Vaginal dysbiosis increases risk of preterm fetal membrane rupture, neonatal sepsis and is exacerbated by erythromycin

Background Preterm prelabour rupture of the fetal membranes (PPROM) precedes 30% of preterm births and is a risk factor for early onset neonatal sepsis. As PPROM is strongly associated with ascending vaginal infection, prophylactic antibiotics are widely used. The evolution of vaginal microbiota compositions associated with PPROM and the impact of antibiotics on bacterial compositions are unknown. Methods We prospectively assessed vaginal microbiota prior to and following PPROM using MiSeq-based sequencing of 16S rRNA gene amplicons and examined the impact of erythromycin prophylaxis on bacterial load and community structures. Results In contrast to pregnancies delivering at term, vaginal dysbiosis characterised by Lactobacillus spp. depletion was present prior to the rupture of fetal membranes in approximately a third of cases (0% vs. 27%, P = 0.026) and persisted following membrane rupture (31%, P = 0.005). Vaginal dysbiosis was exacerbated by erythromycin treatment (47%, P = 0.00009) particularly in women initially colonised by Lactobacillus spp. Lactobacillus depletion and increased relative abundance of Sneathia spp. were associated with subsequent funisitis and early onset neonatal sepsis. Conclusions Our data show that vaginal microbiota composition is a risk factor for subsequent PPROM and is associated with adverse short-term maternal and neonatal outcomes. This highlights vaginal microbiota as a potentially modifiable antenatal risk factor for PPROM and suggests that routine use of erythromycin for PPROM be re-examined. Electronic supplementary material The online version of this article 10.1186/s12916-017-0999-x) contains supplementary material, which is available to authorized users.


Background
Preterm birth and its associated complications are the leading cause of death for children under the age of five worldwide [1]. Survivors often suffer significant motor and sensory deficits, learning disabilities and respiratory disorders [2]. Rupture of the fetal membranes prior to 37 weeks of gestation and before the onset of labour, termed preterm prelabour rupture of the membranes (PPROM), occurs prior to 30% of all spontaneous preterm births [3]. Both pathogenesis of membrane rupture and subsequent maternal and neonatal morbidities are strongly associated with infection [4,5]. It is hypothesised that colonisation of the vagina with pathogenic bacteria activate the local and upper (cervical and fetal membrane) innate immune system [3,6], driving an inflammatory cascade [7][8][9][10] that leads to remodelling and disruption of membrane architecture and premature rupture [11]. In 80% of cases, delivery occurs within 9 days of rupture [12], during which time the uterine cavity, placenta and fetus are exposed to ascending infection and increased risk of chorioamnionitis and funisitis, which are associated with poor maternal and neonatal outcomes [13][14][15][16][17][18][19].
While the vaginal microbiota composition of nonpregnant women is temporally dynamic [20], healthy pregnancy is characterised by a shift towards stable, low-richness and low-diversity community structures dominated by Lactobacillus spp. [21][22][23] that inhibit growth of pathogenic bacteria [24]. However, recent studies have found that the dominance of vaginal bacterial communities by L. iners is a risk factor for preterm birth [25,26]. The absence of Lactobacillus spp. and polymicrobial colonisation of the vagina have long been recognised as a risk factor for PPROM [27], preterm birth [28][29][30][31] and histological chorioamnionitis [32][33][34]. Despite a well-described infectious aetiology and high prevalence of chorioamnionitis in PPROM patients, the few studies to have examined vaginal bacterial composition in women with PPROM are limited to small sample sizes collected only after membrane rupture [35][36][37].
The clinical management of PPROM is challenging, involving an assessment of the balance between prolongation of pregnancy to enable fetal maturation, and risk of infection and subsequent poor neonatal outcomes. As a result, management during this latency period is controversial and varies widely [38]. In many countries, PPROM is managed conservatively with patients receiving steroids to promote fetal lung maturation and prophylactic administration of oral erythromycin at a dose of 250 mg for 10 days [39,40]. In the presence of clinical evidence of chorioamnionitis, patients are treated with intravenous antibiotics and delivery is expedited [40]. The widespread use of erythromycin for PPROM is based upon the short-term neonatal benefits reported in the ORACLE I trial, which compared erythromycin to co-amoxiclav (amoxicillin and clavulanate potassium) or placebo [41]. The continued use of erythromycin for PPROM is contentious considering the lack of identifiable long-term neonatal benefits [42,43], limited coverage of gram negative bacteria [44], Mycoplasma spp. or Ureaplasma spp. [45], rising resistance [46] and recent association with increased risk of cerebral palsy, epilepsy [42,47] as well as asthma and obesity [48].
