The relationship between the gut microbiome and the risk of respiratory infections among newborns.

Background: Emerging evidence points to a critical role of the developing gut microbiome in immune maturation and infant health; however, prospective studies are lacking.
Methods: We examined the occurrence of infections and associated symptoms during the first year of life in relation to the infant gut microbiome at six weeks of age using bacterial 16S rRNA V4-V5 gene sequencing (N = 465) and shotgun metagenomics (N = 185). We used generalized estimating equations to assess the associations between longitudinal outcomes and 16S alpha diversity and metagenomics species.
Results: Here we show higher infant gut microbiota alpha diversity was associated with an increased risk of infections or respiratory symptoms treated with a prescription medicine, and specifically upper respiratory tract infections. Among vaginally delivered infants, a higher alpha diversity was associated with an increased risk of all-cause wheezing treated with a prescription medicine and diarrhea involving a visit to a health care provider. Positive associations were specifically observed with Veillonella species among all deliveries and Haemophilus influenzae among cesarean-delivered infants.
Conclusion: Our findings suggest that intestinal microbial diversity and the relative abundance of key taxa in early infancy may influence susceptibility to respiratory infection, wheezing, and diarrhea.

Introduction

Infections remain the leading causes of mortality in infants globally[1]. The human gut microbiome is becoming increasingly recognized for its critical role in immune function and the inflammatory response[2,3]. A bidirectional relationship emerges following birth whereby the gut microbiome aids the maturation of the immune system and the immune system regulates host–microbe symbiosis[4,5]. The impacts of perturbing these intricate relationships are evident in high-risk infants. For example, among infants with cystic fibrosis, the composition of the gut microbiome is a determinant of colonization with opportunistic pathobionts[6,7]. Likewise, in preterm infants, the gut microbiome is associated with fatal occurrences of necrotizing enterocolitis and infection[2,8,9]. Factors driving the establishment of the gut microbiome, including delivery mode and breast feeding[3,10-13], have also been related to the risk of infections[3,14-18]. Furthermore, the use of antibiotics during pregnancy, which has been found to influence the gut microbiome of offspring[19-21], increased the risk of infant infection-related hospitalizations[22]. Encouraging results from probiotic trials suggest health benefits from altering the gut microbiome, including an enhanced immune response to pathogens[23,24]. While studies have found possible links between early gut microbiome composition and infant infection[3,25], few prospective studies have been conducted, particularly in the general population.

We report on gut microbiome diversity and composition among infants during the critical period of early immune training and the subsequent occurrence of respiratory infections and symptoms, such as wheezing and diarrhea, in the first year of life as part of a prospective study of a cohort of pregnant women and their offspring from the general population in New Hampshire. Here, we measure the fecal microbiome to measure the gut microbiome. Wheeze and diarrhea outcomes for this study include those of any cause. Based on ASV data generated from 16S rRNA sequencing, higher alpha diversity at 6 weeks of age is associated with having an additional respiratory infection or symptom of respiratory infection requiring a prescription medicine, with associations varying by delivery mode. Using next-generation sequencing (NGS), shotgun metagenomics, Veillonella in all deliveries and Haemophilus in cesarean deliveries are among the species in 6-week stool identified as being related to an additional subsequent respiratory infection or symptom of respiratory infection requiring a prescription medicine during an infant’s first year of life.

Methods

Study population

Participants included mother–infant dyads from the New Hampshire Birth Cohort Study from whom we obtained infant stool samples at approximately 6 weeks of age. Pregnant women aged 18–45 were recruited from prenatal clinics in New Hampshire, USA, starting in January 2009, as described previously[26]. Women who were living in the same household served by a private water system since their last menstrual period, had no plans to move, and had a singleton pregnancy were included in the cohort. Participants completed surveys on infant lifestyle questions such as feeding mode, solid food introduction, and daycare. Infant birth characteristics were ascertained from newborn medical records, and maternal characteristics were abstracted from prenatal and delivery records, including age at enrollment, prenatal use of antibiotics, and prepregnancy body mass index in kilograms per meter squared. The Committee for the Protection of Human Subjects at Dartmouth College approved all protocols, and participants provided written informed consent upon enrollment.

Ascertainment of infant health outcomes

Telephone interviews were conducted with infants’ caregivers in the first year of life, i.e., when infants turned approximately 4, 8, and 12 months of age. Caregivers were asked whether their child had any upper respiratory tract infections (RTIs) or associated symptoms (e.g., runny nose, stuffy nose, eye infection, ear infection, influenza, sinus infection, pharyngitis, or laryngitis), lower RTIs (e.g., bronchitis, pneumonia, bronchiolitis (including respiratory syncytial virus (RSV)), or whooping cough), acute respiratory symptoms (e.g., difficulty breathing, wheezing, fever, or cough), or diarrhea since the previous interview. For each positive response, participants were then asked whether the condition lasted more than 2 days, whether the child saw a physician, and whether the child received any prescription medications for the condition.

