Shared and Distinct Genetic Risk Factors for Childhood Onset and Adult Onset Asthma

Childhood and adult onset asthma differ with respect to severity and co-morbidities. Whether they also differ with respect to genetic risk factors has not been previously investigated. Methods We used data from the UK Biobank to conduct genome-wide association studies (GWASs) in 9,433 childhood onset asthma (onset before age 12) and 21,564 adult onset asthma (onset between ages 26 and 65) cases, each compared to 318,237 non-asthmatic controls (older than age 38), and for age of onset in 37,846 asthma cases. Enrichment studies determined the tissues in which genes at GWAS loci were most highly expressed, and PrediXcan, a transcriptome-wide gene-based test, was used to identify candidate risk genes. Findings We detected 61 independent asthma loci: 23 were childhood onset specific, one was adult onset specific, and 37 were shared. Nineteen loci were associated with age of asthma onset. Genes at the childhood onset loci were most highly expressed in skin, blood and small intestine; genes at the adult onset loci were most highly expressed in lung, blood, small intestine and spleen. PrediXcan identified 113 unique candidate genes at 22 of the 61 GWAS loci. Interpretation Genetic risk factors for adult onset asthma are largely a subset of the genetic risk for childhood onset asthma but with overall smaller effects, suggesting a greater role for non-genetic risk factors in adult onset asthma. In contrast, the onset of disease in childhood is associated with additional genes with relatively large effect sizes. Combined with gene expression and tissue enrichment patterns, we suggest that the establishment of disease in children is driven more by allergy and epithelial barrier dysfunction whereas the etiology of adult onset asthma is more lung-centered, with immune mediated pathways driving disease progression in both children and adults. Funding This work was supported by the National Institutes of Health grants R01 MH107666 and P30 DK20595 to H.K.I., R01 HL129735, R01 HL122712, P01 HL070831, and UG3 OD023282 to C.O.; N.S. was supported by T32 HL007605.

2 Background 17 Childhood and adult onset asthma differ with respect to severity and co-morbidities. 18 Whether they also differ with respect to genetic risk factors has not been previously 19 investigated in large samples. The goals of this study were to identify shared and 20 distinct genetic risk loci for childhood and adult onset asthma, and the genes that may 21 mediate the effects of associated variation. 22

Methods 23
We used data from UK Biobank to conduct genome-wide association studies (GWASs) 24 in 37,846 subjects with asthma, including 9,433 childhood onset cases (onset before 25 age 12) and 21,564 adult onset cases (onset between ages 26 and 65), and 318,237 26 subjects without asthma (controls; older than age 38). We conducted GWASs for 27 childhood onset asthma and adult onset asthma each compared to shared controls, and 28 for age of asthma onset in all 37,846 asthma cases. Enrichment studies determined the 29 tissues in which genes at GWAS loci were most highly expressed, and PrediXcan, a 30 transcriptome-wide gene-based test, was used to identify candidate risk genes. 31

Findings 32
Introduction self-reported doctor diagnosed asthma. Of those, 15,519 also had hospital records with 123 an asthma diagnosis (ICD10 codes). 124 We defined childhood and adult onset asthma using strict age of onset criteria 125 that would minimize the likelihood of misclassification, and considered asthma cases 126 with onset < 12 years of age as childhood onset cases (n=9,433; 3,462/9,433 [36·7%] 127 with ICD10 codes) and with onset > 25 and before 66 years of age as adult onset cases 128 (n=21,564; 9,260/21,564 [42·9%] with ICD10 codes). UKB participants without an 129 asthma diagnosis were included as controls (n=318,237). Individuals with COPD, 130 emphysema or chronic bronchitis were excluded from adult onset cases and controls. 131 For the age of onset GWAS, we included an additional 6,849 subjects with asthma with 132 onset between the ages of 12 and 25 years of age (2,797/6,849 [40·8%] with ICD10 133 codes). Characteristics of the sample are shown in Table 1. After genotype quality 134 control, 10,894,596 variants were available for analyses. See appendix (pp 4-7) for 135 additional details on genotype QC and phenotype definitions. 