Genomic adaptation in group B Streptococcus following intrapartum antibiotic prophylaxis and childbirth

ABSTRACT

Through vaginal colonization, group B Streptococcus (GBS) causes severe outcomes including neonatal sepsis and meningitis. Although intrapartum antibiotic prophylaxis (IAP) has reduced neonatal disease rates, GBS can persist in the genitourinary tract even after antibiotic administration. To determine if IAP selects for genomic signatures that enhance GBS survival and persistence, we compared the isolates from individuals before (prenatal) and after (postpartum) IAP/childbirth. Among 34 of the paired strains from participants with persistent colonization, 31 (91.2%) clustered together in a core gene phylogeny, suggesting colonization with highly similar strains before and after IAP. A core-gene mutation analysis, however, identified mutations in 74% (n = 23) of these 31 postpartum genomes when each genome was compared to its respective prenatal genome from the same individuals. Several strains acquired mutations in the same genes, although two postpartum strains accounted for most of the mutations. These two strains were classified as mutators based on high mutation rates and mutations within DNA repair system genes. Changes in biofilm production were observed in a subset of postpartum strains, which is supported by the presence of point mutations in genes linked to survival and colonization. These findings suggest that exposures encountered during pregnancy and childbirth may select for mutations and phenotypes that promote adaptation and survival in vivo. Enhanced survival in the genitourinary tract can lead to persistent colonization, increasing the likelihood of invasive disease in subsequent pregnancies and in newborns (late-onset infections) following IAP cessation.

IMPORTANCE

GBS remains a major cause of neonatal sepsis, pneumonia, and meningitis despite the common use of IAP that aims to eradicate maternal colonization, the main risk factor for neonatal disease. Although IAP has reduced the incidence of early-onset neonatal infections, it has had no impact on late-onset infections in babies between 7 days and 3 months of age. Since colonization is intermittent and GBS has been shown to persist in the genitourinary tract despite antibiotic exposure, more research is needed to understand mechanisms of adaptation. By comparing the genomes of GBS strains recovered before (prenatal) and after (postpartum) IAP and childbirth, this study demonstrates how selective pressures shape GBS evolution, favoring traits that promote survival and persistence. Understanding adaptive traits is essential for improving diagnostics, refining prophylaxis strategies, and guiding the development of more effective prevention practices that can reduce the likelihood of GBS transmission to neonates.

INTRODUCTION

Group B Streptococcus (GBS) is an opportunistic bacterial pathogen that was suggested to colonize up to 35% of pregnant patients, although variation in colonization frequencies has been observed across geographic locations (1). Rectovaginal colonization during pregnancy is the main risk factor for invasive neonatal infection (2, 3) that can lead to sepsis and other adverse outcomes, including preterm birth and stillbirth (4, 5). Following birth, GBS can cause early-onset disease (EOD), presenting as sepsis and/or pneumonia within the first week of life, or late-onset disease (LOD), which occurs between 1 week and 3 months after birth and causes bacteremia and/or meningitis (6).

In the United States, screening for rectovaginal colonization is recommended between 36 0/7 and 37 6/7 weeks’ gestation (7). To prevent neonatal infections, GBS-positive patients are given intrapartum antibiotic prophylaxis (IAP) during labor, for which penicillin is the most effective (3). While IAP has successfully reduced the incidence of EOD, it has not impacted LOD rates nor is it useful for preventing preterm births or stillbirths (8). It can also negatively impact the microbiome (9, 10). Importantly, GBS colonization was found to be intermittent during pregnancy with variation in colonization frequencies across individuals and geographic locations (11, 12). One study of EOD cases, for example, showed that 60% of cases were born to individuals whose GBS prenatal screening test was negative (13), thereby highlighting the need for improved treatment and prevention protocols.

Classifying GBS based on the polysaccharide capsule (cps), which dictates the serotype, and the multilocus sequence type (ST) has identified associations between molecular traits and disease severity (14). Capsule type III ST-17 strains predominate among neonates with invasive disease (14–16) and possess unique features that help withstand oxidative stress, survive inside macrophages, associate with host cells, and elicit proinflammatory responses relative to other STs (17–21). Less definitive epidemiological associations have been observed for other lineages (e.g., STs 19 and 23) despite their high prevalence globally (15). IAP effectiveness was also shown to vary among STs in our prior study (22). STs 17 and 19 more commonly colonized patients before and after IAP while persisting for up to six weeks postpartum, while ST-12 strains were more frequently eradicated after IAP, which was also independently associated with persistent colonization. Despite these associations, the mechanisms that promote GBS persistence in the vaginal tract and evasion of antibiotic-mediated killing remain elusive. This is a serious concern since persistent colonization enhances the risk of invasive disease, especially for subsequent pregnancies and babies after IAP cessation (9), and may partly explain why GBS LOD rates have not decreased in the United States (8, 9).

Altogether, these prior findings led us to hypothesize that IAP imposes a selective pressure on GBS, such that genomic features that promote colonization are selected for, resulting in enhanced fitness and survival in the genitourinary tract. We used whole-genome sequencing (WGS) to compare GBS isolates recovered from pregnant participants sampled both before (prenatal) and 6 weeks after (postpartum) IAP and childbirth (23). This comprehensive genomics analysis has enhanced the understanding of evolution and adaptation among colonizing GBS populations, concurrently providing insight for treatment and prevention improvements.

RESULTS

Antibiotic use and GBS colonization dynamics differ across participants

Isolates (n = 97) were previously collected from 58 participants at two time points during pregnancy and after childbirth (23). The first sampling took place at a prenatal visit between 35 and 37 weeks gestation, which was the previously recommended timing for GBS screening but differs from recent guidelines of 36–37 weeks gestation (7). The second sampling took place up to 11 weeks later at a postpartum visit. Among the 58 participants included in this follow-up study, 53 (91.4%) were given IAP during childbirth, with most receiving ampicillin (n = 26; 49.1%) or penicillin (n = 17; 32.1%) (Table S1).

