Association between MSI and TMB using multivariable linear regression models

PURPOSE:This study aimed to define molecular subtypes of prostate cancer by integrating androgen receptor (AR) signaling, neuroendocrine prostate cancer (NEPC) transcriptional signatures, and genomic alterations to inform biomarker-driven therapies in metastatic castration-resistant prostate cancer. METHODS:We analyzed 8,019 prostate tumors using DNA/RNA sequencing (Caris Life Sciences), classifying them into four molecular subtypes (AR+/NE-, AR-/NE+, AR+/NE+, AR-/NE-). Genomic alterations, cell surface target expression, and overall survival (OS) were evaluated. RESULTS:Of the 8,019 tumors, 87.2% were adenocarcinoma, 1.9% NEPC, and 0.4% had mixed histology; 63% were from primary sites and 36.5% from metastases. The median age was 68 years; 63% were White, 15% Black, and 2.6% Asian or Pacific Islander. Most tumors were classified as AR+/NE- (91%), and 4.6% were AR-/NE+. TP53 and PTEN alterations were enriched in AR-negative subtypes, whereas SPOP mutations were more frequent in AR+ tumors. FOLH1 (prostate-specific membrane antigen) expression was the highest in AR+ tumors, whereas DLL3 expression was elevated in NE+ tumors. Median OS was significantly longer in tumors with high AR signaling (55.0 v 14.0 months, P < .00001) and lower with the NEPC signature (54.3 v 16.1 months, P < .00001). Combined stratification showed the most favorable outcome in AR+/NE- tumors (55.3 months) and the poorest in AR-/NE+ tumors (12.0 months). CONCLUSION:Prostate cancer exhibits distinct molecular subtypes defined by AR signaling activity, NEPC transcriptional profiles, and genomic alterations. These biologically and clinically relevant subgroups provide a framework for precision oncology approaches and inform patient selection for biomarker-driven trials such as the ongoing PREDICT study (ClinicalTrials.gov identifier: NCT06632977).

PURPOSE:Circulating tumor DNA (ctDNA) next-generation sequencing complements tissue-based testing and offers insights into prognosis, treatment selection, and tumor evolution. Despite advances in metastatic castration-resistant prostate cancer (mCRPC) therapies, resistance remains a challenge. This real-world study evaluates longitudinal ctDNA changes following systemic treatments. EXPERIMENTAL DESIGN:We analyzed data from Guardant INFORM, a clinical genomic database linking ctDNA profiles with claims data. Patients with prostate cancer who received androgen receptor (AR) pathway inhibitors (ARPi), poly (ADP-ribose) polymerase inhibitors (PARPi), or taxanes and had ctDNA testing within 3 months before and after treatment discontinuation were included. We evaluated pre- and posttreatment mutational differences and survival outcomes (overall survival, time to treatment discontinuation, and time to next treatment), stratifying by treatment type and AR alterations. RESULTS:From 36,774 patients with prostate cancer, we identified 678 with paired pre-/post-ARPi, 188 with paired pre-/post-PARPi, and 844 with paired pre-/post-taxane ctDNA samples. After ARPi, the most frequent AR alterations included AR amplification (pre%/post%; 10.8%/25.6%), AR L702H (1.3%/7.9%), and AR T878A (2.9%/7.1%). Following PARPi, the most common homologous recombination repair gene alterations were ATM (25%/23.4%), BRCA2 (22.9%/17%), BRCA1 (4%/2.1%), and CDK12 (5.9%/5.9%). After taxane, frequent alterations included TP53 (47.2%→54%), AR (33.2%/49.9%), PIK3CA (9.4%/15.9%), and EGFR (9.6%/14.6%). All treatment cohorts showed a significant increase in mutation burden after therapy (mean increase 2.0-4.2 alterations; P < 0.001). Across all three treatment groups, the presence of AR alterations was consistently associated with inferior clinical outcomes. CONCLUSIONS:Our study revealed dynamic shifts in genetic mutations in patients with mCRPC following ARPi, PARPi, and taxanes. Furthermore, our findings highlight associations between AR alterations and clinical outcomes, emphasizing the potential for personalized treatment strategies.

