Polygenic risk scores enhance the identification of carriers of monogenic forms of idiopathic pulmonary fibrosis

This study demonstrates that incorporating polygenic risk scores (PRS) for idiopathic pulmonary fibrosis into clinical models significantly improves the identification of patients carrying rare deleterious variants, thereby offering a valuable criterion for prioritizing individuals for genetic testing.

The Gist

The Big Picture: Solving the "Pulmonary Fibrosis" Puzzle

Imagine Idiopathic Pulmonary Fibrosis (IPF) as a mysterious, slow-acting fire that scars the lungs. It's a scary disease with no known cure, and doctors often struggle to predict exactly how fast it will burn or who is most at risk.

For a long time, scientists knew there were two main "arsonists" causing this fire:

1. The "Rare" Arsonist: A single, powerful genetic mutation (like a lit match dropped directly into dry grass). This happens in a small group of people and is often inherited from parents.
2. The "Common" Arsonist: A thousand tiny, weak sparks (common genetic variations) that build up over a lifetime. Alone, they do nothing, but together, they create a smoldering risk.

The Problem: Doctors want to find the "Rare" arsonists (the single powerful mutations) because identifying them helps them treat patients better and warn their families. However, finding these rare mutations is like looking for a needle in a haystack. It's expensive and time-consuming to test everyone's DNA.

The Old Strategy: Doctors used to say, "We will only test for the rare mutation if the patient has a family history of the disease or got sick very young."
The Flaw: This misses a lot of people. About 10% of patients have the rare mutation but don't have a family history or didn't get sick young. They get missed.

---

The New Idea: The "Genetic Weather Forecast"

This paper introduces a clever new tool called a Polygenic Risk Score (PRS).

Think of the PRS as a genetic weather forecast.

• If you have a high PRS, it means your "genetic weather" is full of those tiny, common sparks. You are very likely to get the disease because of this background noise.
• If you have a low PRS, your "genetic weather" is clear. You don't have those thousands of tiny sparks.

The Big Discovery: The researchers found a funny inverse relationship.

• If your genetic weather is stormy (High PRS), you probably got the disease from the "Common" sparks. You likely don't have the "Rare" match.
• If your genetic weather is clear (Low PRS), but you still have the disease, it's highly suspicious that you must have dropped that "Rare" match.

The Analogy: Imagine you find a house on fire.

• If the whole neighborhood is on fire (High PRS), it was probably a power grid failure (Common variants). You don't need to look for a specific arsonist.
• If the neighborhood is perfectly safe, but your house is burning, someone must have deliberately set a fire (Rare variant). You need to find that specific person immediately.

What Did They Do?

The researchers took data from nearly 1,400 IPF patients (from the US and UK). They built a computer model that combined two things:

1. The Patient's History: Did they get sick young? Do they have short telomeres (a sign of aging DNA)? Do they have family members with lung issues?
2. The Genetic Weather Forecast (PRS): How many common "sparks" do they carry?

They tested if adding the "Weather Forecast" to the "History" helped them find the "Rare Arsonists" (the people with the dangerous single mutations).

The Results: A Better Net

The results were promising:

• The Old Net: Using only family history and age, the model was okay at catching the "Rare Arsonists," but it missed about 40% of them.
• The New Net: When they added the "Genetic Weather Forecast" (PRS) to the mix, the model got much better. It successfully identified 22% more of the people who carried the dangerous rare mutations.

It's like upgrading from a fishing net with big holes to one with smaller holes. You catch more of the specific fish you are looking for without needing to catch every single fish in the ocean.

Why Does This Matter?

1. Saving Lives: If we find these "Rare Arsonists" earlier, we can monitor them more closely. Some of these mutations cause other problems (like blood or liver issues), so knowing about them helps doctors treat the whole person, not just the lungs.
2. Family Safety: If a patient has a rare mutation, their children and siblings might have it too. This allows families to get tested early and take precautions.
3. Smarter Testing: Instead of testing every single IPF patient (which is expensive), doctors could use this "Weather Forecast" to prioritize who gets tested first. If your forecast says "Clear skies" but you have the disease, you are a top priority for testing.