Forty-one of the 58 (70.7%) participants had persistent colonization, meaning they were positive for GBS at both the prenatal and postpartum samplings even though IAP was given to all but three individuals (n = 38/41; 92.7%). Among these 41 persistently colonized participants, 35 received a β-lactam antibiotic (ampicillin, n = 21; penicillin, n = 13; or cefazolin, n = 1) and three were given clindamycin. Four of the participants also reported taking antibiotics after childbirth. One was treated with cefazolin, and three received either ampicillin (n = 1), a combination of trimethoprim and sulfamethoxazole (Septra) with clindamycin and nitrofurantoin (n = 1), or metronidazole and cephalexin (n = 1) prior to the postpartum sampling. The remaining 17 (29.3%) participants were colonized with GBS only at the prenatal sampling, indicating loss of GBS colonization by the postpartum visit; two of these participants did not receive IAP.

As determined previously (22), the 97 GBS strains belong to 20 STs comprising clonal complex (CC)-1 (n = 25), CC-12 (n = 15), CC-17 (n = 16), CC-19 (n = 17), CC-22 (n = 2), CC-23 (n = 20), and CC-26 (n = 1) plus one singleton (ST-67). Eight distinct cps types were represented, with cps III (n = 27) predominating followed by Ia (n = 21) and V (n = 17); four isolates were non-typeable (NT).

Pangenome analysis reveals a large accessory genome and clustering by CCs

Ninety-two of the 97 genomes were included in the genomics analysis. Three prenatal and two postpartum genomes from five participants were excluded due to poor assembly quality (Table S2), whereas two postpartum strains from two persistently colonized participants (IDs 11 and 14) were not available for WGS. Among the 92 high-quality genomes, 68 represented pairs of strains recovered from 34 participants with persistent colonization (i.e., participants were positive for GBS at both the prenatal and postpartum samplings). Seven unpaired strains from seven persistently colonized participants collected at the prenatal (n = 4) or postpartum (n = 3) sampling were available for inclusion as well as the 17 prenatal strains from participants who had lost GBS by the postpartum sampling.

The GBS pangenome consists of 5,054 unique genes, comprising a small core genome with 1,368 core and 213 soft-core genes (Table S3). It has a large accessory genome containing 945 shell genes and 2,528 cloud genes, which accounts for 68.7% of the total genes. A phylogenetic tree with the highest log likelihood was constructed based on the 1,368 core genes using the maximum likelihood method and Tamura-Nei model of nucleotide substitutions (24) in MEGA12 (25); five sequence clusters were resolved, grouping together with 100% bootstrap support (Fig. S1). Each of the five clusters was associated with a previously defined CC (Fig. 1). The 17 CC-19 strains and 16 CC-17 strains are grouped into two separate clusters as the CC-1 (n = 23), CC-12 (n = 14), and CC-23 (n = 19) strains. Several singletons were detected within each cluster. Construction of a neighbor-net tree in SplitsTree v.19.2 (26), which was based on 17,825 parsimony-informative (PI) sites within the 1,368 core genes, confirmed the clustering observed in the maximum likelihood phylogeny and uncovered evidence for recombination (pairwise homoplasy index [PHI] P-value < 0.001; Fig. S2).

Fig 1.

Evolutionary history of 92 GBS strains isolated from participants before and after IAP and childbirth. The bootstrap consensus tree based on 1,368 core genes (1,268,723 positions) was inferred from 88 replicates (27) with the collapsed branches corresponding to partitions reproduced in less than 50% of replicate trees. The original maximum likelihood tree (Fig. S1) represents the one with the highest log-likelihood between a neighbor-joining (NJ) tree (28) and maximum parsimony (MP) tree. The phylogeny was constructed in MEGA12 (25) and visualized with iTOL (29) while rooting at the midpoint. Participant strain IDs are noted at the end of each branch; persistent strains contain a “.1” (prenatal) or “.2” (postpartum) indicating the sampling time point. The clonal complex (CC) is indicated by colored strips and the circles correspond to the antibiotic(s) used for IAP. Paired strains from the same participant have the same antibiotic usage profiles. IDs in bold red font are the paired prenatal and postpartum strains from four participants (IDs 3, 8, 22, and 43) that do not cluster together in the tree.

Among the 34 pairs of strains from participants with persistent colonization, 30 (88.2%) clustered together in the phylogenetic tree (Fig. 1; Fig. S1). The paired strains from the three participants (IDs 50, 55, and 56) who did not receive IAP were also grouped together. Of the four pairs that did not group together, two (IDs 3 and 8) had prenatal and postpartum strains with different STs, suggesting acquisition of a new strain type following childbirth and IAP. Although the remaining two participants (IDs 22 and 43) had prenatal and postpartum strains with the same ST, only the prenatal and postpartum strains from participant 43 clustered together. Inclusion of only the paired strains in the consensus tree better illustrates that prenatal strain 43.1 is ancestral to postpartum strain 43.2 (Fig. S3).

Acquisition of antibiotic resistance genes is not important for persistent colonization

To determine whether GBS residing in the vaginal tract acquired antibiotic resistance genes (ARGs) to promote colonization, an assembly-based analysis was performed to extract these genes from the 92 genomes. Nine distinct ARGs were identified conferring resistance to tetracyclines (tetL, tetM, tetO, tetW, lmrP), macrolides-lincosamides-streptogramins (MLS) (ermA, lmrP, mreA), fluoroquinolones (norB), and cationic peptides (mprF) (Fig. 2). Both lmrP and mprF were found in all strains, while norB and mreA were found in all but one.

Fig 2.

