Novel eosinophilic gene expression networks associated with IgE in two distinct asthma populations

Abstract

Background:

Asthma represents a significant public health burden; however, novel biological therapies targeting immunoglobulin E (IgE)-mediated pathways have widened clinical treatment options for the disease.

Objective:

In this study we sought to identify gene transcripts and gene networks involved in the determination of serum IgE levels in people with asthma that can help inform the development of novel therapeutic agents.

Methods:

We analyzed gene expression data from a cross-sectional study of 326 Costa Rican children with asthma, aged 6 to 12 years, from the Genetics of Asthma in Costa Rica Study and 610 young adults with asthma, aged 16 to 25 years, from the Childhood Asthma Management Program trial. We utilized differential gene expression analysis and performed weighted gene co-expression network analysis on 25,060 genes, to identify gene transcripts and network modules associated with total IgE, adjusting for age and gender. We used pathway enrichment analyses to identify key biological pathways underlying significant modules. We compared findings that replicated between both populations.

Results:

We identified 31 transcripts associated with total IgE that replicated between the two study cohorts. These results were notable for increased eosinophil-related transcripts (including IL5RA, CLC, SMPD3, CCL23, CEBPE). Pathway enrichment identified the regulation of T cell tolerance as important in the determination of total IgE levels, supporting a key role for IDO1.

Conclusions & Clinical Relevance:

These results provide robust evidence that biologically meaningful gene expression profiles (relating to eosinophilic and regulatory T cell pathways in particular) associated with total IgE levels can be identified in individuals diagnosed with asthma during childhood. These profiles and their constituent genes may represent novel therapeutic targets.

Introduction:

Asthma, allergies and related disorders have doubled in prevalence during recent decades and are among the most common chronic diseases in industrialized countries, with asthma affecting 26 million United States children and adults, and allergies affecting ~42.5% of the population (1–4). A large proportion of asthma cases are characterized by an increase in serum immunoglobulin E (IgE) levels (5), and a dose-response relationship between IgE levels and disease severity has been demonstrated (6). Total IgE is an important therapeutic target, and existing biologic therapies (including omalizumab and recently approved reslizumab and mepolizumab) target IL-4, IL-5, IL-13, and IgE-specific pathways (7). However, therapeutic trials of these biologic agents only demonstrate benefits among subjects with more severe asthma subtypes (7). These findings provide further evidence that asthma comprises a diverse range of immunologic and clinical phenotypes, which require an equally diverse variety of personalized therapies.

Given the high heritability of IgE (estimated to be up to 0.8 in cord blood) (8), a better understanding of the causal genetic factors underlying IgE levels among individuals with asthma may lead to novel IgE-focused treatments for a subset of cases. While many of the genetic and genomic studies of total IgE levels have been performed in genetically heterogeneous populations (9–12), we hypothesized that investigating a more homogeneous population, specifically a semi-population isolate (Costa Rican children with asthma), will enable the improved identification of novel biologic targets. A genetic study of this population previously identified an association between polymorphisms in IL13 and serum IgE, eosinophil count, and asthma exacerbations (13).

The aim of this study was to use a network genomics approach to identify novel differentially expressed genes and gene-networks associated with serum total IgE in two populations with asthma; the ‘Genetics of Asthma in Costa Rica Study’ (GACRS) and ‘The Childhood Asthma Management Program’ (CAMP) cohort. By comparing the semi-isolated GACRS population, which is characterized by a high asthma prevalence (24% in children), a more homogenous genetic background, and a highly allergic population, to the heterogeneous CAMP population, we aimed to improve the robustness and generalizability of our findings (14). These approaches using gene co-expression networks have been useful in identifying gene signatures associated with severe subtypes of asthma, but have not previously been employed in association with IgE levels in asthma(15, 16). Elucidating the biology underlying serum total IgE, particularly as it pertains to asthma severity, is vital to develop improved strategies for the prevention, management, and treatment of asthma and allergic conditions.

