Subjects and sample materials

This study was approved by the Institutional Review Board at Baylor College of Medicine (H-26589) and Harris Health System and was performed with the use of specimens from our universal perinatal database and biospecimen repository (PeriBank database, Institutional Review Board H-26364; Baylor College of Medicine). All pregnant women who come to labor and delivery for delivery are recruited for enrollment into PeriBank, which is Baylor College of Medicine’s universal perinatal database and biospecimen repository. A detailed description of PeriBank and the selection criteria for this study, including the clinical metadata that are extracted, is available in our previous publication on the placental microbiome. Subjects for this study were recruited between August 2011 and November 2012 by trained PeriBank study personnel. As part of the consent process, we discussed with participants the potential risks of participation, including the physical risks that are associated with specimen collection and the possibility that protected health information or deidentified project data that are stored in a public repository could be released accidentally. The protocol and consent form described the precautions that are taken to reduce these risks. Additional protections for participants included coding genomic and metagenomic specimens and sequence data, with the use of controlled-access repositories for extracted nucleic acids and deidentified medical and human genome sequence data, and making efforts to remove any human sequence traces from the microbial data that are deposited in open access databases. If a participant withdrew consent after providing specimens, the remaining specimens and extracted nucleic acids were to be destroyed; however, any metagenomic sequence data that were already published in open access databases could not be retracted.

After written consent was obtained, clinical information was extracted from the medical record and subject interviews. The parent study (H-26589) was a case-cohort study that used 2 case-cohorts. Briefly, for the first case-cohort (preterm birth), the PeriBank database was queried to identify subjects who spontaneously delivered at <37 weeks’ gestation by best obstetric estimate. Spontaneous preterm birth was defined by standard obstetrics definitions as adequately documented gestational age >24 and <37 weeks’ gestation (by last menstrual period with <20-week gestation sonogram or by <14-week sonogram with subsequent second-trimester sonogram) and no evidence of maternal or fetal comorbidity that indicated induction of labor (such as preeclampsia or placenta previa). For the second case-cohort (antenatal infection), the database was queried for women with laboratory evidence of antenatal infection, which included sexually transmitted diseases, urinary tract infections, skin abscesses, appendicitis, periodontal disease, or bacterial meningitis. The control groups for cohort comparison were created by a 1:1 match with previously identified case subjects; women were matched for gestational age (except in the preterm case-cohort), mode of delivery, race/ethnicity, body-mass index, age, and diabetic status. In this manner, we enabled preterm birth cases potentially to carry a diagnosis of a remote history of antenatal infection and vice versa; this is consistent with the principles of a population-based case-cohort study design. Subjects with fatal fetal anomalies, fetal growth restriction, prelabor preeclampsia, placenta previa, and multifetal gestation were excluded because of the likelihood of indicated preterm birth. Women with intrapartum chorioamnionitis that was suspected clinically and had been treated with intrapartum antibiotics and confirmed by positive maternal or infant culture were excluded from this analysis.

For the present study, extracted and analyzed clinical metadata included self-reported and/or documented prepregnancy weight, documented early pregnancy (defined as <20 weeks’ gestation) weight, documented maternal delivery weight, and documented maternal height. Self-reported prepregnancy weight frequently is used in the literature, and the use of self-reported prepregnancy weight has been reported to be within a few pounds of actual prepregnancy weight. The use of first-trimester weight as a surrogate when prepregnancy weight is unknown is also a reasonable substitution when the prepregnancy weight is unknown. Of the 237 subjects in this study, the prepregnancy weight was used in 130 of the women (54.8%). Of the remaining 107 women, the first-trimester weight was used in 98 of 107 pregnancies (91.6%); weights up to 20 weeks’ gestation were used for the remaining 9 women. Ergo, of the 237 women who were included in the analysis herein, 228 of 237 women (96.2%) used prepregnancy or first-trimester documented weights, although only 3.8% arose from documented weight in the early midtrimester (>14 and <20 weeks’ gestation). Patients who had none of these variables were excluded because data were insufficient to calculate GWG. Other extracted clinical metadata included gravidity, parity, the presence or absence of Streptococcus agalactiae , the presence of antepartum infections and antibiotic use (ie, sexually transmitted infection, urinary tract infection, skin abscesses, appendicitis, periodontal disease, or bacterial meningitis), routine demographic variables and maternal health comorbidities (eg, age, race/ethnicity, smoking status, diabetes mellitus, hypertensive, hypothyroid status), and pregnancy outcomes (ie, gestational age at delivery, mode of delivery, presence of gestational hypertensive, and preeclamptic disorders). For the purposes of this study, all classes of diabetes mellitus and all hypertensive disorders were assessed in aggregate.

