Study design and SARS-CoV-2 infection in a Kenyan mother-infant cohort from September 2021 – March 2022
We performed a nested study of Kenyan women and their infants enrolled in a prospective cohort in Nairobi, Kenya who agreed to SARS-CoV-2 testing. Seventy-four mother-infant pairs were included here: 42 women living with HIV (WLHIV), 32 HIV-uninfected women, and their 74 infants: 42 HIV-exposed uninfected (HEU) infants, 32 HIV-unexposed uninfected (HUU) infants (Fig. 1 and Fig. S1). WLHIV are given cotrimoxazole to prevent opportunistic infections during high-risk periods, such as pregnancy and breastfeeding. Additionally, HEU infants are given cotrimoxazole as a prophylactic from the age of 4–6 weeks old until cessation of breastfeeding and confirmed HIV-uninfected, typically around 18 months, though this age may vary (27).
Nasopharyngeal swabs were collected for COVID-19 testing (via TaqPath 2.0 qRT-PCR assay) from September 2021 to March 2022 resulting in a total of 1,262 unique swabs from 148 individuals available for this study (average of 8.5 swabs per person over 7 months). Among these, 20 individuals (13.5%) had a single swab test positive for SARS-CoV-2 (1.6% of the total samples). The 20 individuals testing positive for SARS-CoV-2 included 7 WLHIV, 5 HIV- uninfected women, 6 HEU and 2 HUU; there were four mother-infant dyads in which both mother and infant were SARS-CoV-2 positive (Fig. 1, Table S1). One individual tested positive in October, 3 in late-November, 15 in mid-December, and 1 in January. Six infants negative for SARS-CoV-2 with reported respiratory symptoms at the peak of SARS-CoV-2 cases in Kenya were included as a symptomatic group. Through hybrid capture next-generation-sequencing, we determined of the six symptomatic infants, one infant was infected with WU Polyomavirus, 4 infants were infected with different strains of Human Adenovirus, and two infants we were unable to determine an etiology (Table S2). Additionally, 34 individuals that tested negative for SARS-CoV-2 infection at all timepoints were included as a comparator group: 14 WLHIV, 32 HIV-uninfected women, 7 HEU and 4 HUU (Fig. S2). This cohort is unique in that the women and infants were prospectively sampled longitudinally during the COVID-19 pandemic, allowing us to document changes in the nasopharyngeal microbiome before, during and after SARS-CoV-2 infection.
We first sought to determine which SARS-CoV-2 variants caused infection in the cohort. Through whole genome sequencing, we identified 15 individuals infected with the Omicron variant, one individual infected with the Delta variant, and four individuals for whom there was insufficient genome coverage to successfully determine the variant. We constructed a phylogeny using the 16 SARS-CoV-2 sequences (Fig. 2A). Sequences from three mother-infant dyads included in the phylogeny suggest that the mothers and infants who were SARS-CoV-2 positive at the same time were infected with the same specific strain of SARS-CoV-2.
To compare results of specific SARS-CoV-2 variant transmission in our Nairobi-based cohort to sequences circulating more-widely in Kenya, we performed a sequence network transmission analysis using the sequences from our cohort and 3,016 SARS-CoV-2 sequences from 36 different Kenyan cities and counties submitted to GISAID during the peak of COVID-19 cases both in Kenya and in the study cohort (November 2021 - December 2021; Fig. 2B). November and December 2021 were when most of the study cohort was positive for COVID-19 as well, showing the epidemiology seen in the cohort is comparable to national Kenyan statistics. To examine the introduction of SARS-CoV-2 throughout Kenya, we analyzed the collection dates and locations of samples from GISAID within sub-lineages. There is no clear clustering by geographic area (Fig. 2C), but when collection date information is added, we observe samples from the Nairobi region coinciding with earlier timepoints (Fig. 2D). This would support a scenario where the Nairobi region served as the primary point of SARS-CoV-2 ingress and was spread to the surrounding regions. These data show the cohort samples were consistent with contemporaneous variants circulating in Kenya during the study period.
The nasopharyngeal microbiome community states in women and infants
To determine the community state profiles present in the mothers and infants, we clustered the nasopharyngeal bacterial microbiome abundance at the species level using the k-means method resulting in 6 clusters (Fig. 3). Cluster 2 was the largest cluster with 106 samples and was dominated by Staphylococcus epidermidis (18%) and Enterococcus cecorum (16%). Clusters 1, 3, and 4 were dominated by Dolosigranulum pigrum (23%, 17%, and 62% respectively). Clusters 1 and 3 were also dominated by Moraxella nonliquefaciens (45% and 12% respectively), and cluster 3 additionally was 33% Haemophilus influenzae and 24% Streptococcus pneumoniae. Cluster 5 was predominantly Corynebacterium segmentosum (48%), and cluster 6 was dominantly Corynebacterium propinquum (42%). Multinomial logistic regression indicated that all community state clusters were associated with woman/infant status. Specifically, clusters 1, 3, and 4 were majority infant samples (88%, 92%, and 77% respectively), while clusters 2, 5, and 6 were majority woman samples (90%, 94%, and 59% respectively, Fig. 3, see green versus pink in first row below the abundance plots, p-values ranging from 0.01–0.05). None of the community states were associated with HIV-infection (row 2), timepoint (row 3), SARS-CoV-2 infection (row 4), or antibiotics usage (row 5). Taken altogether, these results indicate that community state is most influenced by woman-infant status, and further analyses should be stratified by woman-infant sample.
