromosomes would contain both selection markers and be able to replicate normally, allowing selection for Ofx^r Rif^r recombinants. In contrast, any odd number of crossovers would result in a linear concatemer containing all of the chromosomal and trans-DNA. The linear reciprocal recombinant product in such cases would contain the wild-type gyrA and rpoB loci, but the absence of these alleles from all of the Ofx^r Rif^r isolates from set C flasks indicated that the linear product could not replicate sufficiently to be detected.

Second, in order to estimate the minimum lengths of trans-DNA needed to account for the allelic signature of each recombinant, it was necessary to specify a DNA donor and recipient; our current data did not provide an unequivocal way of making the distinction. Furthermore, it is possible that different recombinants in the collection of 14 recombinants were derived from different DNA donor and recipient combinations. Therefore, a detailed trans-DNA length estimate is presented for each recombinant, assuming that all recombinants arose from L1:Ofx^r-1 DNA donors (Fig. 1A), and an illustrative estimate is presented for one recombinant, assuming that it was derived from a D:Rif^r-1 DNA donor (Fig. 1B). On this basis, we reasoned as follows.

(i) All recombinants had to have the serovar D-derived rpoB mutant site and the serovar L1-derived gyrA mutant site that LGT brought together in recombinants, allowing their selection with Ofx and Rif; all other polymorphic nucleotides could have been derived from either parental strain in the cross.

(ii) The minimum length of trans-DNA needed to account for the allelic signature of a recombinant was defined as the distance between the most widely separated sequenced polymorphic sites that were derived from the assumed DNA donor strain (Fig. 1). With the exception of some recombinant types, each crossover joining donor and recipient DNA segments could be only regionally localized to a multikilobase segment of DNA between two consecutive sequenced polymorphic sites that had been recombined. Trans-DNA estimates based on these definitions are summarized below.

L1:Ofx^r-1 is the assumed DNA donor.

The type IX recombinant had only ≥123 kb of trans-DNA, but the other recombinants had received ≥336 kb of serovar L1-derived DNA and 10 (71%) of the 14 recombinants had received ≥441 kb of serovar L1-derived DNA comprising ≥40% of the chromosome (Fig. 1A). The actual lengths of serovar L1-derived trans-DNA could be greater than those shown in Fig. 1 if sequencing of loci in the large unsequenced segments (total, ∼600 kb) of DNA flanking ompA reveals currently undetected serovar L1 sequences.

D:Rif^r-1 is the assumed DNA donor.

All 14 independent recombinants had the serovar D ompA allele centered at 779 kb on the chromosome (Fig. 1B). ompA is flanked by two large segments of DNA in which sample loci have not been sequenced: going clockwise, 264 kb between ompA and ori plus 89 kb between ori and recF; and going counterclockwise, 249 kb between ompA and murA. Because the ompA allele was derived from serovar D in all recombinants, the minimal estimates of trans-DNA length must include one of these unsequenced segments plus the DNA from the end of the segment to the sequenced serovar D-derived locus most distal to it. The estimates are 621 to 794 kb in the clockwise direction and 386 to 690 kb in the counterclockwise direction. About 64% of the estimates for serovar D-derived DNA are ≥621 kb (∼60% of the C. trachomatis chromosome).

As an alternative approach to specifying DNA recipients in LGT, we thought that it was likely that recombinants would have the version of the pCT plasmid (29) that was present in the DNA recipient from which they originated. Like the chromosome, pCT has naturally polymorphic nucleotides that are different in different strains. We sequenced for 10 serovar D versus serovar L1 polymorphic nucleotides distributed in two clusters separated by about half of the pCT DNA (see Table S1 in the supplemental material). All 10 marker nucleotides were derived from L1:Ofx^r-1 in all 14 independent recombinants (data not shown). This observation supports identification of the L1:Ofx^r-1 parental strain as the DNA recipient in every recombinant that we analyzed. If this inference is correct, the D:Rif^r-1 parent transferred at least 60% of its chromosome to about two-thirds of the recombinants, but there are testable alternative interpretations (see Discussion).

