An epigenome-wide association study of total serum immunoglobulin E in Hispanic children

Abstract

Background

Total immunoglobulin E (IgE) is a therapeutic target in allergic diseases. DNA methylation in white blood cells (WBCs) was associated with total IgE in an epigenome-wide association study (EWAS) of Caucasians. Whether DNA methylation of eosinophils explains those findings is insufficiently understood.

Methods

We tested for association between genome-wide DNA methylation in WBCs and total IgE in two studies of Hispanic children: the Puerto Rico Genetics of Asthma and Lifestyle Study (PR-GOAL, n = 306) and the Genes-environments and Admixture in Latino Americans (GALA II, n = 573). Whole-genome methylation of DNA from WBCs was measured using the Illumina Infinium HumanMethylation450 BeadChip. Total IgE was measured using the UniCAP 100 system. In PR-GOAL, WBC types (i.e. neutrophils, eosinophils, basophils, lymphocytes, and monocytes) in peripheral blood were measured using Coulter-Counter techniques. In GALA II, WBC types were imputed. Multivariable linear regression was used for the analysis of DNA methylation and total IgE, which was first conducted separately for each cohort, and then combining results from the two cohorts in a meta-analysis.

Results

CpG sites in multiple genes, including novel findings and results previously reported in Caucasians, were significantly associated with total IgE. However, adjustment for WBC types resulted in markedly fewer significant sites. Top findings from this adjusted meta-analysis were in genes ZFPM1 (P=1.5×10⁻¹²), ACOT7 (P=2.5×10⁻¹¹), and MND1 (P=1.4×10⁻⁹).

Conclusions

In an EWAS adjusted for WBC types (including eosinophils), methylation changes in genes enriched in pathways relevant to asthma and immune responses were associated with total IgE among Hispanic children.

INTRODUCTION

Allergic diseases affect millions of people worldwide. In subjects with allergic diseases, secretion of interleukin (IL)-4 and IL-13 by T helper 2 (Th2) cells increases production of total immunoglobulin E (IgE) by B cells. Total IgE is a key intermediate phenotype and a therapeutic target in allergic diseases such as asthma and allergic rhinitis¹.

Genome-wide association studies have identified SNPs linked with total IgE, but their joint estimated effect accounts for a very small proportion of the trait’s heritability^2–5. Environmental exposures affect total IgE^6,7, and thus epigenetic mechanisms such as DNA methylation may explain the “missing heritability” of total IgE.

A recent epigenome-wide association study (EWAS) using DNA from white blood cells (WBCs) reported that methylation in several genes is implicated on total IgE level in Caucasians⁸, with methylation of three genes (LPCAT2, IL5RA, and ZNF22) accounting for ~13% of total IgE variance. Whether those findings are explained by DNA methylation of eosinophils (which play a major role in allergic inflammation) or other WBC is insufficiently understood.

In the United States, Puerto Ricans are heavily affected with allergic diseases, while Mexican Americans have a lower but non-negligible burden from allergies^9–11. We report the first EWAS of total IgE with adjustment for (percentage of) WBC types, as well as the first EWAS of total IgE in Hispanics.

METHODS

Please also see the Online Supplement.

Subject recruitment and study procedures

Puerto Rico Genetics Of Asthma and Lifestyle Study (PR-GOAL)

From March of 2009 to June of 2010, 678 children with and without asthma (defined as physician-diagnosed asthma and at least one episode of wheeze in the prior year) were recruited in San Juan (Puerto Rico), as described elsewhere¹². Eligibility criteria included age 6 to 14 years and having four Puerto Rican grandparents. Of the 562 subjects who had blood samples with sufficient DNA for genome-wide genotyping, 306 subjects had additional data (including WBC count, total IgE, and a genome-wide study of DNA methylation) and were thus included in this analysis. Total serum IgE was measured using the UniCAP 100 system (Pharmacia & Upjohn, Kalamazoo, MI), and total WBC count and percentages of the five possible cell types (i.e. neutrophils, eosinophils, basophils, lymphocytes, and monocytes) in peripheral blood were measured using Coulter-Counter techniques. There were no significant differences in age, gender, asthma status or total IgE between subjects with available DNA who were (n=306) and were not (n=256) included in the current analysis.

