Birth cohorts studied
To investigate possible colonization of the placenta associated with preterm birth, we analyzed 20 placentas from spontaneous preterm births and 20 from full-term deliveries (Table 1 and Additional file 1: Table S1). Placenta tissue samples were processed to remove the external layers in an effort to exclude microbes adhering to the outside of the placenta acquired during delivery. Maternal side and fetal side samples were both collected. Six cases of preterm birth were complicated by both clinical chorioamnionitis (maternal fever > 100.4 and at least one of the following: maternal tachycardia > 100 bpm, fetal tachycardia > 160 bpm, or fundal tenderness) and histological chorioamnionitis (white blood cells detected in fetal membranes). As positive controls, we collected saliva and cervicovaginal fluid from the mothers—both body sites that are known to harbor rich microbial communities [21,22,23]. We also collected several types of negative controls (Additional file 1: Table S2). These included (1) swabs that were opened in the sample processing room and waved in the air (“Air Swab”), (2) empty tubes that were processed through the DNA extraction (“Blank”), and (3) PCR grade water processed in parallel with the samples during amplification and DNA sequencing acquisition, though not DNA purification (“H2O”).
DNA purification
In our previous work, we used two different DNA purification kits to prepare samples, because purification kits are known to be sources of contaminating bacterial DNA [16, 17]. Previously, we used the MO BIO PowerSoil and PSP kits and found that each kit yielded DNA producing distinct bacterial sequences when amplifying negative controls. DNA extracted from placenta samples with the two kits not only resembled negative controls but also matched the particular background of each extraction kit, helping us distinguish the influence of reagent contamination. Thus, in this study, we also used two DNA purification kits (Additional file 1: Table S3). We used the DNeasy PowerSoil kit to match previous work [6, 15], and also, in an attempt to suppress the influence of contamination, we compared a newly available DNA purification kit designed to minimize adventitious DNA, the QIAamp UCP Pathogen Mini kit (UltraClean). As positive controls, we also purified samples expected to contain robust microbial communities from maternal saliva and vaginal swabs. Only one set of vaginal swabs was available, so these were purified using the UltraClean kit.
Absolute abundance of bacterial DNA in samples quantified using 16S rRNA gene-targeted qPCR
We investigated our sample set to determine whether placenta DNA samples contained higher absolute levels of bacterial DNA than negative controls as measured using 16S rRNA gene qPCR (Fig. 1). Samples from saliva and cervicovaginal fluid contained high concentrations of bacterial DNA (low cycle of threshold), as expected (Fig. 1, right two sample sets). Negative controls showed high cycles of threshold, indicating low 16S rRNA gene content (Fig. 1, leftmost three sample sets). Placenta samples contained little 16S rRNA gene DNA, with cycles of threshold similar to negative controls. (Fig. 1, middle). Comparisons between the placenta samples and Blank and H2O samples show that there was no significant difference (Placenta (F)–Blank, p = 0.915; Placenta (F)–H2O, p = 0.392; Placenta (M)–Blank, p = 0.958; Placenta (M)–H2O, p = 0.426; Kruskal-Wallis with Dunn’s post-test) (Additional file 1: Table S4). We found significant differences between placenta samples and Air Swabs (p = 0.001 for fetal side, p = 0.002 for maternal side); however, the negative control Air Swabs yielded a lower CT (higher amounts of bacteria; Additional file 1: Table S4). We found no differences between sample sets when comparing placenta samples separated by term, chorioamnionitis, or extraction kit with the negative controls (Additional file 1: Tables S5-S10).