136 137 Genome-wide association studies 138 We conducted a childhood onset and adult onset GWAS by logistic regression and an 139 age of onset GWAS by linear regression, both using the allele dosages under an 140 enriched tissues. FUMA defines independent loci using linkage disequilibrium (LD) 146 information from the 1000 Genomes project 14 . We specified an r 2 threshold >0·6 147 between genome-wide significant SNPs to represent a single locus. Tissue enrichments 148 were calculated in FUMA using a hypergeometric test to determine overrepresentation 149 of genes mapped to risk loci (by physical distance) among highly expressed genes in 150 each tissue in GTEx relative to all others. 151 Childhood onset and adult onset specific loci were defined as those that were 152 genome-wide significant in either the childhood onset or adult onset GWAS but were not 153 associated with asthma at p<0·05 in the other group, and the 95% confidence intervals 154 (CIs) of the respective ORs did not overlap. All other GWAS loci that were genome-wide 155 signficant in at least one of the two GWASs were considered to be shared. 156 misclassification of COPD as asthma among the adult onset cases, even with exclusion 169 of cases with a reported diagnosis of COPD, emphysma or chronic bronchitis. First, to 170 assure that the adult onset cases did not include a significant proportion of childhood 171 onset asthma in which symptoms remitted in early life but then relapsed in adulthood, 172 we replaced adult onset cases with increasing proportions of randomly selected 173 childhood onset cases, and then tested for association at the two most significant 174 childhood onset specific loci. This procedure was repeated 20 times for each proportion 175 to quantify the sampling variability (appendix pp 7-8). Second, we performed two 176 analyses in which we removed either subjects with ages of asthma onset between 46 to 177 65 years or adult-onset cases and controls with FEV1/FVC <0·70. For each, we 178 compared p-values and ORs to the GWAS including all adult onset cases (appendix pp 179 8-9). 180 181 Predicted transcriptome association test 182 We used the PrediXcan 17 framework to identify genes that may mediate associations 183 between genetic variants and asthma risk. PrediXcan is a software tool that estimates 184 tissue-specific gene expression profiles from an individual's SNP genotype profile using 185 prediction models trained in large references databases of genotypes and tissue-186 specific gene expression profiles. Using these genotype-imputed expression profiles, 187 PrediXcan can perform gene-based association tests that correlate predicted 188 expression levels with phenotypes (e.g., asthma) to identify candidate causal genes 189 from GWAS data. We used a summary version of PrediXcan, which has high 190 concordance with the individual level version (R 2 >0·99) 18 . For predictions,we 191 downloaded elastic net models trained with reference transcriptome data from the GTEx 192 consortium 19 (http://predictdb.org) for 49 tissues (appendix Table 1) . 193 PrediXcan was run separately in the childhood onset and adult onset cases, 194 each with the same controls. Significance was determined using a Bonferroni correction 195 for the 38,608 genes (p<1·29x10 -6 ) that were expressed in the five tissues determined 196 by FUMA to be enriched (skin, lung tissue, whole blood, small intestine and spleen; see 197 Results). We defined childhood onset specific genes as those whose predicted 198 expression is significantly associated with childhood onset asthma and the variants that 199 predict their expression are within childhood onset specific loci. Adult onset specific 200 genes were similarly defined. Because SNPs at shared loci may predict the expression 201 of genes that are associated only in childhood onset or adult onset cases, we also 202 considered genes to be age of onset specific if they were significantly associated with 203 asthma at p<1·29x10 -6 in one age group and not associated with asthma at p<0·05 in 204 the other. All other genes were considered shared. 205

206
Role of funding source 207 The funding source did not have any role in the study design; collection, analysis or 208 interpretation of data; in writing the manuscript; in the decision to submit the paper for 209 publication; or determining who has access to the raw data. The corresponding author 210 had full access to all the data in the study and had final responsibility for the decision to 211 submit for publication. 212 213

214
