Pancreatic adenocarcinoma (PAAD) is one of the deadliest cancers, with very poor survival rates even after surgery. Current pathology-based classification systems do not reliably guide treatment decisions, leaving doctors without clear tools to predict which patients will do better or worse.
Genetic mutations like KRAS (found in 88% of patients) are common but mostly not druggable, and only about 4% of patients carry mutations in genes that can be targeted by existing drugs. This study aimed to find new ways of classifying patients using a type of genetic change called copy number variation (CNV) — how many copies of a gene are present in tumor cells.
The research involved 608 pancreatic cancer patients from a single hospital in China, making it the largest Chinese PAAD cohort study of its kind at the time.
Researchers profiled the complete genomic landscape of 608 PAAD patients, examining three types of genetic changes: somatic mutations (changes that occur in the tumor), germline variants (inherited changes), and copy number variations (CNVs, which measure how many copies of each gene are present).
DNA was extracted from tumor tissue preserved in formalin and paraffin (standard hospital pathology samples), as well as from blood samples. Using CNV data from all 608 patients, the team applied unsupervised clustering — a technique that groups patients based on patterns without pre-set rules — to discover natural patient subgroups.
A CNV score was calculated for each patient using a mathematical formula based on principal component analysis, allowing patients to be ranked from low to high CNV burden. An optimal cutoff was determined to divide patients into two risk groups, and the prognostic value of this score was validated in an independent TCGA dataset of 182 patients.
Patients were divided into two primary groups based on CNV patterns: CNV-G1 (321 patients) and CNV-G2 (287 patients). Patients in CNV-G2 had twice the risk of death compared to CNV-G1, with median survival of 239 days versus 410 days.
Further analysis revealed that amplification (extra copies) of DNA repair genes — specifically those involved in homologous recombination repair (HRR) — was associated with worse outcomes. This was counterintuitive: having more copies of repair genes did not help patients, likely because it enhanced the cancer's ability to fix DNA damage and resist treatment.
Combining CNV clustering and CNV scores, researchers identified three molecular subtypes: repair-deficient (18 patients, best prognosis), proliferation-active (121 patients, intermediate), and repair-enhanced (44 patients, worst prognosis). A five-gene prognostic model built from CNV data could predict relapse with reasonable accuracy in both the Chinese and TCGA validation cohorts.
The five-gene CNV-based prognostic model performed well in predicting which patients would relapse within six months, one year, and at median survival time — information that could help doctors prioritize intensive monitoring or aggressive treatment.
Patients with the repair-deficient subtype had DNA repair patterns suggesting sensitivity to platinum-based chemotherapy and PARP inhibitors, meaning this subtype could guide selection of these specific drug classes. Patients in the proliferation-active subtype had enrichment of RTK (receptor tyrosine kinase) signaling genes, suggesting they might benefit from targeted therapies against these pathways.
The tumor microenvironment also differed across subtypes, with immune cell infiltration patterns that could inform decisions about immunotherapy eligibility.
This study demonstrates that copy number variation — not just point mutations — carries important prognostic and predictive information in pancreatic adenocarcinoma. The molecular subtypes discovered here are independent of standard clinical staging, meaning they add new information beyond what doctors currently use.
The finding that amplified DNA repair genes worsen prognosis has important implications: it suggests that cancers with high HRR gene copy numbers may be particularly resistant to DNA-damaging therapies. Future trials could stratify patients by these subtypes to test targeted approaches.
While the cohort is large, it is from a single hospital, and prospective validation in multicenter trials is needed before this classification enters routine clinical use.