4541 Background: Belzutifan is a hypoxia-inducible factor-2α (HIF-2α) inhibitor approved for RCC that targets a central driver in RCC pathogenesis; however, predictive biomarkers of response remain poorly defined. Methods: Comprehensive DNA (592-gene panel or whole exome) and RNA (whole transcriptome) sequencing were performed by Caris Life Sciences on RCC samples. Clear cell RCC and RCC-NOS with VHL mutations were included. Belzutifan time on treatment (ToT) was derived from insurance claims, and patients were classified as responders or non-responders using the median ToT. Results: A total of 150 belzutifan-treated RCC tumors were included in the analysis, of which 29% were primary kidney tumors and the remainder were from metastatic sites. The study population was classified as responders (n=57) or non-responders (n=93) using a median time on treatment of 85 days (95% CI: 71–114). Baseline characteristics were similar, except responders were younger (median age 57 vs 64 years, p=0.007). Transcriptomic analysis showed similar HIF-2α (median 7.78 vs 7.61) and CA9 (4.82 vs 4.54) expression between responders and non-responders without statistical significance. Responders showed enrichment of PTEN (15.8% vs. 3.3%, p=0.03) and PIK3R1 alterations (7.0% vs. 0.0%, p=0.04). Responders demonstrated numerically higher frequencies of PBRM1 (54.4% vs. 44.3%, p=0.46) and PIK3CA mutations (8.8% vs. 3.3%, p=0.24). Alterations in DNA damage response genes were also numerically more common among responders, including TP53 (14.0% vs. 4.9%, p=0.11) and ATM (3.5% vs. 1.6%, p=0.56), although none of these differences reached statistical significance. In contrast, non-responders exhibited numerically higher frequencies of BAP1 (18.0% vs. 8.8%, p=0.10), SETD2 (26.2% vs. 19.3%, p=0.27), and ARID1A (5.1% vs. 1.8%, p=0.30), all of which were also not statistically significant. Conclusions: HIF2-α RNA expression alone does not predict belzutifan response in RCC. Responders demonstrated alterations in PBRM1, PI3K/AKT/mTOR, and DNA damage response genes, while non-responders showed numerically higher frequencies of chromatin remodeling mutations including BAP1 and SETD2. Study limitation included small sample size and short follow-up. Larger datasets are required to validate these findings.

Baseline characteristics of the proportions of MSI-H and TMB-H

Both the Gleason grading system, which assesses prostate cancer aggressiveness, and the D’Amico risk classification, which predicts the likelihood of tumor recurrence, require post-surgical data to provide results. This highlights the significant research need for non-invasive methods of preoperative prediction of tumor aggressiveness and recurrence risk. To help fill this need, we investigated the correlation between various multiparametric magnetic resonance imaging (mpMRI) parameters and the ISUP 2014 Gleason grade group and D’Amico risk classification for prostate cancer. We prospectively collected pelvic mpMRI data from 561 patients who presented with elevated prostate-specific antigen (PSA) levels at our hospital. After we semi-quantitatively scored primary prostate lesions on a five-point scale according to the PI-RADS v2.1 standard, we post-processed diffusion-weighted imaging (DWI) images to quantitatively measure the apparent diffusion coefficient (ADC) values of the primary lesions. We then calculated Kendall’s tau-b correlation coefficients to assess correlations between the above MR parameters and the ISUP 2014 Gleason grade group and D’Amico risk classification. Our analysis indicates that several mpMRI parameters are correlated with the International Society of Urologic Pathologists (ISUP) 2014 Gleason grade group and the D’Amico risk classification in prostate cancer. In particular, ADC values and PI-RADS scores show strong correlations with the ISUP 2014 Gleason grade group. mpMRI parameters may serve as non-invasive surrogate markers, aiding in preoperative risk stratification for prostate cancer.

