Meta-analysis of exome array data identifies six novel genetic loci for lung function

Version Changes

Revised. Amendments from Version 2

We have added a further limitation to the discussion of the paper outlining a recently highlighted issue regarding the trait transformation undertaken in our replication analyses. We show through sensitivity analyses that our results are not affected by this issue (Supplementary Figure 4), but note that future studies should avoid such a transformation.

Abstract

Background: Over 90 regions of the genome have been associated with lung function to date, many of which have also been implicated in chronic obstructive pulmonary disease.

Methods: We carried out meta-analyses of exome array data and three lung function measures: forced expiratory volume in one second (FEV ₁), forced vital capacity (FVC) and the ratio of FEV ₁ to FVC (FEV ₁/FVC). These analyses by the SpiroMeta and CHARGE consortia included 60,749 individuals of European ancestry from 23 studies, and 7,721 individuals of African Ancestry from 5 studies in the discovery stage, with follow-up in up to 111,556 independent individuals.

Results: We identified significant (P<2·8x10 ^-7) associations with six SNPs: a nonsynonymous variant in RPAP1, which is predicted to be damaging, three intronic SNPs ( SEC24C, CASC17 and UQCC1) and two intergenic SNPs near to LY86 and FGF10. Expression quantitative trait loci analyses found evidence for regulation of gene expression at three signals and implicated several genes, including TYRO3 and PLAU.

Conclusions: Further interrogation of these loci could provide greater understanding of the determinants of lung function and pulmonary disease.

Introduction

Measures of lung function act as predictors of mortality and morbidity and form the basis for the diagnosis of several diseases, most notably chronic obstructive pulmonary disease (COPD), one of the leading causes of death globally ¹. Environmental factors, including smoking and exposure to air pollution play a significant role in lung function; however there has also been shown to be a genetic component, with estimates of the narrow sense heritability ranging between 39–66% ^{2– 5}. Genome-wide association studies (GWAS) of lung function have identified associations between single nucleotide polymorphisms (SNPs) and lung function at over 150 independent loci to date ^{6– 14}. Associations have also been identified in GWAS of COPD ^{15– 19}; however, the identification of disease associated SNPs has been restricted by limited sample sizes. Many signals first identified in powerful studies of quantitative lung function traits, have been found to be associated with risk of COPD, highlighting the potential clinical usefulness of comprehensive identification of lung function associated SNPs ¹³.

Low frequency (minor allele frequency (MAF) 1–5%) and rare (MAF<1%) variants have been largely underexplored by GWAS to date. Exome arrays have been designed to facilitate the investigation of these low frequency and rare variants, predominately within coding regions, in large sample sizes. Alongside a core content of rare coding SNPs, the exome array additionally includes common variation, including tags for previously identified GWAS hits, ancestry informative SNPs, a grid of markers for estimating identity by descent and a random selection of synonymous SNPs ²⁰.

An earlier version of this article can be found on bioRxiv ( https://doi.org/10.1101/164426)

Results

We carried out a meta-analysis of exome array data and three lung function measures: forced expiratory volume in one second (FEV ₁), forced vital capacity (FVC) and the ratio of FEV ₁ to FVC (FEV ₁/FVC). These analyses included 68,470 individuals from the SpiroMeta and CHARGE consortia in a discovery analysis, with follow-up in an independent sample of up to 111,556 individuals. All studies are listed with their study-specific sample characteristics in Table 1, with full study descriptions, including details of spirometry and other measurements described in the Supplementary Note. The genotype calling procedures implemented by each study ( Supplementary Table 1) and quality control of genotype data are described in the Supplementary Methods. We have undertaken both single variant analyses, and gene-based associations, which test for the joint effect of several rare variants in a gene (see Methods for details).

Table 1. Sample characteristics of 11 SpiroMeta and 12 CHARGE studies contributing to the discovery analyses and three studies contributing to the replication analyses.

