Improving immune‐related health outcomes post‐cesarean birth with a gut microbiome‐based program: A randomized controlled trial

Abstract

Background

Infants born via Cesarean section (C‐section) often have a distinct gut microbiome and higher risks of atopic and immune‐related conditions than vaginally delivered infants. We evaluated whether a microbiome‐based program could shift gut microbiome composition and improve microbiome‐associated health outcomes in C‐section born infants.

Methods

This open‐label, randomized, controlled trial included full‐term C‐section‐born infants aged 0–3 months, randomized to an intervention (n = 25) or control arm (n = 29). Over 6 months, the intervention arm received two microbiome reports, personalized recommendations based on their microbiome, educational materials, and coaching calls focused on microbiome health. Parents reported health conditions via surveys. Primary outcome: Difference between study arms in relative abundance of key gut microbiome taxa and functional genes. Other outcomes: Changes in a C‐section index—a taxonomy‐based metric comparing C‐section‐associated taxa to vaginally‐associated taxa—and prevalence of atopic conditions.

Results

Compared to controls, the intervention arm had higher Bifidobacterium (p = .025, q = .121) and higher abundance of genes associated with human milk oligosaccharide degradation (e.g., α‐L‐fucosidase, p = .019, q = .046) at timepoint 2. In the intervention arm, the C‐section index decreased to a level similar to vaginally born infants (p = .807, q = .807). At the end of the intervention, atopic dermatitis prevalence was lower in the intervention arm than in controls (odds ratio, 0.17 [95% CI, 0.023–0.723], p = .031).

Conclusion

A personalized microbiome‐based program can modulate the gut microbiome of C‐section‐born infants and may reduce the risk of atopic conditions (ClinicalTrials.gov: NCT06424691).

Key message.

Cesarean‐born infants often have a distinct gut microbiome and a higher risk of atopic conditions, highlighting the need for early interventions. This study evaluated a personalized microbiome‐based program for these infants, showing modulation of gut microbiome structure and functionality, and lower odds of developing atopic dermatitis. These findings suggest that microbiome‐targeted strategies may help mitigate immune‐related health risks in Cesarean‐born infants, offering a novel approach to early life health optimization.

1. INTRODUCTION

The first 1000 days of life are viewed as a crucial window of immune development, with the gut microbiome playing a key role. ¹ , ² , ³ , ⁴ One major factor shaping gut microbiome composition is delivery mode. ⁵ , ⁶ , ⁷ Vaginally‐born infants typically exhibit a higher abundance of Bifidobacterium, Bacteroides, and Parabacteroides compared to those born via C‐section. ⁵ These first colonizers often drive human milk oligosaccharide (HMO) degradation, resulting in short‐chain fatty acid (SCFA) production and colonization resistance. ⁸ , ⁹ In contrast, C‐section‐born infants are often colonized by skin and hospital‐associated bacteria, such as Staphylococcus, ⁷ Enterococcus, Klebsiella, and Clostridium species. ⁴ , ⁵ , ⁶ These often lack the ability to degrade HMOs or produce SCFAs and are more likely to harbor antibiotic resistance genes. ¹⁰

A C‐section‐associated microbiome signature can persist in children up to the age of four ⁶ , ¹¹ and has been linked to an increased risk of type 1 diabetes, ¹² obesity, ¹³ , ¹⁴ and atopic march. ¹⁵ Atopic march refers to the progression from atopic dermatitis (AD) to food allergies, rhinitis, and asthma, ³ , ¹⁵ , ¹⁶ which can lead to chronic allergic and inflammatory diseases later in life. ¹⁷ Stokholm et al. (2020) found that children who retained a C‐section signature at 1 year had a threefold higher risk of developing asthma by age six. ¹⁸ This stresses the need for strategies to modulate microbiome composition in infants born via C‐section.

We aimed to evaluate the effectiveness of a gut microbiome‐based program in improving gut health and reducing the risk of health conditions in C‐section‐delivered infants. We hypothesized that participation in the program would shift the composition and functionality of the infant gut microbiome and decrease microbiome signatures of C‐section. Additionally, we examined whether a microbiome‐based intervention reduced the prevalence of parent‐reported infant health conditions such as atopic dermatitis.

