Differences in the DNA Methylome of T cells in Adults With Asthma of Varying Severity

Abstract Background DNA methylation plays a critical role in asthma development, but differences in DNA methylation among adults with varying asthma severity or asthma endotypes are less well-defined. Objective To examine how DNA methylomic patterns differ among adults with asthma based on asthma severity and airway inflammation. Methods Peripheral blood T cells from 35 adults with asthma in Beijing, China were serially collected over time (130 samples total) and analyzed for global DNA methylation using the Illumina MethylationEPIC Array. Differential methylation was compared among subjects with varying airway inflammation and severity, as measured by fraction of exhaled nitric oxide, forced expiratory volume in one second (FEV1), and Asthma Control Test (ACT) scores. Results Significant differences in DNA methylation were noted among subjects with different degrees of airway inflammation and asthma severity. These differences in DNA methylation were annotated to genes that were enriched in pathways related to asthma or T cell function and included gene ontology categories related to MHC class II assembly, T cell activation, interleukin (IL)-1, and IL-12. Genes related to P450 drug metabolism, glutathione metabolism, and developmental pathways were also differentially methylated in comparisons between subjects with high vs low FEV1 and ACT. Notable genes that were differentially methylated based on asthma severity included RUNX3 , several members of the HLA family, AGT , PTPRC , PTPRJ , and several genes downstream of the JAK2 and TNF signaling pathway. Conclusion These findings demonstrate how adults with asthma of varying severity possess differences in peripheral blood T cell DNA methylation that contribute to the phenotype and severity of their overall disease.


Patient subjects
Patients > 18 years of age who lived in Beijing, China with a clinical diagnosis of asthma (as de ned by clinician assessment of individuals with a history of episodic symptoms of dyspnea, cough, or wheezing, a positive methacholine challenge test, and exclusion of other respiratory conditions that may mimic asthma) were recruited to participate in the study.Inclusion criteria included either a positive bronchodilation test with an increase in FEV1 of 12% and 200 ml from baseline after 400 µg albuterol inhalation or a positive methacholine challenge test with a decrease in FEV1 of 20% from baseline after methacholine inhalation with concentration < 4 mg/ml.Patients were required to be free of asthma exacerbation for at least one month prior to enrollment.Patients who were current smokers or had a greater than 5 pack year smoking history or have a history of chronic lung disease other than asthma (including COPD, lung cancer, active tuberculosis, or interstitial lung disease) were excluded.All patients had demographic information, asthma history, spirometry, measures of FeNO, and assessment of asthma symptoms, as measured by ACT, taken at baseline.Patients were followed longitudinally up to every month for 12 months.At follow-up visits, spirometry including FEV1, FeNO, and ACT were measured monthly and blood samples were taken at 0, 3, 6, 9, and 12 months of follow up.None of the patients had a recent asthma exacerbation or severe illness at the time the samples were taken.All patients provided written informed consent, and the study was approved by both the Peking University and University of Michigan Institutional Review Board.

Isolation of peripheral blood CD3 + T cells
Peripheral blood CD3 + T cells were isolated using anti-Human CD3 magnetic nanoparticles and a cell separation magnet.Brie y, blood was rst centrifuged and peripheral blood mononuclear cells were isolated using Ficoll-Paque PLUS (Cytiva, Cat No. 17144002, Marlborough, MA) and resuspended in PBS.
Cells were then labeled with anti-human CD3 Particles DM (Cat.No. 552593, BD Biosciences, Franklin Lakes, NJ) and separated using the Cell Separation Magnet (Cat.No. 552311, BD Biosciences).

DNA isolation and methylation analysis
Global DNA methylation patterns were analyzed using the In nium MethylationEPIC array (Illumina, Inc, San Diego, CA), which interrogates over 850,000 methylation sites at single-nucleotide resolution.DNA was isolated using the QlAmp DNA Micro Kit (Cat No. 56304, QIAGEN, Germany) and stored in -80°C before being shipped to the United States for bulk processing.DNA was quantitated using the Qubit high sensitivity DNA assay (ThermoFisherScienti c, Waltham, MA) and assessed for quality using the TapeStation genomic DNA kit (Agilent, Santa Clara, CA).For each sample, 250 ng of genomic DNA was bisul te converted using Zymo EZ DNA Methylation kit (Zymo Research, Irvine, CA) before being hybridized to the In nium MethylationEPIC BeadChip array for analysis.All raw data from the EPIC array were deposited in the National Center for Biotechnology Information Gene Expression Omnibus database under accession number GSE226257.