Discovery studies
SpiroMeta studies	Total sample	n (%) Male	Ever smokers, n (%)	Age, mean (SD)	FEV 1, litres. mean (SD)	FVC, litres. mean (SD)	FEV 1/FVC, mean (SD)
1958 British Birth Cohort (B58C)	5270	2961 (56·2%)	2866 (53·3%)	44·00 (0·00)	3·35 (0·79)	4·29 (1·03)	0·788 (0·09)
Generation Scotland (GS:SFHS)	8164	3413 (41·8%)	3806 (46·6%)	51·59 (13·33)	2·78 (0·87)	3·91 (1·01)	0·710 (0·12)
Cooperative Health Research in the Region of Augsburg (KORA F4)	1447	701 (48·5%)	900 (62·2%)	54·82 (9·66)	3·24 (0·85)	4·20 (1·04)	0·771 (0·07)
CROATIA-Korcula cohort (KORCULA)	791	296 (36·8%)	418 (52·0%)	55·56 (13·69)	2·72 (0·83)	3·29 (0·95)	0·829 (0·10)
Lothian Birth Cohort 1936 (LBC1936)	974	501 (50·6%)	554 (55·9%)	69·55 (0·84)	2·38 (0·67)	3·04 (0·87)	0·787 (0·10)
Study of Health in Pomerania (SHIP)	1681	831 (49·4%)	955 (56·8%)	52·25 (13·43)	3·29 (0·88)	3·88 (1·03)	0·848 (0·07)
Northern Swedish Population Health Study (NSPHS)	880	407 (46·3%)	122 (13·9%)	49·13 (19·96)	2·93 (0·90)	3·53 (1·06)	0·831 (0·09)
Prospective Investigation of the Vasculature in Uppsala Seniors (PIVUS)	836	413 (49·4%)	426 (51·0%)	70·20 (0·17)	2·44 (0·68)	3·20 (0·87)	0·76 (0·10)
Swiss study on Air Pollution and Lung Disease in adults (SAPALDIA)	2707	1379 (50·9%)	1399 (51·7%)	40·86 (10·92)	3·65 (0·83)	4·62 (1·04)	0·794 (0·07)
The Cardiovascular Risk in Young Finns Study (YFS)	434	198 (47·3%)	186 (44·4%)	38·88 (5·07)	3·73 (0·75)	4·68 (0·99)	0·800 (0·06)
Finnish Twin Cohort (FTC)	214	0 (0%)	0 (0%)	68·73 (3·31)	2·18 (0·47)	2·79 (0·58)	0·786 (0·08)
Total	23,398
CHARGE studies (European Ancestry)	Total sample	n (%) Male	Ever smokers, n (%)	Age, mean (SD)	FEV 1, litres. mean (SD)	FVC, litres. mean (SD)	FEV 1/FVC, mean (SD)
AGES-Reykjavik study (AGES)	1566	649 (41·4%)	900 (57·5%)	76·1 (5·62)	2·13 (0·70)	2·87 (0·86)	0·744 (0·09)
Atherosclerosis Risk in Communities Study (ARIC)	10,680	5015 (47·0%)	631 (59·1%)	54·3 (5·70)	2·94 (0·77)	3·98 (0·98)	0·738 (0·07)
Cardiovascular Health Study (CHS)	3967	1737 (43·8%)	2089 (52·7%)	72·8 (5·55)	2·11 (0·66)	3·00 (0·86)	0·702 (0·10)
NHLBI Family Heart Study (FAMHS)	1651	718 (43·5%)	698 (42·3)	53·5 (12·60)	2·91 (0·853)	3·89 (1·05)	0·746 (0·08)
Framingham Heart Study (FHS)	7113	3241 (45·5%)	3780 (53·1)	50·7 (14·12)	3·10 (0·925)	4·09 (1·12)	0·755 (0·08)
Health Aging and Body Composition Study (HABC)	1457	786 (53·2%)	831 (56·5%)	73·7 (2·83)	2·31 (0·66)	3·11 (0·81)	0·741 (0·08)
Health2006 Study	2714	1217 (44·8%)	1577 (58·1%)	49·4 (13·04)	3·13 (0·82)	3·99 (0·99)	0·784 (0·07)
Health2008 Study	687	297 (43·2%)	384 (55·9%)	46·7 (8·22)	3·27 (0·79)	4·13 (0·97)	0·791 (0·06)
Inter99 Study (without pack-years)	1115	549 (49·2%)	1115 (100%)	47·2 (7·76)	3·26 (0·71)	4·12 (0·92)	0·796 (0·07)
Inter99 Study (with pack-years)	4179	2027 (48·5%)	2307 (55·2%)	45·8 (7·95)	3·21 (0·76)	4·10 (0·97)	0·788 (0·08)
Multi-Ethnic Study of Atherosclerosis (MESA)	1323	654 (49·4%)	751 (56·8%)	66·0 (9·8)	2·57 (0·76)	3·51 (0·10)	0·733 (0·08)
The Rotterdam Study (RS)	546	299 (54·8%)	382 (70·0%)	79·4 (5·00)	2·27 (0·68)	3·03 (0·86)	0·750 (0·08)
Total	36,998
CHARGE studies (African Ancestry)	Total Sample	n (%) Male	Ever smokers, n (%)	Age, mean (SD)	FEV 1, litres. mean (SD)	FVC, litres. mean (SD)	FEV 1/FVC, mean (SD)
Atherosclerosis Risk in Communities Study (ARIC)	3180	1183 (37·2%)	1680 (59·1%)	53·6 (5·83)	2·48 (0·65)	3·25 (0·82)	0·765 (0·08)
Cardiovascular Health Study (CHS)	624	232 (37·2%)	340 (54·4%)	73·2 (5·49)	1·76 (0·58)	2·48 (0·80)	0·717 (0·11)
Health Aging and Body Composition Study (HABC)	943	433 (45·9%)	543 (57·6%)	73·4 (2·90)	1·96 (0·57)	2·61 (0·71)	0·749 (0·09)
Jackson Heart Study (JHS)	2143	793 (36·8%)	688 (31·9%)	52·8 (12·6)	2·43 (0·72)	3·02 (0·86)	0·807 (0·09)
Multi-Ethnic Study of Atherosclerosis (MESA)	861	404 (46·9%)	467 (54·2%)	65·6 (9·6)	2·19 (0·66)	2·92 (0·86)	0·756 (0·09)
Total	7721
Replication studies
Study name	Total Sample	n (%) Male	Ever smokers, n (%)	Age, mean (SD)	FEV 1, litres. mean (SD)	FVC, litres. mean (SD)	FEV 1/FVC, mean (SD)
UK Biobank	98,657	45,166 (45·8%)	56,404 (57·2%)	56·7 (7·92)	2·75 (0·80)	3·67 (0·98)	0·75 (0·07)
UK Household Longitudinal Study (UKHLS)	7443	3293 (44·2%)	4509 (60·5%)	53·10 (15·94)	2·89 (0·90)	3·83 (1·08)	0·753 (0·09)
Netherlands Epidemiology of Obesity study (NEO)	5456	2672 (48·0%)	3674 (66·0%)	55·9 (5·9)	3·26 (0·80)	4·26 (1·02)	0·77 (0·07)
Total	111,556

Meta-analyses of single variant associations

We first evaluated single variant associations between FEV ₁, FVC and FEV ₁/FVC and the 179,215 SNPs that passed study level quality control and were polymorphic in both consortia. These analyses identified 34 SNPs in regions not previously associated with lung function, showing association with at least one trait at overall P<10 ^-5, and showing association with consistent direction and P<0·05 in both consortia (full results in Supplementary Table 2, quantile-quantile and Manhattan plots shown in Supplementary Figure 1). We followed up these SNP associations in a replication analysis comprising 3 studies with 111,556 individuals. Combining the results from the discovery and replication stages in a meta-analysis identified six SNPs in total that were independent to known signals and met the pre-defined significance threshold (P<2·8×10 ^-7) overall in, or near to FGF10, LY86, SEC24C, RPAP1, CASC17 and UQCC1 ( Table 2, Supplementary Figure 2). A SNP near to the CASC17 signal (rs11654749, r ²=0·3 with rs1859962) has previously been associated with FEV ₁ in a genome-wide analysis of gene-smoking interactions, although this association was not replicated at the time ²¹; the present analysis provides the first evidence for independent replication of this signal. A seventh signal was also identified in LCT ( Table 2, Supplementary Figure 2); whilst this locus has not previously been implicated in lung function, this SNP is known to vary in frequency across European populations ²², and we cannot rule out that this association is not an artefact of population structure. Our discovery analysis furthermore identified associations (P<10 ^-5) in 25 regions previously associated with one or more of FEV ₁, FVC and FEV ₁/FVC ( Supplementary Table 3).