2. METHODS

2.1. Study design

We conducted an open label, stratified, randomized [1:1], controlled trial to test the effects of a microbiome‐based intervention versus self‐directed control. Our trial was registered with ClinicalTrials.gov (NCT06424691). Participants were recruited between August 2023 and November 2023. The trial was completed as planned, with data collection finalized in May 2024. The full trial protocol is available in Appendix S1, with no significant changes after commencement.

2.2. Setting and population

Major inclusion criteria included: infants born at full term (≥36 weeks' gestation), delivered via C‐section, and between 0 and 3 months of age at the time of enrollment. Additionally, the parents needed to be residents of the United States with access to US postal services.

Exclusion criteria included infants who had previously received probiotic supplements, multiple birth infants, and infants with pre‐existing health conditions with the exception of eczema and rashes. Additional details in Appendix S2.

2.3. Participant flow

Participants were randomized to the intervention arm or self‐directed control. The study lasted approximately 6 months, with microbiome samples collected at time points 1 and 2, and survey data collected at time points 1, 2, and 3. Time point 2 occurred 12–20 weeks after time point 1, and time point 3 followed 8–12 weeks later (Figure 1).

FIGURE 1.

Participant study flow and study design. (A) Progression of participants from recruitment to analysis. (B) Participant flow from enrollment to exit. Participants were randomized into intervention (green) and control (purple) arms. The intervention arm received bi‐weekly educational emails and coaching calls every 4 weeks (green triangles). Data collection occurred at time point 1 (~2 weeks post‐randomization) and time point 2 (10–12 weeks after time point 1). Exit surveys were completed at time point 3.

2.4. Study outcomes

The primary outcome was differences in microbiome composition between study arms, measured by alpha and beta diversity, the mean relative abundance of pre‐defined key taxa in the infant gut microbiome—selected for their known prevalence in early life ¹⁹ , ²⁰ —and the mean abundance of genes related to HMO degradation and SCFA production, which represent core functions of the early life microbiome (Appendix S1). ⁸ , ²¹ The secondary outcome was effects on a birth‐related microbiome signature—C‐section index—between arms, which was compared to a reference group of vaginally delivered infants. An additional outcome was differences in the prevalence of symptoms like gastroesophageal reflux and atopic conditions like atopic dermatitis and food allergy, reported by parents, compared between arms.

2.5. Sample collection

Participants were provided with two microbiome stool sample collection kits, containing a DNA‐free, sterile collection system with an active drying tube (FLOQSwab‐ADT; Copan Diagnostics, USA), ²² instructions, a biosafety bag, and a prepaid return shipping envelope. Parents were instructed to collect a stool sample from a soiled diaper.

2.6. Wellness program contents

Participants in the intervention arm were provided access to a gut microbiome‐based, integrative wellness program aimed at supporting infant gut health. Although this program is commercially available through Tiny Health, it was labeled in a way that no link to the company could be identified during recruitment for all participants and throughout the intervention for the control arm. In addition, Tiny Health did not publish anything related to this study, ensuring that participants in the control arm could not identify the company as its sponsor until after the intervention was complete—when they received their microbiome reports. Detailed methods of the program's content and how these were delivered can be found in Appendix S2.

2.7. Sample and data processing

Samples were received by and processed in a CLIA/CAP‐certified lab, compliant with the Clinical Laboratory Improvement Amendments (CLIA) and accredited by the College of American Pathologists (CAP). Total DNA was extracted using the DNeasy Powersoil Pro kit (Qiagen, Germany). The concentration of extracted DNA at time point 1 ranged from 2.14 to 184 ng/μL and the bacterial‐to‐host ratio (used as a proxy for bacterial load) ²³ ranged from 0.445 to 0.999. After quality control (QC), libraries were prepared using the Nextera kit. Library fragment sizes were then assessed using Agilent TapeStation or Bioanalyzer and quantified by qPCR. All samples and libraries passed QC and were sequenced to at least 10 million reads on a NextGen Illumina Platform. Additional details in Appendix S2.

