Global and local trajectories of infant gut microbiota development in Bangladesh across the COVID-19 pandemic

Study design and sample selection

This study was conducted in Kamalapur, an urban area located in southeast Dhaka characterized by high population density and low socioeconomic status. Thirty healthy pregnant women were enrolled at the International Centre for Diarrheal Disease Research, Bangladesh (icddr,b), and their newborns were followed from birth over a 2-year period (children were born between 26 October 2018 and 01 December 2018, and sampling occurred between 26 October 2018 and 24 November 2020). Infant fecal material collection occurred three times/week during the first 6 months, twice/week in the following 6 months, and once/week throughout the second year. Additionally, one fecal sample per mother was collected shortly after birth. Trained field staff conducted household visits for sample collection and transport to the laboratory, during which detailed metadata questionnaires were completed. These questionnaires were designed to collect longitudinal infant-specific metadata, including demographic characteristics (birth mode, birth place, and household composition), medical history (vaccination status, antibiotic/medication exposure, and disease occurrence), and dietary practices (breastfeeding, formula feeding, and solid food introduction) (Table S1). Further details on metadata collection are provided in Supplemental methods.

Due to incomplete sampling and missing metadata, only 20 of the 30 enrolled infants were included in this study. In total, 1,002 samples were selected for this study, from which 18 were maternal and 984 originated from 20 infants. To explore the longitudinal development of the healthy gut microbiota, we selected approximately 33 samples per infant at predefined time points without reports for antibiotic intake or gastrointestinal disease (~21 samples during the first year, ~12 during the second). For this purpose, 486 (out of 984) samples were selected (i.e., samples representing healthy time points or HTP samples). To assess the effect of external events on the early gut microbiota development, we selected 20–30 additional samples per infant associated with specific life events during the first year of life. For this, 498 (out of 984) ad hoc collected samples (i.e., collected on an ad hoc basis, ADHC samples) were selected (Fig. S2). For the comparative analysis with the Belgian infant population, we included 265 samples from the BABEL cohort (10) consisting of 7 infants. A more detailed description of the study design and sample selection can be found in Supplemental methods.

Fecal microbiota profiling and moisture content measurement

Frozen samples were aliquoted (150–200 mg) at the Rega Institute (Belgium) without thawing and further maintained at −80°C until DNA extraction. Bacterial DNA extraction followed a modification of a published protocol (27), and 16S rRNA amplification and sequencing were performed as previously described (28). Sequencing was performed on 7 runs using the Illumina MiSeq platform (2 × 250 paired-end reads). Full methodological details are provided in Supplemental methods.

Maternal fecal samples (n = 18) were further processed to estimate their moisture percentage by freeze drying using the Alpha 1–4 LSCplus (Martin Christ) freeze dryer and comparing weights before and after lyophilization.

Sequencing data preprocessing

Following demultiplexing with sdm in the LotuS pipeline (29), with no mismatches permitted, the fastq sequences were analyzed separately for each sample using the DADA2 pipeline (v.1.6) (30). Primer sequences and the first 10 subsequent nucleotides were trimmed. Taxonomy was assigned using the formatted RDP (Ribosomal Database Project) training set "rdp_train_set_19” (31) after merging paired sequences and eliminating chimeras. The decontam R package with the prevalence method was used for identification and removal of contaminant sequences (32). In total, 224 bacterial and 2 archaeal amplicon sequence variants (ASVs) were identified in our data set. Samples were rarified to 15,086 reads, and sequences with a relative abundance smaller than 0.0001 across all samples were removed. To ensure a meaningful comparative analysis of this cohort with the previously studied Belgian cohort (10), the Belgian data were re-processed (7 infants with samples up to 365 days of age) using the approach described here. In our updated analysis, the Belgian cohort exhibited two main successional stages of microbiota development in contrast to the three stages reported in the original publication (10). The reduced number observed here likely reflects comparable underlying dynamics, with differences attributed, among other factors, to changes in data processing and taxonomical updates in the available databases since the original publication (Fig. S4).

