Integrating Glaucoma Endophenotypes Improves Polygenic Risk Prediction for Primary Open-Angle Glaucoma Across Ancestries

This study demonstrates that a multi-trait polygenic probability risk score (PPRS) framework, which integrates genetic data from glaucoma endophenotypes like intraocular pressure and optic nerve structure, significantly improves primary open-angle glaucoma prediction accuracy across European and Latino ancestries while revealing distinct trait-specific risk drivers for each population.

The Gist

Imagine your eyes are like a high-tech security system for your home. Glaucoma is a silent thief that slowly steals your vision, often without you noticing until it's too late. The most common type, called Primary Open-Angle Glaucoma (POAG), is tricky because it doesn't just happen for one reason; it's a complex crime involving many different "suspects" (genetic factors).

For a long time, scientists tried to predict who might get this "thief" by looking at a single suspect: Intraocular Pressure (IOP), or how much fluid is pressing against the eye. It's like trying to predict a house fire by only checking the smoke detector, ignoring the faulty wiring or the gas leak.

This new paper introduces a smarter, more comprehensive security system. Here is the breakdown in simple terms:

1. The Old Way: A Single-Lock System

Previously, scientists built a "Polygenic Risk Score" (PRS) like a single key. They looked at thousands of genetic clues related to eye pressure and gave people a score. If your key was "high risk," you were more likely to get glaucoma.

• The Problem: This key only opened one door. It missed other important clues, like the shape of the optic nerve or the thickness of the nerve layer. It was like trying to guess the weather by only looking at the temperature, ignoring humidity and wind.

2. The New Way: The "Super-App" Dashboard

The researchers (led by Dr. Xiaoyi Raymond Gao) built a new tool called a Multi-Trait Polygenic Probability Risk Score (PPRS).

• The Analogy: Imagine a weather app that doesn't just show the temperature. It combines data on temperature, humidity, wind speed, and barometric pressure to give you a single, highly accurate "Storm Probability."
• How it works: Instead of just looking at eye pressure, this new system looks at four different genetic "suspects" simultaneously:
  1. POAG: The general history of glaucoma in the family.
  2. IOP: Eye pressure.
  3. VCDR: The shape of the optic nerve cup (like the size of a crater).
  4. RNFL: The thickness of the nerve fiber layer (like the thickness of a cable).

By combining these four clues into one "Super Score," the system gets a much clearer picture of the risk.

3. The "One Size Does Not Fit All" Discovery

Here is the most fascinating part of the study. The researchers tested this new system on two very different groups of people: Europeans and Latinos.

• The Discovery: They found that the "suspects" driving the crime were different for each group.
  • For Europeans: The biggest driver was Eye Pressure (IOP). It was the main suspect.
  • For Latinos: The biggest driver was the Shape of the Nerve (VCDR). Eye pressure was less of a factor here.
• The Metaphor: Imagine two different houses. In House A (Europeans), the fire starts because of a faulty gas line (Pressure). In House B (Latinos), the fire starts because of old, dry wood (Nerve Shape). If you only check the gas line in House B, you miss the real danger. This new system knows to check both but realizes which one is the "boss" for each specific house.

4. The Results: Catching the Thieves Early

The results were impressive.

• Better Prediction: The new "Super-App" was much better at predicting who would get glaucoma than the old single-key system. It correctly identified high-risk individuals much more often.
• Extreme Risk: If you were in the top 10% of risk scores (the "highest danger zone"), you were 74 times more likely to get glaucoma than someone in the bottom 10% for Europeans, and 49 times more likely for Latinos.
• Capturing the Cases: The new system found about 66% of all glaucoma cases just by looking at the top 20% of the population. That means if doctors screened just the top 20% of people with the highest scores, they could catch the vast majority of future cases early.

Why This Matters

Think of this as moving from a generic warning sign ("Beware of thieves!") to a personalized security plan.

• Personalized: It tells doctors that for a Latino patient, they should pay extra attention to the shape of the optic nerve, while for a European patient, they should watch the pressure closely.
• Early Intervention: Because the system is so good at spotting the "high-risk" people, doctors can start checking their eyes more often before they lose vision.
• Fairness: It proves that genetic risk isn't the same for everyone. By studying diverse groups (like Latinos), the researchers ensured the tool works for more than just one type of person.

In a nutshell: This paper says, "Stop guessing with just one clue. Look at the whole picture, and realize that different people have different risk profiles. By doing this, we can catch glaucoma earlier and save more sight."

Technical Summary

1. Problem Statement

Primary Open-Angle Glaucoma (POAG) is a leading cause of irreversible blindness. While Polygenic Risk Scores (PRSs) offer a method for genetic risk stratification, current approaches face three critical limitations:

• Single-Trait Limitation: Existing PRSs rely on single-trait models (e.g., POAG diagnosis only), failing to capture complementary genetic information from quantitative endophenotypes like Intraocular Pressure (IOP), Vertical Cup-to-Disc Ratio (VCDR), and Retinal Nerve Fiber Layer (RNFL) thickness.
• Methodological Flaws in Multi-Trait Analysis: Previous multi-trait approaches (e.g., MTAG) often assume a uniform variance-covariance matrix across all SNPs for all traits, potentially leading to biased effect size estimates and false positives.
• Limited Genomic Coverage and Ancestry Bias: Many studies use limited genomic coverage (<1M markers) and lack integration of functional annotations. Furthermore, most PRS models are trained on European ancestry data, resulting in poor portability and reduced accuracy in diverse populations (e.g., Latino/Admixed American).

