Pan-Cancer Analysis of Copy-Number Features Identifies Recurrent Signatures and a Homologous Recombination Deficiency...

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Copy-number (CN) alterations are key indicators of cancer's molecular state.
• CN features hold potential for predicting treatment response and patient prognosis.
• Understanding ubiquitous CN signatures across tumor types is crucial for advancing cancer therapy.

Purpose of the Study:

• To analyze a large pan-cancer dataset for copy-number (CN) signatures.
• To characterize prevalent CN signatures and their association with genomic alterations and clinical features.
• To develop and validate a machine learning classifier for detecting homologous recombination deficiency signatures (HRDsig) and assessing their clinical utility.

Main Methods:

• Analysis of CN features in 260,333 pan-cancer samples.
• Examination of 10 CN signatures, their association with genomic alterations, and clinical characteristics.
• Training a machine learning classifier using CN and indel features to detect HRDsig positivity.
• Assessment of clinical outcomes using a real-world clinicogenomic database (CGDB).

Main Results:

• Prevalence of CN signatures across cancer types, linked to tandem duplications, amplifications, genome-wide loss of heterozygosity (gLOH), and HRD.
• A novel HRDsig outperformed gLOH in predicting BRCAness and distinguishing biallelic BRCA and HRRwt samples.
• HRDsig demonstrated high sensitivity in detecting biallelic BRCA in ovarian (93%) and other HRD-associated cancers (80%-87%).
• HRDsig identified more patients than gLOH in ovarian and prostate CGDBs and showed predictive value for PARPi benefit.

Conclusions:

• Tumor CN profiles provide critical insights into cancer processes.
• The study describes 10 CN signatures in a large pan-cancer cohort, including two linked to HRD.
• A machine learning-based HRDsig robustly identified BRCAness, associated with biallelic BRCA, and predicted PARPi benefit in real-world data.