Rapid Divergence of Two Classes of Haemophilus ducreyi

Abstract

Haemophilus ducreyi, the etiologic agent of chancroid, expresses variants of several key virulence factors. While previous reports suggested that H. ducreyi strains formed two clonal populations, the differences between, and diversity within, these populations were unclear. To assess their variability, we examined sequence diversity at 11 H. ducreyi loci, including virulence and housekeeping genes, augmenting published data sets with PCR-amplified genes to acquire data for at least 10 strains at each locus. While sequences from all 11 loci place strains into two distinct groups, there was very little variation within each group. The difference between alleles of the two groups was variable and large at 3 loci encoding surface-exposed proteins (0.4 < K_S < 1.3, where K_S is divergence at synonymous sites) but consistently small at genes encoding cytoplasmic or periplasmic proteins (K_S < 0.09). The data suggest that the two classes have recently diverged, that recombination has introduced variant alleles into at least 3 distinct loci, and that these alleles have been confined to one of the two classes. In addition, recombination is evident among alleles within, but not between, classes. Rather than clones of the same species, these properties indicate that the two classes may form distinct species.

INTRODUCTION

Haemophilus ducreyi is a Gram-negative coccobacillus in the family Pasteurellaceae (2) and the causative agent of chancroid, a sexually transmitted, genital ulcer disease which facilitates transmission of HIV-1 (12, 14, 23, 44). While H. ducreyi was placed in the polyphyletic genus Haemophilus due to its requirement for hemin, Haemophilus species do not form a cohesive group within the Pasteurellaceae (6, 20). The species that are most closely related to H. ducreyi include Actinobacter pleuropneumoniae and Nicoletella semolina (see Fig. S1 in the supplemental material) (16).

Previous attempts to examine the diversity within the species H. ducreyi using immunotyping (36) and outer membrane profiling (32) showed very few differences among strains. Molecular analyses showed some differences, e.g., ribotyping of strains from two outbreaks yielded multiple restriction fragment length polymorphism (RFLP) patterns (19), but designations based on these differences have not been widely adopted or used for epidemiologic studies, in part due to the difficulty in culturing this fastidious organism. The lack of a standard typing system makes epidemiologic studies more difficult, and epidemiologic data for chancroid are therefore limited.

One way to study diversity in H. ducreyi is to assess its population structure. Whereas the lack of recombination in some bacterial species generates stable clones (5), others recombine sufficiently rapidly to obscure clonal relationships (10, 11). While many Haemophilus species are naturally competent, H. ducreyi is considered nontransformable due to defects in the DNA uptake apparatus (16, 34). This would likely decrease rates of recombination and generate stable clones. Initial studies of the clonal structure of H. ducreyi showed limited diversity in specific outer membrane components, including DsrA, DltA, NcaA, Hlp, MOMP/OmpA2, and lipooligosaccharide (45). The DNA sequences and immunoreactivity suggested that H. ducreyi formed two potentially clonal populations, termed class I and class II, a division supported by proteomic studies (33).

However, the relationships between the two classes are entirely unclear, as genetic differences were assessed only at loci encoding surface-exposed antigens. These loci experience positive and diversifying selection, leading to levels of diversity that are not shared by the majority of genes in the genome (24). It is not clear if these two classes represent two distinct clonal populations of H. ducreyi whose genotypic similarities reflect recent periodic selection (25) or two recombining populations whose similarities reflect recent, locus-specific selective sweeps (18). Here we examine the relationships among H. ducreyi strains using sequences of 11 genes, including five genes that encode cytoplasmic or periplasmic proteins and one that encodes a vaccine candidate. This provides a rich data set with which to perform multilocus sequence analysis (41).

MATERIALS AND METHODS

Strains and culture conditions.

Strains were cultured on chocolate agar supplemented with 1% IsoVitaleX. Strains were revived from frozen stocks, incubated at 33 to 35°C with 10% CO₂ for 24 h, and subcultured no more than once prior to PCR.

PCR and sequencing.

