Introduction

Intraocular pressure (IOP) is a modifiable risk factor and predictive measure for glaucoma[1–4] (S1 Fig). Genome-wide association studies (GWAS) have commonly used this endophenotype that exhibits high genetic correlation (rg = 0.71) to glaucoma, as an approach to prioritize genetic variants likely to contribute to disease risk[5]. More than 68 independent loci have been implicated with intraocular pressure by meeting the GWAS significance threshold of association (P < 5x10^-8) [5–8], and a subset have reached genome-wide significance for glaucoma. For these discoveries, like most GWAS results, it has proven challenging to infer the functional consequences of common variant associations beyond cases where protein-altering variants have been directly implicated. Protein-altering variants, generally the strongest-acting genetic variants in medical genetics, include missense substitutions and protein-truncating variants, and understanding their functional consequences provides insight into the therapeutic effects of inhibiting or down-regulating the gene in which they reside[9]. Thus, identifying protein-altering variants that confer protection from disease holds particular promise for identifying therapeutic targets.

Here we leverage two population cohorts that provide complementarity for glaucoma gene discovery (Fig 1). First, UK Biobank has obtained IOP measurements in approximately 128,000 individuals in addition to case-control status for glaucoma from hospital in-patient and verbal questionnaire data in over 500,000 individuals[10–12]. Second, FinnGen has directly genotyped and aggregated disease outcomes in over 176,000 individuals from Finland, an isolated population with recent bottlenecks that offers an unprecedented advantage for studying rare variants in complex diseases[13]. With clinic-based recruitment focused on several areas including ophthalmology, and with 31.1% of the collection above age 70, FinnGen is particularly well-powered for aging-associated endpoints. We, therefore, conduct targeted association analysis with IOP measurements in UK Biobank (N = 82,253, S1 Table) to identify rare protein-altering variants that reduce IOP, and test whether those variants or others in the same genes, also confer protection to glaucoma in FinnGen (6,537 cases and 170,362 controls) and UK Biobank (4,238 cases and 250,660 controls not included in the initial IOP association analysis). The multi-cohort allelic series analysis of protein-altering variants in ANGPTL7 in a total of 10,806 glaucoma patients and over 400,000 controls identifies a significant lowering effect on IOP and protective association with glaucoma. By analyzing putative loss-of-function variants, we find concordant effect directions with the missense substitutions, suggesting that the protective mechanism may reside in the loss of gene function.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. The overview of the study based on 514,050 individuals in the UK Biobank and FinnGen cohorts.

We identified the association between ANGPTL7 and intraocular pressure (IOP) phenotypes using the genome-wide association analysis for rare (0.01% < MAF < 1%) protein-altering variants outside of MHC region in UK Biobank and applied burden and dispersion test (Analysis 1). In FinnGen, we discovered a Finnish enriched allele, p.Arg220Cys, in ANGPTL7 and performed association and subtype analysis of glaucoma (Analysis 2). In the UK Biobank, we replicated the associations between ANGPTL7 and glaucoma using the individuals that are not included in Analysis 1 (Analysis 3). OR corresponds to the odds ratio.

https://doi.org/10.1371/journal.pgen.1008682.g001

Results

We conducted protein-altering variant association analysis with IOP, as measured via corneal-compensated and Goldmann-correlated tonometry, in 82,253 unrelated White British individuals in UK Biobank dataset (S2–S4 Figs, Methods) [14]. Across 41,590 rare (0.01% < MAF < 1%) protein-altering variants outside of the MHC region with genotyping array data in UK Biobank, we performed association analysis to scan for variants with IOP-lowering effects. Specifically, we took the median of the left and right eye IOP measurements, applied quantile normalization and used a generalized linear model implemented in PLINK[13] with age, sex, and the first 4 genotype principal components (PCs) as covariates (Methods). We identified one protein-altering variant significantly associated with lower both IOP measurements below the Bonferroni-corrected P value < 1.0x10^-6, a missense substitution (p.Gln175His) in ANGPTL7 (P = 5.96x10^-9 and 1.07x10^-13, β = -0.20 and -0.16 SD 95% CI: [-0.21, -0.10] and [-0.25, -0.15], -0.53 and -0.67 mmHg for heterozygotes, -3.40 and -2.37 mmHg for homozygotes for corneal compensated and Goldman-correlated IOP, respectively, Fig 1, Tables 1 and 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. ANGPTL7 IOP protein-altering variant association in UK Biobank.

