Pherotypes are driving genetic differentiation within Streptococcus pneumoniae

Pherotype distribution among the pneumococcal population

Traditionally, pneumococcal strains have been characterized by their capsular polysaccharide (serotype) of which pneumococci produce 91 chemically and immunologically distinct variants [23]. Although it has been shown that the serotype defines important epidemiological and virulence properties of pneumococcal isolates [24], it is also recognized that each serotype comprises different clones that may present different properties [25].

The collection of 483 invasive pneumococcal isolates was characterized for the comC allele (pherotype) carried by each isolate. All isolates could be classified either as CSP-1 or as CSP-2 and, in agreement with previous findings, most presented the CSP-1 pherotype (70.6%) [see Additional file 1 - Table S1]. The data was analyzed to determine if the pherotypes were randomly distributed among the population or if there were associations with particular characteristics of the isolates, namely serotype, antibiotic resistance and the genetic lineages identified by pulsed-field gel electrophoresis (PFGE) profiling and MLST.

MLST is a sequence based approach that uses the sequence of internal fragments of housekeeping genes for the purpose of characterizing, typing, and classifying members of bacterial populations. The data derived from MLST can also be used to study the population genetics of bacteria such as Streptococcus pneumoniae [28]. Applying eBURST to MLST data originates subnetworks of isolates with increased probability of sharing a recent common ancestor. These subnetworks define clonal complexes as groups of isolates that share the alleles at no less than six loci with at least another member of that group [29]. MLST from 90 selected strains [30] revealed 57 different sequence types grouped into 39 distinct clonal complexes. The ability of sequence type and clonal complex to predict the pherotype is remarkably high, both with W > 0.97 (Table 1). PFGE and MLST are widely used tools to define bacterial clones, the fact that the groups defined by both these methods show such strong correspondence with pherotype further strengthen the indication that pherotype is a clonal property within the pneumococcal population.

A consistent hypothesis with pherotype clonality is that the role of CSP in triggering competence and its consequences on lateral gene transfer is itself responsible for the distribution of the pherotypes in the pneumococcal population. If this hypothesis is correct and the pherotype is indeed restricting gene transfer within the pneumococcal population, genes that are under recent strong selective pressure and that are known to be horizontally transferred should be associated with pherotype.

Pherotype and antibiotic resistance

The relationship between pherotype and restriction/modification systems

Another important mechanism of lateral gene transfer is bacteriophage transduction [32]. This is an especially important mechanism for the transfer of large DNA fragments that may be restricted in transformation. This is for instance the case of the locus encoding the capsular polysaccharide biosynthesis machinery and of some of the genetic determinants of resistance to tetracycline, chloramphenicol or erythromycin, that are large composite transposons unable to transfer by conjugation, leaving phage transduction as the most likely mechanism of dissemination in the bacterial population, similarly to what was described in other streptococci [33].

Transduction should be independent of CSP activity, but the presence of restriction/modification (R/M) systems was shown to impair horizontal transfer through this mechanism [34]. Pneumococci are unusual in that they posses either one of two complementary R/M systems located in interchangeable genetic cassettes. Strains of S. pneumoniae contain either the dpnI cassete, containing an endonuclease that cleaves only the methylated DNA-sequence 5'GmeATC3' or the dpnII cassette, which includes an endonuclease that cleaves the same sequence when not methylated, together with the corresponding methylase. These mutually exclusive R/M systems were shown to protect against viral infection by viruses produced in cells of the opposite genotype, reducing infection frequency to < 10^-5 [35].

The R/M cassette has a size compatible with horizontal transfer by transformation, so we wondered if the distribution of the R/M cassettes could be correlated to the pherotype and thereby contribute to promote asymmetries of horizontal gene transfer within the pneumococcal population.

To pursue this hypothesis, the R/M cassette carried by pneumococcal isolates previously characterized by MLST was determined. The proportion of CSP-2 isolates with the dpnII cassette (3/23) is lower than the proportion of CSP-1 isolates with that same cassette (25/67) and the association between pherotype and the R/M system is significant (p = 0.037, Fisher exact test), suggesting that phage transduction may be indirectly arbitrated by the pherotype via the R/M systems, such that the spread of large genetic elements that rely on this mechanism of horizontal gene transfer could also be limited by pherotype.