Developing a comprehensive understanding of the vaginal microbiota composition and its response to antibiotic treatment in pregnancies complicated by PPROM is, therefore, of paramount importance for improved diagnostic, preventative and therapeutic strategies. In this study, we examined vaginal microbiota compositions prior to and following PPROM, both before and following erythromycin prophylaxis, and correlated these findings with evidence of funisitis and neonatal sepsis.

Methods
The study was approved by the National Research Ethics Service Committee London-Stanmore of the National Health Service (REC 14/LO/0328), and all participants provided written informed consent.

Study design
We performed a prospective cohort study whereby women with and without risk factors for preterm birth were recruited between 8 and 12 weeks from the preterm surveillance and antenatal clinics of Queen Charlotte's and Chelsea Hospital and Chelsea and Westminster Hospital between January 2013 and August 2014 (n = 250). Exclusion criteria included women under 18 years of age, multiple pregnancies, sexual intercourse or vaginal bleeding within 72 hours of sampling, and HIV or hepatitis C positive status. Under direct visualisation, cervico-vaginal fluid was sampled from the posterior fornix using a BBL CultureSwab MaxV Liquid Amies swab (Becton, Dickinson and Company, Oxford, UK) at each of the following timepoints: 8-12, 19-25, 27-30 and 32-36 weeks of pregnancy gestation. The vaginal swabs were placed immediately on ice before being snap frozen and stored at -80°C within 5 min of collection.
A second cohort were recruited upon presentation with ruptured membranes between October 2013 and June 2015 (n = 87). As per participating hospitals guidelines, PPROM was defined as a rupture of the fetal membranes, diagnosed by pooling of amniotic fluid on speculum examination, prior to 37 weeks gestation more than 24 hours prior to spontaneous preterm delivery or clinically indicated delivery or induction of labour. Swabs were taken upon presentation before erythromycin treatment, 48 hours after erythromycin treatment, and 1 and 2 weeks post-diagnosis. Patients referred from other hospital sites for specialist neonatal care with a prior diagnosis of PPROM who had already started erythromycin were sampled upon arrival and 1 week later, if undelivered.
All patients were treated conservatively as per the United Kingdom Royal College of Obstetrics and Gynaecology guidelines [40] and the policy of the admitting hospital, which involved receiving antenatal steroids for fetal lung maturation (if less than 34 weeks gestation) and oral erythromycin 250 mg four times a day for 10 days. Delivery was expedited or induced in the presence of fetal distress, clinical signs of chorioamnionitis or at 34 completed weeks of gestation. All patients received intrapartum antibiotics in the form of intravenous benzylpenicillin and the eventual mode of delivery was at the discretion of the attending clinician.
Histological examination of the placenta and fetal membranes was performed following PPROM as per routine practice in the study. Chorioamnionitis was defined as the presence of polymorphonuclear cells within the amnion or chorion whilst funisitis was defined by the presence of polymorphonuclear cells (of fetal origin) within the Wharton's jelly of the umbilical cord.
Early onset neonatal sepsis (EONS) is defined as the presence of confirmed or suspected sepsis at ≤3 days after birth for which neonatal antibiotic treatment was prolonged beyond the routine 48 hours of prophylaxis. Confirmed sepsis was established by positive blood cultures whilst suspected sepsis was diagnosed in the presence of clinical suspicion of sepsis (lethargy, apnoea, respiratory distress, hypoperfusion and shock) supported by elevated neonatal C-reactive protein (CRP) (>10 mg/ dl) or blood film suggestive of bacteraemia. Detailed maternal and neonatal metadata were collected for all participants from the hospital case notes and the electronic patient databases Cerner Millennium®, Powerchart® and Badger.net.