Following bisulfite conversion of DNA from WBCs using the EZ DNA methylation kit (Zymo Research, Irvin CA), whole-genome DNA methylation levels were measured using the HumanMethylation 450K BeadChip system (Illumina, San Diego, CA). In brief, 200 ng of bisulfite-converted DNA was whole-genome amplified for 23 hours, followed by end-point fragmentation. The fragmented DNA was precipitated, denatured, and hybridized to the BeadChips for 18 hours at 48°C. The BeadChips were then washed, and the hybridized primers were extended and labeled prior to scanning the BeadChips using the Illumina iScan system (Illumina, San Diego, CA).

Principal components were calculated from genome-wide SNP data, which were genotyped using the HumanOmni2.5 BeadChip platform [Illumina Inc., San Diego, CA), as previously described¹². Genome-wide gene expression in a subset of 121 whole-blood samples was measured at the University of Pittsburgh Genomics and Proteomics Core Laboratories, using the Illumina HumanHT-12 v4 Expression BeadChip, as described in our previous work¹³.

Written parental consent was obtained for participating children, from whom written assent was also obtained. The study was approved by the Institutional Review Boards of the University of Puerto Rico (San Juan, PR), Brigham and Women’s Hospital (Boston, MA) and the University of Pittsburgh (Pittsburgh, PA).

Genes-environments & Admixture in Latino Americans (GALA II)

The GALA II study is a multicenter case-control study of asthma in 1,858 Latino children and adolescents¹⁴. Cases and healthy controls were recruited from five centers throughout the U.S. (Chicago, Bronx, Houston, San Francisco Bay Area, and Puerto Rico). Subjects were eligible if they were 8 to 22 years of age, self-identified all four grandparents as Latino, and had <10 pack-years of smoking history. Asthma was defined as physician-diagnosed asthma and report of symptoms and medication use in the previous two years. Participants were classified into three Latino subgroups (Puerto Rican, Mexican and other Latino), according to the self-reported ancestry of their four grandparents. All participants were genotyped on the Axiom^® LAT1 array (World Array 4, Affymetrix, Santa Clara, CA), as described elsewhere¹⁵. A subset of 573 participants (219 Puerto Ricans, 269 Mexicans and 85 other Latinos) with data for genome-wide DNA methylation and total IgE were included in the current analysis; there were no significant differences in age, gender, asthma status or total IgE between subjects who were (n=573) and were not (n=1,285) included in the analysis. Total serum IgE was measured using the ImmunoCAP™ 100 system (Phadia, Kalamazooo, MI)⁴. WBC DNA was bisulfite-converted and hybridized with Infinium HumanMethylation 450K BeadChips to measure methylation levels.

The study was approved by the Institutional Review Boards of the University of California at San Francisco and all other participating centers. All children and their parents provided written informed assent and consent, respectively.

Preprocessing and quality control for 450K DNA methylation data

We performed the same preprocessing and quality control for 450K DNA methylation data in PR-GOAL and GALA II. We read methylation data from the raw IDAT files using the R package methylumi, and calculated the β-value for each CpG as β = M/(M + U + α), where M and U represent methylated and un-methylated signal intensities at the specific site and α is an arbitrary offset (usually 100) intended to stabilize β-values where fluorescent intensities are low. Next, the methylation data were preprocessed by color balance adjustment, background correction and quantile-normalization with the lumi R package, and followed by Beta-Mixture Quantile (BMIQ) normalization with the wateRmelon package. This normalization strategy first quantile-normalized the intensities of methylation signals among all arrays, and then used BMIQ dilation to normalize the β-values within each array^16–18.