The association statistics for corneal compensated IOP and Goldman-correlated IOP are shown. Variant includes chromosome, position, reference, and alternate allele (hg19). rsID—the rs identifier of the genetic variant. HGVSp—the HGVS amino acid nomenclature. MAF—the minor allele frequency in UK Biobank white British population. Beta—estimated regression coefficient with 95% confidence intervals. P—p-value of association.

https://doi.org/10.1371/journal.pgen.1008682.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. ANGPTL7 allelic series association summary in UK Biobank and FinnGen.

Variant: the rs identifier (rsID), the amino acid nomenclature (HGVSp), and genomic coordinate (CHR for chromosome and POS for the position in hg19), as well as reference (REF) and alternate (ALT) alleles, are shown. Dosage—genotype of individuals for protein and nucleotide sequences. Carrier frequency—carrier frequency in UK Biobank (UK) and FinnGen (Finland) for the respective genotype dosage. N with IOP—number of individuals in UK Biobank with intraocular pressure measurements corresponding to the genotype dosage. Effect size estimates—reported effect size estimates. IOP (mmHg) [95% CI]—unstandardized estimates of effect size on corneal-compensated and Goldmann-correlated intraocular pressure measurements (NB: standardized estimated effect sizes may have lower p-values due to normalization procedure). OR for glaucoma—estimate odds ratio on glaucoma risk for the respective genotype dosage. NS non-significant (p > 0.1). Effect sizes always reported with respect to alternate allele dosage.

https://doi.org/10.1371/journal.pgen.1008682.t002

Based on this finding, we assessed whether any additional rare variant associations in ANGPTL7 were present. We found three additional independent rare protein-altering variants in ANGPTL7 (MAF < 0.25%) including a premature stop-gain allele (p.Arg177Ter, MAF = 0.041%, Table 1). Collectively, these three variants showed a marginally significant association with lower one of the IOP measurements (P = 0.24 and 0.0012 for corneal compensated and Goldman-correlated IOP, respectively, Fig 1), with the protein-altering allele p.Arg140His also showing a marginally significant effect on its own (P = 1.3x10^-3 and 1.9x10^-3 for corneal compensated and Goldman-correlated IOP, respectively, Table 1). Genotyping intensity plots and the concordance of genotype calls from the array and whole-exome sequencing data were manually inspected to ensure high quality and consistent genotyping (S5 Fig, S2 Table, Methods) and alleles were confirmed to be independent (pairwise r² < 10⁻⁴ for each, S3 Table, Methods). The burden and dispersion test showed significant p-values for ANGPTL7 (P = 1.88x10^-7 [burden], 1.43x10^-8 [dispersion]; and 1.44x10^-14 [burden], 2.89x10^-15 [dispersion] for corneal compensated and Goldman-correlated IOP, respectively, Fig 1). Collectively, those four alleles explain 0.03% and 0.07% of phenotypic variation for corneal-compensated and Goldmann-correlated tonometry IOP measures (S4 Table). These signals were consistently observed in corneal-compensated and Goldmann-correlated tonometry IOP measures for both left and right eyes (S5 Table), which is expected as the genetic correlation among those range from 0.75 to 1.0 (S6 Table).

In Khawaja et al. the authors identified a significant association in the same genomic region (chr1p36.22), which they label as the UBIAD1 region [6]. The associated region, defined by recombination rates, encompasses a relatively large area around SNP rs143038218 and includes the ANGPTL7 gene. We asked whether this signal could be explained by our allelic series of protein-altering variants in ANGPTL7. Linkage disequilibrium analysis shows that p.Gln175His has an r² = 0.83 suggesting that the ANGPTL7 protein-altering variant was indeed responsible for the association signal observed in Khawaja et al, whereas p.Arg177Ter, p.Arg140His, and p.Gln136Arg all have r² of approximately 0 and are under linkage-equilibrium as minor alleles are observed in different haplotypes and contribute independently to the association we observe in ANGPTL7 against IOP.

