Fine-mapping and cell-specific enrichment at corneal resistance factor loci prioritize candidate causal regulatory variants


Corneal resistance factor (CRF) is altered during corneal diseases progression. Genome-wide-association studies (GWAS) indicated potential CRF and disease genetics overlap. Here, we characterise 135 CRF loci following GWAS in 76029 UK Biobank participants. Enrichment of extra-cellular matrix gene-sets, genetic correlation with corneal thickness (70% (SE = 5%)), reported keratoconus risk variants at 13 loci, all support relevance to corneal stroma biology. Fine-mapping identifies a subset of 55 highly likely causal variants, 91% of which are non-coding. Genomic features enrichments, using all associated variants, also indicate prominent regulatory causal role. We newly established open chromatin landscapes in two widely-used human cornea immortalised cell lines using ATAC-seq. Variants associated with CRF were significantly enriched in regulatory regions from the corneal stroma-derived cell line and enrichment increases to over 5 fold for variants prioritised by fine-mapping-including at GAS7, SMAD3 and COL6A1 loci. Our analysis generates many hypotheses for future functional validation of aetiological mechanisms.


Corneal properties are clinically important1,2 for: disease prognosis, electing refractive eye surgery, and evaluating intraocular pressure (IOP), a critical factor in glaucoma risk assessment. Genome-wide association studies (GWAS) for measurements acquired in heathy individuals—corneal thickness3,4,5,6,7,8, curvature9,10, resistance factor and hysteresis11, endothelial cells shape, and density7—showed potential to elucidate molecular events shaping these traits and associated disease risks. Statistical methods have emerged to narrow-down the genetic variants most likely to cause observed associations, with high throughput12. Those can provide single-variant resolution, unlocking mechanism hypothesis, even if only for a small fraction of GWAS loci13. To explore this avenue, we leveraged the availability in >75-,000 UK Biobank (UKBB) participants of corneal resistance factor (CRF) measures.

CRF is an empirical measure of the corneal mechanical response to applied force. With cornea hysteresis (CH), it is increasingly computed by devices implementing bidirectional applanation tonometry to measure IOP. The corneal response to deformation is used to derive a cornea compensated IOP measure (IOPcc), which aims to remove the recognised influence of corneal thickness and viscoelasticity on Goldmann-correlated IOP measure (IOPg). CRF, by design, predicts this influence14, correlates more strongly with central cornea thickness (CCT) than CH2 and not or little with IOPcc1,14. CRF encompassing a CCT component—a thinner cornea should be easiest to deform—was apparent in initial CRF genetic investigations7,11. Three well-established CCT loci, FOXO1, ZNF469, and COL6A1 are among the five identified genome-wide significant CRF loci11. CRF and CCT are concomitantly altered in Marfan syndrome patients with ectopia lentis15 or during keratoconus progression1, two conditions with corneal stroma abnomalies16,17. It was proposed11 that CRF could, like CCT5,6, offer genetic clues for keratoconus. Alterations during disease progression uniquely captured by CRF (after adjusting for CCT)18,19 make it a parameter of clinical interest in its own right. The strong CRF genetic associations at ANAPC1 and TCF47,11, major loci for respectively endothelial cell density7 and Fuchs endothelial corneal dystrophy (FECD)20, additionally emphasize a relevance to corneal endothelium health not prominent in the CCT GWAS results.

The cornea has three cellular layers: a stratified epithelium, a collagen-rich stroma which in human accounts for 90% of the cornea and a monolayer of cells which insures essential ion exchanges function, the endothelium21. The extracellular matrix (ECM) stromal components are produced by specialised mesenchymal resident cells, the keratocytes. Fittingly, many genes implicated by CCT GWAS belonged to pathways such as collagenous fibril production and regulation, or hallmarks of the mesenchymal state6. Stromal ECM composition being a strong determinant of cornea strength and viscoelasticity, keratocyte should also be a major target cell type for CRF-GWAS signals. Here, we identified regions of open chromatin, indicators of cis-regulatory activity, in immortalized corneal epithelial and keratocyte cell lines, using the assay for transposase accessible chromatin followed by sequencing (ATAC-seq). Overlaying these and publicly available data with the UKBB CRF-GWAS study, a homogeneous large dataset well suited for fine-mapping, identifies causal regulatory variants and regions active in fibroblastic cells as strong functional candidates. The keratocyte cell line, hTK, represents a suitable system to undertake further functional characterisation at many of those loci.


CRF-GWAS in UK-Biobank

Genome-wide association analysis, using N = 76,029 UKBB participants of white-British ancestry, yielded 135 loci harbouring variants significantly associated (P value < 5 × 10−8) with corneal resistance factor (Fig. 1, Supplementary Data 1). Those include 251 potentially independent variants based heuristically on pairwise linkage disequilibrium measure r2 less than 0.1 (Supplementary Data 1), suggesting multiple association signals at several loci. Variant effects were consistent between this analysis and a GWAS performed in a smaller set of European non-British participants (N = 10,130) (Supplementary Fig. 1, Supplementary Data 2), with the exception of strong but opposite-direction effects for the lead, low-frequency variant, rs112108520 at the ETS1 locus. Forty eight out of the 135 loci map CCT loci, identified in independent and smaller studies6,7,8, including thirteen which have been associated with keratoconus risk5,6,8,22 (Supplementary Data 1). Using the International Genetics Glaucoma Consortium (IGGC)6 CCT summary statistics for participants of European ancestry and linkage disequilibrium (LD) score regression method23, the genetic correlation between CRF and CCT is 70% (SE = 5%). Colocalisation tests further support shared causal variants underlying these two traits at all but two of the 27 loci significantly associated with CCT in the IGGC European subset (Supplementary Data 3). Signals at FNDC3B (loci 31) and RXRA-COL5A1 (locus 68) were attributed to both CRF and CCT, but declared not to be the same. Both loci are complex with several independently associated SNPs6, which may contribute differently to both traits. Two CRF novel loci, ATP1B1-LINC00970 (locus 5) and SLC25A22 (locus 78), together with the previously reported TCF4 (locus 126), map to FECD risk loci24, with close match of association plots (Supplementary Fig. 2). The FECD risk alleles of representative SNPs at these three sites were all associated with decreasing corneal resistance, and their effects on FECD risk appear correlated with their effect on CRF (Supplementary Fig. 2).

Fig. 1: Manhattan plot of the corneal resistance factor GWAS in UK Biobank white-British participants (n = 76,029).

The genome-wide significant threshold (P value = 5 × 10−8) is displayed by the horizontal dotted black line. Only variants with P value < 0.001 are represented. Genomic risk loci overlapping known central cornea thickness and Fuchs corneal dystrophy loci are indicated in blue and orange, respectively. The test statistics inflation factor λgc, 1.147, is mostly due to polygenicity (ratio (LDScore intercept − 1)/(mean(chi2) − 1) = 14.7%; LDscore intercept = 1.057(SE = 0.007)).

More than half of the CRF loci (N = 72) have been reported as IOPg loci in analysis of a larger, fully overlapping UKBB sample25. In the white-British sample, the phenotypic correlation between CRF and IOPg was 59.2% and the genotypic correlation 73.44% (SE = 1.72%). In agreement with expectation, the phenotypic correlation with IOPcc was lower, 5.3%, and fewer CRF loci (N = 18) overlap with reported loci11. Interestingly, two of those, ATP1B1- LINC00970 and ANAPC1 have not been reported for IOPg, but for FECD24 and endothelial cell density, respectively7. This suggests an influence of corneal features on IOPcc, qualitatively different to that affecting IOPg.