Bioinformatics and statistical analysis
All code used to format, clean and plot the data are publicly available in GitHub and are available upon request.Raw red/green IDAT les were read into R using the min Bioconductor package (v1.32.0) [11].
Initial quality control based on detection p-values and signal intensity were performed using the ENmix Bioconductor package (v1.22.4) [12].Probe intensities were background-and dye-corrected using the NOOB background correction [13] followed by strati ed quantile normalization.Probes with a detection p-value < 0.05 in more than 5% of the samples were removed from analysis.By this criterion, 11,936 CpGs were removed from analysis.If more than 5% of the probes in a sample had a detection p-value < 0.05, the sample would be removed from analysis.None of the samples met this criterion.Any known cross-hybridizing probes [14,15], any probe within 2 base pairs of a single nucleotide polymorphism, and probes from chromosome X and Y were also excluded.
Probes were tested for differential methylation using the limma R Bioconductor package (v3.38.3) using linear models whose standard errors were moderated using an empirical Bayes model [16].To assess the relationship between airway in ammation, asthma severity and DNA methylation, methylation levels were analyzed against the subjects' FeNO, FEV1, and ACT after adjustment for covariates.Since most subjects provided multiple samples over time, adjustments were made for repeat measures taken longitudinally from the same subject.Although samples were isolated for CD3 + T cells, we additionally used the FlowSorted.Blood.EPIC package (v1.4.1) to deconvolute samples into constituent cell types (CD8 + T cells, CD4 + T cells, NK cells, B cells, monocytes, neutrophils) using a modi ed version of the Houseman method [17] and these estimates were also used as covariates in the model.The following models were t for each measure of asthma (FeNO, FEV1, ACT), where repeated sampling was accounted for with the subject_id, and the blood cell type deconvolution proportions are continuous covariates: ~ 0 + {asthma measure} + subject_id + sex + CD8T + CD4T + NK + Bcell + Mono + Neu Measures of FeNO, FEV1, and ACT were initially assessed as continuous variables, but because patients' in ammation and asthma severity did not signi cantly change over time (Additional le 1: Figure S2), we binned subjects into top 40% and bottom 40% of each measure based on their mean measures over time.
A difference in methylation (beta value) of ≥ 5% and false discovery rate (FDR) of < 5% was deemed statistically signi cant.Differentially methylated probes were annotated to genic regions and CpG islands, shores, or shelves using the annotatr R Bioconductor package (v1.7.3) [18].Methylation signal was summarized at the gene-level so that the methylation differences were reported as the weighted average of the methylation changes of all CpG sites analyzed for a given gene while retaining minimum FDR.This allowed us to identify differentially methylated genes while adjusting, at least partly, to the uneven distribution of CpGs on different genes, the variable number of CpG sites for different genes, and potential biases within the Epic array itself.The functional signi cance of differentially methylated genes was analyzed using iPathway Guide (Advaita Bioinformatics, Ann Arbor, MI) to identify enrichment of gene ontology (GO) categories and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways and was analyzed over a denominator of all genes assayed by the array.Networks were constructed from signi cant genes within pathways using iPathway Guide.

