Streptococcus agalactiae, or Group B Streptococcus (GBS), is a major pathogen in humans and is the leading cause of neonatal infections in industrialized countries (Stoll et al., 2011; Edmond et al., 2012). Early-onset disease (EOD) (days 0–6) is the result of vertical transmission from a colonized mother during, or just before, delivery (Colbourn and Gilbert, 2007), whereas the pathogenesis of late-onset disease (LOD) (days 7–89) is less well understood (Berardi et al., 2013). GBS belongs to the commensal microbiota that colonize the gastrointestinal and genital tracts of 10–30% of healthy humans and can cause serious infections in neonates (Regan et al., 1991; Yancey et al., 1994). It is now universal practice to screen pregnant women for vaginal and/or rectal colonization with GBS to prevent early-onset infection (ANAES, 2001; Verani et al., 2010; Di Renzo et al., 2015). The aim of this screening strategy is to limit bacterial transmission and prevent EOD by the administration of intrapartum antibiotic prophylaxis. The GBS colonization status can change over the course of pregnancy. Thus, colonization status late in the third trimester has been used as a proxy for intrapartum colonization (at 35–37 weeks of gestation) (Anthony et al., 1978; Lewin and Amstey, 1981; Dillon et al., 1982; Hansen et al., 2004; Verani et al., 2010). In France, GBS screening is performed by swabbing the lower vagina followed by cultivation. This screening strategy must be repeated for each pregnancy, with the exception of a few special cases (women with GBS isolated from the urine during the current pregnancy or who had a previous infant with invasive GBS disease) (ANAES, 2001; Verani et al., 2010; Di Renzo et al., 2015). Moreover, screening for pathogenic bacteria, including GBS, must be performed in cases of imminent delivery risk (suspicion of chorioamnionitis, prolonged membrane rupture, or preterm labor) (ANAES, 2001).

Several methods have been developed to improve the diagnostic and prognostic classification of GBS isolates, such as serotyping, based on capsule polysaccharides, or, more recently, multilocus sequence typing (MLST), based on the sequencing of seven genes, which is expensive and time consuming (Jones et al., 2003; Lindahl et al., 2005; Slotved et al., 2007). These methods have highlighted the involvement of serotype III and sequence-type ST17 in causing more invasive neonate disease (Manning et al., 2009). Other molecular typing methods have been recently developed, such as multilocus variable number of tandem repeat (VNTR) analysis (MLVA), a molecular typing method based on VNTR variability (Haguenoer et al., 2011). This typing method targets six dispersed chromosomal loci. For GBS, it was shown to be more discriminant than MLST, especially for ST17 isolates (Haguenoer et al., 2011). CRISPR (clustered regularly interspaced short palindromic repeats) loci have also been described in GBS and used to genotypically characterize GBS isolates (Lopez-Sanchez et al., 2012; Lier et al., 2015).

Clustered regularly interspaced short palindromic repeats (CRISPR) and CRISPR-associated (Cas) proteins form the CRISPR-Cas system, which protects bacteria against bacteriophages (phages), and more generally, against exogenic mobile genetic elements (MGEs) (Barrangou et al., 2007). This system provides adaptive immunity through specific immunization based on nucleic acid sequences. CRISPR arrays are composed of short (25–40 bp) direct repeats (DR) interspaced by non-repetitive similar-sized sequences called spacers. DRs are highly conserved within a given CRISPR array and are often partially palindromic, having the potential to form hairpin structures (Kunin et al., 2007). Most CRISPR arrays are flanked on one side by an AT-rich sequence, called the leader, and on the other side by a trailer-end sequence, which is located downstream of the terminal direct repeat (TDR), a DR which is often degenerate and/or truncated. The CRISPR-Cas system acts in three stages (Barrangou and Marraffini, 2014). In the adaptation stage, a new spacer, usually derived from a foreign nucleic acid, is incorporated at the leader end of the CRISPR locus, concomitantly to the duplication of the leader-end DR. The expression stage consists on CRISPR RNA (crRNA) biogenesis, starting with transcription of the CRISPR array from a promoter located within the leader, followed by cleavage within the DRs of this long pre-crRNA molecule into short, mature crRNAs. Finally, during the interference stage, proteins target complementary foreign nucleic acids for cleavage within the sequence corresponding to the spacers. Thus, the CRISPR-Cas system provides sequence-specific immunity, whereby specificity is dictated by the sequence of the spacers that have been integrated within the CRISPR array. CRISPR loci thus show a dynamic, rapidly evolving, and polymorphic composition due to the ability of the system to capture novel spacers derived mostly from MGEs, reflecting past encounters with MGEs present in the environment. Spacer acquisition is polarized, as novel spacers are inserted at the leader end of the CRISPR locus. Consequently, sequences present at the trailer end can be considered to be ancestral, providing evidence of ancient MGE invasion events (Barrangou et al., 2007). Overall, spacer polymorphism and polarized acquisition of novel spacers result in the temporal organization of CRISPR loci, making them attractive epidemiological markers for genotyping and phylogenetic analyses (Shariat and Dudley, 2014).

