Shared genetic architecture of hernias: A genome-wide association study with multivariable meta-analysis of multiple hernia phenotypes

Abdominal hernias are common and characterised by the abnormal protrusion of a viscus through the wall of the abdominal cavity. The global incidence is 18.5 million annually and there are limited non-surgical treatments. To improve understanding of common hernia aetiopathology, we performed a six-stage genome-wide association study (GWAS) of 62,637 UK Biobank participants with either single or multiple hernia phenotypes including inguinal, femoral, umbilical and hiatus hernia. Additionally, we performed multivariable meta-analysis with metaUSAT, to allow integration of summary data across traits to generate combined effect estimates. On individual hernia analysis, we identified 3404 variants across 38 genome-wide significant (p < 5×10−8) loci of which 11 are previously unreported. Robust evidence for five shared susceptibility loci was discovered: ZC3H11B, EFEMP1, MHC region, WT1 and CALD1. Combined hernia phenotype analyses with additional multivariable meta-analysis of summary statistics in metaUSAT revealed 28 independent (seven previously unreported) shared susceptibility loci. These clustered in functional categories related to connective tissue and elastic fibre homeostasis. Weighted genetic risk scores also correlated with disease severity suggesting a phenotypic-genotypic severity correlation, an important finding to inform future personalised therapeutic approaches to hernia.


Introduction
A hernia is the abnormal protrusion of a viscus through the wall of the anatomic cavity in which it is normally enclosed. Abdominal wall hernia (AWH) represent the majority of hernia phenotypes, with the lifetime risk for abdominal hernia being 27% in men and 3% in women [1]. Globally, at least 20 million AWHs are repaired every year, and there is an associated annual mortality of 59,800 deaths [2,3]. Surgery is the definitive treatment for symptomatic AWH. However, management is often challenging with significant risk of complications including chronic pain, seroma, haematoma, infection and failure of surgical repair [4,5]. In the case of femoral hernia, diagnostic difficulties lead to up to 40% of cases presenting as bowel strangulation or obstruction requiring emergency repair, which is associated with high mortality [6,7]. Therefore, there is a need to improve understanding of hernia aetiopathology, to guide new therapeutic avenues and improve patient outcomes. Indeed, patients with a family history carry an eight-fold risk of groin hernia and are more likely to suffer from contralateral or recurrent inguinal hernia as well as other hernia pathology including femoral, umbilical, incisional and epigastric hernia [8].
There is evidence for a genetic predisposition to AWH. Groin hernias have previously been shown to cluster in families [9], while the characteristic presence of hernia in several connective tissue disorders including Marfan's, Ehlers Danlos and Cutis laxa suggests an underlying genetic basis relating to impaired homeostasis of the extracellular matrix (ECM) [10][11][12]. Previously, Jorgenson et al. have identified four susceptibility loci for inguinal hernia alone (WT1, EFEMP1, EBF2 and ADAMTS6), each of which may result in aberrant elastic tissue homeostasis mediated via disordered expression of matrix metalloproteinases (MMPs) [13]. A further trans-ethnic GWAS meta-analysis of inguinal hernia identified five further loci including TGFB2, HMCN2 and CDKN3 [14]. Wei et al. attempted to characterise the polygenetic architecture of hernia using individual GWAS analysis of patients with either inguinal, femoral, umbilical or ventral hernia, identifying 57 loci, highlighting AIG1 and CALD1 as candidate genes for shared hernia susceptibility [15]. Interestingly, to our knowledge, GWAS of hiatus hernia has previously not been reported.
Here, we perform a comprehensive analysis to further characterise the shared genetic architecture between four hernia phenotypes including inguinal, femoral, umbilical and hiatus hernia, utilising both individual and combined genome-wide association studies. We further utilised multivariable meta-analysis in metaUSAT [16] a data-adaptive method, robust to the association structure of correlated traits, to perform a unified association test for each SNP across several trait summary statistics.