We then compared placenta samples between term and preterm births. Two comparisons achieved significance, but the direction was inconsistent. For data from the UltraClean kit, comparison of fetal side placenta from preterm and term births showed a significantly lower mean value CT for the preterm births (p < 10−5, Kruskal-Wallis). However, for data from the PowerSoil kit, comparing the maternal side placenta data to controls showed lower mean CT for the term births (p = 0.0396, Kruskal-Wallis). We compared the chorioamnionitis placenta samples to other preterm birth and pooled placenta samples and found no significant differences. We conclude that, unlike the saliva or vaginal samples, the numbers of 16S rRNA gene copies in placental samples were not significantly higher than in negative controls and that there is little difference between preterm, term or chorioamnionitis samples.
Analysis of bacterial DNA using 16S rRNA marker gene sequencing
We next assessed the composition of the 16S rRNA gene sequences in our samples using 16S rRNA marker gene sequencing. For this, we analyzed the V1-V2 region of the 16S rRNA gene, because previous work suggested that the placental microbiome might be derived from oral communities [6], and V1-V2 has been used to characterize oral communities in multiple peer-reviewed publications [24,25,26,27,28,29]. This region of the gene is also relatively short (typically ~ 260 bp), which increases amplification efficiency and so is useful for analyzing low biomass samples [24, 26]. We amplified and sequenced the sample set (Additional file 1: Table S3) and as an additional positive control added a set of synthetic DNAs encoding divergent microbial 16S rRNA gene sequences to verify robust performance of biochemical steps and proper sample tracking [17]. We recovered useable data (> 100 reads) for 303 of 388 samples (Additional file 1: Table S11 and Additional file 2: Figure S1). Of the placenta samples, 62/160 (38.75%) did not meet this threshold. We speculate that clean work up of low biomass samples resulted in suppression of background DNA contamination, leading to fewer 16S rRNA sequencing reads in some samples. The output 16S rRNA gene sequences were clustered using Dada2, and clusters were assigned taxonomically using the SILVA database.
The major lineages detected are summarized in Fig. 2a and detailed further in Additional file 2: Figure S2. Vaginal and oral samples showed well-known bacterial lineages associated with these body sites; vaginal samples contained high proportions of Ureaplasma, Sneathia, and Lactobacillus, and oral samples were high in Neisseria, Porphyromonas, Streptococcus, and Prevotella [21,22,23]. Samples from placenta, and also the negative controls, were high in environmental bacteria commonly associated with contaminants, such as Ralstonia and Pseudomonas[16]. Placenta and negative control samples also contained reads matching to chloroplasts (Streptophyta), possibly derived from pollen in dust. Other bacteria present in both placenta and negative control samples included Prevotella and Enterobacteraceae, which may be either human-associated or environmental. Several vaginal lineages were detected selectively in some of the placenta samples, including Ureaplasma, Sneathia, and Lactobacillus. Further analysis showed that these were found predominantly in the samples from vaginal deliveries and not cesarean deliveries. To test this further, we asked whether vaginal lineages found in placenta matched between mother-placenta pairs more closely than in unmatched mother-placenta pairs; placentas were found to share lineages with their corresponding mother more frequently than with other mothers (p < 10−5 for comparison of Jaccard distances). In contrast, samples from cesarean deliveries showed no such correspondence (p = 0.24 for comparison of Jaccard distances), supporting the idea that vaginal lineages in placentas were derived from the mother during vaginal delivery.
Clustering of samples using unweighted UniFrac showed separation of the saliva and vaginal samples from all other sample types, while negative controls and placenta samples formed an overlapping large cluster (Fig. 2b). Sample clustering was then assessed statistically by PERMANOVA (all p values are in Additional file 1: Table S12), comparing weighted UniFrac, unweighted UniFrac, and Jaccard distances.
Comparison of placenta samples from preterm and term deliveries showed no significant difference. Comparison of vaginal delivery versus cesarean delivery also showed no difference for the UniFrac analyses, but did achieve significance for Jaccard distance. Distances were significantly different between kits (p = 0.021 all samples, fetal, maternal). Comparisons between preterm and term deliveries, performed separately on samples of each kit, showed no differences. Comparisons of vaginal delivery versus cesarean delivery, performed separately on samples purified using each kit, also showed no difference, suggesting that there are no strong distinctions based on delivery type in the data.
