Study of gene–gene interactions for endophenotypic quantitative traits in Chinese asthmatic children

Authors


Dr Christopher W. K. Lam
Department of Chemical Pathology
The Chinese University of Hong Kong
Prince of Wales Hospital
Shatin
Hong Kong

Abstract

Background:  Asthma is a complex disease resulting from interactions between multiple genes and environmental factors. Study of gene–gene interactions could provide insight into the pathophysiology of asthma.

Methods:  We investigated the interactions among 18 single-nucleotide polymorphisms in eight candidate genes for plasma total immunoglobulin E (IgE) concentration and peripheral blood (PB) eosinophil count in 298 Chinese asthmatic children and 175 controls. Generalized multifactor dimensionality reduction and generalized linear model were used to analyze gene–gene interactions for the quantitative traits.

Results:  A significant interaction was found between R130Q in IL13 and I50V in IL4RA for plasma total IgE concentration, with a cross-validation (CV) consistency of nine of 10 and a prediction error of 41.1% (= 0.013). Plasma total IgE concentration was significantly higher in the high-risk than the low-risk groups (< 0.0001). For PB eosinophil count, significant interaction was found between C-431T in TARC and RsaI_in2 in FCERIB, with a CV consistency of nine of 10 and a prediction error of 40.2% (= 0.009). PB eosinophil count was significantly higher in the high-risk group than the low-risk groups (< 0.0001). Generalized linear model also revealed significant gene–gene interaction for the above two endophenotypes with = 0.013 for plasma total IgE concentration and = 0.029 for PB eosinophil count respectively.

Conclusions:  Our data suggest significant interactions between IL13 and IL4RA for plasma total IgE concentration, and this is the first report to show significant interaction between TARC and FCERIB for PB eosinophil count in Chinese asthmatic children.

Asthma is a common and complex chronic respiratory disease resulting from interactions between multiple genes and environmental factors (1, 2). Elevations in plasma immunoglobulin E (IgE) levels, peripheral blood (PB) eosinophil count and bronchial hyper-responsiveness (BHR) are physiological traits that are characteristic of asthma. To date, more than 35 genes have been associated with asthma or related phenotypes (3, 4) but not a single factor or gene was found to have a strong and independent effect on asthma (5, 6). Therefore, searching for the interactions between susceptibility genes and environmental factors will be one of the breakthroughs to dissect this complex disease. We have been the first group to apply multifactor dimensionality reduction (MDR) (7) for asthma phenotypes, and found R130Q of interleukin-13 gene (IL13) together with I50V of interleukin-4 receptor-α gene (IL4RA) and C-431T of thymus and activation-regulated chemokine gene (TARC) respectively, to confer increased risks of asthma and increased plasma total IgE concentration in Chinese children (8). Recently, Battle et al. (9) also reported significant interaction between IL13 and IL4RA for asthma phenotypes in African–Americans using regression analyses.

The advantages of MDR over traditional gene-gene interaction methods like logistic regression, multi-locus linkage disequilibrium and Hardy–Weinberg equilibrium are that it is model-free and can detect high-order gene–gene interaction in relatively small sample size case–control studies (10, 11). However, the limitation of MDR is that it applies to dichotomous phenotypes only. Generalized multifactor dimensionality reduction (GMDR) method (12) is a recent advance of MDR that is applicable to both continuous and dichotomous phenotypes and permits adjustment for discrete and quantitative covariates in various population-based study designs with unbalanced case–control samples.

In this study we genotyped 18 polymorphic markers in eight candidate genes that have been associated with asthma or asthma phenotypes in Chinese asthmatic patients and control subjects (Table 1). These included genes for (1) cytokines: IL13, IL4RA; (2) chemokine: TARC; (3) mediator under innate immunity: CD14; and (4) mediators of atopy and airway inflammation: ADRB2, FCERIB, PTGS2 and CLTA4. Gene–gene interactions for plasma total IgE concentration and PB eosinophil count were examined by GMDR method.

Table 1.   Candidate genes and variants analyzed in this study
ChromosomeCandidate genePolymorphismAssociation
  1. IgE, immunoglobulin E; BHR, bronchial hyper-responsiveness.