The Catch (Limitations)

The authors are honest that this isn't a magic wand yet.

• Cost: Right now, getting the full "weather forecast" (whole-genome sequencing) is expensive and not standard in every doctor's office.
• Imperfect: The model isn't perfect; it's still better than guessing, but it's not 100% accurate.
• Future Work: They need to figure out how to make this cheap and easy enough for regular clinics to use.

The Bottom Line

This study suggests that by looking at the "background noise" of a patient's genes (the common variants), doctors can actually get a better clue about who is carrying the "loud" dangerous mutations. It's a smarter way to hunt for the needles in the haystack, potentially leading to better care and earlier warnings for patients and their families.

Technical Summary

1. Problem Statement

Idiopathic Pulmonary Fibrosis (IPF) is a progressive, fatal lung disease with a complex genetic architecture involving both common variants (polygenic risk) and rare deleterious variants (RDVs) in monogenic genes (e.g., TERT, TINF2, SFTPC).

• The Clinical Gap: Current guidelines for genetic testing rely on clinical criteria (early onset <50 years, family history, severe telomere shortening). However, a significant portion of IPF patients carrying RDVs do not meet these criteria and remain undiagnosed.
• The Hypothesis: There is an inverse relationship between polygenic risk and rare variant carrier status. Patients with a high burden of common risk variants (high Polygenic Risk Score, or PRS) are less likely to carry a high-impact rare variant, and vice versa.
• The Objective: To evaluate whether integrating IPF-specific Polygenic Risk Scores (PRS) with standard clinical variables improves the identification of IPF patients who are carriers of monogenic rare variants, thereby optimizing patient prioritization for genetic sequencing.

2. Methodology

Study Cohorts

• Discovery Cohort: 888 IPF patients from the Pulmonary Fibrosis Foundation Patient Registry (PFF-PR) (USA).
• Validation Cohort: 472 IPF patients from the PROFILE cohort (UK).
• Data Type: Whole-Genome Sequencing (WGS) data was available for both cohorts.

Definitions and Variables

• Qualifying Variants (QVs): Defined as rare variants (MAF < 0.0005) in 13 monogenic PF genes (*TERC, TERT, TINF2, DKC1, RTEL1, PARN, NAF1, ZCCHC8, SFTPC, SFTPA1, SFTPA2, SPDL1, KIF15*). Variants were filtered based on read depth, mapping quality, and predicted deleteriousness (CADD > 15 for missense; ClinVar pathogenicity for TERC).
• Clinical Model: Included three standard variables: Age at onset (<50 years), Telomere Length (TL) (shortest 10th percentile), and Family History of Interstitial Lung Disease (ILD).
• Polygenic Risk Scores (PRS): Three models were constructed using GWAS summary statistics from a large IPF meta-analysis:
  1. Sentinel-PRS: Based on 19 genome-wide significant IPF risk variants.
  2. WG-PRSC+T: Whole-Genome PRS using Clumping and Thresholding (C+T) method.
  3. WG-PRSSBayesRC: Whole-Genome PRS using a Bayesian method (SBayesRC) incorporating functional annotations.

Statistical Analysis

• Association: Logistic regression models assessed the association between PRS and QV carrier status, adjusted for principal components (population stratification) and clinical covariates.
• Predictive Performance: Models were compared using Area Under the Curve (AUC) and Receiver Operating Characteristic (ROC) analysis.
• Model Comparison: The "Integrative Model" (Clinical + PRS) was compared against the "Clinical-Only Model" using DeLong's test.
• Reclassification: Net Reclassification Index (NRI) and Decision Curve Analysis (DCA) were used to evaluate how many carriers were correctly reclassified into high-risk groups when PRS was added.
• Sensitivity Analyses: Performed by excluding the MUC5B locus (the strongest common risk variant) and using stricter definitions of QVs.