Stool sample collection, DNA extraction, sequencing, and profiling

We measured the fecal microbiome of infant stool as a measure of the infant gut microbiome. Infant stool samples were collected at regularly scheduled ~6-week postpartum follow-up appointments as described previously[13,27]. Samples were aliquoted and frozen at −80 °C within 24 h of receipt. A Zymo DNA extraction kit (Zymo Research) was used for DNA extraction from thawed samples, and an OD260/280 nanodrop was used to measure sample quality and purity. The V4-V5 hypervariable region of the bacterial 16S rRNA gene was sequenced using Illumina MiSeq at the Marine Biological Laboratory in Woods Hole, MA. Amplicon sequence variants (ASVs) were then inferred using DADA2[28], and taxonomies were assigned using the SILVA database[29]. Quality control measures were conducted as described previously[13]. A subset of stool samples was also sequenced with NGS and shotgun metagenomics sequencing as previously described[21]. Extracted DNA samples were sheared to a mean insert size of 400 bp using a Covaris S220 focused ultrasonicator. The sequencing libraries were constructed using Nugen’s Ovation Ultralow V2 protocol, and samples were sequenced using Illumina NextSeq. DNA reads were merged and trimmed using KneadData[30] for quality control before species-level taxonomic profiles were generated using Metaphlan2[31].

Statistics and reproducibility

We examined the association between gut microbiome composition and health outcomes ascertained during interval interviews over the subjects’ first year of life. For our analyses, we examined the total number of reported outcomes, specifically upper RTIs and lower RTIs, as well as symptoms such as wheezing with a reported visit to a physician and treatment with a prescription medication. Diarrhea is not typically treated with prescription medications in infants; therefore, we focused the analyses on reports of diarrhea that involved a doctor visit. We imputed missing outcomes if the caregiver completed the interview but a specific question was unanswered using multiple imputation by chained equations and the predictive mean matching method.

For models using 16S data, we aggregated ASVs to the genus level and calculated alpha diversity on read counts per genus using the inverse Simpson index. We then used generalized estimating equation (GEE) for repeated measures with Poisson regression and AR(1) correlation structure to assess the association between log2-transformed 16S-based genus-level alpha diversity and each of the outcomes of interest. For models using metagenomics species data, we calculated the log2-transformed relative abundance using a pseudocount of 5 × 10−20 for zero values. We also used a GEE for repeated measures with Poisson regression and AR(1) correlation structure to estimate relationships with species present in at least 10% of subjects. We applied a false discovery rate (FDR) threshold of 0.1 to adjust for multiple testing[32]. Factors associated with both the gut microbiome and health outcomes were considered potential confounders and included in all GEE analyses. These confounders included maternal prepregnancy body mass index (BMI) (kg/m2), delivery mode (vaginal/cesarean), infant sex (male/female), breast feeding at six weeks (exclusively breastfed/mixed fed or exclusively formula fed), antibiotic use during pregnancy (yes/no), and gestational age (complete weeks). We also conducted stratified analyses by delivery type for both alpha diversity using 16S data and microbial species based on metagenomics data. Due to sample size limitations, we conducted stratified analyses only for species-specific analyses on all outcomes combined and for upper RTIs.

For interpretability, we exponentiated the coefficient values to obtain relative risk (RR) and 95% confidence intervals (CIs). Of the 465 participants included in the 16S analyses, 391 participants (84.1%) had complete data for all potentially confounding variables. Of the 185 participants included in the metagenomics analyses, 160 participants (86.5%) had complete data for all potentially confounding variables. We assumed missing confounder entries were missing at random and used multiple imputation by chained equations and the predictive mean matching method to impute missing data. All analyses were performed using R version 3.4.3 using the functions diversity, mice, and geeglm in the ‘vegan’, ‘mice’, and ‘geepack’ packages.

Table 1

Selected baseline characteristics of mothers and infants in the New Hampshire Birth Cohort Study.

Variable	16S V4–V5 rRNA		Metagenomics
	Sample size	Mean (SD) or no. (%)	Sample size	Mean (SD) or no. (%)
Maternal characteristics
Age at enrollment, mean (SD), years	465	31.9 (4.6)	185	31.9 (4.3)
Body Mass Index before pregnancy, mean (SD), kg/m2	462	25.8 (5.9)	185	25.7 (5.7)
Parity, No. (%)	461		184
0		217 (47.1)		92 (50.0)
1		166 (36.0)		60 (32.6)
2+		78 (16.9)		32 (17.4)
Antibiotic use during pregnancy, No. (%)	427		173
Yes		79 (18.5)		36 (20.8)
No		348 (81.5)		137 (79.2)
Infant characteristics
Delivery mode, No. (%)	465		185
Vaginal		339 (72.9)		129 (69.7)
Cesarean		126 (27.1)		56 (30.3)
Infant sex, No. (%)	464		185
Male		248 (53.4)		107 (57.8)
Female		216 (46.6)		78 (42.2)
Breast feeding at 6 weeks, No. (%)	427		171
Exclusively breast fed		240 (56.2)		92 (53.8)
Mixed fed or exclusive formula fed		187(43.7)		79 (46.2)
Gestational age, Mean (SD), weeks	465	39.1 (1.6)	185	39.0 (1.7)

Of the 465 mothers included in the 16S analyses, maternal BMI was missing for 3 mothers. Parity was missing for 4 mothers, and antibiotic use during pregnancy was missing for 38 mothers. Of the 465 infants included in the 16S analyses, infant sex was missing for 1 infant, and feeding type was missing for 38 infants. Of the 185 mothers included in the metagenomics analyses, parity was missing for 1 mother, and antibiotic use during pregnancy was missing for 12 mothers. Of the 185 infants included in the metagenomics analyses, feeding type was missing for 14 infants.

SD standard deviation, No. frequency, kg kilograms, m meters.