Nasopharyngeal microbiome-epigenome-wide association analysis in infants with severe bronchiolitis

Abstract

Background

Bronchiolitis exposes infants to both acute burdens (e.g., hospitalization in cases of severe bronchiolitis) and increased risks for chronic respiratory sequelae (e.g., asthma). In severe bronchiolitis, recent evidence suggests distinct pathobiological roles of microbiota (e.g., viruses, bacteria) and host responses influenced by genetic and epigenetic factors. However, the relationship of airway microbiota with host DNA methylation (DNAm) in infants with severe bronchiolitis remains unknown.

Methods

In a multi-center prospective cohort of 504 multi-ethnic infants with severe bronchiolitis (age < 1 year), using nasopharyngeal microbiome (exposure) and blood DNAm (outcome, Infinium MethylationEPIC BeadChip, Illumina) data within 24 h of the hospitalization, we conducted microbiome-epigenome-wide association studies (mbEWAS). We examined microbiota-associated differentially methylated CpGs (mbDMCs, false discovery rate [FDR] < 0.05), regions (mbDMRs, FDR<0.05), and DNAm age acceleration. We also determined the associations of DNAm age acceleration with asthma development by age 6 years. Furthermore, we focused on asthma-related pathogenic bacteria—Haemophilus influenzae, Moraxella catarrhalis, and Streptococcus pneumoniae—for functional analyzes by examining serum mbDMR-related proteins (Proseek Multiplex, Olink) and their enriched pathways (FDR < 0.10).

Results

Across 23 common taxa—observed at least in 25% of the infants, we identified 1 mbDMC (S. pneumoniae, cg16594639, chr20: 39528675) and 96 mbDMRs (e.g., S. pneumoniae, chr5:27038497 – 27038802, CDH9; chr6:48068669–48068940, PTCHD4). A higher H. influenzae abundance was associated with DNAm age deceleration, and the deceleration was associated with a higher risk of developing asthma. In 29 mbDMRs of the asthma-related pathogenic bacteria, we identified 156 mbDMR-related proteins (e.g., MMP9, XCL1). These proteins were enriched in immune response-related pathways (e.g., regulation of ERBB signaling and eosinophil chemotaxis and migration pathways).

Conclusions

In this multi-center prospective cohort study of severe bronchiolitis, our mbEWAS suggested the microbiota-host associations that regulate immune responses.

Introduction

Bronchiolitis is the most common lower respiratory infection and the primary reason for hospitalizations among infants in the U.S., resulting in approximately 110,000 hospitalizations (i.e., severe bronchiolitis) each year (1,2). In addition to the acute burden, bronchiolitis also puts infants at risk of chronic respiratory sequelae—approximately 30% develop recurrent wheezing and/or asthma in childhood (3–6). Yet, the pathobiology of bronchiolitis underlying its acute and chronic sequelae remains insufficiently explained, particularly by conventional clinical data.

In the bronchiolitis pathobiology, recent studies have determined independent roles of microbiota (e.g., virus (7), bacteria (8–10), fungi (11)) in the airway or host epigenome (e.g., DNA methylation [DNAm] (12,13), microRNA (14,15)). Additionally, emerging evidence suggests that the microbiota (more specifically, bacteria) modifies the host epigenome by biosynthesizing chemical donors for DNAm, regulating epigenetic-modifying enzymes, and activating intrinsic host-cell epigenetic-modifying processes (16,17). In the airway microbiota, the limited literature has reported not only their influence by host genome (18,19) but also their relationships with targeted DNAm in the upper airway mucosa from pre-schoolers (20) and in cord blood (21). However, little is known about epigenome-wide associations between airway microbiota and blood DNAm, which may reflect airway DNAm signatures (22,23), in infants with severe bronchiolitis and their integrated contribution (e.g., modulating systemic immune responses) to the bronchiolitis pathobiology underlying its acute and chronic respiratory sequelae (e.g., asthma). In particular, current evidence suggests that, in asthma development, Streptococcus pneumoniae, Haemophilus influenzae, and Moraxella catarrhalis in the airway play pathogenic roles (8,24–28).

To address the knowledge gap, in a multicenter prospective cohort of infants with severe bronchiolitis, we conducted a microbiome-epigenome-wide association study (mbEWAS) using nasopharyngeal microbiome and blood DNAm data as well as serum proteome data to investigate how airway microbiota influence host epigenetic changes in early life and how the epigenetic changes influence systemic responses. A better understanding of the disease mechanisms may advance efforts to develop targeted preventive strategies in this high-risk population.

Methods

Study design, setting, and participants

We analyzed data from a multicenter and prospective cohort study of infants (age < 1 year) with severe bronchiolitis. This study, called the 35th Multicenter Airway Research Collaboration (MARC-35) (29–31), is coordinated by the Emergency Medicine Network (EMNet) (32), a research collaboration of 247 hospitals. Using a standardized protocol (29–31), investigators at 17 sites across 14 U.S. states (Table S1) enrolled infants hospitalized with an attending physician diagnosis of bronchiolitis during one of the consecutive bronchiolitis seasons from November 1 to April 30 (2011–2014). Bronchiolitis was defined by the American Academy of Pediatrics guidelines—acute respiratory illness with some combination of rhinitis, cough, tachypnoea, wheezing, crackles, and retractions, regardless of previous breathing problem episodes (33). We excluded infants with a known heart-lung disease, immunodeficiency, immunosuppression, or gestational age < 32 weeks. The institutional review board at each participating hospital approved the study with written informed consent obtained from the parent or guardian.

Data collection

Clinical data were collected via structured interviews and chart reviews (29). All data were reviewed at the EMNet Coordinating Center at Massachusetts General Hospital (Boston, MA, USA), and site investigators were queried about missing data and discrepancies identified by manual data checks. Nasopharyngeal and whole blood specimens were collected by trained site investigators within 24 h of hospitalization using standard protocols (29,34). The details of data collection are described in Methods S1.

