A genome-wide association meta-analysis of self-reported allergy identifies shared and allergy-specific susceptibility loci.

Allergic disease is very common and carries substantial public-health burdens. We conducted a meta-analysis of genome-wide associations with self-reported cat, dust-mite and pollen allergies in 53,862 individuals. We used generalized estimating equations to model shared and allergy-specific genetic effects. We identified 16 shared susceptibility loci with association P<5×10(-8), including 8 loci previously associated with asthma, as well as 4p14 near TLR1, TLR6 and TLR10 (rs2101521, P=5.3×10(-21)); 6p21.33 near HLA-C and MICA (rs9266772, P=3.2×10(-12)); 5p13.1 near PTGER4 (rs7720838, P=8.2×10(-11)); 2q33.1 in PLCL1 (rs10497813, P=6.1×10(-10)), 3q28 in LPP (rs9860547, P=1.2×10(-9)); 20q13.2 in NFATC2 (rs6021270, P=6.9×10(-9)), 4q27 in ADAD1 (rs17388568, P=3.9×10(-8)); and 14q21.1 near FOXA1 and TTC6 (rs1998359, P=4.8×10(-8)). We identified one locus with substantial evidence of differences in effects across allergies at 6p21.32 in the class II human leukocyte antigen (HLA) region (rs17533090, P=1.7×10(-12)), which was strongly associated with cat allergy. Our study sheds new light on the shared etiology of immune and autoimmune disease.

Allergic disease is very common and carries substantial public health burdens. We conducted a meta-analysis of genome-wide association with self-reported cat, dust-mite, and pollen allergy in 53,862 individuals. We used generalized estimating equations to model shared and allergy-specific genetic effects. We identified 16 shared susceptibility loci with P<5×10−8, including 8 loci previously associated with asthma, as well as 4p14 near TLR1, TLR6, and TLR10 (rs2101521: P=5.3×−21); 6p21.33 near HLA-C and MICA (rs9266772: P=3.2×10−12); 5p13.1 near PTGER4 (rs7720838: P=8.2×10−11); 2q33.1 in PLCL1 (rs10497813: P=6.1×10−10); 3q28 in LPP (rs9860547: P=1.2×10−9); 20q13.2 in NFATC2 (rs6021270: P=6.9×10−9); 4q27 in ADAD1 (rs17388568: P=3.9×10−8); and 14q21.1 near FOXA1 and TTC6 (rs1998359: P=4.8×10−8). We identified one locus with substantial evidence for differences in effects across allergies, at 6p21.32 in the class II HLA region (rs17533090: P=1.7×10−12), strongly associated with cat allergy. Our study sheds new light on the shared etiology of immune and autoimmune disease.

Allergies and allergic asthma are among the most common diseases in the industrialized world. In the United States, a nationwide survey showed that over half the population tested positive for sensitization to at least one common allergen, a considerable increase over results collected approximately ten years earlier[1]. The cause of this apparent increase in prevalence is unknown, but the rapidity with which it has occurred implicates an environmental component[2]. Still, estimates of the heritability of allergy are high[3,4], suggesting that understanding the genetic liability underlying these conditions is key to understanding the disease.

A number of genes implicated in allergy and asthma through association and functional studies belong to pathways involved in immune and inflammatory processes, such as innate immunity, adaptive immunity, and allergic inflammation[5]. These genes belong to a range of gene families that encode toll-like receptors, interleukins, chemokines, and various other signaling molecules and transcription factors. Published genome-wide association studies (GWAS) on allergic conditions have focused on asthma and atopic dermatitis, resulting in a substantial number of loci associated with asthma (HLA-DQ, IL33, IL18R1, SMAD3, GSDMA, IL2RB, RORA, GSDMB, IL13, SLC22A5, DENND1B, PDE4D, ORMDL3, IL6R, 5q22.1, 11q13.5)[6-11] and a smaller number with atopic dermatitis (FLG, OVOL1, ACTL9, 5q22.1, 11q13.5, 20q13.33)[12-14]. Studies using other measures of atopy have been less definitive, likely due to limited sample sizes[15-17]; the largest, of allergic rhinitis and IgE sensitization to grass pollen, identified three regions with genome-wide significance (class II HLA, 5q22.1, 11q13.5).

