Polygenic risk score-based phenome-wide association for glaucoma and its impact on disease susceptibility in two large biobanks

Our analysis was conducted in two steps. First, we performed a PheWAS of POAG PRS with all available disease phenotypes in the UK Biobank (discovery set) and in the Penn Medicine Biobank (replication set). Second, we focused on the association of the POAG PRS with ocular factors and incident glaucoma, using a prospective study design within the UK Biobank.

Genotyping and quality control (QC) procedures and imputation followed the standard practices and were performed per cohort-genotyping platform pair. Further details are described in Additional file 1: Method S1.

To quantify genetic risk for glaucoma, we generated a PRS using summary statistics from the GWAS conducted by the International Glaucoma Genetics Consortium (IGGC) [13], which consisted of 12,713,176 variants. The GWAS was conducted with a fixed-effects meta-analysis of 15,229 POAG cases and 177,473 controls of European descent excluding UK Biobank samples and the summary statistics are available via GWAS Catalog [23], under the study accession identifier GCST90011767.

The PRS was constructed using PRS-CS (version 1.0.0) [24], which is a Bayesian polygenic method that infers the posterior mean effect size of each variant based on GWAS summary statistics and the linkage disequilibrium reference panel. The individual PRSs were computed from beta coefficients as the weighted sum of the risk alleles by applying PLINK version 1.90 [25]. Finally, 1,116,933 SNPs that overlapped with the HapMap3 (1,287,078 SNPs) database were utilized for the POAG PRS.

The “PheWAS” R package (version 0.99.5.5) was used to perform PheWAS analyses [26]. In these analyses, POAG PRS was set as the independent variable, and disease phenotypes as dependent variables, with age, sex, genotyping array, and first 10 genetic principal components (PCs) of ancestry as covariates. Disease diagnosis category phenotypes were obtained by mapping the International Classification of Diseases, Ninth or Tenth Revision (ICD-9 and -10) diagnosis codes of the UK Biobank to 1,618 hierarchical phenotypes (PheCodes) categorized into 17 disease categories [26, 27]. We removed phenotypic codes with less than 100 cases and those concerning symptoms, injuries, and poisoning, resulting in 776 phenotypes in 15 disease categories that were included in our analysis. Of these, 767 were eligible for replication analysis in the Penn Medicine Biobank. The number of cases for each phenotype and detailed demographic information in each cohort are summarized in Additional file 1: Table S1. To account for multiple testing, we used a Bonferroni threshold of P < 0.0000644 (0.05/776) when determining significance among the results of our main analyses in the UK Biobank and P < 0.0000652 (0.05/767) for the replication analyses. Next, to obtain insight into comorbid conditions associated with the POAG phenotype, we conducted additional PheWAS analyses using a diagnosis of POAG as the independent variable.

Out of the initial cohort of 502,409 UK Biobank participants across six assessment centers in the United Kingdom, a total of 377,852 participants were available in the data release following thorough QC. Within this subset, cases with accessible ophthalmic data, including spherical power, cylinder power, CH, corneal resistance factor, and IOP were identified and included (n = 83,804). To avoid the effect of surgery on refractive error, IOP, and CH, participants were excluded if they reported having received glaucoma surgery, cornea refractive surgery, cataract surgery, or if they had previous eye injury in either eye (n = 3455). Finally, cases with baseline glaucoma diagnosis were excluded from the analysis (n = 6801). Consequently, the study focused on a refined cohort of 73,548 participants for the investigation of the association between the POAG PRS and the occurrence of incident glaucoma. The flowchart showing participants included for analysis is shown in Additional file 1: Fig. S1.

Medical history of hypertension, diabetes, and dyslipidemia at baseline was based on the self-report collected in an in-person interview at enrollment or on diagnostic and procedure codes in electronic health records. The definitions of baseline dyslipidemia, hypertension, and type 2 diabetes are described in Additional file 1: Table S2.