Results
Following quality control, a total of 62,637 individuals in the UK Biobank possessed a diagnostic and/or operative code for at least one of the hernia subtypes studied. Participants were divided into three hernia cohorts (individual hernia cohorts, overlap hernia cohort, and umbrella hernia cohort) as shown in Fig 1. Each was matched 1:5 to non-hernia controls in UK Biobank based on age (+/-5 years), sex and genotyping platform, while ensuring that control cohorts for the four individual hernia analyses contained completely distinct individuals. Sex distributions for all hernia cohorts are shown in S1 Table.

Individual hernia cohorts
These were four cohorts of participants who had diagnostic and/or operative coding for only one of the four hernia phenotypes. In these analyses, 23,007 individuals had diagnostic or operative codes for inguinal hernia, 1,578 for femoral hernia, 7,432 for umbilical hernia and 36,138 for hiatus hernia. The final individual hernia cohorts were then defined by removing participants with multiple hernia phenotypes (4,216 from the inguinal hernia cohort, 605 from the femoral hernia cohort, 2,076 from the umbilical hernia cohort and 3,841 from the hiatus hernia cohort), to create a phenotypically 'clean' cohort for each hernia subtype.
The final Individual hernia cohorts therefore consisted of the following individuals.

Overlap hernia cohort
The overlap hernia cohort consisted of participants with diagnostic or operative codes for two or more hernia subtypes. There were 5,219 cases which were matched to 26,095 non-hernia controls (total cohort 31,314 individuals).

Umbrella hernia cohort
The umbrella hernia cohort involved all participants who had diagnostic and/or operative codes for any hernia subtype, including those with single or multiple hernia subtypes. The umbrella hernia cohort consisted of 62,637 cases who were matched 313,185 non-hernia controls (total cohort 375,822 individuals). The analytic workflow representing the three analyses implemented to characterise shared genetic underpinnings of AWH are depicted in Fig 2.

Individual hernia cohort analyses
Initially, four separate GWAS analyses were undertaken in order to identify genetic risk loci for inguinal, femoral, umbilical or hiatus hernia in participants affected only by a single hernia subtype (termed 'Individual hernia cohorts'). We cumulatively discovered genomewide significant associations at 3,404 variants across 38 loci (52 independent signals), with Participants with more than one hernia phenotype coding were excluded (dark shading). B: Overlap hernia cohort: 5,219 participants possessed coding for at least 2 hernia phenotypes and were included in the overlap hernia cohort and matched with 26,095 non-hernia controls. All cases with single phenotype coding were excluded (dark shading). C: Umbrella hernia cohort, 62,637 participants had diagnostic or operative coding for any hernia type including those with single or multiple hernia and were matched to 313,185 non-hernia controls.
https://doi.org/10.1371/journal.pone.0272261.g001 24 susceptibility loci for inguinal hernia, one locus for femoral hernia, five loci for umbilical hernia, and eight loci for hiatus hernia (S2-S5 Tables and S1 Fig). Results relating to in silico annotation and candidate gene mapping of individual hernia loci are given in S6-S13 Tables).

Overlap hernia cohort analysis
We performed a further GWAS across participants with diagnostic or operative codes for at least two hernia subtypes (overlap hernia cohort; Fig 1). Significant associations at four loci (six independent signals) were revealed to confer shared susceptibility to multiple individual hernia phenotypes (Table 2 and Figs 3 and 4). The strongest association was observed at 2p16.1 (EFEMP1), closely followed by 1q41 (ZC3H11B) and 11p13 (WT1) which were all identified as shared susceptibility loci on analysis of the individual hernia cohorts. The fourth locus identified was 6q24.2 (AIG1) (rs4896643, p = 3.6×10 −8 , OR = 1.12) also identified in the inguinal individual hernia cohort (p = 7.8×10 −13 , OR = 1.08).

