Identification of an ERCC2 mutation associated mutational signature of nucleotide excision repair deficiency in targeted panel sequencing data

This study develops and validates a targeted panel sequencing-based mutational signature for nucleotide excision repair deficiency caused by ERCC2 mutations, demonstrating its ability to predict improved platinum therapy response and survival in bladder cancer while offering a novel therapeutic targeting strategy for ERCC2-mutant solid tumors across various types.

The Gist

Imagine your body is a massive, bustling city. The DNA inside your cells is the city's master blueprint, containing all the instructions for how to build and run everything. Every day, this blueprint gets damaged by "weather" (sunlight, chemicals) and "construction errors" (natural copying mistakes).

Usually, the city has a highly efficient Repair Crew (called Nucleotide Excision Repair, or NER) that patrols the streets, finds these damaged blueprints, and fixes them before they cause a disaster (cancer).

The Problem: The Broken Repair Crew

In some bladder cancers, a specific part of this repair crew, a worker named ERCC2, gets injured or fired (mutated). Without ERCC2, the city can't fix certain types of blueprint damage. Instead of fixing the errors, the city starts accumulating a unique pile of "trash" or "scars" on the blueprint.

Scientists already knew that if they could look at the entire blueprint (Whole Genome Sequencing), they could see this specific pattern of scars. This pattern tells them: "Hey, the ERCC2 worker is missing! This cancer is broken in a specific way."

Knowing this is huge because it means the cancer is likely very sensitive to a specific type of chemotherapy (platinum-based drugs) that acts like a "bulldozer," destroying the already fragile, broken cells.

The Challenge: The "Spot Check" vs. The "Full Audit"

Here's the catch: Looking at the entire blueprint is expensive, slow, and not done in most hospitals. Instead, doctors usually do a "Spot Check." They only look at a few hundred specific pages of the blueprint (Targeted Panel Sequencing) that are known to be trouble spots.

The big question was: Can we still see the "ERCC2 trash pattern" if we only look at a few pages?
It's like trying to identify a specific type of graffiti artist just by looking at three walls in a city, rather than the whole city. Most scientists thought the answer was "no" because there wasn't enough data.

The Solution: A New Detective Tool

This paper is about a team of scientists who said, "Let's try anyway." They built a new digital detective (a computer model) trained to look at those limited "Spot Check" pages and find the hidden pattern of the broken ERCC2 worker.

Here is how they did it, using simple analogies:

1. Training the Detective: They took thousands of bladder cancer cases where they knew the ERCC2 worker was missing (from full blueprint data). They taught the computer: "Look at these specific pages. When you see this mix of 'T>C' errors and 'long deletion' scars, that's the ERCC2 pattern."
2. The "Signature": They found that even in the limited pages, the ERCC2-deficient tumors left a distinct "fingerprint." It was like finding a specific type of mud footprint that only appears when the ERCC2 worker is gone.
3. The Test: They tested their new detective on patients who only had the "Spot Check" data.
   • Result: The detective was incredibly good at it! It could tell the difference between "ERCC2 broken" and "ERCC2 working" with about 90% accuracy, even without seeing the whole blueprint.

Why This Matters (The "So What?")

1. Saving Lives with Better Targeting:
The study found that patients with this "ERCC2 fingerprint" (even if they didn't have the actual broken gene mutation) responded much better to platinum chemotherapy.

• Analogy: Imagine two houses on fire. One is made of wood (normal cancer), the other is made of dry straw (ERCC2-deficient cancer). If you throw water on the straw house, it goes out instantly. If you throw water on the wood, it might just steam. This test tells the firefighters (doctors) exactly which house is made of straw so they can use the right water pressure (chemo) to save the day.

2. Finding Hidden Gems:
They discovered that this "fingerprint" isn't just in bladder cancer. They found it in rare cases of breast cancer and other tumors where the ERCC2 gene was broken. This suggests that patients with these rare cancers might also benefit from platinum chemotherapy, even if doctors didn't think to try it before.

3. Using What We Have:
Most hospitals already have millions of "Spot Check" results sitting in their computers. This new tool allows them to dig through that existing data and find patients who are likely to respond well to treatment, without needing to run expensive new tests.

The Bottom Line

The scientists created a clever "spot-check" method to find a specific type of cancer weakness. It's like upgrading a metal detector that usually only finds coins, so now it can also find gold rings hidden in the sand, even if you can't dig up the whole beach. This helps doctors treat patients more effectively, using the data they already have.

Technical Summary

1. Problem Statement

• Clinical Context: Identifying DNA repair deficiencies (specifically Nucleotide Excision Repair or NER deficiency) is crucial for selecting targeted therapies, such as platinum-based chemotherapy for NER-deficient tumors.
• The Gap: While mutational signatures derived from Whole Genome Sequencing (WGS) or Whole Exome Sequencing (WES) are the gold standard for detecting these deficiencies, routine clinical practice relies heavily on targeted gene panel sequencing (covering ~400–500 genes).
• The Challenge: It is unclear whether the limited genomic coverage of targeted panels (a few Megabases vs. the whole exome/genome) contains sufficient mutational data to accurately derive NER deficiency signatures, specifically those associated with ERCC2 inactivating mutations in bladder cancer.
• Objective: To develop and validate a composite mutational signature derived specifically from targeted panel data that can identify NER deficiency (associated with ERCC2 mutations) and predict clinical outcomes (platinum sensitivity and survival).