Nasopharyngeal DNA extraction, metagenomic shotgun sequencing, microbiome profiling

The details of DNA extraction, metagenomic shotgun sequencing, microbiome profiling, and quality control (QC) performed according to standard procedures are described in Methods S1 and our previous paper (10). Briefly, considering the low-biomass nature of the study sample, we ensured data quality by confirming that negative controls could be deemed free of contamination, by confirming that positive controls (MSA-1003; ATCC, Manassas, VA, USA) exhibited an expected taxonomic composition, and by maintaining general cleanliness throughout. For microbiome profiling, we prepared libraries using the Illumina DNA Prep library kit (Illumina, San Diego, CA, USA), performed taxonomy profiling with MetaPhlAn3 (35), mapping reads to the marker gene database (mpa_v30_CHOCOPhlAn_201901) using bbmap (36), and generated estimated read counts. To the read counts, we applied a +1 (i.e., pseudocount) shift to mitigate the impact of zeros, calculated relative abundance by dividing the shifted counts by the total sum, performed a centred-log-ratio (CLR) transformation, and used the CLR-transformed values as bacterial abundance for the subsequent analyzes, according to a previous microbiome-genome wide association study (GWAS) (37). The interpretation of effect sizes in CLR transformation is less straightforward than natural log-transformation since the coefficients represent the effect of one taxon’s change relative to all other taxa. However, CLR transformation more appropriately accounts for the compositional nature of microbiome data (e.g., minimize spurious correlations arising from the constant-sum constraint) (38). Furthermore, by using the abundance to compute Aitchison distance and visualizing primary coordinate analysis plots, we also confirmed there were no distinct clusters corresponding to the 4 sequencing runs and the 17 participating sites (Figure S1).

Blood DNA extraction, DNA methylation profiling, and quality control

The details of DNA extraction, DNAm profiling, and QC performed according to standard procedures are described in Supplementary Methods S1 and our previous papers (12,13). Briefly, we performed DNAm profiling using the Illumina Infinium MethylationEPIC BeadChip (Illumina). To ensure quality, we applied multiple sample-level (Figure S2) and probe-level (Figure S3) QC filters (12,13) following the established preprocessing pipeline in the R minfi package (39). After the QC steps, we applied the single-sample normal-exponential using out-of-band probes (ssNoob) procedure to correct for background and dye bias (40). We logit-transformed the measured %DNAm level (i.e., M-value). By using the M-value to compute Euclidean distance and visualizing principal component analysis (PCA) plots, we also confirmed there were no distinct clusters corresponding to the 174 chips and the 17 participating sites (Figure S4).

Serum proteome profiling

The details of proteome profiling and QC performed according to standard procedures are described in Supplementary Methods S1 and our previous papers (13,41,42). Briefly, we performed proteome profiling using the Olink multiplex platform (Proseek multiplex arrays; Olink Bioscience, Uppsala, Sweden) with internal and external control samples in a subset of the MARC-35 participants (n = 112). After the QC steps, we normalized the expression value of each protein to a unit on the log₂ scale (normalized protein expression [NPX]), which is proportional to the protein concentration, and used the NPX values for the subsequent analyzes. We corrected for batch effects between the two assay runs by bridge sample normalization based on overlapping samples measured in both runs. Additionally, by using the normalized NPX to compute Euclidean distance and visualizing PCA plots, we also confirmed there were no distinct clusters corresponding to the 2 assay runs and the 17 participating sites (Figure S5).

Recurrent wheeze and asthma diagnosis

The development of recurrent wheeze by age 3 years and asthma by age 6 years was derived from parent interview data. Recurrent wheeze was defined as having at least 2 corticosteroid-requiring exacerbations in 6 months or at least 4 breathing problems in 1 year that lasted at least 1 day and affected sleep (43–45). Asthma was defined using a commonly used epidemiologic definition (46): physician diagnosis of asthma, with either asthma medication use (e.g., inhaled bronchodilators and inhaled corticosteroids) or asthma-related symptoms (e.g., wheezing and nocturnal cough) in the preceding year (44).

Statistical analysis

The analytic workflow is summarized in Figure 1. The details of the statistical analysis can be found in Methods S1. Analysis used R version 4.1.3 (R Foundation, Vienna, Austria). All P-values were two-tailed, with p < 0.05 considered statistically significant.

Figure 1. Analytic workflow of the study.

In a multicenter prospective cohort study—the 35th Multicenter Airway Research Collaboration (MARC-35), the current study investigated 504 infants hospitalized for bronchiolitis (i.e., severe bronchiolitis) with nasopharyngeal airway (NPA) microbiome and blood DNA methylation (DNAm) data. NPA microbiome data was measured by shotgun metagenomic sequencing, and 23 common taxa were used for the downstream analyzes. Blood DNAm data was measured by Infinium MethylationEPIC array (850 K), and a total of 776,307 CpGs were used for the downstream analyzes. First, we examined associations of NPA microbiota with blood DNAm by using the epigenome-wide association study (EWAS) approach. The microbiome-EWAS (mbEWAS) identified microbiota-associated differentially methylated CpGs (mbDMCs) and regions (mbDMRs). We also examined associations of NPA microbiota with DNAm age and associations of DNAm age with the risk of developing outcomes—recurrent wheeze by age 3 years and asthma by age 6 years. Second, we characterized the mbDMRs in 3 asthma-related pathogenic NPA bacteria—Haemophilus influenzae, Moraxella catarrhalis, and Streptococcus pneumoniae. We determined the 7 blood immune cell type-specific differential DNAm within the mbDMRs. The cell types were inferred by a deconvolution analysis using the DNAm data. In the 112 infants with serum proteome data, including 1,076 proteins measured by using 13 Olink panels, we identified mbDMR-related proteins and examined the pathway enrichment.