Candidate CRF-GWAS target genes selected by the FUMA algorithm26 are listed Supplementary Data 4 alongside the selection criteria (physical location, chromatin interaction evidences, or/and eQTL data). Using this gene list, the top two biological pathways implicated by geneset enrichment test are GO_cc, go_proteinaceous_extracellular_matrix (enrichment P value 1.92 × 10−16) and go_extracellular_matrix (P val 4.17 × 10−15). Both enrichments were highly significant (multi-test corrected 5% threshold of 0.05/10673 = 4.68 × 10−6) and in line with expectation of a major role for the stroma in corneal resistance.

Refinement of CRF association signals using statistical methods

To avoid haplotype structure due to close relatedness, these analyses were performed on the subset of unrelated white-British participants, yielding a smaller but still substantial number of genome-wide significant loci, N = 115, including all those noted as overlapping FECD and keratoconus risk loci in the larger analysis (Supplementary Data 5). The subanalysis newly flags seven loci (association P value nearing genome-wide significance threshold), none of which previously known to affect cornea structure. Conditional and joint multiple-SNP (CoJo) analysis implemented in the program GCTA27 defined multiple causal signals at 22 loci yielding 149 independent associated lead variants (Supplementary Data 6). The stepwise selection of predictor variants implemented being suboptimal in regions with multiple signals in linkage disequilibrium, we also use the exhaustive search for joint combination of alleles method implemented in FINEMAP28. This suggested 186 independent causal signals, largely concordant with those defined by CoJo (Supplementary Data 7). Since the Bayesian method implemented in FINEMAP returns strength of evidence for a variant to contribute to each signal, these causal signals could be fine-mapped to sets of variants, with 95% probability. Sets were named after the variant with the highest posterior inclusion probability (PIP) (Supplementary Data 7); their size ranged from 1 (at loci 16, AC078954.2, and 92, IGF1) to 293 (at locus 40, CWC27). The full list of variants included in these 95% credible sets is given Supplementary Data 8, with PIP alongside the causal evidence support given as Bayes factor (BF).

Subset of highly likely causal variants

Fifty-five variants (Supplementary Data 9) with high PIP (>60%) and/or very strong evidence for being causal (log10(BF) > 3) were examined in greater detail. They comprised five coding variants (Supplementary Data 10), all missense with a high (>20) CADD score, a measure of the predicted deleterious consequence of the amino acid substitution29. All have been previously prioritized in relation to CCT7,8 or keratoconus11, apart from rs77583146, p.Gly165Arg in WNT10A. This latter is 146 bp away from another low-frequency missense variant, rs121908120, reported for CCT and keratoconus risk22, identified here as an independent functional variant candidate. All proteins implicated, WNT10A, ABCA6, GLT8D2, FBN2, and ADAMTS17, have reported function relevant to corneal development or maintenance (Supplementary Data 10).

The majority of prioritised candidate causal variants (50/55), representing 46 independent credible sets at 34 loci, were noncoding (Supplementary Data 9). Over half (in 28 credible sets across 26 loci) are significantly associated with gene transcription or alternative splicing levels in at least one tissue/cell type from public depositories (Supplementary Data 9). To explore whether some of those variants could be causally associated with corneal resistance by modulating gene expression, we applied colocalisation tests for CRF and cis-eQTL using available GTEx30 v7 full summary statistics. Colocalisation is often detected across several tissues for a given gene (Fig. 2), lending support that if observed it could also occur in the cornea—a tissue unrepresented in GTEx. The biological sample with the most evidence of colocalisations (N = 5) was “Cells-transformed fibroblasts”. Whilst more than one gene was implicated at several loci, a unique candidate target gene was supported at six of the investigated loci: DPF3 (at locus 101), SMAD3 (at locus 106), GAS7 (at locus 114), MSL1 (at locus 116), GLT8D1 (at locus Ext3), and TET2 (at locus Ext4). The pertinence of the function for all six genes in the corneal context is summarised Table 1.

Fig. 2: Colocalisation with GTEx v7 eQTL signals at prioritized corneal resistance factor loci.

Genes: candidate target genes based on GTEx eQTL look-up for prioritized candidate causal variants at CRF loci (grey box). Each dot denotes significant association signals for both CRF and GTEx v7 GWAS, colour-coded as independent (grey) or identical (blue if CRF increasing allele increases gene expression, red if it decreases it). The dot size is proportional to the probability of a colocalisation (pp4); pp4 > 0.75 denotes strong evidence in favour of colocalisation.

Table 1 Six noncoding highly likely causal variants for CRF associations colocalising with a single gene expression signal.

For the subset of 30 noncoding SNPs singled-out by a PIP > 0.6, we examined the potential for transcription factor (TF) binding site disruption (summarised in Supplementary Data 11 and presented in full Supplementary Data 12). Two TFs whose binding strengths were predicted to be significantly altered at the most sites (7 out of the 30 investigated sites) were REST and RXRA (Table 2). Chance expectations for such results were estimated to be 9.7% and 7.34%, for REST and RXRA, respectively (Supplementary Note 5). Notably, amongst the seven loci with a prioritised causal variant predicted to affect REST binding affinity, FNDC3, RXRA-COL5A1, and FOXO1 are keratoconus loci and ADAMTS8 and COL6A1, were recently suggested to be11.

Table 2 Prioritised CRF causal candidate variants with potential to disrupt a REST or RXRA binding site.

RXRA is encoded by a gene flanking one of the CRF-GWAS locus (RXRA-COL5A1). Three out of the seven highly likely causal variants predicted to disrupt RXRA binding are located within a RXRA ChIP peak in one of the tissues surveyed by the ENCODE project, including the variant intronic to FOXO1, which is also predicted to alter REST binding and also within an ENCODE REST ChIP peak (Table 2). In addition to RXRA, ten TFs for which binding sites could be disrupted in at least two highly likely causal polymorphic sites are encoded by candidate GWAS target genes (Supplementary Data 4 and 11). For one of those, TCF4, the prioritised variant predicted to alter its binding, rs192498625, locates in the vicinity of the TF encoding gene (intronic one TCF4 isoform, upstream other isoforms). Supporting a (self) regulatory role, among all the tissues examined by the Roadmap Epigenomics Consortium31, the histone modification H3K27ac, considered as a hallmark of active enhancers, is detected around this site, exquisitely in (H1 derived) mesenchymal stem cells32.

Of the prioritised noncoding eQTL variants linked to a unique target gene (Table 1), only rs9913911, predicted to modulate expression of GAS7, was examined (SNP and PIP > 0.6) for transcription factor binding disruption potential. Several factors could be affected (Supplementary Data 12), including the aryl hydrogen receptor class with perfect match between the underlying sequence (with reference A allele at rs9913911) and binding motifs.

Cornea cells-derived ATAC-seq datasets for enrichment analysis

Enrichment of CRF-GWAS variants in regulatory regions of the genome was apparent using annotations generated in a broad range of tissues by the ENCODE, GENCODE, or Roadmap Epigenomics projects33. Enrichment was the lowest for regulatory features derived from blood cells and fetal brain tissues and high (>2.5 fold), for those from lung, heart, fetal muscle, eye, skin, fetal thymus, fibroblasts (Supplementary Fig. 3). That from fetal muscle, which contains fibroblasts and myoblasts of mesenchymal stem cells origin, was the strongest, around 3.3 fold with association P value cut-off of 10−6. These enrichments support GWAS variants impacting prominently regulatory activity in connective tissues and fibroblasts in particular.