Two CRISPR-Cas systems have been characterized in a collection of GBS strains (Lopez-Sanchez et al., 2012). A type II-A system, associated with the CRISPR1 locus, is ubiquitous and functional, whereas a type I-C system, associated with the CRISPR2 locus, is rare and most often incomplete, suggesting little or no activity (Lopez-Sanchez et al., 2012; Makarova et al., 2015). The CRISPR1 locus contains highly conserved 36-bp DRs separated by spacers of approximately 30 bp. Similarities between CRISPR1 spacer sequences and MGEs have been previously reported. Comparative sequence analysis of CRISPR1 loci across numerous strains revealed extensive diversity, due to the acquisition of new spacers, spacer duplications, and spacer deletions, illustrating the dynamics of the system (Lopez-Sanchez et al., 2012; Lier et al., 2015).

Previous studies reported the use of CRISPR1 spacer content and TDRs for the genotyping of GBS strains, showing a strong correlation between MLST and CRISPR-based genotyping (Lopez-Sanchez et al., 2012; Lier et al., 2015). Here, we used this genotyping method to follow vaginal carriage of GBS from 100 women with a minimum 3-month time interval between samples, and show that CRISPR1 polymorphism-based genotyping represents a rapid and discriminating tool for epidemiological studies. Furthermore, as spacer acquisition events occurred in vivo in multiple and independent cases during carriage, notably in response to a probable phage infection, our results strongly suggest that the CRISPR1-Cas system is functionally active for adaptation.

In this work, we explored the in vivo dynamics of human colonization by GBS over several years by comparing the CRISPR1 loci of 205 isolates collected from 100 women over an 11-year period. The aims of this study were to evaluate CRISPR1 locus analysis as an epidemiological marker, to describe the evolution of CRISPR1 arrays in GBS over time during human carriage, and to assess the likelihood of phage infection in vivo, highlighted by alterations of the spacer content, notably additions at the leader end of CRISPR1.

Eighteen women (18/100) provided GBS isolates with a “close” CRISPR1 locus over the sampling time. MLST typing confirmed the genetic relationship between these isolates, as well as MLVA analysis, except for one case which differed only at the SAG21 VNTR locus (woman 75), as discussed below. These isolate pairs arguably share common ancestry, and the observed CRISPR locus modifications are the result of in vivo evolution within patients. Modifications at the leader end (four women) correspond to spacer acquisition which likely occurred in response to MGE invasion events. Two isolates show similarities between novel spacers and plasmids from related species (E. faecalis and S. pseudopneumoniae), consistent with previous studies showing similarity between most GBS spacers and streptococcal genome sequences (Lopez-Sanchez et al., 2012). One isolate acquired a novel spacer showing strong similarity with a gene encoding a phage tail protein, probably originating from a Caudovirales/Siphoviridae phage, as are most Streptococcaceae phages (Domelier et al., 2009). For one woman, the two isolates differed from each other by the polar acquisition of three spacers that show no similarity to any known sequence. These spacers could have been acquired successively, or all at the same time, during the 2.5 years separating the two samplings. Overall, polarized spacer acquisition at the leader end was relatively rare in our population. Indeed, this event occurred in only 4% of the colonized women of the study. The mean time between sampling of GBS isolates showing polarized acquisition was 3.1 years (compared to 2.3 years for all isolates). Metagenomic analyses of the population dynamics of MGE has shown a rapid ability of CRISPR-Cas systems to acquire spacers in both environmental and human niches (Fineran and Charpentier, 2012). In the S. thermophilus type II-A CRISPR system, just one in vitro challenge with a phage was sufficient to result in the acquisition of new spacers (Barrangou et al., 2013). Similarly, in vitro biofilm analysis showed a high rate of evolution for Leptospirillum type II CRISPR arrays, whereby the number of spacers increased from 37 to 197 in 5 months (Tyson and Banfield, 2007; Andersson and Banfield, 2008). Pride et al. (2011) analyzed Streptococcus spp. (S. thermophilus, S. mutans, S. pyogenes, and S. agalactiae) CRISPR loci directly from the saliva of four patients over 18 months by evaluating spacer number by PCR amplification with primers directed at common DRs. These authors showed for each patient a high rate of alteration of spacer number at one CRISPR locus, over time, demonstrating the dynamics of the CRISPR1 system in human. In our study, the rare cases of spacer acquisition might be explained by a low occurrence of MGEs in this environment, or by a low activity of the CRISPR1-Cas system in GBS. Furthermore, as no lytic phage has been isolated yet from GBS, the acquisition of new spacers showing similarity to phage sequences may correspond to adaptation from excised prophages.