2.8. Randomization

To ensure balance across biologically relevant variables, infants were stratified by a combination of sex and age at enrollment (0–30, 31–60, and 61–90 days), resulting in six strata. Additional details in Appendix S2.

2.9. Statistical analyses

Analyses were conducted on the per‐protocol population, defined as participants who completed the exit survey or provided microbiome samples at both time points 1 and 2. Participant characteristics were compared between arms using Fisher's exact test for categorical variables and Mann–Whitney U test for continuous variables. For Mann–Whitney U tests, we report the change‐in‐location parameter as the estimate, as well as the 95% confidence intervals for this parameter. In addition, we report medians as estimates of central tendency.

Microbiome diversity metrics were calculated using the vegan package in R. ²⁴ Shannon diversity and Bray‐Curtis distances were used to assess alpha and beta diversity, respectively. To assess compositional differences both between and within arms, we employed pairwise PERMANOVA tests via adonis2. The pairwise tests were accomplished by subsetting the data to each grouping of interest (i.e., intervention‐sample_1 vs. control‐sample_1, intervention‐sample_1 vs. intervention‐sample_2, etc.) and running the adonis2 test for each. For within‐arm comparisons where paired observations exist, restricted permutations were defined with blocks for the individual. Community dispersion was assessed with betadisper, followed by Tukey's HSD for post hoc comparisons. Non‐metric multidimensional scaling (NMDS) visualizations were generated using metaMDS from the vegan package. All p values for diversity metrics were adjusted for multiple comparisons using the Benjamini‐Hochberg (BH) method (q values, 10 tests).

Taxa and functional gene abundances were compared within study arms using the Wilcoxon signed‐rank test, and between arms using the Mann–Whitney U test. p values were corrected for multiple testing using the BH method (q values, 48 tests for taxa and 24 tests for genes).

Analogous to the approach presented by Stokholm et al. (2020) and independent of the study data presented here, we developed a taxonomy‐based C‐section index to evaluate the C‐section signature, focusing on taxa that are differentially abundant in infants delivered via C‐section compared to those delivered vaginally (Table S1) ¹⁸ identified using a multivariable association analysis (MaAsLin2). ²⁵ To do this, we used proprietary data plus data from Stokholm et al. (2016), Shao et al. (2019), and Reyman et al. (2019). ⁵ , ¹⁵ , ²⁶ The C‐section index was calculated as the sum of C‐section taxa divided by the sum of C‐section and vaginal taxa. C‐section indices were compared between study arms and against external reference groups of vaginally or C‐section delivered infants from Shao et al. (2019) using the Mann–Whitney U test, with multiple testing correction (BH method, 4 tests).

We used logistic regression to compare condition prevalence between study arms before and after the intervention. Conditions reported at time points 2 and 3 were aggregated, with a positive report at either time point 2 or 3 classifying a participant as positive for that condition. Potential confounders, such as age and sex, were tested as random effects using a linear mixed model but were not significant and were excluded from further analyses.

Because AD is a common early‐life condition and often the first step in the atopic march, ²⁷ we conducted an additional analysis comparing taxa abundance and C‐section index in AD versus non‐AD participants using the Mann–Whitney U test, with multiple testing correction (BH method, 3 tests).

For all comparisons, a q value <0.1 was deemed significant, except for condition prevalence analysis, where a nominal p value <0.05 was considered significant.

Sample size was determined through a power calculation using a t‐test to detect a difference in the relative abundance of Bifidobacterium with an effect size of 0.8 (Cohen's d), aiming for 80% power at α = .05. This analysis suggested we would need 25 participants per group. However, Mann–Whitney U tests were used instead after we found the data did not satisfy the distributional assumptions of the t‐test. Additional methods in Appendix S2.

3. RESULTS

3.1. Study recruitment and cohort demographics

The power analysis determined a minimum sample size of 25 infants per arm. A total of 73 parent‐infant dyads were consented into the study. Eighteen participants did not complete the study: 3 of whom officially withdrew and 15 were lost to follow‐up (Figure 1A). One participant from the control arm was excluded due to using an at‐home microbiome test for personal use during the study period. At time point 2, 25 intervention and 29 control participants had completed survey data collection and microbiome sample collection. At time point 3, 18 intervention and 28 control participants completed the exit survey, thereby providing all data points.

