Genomic attributes predict cancer treatment responsiveness

From the Nelson Lab, Human Biology & Clinical Research Divisions

One major challenge of cancer treatment is determining which therapies will be most effective for a patient. Understanding cancer cell vulnerabilities can help doctors select treatments that target those weaknesses to selectively kill cancer cells but not the healthy one. Many cancers have mutations in genes such as BRCA2 that are involved in a double strand DNA break repair process called homology-directed DNA repair (HRR). Here, the cell uses the DNA from the other homologous chromosome as a template to repair the damaged DNA. If one were to induce a type of DNA damage that could only be repaired through HRR, cancer cells lacking HRR deficiency would selectively die. Platinum chemotherapy is one such chemotherapy that induces DNA cross linking mediated double strand breaks, and cancer cells with HRR deficiency respond to this therapy. Exploiting this strategy, knocking down a second player in the DNA repair pathway by using Poly(ADP-ribose) polymerase inhibitors (PARPi) also leaves the cancer cells susceptible to irreversible damage and death. Thus, cancers with an HRR deficiency are responsive to PARPi and platinum chemotherapy.

HRR pathway gene aberrations are common in metastatic prostate cancer, but surprisingly, patients with such aberrant mutations in this DNA repair pathway are not always responsive to these treatments. On an average, only 60% of patients treated with PARPi or platinum chemotherapy respond to treatments. A recent study published in JCI Insight from the lab of Dr. Peter Nelson, a member of the divisions of Human Biology, Clinical Research, and the head of the Prostate Cancer Research Program at Fred Hutch, aimed to improve this statistic. This study was led by Dr. Navonil De Sarkar, a postdoctoral fellow in the Nelson Lab who thought that 60% was “just not good enough”. To increase the ability to predict a patient’s responsiveness to PARPi and platinum chemotherapy treatments, the authors teamed up with Dr. Sayan Dasgupta from the Vaccine and Infectious Disease Division at Fred Hutch and used machine learning to develop a binary classification system that integrated different genomic features to predict a patient’s HRR status. This system, an integrated assessment of HRR deficiency (iHRD), classifies metastatic prostate cancers as functional HRR deficient (iHRD+) or proficient (iHRD-) and was able to more accurately predict which tumors would respond to PARPi and platinum chemotherapy treatments.

Schematic of iHRD classification
Schematic of how iHRD status is determined. Image provided by Dr. Navonil De Sarkar.

The authors evaluated whole exome sequencing from 418 tumors to assess the mutational patterns present in metastatic prostate cancer tumor samples.  Since a few genomic features were already known to correspond with HRR deficiency, the authors thought that if they incorporated all these features together, they might improve the accuracy of this prediction. In total, they evaluated six genomic features in the iHRD machine learning model. Included in these features was a previously identified mutational signature common in HRR deficient tumors called COSMIC signature 3 (CSig3) that has been used as a metric for predicting HRR deficiency and treatment responsiveness. In collaboration with Dr. Eva Corey at the University of Washington, the Nelson group performed in vitro and in vivo functional studies in prostate cancer patient-derived xenografts (PDX) to show the accuracy of iHRD in predicting functional HRR deficiency. The authors further tested whether iHRD status could improve the predicted treatment responsiveness of the PDX models compared to other commonly used metrics, such as CSig3 status alone. In both PDX cell culture and in vivo tumor models treated with or without the platinum chemotherapy drug carboplatin, iHRD improved the predicted treatment response compared to other metrics. Furthermore, the authors retrospectively analyzed a cohort of patients treated with PARPi or carboplatin to determine their iHRD status and whether iHRD status, in turn, predicted the treatment responsiveness of each patient. Again, iHRD outperformed the predictive abilities of other metric systems with an accuracy of close to 90%. The authors further elaborated on CHD1 bi-allelic inactivation, which is a known recurrent genomic aberration in localized prostate cancer. Importantly, CHD1 gene mutation is not considered as a clinical biomarker for HRR deficiency. The Nelson group noticed in the metastatic prostate cancer (mCRPC) cohort that CHD1 bi-allelic mutant tumors associated with iHRD (+) status. They further identified a pair of CRPC PDX models that were CHD1 mutant. Working with Dr. Michael Haffner, they determined that CHD1 bi-allelic inactivation occurred through a combination of genomic copy number loss of one allele and epigenetic silencing of the other through promoter hypermethylation. Remarkably, both of these PDX models were iHRD(+), but CSig3(-) and responded to carboplatin treatment in a dose-dependent manner. These results uncovered a previously less appreciated role for CHD1 in the homology-directed DNA repair process while highlighting the importance of epigenetic regulation in dictating a patient's iHRD status.

The Nelson group was excited about the accuracy of iHRD’s ability to predict treatment responsiveness but envisage further improvements. They believe a more accurate understanding of the drug's mechanism of action and a comprehensive mechanistic understanding of the DNA repair and replication pathways may be the key to developing a more accurate tool and biomarker for precision medicine. In the future, Dr. De Sarkar strongly believes he can improve iHRD treatment responsiveness predictions by incorporating additional genetic and epigenetic features into the model. In addition to incorporating epigenetic status into iHRD, De Sakar plans to enhance the translational aspects of this work too. His goal is to have iHRD be used as a personalized cancer treatment tool to help doctors make informed decisions about treating patients, ultimately providing patients with the best care and giving them the best shot at beating metastatic prostate cancer.

This research was supported by the Prostate Cancer Foundation, the SU2C Prostate Cancer Dream Team, the American Association for Cancer Research, the National Cancer Institute, the Pacific Northwest Prostate Cancer SPORE, Congressionally Directed Medical Research Programs (CDMRP), CDMRP Prostate Cancer Research Program Early Investigator Award and PCF-Valor Young Investigator award.

UW/Fred Hutch Cancer Consortium members Dr. Eva Corey, Dr. Colm Morrissey, Dr. Michael Schweitzer, Dr. Michael Haffner, Dr. Colin Pritchard and Dr. Peter Nelson contributed to this work.

De Sarkar N, Dasgupta S, Chatterjee P, Coleman I, Ha G, Ang LS, Kohlbrenner EA, Frank SB, Nunez TA, Salipante SJ, Corey E, Morrissey C, Van Allen E, Schweizer MT, Haffner MC, Patel R, Hanratty B, Lucas JM, Dumpit RF, Pritchard CC, Montgomery RB, Nelson PS. Genomic attributes of homology-directed DNA repair deficiency in metastatic prostate cancer. JCI Insight. 2021 Dec 8;6(23):e152789. doi: 10.1172/jci.insight.152789. PMID: 34877933; PMCID: PMC8675196.