3. Key Results

Association Between PRS and Carrier Status

• Inverse Relationship: A strong inverse association was found between IPF-PRS and QV carrier status.
  • WG-PRSC+T: OR = 0.65 (95% CI 0.53–0.79).
  • WG-PRSSBayesRC: OR = 0.71 (95% CI 0.59–0.86).
  • Sentinel-PRS: OR = 0.77 (95% CI 0.63–0.94).
• Quintile Analysis: Patients in the highest PRS quintile had significantly lower odds of being carriers compared to those in the lowest quintile (e.g., WG-PRSC+T highest vs. lowest: OR = 0.16).
• Validation: In the PROFILE cohort, the association held specifically for carriers of telomere-related gene variants, though the signal was weaker for overall QVs due to the lack of family history data in this cohort.

Predictive Performance (AUC)

• Clinical-Only Model: Modest performance (AUC = 0.62 in PFF-PR; AUC = 0.58 in PROFILE).
• Integrative Models: Adding PRS significantly improved discrimination.
  • Clinical + WG-PRSC+T: AUC increased to 0.68 (p = 9.54×10⁻⁴ vs. clinical only).
  • Clinical + WG-PRSSBayesRC: AUC increased to 0.66 (p = 0.02).
  • Sentinel-PRS: Did not show a statistically significant improvement over clinical data alone in the discovery cohort (AUC = 0.65, p = 0.09).
• Stratification: The improvement was most pronounced in patients with short telomeres and those with a family history.

Reclassification and Utility

• Reclassification: Adding WG-PRSC+T to the clinical model correctly reclassified 22.8% of carriers into a higher risk category (at a 15% threshold).
• MUC5B Independence: Excluding the MUC5B locus from the PRS calculation did not eliminate the association, confirming that the predictive power extends beyond the single strongest common variant.
• Sensitivity: Models performed better when QVs were defined more stringently (higher deleterious potential), suggesting PRS is most useful for identifying highly penetrant monogenic forms.

4. Key Contributions

1. Validation of Inverse Architecture: Confirmed that in IPF, high polygenic burden and rare monogenic variants are largely mutually exclusive, supporting a "liability threshold" model where different genetic mechanisms drive the disease in different subgroups.
2. Superiority of Whole-Genome PRS: Demonstrated that Whole-Genome PRS (WG-PRS) methods (specifically C+T and Bayesian approaches) outperform simple Sentinel-PRS models in identifying rare variant carriers.
3. Clinical Utility: Showed that integrating PRS with standard clinical history (age, family history, telomere length) significantly improves the AUC for detecting carriers, identifying patients who would otherwise be missed by current guidelines.
4. Stratification Strategy: Proposed a workflow where patients with low PRS (indicating low common variant burden) but clinical IPF should be prioritized for rare variant sequencing, as they are statistically more likely to carry a monogenic driver.

5. Significance and Limitations

Significance:

• Precision Medicine: This study provides a data-driven framework to refine genetic testing criteria for IPF. It suggests that PRS can act as a "triage" tool to prioritize sequencing resources for patients who are currently "sporadic" cases but may harbor rare variants.
• Endotype Identification: Supports the existence of distinct genetic endotypes in IPF (polygenic vs. monogenic), which may have different prognoses and treatment responses (e.g., telomere-related variants often correlate with poorer survival and extrapulmonary complications).
• Cost-Effectiveness: By improving the positive predictive value of genetic testing, this approach could reduce unnecessary sequencing in patients with high polygenic risk who are unlikely to carry rare variants.

Limitations:

• Data Availability: WG-PRS currently requires WGS data, which is not yet routine in clinical practice. The authors suggest low-coverage WGS or imputation from SNP arrays as future cost-effective alternatives.
• Ancestry: The GWAS summary statistics used were primarily of European ancestry, limiting the generalizability of the PRS models to other populations.
• Telomere Measurement: Telomere length was estimated from WGS (TelSeq) rather than the gold standard Flow-FISH, though it showed correlation.
• Clinical Implementation: While statistically significant, the absolute increase in AUC (from 0.62 to 0.68) is moderate. The authors note that the clinical benefit lies in reclassifying specific high-risk subgroups rather than a massive overall shift in diagnostic accuracy.

Conclusion:
The study concludes that incorporating IPF-PRS into clinical models enhances the identification of monogenic IPF carriers. It advocates for a shift in clinical strategy: prioritizing genetic sequencing for IPF patients with low polygenic risk scores, as they represent the population most likely to harbor rare, high-impact deleterious variants.