Vaginal and neonatal microbiota in pregnant women with preterm premature rupture of membranes and consecutive early onset neonatal sepsis | BMC Medicine

Composition of the vaginal microbiota of PPROM patients

The 16S rRNA gene sequencing was performed for 166 vaginal swab samples collected at hospital admission (V0), after antibiotic treatment (V1), and < 24 h before delivery (V2). The overall upstream analyses resulted in 2923 amplicon sequence variants (ASVs) assigned to 20 different phyla and 752 species (Additional file 7: Table S7). Almost half of the vaginal ASVs (1401/2983 ASVs) were observed in V0 samples (n = 78), belonging to 13 phyla and 483 species.

At the time of the hospital admission (V0), 65.4% of the pregnant women harbored a vaginal microbiota dominated by Lactobacillus species (rel. abundance > 50%), mainly L. crispatus (32/78) and L. iners (17/78). Lactobacillus jensenii (3/78) and L. gasseri (1/78) were also dominant in a few samples (Fig. 1). This is in accordance with the data of previous reports comparing the vaginal microbiota in PPROM patients with those of controls, where PPROM patients were less often colonized by a Lactobacillus dominated community compared to controls (56–77% compared to 81–91%) [36, 37].

Fig. 1
figure 1

Microbial composition of vaginal samples of PPROM patients at hospital admission (V0). Heatmap includes taxa with a relative abundance > 10% in at least one sample. Samples where a single taxon shows a relative abundance > 50% of relative abundance are indicated by a specific color code (dominance group). Hierarchical cluster based on Bray–Curtis similarity analysis of vaginal bacterial communities of V0 samples (n = 78)

These communities dominated by Lactobacillus sp. have been termed community states I, III, V, and II [9]. They were characterized by a low diversity as indicated by a Shannon index (H′) below 1.5 (often < 1). Besides Lactobacillus sp., Gardnerella vaginalis was highly abundant and dominant in five samples. In another five samples, G. vaginalis and L. iners were equally abundant and together accumulated > 70% of sequence reads. Similarly, G. vaginalis often co-occurred with L. gasseri, and in four samples, they together comprised > 70% of sequence reads. In contrast, the presence of a high abundance of L. crispatus seems to be incompatible with a high abundance of G. vaginalis. Other species only seldomly become dominating. Malacoplasma girerdii, Corynebacterium pyruviciproducens, Enterococcus faecalis, Fannyhessea vaginae, and Escherichia/Shigella were dominating in one sample each, with only the sample dominated by Malacoplasma girerdii exhibiting an extremely low diversity (H′ = 0.24).

Composition of the vaginal microbiota during antibiotic treatment

Comparison of microbial community composition based on their Bray–Curtis similarities showed that out of 23 samples taken < 2 days after starting of antibiotic administration only 4 exhibited a similarity < 50% to the community present before the antibiotic treatment. This indicates that in most of these cases, the antibiotic treatment time was insufficient to change the bacterial community composition on the 16S rDNA level. In contrast, the microbial community remained relatively stable (> 50% similarity) in only 4 out of those 40 cases where antibiotic was administrated for > 2 days (Additional file 8: Fig. S1).

Then, alpha-diversity indices of 44 pairs of V0 and V1 samples were analyzed (samples being collected at hospital admission or < 2 days thereafter versus samples of patients having been treated with antibiotics longer than 2 days). A significant increase in bacterial diversity (H′; p = 0.0054) was observed (Fig. 2). The increase in bacterial diversity was accompanied by an increase in evenness (Pielou index; p = 0.03) and richness (Chao1 index; p = 0.0084).

Fig. 2
figure 2

Vaginal community diversity indicated by Shannon diversity (H′), Pielou’s evenness (J′), and Chao1 richness. Statistically significant differences between the groups of samples (V0 and V1) are indicated by **p < 0.01 and *p < 0.05. Quartiles and median are shown

Thirty-one out of these 44 patients had a vaginal microbiota dominated by Lactobacillus species before antibiotic treatment (Fig. 1). As reported, antibiotic treatment had a tremendous effect on Lactobacillus spp. abundance in most of these samples (Fig. 3), and similarities between V0 and V1 communities were below 50%. However, in five patients (J01, R11, R18, H15—Lactobacillus species; J18—Escherichia/Shigella), the community remained similar.