Statistical analyses

All statistical analyses and visualization were conducted in R (http://www.R-project.org, v4.3.2), using the following packages: ggplot2 (33) (v3.4.4), vegan (34) (v2.6-4), phyloseq (35) (v1.42.0), microbiome (36) (v1.20.0), and microViz (37) (v0.11.0). To analyze group differences and temporal trends in alpha diversity, linear mixed-effects models were used with the lme4 (38) (v1.1-37) and lsmeans (39) (v2.30-2) R packages. These models account for the non-independence of repeated measurements from the same infants by including subject-level random intercepts. P-values were computed using likelihood ratio tests for overall effects and Kenward-Roger adjusted Wald tests for pairwise group comparisons. To compare median differences between independent or paired groups of continuous variables, we used the Wilcoxon rank-sum (or Mann-Whitney U) and Wilcoxon signed-rank tests for two groups. To evaluate the association between genera prevalence patterns across the Bangladeshi and the Belgian infant cohorts, a Spearman rank correlation analysis was computed. Differences in microbial community composition (beta diversity) between cohorts were assessed using permutational multivariate analysis of variance (PERMANOVA) with the adonis2 function in the vegan R package (Bray-Curtis dissimilarity, 999 permutations). Differences in Segatella prevalence between sample groups across gut microbiota maturation stages were assessed using Fisher’s exact tests while Pearson’s chi-squared tests were conducted on contingency tables that were created to assess the association between different covariates and (i) the occurrence of maturation setbacks and (ii) the presence of samples with high Segatella abundance. Where applicable, multiple testing adjustments were made using the Benjamini-Hochberg method, with a false discovery rate (FDR) threshold set at 0.05.

Additional statistical analyses were performed for (i) the identification of bacterial communities (Dirichlet multinomial mixture modeling [DMM]), (ii) determination of maturation setbacks, (iii) order of appearance of the most abundant bacterial genera, (iv) diversity metrics and multivariate analysis of metadata effects on the microbiota, and (v) analysis of samples with high Segatella abundance. Detailed descriptions of these analyses are provided in Supplemental methods.

The healthy Bangladeshi infant gut undergoes bacterial colonization in three distinct developmental stages

Using 16S rRNA data, we explored gut microbiota maturation dynamics in 20 Bangladeshi infants during the first 2 years of life. We first assessed whether the data set clustered into distinct community types, using the Dirichlet multinomial mixture (DMM) modeling approach. This analysis revealed the presence of three compositionally distinct clusters in which Bangladeshi infant samples were structured throughout the first two years of life, designated as clusters A, B, and C (Fig. 1a). Following the definition previously described by Beller and colleagues (10), these clusters are also referred to as gut microbiota maturation (GMM) stages. Specifically, during the first months of life, infant samples predominantly clustered within stage A, gradually transitioning to stage B, and ultimately reaching stage C around the beginning of the second year of life (Fig. 1b and Fig. S6a). When analyzed alongside the infant cohort, maternal samples mainly clustered within stage C (Fig. 1a), supporting the notion that this transition represents gut microbiota maturation toward an adult-like profile.

At the genus level, our analysis showed an overall increase in bacterial diversity across maturation stages and time (Fig. 1c and d and Fig. S6b). Bifidobacterium dominated the Bangladeshi infant gut throughout the first two years of life. Facultative anaerobes, such as Escherichia/Shigella and Streptococcus, were also highly abundant in maturation stage A, but unlike Bifidobacterium, these taxa declined in stages B and C (Fig. 1c). A similar trend was observed in Belgian infants (10), suggesting that these genera may be indicative of a conserved early-life gut environment, with higher oxygen availability (40). Overall, the Bangladeshi infant gut is initially colonized by opportunistic bacteria and facultative anaerobes, which is consistent with the current understanding of early-life microbial colonization (12, 41–43). Concurrently, we observed the early establishment of Bifidobacterium which could be attributed to breastfeeding and the fact that most infants were vaginally delivered (n = 14; Table S1) (12, 44, 45). Introduction of solid foods in infancy leads to an increase in bacterial richness and the gradual emergence of more obligate anaerobes, such as Dorea and Faecalibacterium, capable of metabolizing structurally diverse dietary compounds. Notably, Bacteroides had a relatively low abundance in Bangladeshi infants and was nearly absent in their mothers (1/18 individuals with a relative Bacteroides abundance of >0.5%; Fig. 1c and Fig. S7), a key difference from Western populations.