In each data set, we removed poor-quality probes with detection P-value > 0.01 in at least 20% of the samples. We also removed 11,656 probes with CpG loci located on sex chromosomes and 19,344 probes that had rs* names or located at 0 distance to known SNPs, according to the Illumina product annotation of Infinium HD Methylation SNP List. This QC process left 454,552 CpGs in both data sets. We then filtered CpG sites with an overall mean β-value > 0.9 or < 0.1 (considering extreme values as noise signals and to reduce multiple tests), which left 188,368 CpGs for data analysis.

Comparison to previous findings in Caucasians (see also Online Supplement)

We compared our results in Hispanic children to those of Liang et al. in Caucasians⁸. The comparison sample consists of three cohorts of Caucasians (Table E1), as follows: (1) The Medical Research Council Asthma (MRCA) panel, a cohort of 355 subjects (183 male) with mean age of 28 years (range 2–61 years), of whom 175 had doctor-diagnosed asthma, (2) a subset of the Poblogaeth Asthma Prifysgol Abertawe (PAPA) study, including 149 subjects (77 male, 34 with doctor-diagnosed asthma) with mean age of 21 years (range, 18 to 30 years), selected equally from the top and bottom deciles of the total IgE distribution from the entire PAPA study of 1,614 unselected volunteers, and (3) a subset of the Saguenay-Lac-Saint-Jean (SLSJ) asthma familial collection, including 160 subjects (80 male) with mean age of 29 years (range 5–79 years), of whom 69 had asthma. MRCA used the Illumina HumanMethylation 27K chip, while PAPA and SLSJ used the Illumina HumanMethylation 450K chip. We used the results presented in the original report, more detailed information can be found in Liang et al.⁸. Of note, we could not directly compare effect sizes across our study and those studies (which used transformed methylation values), but the directions of effect estimates were comparable across studies.

Statistical analysis

Total IgE was transformed to a log10 scale as a natural value in medical studies for data analysis. Linear regression was used for the multivariable analysis of DNA methylation and total IgE, separately for PR-GOAL and GALA II. Both analyses were adjusted for age, sex, asthma, batch (or plate), population subgroup (in GALA II) and the first five principal components (PCs) of the genotypic data, i.e., EWAS regression model (1) is: Log₁₀Ige ~ CpG methylation + Age + Gender + Asthma + Batch + Population + 5 PCs. PCs were calculated using common SNPs (minor allele frequency > 5%) by using EIGENSTRAT¹⁹, separately for PR-GOAL and GALA II. We next performed a meta-analysis to combine summary statistics from PR-GOAL and GALA II, using inverse variance weighting to generate the combined effect size and P-value for each CpG site. A Bonferroni-corrected P < 2×10⁻⁷ was used as the cutoff for genome-wide statistical significance.

We then carried out a similar multivariable analysis of DNA methylation and total IgE in PR-GOAL, with additional adjustment for the percentages of WBC types (only neutrophils, eosinophils, lymphocytes, and monocytes are included, since the sum of the percentages of these four cell types determines the percentage of the fifth possible type) to uncover IgE regulatory pathways that are independent of eosinophil count. Hence, our EWAS regression model (2) is: Log₁₀Ige ~ CpG methylation + Age + Gender + Asthma + Batch + Population + 5 PCs + Percentages of 4 WBC types. Since WBC percentages were not available in GALA II, cell type percentages were inferred based on the method described by Houseman et al.²⁰. In brief, a Lasso regression model for each of the five cell types was first trained from the observations of cell percentage and DNA methylation in PR-GOAL, and then the most predictive 200 CpG sites (see Table E2) for the five cell types were selected as reference CpGs for the application of a multiple imputation method²¹ to impute the cell percentages, which were adjusted for in the multivariable analysis of GALA II. Of note, we validated the performance of the Houseman imputation-based method using data from PR-GOAL. In PR-GOAL, the imputed and true cell percentages were significantly and highly correlated (see Figure E1). Thus, methodologically, this imputation should work for GALA II samples, whose imputed WBC types are likely valid and without significant bias.