Comparisons of sample type showed significant differences between blanks versus saliva, vaginal, and air swab samples. Significant differences were also found when comparing samples of each kit. However, blanks and placenta samples showed no difference between sample type (Fig. 2c). Comparisons of blanks with placenta samples performed on samples from each kit separately also did not show significant differences.
Comparison of placenta samples associated with chorioamnionitis showed no significant difference compared to other placenta samples. An overview of the bacterial lineages detected is in Additional file 2: Figure S3. Recently, Leon et al. [30] published a study of 265 pregnancies, assessing the possible placenta microbiome in term and preterm samples using 16S rRNA amplicon sequencing. They too found no consistent placenta microbiome over background. They did however find representation of Ureaplasma and Mycoplasma in some preterm placentas. In our data, we assessed possible colonization with these organisms, requiring detection by 16S rRNA amplicon sequencing in at least 2/4 samples per placenta to call a lineage as present. Analysis was carried out at the level of operational taxonomic units, our most discriminating measure (Additional file 2: Figure S4). We found Mycoplasma in 2/20 preterm deliveries and none of the term deliveries; we found Ureaplasma in 3/20 preterm deliveries and 1/20 term deliveries. Thus, although there was no statistically significant difference between our preterm and term samples, detections of Mycoplasma and Ureaplasma in the 16S rRNA amplicon data from preterm samples are potentially consistent with results of Leon et al. However, we note that 5/6 cases are in vaginally delivered placentas and the organism could be detected in the vaginal sample from the same woman (Additional file 2: Figure S4). Thus, contamination with vaginal microbes during delivery is also a possible explanation for the origin of Ureaplasma and Mycoplasma in the preterm placenta samples.
The exception was a unique Ureaplasma amplicon sequence variant detected in two placenta samples and not in the corresponding vaginal swab from a preterm, cesarean section, and chorioamnionitis-associated birth (subject 265). Ureaplasma has been associated with chorioamnionitis, and thus in this case may represent detection of an authentic infection in subject 265 [31], paralleling results of Leon et al. [30].
We conjecture that kit contamination was a less significant factor here than in Lauder et al. [15] and that sequences from other sources, such as laboratory water (e.g., Ralstonia) and maternal vagina were the major source of contamination in this study. Thus, our analysis of 16S rRNA marker gene sequencing did not disclose any consistent distinction between placenta samples and negative controls.
Analysis of total DNA content using shotgun metagenomic sequencing
Previous studies have used shotgun metagenomic sequencing to analyze the placenta microbiome [6], so we decided to also compare shotgun sequencing for this sample set. A major disadvantage of using shotgun metagenomics to interrogate human tissue is that the vast majority of sequencing effort will be spent on sequencing human DNA. We thus focused on sequencing a subset of our samples, so as to obtain more sequencing depth. We chose to sequence only the samples extracted with the UltraClean kit because that set included the vaginal swabs.
Deep sequencing produced a large number of reads for each sample type: 1.4 billion reads for placenta, 31 million for negative controls, 190 million for saliva, and 72 million for vaginal samples (Fig. 3). Filtering out sequences matching the human genome removed large proportions of reads: 99.8% of the placenta samples, 19.3% of the negative controls, 76.0% of the saliva, and 97.8% from the vaginal swabs. The large proportion of human sequences removed from the placenta samples left only an average of 35,600 total reads per sample to interrogate for microbial composition.
Reads were then filtered a second time using Komplexity to mask low-complexity reads likely derived from host repetitive DNA elements such as microsatellites that eluded the first filter [32]. Next, the reads were assigned taxonomically using Kraken [33]. The reads were filtered a final time to remove those classified as Chordata and Arthropoda, which were likely human sequences that eluded earlier filtering steps, and Apicomplexa, unlikely placental colonists, because database genomes of this group are known to be problematic [34]. Finally, Kraken assignments with three or fewer classified reads (summed over the full sample set) were removed.