If all 14 recombinants analyzed had the version of pCT present in the recipients from which they originated, why were recombinants derived from D:Rif^r-1 DNA recipients not observed? While it is possible that the serovar D strain was the exclusive DNA donor in the crosses and the serovar L1 strain was the exclusive recipient, as an alternative, we suggest that the absence of recombinants derived from serovar D strain DNA recipients resulted from the relatively poor ability of the D:Rif^r-1 parent to attach to host cells compared to the ability of the L1:Ofx^r-1 parent. Thus, the frequency of serovar D-like recombinants with regard to attachment ability might have been gradually reduced relative to the frequency of serovar L1-like recombinants during passaging, resulting in the fact that serovar D-like recombinants were not included in the sample of 14 recombinants that we analyzed. A way to recover the hypothetical missing recombinants is described in the Discussion.

Another especially notable feature of our data is the nonrandomly high concentration of exchanges associated with the rpoB gene. All of the 14 independent recombinant clones that we analyzed had the gyrA G249 mutant and rpoB T1400 mutant nucleotides, and since a crossover anywhere in the ∼143 kb of DNA between these nucleotides could have created an Ofx^r Rif^r recombinant, the roughly estimated average density of crossovers in this segment was 14.0/14.3 × 10⁴ or 9.79 × 10⁻⁵ per base pair. Table 1 shows that 6 of the 14 exchanges within the 143-kb region occurred in the 553-bp segment bounded by rpoB nucleotides 1400 and 1953. The density of crossovers in this segment was 6/5.53 × 10² or 1.08 × 10⁻² per base pair, which was about 110 times greater than the average density for the 143-kb region. Since an exchange anywhere in the 143-kb region between the mutant sites would have conferred resistance to both Ofx and Rif to a recombinant, the nonrandomly high concentration of exchanges within the rpoB gene suggests that there is a mechanistic predisposition to recombination rather than selection for doubly resistant recombinants that have exchanges in rpoB. Below we show how this result obtained with our in vitro LGT methodology affects possible interpretations of nonrandomly high concentrations of crossovers (“hot spots”) in clinical isolates (see Discussion) (10).

TABLE 1.

Polymorphic nucleotides that allowed detection of intragenic recombination in the rpoB, gyrA, and gyrB chromosomal genes

Strain or type	n	Serovar from which indicated allele derivea
		rpoB nucleotide:													gyrB nucleotide:			gyrA nucleotide:
		674	802	950	975	996	1044	1400	1560	1714	1953	2226	2428	2442	1779	2004	2371	237	249	492	909	987	1088
Strains
D:Rifr-1		D	D	D	D	D	D	D	D	D	D	D	D	D	D	D	D	D	D	D	D	D	D
L1:Ofxr-1		L1	L1	L1	L1	L1	L1	L1	L1	L1	L1	L1	L1	L1	L1	L1	L1	L1	L1	L1	L1	L1	L1
Types
III	1	D	D	D	D	D	D	D	L1	L1	L1	L1	L1	L1
V	2	D	D	D	D	D	D	D	L1	L1	L1	L1	L1	L1
V-A	1	D	D	D	D	D	D	D	D	L1	L1	L1	L1	L1
V-B	1	D	D	D	D	D	D	D	D	D	L1	D	D	D
VII	1														L1	L1	L1	L1	L1	D	D	D	D
VIII	1														L1	L1	D	D	L1	D	D	D	D

^aD, the nucleotide allele is present in the serovar D parent in the cross; L1, the nucleotide allele is present in the serovar L1 parent in the cross (see Table S1 in the supplemental material). The type III and two type V recombinants had a crossover between rpoB nucleotides 1400 and 1560, the type V-A recombinant had a crossover between rpoB nucleotides 1560 and 1714, and the type V-B recombinant had a crossover between rpoB nucleotides 1714 and 1953. The type VII recombinant (Fig. 1A) had a crossover between gyrA nucleotides 249 and 492. In the type VIII recombinant (Fig. 1A) the presence of at least two other segments of serovar L1-derived DNA indicates that the presence of the gyrA nucleotide 249 mutant allele resulted from insertion of a short segment of serovar L1-derived DNA by means of recombination rather than from a new spontaneous T249→G mutation.