Genome-wide association studies of asthma 215 We first conducted GWASs of childhood onset and adult onset asthma. These studies 216 revealed 61 independent loci associated with asthma, 52 were significant in the 217 childhood onset asthma GWAS and 19 were significant in the adult onset asthma 218 GWAS (p<5x10 -8 ) (Figure 1). The GWAS results were robust to inclusion of varying 219 numbers of PCs (10, 14, or 20) (appendix Figure 1) and to limiting the sample to cases 220 with diagnoses based on ICD10 codes (appendix Figure 2). Twenty-eight of the 61 loci, 221 in 27 chromosomal regions, were not previously reported in the GWAS catalog 20 , 222 including one study also conducted in UKB subjects but focused on the phenotype 223 asthma+allergies 21 . Among the 28 new loci, 17 were significant in the childhood onset 224 GWAS, one was significant in the adult onset GWAS, and 10 were significant in both 225 (Table 2). Some of these loci contained genes that have been associated with asthma 226 in candidate gene studies (e.g., FADS2 22 , MUC5AC 23,24 and TBX21 25 ), which provide 227 independent validation of our GWAS findings. The lead SNP or an LD surrogate SNP 228 (r 2 >0·40) was reported for 23 of the 28 new loci in the TAGC GWAS 26 . SNPs at 16 of 229 the 23 loci were associated with asthma in TAGC (p<0·05) and had the same direction 230 of effects as in the UKB GWASs (appendix Table 2). To test for potential effects of 231 asthma diagnostic criteria on the GWAS results, we conducted a GWAS comparing 232 individuals with asthma based on self-reported doctor diagnosis alone compared to 233 those with self-reported doctor diagnosis plus ICD10 codes. There were no differences 234 detected between these two groups (appendix Figure 3), indicating that differences 235 between these groups are not influencing the GWAS results discussed above. 236 As in previous GWASs comprised largely of children, the most significant locus in 237 the childhood onset GWAS is at 17q12 27 , with the lead SNP in the GSDMB gene 238 ( Figure 1A). However, the estimated ORs for the lead SNPs at three other loci were 239 similar to or larger than the lead SNP at the 17q locus. The lead SNPs in IL1RL1 at 240 2q12.1 and in EMSY at 11q13.5 had effect sizes on childhood onset asthma similar to 241 the lead SNP in GSDMB on 17q ( To address this possibility, we repeated the childhood onset GWAS after 249 excluding 3,205 childhood onset cases and 5,785 controls who reported having a 250 history of allergic rhinitis, AD or food allergy. As expected in a smaller sample the p-251 values were overall larger, but the ORs were strikingly similar (appendix Figure 4 and 252 Table 3). Even though the OR at the FLG locus on 1q21.3 decreased from 1·97 (95% 253 CI 1·82, 2.13) to 1·61 (95% CI 1·49, 1·74), it remained both highly significant 254 (p=2·45x10 -19 ) and the largest OR for childhood onset. These results suggest both a 255 critical role for the allergic diathesis in the development of asthma in childhood and a 256 shared architecture between allergic disease and childhood onset asthma, as previously 257 discussed 34,35 . 258 The most significant association in the adult onset GWAS was in the HLA region, 259 with independent associations at the HLA-C/B (6p21·33) and HLA-DR/DQ (6p21·32) 260 loci ( Figure 1B). Compared to the childhood onset GWAS, effect sizes were quite small 261 in adult onset cases, with ORs reaching 1·1 at only five loci (2q12.1, 6p21.33, 6p21.32, 262 9p24.1, 10p14) ( Table 2). 263 Among the 61 asthma loci, 23 were specific to childhood onset asthma and one 264 was specific to adult onset asthma (Table 2A-B). Regional association plots for the 23 265 loci with childhood or adult onset specific effects are shown in appendix Figure 5. 266 Among the remaining 38 shared loci (Table 2C), mean ORs were larger in the childhood 267 onset cases at all but six loci (permutation test p<10 -4 ; appendix pp 8-9), indicating that 268 both more loci contribute to childhood onset asthma and even among shared loci, effect 269 sizes are larger in childhood onset asthma cases ( Figure 2). Colocalization analyses 270 using GWAS-PW 36 yielded results that supported our classification of age of onset 271 specific and shared loci (appendix Table 4). 