Umbrella hernia cohort analysis
A sixth GWAS of all UK Biobank participants affected by hernia, single or multiple ('umbrella hernia cohort') demonstrated 19 genome-wide significant loci representing 25 independent signals. Eleven loci were previously identified in the individual or overlap cohorts (

In silico annotation of overlap and umbrella cohort analyses
We used FUMA SNP2GENE [17] to annotate the overlap and umbrella hernia cohorts. In the overlap hernia cohort, 187 genome-wide significant candidate SNPs were identified by FUMA to be in LD (r 2 > 0.6) with the lead variant at each of the four loci. No exonic variants were identified, however, six intronic / intergenic variants had predicted deleterious effects and were in high LD with the index variant at each overlap hernia loci, including three at locus 1q41 (ZC3H11B) and two at locus 2p16.1 (EFEMP1) (S14 Table).
Analysis of the umbrella hernia cohort yielded 877 genome-wide significant candidate SNPs in LD with the lead SNP at the 19 loci. Thirty-eight intergenic or intronic variants were predicted to be functional (S15 Table), and 18 high LD exonic variants were discovered (S16 Table), 15 of which were at the MHC locus (6p22.2). Of these, four variants resulted in substitutions predicted to have damaging (PolyPhen) and deleterious (SIFT) consequences on BTN2A1. Of the non-MHC exonic variants, rs17855988 results in a pGly581Arg substitution in ELN that is predicted by SIFT [18] with low confidence to have a deleterious consequence on elastin function (S16 Table).

Multivariable meta-analysis of individual hernia phenotypes
We additionally performed multivariable meta-analysis of the four individual hernia traits in metaUSAT [16] across a total of 57, 418 individual hernia cases and 287,090 matched controls in UK Biobank. metaUSAT enables joint analysis of summary statistics from existing GWAS such that statistical power is augmented more so than for multiple univariable analyses alone. Designed to be robust to the association structure of correlated traits, metaUSAT may provide further insight into a shared genetic architecture for multiple hernia phenotypes.
Of note, tissue expression analysis in MAGMA [19] revealed Adipose Visceral Omentum to be most enriched whilst Adipose Subcutaneous tissue to be fourth most enriched (p = 1.11×10 −3 ). GTEx v8.0 30 general tissue types analysis confirmed this strong enrichment for Adipose tissue (p = 6.31×10 −4 , most enriched) (S7 Fig). Genetic risk score. We also implemented genetic risk score methodology in order to explore a hypothetical correlation between phenotypic severity and genotypic burden. A weighted genetic risk score for surgically managed cases versus non-surgically managed cases for both individual and combined hernia analyses was constructed from the lead independent variants from association analyses. All hernia patients who had undergone surgery had a higher wGRS compared to non-surgically managed hernia patients (Table 5).

Genetic correlations of individual hernia phenotypes
Estimated genetic correlation between individual hernia subtypes was evaluated with LDSC [22] using GWAS summary statistics from individual hernia GWAS analyses. Genetic Table 4. 24 genome-wide significant loci discovered in the metaUSAT multivariable meta-analysis of inguinal, femoral, umbilical, hiatus hernia in 57,418 cases and 287,090 controls in UK Biobank. Statistically significant signals from the metaUSAT analysis are shown in the left-hand column. The central column shows the association p-values for those SNPs in the six original GWAS analyses, with the direction of effect indicated by a + or-sign. Candidate genes are those selected from the prioritised genes (using the four mapping strategies described previously for all GWASdiscovered loci) or genes in proximity as identified within the UCSC genome browser.  correlation was greatest between inguinal and femoral hernia subtypes with r g 0.60, p = 0.011, followed by umbilical and hiatus hernia which yielded r g of 0.21 with p = 0.0041 (Table 6). Umbilical-inguinal hernia also showed positive correlation with r g 0.19, p = 0.029. Evidence of genetic correlations was not observed between inguinal-hiatus or femoral-umbilical hernia phenotypes.