Resulting sequence data were analysed using the MiSeq standard operating procedure pipeline of the Mothur package [50]. Sequence alignment was performed using the Silva bacterial database (www.arb-silva.de/), and classification was performed using the Ribosomal Database Project (RDP) database reference sequence files and the Wang method [51]. The RDP MultiClassifier script was used for determination of operational taxonomic unit taxonomies (phylum to genus) and species-level taxonomies were determined using USEARCH [52]. To avoid sequencing bias, data were resampled and normalised to the lowest read count (n = 6940).
Public access to sequence data and accompanying metadata can be obtained from the Sequence Read Archive of the European Nucleotide Archive (PRJEB21325).

Quantitative bacteriology
The total number of 16S rRNA gene copies per swab was measured as a representation of the total bacterial load. A bacterial DNA template was used for broad coverage quantitative real-time PCR using the Bact-QUANT method [53]. For this, a tenfold standard curve (30 to 3,000,000 copies) of Escherichia coli 16S DNA (Sigma, D4889) was generated and each standard was combined with 5 μl of sample DNA templates and platinum PCR-supermix UDG containing 50 nM Rox (Life Tech, cat. no. 11730-017), BactQUANT forward primer sequence (5' CCT ACG GGA GGC AGC A), BactQuant reverse primer sequence (5' GGA CTA CCG GGT ATC TAA TC) and BactQUANT probe (Life Tech, cat. no. 4316034, 6000pmol scale) sequence 5' 6FAM-CAG CAG CCG CGG TA-MGBNFQ. Template-free PCR controls and sham digest controls were included in each run. Bacterial load was displayed as copy number per swab corrected for variation of 16S rRNA gene copy proportional to bacterial species abundance in each swab. For this, 16S rRNA gene copy number of bacterial species comprising >95% of sequence reads for each swab was identified using the Michigan rrn database (https:// rrndb.umms.med.umich.edu/) and weighted for relative abundance. Bacterial load values were then normalised to the weighted copy number. Where operon copy number was not available at the species level, average copy number at the genera level was used.

Statistical analysis
Examination of statistical differences between vaginal microbiota was performed at genera and species taxonomic levels using the Statistical Analysis of Metagenomic Profiles software package (STAMP) [54]. Samples were classified into eight vaginal microbiota groups (VMGs) according to Ward's linkage hierarchical clustering analysis of bacterial species using a clustering density threshold of 0.75 with the 50 most abundant species displayed. Clusters were then sub-grouped based on Lactobacillus abundance into Lactobacillus dominant (>80%), intermediate abundance (33-78%) and depleted/dysbiotic (<10%).
The significance of differences between richness and diversity measures, bacterial load and relative abundance of species and genera data between patient groups was assessed using one-way ANOVA with Dunn's multiple comparisons and the Mann-Whitney t-test where appropriate.
The linear discriminant analysis with effect size (LEfSe) method [55] was used to identify differentially abundant taxonomic features between patient groups of interest. An α value of 0.05 was used for the factorial Kruskal-Wallis test between classes and a minimum threshold of 2.0 was used for the logarithmic latent discriminatory analysis (LDA) score for discriminative features for all LEfSe plots.
Transition of the VMGs following administration of erythromycin was visualised using a Sankey plot created in the JavaScript Sankey Diagram package from Google Charts (https://developers.google.com/chart/interactive/docs/gallery/sankey). A statistical comparison of data from paired samples obtained before and after erythromycin was administered was performed using a Wilcoxon signed rank test.