Sequences of the lspA2, murC, ncaA, pal, recA, and wecA loci were obtained from fragments amplified by PCR using the following primers: lspA2, 5′-AAGTTTCAGCAAGAGCGGC-3′ and 5′-TATTGGCTGCAAGCTCTG-3′ (43); murC, 5′-GGCATTGCCGAAGTGTTATT-3′ and 5′-TATGATGGCCCGGTACATTT-3′; ncaA, 5′-GGTTGATTATGTCGAATAATTTG-3′ and 5′-CTAAGCGCGTAAAAATTCGATG-3′ (45); pal, 5′-AGTAGTTCATCAGGTAAAACAGATG-3′ and 5′-AAATTAGTACTCTAATACTGCACGG-3′ (40); recA, 5′-CATTATGGCAGCGGATAAAAA-3′ and 5′-TCCTCAAACGCTTCATCAAA-3′; wecA, 5′-CCGGAATCACCCATAAACAC-3′ and 5′-GATATGCGTTTGCCTTGGTT-3′. The specific locations of amplified regions relative to the indicated gene are indicated in Fig. S2 in the supplemental material. Genomic DNA was obtained using GeneReleaser according to the manufacturer's instructions. Sequencing was done at the University of Iowa DNA Facility, Eurofins MWG Operon, or the Yale University DNA Analysis Facility. All other sequences were obtained from NCBI as follows: dsrA (AF187001-AF187009, AY606120-AY606127, AY612644), sapA (HQ630266-HQ630274), fgbA (HQ630257-HQ630265), hgbA (AY606114-AY606118, AY603046-AY603049), mtrC (HQ712073-HQ712081), and ncaA (AY606128-AY606130, AY612645-AY612647). Sequences for strain 35000HP were obtained from its genomic sequence (NC_002940).

Genome sequences.

The sequences of Bacillus anthracis Ames (NC_003997), Bacillus cereus 03BB102 (NC_012472), Bordetella pertussis Tahoma I (NC_002929), Bordetella parapertussis 12822 (NC_002928), Escherichia coli O157:H7 Sakai (NC_002695), Haemophilus ducreyi 35000HP (NC_002940), Klebsiella pneumoniae 342 (NC_011283), Klebsiella variicola At-22 (NC_013850), Mycobacterium tuberculosis H37Ra (NC_009525), Mycobacterium bovis BCG (NC_008769), Shigella dysenteriae Sd197 (NC_007606), Yersinia pestis CO92 (NC_003143), and Yersinia pseudotuberculosis YIPIII (NC_010465) were downloaded from the NCBI database. The genes were identified using the accompanying annotation.

Computational analyses.

DNA sequence alignments were performed using MUSCLE (8). Substitution rates were calculated using the method of Li et al. (26). Mean substitution rates between entire genomes were calculated as weighted averages of synonymous substitution rate (K_S) values for orthologues. Orthologues were identified as reciprocal best matches in BLAST searches, requiring at least 80% sequence identity. Codon usage bias was calculated as Karlin's E (22), using all genes to represent mutational bias and Sharp's set of 40 translational genes (37) to represent the endpoint of codon selection. Phylogenies were calculated by the Dollo parsimony method (9) or by the maximum likelihood method using PhyML (17).

Nucleotide sequence accession numbers.

Sequences generated by this study (Table 1) have been deposited in GenBank and assigned accession numbers JF806363 to JF806413.

Table 1.

Strains of Haemophilus ducreyi used in this study

Strain	Class	Site of isolation (yr)	Sequence origina
			dsrA	fgbA	hgbA	lspA2	mtrC	murC	ncaA	pal	recA	sapA	wecA
35000HP	I	Winnipeg, Canada (1975)	A	A	A	A	A	A	A	A	A	A	A
HD183	I	Singapore (1982)	ND	A	ND	N	A	N	N	N	N	A	N
HD188	I	Kenya (1982)	ND	A	ND	N	A	N	N	N	N	A	N
85-023233	I	New York, United States (1985)	ND	A	ND	N	A	N	N	N	N	A	N
C111	I	Nairobi, Kenya (unknown)	ND	ND	A	N	ND	N	N	N	N	ND	N
6644	I	Massachusetts, United States (1989)	ND	A	ND	N	A	ND	N	N	N	A	N
82-029362	I	California, United States (1982)	ND	A	ND	ND	A	N	N	ND	ND	A	ND
HMC46	I	Kenya (1995)	A	ND	A	ND	ND	ND	ND	ND	ND	ND	ND
HMC56	I	Dominican Republic (1995)	A	ND	A	ND	ND	ND	ND	ND	ND	ND	ND
HMC60	I	Florida, United States (1989)	A	ND	A	ND	ND	ND	ND	ND	ND	ND	ND
SSMC57	I	Bangladesh (unknown)	ND	ND	A	ND	ND	ND	ND	ND	ND	ND	ND
DMC111	II	Bangladesh (unknown)	A	ND	A	ND	ND	ND	A	ND	ND	ND	ND
SSMC71	II	Bangladesh (unknown)	A	ND	A	ND	ND	ND	A	ND	ND	ND	ND
CIP542	II	Hanoi, Vietnam (1954)	A	A	A	N	A	N	A	N	N	A	N
HMC112	II	Unknown (1984)	A	A	A	N	A	N	A	N	N	A	N
DMC64	II	Bangladesh (unknown)	A	A	A	N	A	N	A	N	N	A	N
33921	II	Nairobi, Kenya (unknown)	A	A	ND	N	A	N	A	N	N	A	N