GWG calculation

GWG was calculated by subtracting the prepregnancy or early pregnancy weight (defined as <20 weeks’ gestation) from the delivery weight. Baseline body mass index (BMI) was determined with the use of prepregnancy or early pregnancy (<20 weeks’ gestation) weight. As noted previously, 96.2% of study subjects had a documented prepregnancy or first-trimester maternal weight. To calculate whether GWG exceeded recommendations, we used previously validated, BMI-specific formulas that were derived from the Institute of Medicine. Excess weight gain was defined as weight gain that was in excess of the maximum recommended weight gain for a given gestational age and prepregnancy BMI. Specifically, the Institute of Medicine recommends a maximum of 4.4 pounds in the first trimester for all women, regardless of BMI; thus, this was used as the first-trimester component and was added to the maximum recommended weekly weight gain, which is BMI-specific. These values are 1.3, 1, 0.7, and 0.6 pounds per week for underweight, normal weight, overweight, and obese women, respectively. This rate was multiplied by the number of weeks of pregnancy from the second trimester through delivery to yield the formulas shown in Table 2 , which shows the maximum recommended GWG by BMI ; GWG that exceeded the maximum recommended GWG was defined as excessive GWG . The women were then assigned to 2 cohorts: those with excess GWG (excess GWG) and those without excess GWG (no excess GWG).

Placental samples

All samples were collected by placental pathology-trained personnel according to the PeriBank Manual of Procedures. As described in the parent publication, after the placenta is delivered by the obstetrics provider, it is immediately passed off to PeriBank personnel in a clean container. In a separate room, 4- 6 1 cm × 1 cm × 1 cm cuboidal sections are excised circumferentially from separate areas of the placenta; each section was located approximately 4 cm from the cord insertion site by PeriBank personnel who wore facial masks and gloves and used sterile scalpels and tissue forceps. These clean and sterile conditions are part of the standardized technique that is used in the banking of all PeriBank specimens. After excision, the placental surfaces, namely the maternal decidua and the fetal chorionamnion, were excised and discarded to avoid potential contamination from maternal, fetal, or environmental surface exposure during delivery (vaginal or cesarean) or at any time during transport.

Metagenomics sequencing

This study was a subgroup analysis of our initial case-cohort study of the placental microbiome among subjects with antepartum infections or subjects with spontaneous preterm delivery (<37 weeks’ gestation). In the previous study, after we identified subjects with any antepartum infections or spontaneous preterm birth, we then queried the database for 1:1 matches for women without antepartum infection or preterm birth. As described in our previous publication, the placental samples from 48 women were selected randomly from the overall case-cohort to undergo whole-genome shotgun sequencing. Approximately one-fourth of these samples were from the preterm birth case-cohort; one-fourth of the samples were from the remote antenatal infection case-cohort, and the remaining one-half of the samples were from the control subjects. Of note, the nested 48 subjects did not significantly differ from their derived cohort counterparts by any measure other than their case definition (not for maternal age, mode of delivery, race/ethnicity, BMI, maternal comorbidities, or tobacco use; Student t test on continuous variables and chi-squared analysis for categoric variables with lack of significance designated to be P > .05; 7). The present study used women (n = 237 of 320) for whom pregestational weight or early gestational weight was documented.

Metagenomic DNA extraction and sequencing

All samples were collected under uniform protocol (PeriBank) by placental pathology-trained personnel who wore facial masks and gloves, as described earlier. Approximately 100-150 mg of frozen placental tissue were taken from each sample and were homogenized and heated then processed with the use of PowerSoil DNA Isolation Kits (MO-BIO Laboratories, Carlsbad, CA). Briefly, genomic DNA was isolated according to the standard protocol from the MO-BIO PowerSoil DNA Isolation Kit (MO-BIO Laboratories). DNA was quantified by Nanodrop assay (Thermo Fisher Scientific, Wilmington, DE) and Picogreen assay (Invitrogen, Carlsbad, CA) and extracted DNA samples were stored at –20°C. Bacterial genomic DNA that had been isolated from placental samples was sheared to approximately 700 base pairs, and Illumina adapters (Illumina Inc, San Diego, CA) were ligated to the sheared fragments for whole genome shotgun sequencing, and analysis. DNA fragments were then paired-end sequenced (150 base pair [bp] reads) in multiplexed pools on the Illumina HiSeq 2500 platform (Illumina Inc) in the Baylor College of Medicine-Human Genome Sequencing Center, as described in our previous papers, including the parent publication on the placental microbiome. Because of the concern for bias in low-abundance samples, all specimens were thrice extracted and thrice run to enable triplicate extraction and sequencing. Sequences from all 3 extractions were combined for analysis.

16S recombinant DNA-based metagenomic sequencing and data analysis

Variable region amplicons of the bacteria 16S rRNA gene were sequenced as previously described. For whole genome shotgun sequencing and analysis, bacterial genomic DNA isolated from placental samples was sheared to approximately 700 bp and Illumina adapters were ligated to the sheared fragments. DNA fragments were then paired-end sequenced (150 bp reads) in multiplexed pools on the Illumina HiSeq 2500 platform. From 30 ng/μL of DNA, variable region amplicons were sequenced as previously described. The variable regions 1-3 of the 16S rRNA gene were amplified by polymerase chain reaction with bar-coded universal primers 27F (F, forward primer) and 534R (R, reverse primer) that contained linker sequences for 454-pyrosequencing. Samples that contained no template and those that contained known 16S rDNA sequences were used as positive and negative controls in these polymerase chain reaction steps. Variable-region amplicons were quantified by Picogreen assay and pooled in equimolar amounts. Multiplex sequencing was performed at the Baylor College of Medicine–Human Genome Sequencing Center on a 454-FLX Titanium sequencer (454 Life Sciences, Banford, CT). Raw 16S sequences were demultiplexed with quantitative insights into microbial ecology. Taxonomy was assigned with the use of the Ribosomal Database Project (RDP) classifier (version 2.2, re-trained with GreenGenes; Center for Microbial Ecology, Michigan State University, East Lansing, MI).