We selected three common self-reported allergy phenotypes -- pollen allergy, dust mite allergy, and cat allergy -- for which comparable data were available in the 23andMe participant cohort[18] and in a cohort of mothers from the Avon Longitudinal Study of Parents and Children (ALSPAC)[19] (Table 1). We used generalized estimating equations (GEE) to jointly model genetic effects across all three phenotypes. The GEE approach accounts for the correlations between the phenotypes, and enables us to estimate both shared and allergy-specific effects. We first performed a genome-wide meta-analysis of GEE tests for shared effects. Then, for a set of 3725 markers with nominal evidence of association with at least one allergy, we performed GEE tests for allergy-specific effects (Supplementary Table 1).

Table 1

Demographic characteristics of cohorts.

	23andMe		ALSPAC
	N	%	N	%
Total	46646	100.0	7216	100.0
Gender
Male	26344	56.5	0	0.0
Female	20302	43.5	7216	100.0
Age
age <= 30	4300	14.6	4829	67.0
30 < age <= 45	8088	31.1	2382	33.0
45 < age <= 60	6282	25.6	0	0.0
60 < age	6428	28.7	0	0.0
Allergy status
cat allergy	10509	22.5	704	9.8
dust mite allergy	9815	21.0	964	13.4
pollen allergy	16133	34.6	1201	16.6
Number of allergies
three allergies	4947	10.6	328	4.6
any two allergies	6228	13.3	536	7.4
any one allergy	9160	19.6	813	11.3
no allergies	26311	56.4	5539	76.8

In the GEE meta-analysis for shared effects across allergies, we identified 16 genome-wide-significant loci with P<5×10−8 (Table 2, Fig. 1, Supplementary Fig. 1). Of these, 8 had P<5×10−8 in the 23andMe cohort and P<0.05 in the ALSPAC cohort (Supplementary Tables 2 and 3). We identified 6 loci with suggestive evidence for association (5×10−8 < P < 1×10−6) (Supplementary Note). Many of these loci have previously been associated with other immunity related phenotypes, and 8 have been associated with asthma in previous GWAS (Supplementary Note). While we describe loci by their proximal genes, in most cases we have no functional evidence for a specific target, and these variants may affect regulation of more distant genes.

Table 2

Index SNPs from meta-analysis for shared effects.

SNP	region	position	alleles	RAF	OR	95% CI	P	gene context
rs2101521	4p14	38811551	A/G	0.766	1.15	[1.11,1.18]	5.3×10−21	TLR1-[]--TLR6
rs1438673	5q22.1	110467499	T/C	0.498	1.12	[1.09,1.14]	2.3×10−20	WDR36-[]--CAMK4
rs2155219	11q13.5	76299194	G/T	0.511	1.11	[1.09,1.14]	1.6×10−19	C11orf30--[]--LRRC32
rs10189629	2q12.1	102879464	A/C	0.857	1.16	[1.12,1.20]	1.8×10−16	IL1RL2--[]--IL1RL1
rs6906021	6p21.32	32626311	T/C	0.475	1.10	[1.07,1.13]	7.1×10−15	HLA-DQA1--[]-HLA-DQB1
rs9266772	6p21.33	31352113	T/C	0.193	1.11	[1.08,1.14]	3.2×10−12	HLA-C[]--MICA
rs7720838	5p13.1	40486896	G/T	0.580	1.08	[1.06,1.11]	8.2×10−11	[]PTGER4
rs10497813	2q33.1	198914072	T/G	0.483	1.08	[1.05,1.10]	6.1×10−10	[PLCL1]
rs9860547	3q28	188128979	G/A	0.462	1.08	[1.05,1.10]	1.2×10−9	[LPP]
rs7032572	9p24.1	6172380	A/G	0.167	1.12	[1.08,1.16]	1.7×10−9	RANBP6[]--IL33
rs6021270	20q13.2	50141264	C/T	0.939	1.16	[1.11,1.23]	6.9×10−9	[NFATC2]
rs9303280	17q12	38074031	T/C	0.517	1.07	[1.05,1.10]	8.9×10−9	[GSDMB]
rs17228058	15q22.33	67450305	A/G	0.240	1.08	[1.05,1.11]	1.2×10−8	[SMAD3]
rs962993	10p14	9053132	T/C	0.576	1.07	[1.05,1.10]	1.5×10−8	GATA3[]
rs17388568	4q27	123329362	G/A	0.275	1.08	[1.05,1.10]	3.9×10−8	[ADAD1]
rs1998359	14q21.1	38077148	C/G	0.246	1.08	[1.05,1.12]	4.8×10−8	FOXA1--[]TTC6
rs6473223	8q21.13	81268155	C/T	0.358	1.07	[1.04,1.10]	7.7×10−8	TPD52[]ZBTB10
rs10174949	2p25.1	8442248	A/G	0.724	1.07	[1.05,1.10]	1.0×10−7	[]ID2
rs7203459	16p13.13	11230703	C/T	0.734	1.07	[1.04,1.10]	2.0×10−7	[CLEC16A]
rs2107357	16p12.1	27410829	G/A	0.138	1.09	[1.06,1.13]	3.3×10−7	IL4R--[]--IL21R
rs2056417	1p36.22	10581658	A/G	0.694	1.07	[1.04,1.10]	3.7×10−7	[PEX14]
rs10893845	11q24.3	128186882	T/G	0.493	1.06	[1.04,1.09]	6.4×10−7	[]ETS1

Region, cytogenetic band; position, build 37 map position of SNP; alleles, low/high risk alleles on genomic reference strand; RAF, risk allele frequency across all study participants; OR, meta-analysis odds ratio for the risk allele; CI, confidence interval; gene context, gene(s) spanning or flanking (<1mb) the index SNP; brackets indicate the position of the SNP and dashes indicate distance to a flanking gene (‘-‘, >1kb; ‘--‘, >10kb; ‘---‘, >100kb).

Figure 1

Manhattan plot of meta-analysis results for shared effects. The plotted values represent the most-significant scores from the meta-analyses of cat, pollen, and dust mite allergy, with all results with P<10 −4 recomputed using generalized estimating equations to assess effects shared across allergens. Results with P<5×10−8 are shown in red. Gene labels are provided for cross referencing with other results and are not intended to suggest that we have established a causal basis for the observed associations.

To ensure that the results were not confounded by age or differences between genotyping platforms, we tested the index SNPs for platform effects and for interactions with age within the 23andMe cohort, but no tests yielded strong evidence for interaction after adjusting for multiple comparisons (Supplementary Table 4). We also tested for pairwise interactions among the 22 loci, but none were significant after adjustment for multiple comparisons (Supplementary Table 5).

We assessed whether these associations were supported in a companion study of allergic sensitization[20] (Table 3, Supplementary Table 6). All 22 loci had effects in the same direction, and 10 of our 16 genome-wide-significant loci were supported with P<0.05 in the sensitization study. We also annotated our findings based on linkage disequilibrium (LD) with results from published GWAS, coding variation, monocyte expression quantitative trait loci (eQTL)[21], and putative regulatory regions identified by the ENCODE project (Table 3; Supplementary Tables 7, 8, 9, and 10).

Table 3

Summary of supporting evidence for allergy loci.

SNP	region	repla	atopyb	autoc	nsSNPd	eQTLe	gene contextf
rs2101521	4p14	***			*	TLR6	TLR1-[]--TLR6
rs1438673	5q22.1	***	**				WDR36-[]--CAMK4
rs2155219	11q13.5	***	***	**			C11orf30--[]--LRRC32
rs10189629	2q12.1	**	*				IL1RL2--[]--IL1RL1
rs6906021	6p21.32	***					HLA-DQA1--[]-HLA-DQB1
rs9266772	6p21.33	**					HLA-C[]--MICA
rs7720838	5p13.1	*		*			[]PTGER4
rs2117339	2q33.1			*		PLCL1	[PLCL1]
rs9860547	3q28	***		***		BCL6	[LPP]
rs7032572	9p24.1		**				RANBP6[]--IL33
rs6021270	20q13.2						[NFATC2]
rs9303280	17q12		***	***	*	IKZF3	[GSDMB]
rs17293632	15q22.33			*			[SMAD3]
rs962993	10p14	*					GATA3[]
rs17388568	4q27	***	*	***			[ADAD1]
rs9671863	14q21.1						FOXA1--[]TTC6
rs2202749	8q21.13	***					TPD52[]ZBTB10
rs13416555	2p25.1	***				ID2	[]ID2
rs7203459	16p13.13		*	*			[CLEC16A]
rs2107357	16p12.1						IL4R--[]--IL21R
rs2056417	1p36.22					PEX14	[PEX14]
rs970924	11q24.3						[]ETS1

Strength of replication for allergic sensitization[20]:

P<0.05,

P<0.005,

P<0.0005 (Supplementary Table 4).

Nearby (<500kb, r2>0.5) GWAS findings for atopy phenotypes (Supplementary Table 7).

Nearby (<500kb, r2>0.5) GWAS findings for autoimmune disease phenotypes (Supplementary Table 7).

nsSNP: * r2>0.5 with nonsynonymous SNP (Supplementary Table 8).

Association with expression of gene listed (Supplementary Table 9).

As defined in Table 2.

We examined evidence for association in our meta-analysis at other loci previously associated with either asthma or atopic dermatitis (Supplementary Tables 11 and 12). We have nominal support (P < 0.05, consistent risk allele) for 7 additional asthma loci (IL6R, GAB1, RAD50, IL13, IKZF4, RORA, and IL2RB), with a false discovery rate (FDR) of 0.04 across these variants. For atopic dermatitis, we have nominal support at 5 additional loci (IL13, KIF3A, CARD11, MIR1208, and NCF4), with an FDR of 0.07 for this group. These results indicate substantial overlap among these phenotypes beyond the loci meeting our criteria for significant and suggestive associations.

To test for allergy-specific genetic effects, we included interaction terms for specific allergies in our GEE models (Supplementary Table 1). We found one locus with strong evidence of allergy-specific effects, on chromosome 6 in the MHC region spanning the HLA-DRA, -DRB, -DQA1, and -DQB2 genes (Fig. 2, Supplementary Fig. 2). Index SNP rs17533090 had a combined P=1.7×10−12 for interaction with allergy type. Effects for the three allergies were consistent across cohorts, and indicated that this locus was specifically associated with cat allergy (Fig. 3). Among SNPs in Table 2, only rs2101521 showed evidence for an allergy-specific effect (unadjusted P=0.0011), which was weak compared to the evidence for a shared effect (P=5.3×10− 21).

Figure 2

Manhattan plot of meta-analysis results for interactions with allergen. Results with P<5×10−8 are shown in red. Interaction tests were performed for markers with P<1e-4 for association with at least one of cat, pollen, or dust mite allergy.

Figure 3

Marginal effect sizes and 95% confidence intervals for rs17533090 for cat, pollen, and dust mite allergy, in the 23andMe and ALSPAC cohorts. Effects are odds ratios for the high risk G allele of rs17533090.

We performed an exploratory analysis to assess associations of allergy loci with symptoms of allergic rhinitis, allergic contact dermatitis, and allergic asthma in the 23andMe cohort. We reclassified cases based on reported symptoms, and used the GEE approach to jointly model genetic effects across symptoms (Supplementary Tables 13 and 14). All effects were in the same direction at all index SNPs. At most loci, we did not see evidence for differential effects (P>0.05 for interaction). The exception was at GSDMB rs9303280, which was most strongly associated with asthma (P=0.000035 for interaction). Effect sizes for contact dermatitis symptoms tended to be smaller than for asthma (20/23, P=0.0004) or rhinitis (18/23, P=0.007). Effect sizes for asthma tended to be larger than for rhinitis, but not significantly so (14/23, P=0.30). However, association tests for rhinitis were more significant than for asthma at most loci (18/23, P=0.007), often by several orders of magnitude. Thus, while asthma may be a slightly more specific atopy phenotype, rhinitis appears to be more powerful for discovery of atopy loci in cohort studies, because it is more sensitive and captures more individuals who report allergies.

Genes implicated in our GWAS highlight key pathways in the etiology of common allergy. In the 4p14 region near rs2101521, TLR1 (Toll-Like Receptor 1) and TLR6 (Toll-Like Receptor 6) encode pattern-recognition receptors whose role in recognizing external pathogens and activating appropriate immune responses lies at the interface between innate immunity and immunoregulation. Candidate gene studies have identified associations between TLRs and asthma[22-24], and with grass sensitization and rhinitis[17]. However, this region has not been reported as significant in a genome-wide analysis.

We see substantial overlap between loci associated with allergy and loci previously linked to autoimmune disease. In the 5p13.1 region, index SNP rs7720838 is upstream of PTGER4, or Prostaglandin E receptor 4, previously implicated as a candidate asthma locus[25]. This SNP is close to a reported association with ankylosing spondylitis (rs10440635, r2=0.94)[26]. Variants affecting PTGER4 expression have also been associated with Crohn’s disease[27], and mouse studies point to a role in initiating skin immune responses[28,29]. In the 2q33.1 region, our eQTL analysis suggests that index SNP rs10497813 is associated with expression of PLCL1, or phospholipase C-like 1, involved in inositol 1,4,5-triphosphate intracellular signaling[30]. Variation in PLCL1 (rs6738825, r2=0.97) has also been associated with Crohn’s disease[31].

Several novel allergy loci are in or near genes involved in T helper cell differentiation. Index SNP rs9860547 in the 3q28 region falls in the LPP gene (lipoma-preferred partner). A nearby variant in LPP (rs1464510, r2=0.70) has been associated with celiac disease[32,33] and vitiligo[34]. Our eQTL analysis suggests that our association may be mediated by an effect on expression of BCL6 (B-cell lymphoma 6), a transcription factor that represses STAT6-mediated response to IL-4 and IL-13, and IgE class switching[35], and inhibits type 2 T helper (Th2) cell differentiation in a mouse model[36]. In the 20q13.2 region, index SNP rs6021270 is in the NFATC2 gene, encoding a component of the NFAT (nuclear factor of activated T cells) transcription complex, which plays an important role in regulating Th cell differentiation[37]. Variation in NFATC2 has not been linked to any allergic or autoimmune phenotype, however, mice lacking NFATc2 show increased lung inflammation in experimentally induced allergic asthma[38,39]. In the 4q27 region, index SNP rs17388568 falls in the ADAD1 gene but evidence for association spans the nearby IL2 and IL21 genes. This same SNP has been associated with type I diabetes autoantibodies[40] and ulcerative colitis[41], and a nearby SNP in strong LD (rs2069772, r2=0.91) has been suggestively associated with allergic rhinitis[17]. IL-2 and IL-21 cytokines are involved in regulation of multiple Th cell types; IL-21 is up-regulated in Th2 and Th17 cells and inhibits IL-2, while IL-2 is required for Th1 differentiation and inhibits differentiation of Th17 cells[42].

In the 14q21.1 region, index SNP rs1998359 is upstream of FOXA1, a member of the forkhead box transcription factor family. The closely related FOXA2 and FOXA3 transcription factors have roles in regulation of Th2 mediated inflammation and mucus production in allergic airway disease[43]; while a similar role of FOXA1 has not been established, FOXA1 and FOXA2 are known to have overlapping patterns of expression in respiratory epithelium[44].

In the 6p21.33 region, HLA-B and HLA-C are major histocompatibility complex (MHC) class I molecules expressed on most cell types, responsible for the display of intracellular peptides to T cells. MICA belongs to a family of non-classical MHC molecules that resemble the class I molecules and are thought to be involved in innate antitumor and antiviral surveillance[45]. Alleles of HLA-B are associated with severe allergic reactions such as abacavir hypersensitivity and Stevens-Johnsons syndrome[46,47]. SNPs in these three genes have been associated with a number of immune-system-related phenotypes such as psoriasis and HIV-1 control[48,49].

Previous studies have suggested associations between specific allergen sensitivities and HLA class II alleles[50]. However, these studies have been small and have reported inconsistent results[51]. Our finding of a specific association with cat allergy is the first demonstration of allergen specificity in a GWAS context.

We assessed directionality of effects in cases where our index SNPs are in strong LD (r2 > 0.5) with SNPs previously associated with autoimmune disease (Supplementary Table 15). At some loci (LRRC32, PTGER4, PLCL1, SMAD3, ADAD1, CLEC16A), autoimmune disease and allergy are associated with the same risk alleles. At others (GSDMB, LPP), the risk allele for autoimmune disease appears to be protective for allergy. Many autoimmune diseases are associated with increased activation of type 1 T helper (Th1) responses, while allergy has been associated with Th2 activity[52]. Our results may help to identify elements that influence the balance of Th1 versus Th2 activity, versus elements that contribute to both responses.

Self reported allergy status can be unreliable[53], and the surveys we used were not standardized or validated. In the 23andMe cohort, the high proportion of allergy cases likely reflects responder bias in completing the allergy survey. The ALSPAC cohort was assessed during pregnancy, which can alter allergic disease status[54]. These limitations should not compromise the validity of our genetic associations, but they make functional interpretation more challenging.

Our results demonstrate that self reported allergy can be used to identify disease susceptibility loci, with results consistent with studies of more narrowly defined allergy manifestations and allergic sensitization. Self-directed web-based data collection in the 23andMe cohort yielded results largely consistent with traditional survey methods used in the ALSPAC cohort. Our findings reinforce and extend evidence for a shared genetic etiology of allergic and autoimmune disease, with novel allergy susceptibility loci near LPP/BCL6, HLA-C/MICA, PTGER4, and PLCL1, all previously associated with autoimmune disease. Our findings also highlight the role of the Th2 cell lineage in pathogenesis of allergy, with associations in or near key Th2 genes including ID2, BCL6, GATA3, IL13, IL33, TSLP, and IL1RL1. An important next step will be to more carefully characterize the extent to which individual associations lead to a global predisposition to allergy, versus effects on specific targets such as skin, lung, or mucosa.

Methods

23andMe Cohort

Participants in the 23andMe cohort were customers of 23andMe, Inc., a personal genetics company, who had been genotyped as part of the 23andMe Personal Genome Service®. Individuals included in the analysis were selected for having >97% European ancestry, as determined through an analysis of local ancestry via comparison to the three HapMap 2 populations[55]. A maximal set of unrelated individuals was chosen for the analysis using a segmental identity-by-descent (IBD) estimation algorithm[56]. Individuals were defined as related if they shared more than 700 cM IBD, including regions where the two individuals share either one or both genomic segments identical-by-descent. This level of relatedness (roughly 20% of the genome) corresponds approximately to the minimal expected sharing between first cousins in an outbred population. The study protocol and consent form were approved by the external AAHRPP-accredited Institutional Review Board, Ethical & Independent Review Services (E&I Review). For a small number of participants (167) under the age of 18, consent was provided by a parent, guardian, or legally authorized adult.

DNA extraction and genotyping were performed on saliva samples by National Genetics Institute (NGI), a CLIA-certified clinical laboratory and subsidiary of Laboratory Corporation of America. Samples were genotyped on one of two platforms. About 35% of the participants were genotyped on the Illumina HumanHap550+ BeadChip platform, which included SNPs from the standard HumanHap550 panel augmented with a custom set of approximately 25,000 SNPs selected by 23andMe. Two slightly different versions of this platform were used, as previously described[18]. The remaining 65% of participants were genotyped on the Illumina HumanOmniExpress+ Bead Chip. This platform has a base set of 730,000 SNPs. This was augmented with approximately 250,000 SNPs to obtain a superset of the HumanHap550+ content, as well as a custom set of about 30,000 SNPs. Every sample that failed to reach a 98.5% call rate for SNPs on the standard platforms was re-analyzed. Individuals whose analyses failed repeatedly were re-contacted by 23andMe customer service to provide additional samples, as is done for all 23andMe customers.

Participant genotype data were imputed against the August 2010 release of 1000 Genomes reference haplotypes[57]. First, we used Beagle[58] (version 3.3.1) to phase batches of 8000–9000 individuals across chromosomal segments of no more than 10,000 genotyped SNPs, with overlaps of 200 SNPs. We excluded SNPs with minor allele frequency < 0.001, Hardy-Weinberg equilibrium P<10−20, call rate < 95%, or with large allele frequency discrepancies compared to the 1000 Genomes reference data. We then assembled full phased chromosomes by matching the phase of haplotypes across the overlapping segments. We imputed each batch against the European subset of 1000 Genomes haplotypes using Minimac[59] (2011-10-27), using 5 rounds and 200 states for parameter estimation. Analyses were limited to 7.4 million SNPs with imputed r2 > 0.5 averaged across all batches, and r2 > 0.3 in every batch.

23andMe participants were able to fill out web-based questionnaires whenever they logged into their 23andMe accounts. Allergy information was derived primarily from an “Allergies and Asthma” survey (Supplementary Note). The survey covers allergic reactions to 38 common allergens, including foods, plants, animals, molds, latex, dust mites, medicines, and vaccines. Cases were defined as those who reported a positive allergy test, difficulty swallowing or speaking, hives, itchy mouth, itchy eyes, itchy nose, or asthma in response to that allergen. Controls were defined as individuals who did not meet these criteria. For pollen allergy, we aggregated reports of allergies to grasses, trees, or weeds. At the time of the analysis, 30% of 171,274 23andMe research participants had taken this allergy survey. We incorporated 7635 additional controls who reported having neither seasonal nor environmental allergies in a medical history survey, or who reported not currently having allergies in an asthma survey. The final analysis included 46,646 participants.

ALSPAC Cohort

The Avon Longitudinal Study of Parents and Children[19] is a large birth cohort that has recruited 14,541 pregnant women resident in Avon, UK with expected dates of delivery 1st April 1991 to 31st December 1992. Mothers enrolled in the study, filled out a questionnaire at the end of the third month of pregnancy, which included questions on allergies. Mothers were asked if they were allergic to cat, pollen or dust with the option each time of indicating yes or no. The questions did not specify current or past allergy.

Centre National de Génotypage (CNG) carried out DNA genotyping on the Illumina human660W-quad array and genotypes were called with Illumina GenomeStudio. PLINK[60] (v1.07) was used to carry out quality control measures on an initial set of 10,015 subjects and 557,124 directly genotyped SNPs. SNPs were removed if they displayed more than 5% missingness or a Hardy-Weinberg equilibrium P<10−6. Additionally SNPs with a minor allele frequency of less than 1% were removed. Samples were excluded if they displayed more than 5% missingness, had indeterminate X chromosome heterozygosity or extreme autosomal heterozygosity. We restricted the analysis to individuals with European ancestry; samples showing evidence of population stratification were identified by multidimensional scaling of genome-wide identity-by-state pairwise distances using the four HapMap populations as a reference, and then excluded. Cryptic relatedness was assessed using a Pi hat of more than 0.125 which is expected to correspond to roughly 12.5% alleles shared IBD or a relatedness at the first cousin level.

A total of 8,340 subjects and 526,688 SNPs passed these quality control filters. We imputed autosomal SNPs against the HapMap[61] CEU population (release 22) using MaCH[62] (v1.0.16, Li 2010). A combination of MaCH and Minimac[59] (v4.4.3, 2010-12-13) was used to impute X chromosome genotypes against the HapMap CEU population (release 21). Analyses were limited to 2.5 million SNPs with imputed r2 > 0.3. Out of 8,340 subjects with genotype data, 7,216 had allergy phenotype data and were used in the GWAS.

Ethical approval for the study was obtained from the ALSPAC Law and Ethics Committee and the Local Research Ethics Committees.

Single Phenotype GWAS

We performed traditional genome-wide tests for association with each of the three allergy phenotypes, using logistic regression, assuming an additive model for genetic effects. The 23andMe analyses were adjusted for age, gender, and the top 5 principal components of the genotype data matrix. The ALSPAC analyses were not adjusted for covariates. ALSPAC GWAS results were remapped to NCBI Build 37 using the liftOver tool[63]. For each allergy phenotype, we used METAL[64] to perform an inverse-variance-weighted fixed-effects meta-analysis across 2.4 million SNPs in the intersection of the 23andMe and ALSPAC results. We applied genomic control corrections to the individual GWAS result sets (23andMe: λ=1.06 to 1.08; ALSPAC: λ=1.00 to 1.02). The meta-analysis results showed no inflation (λ=1.001 to 1.004).

Multiple Phenotype GWAS

We jointly modeled association across the three allergens using generalized estimating equations (GEE)[65]. We used an unstructured correlation matrix for the three outcomes. In each cohort, we first fit GEE models with the same covariates used in the single phenotype GWAS, with additional terms for interactions between each covariate and allergen, and a single shared genotype effect, using a fast approximate method[66]. Results were adjusted for genomic control (23andMe: λ=1.07; ALSPAC: λ=1.02). We performed an inverse-variance-weighted fixed-effects meta-analysis of the shared effects across the 23andMe and ALSPAC cohorts. Then, for a subset of 3725 SNPs with either a single-phenotype P<10−4 with any allergy, or P<10−4 in the approximate GEE meta-analysis, we refit GEE models using the R package geepack[67]. In addition to refitting the shared effects, we incorporated interactions between genotype and allergy type, and used analysis of deviance to assess significance of the interactions. We used Fisher’s method[68] to compute combined P values from the 23andMe and ALSPAC interaction tests. This test combines evidence for an interaction in each cohort but does not assess directional consistency of the interactions.

Heterogeneity Assessment

In the 23andMe cohort, we assessed genotyping platform effects by logistic regression of platform against 5 principal components and the imputed allele dosage, and performed a likelihood ratio test to assess significance of the allele dosage term. We assessed the index SNPs for age effects in the 23andMe cohort by fitting a GEE model with an age by dosage interaction, and testing significance with a Wald test on the interaction term. We tested index SNPs for heterogeneity across cohorts using Cochran’s Q statistic, and used I2 to measure the extent of heterogeneity[69]. We determined confidence intervals for I2 using the non-central χ2 method (Supplementary Table 4). While several SNPs have large I2, confidence intervals are very wide and remain consistent with the null hypothesis of no heterogeneity.

Assessment of SNP Interactions

In the 23andMe cohort, we fit GEE models assuming shared effects across allergy types with allele dosages and interactions for all pairwise combinations of the 22 shared-effect index SNPs and rs17533090 (Supplementary Table 5). We used Wald tests to assess significance of the interaction terms. Given a conditioning SNP1 and tested SNP2, we also computed a joint test of both the main effect of SNP2 and the interaction SNP1×SNP2 being equal to zero.

Functional Annotation

We used publicly available bioinformatic resources to annotate putative associations. Generally, we required that an annotated variant be within 500 kb of our index SNP, with r2>0.5 based on the European subset of 1000 Genomes haplotypes. We used the NCBI Gap Plus resource to identify nearby GWAS findings (Supplementary Table 7). We used tables from the UCSC Genome Browser to identify nearby coding SNPs (Supplementary Table 8). We identified nearby expression quantitative trait loci (eQTL) from a study of monocytes[21] (Supplementary Table 9). We also used HaploReg[70] to identify nearby annotations from the ENCODE project[71] (Supplementary Table 10). Finally, we took all reported associations with asthma or atopic dermatitis from the NHGRI GWAS Catalog[72], and looked up our corresponding meta-analysis results (Supplementary Tables 11 and 12).

Assessment of SNP Effects on Allergy Symptoms

In the 23andMe cohort, we reclassified cases based on self report of allergic symptoms representing allergic rhinitis, allergic asthma, and/or allergic contact dermatitis (Supplementary Methods). We performed a GEE analysis across these multiple outcomes, including the same controls used in the GWAS (Supplementary Tables 13 and 14). The model included the same covariates used in the GWAS (age, gender, 5 principal components), as well as interactions between these covariates and symptom type.