Table 2. Novel loci associated with lung function traits.

Results are shown for variant in novel loci associated (P<2·7×10 ^-7) with lung function traits in a two stage meta-analysis consisting of up to 68,470 individuals from the SpiroMeta and CHARGE Consortia in the discovery analyses, with follow-up in up to 111,556 individuals from UK Biobank, UKHLS and NEO. For each SNP, the result for the trait-smoking-ancestry combination which resulted in the most statistically significant association is given. The results for these SNPs and all three traits are shown in Supplementary Table 12. Beta values from SpiroMeta (β _Sp) reflect effect-size estimates on an inverse-normal transformed scale after adjustments for age, age ², sex, height and ancestry principal components, and stratified by ever smoking status (Analysis of All individuals only). Beta values from CHARGE (β _CH) reflect effect-size estimates on an untransformed scale (litres for FEV ₁ and FVC; ratio for FEV ₁/FVC). Samples sizes (N), Z-statistics (Z) and two-sided P-values (P) are given for the combined discovery analysis and the replication analysis. Two-sided P-values are also given for the full two-stage combined analyses (discovery + replication).

								Consortium results		Combined discovery meta-analysis			Replication			Two-stage combined
SNP	Chr:Pos	(Nearest) gene(s)	Trait	Smoking	Ancestry	Effect/other allele	Effect allele frequency (Discovery)	β CH	β Sp	N disc	Z disc	P disc	N rep	Z rep	P rep	P meta
rs2322659	2:136555659	LCT (nonsynonymous)	FVC	All Individuals	EA Only	T/C	23·5%	27·34	0·032	55,591	5·597	2·18×10 -8	12,899	2·286	0·0223	1·70 ×10 -9
rs1448044	5:44296986	FGF10(dist=8111), NNT(dist=591,318)	FVC	Ever Smokers	EA+AA	A/G	35·6%	18·63	0·057	30,966	4·813	1·49 ×10 -6	64,400	4·805	1·55 ×10 -6	2·22 ×10 -11
rs1294421	6:6743149	LY86(dist=87,933), RREB1(dist=364,681)	FEV 1/ FVC	All Individuals	EA+AA	T/G	36·8%	-0·222	-0·038	68,099	-5·479	4·27 ×10 -8	111,556	-8·171	3·06 ×10 -16	9·74 ×10 -23
rs3849969	10: 75525999	SEC24C (intronic)	FEV1	All Individuals	EA+AA	T/C	29·4%	13·10	0·036	68,116	4·767	1·87 ×10 -6	111,556	5·042	4·60 ×10 -7	4·99 ×10 -12
rs1200345	15: 41819716	RPAP1 (nonsynonymous)	FEV1/ FVC	All Individuals	EA only	C/T	48·8%	-0·217	-0·025	60,381	-4·586	4·51 ×10 -6	111,556	-5·725	1·03 ×10 -8	2·33 ×10 -13
rs1859962	17: 69108753	CASC17 (intronic)	FEV 1	All Individuals	EA only	G/T	48·2%	15·39	0·026	60,395	4·876	1·08 ×10 -6	111,554	4·612	3·99 ×10 -6	4·10 ×10 -11
rs6088813	20: 33975181	UQCC1 (intronic)	FVC	All Individuals	EA+AA	C/A	36·7%	-16·16	-0·023	68,115	-4·634	3·58×10 -6	111,556	-7·688	1·50 ×10 -14	4·90 ×10 -19

Generally, the observed effect of the SNPs at the novel signals were similar in ever and never smokers; the exception was rs1448044 near FGF10, which showed a significant association with FVC only in ever smokers in our discovery analysis (ever smokers P=1·49×10 ^-6; never smokers P=0·695, Supplementary Table 4 and Supplementary Figure 3). In the replication analysis, however, this association was observed in both ever and never smokers (ever smokers P=3·14×10 ^-5; never smokers P=1·40×10 ^-4, Supplementary Table 5). For rs1200345 ( RPAP1) and rs1859962 ( CASC17), associations were most statistically significant in the analyses restricted to individuals of European Ancestry ( Supplementary Table 4 and Supplementary Figure 3), as was the association with rs2322659 ( LCT), giving further support that this association may be due to population stratification.

Meta-analyses of gene-based associations

We undertook Weighted Sum Tests (WST) ²³ and Sequence Kernel Association tests (SKAT) ²⁴ to assess the joint effects of multiple low frequency variants within genes on lung function traits. In our discovery analyses of all 68,470 individuals, we tested up to 14,380 genes that had at least two variants with MAF<5% and met the inclusion criteria (exonic or loss of function [LOF], see Methods for definitions) in both consortia. The SKAT analyses identified 16 genes associated (P<0·05 in both consortia and overall P<10 ^-4) with FEV ₁, FVC or FEV ₁/FVC ( Supplementary Table 6), whilst the WST analyses identified 12 genes ( Supplementary Table 7). There was one gene ( LY6G6D) that was identified in both analyses. These genes were followed up in UK Biobank, with two genes, GPR126 and LTBP4, showing evidence of replication in the exonic SKAT analysis (P<3·5×10 ^-6); however conditional analyses in UK Biobank showed that both these associations were driven by single SNPs, that were identified in the single variant association analyses and have been previously reported in GWAS of these traits ( Supplementary Table 6 and Supplementary Table 7).

Functional characterization of novel loci

In order to gain further insight into the six loci identified in our analyses of single variant associations (excluding LCT), we employed functional annotation and assessed whether identified SNPs in these regions were associated with gene expression levels. One of the identified novel SNPs was nonsynonymous, three intronic and two were intergenic. We found evidence that three of the SNPs may be involved in cis-acting regulation of the expression of several genes in multiple tissues ( Supplementary Table 8).

