Genital tract infections, the vaginal microbiome and gestational age at birth among pregnant women in South Africa: a cohort study protocol

Data sources and variables

The source data are case report forms recording questionnaire data for the enrolment, third trimester and postnatal visits, and forms for laboratory and specimen results, which are stored in REDCap, a secure web-based database30 (online supplemental file), hosted by the Foundation for Professional Development, Pretoria, South Africa.

Specimen collection

At the enrolment and the third trimester visits, a study nurse collects two vaginal smears using inoculation loops and air dries them on glass slides. She then collects vaginal specimens by inserting swabs into the vagina up to a mark at 4 cm and rotating around the vaginal wall. Five swabs are collected in the following order: one Cepheid GeneXpert Xpert Vaginal/Endocervical Swab in a tube with Xpert Swab Transport Reagent (Cepheid); two Qiagen digene Female Swabs in a single tube with digene Specimen Transport Medium (Qiagen, Hilden, Germany); and two dry FLOQSwabs (COPAN, Brescia, Italy) each in a separate sterile tube (figure 1).

Transport and storage of specimens

Vaginal smear glass slides are stored and transported in plastic slide carriers at room temperature. All vaginal swabs are initially stored at the clinic in a refrigerator (2–8°C with daily temperature checks). All vaginal swabs, except the Xpert swab, which is tested on-site, are transported on ice packs once a week by overnight road courier to the laboratory at the Department of Medical Microbiology, University of Pretoria, where they are also stored in a refrigerator until DNA extraction.

Microbiological analyses

The Xpert vaginal swabs are tested on-site using the Xpert CT/NG assay (Cepheid) to detect C. trachomatis and N. gonorrhoeae, as per manufacturer’s instructions. At the University of Pretoria, air-dried vaginal smears are heat fixed and Gram stained.36 Two qualified people record the Nugent scores (0–3: normal; 4–6: intermediate; 7–10: bacterial vaginosis) and the presence of yeasts.37 In case of discrepancies, a third person assesses the slide and consensus is reached by discussion. At the University of Pretoria, one vaginal FLOQSwabs is used for PCRs. The genomic DNA is extracted using the High Pure PCR Template Preparation Kit (Roche Diagnostics, Mannheim, Germany) as per manufacturer’s instructions. Real-time PCR assays are then performed using the LightCycler 480 Probes Master Kit (Roche Diagnostics) on the LightCycler 480 II instrument (Roche Diagnostics). Previously published primer and hydrolysis probe sequences and cycling conditions are used for detection and quantification of M. genitalium,38 M. hominis,39 U. parvum,40 U. urealyticum,40 T. vaginalis41 and Candida spp.42 43 The load for each assayed microorganism detected in vaginal swab specimens by real-time PCR or GeneXpert is obtained from the cycle threshold value.

Vaginal microbiome laboratory analyses

The vaginal swabs stored in Qiagen digene Specimen Transport Medium will be used for DNA extraction and subsequent 16S ribosomal RNA (rRNA) amplicon sequencing targeting the V3–V4 hypervariable regions for vaginal microbiota analyses at the Division of Medical Microbiology, University of Cape Town.

A commercial DNA extraction kit will be used and a bead-beating step included.44 A DNA isolation control will be prepared from an unused vaginal swab specimen during this process. Two PCR rounds will be conducted to prepare amplicon libraries.45 The aim of the first PCR round is to amplify 16S rRNA gene V3–V4 regions using the 319F 5′-ACTCCTACGGGAGGCAGCAG-3′ forward primer and 806R 5′-GGACTACHVGGGTWTCTAAT-3′ reverse primer. The aim of the second PCR round is to barcode the V3–V4 amplicons by a dual-index approach, permitting multiplexing of up to 384 samples (including controls). Amplicon concentrations for all sample libraries are measured and normalised to form a mixed loading library. The libraries will be sequenced on an Illumina MiSeq instrument (Illumina, San Diego, California, USA), 2×300 bp. To quantify the number of 16S rDNA copies per swab, a quantitative PCR using the same forward and reverse primers as described above will be used. Samples from enrolment and third trimester visits from the same woman will be processed in the same run.

Vaginal microbiota bioinformatics

Raw sequencing reads will be processed using an established bioinformatics pipeline.46 Taxonomic assignment of amplicon sequence variants will be done in DADA247 with SILVA48 as the reference database. Vaginal microbiome composition data will be visualised in heatmaps and diagrams. For each vaginal sample, we will calculate diversity measures (alpha diversity), relative abundances and estimated concentrations of key vaginal bacteria and bacterial groups, as described.46 We will use the entire sequencing dataset to design vaginal microbiota types by hierarchical clustering, and each sample will be assigned to one vaginal microbiota type.

Study timeline

Enrolment began on 28 March 2021, with an estimated date for reaching the target sample size in August 2023. Follow-up of all participants until the postnatal visit is expected to be completed by March 2024.