Methods:

GACRS population and study design

Participants in the GACRS are members of a Hispanic population semi-isolate from the Central Valley of Costa Rica, which has one of the highest rates of asthma worldwide (17). From February 2001 to August 2008, questionnaires were sent to the parents of 16,912 children (ages 6–14 years) enrolled in 140 Costa Rican schools; 9,180 (54.3%) questionnaires were returned. Children were eligible for the study if they had asthma (physician-diagnosed asthma and ≥2 respiratory symptoms or asthma attacks in the prior year) and a high probability of having ≥6 great-grandparents born in the Central Valley of Costa Rica. A total of 1,189 children with asthma and their parents (i.e. 383 parent-child trios) were enrolled. 383 children provided blood samples, on which whole-blood RNA gene expression profiling was performed, using the Illumina HumanHT-12v4 microarray. All participating children completed a protocol including a slightly modified and translated version of the questionnaire used in the Collaborative Study on the Genetics of Asthma (18). Serum total IgE level was determined using the UniCAP 250 system (Pharmacia & Upjohn, Kalamazoo, Mich), with samples measured in duplicate, conducted at the same time as gene expression profiling. Written parental consent and child’s assent were obtained. The study was approved by the Institutional Review Boards of the Hospital Nacional de Ninos (San Jose, Costa Rica) and Brigham and Women’s Hospital (Boston, MA, USA) (IRB Protocol Number 2000P001130).

CAMP population and study design

CAMP is a multi-center, randomized, double-masked, clinical trial designed to determine the long-term (~16.5 years of follow-up) effects of three inhaled treatments for mild to moderate asthma in children aged 5 to 12 at baseline: placebo, nedocromil, or budesonide.(19) Asthma was defined based on having both a clinical history (with presence of symptoms or use of an inhaled bronchodilator at least twice weekly or the use of daily medication for asthma) and airway responsiveness to methacholine (as measured by 20% decline in forced expiratory volume (FEV1) at doses ≤12.5 mg/mL of methacholine). A follow-up study to the primary trial extracted blood samples from 620 CAMP subjects at early adulthood (ages 16 to 25, a mean 11.6 years (range 10–13.5 years) after enrollment) for gene expression profiling using the same Illumina HumanHT chip and concurrently measured serum total IgE (two individuals for whom IgE was measured >18 months from gene expression profiling were removed from the analysis). Of these, 610 subjects had complete phenotype data for downstream analysis.

Gene expression profiling

In both populations, whole-blood gene expression was assayed with 47,009 probes on the Illumina HumanHT-12 v4 Expression BeadChip (Illumina, Inc., San Diego, CA, USA) that passed stringent and commonly used quality control metrics (20). Expression data were log2-transformed and quantile-normalized using the R package “lumi” (version 2.22) and Bioconductor (21). Principal components of gene expression were generated for both GACRS and for CAMP. Prior to downstream statistical analyses, a standard non-specific variance filter was applied to the expression data using the R package “genefilter”. Probes not annotated with a valid Entrez gene identifier or Human Genome Organization gene symbol and probes with inter-quartile ranges (IQR) of expression variance below the 50^th percentile were excluded,(22) ensuring that the most informative probes were used for analysis. The data were then collapsed to assign a single probe to each gene based on the largest IQR of expression variance. This resulted in a total of 25,059 and 24,971 gene probes for the GACRS and CAMP analysis respectively. All 24,971 genes in the CAMP dataset were also represented in the larger GACRS dataset.

Statistical analyses

All analyses were first conducted in the GACRS, and then conducted independently in the CAMP population for validation of the significant findings using a parallel analysis design and identical methods. As in previous study designs (23, 24), we aimed to compare findings that replicated in both a relative genetic isolate (GACRS) as well as an outbred population (CAMP) (19).

Differential gene expression analysis

Generalized linear regression models (GLM)(25) were constructed to identify differentially expressed genes associated with log-transformed total serum IgE levels. We adjusted for age (due to the age differences in our two cohorts), sex (due to prior genetic associations with IgE (25)), and the first two principal components of gene expression to account for any batch effects (all concurrent with serum total IgE measurements). The Benjamini– Hochberg procedure was used to control the false discovery rate (FDR) with q-value set at 0.05 (26).