In our study, we observed two types of modifications of the internal CRISPR1 array: spacer acquisitions, and deletions. Deletions always occurred in an internal region of the locus or close to the last spacer, as previously described (Deveau et al., 2008; Horvath et al., 2008; Lopez-Sanchez et al., 2012). These deletions never concerned spacers in contact with the leader or TDR. Deletions and spacer duplications may result from slipped misalignment during DNA replication (Lopez-Sanchez et al., 2012). For two isolates (women 20 and 79), the acquired spacers were not already present in the locus, and acquisition could be due to homologous recombination, which has been described in GBS (Bisharat et al., 2004; Brochet et al., 2008; Lopez-Sanchez et al., 2012). In two cases (women 14 and 42), internal spacer acquisition -including a spacer duplication case- was seemingly combined with the deletion of an adjacent spacer. A similar phenomenon probably also occurred for the intermediary isolate of woman 67, from whom three isolates were studied; the first and third isolates were identical, while the intermediate isolate differed by a duplication and deletion of one spacer.

We observed a correlation between GBS phylogenetic lineages and the ability to integrate new spacers in CRISPR1, suggesting an impact of the genetic background on adaptation efficiency. Isolates with polarized acquisition belong only to ST1, ST19, or ST6, whereas internal locus modifications were observed for isolates belonging to CC1, CC19, CC23, CC10, and ST26 (Table 2). We observed no CRISPR1 modification for CC17 isolates, neither in at the leader end nor in the middle of the locus, despite their relatively large representation (26 isolates from 18 women). Moreover, CC17 isolates showed less spacer variation at CRISPR1 locus, suggesting lower activity of CC17 CRISPR system. This result is consistent with previous observations which also highlight a significantly lower number of spacers in CC17 isolates in comparison with isolates belonging to other lineages (Lier et al., 2015). Nevertheless, Domelier et al. (2009) demonstrated the unique character of CC17 phages, in addition to a considerable prophage diversity in ST17 isolates. This is consistent with previous studies which compared inter-species CRISPR loci, reporting a strong inverse correlation between the number of spacers and prophages within a given genome (Nozawa et al., 2011). Similarly, CC23 isolates displayed less spacer variation, as previously described, and shared a relatively high number of common spacers (Lier et al., 2015). We found that internal spacer duplication was more frequent among CC23 isolates. Notably, spacer 247 was duplicated in 54% of CC23 isolates and was highly represented (82% of CC23 isolates). This spacer shares similarity with S. dysgalactiae phage phi 3396, which likely originated from a S. pyogenes phage (Davies et al., 2007).

Furthermore, we noticed an association between phylogenetic affiliation and leader sequence. The leader sequence is highly conserved in GBS, as well as in other bacteria (Yosef et al., 2012; Lier et al., 2015; Wei et al., 2015). Notably, the sequence at the leader-repeat junction is critical for adaptation by CRISPR-Cas systems, as shown in Escherichia coli type I-E (Yosef et al., 2012) and S. thermophilus type II-A (Wei et al., 2015). Among our isolates, the leader-repeat junction was modified in 8% of them (17/205), and isolates with leader-repeat junction modifications mostly belonged to CC10 (82%, 14/17). This modification was frequent and occurred in 67% of the CC10 isolates (14/21), in agreement with prior observations (Lier et al., 2015).

Our study showed a strong congruence between the various typing methods used, namely CRISPR typing, MLST, and MLVA. We explored other regions of the GBS genome by MLVA and MLST (exploring six and seven loci distributed throughout the chromosome, respectively) to confirm the relationship between the isolates. For one woman (woman 70), the two isolates carried different CRISPR1 loci but displayed the same MLST and MLVA profiles, suggesting a genetic link. Their CRISPR1 arrays contained five identical ancestral spacers and the same TDR, confirming common ancestry. However, spacers located near the leader showed high variability, likely due to an independent recent evolution. For these isolates, whole genomic analysis could be applied to evaluate their relationship.