We next asked whether any of these putative IOP-lowering genetic variants showed effects consistent with reducing glaucoma risk. We focused on unrelated White British individuals that do not have IOP measures (4,238 cases and 250,660 controls, S6 Fig). For p.Gln175His in ANGPTL7, using logistic regression analysis with age, sex, and principal components (PC1-PC4) as covariates, we estimated that the variant lowers glaucoma risk by 34% (P = 0.00543; OR = 0.66 [95% CI: 0.366–0.954], Table 2). The three additional protein-altering variants did not significantly confer protection against glaucoma (burden test P = 0.77). This is consistent with power calculations, using Genetic Power Calculator [15], where our power to detect association for rare variants with a composite allele frequency of 0.345% and a binary trait in 4,238 cases and 250,660 controls at alpha = 0.05, and 0.001, i.e. P < .05 and .001, with OR of 0.7 (30% reduction in risk) is equal to 62% and 15.2%, respectively.

We then sought evidence from the FinnGen dataset that either the same or novel Finnish-enriched protein-altering variants would confirm the association of ANGPTL7 with protection from glaucoma. This additional ANGPTL7 association data, obtained in 6,537 glaucoma patients and 170,362 controls, provided strong support that protein-altering variants in ANGPTL7 protect against glaucoma (case definitions described in S7 Table). Specifically, we found that the p.Gln175His substitution has nominal evidence of association (P = 0.009, OR = 0.47) despite the variant only being present at a minor allele frequency of 0.1% in this Finnish cohort (8-fold depleted compared to UK Biobank). The remaining protein-altering variants in ANGPTL7 tested in UK Biobank were not found in the FinnGen dataset. Confirmation of an ANGPTL7 effect on glaucoma risk was seen in data from an independent Finnish-specific protein-altering missense substitution, p.Arg220Cys, which was strongly associated with protection from glaucoma (P = 2.0x10^-12, OR = 0.71 [95% CI: 0.64–0.78], S7 Fig). Of note, this observation is advantaged by the property that p.Arg220Cys is found at a greater than 50-fold increased frequency in Finland compared with other populations [16], reinforcing the value of isolated, bottlenecked populations in which the allele frequency spectrum is intensely concentrated on the minority of variants passing through the bottleneck.

While registry-based diagnoses in FinnGen do not yet contain detailed ophthalmologic records, a subset of 3,375 glaucoma patients had been recorded in special health care as having primary open-angle glaucoma (POAG). In this sub-group, a stronger effect was observed (P = 1.3x10^-8, OR = 0.68) versus those glaucoma cases without a definitive record of POAG (OR = 0.77 [95% CI: 0.67–0.88]), reminiscent of the stronger risk effects seen at the myocilin (MYOC) gene and other established genes in the POAG subgroup [17]. Furthermore, we find protective association to exfoliation glaucoma (P = 6.7x10^-5, OR = 0.64), primary angle-closure glaucoma (P = 0.0016, OR = 0.59), and normotensive glaucoma (P = 0.07, OR = 0.78, case n = 653, Fig 1).

Given the Finnish enrichment of the known strong glaucoma risk allele, p.Gln368Ter, in MYOC (MAF in Finland = 0.3%, MAF in Non-Finnish European = 0.16%, reference sequence: NM_00026), we next asked whether carriers have risk reduced if they carry ANGPTL7 p.Arg220Cys. In FinnGen, we estimate that 7.0% of carriers for MYOC p.Gln368Ter variant is POAG cases in comparison to 2% for non-carriers. In the presence of ANGPTL7 p.Arg220Cys, only 1.3% of individuals are POAG cases, and only 2 of 86 (2.3%) who carry both MYOC risk and ANGPTL7 protective variants were POAG cases (S8 Table). This suggests ANGPTL7 protection extends to the MYOC risk group but the small counts preclude any definitive statement regarding interaction (P = 0.318, for interaction term in a logistic regression model)—given the limited number of double-carriers, larger case-control series are needed to refine our understanding as to whether ANGPTL7 p.Arg220Cys variant modifies the glaucoma risk conferred by p.Gln368Ter in MYOC.