At time point 1, participant characteristics were not different between arms (Table S2). The percentage of infants receiving a majority of breast milk decreased in both arms but was not different between arms at any time point (time point 1: p = .126, time point 2: p = .576, time point 3: p = .541). All infants were introduced to solids during the study. The proportion of infants consuming more than 10% solids was not different between arms at time point 2 (p = .535) but became different at time point 3 (p = .011), with a higher percentage in controls.

Since probiotic usage before recruitment was an exclusion criterion, both arms were not different at time point 1 (p = 1). During the intervention, a potential recommendation was a Bifidobacterium probiotic. By time point 2, usage increased to 64% in the intervention arm while remaining unchanged at 3% in the control arm (p < .001). At time point 3, usage in the intervention arm was 67%, while the control arm saw a modest increase to 7% (p < .001) (Table S2). At time points 2 and 3, use of Lactobacillus probiotics was not different between arms.

No harms or unintended effects were reported by participants during the trial period, and no serious adverse events occurred.

3.2. Microbial composition shifts after the intervention

We observed a strong shift in microbial community structure in the intervention arm that was not present in the control arm. At the genus level, Bifidobacterium was the most prevalent genus in both arms (Figure 2A). While there were no differences in Shannon diversity index between the two arms at time point 1, we observed a difference at time point 2 (point estimate: 0.891, [95% CI, 0.029–1.637], p = .042, q = .085), driven by a decrease observed only in the intervention arm (point estimate: −0.768, [95% CI, −1.375 to −0.368], p = .014, q = .057) (Figure 2B, Table S3). For beta diversity, we observed a marked shift in gut microbiome composition from time point 1 to time point 2 in the intervention arm only (Figure 2C,D and Figure S1, PERMANOVA p < .001, q < .001). The shift observed in the intervention arm was consistent between subjects, as indicated by the direction and magnitude of centroid movement (Figure 2D) and the significant change in MDS1 values (p < .001, q < .001) (Figure 2E). In contrast, subjects in the control arm displayed more subtle and heterogeneous shifts, with the microbiome of these infants moving in divergent directions (Figure 2C) and with no significant overall change (Figure 2E). These observed differences in microbiome composition indicate a unifying compositional shift in the intervention arm that was absent in the control arm.

FIGURE 2.

Changes in microbial community structure and alpha and beta diversity post‐intervention. (A) Relative abundance of the 20 most abundant microbial taxa at the genus level. Less abundant taxa were grouped and are shown as “Other”. Each vertical bar represents an individual sample, and sample order is consistent across top and bottom panels to allow for individual‐level comparisons. (B) Shannon diversity index for control and intervention arms. (C, D) Non‐metric multidimensional scaling (NMDS) plots based on Bray‐Curtis distances, illustrating the clustering of microbial communities for each sample in (C) control and (D) intervention arm. Ellipses represent the standard error around group centroids for each time point, and the black arrows connect group centroids between time point 1 and time point 2 with the arrowhead pointing to time point 2. Samples from the same individual are joined by gray arrows with the arrowhead pointing to time point 2. (E) Boxplots of MDS1 from a NMDS analysis based on Bray‐Curtis dissimilarities. Boxplots show the median, interquartile range, and individual data points for each arm and time point. Points from the same individual are connected by lines to illustrate within‐subject changes over time. The Wilcoxon signed‐rank test was used to assess statistical significance between time points, and the Mann–Whitney U test and PERMANOVA were used to assess statistical significance between arms. p values were adjusted for multiple comparisons using the BH method (q values, 10 tests). *q < .1, ***q < .01; n.s., not significant.

