(P 100) COPY NUMBER ALTERATIONS IN CDKN2A AND CDKN2B ARE ASSOCIATED WITH WORSENED OVERALL SURVIVAL IN PATIENTS WITH GIST

Objective: Gastrointestinal Stromal Tumor (GIST), the most common sarcoma of the GI tract, is driven by activating mutations in multiple genes, including KIT and PDGFRA. Tyrosine kinase inhibitors (TKIs) have been developed to treat GIST, but these are not curative. The objective response rates for second-line therapies and beyond are approximately 5 to 10%. We sought to determine additional genetic factors that may influence overall survival (OS) in GIST patients and to identify possible molecular therapeutic targets for future treatment strategies.

Methods: We utilized the publicly available 2022 MSK-IMPACT sarcoma dataset on cBioPortal, which contains 395 GIST with clinicopathologic and genomic profiling data. We performed univariate and multivariate Cox Proportional Hazard regression analysis to identify covariates that predicted OS. Based on these results, we then examined discrete copy-number alteration (CNA) data for this cohort. Feature selection was performed by negatively selecting genes with near-zero variance. Using these features, principal component analysis (PCA) and K-means clustering (k=3) were performed to identify different clusters of patients based on CNAs. These groupings were then stratified by KIT vs. non-KIT mutation status and Kaplan-Meier estimates were performed.

Results: Overall, 336 of the 395 patients in the cohort had complete OS data. Patient sex (male, Hazard Ratio (HR) 2.36 [1.41-3.97], p=0.001), presence of metastatic disease (HR 3.05 [1.86-5.01], p< 0.001), number of mutations (>2 mutations, HR 1.94 [1.06-3.57], p=0.023), and fraction CNA (HR 12.42 [3.49-44.22], p< 0.001) were significant predictors of OS on multivariate analysis. Gene feature selection of CNA data revealed that CNAs in cell-cycle genes (i.e., losses of CDKN2A, CDKN2A p16INK4A, CDKN2A p14ARF, CDKN2B) had the most variance between patients. K-means clustering identified three subgroups: 1) no CNAs (n=265, 78.8%); 2) CDKN2A loss only (n=11, 3.3%); and 3) CDKN2A and CDNK2B losses (CDKN2A/B; n=60, 17.9%) (Panel A). Kaplan-Meier estimates demonstrated that patients with CDKN2A/B losses had significantly worse survival (median OS 32.6 mos) compared to those with only CDKN2A losses and those without CNAs (median OS not reached for both subgroups; p< 0.001) (Panel B). When stratifying by driver mutation, patients with non-KIT mutations (n=76, 22.6%) were significantly less likely to harbor any CNAs compared to those with KIT mutations (n=260, 77.4%; p< 0.001) (Panel A). When selecting for patients with metastatic disease (n=108), those with CDKN2A/B CNAs had significantly worse survival (median OS 16.3 mos; n=33) compared to those lacking CDKN2A/B CNAs (median OS not reached; n=70, P< 0.001) (Panel C). Finally, after further stratifying for those harboring KIT driver mutations (n=88), patients with CDKN2A/B CNAs (median OS = 20.7 mos; n=30) had significantly worse survival compared to those without any CNAs (median OS not reached; n=53, P=0.0016) (Panel D).

Conclusion: These results demonstrate that CNAs may be an unappreciated factor that may help stratify risk in patients with GIST. Copy number losses in cell cycle related genes, including CDKN2A and CDKN2B, are associated with worse overall survival, even when controlling for factors such as metastatic disease and oncogenic driver mutations. CDKN2A/B act as tumor suppressors via inhibition of cyclin dependent kinases 4 and 6 (CDK4/6) in the cell cycle. While a previous trial examining the efficacy of single agent palbociclib, a CDK4/6 inhibitor, was disappointing, our new results suggest a potential rationale for developing a clinical trial to investigate combination therapy with a KIT-targeted agent and a CDK4/6 inhibitor in KIT mutant GIST patients with CDKN2A/B losses.