We then investigated the levels of open chromatin regions enrichment, using published and purposely generated ATAC-seq datasets, in order to examine that for keratocytes, the specialised fibroblasts that populated the corneal stroma. Available datasets selected (Supplementary Data 13) include those derived from: primary cornea epithelial cells (CEC), cranial neural crest cells (cornea endothelial cells and keratocytes having a neural crest origin), adult and neonate skin fibroblasts (DermFb and nDF, respectively), primary retinal pigmented epithelium (RPE), immortalised retinal pigmented epithelium cells (RPE_CellLine) and two, a priori negative, controls, lymphoblastoid and myelogenous leukemia cells. We performed ATAC-seq in the telomerase-immortalised human cornea cell lines hTK and hTCEpi, derived respectively from stromal34 and epithelial35 corneal tissues. The greatest enrichment for the CRF-GWAS results, using association P value threshold of 10−8, was 1.82 and significant (p = 4.28 × 10−8) in open chromatin regions (OCR) of the keratocyte cell line, hTK, after those present in the hTCEpi cell line have been subtracted out, here named hTK-specific OCR (Table 3). Of note, the enrichment is sensitive to the ATAC-seq peak calling and subtraction method adopted (Supplementary Note 3). When considering hTK regulatory features without filtering out those present in hTCEpi, enrichment is decreased to a level similar to that of adult or neonate dermal fibroblasts, all of which were significant. Enrichments using open chromatin regions derived from the two blood derived cells lines were not significantly different than those expected by chance, nor were those derived from the epithelial cells hTCEpi or RPE. The enrichment in primary cornea epithelium cells (CEC)-derived features, although lower than those from the fibroblastic cells, was significant and hence cautions on hTCEpi representing faithfully the primary tissue state. The open chromatin regions annotated by CRF-associated variants overlap across the four significantly enriched datasets (Fig. 3a), with detailed information in Supplementary Data 14. Whether those variants belong to 95% credible causal sets narrows down candidate causal regions as illustrated Fig. 3b for two loci. At locus 115, ALDH3A1, two distinct OCR are mapped but only one, in epithelial corneal cells and dermal fibroblasts, by a credible set variant. At locus 114, GAS7, a variant from each of the two independent credible causal sets map distinct regulatory regions, the former active in dermal cells only, the latter in all fibroblastic cells (nDF, DermFb, and hTK). ATAC-seq profiles from the in-house datasets around this second region show it to be inaccessible in hTCEpi cells (Fig. 3c). The variant tagging this region, rs9913911, associates with GAS7 transcription levels in skin-derived cultured fibroblasts (Fig. 2 and Fig. 3d) with the CRF-and gene expression-increasing allele, T, displaying strongest arnt::ahr binding potential (Fig. 3d).

Table 3 Cell-specific open chromatin regions enrichment analyses for CRF-associated variants.
Fig. 3: Regulatory genomic annotations for associated CRF-GWAS variants.

a Overlap of open chromatin regions mapped to GWAS variants (P-value threshold > 10−8 or tagging, r2 > 0.8, variants) across four ATAC-seq datasets significantly enriched in CRF-GWAS variants. Cell origins: adult cornea epithelium primary tissue (CEC); skin fibroblasts (DermFb); neonate skin fibroblasts (nDF); immortalized corneal keratocytes hTK. b Overlap details at two CRF loci. Variants named are those selected for enrichment analysis based on P-value threshold (tag variant), bold indicates that they belong to 95% credible sets of causal variants and fall in OCR themselves (*) or tag another credible set variant that do. **different credible set variants map to OCR: rs4646785 in CEC and nDF, rs12939864 in DermFb. c ATAC-seq profiles in immortalised corneal epithelial and stromal cell lines (respectively, hTCEpi and hTK, each in duplicate) around variant rs9913911, prioritised causal variant at locus 114, credible set 2. Screenshot from UCSC genome browser with annotation coordinates used in enrichment analysis in top tracks. d eQTL data from GTEx v8 and predicted significantly disrupted transcription factor binding motif at rs9913911 (T > C).

Enrichment is higher, but did not reach significance threshold, when only using the much reduced set of fine-mapped variants with strong causality support (Table 3). The highest enrichment was obtained for hTK-specific open chromatin region (OR = 5.68, 95% CI [1.81–17.8], p = 2.9 × 10−3), closely followed by those for DermFb and hTK. In total, eight highly likely causal variants tag OCRs in either of these tissue/cells (Supplementary Data 15). Five locate within an OCR themselves, including the three variants proposed to cause association via modulation of SMAD3, GAS7, and MSL1 expression (Table 1). The prioritized intronic SMAD3 and intergenic between COL6A1 and COL6A2 variants, locate within regions of accessible chromatin in the corneal fibroblast cell line (Supplementary Fig. 4). For all but one of the selected tag variants outside a regulatory annotation, at least one linkage disequilibrium-tagged variant within OCRs belong to the credible set, with PIP ranging from 0.014 to 0.15 and log10(BF) from 1.56 to 2.36.


GWAS analysis in the UKBB white-British sample led to 130 novel genome-wide significant CRF loci, many with multiple independent causal signals, providing a rich resource to investigate regulatory mechanisms. Although not strictly replicated given the current absence of a sample of similar size, the signals show concordance of effects in European non-British UKBB participants. We focused particularly on identifying and describing the subset of genetic variations with a strong statistical support of being causal. This set is reduced given pervasive linkage disequilibrium in the genome and leaves unexamined many associations that may be more prominent and important for an understanding of diseases or key regulatory pathways. However, this subset provides a foundation for mechanistic insights. We showed that some of those could be pursued in the TERT-immortalised keratocyte cell line hTK, which we selected as a human cell line a priori the most relevant to corneal thickness and resistance.

The multidimensionality of the CRF measure complicates interpretation of results. While CRF was designed to capture cornea properties impacting on IOPg measures, it may still be influenced by true IOP. Reciprocally, our data support that IOPcc is not devoid of relationship with corneal features, in particular those other than CCT, which agrees with observations that IOPcc, not IOPg, is sensitive to change in FECD36. Glaucoma GWAS results should in principle help disentangling effects on IOP from those on cornea per se. However, properties of the corneal cells, keratocytes in particular, are shared by other ocular tissues important in glaucoma pathology: the trabecular meshwork cells or the connective tissue of the optic nerve head37,38. Strong associations with IOPg, IOPcc and glaucoma for the GAS7 variant for example suggest that its effect on CRF is via IOP. Yet, its annotation to open chromatin region in corneal fibroblastic cells show that biological horizontal pleiotropy, a common feature of genetic variant effects, cannot be disproved. Disentangling CRF and IOPcc relationship using Mendelian Randomisation (MR) is one focus of CRF and CH UKBB GWAS39 analyses published whilst our manuscript was under review. Pitfalls and assumptions underlying MR techniques40 make advancing our understanding of genetic effects a pressing need.

CRF causal variants appear well suited to inform on corneal stromal cell biology as evidenced by enrichment analyses with the ECM pathways topping the list of biological genesets and with regulatory annotations generated in fibroblasts and mesenchymal tissues topping that of regulatory annotations. Connective tissue disorders genes FBN2, ADAMTS17, and SMAD3 affected by fine-mapped variants, as well as lower enrichment of regions active in cornea epithelial cells compared to that in fibroblastic cells lines, and improved enrichment when OCR shared with hTCEpi are removed from the hTK dataset, also support this notion.

