Discovery and Mediation Analysis of Cross-Phenotype Associations Between Asthma and Body Mass Index in 12q13.2

Abstract

Twin studies suggest that shared genetics contributes to the comorbidity of asthma and obesity, but candidate-gene studies provide limited evidence of pleiotropy. We conducted genome-wide association analyses of asthma and body mass index (BMI; weight (kg)/height (m)²)) among 305,945 White British subjects recruited into the UK Biobank in 2006–2010. We searched for overlapping signals and conducted mediation analyses on genome-wide-significant cross-phenotype associations, assessing moderation by sex and age at asthma diagnosis, and adjusting for confounders of the asthma-BMI relationship. We identified a genome-wide-significant cross-phenotype association at rs705708 (asthma odds ratio = 1.05, 95% confidence interval: 1.03, 1.07; P = 7.20 × 10⁻⁹; and BMI β = −0.065, 95% confidence interval: −0.087, −0.042; P = 1.30 × 10⁻⁸). rs705708 resides on 12q13.2, which harbors 9 other asthma- and BMI-associated variants (all P < 5 × 10⁻⁵ for asthma; all but one P < 5 × 10⁻⁵ for BMI). Follow-up analyses of rs705708 show that most of the BMI association occurred independently of asthma, with consistent magnitude between men and women and persons with and without asthma, irrespective of age at diagnosis; the asthma association was stronger for childhood versus adult asthma; and both associations remained after confounder adjustment. This suggests that 12q13.2 displays pleiotropy for asthma and BMI. Upon further characterization, 12q13.2 might provide a target for interventions that simultaneously prevent or treat asthma and obesity.

Abbreviations

BMI

body mass index

CI

confidence interval

GWA

genome-wide association

LD

linkage disequilibrium

SNP

single nucleotide polymorphism

Asthma and obesity affect millions of people globally (1–3), contributing substantially to morbidity and health costs. The rising prevalences of both diseases since the 1980s (4, 5) have suggested a link between them (6). Cross-sectional studies confirm the asthma-obesity association (7–10), and prospective studies suggest that obesity is a risk factor for asthma (11, 12). Asthma is also a risk factor for obesity (13), and sex and age at asthma diagnosis might moderate these relationships (13). However, the mechanism(s) underlying these epidemiologic observations are not completely understood.

Pleiotropy might contribute to the comorbidity of asthma and obesity (14). Twin studies show that asthma and obesity, as measured by body mass index (BMI), are genetically correlated (15, 16) and thus might share a portion of their genetic architectures. The posited pleiotropic loci might provide insights into the mechanistic underpinnings of this comorbidity and potentially serve as targets for interventions that simultaneously prevent or treat both diseases. Discovery of pleiotropic drug targets for asthma and obesity is vital to addressing the needs of patients with this comorbidity, because available asthma medications are less effective among obese persons with asthma (17).

Previous empirical searches for pleiotropic loci for asthma and obesity-related traits primarily used candidate-gene approaches (18–24), and their findings were limited by key methodological constraints. First, some tested for linkage/association with obesity/BMI only among people with asthma. This precludes assessments of cross-phenotype associations (i.e., marginal associations with both phenotypes). Second, most studies failed to account for the relationship between asthma and obesity/BMI in the study populations and confounders of this relationship. This is essential for distinguishing between cross-phenotype associations due to biological, mediated, or spurious pleiotropy, which, in turn, has important implications for clinical/public health interventions targeting pleiotropic loci (25). Third, candidate-gene approaches are limited by gaps in our understanding of the genetic architectures of asthma and obesity. Indeed, the most comprehensive study (21)—assessing 38 genomic regions from previous univariate genome-wide association (GWA) studies of BMI, obesity, and asthma—found insufficient evidence of cross-phenotype associations. Therefore, variants displaying pleiotropy for asthma and BMI/obesity might reside in loci not yet known to be associated with these phenotypes or might have gone undetected due to constraints of prior research.

To explore this hypothesis, we conducted GWA analyses of asthma and BMI within the UK Biobank and searched for cross-phenotype associations. We dissected genome-wide-significant cross-phenotype associations via mediation analyses, controlling for the asthma-BMI relationship in our study population and multiple confounders of that relationship, to distinguish between cross-phenotype associations due to biological, mediated, or spurious pleiotropy. We characterized these cross-phenotype associations further by assessing potential moderation by sex and age at asthma diagnosis.

METHODS

Data access and ethical approval

This research was conducted using the UK Biobank Resource (application number 27477). The UK Biobank study was conducted under generic approval from the National Health Services’ National Research Ethics Service. The present analyses were approved by the Human Investigations Committee at Yale University (institutional review board protocol number 2000022579).

