DNA methylation in nasal epithelium, atopy, and atopic asthma in children: a genome-wide study

Abstract

Background: Epigenetic mechanisms could alter the airway epithelial barrier and ultimately lead to atopic diseases such as asthma. We aimed to identify DNA methylation profiles that are associated with-and could accurately classify-atopy and atopic asthma in school-aged children.

Methods: We did a genome-wide study of DNA methylation in nasal epithelium and atopy or atopic asthma in 483 Puerto Rican children aged 9-20 years, recruited using multistage probability sampling. Atopy was defined as at least one positive IgE (≥0·35 IU/mL) to common aeroallergens, and asthma was defined as a physician's diagnosis plus wheeze in the previous year. Significant (false discovery rate p<0·01) methylation signals were correlated with gene expression, and top CpGs were validated by pyrosequencing. We then replicated our top methylation findings in a cohort of 72 predominantly African American children, and in 432 children from a European birth cohort. Next, we tested classification models based on nasal methylation for atopy or atopic asthma in all cohorts.

Findings: DNA methylation profiles were markedly different between children with (n=312) and without (n=171) atopy in the Puerto Rico discovery cohort, recruited from Feb 12, 2014, until May 8, 2017. After adjustment for covariates and multiple testing, we found 8664 differentially methylated CpGs by atopy, with false discovery rate-adjusted p values ranging from 9·58 × 10^-17 to 2·18 × 10^-22 for the top 30 CpGs. These CpGs were in or near genes relevant to epithelial barrier function, including CDHR3 and CDH26, and in other genes related to airway epithelial integrity and immune regulation, such as FBXL7, NTRK1, and SLC9A3. Moreover, 28 of the top 30 CpGs replicated in the same direction in both independent cohorts. Classification models of atopy based on nasal methylation performed well in the Puerto Rico cohort (area under the curve [AUC] 0·93-0·94 and accuracy 85-88%) and in both replication cohorts (AUC 0·74-0·92, accuracy 68-82%). The models also performed well for atopic asthma in the Puerto Rico cohort (AUC 0·95-1·00, accuracy 88%) and the replication cohorts (AUC 0·82-0·88, accuracy 86%).

Interpretation: We identified specific methylation profiles in airway epithelium that are associated with atopy and atopic asthma in children, and a nasal methylation panel that could classify children by atopy or atopic asthma. Our findings support the feasibility of using the nasal methylome for future clinical applications, such as predicting the development of asthma among wheezing infants.

Funding: US National Institutes of Health.

Figure 1 –. Epigenome-wide association study (EWAS) of atopy in nasal epithelium

Top panel shows the Manhattan plot, with –log₁₀(P) on the y-axis and chromosome position in the x-axis; red line indicates FDP-P<0·01. Bottom panel shows the volcano plot, with –log₁₀(FDR P-value) in the y-axis and effect size in the x-axis. Hypermethylated CpGs (i.e. higher methylation level in participants with atopy compared to those with no atopy) are shown in red; hypomethylated CpGs in blue; non-significant (FDR P>0·01) in black.

Figure 2 –. Integration of Epigenome-wide Association Study (EWAS) and Transcriptome-wide Association Study (TWAS) results for atopy

Figure shows the -log₁₀(FDR P-value) for DNA methylation (EWAS) in the x-axis, and for gene expression (TWAS) in the y-axis; positive values indicate hypermethylation or over-expression, and negative values indicate hypomethylation or under-expression. Pairs where both EWAS and TWAS results are non-significant (FDR P>0·01) are shown in black. Significantly hypermethylated CpGs with under-expressed corresponding genes are shown in purple (right lower quadrant; 779 CpGs, 514 genes); hypomethylated and over-expressed genes shown in dark blue (left upper quadrant; 1506 CpGs, 815 genes).

Figure 3 –. Pathway analysis of nasal epigenomics and atopy

Left panel shows the top 20 enriched canonical pathways for our analyses of atopy. Blue bars depict -log(P-value) and orange symbols depict enrichment ratios. Right panel shows the overlap between these top 20 pathways. Only connections between pathways with ≥5 genes in common are shown. Pathway analysis performed including all genes that were significant (FDR P<0·01) in the EWAS, TWAS and eQTM analyses (n=724 genes). See supplementary table 6 for further details.

Figure 4 –. DNA methylation panel and classification/prediction of atopy in study cohorts

Top panels show the receiver operating characteristic (ROC) curves for Puerto Rico (left), Yang (middle), and PIAMA (right) using three different statistical approaches (GLMNET, GBM, RF; see Methods for details). Middle row shows heat maps for the three cohorts using the 30 CpG in the GLMNET model. Bottom row shows classification tables for atopy in the three cohorts. Results shown are from using the same 30-CpG panel, coefficients, and probability cut-offs trained from the Puerto Rico cohort data, then applied to both independent cohorts. Sens: sensitivity. Spec: specificity. PPV: positive predictive value. NPV: negative predictive value. Accuracy: (True positives + true negatives) / (total N).