SNP rs1200345 in RPAP1 is a nonysynomous variant, predicted to be deleterious by both SIFT (deleterious) and Polyphen (possibly damaging) ( Supplementary Table 9); RPAP1 is ubiquitously expressed, with high levels of protein detected in the lung ( Supplementary Table 10). SNP rs1200345 or proxies (r ²>0·8) were also found to be amongst the most strongly associated SNPs with expression levels of RPAP1 in several tissues, including lung, and with a further six genes in lung tissue ( Supplementary Table 8), including TYRO3, one of the TAM family of receptor tyrosine kinases. TYRO3 regulates several processes including cell survival, migration and differentiation and is highly expressed in lung macrophages ( Supplementary Table 10). Evidence of association with gene expression was found at two more of the novel signals (sentinel SNPs rs3849969 and rs6088813), implicating a further 16 genes. Of note, in blood expression quantitative trait loci (eQTL) databases, a proxy of a SNP in complete linkage disequilibrium (r ²=1) with rs3849969 (rs3812637) was an eQTL for plasminogen activator, urokinase ( PLAU).

Discussion

We undertook an analysis of 68,470 individuals from 23 studies with data from the exome array and three lung function traits, following up the most significant single SNP and gene-based associations in an independent sample of up to 111,556 individuals. There were six SNPs which reached P<10 ^-5 in the discovery stage meta-analysis of single variant associations, and subsequently met the Bonferroni corrected significance threshold for independent replication (P<1⋅47×10 ^-3, corrected for 34 SNPs being tested). In the combined analyses of our discovery and replication analyses, these six SNPs met the exome chip-wide significance threshold (P<2⋅8×10 ^-7). One of the SNPs is in a region that has previously been implicated in lung function (near KCJN2/SOX9) ²¹, whilst the remaining five SNPs, although all common, have not previously been identified in other GWAS of lung function. In a recent 1000 Genomes imputed analysis of lung function (which includes some of the studies contributing to the present discovery analysis), all of these SNPs showed at least suggestive association (2·97×10 ^-3>P>1·28×10 ^-5) with one or more lung function trait, but none reached the required threshold (P<5×10 ^-6) to be taken forward for replication in that analysis ¹².

We further identified a seventh association with rs2322659 in LCT (MAF=23⋅5%; combined discovery + replication P=1⋅70×10 ^-9). Given SNPs in this region are known to vary in frequency across European populations, we cannot dismiss the possibility that this association may be confounded by population stratification; hence we do not report this signal as a novel lung function locus. For SNPs at 7 loci that have been shown to have differences in allele frequency between individuals from different regions of the UK ²⁵, and subsequently European populations (including the LCT locus), we undertook a look-up of associations with lung function in our discovery analyses. and subsequently across European populations ²⁶. Aside from the association between the LCT locus and FVC, no significant associations were observed between SNPs at these loci and any lung function trait, in either the analyses restricted to European Ancestry (EA) individuals, or in the analysis of EA and African Ancestry (AA) individuals combined ( Supplementary Table 11); this suggests population structure was generally accounted for adequately in our analyses.

One of the novel signals was with a nonsynonymous SNP, rs1200345 in RPAP1, (MAF=48⋅8%; P=2⋅33×10 ^-13), which is predicted to be deleterious. This SNP and proxies with r ²>0·8 were also associated with expression in lung tissue of seven genes, including RPAP1 and the TAM receptor TYRO3. TAM receptors play a role in the inhibition of Toll-like receptors (TLRs)-mediated innate immune response by initiating the transcription of cytokine signalling genes (SOCS-1 and 3), which limit cytokine overproduction and inflammation ^{27, 28}. It has been shown that influenza viruses H5N1 and H7N9 can cause downregulation of Tyro3, resulting in an increased inflammatory cytokine response ²⁸.

Three further signals were with intronic SNPs in SEC24C (MAF=29⋅4%; P=4⋅99×10 ^-12), CASC17 (MAF=48⋅2%; P=4⋅10×10 ^-11), and UQCC1 (MAF=36⋅7%; P=4⋅90×10 ^-19). Two of these intronic SNPs have previously been implicated in GWAS of other traits: rs1859962 in CASC17 with prostate cancer ²⁹ and rs6088813 in UQCC1 with height ³⁰. The CASC17 locus, near KCNJ2/SOX9 has also previously been implicated in lung function, showing significant association with FEV ₁ in a genome-wide analysis of gene-smoking interactions; however, this association was not formally replicated ²¹. Whilst the individuals utilised in the discovery stage of this analysis overlap with those included in this previous interaction analysis, the replication stage of the present study provides the first evidence of replication for this signal in independent cohorts. In the present analysis, there was no evidence that the results differed by smoking status.

SNPs rs6088813 in UQCC1 and rs3849969 in SEC24C were identified as eQTLs for multiple genes. Whilst our eQTL analysis did not include formal tests of colocalisation, a SNP in complete linkage disequilibrium with rs3849969 (rs3812637, r ²=1) was associated with expression of PLAU in blood. The plasminogen activator, urokinase (PLAU) plays a role in fibrinolysis and immunity, and with its receptor (PLAUR) is involved in degradation of the extra cellular matrix, cell migration, cell adhesion and cell proliferation ³¹. A study of preterm infants with respiratory distress syndrome, a condition characterised by intra-alveolar fibrin deposition, found PLAU and its inhibitor SERPINE1 to be expressed in the alveolar epithelium, and an increased ratio of SERPINE1 to PLAU was associated with severity of disease ³². Studies in mice have also shown that increased expression of Plau may be protective against lung injury, by reducing fibrosis ³³. PLAU has also been found to be upregulated in lung epithelial cells subjected to cyclic strain ³⁴ and in patients with COPD and lung cancer, PLAU was found to be expressed in alveolar macrophages and epithelial cells ³¹.

