A 36-gene predictive score of anti-cancer drug resistance anticipates cancer therapy outcomes microbiologystudy

In 1937, President Franklin Roosevelt signed the National Cancer Act, launching a nationwide effort to combat the disease. Eighty-seven years later, despite significant progress, cancer treatment often falls short, with 50% to 80% of patients not responding to treatment and more than 600,000 cancer deaths annually in the United States.

What if clinicians could predict the success of any cancer treatment, ensuring each patient receives the most effective care?

The challenge lies in the diverse nature of the disease. There are hundreds of different types of cancers, characterized by the specific type of cell from which they originate. Even patients with the same cancer type require personalized treatments due to unique factors like genetic predisposition, lifestyle and immune response.

The therapeutic outcomes—from complete remission to resistance to treatment—are unpredictable because cancer cells can develop resistance to drugs through genetic mutations, rendering therapy ineffective.

To tackle this complexity, a research team at the University of Alabama at Birmingham led by Anindya Dutta, Ph.D., professor and chair of the UAB Department of Genetics, sought to identify patterns within this apparent randomness.

Leveraging established cancer cell databases—including the Genomics of Drug Sensitivity in Cancer, or GDSC, the Cancer Therapeutics Response Portal, or CTRP, and the Catalogue of Somatic Mutations in Cancers, or COSMIC—the team investigated “whether gene expression levels correlate with drug response” across various cancer cell lines.

GDSC and CTRP provide information on how sensitive different cell lines are to various anti-cancer drugs, whereas COSMIC catalogs their gene expression. Divya Sahu in Dutta’s lab studied 777 cancer cell lines that were present in both databases and found 36 genes linked with anti-cancer drug resistance.

One of these genes, FAM129B, was found to be particularly important in drug resistance by cancer cells. This finding aligns with previous experimental studies on FAM129B, validating the efficacy of the analytical approach employed in this UAB study, now published in npj Precision Oncology.

The research group developed a combined score, named UAB36, using the 36 genes most linked to drug resistance. They found that the polygenic score UAB36 showed superior correlation with relative resistance to various anti-cancer drugs compared to existing polygenic scores.

The researchers applied UAB36 to predict the expression of genes linked with the resistance of breast cancer against tamoxifen, a drug widely used for breast cancer treatment. UAB36 consistently showed a higher efficacy compared to a single gene approach. UAB36 also outperformed established gene signatures like ENDORSE and PAM50 in its correlation with tamoxifen resistance in breast cancer cells.

The study crossed from the cell-line studies to application as a prognostic tool when the researchers used the UAB36 score to predict patient outcome in three different cohorts of actual breast cancer patients treated with tamoxifen. They found that patients with high UAB36 scores showed poorer survival independent of patient’s age and tumor stage, consistent with the expectation that this score predicts higher resistance to tamoxifen.

The tumors with high UAB36 showed enrichment of gene sets associated with multiple drug resistance. This establishes UAB36 as a promising biomarker for predicting anti-cancer drug resistance and poor survival.

UAB36 has potential as a tool for personalized medicine, helping identify patients at higher risk of tamoxifen resistance and poor survival, suggesting that these patients will benefit from alternative treatment strategies. The study provides a map to help doctors choose the best cancer treatment and predict outcomes for each patient, though this has to be validated by a prospective clinical trial.

“This approach should provide promising polygenic biomarkers for resistance in many cancer types against specific drugs and can be improved further by incorporating machine-learning methods in the analysis,” Dutta said.