It should also be noted that the type V-B and type VIII recombinants (Fig. 1A) may have gene conversions in the rpoB and gyrA genes, respectively. In the type V-B recombinant, it appears that a ≤512-bp segment of serovar L1-derived DNA marked by a serovar L1 nucleotide (rpoB nucleotide 1953) was inserted into a much longer segment of serovar D-derived DNA (Table 1 and Fig. 1A). Similarly, a ≤257-bp segment of serovar L1 DNA marked by a single serovar L1 nucleotide (gyrA nucleotide 249) was inserted into serovar D-derived DNA. The estimated rate of spontaneous mutation, ∼10⁻⁹ per nucleotide per division (8), makes it highly unlikely that these nucleotide changes resulted from mutations, especially because the presence of other unselected segments of serovar L1 DNA in the isolates identified them as LGT recombinants. In a formal sense, a double crossover inserted a short serovar L1-derived DNA segment into serovar D-derived DNA in each recombinant. Two crossovers that are so narrowly separated would not be unexpected in a gene conversion tract in which a short DNA segment was copied from a serovar L1 template in the course of resolving a Holliday junction during recombination; CT501 and CT502 are orthologues of the Holliday junction (18) helicase motor protein and resolvase, respectively. These observations augment previous suggestions for the possible involvement of gene conversion in the in vitro (8) and in vivo (24) origin of azithromycin resistance in C. trachomatis, as well as in the origin of ompA variants (23) and paired exchanges in recombination “hot spots” (10) (see Discussion) in certain in vivo LGT recombinants.

DISCUSSION

The recombinant analysis summarized above confirmed our initial description of in vitro LGT in C. trachomatis (8) and is a substantial advance because it shows that in vitro LGT can readily be detected in mixed infections with different strains. The LGT mechanism has not been identified, but the present study opens the way to accomplishing this. Furthermore, various publications, exemplified by the paper of Sayada et al. (32), have suggested that at least several percentages of many millions of clinical infections involve more than one strain of C. trachomatis; the results described here should enable investigation of possible relationships between in vitro LGT and the high frequency of LGT recombinants in clinical isolates (see below).

Trans-DNA lengths and LGT mechanisms.

Estimates of the lengths of trans-DNA segments in LGT recombinants may provide valuable clues concerning the LGT mechanism. Such an analysis with in vivo recombinants is made uncertain by the possible occurrence of multiple LGT events during successive infections in the lineages of the recombinants, perhaps during very long periods of time; secondary LGT events might disrupt segments of DNA transferred during previous LGT events. An important feature of the LGT system described here is the ability to produce many recombinants that allow estimation of the lengths of DNA segments that are transferred between chlamydiae in single LGT episodes. We think this because we previously showed (8) that (i) the frequency of LGT recombinants was ∼10⁻⁴ to 10⁻³ in mixed infections with two different kinds of mutants isolated from our serovar L1 strain and (ii) simultaneous infection of cells with three different kinds of mutants resulted in the appearance of recombinants that possessed selection markers derived from all three parental strains during one course of development. Such recombinants must have resulted from two LGT events, and their frequency (10⁻⁶) suggests that perhaps 1 in 1,000 recombinants produced in the cross described here would be the product of more than one LGT event. Therefore, we used the estimated lengths of trans-DNA (Fig. 1) as possible clues to the nature of the LGT mechanism in C. trachomatis. Since we could not unequivocally identify the DNA donor and recipient from which each recombinant originated, trans-DNA estimates were derived by using the assumption that each parental strain could be either the donor or the recipient.

We think that the estimates of trans-DNA length are accurate because the DNA sequencing data indicate the monoclonality of our sequenced recombinants; only one parental nucleotide allele was detected at each of the 107 polymorphic sites that we studied (see Table S1 in the supplemental material). Therefore, we found it difficult to challenge the estimates of minimal trans-DNA length as we have defined it: a DNA segment in which segments of DNA from both parental strains in the cross alternate and that is bounded by the most widely separated sequenced polymorphic sites derived from the assumed DNA donor. If this definition is correct, trans-DNA lengths in the majority of C. trachomatis recombinants comprised ≥40% (≥441 kb) of the chromosome if L1:Ofx^r-1 was the assumed donor and ≥60% (621 kb) of the chromosome if D:Rif^r-1 was the assumed DNA donor. How are trans-DNA lengths in these size ranges related to trans-DNA lengths in the three main kinds of LGT that have been observed in other microbes?