1q25.2-25.3PTGS2C-4231A;
A-1195G;
PTGS2.5209T→G;
PTGS2.8473T→C
Asthma, atopy and lung function
2q33CTLA4−1147CT; +49 A/G;
CT60; JO31;
JO30; JO27_1
Atopy, asthma
5q31-33IL13R130QIgE, asthma
5q31CD14C-159TIgE
5q32-34ADRB2R16G; E27QIgE, BHR, treatment response, nocturnal asthma
11q13FCERIBRsaI_in2; RsaI_ex7Atopy, IgE, BHR
16p12IL4RAI50VAtopy, IgE, atopic asthma
16q13TARCC-431TAtopy, asthma

Methods

Study population

This study recruited unrelated children aged between 5 and 18 years with chronic stable asthma, diagnosed according to the American Thoracic Society guideline (13), from pediatric outpatient clinics of a university teaching hospital in Hong Kong. Both parents of these subjects are ethnic Chinese. Age- and gender-matched Chinese control subjects were selected among the children attending the hospital for minor complaints, who did not have allergic or immunological disease, and did not require any drug treatment at the time of this study. All subjects were free from any self-reported symptoms of infection for  the last 4 weeks before study. Subjects or their parents gave written consent, and the Clinical Research Ethics Committee of our university approved this study.

Atopic markers in peripheral blood

Plasma total IgE concentration was measured by microparticle immunoassay (IMx analyzer; Abbott Laboratories, Abbott Park, IL, USA) and specific IgE antibodies to crude Dermatophagoides pteronyssinus, cat, dog, mixed cockroaches and mixed molds were measured by fluorescent enzyme immunoassay (AutoCAP system; Phadia AB, Uppsala, Sweden). Plasma total IgE concentration was analyzed as a quantitative trait following logarithmic transformation (IgElog). Specific IgE concentration ≥0.35 kIU/l was considered positive. Children were classified as atopic if they had at least one type of allergen-specific IgE antibody (14). Eosinophils in PB were enumerated using a Coulter STKS counter (Beckman-Coulter, Miami, FL, USA).

Spirometry

Patients underwent spirometric assessment (COMPACT II, Vitalograph, Buckingham, England) to measure their forced expiratory volume in 1 s and forced vital capacity. These results were compared with local references (15).

Genotyping

Polymerase chain reaction (PCR) and restriction fragment-length polymorphism (RFLP) were used to genotype the 18 single-nucleotide polymorphisms (SNPs) as listed in Table 1. Detailed protocols of PCR and RFLP are described in supplementary Table S1. The results of these RFLP assays were validated by direct sequencing of the SNPs using Big Dye Terminator Cycle sequencing kits with an ABI-3100 autosequencer (Applied Biosystems, Foster City, CA, USA) in 30 randomly selected samples.

Statistical analysis

The results of plasma total IgE and allergen sensitization as risk factors for the development of asthma between different groups were compared using Student t-test, chi-squared or Fisher exact test as appropriate. Allele frequencies were estimated by gene-counting method, and chi-squared test was used to examine the Hardy–Weinberg equilibrium. Association of asthma and plasma total IgE with 18 SNPs were analyzed by Pearson chi-squared test.

Generalized multifactor dimensionality reduction (version 0.7; Department of Psychiatry and Neurobehavioral Sciences, University of Virginia, Charlottesville, VA; http://www.healthsystem.virginia.edu/internet/addiction-genomics/software/; accessed 31 January 2008) was used to determine high- and low risk on the basis of the work of Lou et al. (12). GMDR is an advancement of MDR (7) that permits adjustment for discrete and quantitative covariates and is applicable to both continuous and dichotomous phenotypes. This method uses the same data-reduction strategy as the original MDR method (7), but score-based statistics using maximum-likelihood estimates is introduced to classify multifactor cells into two different groups (12). We determined statistical significance by comparing the average prediction error from the observed data with the distribution of average prediction errors under the null hypothesis of absence of association derived empirically from 5000 permutations. The null hypothesis was rejected when the P-value derived from the permutation test was 0.05 or lower.

