The respiratory microbiome is linked to the severity of RSV infections and the persistence of symptoms in children

Abstract

Respiratory syncytial virus (RSV) is the leading cause of infant respiratory infections and hospitalizations. To investigate the relationship between the respiratory microbiome and RSV infection, we sequence nasopharyngeal samples from a birth cohort and a pediatric case-control study (Respiratory Syncytial virus Consortium in Europe [RESCEU]). 1,537 samples are collected shortly after birth ("baseline"), during RSV infection and convalescence, and from healthy controls. We find a modest association between baseline microbiota and the severity of consecutive RSV infections. The respiratory microbiota during infection clearly differs between infants with RSV and controls. Haemophilus, Streptococcus, and Moraxella abundance are associated with severe disease and persistence of symptoms, whereas stepwise increasing abundance of Dolosigranulum and Corynebacterium is associated with milder disease and health. We conclude that the neonatal respiratory microbiota is only modestly associated with RSV severity during the first year of life. However, the respiratory microbiota at the time of infection is strongly associated with disease severity and residual symptoms.

Keywords: 16S; RSV; airway; birth cohort; case-control; microbiota; nasopharynx; respiratory; severity.

Graphical abstract

Figure 1

Study overview Experimental setup to assess (1) the relationship between microbiota profiles preceding RSV infection (early-life) on later-life RSV occurrence and severity, (2) microbiota profiles during/after RSV infection compared to healthy controls, and (3) severity-associated changes in microbiota profiles. Samples were included from two studies, conducted across five European study sites. From all newborns/infants, questionnaires were collected on demographics, RSV risk factors, and measures of RSV severity. In addition, (longitudinal) nasopharyngeal samples were collected, resulting in matched samples collected before (birth cohort only), during, and after RSV infection (birth cohort and case-control study). All samples were subjected to 16S-rRNA-sequencing to characterize bacterial microbiota profiles. Linear/logistic mixed-effects regression models were employed, allowing us to adjust for relevant covariates and account for study site (random effect) (Methods). N, number of individuals; n, number of samples.

Figure 2

Microbiota diversity, cluster membership, and composition during RSV infection and at convalescence (A) ASV-level Shannon diversity (non-rarefied) between study groups. Boxplots represent the 25^th and 75^th percentiles (lower and upper boundaries of boxes, respectively), the median (middle horizontal line), and measurements that fall within 1.5 times the interquartile range (IQR; distance between 25^th and 75^th percentiles; whiskers). Statistical significance was assessed using linear mixed-effects models with Shannon diversity as outcome, age, gender, sequencing depth (scaled), and health status (healthy controls, RSV, or RSV convalescent) as fixed effects and study site as random effect. Time between RSV infection and convalescence (fixed effect) and subject ID were additionally included for comparisons between RSV and RSV convalescence. (B) Principal coordinate analysis (PCoA) based on Bray-Curtis dissimilarities showing the nasopharyngeal microbiota composition in healthy controls (n = 52), acute RSV (n = 374), and (matched) samples collected at convalescence (n = 338). Percentages in brackets denote the total variance explained by the first two principal coordinates. Each data point (dot) indicates a nasopharyngeal sample colored by study group. Ellipses denote the standard deviation of data points for each group. The 10 highest ranking ASVs over the first days of life were simultaneously visualized (triangles). R² and statistical significance of the association between health status and the overall microbiota composition was assessed using PERMANOVA tests (1,000 permutations, adjusting for age, gender, and study site [restricted permutations]). (C) Mosaic plot showing cluster membership in healthy controls, acute RSV, and convalescent samples. (D) Adjusted odds ratios (aORs) for cluster membership during RSV infection or convalescent phase (categorical variable; predictor), adjusted for age, gender (fixed effects), and study site (random effect), with health status as an outcome variable. Two models were simultaneously visualized: model (1) RSV infection vs. health and model (2) RSV convalescence vs. health. HAE9 cluster was not visualized, as it was highly prevalent during both RSV infection and convalescence, but absent in healthy controls. Whiskers denote 95% confidence intervals (CIs; Wald method). Asterisks denote statistical significance (NS, not significant [p > 0.05]; ∗, p ≤ 0.05; ∗∗, p ≤ 0.01; ∗∗∗, p ≤ 0.001). (E and F) Log₂ fold change (FC) of features (genera [E]/ASVs [F]) based on MaAsLin2 (linear mixed-effects model) with health status (healthy controls, RSV infection, and convalescence) as variable of interest, adjusted for age, gender (fixed effects), and study site (random effect) and log₂-transformed relative abundance as outcome. Only features present in ≥5% of samples at >0.1% relative abundance were tested. ASVs with a q ≤ 0.05 are depicted. Whiskers denote 95% confidence intervals (CIs; Wald method). Asterisks denote statistical significance (∗, q ≤ 0.05; ∗∗, q ≤ 0.01; ∗∗∗, q ≤ 0.001).