As previously noted (8), C. trachomatis phages that might transduce DNA have been sought but not reported, which makes it highly unlikely that phage-mediated transduction is the mechanism of the in vitro LGT that we describe here. The estimated minimal trans-DNA lengths in the great majority of the recombinants that we analyzed (Fig. 1) exceed the DNA packaging capacities of all known phages except phage G of Bacillus megaterium, which has a 670-kb genome (13).

C. trachomatis has a low-copy-number 7.5-kb plasmid, pCT (29), that has eight open reading frames, none of which has homology to genes in plasmids that are known to be involved in conjugation or in transfer of chromosomal genes in microbes. The strong suggestion that pCT might not be involved in chromosomal LGT in C. trachomatis (see below) prompted us to hypothesize that chromosomal LGT recombinants possess the version of the pCT plasmid that was initially present in the DNA recipients from which they were derived. If this proposal is correct, all of the recombinants that we analyzed originated from L1:Ofx^r-1 DNA recipients. This may not necessarily mean that recombinants derived from D:Rif^r-1 recipients were not produced in the mixed infection. Long after the recombinants were isolated, we discovered that the D/UW-3/Cx strain that we used and the D:Rif^r-1 mutant derivative attach to host cells much less efficiently than our serovar L1 strain and its mutant derivatives when infection is performed with the “rock and rest” procedure (8) that we used to passage the progeny of the mixed infection in this study. After measuring the different attachment rates of the parental strains (data not shown), we could predict that descendants of “serovar L1-type” recombinants that attached efficiently would outnumber “serovar D-type” attachment-inefficient recombinants initially present in P-0 IFU harvests by at least 20-fold after the four or five rock-and-rest passages preceding DNA isolation. A further relative reduction would have occurred when IFU were plated out for cloning by limiting dilution. We verified that the attachment of D:Rif^r-1 is greatly increased by centrifugation at 1,600 × g for 60 min at 37°C (26). Stored remainders of the P-0 IFU harvests from our set C mixed infections in the cross will be passaged and cloned with the aid of centrifugation in an attempt to find recombinants that attach relatively poorly, presumably because they possess D:Rif^r-1-derived determinants of poor attachment ability. If recombinants possess the version of pCT that is present in the DNA recipients from which they originated, poorly attaching recombinants recovered as described above might have the version of pCT that is present in the D:Rif^r-1 parent in the cross.

The experiment described above illustrates one promising aspect of the LGT experimental system that we describe here. It posits the existence of different alleles of undescribed attachment efficiency genes that can be regionally mapped in recombinants produced by crossing strains with different abilities to attach. Specific segments of DNA that are characteristically derived from the same parental strain and that are associated with either efficient or inefficient attachment might include such genes. The potentially broad applicability of this approach to mapping functionally defined genes is evident.

Crosses with pCT-deficient isolates of D:Rif^r-1 and L1:Ofx^r-1 or other existing mutants of various strains (19, 26) could be used to determine if pCT is needed for LGT. But it is also interesting to ask if LGT involving the pCT plasmid can also occur, perhaps separately from LGT of chromosomal DNA. The unproven working assumption that pCT was not transferred between chlamydiae in our crosses can be probed experimentally. Our's appears to be the first published use of pCT as a genetic marker in observations of LGT in C. trachomatis, and it opens a field of experimentation in which mixed infections with pCT⁺ and pCT⁻ strains (19, 26) can be used to determine if pCT can be transferred between chlamydiae and to ascertain the phenotypic effects of such a transfer. The relatively poor attachment ability of pCT⁻ strains and their greatly reduced accumulation of glycogen and reduced stainability with iodine (19, 26) could be exploited to detect pCT transfer in such experiments.