Standard analysis of variance test was used to compare plasma total IgE concentration among different groups according to GMDR classification while Student t-test was used for subgroup analyses. Kruskal–Wallis test was used for analyzing PB eosinophil count among different groups according to GMDR classification while subgroup comparisons were done by Mann–Whitney U-test.

Generalized linear model using spss v14.0 (SPSS Inc., Chicago, IL, USA) was performed to confirm the results from GMDR analyses. A P value of <0.05 was considered statistically significant.

Results

Clinical and laboratory characteristics

Two hundred and ninety-eight asthmatic children and 175 controls were recruited, with their mean (standard deviation) ages being 10.4 (3.7) and 11.0 (3.8) years respectively (= 0.101). The demographic and clinical characteristics of these subjects are shown in supplementary Table S2. Increased plasma total IgE concentration, sensitization to D. pteronyssinus, cat, mixed cockroaches and mixed molds and elevated PB eosinophil count were the significant risk factors for asthma diagnosis.

Asthma and individual polymorphisms

The allele frequencies for the 18 SNPs tested are shown in Table 2. Except E27Q, all SNPs examined followed Hardy–Weinberg equilibrium. The genotyping completion rate for each SNP exceeded 97%. Apart from PTGS2.8473T/C (= 0.023), there was no significant association between asthma diagnosis and these SNPs (Table 2).

Table 2.   Allele and genotype frequencies of the SNPs
PolymorphismAsthmaticsControlsP*
  1. SNP, single-nucleotide polymorphisms.

  2. q: Minor allele frequency.

  3. *Compared by Pearson chi-squared test.

  4. †Compared by chi-squared test with Yates’ correction.

−4231C/An = 297n = 173 
CC264155 
CA3318 
AA000.972†
q0.060.050.818
−1195A/Gn = 292n = 173 
AA6536 
AG14392 
GG84450.676
q0.470.470.847
PTGS2.5209T/Gn = 298n = 173 
TT270153 
TG2818 
GG020.164†
q0.050.060.273
PTGS2.8473T/Cn = 298n = 175 
TT191125 
TC9940 
CC8100.023
q0.190.170.410
C-1147Tn = 291n = 175 
CC216129 
CT6843 
TT730.855
q0.140.140.970
+49 A/Gn = 272n = 171 
AA4021 
AG11975 
GG113750.748
q0.370.340.473
CT60n = 293n = 173 
GG183109 
GA9758 
AA1360.878
q0.210.200.782
JO31n = 295n = 172 
GG15092 
GT12369 
TT22110.825
q0.280.260.542
JO30n = 296n = 173 
GG15690 
GA11869 
AA22140.965
q0.280.280.825
JO27_1n = 294n = 174 
TT14689 
TC11768 
CC31170.939
q0.300.290.715
R130Qn = 273n = 141 
RR9454 
RQ13670 
QQ43170.531
q0.410.370.291
C-159Tn = 269n = 128 
CC5526 
CT13477 
TT80380.653
q0.450.460.915
R16Gn = 295n = 173 
RR10151 
RG13589 
GG59330.462
q0.430.450.568
E27Qn = 294n = 173 
EE1921 
EQ4319 
QQ2321330.073
q0.140.180.113
RsaI_in2n = 274n = 167 
CC170104 
CT8957 
TT15600.651
q0.220.210.710
RsaI_ex7n = 291n = 167 
TT267154 
CT2313 
CC100.749
q0.040.040.768
I50Vn = 295n = 167 
II7949 
IV15980 
VV57380.448
q0.460.470.899
C-431Tn = 266n = 161 
CC9568 
CT14183 
TT30100.142
q0.380.320.087

GMDR analysis

Table 3 shows the results of cross-validation (CV) consistency and prediction error for plasma total IgE concentration by GMDR analysis of the data set of patients and control subjects. One two-locus model had a minimum prediction error of 41.1% (= 0.013) and a maximum CV consistency of nine of 10. This two-locus model consisted of the R130Q in IL13 and I50V in IL4RA (Fig. 1). As there were only three patients and five controls in the genotype combination of IL13 130Q/Q with IL4RA 50V/V, this combination was not classified into either high- or low-risk groups (unclassified). On the other hand, we were able to classify the combinations of IL4RA 50I/I with IL13 130R/R and 130R/Q as the low-risk group, and all others as the high-risk groups. In this model, plasma total IgE concentration was significantly different among the four groups (< 0.0001; Fig. 2). IgElog were significantly higher in high-risk patients than in low-risk patients (mean: 2.65 vs 2.38; = 0.003).