Fig. 3
figure 3

Changes in the dominant bacterial taxon of vaginal samples during antibiotic treatment. A taxon is defined as dominant if it is present in a relative abundance > 50%. Only samples of those 44 patients are shown, where both samples during hospital admission (V0) and samples taken between > 48 and < 170 h of antibiotic treatment (V1) were available

The microbiota analysis showed 14 V1 samples being dominated by Ureaplasma parvum (Figs. 3 and 4), with 8 of them exhibiting a low diversity (H′ < 1). Although U. parvum was frequently observed in V0 samples, it was not dominant in any vaginal microbiota before antibiotic treatment (Figs. 1 and 3). A few cases were dominated by Bifidobacterium breve, Corynebacterium jeikeium, Escherichia/Shigella, Metamycoplasma hominis, Prevotella bivia, P. melaninogenica, or U. urealyticum.

Fig. 4
figure 4

Microbial composition of vaginal samples of PPROM patients during antibiotic therapy (V1). Heatmap includes taxa with a relative abundance > 10% in at least one sample. Samples where a single taxon shows a relative abundance > 50% of relative abundance are indicated by a specific color code (dominance group). Hierarchical cluster based on Bray–Curtis dissimilarity analysis of vaginal bacterial communities of V1 samples (n = 44)

Composition of vaginal microbiota at the time of delivery

A total of 63 samples were collected right before delivery. Twenty-nine of these V2 samples were collected < 2 days after admission to the hospital (V2e). They typically were similar to the respective V0 samples (> 50%) and only 4 V2e had a community composition considerably different from that observed at hospital admission (J07, J35, R04, and R13; < 33% similarity). In accordance, twelve V2e samples were dominated by L. crispatus (12/29; 41%) and another six samples by L. iners, L. jensenii, or L. gasseri (6/29; 21%) (Additional file 9: Fig. S2).

Thirty-four V2 samples were collected after the patient had received antibiotics for at least 2 days. For 22 of those patients, a sampling set of V0 (at admission), V1 (between 2 and 6 days of antibiotic treatment), and V2l (after extended antibiotic treatment) was available (Additional file 10: Fig. S3). The microbial composition remained stable in 4 patients at V1 (R11, R18, J01, and J18, V0–V1 similarity > 50%), even though the extended antibiotic treatment resulted in the eradication of Lactobacillus spp. from those communities. The only community that remained stable (J18, V0–V1, and V0–V2 similarity > 50%) was dominated by Escherichia/Shigella already at hospital admission.

Interestingly, V2l samples of patients that received antibiotics until the time of birth did not change dramatically in microbial composition (43.9 ± 19.5% similarity) upon extended antibiotic treatment (between V1 and V2l samples), and 6 of 10 samples (H11, H18, J29, J46, R03, and R09) kept a similarity > 50% (Additional file 10: Fig. S3). In contrast, the microbial composition changed more dramatically (V1–V2, similarity 24.0 ± 22.7%; p = 0.042) in patients that stopped the antibiotic treatment at least 2 days before delivery (V2r) (Additional file 10: Fig. S3). However, in only one case (J42), the microbial composition recovered to that observed before the antibiotic treatment, and V2r became dominated by L. crispatus (Additional file 9: Fig. S2).

Composition of neonatal microbiota

Samples collected from 84 neonates were processed for sequencing. Ninety-seven of 309 samples (31.4%) did not yield an amplification product or a sufficient number of qualified reads. A total of 60 meconium, 58 rectal swabs, 55 pharyngeal swabs, and 39 umbilical cord blood samples passed through the quality control and were analyzed (Additional file 11: Table S8). Overall, the neonatal microbiota showed a heterogeneous microbial composition with multiple species in high relative abundance (Fig. 5).