Examining associations of nasopharyngeal microbiota with blood DNAm

We first determined common nasopharyngeal bacterial taxa, in all taxonomic levels, that were identified at least in 25% of infants in the analytic cohort. We included not only species-level taxa but also taxa in higher taxonomic levels, as in previous mbGWASs (37,47–49) and mbEWAS (50), to ensure lower sparsity, which contributes to model stability, as well as greater classification accuracy, despite providing a less specific biological interpretation. Second, in the common taxa, we examined associations of bacterial abundance (explanatory variable) with individual DNAm levels (M-value, response variable) by linear regression models. The models were adjusted for covariates and surrogate variables, using the R meffil package (51), to mitigate the potential effects of factors that were also associated with DNAm levels. The covariates included sex, age, race and ethnicity, prematurity, C-section delivery, breastfeeding, cigarette smoke exposure at home, antibiotic use, and RSV infection at hospitalization as well as inferred seven blood immune cell types—neutrophils, eosinophils, monocytes, natural killer (NK) cells, helper T cells, cytotoxic T cells, and B cells—inferred by the DNAm data using the R EpiDISH package (52). Of the seven, we excluded NK cells from the covariates to avoid perfect collinearity. The surrogate variables represented variations in the DNAm data that were independent of the bacterial abundance and the covariates in the linear regression models (e.g., batch effects, cell type effects) and were estimated by using the R SmartSVA package (53). Third, by using the P-values, we computed the Benjamin-Hochberg false discovery rate (FDR) that allows for the interpretation of statistical significance in the context of multiple hypothesis testing (54). The procedure was separately applied for each taxon to account interdependency of abundances among hierarchically related taxa (e.g., Streptococcaceae, Streptococcus, Streptococcus pneumoniae) to mitigate the risk of type II errors. We determined CpGs with FDR < 0.05 as microbiota-associated differentially methylated CpGs (mbDMCs). Fourth, to confirm the robustness of using a pseudocount of 1 in the CLR transformation, we performed a sensitivity analysis by repeating mbEWAS for three asthma-related pathogenic bacteria using pseudocounts of 0.1 and 0.5, and computed Pearson’s correlation coefficients (r) based on the effect sizes. Sixth, by combining the spatially correlated P-values, we identified microbiota-associated differentially methylated regions (mbDMRs) by the comb-p method (55) using the R ENmix package (56). mbDMRs were identified as regions with a multiple testing corrected Šidák P-value below 0.05 and covering at least 5 CpG sites. Seventh, to ensure that findings from the comb-p method in our main analysis were comparable, we conducted a sensitivity analysis using an alternative DMR finding method—DMRcate (57). Lastly, we visualized the regional differences between P-values for CpGs located within and ±10 kb around the mbDMRs.

Computing DNAm age and determining associations with nasopharyngeal microbiota and outcomes

To examine biological aging in infants, we computed DNAm ages (year) by Horvath’s formula (58), the most widely cited method, with 334 CpGs and Wu’s (59) formula, which is specific to children, with 107 CpGs using the R methylclock package (60). Between Horvath’s and Wu’s formula, 11 CpGs overlapped (Figure S6). We also computed DNAm age acceleration (i.e., residual) by regressing DNAm age on chronological age in linear regression models. Additionally, we examined associations of bacterial abundance with DNAm age acceleration by linear regression models adjusted for the covariates—sex, age, race and ethnicity, prematurity, C-section delivery, breastfeeding, cigarette smoke exposure at home, antibiotic use, and RSV infection at hospitalization. We also examined associations of DNAm age acceleration with the risk of developing the outcomes by Cox proportional hazards models for recurrent wheeze and logistic regression models for asthma. The models were adjusted for the potential confounders—sex, age, race and ethnicity, prematurity, C-section delivery, breastfeeding, cigarette smoke exposure at home, parental history of asthma, antibiotic use, and RSV infection at hospitalization.

Examining differentially methylated cell types, associations of blood DNAm with serum proteins, and pathway enrichment of the proteins

In the three asthma-related nasopharyngeal pathogenic bacteria, we investigated whether CpGs within the mbDMRs were differentially methylated in the inferred seven blood immune cell types by CellDMC function from the R EpiDISH package (52) using DNAm levels (beta value). Also, we examined the relationship between the blood DNAm changes and systemic responses in a subset of the analytic cohort with available serum proteome data (n = 112). We first determined the relationships of DNAm levels (M-value) at each individual CpG site within the mbDMRs with circulating protein levels by using linear regression models adjusted by the covariates—sex, age, race and ethnicity, antibiotic use, prematurity, C-section delivery, breastfeeding, cigarette smoke exposure at home, antibiotic use, and RSV infection at hospitalization. We determined proteins with FDR <0.05 as mbDMR-related proteins. Lastly, we examined the enrichment of the mbDMR-related proteins with UniProt IDs in pathways based on Biological Process in Gene Ontology (GO) by an over-representation analysis using the R clusterProfiler package (61), setting the included proteins as the background. Of the analyzed 1,950 pathways, we focused on 1,014 pathways categorized under the descendants of GO:0002376 immune system process, GO:0008152 metabolic process, GO:0016032 viral process, and GO:0050896 response to stimulus.

Results

Of the 1,014 infants enrolled in the MARC-35 cohort, the current study focused on 504 infants with nasopharyngeal microbiome and blood DNAm data (Figure S2). The analytic and non-analytic cohorts did not differ in baseline patient characteristics (p ≥.05; Table S2), except for age, breastfeeding, body weight, and RSV infection. Among the analytic cohort, the median age was 3 (interquartile range [IQR] 2–6) months, 61% were male, and 22% were non-Hispanic white. Overall, 76% of infants had RSV, 28% developed recurrent wheeze by age 3 years, and 19% developed asthma by age 6 years (Table 1). With nasopharyngeal microbiome profiling, a total of 11 phyla, 21 classes, 39 orders, 68 families, 102 genera, and 225 species were identified. Of these, the following 23 common taxa were used in the subsequent analyzes: Phylum: Actinobacteria, Firmicutes, Proteobacteria; Class: Actinobacteria, Bacilli, Betaproteobacteria, Gammaproteobacteria; Order: Lactobacillales, Pasteurellales, Pseudomonadales; Family: Carnobacteriaceae, Moraxellaceae, Pasteurellaceae, Streptococcaceae; Genus: Haemophilus, Moraxella, Streptococcus; Species: Haemophilus influenzae, Moraxella catarrhalis, Moraxella nonliquefaciens, Streptococcus mitis, Streptococcus oralis, Streptococcus pneumoniae (Figure S7). With blood DNAm profiling, a total of 863,904 CpG sites were measured. Of these, 776,307 CpG sites passed stringent QC and were included in the subsequent analyzes (Figure S3). With serum proteome profiling, of the 1,196 proteins identified, 1,076 unique proteins (measured in 1,108 Olink assays) were included in the subsequent analyzes.