Gene network analysis

Gene expression modules have previously been shown to be enriched for biological functions. We hypothesized that by using Weighted Gene Co-expression Network Analysis (WGCNA) (27) we would be able to refine the results of the regression analysis to identify networks of highly significant genes that are expressed together in groups or ‘modules’ of co-expressed genes. These modules can then be related to phenotypic and baseline traits. WGCNA was used to generate a network describing correlation patterns between the genes. Using hierarchical clustering based on topological overlap and applying a soft thresholding power of 5, chosen to achieve a scale-free topology fitting index >0.9 (explaining over 90% of the variance), the interconnectedness between all possible gene pairs was quantified and they were assigned to gene co-expression modules. A dynamic clustering approach was utilized with a deep split of 0.8, and a minimum module size of 30 was specified, in order to further refine the degree of detail each module contained without influencing module structure, based on recommended defaults (27). We then removed all uncorrelated transcripts that were assigned to the grey module, a module typically reserved for transcripts that do not map to a discrete module. Each module was summarized by an eigenvalue based on its first principal component, and the association between the module eigenvalues and log₁₀-transformed serum total IgE, after adjustment for age at gene expression profiling and gender, was computed using multivariable linear regression.

Pathway enrichment analysis

Pathway enrichment analysis using Gene Ontology (GO) terms was performed using an online tool from the GO Consortium (http://geneontology.org/page/go-enrichment-analysis), relying on the PANTHER Overrepresentation Test (release 20171205) (28) with an alpha<0.01, fold enrichment >5, and containing at least 2 genes transcripts from our study included in the GO category. We identified key biological pathways among (i) the differentially expressed genes identified using the multivariate linear regression models, (ii) the gene transcripts of WGCNA-generated modules associated with serum IgE.

Overrepresentation analysis

We conducted gene set overrepresentation analysis using replicated gene transcripts common to both the GACRS and CAMP population, in comparison with the Molecular Signatures Database version 6.1 (http://software.broadinstitute.org/gsea/msigdb/).(29) We specifically used hypergeometric tests with FDR to correct for multiple hypothesis testing and compared our dataset with the C7 database, which comprises 4872 immunologic gene sets that were curated based on published human and murine studies of the immune system from the Gene Expression Omnibus (30).

Results:

Study populations

The baseline characteristics of the GACRS and CAMP study populations are described in Table 1. The GACRS comprises a Hispanic/Latino population, while CAMP was predominantly white European (68%), and only 10% Hispanic/Latino. The proportion of males and females in the two cohorts were similar, with a higher percentage of males in both. The age of asthma onset was young in both cohorts: 2.4 years in the GACRS and 3.0 years in CAMP. Gene expression profiling and the measurement of serum IgE levels were performed on blood samples taken in childhood (mean: 9.1 years) in the GACRS, and in early adulthood in CAMP (mean: 20.4 years). Importantly, there was no significant difference in serum IgE levels between the two cohorts, despite the age difference.

Table 1:

Demographics of GACRS population and CAMP populations

	GACRS	CAMP	P-Value
Number of participants	326	610
Sex, n (%)
Female	140 (43%)	226 (37%)	0.10
Male	186 (57%)	384 (63%)
Race/Ethnicity
Hispanic/Latino	326 (100%)	59 (10%)	na
American Indian or Alaskan Native	0 (0%)	4 (1%)
Black or African American	0 (0%)	94 (15%)
East/Southeast Asian	0 (0%)	4 (1%)
European	0 (0%)	417 (68%)
Multiple Races	0 (0%)	32 (5%)
Age at gene expression profiling, mean (SD)	9.1 (1.8)	20.4 (2.2)	<0.001
Age of asthma onset (years), mean (SD)	2.4 (2.2)	3.0 (2.4)	<0.001
log10IgE (kU/L), mean (SD)	2.5 (0.7)	2.5 (0.6)	0.74
Atopic Dermatitis	38 (25%)	174 (29%)	0.44
Allergic Rhinitis	300 (92%)	310 (51%)	<0.001
Inhaled Corticosteroid Use	173 (53%)	125 (23%)	<0.001