272 Finally, to directly test for loci associated with asthma age of onset, we 273 conducted a third GWAS including all asthma cases in UKB who met our inclusion 274 criteria (n=37,846). In this analysis, 19 loci were associated with age of onset (p<5x10 -8 ) 275 (appendix Figure 6 and Table 5). Age of onset loci overlapped with both the childhood 276 and adult onset specific and shared loci, and asthma risk alleles at all but 2 loci were 277 associated with earlier age of onset (11q12 in the FADS2 gene and 12q13.11 near the 278 VDR gene) (Table 2; Figure 2). SNPs at the 1q21.3 locus (FLG) had the largest effect 279 on age of onset, with each copy of the asthma risk allele (rs61816761) associated on 280 average with 4·57 (SE 0·43) years earlier onset compared to individuals without the risk 281 allele (p=8·15x10 -27 ). At the 17q12 locus (rs4795399) each copy of the risk allele was 282 associated on average with 2·29 (SE 0·13) years earlier onset compared to individuals 283 without the risk allele (p=6·76x10 -65 ). Examples of significant age of onset effects at 284 these and other loci are shown in Figure 3. Overall, both childhood onset specific and 285 shared asthma risk loci were associated with younger ages of onset, and alleles at loci 286 associated with younger ages of onset had larger effects compared to alleles at loci 287 associated with later ages of onset. These results are consistent with a previous GWAS 288 of age of asthma onset in European ancestry subjects that reported five genome-wide 289 significant and three suggestive significant loci, all associated with earlier age of 290 onset 37 . Six of those 8 loci are also associated with age of onset in the UKB GWAS 291 (appendix Table 6). 292 293 SNP-based heritability (h 2 ) 294 We used LD score regression to estimate the hertiabilities of childhood onset asthma, 295 adult onset asthma, and age of asthma onset. Consistent with the number of associated 296 SNPs and their effect sizes, estimated heritabilities were 0·33 for childhood onset 297 asthma, 0·098 for adult onset asthma, and 0·14 for age of asthma onset. After 298 excluding the significant SNPs (Table 2), estimates were reduced to 0·21, 0·082, and 299 0·087, respectively, indicating that the associated SNPs account for 0·11 of the 300 variance in childhood onset asthma risk, 0·016 of the variance in adult onset risk, and 301 0·049 of the variance in age of asthma onset risk. These results reflect the more 302 significant role for genetic variation in risk for childhood onset compared to adult onset 303 asthma and, conversely, the larger role for environmental variation in risk for adult onset 304 compared to childhood onset asthma. 305 306 Tissue-specific expression of genes at associated loci 307 Using an unbiased approach, we asked whether the tissue-specific expression of genes 308 that map to the 52 childhood onset loci differed from the tissue-specific expression of 309 genes mapped to the 19 adult onset loci. Genes at childhood onset loci ( Figure 1A) 310 were most highly expressed in skin, whole blood, and small intestine (lower ileum) 311 compared to all other tissues, whereas genes at adult onset loci ( Figure 1B) were most 312 highly expression in lung, whole blood, small intestine (lower ileum), and spleen 313 (enrichment for higher expression, p<10x10 -3 ) (appendix Figure 7 and Table 7). These 314 patterns suggest both overlapping and distinct underlying mechanisms associated with 315 asthma that begins in childhood and asthma with onset in adulthood. 316 317 Predicted transcriptome-wide association test 318 To better understand molecular mechanisms and to narrow the list of candidate causal 319 genes at associated loci, we focused on the five tissues that most highly expressed the 320 genes at childhood onset or adult onset loci: skin, lung tissue, whole blood, small 321 intestine, and spleen. We used PrediXcan 17 to identify genes whose expression is 322 predicted by variants associated with asthma in the childhood onset or adult onset 323 GWAS and potentially mediate the effects of associated SNPs on asthma risk. 324 This analysis identified 113 unique, candidate causal genes at 22 of the 61 325 included 39 genes associated with childhood onset asthma at eight of the childhood 327 onset specific loci and 76 genes associated with childhood and/or adult onset asthma at 328 13 of the shared loci. Variants at the one adult onset specific locus at 2q22·3 did not 329 predict the expression of any genes in the five tissues. 