3.3. Changes in key bacterial taxa in the intervention arm

Next, we compared the relative abundance of key infant taxa across arms and samples (Table S4). Between time points, there was an increase in Bifidobacterium both for the intervention (medians: time point 1 = 31%, time point 2 = 82%, estimate = 36%, [95% CI, 15%–53%], p = .001, q = .007) and the control arm (medians: time point 1 = 45%, time point 2 = 55%, estimate = 8.1%, [95% CI, 2%–22%], p = .014, q = .084) (Figure 3A), but the intervention arm reached higher levels than the control arm (estimate = −18%, [95% CI, −37% to −12%] p = .025, q = .121) (Figure 3A). For infant‐associated Bifidobacterium species, that is, species that have the genetic capacity to degrade HMOs from breast milk—namely B. infantis, B. longum, B. breve, and B. bifidum, B. infantis increased in the intervention arm (medians: time point 1 = .4%, time point 2 = 61%, estimate = 41%, [95% CI, 29%–59%], p < .001, q = .001) and was more abundant than in the control arm at time point 2 (medians: Intervention = 61%, Control = 3.4%, estimate = −27%, [95% CI, −61% to −6%], p = .001, q = .013) (Figure S2). The levels of the other infant‐associated Bifidobacterium species were not different between arms at time points 1 and 2 (Figure S2).

FIGURE 3.

Key taxa and HMO‐degrading gene abundances change over time and between control and intervention arm. Relative abundance of (A) Bifidobacterium and (B) Enterobacteriaceae; and gene abundance (measured in RPKM) of (C) α‐L‐fucosidase and (D) 2,3‐2,6‐a‐sialidase. Boxplots show the median, interquartile range, and individual data points for each arm and time point. The Wilcoxon signed‐rank test was used to assess statistical significance between time points, and the Mann–Whitney U test was used to assess statistical significance between arms. p values were adjusted for multiple comparisons using the BH method (q values, 48 tests for taxa and 24 tests for genes). *q < .1, **q < .05, ***q < .01, n.s., not significant.

We next inspected the relative abundance of opportunistic pathogens commonly found at high levels in the gut microbiome of infants born by C‐section (Table S4). The relative abundance of Enterobacteriaceae decreased from time point 1 to time point 2 in both the control (medians: time point 1 = 15.9%, time point 2 = 3.1%, estimate = −10%, [95% CI, −16% to −5%], p < .001, q = .002) and intervention arm (time point 1 = 16.3%, time point 2 = 3.7%, estimate = −20%, [95% CI, −30% to −8%], p < .001, q = .001) (Figure 3B). For Escherichia coli, a decrease between time point 1 and time point 2 was observed only in the intervention arm (medians: time point 1 = 5.5%, time point 2 = 1.4%, estimate = −4.5%, [95% CI, −15% to −1%], p = .009, q = .059) (Figure S3). At any time point, no differences between arms were observed for these or the other analyzed opportunistic pathogens (Figure S3).

3.4. The intervention led to functional shifts in HMO degradation and SCFA production genes

We assessed the abundance of microbial genes associated with HMO degradation (Table S5) and SCFA production (Table S6). At time point 2, the intervention arm showed a higher abundance of HMO degradation genes compared to the control arm (Table S7). This included genes for α‐L‐fucosidase (medians: Intervention = 3157, Control = 1638, estimate = −1041, [95% CI, −2056 to −163], p = .019, q = .046) (Figure 3C), 2,3‐2,6‐a‐sialidase (medians: Intervention = 3157, Control = 1638, estimate = −778, [95% CI, −1273 to −136], p = .012, q = .033) (Figure 3D), and other HMO‐degrading enzymes (medians: Intervention = 4132, Control = 2789, estimate = −1204, [95% CI, −2011 to −469], p = .002, q = .012) (Figure S3F). The abundance of butyrate‐related genes decreased from time point 1 to time point 2 only in the intervention arm (medians: time point 1 = 428, time point 2 = 98, estimate = −224, [95% CI, −379 to −90], p = .002, q = .012), and it was lower in the intervention arm than in the control arm at time point 2 (medians: Intervention = 98, Control = 378, estimate = 123, [95% CI, 1.2 to 338], p = .046, q = .091) (Figure S4B). No differences between arms were observed for the other SCFA genes (Figure S4A,C and Table S7).

3.5. The intervention reduced C‐section index to levels comparable to those of vaginally‐born infants

We then assessed the intervention's impact on C‐section index, a metric that represents the ratio of taxa associated with C‐section births to those associated with vaginal births (Table S1). Higher index values reflect a microbiome profile typical of C‐section‐delivered infants.