Results
Thirty-seven adult asthmatics from Beijing, China were initially screened.As two subjects did not provide blood samples, 35 subjects were included in the analysis and their baseline characteristics are shown in Table 1.The mean age of the subjects was 42 years, with the mean reported age of asthma onset of 36.9 years.Patients were followed longitudinally for up to a year and blood samples from subjects were taken every three months when available.From 35 participants, 130 blood specimens were collected and used for DNA methylomic analysis (Additional File 1: Figure S1A).Although we speci cally isolated T cells using anti-CD3 magnetic beads for DNA methylomic analysis, cell type deconvolution based on the DNA methylation patterns was also performed, which con rmed that the DNA was mostly derived from CD4 + and CD8+ (CD4 > CD8) T cells as compared to NK, B cells, monocytes, and neutrophils (Additional le 1: Figure S1B).Stability of DNA methylation and phenotype over time One of our initial goals was to determine whether DNA methylation might change as asthma symptoms change over time for each subject.We also sought to determine whether there were linear correlations between clinical indices of asthma and overall patterns of DNA methylation.However, longitudinal follow-up of the cohort revealed that levels of FEV1, FeNO, and ACT scores remained, for the most part, relatively stable for most subjects throughout the study (Additional le 1: Figure S2).In fact, none of the subjects developed acute exacerbations of their asthma (de ned by hospitalizations or need for systemic corticosteroids) that would cause large shifts in FEV1, ACT, or FeNO.Thus, when we initially attempted to associate DNA methylation patterns with absolute values of FEV1, FeNO, and ACT, statistical modeling demonstrated a poor correlation between DNA methylation and changes in subjects' FEV1, FeNO, or ACT over time, with no differences in methylation that met the false-discovery threshold of less than 0.05.In fact, when examining overall DNA methylation patterns by principal components analysis, greater variances in DNA methylation were noted between subjects than within a given subject over time (Additional le 1: Figure S3).

Differential methylation associated with differences in FeNO, baseline FEV1, and ACT
Although initial attempts to linearly correlate DNA methylation with absolute levels of FeNO, FEV1 and ACT scores were unrevealing, we did note that the cohort consisted of asthmatics with a broad range of underlying severity and control.When each subject's FEV1, FeNO, or ACT was examined over the entire period of study, a clear segregation was noted among those at the top 40% and those at the bottom 40% of the cohort.Those in the bottom 40% of pre-bronchodilator FEV1 demonstrated an FEV1 consistently less than 80% of predicted (de ned clinically as obstructive physiology) as compared to those whose FEV1 was greater than 80% of predicted (non-obstructive) (Fig. 1A).Similarly, those in the top 40% of FeNO had an FeNO greater than 25 parts per billion (ppb), which is often associated with a type 2 endotype, versus those with low FeNO (< 25 ppb) (Fig. 1B).The same could be used to de ne those with high ACT scores (considered well-controlled) versus those with lower ACT (Fig. 1C).Those subjects with high FeNO (or high FEV1 or high ACT) demonstrated levels of FeNO (or FEV1 or ACT) that remained high throughout the study and vice versa.Our cohort could thus be segregated into populations of either high vs low FeNO, high vs low FEV1 (non-obstructive vs obstructive lung function), or high vs low ACT.Finally, measures of FeNO, FEV1, and ACT correlated poorly with one another (Fig. 1D-F), indicating that each of these measurements represent distinct, non-overlapping measures of airway in ammation, asthma severity, and control, respectively.Subjects with high FeNO, for example, did not necessarily demonstrate diminished lung function or poor asthma control.
Using these dichotomous groupings, we compared the DNA methylation pro les among samples from patients with high vs low FeNO, FEV1, and ACT.Adjusted for repeat sampling, this analysis identi ed many differentially methylated CpG sites, this time that were statistically signi cant by false discovery rate (FDR), between high vs low FeNO, high vs low FEV1, or high vs low ACT (Fig. 2A-C).Many of the top differentially methylated sites were annotated to genes, as shown.Only a small percentage of the differentially methylated CpG sites for a given comparison was found in common to be differentially methylated in another comparison (Fig. 2D-E), indicating that each of these distinct clinical parameters (FeNO, FEV1, or ACT score) was associated with a different set of differentially methylated CpG sites.