The CRF-associated variants enrichment of OCR annotations in hTK was, arguably, not substantially greater than that obtained using skin fibroblast cell lines, and both display the most significant enrichments. This may not be too surprising as we expect many of the GWAS variants to be concerned with maintenance or establishment of the fibroblastic state, and Mendelian disorders associated with thin cornea (e.g. Brittle Cornea and Ehlers–Danlos Syndromes) are also often characterised by general connective tissue dysfunction41,42,43. Yet, effects of genetic variants could be tissue-specific and broadening cell-type repertoire is warranted for mechanistic insights. Prioritised causal variants falling into OCRs in both the dermal and corneal fibroblastic cell lines, rs8127032 (locus 135), intergenic between COL6A1 and COL6A2, or the intronic variants at SMAD3 (locus 106) are appealing candidates to start exploring mechanism and cell-specificity questions further. Type VI collagen is a major component of the human cornea stroma44 with a suggested role in cornea tensile strength45. It is a component of the ECM in muscle, vessels and skin—tissues which display the most prominent features of type VI collagenopathies. Recent exome sequence analysis in UKBB46 reported significant large effects on CRF of a burden of rare loss of function coding variants in COL6A1; rs8127032 could shed light on cis-regulatory control of this gene. Drawing on the ever-increasing regulatory annotations of the genome, particularly on experimentally defined trans-acting factors binding sites, could further shape precise hypotheses to be tested. Two of the three prioritized variants at SMAD3 locate within AP-1 components (FOS/JUN/JUND/FOSL2) binding sites in the ENCODE 3 data. AP-1 and SMAD3 are important mediators of TGF-β dependent gene regulation, including that of genes involved in ECM homeostasis47,48.

Given that the activity of many regulatory elements will be development stage and/or environment dependent—situations unlikely to be all encapsulated in available cells and tissues, we also attempted to complement enrichment analyses of regulatory features using in silico prediction of the disruptiveness of prioritised variants on TF binding sites. Suggestion of RXRA as potential link between several of the prioritised noncoding variants is supported by importance of retinoic acid signalling during cornea development and isolated keratocyte phenotypes including ECM composition49. Functional follow up will be required to ascertain relevant links especially as we used a lenient threshold for motif matching. This was chosen for discovery purpose given the lack of accuracy of in silico predictions compounded with possible functional range of transcription factors binding-affinities.

Our analyses converged nicely within locus 114 to pinpoint one noncoding causal candidate, rs9913911, giving clues as to how it could exert its effect—modulating transcription levels of GAS7 by affecting the binding of a trans-acting factor. The basic helix-loop-helix transcription factors Ahr, ARNT (HIF1β), or ARNT2, which mediate response to developmental and environmental stimuli, appeared strong contenders to be affected trans-acting factors. Down-regulation by hypoxia of GAS7, ARNT2, and HIF1α transcript levels in monocytes50 opens a possible hypoxia response connection.

Overall, our analysis provides many leads to plan hypothesis-driven experimental analyses, which are ultimately required to demonstrate causality and function of CRF-GWAS variants. It also provides ground for further investigations as in addition to the limitations already alluded to, including a focus on transcription factor-dependent causal mechanisms, we did not investigate corneal endothelial cells, an important target tissue for corneal dystrophies for which CRF appeared to be a good endophenotype, and the cornea immortalized-cell lines studied were derived from a unique donor (N = 1).


Study population

The UK Biobank is a large-scale prospective study established by the Medical Research Council, Department of Health, Wellcome Trust, Scottish Government and North-West Regional Development Agency51. The study was conducted with the approval of the North-West Research Ethics Committee (Reference: 06/MRE08/65).

Between 2006 and 2010, close to 500,000 people (273,467 female/229,175 male) were recruited. Genome-wide genotype data and imputations on the full set were made available to the international community with primary genotype quality controls and analyses such as ancestry grouping and detection of close kinship ( and ref. 52). A detailed description of the study and its access process are available online (

Phenotypes and genotypes used in this manuscript were obtained through approved application number 19655. Analyses presented are those performed with the largest ethnic subset of participants, that of European ancestry, passing phenotypic and genotypic quality controls. The main analysis was carried out in participants of the white-British ancestry subset defined by Bycroft et al.52.

A nonoverlapping subset of UKBB participants of European ancestry, among those who self-reported “other white background” or “Irish” ancestry, was used to perform independent genetic analysis. A genetically homogeneous subset for those participants was created following the method used to define the subset of white-British ancestry. Briefly, participants genetically European, based on the principal component analysis (PCA) using 1000 Genomes project Phase I reference samples performed by Bycroft et al.52, were selected if within five standard deviations from the self-reported European sample mean for each of the three first principal components (discriminating European, Asian and African ancestry super-groups). Multidimensional scaling was then performed based on the kinship matrix derived from the subset of individuals with corneal measures and not selected as white-British, with LD pruned genotyped markers of call rate >99%, MAF > 1%, and not in regions of high LD52 using KING53. Eight participants with outlying values on PC 5 and 6 were removed, leaving 10130 individuals. Final clustering was performed using the 500 reference samples of European origins from the 1000 Genomes Project Phase3 which were not closely related (of 3rd or greater degree of kinship) and with markers intersecting those present in the selected sample. Final PCs (N = 10) to use as covariates for the association analysis were generated by projecting the selected samples (N = 10,130) to the PC space of reference samples using KING53.

Ocular response analyser measures

Eye examinations were performed on 127,453 individuals, close to 25% of the UK Biobank participants, at six assessment centres, and include bidirectional applanation tonometry using a Reichert Ocular Response Analyser (ORA). The ORA produces parameters derived from the measured applanation pressures, p1 and p2 where p1 is the pressure at which the cornea flattens as air-pressure is applied, and p2 the pressure at which applanation reoccurs after the air-pressure is released14. A Golmann-correlated IOP measure (IOPg) is calculated as average of p1 and p2, and a corneal hysteresis measure (CH), is calculated as the difference between the two pressures. Corneal resistance factor (CRF) and a cornea compensated IOP measure, IOPcc, are derived as linear combinations of p1 and p2 following proprietary formulae. Those are devised to minimize the changes in IOP measurements before and after LASIK refractive surgery14. All measures are expressed in mm of mercury.

CRF analysis was performed on the average of the left and right eye measurements, following some filtering of the available measures. The 1041 UK Biobank participants with extreme inter-ocular differences (greater than population mean difference + 3 standard deviations) were removed. Additionally, 5001 participants were excluded as having self-reported or being linked to an ocular condition that could affect the measurements accuracy. Those were self-reported recent eye surgery (code 5181 in corresponding UK Biobank data-field), refractive laser surgery (code 5325), cataract surgery (code 5324), glaucoma high pressure surgery or laser treatment (codes 5326 and 5327), corneal graft surgery (code 5328), eye injury (code 5419); and the following electronic health records (data-fields 41202-41205):keratoconus (ICD10: H18.6; ICD9: 3716) or cornea disorders (ICD10: H18.4-9).

Finally, the samples failing the centrally performed quality controls for heterozygosity or/and missingness52, or having a mismatch between self-reported and genotype-derived gender or showing putative sex chromosome aneuploidy as well as individuals who have withdrawn from the study at the time of analysis were removed.

A total of 76,029 white-British participants and 10,130 European not white-British were thus available for GWAS analyses. The CRF distribution and characteristics in the white-British dataset are presented in Supplementary Note 1, including effect of age and sex which were adjusted for in the GWAS analysis.