Table 3.   Multi-loci interaction model for plasma total IgE concentration by GMDR method
Loci number and combinationCross-validation consistencyPrediction error (%)P-value based on 5000 permutations
  1. GMDR, generalized multifactor dimensionality reduction; IgE, immunoglobulin E.

1 locus: I50V843.630.033
2 locus: I50V, R130Q941.090.013
3 locus: I50V, R130Q, JO30438.120.001
4 locus: I50V, R130Q, C-159T, R16G243.310.055
5 locus: I50V, C-431T, C-159T, R16G, +49A/G245.040.142
Figure 1.

 Best 2-loci model for plasma total immunoglobulin E concentration derived from analyses of 18 variants. High-risk genotypes are in dark and low-risk genotypes are in grey. The white box is unclassified. The figures above the bars (left bars represent asthmatics while right bars are controls) are score while those in bracket are the subject number.

Figure 2.

 Box plot of logarithmic plasma total immunoglobulin E concentrations in high-risk and low-risk groups defined by the interaction of IL13 and IL4RA model (< 0.0001 by anova while subgroup analyses were done by Student t-test with < 0.001 for high-risk controls vs low-risk patients and = 0.003 for low-risk vs high-risk patients).

Table 4 shows the results of CV consistency and prediction error for PB eosinophils by GMDR analysis. One two-locus model had a minimum prediction error of 40.2% (= 0.009) and a maximum CV consistency of nine of 10. This two-locus model consisted of the C-431T in TARC and RsaI_in2 in FCERIB (Fig. 3). As there was no patient and control in the genotype combination of TARC -431T/T with FCERIB RsaI_in2T/T, this combination was not classified into either high- or low-risk groups (unclassified). On the other hand, we were able to classify the combinations of TARC -431C/C with FCERIB RsaI_in2C/T, TARC -431C/T with FCERIB RsaI_in2T/T and TARC -431T/T with FCERIB RsaI_in2C/C as high-risk groups, and all others as the low-risk groups. In this model, the PB eosinophil count was significantly different among the four groups (< 0.0001; Fig. 4). Counts were significantly higher in high-risk patients than in low-risk patients (median: 9.0%vs 7.0%; = 0.017).

Table 4.   Multi-loci interaction model for peripheral blood eosinophils by GMDR method
Loci number and combinationCross-validation consistencyPrediction error (%)P-value based on 5000 permutations
  1. GMDR, generalized multifactor dimensionality reduction.

1 locus: R130Q844.620.094
2 locus: C-431T, RsaI_in2940.220.009
3 locus: C-431T, RsaI_in2, R16G745.290.145
4 locus: C-431T, RsaI_in2, R16G, -1195A/G840.440.022
5 locus: C-431T, RsaI_in2, R16G, -1195A/G, JO27_1742.090.113
Figure 3.

 Best 2-loci model for peripheral blood eosinophils derived from analyses of 18 variants. High-risk genotypes are in dark, and low-risk genotypes are in grey. The white box is unclassified. The figures above the bars (left bars represent asthmatics while right bars are controls) are score while those in bracket are the subject number.

Figure 4.

 Box plot of peripheral blood eosinophil in high-risk and low-risk groups defined by the interaction of TARC and FCERIB model (< 0.0001 by Kruskal–Wallis test while subgroup analyses were done by Mann–Whitney U-test with = 0.030 for high-risk controls vs low-risk patients and = 0.017 for low-risk vs high-risk patients).

Generalized linear model

A significant interaction between R130Q in IL13 and I50V in IL4RA on plasma total IgE concentration was also found by generalized linear model (= 0.013), adjusting for age and gender as covariates. Significant interaction between these two SNPs was also found in patient group (= 0.038). Among the 18 SNPs, only R130Q showed a significant association with plasma total IgE (= 0.035).