Annotation of differential methylation sites to genic regions
The differentially methylated probes or CpG sites for each comparison were next annotated by location within the genome and whether they were located within CpG islands, shores, or shelves.
Hypermethylation in CpG islands is often associated with suppression of gene expression.Compared to the low FeNO group, the high FeNO group demonstrated a smaller proportion of hypermethylated and a larger proportion of hypomethylated CpG sites within CpG islands (Fig. 3A).Samples from individuals with high FEV1 also had a high proportion of hypomethylated sites among CpG islands and a smaller proportion of hypomethylated sites among inter-CpG islands/shores/shelves.Finally, within the high FeNO group, a smaller proportion of hypermethylated CpG sites were found in promoters, 5-untranslated regions (UTR), and exons, whereas a greater proportion of hypermethylated sites were found within 3'-UTR and intergenic regions (Fig. 3B).A similar pattern was seen among the high FEV1 and high ACT groups, where a smaller proportion of hypermethylated CpG sites were found in promoters, 5'-UTR, and exons and a greater proportion of hypermethylated CpG sites were found among introns and intergenic regions.Together, these data suggest that the differential DNA methylation is not distributed uniformly throughout the genome, but instead, in select regions of certain genes that may ultimately in uence the regulation of those genes.

Enrichment of pathways from differentially methylated genes
As each of these CpG sites or probes were annotated to genes, we next examined the functional relevance of these genes and enrichment of functional pathways that were differentially methylated in each comparison.As each gene often had multiple CpG probes, we took the average methylation difference among all probes with FDR < 0.05 to determine whether genes were hyper-or hypomethylated.
Using iPathway, we found that some of the most highly enriched KEGG pathways among the high vs low FeNO differentially methylated genes include "Asthma", "Antigen processing and presentation", "Cytokine-cytokine receptor interaction", and "Th1 and Th2 cell differentiation" (Fig. 4A). Figure 4B highlights some of the differentially methylated genes within the "Asthma" KEGG pathway and include IL-4, eotaxin, and major histocompatibility complex (MHC) class II genes.GO analysis further revealed enrichment in molecular functions related to MHC class II protein assembly, T-cell activation, IL-1 production, and leukocyte adhesion (Fig. 4A).Common differentially methylated genes found among these GO terms include certain HLA family genes, TLR4, and IFNG (Fig. 4C).The relationship between these genes can be further viewed by network analysis (Fig. 4D).iPathway also identi ed differential methylation of upstream mediators such as PTPRC, that might regulate other genes downstream (Additional le 1: Figure S4A) or differential methylation of a family of genes including olfactory receptors (Additional le 1: Figure S4B), some of which have been implicated in asthma [19,20].Finally, other enriched pathways, such as "Neuroactive ligand-receptor interaction", which may on the surface seem less relevant, include differentially methylated genes such as P2RY11 (part of the purinergic pathway), AGT, and NMUR2 that are centered in network hubs (Fig E4C ); these genes have also been implicated in other studies as important in asthma [21][22][23][24].
When performing pathway analysis of differentially methylated genes between high vs low FEV1 comparison, different pathways compared to that found in the FeNO comparisons were noted (Fig. 5A).That these pathways were different was not surprising, as most of the differentially methylated CpG sites found in the high vs low FEV1 comparison were different than that found in the FeNO comparison (Fig. 2D).Some of these pathways including "Drug metabolism -other enzymes", "Chemical carcinogenesis -DNA adducts", and "metabolism by xenobiotics by cytochrome P450", which include genes involved in cytochrome P450 metabolism (e.g.CYP1A1, CYP2E1, CYP2A13), glutathione metabolism (GST family genes), and glucuronidation (UGT family of genes); these pathways have all reported to have some role in asthma [25][26][27][28].Other enriched pathways and GO terms continue to focus on T cell functions including "Th17 cell differentiation", "Regulation of T cell activation", "Adaptive immune response", "leukocyte cell-cell adhesion", and "T cell activation".Although these GO terms and pathway names bear similarities to that found in the FeNO comparison group, the differentially methylated genes within these pathways were different, including several genes involved in Th17 differentiation (Fig. 5B and Additional le 1: Figure S5A).Diverse chemokines (CXCL10, CCR2, CCR10, IL6, IFNG) and complement (C3 and C4A) genes were also more prominently enriched in this comparison as compared to the FeNO comparison.Finally, iPathway identi ed JAK2 and TNF as upstream regulators of many of these differentially methylated genes (Fig. 5C-D).
Pathway analysis of the differentially methylated genes associated with the high vs low ACT comparison identi ed pathways and enriched GO terms that were, yet again, different from those in the FeNO and FEV1 comparisons.These include KEGG pathways such as "Hypertrophic cardiomyopathy", "Cell adhesion molecules", and "Hematopoietic cell lineage" as well as GO terms such as "Embryonic skeletal development" and "Embyronic skeletal system morphogenesis" (Fig. 6A).These latter GO terms particularly highlight developmental genes within the HOX family and SMAD3 as differentially methylated (Fig. 6B).Although the KEGG pathway "Hypertrophic cardiomyopathy" at rst glance would seem unusual for a study investigating T cells in asthma, many of the highlighted genes include integrins (ITGB5, ITGA6), desmin (DES), endothelin (EDN1) and angiotensinogen (AGT) (Fig. 6C) that have all been implicated in asthma or airway remodeling [24,[29][30][31][32].Other pathways such as "Antigen processing and presentation" and "Graft-versus-host disease" and GO terms such as "Antigen processing" were like those highlighted as differentially methylated in the FeNO and FEV1 comparisons (Fig. 6D and Additional le 1: Figure S6).Finally, the high vs low ACT comparison also identi ed novel genes that form interrelated networks (Fig. 6E).