^aA, previously available sequence; N, new sequence generated in this study; ND, no data available.

RESULTS

Two classes of strains.

The sequences of multiple class I and class II strains of H. ducreyi were previously published for the dsrA and hgbA genes (45). For the ncaA gene, the sequences for six class II strains were available (45), whereas a sole class I sequence was provided by the genome sequence of strain 35000HP (4). To expand this data set, we determined the sequences of the wecA, pal, murC, lspA2, and recA loci for 9 class I and class II strains and the ncaA locus for 6 class I strains of H. ducreyi (Table 1). This data set was further augmented by the sequences of the mtrC, sapA, and fgbA genes (3, 29, 35), kindly provided by Margaret Bauer. Thus, the total data set included sequences at 11 unlinked loci for at least 10 strains with broad global distributions (Table 1). This number exceeds the standard of 7 loci used in conventional multilocus sequence typing (41) and is thus sufficient to illuminate the overall relationships among these strains.

Multiple alignments of the 11 genes provided a total of 10,757 sites, of which 10,122 (94.1%) were invariant (Table 2; Fig. 1). The 635 variant sites included 595 sites (93.7%) which solely distinguished class I strains from class II strains, wherein strains were monotypic within each class at these sites. Of the 595 sites, 332 occurred in the dsrA gene and 263 were distributed in the other 10 genes. Twenty-seven of the other 40 variant sites occurred in the dsrA gene, and 13 were distributed across the wecA, recA, sapA, murC, mtrC, and hgbA genes. The pal, fgbA, lspA2, and ncaA genes lacked any variant sites beyond those that distinguished class I strains from class II strains (Fig. 1).

Table 2.

Summary of nucleotide sequence information

Gene	Position (min)a	Length (nucleotides)b	Ec	Substitution rated		Location; function	Source or reference
				KA	KS
sapA	996773–998523 (59)	1,683	0.468	0.0032	0.0199	Periplasm; peptide transport	29
wecA	1551061–1551456 (91)	398	0.432	0.0000	0.0207	Cytoplasm; synthesis of common antigen	This study
pal	1485485–1485823 (87)	367	0.770	0.0033	0.0250	Outer membrane; peptidoglycan-associated lipoprotein	This study
murC	654024–654714 (38)	694	0.561	0.0056	0.0310	Periplasm; UDP-N-acetylmuramate-l-alanine ligase	This study
recA	325798–326862 (19)	1,073	0.693	0.0099	0.0433	Cytoplasm; recombination-promotion	This study
fgbA	136574–137285 (8)	572	1.144	0.0090	0.0490	Outer membrane; fibrinogen binding	3
mtrC	1258031–1259530 (74)	1,505	0.547	0.0109	0.0561	Periplasm; efflux of antimicrobial peptides	35
hgbA	1686700–1689521 (99)	2,850	0.467	0.0211	0.0851	Outer membrane; hemoglobin binding	45
lspA2	925194–925452 (54)	459	0.449	0.1705	0.4215	Outer membrane; antiphagocytic protein	This study
ncaA	1614393–1614596 (95)	203	0.613	0.0688	0.5863	Outer membrane; collagen adherence	This study; 45
dsrA	603307–604195 (35)	910	0.726	0.0408	1.2232	Outer membrane; serum resistance	45
Total		10,757		0.0112e	0.0518e

^aPosition of orthologous region on the H. ducreyi 35000HP chromosome; numbers in parentheses indicate minutes on a scale of 0 to 100.