Bangladesh is a country with high breastfeeding rates (46, 47), and in this cohort, all infants were breastfed at similar frequencies. Most infants were exclusively breastfed for approximately the first 6 months, after which solid foods were introduced alongside breastfeeding. Prolonged breastfeeding likely contributed to the dominance of Bifidobacterium in the gut microbiota across all developmental stages. Although 16S rRNA V4 gene marker sequencing generally lacks species- and strain-level resolution, the most abundant Bifidobacterium ASV in our data set was classified as B. longum when compared to RDP taxonomy, indicating that at least some of the dominant ASVs correspond to this species, in line with observations from other Bangladeshi infant cohorts (24, 48). As this species is capable of degrading both HMOs and complex polysaccharides when breastfeeding is combined with solid food (24) and given that dietary diversity has been associated with bifidobacteria abundance (49), this may explain the maintained high relative abundance of the genus following the introduction of solid food. This compositional pattern aligns with a recent study reporting that, during early infancy, microbiota in many Asian populations, including Bangladesh, are enriched in Bifidobacterium, reflecting early-stage development (50).

Although Segatella was initially absent in early life, its relative abundance increased as the gut microbiota matured toward GMM stage C (Fig. 1c), mirroring trends in non-Western and less industrialized infant and adult populations, where species of the Prevotellaceae family dominate the gut microbiome (13, 16, 17, 19, 24). Compared to infants, members of the Prevotellaceae family were enriched in several maternal fecal samples (15/18 individuals harbored Segatella alone or in combination with Leyella and/or Prevotellamassilia; mean combined relative abundance: 30.5%) (Fig. S7). Analysis of the maternal samples revealed a high water content, averaging 88.6% (Fig. S7). High fecal moisture reflects looser stool consistency, which corresponds to higher Bristol Stool Scale (BSS) scores and can serve as a proxy for faster colonic transit time (51). Elevated stool water content has been associated with the Prevotella enterotype (as described by Arumugam et al. [19]) in a Belgian adult cohort (52). This enterotype linked to fiber-rich diets common in non-Western settings (53) is characterized by a microbial community in which Prevotella (now Segatella) plays a major role in carbohydrate degradation (54). Consumption of non-fermentable fiber increases stool water content and plasticity, contributing to increased water retention in the gut (52, 55).

Alpha diversity differed significantly across GMM maturation stages (linear mixed-effects models (LMM), P < 0.001), with a stepwise increase in microbiota richness and diversity from stage A to B (LMM, P< 0.001) and from B to C (LMM, P < 0.001), supporting a structured, sequential maturation process (Fig. 1d; Table S3).

Despite occasional regressions to earlier maturation stages (Fig. 1e)—potentially due to illness, medication, or other factors—we observed a conserved and structured trajectory of microbiota maturation over time in three distinct GMM stages across infants (Fig. 1e, n = 20, Kendall test, Kendall’s corrected w = 0.91, and P = 9.9e−5). This finding highlights the resilience of the Bangladeshi infant gut microbiome in its ability to maintain a structured and predictable developmental pattern despite occasional regressions.

Bangladeshi and Belgian infant microbiota show similar developmental patterns with distinct microbial signatures

To investigate gut microbiota maturation patterns across different geographical and lifestyle settings, we compared the gut microbiota maturation patterns of the Bangladeshi infant cohort (BBGUT) with a previously studied Belgian infant cohort (BABEL) (10) considering only samples representing healthy time points for both populations. The Bangladeshi and the Belgian data sets were generated and processed using the same analytical approach for comparability, with analysis limited to the first year of life, as it is the data available for the Belgian cohort.