4543 Background: Clear cell renal cell carcinoma (ccRCC) is commonly driven by VHL loss, resulting in HIF-2α stabilization and upregulation of targets such as carbonic anhydrase IX (CA9). Clarifying the relationship between CA9 and HIF-2α expression is critical to guide rational therapeutic strategies. Methods: DNA (592-gene panel or whole exome) and RNA (whole transcriptome) sequencing was performed on ccRCC and RCC–not otherwise specified, (NOS) tumors with VHL alterations via Caris Life Sciences for integrated genomic and transcriptomic profiling. The primary objective was to characterize the expression patterns of HIF-2α and CA9 across RCC tumors and co-occurring genomic alterations. Overall survival (OS) as measured from initial diagnosis. Results: A total of 1,935 RCC tumors were analyzed, including 967 (50%) ccRCC and the remainder RCC–NOS with VHL alterations; 49% were derived from the primary site. Expression of HIF-2α and CA9 were similar across racial and ethnic subgroups. Compared to primary kidney tumors, lower HIF-2α expression was observed in lymph node and liver metastases (both p <0.0001) and lower CA9 expressions were observed in CNS and bone metastases (both p <0.0001). Increased either HIF-2α or CA9 expressions were associated with high angiogenic and T-effector gene signatures as defined in IMmotion150 (all p < 0.001). HIF-2α expression positively correlated with CA9 expression (r=0.43, p <0.001). Tumors with highest quartile HIF-2α expression (Q4) had increased PBRM1 and decreased BAP1, TP53, and TERT mutation frequencies (Table). High CA9 expression (Q4) had increased PBRM1 and decreased TP53 mutations. High HIF-2α (Q4) and CA9 (Q4) expressions were associated with improved OS compared with Q1 (HIF-2α: median OS 97 vs 64.7 months, HR 0.56, p < 0.0001; CA9: median OS 90 vs 49 months, HR 0.63, p = 0.002). Conclusions: High expression of HIF-2α and CA9 identifies a canonical VHL/HIF-driven RCC subtype marked by enrichment of PBRM1 mutations, depletion of aggressive genomic alterations ( BAP1 , TP53 , TERT ), and improved OS. Strong associations with angiogenic and immune-related gene signatures further support a coherent HIF-dependent phenotype. Key gene mutation frequency by HIF-2 α/CA9 expression. HIF-2α (Q1) HIF-2α (Q4) P value CA9 (Q1) CA9 (Q4) P value PBRM1 33% 48% <0.001 22% 46% <0.001 BAP1 29% 8% <0.001 22% 19% 0.43 TP53 13% 7% 0.012 16% 8% 0.04 TERT 13% 6% 0.005 12% 8% 0.37

Overall workflow for investigating genes that affect discrepancy between MSI and TMB

Discrepancy in proportions of MSI-H and TMB-H

528 Background: Belzutifan is a HIF-2α inhibitor approved for the treatment of advanced RCC, targeting the hypoxia signaling pathway central to tumor progression. While initial clinical trials demonstrated efficacy, predictive biomarkers remain undefined. Methods: Comprehensive DNA (592-gene panel or whole exome) and RNA (whole transcriptome) sequencing were performed on RCC samples via Caris Life Sciences. Patient samples with clear cell histology or VHL mutation in non-clear cell histology were considered for the study. Time-on-treatment (TOT) data for Belzutifan were extracted from insurance claims, and patients were stratified as responders or non-responders based on the median TOT. Results: In a cohort of 2,538 RCC patients, 240 patient samples were identified to have received Belzutifan (majority treated after biopsy) and were stratified into responders (n = 109) and non-responders (n = 131) based on median TOT (77 days; 95% CI: 62–84 days). No statistically significant differences were observed between the two groups across a range of clinical and demographic variables. Median age was 60 years in responders and 62 in non-responders. The proportion of metastatic biopsies was 61.1% in non-responders vs 58.7% in responders. Genomic profiling revealed high prevalence of VHL mutations in both groups (94.4% in responders vs. 92.8% in non-responders). Responders exhibited enrichment for alterations in the PI3K/AKT/mTOR pathway, including PIK3CA (12.4% vs. 3.3%), PIK3CB (6.7% vs. 1.9%), PIK3R1 (4.8% vs. 0.8%), and PTEN (13.2% vs. 9.8%). Mutations in DNA damage response genes were also more common in responders, particularly STAG2 (6.7% vs. 0%) and ATM (4.8% vs. 0.8%). In contrast, non-responders showed higher frequencies of BAP1 (17.7% vs. 12.2%), ARID1A (8.3% vs. 2.9%), NF2 (4.0% vs. 1.9%), and KDM5C (14.6% vs. 11.8%). None of these differences were statistically significant, and no single mutation was unique to either group, suggesting response may depend on combined or pathway-level alterations rather than individual genes. Transcriptomic analysis of HIF pathway genes revealed subtle numerical differences: slightly higher EPAS1 expression in responders (median: 7.73 vs. 7.54) and CA9 (4.82 vs. 4.46). Expression of other genes such as HIF1A, ARNT, ARNT2, and HIF3A showed minor variation between groups, with HIF1A and HIF3A slightly numerically higher in non-responders. Conclusions: Belzutifan response in RCC may relate to PI3K/AKT/mTOR and DNA damage response alterations, whereas resistance may involve chromatin remodeling mutations. Minimal clinical or demographic differences support broad trial inclusion, though subtle HIF-pathway and mutational patterns suggest molecular markers warranting prospective validation.