Next, we conducted a meta-analysis of total IgE in PR-GOAL and GALA II, after additional adjustment for WBC types. Genes containing the top 1000 CpG sites associated with total IgE in this meta-analysis were examined in a KEGG pathway enrichment analysis, using Fisher’s exact test, the multiple testing p-values were adjusted with FDR. Moreover, we performed an expression Quantitative Trait Methylation (eQTM) analysis for the top 20 findings from the meta-analysis adjusted for WBC types (using PR-GOAL data), to test for association between DNA methylation and cis-gene (< 1M bases) expression in whole blood, adjusting for age, sex, asthma status (and WBC types), i.e. Gene expression ~ CpG methylation + Age + Gender + Asthma + 5 PCs +(Percentages of 4 WBC types).

RESULTS

The characteristics of participants in the Puerto Rico Genetics Of Asthma and Lifestyle Study (PR-GOAL) and the Genes-environments & Admixture in Latino Americans Study (GALA II) are shown in Table I. Compared to participants in GALA II, those in PR-GOAL were younger and had a higher total IgE. DNA methylation levels were similar across studies (Figure E2).

Table I.

Characteristics of participants in PR-GOAL^** and GALA II^***

Study	PR-GOAL (n=306)	GALA II (n=573)
Age (years)	9 (6–15)	13 (8–22)
Gender (female)	131 (42.8%)	287 (50.1%)
Asthma	173 (56.5%)	311 (54.3%)
Total IgE (IU/mL)	694 (2–10178)	371 (1–8545)
Average global genetic ancestry
European	63%	53%
African	25%	17%
Native American	12%	30%

^*Data shown as mean (range) for continuous variables and number (percentage) for categorical variables

^**PR-GOAL: Puerto Rican Genetics of Asthma and Lifestyle Study

^***GALA II: Genes-environments & Admixture in Latino Americans Study

We conducted a multivariable linear regression analysis of DNA methylation and total IgE, first separately for each cohort, and then combining results from the two cohorts in a meta-analysis. The meta-analysis identified 1,326 CpGs associated with total IgE at P < 2×10⁻⁷, and 2,992 CpGs at FDR < 0.01 (See Manhattan and QQ plots in Figure E3). Table E3 lists the top 200 CpG sites (by P-value) in this meta-analysis, as well as their results of separate analysis for each cohort. The direction of association between CpG sites and total IgE was consistent across studies. PR-GOAL had 640 significant CpGs with P < 2×10⁻⁷, while GALA II had 404 significant CpGs, and their correlation of overall effect size or -log₁₀(P-value) were also significant (Figure E4). In a sensitivity analysis, we performed a mega-analysis by combining the datasets from PR-GOAL and GALA II into a single dataset, using a similar approach to that for the meta-analysis, with additional adjustment for study site. This mega-analysis yielded similar results to those from the meta-analysis (Table E3).

Next, we compared the results of our meta-analysis unadjusted for WBC types with those from an EWAS of total IgE in Caucasians⁸. In spite of differences in ethnicity and age, most of the top significant CpG sites in our EWAS had the same direction of association as in Caucasians (Table E3). Of the top 200 CpG sites in our meta-analysis, 189 were also associated with total IgE (at P < 0.01) in either the PAPA or SLSJ study; and all 9 CpG sites that overlapped with those in the MRCA study (which had fewer CpG sites from the Illumina 27K Chip) were significantly associated with total IgE in that study. Moreover, we confirmed the top findings from the previous EWAS in Caucasians⁸ in our study (Table E4).

We then repeated the multivariable meta-analysis of our study cohorts after adjustment for WBC types. Fig 1 shows Manhattan and QQ plots of results in this adjusted meta-analysis, the number of significant CpG sites is markedly reduced comparing to the unadjusted results (Figure E3).

Fig 1.