All participants who had available spherical equivalent (SE), Goldmann-correlated IOP (IOP_g), and CH were used for this analysis. Participants underwent a detailed ophthalmologic examination in which non-cycloplegic refraction status was determined using an autorefractor (Tomey, Nagoya, Japan). The SE of refractive error was defined as the sphere component plus half the cylinder component. After measuring visual acuity and refraction, CH, corneal resistance factor, and IOP_g were measured with the Reichert Ocular Response Analyzer (ORA, Reichert, Inc., Depew, NY, USA). The average of the ORA pressure values was calibrated against Goldmann applanation tonometer measures to derive IOP_g. Participants who had eye surgery within the previous 4 weeks or those with possible eye infections were precluded from the determination of IOP. Measurements were excluded from analysis if the participants had a history of corneal surgery, refractive surgery, or injuries in either eye; additionally, participants with a history of glaucoma surgery were excluded and left eyes with missing data were excluded.

Participants were identified as having glaucoma if they self-reported glaucoma on the eye problems/disorders (UK Biobank data field: 6148) or non-cancer illness (UK Biobank data field: 20002), or had an ICD-9/10 diagnosis code for POAG, other glaucoma, or glaucoma, unspecified, based on the use of these codes in prior in population-based and registry-based studies of genetic risk for glaucoma [28, 29]. (Additional file 1: Table S2).

In the prospective study conducted within the UK Biobank, measures from the left eye were used as the outcome variable, consistent with previous reports [30, 31]. In the UK Biobank study, the measurement of the left eye was done, as per the study protocol, following the collection of right eye data. This sequencing choice suggests that left eye data in our cohort may be less susceptible to artifacts, such as blinking. Reference was made to previous authoritative articles that reported IOP and CH in the UK Biobank, where analyses were based on left-eye data. Continuous variables are reported as means with standard deviations (SD) and categorical variables as frequencies and proportions. In some analyses, individuals were classified according to the degree of myopia. Myopia overall was defined as having a SE of ≤ − 0.5D and mild myopia, moderate myopia, and high myopia as − 3.0D < SE ≤ − 0.5D, − 6.0D < SE ≤ − 3.0D, and SE ≤ − 6.0D, respectively. Regarding CH, participants with a value of ≤ 10.1 mmHg were considered to have low CH; this cutoff was determined to be significantly associated with glaucoma prevalence in a previous study based on UK Biobank data [31]. Those with a value > 10.1 mmHg were classified as having high CH. In terms of IOP, individuals were categorized into two groups: those with ocular hypertension (IOP_g ≥ 21 mmHg) and those without (IOP_g < 21 mmHg).

As done in previous studies, we categorized study participants based on the PRS into low, intermediate, or high genetic risk groups [32, 33], and also further classified the top 1% of the PRS distribution as a very-high-risk group in light of the curve of cumulative incidence of disease prevalence over the PRS distribution (Fig. 1). Thus, participants were categorized into the following four risk subgroups: low, 0–19th percentile; intermediate, 20–79th percentile; high; 80–98th percentile; and very high, 99th percentile.

The baseline clinical characteristics of participants were compared using analysis of variance (ANOVA) for continuous variables and the Chi-square test for categorical variables. To evaluate the association of glaucoma genetic risk with SE, IOP_g, and CH, we applied multivariable logistic and linear regression analyses after adjusting for age, sex, genotyping array, and the first 10 PCs of ancestry, and other factors determined on the basis of previous studies [30, 31]. We performed a Cox proportional hazards regression analysis, utilizing genetic risk and ocular factors, with age at glaucoma onset or age at the last clinical visit as time variables and the diagnosis of glaucoma as a status and calculated hazard ratio (HR). Kaplan–Meier survival curves and standardized cumulative incidence according to categories of PRS for POAG were plotted to assess potential disparities in survival among genetic risk groups. Subsequently, we conducted joint association analyses to investigate the interplay between genetic and ocular factors on the development of incident glaucoma. Statistical tests were two-sided, and P < 0.05 was considered statistically significant. All statistical analyses were conducted using R version 3.9.0.

