Study population
The Linda Kizazi Study was a prospective birth cohort study of 211 mother-infant pairs in Nairobi, Kenya from 2018–2022 (see methods)23, 42. Between January 1-December 31, 2020, a subset of 63 mothers and 72 infants consented to SARS-CoV-2 testing and had longitudinal stool samples collected for bacterial microbiome and virome sequencing (Fig. 1A-B). Thirty-one mothers (women) were living with HIV (WLHIV) and receiving ART and 32 were HIV-negative (Fig. 1); 35 infants were HEU and 37 were HUU. SARS-CoV-2 infection timing was determined by serology testing of plasma samples collected approximately quarterly; 23 mothers and 12 infants were positive for SARS-CoV-2 antibodies at any time during study follow-up, which was prior to the availability of SARS-CoV-2 rapid tests or vaccines in Kenya23, 42.
When stratified by SARS-CoV-2 seropositivity, we found that there was no significant difference in infant age, gestational age at birth, time from weaning, delivery route or maternal CD4 count; however, antibiotic use, maternal age and post-partum time at sample collection were significantly different (Table 1). When stratified by HIV-status, we found no difference by maternal post-partum time, infant age, gestational age at birth, time from weaning or delivery route; however maternal age at study enrollment was significantly higher among WLHIV (Table 1).
Table 1
Population characteristics
Characteristics | Controls | SARS-CoV-2 | HIV-negative | WLHIV | P-value (SARS-CoV-2; HIV status) |
Maternal age (median, IQR, range) | 27 (6, 22) | 30 (6.5,18) | 26 (8, 22) | 30 (6, 16) | 0. 0097; 0.0009 |
Maternal post-partum month (median, IQR, range) | 9 (9, 21) | 6 (7, 18) | 9 (6.75, 17) | 6 (10, 17) | 0.0025;0.28 |
Infant age in months (median, IQR, range) | 9 (6, 17) | 6 (7, 17) | 9 (6, 17) | 9 (9, 17) | 0.096; >0.99 |
Gestational age (median, IQR, range) | 38 (0,10) | 38 (1.25, 3) | 38 (0,3) | 38 (1, 10) | 0.28; >0.99 |
Time-since-weaning (median, IQR, range) | 6 (9, 16) | 3 (6.25, 15) | 6 (9, 16) | 3 (9, 15) | 0.10; 0.81 |
Antibiotics usage, no. (%) | 27 (48%) | 29 (51%) | 27 (10%) | 29 (62%) | 0.0004; 0.883 |
Vaginal delivery no. (%) | 94 (91%) | 30 (100%) | 65 (93%) | 59 (94%) | 0.21; >0.99 |
Maternal CD4 count (median, IQR, range) | 546 (252.5, 1084) | 621 (400, 847) | NA | 591 (316.8, 1084) | 0.56; NA |
DNA virome and bacterial microbiome differences between mothers and their infants
We performed metagenomic sequencing for bacteria and DNA viruses on 306 stool samples collected longitudinally from the study participants (Fig. 1B). One maternal sample for the bacterial analysis and 3 infant samples for the DNA viral analysis were excluded due to low sequencing read depth (< 200K reads). An average of 14,147,662 ± 13,612,749 bacterial metagenomic reads per sample and 4,498,593 ± 3,826,239 DNA viral metagenomic reads per sample were analyzed.
Previous studies have shown that adult and infant gut microbiomes differ substantially27, 43–46. Using linear mixed effects models (LME) and PERMANOVA to compare microbiome diversity, we found that maternal and infant samples had significantly different bacterial richness (p = 0.0002) and alpha diversity (p < 0.0001), but not beta diversity (PERMANOVA, p = 0.85). DNA virome richness also differed between mothers and infants (p = 0.006), while virome alpha and beta diversity did not (p = 0.13; p = 0.62). When comparing all infant and maternal sample Bray-Curtis distances without controlling for time, we found significant differences for bacterial and viral beta diversity (p < 0.001, Fig. 2A-B). Microbiome Bray-Curtis dissimilarity distance between mothers and infants decreased over time regardless of SARS-CoV-2 or maternal HIV status (p < 0.0001), indicating infants were converging toward an adult-like microbiome configuration that is resilient to these viral infections (Fig. 2C, Supplementary Fig. 1A-B). In contrast, the DNA virome beta diversity between mother and infants did not change over time (p = 0.23) and maintained high dissimilarity throughout the early life period, regardless of SARS-CoV-2 and maternal HIV status (median maternal-infant dissimilarity = 0.99, Fig. 2D, Supplementary Fig. 1C-D).