First, we examined differences in the prevalence of the most abundant bacterial genera in the two cohorts (Fig. 2a). Our analysis revealed that only about 33% of the most abundant genera (5 out of 15) detected in each cohort exhibited similar prevalence trends over the first year of life in both cohorts, supported by a significant positive correlation in mean prevalence over time (Fig. 2b). While some of these bacteria followed similar patterns in acquisition and persistence in both cohorts during early life, the relative abundance dynamics differed over time for most genera (Fig. S8a and b). Cohort differences included the presence of Segatella, found exclusively in Bangladeshi infants (Fig. 2a, BBGUT), supporting the dominance of Prevotellaceae members in non-Western regions over Bacteroides (13, 53, 56). Certain lactic acid bacteria (Limosilactobacillus, Ligilactobacillus) were also exclusive to Bangladesh (Fig. 2a, BBGUT), whereas bacteria taxonomically assigned to Parabacteroides, Thomasclavelia, and Sutterella genera were predominantly found in Belgium (Fig. 2a, BABEL). Additionally, Streptococcus was more prevalent and abundant in Bangladeshi infants (Fig. 2a, BBGUT; Fig. S8a) together with a richer community of Bacillota genera, particularly lactic acid bacteria such as Lactobacillus. Elevated Bacillota diversity has been previously reported in Bangladeshi children and infants (16, 57), aligning with our observations. These differences suggest variations in microbial signatures due to geographical factors and the bio-cultural environment, such as divergent breastfeeding practices (e.g., continuation of breastfeeding alongside solid food in Bangladeshi infants, which was not the case for Belgian infants). Beta diversity analysis further supported the differences in community composition between the two cohorts (permutational multivariate analysis of variance, Adonis test, R² = 0.059, P = 0.001; Table S4).

We next assessed alpha diversity patterns to determine whether microbiota diversity (genus-level bacterial diversity) evolved similarly in both cohorts and found comparable trajectories over time (Fig. 2c; Table S5). Bacterial GMM stages aligned well with infant age across the first year (Fig. 2c, bottom), progressing from A to A–B in BABEL and from A to A–B to A–B–C in BBGUT. To determine whether these observations resulted from age-binning effects, we performed a month-by-month analysis of alpha diversity (Fig. S9). Differences in alpha diversity were statistically significant only during the first 10 days of life (Shannon diversity and observed richness, Wilcoxon rank-sum tests, adjusted P < 0.01) while subsequent time points showed similar patterns, supporting our initial conclusion that alpha diversity follows a consistent trajectory over time.

To further determine if infant gut microbiota development follows a consistent pattern of bacterial colonization in two geographically distinct populations, we analyzed the temporal sequence, or the order of appearance, of the 15 most dominant bacterial genera during the first year of life for both cohorts (Fig. 2d). We defined genera as appearing when their relative abundance exceeded 0.5% and were ranked accordingly for each infant in each cohort (Fig. S10). Individual analysis revealed that the order of appearance was highly consistent across individuals for both BBGUT (n = 20, Kendall test, corrected Kendall’s w = 0.59, P = 9.99e−05) and BABEL cohorts (n = 7, Kendall test, corrected Kendall’s w = 0.49, P = 9.99e−05), consistent with previous reports (10).

Despite cohort-specific compositional differences, we observed a significant positive correlation in the sequence of bacterial appearance during early life when comparing infants from Belgium and Bangladesh (Fig. 2d). These findings suggest a conserved pattern of microbial succession across geographically and culturally distinct infant populations (Kendall test, corrected Kendall’s w = 0.79, P = 0.016). We found that lower ranking colonizers, mostly facultative anaerobes such as Pseudomonadota members, appeared early in life, co-occurring with Actinomycetota, and followed by the later appearance of more strictly anaerobic bacteria, such as Bacillota (Fig. S10). While the general order of appearance is similar between the two cohorts, we observed notable differences, including a simultaneous early appearance of most genera in the BBGUT cohort vs a more gradual succession over time in BABEL (detailed overview in Fig. S10).

Disease-related events are associated with maturation setbacks

We analyzed a range of demographic factors (delivery mode/place, gestational age, hygiene levels, sibling presence), infant-specific variables (infant ID, age), and medical history (vaccination, gastrointestinal disease occurrence, antibiotic use), to identify the key covariates associated with compositional variation in the Bangladeshi cohort across all samples (n =984). Overall, the cumulative explanatory power of the non-redundant variables in our multivariate distance-based redundancy analysis (dbRDA) collectively explained approximately 24% of the variation in the microbial community composition (Fig. 3a; Table S6). Among significant covariates, infant identity, reflecting interindividual variation, had the highest explanatory power (n = 984, dbRDA with Bray-Curtis dissimilarity, R² =0.12, adjusted P < 0.01), followed by age (R² = 0.09, adjusted P < 0.01), diet (R² = 0.013, adjusted P < 0.01), diarrhea occurrence (R² = 0.006, adjusted P < 0.01), and antibiotic use (R² = 0.006, adjusted P < 0.05) (Fig. 3a; Table S6). Despite our findings being broadly consistent with previous observations in Belgian infants (10), medical history-related variables, such as diarrhea and antibiotic use, contributed more substantially to the explained variation in BBGUT, likely reflecting higher disease and medication exposure compared to BABEL. Whereas the tested covariates explained nearly 40% of variance in Belgian infants (10), they accounted for only 24% in the Bangladeshi cohort.