GWAS analysis

Single-variant associations were performed by testing for an additive allelic effect at each (HRC + UK10K) imputed genotypes in the white-British subset defined by Bycroft et al.52, and, separately, in the independent sample of European not white-British UKBB participants described in the Study population section. Only the well imputed (INFO > 0.6) and common to low-frequency (MAF > 0.5%) variants were tested. The primary GWAS with all measured individuals were performed using a linear mixed model, accounting for population structure and (cryptic) relatedness, implemented in the software BOLT_LMM v1.354,55. Covariates fitted in the model were: age, sex, assessment centre, genotyping array, genotyping batch, and principal components of ancestry (20 for the white-British analysis, 10 for the European not white-British analysis). Fine-mapping analyses were performed on GWAS results obtained from the restricted set of measured individuals who were not closely related (pairwise kinship coefficient greater than 0.025 calculated using KING53), amounting to N = 72,301 individuals. GWAS there was performed using PLINK256.

Preliminary functional annotation and gene mapping

SNPs annotations, gene mapping, and geneset enrichment test were performed with the functional mapping and annotation of genome-wide association studies (FUMA) platform v1.3.526. UKBB release 2, with a subset of 10,000 white-British participants, was selected as the linkage disequilibrium (LD) reference for the annotations. Those included the determination of independent lead SNPs in a locus, defined heuristically by FUMA using a LD measure r2 cut-off of 0.1. Loci were defined by the coordinates of variants in LD (r2 > =0.6) with lead variant (lowest P-value) or independent lead variants and merging of LD blocks located less than 250 kb to each other. The SNP2GENE function in FUMA was used to annotate putative target genes in an exhaustive manner. A first annotation include physical mapping. Other candidate genes, which can locate outside loci limits, are those whose expression has been significantly associated with GWAS variants based on a comprehensive source of eQTL repositories at time of query (August 2019-GTEx, eQTLGen, Common Mind Consortium, Blood eQTL, BIOS, and BRAINEAC) or based on chromatin interaction maps available at time of query (PsychENCODE, FANTOM5, Hi-C data GSE87112 and Giusti-Rodriguez_et_al_2019) with tissue or cell type where interactions have been performed. The references for those resources compiled by FUMA can be found at

The geneset enrichment test implemented in MAGMA57 v1.06 was also performed within FUMA, with genesets from MsigDB v6.2. Significance threshold is based on Bonferroni correction.

GWAS signals colocalisation

Bayesian colocalisation analysis as proposed by Giambartolomei et al.58 was performed using the default priors of the coloc.compute function of the R package gtx 2.1.6 ( This framework provides posterior probabilities for each possible scenario within the region tested: neither trait has a genetic association (pp0), only trait1 has (pp1), only trait2 has (pp2), both trait1 and trait2 are associated but with different causal variants (pp3), and both traits share causal association signal(s) (pp4). A pp4 > 75% indicates strong evidence that both GWAS signals colocalise.

Multiple signals detection and fine-mapping

The genome-wide complex trait analysis (GCTA) software package (C++), version 1.9.1 beta3, implements a multi-SNP stepwise model selection, which can indicate multiple functional variants within a region and estimates SNP effect conditional on effect of nearby selected variants27. The stepwise model selection (--cojo-slct) was run with the default P-value threshold of 5 × 10−8, collinearity threshold of 0.9, with genotypes (hard calls from imputed dosages as implemented in plink2) and phenotype files used for input. A computationally efficient Bayesian alternative variant selection method implemented (C++) in FINEMAP28, v1.3.1, was also performed. The shotgun stochastic search algorithm was used (-sss argument) using the number of causal SNPs with associated posterior probabilities determined from a prerun with the maximal number of causal SNPs set to 10 (--n-causal-snps = 10). The data required for each investigated locus are the corresponding GWAS summary statistics and a regional LD matrix. This latter was computed from the imputed genotypes posterior probabilities stored in UKBB bgen files using LDstore59. FINEMAP reported a set of CRF-associated variants for each detected independent association signal, with 95% probability of containing the causal variant—so called 95% credible set. Attached to each variant are two useful metrics to rank the candidate variants: its posterior probability to be included in the credible set (PIP), with note that two variants in complete LD will have the same PIP score, and a Bayes factor (log10 scale) quantifying how likely the variant is to be causal rather than non--causal, with log10 Bayes factor (log10(BF)) greater than 2 deemed decisive evidence.

TF binding sites prediction

We used motifbreakR60 to identify variants predicted to strongly alter binding of a transcription factor (TF) based on position probability matrices (PPM). Two sources of TF motifs encompassing 14 public collections (including JASPAR, HOCOMOCO, ENCODE, HOMER and FactorBook) were used (MotifDb and motifbreakR_motif). Analysis was carried out with the program default settings apart from the P-value threshold to declare TF binding site matching either of the allelic configuration set to 5 × 10−4 and the relative entropy scoring method set to information content algorithm (method = “ic”) as performed in ref. 61.

Accurate P-values for each allele match were calculated for a subset of variants of interest using the function calculatePvalue() implementing the algorithm developed by Touzet and Varre62. All but three prioritised variants, those corresponding to indels rs141144358, rs200584273, and rs5877786, could be analysed. To derive probability of the findings by chance, sampling of GWAS variants matched to the query SNPs based on allele frequency, number of SNPs in LD, distance to nearest gene and gene density63 was generated 10,000 times as described in more details in Supplementary Note 5.


Assay for transposase accessible chromatin followed by sequencing (ATAC-seq) was performed on two cornea immortalised cell lines generated by Prof J Jester and collaborators and kindly gifted to us by Dr Che Connon (University of Newcastle): the hTCEpi cell line derived from primary human cornea epithelial cells35 and the hTK cell line derived from primary cornea keratocytes34, resident cells of the cornea stroma. hTCEpi and hTK cells were cultured respectively in KGMTM-2 basal medium with KGMTM-2 Single-QuotsTM supplements (Lonza) and Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum, with addition of 1% penicillin and streptavidin in both media. Two replicates of hTCEpi and hTK cell batches were harvested at 70–90% confluence, and 50,000 cells processed per experiment. Nuclei preparation and transposase digestion followed the ATAC-seq protocol described in ref. 64 and are further detailed in Supplementary Note 2, together with library preparation and sequencing. We obtained over 75 million usable uniquely mapped paired-end reads for each sample (min 78,877,410– max 192,169,752) following removal of reads mapping to the mitochondrial genome, or to more than one genomic region or duplicated (Supplementary Note 2). The genomic coverage of replicates (two independent cultures) was highly reproducible with Pearson’s correlation coefficients of 0.958 and 0.987 respectively for hTCEpi and hTK cells. Peak calling was performed using the findPeaks command from the HOMER software package65 as detailed in Supplementary Note 2, and replicates combined following algorithms detailed Supplementary Note 3. Sets of ATAC-seq peaks present in corneal keratocyte but not in the epithelium derived cells lines and vice versa were also derived using a few calling variations (Supplementary Note 3).

Published datasets from relevant or control human cells were obtained from NCBI Gene Expression Omnibus (GEO) and their quality checked in a similar way to that used for the in-house datasets. Briefly, the original fastq reads were downloaded and sequences aligned using bowtie2 (v, as paired ended alignments for all but the K562 cell lines data (Supplementary Data 13). In general, read numbers were small for individual datasets and pooling was performed to create a better set for confident peaks calling (Supplementary Data 13). Fragments length plots, on which the Supplementary Data 13 fragment quality criterion is based, are presented Supplementary Note 4. Resemblance between tissues/cells ATAC-seq sets was evaluated by unsupervised clustering of the original BAM files using the plotCorrelation option of the deeptools suite66 in Python and illustrated Supplementary Fig. 5.