Antibiotic resistance genes (ARGs) were detected in the 92 GBS genomes. ARGs were extracted from the 92 high-quality genome assemblies using five different ARG databases. The x-axis shows the patient strains with a “.1” (prenatal) and “.2” (postpartum) after the “Participant ID” indicating the sampling time point. The gene profiles for the paired isolates from the same 32 participants with isolates of the same sequence type (ST) are shown next to each other on the left side of the vertical blue line. The two sets of paired strains (IDs 3 and 8) with different STs along with seven unpaired strains are shown between the blue and red vertical lines. The ARG profiles for the 17 “lost” strains, which were recovered only at the prenatal sampling and did not persist, are to the right of the red vertical line. ARGs are indicated on the y-axis that confer resistance to tetracyclines (tetL, tetM, tetO, tetW, lmrP), macrolides-lincosamides-streptogramins (ermA, lmrP, mreA), fluoroquinolones (norB), and cationic peptides (mprF). The color gradient shows the average percent identity of a gene across the five databases, with black representing the highest (100%) identity and light green representing the lowest (<10%) identity.

Among the variably present ARGs, tetM predominated and was detected in 63 (68.5%) genomes, including 47 of the 68 (69.1%) genomes from the 34 participants with persistent colonization. A similar proportion (n = 11; 64.7%) was detected in 17 prenatal strains that were lost following IAP. Although the prenatal and postpartum strains from participant 22 had the same ST, tetM was only detected in the prenatal strain (22.1). A discrepancy was also observed between the paired genomes from participant 3, although this finding is less surprising in that the two strains had different STs, strongly suggesting the acquisition of a new strain by the postpartum visit.

Specific virulence genes or gene profiles are not linked to persistent colonization

The 92 genome assemblies were also interrogated for known virulence genes; 50 distinct genes were detected with 65–100% identity and classified based on function (Table S4). Genes that are important for adherence and invasion of host cells (n = 18, 36%), immune modulation (n = 17, 34%), metabolism (n = 1, 2%), toxin production (n = 13, 26%), and dissemination (n = 1, 2%) were found. The most variable genes were those that are used for capsule (cps) production and formation of pili (pil) (Fig. 3).

Fig 3.

Virulence genes were detected in 92 GBS genomes. The x-axis shows the participant ID with the “.1” (prenatal) and “.2” (post-partum) indicating the sampling time point. Gene names are displayed on the y-axis. The gene profiles for the paired isolates from 32 participants with isolates of the same sequence type (ST) are next to each other, left of the blue line. The seven unpaired strains plus the two paired strains with different STs from participants 3 and 8 are shown between the purple and red vertical lines. Profiles for the 17 “lost” strains are to the right of the red line. The color gradient displays the average percent identity of a gene, with black representing 100% identity and mint green representing <10% identity.

Among 32 of the 34 participants persistently colonized with strains of the same ST, several discrepancies in gene presence were observed between the prenatal and postpartum strains per pair. In most cases, the differences involved genes with values that fell below the identity and coverage thresholds, indicating that they are present but may be variable. The paired strains from participant 22, however, had multiple differences. Most notably, distinct capsule (cps) types (serotype) were observed in the prenatal (cpsV) and postpartum (cpsVI) strains, as noted previously (22), and could indicate capsular switching, colonization with more than one strain type, or acquisition of a new strain by the postpartum sampling. Not surprisingly, participants 3 and 8 with prenatal and postpartum strains belonging to different STs also had distinct virulence gene profiles.

To further compare the virulence gene profiles among the paired strains from the participants with persistent colonization, a principal component analysis was performed. No distinct clustering patterns were observed for the prenatal versus postpartum strains (Fig. S4A) or after stratifying by antibiotic, cps type, or colonization phenotype (persistent or lost) (Fig. S4B through E). In addition, no difference was observed between colonization phenotype and the number of adherence (n = 1,031, chi-square P = 0.10) or immune modulatory (n = 1,231, chi-square P = 0.52) genes. These data suggest that neither possession nor acquisition of specific virulence genes was linked to persistence in the genitourinary tract after IAP and childbirth.

The number of point mutations varies across persistent strains

To detect nucleotide-level variation, we conducted a reads-based analysis of the 1,368 core genes in the 64 paired prenatal and postpartum strains that shared STs. In this analysis, each prenatal genome was used as the reference genome for its paired postpartum genome to determine the number and type of mutations acquired. No mutations were detected in eight (25.0%) postpartum genomes, which have identical core gene sequences as their paired prenatal genomes. Moreover, only 2, 3, or 4 total mutations were detected in the postpartum genomes from three participants who did not receive IAP when compared to their respective prenatal isolates. The remaining 24 postpartum genomes, however, have at least one mutation relative to the prenatal genomes (Table S5). Although a range of 1 to 6,564 mutations was detected per genome, the median was only 2.5 mutations per genome due to the presence of three outliers that skewed the distribution (Fig. 4A). These outlier genomes contained 204, 262, and 6,454 total mutations. Omitting the three outliers resulted in an average of five mutations per genome, with a median of 2 and range of 1–16 mutations for the remaining 21 postpartum genomes (Fig. 4B).

Fig 4.

Number of core gene mutations in the postpartum GBS genomes with at least one mutation. Both plots show the number of mutations (x-axis) by the number of genomes (y-axis) for (A) all 24 postpartum genomes, using a bin width, or interval size, of 10. (B) The subset of 21 postpartum genomes after removing three outliers with the greatest number of mutations with a bin width of 3. The eight genomes without any mutations were omitted from these plots.

Placement of the three outlier genomes within the maximum likelihood core gene phylogeny alongside the prenatal genomes (Fig. 1) demonstrates that only two of the three strain pairs are closely related. Except for strain 22.2, which had the greatest number of mutations, the other two outlier strains, 36.2 and 43.2, grouped together with their respective prenatal genomes. The pair of strains from participant 36 shares a node in the phylogeny, while the prenatal strain 43.1 is on an adjacent ancestral branch to its paired postpartum strain, 43.2 (Fig. S1). These data suggest that the genetic backbones of these two outlier postpartum genomes are highly similar to their respective prenatal genomes despite the high number (>200) of core gene mutations detected. By contrast, placement of the 22.2 postpartum genome in the phylogeny relative to its paired prenatal genome (22.1) suggests that strain 22.1 is not closely related or ancestral to strain 22.2. Additional support for this suggestion comes from the gene analysis showing differences in the presence of some resistance and virulence genes between the pair of strains (Fig. 2 and 3). Together, these data indicate that participant 22 likely acquired a new strain of the same ST and CC with a distinct cps type by the postpartum visit.