Fig. 5
figure 5

Microbial composition of neonatal samples. Heatmap includes taxa with a relative abundance > 10% in at least one sample. Hierarchical clusters are based on the Bray–Curtis dissimilarity analysis of bacterial composition. The heatmap includes meconium samples (ME, n = 60), rectal swab samples (RE, n = 58), pharyngeal swab samples (PH, n = 55), and umbilical cord blood samples (UC, n = 39)

A comparison of the Bray–Curtis similarities between the vaginal microbiota at the time of birth and the respective neonatal microbiota showed that, specifically, the umbilical cord blood samples of neonates born by vaginal delivery had a community composition highly similar to those of the mothers. This indicates that the microbial community on umbilical cord blood samples are remnants of the maternal microbiota from the umbilical cord surface rather than the microbial content of the neonatal bloodstream. Accordingly, bacterial communities determined from the umbilical cord blood of neonates born by Cesarean section were significantly less similar to those of the mother (p = 0.0184) (Fig. 6). In contrast, meconium samples from neonates born by vaginal delivery had a lower similarity to the V2 microbial community compared to other neonatal sites sampled. Therefore, meconium microbiota has to be considered as minimally contaminated in dependency of birth mode and as a reliable microbiota source representing a possible intrauterine bacterial colonization after PPROM.

Fig. 6
figure 6

Similarities in microbial community structure between the vaginal microbiota at the time of birth and neonatal microbiota. The Bray–Curtis similarities are given based on standardized species abundance data. a Similarity between V2 samples and umbilical cord blood (n = 18), pharyngeal swabs (n = 24), rectal swabs (n = 22), and meconium samples (n = 23) of neonates delivered by spontaneous birth. b Similarity between V2 samples and umbilical cord blood (n = 16), pharyngeal swabs (n = 23), rectal swabs (n = 29), and meconium samples (n = 32) of neonates delivered by C-section. The difference in similarity to V2 samples between neonatal communities of different origins was analyzed by the Kruskal–Wallis test, and a significant difference in pairwise comparisons is indicated with *p < 0.05, **p < 0.01, and ***p < 0.001

Whereas the vaginal microbiota of early delivery mothers (V2e, < 2 days antibiotic treatment) and late delivery mothers (V2m > 2 days antibiotic treatment) were in fact significantly different (Table 1), there was no significant difference in the meconium communities of the respective children indicating that antibiotics given to the mothers do not exert a significant effect on the meconium communities.

Table 1 Pairwise Results for PERMANOVA based on the Bray–Curtis similarities of vaginal and meconium communities calculated from standardized relative species abundance data of early (< 48 h) and late (> 48 h) delivery mothers and their neonates

However, even though vaginal and meconium communities were significantly different and a significant effect of antibiotic treatment of the mother was not observed on meconium communities, some correlations between vaginal and meconium samples were obvious, such as the dominance of L. crispatus in the meconium samples of neonate H21, or even, the dominance of U. parvum and Escherichia/Shigella in meconium samples of neonates J13 and J18, respectively (Additional file 12: Fig. S4). Besides maternal vaginal microbiota, skin inhabitants such as Cutibacterium acnes and typical gastrointestinal organisms, such as Helicobacter pylori, were observed in meconium samples (Additional file 12: Fig. S4).

We observed above that communities characterized from umbilical cord blood samples exhibited some similarities with maternal vaginal samples. To analyze if there were or were no significant differences between umbilical cord blood samples as well as pharyngeal and rectal swab communities, these were compared to those of the mothers at the time of delivery. The 2-way PERMANOVA (Additional file 13: Table S9) revealed that V2 communities differed from the corresponding pharyngeal, rectal, and umbilical cord blood samples. However, there was no statistically significant difference between the neonatal samples. This holds both for early delivery and late delivery samples. As observed for meconium communities, a significant difference between early and late delivery communities was only evident for vaginal but not for any neonatal communities.