Study population

Our study population (Table 1) consisted of 305,945 White British subjects from the UK Biobank, a population-based prospective cohort study of approximately 500,000 subjects. Recruitment and data collection have been described previously (26–28). Briefly, participants aged 40–69 years were recruited at 22 centers across the United Kingdom between 2006 and 2010. Clinical and anthropometric measures, self-reported health and lifestyle factor data, and blood samples for subsequent DNA extraction and direct genotyping were collected. Genotyping, performed by Affymetrix (Santa Clara, California), used 2 arrays—the UK Biobank Axiom Array and the UK BiLEVE Array—which share over 95% of marker content (28). We analyzed subjects who passed our sample and genetic quality control procedures (Web Appendix 1 and Web Figure 1, available at https://academic.oup.com/aje) and had complete asthma and BMI data at baseline.

Table 1.

Characteristics of Our Study Population, UK Biobank, 2006–2010^a,b

		Mediation Analysis (n = 202,922) d
Characteristic	GWA Analysis c (n = 305,945)		Nonasthmatic (n = 181,304)		Childhood Asthmatic e (n = 4,817)		Adult Asthmatic f (n = 16,801)
	No.	%	No.	%	No.	%	No.	%
Asthma
No	270,572	88.4
Yes	35,373	11.6
Weight status			NC	NC	NC		NC	NC
Underweight	1,510	0.5
Normal weight	99,701	32.6
Overweight	130,844	42.8
Obese	73,890	24.2
BMIg	27.40 (4.80)		27.19 (4.69)		27.27 (4.77)		28.23 (5.44)
Age at baseline, yearsg	56.90 (8.00)		56.15 (8.03)		54.35 (8.16)		55.46 (8.20)
Sex
Female	164,539	53.8	103,080	56.9	2,088	43.3	10,850	64.6
Male	141,406	46.2	78,224	43.2	2,729	56.7	5,951	35.4
Was breastfed	NC	NC
No			51,589	28.5	1,397	29.0	5,279	31.4
Yes			129,715	71.6	3,420	71.0	11,522	68.6
Exposure to maternal smoking	NC	NC
No			127,995	70.6	3,353	69.6	11,398	67.8
Yes			53,309	29.4	1,464	30.4	5,403	32.2
Smoking status at asthma diagnosis	NC	NC
Never a regular smoker			126,364	69.7	4,813	99.9	11,982	71.3
Former regular smoker			42,242	23.3	0	0.0	2,480	14.8
Current regular smoker			12,698	7.0	4	0.1	2,339	13.9

Abbreviations: BMI, body mass index; GWA, genome-wide association; NC, not considered.

^a Table values are n and column percentages, except where otherwise indicated; some percentages do not sum to 100% due to rounding.

^b NC indicates that that characteristic was not considered in that analysis.

^c Subjects included in GWA analysis are of White British ancestry (UK Biobank field 22006.0.0), passed our sample and genetic quality control procedures (detailed in the Web Appendix 1 and summarized in Web Figure 1), and had complete phenotype data for asthma and BMI at baseline.

^d Subjects included in the mediation analyses are those with complete data on rs705708, asthma, BMI, and all potential confounders (Web Figure 11).

^e Childhood asthmatic subjects are those with a reported age at asthma diagnosis of 1–10 years.

^f Adult asthmatic subjects are those with age at asthma diagnosis of 11–70 years.

^g Values for BMI and age are mean (standard deviation). BMI was calculated as weight (kg)/height (m)².

Phenotype definitions

All phenotypic data analyzed were collected at baseline. Further details and covariate definitions are in Web Appendix 1.

BMI.

Quantitative BMI (UK Biobank field 21001.0.0) was calculated (weight (kg)/height (m)²) using height and weight collected by trained staff. For qualitative BMI analyses, participants were classified as underweight (BMI < 18.5; n = 1,510), normal weight (18.5 ≤ BMI < 25.0; n = 99,701), overweight (25.0 ≤ BMI < 30.0; n = 130,844), or obese (BMI > 30.0; n = 73,890), in line with the internationally recognized World Health Organization criteria (29).

Asthma.

Asthma (yes/no) was ascertained via the question “Has a doctor ever told you that you had any of the following conditions?” Participants selecting asthma from the list of conditions (fields 6152.0.1–6152.0.4) were classified as asthmatic (n = 35,373), and all others as nonasthmatic (n = 270,572).

Age at asthma diagnosis.

Self-reported age at asthma diagnosis (field 3786.0.0) was available for 31,366 of 35,373 people with asthma. Two subgroups (defined a priori to capture asthma before and after puberty; Web Appendix 1) were created: childhood asthmatic (n = 6,832; age at diagnosis: ≤10 years) and adult asthmatic (n = 24,534; age at diagnosis: >10 years).

Association analyses

Quality control was performed within PLINK (30). Genetic correlation was estimated using variance components analysis in BOLT-REML (31). GWA analyses were performed using linear mixed models in BOLT-LMM (32), which analyzes both quantitative and case-control traits (Web Appendix 1). BOLT-LMM computes association statistics using a standard infinitesimal mixed model first, and then with a Bayesian Gaussian mixture model, if the prediction accuracy of the best-fit Gaussian mixture model exceeds that of the infinitesimal model by at least a specified amount (32). Results presented for each phenotype are from the last model fitted (see BOLT-LMM calibration statistics and diagnostics in Web Table 1 and Web Figures 2–9). For case-control traits, estimated single-nucleotide polymorphism (SNP) effect sizes were converted on the quantitative scale to traditional odds ratios via the approximation: , where is the case fraction (32). The corresponding standard errors were also divided by . All models included a single term for the SNP, representing the number of minor alleles and assuming an additive mode of inheritance, and adjusted for age (years; field 21003.0.0) and sex (male/female; field 31.0.0), which were significantly associated with asthma and BMI (Web Tables 2 and 3).

The genome-wide significance, genome-wide suggestive, and nominal significance thresholds were α = 5 × 10⁻⁸, 5 × 10⁻⁵, 0.05, respectively. Manhattan plots were created using the qqman package in R (R Foundation for Statistical Computing, Vienna, Austria) (33). Regional plots were created in Locus Zoom (34) and adapted to display associations with both phenotypes and PLINK-derived empirical estimates of linkage disequilibrium (LD).

Model building for mediation analyses

The asthma-BMI relationship was assessed using linear models. A cross-product term was included to test for sex × asthma interaction. Case-only analyses were used to assess moderation by age at asthma diagnosis.

Age, sex, smoking status at asthma diagnosis, maternal exposure to smoking, and breastfeeding status were evaluated as mediator-outcome confounders based on their previously reported associations with and temporal ordering in relation to asthma diagnosis and baseline BMI. For each confounder, bivariate associations with BMI, for all subjects and by asthma, were tested using simple linear regression (Web Table 3). Bivariate associations with asthma (Web Table 2) and with rs705708 (Web Table 4) were tested using the χ² test for categorical variables and t test or 1-way analysis of variance for continuous variables.

Additional model-building steps included tests for 2-way interaction between rs705708 and asthma (exposure-mediator interaction) and rs705708 and sex by including appropriate cross-product terms in mediator and/or outcome models. All analyses were performed in R (R Foundation for Statistical Computing) (35). Further details are in Web Appendix 1.

Mediation analyses

Given the temporal ordering of the phenotypes (asthma diagnosis occurred prior to the baseline BMI measure; Web Figure 10), mediation analyses considered asthma the mediator and aimed to decompose estimates of the total effect of rs705708 on BMI into direct and indirect effect estimates. Mediation analyses were performed using the “mediation” R package (36). We included (n = 202,922) subjects with complete data on rs705708, asthma, BMI, and all confounders (Web Figure 11). To account for moderation of the asthma-BMI relationship by age at asthma diagnosis, 2 separate analyses were performed. Analysis 1 included subjects with a childhood diagnosis of asthma (n = 4,817), analysis 2 included subjects with a adult diagnosis of asthma (n = 16,801), and both included a common set of nonasthmatic subjects (n = 181,304). For each analysis, minimal-adjustment (model set 1) and full-adjustment (model set 2) mediator (asthma) and outcome (BMI) models were fitted (see model equations in Web Appendix I). Direct and indirect SNP effects were estimated as described by Imai et al. (37). Moderated mediation by sex was assessed in both analyses because of the observed interactions between sex, asthma, and age at diagnosis in their relationships with BMI. Confidence intervals (95%) for all effect estimates are bias-corrected and accelerated intervals (38) estimated via nonparametric bootstrapping. Significance of direct and indirect effect estimates was evaluated against = 0.05.

RESULTS

GWA analyses

BMI.

Of 528,797 SNPs tested, 1,699 were genome-wide significant for quantitative BMI (Web Figure 12), and 488 were significant for qualitative BMI (Web Figure 13; case-control GWA of overweight/obese vs. normal-weight/underweight). The genome-wide significant SNP set for quantitative BMI included 452 (92.6%) of the qualitative BMI hits, and all but 7 of the remaining SNPs were at least genome-wide-suggestive. Therefore, quantitative BMI results were used to search for cross-phenotype associations with asthma.

Asthma.

In case-control analyses including all asthmatic persons, 1,457 SNPs were genome-wide-significant (Web Figure 14). Most of these associations did not differ by age at asthma diagnosis. Indeed, in case-only gene × environment interaction analyses comparing childhood versus adult asthmatic subjects (Web Appendix 1, Web Figures 15–17), genome-wide-significant evidence of heterogeneity was detected for only 107 (7.3%) of the 1,457 asthma-associated SNPs. In total, 124 SNPs had genome-wide-significant gene × environment interaction P values, and 107 (86.3%) of these were included in the genome-wide significant set for all asthmatic subjects. Therefore, results for analyses including all asthmatic subjects were used to search for cross-phenotype associations with BMI.

Cross-phenotype associations

Asthma and quantitative BMI were positively genetically correlated in our study population (r_g = 0.1439, standard error = 0.0111), suggesting potential cross-phenotype associations. Indeed, 652 (45%) of the 1,457 genome-wide-significant SNPs for asthma were at least nominally significant for BMI (Figure 1A); and 444 (26%) of the 1,699 genome-wide-significant SNPs for BMI were at least nominally significant for asthma (Figure 1B). Of note, 25 SNPs were genome-wide significant for both phenotypes: 24 SNPs in chromosome 6 near or in the major histocompatibility complex region and 1 SNP in chromosome 12 (12q13.2). None of the 24 chromosome-6 SNPs was significantly associated with asthma after conditioning on the top asthma GWA signal (Affx-37,072,023 in the major histocompatibility complex, class II, DQ α 1 gene (HLA-DQA1), which was not associated with BMI (data not shown)). Therefore, only the genome-wide-significant cross-phenotype association in 12q13.2—rs705708—was examined further.

Figure 1.

Overlap in genome-wide association analysis signals between asthma and body mass index (BMI), UK Biobank, 2006–2010. A) In genome-wide association (GWA) analyses of asthma, 1,457 single nucleotide polymorphisms (SNPs) achieved genome-wide significance. Of these SNPs, 652 (45%) were at least nominally significant for BMI. B) In GWA analyses of BMI, 1,699 SNPs achieved genome-wide significance. Of these SNPs, 444 (26%) were at least nominally significant for asthma. Asthma results are from analyses comparing 35,373 asthmatic subjects versus 270,572 nonasthmatic subjects in our study population of White British ancestry, with P values from the BOLT-LMM (32) infinitesimal mixed model (the last model fitted, Web Table 1). BMI results are for the quantitative BMI analysis (n = 305,945), with P values from the BOLT-LMM Gaussian mixture model (the last model fitted, Web Table 1). Both analyses adjusted for age and sex. The genome-wide significance, genome-wide suggestive, and nominal significance thresholds for each analysis were 5 × 10⁻⁸, 5 × 10⁻⁵ and .05, respectively. No association was defined as P > 0.05.

Web Figure 18 shows that rs705708 is 1 of 10 12q13.2 SNPs displaying cross-phenotype associations with asthma and BMI. Among the other 9 SNPs, there are 7 genome-wide-significant associations with asthma and 8 genome-wide-suggestive associations with BMI. These SNPs are in at least moderate LD with rs705708 (D´ = 0.55–0.96; r² = 0.21–0.55; Web Tables 5 and 6), and the magnitude and direction of their associations with each phenotype are consistent with those of rs705708 (Table 2). Consistent with regional LD patterns, conditional analyses show that the BMI and asthma-associations at these SNPs are not independent of one another (data not shown).

Table 2.

Cross-Phenotype Associations in 12q13.2 Among Subjects Included in the Genome-Wide Association Analyses (n = 305,945), UK Biobank, 2006–2010^a

SNP	Gene	Base Pair	Effect/Reference Allele	EAF	Asthma			BMI
					OR	95% CI	P Value b	β c	95% CI	P Value d
rs2069408	CDK2	56,364,321	G/A	0.3388	1.04	1.02, 1.06	3.30 × 10−6	−0.061	−0.084, −0.037	5.40 × 10−7
rs1873914	RAB5B	56,379,427	C/G	0.4237	1.06	1.04, 1.08	2.40 × 10−12	−0.045	−0.068, −0.023	7.90 × 10−5
rs705702e	SUOX	56,390,636	G/A	0.3376	1.07	1.05, 1.09	3.10 × 10−14	−0.053	−0.077, −0.030	1.10 × 10−5
rs10876864e	SUOX	56,401,085	G/A	0.4279	1.06	1.04, 1.08	1.50 × 10−12	−0.050	−0.072, −0.027	1.60 × 10−5
rs1701704	IKZF4	56,412,487	G/T	0.3433	1.07	1.05, 1.09	1.50 × 10−14	−0.062	−0.085, −0.038	3.70 × 10−7
rs2456973	IKZF4	56,416,928	C/A	0.3432	1.07	1.05, 1.09	1.50 × 10−14	−0.061	−0.084, −0.037	6.00 × 10−7
rs11171739e	ERBB3	56,470,625	C/T	0.4337	1.06	1.04, 1.07	8.80 × 10−11	−0.050	−0.073, −0.028	1.10 × 10−5
rs2292239	ERBB3	56,482,180	T/G	0.3470	1.07	1.05, 1.08	4.50 × 10−13	−0.061	−0.084, −0.037	4.20 × 10−7
rs705708	ERBB3	56,488,913	A/G	0.4712	1.05	1.03, 1.07	7.20 × 10−9	−0.065	−0.087, −0.042	1.30 × 10−8
rs11171747e	ESYT1	56,518,408	T/G	0.6180	1.04	1.02, 1.05	2.90 × 10−5	−0.059	−0.036, −0.082	4.50 × 10−7

Abbreviations: BMI, body mass index; CDK2, cyclin dependent kinase 2; CI, confidence interval; EAF, effect allele frequency; ERBB3, Erb-B2 receptor tyrosine kinase 3; ESYT1, extended synaptotagmin 1; IKZF4, Ikaros family zinc finger protein 4; OR, odds ratio; RAB5B, Ras-related protein Rab-5B; SNP, single-nucleotide polymorphism; SUOX, sulfite oxidase.

^a Results shown for SNPs with P < 5 × 10⁻⁵ for asthma and P < 0.05 for BMI.

^b P value from BOLT-LMM (32), calculated using the standard infinitesimal mixed model, the last model fitted (Web Table 1).

^c β values represent the change in BMI per minor allele of rs705708; BMI calculated as weight (kg)/height (m)².

^d P value from BOLT-LMM, calculated using the Gaussian mixture model, the last model fitted (Web Table 1).

^e For intergenic SNPs, the nearest gene is listed, with priority given to genes directly downstream of variant.

Asthma-BMI relationship in our study population

Mediation analyses of rs705708 were motivated by the relationship between a prior asthma diagnosis and baseline BMI in our study population (Table 3). Among all subjects (n = 305,945), mean BMI was greater among asthmatic subjects than nonasthmatic subjects (28.20, 95% confidence interval (CI): 28.15, 28.25 vs. 27.33, 95% CI: 27.15, 27.51; P = 3.25 × 10⁻²³⁰). The magnitude of this BMI difference was greater among women than men (P = 1.51 ×10⁻⁷³). The asthma-BMI relationship also differed by age at asthma diagnosis. In case-only analyses (n = 31,366 asthmatic subjects), mean BMI was greater among persons with asthma diagnosed in adulthood than among those diagnosed in childhood (28.31, 95% CI: 28.24, 28.38 vs. 27.38, 95% CI: 27.25, 27.51; P = 3.90 × 10⁻³⁷). The magnitude of this BMI difference was also greater in women than in men (P = 1.44 × 10⁻³).

Table 3.

Body Mass Index of Asthmatic and Nonasthmatic Persons Included in the Genome-Wide Association Analyses (n = 305,945), Overall and According to Sex, UK Biobank, 2006-2010^a,b

Subgroup	Nonasthmatic (n = 270,572)		All Asthmatic (n = 35,373) c		P Value d , e	Childhood Asthmatic (n = 6,832) f		Adult Asthmatic (n = 24,534) g		P Value e , h
	Mean	95% CI	Mean	95% CI		Mean	95% CI	Mean	95% CI
All subjects	27.33	27.15, 27.51	28.20	28.15, 28.25	3.25 ×10−230	27.38	27.25, 27.51	28.31	28.24, 28.38	3.90 ×10−37
Women	26.88	26.86, 26.9	28.16	28.10, 28.22	4.92 ×10−251	27.09	26.89, 27.29	28.29	28.21, 28.37	3.61 ×10−22
Men	27.79	27.76, 27.82	28.10	28.02, 28.18	3.66 ×10−15	27.59	27.43, 27.75	28.31	28.20, 28.42	2.17 ×10−19

Abbreviations: BMI, body mass index; CI, confidence interval.

^a BMI values were calculated as weight (kg)/height (m)².

^b Results for all subjects are adjusted for age and sex; results for men and women are adjusted for age only.

^c All asthmatic subjects are those reporting an asthma diagnosis (UK Biobank data field 6152.0); only 31,366 of these subjects had data on age at asthma diagnosis (UK Biobank data field 3786.0).

^d P value from linear regression comparing BMI of all asthmatic versus nonasthmatic persons.

^e Mean BMI differences between asthmatic and nonasthmatic subjects and between adult and childhood asthmatic subjects were moderated by sex (P for interaction = 1.51 × 10⁻⁷³ and P for interaction = 1.44 × 10⁻³, respectively).

^f Childhood asthmatic subjects are those with age at asthma diagnosis of 1–10 years.

^g Adult asthmatic subjects are those with age at asthma diagnosis of 11–70 years.

^h P value for linear regression comparing BMI of childhood versus adult asthmatic subjects.

Mediation analyses

Table 4 summarizes the multivariable-adjusted associations between rs705708 and asthma estimated in mediation models. The direction of the estimated SNP effect on asthma in model set 1 and model set 2 was consistent with that of GWA estimates (Table 2). However, the rs705708-asthma association was stronger in analysis 1 (model set 2, odds ratio = 1.10, 95% CI: 1.05, 1.14) than analysis 2 (model set 2, odds ratio = 1.04, 95% CI: 1.01, 1.06). This is consistent with our case-only gene × environment interaction analysis results (Web Table 7).

Table 4.

Multivariable-Adjusted Associations Between rs705708 and Asthma Based on Mediation Models, UK Biobank, 2006–2010^a

Model	OR	95% CI	P Value b
Analysis 1 (n = 186,121)c
Model set 1	1.10	1.05, 1.14	1.58 × 10−5
Model set 2	1.10	1.05, 1.14	1.73 × 10−5
Analysis 2 (n = 198,105)d
Model set 1	1.04	1.02, 1.06	7.03 × 10−5
Model set 2	1.04	1.01, 1.06	9.55 × 10−5

Abbreviations: CI, confidence interval; OR, odds ratio.

^a Mediation model equations are available in Web Appendix 1.

^b P value from logistic regression.

^c Analysis 1 includes asthmatic subjects diagnosed in childhood (n = 4,817) and the common set of nonasthmatic subjects (n = 181,304).

^d Analysis 2 includes asthmatic subjects diagnosed in adulthood (n = 16,801) and the common set of nonasthmatic subjects (n = 181,304).

Table 5 summarizes our estimates of the total effects, controlled direct effects, and controlled indirect effects (effects through asthma) of rs705708. In both analyses, most of the estimated total effect of rs705708 on BMI is direct (model set 2, controlled direct effect estimate = −0.0655, 95% CI: −0.0976, −0.0364 in analysis 1 and −0.0582, 95% CI: −0.0894, −0.0305 in analysis 2). The magnitude and direction of these effect estimates did not differ between persons with and without asthma or between men and women (Web Appendix 1). The estimated controlled indirect effect was small in absolute magnitude in both analyses, and its direction reflected the sex-specific asthma-BMI relationships observed herein (Table 3 and Web Table 8). Notably, in analysis 1, the controlled indirect effect estimate was masked when considering the population average because the direction of the effect estimate was negative in men but positive in women.

Table 5.

Estimates of the Total, Direct, and Indirect Effects of rs705708 on Body Mass Index Based on Mediation Models, UK Biobank, 2006–2010^a

	Analysis 1 (n = 186,121) b						Analysis 2 (n = 198,096) c
Effect Component	Model Set 1			Model Set 2			Model Set 1			Model Set 2
	β d	95% CI	P Value	β d	95% CI	P Value	β d	95% CI	P Value	β d	95% CI	P Value
Total effect	−0.06600	−0.09558, −0.03606	<0.001	−0.06556	−0.09751, −0.03639	<0.001	−0.0574	−0.08695, −0.02635	<0.001	−0.05601	−0.08632, −0.02746	<0.001
Controlled direct effecte	−0.06600	−0.09579, −0.03615	<0.001	−0.06545	−0.09762, −0.03641	<0.001	−0.0598	−0.08985, −0.02943	<0.001	−0.05824	−0.08944, −0.03053	<0.001
Controlled indirect effect (population average)	−0.00002	−0.00040, 0.00056	0.920	−0.00011	−0.00042, 0.00050	0.990	0.00241	0.00004, 0.00569	0.048	0.00223	0.00003, 0.00605	0.046
P value for moderation of indirect effect by sex			<0.001			<0.001			0.106			0.140
Controlled indirect effect (among men)	−0.00073	−0.00162, −0.0002	<0.001	−0.00071	−0.00148, −0.00023	0.006
Controlled indirect effect (among women)	0.00023	0.00019, 0.00143	0.006	0.00082	0.00015, 0.00140	0.006

Abbreviations: BMI, body mass index; CI, confidence interval; SNP, single nucleotide polymorphism.

^a Mediation model equations and further details are available in Web Appendix 1. The total effect estimate is the sum of the controlled direct and indirect effect estimates (population averages). Direct and indirect SNP effect estimates were averaged across values of the mediator (asthma) because we did not detect significant evidence of SNP × asthma interactions (P = 0.971 and P = 0.167, respectively, when considering subjects in Analysis 1 and Analysis 2; Web Appendix 1).

^b Analysis 1 includes those with asthma diagnosed in childhood (n = 4,817) and the common set of nonasthmatic subjects (n = 181,304).

^c Analysis 2 includes those with asthma diagnosed in adulthood (n = 16,801) and the common set of nonasthmatic subjects (n = 181,304).

^d β values represent the change in BMI per minor allele of rs705708; BMI calculated as weight (kg)/height (m)².

^e Controlled direct effect estimates did not differ between men and women in either analysis; table values represent the population averages.

DISCUSSION

We have presented the findings of a comprehensive search for genetic loci displaying pleiotropy for asthma and BMI. Relative to previous investigations, our study employed a hypothesis-free GWA approach, expanding the search space beyond loci with previously reported asthma and/or BMI/obesity associations. This allowed us to discover a locus (12q13.2) harboring genome-wide-significant associations with both phenotypes, which to our knowledge, was previously associated with asthma (39) but not with BMI at the time of study.

We characterized the 12q13.2-BMI association by: 1) conducting mediation analyses that assessed to what extent the asthma-BMI relationship explained the association; and 2) considering potential moderation by sex and by age at asthma diagnosis. Our results suggest that most of the estimated total effect of 12q13.2 (tagged by rs705708, our sentinel cross-phenotype association) on BMI is independent of asthma and remains after adjustment for common determinants of asthma and BMI. Furthermore, while sex and age at asthma diagnosis moderate the asthma-BMI relationship in our study population (Table 3), the magnitude and direction of the estimated direct effect on BMI is consistent between men and women and between nonasthmatic and asthmatic subjects, irrespectively of age at diagnosis (Table 5). With respect to the 12q13.2-asthma association, we show that it is consistent between men and women (Web Appendix 1) and slightly stronger in childhood versus adult asthmatic subjects (Table 3 and Web Table 7). Altogether, these robust associations suggest that 12q13.2 displays pleiotropy for asthma and BMI.

Through a clinical or public health lens, this pleiotropic locus is of interest because it might provide opportunities for intervening to treat asthma and obesity simultaneously, rather than treating each disease separately. It might also stimulate joint prevention efforts: For example, genetic screening for 12q13.2 variants could identify asthmatic subjects at risk of developing obesity (25). However, for this to become possible, additional studies are needed to thoroughly characterize our candidate pleiotropic locus.

First, whether or how 12q13.2 contributes to asthma-obesity comorbidity needs to be clarified. Herein, the 12q13.2 alleles were positively associated with asthma and negatively associated with BMI/obesity (see Web Table 9 for qualitative BMI results), which is counter to the overall epidemiologic expectation for these phenotypes. Such discrepancies between the direction of genetic and phenotypic associations have been reported for other comorbidities (40) and might be explained by the antagonistic pleiotropy hypothesis for senescence (41, 42). To extend longevity, genetic predisposition to one disease (e.g., asthma) could be accompanied by selection for genetic protection from associated comorbidities (e.g., obesity). While antagonistic pleiotropy might indeed underlie our findings, it is also possible that the 12q13.2-BMI relationship varies with age (as reported for other BMI/obesity loci, e.g., as in Graff et al. (43)) and that the direction of association differs between pediatric and adult populations. Alternatively, 12q13.2 could be associated with increases in BMI upon asthma diagnosis (which we could not investigate because data on BMI at time of or before asthma diagnosis was not available in the UK Biobank). Therefore, analyses assessing the relationship between 12q13.2 and BMI trajectory should be an important component of future research.

It will also be important to identify the causal variant(s) and unravel the mechanism(s) by which 12q13.2 variants affect BMI and asthma. Regarding the first point, we note that locus 12q13.2 (chr12:56,372,585-56,482,185; GRCh37/hg19 human genome assembly) is broad and multigenic, and our cross-phenotype associations span multiple genes (cyclin dependent kinase 2 (CDK2), Ras-related protein Rab-5B (RAB5B), sulfite oxidase (SUOX), Ikaros family zinc finger protein 4 (IKZF4), ribosomal protein S26 (RPS26), Erb-B2 receptor tyrosine kinase 3 (ERBB3), and extended synaptotagmin 1 (ESYT1)). Notably, rs705708, our sentinel cross-phenotype association and top regional BMI signal, resides in ERBB3, but the top regional asthma signal (rs2456973) resides in IKZF4, a known asthma gene (39). This could mean that each variant tags a trait-specific causative variant within different genes. Alternatively, rs705708 and rs2456973 could be in LD with the same causative variant in ERBB3, IKZF4, or another gene in 12q13.2. Nonetheless, empirical LD estimates suggest that the alleles at 12q13.2 are in at least moderate LD with one another. Therefore, even if there are multiple causal variants, they are likely co-inherited. Regarding the second point, we note that the relationship between 12q13.2 variants and these phenotypes is biologically plausible. With respect to rs705708, Genotype-Tissue Expression (GTEx) project data (44) show that the T allele is associated with greater RPS26 expression in lung and subcutaneous and visceral adipose tissue. With respect to ERBB3 and its encoded protein (receptor tyrosine kinase ErbB3, a member of the epidermal growth factor family of receptors), studies in bronchial epithelial cell lines show ErbB2/ErbB3 signaling supports development and regulation of the airway permeability barrier (45). Moreover, epidermal growth factor receptor pathway misactivation is associated with airway proliferation and mucus overproduction/hypersecretion (46). Additionally, downregulation of ErbB2/ErbB3 in the presence of insulin promotes adipose tissue differentiation (47). Still, mechanistic explanations about the role of 12q13.2 variants in regulating BMI and asthma require additional functional data.

Our study had several strengths—including our use of a large population-based cohort that did not select subjects based on their asthma status and our series of follow-up/sensitivity analyses (Web Appendix 1)—but it is not without limitations. First, our main asthma definition was based on a single questionnaire item: self-report of physician-diagnosed asthma. While self-reported asthma is commonly used in population-based studies, has been rigorously evaluated, and displays high specificity, it also displays low sensitivity (48–50). Therefore, we cannot discard the possibility of misclassification of asthmatic cases and controls in our study. Because misclassification is likely nondifferential with respect to genotype, any resulting bias in the SNP effect estimates for asthma is likely toward the null. Such a bias might have, in turn, underestimated the indirect effect of 12q13.2 on BMI through asthma (51). Second, our main asthma phenotype is an “asthma ever” phenotype, capturing asthmatic subgroups with different disease trajectories (e.g., past vs. current asthmatic subjects) and ages at diagnosis. In choosing this definition, we assumed there are commonalities in the genetic architecture of different asthmatic subgroups, which is supported by previous studies (52) and our case-only gene × environment analyses comparing childhood versus adult asthma. Nevertheless, we acknowledge that a heterogeneous asthma definition might have limited the power of our GWA analyses and that the asthma-BMI relationship could vary across asthmatic subgroups that were not explored in the present analyses. Still, in sensitivity analyses that refined our asthma phenotype further (e.g., to distinguish between asthma and chronic obstructive pulmonary disease; Web Appendix 1, Web Table 10, and Web Figures 19–21) our findings were robust to heterogeneity in the asthma definition. Third, our study used age at diagnosis to distinguish between asthma diagnosed before and after puberty (“childhood” vs. “adult” asthma) because Tanner stage at the time of asthma diagnosis was not available for our study population. Notwithstanding potential misclassification, we observed differences in the epidemiologic profile of asthma by pubertal status that are consistent with previous research (e.g., greater percentage of women among those diagnosed with asthma as adults than those diagnosed as children (Table 1)). We acknowledge that, by creating age-at-diagnosis subgroups that capture pubertal status, we adopted a cutoff for adult asthma lower than that used in some previous studies. However, to our knowledge, there is no standard age cutoff for adult asthma (53). Nonetheless, in sensitivity analyses, we raised it to age 20 years and found no appreciable differences in our results (data not shown). Fourth, our asthma and BMI analyses included underweight subjects (n = 1,510), and underweight could be a sign of illness. Nonetheless, sensitivity analyses showed that neither our effect estimates, nor our conclusions would have been different had we excluded these subjects (data not shown). Last, our mediation analyses considered asthma the mediator because BMI was measured at baseline and asthma diagnosis typically occurred prior to baseline in the UK Biobank (Web Figure 10). However, we note that subjects’ baseline BMI likely reflects historical BMI, and, to account for this, we would have had to incorporate data on BMI at the time of or before asthma diagnosis, which was not available for our study population.

Notwithstanding these limitations, our study positions 12q13.2 as a novel candidate pleiotropic locus for asthma and BMI. The present work lays the foundation for future research, which we hope will provide mechanistic insights into the relationship between 12q13.2 and asthma and BMI/obesity. This will be essential to realizing the potential clinical and public health impact of our candidate pleiotropic locus.