Common genes, networks, and pathways among all clinical comparisons
Although comparisons between high vs low FeNO, FEV1, and ACT produced mostly distinct sets of differentially methylated CpG sites and genes, several CpG sites and genes that were consistently found to be differentially methylation among all clinical comparisons.As noted in Fig. 2, there were 42 probes in common that were differentially methylated among all comparisons.When clinical severity and directionality of methylation changes were considered, 13 probes were hypermethylated in the more severe or more symptomatic group (high FeNO, low FEV1, low ACT) and 11 probes were hypermethylated in the less severe, less symptomatic group (Fig. 2E).Since genes typically have multiple CpG sites and probes, one can also examine the genes, differentially methylated in at least one probe or CpG site, that were in common in all comparisons.Analysis of those genes revealed three pathways and several GO terms, almost all of which focus on antigen processing and presentation and T cell function, that stand out (Fig. 7).Many of these genes include genes of the HLA family, and upstream mediators including protein tyrosine phosphatase receptor type C (PTPRC) and type J (PTPRJ) that are known regulators of B and T cell antigen signaling [33,34].

Discussion
In this study of adults with variable asthma severity, we identi ed extensive genome-wide differences in DNA methylation in CD3 + T cells of individuals with asthma that correlated with markers of airway in ammation, lung function, and asthma control.The differences in DNA methylation were widespread throughout the genome but varied depending on the comparison.Many of the genes found to be differentially methylated are known to be involved in antigen presentation or modulation of T cell differentiation, supporting the potential importance of DNA methylation changes in asthma.Although some genes have been shown in the literature to be clearly associated with asthma, other genes are novel, potentially providing new insights into asthma pathophysiology.
Changes in DNA methylation have been well-described in children who develop asthma [1,2,5,6] and represent a fundamental mechanism that support the fetal-origin hypothesis of the disease.However, asthma is a complex disease where patients develop asthma at different ages of onset and exhibit varying levels of airway in ammation and severity.Although changes in DNA methylation are wellassociated with asthma development [1-3, 6, 10], the role of DNA methylation in modulating asthma severity is less appreciated.By examining multiple facets of asthma severity and phenotype including FeNO, pre-bronchodilator FEV1, and symptom control by ACT, we were able to identify methylomic changes that were associated with each of these clinical variables.That methylation differences with each of these comparisons produced a unique set of differentially methylated probes and genes speaks to how each of these variables -FeNO, FEV1, and ACT -serve as different measures of asthma severity and pathophysiology.That many of the identi ed differentially methylated genes were different from those described in studies of children highlighting potential differences between adult and childhood asthma pathophysiology and the role methylation plays in asthma development vs severity.
As compared to many other DNA methylomic studies in asthma that utilize whole blood or peripheral blood mononuclear cells, we speci cally isolated CD3 + T cells for our study, as DNA methylation patterns are often cell-type speci c and identifying changes in methylation attributable to a speci c cell type may provide greater insights into the signi cance of these changes.Because anti-CD3 + magnetic bead isolation is subject to variability in technique, we additionally analyzed samples by cell-type deconvolution, which showed that although our samples consisted mostly of CD4 and CD8 T cells, a small percentage included NK cells, B cells, monocytes, and neutrophils.Our model thus used cell-type deconvolution as a covariate.Although other reference-free methods for cell-type deconvolution [35] could also have been used, the combination of CD3 + magnetic bead isolation and the cell-type deconvolution we employed was able to identify methylation differences with genes involved in antigen presentation and T cell function and activation.More speci cally, methylation differences between those with high and low FeNO, which is often used clinically as a marker of type 2 airway in ammation, were associated