Tissue-/cell-specific functional annotation enrichment analysis

We use the GWAS Analysis of Regulatory or Functional Information Enrichment with LD correction (GARFIELD)33 to perform annotation enrichment analysis. It tests using logistic regression whether associated variants (or tagged variants) map significantly more frequently to regions of interest than the nonassociated variants. The method takes into account variations in LD patterns and gene density across the genome as well as the correlations between functional annotations when multiple sets of annotation are tested together. Genomic locations tagged by the GWAS SNPs at different P-value thresholds were investigated, as well as those tagged by the restricted set of identified SNPs with strong statistical support for being causal (log10(BF) greater or equal to 3). To perform feature enrichments for those SNPs, a file was created with dummy P-values, above threshold (>10-8) for all variants except those of interest which were allocated an under threshold P-value.

Analysis was first carried out using the functional annotations provided within the package, those from Roadmap Epigenomics31 and Encyclopedia of DNA Elements (ENCODE)67. We then performed enrichment analysis for cell-specific chromatin accessible regions by using custom annotations corresponding to the ATAC-seq datasets described in the ATAC-seq section.

We prepared GARFIELD variants-related input files anew so that all the variants analysed (Imputation Score > 0.6 and MAF > 0.005) and their pairwise LD measures matched the GWAS dataset (UK Biobank white-British ancestry). Those input files contain information on the distance to the nearest transcription start site for each variant, the lists of SNPs that can be clumped together (r2 > 0.1) for the first step of greedy pruning of GWAS SNPs and the lists of SNPs that are proxies to the GWAS SNPs (r2 > 0.8) for scoring overlap with an annotated feature. Enrichment P-values reported by GARFIELD are adjusted with respect to multiple testing using a Bonferonni correction and an effective number of tests carried out based on the correlation structure of the annotation features tested. For the ATAC-seq datasets tested, which were analysed all together, the significance threshold (false positive rate of 5%) was 0.00055.

URLs: UK Biobank Access Management System,; ENCODE,; FUMA,; GCTA,; GENCODE; MotifBreakR,; UCSC genome browser,; GARFIELD,; GEO,; GTEx portal,; Roadmap Epigenomics Consortium; KING; MSigDB

Data availability

The UK Biobank resource, open to all bona fide health researchers, is available upon request through their access management system. The summary statistics for the analysis of CRF in participants of white-British ancestry can be downloaded from Novel ATAC-seq data generated in the hTCEpi and hTK cells have been deposited on GEO under accession GSE150064.


  1. 1.

    Garcia-Porta, N. et al. Corneal biomechanical properties in different ocular conditions and new measurement techniques. ISRN Ophthalmol. 2014, 724546 (2014).

    PubMed  PubMed Central  Article  Google Scholar 

  2. 2.

    Kotecha, A. What biomechanical properties of the cornea are relevant for the clinician? Surv. Ophthalmol. 52, S109–14 (2007).

    PubMed  Article  Google Scholar 

  3. 3.

    Lu, Y. et al. Common genetic variants near the Brittle Cornea Syndrome locus ZNF469 influence the blinding disease risk factor central corneal thickness. PLoS Genet. 6, e1000947 (2010).

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  4. 4.

    Vitart, V. et al. New loci associated with central cornea thickness include COL5A1, AKAP13 and AVGR8. Hum. Mol. Genet. 19, 4304–4311 (2010).

    CAS  PubMed  Article  Google Scholar 

  5. 5.

    Lu, Y. et al. Genome-wide association analyses identify multiple loci associated with central corneal thickness and keratoconus. Nat. Genet. 45, 155–163 (2013).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  6. 6.

    Iglesias, A. I. et al. Cross-ancestry genome-wide association analysis of corneal thickness strengthens link between complex and Mendelian eye diseases. Nat. Commun. 9, 1864 (2018).

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  7. 7.

    Ivarsdottir, E. V. et al. Sequence variation at ANAPC1 accounts for 24% of the variability in corneal endothelial cell density. Nat. Commun. 10, 1284 (2019).

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  8. 8.

    Choquet, H. et al. A multiethnic genome-wide analysis of 44,039 individuals identifies 41 new loci associated with central corneal thickness. Commun. Biol. 3, 301 (2020).

    PubMed  PubMed Central  Article  Google Scholar 

  9. 9.

    Han, S. et al. Association of variants in FRAP1 and PDGFRA with corneal curvature in Asian populations from Singapore. Hum. Mol. Genet. 20, 3693–3698 (2011).

    CAS  PubMed  Article  Google Scholar 

  10. 10.

    Fan, Q. et al. Genome-wide association meta-analysis of corneal curvature identifies novel loci and shared genetic influences across axial length and refractive error. Commun. Biol. 3, 133 (2020).

    PubMed  PubMed Central  Article  Google Scholar 

  11. 11.

    Khawaja, A. P. et al. Genetic variants associated with corneal biomechanical properties and potentially conferring susceptibility to keratoconus in a genome-wide association study. JAMA Ophthalmol. 137, 1005–1012 (2019).

    PubMed Central  Article  Google Scholar 

  12. 12.

    Spain, S. L. & Barrett, J. C. Strategies for fine-mapping complex traits. Hum. Mol. Genet. 24, R111–R119 (2015).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  13. 13.

    Huang, H. et al. Fine-mapping inflammatory bowel disease loci to single-variant resolution. Nature 547, 173–178 (2017).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  14. 14.

    Luce, D. A. Determining in vivo biomechanical properties of the cornea with an ocular response analyzer. J. Cataract Refract Surg. 31, 156–162 (2005).

    PubMed  Article  PubMed Central  Google Scholar 

  15. 15.

    Kara, N. et al. Corneal biomechanical properties and intraocular pressure measurement in Marfan patients. J. Cataract Refract Surg. 38, 309–314 (2012).

    PubMed  Article  PubMed Central  Google Scholar 

  16. 16.

    Meek, K. M. et al. Changes in collagen orientation and distribution in keratoconus corneas. Invest. Ophthalmol. Vis. Sci. 46, 1948–1956 (2005).

    PubMed  Article  PubMed Central  Google Scholar 

  17. 17.

    Sultan, G. et al. Cornea in Marfan disease: orbscan and in vivo confocal microscopy analysis. Invest. Ophthalmol. Vis. Sci. 43, 1757–1764 (2002).

    PubMed  PubMed Central  Google Scholar 

  18. 18.

    Galletti, J. G., Pfortner, T. & Bonthoux, F. F. Improved keratoconus detection by ocular response analyzer testing after consideration of corneal thickness as a confounding factor. J. Refract Surg. 28, 202–208 (2012).

    PubMed  Article  PubMed Central  Google Scholar 

  19. 19.

    Johnson, R. D., Nguyen, M. T., Lee, N. & Hamilton, D. R. Corneal biomechanical properties in normal, forme fruste keratoconus, and manifest keratoconus after statistical correction for potentially confounding factors. Cornea 30, 516–523 (2011).

    PubMed  Article  PubMed Central  Google Scholar 

  20. 20.

    Baratz, K. H. et al. E2-2 protein and Fuchs’s corneal dystrophy. N. Engl. J. Med. 363, 1016–1024 (2010).

    CAS  PubMed  Article  PubMed Central  Google Scholar 

  21. 21.

    Hassell, J. R. & Birk, D. E. The molecular basis of corneal transparency. Exp. Eye Res. 91, 326–335 (2010).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  22. 22.