The C‐section index went down in the intervention arm between time points 1 and 2, a change not observed in the control arm (Figure 4A–C). Comparing the C‐section index of both arms to infants from Shao et al. (2019), we found that at time point 1, both arms had higher C‐section indices than vaginally‐born infants (control: p = .005, q = .007; intervention: p < .001, q < .001). By time point 2, only the control arm remained different (p = .013, q = .052) (Figure 4C, Table S8), suggesting the intervention shifted the microbiome of C‐section infants closer to that of vaginally‐born infants.

FIGURE 4.

Reduced C‐section index in the intervention arm. (A, B) Non‐metric multidimensional scaling (NMDS) plots based on Bray‐Curtis distances shaded by C‐section index. Plots show the clustering of microbial communities for time point 1 (circles) and time point 2 (triangles) in (A) the control arm and (B) the intervention arm. Ellipses represent standard error around the group centroids for each time point. (C) C‐section index across time points 1 and 2 in the control and intervention arms, shown alongside a group of C‐section‐delivered and vaginally‐delivered infants from Shao et al. (2019): Time point 1 at 21 days, time point 2 between 100 and 365 days. The Wilcoxon signed‐rank test was used to assess statistical significance between time points, and the Mann–Whitney U test was used to assess statistical significance between arms. p values were adjusted for multiple comparisons using the BH method (q values, 4 tests). *q < .1, ***q < .01, n.s., not significant.

3.6. Reduced prevalence of health conditions in the intervention arm

Next, we assessed the intervention's effects on the prevalence of conditions before and after the intervention. At time point 1, reflux was the most reported condition, followed by atopy (i.e., infants with AD, food allergies, cow's milk protein allergy, or sensitivities through breast milk) (Table S9). No differences in any reported conditions were observed between arms at time point 1 (Table S10). Considering both time points 2 and 3, the intervention arm contained fewer infants with reported conditions overall (p = .004), as well as fewer cases of atopy (p = .034) and AD (p = .031) (Figure 5A, Table S9 and Table S11). Together, this suggests the intervention contributed to reducing the prevalence of atopic conditions, despite its short duration and relatively small size.

FIGURE 5.

Infants with atopic dermatitis have a higher C‐section index and altered gut microbiome composition. (A) Percentage of infants with different health conditions across time points. The left panel shows the percentage of infants with any condition, and the central panel shows the percentage of infants with atopic dermatitis (AD) and/or any food reaction. (B–D) Association between AD status at time point 2 and (B) C‐section index, relative abundance of (C) Bifidobacterium and (D) Enterobacteriaceae. The Mann–Whitney U test was used to assess statistical significance between the AD and non‐AD groups. p values were adjusted for multiple comparisons using the BH method (q values, 3 tests). *q < .1, **q < .05.

3.7. Infants with atopic dermatitis have an elevated C‐section index and altered levels of key bacterial taxa

We further examined differences among infants with and without AD independent of the study arm. When participants were stratified according to their AD status, there were no demographic or baseline differences between the groups (Table S12). At time point 2, infants with AD had a higher C‐section index compared to those without AD (point estimate, 0.130, [95% CI, 0.027 to 0.550], p = .006; q = .019) (Figure 5B and Table S13). Bifidobacterium levels were higher (point estimate, −0.233, [95% CI, −0.668 to 0.000], p = .056; q = .058) (Figure 5C) and Enterobacteriaceae levels were lower (point estimate, 0.030, [CI, −0.002 to 0.088], p = .058; q = .058) in infants without AD (Figure 5D). These findings suggest a link between a C‐section‐like microbiome and the presence of AD.

4. DISCUSSION

Infants born via C‐section often exhibit distinct gut microbiome composition, are at higher risk of developing atopic conditions, and have elevated risk of chronic inflammatory conditions later in life. This study explored whether an integrated approach of gut microbiome‐focused education, coaching, and reports with tailored recommendations could improve infant gut health and potentially mitigate these risks. We found that our intervention led to significant changes in the gut microbiome, with notable shifts in both microbial taxonomy and functions, that made the gut microbiome profile more similar to that of vaginally born‐infants. The intervention also resulted in fewer reports of atopic conditions. Together, these results suggest that targeted microbiome interventions can be effective and may lower the risk of future health conditions in C‐section‐born infants.