To analyze in more detail the differences between the bacterial communities, the difference in the distribution of taxa was compared between vaginal (V2) and neonatal samples (ME, PH, RE, NB). Clearly, V2 was the most distinct, and various bacterial species and genera were differentially distributed (Additional file 14: Table S10).

Among others, different Anaerococcus species (A. hydrogenalis/senegalensis, A. lactolyticus, A. murdochii, A. obesiensis, and A. octavius), Campylobacter species (C. hominis, C. ureolyticus), Dialister species (D. microaerophilus, D. propionifaciens), Peptoniphilus species (P. coxii, P. gorbachii, P. grossenis, P. harei, and P. lacrimalis), or Prevotella species (P. bergensis, P. bivia, P. buccalis, P corporis, and P. timonensis) were significantly higher abundant in vaginal communities compared to the neonatal communities, whereas C. acnes, Acinetobacter junii, or Staphylococcus spp. were less abundant. Meconium samples were characterized by specific taxa enriched in this niche and Bacteroides ovatus, Bifidobacterium longum, Parabacteroides merdae, Phocaeicola vulgatus, Agathobacter rectale, Faecalibacillus, Phascolarctobacterium, and Romboutsia were of significantly higher abundance in meconium communities compared to others neonatal communities and to maternal vaginal communities, as well (Fig. 7).

Fig. 7
figure 7

Abundance of selected bacterial taxa which were differentially distributed between meconium (ME), umbilical cord blood (UC), pharyngeal swabs (PH), rectal swabs (RE), and the vaginal microbiota at birth (V2). a Taxa of significantly higher abundance in vaginal samples compared to neonatal samples. b Taxa of significantly higher abundance in meconium samples compared to other compartments tested. The difference in similarity was analyzed by the Kruskal–Wallis test, and pairwise comparisons were performed by the Dunns’s post hoc test with *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001

The PERMANOVA pairwise analysis between spontaneous vaginal delivery and C-section delivery did not show any significant influence of delivery type on neonatal microbial composition for any of the neonatal sampling groups tested. However, there were also cases where a transfer of vaginal communities to the neonate was clearly observed. As an example, all UC, PH, and RE samples of neonates H06 and R07 were dominated by U. parvum as was the mother’s vaginal microbiota at birth and similar observations were made for a possible transfer of Escherichia/Shigella in neonate R04 and of L. crispatus in neonate H05.

Prediction of early-onset neonatal sepsis

Sixteen PPROM patients delivered 18 neonates which developed EONS (Additional file 2: Table S2). To evaluate if the vaginal community of the mother could be predictive for EONS, we analyzed if the V0, V1, or V2 communities differed at the species level between mothers giving birth to at least one child suffering from EONS and mothers giving birth to children not suffering from EONS (Table 2).

Table 2 Results for pairwise PERMANOVA based on the Bray–Curtis similarities of vaginal communities calculated from standardized relative species abundance data of mothers giving birth to at least one neonate later on suffering from EONS and mothers giving birth to neonates not suffering from EONS

Whereas PERMANOVA did not indicate any difference when comparing V0 and V1 communities (p = 0.837 and 0.253, respectively), the difference in the overall community structure was significant when V2 samples were compared (p = 0.044). The difference was even more pronounced when only mothers having received antibiotics for an extended period of time were considered (V2l, p = 0.035). On the single taxa level, Anaerococcus obesiensis, Anaerococcus lactolyticus, Campylobacter ureolyticus, and Howardella trended to be underrepresented in samples of mothers where a child suffers later from EONS, whereas Escherichia/Shigella, Enterococcus faecalis, Facklamia, Winkia neulii, S. aureus, and Eremecoccus trended to be overrepresented in mothers of EONS (Table 3) (Additional file 15: Fig. S5); however, higher sample numbers are necessary to further prove these results.