11119 Background: Palliative care (PC) is a key component of high quality cancer care, especially at the end of life. However, real world national data describing inpatient PC utilization remain limited. We evaluated patterns of PC use and associated outcomes among decedent patients with metastatic cancer in the U.S. Methods: We conducted a cross-sectional analysis of the 2021–2023 National Inpatient Sample of decedent patients with metastatic cancer. Palliative care, cancer type, and metastatic status were identified using ICD-10 codes. Patient characteristics included age, sex, race, comorbidity burden, insurance status, and household income. Hospital characteristics included location (urban vs rural) and ownership (public, private not-profit, and private invest-own). Outcomes included length of stay (LOS), inpatient procedures, and total costs. Factors associated with palliative care use and adjusted mean differences in outcomes were estimated. Results: An estimated 52,841 hospitalizations of decedent patients with metastatic cancer were identified, with 34,955 (66.2%) involving palliative care. PC utilization varied by cancer type (Table) and increased over time (aOR per year 1.15; 95% CI 1.12–1.18). PC use was independently associated with older age, female sex, private insurance, higher neighborhood income, and care at private not-for-profit hospitals (all p<0.001). PC was associated with fewer inpatient procedures (adjusted mean difference −1.38; 95% CI −1.45 to −1.31), shorter LOS (−1.25 days; 95% CI −1.45 to −1.06), and lower total costs (−$10,557; 95% CI −$11,421 to −$9,694). Reductions in procedures, LOS and costs were significant across most cancer types. (Table). Conclusions: Inpatient PC was associated with lower care intensity and costs across cancer types among decedent patients with metastatic cancer. Yet, about one third never received PC, highlighting a major end-of-life care gap and the need for earlier, more equitable integration to improve value based oncology care. Palliative care utilization and outcomes by major cancer type. Cancer Type PC use Adjusted Procedures difference Adjusted Cost difference ($) Adjusted LOS difference (days) Breast 2646 (66.7%) -1.47 (-1.70,-1.24) -12512 (-16301, -8722) -1.90 (-2.63,-1.17) Colorectal 2932 (65.8%) -1.75 (-1.49,-2.00) -12231 (-15261, -9201) -1.09 (-1.83,-0.35) Lung 8942 (67.9%) -1.14 (-1.28,-1.00) -6341 (-7894, -4968) -0.90 (-1.24,-0.56) Prostate 1897 (62.3%) -1.24 (-1.52,-0.97) -9736 (-12915, -6558) -0.88 (-1.82,0.05)* Pancreas 2574 (68.4%) -1.46 (-1.72,-1.21) -7396 (-9525, -5266) -0.39 (-0.94,0.17)* Kidney 936 (66.3%) -1.72 (-2.20,-1.23) -12353 (-17623, -7082) -1.93 (-3.11,-0.75) Liver 1362 (67.8%) -0.87 (-1.24,-0.51) -8896 (-13426, -4367) -1.27 (-0.33,-2.20) All p<0.001 unless noted with *.

Stratified analyses of the association between TMB and MSI by race across 22 cancer types in GENIE data

Significant genes contributing to MSI-H among 14 cancer cases: results from GENIE

Stratified analyses of the association between TMB and MSI by race across three cancer types in TCGA data

Stratified analyses of the association between TMB and MSI among the Asian population across three cancer types in KM data

Clinical and demographic characteristics of patients from GENIE, KM, and TCGA