The final two signals were with common intergenic SNPs close to LY86 (MAF=36⋅8%; P=9⋅74×10 ^-23) and FGF10 (MAF=35⋅6%; P=2⋅22×10 ^-11). LY86 (lymphocyte antigen 86) interacts with the Toll-like receptor signalling pathway, to form a heterodimer, when bound with RP105 ³⁵. The sentinel SNP in the present analysis (rs1294421) has previously shown association with waist-hip ratio ³⁶, whilst an intronic SNP within LY86 (rs7440529, r ²=0·005 with rs1294421) has been implicated in asthma in two studies of individuals of Han Chinese ancestry ^{37, 38}. FGF10 is a member of the fibroblast growth factor family of proteins, and is involved in a range of biological processes, including embryonic development and morphogenesis, cell growth and repair, tumor growth and invasion. Specifically, the FGF10 signalling pathway is thought to play an criticial role in the development of the lung and in lung epithelial renewal ³⁹. A deficiency in Fgf10 has been demonstrated to lead to a fatal disruption of branching morphogenesis during lung development in mice ⁴⁰.

Our discovery analyses included individuals of both EA and AA. Two of the identified six novel signals showed inconsistent effects in the AA and EA individuals. For these SNPs, the associations in AA individuals were not statistically significant, and we report associations from the analysis restricted to EA individuals only. For the remaining four SNPs similar effects were observed in both the EA and AA individuals ( Supplementary Figure 3). We also examined the effects of the novel SNPs in ever smokers and never smokers separately and found these to be broadly similar, with the exception of rs1448044 in FGF10, which in the discovery analysis showed significant association with FVC in ever smokers, whilst showing no association in never smokers (P=0·695). However, in our replication stage analyses, similar effects were seen in both ever and never smokers for this SNP, and the combined analysis of discovery and replication stages for this SNP, including both ever and never smokers, met the exome chip-wide significance level overall (P=4·22×10 ^-9). We also considered whether this signal could be driven by smoking behaviour in our discovery stage as our primary analyses in SpiroMeta did not adjust for smoking quantity. We undertook a look-up of this SNP in the publicly available results of a GWAS of several smoking behaviour traits ⁴¹; there was only weak evidence that this SNP was associated with ever versus never smoking (P=0·039), and no evidence for association with amount smoked (cigarettes per day, P=0·10).

Through the use of the exome array, we aimed to identify associations with low frequency and rare functional variants, thereby potentially uncovering some of the missing heritability of lung function. However, whilst our discovery analyses identified single SNP associations with 23 low frequency variants ( Supplementary Table 2), we did not replicate any of these findings. Eleven of these 23 SNPs we were unable to follow-up in our replication studies, due to them either being not genotyped, or monomorphic. Overall, limited statistical power is likely to explain our lack of convincing single variant associations with rare variants, in particular if those variants exhibit only modest effects ⁴². We additionally investigated the joint effects of low frequency and rare variants within genes, on lung function trait, by employing SKAT and WST gene-based tests. These analyses identified associations with a number of genes that could not attributed to the effect of a single SNP. Replication of these gene-based signals proved difficult however, as again a number of SNPs included in the discovery stage of these analyses were monomorphic, or had not been not genotyped in the replication studies. This lead to a disparity in the gene unit being tested in our discovery and replication samples; hence interpretation of these results was not clear-cut. In the end, we were able to replicate only findings with common SNPs. This finding is in line with several other studies of complex traits and exome array data that have been unable to report robust associations with low frequency variants ^{43– 45} and it is clear that future studies will require increasingly larger sample sizes in order to fully evaluate the effect of variants across the allele frequency spectrum. The identification of common SNPs remains important, however, as such findings have the potential to highlight drug targets ⁴⁶, and these variants collectively could have utility in risk prediction.

In our replication analyses using UK Biobank, we applied adjustment for covariates including ancestry principal components, before undertaking inverse-normal transformations of the lung function phenotypes. Association analyses were then performed using these transformed phenotypes. It has recently been shown that such transformation has the potential to introduce correlations between principal components and phenotypes ⁴⁷; we undertook sensitivity analyses for the six reported SNPs by repeating the association analyses with phenotypes that had been transformed without prior adjustment, with covariate adjustment made as part of the SNP-trait association test. We found there to be some difference in P-values for some SNP-trait combinations; however, the six novel SNP associations we report all met the replication P-value threshold (P<1·47×10 ^-3) in the sensitivity analyses ( Supplementary Figure 4). This issue may also be relevant to the gene-based tests; however no replicated novel gene-based associations were identified in this study. Future studies should avoid undertaking adjustment for principal components of ancestry prior to trait transformation, in order to avoid this potential bias.

This study has identified six common SNPs, independent to signals previously implicated in lung function. Additional interrogation of these loci could lead to greater understanding of lung function and lung disease, and could provide novel targets for therapeutic interventions.

Methods

Study design, cohorts and genotyping

The SpiroMeta analysis included 23,751 individuals of EA from 11 studies, and the CHARGE analysis comprised 36,998 EA individuals and a further 7,721 individuals of AA from 12 studies. Follow-up analyses were conducted in an independent sample of up to 111,556 individuals from UK Biobank (2015 interim release), the UK Household Longitudinal Study (UKHLS) and the Netherlands Epidemiology of Obesity (NEO) Study ( Figure 1). All studies (excluding UK Biobank) were genotyped using either the Illumina Human Exome BeadChip v1 or the Illumina Infinium HumanCoreExome-12 v1·0 BeadChip. UK Biobank samples were genotyped using the Affymetrix Axiom UK BiLEVE or UK Biobank arrays.

Figure 1. Study design.

Statistical analyses

Consortium level analyses: Within the SpiroMeta Consortium, each study contributing to the discovery analyses calculated single-variant score statistics, along with covariance matrices describing correlations between variants, using RAREMETALWORKER ⁴⁸ or rvtests ⁴⁹. For each trait, these summary statistics were generated separately in ever and never smokers. Traits were adjusted for sex, age, age ² and height, and inverse normally transformed prior to association testing. For studies with unrelated individuals, SNP-trait associations were tested using linear models, with adjustments made for the first 10 ancestry principal components, whilst studies with related individuals utilised linear mixed models to account for familial relationships and underlying population structure.

Within the CHARGE Consortium, each study generated equivalent summary statistics using the R package SeqMeta ⁵⁰. For each trait, summary statistics were generated in ever and never smokers separately, and in all individuals combined. The untransformed traits were used for all analyses, adjusted for smoking status and pack-years, age, age ², sex, height, height ², centre/cohort. Models for FVC were additionally adjusted for weight. Linear regression models, with adjustment for principal components of ancestry were used for studies with unrelated individuals, and linear mixed models were used for family-based studies.