Diversity and richness of the nasopharyngeal bacterial microbiome in women and infants
We next considered within-person and between-person microbiome variation between women, infants, and family dyads. Beta diversity was greater when comparing samples between different women than when comparing all samples from an individual woman (p-value = 3.11e-34; Fig. 4A). Similarly, beta diversity was significantly greater when comparing samples between different infants than when comparing all samples from an individual infant (p-value = 7.94e-8; Fig. 4B). In addition, when comparing women and infants using Bray Curtis distance, we observed high dissimilarity between women and infants; however, related women and infants were more like one another than unrelated women and infants (p-value = 1.6e-2; Fig. 4C). Together, these results show that infants and women have lower within-person and within-family variation than between-person variation.
We then considered the differences in bacterial biomass (assayed by 16S qPCR), richness, and alpha diversity between infants and women. Infants had significantly higher bacterial biomass than women (p = 1.39e-10; Fig. 4D). However, women had significantly higher richness (p-value = 7.50e-16) and alpha diversity than infants (p-value = 1.07e-16), which was consistent over time (Fig. 4E and Fig. 4F). Additionally, PCoA analysis of weighted Bray-Curtis dissimilarity showed clearly distinct clustering of women’s samples apart from infant’s samples (PERMANOVA, p-value = 0.001; Fig. 4G). Taken together, these data suggest the infant nasopharyngeal microbiome is more densely populated with fewer bacterial species, whereas the microbiome of adult women is richer and more diverse with less overall biomass, and that these differences were stable over time.
Impact of HIV on nasopharyngeal microbiome
We next sought to determine whether HIV infection has an impact on the nasopharyngeal microbiome of women and infants. HIV-uninfected women consistently had higher alpha diversity than WLHIV. However, the difference was small and not statistically significant (p-value = 0.1614; Fig. 5A). PCoA analysis of weighted Bray-Curtis dissimilarity (Fig. 5B), did not show clustering based on HIV status (PERMANOVA, p-value = 0.345) in women. Additionally, there was no difference in bacterial biomass between WLHIV and HIV-uninfected women (p-value = 0.8717; Fig. 5C). Similarly, in infants, we saw no change in alpha diversity (p-value = 0.5108) between HEU and HUU infants (Fig. 5D), and there was no clustering by HIV-exposure status (PERMANOVA, p-value = 0.795; Fig. 5E). There was also no difference in bacterial biomass between HEU and HUU infants (p-value = 0.2989; Fig. 5F). Together, these results suggest living with, or exposure to, maternal HIV infection treated with optimized, long-term antiretroviral therapy does not have an impact on the nasopharyngeal microbiome.
Longitudinal dynamics of the nasopharyngeal microbiome and the impact of SARS-CoV-2
We then examined whether temporal changes in the linear trajectories (slopes) of alpha diversity and richness of the nasopharyngeal microbiome were associated with incident SARS-CoV-2 infection, HIV exposure, or antibiotic use. Analyses were conducted separately for women and infants. Among women, neither richness nor alpha diversity changed significantly over time and there was no association between SARS-CoV-2 infection, or antibiotics usage (p-values ranging from 0.0897–0.9277). There was a significant association between richness and HIV status in WLHIV (p = 0.0494), but the relationship was not significant in alpha diversity. In infants, there was also no significant change over time in richness or alpha diversity, and neither SARS-CoV-2 infection nor HIV infection were associated with change in alpha diversity or richness (p-values ranging from 0.1002–0.8819). Convergence of the infant nasopharyngeal microbiome to that of women is expected, but the period in which the infants were sampled was too short to detect the infant microbiome reaching adult maturity. There was an association between alpha diversity and antibiotics usage (p = 0.0289), but this significance was not seen in bacterial richness. Together these data suggest that the infant and maternal nasopharyngeal microbiomes were stable throughout the period of observation and were not influenced by SARS-CoV-2 infection, maternal HIV status or infant HIV exposure or antibiotic use.
While the overall dynamics were not impacted by SARS-CoV-2, we next compared alpha diversity in pre-infection timepoints to the first SARS-CoV-2 positive timepoints to assess transient changes in the nasopharyngeal microbiome. We also compared the nasopharyngeal microbiome before, during, and after SARS-CoV-2 infection between SARS-CoV-2 positive individuals to controls that never acquired SARS-CoV-2 infection. To do so, a longitudinal sampling timeline was constructed using the nasopharyngeal swabs, with two timepoints from the 2 weeks pre-infection, one timepoint during documented SARS-CoV-2 infection and one timepoint an average of 38 days post-infection (Fig. S2A). All individuals positive for SARS-CoV-2 had no reported respiratory symptoms at the time of infection. For the controls, the timing was similar, but we substituted the SARS-CoV-2 infection sample with either a timepoint at which the mother or infant had reported respiratory symptoms (Fig. S2B), or a matched calendar timepoint if there were no reported respiratory symptoms (Fig. S2C).