330 The predicted genes most significantly associated with childhood onset asthma 331 were at the 17q12 locus (Z score >10) in skin (ORMDL3, ERBB2, PGAP3, GSDMA, 2 332 long noncoding RNAs), lung (ORMDL3, GSDMB, GSDMA, PGAP3, PNMT), blood 333 (ORMDL3, GSDMB, IKZF3, MED24), small intestine (GSDMA, GSDMB, PGAP3), and 334 spleen (ORMDL3, GSDMB, ZPBP2, MED24). Some genes were predicted to be more 335 highly expressed in individuals with asthma (e.g., ORMDL3, GSDMB, PGAP3, ERBB2), 336 while others were predicted to be less expressed in individuals with asthma (e.g., 337 GSDMA, MED24, IKZF3) ( Figure 4A).This pattern of expression reflects the broad 338 regulatory effects of SNPs and tissue specificifity of gene expression at this locus 27 . The 339 childhood onset asthma locus at 1q21.3 includes genes essential for epidermal 340 differentiation and maintaining essential barrier function. The predicted expression of 341 nine genes at this locus were associated with childhood onset asthma. Higher predicted 342 expression of CRNN, CRCT1 and THEM5 in skin, of PSDM4 in lung and blood, and of 343 LINGO4 in skin, lung, and blood were associated with increased asthma risk. Lower 344 predicted expression of SPRR2D in skin, of S100A12 in lung and blood, of FLG in skin, 345 lung and spleen, and of TDRKH in skin, lung, blood and spleen were associated with 346 increased asthma risk. S100A12 has been previously implicated in asthma 38 and FLG 347 variants have been associated with atopic dermatitis and food allergies, and asthma in 348 the context of other allergic diseases 21,28-34 . Other childhood onset specific genes previously implicated in asthma but not previously reported in asthma GWASs are 350 CCL20 39 at 2q36.3 and TLR10 40 at 4p14 in whole blood, and TLR6 41,42 at 4p14, AP5B1 351 at 11q13.1 and SERPINB7 43 at 18q21·33 in skin. 352 The 5q31.1 region had independent loci that were both childhood onset specific 353 and shared in the GWASs. Although the predicted expression of all eight asthma genes 354 at this extended locus were shared, five genes were more significantly associated with 355 childhood onset asthma: higher predicted expression of RAD50 in skin but lower 356 predicted expression of SEPT8 in skin and lung, IL4 in skin, lung and blood, and AFF4 357 small intestine were associated with increased risk for asthma ( Figure 4B). The 358 remaining three genes had similar associations with childhood and adult onset asthma, 359 with predicted lower expression of IRF1 in skin and spleen and predicted higher 360 expression of PDLIM4 in skin and SLC22A5 in all five tissues associated with increased 361 asthma risk. IL4, RAD50, SLC22A5 and PDLIM4 have been highlighted in previous 362 asthma GWAS 26,44 , and KIF3A was identified in a GWAS of the atopic march 45 and 363 associated with childhood onset asthma in a candidate gene study 46 . 364 In contrast to all other loci, predicted expression of 44 genes at two independent 365 shared loci in the HLA region (6p21.32 and 6p21.33; referred to as the HLA region from 366 hereon in) were associated with childhood asthma only (n=1; in skin and lung), adult 367 onset asthma only (n=3; in skin only), or both (n=39; in multiple tissues). ( Figure 5). The 368 sheer number of genes in this region with predicted expression associated with asthma, 369 the generally broad tissue expression patterns, and three associated with asthma only 370 in adult onset cases are consistent with this locus being among the two most signficant 371 loci in nearly all asthma GWASs, and the most significant locus in a previous small 372 GWASs of adult onset asthma 7 and in adults with asthma 44,47 . 373 Among the remaining 37 shared loci ( Figure 2B), SNPs at 12 predicted the 374 expression of 23 unique genes, all of which were associated with both childhood onset 375 and adult onset asthma. These include IL18R1, IL18RAP, and ILRL2 at 2q12·1 in 376 multiple tissues, TSLP at 5q22.1 in skin, SMAD3 at 15q22·33 in skin and blood, LRP1 377 at 12q13·3 in skin, IL4R at 16p12.1 in blood, and CLEC16A at 16p13.13 in lung. We report here the first large GWAS of both childhood and adult onset cases. To both 383 maximize our power to detect differences and minimize the likelihood of 384 misclassification, we considered doctor diagnosed asthma before the age of 12 years 385 as childhood onset asthma and doctor diagnosed asthma after the age of 25 years as 386 adult onset asthma. These GWASs revealed 61 independent asthma loci, 23 specific to 387 childhood onset, one specific to adult onset, and 37 shared; with overall larger effect 388 sizes for childhood onset asthma at nearly all loci. Moreover, the predicted expression 389 of 41 of the 113 implicated genes were associated specifically with childhood onset 390 asthma, compared to the predicted expression of three genes associated specifically 391 with adult onset asthma. Our findings of more childhood onset asthma loci and 392 potentially causal genes, and the larger effect sizes of risk alleles in childhood onset 393 cases are particularly striking given that there were nearly 2·5-times more adult onset 394 than childhood onset cases in this study. Thus, despite having substantially less power 395 to detect loci specific to childhood onset asthma, our analyses revealed many more 396 childhood onset asthma loci. Similarly, the asthma risk alleles at 19 loci that were 397 significant in the age of onset GWAS were all associated with younger age of onset. 