Linear mixed effects and PERMANOVA models

We used linear mixed-effects (LME) models (R package nlme, version 3.1–148) to compare changes in richness and alpha diversity across post-partum time (mothers) or month of life (infants) while accounting for repeated measurements. We used PERMANOVA (vegan, permute and Adonis in R) to assess changes in beta diversity as measured by unweighted and weighted Bray-Curtis distances. Person code (mother or infant), maternal HIV status, post-partum time/month of life (age), time since weaning, antibiotic use, time since SARS-CoV-2 infection were included in the LME and PERMANOVA models to account for exposures and/or possible confounding. For infants, time-since-weaning is an important time variable in addition to age, as introduction to solids has a large impact on the gut microbes in infants^38,39. Because infant weaning was collinear with age (Pearson r = 0.94, p < 0.0001), we included them in models separately. For the models we used two different methods to classify SARS-CoV-2 samples: first, we used SARS-CoV-2 infection status at sample collection time and second, we used if patients were ever versus never SARS-CoV-2 seropositive throughout follow-up in the year 2020. This method allowed us to code samples as controls, pre-SARS-CoV-2 infection, and post-SARS-CoV-2 infection in the models. We also measured time since SARS-CoV-2 infection for women who were ever seropositive to see if there were differences in the gut microbiota over time after infection. We first ran LME and PERMANOVA (Adonis) models with to assess whether outcomes differed between mothers and infants and inform if subsequent analyses should be stratified by person. After stratifying by person, we included an interaction between maternal HIV status and SARS-CoV-2 status to assess the need for further stratification by HIV status in women.

To assess if there was a change in the richness, alpha and beta diversity of SARS-CoV-2 women in comparison to controls, we modeled an interaction term between SARS-CoV-2 and time postpartum/infant month of life. We also removed the interaction term to re-run the LME and PERMANOVA models to test if for differences in alpha diversity, richness and beta-diversity due to the co-variables (metadata) within the model. For the women who were ever SARS-CoV-2 seropositive, we created interactive models for SARS-CoV-2 infection status at time of collection and time to compare the changes that occur in the alpha diversity and richness before and after seroconversion. We then wanted to analyze if there was a change in the first ever positive SARS-CoV-2 samples from all SARS-CoV-2 negative samples. Only the first positive samples were kept in the models. Linear and PCoA plots were plotted using ggplot2 (version 3.3.1).