Breastfeeding is a powerful determinant of gut microbiome composition, ¹⁹ , ²⁰ playing a critical role in correcting imbalances caused by C‐section delivery or antibiotic use. ²⁸ In this study, most infants were exclusively breastfed, and those who were mixed‐fed still primarily received breast milk. This potentially explains the increase in Bifidobacterium observed in both arms and the reduction of Enterobacteriaceae, which has been previously observed as part of normal gut maturation in the context of predominant breastfeeding. ²⁰ Larger studies may help to better understand these interactions and optimize interventions to better align with various feeding practices.

Not all breastfed infants may be able to fully benefit from breast milk, as effective HMO degradation depends on the presence of specific Bifidobacterium species. ⁸ In this context, the increase in B. infantis and lower alpha diversity ²⁹ observed in the intervention arm stands out, as B. infantis is able to comprehensively degrade HMOs from breast milk ³⁰ leading to domination of the community. ³¹ This is also reflected in the increased abundance of HMO degradation genes in the gut microbiome of the intervention arm. We interpret these changes as a favorable turn towards a more specialized microbial community. The presence of expert HMO‐degraders suggests utilization of HMOs, which results in the production of beneficial metabolites that mature the immune system ³² and contribute to colonization resistance. ³³ , ³⁴ , ³⁵ The increase in B. infantis is in part attributable to probiotics: infants in the intervention arm with less than 80% relative abundance of Bifidobacterium were recommended several choices of probiotics containing clinically validated infant‐type Bifidobacterium strains. Based on survey data, 64% of infants in the intervention arm took a Bifidobacterium probiotic between time point 1 and time point 2, compared to only 3% in the control arm, where no recommendations were provided.

We also observed a significant decrease in genes associated with butyrate production at time point 2, but only within the intervention arm. While this may initially appear counterintuitive, we believe this reduction brought butyrate gene abundance into a range that likely reflects the normal developmental stage of the infant gut microbiome at this early age, when breast milk (or formula) is still the primary source of nutrition. Butyrate‐producing bacteria that specialize in metabolizing complex starches and fibers present in solid foods—such as Bacteroides, Faecalibacterium prausnitzii, Blautia, and Roseburia—tend to markedly increase with the cessation of breastfeeding. ²¹ , ³⁶ Thus, a low abundance of butyrate‐producing genes at this stage is expected and may reflect a more age‐appropriate maturation of the microbial community, as accelerated maturation has been linked to increased risk of asthma ¹⁸ , ³⁷ and AD. ³⁸

In terms of health outcomes, the intervention arm reported significantly fewer cases of atopy and AD, suggesting that the intervention translated into tangible health benefits even within the short follow‐up period. The intervention also successfully reduced the C‐section index, bringing the gut microbiome profiles of C‐section‐delivered infants closer to those of vaginally‐born infants. In contrast, the C‐section index in the control arm remained significantly elevated at time point 2. When grouped according to their AD status across arms, infants with AD also had a markedly higher C‐section index. Our findings differ from those of Shao et al. (2019), who reported no significant difference in the C‐section index between C‐section and vaginally‐delivered infants at a similar age. Such discrepancies may stem from differences in sampling and bioinformatics processes, feeding practices, and geography between the studies. To control for technical variation stemming from bioinformatics pipelines, we reprocessed all raw reads using the same bioinformatics pipeline. Regarding feeding practices, our cohort had a higher proportion of exclusively breastfed infants, though breastfeeding rates were balanced across study arms. Participants in the comparison study were recruited in the UK, while our study spanned multiple U.S. locations, and differences in lifestyle as well as environmental exposures (including bacteria in the environment) are expected.