Differential gene expression analysis

A total of 81 gene transcripts were significantly associated with serum IgE levels in the GACRS cohort after adjustment for age at expression profiling, gender, and the first two principal components, and controlling for the FDR. 69 of these gene transcripts (85%) were positively associated with serum total IgE and 12 (15%) were negatively associated (Figure 1). Differential gene expression analysis showed no association of these genes with atopic dermatitis or allergic rhinitis. Of these 81 gene transcripts, 42 (54%) were replicated in CAMP with a concordant direction of effect; all 42 were overexpressed with increasing IgE levels (Figure 2, Supplementary Table 1). We then conducted a sensitivity analysis, adjusting for inhaled corticosteroid use in both populations, with 36 (86%) transcripts still significant (Supplementary Table 1). Of note, all of the transcripts identified in CAMP remained significantly associated with IgE after adjusting for original treatment group from the CAMP randomized trial. The 42 transcripts replicated in CAMP and GACRS were nominally enriched (p<0.01, fold enrichment>5, number of genes>1) for 32 biological processes (Supplementary Table 2). Although none of these 32 processes were robust to the most stringent correction for multiple testing, Bonferroni correction, a number had biologically plausible functional annotations, particularly those related to immune responses, such as ‘regulation of T cell tolerance induction’.

Figure 1; Number of genes significantly associated with serum total IgE in the GACRS and CAMP populations by differential gene expression.

BH FDR = Benjamini-Hochberg adjusted false discovery rate.

Figure 2: Heatmap comparing differential gene expression of the 42 replicated genes by IgE (top versus bottom quartile) in the (A) GACRS and (B) CAMP populations.

All individuals were included in the differential gene expression analysis; however the top quartile IgE (green) and the bottom quartile IgE (yellow) were displayed for comparison purposes.

Gene network analysis

The gene expression network analysis was first performed in the GACRS. The samples were clustered, resulting in the exclusion of three samples identified as outliers. From the 25,059 transcripts, hierarchical clustering based on topological overlap then designated the 10,073 transcripts into 32 modules (43%) (Supplementary Figure 1A). Five modules, as summarized by their eigengenes, were associated with serum IgE level (alpha<0.05) (Supplementary Figure 1A). Only one module, containing 103 gene transcripts, was robust to correction for multiple testing (p=4.8×10⁻¹¹). This module was also associated with the more stringent subtype of asthma (p=0.002), defined by physician diagnosis and the presence of airway hyperreactivity on methacholine challenge, but was not associated with allergic rhinitis or atopic dermatitis. The module was also not associated with inhaled corticosteroid use, but had a nominal association with oral steroid use in the last year (p=0.04) Pathway enrichment analysis of these 103 transcripts identified 52 GO-defined biological processes (Supplementary Table 3), including 12 of the 32 processes identified in the differential gene expression analysis and additional processes relating to immune function.

WGCNA was then applied to the CAMP dataset using the same network construction approach; one sample was excluded as an outlier. Within the remaining 611 samples the gene transcripts were clustered into 40 modules including 15,120 transcripts (61%). Fourteen modules were associated with serum IgE levels after adjustment for age and gender (p<0.05), and one module, was robust to correction for multiple testing (p=6.8×10⁻¹⁴) (Supplementary Figure 1B). This module was also associated with inhaled corticosteroid use (p=0.0003). This module contained 38 transcripts that were enriched for 25 biological processes (Supplementary Table 4).

To determine any genomic and biological similarities between the significant GACRS and CAMP modules, the overlap in terms of gene transcripts and enriched biological processes was computed. Of the 103 transcripts in the significant GACRS module and the 38 transcripts in the significant CAMP module, all 38 were common to both (Supplementary Table 5). Of the 52 and 25 enriched GO-defined biological processes from the GACRS and CAMP modules, respectively, 14 were common to both (Supplementary Table 5). Finally the similarities between the differential gene expression analysis and the WGCNA analyses were compared (Figure 3). Thirty-one transcripts (Table 2), and twelve enriched biological processes, including a number related to immune response, were common to both sets of analyses in both cohorts (Table 3), and further confirmed the similarities in results between the two independent populations. Based on the literature, eight of the thirty-one gene transcripts have been previously linked to IgE, while the remainder represent novel associations (Table 2).