Manhattan and QQ plots for the results of meta-analyses adjusting for WBC types. The Manhattan plot shows the chromosomal locations of –log₁₀(P-value) for association at each CpG site; the solid line illustrates a threshold of P-value=2×10⁻⁷ and the dashed line illustrates a threshold of FDR=0.01. The QQ plot shows observed versus expected –log₁₀(P-value) for association at all loci. The gray area illustrates the 95% confidence band.

Table II lists the top twenty CpG sites, and four (cg09249800, cg04983687, cg12614529 and cg21220721) of them were significantly associated with total IgE at P < 2×10⁻⁷, and the top two CpGs overlap with those from the meta-analysis without adjusting for WBC types. Partial regression plots in Fig 2 show the associations between CpG methylation values and total IgE, conditional on all other covariates, for the top four genes. Table E5 lists the top 200 CpG sites as well as the results of this cell-adjusted analysis for each cohort, and similar results were obtained when a mega-analysis was conducted instead of a meta-analysis.

Table II.

Top twenty CpG sites associated with total IgE in the meta-analysis adjusting for WBC types

ProbeID	Gene	CHR	Position	PR-GOAL		GALA II		Meta-analysis
				Effect*	P-value	Effect	P-value	Effect	P-value	FDR
cg04983687	ZFPM1	chr16	88558223	−5.27	4.69×10−7	−3.82	3.61×10−7	−4.31	1.53×10−12	2.88×10−7
cg09249800	ACOT7	chr1	6341287	−6.81	5.32×10−8	−6.09	1.06×10−4	−6.53	2.55×10−11	2.41×10−6
cg12614529	MND1	chr4	154269418	−5.25	7.38×10−5	−4.67	4.39×10−6	−4.88	1.40×10−9	8.76×10−5
cg21220721	ACOT7	chr1	6341230	−5.31	1.26×10−7	−1.84	1.92×10−3	−2.74	8.45×10−8	3.98×10−3
cg12227660	MIR202	chr10	135061670	−4.26	1.21×10−3	−3.89	9.19×10−5	−4.02	3.96×10−7	1.49×10−2
cg11699125	ACOT7	chr1	6341327	−5.77	5.34×10−6	−3.19	1.43×10−2	−4.52	6.68×10−7	1.88×10−2
cg01000631	EVL	chr14	100610667	−4.65	1.68×10−4	−3.88	1.09×10−3	−4.25	6.97×10−7	1.88×10−2
cg08940169	ZFPM1	chr16	88540241	−5.49	8.37×10−5	−4.13	4.14×10−3	−4.83	1.43×10−6	3.38×10−2
cg01458054	HELZ2	chr20	62200603	−3.34	2.75×10−3	−2.99	1.78×10−4	−3.11	1.66×10−6	3.48×10−2
cg15017119	SPRY2	chr13	82585793	−4.84	7.04×10−5	−2.96	5.45×10−3	−3.78	2.49×10−6	4.06×10−2
cg00405825	HIPK2	chr7	139474807	4.10	2.84×10−3	3.35	2.47×10−4	3.58	2.54×10−6	4.06×10−2
cg25087851	PTGDR2	chr11	60623918	−4.43	2.41×10−3	−4.30	3.29×10−4	−4.35	2.58×10−6	4.06×10−2
cg12103951	WAPL	chr10	88162314	0.77	7.00×10−2	1.28	1.18×10−5	1.12	3.55×10−6	5.15×10−2
cg12427941	TJAP1	chr6	43457177	5.40	7.88×10−3	4.56	3.11×10−4	4.79	7.99×10−6	1.08×10−1
cg06623197	MTMR3	chr22	30400763	2.99	5.23×10−3	1.64	3.59×10−4	1.86	1.18×10−5	1.48×10−1
cg12296550	HLA-DQB2	chr6	32728862	0.54	5.89×10−2	0.91	4.77×10−5	0.77	1.26×10−5	1.48×10−1
cg15700636	PRG2	chr11	57156050	−5.59	9.80×10−4	−4.41	4.07×10−3	−4.94	1.41×10−5	1.57×10−1
cg13074835	NFKBIL1	chr6	31515196	−2.05	1.22×10−1	−3.32	3.72×10−5	−2.98	1.51×10−5	1.58×10−1
cg15492834	C1orf226	chr1	162351057	−5.26	1.55×10−3	−3.11	1.79×10−3	−3.68	1.66×10−5	1.64×10−1
cg20481287	ETV6	chr12	12224457	−2.10	3.73×10−2	−2.58	1.95×10−4	−2.43	2.14×10−5	2.02×10−1

^*Effect is the regression parameter for methylation data (beta value) at each CpG site

Fig 2.