Access to genotype data in over 330,000 individuals from the UK and 176,000 from the Finnish group enabled us to identify rare protein-altering homozygotes. In UK Biobank, we found 28 individuals homozygous for the p.Gln175His allele, consistent with Hardy-Weinberg expectation (n = 22.6), where we estimated a -3.40 and -2.37 mmHg drop for corneal compensated and Goldman-correlated IOP, respectively, compared to the mean IOP levels. Furthermore, the oldest reached the age of 80 and one of the 28 died (age 65). In FinnGen we found 343 individuals homozygous for the p.Arg220Cys allele, the oldest reached the age of 98, with no depletion of homozygotes compared with Hardy-Weinberg equilibrium expectation. There was no significant association of the homozygous genotype with a decreased lifespan. We did not observe homozygous p.Arg220Cys association across disease endpoints in FinnGen (Bonferroni-corrected p > 0.05, S1 Data). Together this indicates that having two copies with p.Gln175His or p.Arg220Cys in ANGPTL7 is compatible with a normal lifespan. To assess the potential impacts of those protein-truncating variants on reproductive fitness, we assessed the association of p.Gln175His with the number of live births and the number of children fathered and found no significant association (P > 0.05/4, S9 Table). Through phenome-wide association analysis (PheWAS), we did not find any significant association for non-eye phenotypes (P>1.0 x 10⁻⁵ for both in UK Biobank and FinnGen, S10 Table, S2 and S3 Data). Hence, we did not find any severe medical consequences that would be of obvious concern in developing a therapeutic to mimic the effect of these alleles.

ANGPTL7, a five-exon protein-coding gene, encodes the Angiopoietin-related protein 7, which is expressed in several human tissues including the trabecular meshwork, cornea, and retina [18–20]. We examined proteomics expression data in normal tissues and cell lines from ProteomicsDB and MOPED[21,22] and found vitreous humor tissue-specific expression of ANGPTL7 (log10 ppm = 1, S8 Fig).

Discussion

This study establishes strong genetic evidence for the involvement of ANGPTL7 in glaucoma risk in which a powerful allelic series, including multiple low-frequency missense substitutions and a single premature stop-gain substitution, is conclusively associated with reduced disease risk and endophenotype-lowering effects. Our results highlight the benefit of rare protein-altering variant analysis using multiple large cohorts, especially when the population history of the participating cohort experienced a bottleneck, which enables enrichment of rare alleles as we report with the ANGPTL7 p.Arg220Cyc allele [13]. In Finland, the most common glaucoma subtypes are POAG and the secondary exfoliation glaucoma. The main difference in glaucoma prevalence is that in Finland the exfoliation glaucoma is much more prevalent (31%) than in the UK [23]. The prevalence of POAG is similar in Finland than in other European populations. The prevalence is heavily affected by age. In one Finnish cohort study, among individuals aged 70 years or older, the prevalence of POAG was approximately 7% [24]. Relative similar prevalence for POAG is reported in European populations [25]. Many patients with POAG are undiagnosed so the prevalence is affected by sampling methods (i.e. cohort or diagnosis reported). The population cohorts from founder populations enable future recall studies focusing on individuals homozygous for the allele, which can eventually improve our understanding of the mechanism by which ANGPTL7 disruption leads to protection to glaucoma risk and lowering of IOP. The discovery of two independent protein-altering alleles with directionally consistent effects from the two analyzed populations increases our confidence in the gene’s causal link to glaucoma.

Recent studies have associated ANGPTL proteins with cardiometabolic phenotypes [26–30]. Although it has been proposed that ANGPTL7 levels are increased in obesity (and reduced after physical exercise), we do not observe any evidence of genetic association in either UK Biobank or FinnGen to support this hypothesis [31].

ANGPTL7 overexpression in primary human trabecular meshwork cells was found to alter the expression of relevant trabecular meshwork proteins of the extracellular matrix, including fibronectin, collagens type I, IV, and V, myocilin, versican, and MMP1, and ANGPTL7 protein was increased as the disease progressed in POAG beagle dogs [18]. The tissue-specific protein expression data suggest that further work in dissecting the role of ANGPL7 in all possible cell types in the eye is warranted.