Major lineages detected for each sample are shown in Fig. 4 and detailed further in Additional file 2: Figure S5; placenta samples only are summarized in Additional file 2: Figure S6. Saliva samples showed the expected oral bacterial lineages, including Streptococcus and Prevotella. Vaginal lineages were dominated by Lactobacillus and Gardnerella. Placenta samples and the negative controls contained high proportions of Ralstonia, a bacteria known to be a frequent reagent contaminant [16], which was also detected in the 16S rRNA marker gene sequence data. Thus, the major placenta bacterial lineage in the shotgun sequencing data could be attributed to contamination.
The next two most abundant lineages in placenta, Alteromonas mediterranea and Methanosarcina mazei, were less abundant in controls, but both appear to be artefactual detections. Analysis of reads aligning to Alteromonas mediterranea showed that the 5050 classified reads aligned to only 0.221% of the 4.6 million base pair genome, and some of these reads aligned to human DNA as well as the bacterial genome. Methanosarcina mazei also showed sparse coverage of the genome, with 1159 reads aligning to only 0.0098% of the 4.1 million base pair genome, and the sequences that mapped were of low complexity (Additional file 1: Table S13). Low-complexity sequences are known to be problematic in quality control and classification steps, and in some cases, these low-complexity samples aligned to human DNA as well as the bacterial genomes, likely explaining their presence.
A few placenta samples were also higher in Vibrio bacteria than were negative controls. We used DNA from Vibrio campbellii, a marine bacteria, as a positive control during sequence acquisition. Reads matching Vibrio campbellii formed over 2% of nonhuman placental reads and up to 31% of a single sample. We thus infer that barcode misreading during sequence acquisition [17] accounted for the detection of this organism.
We next asked whether any distinctions could be found among the different types of placenta samples. When we extracted lineages unique to placenta after excluding those found in negative controls (Additional file 1: Table S14), we recovered very few reads. The most abundantly represented taxon, Acidiplasma cupricumulans, was identified with only 12 reads in 11 samples total. Acidiplasma cupricumulans is an acidophile from the phylum Archaea and not a likely candidate for a placental colonist. Other organisms showed even lower read numbers, leaving us doubtful that any represented authentic placental colonists.
After filtering out probable artifacts and barcode misattributions (Alteromonas mediterranea, Methanosarcina mazei, and Vibrio campbellii), comparison of preterm and term deliveries showed a significant difference (p = 0.002, PERMANOVA, Jaccard distance). However, none of the taxa identified show more than 24 reads aligning to the target genome. (Additional file 1: Tables S15 and S16). The most abundant organism identified, Kytococcus sedentarius, was identified largely in a single preterm, fetal side placenta sample (20 out of 24 classified reads) that was from a pregnancy with preterm premature rupture of membranes (PPROM) and chorioamnionitis. Kytococcus sedentarius was discovered as a marine bacteria and is suggested to also be an opportunistic human pathogen [35]. Reads classified as Kytococcus sedentarius were also found in other samples including Saliva, Blanks, and H2O at similar or higher levels, suggesting this bacteria could be derived from reagent contamination. We thus find that all differential detections are of extremely low abundance and are sample specific. Comparison of vaginal deliveries and cesarean deliveries showed no distinction (p = 0.204, PERMANOVA, Jaccard distance). Comparison of fetal versus maternal side placenta samples showed no significant distinction (p = 0.169, PERMANOVA, Jaccard distance). Comparison of placenta samples from cases of chorioamnionitis to other placenta samples also showed no differences (Additional file 2: Figure S6). Thus, we conclude that analysis by shotgun metagenomic sequencing also did not disclose a detectable placenta microbiome, and consistent with the lack of signal, we did not see convincing biological distinctions in the data.