The postpartum genomes have a variety of mutations and variable mutation rates

After excluding the extreme outlier, postpartum genome 22.2, we characterized the remaining 571 mutations in the 21 postpartum genomes with at least one mutation (Table S5). Most (n = 445; 77.9%) of these mutations were in coding regions, although the type of mutations varied across genomes. In all, single nucleotide polymorphisms (SNPs) (n = 411; 72.0%) were most common, with 350 of them mapping to coding regions. Eighty-three (14.5%) insertions, 48 (8.4%) deletions, and 29 (5.1%) complex mutations, defined as a combination of SNPs, insertions, and/or deletions (indels), were also detected. Among the 350 characterized SNPs within coding regions, most were classified as missense (n = 205; 58.6%) or synonymous (n = 129; 36.9%) mutations. Sixteen (4.6%) nonsense mutations were also identified.

Given the wide range and frequency of mutations observed, mutation rates were calculated based on the total assembly length per genome. The mutation rate, which represents the number of mutations per number of base pairs (bp), was determined for each postpartum genome assembly relative to its respective paired prenatal genome assembly (Table S5). Notably, the average mutation rates for the two outlier genomes were 9.37 × 10⁻⁵ (ID = 36.2) and 1.32 × 10⁻⁴ (ID = 43.2), which are 53- and 75-fold higher, respectively, than the average mutation rate (1.77 × 10⁻⁶ mutations/bp) for the remaining 29 genomes (including those without any detectable mutations).

Among the three persistently colonized participants (IDs 50, 55, and 56) who did not receive IAP, the average mutation rate was 1.50 × 10⁻⁶ mutations/bp, which is 6.5 times lower than the average rate (1.16 × 10⁻⁴) for the remaining 28 participants who were given IAP, and 1.2 times lower after excluding the two outlier genomes (36.2 and 43.2). Although the difference in means was not significant (Student’s t-test P = 0.64 and P = 0.18, respectively), the mutation rate for strains without IAP exposure includes postpartum strain 56.2 (n = 4 mutations) from one of four individuals reporting antibiotic use (cefazolin) after childbirth.

Specific genes and lineages are more likely to acquire mutations in vivo

To define which genes possessed SNPs that could potentially alter functionality, we mapped the 350 characterized SNPs found within coding sequences to 79 known genes representing 78 gene products (Table S6). Although 237 (67.7%) SNPs were detected in unknown (unclassified) genes, the gene product was known for most and represented 148 different proteins. Twenty of these products are hypothetical proteins. The two outlier postpartum genomes from pairs 36 and 43 possessed most of the characterized SNPs (n = 292; 83.4%), which were detected within 62 known and 198 unclassified genes for an average of 1.1 SNP/gene. After excluding both outliers from the analysis, the 58 characterized SNPs from the remaining 29 genomes were found within 19 known and 39 unclassified genes, yielding a similar average of 1.0 SNP/gene. It is notable that 27 genes have more than one SNP, although 14 of these are in unclassified genes. Among these 27 genes, missense (nonsynonymous) and/or synonymous mutations were most common, however, nonsense mutations were detected in two genes, mutY (DNA repair) and an unknown gene (1,170 bp) encoding a sugar transferase. Importantly, more than one genome had mutations within the same 14 genes such as asnS, ezrA, mutY, srr1, and vicK, which encode for products used in tRNA function, cell division, DNA repair, adherence, and cell signaling, respectively.

Variation in the number of mutations was also observed by bacterial genotype. Overall, the postpartum genomes belonging to CC-19 (n = 121) and CC-23 (n = 199) had the greatest number of SNPs (Table S7), yet these numbers are skewed by inclusion of the two outliers. Excluding the CC-19 and CC-23 outliers from the analysis still showed that the CC-23 strains have the most SNPs (n = 21), ranging from 0 to 10 per strain, followed by strains of CC-12 and CC-1, which contain 15 and 12 SNPs, respectively. The remaining six CC-19 postpartum strains had only seven SNPs (0–3 SNPs per strain), whereas the six CC-17 postpartum strains acquired the fewest SNPs overall (n = 3) with 0–1 SNP per strain. Among the three SNPs identified in the CC-17 strains, only one was nonsynonymous, whereas the strains belonging to CCs 23, 12, and 1 collectively had twice as many nonsynonymous (n = 32) SNPs as synonymous (n = 16) SNPs. Inclusion of the CC-23 outlier in this analysis revealed 1.7 times more nonsynonymous mutations than synonymous mutations in CCs 1, 12, and 23 combined. Genomes representing these three CCs also had 1.8 times more nonsynonymous mutations than those detected in all of the CC-17 and CC-19 genomes (and includes the CC-19 outlier genome).

Mutations in DNA repair and proofreading genes uncover GBS “mutator” strains

Because the two outlier postpartum genomes have higher-than-average mutation rates, a common trait of “mutator” strains (30), we sought to determine whether either genome had acquired mutations in genes encoding mismatch repair (MMR) and/or proofreading systems. Indeed, mutators are known to arise at a low frequency in the population due to mutations within genes encoding DNA replication and proofreading machinery, MMR pathways, and the 8-oxo-dG (GO) system (30, 31). We therefore compared the genes involved in these pathways among the prenatal and potential mutator strains from participants 36 and 43. The six MMR pathway genes (7,467 bp region) were extracted from each genome and were aligned using MUSCLE (32). These genes encode DNA MMR proteins, MutS and MutL, a cold shock protein, an MFS transporter/LmrP homolog, the Holliday junction branch migration protein (RuvA), and the DNA-3-methyladenine glycosylase I. Relative to its paired 36.1 prenatal genome, the postpartum 36.2 genome has a deletion in the error-prone DNA polymerase IV gene (dinB_2) as well as a missense SNP in mutY (A/G-specific adenine glycosylase) of the GO system (Table S6). Comparatively, genome 43.2 has a missense SNP in the DNA topoisomerase gene (parE) and a 1 bp deletion downstream of the MMR loci relative to its paired 43.1 strain. Although the MMR loci were identical in the remainder of the paired prenatal and postpartum genomes, postpartum genome 26.2 also possesses a missense SNP in parE relative to the prenatal 26.1 genome. However, the mutation rate (n = 10 mutations) was not markedly different than the average. Altogether, the increased mutation rates in genomes 36.2 and 43.2, along with the identification of mutations in regions important for DNA repair and proofreading, meet the criteria (30, 33) for classifying the two outlier strains as “mutators.”