398 Finally, we showed that the SNP-based heritability of childhood onset asthma is over 3-399 times larger than the SNP-based heritability of adult onset asthma. These findings are 400 consistent with previous studies showing decreased estimates of asthma heritability 401 with increasing age of onset 48 , age of onset SNPs associated with earlier age of 402 onset 37 , and an additive, unweighted genetic risk score comprised of 15 SNPs at eight 403 asthma-associated loci associated with earlier age of onset 49 . Our study further shows 404 that genetic risk for adult onset asthma is largely a subset of the genetic risk loci for 405 childhood onset asthma, but with overall smaller effect sizes, consistent with a larger 406 role for environmental risk factors in adult onset asthma. 407 Despite the overlap of adult onset and childhood onset loci, distinct mechanisms 408 contributing to each were suggested by tissue enrichments: childhood onset loci were 409 enriched for genes with highest expression in skin whereas adult onset loci were 410 enriched for genes with highest expression in lung and spleen; both were enriched for 411 genes highly expressed in whole blood and small intestine. The highlighting of skin as a 412 target tissue for childhood onset asthma supports the widely held idea that asthma in 413 childhood is due to impaired barrier function in the skin and other epithelial surfaces. 414 This model proposes that compromised epithelial barriers promote sensitization to food 415 and airway allergens and to wheezing illnesses in early life 35,50 . In fact, childhood onset 416 specific loci identified here have been associated with atopic dermatitis or food allergies, 417 such as FLG on 1q21·3 with the atopic march 45 , atopic dermatitis 31-33 and food 418 allergies 28-30 , KIF3A on 5q31.1 and AP5B1/OVOL1 on 11q13.1 with the the atopic 419 march 45 and atopic dermatitis 51 , SERPINB7 on 18q21.33 with food allergies 43 , and 420 CRNN (cornulin) on 1q21·3 with atopic dermatitis concomitant with asthma and reduced 421 expression in atopic dermatitis-affected skin 52 . Variants at those loci were all associated 422 with earlier age of asthma onset. We further show that these loci are associated with 423 childhood onset asthma, even after exclusion of cases with a history of allergic 424 diseases. In contrast, the enrichment for genes highly expressed in lung and spleen at 425 adult onset loci suggests a more lung-centered, and potentially immune mediated, 426 etiology for asthma with onset later in life. The prominant role of the HLA region in the 427 adult onset asthma GWAS and the fact that predicted expression of three HLA region 428 genes was associated only with adult onset asthma further highlights a central role for 429 immune processes driving asthma pathogenesis in adults. The fact that both childhood 430 onset and adult onset asthma loci were enriched for genes that are most highly 431 expressed in whole blood cells and small intestine further indicate a shared immune 432 etiology, as suggested from a large GWAS that included both children and adults 26 . 433 Combining GWAS with a transcriptome-wide association test that uses 434 combinations of associated SNPs to predict gene expression in different tissues 435 revealed significant complexity at the two most highly associated asthma loci. SNPs at 436 the 17q12 locus predicted expression of 18 childhood onset asthma genes and SNPs at 437 the HLA region predicted expression of 42 genes: three were associated with adult 438 onset asthma and most were not HLA genes per se. In this regard, it is notable that the loci with autoimmune diseases, are predicted to have reduced expression in both 441 childhood onset and adult onset asthma. Instead, HLA genes with less clear functions 442 have increased predicted expression in asthma ( Figure 5). These results strengthen the 443 argument that multiple genes contribute to asthma risk at the HLA and 17q12 loci and 444 probably account for the highly significant GWAS p-values observed at these loci in 445 nearly all studies. It is also likely that these genes have both tissue specific and broad 446 effects in epithelium, lung, and immune tissues. 