Concordant with previous work of ours and others, all 16S data were quality filtered to attain high-quality reads. Sequences from each file were preprocessed; thereby, sequences with length <200 nucleotides or >1000 nucleotides and sequences with minimum average quality score <20 and sequences without precisely matched 5′ primers were removed. Sequences were assigned thereafter to their sample of origin based on barcode matches and the 5′ primer, and the barcodes were excised after the filtering. RDP classifier (version 2.2 [re-trained using the release Feb. 4, 2011, of the GreenGenes taxonomy ]) was used to assign taxonomy for each read at various levels. We used reads with confidence level >50% at lowest level of assignment for further analysis. The read counts at each taxonomy level (phylum, order, family, genus) were normalized to relative abundance.

For operational taxonomic unit (OTU) analysis, quality, and RDP-filtered reads were analyzed with the quantitative insights into microbial ecology pipeline and AbundantOTU and to assign each read to OTUs. Quantitative insights into microbial ecology performed OTU picking, with error correction and OTU centroid choice at 97% similarity. ChimeraSlayer was used to identify and remove chimeric sequences from the following analyses: Artifact caused by sequencing error was removed with the use of AbundantOTU as a denoising approach. OTUs were analyzed by RDP classifier (version 2.2) and well to find out the corresponding taxonomy. The number of reads assigned to each OTU was counted and normalized. The generated OTU table, combined with the clinical metadata, was used as input for follow-up analysis.

Principal coordinate analysis (PCoA) was used to measure beta diversity on normalized OTU tables with Unifrac distance. Briefly, these plots demonstrate the degree to which each placenta’s microbiome differs from that of other placentas. With Unifrac, all taxa that are identified are placed on a phylogenetic tree. A branch that leads to taxa from both samples is marked as shared, and those taxa that appear only in one sample are marked as unshared. Distance in this sense is defined as the sum of unshared branch lengths over the sum of all branch lengths (both shared and unshared). Each sample is then plotted in its own dimension at the appropriate distance for each other point on the plot. The aggregate data from the 2 groups of placentas (obese or nonobese women or women with and without excess GWG, respectively) are used to calculate the overall degree of difference between the 2 groups, which is reflected in the permutational analysis of variance (PERMANOVA). Boruta feature selection and linear discriminate analysis effect size algorithms were used to detect significant features (taxonomy and pathway abundance) that differentiate groups and rank these features by effect size.

Whole genome shotgun metagenomic analysis

As anticipated, there was a large amount of human DNA sequence in our placental metagenomics datasets. Thus, to process the whole genome shotgun (WGS) reads, human contamination was identified and removed with Best Match Tagger according to the steps described in the HMP Human Sequence Removal Standard Operating Procedure and elsewhere. This ensured quality and privacy. Metabolic pathway reconstruction was inferred directly based on short reads with the HMP Unified Metabolic Analysis Network. Quality filtered reads were uploaded to metagenomic-RAST server (MG-RAST) for taxonomy identification with the use of WGS reads. Taxonomy assignment was based on best hit classification, with the use of the M5NR database as an annotation source with a maximum e-value cutoff of 1, minimum percent identity cutoff of 50%, and minimum alignment length cutoff of 15. MG-RAST was used to generate low level Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway abundance for comparison between case and control-cohort samples. MG-RAST–generated taxonomy and pathway abundance were further normalized by the number of reads that were mapped to each sample. HUMAnN (the HMP Unified Metabolic Analysis Network) was used to KEGG-based community functions for metabolic reconstruction from WGS reads.

Metabolic pathway reconstruction directly inferred the KEGG-based community functions as described elsewhere. For linear discriminate analysis effect size analysis, the alpha value that was used for the factorial Kruskal-Wallis test was 0.05. The threshold used on the logarithmic linear discriminate analysis score for discriminative feature is 2.0, as we previously demonstrated.

Statistical analysis

Descriptive findings of clinical data were reported. Where appropriate, χ ² testing was performed to assess categoric variables; the Student t test was performed on continuous variables with normal distribution, and the Wilcoxon rank-sum test was performed on continuous variables without normal distribution. A probability value of < .05 was considered statistically significant. Statistical analysis was performed with SPSS software (version 22.0; SPSS Incorporated, IBM Corp, Armonk, NY). The significance of PCoA plots were analyzed with PERMANOVA, which uses distance metrics to confirm the strength and statistical significance of sample groupings.

All sequencing files have been submitted to the Sequence Read Archive.