Within each consortium we used the score statistics and variance-covariance matrices generated by each study to construct both single variant and gene-based tests using either RAREMETAL ⁴⁸ (SpiroMeta) or SeqMeta ⁵⁰ (CHARGE). For single variant associations, score statistics were combined in fixed effects meta-analyses. Two gene-based tests were constructed: first, the Weighted Sum Test (WST) using Madsen Browning weightings ²³, and secondly, the Sequence Kernel Association Test (SKAT) ²⁴. We performed the SKAT and WST tests using two subsets of SNPs: 1) including all SNPs with an overall consortium-wide MAF<5% that were annotated as splicing, stopgain, stoploss, or frameshift (loss of function [LOF] analysis), and 2) including all SNPs meeting the LOF analysis criteria in addition to all other nonsynonymous variants with consortium wide MAF<5% (exonic analysis). Variants were annotated to genes using dbNSFP v2·6 ⁵¹ on the basis of the GRCh37/hg19 database.

For both single variant and gene-based associations, consortium-level results were generated for ever smokers and never smokers separately, and in all individuals combined. Within the CHARGE Consortium, results were combined separately for the EA and AA studies and also in a trans-ethnic analysis of both ancestries.

Combined meta-analysis: The single variant association results from the SpiroMeta and CHARGE consortia were combined as follows: The genomic inflation statistic (λ) was calculated for SNPs with consortium-wide MAF>1%; where λ had a value greater than one, genomic control adjustment was applied to the consortium level P-values. The consortium-level results were then combined using sample size weighted z-score meta-analysis. The λ was again calculated for the meta-analysis results and genomic control applied, as appropriate. λ values at the consortium and meta-analysis level are shown in Supplementary Table 13. Since we were interested in identifying low frequency and rare variants, we applied no MAF or minor allele count (MAC) filter. We identified SNPs of interest as those with an overall P<10 ^-5 and a consistent direction of effect and P<0·05 observed in both consortia. Rather than using a strict Bonferroni correction for defining the significance threshold, we adopted the more lenient P<10 ^-5 threshold in order to increase the power to detect variants with modest effect in our discovery analyses, whilst the requirement for consistency in results from the two consortia aimed to limit false positives. All SNPs meeting these thresholds were followed up in independent replication cohorts. Where we identified a SNP within 1Mb of a previously identified lung function SNP, we deemed the SNP to represent an independent signal if it had r ²<0·2 with the known SNP, and if it retained a P <10 ^-5, when conditional analyses were carried out with the known SNP, or a genotyped proxy, using data from the SpiroMeta Consortium, or UK Biobank. Our primary meta-analysis included all individuals; we additionally carried out analyses in smoking subgroups (ever and never smokers), and in the subgroup of individuals of European ancestry only.

For genes which contained at least 2 polymorphic SNPs in both consortia, we combined the results of the consortium level gene based tests using either z-score meta-analysis (for the WST analysis) or Fisher’s Method for combining P-values (in the case of SKAT). We identified genes of interest as those with P<0·05 observed in both consortia and an overall P<10 ^-4, thresholds again chosen to limit both false positive and false negative findings. As in the analyses of single variant associations, our primary meta-analyses included all individuals, with secondary analyses undertaken in smoking and ancestry specific subgroups.

Replication analyses: All SNP and gene-based associations were followed up for the trait with which they showed the most statistically significant association only. For associations identified through the smoking subgroup analyses, we followed up associations in the appropriate smoking strata; however, no ancestry stratified follow-up was undertaken as replication studies included only a sufficient number of individuals of European Ancestry.

Single variant associations in UK Biobank were tested in ever smokers and never smokers separately, and stratified by genotyping array (UK BiLEVE array or UK Biobank array) using the score test as implemented in SNPTEST v2·5b4 ⁵². Traits were adjusted for age, age ², height, sex, ten principal components and pack-years (ever smokers only), and the adjusted traits were inverse normally transformed. Correlations between principal components and transformed phenotypes may be introduced where adjustment is made prior to transformation. In this analysis, we found any introduced correlations to have no impact on the conclusion of our replication analyses; however future studies should apply transformation of phenotypes prior to covariate adjustment, to avoid this issue. For UKHLS, analyses were undertaken analogously to the SpiroMeta discovery studies using RAREMETALWORKER, while for NEO, analyses were undertaken in the same way as was done in the CHARGE discovery studies using SeqMeta. The single variant results from all replication studies were combined using sample size weighted Z-score meta-analysis. Subsequently, we combined the results from the discovery and replication stage analyses and we report SNPs with overall exome-wide significance of P<2·8×10 ^-7 (Bonferroni corrected for the original 179,215 SNPs tested).

We followed up genes of interest (P<10 ^-4) using data from UK Biobank only. Summary statistics for UK Biobank were generated using RAREMETALWORKER, with gene-based tests then constructed using RAREMETAL. Finally, we combined the results from the discovery analysis with the replication results in an overall combined meta-analysis using either z-score meta-analysis (WST) or Fisher’s Method (SKAT). We declared genes with overall P<3·5×10 ^-6 (Bonferroni corrected for 14,380 genes tested) in our combined meta-analysis to be statistically significant. For these statistically significant genes, we carried out additional analyses using the UK Biobank data in which we conditioned on the most significantly associated individual SNP within that gene, to determine whether this was a true gene-based signal, or whether the association could be ascribed to the single SNP (if the conditional P<0·01, then association was deemed to not be driven by the single SNP).

Characterization of findings

In order to gain further insight into the loci identified in our analyses of single variant associations, we assessed whether these regions were associated with gene expression levels in various tissues (FDR of 5%, or q-value<0·05), by querying a publically available blood eQTL database ⁵³ and the GTEx project ⁵⁴ for the sentinel SNPs, or any proxy (r ²>0·8). We further assessed SNPs of interest (and proxies) within a lung eQTL resource based on non-tumour lung tissues of 1,111 individuals ^{55– 57}. Descriptions of these resources and further details of the look-ups are provided in the Supplementary Methods. Moreover, all sentinel SNPs and proxies with r ²>0.8 were annotated using ENSEMBL’s Variant Effect Predictor (VEP) ⁵⁸; potentially deleterious coding variants were identified as those annotated as ‘deleterious’ by SIFT ⁵⁹ or ‘probably damaging’ or ‘possibly damaging’ by PolyPhen-2 ⁶⁰. For all genes implicated through the expression data or functional annotation, we searched for evidence of protein expression in the respiratory system by querying the Human Protein Atlas ⁶¹.