BLAST analysis (8) that revealed five C. trachomatis sequences that were orthologous to genes known to be involved in DNA uptake suggested the possibility that the LGT might occur by natural DNA transformation. So far, studies of cotransfer of linked genetic markers by means of natural DNA transformation in other microorganisms have reported transfer of DNA segments that are ∼53 kb long (37) (see below), which is much smaller than the minimal trans-DNA segments in our recombinants. This suggests that in vitro LGT in C. trachomatis may usually or always occur by means of conjugation, which often transfers hundreds to ∼2,000 kb of DNA in Escherichia coli. It is interesting that the C. trachomatis ihfA gene (CT267) is an orthologue of the integration host factor that physically links DNA that is processed for transfer to the cell apparatus that exports it out of DNA donors during conjugation (34). This suggestion is also compatible with the presence in at least 5 or 6 of a group of 10 clinically isolated recombinants of large segments of DNA derived from different C. trachomatis strains (10). Assuming that either strain could have been the DNA donor in an LGT event, the minimum lengths of trans-DNA needed to explain the allelic signatures of these in vivo recombinants are either the ∼430 kb between the CT049 and pmpC loci or the ∼250 kb between the pmpF and rs2 loci. Both estimates are in the range reported here for trans-DNA segments in in vitro recombinants.

However, in 7 of the 10 recombinant clinical isolates a small segment of DNA from one strain was also inserted into a much larger segment of DNA from another strain (10). DNA from the same strain source as the small insert was not found elsewhere in the DNA sequenced in the recombinants. It was proposed that the small size of the insertions indicated that the origin of the recombinants was natural DNA transformation. However, the impression that natural DNA transformation typically involves shorter trans-DNA segments than we found may result from the fact that previously described studies of transformation with linked markers were not designed to detect trans-DNA lengths greater than ∼50 kb. Further studies of natural transformation of linked polymorphic markers that sample hundreds of kilobases of the donor chromosomes, as we did, may reveal trans-DNA segments as large as those we report for C. trachomatis. The proposal that the small inserts in ompA-neighboring “hot spots” resulted from natural DNA transformation may prove to be correct, but we suggest that the amounts of trans-DNA derived from the donors of the small inserts may have been underestimated in the in vivo recombinants in the study of Gomes et al. (10). Unsequenced 250- to 300-kb segments flank the short rs2/ompA/IGR segment in which the trans-DNA inserts are located; sequencing within these large flanking regions may reveal additional segments of trans-DNA having the same strain origin as the small inserts. For example, our type VIII in vitro recombinant (Fig. 1A) shows how a small insert of serovar L1-derived DNA in the gyrA gene was accompanied by other segments of serovar L1-derived DNA that would have been undetected if we had not sequenced DNA within 130 kb on one side of gyrA and 318 kb on the other side. Additional sequencing of DNA within a few hundred kilobases of the ompA-neighboring hot spots in recombinant clinical isolates should allow improved estimation of the amounts of trans-DNA from the strains that donated the small inserts.

The in vitro LGT system described here provides an experimental means for investigating possible mechanisms of LGT. Conjugation is the only LGT mechanism known to involve trans-DNA segments in the size range that we found in C. trachomatis recombinants and that was reported by Gomes et al. (10). Our LGT system could be used to produce recombinants that could be examined for characteristics of conjugational LGT other than trans-DNA length. One such additional characteristic is the existence of an origin of DNA transfer (“oriT” in E. coli). The presence of oriT and of a distinct orientation of DNA transfer might be manifested as gradients of genetic marker frequencies among recombinants that were affected by the combinations of selection genes used to isolate recombinants. Such experiments could be undertaken now with available strains that have selectable mutations in known genes (gyrA, rpoB, and 23S rRNA) that are at well-separated locations in the C. trachomatis chromosome. These experiments would be facilitated by the use of additional selectable mutant genes whose locations are known (31, 39). The existence of gene oriT and of oriented DNA transfer in natural DNA transformation has not been reported.

Nature of recombination hot spots.