Table 3 Bacterial taxa differentially distributed in the vaginal microbiota of EONS and non-EONS cases

It was then analyzed if also the neonatal meconium and pharyngeal and rectal swab communities differ between EONS and non-EONS cases. PERMANOVA did not indicate a significant difference between the meconium communities of EONS and non-EONS neonates (p = 0.188) whereas significant differences were observed in pharyngeal (p = 0.006) and rectal swab (p = 0.033) communities. Birth mode had no significant influence on the community structures (meconium, p = 0.402; pharyngeal swab, p = 0.646; rectal swab, p = 0.695).

An analysis of the pharyngeal swab taxa differentially distributed between EONS and non-EONS neonates showed that taxa of increased relative abundance in non-EONS cases are typical skin inhabitants such as S. epidermidis or C. acnes (Table 4). Whether this increase in the relative abundance of skin organisms is due to an overall lower bacterial load and hence a relatively higher impact of skin-derived bacteria remains to be established. There were no taxa clearly differentially distributed on rectal swabs. In contrast, a higher prevalence of Bifidobacterium longum and Agathobacter rectalis was indicated in meconium samples, as putative protective organisms (Table 4). As mentioned above, higher sample numbers are necessary to further prove these results.

Table 4 Bacterial taxa differentially distributed in the neonatal microbiota of EONS and non-EONS cases

The potential use of vaginal and meconium communities as possible biomarkers to discriminate and predict EONS and non-EONS cases was tested using three different modeling techniques (logistic regression (LR), linear support vector machine (SVM), and random forest classifier (RF)). Best performances were achieved with logistic regression modeling for vaginal and pharyngeal swab communities, and with support vector machine modeling for meconium communities using cube root transformed data.

Areas under the receiver operating characteristic curves (ROC-AUC) of up to 0.675, 0.736, 0.672, and 0.685 could be achieved for V2 communities (all samples), V2l communities (vaginal communities under antibiotic treatment for > 6 days), meconium communities, and pharyngeal swab communities, respectively (Fig. 8). The 5-feature pharyngeal model was capable to correctly classify 86% of EONS cases. However, only 58% of non-EONs cases were correctly classified. Here, the best performance was achieved using the 100 features meconium model, which could correctly classify 78% of EONS and 84% of non-EONS cases. Also, the vaginal communities showed a reasonable predictive power for classifying EONS cases, with a 64% true positive and a 69% true negative rate. Communities having been subject to extended antibiotic treatment showed a slightly enhanced predictive power (67% true positive and a 77% true negative rate) (Additional file 16: Fig. S6).

Fig. 8
figure 8

Using microbial communities for the prediction of EONS. ROC curves for a logistic regression (a, b, d) and a support vector machine classifier (d) trained on 5–100 taxa of vaginal (a all V2 samples and b V2l samples > 6 days antibiotic treatment), meconium (c), and pharyngeal swab communities (d). The amount of taxa used (Var), the area under the curve (AUC), and the confidence interval (CI) are given as inserts

Then, three- to four-taxon models were evaluated on a training cohort consisting exclusively of samples from the two trial sites Halle and Rostock hospitals. Models were validated using an independent cohort of samples from the third trial site Jena hospital. For all V2 samples, the model comprising Escherichia/Shigella and Facklamia (risk taxa) as well as Anaerococcus obesiensis and Campylobacter ureolyticus (protective taxa) yielded AUCs of 0.674 and 0.788, respectively. Only a few samples were available to perform this analysis for V2l samples. However, also here, high AUC values of 0.975 and 0.79 were obtained using the risk taxa Escherichia/Shigella and Winkia neuii and the protective taxon Anaerococcus obesiensis as features. Finally, EONs cases could also be predicted at a reasonable rate from neonatal meconium communities with the protective taxa Bifidobacterium longum, Agathobacter rectale, and S. epidermidis as features yielding AUCs of 0.592 and 0.753 for training cohort and validation cohort, respectively, while AUCs of 0.706 and 0.664 for prediction of EONS by the pharyngeal community protective taxa S. epidermidis, C. acnes, and Neisseriales were observed (Additional file 17: Fig. S7).

Read more here: Source link