It is worth noting that the two mutator strains had been exposed to penicillin, which was reportedly used for IAP in participants 36 and 43 during childbirth (23). Among the 16 participants given ampicillin and the 10 receiving penicillin, 28 and 310 SNPs were detected in the postpartum strains with an average of 6 and 62 SNPs, respectively (Table S7). After excluding the two mutators, the number of SNPs detected among the eight participants who received penicillin decreased from 310 to 18 SNPs. No difference in the presence of SNPs was observed in the genomes of participants who were given penicillin versus ampicillin (Fisher’s exact test P = 0.68). Similarly, no difference in the average number of SNPs was observed when comparing those genomes from participants who had received ampicillin versus penicillin (Student’s t-test P = 0.17). While comparisons could not be made for the other antibiotics due to low numbers, the three genomes from individuals who did not receive IAP only had 1, 3, and 4 total mutations with an average of 1 SNP/strain. The strain with four mutations, however, had been exposed to cefazolin after childbirth and prior to the postpartum sampling.

Variation in biofilm production was observed in the postpartum strains

Since biofilm formation is a common colonization strategy and mutations were found in genes important for adherence, we examined this phenotype in the paired strains. A significant increase in biofilm production was observed in four (12.9%) of the 31 postpartum strains relative to each respective prenatal strain, while a significant decrease was observed in two (6.5%) strains (Fig. 5). Among the four postpartum strains with increased biofilm production, no mutations were found in the genomes from participants 39 and 46, however, genome 37.2 has nonsynonymous SNPs in ezrA and two unknown genes encoding a cation-anion antiporter (1,128 bp) and phosphoglycerate dehydrogenase (1,182 bp). Genome 34.2 has missense SNPs in an ABC transporter ATP-binding protein/permease (1,740 bp) and a tyrosine-protein kinase (699 bp), as well as a nonsense mutation in a replication initiation protein (999 bp). By contrast, the two postpartum strains from participants 26 and 27, with significantly reduced biofilm production relative to the prenatal isolates, had 16 and 2 total mutations, respectively. Notably, postpartum genome 26.2 has nonsynonymous mutations in genes encoding binding and surface proteins like the penicillin-binding protein (Pbp2B) and capsule protein (CpsC), as well as those that are important for growth, such as manY and ccpA (Table S6). Among the two mutators, one (36.2) had increased biofilm production, and one (43.2) had decreased production relative to the respective prenatal strains (Fig. S5), although the difference between pairs was not significant.

Fig 5.

Biofilm formation between prenatal and postpartum strains from 30 participants with persistent colonization. Biofilm formation was assessed across paired strains by measuring the absorbance (OD₅₉₅) values (y-axis) and averaging the biological replicates (n = 3) that were each conducted in triplicate. Prenatal (mint green) and postpartum (dark blue) isolates are shown by participant ID (x-axis). One-tailed, paired t-tests were calculated to identify significant differences in biofilm formation between the paired strains (* = P < 0.05, ** = P < 0.01). Each measurement was normalized to a negative media control. Error bars display the standard deviation.

DISCUSSION

Through this WGS analysis, we have identified a range of genomic alterations in GBS strains recovered after IAP and childbirth compared to strains isolated during prenatal visits from the same individuals. Although the pangenome analysis of 1,368 core genes uncovered five distinct sequence clusters, most of the paired prenatal and postpartum strains with identical STs clustered together in the phylogeny. This finding, along with a comprehensive evaluation of point mutations, indicates that in most cases the strains that persisted are highly similar to those recovered prior to IAP and childbirth. Indeed, eight of the 32 participants had postpartum strains with no core gene mutations relative to the prenatal strains, and another 12 participants had ≤3 mutations, thereby demonstrating how stable the colonizing GBS population is despite antibiotic exposure.

Even though various antibiotic resistance mechanisms are known to enhance bacterial survival in the presence of antibiotics, our assembly-based genomic analysis failed to detect newly acquired resistance genes in the postpartum strains relative to the highly similar prenatal strains from the same participants. Hence, resistance to the antibiotics used for IAP is not a mechanism of persistent colonization in this study population. Among the paired postpartum genomes from the 26 (83.9%) participants who received β-lactam antibiotics for IAP, none had β-lactam resistance genes, which is not surprising since GBS remains universally susceptible to β-lactam antibiotics (9). Resistance to MLS antibiotics is more common in GBS (9), yet the three (9.7%) participants given clindamycin for IAP had postpartum strains that were neither phenotypically resistant (23) nor possessed macrolide resistance genes. A similar analysis also demonstrated that acquisition of specific virulence genes was not associated with persistence, since distinct variably present genes were not detected in the postpartum versus prenatal strains. While these findings indicate that horizontally acquired resistance and virulence genes are not a mechanism of GBS adaptation in this patient population, there are limitations associated with assembly-based analyses. Genes labeled as absent, for instance, may have been present below the detection threshold due to poor coverage or allelic variation or may exist in a region that is more difficult to sequence. Since the only discrepancies observed were in three pairs of strains that did not cluster together in the core gene phylogeny (Fig. 1), however, it is more likely that these were genetically distinct strains. Indeed, the postpartum strain could have been acquired following IAP/childbirth or could have been present at the prenatal visit but not selected for characterization. Since only a single colony was selected from each sample in the original study (23), these participants may have been unknowingly colonized with more than one strain type. A meta-analysis involving 18 studies from 14 countries determined that 10% of individuals with asymptomatic colonization were colonized with strains belonging to more than one GBS serotype (34).