Figure 3;

Comparison of the total IgE gene expression profiles generated using differential gene expression and network approaches

Table 2:

Replicated gene transcripts (n=31) associated with total IgE using two different statistical approaches in two distinct cohorts with asthma: GACRS and CAMP

Gene Symbol	Gene Name	Chr	GACRS P-Value	CAMP P-Value	Sig. adjusted for ICS	Linked with eosinophils
						Esnault (37)	Nakajima(38)	MSigDB(29)
PRSS33	protease, serine, 33	16	2.8×10−9	2.4×10−10	■	●
OLIG2	oligodendrocyte lineage transcription factor 2	21	1.7×10−8	2.3×10−12	■	●
TFF3	trefoil factor 3 (intestinal)	21	2.4×10−8	6.7×10−13	■		●
CLC*	Charcot-Leyden crystal protein	19	3.8×10−8	4.0×10−9	■	●	●
CCL23	chemokine (C-C motif) ligand 23; macrophage inflammatory protein 1-alpha (MIP-1-alpha)	17	1.5×10−7	3.0×10−10	■		●	●
EMR4P	adhesion G protein-coupled receptor E4, pseudogene	19	1.5×10−7	7.1×10−6	■	●
OLIG1	oligodendrocyte transcription factor 1	21	1.5×10−7	7.8×10−7	■	●
GPR44*	prostaglandin D2 receptor 2, G protein-coupled receptor 44, (aka CRTH2)	11	2.0×10−7	2.1×10−8	■	●	●
FLJ43093	Unknown	6	2.5×10−7	2.0×10−8	■
CEBPE	CCAAT/enhancer binding protein (C/EBP), epsilon	14	2.7×10−7	2.6×10−13	■			●
PYROXD2	pyridine nucleotide-disulphide oxidoreductase domain 2 (PYROXD2)	10	3.3×10−7	3.8×10−10	■	●
SORD2	sorbitol dehydrogenase	15	1.5×10−6	4.9×10−5	■
SIGLEC8*	sialic acid binding Ig-like lectin 8	19	1.6×10−6	5.0×10−9	■	●
PIK3R6*	phosphoinositide-3-kinase, regulatory unit 6	17	2.9×10−6	2.0×10−8	■	●
LGALS12	galectin 12	11	6.1×10−6	3.7×10−4	■	●
GFOD1	glucose-fructose oxidoreductase domain containing 1	6	8.7×10−6	1.8×10−3	■
CCR3*	chemokine (C-C motif) receptor 3	3	1.8×10−5	9.2×10−5	■	●	●
IL5RA*	interleukin 5 receptor, alpha	3	3.5×10−5	0.020	■	●	●
SMPD3	sphingomyelin phosphodiesterase 3, neutral membrane	16	5.6×10−5	2.2×10−6	■			●
INPP1	inositol polyphosphate-1-phosphatase	2	1.5×10−4	6.4×10−5	■
EMR1	adhesion G protein-coupled receptor E1 (alias ADGRE1)	19	2.5×10−4	5.5×10−3	■		●
ACOT11	acyl-CoA thioesterase 11	1	2.6×10−4	2.1×10−4	■
CAMK1	calcium/calmodulin-dependent protein kinase I	3	3.0×10−4	6.8×10−5	■	●	●
VSTM1	V-set & transmembrane domain containing 1	19	4.6×10−4	1.9×10−6	■	●
RNF14	ring finger protein 14	5	4.6×10−4	3.1×10−5	■
SPNS3	spinster homolog 3 (Drosophila)	17	6.2×10−4	6.6×10−9	■	●
ZBTB42	zinc finger and BTB domain containing 42	14	7.5×10−4	2.7×10−8	■
IDO1*	indoleamine 2,3-dioxygenase 1 (aka INDO)	8	1.1×10−3	2.4×10−10	■		●	●
LOC642639	Unknown	9	1.2×10−3	1.7×10−3	■
LTC4S*	leukotriene C4 synthase	5	2.5×10−3	6.9×10−6	■	●
CYP4F12	cytochrome P450, family 4, subfamily F, polypeptide 12	19	4.2×10−3	2.0×10−5	■	●

^*Known association with IgE.

^●known association with eosinophils.

GACRS=Genetics of Asthma in Costa Rica Study

CAMP=Childhood Asthma Management Program

MSigDB=Molecular Signature Database.

^■Significant in both cohorts after adjustment for recent inhaled corticosteroid (ICS) use.

Table 3;

Gene-ontology defined biological processes associated with total IgE using two different statistical approaches in two different cohorts: GACRS and CAMP cohorts.