There was a significant interaction between C-431T in TARC and RsaI_in2 in FCERIB on the PB eosinophil count found by generalized linear model (= 0.029), adjusting for age and gender as covariates. Significant interaction between these two SNPs was also found in patient group (= 0.040) (All SNPs alone were not associated with PB eosinophil count).

Discussion

By using GMDR and generalized linear models, this study suggests significant gene-gene interactions between (i) IL13 and IL4RA for plasma total IgE concentration and (ii) TARC and FCERIB for PB eosinophil count in Chinese asthmatic children. Genome screens with classical linkage and fine mapping approaches suggest that susceptibility to asthma is determined by many genes that have a moderate effect. Palmer et al. (16) using Gibbs sampling (Markov Chain Monte Carlo)-based segregation analysis of asthma-associated quantitative traits found that they are under both major loci effect and polygenic influence. Thus, a multi-loci approach should be used to analyze the quantitative traits of a complex disorder. Our finding of epistatsis between IL13 and IL4RA for plasma total IgE concentration is consistent with the similar results in African–Americans (9) and Dutch (17). We have previously reported that R130Q of IL13 was associated with total IgE (14) and similar result was reported by Hunninghake et al. (18). IL4RA has previously been associated with numerous atopic conditions (19, 20) and our group first reported significant associations between TARC C-431T and atopy, aeroallergen sensitization and peripheral eosinophilia in Chinese children (21). FCERIB has also been associated with atopy, aeroallergen sensitization, and PB eosinophil count in Caucasian (22, 23). In this study, we found significant gene-gene interactions between two of 18 SNPs, demonstrating the sensitivity and power of GMDR method in detecting polygenic interactions for endophenotypic quantitative traits among different loci in multiple asthma and atopy candidate genes.

Asthma is a complex disease characterized by airway obstruction, airway inflammation and BHR to a variety of stimuli (24). The development of allergic asthma exists in three phases, namely the induction phase, the early-phase asthmatic reaction and the late-phase asthmatic reaction. Each phase is characterized by the production and interplay of various cell-derived mediators (25, 26). Both mast cells and basophils express the IgE receptor FCERI and degranulate after re-exposure to allergens. The release of multi-functional cytokines and mediators enhance lymphocyte recruitment and T helper 2 (TH2) cytokine productions (27). Both IL-13 and IL-4 are produced by TH2 cells and capable of inducing immunoglobulin class switching of B cells to produce IgE after allergen exposure. These cytokines share a common receptor component, IL-4Rα (28, 29). The Q130 polymorphism of IL13 had a lower affinity with the IL-13 receptor α2 chain, a decoy receptor, causing less clearance. This variant was also demonstrated to have enhanced stability in both human and mouse plasma (30). Another study showed that the combination of IL4RA V50 and R551 variants resulted in the expression of an IL-4Rα with enhanced sensitivity to IL-4 (31). The co-existence of both functional SNPs probably interacts to augment this IL-13/IL-4Rα signaling. In fact, the R130Q polymorphism is located within the D helix, close to the C-terminal region of IL-13 (32). It is possible that, as in IL-4 binding to IL-13R, the D helix of IL-13 interacts with the IL-4RA subunit and the R130Q polymorphism influences such interaction (33). The risk association of these gene–gene combinations was consistent with the complementary role of both molecules in IgE switching. It is therefore possible that different polymorphisms in IL4RA and IL13 contribute to the complex regulation of atopy or asthma phenotypes.

Eosinophils are the central effector cells in late phase airway inflammation (25, 34). Thymus and activation-regulated chemokine (TARC; CCL17) is a key chemokine for attracting TH2 lymphocytes into the site of allergic inflammation (35). Our recent studies suggested that plasma TARC concentration was significantly associated with chronic stable asthma (36) and the severity of asthma exacerbation in children (37). Increased TARC production was also associated with a state of systemic allergic inflammatory response. The resulting TH2 activation could lead to the increase in IgE production and recruitment of eosinophils from bone marrow under the actions of the important TH2 cytokines IL-4 and IL-5, respectively (38). TARC C-431T is functional (39), and several studies reported that C-431T was associated with increased plasma TARC concentration (21, 40). In our earlier study (21), we also found that subjects with TT genotype had significantly higher PB eosinophil percentage. This is consistent with our 2-loci model by GMDR analysis, i.e. TARC -431T/T with FCERIB RsaI_in2C/C as the high-risk group. The other two combinations for high-risk group were TARC -431C/C with FCERIB RsaI_in2C/T and TARC−431C/T with FCERIB RsaI_in2T/T. These two combinations are ‘checkerboard’ genotypes (containing exactly one heterozygote and one homozygote), which are the properties of a purely epistatic model.