While this study presents promising results, we acknowledge its limitations. First, the relatively small sample size may limit the generalizability of the findings. Additionally, the short follow‐up period may not capture the long‐term effects of the intervention on the gut microbiome and health outcomes. We did not collect stool samples at time point 3, where some of the differences in health conditions were observed. To reduce concerns about potential bias introduced by attrition at time point 3, we compared baseline characteristics of participants who completed the study between the two arms and found no significant differences in demographics or clinical factors. Finally, most infants in our study were exclusively breastfed, limiting our ability to evaluate the intervention's impact on formula‐fed infants.

Future studies should have longer follow‐up, including medical diagnosis of health conditions, and enroll formula‐fed infants. In addition, this study was not prescriptive, as it allowed parents to make their own choices regarding recommendations and product brand selection. While this enhances real‐world applicability, it also introduces variability. Despite these limitations, we view the levels of statistical significance of the presented findings in this small human cohort as remarkable and highly encouraging for the field.

In conclusion, this study provides evidence that a personalized microbiome intervention can significantly influence the gut microbiome composition and functionality of C‐section‐born infants. This approach is different from previous interventions that mainly focused on the effects of probiotics on gut microbiome composition or specific health conditions. ³⁹ , ⁴⁰ , ⁴¹ Our approach is the first to leverage a comprehensive microbiome intervention with a variety of actions and personal coaching. Whether similar results can be obtained without coaching and simplified interventions is a topic of future study. The significant alterations in microbiome composition and functionality as well as reduction in atopic conditions observed within the short time frame of this study highlight the potential of our intervention strategy to potentially prevent chronic health conditions.

AUTHOR CONTRIBUTIONS

Pamela A. Nieto: Writing – original draft; writing – review and editing; visualization. Claudia Nakama: Investigation; project administration; visualization; writing – original draft; writing – review and editing. Julian Trachsel: Formal analysis; methodology; visualization; writing – review and editing. David Goad: Data curation; methodology; software. Taylor K. Soderborg: Investigation; writing – review and editing. Danielle Shea Tan: Investigation; writing – review and editing. Amy Orlandi: Investigation; writing – review and editing. Qian Yuan: Writing – review and editing. Elisa Song: Writing – review and editing. Noel T. Mueller: Writing – review and editing. Ruben A. Mars: Conceptualization; formal analysis; writing – review and editing. Cheryl Sew Hoy: Conceptualization; supervision; writing – review and editing; funding acquisition. Kimberley V. Sukhum: Conceptualization; investigation; formal analysis; methodology; supervision; visualization; writing – review and editing.

FUNDING INFORMATION

This study was financed and sponsored by Seeding, Inc. doing business as Tiny Health.

CONFLICT OF INTEREST STATEMENT

P.A.N., C.N., J.T., D.G., T.S., D.S.T., A.O., and K.V.S. are employees of Tiny Health, and except for A.O., all hold stock options in the company. K.V.S. and T.S. have been paid by Tiny Health to attend meetings. K.V.S. and D.G. have planned and pending patents related to the research. E.S. serves as the Chief Medical Officer (CMO) of Tiny Health. N.T.M. is a scientific advisor to Tiny Health. R.A.M. is a founding advisor to Tiny Health, a role approved by the Medical‐Industry Relations Committee of Mayo Clinic, fully independent of his employment at Mayo Clinic. C.S.H. is the CEO and founder of Tiny Health. Q.Y. declares no competing interests.

ETHICS STATEMENT

This trial was reviewed and approved by Sterling Institutional Review Board (ID:10947). All participants provided written informed consent before participating in the study. The study was carried out in accordance with the principles of the Declaration of Helsinki. The trial was registered at ClinicalTrials.gov, number NCT06424691.

Supporting information

Appendix S1.

Appendix S2.

Appendix S3.

Appendix S4.

Appendix S5.

Nieto PA, Nakama C, Trachsel J, et al. Improving immune‐related health outcomes post‐cesarean birth with a gut microbiome‐based program: A randomized controlled trial. Pediatr Allergy Immunol. 2025;36:e70182. doi: 10.1111/pai.70182

DATA AVAILABILITY STATEMENT

Deidentified individual participant data and microbial reads generated and analyzed during this trial are available in the National Center for Biotechnology Information (NCBI) repository under bioproject accession PRJNA1206145.