Seven of 10 independent clinical C. trachomatis isolates had one of two crossovers that were concentrated in a 44-bp region and a 254-bp segment immediately 3′ of the ompA locus (10) and were designated recombination “hot spots.” To what extent does the high incidence of such in vivo recombinants result from crossovers in especially recombinogenic sequences and to what extent does it result from in vivo selection? Use of the in vitro LGT system may allow evaluation of these alternatives. Intrinsically recombinogenic segments would be expected to have nonrandomly high frequencies of exchange among both in vivo and in vitro recombinants. In contrast, in vivo “hot spots” that are not especially recombinogenic might not have high frequencies of exchange among in vitro recombinants if their high frequencies among in vivo recombinants result mainly from selection. We found that none of the 14 in vitro recombinants shown in Fig. 1 had an exchange in the ompA-associated segments designated recombination hot spots by Gomes et al. (10) (data not shown). The difference between 7/10 in vivo recombinants and 0/14 in vitro recombinants with exchanges in designated hotspots is highly significant (P = 0.007, Fisher's exact test). But this difference may have resulted from infrequent cotransfer of the unselected ompA locus with a gyrA or rpoB parental selection marker. We suggest that crosses could be performed with parental strains in which ompA is closer to a selection marker than it was in the cross described here; e.g., the D:Rif^r-1 mutant used in the present study could be crossed with the Lin^r L1:Lin^r-1 mutant used previously (8) because the ompA locus would be only about 95 kb from the mutant 23S rRNA loci that confer resistance to L1:Lin^r-1.

A very low frequency of in vitro Lin^r Rif^r recombinants that have exchanges in designated in vivo hot spots would suggest that in vivo positive selection for hot spot recombinants strongly contributes to the high frequency of such recombinants among clinical isolates. As noted in the introduction, there is evidence for in vivo selection for mutants in which the ompA coding sequence is altered, but comparable MOMP alterations were not reported in clinical isolates that had exchanges in the ompA-associated hot spots (10). This raises the possibility that recombination in hot spots increases fitness by means other than alteration of the MOMP. A search for altered genic expression resulting from hot spot crossovers would be valuable because such crossovers evidently abetted selection in a large proportion of clinical isolates.

If the 553-bp segment in rpoB that contained six crossovers in only 14 in vitro LGT recombinants is intrinsically predisposed to recombination (i.e., is a “mechanistic hot spot”), it should have a nonrandomly high concentration of crossovers in in vivo LGT recombinants. This could be ascertained by sequencing the 553-bp segment in clinical isolates that have a crossover between ompA and, e.g., omcB (3) or pmpC (9, 10), which are only ∼156 and ∼126 kb, respectively, from rpoB and should permit frequent cotransfer of rpoB and omcB or pmpC. Among such recombinants, the frequency of recombinants having crossovers in the 553-bp rpoB segment would not be inflated by selection for rpoB phenotypes. Observation of a nonrandomly high frequency of crossovers in the rpoB segment among in vivo recombinants would validate the use of the segment as a positive “mechanistic hot spot” control for the 14 in vitro LGT recombinants that lacked crossovers in the ompA-associated “hot spots” (10).

Overall, the evidence strongly indicates that LGT is an important process in the evolution of C. trachomatis strains. The in vitro LGT system that we describe here might allow the transfer of alleles of genes thought to differently affect cytotoxicity, tissue tropism, etc., into various genetic backgrounds in order to evaluate the perceived associations between alleles and functions. We show above that it should be feasible to accomplish such transfers between serovars and biovars; serovar L1 strains belong to the LGV biovar, and serovar D strains belong to the urogenital biovar. Ultimately, it would be desirable to determine the effects of allele transfers on function in animal models. It has been suggested (20) that the use of nonhuman primates as model hosts for chlamydiae that infect humans would be prohibitively difficult and expensive, while improved mouse models, including “humanized mice,” might be valuable, affordable alternatives. Such mouse disease models would permit the use of the much-studied MoPn strain of Chlamydia muridarum. It might be possible to use the in vitro LGT system to introduce diverse human alleles of genes that have functional interest into C. muridarum and determine the effects of the allelic substitutions. In this respect, it is encouraging that the order of genes on the chromosome is highly conserved in C. trachomatis and C. muridarum and that their inclusions fuse in cells simultaneously infected with these two species (25).