The UK Biobank (discovery set) sample comprised 53.7% female (mean age: 56.9 ± 7.8) individuals of white-British ancestry, and the Penn Medicine Biobank (replication set) sample included 45.2% female (mean age: 57.4 ± 16.2) of white-European ancestry. In the PheWAS analyses, higher genetic predisposition to POAG was exclusively associated with ocular disease phenotype. In the discovery set, we observed significant associations between POAG PRS and 10 disease phenotypes, which remained significant after Bonferroni corrections; these phenotypes were glaucoma, POAG, celiac disease, primary angle-closure glaucoma, myopia, unspecified diffuse connective tissue disease, disorders of iris and ciliary body, cataract, retinal detachments, and senile cataract. Disease phenotypes categorized as involving sensory organs, which includes eye-related disorders, displayed positive associations with POAG PRS, while celiac disease and unspecified diffuse connective tissue disease showed negative associations (Fig. 2A). Of the 10 disease phenotypes that exhibited significant associations in the discovery set, eight demonstrated the same direction of effect in the replication set; however, only two, namely glaucoma and POAG, remained significant after Bonferroni’s correction (Fig. 2B). A summary of the 776 disease phenotypes included in the PheWAS and their odds ratios (ORs) are presented in Additional file 2: Table S3. In contrast, when we conducted PheWAS analyses with a diagnosis of POAG as the independent variable, we identified a notable overrepresentation of multiple systemic disease phenotypes, across the spectrum of disease categories (Additional file 1: Fig. S2 and Additional file 2: Table S4).

Demographic baseline information for all 73,548 participants and systemic and ocular characteristics for each PRS group are presented in Table 1. Overall, the prevalence of systemic diseases, including diabetes, hypertension, and dyslipidemia, and the proportion of current smokers and alcohol consumers, were not significantly different across the POAG genetic risk groups. Participants at higher genetic risk for glaucoma tended to have a lower body mass index (BMI) than those at low genetic risk (P < 0.001). Importantly, there were distinctive differences in the ocular characteristics across the genetic risk groups, with the trend showing that higher genetic risk accompanied higher IOP, lower SE, and lower CH (all P < 0.001). Compared with the low genetic risk group, the very high-risk category was associated with high IOP in multivariable logistic regression analyses (OR, 3.84; 95% confidence interval (CI) 2.98–4.89), myopia (OR, 1.28; 95% CI 1.06–1.52) and low CH (OR, 1.27; 95% CI 1.06–1.51). (Table 2; Results of multivariable linear regression analyses are shown in Additional file 1: Table S5).

The median duration of follow-up was 11.1 years, during which 1300 participants were identified as incident glaucoma cases. Compared with low genetic risk, higher genetic risk was associated with a higher risk of incident glaucoma during follow-up (P for trend < 0.001) (Table 3; Kaplan–Meier survival curves and cumulative incidences are shown in Fig. 3 and Additional file 1: Fig. S3 and S4). After adjusting for potential confounding factors, a 1-SD increase in POAG PRS was associated with 64% greater risk of incident glaucoma (HR, 1.64; 95% CI 1.54–1.75; P < 0.001). Notably, participants in the very high-risk group were 8.76 times more likely to have glaucoma than those in the low-risk group (Table 3).

In comparison to individuals exhibiting IOP < 21 mmHg, those with IOP ≥ 21 mmHg demonstrated an increased risk of incident glaucoma (HR, 6.74; 95% CI 5.89–7.71; P < 0.001). Notably, participants with moderate to high myopia showed an increased risk of glaucoma compared to those with non to mild myopia (HR, 1.59; 95% CI 1.36–1.88; P < 0.001). Furthermore, participants with CH ≤ 10.1 mmHg showed an increased risk of glaucoma compared to those with CH > 10.1 mmHg (HR, 1.24; 95% CI 1.09–1.42; P < 0.001) (Table 4 and results of continuous scale are shown in Additional file 1: Table S6).

To explore the effect of ocular risk factors on glaucoma risk according to genetic risk, we stratified the ocular factors by PRS category (Fig. 4). We observed a monotonic association between increasing PRSs and ocular factors and a higher risk of incident glaucoma. In particular, the participants with the very high-PRS and high IOP had the highest risk for incident glaucoma (HR, 28.85; 95% CI 16.36–50.88; P < 0.001). In addition, the participants with the very high-PRS and moderate-high myopia exhibited the substantial risk for incident glaucoma with HR of 11.88 (95% CI 5.45–25.88), and those with very high-PRS and low CH with HR of 9.20 (95% CI 5.23–16.16; P < 0.001).