As was indicated in the reads-based analysis of 1,368 core genes, the accumulation of point mutations appears to be most important for GBS adaptation. In all, 0 to 262 mutations were detected in the 31 postpartum genomes relative to their respective prenatal genomes (Table S5), with two genomes possessing most (n = 466; 81.6%). It is important to note that each mutation was manually confirmed to rule out sequencing artifacts, which can occur in sequencing analyses, and each was determined to be in an area of >10× read coverage. After excluding the two outlier genomes, we uncovered 105 mutations in the 29 genomes with a range from 0 to 16 mutations per genome, which is slightly greater than the total number of mutations (n = 29) observed in a prior study comparing paired strains from 19 pregnant patients and their sick neonates (35). In the paired neonatal strains, mutations were detected within covR that encodes a two-component response regulator linked to virulence (36) and other virulence-associated genes. Although covR mutations were not observed in our postpartum genomes, several strains had mutations in genes encoding transcriptional regulators and ABC-transporters as well as colonization factors like accessory Sec system secretion proteins (asp2, gtfA) (37), serine-rich repeat (srr) glycoprotein adhesins (38), and sortases (srtC) linked to pilus formation (39). It is therefore possible that these mutations enhance fitness in vivo while promoting adaptation to the genitourinary tract. Since epistasis, or the impact of bacterial genotype on the fitness effects of a mutation, has been described (40), it is also noteworthy that the three CCs (i.e., 1, 19, and 23), most commonly associated with maternal colonization (15), had the greatest number of mutations overall (Table S7). While CC-17, the lineage associated with neonatal disease, is also commonly linked to maternal colonization (14–16), the CC-17 postpartum strains examined herein had the fewest mutations overall, indicating a pattern of negative or purifying selection for strains of this lineage. On the contrary, most mutations in the colonizing lineages, CCs 1, 19, and 23, were categorized as nonsynonymous SNPs, suggesting a pattern of positive/diversifying selection and a trend of adaptive evolution. Despite these findings, future experiments are critically important to confirm the impact of each mutation on GBS fitness and survival in different strain backgrounds.

It is also notable that functional changes in biofilm formation were observed in several postpartum strains relative to their respective prenatal strains. Despite the wide range in biofilm levels observed, which is consistent with studies showing differences across lineages, pilus types (41), and culture conditions (42), only four of the postpartum strains had a significant increase in biofilm levels relative to their respective prenatal strains. Among these, no mutations were found in two genomes, and the remaining two lacked mutations in genes commonly linked to biofilms or adherence. Mutations in genes encoding products linked to cell signaling (PstA, tyrosine-protein kinase) and cell division (EzrA) were detected and could influence biofilm formation (43, 44); however, these have not been evaluated in GBS. Even though the 2.6-fold increase in biofilm production was not significant for mutator strain 36.2 (Fig. S7), mutations were detected in similar signaling genes as well as srtC (class C sortase), Sec-system proteins (GtfA, Asp2), and a serine-rich glycoprotein adhesin (srr1) that has been linked to persistent vaginal colonization in mice (45). Comparatively, postpartum strain (26.2) had the largest decrease in biofilm production and harbors 16 mutations in genes linked to DNA replication, protein synthesis, metabolism, and cell signaling, which could impact post-biofilm growth, adaptation, and survival. A missense SNP in the capsular polysaccharide biosynthesis protein gene, cpsC, was also detected. Given the tradeoff between capsule production and adherence, which has been described in S. pneumoniae (46), it is reasonable to speculate that this mutation may promote capsule production while simultaneously blocking adhesins that could be required for aggregation. Together, these data demonstrate the need for additional mutagenesis and functional studies, as well as the use of more sophisticated biofilm quantification methods, to confirm the impact of these mutations on biofilms, fitness, and persistent colonization.

The high frequency of mutations observed in the two outlier postpartum genomes, which have 204 and 262 mutations relative to their respective prenatal genomes, led to their classification as mutators. Support for this classification comes from the placement of each pair in the core gene phylogeny (Fig. 1), the increased mutation rates in both postpartum genomes relative to their prenatal genomes, and the identification of mutations in DNA replication and proofreading machinery genes (30, 33). Mutators represent a subpopulation of bacterial cells with a higher-than-average mutation rate (30) that typically arise in populations undergoing adaptation to stressful conditions (47, 48). They serve as a reservoir of mutations that can enhance fitness along with the genetic diversity of bacterial populations (49). A study of S. pneumoniae, for instance, described increased mutation rates after exposure to subinhibitory concentrations of penicillin (50), which facilitated adaptation. While mutators have been described in S. pneumoniae (51) and other streptococcal species (e.g., Streptococcus pyogenes [52], S. mutans [31], and S. iniae [53), this study is the first to describe mutators in GBS.

Importantly, mutators have been linked to persistent bacterial infections, such as those caused by Pseudomonas aeruginosa (54) and Staphylococcus aureus (55) in the lungs of cystic fibrosis (CF) patients. They arise at low frequencies due to DNA polymerase errors, dysfunctional proofreading mechanisms, and/or failure of mismatch/error correcting machinery, including the MMR pathway (30, 33). A prior study of pathogenic Escherichia coli and Salmonella enterica, for example, reported the incidence of mutators with defective mutS alleles to be 1% (56). MutS is a critical protein that recognizes mismatches (57). Impairment of genes in the MMR pathway can contribute to increased mutation rates, recombination (56), and horizontal gene transfer, which can subsequently enhance the spread of antibiotic resistance determinants (33). In S. aureus, mutator strains recovered from the CF lung had high frequencies of mutations in macrolide target sites (55), highlighting their role in the emergence of macrolide resistance. Besides promoting adaptation in the presence of antibiotics, many of these S. aureus mutators also had mutations in mutS. As was suggested for S. aureus (55), our discovery of mutations within or near the MMR-encoding region in both GBS mutators could cause disruptions in the functionality of this system. It is also worth noting that both mutators belong to two common lineages (CCs 19 and 23) associated with colonization during pregnancy (15). Nevertheless, we cannot rule out the possibility that the prenatal strains have other unique traits that made them more prone to mutation within DNA repair system genes. It is also possible that both strain types were present, but only one was selected for characterization from each participant per sampling. Although co-colonization with multiple strain types has been reported (34), the high degree of genetic relatedness among the strain pairs is suggestive of evolution at some point between the two samplings. Additional experiments and cohort studies are therefore needed to determine how GBS mutators emerge in different populations upon exposure to stressors such as antibiotics. GBS strain traits (e.g., serotypes, CCs) vary across geographic locations (15), with many countries lacking IAP protocols altogether (58). Hence, different selection pressures in distinct locations will impact the type and number of mutations that arise in colonizing GBS populations.