The gene–gene interaction between TARC and FCERIB for eosinophilia suggests that these molecules might have interacting pathways that influence PB eosinophils. In fact, in the studies of anti-IgE antibody omalizumab on patients with allergic asthma (41) and seasonal allergic rhinitis (42), it was found that apart from decreasing free IgE levels and down-regulating the expression of FCERI on human basophils, mast cells and other inflammatory cells (43), omalizumab also has profound and significant effects on circulating and tissue eosinophils. Omalizumab decreases the levels of circulating IgE by binding to the constant region (cε3) of the IgE molecule, which prevents free IgE from interacting with IgE receptors (FCERI and FCERII). The effect of omalizumab on eosinophils might be explained in part by inhibition of the allergen-IgE-mast cell response, which prevents mast cell activation, cytokine release and subsequent eosinophil chemotaxis. Korzycka-Zaborowska et al. (44) have reported that the intronic variant of RsaI_in2 in FCERIB is related to atopy and may influence its development. Together with the functional C-431T in the promoter of TARC, we speculate that the interaction between FCERIB and TARC would affect allergen presentation, TH2-cell activation and subsequent eosinophil chemotaxis.

Asthma does not follow a typical pattern of Mendelian inheritance that can be explained by a linear model (45). The epistatic nature of our interaction models indicate that the influence of each genotype on plasma total IgE and PB eosinophil count appear to be dependent on the genotypes at each of the other loci. Sorting out the functional basis of the interactions in these 2-loci2-loci models remains an interpretive challenge.

Eosinophilia is thought to be under genetic control, and linkage peaks have been mapped to the 5q31-33 (46, 47). As reviewed by Adamko et al. (48), granulocyte-macrophage colony-stimulating factor (GM-CSF), IL-3 and IL-5 are the key cytokines in the induction and regulation of eosinophils. Genetic studies have been performed on these genes with asthma outcomes (49–51), and Kabesch et al. (52) reported a significant gene–gene interaction between IL-5 and GM-CSF with atopic outcomes. However, no data on eosinophil production, eosinophils in PB or eosinophilia in the lung was available. Thus, the gene(s) causing eosinophilia in asthma remain undetermined. This is the first report to show gene-gene interaction for PB eosinophil count in asthmatic children.

In this study, PB eosinophils have been expressed both as % of total leukocytes and absolute count. Both values were significantly higher in asthmatics than in controls. However, in the gene–gene interaction for PB eosinophils, we reported PB eosinophils in terms of % of total leukocytes which may account for any increase in other leukocytes such as neutrophils during asthma inflammation. Besides, we also analyzed the gene–gene interaction for PB eosinophils in absolute count and similar results were obtained.

One of the limitations in this study was that only one to several SNPs per candidate gene was chosen. These numbers of SNPs were not sufficient to capture most genetic information of the candidate genes. Future studies should elaborate on our findings by including more haplotype tagging or functional SNPs in the studied candidate genes.

In summary, by using GMDR, this study demonstrates significant gene–gene interactions between IL13 and IL4RA for plasma total IgE concentration, and TARC and FCERIB for PB eosinophil count in Chinese asthmatic children. These results are consistent with those obtained from generalized linear analyses. Increased plasma total IgE concentration and elevated PB eosinophils are results of polygenic effect and environmental factors. These interaction data can provide useful insights into the genetic architecture of asthma for understanding the disease better, fostering development of novel and effective therapeutic strategy.

Acknowledgments

We thank Dr Ming D. Li and Guo-Bo Chen of Section of Neurobiology, University of Virginia, Charlottesville, for providing the GMDR software and advising us on the use of this program. This study was supported by a Direct Grant for Research of the Chinese University of Hong Kong.

Ancillary