Additionally, the contribution of different metadata covariates to the variation of the most abundant bacterial genera of the BBGUT cohort (n = 984, multivariate stepwise distance-based redundancy analysis using Euclidean distance on composition) was also evaluated. Here, we found age to be the variable with the highest explanatory power in over 50% of the genera (8 out of 15) (Fig. 3b; Table S7). For only four taxa, namely, Megasphaera, Segatella, Veillonella, and Limosilactobacillus, diet (i.e., introduction to solid food) explained most variation in their abundance patterns. Interestingly, Streptococcus, one of the common bacterial genera in healthy early-life samples, was associated with diarrhea episodes.

Several occasional regressions (i.e., setbacks) to earlier GMM stages were observed during the gut microbiota maturation of the BBGUT cohort despite the conserved process generally followed in healthy samples (Fig. 1e). To better understand these fluctuations, we assessed the influence of external factors on gut microbiota maturation by considering both the setback changes and the level of maturation at the time of the change. The latter was quantified by calculating a maturation score (see Supplemental methods) considering the presence or absence of key bacterial genera. Only samples that exhibited a setback to a lower maturation stage, along with a corresponding decrease in maturation score, were included in the analysis (n = 31, across all samples; Fig. S11). This approach allowed for a more precise evaluation of external factors influencing gut microbiota development (example of infant 15 in Fig. 3c), revealing 31 setback instances throughout our cohort. To further assess which factors could explain these disruptions, we evaluated the metadata of these 31 samples and explored their associated health status. Instances with an observed maturation setback comprised only 3.1% of the total healthy time points (15/486) and 12.6% of the total disease-associated events (16/127) (Fig. 3d; Table S8). A chi-squared test confirmed that the occurrence of maturation setbacks is significantly associated with health status [χ² (1) =17.045, P = 3.65e−05] and positively associated with disease (Table S8). This positive association between maturation setbacks and disease-associated samples suggests a link between disease occurrence and maturation disruption. Notably, we observed a maturation setback for around 24% of the total disease events for which both antibiotic administration and gastrointestinal signs were reported at the same time (9/37). However, a specific association with these types of events and maturation setbacks was not confirmed statistically. Samples associated with setback events also exhibited lower microbial alpha diversity compared to samples from healthy timepoints preceding the setbacks (Fig. S12).

Changes in the development of a healthy gut microbiota coincide with the first COVID-19 lockdown period in Bangladesh

While analyzing the monthly progression of gut microbiota maturation in the BBGUT cohort (considering only HTP samples), we noticed a qualitative decrease in the relative abundance of some taxa during the second year (Fig. 4a). To further assess these shifts, we compared period-specific changes in bacterial relative abundances and identified several genera with significantly greater fluctuations after 16 months of age (Fig. S13; Table S9). We, therefore, divided the second year into two periods, months 13–16 and months 17–24 of infant age. As all infants were born within a 36-day window, this boundary also coincides with the onset of the national COVID-19 lockdown in Bangladesh (26 March 2020; Fig. S14), meaning that infant age and calendar period cannot be distinguished in this cohort. For this analysis, we focused on samples representing healthy time points due to a limited number of samples associated with disease in year 2 (n = 17/175 samples). Comparing the two periods revealed a statistically significant decline in the relative abundance of several common bacteria, including Segatella, Bacteroides, Veillonella, and Megasphaera, observed during months 17–24 (Fig. 4b; Table S10). Only Blautia, a genus suggested to have probiotic potential (58) and Streptococcus, showed an increase in relative abundance during this time. Although changes in these bacterial taxa were observed, the current study design does not allow us to determine whether they were driven by pandemic-related factors.