Table 1.

Baseline patient characteristics and clinical course of infants hospitalized for bronchiolitis

Characteristics	Overall (n = 504)
Demographics
Age (month), median (IQR)	3 (2–6)
Male sex	305 (61)
Race/ethnicity
Hispanic	150 (30)
Non-Hispanic white	109 (22)
Non-Hispanic black	228 (45)
Other	17 (3)
C-section delivery	321 (64)
Prematurity (32–36.9 weeks)	89 (18)
History of eczema	82 (16)
Previous breathing problems (count)
1	82 (16)
≥2	29 (6)
Number of other children at home
1	185 (37)
≥2	206 (41)
Mostly breastfed during 0–2.9 months	195 (39)
Ever attended daycare	126 (25)
Cigarette smoke exposure at home	85 (17)
Antibiotic use*	160 (32)
Corticosteroid use*	72 (14)
Maternal age at delivery (year), median (IQR)	27 (23–32)
Maternal antibiotics use during pregnancy	138 (27)
Parental history of eczema	101 (20)
Parental history of asthma	164 (33)
Clinical presentation at index hospitalization
Weight (kg), median (IQR)	6.2 (5.0–8.0)
Respiratory rate (per minute), median (IQR)	48 (40–60)
Oxygen saturation
<88%	23 (5)
88–89.9%	19 (4)
90–93.9%	82 (16)
≥94%	369 (73)
Blood testing
Blood eosinophilia (≥4%)	28 (8)
Total IgE (kU/L), median (IQR)	5.0 (1.9–15.8)
Allergic (specific IgE) sensitization	107 (21)
Food sensitization	102 (20)
Aeroallergen sensitization	7 (1)
Viral testing
RSV positive	382 (76)
RV positive	94 (19)
Clinical course
Positive pressure ventilation use†	23 (5)
Recurrent wheeze by age 3 years‡	140 (28)
Asthma by age 6 years§	94 (19)

Note: Data are the number (percentage) of children unless otherwise indicated. Percentages may not equal 100 because of rounding and missingness.

Abbreviations: IgE, immunoglobulin E; IQR, interquartile range; RSV, respiratory syncytial virus; RV, rhinovirus.

^*Defined as the use of systemic corticosteroids or antibiotics between birth and the index hospitalization.

^†Defined as the use of invasive and/or non-invasive mechanical ventilation (e.g., continuous positive airway pressure ventilation) during the index hospitalization.

^‡Recurrent wheeze was defined as having at least two corticosteroid-requiring exacerbations in six months or at least four wheezing episodes in one year that last at least one day and affect sleep.

^§Asthma was defined as physician-diagnosis of asthma by age 6 years, plus either asthma medication use (e.g., albuterol inhaler, inhaled corticosteroids, montelukast) or asthma-related symptoms in the preceding year.

Nasopharyngeal microbiome-epigenome-wide analysis identified mbDMCs and mbDMRs

In the mbEWAS, the confounding and batch effects were well-controlled, with minimal inflation in the common taxa (median λ genomic control = 1.00 [IQR 0.99–1.02]; Figure S8A). Among the 23 common taxa, a total of one mbDMC—cg16594639, chr20: 39528675—was identified in S. pneumoniae (FDR < 0.05; Figure S8B and Supporting Information S1). The mbDMC, along with CpGs with $p<5.0\times 10^{-6}$ (Table S3), did not show multimodal distributions (Figure S9). Additionally, these associations were consistent when CLR-transformed abundances with pseudocounts of 0.1 and 0.5 were used, in three asthma-related nasopharyngeal pathogenic bacteria—H. influenzae, M. catarrhalis, and S. pneumoniae—which were included among the 23 common taxa (Figure S10). Furthermore, in the region-based analysis, 96 mbDMRs (median 10 mbDMRs per taxon [IQR 7–12]) were identified by using the comb-p method, while the mbDMC was not included in the mbDMRs. Of the mbDMRs, 15 mbDMRs were assigned to gene symbols that overlapped with the symbols identified by using the DMRcate method (e.g., S. pneumoniae, chr5:27038497–27038802, CDH9; chr6:48068669–48068940, PTCHD4; Šidák p-value < 0.05 [comp-b], FDR < 0.05 [DMRcate]; Table 2 [for the three asthma-related nasopharyngeal pathogenic bacteria], Supporting Information S2, Table S4, Figures. S11 and S12).

Table 2.

mbDMRs in blood from infants with severe bronchiolitis in three asthma-related nasopharyngeal pathogenic bacteria

Taxa	mbDMRs* (chromosome:position)	Width (bp)	N of CpGs	Šidák P-value	Nearest genes	Direction of effect†	Overlap with promoter	Distance to TSS (bp)	Evidence for open chromatin‡
Haemophilus	chr5:1867863–1868579	717	7	6.86×10−12	IRX4	+	No	13982	No
influenzae	chr1:236523172–236523549	378	9	3.94×10−08	EDARADD	+	No	31130	No
	chr20:45308212–45308649	438	11	1.23×10−07	SLC13A3	-	No	4175	No
	chr22:45413764–45414162	399	5	8.65×10−06	PHF21B	-	No	4954	Yes
	chr9:122226772–122227271	500	6	5.68×10−05	BRINP1	+	No	92032	No
	chr15:93073720–93073939	220	8	4.32×10−04	C15orf32	+	No	41254	Yes
	chr10:35605083–35605502	420	7	1.61×10−03	CCNY	-	No	17299	No
	chr17:77793196–77793447	252	5	3.08×10−03	CBX4	-	No	19280	No
	chr7:154892616–154892919	304	5	3.96×10−03	HTR5A	-	No	26348	No
	chr21:34886881–34887163	283	5	9.56×10−03	GART	+	No	19857	Yes
	chr21:35831870–35832180	311	9	9.92×10−03	KCNE1	-	Yes	0	No
	chr7:134779371–134779544	174	5	1.16×10−02	CYREN	+	No	50221	No
	chr1:1013244–1013513	270	7	1.90×10−02	RNF223	+	No	556	Yes
	chr8:143549597–143549751	155	5	2.20×10−02	ADGRB1	-	No	3920	Yes
Moraxella	chr8:103371204–103371502	299	6	7.83×10−06	UBR5	+	No	53115	Yes
catarrhalis	chr1:54805999–54806254	256	7	2.50×10−04	SSBP3	+	No	65514	No
	chr20:31637862–31638048	187	5	4.87×10−04	BPIFB3	+	No	2181	No
	chr8:94949946–94950235	290	6	3.92×10−03	PDP1	+	No	15358	No
	chr5:79069823–79070007	185	7	1.10×10−02	CMYA5	+	No	83864	No
Streptococcus	chr1:3858262–3858770	509	8	1.78×10−08	LINC01134	+	No	38404	No
pneumoniae	chr5:27038497–27038802	306	7	5.41×10−08	CDH9	+	Yes	0	No
	chr17:60422317–60422550	234	7	4.11×10−07	EFCAB3	-	No	22028	No
	chr6:48068669–48068940	272	8	4.39×10−05	PTCHD4	+	No	9703	No
	chr19:5838723–5838999	277	8	6.85×10−04	FUT6	-	Yes	0	Yes
	chr19:55461589–55461971	383	6	7.71×10−04	NLRP7	+	Yes	0	Yes
	chr4:73981992–73982299	308	7	3.29×10−03	ANKRD17	+	No	37231	No
	chr7:120989349–120989649	301	5	3.63×10−03	FAM3C	+	No	19428	No
	chr3:46717947–46718208	262	7	1.54×10−02	ALS2CL	-	No	391	No
	chr17:7307335–7307487	153	7	2.91×10−02	TMEM256	+	Yes	0	Yes