GO biological process	# Genes in GO Group	Differential Gene Expression		WGCNA
		GACRS & CAMP		GACRS		CAMP
		Fold Enrichment	Raw P value	Fold Enrichment	Raw P value	Fold Enrichment	Raw P value
regulation of leukocyte mediated immunity (GO:0002703): CLC, PIK3R6	167	9.7	0.0038	6.3	0.0014	10.5	0.0030
regulation of leukocyte proliferation (GO:0070663) : IDO1, CLC	219	7.4	0.0080	5.8	0.0007	8.0	0.0064
regulation of lymphocyte mediated immunity (GO:0002706): CLC, PIK3R6	126	12.9	0.0017	6.7	0.0034	13.9	0.0014
regulation of lymphocyte proliferation (GO:0050670) : IDO1, CLC	210	7.7	0.0071	6.0	0.0006	8.4	0.0057
regulation of mononuclear cell proliferation (GO:0032944) : IDO1, CLC	211	7.7	0.0072	6.0	0.0006	8.3	0.0058
regulation of T cell tolerance induction (GO:0002664): IDO1, CLC	12	89.9	0.0003	35.1	0.0019	97.4	0.0003
regulation of tolerance induction (GO:0002643): IDO1, CLC	18	60.0	0.0006	23.4	0.0040	64.9	0.0005
neuron fate commitment (GO:0048663): OLIG1, OLIG2	70	15.4	0.0079	12.0	0.0004	16.7	0.0067
oligodendrocyte differentiation (GO:0048709) : OLIG1, OLIG2	65	16.6	0.0068	13.0	0.0003	27.0	0.0002
positive regulation of glial cell differentiation (GO:0045687): OLIG2, CLC	33	32.7	0.0019	19.1	0.0007	35.4	0.0016
positive regulation of gliogenesis (GO:0014015) : OLIG2, CLC	51	31.7	0.0001	16.5	0.0001	22.9	0.0037
regulation of glial cell differentiation (GO:0045685) : OLIG2, CLC	59	18.3	0.0057	14.3	0.0002	19.8	0.0049

Overrepresentation Analysis

To further clarify the immunologic relevance of the 31 validated transcripts, we compared our gene set to a database of immunologic signatures, yielding three gene sets that were enriched (Supplementary Table 6). These three gene sets encompassed gene transcripts upregulated in neutrophils (GPR44, CCR3, IL5RA, CLC, SIGLEC8; FDR q value<0.001), bone-marrow derived macrophages (EMR1, CCL23, OLIG2, OLIG1; q<0.011), and eosinophils (SMPD3, CCL23, IDO1, CEBPE; q<0.011).

Discussion:

Novel biological therapies targeting total IgE-mediated pathways have widened clinical treatment options for asthma. A better understanding of the biology underlying total IgE levels should help refine and improve these therapeutic approaches. We used differential gene expression approaches and network analysis, followed by a variety of pathway enrichment techniques, to identify novel gene transcripts co-regulated with determinants of serum IgE levels in individuals with asthma. The combination of statistical approaches allowed us to identify the most biologically relevant transcripts working in combination to influence total IgE. Crucially, we had access to two independent populations and achieved a high level of replication, with 52% of transcripts replicated using GLM, and 37% using WGCNA. Combining the results of these two methods yielded ultimately 31 biologically- relevant and informative transcripts that were associated with serum total IgE. It is also notable that the key gene co-expression network module associated with IgE, was also associated with the more stringent definition of asthma based on physician diagnosis and methacholine challenge, suggesting that these 31 replicated gene transcripts reflect not simply an IgE-related gene signature but also an allergic asthma signature. This network module was associated with the use of inhaled corticosteroid in the CAMP cohort, though not in the GACRS cohort. While it is difficult to determine if this is due to confounding with severity of asthma, we reported that all of transcripts reported in Table 2 remained significantly associated with IgE levels by differential gene expression even after adjusting for the use of inhaled corticosteroids.