Discussion
We performed a six-stage genome-wide association study (GWAS) of multiple hernia phenotypes with additional multivariable meta-analysis using metaUSAT in order to characterise the shared genetic underpinnings of common hernia phenotypes. We identified 38 susceptibility loci (11 previously unreported) associated with inguinal, femoral, umbilical or hiatus hernia among an umbrella cohort of 62,637 individuals derived from UK Biobank. Five biologically relevant loci were discovered on individual hernia analyses to confer shared susceptibility to multiple hernia phenotypes including 1q41 (ZC3H11B), 2p16.  further proof of principle with regards to the shared genetic underpinnings of a common hernia phenotype. The four remaining susceptibility loci for hiatus hernia have previously not been reported.
It has been postulated that dysregulation of elastic tissue biology mediated via matrix metalloproteinases (MMPs) is central to the pathophysiology of hernia development. Jorgenson and colleagues previously identified 4 inguinal hernia susceptibility loci purported to result in reduced MMP activity: WT1, EFEMP1, EBF2 and ADAMTS6 [13]. Recently, Wei et al. replicated these associations and further implicated AIG1 and CALD1, which were identified as biologically relevant genes in their individual GWAS's of hernia phenotypes [15]. We have extended these results to focus on the shared biology of abdominal wall hernias. Our study provides further evidence that these loci each impart susceptibility to multiple hernia phenotypes, supported by the observation that these demonstrated some of the strongest associations in the combined cohorts, while also being prioritised in the multivariable meta-analysis.

Genes of interest
The 1q41 locus (ZC3H11B) was strongly associated across three individual hernia phenotypes, the combined hernia cohorts, and importantly was also prioritised in metaUSAT meta-analysis. ZC3H11B, a zinc finger CCH domain-containing protein, was previously identified by Wei et al. [15] in association with inguinal, femoral, umbilical and ventral hernia. ZC3H11B has also been associated with myopia endophenotypes, including axial length, refractive error, and corneal astigmatism [23]. It is thought that accelerated connective tissue remodelling of the posterior sclera leads to axial elongation, a key feature of myopia [24], thereby implicating ZC3H11B in a number of suspected connective tissue diseases. Intriguingly, at least two Marfan-like syndromes have been described with co-existing myopia and hernia [25,26].
Locus 2p16.1 (EFEMP1) imparted susceptibility to inguinal and hiatus hernia phenotypes, was significantly associated in both overlap and umbrella cohort analyses, and was also identified in metaUSAT meta-analysis. EFEMP1 encodes fibulin-3, a secreted extracellular matrix

PLOS ONE
Shared genetic architecture of hernias: GWAS & multivariable meta-analysis of multiple hernia phenotypes glycoprotein, which has been shown to downregulate matrix metalloproteinases (MMPs) 2 and 3, whilst simultaneously upregulating tissue inhibitor of metalloproteinase-3 [27]. As well as collagen, fibulin-3 binds tropoelastin [28], the monomeric unit of elastin fibres. EFEMP1 knockout mice show depleted elastic fibres within fascia and invariably develop inguinal hernia, adding further strength to the evidence for its importance in AWH pathophysiology [29]. True pleiotropy is further substantiated by the finding that EFEMP1 has previously been identified by our group as a candidate gene conferring susceptibility to carpal tunnel syndrome [30] and varicose veins [31], disorders also thought to be underpinned by opposing impairments in extracellular matrix homeostasis. Additionally, EFEMP1 has recently been implicated in conferring susceptibility to pelvic organ prolapse [32] and intriguingly is also associated with anthropometric measures of height [33] and abdominal circumference [27], which have also been associated with ZC3H11B [34].
Additionally, two signals in proximity to ADAMTS16 were discovered as genome-wide significant in metaUSAT analysis and in the umbrella cohort GWAS analysis. The ADAMTS family are a group of metalloendopeptidases, related to MMPs, serving to synthesise collagen from procollagen [27]. Variants in ADAMTS16 have been associated with urinary incontinence [35], a manifestation of pelvic floor dysfunction, which have been shown independently to lead to a higher prevalence of hiatus and inguinal hernia [36]. Several mutations have been described in the other 18 ADAMTS superfamily genes, which result in distinct human genetic disorders [37]. For example, mutations in ADAMTS2 are responsible for dermatosparactic type Ehlers-Danlos Syndrome (type VIIC) [38], typified by extreme skin fragility, joint laxity, and umbilical hernia. ADAMTS4 shows significant aggrecanase activity and is implicated in articular cartilage degradation and arthritis [39], and ADAMTS4 mRNA and protein have been found to be highly expressed in herniated lumbar intervertebral discs [40].