Abbreviations: bp: base pair; mbDMR: microbiome-associated differentially methylated regions; TSS: transcription starting site.

^*The DMRs were identified by the comb-p. A region was identified as a DMR if it had a Šidák P-value <0.05 and at least 5 CpG sites.

^†Direction of effect was obtained for the main effect of nasopharyngeal bacterial abundance by computing the average of coefficients across the CpG sites.

^‡Evidence for open chromatin was obtained by single-cell ATAC-seq peak in human peripheral blood mononuclear cells from a healthy donor.

Nasopharyngeal microbiota was associated with DNAm age deceleration

In infants with severe bronchiolitis, the mean DNAm age acceleration was −0.036 years (standard deviation [SD], 0.342 years) in Horvath’s formula and −0.077 years (SD, 0.385 years) in Wu’s formula. Among the common taxa, higher abundances in 4 taxa were associated with Wu’s DNAm age deceleration (e.g., H. influenzae; $-1.46\times 10^{-2}$ years for each CLR-transformed abundance increment; 95% CI, $-2.11\times 10^{-2}–-8.17\times 10^{-3}$; FDR < 0.001; Table S5) while no taxa were associated with Horvath’s DNAm age acceleration (FDR ≥0.05, Table S5). Additionally, Wu’s DNAm age acceleration was associated with a lower risk of developing asthma by age 6 years (adjusted odds ratio, 0.34 for one-year DNAm age acceleration increment; 95% CI, 0.15–0.72, p = 0.006; Table S6).

Cell-type specific associations within mbDMRs, mbDMR-related proteins, and the pathway enrichment in asthma-related nasopharyngeal pathogenic bacteria

Of the 96 mbDMRs, 29 mbDMRs were identified in the three asthma-related nasopharyngeal pathogenic bacteria. Additionally, CpGs within the 29 mbDMRs were suggestive of differentially methylated across inferred blood immune cell types (nominal p < 0.05, Figure 2). For example, in S. pneumoniae, CpGs within CDH9 were hypermethylated in eosinophils and CpGs within NLRP7 were hypomethylated in monocytes. Additionally, the DNAm levels in the CpGs within 29 mbDMRs were associated with levels of 156 proteins (i.e., mbDMR-related proteins, FDR < 0.05, Figure 3). For example, DNAm levels in CpGs within mbDMRs (e.g., S. pneumoniae, EFCAB3 and FUT6) were negatively associated with MMP9 (matrix metallopeptidase 9) levels and were positively associated with serum XCL1 (lymphotactin) levels. Furthermore, the 148 mbDMR-related proteins with UniProt IDs were enriched in immune response-related pathways (e.g., regulation of ERBB signaling pathway, eosinophil chemotaxis and migration pathways, FDR < 0.10, Table 3). Lastly, our identified links between CpGs within mbDMRs and mbDMR-related proteins have not been reported in previous protein quantitative trait methylation studies (62,63) in adults with a different proteome assay—SOMAscan (SomaLogic Operating Co., Inc., CO, USA). In contrast, of the 156 mbDMR-related proteins, 44 proteins (e.g., MMP9, XCL1) have been identified in our previous EWAS for recurrent wheeze and childhood asthma as proteins related to the DMRs (13) (Figure 3). Similarly, 12 proteins (e.g., MMP9) have been identified in two recent protein-wide association studies using SOMAscan for adult asthma (64,65).

Figure 2. Cell-type specific associations of three asthma-related nasopharyngeal pathogenic bacteria with blood DNA methylation at CpGs within mbDMRs.

The heatmap summarizes the associations of three nasopharyngeal asthma-related nasopharyngeal pathogenic bacteria (i.e., Haemophilus influenzae, Moraxella catarrhalis, Streptococcus pneumoniae) abundances with DNA methylation (DNAm) levels (i.e., beta values) at cytosine-phosphate-guanine (CpG, cg) sites within the microbiome-associated differentially methylated regions (mbDMRs, annotated by their nearest gene symbols) identified in Table 2, according to blood cell types. The seven blood immune cell types—neutrophils, eosinophils, monocytes, natural killer (NK) cells, helper T cells, cytotoxic T cells, and B cells—were inferred by a deconvolution analysis using blood DNAm data. We identified suggestively differentially methylated CpG sites with statistically significant cell-type-specific effects (nominal P <0.05) within the DMRs and included the mbDMRs with two or more significant effects in the heatmap. The tab colors represent effect estimates. The effect estimates in the overall column were determined in microbiome epigenome wide association study using DNAm levels—M-values (Figure S8B, Table S3, and Supporting Information S1). Daggers in the tabs denote nominal P-values.