Pherotype is a marker of population segregation

MLST data has been used to characterize the clonality of bacterial populations and to explore the impact of recombination and mutation in bacterial evolution [4]. For S. pneumoniae the recombination rate has been estimated to be 3-10 times the mutation rate per locus [28, 36]. To test if the pherotype could be limiting the genetic exchanges within pneumococci, we took the simple approach of testing among all pairs of sequence types that diverge at the allele of a single locus (single-locus variants - SLV) and that should represent the initial stages of diversification dominated by recombination, if the allele that differed was more frequent among sequence types sharing the same pherotype or among isolates of a different pherotype. Considering the observed SLV pairs in our study, the probability that the changing allele came from a different pherotype is 0.11. In a panmictic population, the expected probability would be 0.38 (p < 10^-4, permutation test), again suggesting that recombination between pherotypes is reduced.

A different approach to test if the pherotype is a marker of genetic isolation consists of calculating the probability that pairs of isolates with increasing levels of genetic divergence have of belonging to different pherotypes. Figure 1 shows that the closest pairs of isolates have a significantly lower probability of having different pherotypes. When genetic divergence increases, the probability of differing in pherotype also increases, reaching the levels expected by chance when isolates differ in more than three alleles. Again, these results show that isolates that are phylogenetically closely linked have an increased likelihood of sharing the same pherotype.

Infinite allele model

The structured nature of the pneumococcal population and the geographically limited origin of our sample could explain, at least partially, the segregation of pherotypes seen in Figure 1 and the high Wallace indices of Table 1. To address this issue a MLST infinite allele model was used to test the effect of restricting or promoting recombination between the two pherotype defined subpopulations. This modeling approach was previously shown to reproduce the clonal structure of the pneumococcal population [36, 41] and provides a possibly more realistic null hypothesis for the distribution of phenotypes in the population. The model was expanded to include a new locus with two possible alleles: CSP-1 and CSP-2. This extra locus recombines with the same rate as the MLST loci and the frequency of each allele is kept constant and equal to 70 and 30% of CSP-1 and CSP-2 respectively, corresponding to the observed values in natural populations. Additionally, a new parameter IPR was introduced, that controls the probability of inter-pherotype recombination. If pherotype differences would not prevent or promote recombination, the observed frequencies of each pherotype in the population would lead to a probability of inter-pherotype recombination of 0.42. Figure 2A shows that even in the absence of a pherotype effect on recombination, high Wallace values of clonal complex predicting pherotype are expected. This result is intuitive since the recent common ancestry of strains belonging to the same clonal complex would also cause them to share the same pherotype. Still, there is a marked shift to higher Wallace values when the probability of inter-pherotype recombination decreases (IPR = 0.1 in Figure 2A). On the other hand, if genetic exchange between pherotypes is favored, in spite of their different prevalence in the population (IPR = 0.9 in Figure 2A), a shift towards lower W_CC→ST values is observed. When systematically varying IPR and computing the probability density for the observed Wallace coefficients (Figure 2B), one concludes that a value of 0.2 is 2-3 times more likely to explain the observed values than an IPR of 0.42, expected in case of no CSP effect in recombination. Since the more probable IPR is lower than expected if the two pherotype populations were recombining freely, these results strengthen the proposal that recombination is promoted within individuals sharing the same pherotype, promoting the divergence of two subpopulations of S. pneumoniae.

Addendum in proof

We recently became aware of a study (Omar Cornejo, personal communication) that has addressed the same issue discussed here. In contrast to our findings, the authors failed to detect any differentiation between the two pherotype defined populations. The reasons behind this discrepancy of results is not clear and further studies are needed to reconcile these apparently contradictory findings.

Bacterial strains, growth conditions, PFGE and MLST

A collection of 483 invasive pneumococcal isolates recovered during the period of 1999 to 2002 in Portugal were obtained from the Faculdade de Medicina de Lisboa collection. The serotype, PFGE type, MLST characterization and antibiotic susceptibility of these strains were collected from previous studies[25, 30, 54]. Briefly, all S. pneumoniae strains were grown in a casein-based semi-synthetic medium (C+Y) at 37°C without aeration or in tryptic soy agar (TSA) (Oxoid, Hampshire, England) supplemented with 5% (v/v) sterile sheep blood incubated at 37°C in 5% CO₂. Antimicrobial susceptibility, serotyping and PFGE analysis was performed for all isolates. MLST analysis was performed for at least one isolate in each major PFGE cluster (n = 90) and revealed 57 different sequence types (ST) corresponding to 39 different lineages by eBURST analysis.