Within the pathway enrichment analyses of the total IgE-associated transcripts, we identified the immune relevant GO category, “regulation of tolerance induction,” which was highlighted due to the upregulation of Charcot-Leyden Crystal galectin (CLC) and indoleamine-pyrrole 2,3-dioxygenase (IDO1) in association with increased total IgE levels. Both genes are expressed in eosinophils and their activity is stimulated following binding of IgE. CLC levels have been positively associated with eosinophilic airway inflammation and aspirin induced asthma (31, 32) and have also been associated with expansion of T regulatory cells(33). IDO1 encodes an enzyme involved in tryptophan catabolism and suppresses helper and effector T cells while activating regulatory T cells (34). These findings are notable given that regulatory T cell activity is thought to be deficient in allergic asthma (35), yet we found a positive association of increasing IgE with increased IDO1 and CLC. We hypothesize that this represents a regulatory mechanism induced to curb the allergic immune response in the setting of asthma. This hypothesis is supported by findings describing a spectrum of increased IDO1 activity that is higher among atopic individuals compared to healthy controls, and highest among asymptomatic atopic subjects compared to symptomatic ones, suggesting that IDO1 may be involved in inducing a state of clinical unresponsiveness in allergic individuals.(36) Further study is needed to clarify the roles of IDO1 and serum IgE in asthma.

Overrepresentation analysis suggested the presence of various immunologic signatures such as enrichment of neutrophil, macrophage, and eosinophil-associated genes. In particular, comparison with a gene set of eosinophils compared to Th1 cells showed that 4 of our 31 transcripts overlapped, including SMPD3, CCL23, IDO1, and CEBPE. These findings support the involvement of eosinophil-associated genes in total IgE regulation. At least 20 of these 31 validated transcripts were previously found to be upregulated in eosinophils (37, 38). Esnault et al. compared gene expression in sputum eosinophils from individuals with mild asthma before and after allergen challenge, and found 365 upregulated transcripts (37). Sixteen of our 31 transcripts overlapped with that set, providing further validation for the genes as being strongly associated with the asthmatic response to allergen (Table 2). Esnault et al. also conducted experiments comparing the changes in allergen-induced gene expression in broncheoalveolar lavage cells, before and after treatment with mepolizumab, in order to block IL-5 pathways. Of the gene transcripts that were upregulated after allergen exposure but down-regulated after mepolizumab treatment, 10 of our transcripts overlapped, suggesting involvement in IL-5 related pathways of inflammation in asthma (CYP4F12, EMR4P, IL5RA, LGALS12, LTC4S, PIK3R6, PRSS33, SIGLEC8, SPNS3, VSTM1). In another gene expression analysis of eosinophils compared with mast cells, neutrophils, and mononuclear cells, Nakajima et al. identified 30 eosinophil-specific genes (38); 9 of our 31 transcripts overlapped with that set, further supporting the association between eosinophilic inflammation and total IgE levels. While we found high levels of replication between our populations and with other microarray data, we also compared our findings with GWAS studies of IgE and were unable to find similarities (24, 39). However, given the limited overlap between IgE associated GWAS findings and those of asthma (39), it is not surprising that our replicated transcripts associated with IgE in asthmatics did not coincide with these prior association signals.

The most notable aspect of this analysis is the high degree of replication of the identified transcripts, using two distinct statistical methods in two different patient populations, as well as in comparison to gene sets described in the literature. Some of these genes, such as IL5RA and GPR44 (CRTH2, receptor for Prostaglandin D2), represent known therapeutic targets, with IL-5 receptor antagonists showing some efficacy in uncontrolled eosinophilic asthma (40). CCR3 (eotaxin-3), a chemokine receptor, has been associated with allergic asthma, though CCR3 targeted therapies appear to be less efficacious in humans than in mouse models (41, 42). In addition, the leukotriene pathway has been a therapeutic target for asthma for decades, though specific inhibitors for LTC4S (leukotriene C4 synthase) have only recently been developed and evaluated in mouse models (43, 44). The identification of these known targets using our analytical approach corroborates the possibility that the other transcripts in this set may be equally useful as therapeutic targets.