Figure 3. Relationships between blood DNA methylation within mbDMRs in asthma-related nasopharyngeal pathogenic bacteria and serum proteins.

The heatmap summarizes the relationships of blood DNA methylation (DNAm) levels (i.e., M-values) at cytosine-phosphate-guanine (CpG, cg) sites within microbiome-associated differentially methylated regions (mbDMRs, annotated by their nearest gene symbols), in three asthma-related nasopharyngeal pathogenic bacteria, with serum protein levels (annotated by Olink ID followed by protein symbols). Of the CpG-protein pairs with statistically significant associations (false discovery rate [FDR] <0.05, mbDMR-related proteins), we included the mbDMRs with 2 or more significant effects and the proteins with 5 or more significant effects in the heatmap. No mbDMR-related proteins were identified in Moraxella catarrhalis. Serum proteins were clustered and sorted based on the similarities of the effect estimates by hierarchical clustering using Euclidean distances and the Ward method. In the heatmap tabs, colors represent effect estimates, and daggers denote FDR.

The tabs on the left side of the heatmap represent serum proteins previously reported to be associated with asthma in proteome wide association studies (Asamoah et al., EBioMedicine, 2024, https://doi.org/10.1016/j.ebiom.2023.104936; Smilnak et al., Allergy. 2024, https://doi.org/10.1111/all.16000), as well as in a previous study that used the same blood DNAm data from the MARC-35 study (Li et al., Nat Commun. 2025, https://doi.org/10.1038/s41467-025-57288-6).

Table 3.

Pathway enrichments for mbDMR-related proteins in three asthma-related nasopharyngeal pathogenic bacteria

Term	Pathway description	N of protein*	P-value	FDR	Matching protein symbols in the pathway
GO:0006955	immune response	63/335	1.91×10−04	7.77×10−02	CRTAM, GZMA, CSF3, CD200R1, KYNU, IL1RN, XCL1, CEACAM8, PTX3, CCL3, MARCO, IL7, TNFSF10, OSM, CCL4, CCL4L1, CCL11, TNFSF14, TNFSF11, IL12B, CCL23, S100A12, CCL25, LTA, CCL2, GRN, AZU1, IL1R1, PRTN3, IL1RL1, CCL16, CEACAM1, FASLG, CD27, CD160, TNFRSF4, GZMB, ANXA1, CLEC4C, CD28, CLEC4D, PIK3AP1, SH2D1A, LAMP3, TANK, LAG3, YES1, NCF2, CLEC5A, ARG1, DEFA1, DEFA1B, LCN2, CBL, SKAP1, SEMA7A, CD58, ENPP2, CDH17, FCAR, CCL27, IL32, TNF
GO:0048247	lymphocyte chemotaxis	12/32	3.39×10−04	7.77×10−02	XCL1, CCL3, CCL4, CCL4L1, CCL11, TNFSF14, CCL23, CCL25, CCL2, CCL16, DEFA1, DEFA1B
GO:1901184	regulation of ERBB signaling pathway	8/16	3.56×10−04	7.77×10−02	TGFA, MMP9, EGFR, CEACAM1, EGF, FASLG, CBL, LGMN
GO:0006952	defense response	62/338	5.03×10−04	7.77×10−02	CRTAM, CD200R1, KYNU, IL1RN, OLR1, XCL1, PTX3, CCL3, MARCO, OSM, CCL4, CCL4L1, CCL11, TNFSF11, HGF, IL12B, CCL23, S100A12, CCL25, LTA, MMP9, CCL2, GRN, CDH5, SELE, AZU1, MPO, IL1R1, PRTN3, IL1RL1, CCL16, CEACAM1, CD48, FASLG, EPHA2, CD160, TNFRSF4, GZMB, ANXA1, TPSAB1, CLEC4C, CD28, CLEC4D, PIK3AP1, SH2D1A, TANK, LAG3, YES1, NCF2, CLEC5A, GHRL, ARG1, DEFA1, DEFA1B, LCN2, ITGB1, SEMA7A, CD58, PLA2G10, IL32, LSP1, TNF
GO:0071347	cellular response to interleukin-1	13/38	5.43×10−04	7.77×10−02	IL1RN, XCL1, CCL3, CCL4, CCL4L1, CCL11, CCL23, CCL25, CCL2, IL1R1, MMP2, CCL16, TANK
GO:0009620	response to fungus	8/17	5.98×10−04	7.77×10−02	PTX3, S100A12, MPO, CLEC4C, CLEC4D, ARG1, DEFA1, DEFA1B
GO:0070555	response to interleukin-1	15/48	6.12×10−04	7.77×10−02	IL1RN, XCL1, CCL3, CCL4, CCL4L1, CCL11, CCL23, CCL25, CCL2, SELE, IL1R1, MMP2, CCL16, ANXA1, TANK
GO:0048245	eosinophil chemotaxis	9/21	6.27×10−04	7.77×10−02	XCL1, CCL3, CCL4, CCL4L1, CCL11, CCL23, CCL25, CCL2, CCL16
GO:0002548	monocyte chemotaxis	14/44	7.66×10−04	7.77×10−02	PDGFB, XCL1, CCL3, CCL4, CCL4L1, CCL11, TNFSF11, CCL23, S100A12, CCL25, CCL2, CCL16, ANXA1, LGMN
GO:0071346	cellular response to type II interferon	14/44	7.66×10−04	7.77×10−02	XCL1, CCL3, CCL4, CCL4L1, CCL11, IL12B, CCL23, CCL25, CCL2, CCL16, FASLG, ARG1, CD58, TNF
GO:0072677	eosinophil migration	9/22	9.44×10−04	8.29×10−02	XCL1, CCL3, CCL4, CCL4L1, CCL11, CCL23, CCL25, CCL2, CCL16
GO:0050727	regulation of inflammatory response	23/93	9.82×10−04	8.29×10−02	CD200R1, XCL1, CCL3, OSM, TNFSF11, HGF, IL12B, S100A12, LTA, MMP9, GRN, CDH5, SELE, IL1R1, IL1RL1, ANXA1, CD28, PIK3AP1, YES1, GHRL, SEMA7A, PLA2G10, TNF
GO:0045087	innate immune response	36/172	1.09×10−03	8.49×10−02	CRTAM, KYNU, XCL1, PTX3, CCL3, MARCO, CCL4, CCL4L1, CCL11, IL12B, CCL23, S100A12, CCL25, CCL2, GRN, CCL16, CEACAM1, FASLG, CD160, GZMB, ANXA1, CLEC4C, CLEC4D, PIK3AP1, SH2D1A, TANK, LAG3, YES1, NCF2, CLEC5A, ARG1, DEFA1, DEFA1B, LCN2, CD58, TNF
GO:0034341	response to type II interferon	15/51	1.25×10−03	8.63×10−02	KYNU, XCL1, CCL3, CCL4, CCL4L1, CCL11, IL12B, CCL23, CCL25, CCL2, CCL16, FASLG, ARG1, CD58, TNF
GO:0098542	defense response to other organism	42/212	1.28×10−03	8.63×10−02	CRTAM, KYNU, XCL1, PTX3, CCL3, MARCO, CCL4, CCL4L1, CCL11, IL12B, CCL23, S100A12, CCL25, LTA, CCL2, GRN, AZU1, MPO, PRTN3, CCL16, CEACAM1, FASLG, EPHA2, CD160, GZMB, ANXA1, CLEC4C, CLEC4D, PIK3AP1, SH2D1A, TANK, LAG3, YES1, NCF2, CLEC5A, ARG1, DEFA1, DEFA1B, LCN2, CD58, PLA2G10, TNF
GO:0042058	regulation of epidermal growth factor receptor signaling pathway	7/15	1.44×10−03	9.12×10−02	TGFA, MMP9, EGFR, CEACAM1, EGF, FASLG, CBL