2. Methodology

Data Sources

The study utilized multiple cohorts to train, validate, and test the model:

• Training Set: AACR GENIE dataset (v16.1), specifically Bladder Urothelial Carcinoma (BLCA) samples sequenced with MSK-IMPACT panels (n=1,609). Samples were filtered to include only ERCC2 Wild Type (WT) or pathogenic ERCC2 Mutant (MUT) cases.
• Validation Sets:
  • GENIE Subsets: BLCA samples from DFCI-Oncopanel (n=549) and other non-MSK panels (n=498).
  • Downsampled WES Cohorts: Three independent cohorts (DFCI-MSKCC, Philadelphia, Aarhus) with pre-existing WES data. These were computationally downsampled to match the gene coverage of the MSK-IMPACT 505 panel to test the model's performance on "panel-like" data derived from WES.
  • Cross-Tumor Validation: Non-bladder solid tumors with known pathogenic ERCC2 mutations (n=44).

Feature Engineering & Model Development

• Mutational Spectra: Single Base Substitutions (SBS) and Insertions/Deletions (ID) were categorized into 96 and 83 mutation types, respectively.
• Feature Selection: A Recursive Feature Elimination (RFE) process using a Gradient Boosting Machine (GBM) classifier was performed.
  • Initial Features: Cosine similarities to known COSMIC signatures (SBS1, 2, 5, 8, 13, 29, 40; ID1-5, 8-10) and Tumor Mutational Burden (TMB) metrics.
  • Harmonization: TMB values were normalized and standardized (z-scores) to account for differences in panel sizes across platforms.
• Optimal Feature Subset: The final model utilized six features:
  1. SBS Signature 2
  2. SBS Signature 5 (associated with ERCC2)
  3. SBS Signature 13
  4. ID Signature 8 (associated with large deletions)
  5. Total SBS count
  6. Total ID count
• Classification: A GBM classifier was trained to predict ERCC2 mutation status. An optimal probability cut-off (>0.048) was determined to maximize sensitivity and specificity.

Statistical Analysis

• Enrichment Analysis: Binomial tests were used to determine if mutations were enriched in panel-covered regions compared to the whole genome.
• Survival Analysis: Kaplan-Meier curves and Cox Proportional Hazards models were used to assess Overall Survival (OS) and response to neoadjuvant chemotherapy (NAC) based on the signature score.

3. Key Results

A. Enrichment of Mutations in Panel Regions

• Analysis of TCGA WGS data revealed that ERCC2 mutant tumors have a significant enrichment of SBS and ID mutations within the genomic regions covered by targeted panels compared to ERCC2 WT tumors.
• Specifically, T>C mutations (dominating SBS5) and >5bp deletions (dominating ID8) were highly enriched in ERCC2 MUT cases within panel regions, confirming that panel data captures the specific mutational footprint of NER deficiency.

B. Model Performance

• Training (MSK-IMPACT): The model achieved 89.4% accuracy, 98.2% sensitivity, and 88.8% specificity with an AUC of 0.98.
• External Validation (DFCI-Oncopanel): Achieved 91.6% accuracy and an AUC of 0.90.
• Other Panels: Achieved 88.5% accuracy and an AUC of 0.90 on non-MSK/DFCI panels.
• Downsampled WES Validation: When applied to downsampled WES data, the panel-based signature showed substantial agreement (Cohen's Kappa = 0.714) with the original WES-based signature, successfully identifying the majority of ERCC2 MUT cases.

C. Clinical Correlation (Bladder Cancer)

• Platinum Sensitivity: In a pooled cohort of muscle-invasive bladder cancer (MIBC) patients receiving neoadjuvant cisplatin, a high panel-based ERCC2mut signature score was significantly associated with pathologic complete response (downstaging to pT0/Ta/Tis/N0), even in patients without a detectable ERCC2 mutation (p=0.021).
• Survival: In primary tumor samples from the GENIE dataset, patients with a high signature score had significantly better overall survival (HR = 0.55, p < 0.001). This association persisted even after excluding known ERCC2 mutant cases, suggesting the signature identifies a broader NER-deficient phenotype.

D. Cross-Tumor Applicability

• The signature was tested on 44 non-bladder solid tumors harboring pathogenic ERCC2 mutations.
• 70% (31/44) of these non-bladder cases exhibited the characteristic mutational signature, suggesting that NER deficiency and potential platinum sensitivity extend beyond urothelial carcinoma.

4. Significance and Contributions

1. Bridging the Clinical-Research Gap: The study demonstrates that clinically available targeted panel data, often considered insufficient for complex mutational signature analysis, can be effectively leveraged to detect NER deficiency. This allows for the retrospective analysis of vast clinical archives where only panel data exists.
2. Biomarker for "Silent" NER Deficiency: The model identifies NER deficiency not only in ERCC2 mutant cases but also in ERCC2 Wild Type cases that exhibit the mutational signature. This expands the pool of patients potentially eligible for platinum-based therapies or NER-targeted drugs (e.g., irofulven analogs).
3. Prognostic Value: The signature serves as an independent prognostic marker for improved survival and response to neoadjuvant chemotherapy in bladder cancer.
4. Pan-Cancer Implications: The detection of this signature in non-bladder cancers with ERCC2 mutations suggests a novel therapeutic avenue for rare solid tumors that might otherwise be overlooked for platinum therapy.
5. Methodological Framework: The paper provides a reproducible workflow (code available on GitHub) for deriving composite mutational signatures from limited genomic data, which can be adapted for other DNA repair deficiencies (e.g., Homologous Recombination Deficiency).

Conclusion

The authors successfully developed a robust, panel-based composite mutational signature that accurately identifies NER deficiency driven by ERCC2 inactivation. This tool enables the identification of bladder cancer patients who are likely to respond to platinum therapy and have improved survival, even in the absence of a direct ERCC2 mutation, and offers a pathway to re-evaluate historical panel sequencing data for therapeutic stratification.