For five patients, isolate pairs having identical or close CRISPR1 loci had the same ST but different MLVA profiles. Among them, the second isolate of one woman (woman 75) acquired three spacers at the leader end of CRISPR1. The only difference between MLVA profiles were found within the SAG21 VNTR and may be due to repeat sequence acquisition or deletion over time. Of note, SAG21 is the most variable VNTR among those used for MLVA (Haguenoer et al., 2011) and is perhaps too variable for MLVA typing for middle- to long-term carriage of GBS, and typing over a long period. The rate of evolution of this locus has not been explored yet.

These results need to be interpreted with caution. Although spacer content of the CRISPR1 locus is generally specific for isolates for a given woman, some isolates from different women (isolates from women 5 and 6, and those from women 68 and 18) and belonging to CC23 had identical CRISPR1 loci, as observed previously (Lier et al., 2015). Similarly, some isolates belonging to ST17 had close CRISPR1 loci, with a change in just the most recent spacer (isolates from women 99, 26, and 87). Identical CRISPR1 loci among strains of different origin and belonging to ST17 were observed previously by Lopez-Sanchez et al. (2012). This may be due to the circulation of highly prevalent clones of CC23 or CC17, even if there is no evidence of an epidemiological link between the different isolates, except pairwise.

Another aim of this study was to evaluate the evolution of GBS during carriage using these three typing methods. Among the 100 women studied, 78% presented carriage of GBS strains with the same profile, irrespective of the molecular method used (CRISPR1 typing, MLVA, or MLST), and 83% presented carriage of the same GBS strain with the same or close CRISPR1 array. Similar GBS carriage has been previously described (Hansen et al., 2004; Manning et al., 2008). Hansen et al. (2004) followed carriage for 21 months (during pregnancy and up to 1 year after delivery) by pulsed-field gel electrophoresis (PFGE). In their study of 35 carriers, 86% of women were colonized by just one clone, and the others by two clones simultaneously, and no clonal shift was observed (Hansen et al., 2004). Manning et al. (2008) explored perinatal GBS colonization (during pregnancy, and 6 weeks after delivery) by serotyping, GBS capsular gene cluster genotyping, and MLST. Comparison of the three typing methods revealed differences, with clonal replacement in 18% of women, and a change of ST for 10% (Manning et al., 2008). However, these longitudinal studies were performed during a short period (maximum 21 months between two samples). Our study, based on a longer period, extends previous findings on the predominance of one clone during carriage, using highly discriminant typing methods. One limitation of our study is that the stored isolates may represent the predominant strain in cases where two or more coexisted in the sample. This approach does not provide a complete overview of bacterial diversity over time and we cannot exclude the possibility of multiple populations, particularly if a strain was predominant. Indeed, carriage of a double population has been shown previously (Hansen et al., 2004; Perez-Ruiz et al., 2004; Brzychczy-Włoch et al., 2014). This double population may explain our observations for woman 67, where the second isolate carried a different CRISPR1 locus than that of the first and third isolates. The discriminatory power provided by CRISPR1 locus analysis could make this a simple and relatively rapid typing method to explore GBS vaginal carriage.

Clustered regularly interspaced short palindromic repeats-based typing provides a glimpse of GBS vaginal carriage at the molecular level. This method has the advantage of being linked to the genetic lineage of the isolate through analysis of the TDR and ancestral spacers. Furthermore, it allows exploration of the recent evolution of the isolate, especially encounters with MGEs, through analysis of the leader end. The CRISPR-based typing method appears to be a useful tool to compare GBS isolates, thanks to its high discriminatory power and ease of use. However, further studies are required before CRISPR can be used more generally as a typing method for GBS. In addition, it is now evident that the GBS CRISPR1 locus evolves in vivo. We observed polarized acquisitions in response to MGEs (plasmids or phages), highlighting the dynamics of the system. The ability of the CRISPR1 array to evolve may be linked to phylogenetic affiliation, exemplified by the apparently lower ability of the hypervirulent strain ST17 to acquire novel spacers, compared to the higher ability of ST6 isolates. This potential link is yet to be explored.

Conceived and designed the work: CB and PL. Performed manipulations: CB and AP. Drafted the paper: CB, PH, and PL. Critically revised the manuscript: PH, FP, and LM. All authors read and approved the final version of the manuscript.

PH is an employee of DuPont, and co-inventor on several patent applications related to CRISPR-Cas systems. The other authors declare that the research was conducted in the absence of any commercial or financial relationship that could be construed as a potential conflict of interest.