Other genes from our eosinophil-related gene sets have known associations with allergic asthma and may also have therapeutic or diagnostic potential, including CCL23, CLC, CEBPE, CAMK1, IDO1, SIGLEC8, SMPD3, and SPNS3 (45–52). Potential therapeutic targets include cell surface receptors, such as Siglec8, an inhibitory receptor on eosinophils with polymorphisms associated with asthma (46–48). CCL23 is an eosinophilic chemokine, that has been identified as a possible biomarker for anti-IL13 targeted therapies (49, 50). Several of these are enzymes expressed within eosinophils, such as CLC, IDO1, SMPD3, and SPNS3, and the latter two are both involved in the sphingolipid pathways, known to mediate airway hyperresponsiveness and mast cell activation (51, 52). Another gene CAMK1 has been implicated as an asthma control gene(20), but has not been identified previously with IgE or an allergy-related phenotype among individuals diagnosed with asthma. Of the calmodulin dependent kinases, CAMK2 and CAMK4 have been implicated in bronchoconstriction and smooth muscle proliferation (53, 54), suggesting that CAMK1 may also play a role in asthma pathogenesis.

We also report the expression of a number of novel genes, including EMR1, EMR4P, and PIK3R6, which were significant in our data and have been shown to be upregulated in eosinophilic asthma in other studies. These genes have no known mechanistic association with total IgE or asthma but have been implicated in allergic processes such as allergic rhinosinusitis and anaphylaxis (55, 56). EMR1 and EMR4P encode G-protein coupled receptors that may be particularly useful as therapeutic targets. While these genes do not have known roles in IgE or asthma, their replication in both of our cohorts as well as other published gene sets, implicates them as potentially novel biomarkers or therapeutic targets.

Our study design and analysis does have some limitations. The cross-sectional nature of the expression and phenotype data limit the ability to make true causal inferences. The use of whole blood gene expression profiles may not be as physiologically relevant as lung tissue- specific profiling; however, the use of less invasive sampling methods allows for greater sample sizes and more robust replication. While it could prove useful to split the whole blood expression data into specific cell populations, these techniques are in development and still face significant limitations, including concerns about accuracy and ability to detect expression profiles of cell subsets (57, 58). Recent studies support the utility of whole blood transcriptomics in the setting of asthma (59, 60). We cannot exclude the possibility that these transcripts play a role in atopic disease in general; however the WGCNA analysis showing an association with a stricter definition of asthma in the GACRS population, and a lack of association with either atopic dermatitis or allergic rhinitis, supports a more prominent role in allergic asthma. Also, GO based enrichment analyses are subject to annotation bias and can often yield high level information about biological processes that are ubiquitous to all cellular function; thus we attempted to improve the biological relevance of this analysis by placing more stringent criteria on fold enrichment and the number of genes required.

This analysis benefits from the GACRS population being a highly homogeneous relative genetic isolate with consistent and controllable environmental exposures, detailed clinical information, and minimal population stratification. We have abundant data demonstrating that the genetic findings in the GACRS cohort are generalizable to CAMP. There are over 15 studies of validated susceptibility loci replicated in GACRS and CAMP with similar effect size and direction of effect, whether identified by GWAS (61–67), by linkage and positional cloning (68–70), or by candidate gene testing (23, 65, 71–74). However, we acknowledge that there are ethnic, genetic and environmental differences between the GACRS population and the North American CAMP population, which may have limited our ability to find additional shared gene probes. Focusing on only replicated gene transcripts may have missed important gene probes relevant only in a single population and could have limited the detection of novel biological targets. In addition, the gene expression profiling was performed on childhood samples from GACRS and adulthood samples from CAMP. Given the striking overlap in our findings over both ethnicity and age, the generalizability of these results is a strength rather than a limitation.

In conclusion, we present robust and reproducible evidence for consistent gene expression profiles of serum total IgE in individuals with asthma. Indeed we found similar profiles with clear biological relevance regardless of the statistical approach used; multivariable regression or network-based approaches. Pathway enrichment identified an important role for regulation of T cell tolerance. In addition to gene transcripts with known or suspected relationships with IgE related pathways, we identified a number of novel transcripts. Many of the genes encoding these transcripts have been reported in other studies of asthma suggesting that they may represent novel therapeutic targets or biomarkers. Further functional validation of these transcripts and their roles in influencing total IgE and asthma will be essential to developing novel strategies for improved primary prevention, management and treatment of asthma.

Abbreviations:

IgE

Immunoglobulin E

GACRS

Genetics of Asthma in Costa Rica Study’

CAMP

The Childhood Asthma Management Program

IQR

inter-quartile ranges

GLM

Generalized linear regression models

WGCNA

Weighted Gene Co-expression Network Analysis

FDR

false discovery rate

GO

Gene Ontology