^bLength of multiply aligned sequences, including insertions relative to the H. ducreyi 35000HP genome.

^cKarlin's E metric of codon usage bias (22).

^dDivergence between typical class I and class II alleles; only regions with the specified protein-coding gene were analyzed.

^eWeighted average of the eight genes with a K_S of <0.1.

Fig. 1.

Variant and informative sites among 10 H. ducreyi genes. Sites that do not conform to the strict pattern discriminating class I and class II strains are indicated by gray boxes. Site positions, read vertically, are numbered from the beginning of the multiply aligned sequences. The site of the first base of the multiply aligned sequences in the H. ducreyi 35000HP genome is provided in Table 2.

Each of the 11 genes supports the division of strains into two classes (Fig. 1). A parsimony tree was constructed using the 276 variant and informative sites at 10 loci, excluding the dsrA locus (see below). This tree shows strong separation of the two classes with very little homoplasy (Fig. 2). The consistency index of the tree is 0.997 with only a single site showing a reversion. Moreover, class identity is preserved across all loci; in no case did a strain belong to one class at one locus and a different class at another locus, suggesting that gene flow between class I and class II strains is very low. Total variation within the classes is small (2 steps and 5 steps for class I and II, respectively) (Fig. 2A) compared to the discrimination between the classes (263 steps) (Fig. 2B). Since members of each class were isolated over multiple decades and across multiple continents (Table 1), neither represents a single outbreak. Rather, the lack of variability within each class is consistent with recent periodic selection events purging variant alleles within each class.

Fig. 2.

Maximum parsimony phylogeny of strain using variant and informative sites at 10 loci (data in Fig. 1). (A) Relationships among class I and class II strains; (B) relationships among strains with all branches drawn to scale. Vertical marks denote changes.

Variation in levels of diversity.

While alleles at all 11 loci examined distinguish between class I and class II strains (Fig. 1 and 2), the magnitude of this difference varies between genes (Table 2). Previously published data showed large differences between class I and class II alleles at the dsrA and ncaA (45) loci. Inspection of Table 2 shows that these genes, as well as the lspA2 locus, have unusually high levels of divergence at both synonymous (K_S) and nonsynonymous (K_A) sites. In contrast, the majority of genes showed more modest differences between the classes, with divergence at synonymous sites being 10-fold lower for genes encoding cytoplasmic or periplasmic proteins than for the three aforementioned genes which encode surface-exposed proteins.

If high divergence between class I and class II strains at the lspA2, ncaA, and dsrA genes were the result of a relaxation of selection, one would expect comparable divergence at synonymous (K_S) and nonsynonymous (K_A) sites (30). Similarly, if the high divergence were due to positive selection acting on spontaneous mutations, K_A would increase without a commensurate increase in K_S (47). However, since both K_A and K_S were elevated, and the K_A/K_S ratios of these genes are very similar to those of genes with low divergence, we posit that recombination has introduced variant alleles into class I, class II, or both sets of strains from a more distantly related donor. This is not surprising, as recombination is a common source of variant alleles at genes encoding surface-exposed proteins which are subject to positive, diversifying, or frequency-dependent selection (24, 46).

Recombination with the dsrA gene of strain 35000HP.

The 910 aligned sites within the dsrA genes include 359 variant positions, 332 of which simply distinguish class I and class II strains. These sites are distributed across the length of the aligned region (Fig. 3, class A sites). The other 27 sites fall into 4 classes, all but one of which distinguish strain 35000HP from other class I strains. Notably, 25 of 26 of these sites are significantly clustered between bases 157 and 567 in the central region of the dsrA gene (P = 1.6 × 10⁻⁷; χ² test, 1 df). There is no enrichment for class A sites in this region. Setting aside the possibility that the 35000HP genome sequence data are of very poor quality at this locus in this region, the data strongly suggest that recombination has introduced the variant sites into the 35000HP dsrA allele. It is for this reason that we excluded dsrA sequences when constructing phylogenies (Fig. 2). Given that the difference between the 35000HP alleles and other class I alleles is modest relative to the difference between alleles in class I and class II strains (Fig. 2), the data suggest that variability exists within class I strains and that class I strains do recombine with each other. Therefore, class I and class II strains do not represent strictly clonal groups. Recombination may not be detectable among other class I strains owing to the paucity of variation among them.