PLOS ONE
metaUSAT met-analysis yielded a further candidate gene at 1q41 (TGFB2) which was also the most statistically significant locus in the umbrella hernia analysis. TGFB2 encodes the protein Transforming Growth Factor β2 (TGF β2) which is observed to be upregulated in Marfan syndrome, Loeys-Dietz syndrome, and cutis laxa, which are associated with aneurysmal changes with histological features including smooth muscle cell apoptosis [41]. Furthermore, TGFB2 haploinsufficiency pathologically activates the TGF-β signalling pathway [41], leading to Loeys-Dietz syndrome type 4 [42], which is characterised by arterial vasculopathy (arterial aneurysms, dissection and tortuosity), and other widespread connective tissue pathology, including hernia [43]. Like ZC3H11B, TGFB2 has previously been implicated in ophthalmic pathology including glaucoma endophenotypes, with roles in intraocular pressure [44], central corneal thickness [45], as well as FEV1/FVC ratio [46] and severe chronic obstructive pulmonary diseaseO [47], of which the latter has also been suggested as an independent risk factor for hernia pathology and severity [48,49].
Among the further candidates discovered on multivariable meta-analysis, CEP72 at 5p15.33 encodes a centriolar satellite protein necessary for regulating microtubule-organising activity and centrosome integrity [50]. Using comparative genomic hybridisation, Choi et al. discovered copy number increases at 5p15.33 in patients with ruptured intracranial aneurysms [51]. The CEP72 region has also been implicated in a genome-wide meta-analysis of Barrett's oesophagus and oesophageal adenocarcinoma [52], for which hiatus hernia is a major risk factor [53]. Indeed, the size of a hiatus hernia is significantly associated with progression of Barrett's oesophagus to high-grade dysplasia or malignancy [54]. To this end, a tangible and biologically plausible contributor to shared hernia risk has been identified through multivariate meta-analysis.
GDF7 was discovered to associate with hernia in the umbrella and metaUSAT analyses, with the lead variant rs3072 demonstrating strong functionality as a robust eQTL for GDF7 in GTEx aorta tissue (P eQTL = 5.4×10 −9 ). GDF7 encodes BMP12, part of the bone morphogenetic protein pathway, and is heavily implicated in Barrett's oesophagus [55] with several studies identifying polymorphisms in the TBX-GDF7 genomic region [56,57]. GDF7 has been identified through GWAS to associate with eight traits, three of which are characterised by connective and elastic tissue dysfunction: pelvic organ prolapse [32], abdominal aortic aneurysm [58], and diverticular disease [59]. The T allele of rs7255 (which is in high LD with lead SNP rs3072) was also found to confer risk of Barrett's oesophagus in the GWAS by Gharahkhani et al. [52].
The umbrella analysis further revealed association at 3q22.2 locus which was not identified in any other analyses presented. Interestingly, de novo deletions at 3q22.1 result in a syndromic presentation of bilateral inguinal hernia [60] and an interstitial deletion of 3q23 has been described to result in BPES syndrome, characterised by diaphragmatic hernia [61]. This region on the long arm of chromosome 3 may therefore be of considerable interest in multiple hernia pathobiology.

Genetic risk scoring
Our simple weighted genetic risk score correlated with disease severity, with patients undergoing surgery having a higher genetic burden than those managed non-surgically across all individual hernia subtypes and overlap analyses. These data provide an important proof-ofprinciple of genetic risk scoring in personalising risk in this highly prevalent disease. Further work to validate the risk score in an independent cohort is required.