Partial regression plots of CpG methylation and total IgE for the top four genes in the meta-analysis adjusting for WBC types. The x and y axis respectively stand for the residuals of regressing methylation and IgE against all other covariates, which are adjusted in the multivariable regressions. The solid and hollow circles indicate GALA II and PR-GOAL samples.

We then performed a KEGG pathway enrichment analysis on the genes closest to the top 1000 CpG sites from the meta-analysis adjusted for WBC types. Of the seven pathways enriched for these genes at FDR ≤ 0.01 (Table III), the first one (“hsa05310 Asthma”) is highly related to total IgE. Some significant pathways are relevant to immune responses, such as “hsa04612 Antigen processing and presentation”, “hsa05320 Autoimmune thyroid disease” and “hsa04672 Intestinal immune network for IgA production”.

Table III.

KEGG pathways enriched by genes closest to the top CpG sites in the meta-analysis adjusting for WBC types

Pathway	Odds Ratio	P-value	FDR	Included.Genes
hsa05310 Asthma	7.41	1.72×10−05	0.002	PRG2,HLA-DOA,HLA-DQA1,HLA-DQB1,IL4,HLA-DRB5,IL13,HLA-DQA2,HLA-DPA1
hsa05330 Allograft rejection	6.06	6.46×10−05	0.004	HLA-DOA,HLA-DQA1,HLA-G,HLA-DQB1,IL4,HLA-DRB5,HLA-DQA2,GZMB,HLA-DPA1
hsa04612 Antigen processing and presentation	4.15	2.02×10−04	0.009	HLA-DOA,PSME3,HSPA2,HLA-DQA1,HLA-G,HLA-DQB1,CTSB,TAP2,HLA-DRB5,HLA-DQA2,HLA-DPA1
hsa05320 Autoimmune thyroid disease	4.88	2.66×10−04	0.009	HLA-DOA,HLA-DQA1,HLA-G,HLA-DQB1,IL4,HLA DRB5,HLA-DQA2,GZMB,HLA-DPA1
hsa05332 Graft versus-host disease	5.22	3.89×10−04	0.01	HLA-DOA,HLA-DQA1,HLA-G,HLA-DQB1,HLA-DRB5,HLA-DQA2,GZMB,HLA-DPA1
hsa04750 Inflammatory mediator regulation of TRP channels	3.29	4.25×10−04	0.01	ADCY4,PRKCE,PIK3CG,PLA2G4C,IL1RAP,ADCY8,TRPM 8,ADCY5,PRKCA,PPP1CA,PRKCB,PIK3CB,ADCY3
hsa05323 Rheumatoid arthritis	3.43	5.06×10−04	0.01	ATP6V0A2,HLA-DOA,HLA-DQA1,TNFSF11,CSF2,HLA-DQB1,ATP6V0E2,HLA-DRB5,TGFB3,HLA-DQA2,ANGPT1,HLA-DPA1