with methylation differences in IL4, CCL11 (eotaxin), and many HLA genes that are involved in antigen presentation and MHC class II assembly.Coincidentally, the cutoffs between subjects with high vs low FeNO in our cohort fell near 25 ppb, which is the cutoff commonly used in clinical guidelines [36] to distinguish type 2 vs non-type 2 asthma, suggesting that these methylation differences may be important in distinguishing different clinical phenotypes or endotypes of disease.The lack of available data on subjects' peripheral eosinophil levels, total IgE level, or speci c aeroallergen sensitivity testing is however, a limitation and prevents extrapolation of our ndings from FeNO with a true "type 2 high" phenotype.
Although comparisons of methylation patterns between subjects with high vs low FEV1 also resulted in methylation differences in genes associated with T cell function and activation, there were also other genes that were differentially methylated, including many related to drug metabolism and cytochrome P450.The signi cance of this is unknown but may indicate a mechanism by which environmental toxins can affect asthma and in uence lung function.Other genes identi ed as differentially methylated in the high vs low FEV1 comparison include those involved in in ammation and are downstream targets of JAK2 and TNF.These include genes such as IL6, interferon γ (IFNG), SOCS3, and NOS3.Complement genes (C3 and C4A) and genes involved in TH17 differentiation were also highlighted.These ndings indicate how much methylation differences in these genes that regulate and modulate in ammation may in uence lung function.Finally, it was interesting to note that individuals with asthma with different levels of symptom control, as measured by ACT, had differential methylation in yet other genes, some of which were involved in development and airway remodeling, such as HOX family genes, integrins (ITGB5, ITGA6), desmin (DES), and endothelin (EDN1), suggesting that differential methylation of these genes may have an important role in daily symptoms and asthma control.As asthma patients with diminished lung function and persistent symptoms often exhibit pathologic evidence of airway remodeling, the differential methylation of these in ammatory and developmental genes may predispose patients to developing airway remodeling.Some of the genes that were differentially methylated were observed to be differentially methylated in other studies of asthma as well, such as RUNX3 and IL4 [6].This supports the generalizability of our ndings.Certain genes, however, were found unique to our study but differentially methylated across all our comparisons.These include certain HLA family genes, PTPRC, and PTPRJ.These latter two genes are protein tyrosine phosphatases that have been described as master regulators that are differentially expressed in transcriptomic studies of children with severe asthma [34,37].These genes have also been shown by others to be regulated by DNA methylation and our study is the rst to identify the differential methylation of these genes to be associated with asthma severity.AGT, which codes for angiotensinogen, was another gene present in the center of many of our network analyses and shown among all our clinical comparisons as differentially methylated.
Our study was unique in that we studied a cohort of asthma patients from China, as compared to other population-based DNA methylomic studies that focus on European or North American cohorts.Differences in local environment and ancestral origin may account for some of the differences in DNA methylation in our study compared to others though there were many genes and pathways that were also similar.We excluded patients with smoking history to eliminate potential confounding effects of patients with COPD or asthma-COPD overlap.Although the total size of our cohort was small, longitudinal collection of blood samples from the same patient enhanced the power of our study and all analyses were adjusted for repeat measurements from the same patient.