Genetic correlations of individual hernia phenotypes
We found strong positive genetic correlations between femoral and inguinal hernia, with r g 0.60, further supporting the notion of shared genetic architecture. It is possible however, that due to the small femoral hernia sample size, these findings may have been spurious. However, robust correlations were also observed between umbilical-hiatus and umbilical-inguinal hernia, of which the former was most statistically significant (r g 0.21, P = 0.0041). Given the large sample sizes of the inguinal and hiatus hernia cohorts, it is interesting that a correlation was not observed between these phenotypes. This may reflect the fact that hiatus hernia occurs through the diaphragm, which is derived embryologically from the septum transversum and not somitic and lateral plate mesoderm [62].

Limitations
Our primary aim was to identify shared susceptibility across multiple hernia phenotypes which is made possible in a large Biobank-scaled cohort. The lack of a replication cohort for these results is a clear limitation, however, this was somewhat mitigated by the use of stringent quality control and case definitions as well as the implementation of four distinct analytic strategies. Secondly, the use of unselected biobank data inevitably results in imbalance between the phenotypes. This means that hiatus and inguinal hernia, which are substantially more common in the UK population, were overrepresented in our dataset as they accounted for approximately 90% of the total cohort and were therefore more powered in the joint analyses. A larger cohort with greater balance across all four hernia phenotypes may prove useful in further defining shared genetic susceptibility loci as well as uncovering new associations. Finally, we acknowledge the limitations of restricting the GWAS analyses to a cohort of white British ancestry, and that the genetic loci for hernia susceptibility identified in this study may not be applicable to individuals of other ancestries.

Conclusions
In conclusion, the distinct analytic approaches to examine the shared genetic architecture of the four hernia subtypes allowed us to discover new insights into the biology of abdominal wall hernias. We discovered new genetic associations that were not found on traditional single-trait association analyses. By segregating the four hernia cohorts in UK Biobank to avoid overlap, we can have confidence in the validity of several loci that were discovered across multiple hernia phenotypes. This is the case for the twelve loci that demonstrated the greatest degree of overlap across the different analyses, and the resulting clustering of many of these loci across functionally related ontologies. Furthermore, the enrichment of biological pathways previously implicated in hernia pathobiology provides further compelling evidence to support the veracity of these loci, and for a shared genetic susceptibility to hernia.

Overview
We performed four individual genome-wide association studies (GWAS) of hernia subtypes (inguinal, femoral, umbilical and hiatus hernia) of 488,377 UK participants, aged 40-69 years at the time of recruitment, who provided written consent to be prospectively enrolled into the UK Biobank multicentre cohort from 2006-2010. The full characteristics of the UK Biobank cohort are described in full elsewhere [63,64].

Ethics statement
UK Biobank obtained ethical approval from the North West Multi-Centre Research Ethics Committee (MREC) (11/NW/0382). This study was conducted under UK Biobank project ID 22572.
We looked for evidence of shared genetic underpinnings between the four distinct hernia phenotypes with a further GWAS analyses of participants with multiple hernia phenotypes, as well as a sixth combined cohort of participants with individual or multiple hernia phenotypes. Additionally, we undertook multivariable meta-analysis to characterise shared susceptibility loci, by aggregating information across the four correlated traits.

Cohort definitions
The UK Biobank population was divided into three hernia cohorts as described below. Each was matched 1:5 to non-hernia controls.
i. Individual hernia cohort. This comprised of participants who had diagnostic and/or operative codes for just one of four hernia phenotypes studied. That is, either inguinal, femoral, umbilical or hiatus hernia. Participants having more than one hernia type were excluded.
ii. Overlap hernia cohort. This cohort consisted of participants with at least two of the four hernia phenotypes studied. Participants affected by a single hernia phenotype were excluded.
iii. Umbrella hernia cohort. This encompassed all participants with any of the hernia phenotypes studied. As such, this cohort was comprised of cohorts (i) plus (ii).
The full list of diagnostic and operative codes used are shown in S20 Table. Genotyping UK Biobank participants were genotyped sequentially, initially with the Affymetrix BiLeve Axiom array (805,426 directly genotyped variants) and Affymetrix UK Biobank Axiom arrays (825,927 genotyped variants), which share 95% marker content. The present study is based on the third release of the UK Biobank cohort (July 2017), which contained the complete set of genotypes for the 488,377 participants.