To better understand the community structures, we applied k-means clustering that derived 5 distinct gut bacterial community profiles (Fig. 3A). Community groups 2 (composed of Faecalibacterium prausnitzii (22%) and Prevotella copri (15%) abundance) and group 4 (composed of Faecalibacterium prausnitzii (12%), Bifidobacterium adolescentis (11%), Collinsella aerofaciens (10%) abundance) had significantly more maternal samples than infants in comparison to other community groups (Supplementary Fig. 1E-F, p-values between 0.005 and < 0.0001). Likewise, 2274 active metabolic pathways were differentiated between mothers and infants (Supplementary Fig. 1G). Due to high interpersonal variation in the viromes, k-means clustering could not identify distinct virome community clusters. In general, mothers had significantly higher median abundance of Microviridae, Inoviridae, and Suoliviridae (65%, 1.1% and 0.2% respectively) than infants (4.2%, 0% ,0.03% respectively; p < 0.0001) (Fig. 3B). Infants had significantly higher median abundance of Anelloviridae and Genomoviridae (6% and 2% respectively) compared to mothers (0.01% and 0.4% respectively; p < 0.0001).
Multivariable analyses identified 30 bacteria taxa which differentiated mother (e.g., Bifidobacterium adolescentis and several Ruminococccus species) from infants (e.g., Bifidobacterium bifidum, Escherichia coli, and several Streptococcus and Veillonella species) (Fig. 3C). We also identified 11 viral contigs (mostly Anelloviridae) differentially associated with infants (Fig. 3D). Within just infants, we found several bacterial taxa and viral contigs associated with time and SARS-CoV-2 infection (Supplementary Fig. 1H-K). Longibacterium sp. KGMB06250, Faecalibacillus intestinalis and several Genomoviridae contigs were more abundant in SARS-CoV-2 infected infants (Supplementary Fig. 1J-K). Taken together, these findings indicate that robust microbiome and virome signatures differentiated maternal samples from infants.
Assessing changes in microbiome and virome trajectory after SARS-CoV-2 infection
Given the dynamic nature of the microbiome, we tested the hypothesis that SARS-CoV-2 infection affected microbiome trajectory over time by LME modeling. Since a subset of women in this study were living with HIV, we assessed if there was an interaction between HIV and SARS-CoV-2. However, because there was no significant interaction between HIV status and SARS-CoV-2 infection on microbiome and virome richness and diversity (Supplementary Fig. 1L), women were not stratified by HIV status in further analyses. Interaction between HIV exposure and SARS-CoV-2 could not be assessed for infants due to the limited number of HEU infants with SARS-CoV-2 infection (n = 7).
We designed interaction LME models to test if the trajectory of microbiome was altered due to SARS-CoV-2 infection over time in women. We found the trajectory changes in bacterial alpha diversity trended differently after infection (post-SARS-CoV-2) compared to SARS-CoV-2 negative women throughout study follow-up (controls), though this did not reach statistical significance (alpha diversity p = 0.082, Fig. 4A). We then compared samples prior to infection (pre-SARS-CoV-2) and after infection (post-SARS-CoV-2) by including an interaction term between SARS-CoV-2 infection and time since infection. Alpha diversity increased over post-partum time before SARS-CoV-2 infection, but then markedly reversed to decrease significantly after infection (p = 0.015; Fig. 4B). Changes in beta diversity and bacterial richness over time were not associated with SARS-CoV-2 infection (Fig. 4C, Supplementary Fig. 2A). To assess the significance of individuals variable factors in women, we removed the interaction and found post-partum time and antibiotic use to be significant for changes in beta diversity across all women (post-partum time p = 0.020, antibiotic use p = 0.030, Supplementary Fig. 2B-C).