Fig. 3.

Distribution of variant and informative sites within the dsrA gene. Classes of sites are defined as follows: A, class I and class II strains are distinguished; B, 35000HP is different from all other class I strains; C, 35000HP is variant from other strains; D, 35000HP is different from other class I strains but the same as class II strains (parallelism); E, variant class II. A horizontal bar indicates the regions containing 25 of 26 class B, C, and D sites. Count, number of sites in each class.

Rearrangements.

Beyond the substitutions outlined above, four of the loci examined herein show rearrangements between class I and class II alleles. The lspA2 gene shows a 184-bp insertion/deletion between class I and class II alleles. This could represent a recombination event, either from externally acquired DNA or an internal rearrangement of the chromosome. LspA1, LspA2, and LspB are used by H. ducreyi to avoid phagocytic uptake and are critical virulence factors (42). The LspA1 and LspA2 predicted protein sequences are 86% identical to each other (43). The sequence amplified here for lspA2 represents an area of the gene unique to the lspA2 class I allele in 35000HP. Although class II lspA2 sequences are more similar to class I lspA2 sequences than they are to the class I lspA1 gene, only a complete genome sequence will confirm the orthology of this gene. Second, the fgbA gene, which encodes another surface-exposed product, shows two in-frame insertion/deletion events of 174 and 21 bp (58 and 7 residues). Third, the ncaA alleles are homologous only at the extreme 3′ end. The entirely different 5′ portions of the class I and class II genes reflect illegitimate recombination with an unknown donor. Lastly, the 5′ ends of the dsrA alleles are also quite different, again bespeaking illegitimate recombination. An additional 1-bp deletion is evident in the coding region of the mtrC gene of the class II strain 6644. In all cases, the recombinant regions were excluded from the calculations of substitution rates reported in Table 2. The appearance of these complex variants in genes encoding surface proteins supports the hypothesis that the encoded proteins experience selection for these differences.

DISCUSSION

Variant lineages were created by recombination.

Although the data set presented herein is limited to a few strains, it does shed significant light on the relationships between class I and class II strains of H. ducreyi. The 11 loci around the H. ducreyi chromosome fall into two groups: the sapA, wecA, pal, mtrC, murC, recA, fgbA, and hgbA genes have low diversity, with an average K_S between class I and class II strains of 0.052 (Table 2). In contrast, the unlinked lspA2, ncaA, and dsrA genes are much more divergent, with an interclass K_S ranging from 0.42 to 1.22. The data are consistent with two scenarios (see Fig. S3 in the supplemental material). First, one may consider the two classes to be very closely related, whereby the low-diversity genes would typify the genetic differences between the strains. Here, multiple recombination events with outside strains introduced variant alleles into the genes encoding surface-exposed proteins. Alternatively, the H. ducreyi classes may be quite different, with the dsrA, lspA2, and ncaA genes being typical of other chromosome genes. Here, the similarity at the remaining genes would indicate recent recombination between the two classes at these loci. We reject this alternative since the high variation in K_S values among high-diversity genes is more consistent with recombination with a variety of sources than with high intragenomic variation in substitution rates. While such variation could result from strong variability in levels of codon selection among genes (38, 39), these genes do not encode functions which show strong codon selection (37), and genes with high and low K_S values are similarly weakly biased (Table 2). Therefore, we conclude that the two classes are relatively closely related, with variant alleles having been introduced by lateral transfer. While recombination could occur at any locus, recombinants would most likely be detected at loci experiencing strong selection for variant alleles, such as those encoding exposed proteins interacting with the immune system. Recombination need not have occurred at all loci encoding surface-exposed proteins (e.g., the hgbA gene). Given the rate of synonymous substitution of 1% per million years (28, 31), the two classes diverged ∼5 million years ago, predating the separation of the human and chimpanzee lineages (21).

Are H. ducreyi classes different species?