Next, we performed eQTM analyses for the twenty CpG sites in Table II and their cis-genes, using the DNA methylation and gene expression data from PR-GOAL. From the analysis unadjusted for WBC types, cg25087851 methylation was significantly associated with expression of the gene for the prostaglandin D2 receptor 2 (PTGDR2, P = 5.16×10⁻¹⁷, see Table E7)). cg25087851 is 473 nucleoside bases upstream of PTGDR2, which encodes a G-protein-coupled receptor that is preferentially expressed in CD4+ effector Th2 cells. This receptor mediates pro-inflammatory chemotaxis of eosinophils, basophils, and Th2 lymphocytes generated during allergic inflammation²², and a PTGDR2 antagonist reduces eosinophilic airway inflammation (“experimental asthma”) in murine models²³. DNA methylation of another CpG site (cg12227660) was significantly associated with expression of the gene polyamine oxidase (PAOX, P = 2.04×10⁻¹³); cg12227660 is 131,071 bases upstream of PAOX. PAOX particles can be used as a drug delivery vehicle for the treatment of acute inflammatory diseases such as asthma and acute lung injury^24,25. After adjusting for cell types, the observed associations between methylation of CpG sites and gene expression became non-statistically significant, following a pattern similar to that for the EWAS results (Table E7).

GALA II participants belong to three Latino subgroups, captured by PCA of genotypic data, while the effect of population composition on DNA methylation is negligible, according to PCA of methylation data (Figure E5). Since PR-GOAL participants are all Puerto Rican, we conducted a secondary EWAS of total IgE after selecting only Puerto Ricans from GALA II (n = 219). Figure E6 shows the Manhattan and QQ plots for all CpG sites, and the results of top 200 sites are listed in Table E8 for the EWAS without cell type adjustment and Table E9 for the EWAS with cell type adjustment. Findings in this analysis in Puerto Ricans were similar to those in all Hispanics, the correlations of resultant effect size or −log10(Pval) were significant (Figure E7); and the 4 significant CpG sites in the meta-analyses with adjustment for WBC types were the same as those in all Hispanics.

DISCUSSION

Our meta-analysis without adjustment for WBC types identified 1,326 CpG sites that achieved a genome-wide significant association with total IgE at P < 2×10⁻⁷. There are some novel findings on top of the results (Table E3), including methylation in the genes for microRNA 202 (MIR202), lipin 1 (LPIN1), and prostaglandin D2 receptor 2 (PTGDR2). MIR202 is differentially expressed in cultured airway epithelial cells of monkeys in response to ozone exposure and lipopolysaccharide (LPS) treatment²⁶, and is decreased in bronchoalveolar lavage (BAL) of patients with sarcoidosis²⁷. LPIN1, which has been implicated in lipid metabolism and rhabdomyolisis, encodes a magnesium-ion-dependent phosphatidic acid phosphohydrolase enzyme. Expression of LPIN1 is low in subjects with atopic eczema, who often have other atopic diseases such as allergic asthma and allergic rhino-conjunctivitis²⁸.

After additional adjustment of our meta-analysis for WBC types, there was a marked reduction in the number of CpG sites associated with total IgE. Since only eosinophils were consistently and significantly associated with total IgE in our study (Table IV), our overall findings suggest that DNA methylation in eosinophils explain most results for DNA methylation and total IgE in the analyses unadjusted for WBC types, thus emphasizing the importance of adjusting for cell types in future studies.

Table IV.

Association between WBC type percentages and total IgE

IgE ~ Cell Percentage	PR-GOAL						GALA II
	Control		Case		All		Control		Case		All
	Effect	P-value	Effect	P-value	Effect	P-value	Effect	P-value	Effect	P-value	Effect	P-value
Neutrophil	−0.007	0.216	−0.009	0.091	−0.008	0.050	−0.007	0.061	−0.004	0.319	−0.008	0.007
Lymphocyte	−0.011	0.076	−0.001	0.885	−0.008	0.057	0.001	0.792	−0.009	0.055	−0.002	0.573
Monocyte	0.045	0.128	−0.023	0.419	0.007	0.756	0.021	0.237	−0.012	0.566	−0.008	0.575
Eosinophil	0.116	2.1×10−15	0.040	6.0×10−05	0.067	2.1×10−16	0.112	5.1×10−11	0.090	4.2×10−15	0.108	1.0×10−29
Basophil	0.491	0.047	−0.058	0.756	0.194	0.200	0.449	0.010	0.115	0.519	0.258	0.050