Next, we compared maturation scores between infants before and after 16 months of age (calculated based on HTP samples), accounting for bacterial rankings and presence across monthly samples throughout their second year of life as previously discussed (Fig. 3b). Although there was no significant overall decrease in the maturation score during months 17–24 (Fig. 4c), and alpha diversity indices similarly showed no significant change between the two age periods (linear mixed model, P = 0.44 [Shannon index], P = 0.43 [observed richness]; Table S3), individual responses varied, with some infants showing signs of plateaued or reduced maturation. While these results do not support a consistent, cohort-wide delay in microbiota maturation (Fig. 4b), we identified temporal perturbations during this period, as reflected by changes in the relative abundance of specific bacterial taxa (Fig. 4c).

The role of Segatella bacteria in the gut microbiota maturation of Bangladeshi infants remains unclear

Temporal changes in Segatella abundance in the gut microbiota of BBGUT infants were observed, with notable fluctuations in healthy samples (Fig. 4a). Segatella was first detected toward the end of the first year of life, but contrary to the stable prevalence typically seen in other Bangladeshi (24) and non-Western populations (17), its average abundance declined sharply between months 17 and 24 (Fig. 4a and b). This observation prompted us to further investigate the development of Segatella, as it was the dominant genus within the Prevotellaceae family present in this cohort and one of the most abundant bacterial genera in maternal samples (Fig. S7).

Further analysis revealed that the observed increase in Segatella prevalence was primarily driven by disease-associated samples in maturation stage C (Fisher’s exact test, P < 0.001; Fig. S15a and b). Thus, we focused on a subset of 33 samples exhibiting considerably high Segatella abundances (≥25%) during the first 2 years of life (Fig. 5a). We first examined the proportion of these high Segatella abundance (HSA) samples over time and found that they began to appear around month 8 (Fig. 5a). Although these samples were more prevalent in HTPs overall, Segatella-rich samples appeared earlier in samples associated with disease (Fig. 5a). In the first year of life, the first detection occurred during disease-related time points for 45% of infants (9 out of 20), compared to only 25% of infants (4 out of 20) whose first Segatella-rich samples appeared during HTPs (Fig. 5b).

When categorizing HSA samples by health status, they represented 4% of all healthy samples and 11% of all disease samples in the cohort (Fig. 5c; Table S11). Notably, within the disease-related group, most HSA samples were associated with antibiotic administration (Fig. 5c). A chi-squared test confirmed a significant association between the health categories and the occurrence of HSA samples [χ² (1) = 8.66, P = 0.003] (Table S11). Specifically, we observed a positive association between disease-related samples and elevated Segatella abundance, suggesting a link between Segatella spikes and disease occurrence.

Following this observation, we explored whether certain Segatella ASVs were enriched in disease-associated samples compared to those linked to healthy time points and found no significant differences (Fig. S16). To gain deeper insight into the taxonomical classification of Segatella sequences in this study, using the latest RDP taxonomy (training set 19), Segatella ASVs were identified as Segatella copri and Segatella hominis (Table S12). Within-individual comparison of HSA event samples against the closest preceding healthy time point (n = 33 paired events from 16 infants) identified Bifidobacterium as the only genus with a significant reduction in relative abundance during HSA events (Wilcoxon signed-rank test, adjusted P < 0.001; Fig. S17).

Conclusion

This study provides a densely sampled longitudinal characterization of gut microbiota development in an LMIC setting, offering detailed insights into the maturation of the infant gut microbiome in Bangladesh. By comparing gut microbiota development to a Western infant cohort, we identified both conserved trajectories across geographies and cohort-specific compositional differences reflecting the influence of lifestyle and the environment. We are the first to characterize early-life gut microbiota development in a Bangladeshi infant population during COVID-19, providing insights into microbiota dynamics under these unique circumstances and highlighting the need for appropriate study designs and continued longitudinal monitoring to evaluate potential long-term impacts. In addition, we identified a link between Segatella and disease, adding to the growing and sometimes conflicting literature on its role in human health. Together, these findings offer new perspectives on gut microbiota development across diverse populations and emphasize the significance of environmental, cultural, and global disruption factors in shaping the infant gut microbiome. Looking ahead, extending these observations into later childhood and larger cohorts as well as integrating higher-resolution multi-omics approaches will be critical to deepen our understanding and to guide interventions that support infant health in LMICs.