Figure 3

Associations between microbiota diversity, stability and composition, and RSV infection severity (A) ASV-level Shannon diversity (non-rarefied) in healthy controls (n = 52) compared to mild (RESViNet score 0–7; n = 218), moderate (8–13; n = 106), and severe RSV (14–20; n = 47). Statistical significance was assessed using linear mixed-effects models with Shannon diversity as outcome, age, gender, sequencing depth (scaled), and RSV severity (healthy, mild, moderate, and severe RSV) as fixed effects and study site as random effect. (B) Principal coordinate analysis (PCoA) depicting the overall nasopharyngeal microbiota composition in healthy controls, mild, moderate, and severe RSV. The R² and statistical significance across all severity groups was estimated and depicted in the upper left corner. Pairwise differences between groups were additionally modeled, adjusting for age, gender, and study site (restricted permutations). See legend Figure 2B. (C) Bray-Curtis dissimilarity between paired RSV infection and convalescent samples, stratified by mild, moderate, and severe disease (n = 179, n = 89, and n = 32, respectively). Statistical significance assessed using a linear mixed-effects model with severity as outcome of interest, adjusted for age at RSV infection, gender, and time between RSV infection (fixed effects) and study site (random effect), and Bray-Curtis dissimilarity as outcome. (D) Adjusted odds ratios (aORs) for cluster membership during mild, moderate, or severe RSV infection (categorical variable; predictor), adjusted for age, gender (fixed effects), and study site (random effect), with severity as an outcome variable. Three logistic mixed-effects regression models were simultaneously visualized, comparing cluster membership in (1) mild RSV vs. health, (2) moderate RSV vs. health, and (3) severe RSV vs. health. Although the HAE9 cluster was not visualized, it was highly associated with RSV infection and convalescence, as it was absent in healthy controls. Whiskers denote 95% confidence intervals (CIs; Wald method). Asterisks denote statistical significance (NS, not significant [p > 0.05]; ∗, p ≤ 0.05; ∗∗, p ≤ 0.01; ∗∗∗, p ≤ 0.001). (E and F) Log₂ fold change (FC) of features (genera [E]/ASVs [F]) based on MaAsLin2 (linear mixed-effects model) with RSV severity (compared to healthy controls) as variable of interest, adjusted for age, gender (fixed effects), and study site (random effect) and log₂-transformed relative abundance as outcome. Only features present in ≥5% of samples at >0.1% relative abundance were tested. Asterisks denote statistical significance (∗, q ≤ 0.05; ∗∗, q ≤ 0.01; ∗∗∗, q ≤ 0.001).

Figure 4

Associations between microbiota profiles at RSV convalescence and remaining symptoms (A and B) Log₂ fold change (FC) of features (genera [A]/ASVs [B]) based on MaAsLin2 (linear mixed-effects model) with symptoms (yes/no cough, wheeze, blocked/runny nose, and any symptoms) as variable of interest, adjusted for age, gender, and time between RSV infection and convalescence (fixed effects) and study site (random effect) and log₂-transformed relative abundance as outcome. Only features present in ≥5% of samples at >0.1% relative abundance across RSV and RSV convalescence samples were tested. Whiskers denote 95% confidence intervals (CIs; Wald method). Asterisks denote statistical significance (∗, q ≤ 0.05; ∗∗, q ≤ 0.01; ∗∗∗, q ≤ 0.001).

Figure 5

Early-life microbiota development and cluster membership (A) ASV-level Shannon diversity (non-rarefied) for baseline samples over age. Statistical significance was assessed using a linear mixed-effects model, including Shannon diversity as outcome, the natural log of age, sequencing depth, season of birth, gender, and presence of siblings as fixed effects and study site as random effect. (B) Principal coordinate analysis based on Bray-Curtis dissimilarities showing the nasopharyngeal microbiota composition over the first days of life (<3 days, n = 97; 3–4 days, n = 113; 5–6 days, n = 256; 7–8 days, n = 255; >8 days, n = 51). The R² and statistical significance of age categories was estimated (PERMANOVA; restricted permutations within study site) and depicted in the upper right corner. See legend Figure 2B. (C) Mosaic plot showing cluster membership over the first days of life, stratified by age (<5 days vs. ≥5 days). Only clusters including at least 2% of samples are shown (CDGa/b, STR4, STA1, and MOR2 clusters). (D) Adjusted odds ratios (aORs) for cluster membership at baseline (categorical variable; predictor), adjusted for age at sampling, adjusted for age at first RSV infection, gender, season of birth, presence of siblings (fixed effects), and study site (random effect), with medically attended RSV infection yes/no as an outcome variable. Whiskers denote 95% confidence intervals (CIs; Wald method). Asterisks denote statistical significance (NS, not significant [p > 0.05]; ∗, p ≤ 0.05; ∗∗, p ≤ 0.01; ∗∗∗, p ≤ 0.001). (E) Area-under-the-curve (AUC) receiver operating curves (ROCs) to evaluate the random forest classifier to discriminate between medically attended (N = 85) and not medically attended RSV (N = 100). The model includes 39 ASVs and age (in days) at first RSV infection as predictors. Curves were calculated for out-of-bag (OOB) and 5-fold cross-validated (CV) predictions, giving similar results, validating the use of OOB estimates for subsequent analyses. 95% Confidence intervals were calculated using the DeLong method as implemented in the pROC package. (F) Mean absolute permutation-based Shapley additive explanations (SHAP) values of the 10 most important features of the model described in (E). (G) SHAP local explanation summary plot with individual SHAP values for each subject (N = 137 per feature). Each dot has three characteristics: (1) vertical location indicates the feature it is depicting, (2) the color shows whether that feature was high or low for a given subject (scaled relative abundance or age), and (3) horizontal location shows whether that value caused a higher or lower prediction. Higher SHAP values indicate a positive contribution to the likelihood of developing a medically attended RSV infection.