Comprehensive Evaluation of the Genetic Basis of Keratoconus: New Perspectives for Clinical Translation

Abstract

Purpose: Keratoconus (KC) is a corneal disorder with complex etiology, apparently involving both genetic and environmental factors, characterized by progressive thinning and protrusion of the cornea. We aimed to identify novel genetic regions associated with KC susceptibility, elucidate relevant genes for disease development, and explore the translational implications for therapeutic intervention and risk assessment.

Methods: We conducted a genome-wide association study (GWAS) that integrated previously published data with newly generated genotyping data from an independent European cohort. To evaluate the clinical translation of our results, we performed functional annotation, gene prioritization, polygenic risk score (PRS), and drug repositioning analyses.

Results: We identified two novel genetic loci associated with KC, with rs2806689 and rs807037 emerging as lead variants (P = 1.71E-08, odds ratio [OR] = 0.88; P = 1.93E-08, OR = 1.16, respectively). Most importantly, we identified 315 candidate genes influenced by confirmed KC-associated variants. Among these, MINK1 was found to play a pivotal role in KC pathogenesis through the WNT signaling pathway. Moreover, we developed a PRS model that successfully differentiated KC patients from controls (P = 7.61E-16; area under the curve = 0.713). This model has the potential to identify individuals at high risk for developing KC, which could be instrumental in early diagnosis and management. Additionally, our drug repositioning analysis identified acetylcysteine as a potential treatment option for KC, opening up new avenues for therapeutic intervention.

Conclusions: Our study provides valuable insights into the genetic and molecular basis of KC, offering new targets for therapy and highlighting the clinical utility of PRS models in predicting disease risk.

Figure 1.

(A, B) Regional plots for the novel keratoconus association signals identified in this study: rs2806689 (A) and rs807037 (B). The plots display −log₁₀ P values for SNPs within a region flanking 500 kb on both sides of the top SNPs. Lead variants are highlighted in purple, and the r² values of other variants are represented by different colors. Genes in the region and their transcription direction (arrows) are indicated. The y-axis shows the recombination rates, aligned with chromosomal positions on the x-axis.

Figure 2.

PRS models assessed in this study. (A–C) Receiver operating characteristic (ROC) curves (left) and distribution of PRS values for KC cases and controls in the Spanish testing cohort (right) are shown for the models considering KC data from Hardcastle et al. (A), CCT data from Choquet et al. (B), and KC and CCT combined data from both studies (C).

Figure 3.

Schematic representation of the computational process for functionally annotating and prioritizing KC-associated signals. The meta-analysis revealed 119 independent SNP associations with KC (P < 5E-08, r² < 0.6). A total of 4001 candidate SNPs were considered due to their linkage with the initial 119 variants (r² ≥ 0.6). The candidate SNPs were further categorized based on their impact on genes, distinguishing between coding and non-coding effects. Each gene affected by these SNPs received a score according to the following criteria: (1) For genes influenced by SNPs with coding effects, points were assigned as follows—1 point for a probability of loss-of-function intolerance (pLI) ≥ 0.9, 1 point for any associated SNP with a deleteriousness score (Combined Annotation Dependent Depletion [CADD]) ≥ 10, and 2 points for CADD ≥ 20; also, depending on the type of SNP effect, 1 point for synonymous or non-reading frame-changing mutations and 2 points for non-synonymous or reading frame-changing mutations. (2) For genes influenced by SNPs with non-coding effects, points were allocated based on CADD and pLI values as before—1 point for a non-coding residual variation intolerance score (ncRVIS) < 0, 1 point if the gene was protein coding, 1 point if there was an expression quantitative trait locus (eQTL) effect, 1 point if the associated SNP had a RegulomeDB (RDB) score of 3/4 and 2 points for scores of 1/2, and 1 point if the gene was mapped by chromatin interaction (CI) mapping. According to the scores, the genes were classified into three tiers: tier 1, score 6 to 9 for non-coding effects and 4 or 5 for coding effects; tier 2, score 3 to 5 for non-coding effects and 2 or 3 for coding effects; and tier 3, score 0 to 2 for non-coding effects and 0 or 1 for coding effects. Subsequently, genes in tiers 1 and 2 underwent PPI and biological pathway enrichment analyses using STRING and were subjected to drug repositioning analysis. In the STRING-derived image, confirmed genes associated with KC are depicted in red, and those associated with intraocular pressure measurement are shown in blue.