Since a high frequency of nonsynonymous SNPs was detected in the postpartum genomes, which could influence protein function, and multiple strains had mutations in the same genes, it is likely that some genes evolve in parallel under selection, leading to enhanced fitness and similar adaptations (59). We cannot conclude, however, that the mutations arose solely due to IAP exposure, as we are unable to compare the impact of IAP versus no IAP on genomic changes in this study. IAP was the standard protocol for GBS-positive participants; hence, only three participants with persistent colonization were not given IAP. Although the postpartum strains from these three participants had a lower average mutation rate (1.50 × 10⁻⁶) than the average of the 29 strains with IAP exposure (1.16 × 10⁻⁴), the difference was not significant (t-test, P = 0.74). Since all participants did not have the same antibiotic exposures, this variation could impact both the selection and mutation rate estimates as well. Not all participants were given the same antibiotic or the same number of doses, for instance, while a subset reported postpartum antibiotic use prior to the second sampling. Moreover, we were unable to account for other factors (e.g., changes in the vaginal microbiome or host susceptibility, over-the-counter medication use, complications during childbirth, diet) that may impact the selection and evolution of GBS in vivo. These limitations highlight the need for long-term studies with a larger and more diverse patient population, as well as more serially collected samples from participants with identical IAP exposures (e.g., concentration, doses, and type of antibiotic). This type of sampling scheme could have helped determine whether ancestral intermediaries exist in the evolution of strain 22.2, for example, which shares an ST with the prenatal 22.1 strain but differs in gene content. Alternatively, it could pinpoint the timing when a new strain type was acquired by the participant after IAP and childbirth.

In all, this analysis highlights the need for future studies that will examine potential mechanisms that could be linked to persistent colonization and to understand how mutators emerge in populations under selection. Even though the biofilm assays revealed functional differences in some strains collected before and after IAP and childbirth, additional experiments are needed to determine if the mutations identified in this study can impact this phenotype or others. The lack of this mechanistic confirmation prevents us from confirming a causal relationship between IAP and mutation rates. Hence, larger prospective clinical studies are critically important to determine how IAP and other selective pressures encountered by GBS during childbirth contribute to mutation rates and persistence in vivo in order to further define their role in colonization and disease. Despite these findings, it is important to note that the prophylactic use of antibiotics to protect newborns from invasive GBS infections far outweighs the risks imposed by evolutionary events that may enhance colonization.

Conclusions

The use of WGS to compare paired GBS strains from individuals before and after IAP/childbirth illustrates how stable the colonizing GBS population can be and, in some cases, identified specific mutations that may be important for adaptation to and persistence in the genitourinary tract. Acquisition of virulence and antibiotic resistance genes was not a common mechanism of adaptation, though several missense mutations were identified in genes associated with colonization and adherence. Two strains with the greatest number of mutations were classified as mutators based on above-average mutation rates and mutations in genes linked to DNA repair functions. Together, these data suggest that pressures related to antibiotic use via IAP and childbirth may select for mutations and phenotypes that promote survival in vivo. Enhanced survival and persistent maternal colonization can increase the likelihood of transmission to newborns following treatment cessation and cause complications in subsequent pregnancies. Future studies should aim to confirm the potential mechanisms identified herein and define how mutators emerge in populations under selection.

MATERIALS AND METHODS

Bacterial isolate selection and characterization

A subset of 97 GBS isolates recovered from vaginal-rectal swabs from 58 pregnant participants (23) was included in this analysis (Table S1). Isolates were recovered from the same participant during a prenatal (35–37 weeks gestation) visit and during a postpartum visit, which took place six weeks after childbirth. Isolates were previously characterized by MLST, serotyping, and cps typing (22). Participants with GBS at both samplings were considered to have persistent colonization resulting in 80 strains from 41 participants. Seventeen participants had GBS only at the prenatal sampling and were considered to have lost the pathogen by the postpartum visit.

Whole-genome sequencing (WGS)

A set of 41 of the 97 isolates was sequenced previously, and the raw reads were downloaded from the NCBI Sequence Read Archive (SRA). The remaining 56 isolates were sequenced for this study (Table S2). Following DNA extraction of GBS grown overnight in Todd-Hewitt broth (THB) at 37°C + 5% CO₂ using the E.Z.N.A. (Omega Bio-tek Inc., Norcross, GA, USA) or Wizard HMW (Promega, Madison, WI, USA) kits, libraries were constructed at the Michigan Department of Health and Human Services using the Nextera XT library prep kit (Illumina, San Diego, CA, USA). Sequencing was performed using the MiSeq (Illumina) with 2 × 250 bp paired-end reads. Raw reads were trimmed with Trimmomatic v0.39 (60) using “gentle trimming” parameters to trim adapters and remove sequences with average quality scores <15 or that were <36 nucleotides long. The paired-end reads, or single-end reads for SRR517012 that was sequenced using an unpaired method, were assessed for quality using FastQC v0.11.9 (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). High-quality genomes were defined as those that failed <10% of quality modules.