    Cuellar-Partida, G. et al. WNT10A exonic variant increases the risk of keratoconus by decreasing corneal thickness. Hum. Mol. Genet. 24, 5060–5068 (2015).

    CAS  PubMed  Article  PubMed Central  Google Scholar 

  23. 23.

    Bulik-Sullivan, B. K. et al. LD Score regression distinguishes confounding from polygenicity in genome-wide association studies. Nat. Genet. 47, 291–295 (2015).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  24. 24.

    Afshari, N. A. et al. Genome-wide association study identifies three novel loci in Fuchs endothelial corneal dystrophy. Nat. Commun. 8, 14898 (2017).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  25. 25.

    Gao, X. R., Huang, H., Nannini, D. R., Fan, F. & Kim, H. Genome-wide association analyses identify new loci influencing intraocular pressure. Hum. Mol. Genet. 27, 2205–2213 (2018).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  26. 26.

    Watanabe, K., Taskesen, E., van Bochoven, A. & Posthuma, D. Functional mapping and annotation of genetic associations with FUMA. Nat. Commun. 8, 1826 (2017).

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  27. 27.

    Yang, J., Lee, S. H., Goddard, M. E. & Visscher, P. M. GCTA: a tool for genome-wide complex trait analysis. Am. J. Hum. Genet. 88, 76–82 (2011).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  28. 28.

    Benner, C. et al. FINEMAP: efficient variable selection using summary data from genome-wide association studies. Bioinformatics 32, 1493–1501 (2016).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  29. 29.

    Rentzsch, P., Witten, D., Cooper, G. M., Shendure, J. & Kircher, M. CADD: predicting the deleteriousness of variants throughout the human genome. Nucleic Acids Res. 47, D886–D894 (2019).

    CAS  PubMed  Article  PubMed Central  Google Scholar 

  30. 30.

    Consortium, G. T. et al. Genetic effects on gene expression across human tissues. Nature 550, 204–213 (2017).

    Article  Google Scholar 

  31. 31.

    Bernstein, B. E. et al. The NIH roadmap epigenomics mapping consortium. Nat. Biotechnol. 28, 1045–1048 (2010).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  32. 32.

    Ward, L. D. & Kellis, M. HaploReg v4: systematic mining of putative causal variants, cell types, regulators and target genes for human complex traits and disease. Nucleic Acids Res. 44, D877–D881 (2016).

    CAS  PubMed  Article  Google Scholar 

  33. 33.

    Iotchkova, V. et al. GARFIELD classifies disease-relevant genomic features through integration of functional annotations with association signals. Nat. Genet. 51, 343–353 (2019).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  34. 34.

    Jester, J. V. et al. Myofibroblast differentiation of normal human keratocytes and hTERT, extended-life human corneal fibroblasts. Invest. Ophthalmol. Vis. Sci. 44, 1850–1858 (2003).

    PubMed  Article  Google Scholar 

  35. 35.

    Robertson, D. M. et al. Characterization of growth and differentiation in a telomerase-immortalized human corneal epithelial cell line. Invest. Ophthalmol. Vis. Sci. 46, 470–478 (2005).

    PubMed  Article  Google Scholar 

  36. 36.

    Clemmensen, K. & Hjortdal, J. Intraocular pressure and corneal biomechanics in Fuchs’ endothelial dystrophy and after posterior lamellar keratoplasty. Acta Ophthalmol. 92, 350–354 (2014).

    PubMed  Article  Google Scholar 

  37. 37.

    Gould, D. B., Smith, R. S. & John, S. W. Anterior segment development relevant to glaucoma. Int. J. Dev. Biol. 48, 1015–1029 (2004).

    PubMed  Article  Google Scholar 

  38. 38.

    Marshall, G. E., Konstas, A. G. & Lee, W. R. Collagens in ocular tissues. Br. J. Ophthalmol. 77, 515–524 (1993).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  39. 39.

    Simcoe, M. J. et al. Genome-wide association study of corneal biomechanical properties identifies over 200 loci providing insight into the genetic aetiology of ocular diseases. Hum. Mol. Genet. 29, 3154–3164 (2020).

    PubMed  PubMed Central  Article  Google Scholar 

  40. 40.

    Bowden, J. Misconceptions on the use of MR-Egger regression and the evaluation of the InSIDE assumption. Int. J. Epidemiol. 46, 2097–2099 (2017).

    PubMed  PubMed Central  Article  Google Scholar 

  41. 41.

    Abu, A. et al. Deleterious mutations in the Zinc-Finger 469 gene cause brittle cornea syndrome. Am. J. Hum. Genet. 82, 1217–1222 (2008).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  42. 42.

    Burkitt Wright, E. M. et al. Brittle cornea syndrome: recognition, molecular diagnosis and management. Orphanet J. Rare Dis. 8, 68 (2013).

    PubMed  PubMed Central  Article  Google Scholar 

  43. 43.

    Villani, E. et al. The cornea in classic type Ehlers-Danlos syndrome: macro- and microstructural changes. Invest. Ophthalmol. Vis. Sci. 54, 8062–8068 (2013).

    PubMed  Article  PubMed Central  Google Scholar 

  44. 44.

    Zimmermann, D. R., Trueb, B., Winterhalter, K. H., Witmer, R. & Fischer, R. W. Type VI collagen is a major component of the human cornea. FEBS Lett. 197, 55–58 (1986).

    CAS  PubMed  Article  PubMed Central  Google Scholar 

  45. 45.

    Cho, H. I., Covington, H. I. & Cintron, C. Immunolocalization of type VI collagen in developing and healing rabbit cornea. Invest. Ophthalmol. Vis. Sci. 31, 1096–1102 (1990).

    CAS  PubMed  PubMed Central  Google Scholar 

  46. 46.

    Van Hout, C. V. et al. Whole exome sequencing and characterization of coding variation in 49,960 individuals in the UK Biobank. Nature 586, 749–756 (2020).

    CAS  PubMed  Article  PubMed Central  Google Scholar 

  47. 47.

    Verrecchia, F. et al. Smad3/AP-1 interactions control transcriptional responses to TGF-beta in a promoter-specific manner. Oncogene 20, 3332–3340 (2001).

    CAS  PubMed  Article  Google Scholar 

  48. 48.

    Chung, K. Y., Agarwal, A., Uitto, J. & Mauviel, A. An AP-1 binding sequence is essential for regulation of the human alpha2(I) collagen (COL1A2) promoter activity by transforming growth factor-beta. J. Biol. Chem. 271, 3272–3278 (1996).

    CAS  PubMed  Article  Google Scholar 

  49. 49.

    Gouveia, R. M. & Connon, C. J. The effects of retinoic acid on human corneal stromal keratocytes cultured in vitro under serum-free conditions. Invest. Ophthalmol. Vis. Sci. 54, 7483–7491 (2013).

    CAS  PubMed  Article  Google Scholar 

  50. 50.

    Bosco, M. C. et al. Hypoxia modifies the transcriptome of primary human monocytes: modulation of novel immune-related genes and identification of CC-chemokine ligand 20 as a new hypoxia-inducible gene. J. Immunol. 177, 1941–1955 (2006).

    CAS  PubMed  Article  PubMed Central  Google Scholar 

  51. 51.

    Sudlow, C. et al. UK biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med. 12, e1001779 (2015).

    PubMed  PubMed Central  Article  Google Scholar 

  52. 52.

    Bycroft, C. et al. The UK Biobank resource with deep phenotyping and genomic data. Nature 562, 203–209 (2018).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  53. 53.