To assess the statistical significance of dysbiosis and microbiome groups, we performed linear regression analysis in the R programming environment. Specifically, we used the function lmer() (R package lme4 version 1.1-7, http://CRAN.R-project.org/package=lme4) where paired samples were present and lm() where no paired samples were present. For each analysis, a false discovery rate adjustment (Benjamin and Hochberg) was applied to correct P values. In total, four analyses were carried out as follows: (i) Analysis of differences in microbiome composition between patient groups and dysbiosis/microbiome groups. An indicator variable is created, where the indicator is 1 for samples that could be assigned to the given dysbiosis or microbiome group and the indicator is 0 for all other samples. This indicator variable is regressed against pairs of patient groups adjusted for maternal age, ethnicity, body mass index, smoking status, cervical stitch and progesterone treatment. Where paired samples were available, the model also included the patient ID modelled as a random effect. (ii)Analysis of differences in microbiome composition between PPROM patients (n = 16) before erythromycin was given and 48 hours thereafter. As above, an indicator variable is created and regressed against the time point. As all samples are paired, no additional predictors are included in the model. (iii)Analysis of differences in microbiome composition between healthy patients and patients with chorioamnionitis or funisitis. As above an indicator variable is created and regressed against the chorioamnionitis/funisitis status adjusted for maternal age, ethnicity, body mass index, smoking status, cervical stitch, progesterone usage and latency. (iv) Analysis of differences in microbiome composition between PPROM patients with and without EONS. As above, an indicator variable is created and regressed against the EONS status adjusted for maternal age, ethnicity, body mass index, smoking status, cervical stitch and progesterone usage.

Results
Baseline clinical characteristics of the study cohort A total of 250 pregnant women were prospectively recruited while attending a prematurity surveillance clinic after a history of preterm birth or mid-trimester loss and followed up at 8-12, 19-25, 27-30 and 32-36 weeks of gestation. The majority of these patients delivered at term (n = 212, 85%). Of those who delivered preterm after <37 weeks of gestation (38/250, 15%), 15 experienced PPROM and were sampled at a mean gestational age of 30 +1 weeks. Samples collected at 27-30 weeks (mean 29 weeks) from 20 patients who delivered at term following uncomplicated pregnancy were used as gestational age matched controls. An additional 87 women presenting acutely with PPROM at a mean gestational age of 28+1 weeks were recruited; 39 were sampled before erythromycin prophylaxis, 48 after erythromycin only. Of these women, 16 were sampled both before and after erythromycin treatment.

Vaginal microbiota composition prior to and following PPROM
A total of 4,542,391 high quality reads were generated in the study, with a mean read count per sample of 27,530 (range 6940 to 163,610). After removing singletons and rare operational taxonomic units (<10 reads per sample), a total of 37, 84 and 332 taxa were identified in vaginal swabs collected from controls, women who subsequently developed PPROM and women sampled at time of PPROM presentation, respectively.
Vaginal microbial communities isolated from control women sampled at 28 weeks of gestation were characterised by low richness and diversity and dominance by Lactobacillus spp. (Fig. 1a-d). In contrast, women sampled antenatally who subsequently experienced PPROM had a higher proportion of microbiota profiles characterised by intermediate or low Lactobacillus spp. dominance and high diversity (7/15, P = 0.011, Fisher's exact). Significant differences in the proportion of Lactobacillus spp. dominant, intermediate and depleted communities remained when analyses were adjusted for potential confounders, including maternal age, ethnicity, body mass index, smoking status and treatment interventions (cervical stitch or progesterone therapy) (Additional file 1: Table S2).
To identify vaginal bacteria associated with PPROM risk, we used LEfSe on 16S rRNA data collected from control patients and women sampled antenatally prior to PPROM. Samples from the latter cohort were enriched in bacteria classes Bacteroidales, Fusobacteriales and Clostridiales, whereas increased Lactobacillales was predictive of normalterm delivery (Fig. 2a-d).

Impact of membrane rupture on the vaginal microbiome
Membrane rupture prior to erythromycin prophylaxis was associated with a reduction in bacterial load compared to both control (P = 9.37 × 10 -8 ) and PPROM cases (P = 0.0011) sampled prior to membrane rupture (Fig. 1e). Although a similar proportion of Lactobacillus spp. depleted, high-diversity communities were observed post-rupture (Fig. 1b), membrane rupture was associated with a significant increase in richness (P = 0.000001, Fig. 1c) and alpha diversity (P = 0.00194, Fig. 1d). LEfSe analysis identified 20 genera positively associated with membrane rupture (Additional file 1: Figure S2) including Prevotella, Staphylococcus, Aerococcus and Streptococcus spp., and a negative association with Lactobacillus spp.  Table S2) that persisted for up to 1 week of treatment. Treatment beyond 1 week was associated with a reduction of intermediate communities and an increase in Lactobacillus spp. dominance. However, the proportion of dysbiotic vaginal microbiota remained constant. Bacterial richness (Fig. 1c), diversity (Fig. 1d) and load (Fig. 1e) remained unchanged following erythromycin treatment at all time points. Similar results were  Reduced Lactobacillus spp. abundance is accompanied by a reciprocal increase in c richness and d diversity. e Bacterial load is significantly higher pre-membrane rupture in comparison to post-membrane rupture (P = 9.37 × 10 -8 ) but remains stable thereafter, despite ongoing erythromycin treatment. PPROM preterm prelabour rupture of the fetal membranes observed in paired samples taken before and 48 hours after erythromycin treatment (n = 16, Fig. 3a-d). A subanalysis of women with Lactobacillus spp. dominance prior to erythromycin exposure (n = 10) showed that treatment was associated with a shift towards an intermediate or dysbiotic community structure in 80% of cases (8/10, P = 0.009, Fisher's exact, Fig. 3a, Additional file 1: Table S3), a significant decrease in Lactobacillus spp. relative abundance (P = 0.0039, Wilcoxon matched pairs, Fig. 3h) and increased diversity (P = 0.0098, Wilcoxon matched pairs) (Fig. 3e,f ). In contrast, erythromycin treatment was associated with a reduction in both richness (P = 0.031) and diversity (P = 0.031, Wilcoxon matched pairs) in samples collected from women with Lactobacillus spp. depletion prior to treatment despite unchanged bacterial load (Fig. 3g).   Table S4]. Examination of vaginal microbiota composition in samples obtained just prior to delivery showed that chorioamnionitis with funisitis was associated with enrichment for Prevotella, Sneathia, Peptostreptococcus and Catonella spp. and reduced Lactobacillus spp. levels (P = 0.0025) compared to women with normal histology (Fig. 4a-c, Additional file 1: Table S5). Chorioamnionitis with funisitis was associated with increased vaginal alpha diversity (P = 0.0134, Mann-Whitney, Fig. 4e), but not richness (Fig. 4d) or bacterial load (Fig. 4f ). Maternal CRP (P = 0.000016) and white cell count (WCC) (P = 0.0016, Mann-Whitney U) (Fig. 4g,h) were elevated in patients with chorioamnionitis with funisitis and both were significantly correlated with vaginal bacterial alpha diversity (WCC; rho = 0.54, P = 0.0001 and CRP; rho = 0.45, P = 0.0013) (Fig. 4i,j).

Vaginal microbiota and early onset neonatal sepsis
A total of 16 cases of EONS (22%, 16/72) were identified from the cohort once intrauterine deaths (n = 2) and cases with insufficient or missing neonatal metadata were removed (n = 13). Neonatal CRP (P = 1.79 × 10 -8 , Mann-Whitney), duration of neonatal antibiotics (118 hours vs. 44 hours, P = 3.12 × 10 -8 , Mann-Whitney), maternal CRP (P = 0.0008, Mann-Whitney U) and presence of chorioamnionitis +/-funisitis (P = 0.0051, chi-squared) were all significantly higher compared to PPROM cases without EONS. Gestational age at delivery [31 +2 (30 +1 -32 +2 ) vs. 27 +6 (26 +1 -29 +3 ) weeks of Lactobacillus spp. depletion (n = 2) whilst communities initially low in Lactobacillus spp. remain so. Erythromycin treatment did not reduce b richness, c diversity or d bacterial load. A sub-analysis showed that erythromycin treatment was associated with decreased e richness (P = 0.03) and f diversity (P = 0.03) in communities initially depleted in Lactobacillus spp., treatment but was associated with a significant increase in diversity (P = 0.01) in communities initially dominated by Lactobacillus spp. g Bacterial load was similar between subgroups. h Lactobacillus spp. abundant communities experienced a 40% reduction in Lactobacillus spp. post treatment (P = 0.004, Wilcoxon signed rank test). Inv. inverse, VM vaginal microbiota Fig. 4 Vaginal dysbiosis is associated with increased risk of chorioamnionitis with funisitis following PPROM. a Differentially abundant vaginal taxa detected using LEfSe analysis in samples taken prior to delivery in women with normal histology or chorioamnionitis with funisitis. b The vaginal microbiome of patients with normal histology was enriched with Lactobacillus spp. whereas those with chorioamnionitis with funisitis were enriched with Fusobacteriales including Sneathia, Bacteroidales, Peptostreptococcus and Catonella species. c Increased prevalence of Lactobacillus depleted/dysbiotic communities with significantly reduced Lactobacillus abundance was observed in the chorioamnionitis with funisitis cohort (P = 0.0025, Fisher's exact). d No difference was detected in vaginal microbial richness between groups. However, e diversity was increased in cases complicated by chorioamnionitis with funisitis. f Bacterial load was comparable between histological groups. g, h Serum markers of maternal infection and inflammation were significantly elevated in cases of chorioamnionitis, +/-funisitis (CRP: P = 0.000016, WCC: P = 0.0016). i, j Both WCC (rho = 0.54, P = 0.0001) and CRP (rho = 0.45, P =0.0013) were positively correlated with vaginal bacterial alpha diversity. CRP C-reactive protein, LDA latent discriminatory analysis, LEfSe linear discriminant analysis with effect size; WCC white cell count gestation, P = 0.0046, Mann-Whitney], birth weight [1707 (1519-1896) vs. 1100 (790-1411) g, P = 0.0037, Mann-Whitney] and Apgar scores at 1 (P = 0.046), 5 (P = 0.01) and 10 minutes (P = 0.03, all Mann-Whitney) were significantly lower in the EONS group. Neonatal gender, mode of delivery, arterial cord pH and latency between membrane rupture and delivery were comparable (Additional file 1: Table S6). Vaginal swabs collected closest to the time of delivery were enriched for Catonella spp. and Sneathia spp. in cases developing EONS, whilst Lactobacillus crispatus was overrepresented in the maternal vaginal microbiota of neonates who did not develop EONS (Fig. 5a-d). Other species previously associated with EONS, including Streptococcus agalactiae, Fusobacterium nucleatum and Escherichia coli, were frequently observed in vaginal samples collected from EONS-complicated pregnancies but not in samples isolated from uncomplicated controls (Fig. 4c). Similar results were obtained when the analysis was repeated only for those mothers who delivered at 28 weeks or sooner (n = 27) (Additional file 1: Figure S3).

Discussion
Infection is strongly associated with PPROM and as a result, empiric antibiotic therapy is routinely used, particularly in high-income countries [58]. In this study, we show Fig. 5 Vaginal dysbiosis is a risk factor for early onset neonatal sepsis (EONS) following PPROM. a Cladogram and b LDA of differentially abundant species detected in vaginal swab samples collected just prior to delivery in cases who did or did not subsequently develop EONS. Lactobacillus crispatus was void from EONS samples, which were enriched with bacteria from genera Sneathia and Catonella. c Distribution of the 21 most abundant bacterial species across the two patient groups. Lactobacillus crispatus was present only in mothers whose babies did not develop EONS. d Relative abundance of differentially expressed genera: Lactobacillus (P = 0.001), Leptotrichiaceae (P = 0.0291) and Lachnospiraceae (P = 0.0114). EONS early onset neonatal sepsis, LDA latent discriminatory analysis, PPROM preterm prelabour rupture of the fetal membranes that vaginal dysbiosis is present prior to the rupture of fetal membranes in approximately a third of cases and is associated with both chorioamnionitis with funisitis and with EONS. Reported benefits of antibiotic treatment following PPROM are often attributed to the prevention of neonatal infection caused by ascending colonisation of pathogenic bacteria originating from the vagina [59]. We, therefore, hypothesised that prophylactic erythromycin would lead to a reduction of vaginal bacterial load, diversity and richness. However, treatment was associated with a shift towards vaginal dysbiosis, particularly in women initially colonised predominately by Lactobacillus species. Our sub-analysis showed that in women with Lactobacillus spp. dominance, erythromycin exposure was associated with a shift towards a dysbiotic community structure in most cases. In contrast, erythromycin treatment was associated with a reduction in both richness and diversity in women with a Lactobacillus spp. depleted vaginal microbiota. There are, therefore, two groups of women who experience PPROM, for one of which erythromycin therapy is detrimental and for the other potentially beneficial. This has important implications for the continued use of prophylactic erythromycin in the context of PPROM, as is currently recommended by the World Health Organization (WHO) [60] and professional bodies throughout the world including the United Kingdom [40], Canada [61], Germany [62], Australia and New Zealand [63].
Lactobacillus-depleted high-diversity vaginal bacterial communities have been identified as risk factors for preterm birth in prospective studies using both culturedependent [29,64] and culture-independent [30,31] methods. The pathophysiology linking vaginal dysbiosis to activation of inflammation and untimely stimulation of prolabour pathways in gestational tissues is well documented [6,65,66]. Our results indicate that around one third of patients have vaginal dysbiosis prior to membrane rupture, providing further evidence for ascending vaginal infection in the pathophysiology of PPROM and preterm birth. Haematogenous infection of the gestational tissues leading to rupture may be responsible for a small proportion of PPROM cases. However, a non-infectious mechanism is likely responsible for the remainder. Therefore, patientspecific selection of targeted antibiotic therapy may improve efficacy and patient outcomes.
Moreover, vaginal dysbiosis just prior to delivery was strongly associated with both chorioamnionitis with funisitis and maternal serum markers of infection and inflammation. Cross-sectional and longitudinal analyses showed that erythromycin failed to resolve this dysbiosis within 1 week of treatment, which coincides with delivery in approximately 80% of cases, and instead was associated with a significant and persistent increase in dysbiotic community structures. This increase was particularly apparent in women with initial colonisation of Lactobacillus species. In women with pre-existing dysbiosis, erythromycin was associated with a reduction in species richness and diversity. However, communities continued to be depleted of Lactobacillus species, indicating there was a restructuring of the highly diverse compositions. Erythromycin treatment beyond 1 week was associated with a recovery of pre-treatment levels of Lactobacillus species dominance. However, the proportion of dysbiotic communities remained unchanged throughout the treatment course.
The use of erythromycin treatment for PPROM is largely driven by results from the ORACLE I trial, which reported prolongation of pregnancy for 48 hours (34.8% vs. 40.7%, P = 0.004), reduced need for supplemental oxygen (31.1% vs. 35.6%, P = 0.02) and a 2.2% reduction in composite neonatal morbidity (neonatal death, chronic lung disease or major cerebral abnormality; 11.2% vs. 14.4%, P = 0.02) in women randomised to orally administered erythromycin prophylaxis compared to placebo [67]. The trial also included randomisation arms of co-amoxiclav and co-amoxiclav plus erythromycin, which were both associated with a significant increase in the risk of necrotising enterocolitis (1.9% vs. 0.5%, P = 0.001 and 1.7% vs. 0.5%, P = 0.005, respectively). The decision to test these antibiotics in the trial was based upon their broad spectrum, complementary ranges of activities, comparatively minimal contraindications in pregnancy and the opportunity to test a macrolide and β-lactam antibiotic. Beneficial outcomes associated with erythromycin treatment in PPROM are often attributed to its assumed inhibition of ascending vaginal infection, but this seems unlikely considering erythromycin concentration in the vaginal lumen following oral dosing is low [68], reaching a mean inhibitory concentration effective against Lactobacillus species [44,69,70], but not against most other species known to colonise the vagina [44]. This provides a possible explanation for the reduction of Lactobacillus spp. and increased diversity and richness of bacterial communities observed in our study following erythromycin treatment that occurred without a reduction in overall bacterial load.