Abbreviations: FDR, false discovery rate; GO, Gene Ontology.

Enriched pathways were identified by applying the 156 microbiome-associated differentially methylated region (mbDMR)-related proteins of three asthma-related nasopharyngeal pathogenic bacteria (i.e., Haemophilus influenzae, Moraxella catarrhalis, Streptococcus pneumonia) to the over-representation analysis based on Biological Processes in GO. Pathways categorized under the descendants of GO:0002376 immune system process, GO:0008152 metabolic process, GO:0016032 viral process, and GO:0050896 response to stimulus with FDR <0.10 were selected.

^*Listed as the number of observed / background proteins.

Discussion

In a multicenter prospective cohort study of infants with severe bronchiolitis, we applied the EWAS approach to nasopharyngeal microbiome and blood DNAm data. We identified 1 mbDMC—S. pneumoniae, cg16594639, chr20: 39528675) and 96 mbDMRs (e.g., S. pneumoniae, chr5:27038497–27038802, CDH9; chr6:48068669–48068940, PTCHD4) and an association of a higher H. influenzae abundance with DNAm age deceleration. Additionally, in the three asthma-related nasopharyngeal pathogenic bacteria, by using serum proteome data, we determined mbDMR-related proteins (e.g., MMP9, XCL1) and protein enrichment in immune response pathways (e.g., ERBB and eosinophil-related pathways). To the best of our knowledge, this is the first investigation of mbEWAS using nasopharyngeal microbiome and blood DNAm data.

To examine the pathobiology of bronchiolitis underlying its acute and chronic respiratory sequelae, recent research has suggested important roles of airway microbiota and host epigenome. Our mbEWAS using nasopharyngeal microbiome and blood DNAm data is in agreement with studies that have evaluated the associations of microbiota with host DNAm signatures as well as airway diseases (e.g., allergic rhinitis or asthma development) (17,20,21,50). For example, a birth-cohort study has shown that hypopharyngeal microbial richness at age 1 month was negatively associated with levels of a DNAm module, comprising DMCs for allergic rhinitis, in the nasal cavity at age 6 years and that the richness was higher in children with allergic rhinitis at age 6 years (20). Another birth-cohort study has reported that higher levels of a DNAm module in the cord blood mononuclear cells, comprising DMCs for lower interferon-γ:IL-13 ratio in their mothers’ prenatal serum, was associated with nasal microbial composition (e.g., lower Moraxella abundance) at age 2 years and with a higher risk of developing asthma at age 2–9 years (21). Additionally, a mbEWAS has identified 2 mbDMCs using gut microbiome and blood DNAm data in adults (50). The current study builds on these earlier reports and extends them by conducting airway mbEWAS. A better understanding of airway microbiome-DNAm associations should offer an evidence-base for prevention strategies against the chronic respiratory sequelae (e.g., asthma) in infants with severe bronchiolitis.

Although the exact mechanisms underlying the associations of nasopharyngeal microbiota with blood DNAm warrant further investigation, the current study identified mbDMC—S. pneumoniae, cg16594639) and mbDMRs (e.g., S. pneumoniae, chr5:27038497–27038802, CDH9, cadherin 9; chr6:48068669–48068940, PTCHD4—a suppresser of the hedgehog signaling pathway, a homolog of PTCH1 (66)). Although no previous studies have reported the biological role or relevance of cg16594639, growing evidence suggests the link between bacteria and cadherin and hedgehog signaling as well as asthma development (67–71). For example, a recent study has reported that S. pneumoniae that infected living human lung tissue degraded cadherin in vascular endothelium (67). Another study has reported the parent-of-origin effect of rs10050550—a tag-SNP of CDH9—with childhood asthma (68). Additionally, the previous literature has shown that conventionalized mice had a lower expression of ptch1 compared to germ-free mice in the small intestine (69). Furthermore, a GWAS in adult asthma has demonstrated that SNPs near PTCHD4 were validated across 3 cohorts and reached genome-wide significance in a meta-analysis for oral corticosteroid bursts (71). These data collectively suggest the relationships between airway microbiota, host epigenetic, and asthma development.