Our four genome-wide significant findings in the meta-analysis adjusted for WBC types were not among the top 62 top loci associated with total IgE (at FDR <0.005, with consistent direction of association across cohorts) in the meta-analysis of three cohorts included in the prior EWAS of total IgE in Caucasians⁸. These four CpG sites were located in the genes for zinc-finger protein multi-type 1 (ZFPM1), acyl-CoA thioesterase 7 (ACOT7), and meiotic nuclear division 1 (MND1). ZFPM1 suppresses expression of interleukin 4 (IL-4) and induces expression of IFN-γ in CD4+ T cells, thus reducing Th2 cell differentiation²⁹. Moreover, WBC DNA methylation in both ZFPM1 and ACOT7 has been associated with asthma³⁰, and WBC DNA methylation of ZFPM1 was consistently associated with atopy (defined as at least one positive IgE to allergens) or high total IgE (≥200 kU/L) across two Caucasian cohorts included in a prior study³¹. In that study, DNA methylation of acyl-CoA thioesterase 7 (ACOT7) was significantly associated with atopy or high total IgE in the discovery cohort but not in a replication cohort. Thus, herein we first report consistent evidence of a genome-wide significant association between ACOT7 methylation and total IgE.

Of the top findings in our analysis unadjusted for cell types, CpG sites in ZFPM1, ACOT7, MIR202, EVL, PTGDR2 and PGR2 remained among the top twenty results in the analysis adjusted for WBC types. While not genome-wide significant, the WBC-adjusted analysis revealed some new genes among our top twenty findings, including MND1, sprouty RTK signaling antagonist 2 (SPRY2), major histocompatibility complex, class II, DQ beta 2 (HLA-DQB2) and ETS variant 6 (ETV6) genes. SPRY2 differentially regulates lymphoid and myeloid cell functions relevant to allergic asthma³². HLA-DQB2 is biologically relevant to immune responses and has previously associated with asthma³³. ETV6-ABL gene rearrangement can result in chronic myeloid leukemia associated with eosinophilia³⁴.

We recognize several study limitations. First, WBC types were not measured in GALA II. Although we applied validated methods to perform imputation of WBC percentages, the imputation uncertainty may still have reduced our statistical power. It is unlikely to have biased our results. Second, WBC types such as lymphocytes include multiple cell types (e.g., Th1, Th2, Th17) and we lack data to identify whether methylation of a specific cell type accounts for our findings in the meta-analysis adjusted for cell types. Third, we had data for gene expression in whole-blood (and not WBCs) in only a subset of participants in PR-GOAL, likely reducing our statistical power for the eQTM analysis of correlation between DNA methylation and gene expression. Finally, we lack functional analyses, but our top CpG sites are located within biologically plausible genes, as demonstrated by combined evidence from prior publications, and the results from our enrichment and eQTM analyses.

In summary, we demonstrate novel WBC DNA methylation changes associated with total IgE in an EWAS of Hispanic children, while replicating prior findings in Caucasians⁸. Our findings also emphasize the importance of adjusting for cell types (particularly eosinophils) in future studies of WBC DNA methylation and total IgE or related phenotypes (e.g., asthma).

Clinical Implications.

Among Hispanic children, methylation changes in the genes ZFPM1, ACOT7, and MND1 are associated with total IgE. ACOT7 and ZFPM1 have been implicated in the pathogenesis of asthma or atopy.

Capsule Summary.

In an EWAS adjusted for WBC types, few DNA methylation changes were significantly associated with total IgE among Hispanic children. Future EWAS of total IgE should account for DNA methylation of WBC types, particularly eosinophils.

Abbreviations

EWAS

epigenome-wide association study

IgE

immunoglobulin E

WBC

white blood cell

PR-GOAL

Puerto Rico Genetics of Asthma and Lifestyle Study

GALA II

Genes-environments and Admixture in Latino Americans study

eQTM

expression Quantitative Trait Methylation

PC

principle component