When combined with the previously-reported associations with IOP and glaucoma, our results provide compelling genetic evidence of the role of ANGPTL7 in glaucoma and its subtypes including exfoliation, primary open-angle, and primary angle-closure, which may come in contrast to prior findings with lack of overlap between POAG risk and IOP loci [32]. In the context of the other established variants in glaucoma, including the protein-truncating variants in MYOC, p.Gln175His and the 57-fold Finnish-enriched p.Arg220Cys variants in ANGPTL7 exert a comparable protective effect. While our genetic discovery provides compelling evidence of involvement of ANGPTL7 in glaucoma, several important questions remain to be answered before its eventual clinical translation. First, we were not able to assess whether the missense variants are complete loss-of-function, partial loss-of-function variants, dominant negative, or gain of function given the data we have at hand. Although we do have a predicted protein-truncating variant, p.Arg177Ter, with nominal evidence of association to IOP and an estimated effect consistent with the missense substitutions, it is challenging to draw conclusions about its functional consequence from in silico predictions, as we have reported in earlier studies assessing when PTVs trigger degradation pathways like nonsense-mediated decay [33]. Second, it is unclear in which cell types these variants are acting on to confer the protective and IOP lowering effects. We anticipate that ANGPTL7 may be acting in the trabecular meshwork given its high expression in both adult and fetal trabecular meshwork (> 3000 FPKM) [34], we see high expression in both adult and fetal cornea (>200 FPKM), which introduces some challenges as how we interpret its functional role, and we hypothesize that given its high expression in cornea it may be one reason why we see stronger evidence of association in IOP Goldman correlated measures compared to corneal compensated IOP. Additionally, future studies should assess whether ANGPTL7 variants modify the progression of glaucoma, for example, whether ANGPTL7 carriers are less likely to go from glaucoma diagnosis to potential surgery. Although we are aggregating these data, we are thus far unable to draw definitive conclusions.

Because of the strong protective effect associated with the ANGPTL7 protein-altering variants (S9 Fig), further studies of ANGPTL7 inhibition and the specific action of these variant proteins should be useful in understanding the mechanism by which glaucoma protection occurs and whether this reveals a promising therapeutic opportunity similar to that which has been realized from the examples of PCSK9, APOC3 and cardiovascular disease [35–37]. Given the rapidly evolving field of gene editing and siRNA, we can only speculate that if the effect is truly loss-of-function and that gene inhibition is an appropriate strategy then these therapeutic modalities will be especially relevant. Therapeutic delivery is also a complicated challenge. Although injection to the eye is currently commonplace in practice, it is unclear whether different therapeutic modalities, e.g. antibody, siRNA, CRISPR, base-editing would be appropriate, and whether the duration of the treatment would be sufficiently durable to be effective to prevent extremely frequent injections or competitive against current therapeutic modalities. New drug delivery technologies are of interest and it is clear that a durable and efficient mode of delivery that mimics the protective effect of these mutations is an attractive strategy. Our genetic data from ANGPTL7 homozygotes with up to a 69% risk reduction for all glaucoma and 80% risk reduction for primary open-angle glaucoma suggest that this is likely to be a safe and effective strategy for therapeutic intervention.

Methods

Supporting information

Acknowledgments

This research has been conducted using the UK Biobank Resource under Application Number 24983, “Generating effective therapeutic hypotheses from genomic and hospital linkage data” (http://www.ukbiobank.ac.uk/wp-content/uploads/2017/06/24983-Dr-Manuel-Rivas.pdf). Based on the information provided in Protocol 44532 the Stanford IRB has determined that the research does not involve human subjects as defined in 45 CFR 46.102(f) or 21 CFR 50.3(g). All participants of UK Biobank provided written informed consent (more information is available at https://www.ukbiobank.ac.uk/2018/02/gdpr/). For the Finnish Institute of Health and Welfare (THL) driven FinnGen preparatory project (here called FinnGen), all patients and control subjects had provided informed consent for biobank research, based on the Finnish Biobank Act. Alternatively, older cohorts were based on study specific consents and later transferred to the THL Biobank after approval by Valvira, the National Supervisory Authority for Welfare and Health. Recruitment protocols followed the biobank protocols approved by Valvira. The Ethical Review Board of the Hospital District of Helsinki and Uusimaa approved the FinnGen study protocol Nr HUS/990/2017. The FinnGen preparatory project is approved by THL, approval numbers THL/2031/6.02.00/2017, amendments THL/341/6.02.00/2018, THL/2222/6.02.00/2018 and THL/283/6.02.00/2019. All DNA samples and data in this study were pseudonymized. We thank all the participants in the UK Biobank and Finnish Biobanks used in FinnGen studies. We thank Sirpa Soini, Fatima Rodriguez and David Amar for invaluable feedback on the manuscript.