447 The new loci identified in our study include the first adult onset asthma specific 448 association at 2q22.3. The lead SNP at 2q22.3 is intergenic between TEX41 and 449 ACVR2A. The predicted expression of ACVR2A was not associated with asthma in our 450 study, despite it being expressed in lung, blood, small intestine and spleen. TEX41 was 451 not expressed in any of the five tissues investigated. Interestingly, a GWAS also 452 performed in UKB subjects implicated variants near TEX41 in heavy vs. never smoking 453 behavior 53 . However, even after removing adult onset cases and controls with reported 454 'ever smoking', the p-value for this SNP remained significant and the OR slightly 455 increased (OR 1·077 [95% CI 1·05, 1.1], p=2·26x10 -8 ; n=12,132 cases and 176,704 456 controls). Variants in or near this gene, which encodes a lincRNA, have been 457 associated with cardiovascular and immune mediated traits 54 , making this a potentially 458 interesting candidate gene for adult onset asthma. 459 Our study had limitations. First, diagnoses of asthma and allergic disease in 460 study subjects were from self reported doctor diagnosis and medical records (ICD10 461 codes). Thus, it is possible that diagnoses, age of onset, or both are misspecified in 462 some subjects. On the one hand, the large sample size and our ability to replicate nearly all previously reported asthma loci (appendix Tables 2 and 9) suggest that our 464 analyses were robust to any inaccuracies in the data. On the other hand, it is possible 465 that subjects with adult onset asthma included cases with poor recall of childhood onset 466 asthma in which symptoms remitted and then relapsed later in life 55 or misclassified 467 cases of COPD among the older age groups. Our sensitivity analysis suggested that if 468 even as few as 5% of the adult onset cases were misclassified we should have 469 observed some signal of association at childhood onset loci, which we did not. The fact 470 that the odds ratios for asthma at shared loci are relatively similar from approximately 471 age 25 to age 65 ( Figure 3) and that we do not detect any association signal at the 472 major COPD locus on chromosome 15q25.1 (appendix Figure 10), further suggests that 473 there is negligible misclassification of cases in the older age groups. Second, although 474 we used stringent criteria to classify loci as childhood or adult onset specific, we can't 475 exclude the possibility that in infinitely large sample sizes the effect sizes of some of 476 these loci will have 95% CIs that overlap or the association p-value will become smaller 477 than 0·05. Conversely, some of the shared loci with modest p-values in the adult onset 478 cases may not be true risk loci for asthma with onset at older ages. Third, the gene 479 expression data used to predict candidate target genes included heterogeneous tissues 480 and were collected mostly from adults. As a result, our study may have missed relevant 481 genes whose expression is developmentally regulated or environment specific. Our 482 finding of candidate genes at only 22 of the 61 asthma loci may be due in part to the 483 importance of both in asthma pathogenesis. Moreover, all inference based on gene 484 expression is using imputed expression. It is possible, therefore, that some relevant 485 genes were more difficult to impute and not included in our analysis, although a recent comparative study showed that PrediXcan is a more robust method for prediction of 487 gene expression than other related methods 56 . Fourth, because of the ethnic 488 composition of UKB, this study was limited to individuals of European ancestry only. As 489 a result, we could not evaluate the genetic risk architecture or assess the effects of age 490 of onset specific loci in other populations. 491 In the largest asthma GWAS to date, we show that genetic risk loci for adult 492 onset asthma is largely a subset of the loci associated with childhood onset asthma, 493 with overall smaller effect sizes for onset at later ages. These data suggest that 494 childhood onset specific loci and those associated with age of onset play a role in 495 disease initiation, whereas the other associated loci reflect shared mechanisms of 496 disease progression. The differences in the target tissues that most highly express the 497 genes at associated loci and the predicted expression of genes at age specific and 498 shared loci provides additional genetic and molecular evidence for both shared and 499 distinct pathogenic mechanisms in childhood onset and adult onset asthma. It is 500 therefore possible that the most effective treatments will also differ between these two 501 groups, and that strategies for precision medicine should be further personalized to 502 account for age of asthma onset.