Objective 1: association between specified exposures and pregnancy outcomes

1. We will examine a primary set of microorganisms as exposures, detected at either enrolment or at the third trimester visit: M. genitalium, M. hominis, U. urealyticum, U. parvum, T. vaginalis and Candida spp are the microorganisms for which women did not receive diagnostic tests and treatment during study visits. We will use the mean and SD for the continuous outcome (gestational age) and absolute and relative frequencies for the binary outcomes (all secondary outcomes). Gestational age at birth for each exposure will be compared using Student’s t-test and a mean difference with 95% CIs (continuous outcome) and Fisher’s exact test and a risk difference with 95% CIs (binary outcomes).

   To control for confounding, multivariable regression models will be fitted to all outcomes with a set of prespecified potential confounders (age, educational level, alcohol consumption, HIV infection, prior preterm birth) for all organisms. For analyses of M. genitalium, T. vaginalis and Candida spp, we will also control for bacterial vaginosis (Nugent score 7–10). The other genital mycoplasmas can be identified from 16S rRNA amplicon sequencing in women with vaginal dysbiosis, so are sometimes considered part of bacterial vaginosis. For these organisms, we will conduct descriptive analyses, stratified by the presence of bacterial vaginosis. For continuous confounders, a linear relationship will be assumed by default but transformations (eg, log) or more flexible approaches (eg, splines or fractional polynomials) will be considered if there is evidence for non-linearity. For the continuous outcome, we will use linear mixed-effects regression models (including data from either visit and the participant as random effect) and report the result as mean difference with 95% CIs. For the binary outcome, we will use logistic mixed-effects regression and report the result as ORs with 95% CIs.

2. Comparisons for associations with timing of detection, other microorganism exposures and birth outcomes will be considered secondary analyses. Associations between vaginal microbiota composition and pregnancy outcomes will be assessed. We will use compositional multivariable analysis methods to identify bacterial taxa that are differentially abundant between binary pregnancy outcome groups at the level of individual taxon relative abundances. We will use mixed-effects models (with the individual participant as the random effect and including data from both visits) to assess associations between continuous and binary pregnancy outcome and the following fixed effects derived from the vaginal microbiota data: alpha diversity, vaginal microbiota types and absolute abundances of predefined bacterial groups.46 These models will be adjusted for confounding as described in the previous paragraph.

Objective 2: association between quantified microorganism load and pregnancy outcomes

We will investigate the hypothesis that the quantity of microorganisms with inflammatory potential is associated with gestational age at birth. For this, we will analyse the vaginal microbiota data jointly with sexually transmitted infections and Candida spp diagnostic test results during pregnancy (these will be considered as additional covariates in the above-mentioned regression models). We will develop a ‘vaginal inflammation index’ based on quantification of the vaginal microbiota and their inflammatory potential49 and of yeasts. This vaginal inflammation index will also be analysed as a fixed effect in mixed-effects models with pregnancy outcomes as the outcomes; these models will not include any of the infection parameters that were used to design the index.

Objective 3: classification and regression tree analysis for the primary outcome

We will conduct exploratory analyses to examine the combination of microorganisms that best predicts earlier gestational age at birth using classification and regression tree analysis.50 This method belongs to the family of decision tree machine learning algorithms and allows for non-parametric analyses of a large number of binary, categorical or continuous predictors. They are typically easy to interpret and can detect predictors with small marginal effects when there are strong interaction effects. We will make use of the predictive potential for gestational age at birth of all sexually transmitted and genital tract microorganisms, including individual bacterial taxa or bacterial groups identified by 16S rRNA gene amplicon sequencing (as binary or continuous variables) and confounding variables identified in objective 1. We will present variable importance scores and curves of marginal effects to show how prediction of the outcome changes at different levels of the exposure of each variable in the model. To avoid overfitting, we will consider bootstrap aggregating via random forests.51

Data management

Each potential participant screened for eligibility is assigned a unique participant identification number, which does not include any personal identifying information. Personal identification numbers are used to link records, specimens and laboratory test results of the participants. Data are stored in a REDCap database,30 which is only accessible to authorised project staff. Paper records are kept in lockable fire-resistant filing cabinets. Laboratory records and journals are kept at the University of Pretoria and University of Cape Town. Forms with personal identifying information are kept separately from demographic, clinical and other data. The data manager maintains a separate, access-controlled database that links the personal identification number with identifying information. Data quality checks are conducted by study staff on-site and data administrators at the office of the Foundation for Professional Development. All study data are stored securely at the offices of the Foundation for Professional Development in East London for up to 5 years after the completion of the study or as required by the institutional review board.

Confidentiality

The research team is trained to adhere to guidelines on the Protection of Human Research Participants and Good Clinical Practice and fully protects the confidentiality of participants. Besides the measures described under data management, interviews are conducted in a private setting. In reports and publications, data will not be presented in a way where it could be linked to individual participants.