One of the limitations of the study was that we did not observe robust longitudinal changes in DNA methylation that correlated with a given subject's FEV1, FeNO, or ACT over time, as we initially set out to do.Instead, the DNA methylation patterns for each patient remained relatively stable, and greater variations in DNA methylation were present between patients as compared to changes in methylation over time within subjects.This suggests that DNA methylation may be a stable epigenetic mark though this may also be a consequence of the fact that the degree of asthma severity in most of the subjects remained relatively stable throughout longitudinal follow-up and no subjects developed acute exacerbations during the study period.Not surprisingly then, attempts to linearly correlate DNA methylation patterns with absolute values of FeNO, FEV1, or ACT was poor.This, however, did not preclude our ability to nd major differences in DNA methylation when dichotomous groups of high vs low FeNO, FEV1, or ACT were compared.These ndings suggest that DNA methylation may be responsible for fundamental differences in baseline asthma phenotype/endotype but it is unclear if it is predictive of day-to-day asthma variation.Future studies on a larger cohort or that focuses on subjects with more variable asthma control will allow us to examine whether DNA methylation might also change over time in patients with labile disease.
Although we observed robust patterns of differential methylation, the biological signi cance of these differences in methylation is unknown.In our comparisons, we observed shifts in the distribution of methylation differences, with more hypomethylation in CpG islands and shores, for example, among subjects with high FeNO.This suggests that there may be an increase in gene expression among patient with high FeNO, which would be congruent with the presence of greater airway in ammation, though this remains to be determined experimentally.To summarize the net effect of methylation at the gene level, we calculated the weighted average of the methylation changes of all CpG sites for a given gene while maintaining the minimum FDR.Calculating the average methylation for a gene has the advantage of avoiding bias that may occur with genes that possess variable number of CpG sites assayed, uneven distribution of CpG sites, and inherent genomic biases within the Epic array.However, the effects of methylation are complex and often context dependent and this approach of taking the weighted average may obscure important methylation differences that only occur in speci c genomic regions.Furthermore, this approach does not completely remove bias from genes that possess a greater number of CpG sites and sequencing-based approaches such as whole genome bisul te sequencing, though more costly, would provide a more comprehensive picture of methylation differences.Gene pathway enrichment analysis was performed over the background of all genes analyzed by the Epic array, but ultimately, the functional consequences of the observed methylation changes are unknown.The genes, networks, and pathways that we highlight in this study require future studies to determine their mechanistic relevance.
In conclusion, differential patterns of DNA methylation were observed in the peripheral blood T cells of patients with different degrees of asthma severity.These differences occurred in many genes associated with T cell function, antigen presentation, and leukocyte activation, indicating a potential role for DNA methylation to modulate the immunologic pathways critical to asthma.Novel differentially methylated genes were also identi ed that may in uence T cell function or affect asthma pathobiology and warrant future studies.Together, these data support the potential for this epigenetic mechanism as a critical determinant of asthma phenotype and severity.

Declarations
Ethics approval and consent to participate: All patients provided written informed consent.The study was approved by both the Peking University and University of Michigan Institutional Review Board.
Consent for publication: Not applicable    UpSet plots were generated from the number of differentially methylated probes or CpG sites among all comparisons and the number of probes that were found in overlap between the three comparisons.
(E)Upset plots were generated listing the number of overlapping and non-overlapping probes that were hypermethylated among the high (vs low) FeNO group, low (vs high) FEV1 group, and low (vs high) ACT, which we collectively de ned as those having a more severe and more symptomatic phenotype.Similar UpSet plots were generated from the number of overlapping and non-overlapping hypermethylated probes from the low (vs high) FeNO, high (vs low) FEV1, and high (vs low) ACT groups, which are de ned as individuals with a less severe, less symptomatic phenotype.Figure 6

Figures
Figures

Figure 1 Comparison
Figure 1

Figure 2 Number
Figure 2

Figure 3 The
Figure 3

Table 1
Demographic and Clinical Characteristics of Participants at Baseline