We also observed that a higher H. influenzae abundance was associated with DNAm age deceleration in Wu’s formula, which is specific to children, and that Wu’s DNAm deceleration in Wu’s formula was associated with a higher risk of developing asthma by age 6 years. Consistently, recent evidence supports the relationships between microbiota, DNAm age, and asthma (72–74). For example, a two-sample Mendelian randomization study has revealed causal effects of lower gut microbiota abundances (e.g., Hemophilus) with DNAm age deceleration (72). Additionally, a birth-cohort study has shown that DNAm age deceleration, estimated by the Horvath’s formula, was associated with a lower prevalence of current asthma in blood from mid-childhood children (73) and in nasal cavity from early adolescents (74). These findings suggest that DNAm age deceleration may favorably influence the microbiota and asthma development in school-age children and are inconsistent with our findings in infants. However, at least during the neonatal period, another birth-cohort study has reported an unfavorable influence—DNAm gestational age deceleration in cord blood was observed in neonates exposed to maternal pregnancy complications (e.g., gestational diabetes) compared to unexposed (75). Taken together, these studies indicate the important roles of DNAm deceleration (i.e., immaturity) during early infancy in airway microbiota and asthma development.

Our data also identified, in the three asthma-related nasopharyngeal pathogenic bacteria, mbDMR-related proteins (e.g., MMP9—a protease involved in the degradation of extracellular matrix collagen, XCL1—a chemokine involved in lymphocyte recruitment and enhanced antigen presentation (76)). Additionally, we identified protein enrichment in immune response-related pathways (e.g., regulation of ERBB signaling [involving MMP9], and eosinophil chemotaxis and migration pathways [involving XCL1]). Consistently, recent evidence suggests important roles of asthma-related nasopharyngeal pathogenic bacteria in regulating host responses and asthma (77–80). For example, the previous literature on MMP9 has shown that S. pneumoniae induced MMP9 production in peripheral bone marrow cells from healthy adults (81) and that MMP9 levels in bronchoalveolar (BAL) fluids from adults with severe asthma were higher than in those from control subjects (82). Additionally, research on ERBB2 (also known as HER2) has reported distinct microbial profile between HER2+ and HER2- breast cancer (83). Furthermore, a GWAS for adult asthma has also reported the association of rs869402—a tag-SNP of ERBB2, located in the 17q12 within the 17q12–21, the most widely replicated locus for asthma (84,85). Regarding XCL1 and eosinophils, a recent study has shown that a Bordetella bronchiseptica mutant increased secretion of XCL1 in murine bone marrow-derived eosinophils (76), a key player in childhood asthma. Altogether, these studies support that airway microbiota and blood DNAm integrally contribute to systemic immune-related responses and chronic respiratory sequelae (e.g., asthma).

Although the current study carefully prepared omics data (e.g., thorough QCs), performed mbEWAS with considerable efforts to mitigate bias (e.g., adjusting for deconvoluted cell types), and represents the first airway mbEWAS, the current study has several potential limitations. First, the cross-sectional design limits our ability to determine the exact causal relationship between the airway microbiota, blood DNAm, and serum protein signatures. Therefore, these relationships warrant further investigation in future longitudinal studies. Second, although bronchiolitis involves the microbiota and inflammation in both upper and lower airways, this study relied on upper airway data. However, research in children has demonstrated that upper airway data reliably represent the microbiome profile of lower airway (86), which are difficult to collect in infants due to their invasiveness. Third, DNAm profiling was performed using blood specimens, limiting our inferences to other specimens, particularly the airway. However, an epigenome-wide meta-analysis for childhood asthma has shown that DMCs in blood DNAm data were also identified in airway DNAm data (22,23). Fourth, our mbEWAS, with a considerable sample size of 504, identified only one mbDMC—similar to a previous gut-mbEWAS (50) that reported only two mbDMCs. This sample size may not be sufficient to identify additional mbDMCs with relatively small effect sizes. In addition, although our nasopharyngeal microbiome and blood DNAm data were collected at the same time point, the limited findings on the mbDMC may indicate a delayed timing of the blood DNAm establishment responding to the airway microbiota colonization or infection, which needs to be validated by future experimental studies. Fifth, in the identified mbDMR-related proteins, the mbDMRs were not in the genetic regions coding for the proteins. Therefore, DNAm within the mbDMRs may indirectly affect protein levels by modulating other proteins upstream. Sixth, although this study lacked both mechanistic experiments and independent replication cohorts, due to the unavailability of such cohorts, our clinically and biologically plausible findings generate hypotheses that warrant the validation for the mbDMCs, mbDMRs, and mbDMR-related proteins. Lastly, although our study sample consisted of racially, ethnically, and geographically diverse infants, all children were hospitalized for severe bronchiolitis. This case-only design may limit the internal and external validity of our findings. Regarding internal validity, our findings may be biased due to conditioning on hospitalization for bronchiolitis (i.e., collider bias (87)) since hospitalization could be influenced by both airway microbiota and blood DNAm. Regarding external validity, our inferences must be cautiously generalized beyond infants with severe bronchiolitis. Nevertheless, our findings remain highly relevant to >110,000 infants hospitalized for bronchiolitis annually in the U.S. (1). To address this limitation, future mbEWAS conducted in the general population is warranted.

In conclusion, by using the EWAS approach in a multicenter cohort of infants with severe bronchiolitis, we identified associations of the nasopharyngeal microbiota with blood DNAm signatures, especially in the three asthma-related nasopharyngeal pathogenic bacteria. For example, in S. pneumoniae, we identified one mbDMC—cg16594639, chr20: 39528675—and mbDMRs (e.g., S. pneumoniae, chr5:27038497–27038802, CDH9; chr6:48068669–48068940, PTCHD4) and the inverse association with DNAm age acceleration. Additionally, we determined the mbDMR-related proteins (e.g., MMP9, XCL1) and the enrichment in immune response pathways (e.g., ERBB and eosinophil-related pathways). Our findings should facilitate further research into the relationship between the airway microbiota, host epigenome, systemic responses, and disease pathobiology of infant bronchiolitis. Furthermore, this facilitation will advance the development of targeted therapeutic measures (e.g., airway microbiota modification influencing DNAm and related immune responses) against the chronic respiratory sequelae (e.g., asthma) (88) in this large patient population with a high morbidity burden.

Data availability:

The blood DNAm data that support the findings of this study will be available on the NIH/NIAID ImmPort under Accession ID: SDY2306 through controlled access to be compliant with the informed consent forms of MARC-35 study and the genomic data sharing plan. Additionally, the mbEWAS summary statistics generated in this study are available in Supplementary Data 1.