Detection of the pherotype and endonuclease restriction phenotype by PCR

CSP-1 and CSP-2 gene fragments were amplified using multiplex PCR with primers CSP_up (5'-TGA AAA ACA CAG TTA AAT TGG AAC-3'), CSP1_dn (5'-TCA AGA AAG GAT AAA GGT AGT CCT C-3') and CSP2 _dn (5'-TAA AAA TCT TTC AAT CCC TAT TT-3'), which allowed the amplification of fragments of 620 bp for the CSP-1 allele and 340 bp for the CSP-2 allele. dpnI and dpnII genotype was also detected by multiplex PCR with primers DpnI_up (5'-GAA GTA GGA GAT AAA TTG CCA GAG), DpnII_up (5'-TAC GAA TGA TGG GAA TAC TGT G-3') and Dpn_dn (5'-TGT CCT CAA TGC CGT ATT AAA TC-3'), with the expected products of 342 bp and 421 bp for dpnI and dpnII, respectively. Template DNA was prepared by diluting 9 μl of an overnight culture in 441 μl of water and boiling this mixture for 2 minutes. The PCR reactions were performed in 50 μl of final volume containing 20 μl of template solution, 1× reaction buffer (Biotools, Madrid, Spain), 10 mM dNTPs (Fermentas, Vilnius, Lithuania), 20 pmol of each of the primers and 1.25 U GoTaq Polymerase (Invitrogen, Carlsbad, California). The same PCR program was used consisting of 30 cycles of denaturation at 95°C for 1 min, annealing at 55°C for 30 sec, and primer extension at 72°C for 1 min. Followed by 10 min incubation at 72°C to complete extension.

Data analysis

Statistical association between serotypes, PFGE clusters, antimicrobial resistance or endonuclease restriction phenotype and pherotype where characterized by odds ratios (OR) with 95% confidence intervals (CI) computed through the Fisher method implemented in the epitools package for the R language. OR significance was evaluated with the Fisher exact test. The resulting p-values were corrected for multiple testing by controlling the False Discovery Rate (FDR) under or equal to 0.05 through the linear procedure of Benjamini and Hochberg [55].

Wallace coefficients (W) and respective 95% confidence intervals were computed as previously described [26, 27].

The relationship between cross-pherotype pair frequency and the number of divergent alleles between STs was validated for statistical significance by permutation tests. The latter consisted in repeating the computation of frequencies of cross-pherotype strain pairs for 1,000 times, randomly shuffling the pherotype assignment of the strains before each repetition. The p-values were obtained from the fraction of the 1,000 random runs where the cross-pherotype pair frequency was lower than the respective values with the correct pherotype assignment. A permutation test was also performed to evaluate the significance of the probability that a divergent allele in an SLV pair was donated from a strain with a different pherotype. In this case, in each of the 1,000 runs, the divergent allele was randomly sampled from the corresponding locus in the collection of STs. The determination of π, F_ST, K*_ST and S_nn for the analysis of sequence data was done using the DNASP v4.50.3 program. The values of K*_ST and S_nn were used to assess population differentiation in combination with permutation tests (1,000 permutations).

Neutral Multilocus Infinite Allele Model

The model presented by Fraser et al. [36] was expanded to include an additional CSP locus and a new IPR parameter. The CSP locus has only two possible alleles, CSP-1 and CSP-2 that can interchange by recombination but are not affected by mutations. The parameter IPR defines the inter-pherotype recombination probability. The model was simulated with the parameter values determined in [36] for the pneumococcal population. Namely, the population size was 1,000, the population mutation and recombination rates were 5.3 and 17.3, respectively. All the analyses were repeated with a population recombination rate reduced in 50% and the results were qualitatively similar. All simulations were run for 1,000 generations, after which the sequence type diversity was stable, as measured by the Simpson's index of diversity [56]. At each generation, 70% of the selected individuals were CSP-1 and 30% were CSP-2. For each value of parameter IPR, 1,000 independent simulations were carried out. Wallace coefficients for ST and CC predicting CSP type were calculated for each of the final 1,000 populations. Probability density functions for the Wallace distributions were determined by kernel density estimation with a Gaussian kernel function. All simulations and computations were done in Matlab version 7.7.