When assessing whether SARS-CoV-2 infection changed infants’ bacterial microbiome trajectory, we found that the richness, alpha diversity and beta diversity of SARS-CoV-2 seropositive infants were not statistically significantly different compared to controls when measuring time by month of life (alpha diversity p = 0.10, richness p = 0.40, weighted beta diversity p = 0.14; Fig. 4D-E, Supplementary Fig. 2D), or by time since weaning (richness p = 0.63, alpha diversity p = 0.20, weighted beta diversity p = 0.093). Because changes in the gut microbiome are associated with early life development47, 48, we also analyzed infant samples without modeling for SARS-CoV-2 interactions. Bacterial richness, alpha diversity and beta diversity were associated with changes both by month-of-life and time since weaning (richness p < 0.0001, alpha diversity p < 0.0001, beta diversity by month-of-life p < 0.001, beta diversity by time-since-weaning p < 0.001; Supplementary Fig. 2E). These findings strongly suggest SARS-CoV-2 infection in microbiome changes over time in adult women but not in infants, potentially due to stronger drivers of microbiome maturation associated with infant development.
Virome alpha diversity, beta diversity and richness over time were generally not associated with SARS-CoV-2 infection in either SARS-CoV-2 seropositive or seronegative women (alpha diversity p = 0.67, beta diversity p = 0.95, richness p = 0.25, Fig. 4F-G, Supplementary Fig. 2F). However, all women had increasing richness over time postpartum (p = 0.017), and women that remained seronegative for SARS-CoV-2 had a higher alpha diversity compared to women who were ever seropositive for SARS-CoV-2 (p = 0.046, Supplementary Fig. 2F-G). Infant models with interaction between SARS-CoV-2 status and month-of-life showed no significant associations between virome richness (p = 0.28), alpha diversity (0.075) and beta diversity (p = 0.85, Fig. 4I-J, Supplementary Fig. 2H). Similar to the gut microbiome developmental dynamics, when omitting the interaction from the model, infant virome richness, alpha diversity and beta diversity were significantly associated with changes by month-of-life (richness p = 0.010, alpha diversity p = 0.034, beta diversity p = 0.001, Supplementary Fig. 2I). These findings are consistent with the dynamic changes in the gut virome during early infant development47, 49. Together, these results indicate that the SARS-CoV-2 infection does not substantially alter the gut virome.
Assessing short term impacts of SARS-CoV-2 infection
We next considered whether SARS-CoV-2 might have a more pronounced impact on the microbiome during or immediately after infection. To test this, we compared only the first SARS-CoV-2 seropositive samples to all seronegative samples (including samples prior to seroconversion and all samples from seronegative, uninfected controls). This approach allowed us to identify changes immediately post-infection that might be masked by the overall microbiome stability over time (i.e., prior trajectory analysis) and is more comparable to existing cross-sectional studies. Among women, we did not find significant differences between SARS-CoV-2 seronegative and the first seropositive samples in bacterial richness (p = 0.94), alpha diversity (p = 0.22), or weighted beta diversity (p = 0.095; Fig. 5A-B, Supplementary Fig. 3A). Findings were similar for infants (richness p = 0.93, alpha diversity p = 0.61, month-of-life weighted beta diversity p = 0.76, time-since-weaning weighted beta diversity p = 0.48, Fig. 5C-D, Supplementary Fig. 3B). There were also no significant differences between women’s seronegative and first seropositive samples in viral DNA richness (p = 0.70), alpha diversity (p = 0.29), or weighted beta diversity (p = 0.49, Fig. 5E-F, Supplementary Fig. 3C). Likewise for infants, when modeling for month-of-life, viral richness (p = 0.066), alpha diversity (p = 0.28), and beta diversity (p = 0.95), of their first SARS-CoV-2 positive samples were not significantly different (Fig. 5G-H, Supplementary Fig. 3D). Thus, there was no evidence that SARS-CoV-2 infection significantly alters the gut microbiome and virome in mothers and infants immediately after infection.