Data availability

Summary level results for all analyses are available on OSF: https://doi.org/10.17605/OSF.IO/NSDPJ ⁶²

Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0).

This research has been conducted using the UK Biobank Resource. The genetic and phenotypic UK Biobank data are available upon application to the UK Biobank ( https://www.ukbiobank.ac.uk/) to all registered health researchers. These data are from Understanding Society: The UK Household Longitudinal Study (UKHLS), which is led by the Institute for Social and Economic Research at the University of Essex and funded by the Economic and Social Research Council. The data were collected by NatCen and the genome wide scan data were analysed by the Wellcome Trust Sanger Institute. Information on how to access the data can be found on the Understanding Society website https://www.understandingsociety.ac.uk/.

Funding Statement

MDT has been supported by MRC fellowships G0501942 and G0902313. MDT and LVW are supported by the MRC (MR/N011317/1). IPH is supported by the MRC (G1000861). ALW and SJL are supported by the Intramural Research Program of the NIH, National Institute of Environmental Health Sciences (ZIA ES 043012). We acknowledge use of phenotype and genotype data from the British 1958 Birth Cohort DNA collection, funded by the Medical Researanch Council grant G0000934 and the Wellcome Trust grant 068545/Z/02. APM was a Wellcome Trust Senior Fellow in Basic Biomedical Science (grant number WT098017) and was also supported by Wellcome Trust grant WT064890. EI is supported by the Swedish Research Council (2012-1397), Knut och Alice Wallenberg Foundation (2013.0126) and the Swedish Heart-Lung Foundation (20140422). JK is supported by Academy of Finland Center of Excellence in Complex Disease Genetics grants 213506, 129680 and Academy of Finland grants 265240, 263278. The Finnish Twin Cohort is supported by the Welcome Trust Sanger Institute, UK. The Lothian Birth Cohort is supported by Age UK (The Disconnected Mind Project), the UK Medical Research Council (MR/K026992/1) and The Royal Society of Edinburgh. ÅJ is supported by the Swedish Society for Medical Research (SSMF), The Kjell och Märta Beijers Foundation, The Marcus Borgström Foundation, The Åke Wiberg foundation and The Vleugels Foundation. UG is supported by Swedish Medical Research Council grants K2007-66X-20270-01-3 and 2011-2354 and European Commission FP6 (LSHG-CT-2006-01947). SHIP is part of the Community Medicine Research net of the University of Greifswald, Germany, which is funded by the Federal Ministry of Education and Research, the Ministry of Cultural Affairs as well as the Social Ministry of the Federal State of Mecklenburg-West Pomerania, and the network ‘Greifswald Approach to Individualized Medicine (GANI_MED)’ funded by the Federal Ministry of Education and Research, and the German Asthma and COPD Network (COSYCONET) (grant no.01ZZ9603, 01ZZ0103, 01ZZ0403, 03IS2061A, BMBF 01GI0883). ExomeChip data have been supported by the Federal Ministry of Education and Research (grant no. 03Z1CN22) and the Federal State of Mecklenburg-West Pomerania. The University of Greifswald is a member of the Caché Campus program of the InterSystems GmbH. UKHLS is supported by grants WT098051 (Wellcome Trust) and ES/H029745/1 (Economic and Social Research Council). Y.B. holds a Canada Research Chair in Genomics of Heart and Lung Diseases. Lies Lahousse is a Postdoctoral Fellow of the Research Foundation - Flanders (FWO grant G035014N). The Rotterdam Study is funded by Erasmus Medical Center and Erasmus University, Rotterdam, the Netherlands Organization for Scientific Research (NOW), the Netherlands Organization for the Health Research and Development (ZonMw), the Research Institute for Diseases in the Elderly (RIDE), the Ministry of Education, Culture and Science, the Ministry for Health, Welfare and Sports, the European Commission (DG XII), and the Municipality of Rotterdam. Genotyping in the Rotterdam study was supported by Netherlands Organization for Scientific Research (NOW grants 175.010.2005.011 ; 911-03-305 012), the Research Institute for Diseases in the Elderly (RIDE2 grants 014-93-015) and Netherlands Genomics Initiative (NGI)/Netherlands Consortium for Healthy Aging (NCHA grant050-060-810). MESA/MESA SHARe is supported by HHS (HHSN268201500003I), NIH/NHLBI (contracts N01-HC-95159, N01-HC-95160, N01-HC-95161, N01-HC-95162, N01-HC-95163, N01-HC-95164, N01-HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168, N01-HC-95169) and HIH/NCATS (contracts UL1-TR-000040, UL1-TR-001079, UL1-TR-001881, DK063491). MESA SHARe is funded by NIH/NHLBI contract N02-HL-64278, MESA Air is funded by US EPA (RD831697) and MESA Spirometry funded by NIH/NHLBI (R01-HL077612). SSR and BMP are supported by NIH/NHLBI grant rare variants and NHLBI traits in deeply phenotyped cohorts (R01-HL120393). Cardiovascular Health Study: This CHS research was supported by NHLBI contracts HHSN268201200036C, HHSN268200800007C, HHSN268201800001C, HHSN268200960009C, N01HC55222, N01HC85079, N01HC85080, N01HC85081, N01HC85082, N01HC85083, N01HC85086; and NHLBI grants U01HL080295, R01HL068986, R01HL087652, R01HL105756, R01HL103612, R01HL120393, and R01HL130114 with additional contribution from the National Institute of Neurological Disorders and Stroke (NINDS). Additional support was provided through R01AG023629 and R01HL085251 from the National Institute on Aging (NIA). The provision of genotyping data was suprovidedpported in part by the National Center for Advancing Translational Sciences, CTSI grant UL1TR001881, and the National Institute of Diabetes and Digestive and Kidney Disease Diabetes Research Center (DRC) grant DK063491 to the Southern California Diabetes Endocrinology Research Center. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The Atherosclerosis Risk in Communities (ARIC) study is carried out as a collaborative study supported by the National Heart, Lung, and Blood Institute (NHLBI) contracts (HHSN268201100005C, HHSN268201100006C, HHSN268201100007C, HHSN268201100008C, HHSN268201100009C, HHSN268201100010C, HHSN268201100011C, and HHSN268201100012C). Funding support for “Building on GWAS for NHLBI-diseases: the U.S. CHARGE consortium” was provided by the NIH through the American Recovery and Reinvestment Act of 2009 (ARRA) (5RC2HL102419). DOMK received funding from the Dutch Science Organisation (ZonMW-VENI Grant 916.14.023). The genotyping in the NEO study was supported by the Centre National de Génotypage (Paris, France), headed by Jean-François Deleuze. The NEO study is supported by the participating Departments, the Division and the Board of Directors of the Leiden University Medical Center, and by the Leiden University, Research Profile Area Vascular and Regenerative Medicine. SAPALDIA was supported by the Swiss National Science Foundation (grants no 33CS30-148470/1, 33CSCO-134276/1, 33CSCO-108796, , 324730_135673, 3247BO-104283, 3247BO-104288, 3247BO-104284, 3247-065896, 3100-059302, 3200-052720, 3200-042532, 4026-028099, PMPulDP3_129021/1, PMPDP3_141671/1), the Federal Office for the Environment, the Federal Office of Public Health, the Federal Office of Roads and Transport, the canton's government of Aargau, Basel-Stadt, Basel-Land, Geneva, Luzern, Ticino, Valais, and Zürich, the Swiss Lung League, the canton's Lung League of Basel Stadt/ Basel Landschaft, Geneva, Ticino, Valais, Graubünden and Zurich, Stiftung ehemals Bündner Heilstätten, SUVA, Freiwillige Akademische Gesellschaft, UBS Wealth Foundation, Talecris Biotherapeutics GmbH, Abbott Diagnostics, European Commission 018996 (GABRIEL), Wellcome Trust WT 084703MA. The Novo Nordisk Foundation Center for Basic Metabolic Research is an independent Research Center at the University of Copenhagen partially funded by an unrestricted donation from the Novo Nordisk Foundation (www.metabol.ku.dk). Generation Scotland received core support from the Chief Scientist Office of the Scottish Government Health Directorates [CZD/16/6] and the Scottish Funding Council [HR03006]. Genotyping of the GS:SFHS samples was carried out by the Genetics Core Laboratory at the Edinburgh Clinical Research Facility, University of Edinburgh, Scotland and was funded by the Medical Research Council UK.. The Croatia KORCULA study was supported by the Ministry of Science, Education and Sport in the Republic of Croatia (108-1080315-0302). JD, JCL, WG and GTOC are supported by NIH/NHLBI Contract HHSN268201500001I. Genotyping, quality control and calling of the Illumina HumanExome BeadChip in the Framingham Heart Study was supported by funding from the National Heart, Lung and Blood Institute Division of Intramural Research (Daniel Levy and Christopher J. O’Donnell, Principle Investigators). The AGES study is supported by the NIH (N01-AG012100), the Iceland Parliament (Alþingi) and the Icelandic Heart Association. HABC was supported by NIA contracts N01AG62101, N01AG62103, and N01AG62106; NIA grant R01-AG028050, and NINR grant R01- NR012459 and was supported in part by the Intramural Research Program of the NIH, National Institute on Aging. The HABC genome-wide association study was funded by NIA grant 1R01AG032098- 01A1 and genotyping services were provided by the Center for Inherited Disease Research (CIDR). CIDR is fully funded through a federal contract from the National Institutes of Health to The Johns Hopkins University, contract number HHSN268200782096C. We thank the Jackson Heart Study (JHS) participants and staff for their contributions to this work. The JHS is supported by contracts HHSN268201300046C, HHSN268201300047C, HHSN268201300048C, HHSN268201300049C, HHSN268201300050C from the National Heart, Lung, and Blood Institute and the National Institute on Minority Health and Health Disparities. JGW is supported by U54GM115428 from the National Institute of General Medical Sciences.