Quality control
Quality control (QC) was performed using PLINK v1.919 and R v3.3.1. The full details of quality control (QC) has been previously described [30]. Briefly, SNPs with low call rates (<98%) were initially excluded. Samples with heterozygosity >3.5 SD from the mean, discordant sex information, or who were not of white British ancestry (ethnic outliers) were excluded. A linear mixed model implemented in BOLT-LMM enabled the inclusion of related participants. SNP-level QC was further performed based on deviations from Hardy-Weinberg equilibrium (p < 10 −4 ), minor allele frequency (MAF) <0.01, as well as on visual inspection of autosomal heterozygosity against call rate.

Imputation
Phasing and imputation of UK Biobank was performed centrally using a 1000 Genomes Consortium Phase 3 reference panel in SHAPEIT3, and has been detailed elsewhere [65].

Association analyses in BOLT-LMM
In the UK Biobank, GWAS was performed across 8,944,547 imputed SNPs (547,011 directly genotyped (MAF � 0.01) and 8,397,536 imputed SNPs (MAF � 0.01, INFO score � 0.90) using a linear mixed non-infinitesimal model implemented in BOLT-LMM v2.323 [66]. The reference human genome assembly used was GRCh37 (hg19) and linkage disequilibrium scores were obtained from participants of European-ancestry extracted from the BOLT-LMM package. Covariates included in the model were genetic sex and genotyping platform. Association testing was implemented by linear regression assuming an additive allelic effect using imputed allelic dosages. Conditional regression analysis was performed in BOLT-LMM for the top signal at each significant locus (except the MHC region), and repeated until no further residual significant signals remained.

Functional annotation of SNPs
Annotation of associated SNPs was performed in FUMA SNP2GENE v1.3.6 [17], using summary statistics from the UK Biobank discovery cohort and default settings. As such, genomic location and effect/non-effect allele were used to collate functional annotation data from established genetic annotation databases, including ANNOVAR [67], RegulomeDB [68], CADD [69], and 15-core chromatin state categories [70]. Exonic SNPs were investigated further using gnomAD and Ensembl genome browsers to uncover putative functionality [71].

Multivariable meta-analysis in metaUSAT
The metaUSAT [16] multivariable method was used as an auxiliary meta-analysis method to further characterise potential regions of shared hernia susceptibility between the four individual hernia traits. metaUSAT performs a unified association test for each SNP, using the estimated correlation matrix to test association, across several trait summary statistics. metaUSAT is data-adaptive and was established to be robust to the association structure of correlated traits (less affected by the true (unknown) association structure) and is not dependent on individual-level data [16] Unlike other multi-trait meta-analysis approaches, metaU-SAT does not assume homogeneity of effects across traits. metaUSAT outputs an approximate asymptomatic P-value for the meta-analysis association and has been shown to maintain a low type I error in simulation experiments [16]. The metaUSAT meta-analysis was performed across the four individual hernia cohorts (total 57,418 cases and 287,090 controls) and 8,896,286 SNPs. The genome-wide significant threshold for the metaUSAT association was set a p < 5×10 −8 .