In women, there was no significant difference in alpha diversity (p-value = 0.7746) when comparing pre-infection timepoints to the timepoint of SARS-CoV-2 infection (Fig. 6A). These results were also supported by PCoA of weighted Bray-Curtis dissimilarity, which did not show clustering of negative samples from never positive women, negative samples from women with SARS-CoV-2 infection, or SARS-CoV-2 positive samples (PERMANOVA, p-value = 0.276, Fig. 6B). There was also no significant difference (p-value = 0.5549) in bacterial load when comparing womens’ negative samples to positive SARS-CoV-2 samples (Fig. 6C). Similarly, there was no statistically significant difference in alpha diversity among infants when comparing negative timepoints to SARS-CoV-2 infected timepoints (p-value = 0.7756). Nor was there a difference between negative timepoints and symptomatic timepoints (p-value = 0.2060) (Fig. 6D). There was no clustering of negative samples from never positive infants, negative samples from infants with SARS-CoV-2 infection, SARS-CoV-2 positive samples, or samples from timepoints with reported respiratory symptoms (PERMANOVA, p-value = 0.559, Fig. 6E). Lastly, we compared bacterial load at pre-infection timepoints, SARS-CoV-2 timepoints, and symptomatic timepoints (Benjamini-Hochberg corrected p-values = 0.3755) in infants (Fig. 6F). Taken together, we conclude that neither SARS-CoV-2 infection nor undiagnosed respiratory infections caused dysbiosis of the nasopharyngeal bacterial microbiome from both negative samples and never positive samples.
Though we did not detect a change in the nasopharyngeal microbiome at the time of SARS-CoV-2 infection, SARS-CoV-2 can cause long term sequalae led us to ask whether there were effects on the microbiome after infection. The next available subsequent follow-up sample (average 38 days post-infection) was termed the “recovery” timepoint, and all recovery timepoints were negative for SARS-CoV-2 RNA. The alpha diversity of the nasopharyngeal microbiome remained stable in women even after SARS-CoV-2 recovery (p-value = 0.8874, Fig. 7A). PCoA on weighted Bray-Curtis dissimilarity showed no discernable clustering (PERMANOVA, p = 0.095) in women when comparing negative, SARS-CoV-2 samples and recovery samples (Fig. 7B). There was also no statistical difference (p-value = 0.7021) between SARS-CoV-2 infected timepoints and recovery timepoints when looking at bacterial load (Fig. 7E). Neither infants infected with SARS-CoV-2 (p-value = 0.7789, Fig. 7D) nor infants who were symptomatic for respiratory illness showed a difference in alpha diversity (p-value = 0.5368) from the time of infection to post infection (Fig. 7E). No clustering was apparent with PCoA on weighted Bray-Curtis dissimilarity between negative, SARS-CoV-2 samples, and symptomatic infants, and their respective recoveries (Fig. 7F). Finally, we looked at the differences in bacterial load between infected samples and all available recovery samples. There was no statistical difference in infants between both SARS-CoV-2 infected timepoints and symptomatic timepoints and their respective recoveries (p-values = 0.3357 and 0.2468 respectively, Fig. 7G and Fig. 7H). Taken together, these results show that SARS-CoV-2 infection does not have a lasting impact on the nasopharyngeal microbiome and suggest the nasopharyngeal microbiome is resilient in both mothers and infants, maintaining stability before, during, and after SARS-CoV-2 infection.
Specific bacterial species that differentiate the nasopharyngeal microbiome in women and infants
Lastly, we performed a multivariate analysis using MaAsLin2 and used novel machine-learning methods to identify discriminating taxa among women and infants and by SARS-CoV-2 infection status. We identified 197 statistically significant discriminating bacterial species (Table S3) that differed based on mother-infant status (Fig. 8A). Of these, Moraxella nonliquefaciens, Moraxella catarrhalis Haemophilus influenzae, Streptococcus pneumoniae, Corynebacterium propinquum and Dolosigranulum pigrum were more common in infants, while Staphylococcus epidermidis, Staphylococcus aureus, Cutibacterum acnes, Corynebacterium segmentosum, and Corynebacterium macginleyi were more common in mothers.
We also used novel machine learning methods to identify discriminating taxa associated with SARS-CoV-2 infection. AUC scores of the best models for combined and stratified datasets confirmed the models should be stratified between women and infants (Fig. 8B). We identified 84 taxa that differed significantly at the familial level according to SARS-CoV-2 infection status (Table S4) and plotted the mean Shapley Additive Explanations (SHAP) values for both women and infants (Fig. 8C). Together, these data corroborate our previous finding that the microbiome composition of women is different than that of infants and suggest that SARS-CoV-2 may impact the nasopharyngeal microbiome through smaller compositional changes at the familial taxa level.