Sequences were assembled de novo using SPAdes v3.13.1 (kmers 21, 33, 55, 77, 99, 127) with mismatch error correction (61) and quality checked using QUAST v5.0.2 (62) and MultiQC v1.15 (63) with default parameters. High-quality assemblies, which were considered if the N50 and L50 scores were >15,000 and <50, respectively, were annotated using Prokka v1.14.6 with the –proteins option that allows for the use of a custom database (64). This database was curated from 25 closed S. agalactiae genomes downloaded from NCBI to ensure proper annotations (Table S8). STs were assigned using PubMLST (65), though previously determined STs (22) were used for three strains with poor assembly quality for one of the seven MLST loci.

Pangenome and phylogenetic analysis

Roary v3.11.2 was used to create a multiFASTA alignment of core genes from the 92 high-quality assemblies using the -i (blastp of 95%) and -e (PRANK aligner 170427) parameters (66). Maximum likelihood phylogenies were generated using RAxML v8.2.12 (parameters -m GTRGAMMA, 500 bootstrap replicates) (67). The core-gene alignment was input into MEGA12 (25) to construct a phylogeny using the maximum likelihood method and Tamura-Nei model (24) of nucleotide substitutions, which used up to four parallel computing threads. The tree with the highest log likelihood was used, and the percentage of replicate trees was determined adaptively. The initial tree was selected by choosing the tree with the superior log-likelihood between a neighbor-joining (NJ) tree (28), which uses a matrix of pairwise distances, and a maximum parsimony (MP) tree with the shortest length among 10 MP tree searches. Phylogenies were visualized and annotated using the Interactive Tree of Life v7 (IToL, https://itol.embl.de/) (29). Neighbor-net trees were generated in SplitsTree v4.19.0 (26), and the PHI test was used to detect recombination with a window size of 100 and a k of 1; a P-value of <0.05 was considered significant. Analysis of the six MMR loci was performed by manually extracting them from each genome assembly using Geneious v2022.2.2 (https://www.geneious.com/) followed by a MUSCLE (32) nucleotide alignment.

Gene and nucleotide-level analyses

ARGs and virulence genes were extracted from the 92 assemblies with ABRicate (https://github.com/tseemann/abricate) using five ARG databases (CARD [68], ARG-annot [69], MEGARes [70], Resfinder [71], NCBI AMRFinderPlus [72]) and the virulence-finder-database (VFDB) (73), respectively. A gene was considered present if it had ≥10% identity and ≥80% coverage and for the ARGs, if it was detected in two or more databases.

Core genes were extracted from the assemblies of each pair of strains recovered from participants with persistent colonization (n = 64, 32 pairs). For each pair, the assembled prenatal genome was used as the reference, while the trimmed reads from the postpartum isolate were mapped against the reference. Each core genome was interrogated for mutations including SNPs, insertions, deletions, multiple nucleotide polymorphisms (MNPs), and complex mutations using Snippy v4.6.0 (https://github.com/tseemann/snippy). For discrepancies in gene presence/absence within a pair of strains, Geneious v2022.2.2 was used to confirm ≥15× read coverage and the upstream and downstream regions were mapped with respect to a given gene to confirm location and completeness of the gene within the contig. Core gene mutations were validated using a ≥10× read coverage cutoff.

Biofilm assays

Biofilm assays were performed three times in triplicate. Values were normalized to blank media controls as we described (41) previously with the modification of stagnant growth for 24 h. The magnitude (fold change) of the difference in biofilm formation was determined for each postpartum strain relative to each respective prenatal strain.

Data analysis and statistics

Raw WGS output was managed in Microsoft Excel. R v4.1.2 (https://www.r-project.org/), and RStudio v2022.07.1+554 (https://www.rstudio.com/) was used for importing data (readr v2.1.4), data wrangling (devtools v2.4.5, dplyr v1.1.3, forcats v1.0.0, plyr v1.8.8, tidyr v1.3.0, tidyverse v2.0.0), and visualization (ggplot2 v3.4.3, viridis v0.6.4) (https://cran.r-project.org/). R package factoextra v1.0.7 was used to perform and visualize PCA with 95% confidence ellipses for virulence genes using the prcomp() and fviz_pca() functions (https://cran.r-project.org/). Statistically significant associations were classified at P ≤ 0.05. For all SNP-only analyses, SNPs were filtered to exclude those that occurred in uncharacterized and/or non-coding regions. As characterized by Snippy, nonsynonymous SNPs included those classified as “initiator codon variant,” “missense,” “start lost,” “stop lost,” and “stop gained” (nonsense), while synonymous SNPs included those referred to as “stop retained” and “synonymous variant.” Chi-square and Fisher’s exact tests were performed in Epi Info v7 (Centers for Disease Control and Prevention, Atlanta, GA, USA) to compare categorical variables across groups, while Student t-tests were used to analyze differences in continuous variables including the mean biofilm levels between isolate pairs and mutation rates. A P-value of <0.05 was considered statistically significant.

DATA AVAILABILITY

All data sets are available in the supplemental tables, and raw sequencing reads for each genome have been deposited in GenBank under several BioProjects listed in Table S2 unless they were uploaded previously. R scripts are at https://github.com/macyEpell and via Zenodo (https://doi.org/10.5281/zenodo.18408634).

ETHICS APPROVAL

The parent study was conducted in November 1998-April 2000 and involved obtaining informed consent from each patient prior to enrollment. It was approved by the Conjoint Medical Ethics Committee of the University of Calgary and the Ethics Board of the participating University of Alberta Hospitals. All isolates and patient data were de-identified and given unique identifying numbers prior to arriving at our laboratory for characterization. The retrospective genomic characterization of these isolates is not considered human subjects research according to exemption 4(ii), as no identifying information is available to establish a link with any of the patients in the original parent study, and no attempts to contact the original patients are possible. All isolates have been double-blinded with unique identifiers assigned at the time of collection and upon receipt in our laboratory.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/mbio.03966-25.

Supplemental Figures. mbio.03966-25-s0001.pdf.

Figures S1 through S5.

Supplemental Tables. mbio.03966-25-s0002.xlsx.

Tables S1 through S8.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.