The data described above suggest that H. ducreyi classes are genotypically distinct and that recombination is not evident between the classes, while recombination is evident within class I. These properties have been used to delineate different bacterial species (7). It is currently not clear if class I and class II strains are phenotypically distinct in a significant manner; additional genome sequencing of a class II strain would shed light on this question. If so, then the two classes of H. ducreyi may form two distinct species, rather than simply forming two clonal groups. When genes affected by recombination with outside genomes (dsrA, ncaA, lspA2) are ignored, the average divergence at synonymous sites between class I and class II strains of H. ducreyi is 0.052 (Table 2). While this difference is modest when considering nonpathogenic species, different species of pathogens are often quite similar to sibling species, as significant phenotypic differences in pathogenicity or host range motivate the taxonomic distinction. For example, Mycobacterium tuberculosis is very closely related to M. bovis, with an average K_S of only 0.003 (Table 3). The difference between the two classes of H. ducreyi is more than 10 times that amount. There are a number of pathogens which are more closely related to sibling species than the two H. ducreyi classes are to each other (Table 3). The long time of separation (5 million years) for this sexually transmitted species also suggests that the two classes of strains may have different lifestyles. A complete genome sequence of a class II strain will assist in the identification of other loci which distinguish these two taxonomic groups.

Table 3.

Average divergence at synonymous sites among pairs of bacterial taxa

Taxon 1	Taxon 2	No. of genes	Mean KSa
Mycobacterium tuberculosis H37Ra	M. bovis BCG	3,704	0.003
Yersinia pestis CO92	Y. pseudotubculosis YPIII	3,404	0.014
Bordetella pertussis Tahoma I	B. parapertussis 12822	3,156	0.029
Klebsiella pneumoniae 342	K. variicola At-22	4,665	0.033
Escherichia coli O157:H7 Sakai	Shigella dysenteriae Sd197	3,066	0.049
Haemophilus ducreyi class I	H. ducreyi class II	8b	0.052
Bacillus anthracis Ames	B. cereus 03BB102	4,646	0.067

^aAverage K_S weighted by gene length.

^bExcluding the 3 genes in Table 2 with aberrantly high K_S values.

Epidemiologic typing of H. ducreyi.

Analysis of clonal complexes or clusters typically involves examination of a set of 7 genes under neutral selective pressure using multilocus sequence analysis (MLSA) (15, 41). The inclusion of multiple loci prevents one's conclusions from being influenced to a high degree by unusual evolutionary events that may affect an individual gene. Because six of the genes presented here (dsrA, hgbA, ncaA, fgbA, pal, and lspA2) encode gene products that interact with the host, they may experience selective pressures atypical of the genome as a whole. Therefore, we expanded the conventional data set of 7 genes to a total of 11 genes, including genes encoding cytoplasmic and periplasmic proteins. Within this expanded data set, all 11 genes supported the same population structure: H. ducreyi strains partition into two, well-separated groups, with no evidence of recombination between them. Thus, this set, or even a subset, of genes would allow for reliable differentiation of clinical isolates and make strain typing for epidemiologic studies more feasible. Among the 11 genes analyzed here, recA, sapA, wecA, mtrC, and murC encode periplasmic or cytoplasmic proteins that are likely to experience little selective pressure from the host and would be most reliable in predicting the class to which a novel strain of H. ducreyi belongs. Using MLSA to classify H. ducreyi strains offers several advantages, including the possibility of performing the assay with PCR-amplified material from clinical specimens (27), eliminating the need for culturing this fastidious organism. Moreover, the lack of recombination increases the clarity of class identification, allowing fewer loci to be used to achieve accurate typing.

Impact of diversity on vaccine development.

The segregation of H. ducreyi strains into two distinct classes affects strategies of vaccine development. Potential vaccines must offer protection against both classes, yet the differences between them will likely complicate the process of developing a single vaccine. For example, a monovalent HgbA vaccine has shown promise in a pig model, where immunization with class I HgbA (HgbA_I) and Freund's adjuvant provided protection against a challenge with class I strains (1). However, immunization with HgbA_I and monophosphoryl lipid A, an adjuvant approved for use in humans, resulted in protection against class I strains but not class II strains (13). This is not surprising, given the differences in the HgbA sequences in class I and II strains (45). Genes encoding surface proteins show large differences between class I and class II strains, including substitutions and insertion/deletion events, which can alter or eliminate epitopes. If the restriction of alleles to either class I or class II strains that we observe can be generalized to the entire genome, as suggested above, then development of a single vaccine that is maximally efficacious for both classes may be difficult or impossible to achieve.