2. Methodology

The study employed a two-stage design to develop and validate a Multi-Trait Polygenic Probability Risk Score (PPRS) framework.

Study Populations:

• Discovery/Training: Utilized large-scale GWAS summary statistics for POAG, IOP, VCDR, and RNFL from diverse sources (IGGC, UK Biobank, etc.).
• Target Cohorts:
  • UK Biobank (UKB): 324,713 individuals of European ancestry (2,168 POAG cases).
  • Mexican American Glaucoma Genetic Study (MAGGS): 4,549 individuals of Latino ancestry (275 POAG cases).

PRS Construction:
The authors constructed four distinct trait-specific PRSs using SBayesRC, a method that leverages functional genomic annotations (96 annotations, including regulatory elements and chromatin states) and >7 million variants to improve prediction accuracy:

1. PRS-POAG: Based on POAG diagnosis.
2. PRS-IOP: Based on Intraocular Pressure.
3. PRS-VCDR: Based on Vertical Cup-to-Disc Ratio.
4. PRS-RNFL: Based on Retinal Nerve Fiber Layer thickness (derived from a new GWAS conducted by the authors).

Note: A traditional Clumping and Thresholding (C+T) PRS and an MTAG-derived PRS were also constructed for benchmarking.

The PPRS Framework:
Instead of collapsing traits into a single score, the authors integrated the four trait-specific PRSs into a penalized (L2) logistic regression model.

• Inputs: The four standardized PRSs, adjusted for age and sex.
• Output: A predicted probability of POAG (the PPRS).
• Validation: Models were trained using stratified 10-fold cross-validation. Performance was evaluated using Area Under the Curve (AUC), Odds Ratios (OR) across deciles, and risk stratification metrics.

3. Key Contributions

• Novel Framework: Introduction of a multi-trait PPRS that preserves the individual contribution of each endophenotype, allowing for the dissection of ancestry-specific genetic architectures.
• Enhanced Genomic Resolution: Utilization of SBayesRC with >7 million variants and 96 functional annotations, significantly outperforming traditional C+T methods.
• Cross-Ancestry Validation: Successful application and validation in both European (UKB) and Latino (MAGGS) cohorts, addressing the lack of diversity in glaucoma genetics.
• Ancestry-Specific Insights: The framework revealed that different endophenotypes drive genetic risk in different populations, a nuance lost in single-score or MTAG approaches.

4. Key Results

Predictive Performance (AUC):

• European Ancestry (UKB):
  • Baseline (Age/Sex): 0.721
  • Best Single-Trait (SBayesRC-POAG): 0.790
  • Full PPRS Model: 0.814 (Significant improvement over baseline and single-trait models).
• Latino Ancestry (MAGGS):
  • Baseline (Age/Sex): 0.753
  • Best Single-Trait (SBayesRC-POAG): 0.782
  • Full PPRS Model: 0.802 (Significant improvement over baseline and single-trait models).

Risk Stratification:

• Extreme Deciles: Individuals in the top PPRS decile had drastically higher odds of POAG compared to the bottom decile:
  • Europeans: 74.4-fold increase (OR = 74.4).
  • Latinos: 49.3-fold increase (OR = 49.3).
• Case Capture: The top quintile of the PPRS captured 65.7% of POAG cases in Europeans and 62.2% in Latinos, a substantial improvement over previous studies (which typically captured ~40-50%).

Ancestry-Specific Genetic Drivers:
Multivariate analysis revealed distinct drivers of risk:

• Europeans: IOP was the strongest predictor (OR = 1.63, P = 5.37 × 10⁻⁸⁹).
• Latinos: VCDR was the strongest predictor (OR = 1.64, P = 2.04 × 10⁻¹¹), while IOP showed a non-significant trend (P = 0.07). This aligns with epidemiological data showing a higher prevalence of normal-tension glaucoma in Latino populations.

5. Significance and Clinical Implications

• Improved Early Detection: The PPRS framework significantly enhances the ability to identify high-risk individuals, potentially facilitating earlier screening and intervention before irreversible vision loss occurs.
• Personalized Medicine: By decomposing polygenic risk into trait-specific components, the study demonstrates that a "one-size-fits-all" PRS is insufficient. Risk stratification strategies must be tailored to ancestry, as the biological pathways (IOP-driven vs. structural VCDR-driven) differ between populations.
• Methodological Advancement: The study proves that integrating multiple endophenotypes at the prediction model level (rather than the GWAS summary statistic level) avoids the statistical pitfalls of MTAG while capturing complex genetic architectures.
• Public Health Impact: Identifying the top decile of risk could allow for targeted resource allocation in screening programs, potentially reducing the global burden of glaucoma-induced blindness.

Limitations Noted:
The study was limited to European and Latino populations; future work needs to include African, East Asian, and South Asian cohorts. Additionally, the sample size for the Latino cohort (MAGGS) was smaller, which may limit the precision of effect estimates for that group. The model currently focuses on common variants, excluding rare variants.