The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Supplementary material

Supplementary Information: File includes Supplementary Note, Supplementary Methods, Supplementary Figures and Supplementary Tables, as detailed below.

Supplementary Note includes individual study descriptions.

Supplementary Methods includes details of study level quality control procedures and eQTL analyses.

Supplementary Figures:

Supplementary Figure 1 - Quantile-quantile (QQ) and Manhattan plots for consortium-wide analyses, and the combined meta-analysis.

Supplementary Figure 2 - Region Plots for novel loci.

Supplementary Figure 3 - Forest Plots for novel loci.

Supplementary Figure 4 - Trait Transformation Sensitivity Analysis.

Supplementary Tables:

Supplementary Table 1 - Details of study specific genotyping platform, genotype calling procedure and software.

Supplementary Table 2 - Association results for all SNPs identified in single variant association discovery analyses (P<10 ^-4).

Supplementary Table 3 - Association results for SNPs identified in single variant association discovery analyses (P<10 ^-4), located in known lung function regions.

Supplementary Table 4 - Single variant association result for the seven novel signals, in smoking and ancestry subgroups.

Supplementary Table 5 - Single variant association result for rs1448044 and FVC in ever smokers and never smokers separately, and in all samples combined.

Supplementary Table 6 - Association results for all genes identified in discovery SKAT analyses (meta-analysis P<10 ^-4).

Supplementary Table 7 - Association results for all genes identified in discovery Weighted sum test (WST) test analyses (P<10 ^-4).

Supplementary Table 8 - Evidence for the role of novel variants identified in single variant association analyses as eQTLs.

Supplementary Table 9 - SIFT/Polyphen predictions for sentinel SNPs and proxies (r2>0.8).

Supplementary Table 10 - Protein and RNA expression results all implicated genes from the single variant association analyses.

Supplementary Table 11 - Look-up of association results for SNPs at 7 of the 12 loci which showed allele frequency differences between individuals from different regions in the UK.

Supplementary Table 12 - All traits results for the seven novel lung function loci.

Supplementary Table 13 - Genomic Inflation Factors: consortium and meta-analysis level.