Gene set, tissue-specific, and pathway enrichment analysis
Gene-set analysis were performed in MAGMA v1.07 [19] across 15,496 gene sets obtained from MSigDB v8.0 [21] with p < 3.23×10 −6 deemed significant. Enrichment of the overlap between GWAS variants and those reported in previous GWAS within the NIH GWAS Catalog were also examined [73], with enrichment P-values for the proportion of overlap in the genes determined. Tissue-specific analysis was also performed in MAGMA v1.07 to assess gene expression from 30 tissue types in GTEx v8 [17]. The gene set and tissue expression analyses described above were then repeated within FUMA GENE2FUNC v1.3.5d [17], to specifically examine the functionality of genes prioritised directly from the four candidate gene mapping approaches. Gene set enrichment analyses of the gene sets within MSigDB v8.0 [21] were tested, and gene property and tissue enrichment analyses within GTEx [17] consortium tissue was also performed distinctly for the prioritised hernia associated genes.
Using eXploring Genomic Relations [20] (XGR) software, pathway enrichment analysis of the prioritised genes was performed to highlight canonical pathways that were enriched. A hypergeometric distribution test was performed and adjusted FDR < 0.05 used to highlight prioritised gene sets.

Genetic risk score
Weighted genetic risk scores (wGRS), based on the lead independent variants at each genomewide significant locus, were constructed for each of the six GWAS. For each of the Individual cohorts, overlap and umbrella cohort summary statistics, wGRS were compared between all cases managed surgically and those that were not surgically managed. Surgical cases were defined as those with OPCS (Office of Population Censuses and Surveys Classification of Interventions and Procedures) or self-reported operative codes.
The following formula was implemented: where i is the lead SNP at each genomic risk locus, n is the total number of lead SNPs in the GWAS, Wi is the weighting for each of the SNPs (the natural logarithm of the odds ratio for each effect allele), and Xi is the number of effect alleles each individual possesses for each SNP. Each individual's risk allele was used to compute a SNP dosage (QCTOOL v2). wGRS calculations and statistical tests between the different subgroups was performed in R v3.3.

Genetic correlations of individual hernia phenotypes
To investigate potential genetic correlation between the four individual hernia subtypes, linkage disequilibrium score regression (LDSC) was performed between the four individual hernia GWAS analyses. This method evaluates genetic correlation between traits based on a fitted linear model of Z-scores calculated using GWAS summary statistics [22]. For polygenic traits with shared genetic architecture, SNPs with high LD would, on average, be expected to have higher Z-scores than those with low LD. As this was an exploratory analysis, P-values were not corrected for multiple testing.
Supporting information S1  Table. Genes mapped to the inguinal hernia-associated loci using the four mapping strategies. 101 unique genes (169 total) were mapped to 21 of 24 inguinal hernia susceptibility loci by one or more gene mapping strategies. 53 genes were mapped via positional mapping, 42 genes were mapped via eQTL mapping, 64 genes were mapped using MAGMA and 3 genes were mapped using summary-based mendelian randomisation. Overlap between the four different mapping strategies is shown (and highlighted in pink). (PDF) S11 Table. Genome-wide gene-based association analysis for umbilical hernia in MAGMA.
Three protein-coding genes met the threshold for genome-wide significance (p<2.64x10-6, 0.05/18,916) in this analysis. The one gene that lays within the realms of the genome-wide significant susceptibility loci and are highlighted in red. (PDF) S12 Table. Genome-wide gene-based association analysis for hiatus hernia in MAGMA. 26 protein-coding genes met the threshold for genome-wide significance (p<2.64x10-6, 0.05/ 18,918) in this analysis. 11 of the 26 genes lay within the realms of the FUMA-defined genome-wide significant susceptibility loci and are highlighted in red. (PDF) S13 Table. Genes mapped to the hiatus hernia-associated loci using the four mapping strategies. 15 unique genes (20 total) were mapped to 5 of 8 hiatus hernia susceptibility loci by one or more gene mapping strategies. 5 genes were mapped via positional mapping, 4 genes were mapped via eQTL mapping, 11 genes were mapped using MAGMA and no genes were mapped using summary-based mendelian randomisation. Overlap between the four different mapping strategies is shown (and highlighted in pink). (PDF) S14 Table. Predicted functional intronic and intergenic variants associated with overlap hernia. Six genome-wide significant intronic and intergenic variants predicted to be deleterious according to a CADD � 12.37 and associated with overlap hernia as identified by FUMA SNP2GENE. (PDF) S15