Despite adverse effects on vaginal microbiota composition, reported improvements in neonatal and maternal outcomes following erythromycin treatment for PPROM may be attributable to anti-bacterial activity at other gestational tissue sites (e.g. the placenta) or to other modes of action. Erythromycin is used primarily as an antiinflammatory for the treatment of chronic inflammatory lung disease (panbronchiolitis) [71,72] and has been shown to have tocolytic action in vitro [73]. Considering that the placental transfer of erythromycin into the fetal circulation is low (approximately 2%) [74], neonatal benefits are more likely due to action on maternal tissues and subsequent inhibition of inflammatory mediators that could cross the placenta. Nevertheless, vaginal delivery acts as a high-dose inoculum to the neonate, which shapes the composition of the early infant gut microbiome [25,[75][76][77][78], which is in turn linked to short-term and long-term health outcomes [79,80]. Therefore, aberrant augmentation of vaginal bacterial communities towards dysbiosis just prior to delivery is undesirable and may contribute to poor neonatal outcomes. In our study, vaginal dysbiosis and enrichment of Sneathia spp. and other potential pathogenic bacteria (e.g. Streptococcus agalactiae) just prior to delivery were observed in cases subsequently developing EONS. Sneathia spp. are often associated with bacterial vaginosis [81] and their colonisation of the vagina has been linked to various adverse pregnancy outcomes including septic abortion [82], neonatal bacteraemia [83], neonatal meningitis [84] and chorioamnionitis [85]. Vaginal bacterial communities isolated from cases subsequently developing EONS were almost entirely void of Lactobacillus crispatus. Colonisation of L. crispatus is highly stable in healthy pregnancies from similar cohorts to those studied here [22,25] and dominant colonisation in early pregnancy is associated with protection against preterm birth in those women at high risk [25]. Our data indicate that L. crispatus may also provide protection against subsequent development of EONS.
Although our study size is limited by the practicalities and costs associated with prospectively recruiting large numbers of women subsequently experiencing PPROM, it represents a unique assessment of vaginal microbiota prior to rupture of fetal membranes and is the largest study of the vaginal microbiota in the context of PPROM to date. Given the observational nature of the study, it was not possible, in the context of UK National Health Service care, to longitudinally sample a cohort of women following PPROM who did not receive erythromycin as part of treatment guidelines issued by the WHO, National Institute for Health and Care Excellence (NICE), the Royal College of Obstetricians and Gynaecologists and the recruiting hospital. Therefore, it is difficult to separate the potential temporal impact of membrane rupture on shaping vaginal community structure from the pharmacological effect of erythromycin. Amniotic fluid is highly alkaline with a pH of 7.1-7.3 and contains antimicrobial peptides [86] that may account for the reduction in bacterial load following rupture observed in our study, prior to erythromycin treatment. However, even if erythromycin is not the primary driver of dysbiosis, our data show that it fails to improve the composition of the vaginal microbiome by eradicating potential pathogens or reducing overall bacterial load and is detrimental for individuals with a Lactobacillus spp. dominated microbiome.

Conclusions
Our data show that the composition of the vaginal microbiome is a risk factor for subsequent PPROM and is associated with adverse short-term maternal and neonatal outcomes. Erythromycin prophylaxis promotes Lactobacillus spp. depletion and increased the bacterial diversity of the microbiota, which correlates with increased prevalence of chorioamnionitis, funisitis and EONS. There is an urgent need to review the continued use of prophylactic erythromycin in the context of PPROM and to develop alternative treatment strategies, such as patient-specific therapy, selective antibiotics or different routes of administration, to resolve the vaginal dysbiosis associated with PPROM and improve maternal and neonatal outcomes.

Additional file
Additional file 1: Figure S1. Characteristics of vaginal microbiota groups (VMGs) defined using Ward clustering. Figure S2. Bacterial taxonomic groups discriminate between normal-term delivery and samples taken following membrane rupture. Figure S3. Bacterial taxonomic groups associated with early onset neonatal sepsis (EONS) following PPROM, for neonates delivered at or before 28 weeks gestation (n = 27). Table S1. Bacterial diversity, richness and relative abundance of Lactobacillus spp. for VMGs 1-8. Table S2. Linear regression analysis comparing proportion of Lactobacillus spp. dominance across all patient groups corrected for potential confounders. Table S3. Linear regression analysis comparing proportion of Lactobacillus spp. dominance in paired samples before and after 48 hours erythromycin treatment. Table S4. Maternal and neonatal factors in the presence and absence of chorioamnionitis +/-funisitis. Table S5. Linear regression analysis comparing proportion of Lactobacillus spp. dominance in cases with and without chorioamnionitis +/-funisitis. Table S6