    Manichaikul, A. et al. Robust relationship inference in genome-wide association studies. Bioinformatics 26, 2867–2873 (2010).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  54. 54.

    Loh, P. R., Kichaev, G., Gazal, S., Schoech, A. P. & Price, A. L. Mixed-model association for biobank-scale datasets. Nat. Genet. 50, 906–908 (2018).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  55. 55.

    Loh, P. R. et al. Efficient Bayesian mixed-model analysis increases association power in large cohorts. Nat. Genet. 47, 284–290 (2015).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  56. 56.

    Chang, C. C. et al. Second-generation PLINK: rising to the challenge of larger and richer datasets. Gigascience 4, 7 (2015).

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  57. 57.

    de Leeuw, C. A., Mooij, J. M., Heskes, T. & Posthuma, D. MAGMA: generalized gene-set analysis of GWAS data. PLoS Comput. Biol. 11, e1004219 (2015).

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  58. 58.

    Giambartolomei, C. et al. Bayesian test for colocalisation between pairs of genetic association studies using summary statistics. PLoS Genet. 10, e1004383 (2014).

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  59. 59.

    Benner, C. et al. Prospects of fine-mapping trait-associated genomic regions by using summary statistics from genome-wide association studies. Am. J. Hum. Genet. 101, 539–551 (2017).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  60. 60.

    Coetzee, S. G., Coetzee, G. A. & Hazelett, D. J. motifbreakR: an R/Bioconductor package for predicting variant effects at transcription factor binding sites. Bioinformatics 31, 3847–3849 (2015).

    CAS  PubMed  PubMed Central  Google Scholar 

  61. 61.

    Ulirsch, J. C. et al. Interrogation of human hematopoiesis at single-cell and single-variant resolution. Nat. Genet. 51, 683–693 (2019).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  62. 62.

    Touzet, H. & Varre, J. S. Efficient and accurate P-value computation for Position Weight Matrices. Algorithms Mol. Biol. 2, 15 (2007).

    PubMed  PubMed Central  Article  Google Scholar 

  63. 63.

    Pers, T. H., Timshel, P. & Hirschhorn, J. N. SNPsnap: a Web-based tool for identification and annotation of matched SNPs. Bioinformatics 31, 418–420 (2015).

    CAS  PubMed  Article  PubMed Central  Google Scholar 

  64. 64.

    Buenrostro, J. D., Wu, B., Chang, H. Y. & Greenleaf, W. J. ATAC-seq: a method for assaying chromatin accessibility genome-wide. Curr. Protoc. Mol. Biol. 109, 1–9 (2015). 21 29.

    Article  Google Scholar 

  65. 65.

    Heinz, S. et al. Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol. Cell 38, 576–589 (2010).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  66. 66.

    Ramirez, F. et al. deepTools2: a next generation web server for deep-sequencing data analysis. Nucleic Acids Res. 44, W160–W165 (2016).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  67. 67.

    Consortium, E. P. An integrated encyclopedia of DNA elements in the human genome. Nature 489, 57–74 (2012).

    Article  CAS  Google Scholar 

  68. 68.

    Lessard, J. et al. An essential switch in subunit composition of a chromatin remodeling complex during neural development. Neuron 55, 201–215 (2007).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  69. 69.

    Lange, M. et al. Regulation of muscle development by DPF3, a novel histone acetylation and methylation reader of the BAF chromatin remodeling complex. Genes Dev. 22, 2370–2384 (2008).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  70. 70.

    Zeng, L. et al. Mechanism and regulation of acetylated histone binding by the tandem PHD finger of DPF3b. Nature 466, 258–262 (2010).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  71. 71.

    Choi, S. I. et al. Involvement of TGF-{beta} receptor- and integrin-mediated signaling pathways in the pathogenesis of granular corneal dystrophy II. Invest. Ophthalmol. Vis. Sci. 51, 1832–1847 (2010).

    PubMed  Article  Google Scholar 

  72. 72.

    Busch, C. et al. Ocular findings in Loeys-Dietz syndrome. Br. J. Ophthalmol. 102, 1036–1040 (2018).

    PubMed  Article  Google Scholar 

  73. 73.

    Sethi, M. K. et al. Identification of glycosyltransferase 8 family members as xylosyltransferases acting on O-glucosylated notch epidermal growth factor repeats. J. Biol. Chem. 285, 1582–1586 (2010).

    CAS  PubMed  Article  Google Scholar 

  74. 74.

    Ju, Y. T. et al. gas7: A gene expressed preferentially in growth-arrested fibroblasts and terminally differentiated Purkinje neurons affects neurite formation. Proc. Natl Acad. Sci. USA 95, 11423–11428 (1998).

    CAS  PubMed  Article  Google Scholar 

  75. 75.

    Hung, F. C., Shih, H. Y., Cheng, Y. C. & Chao, C. C. Growth-arrest-specific 7 gene regulates neural crest formation and craniofacial development in zebrafish. Stem Cells Dev. 24, 2943–2951 (2015).

    CAS  PubMed  Article  Google Scholar 

  76. 76.

    Smith, E. R. et al. A human protein complex homologous to the Drosophila MSL complex is responsible for the majority of histone H4 acetylation at lysine 16. Mol. Cell Biol. 25, 9175–9188 (2005).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  77. 77.

    Sardina, J. L. et al. Transcription factors drive Tet2-mediated enhancer demethylation to reprogram cell fate. Cell Stem Cell 23, 905–906 (2018).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  78. 78.

    Choudhry, H. & Harris, A. L. Advances in hypoxia-inducible factor biology. Cell Metab. 27, 281–298 (2018).

    CAS  PubMed  Article  Google Scholar 

Download references


The research was supported by Medical Research Council University Unit programme grants MC_UU_00007/10 and MC_UU_00007/2. N.D. was supported by the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No 642934, Chromatin3D. X.J. tuitions fees were sponsored by the School of Basic Medical Sciences, Zhejiang University. We thank the UK Biobank Resource, approved under application 19655 and are grateful to David Clark and Jim Wilson for the management of this application. We thank members of the International Glaucoma Genetics Consortium (IGGC) for making summary statistics for CCT publicly available and the members of the UK Biobank Eye consortium who selected and implemented quantitative eye measurements in the UK Biobank. We are grateful to the joint MSc Biomedical Sciences programme between the Universities of Edinburgh and Zheijang which supported X.J. Finally, we are grateful to Martin Mikl from the Weizmann Institute for help with the ATAC-seq data and the IGMM IT core team, in particular John Ireland.

Author information




V.V. and W.B. conceived and supervised the study. X.J. performed GWAS, fine-mapping and transcription factors binding predictions, N.D. generated ATAC-seq profiles in the immortalized cornea cell lines and performed quality control and data preparation with Y.K. and E.P.-C. performed data preparation and regulatory annotations enrichments, T.B. performed GWAS and colocalisation analysis with GTEx data, Y.K. performed quality check of publicly available ATAC-seq datasets and prepared data for analysis and deposition to GEO, V.V. wrote a first draft of the manuscript and all contributed to data interpretation, contributed critical reviews and edits to subsequent drafts.

Corresponding author

Correspondence to Veronique Vitart.

Ethics declarations

Competing interests

The authors declare no competing interests

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Jiang, X., Dellepiane, N., Pairo-Castineira, E. et al. Fine-mapping and cell-specific enrichment at corneal resistance factor loci prioritize candidate causal regulatory variants. Commun Biol 3, 762 (2020).

Download citation


By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.


Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing