3382 - Decipher Immune Microenvironment Signatures Predict Radiation Toxicity In Prostate Cancer: A Single-Institution Analysis

Purpose/Objective(s): The Decipher genomic classifier is used for risk stratification in prostate cancer but its component biomarkers have not been evaluated as predictors of radiation toxicity. We investigated whether tumor immune microenvironment features captured by Decipher predict treatment-related adverse events following definitive radiation.

Materials/Methods: We reviewed 91 prostate cancer patients treated with definitive radiation at our institution (2024-2025) with minimum 6-month follow-up and prior Decipher testing. Modalities included IMRT (n=59), SBRT (n=25), and brachytherapy (n=7). Toxicity was graded per CTCAE criteria. The primary endpoint was unique toxicity events per patient; secondary was grade =2 toxicity. 18 Decipher-derived markers were individually assessed. 13 markers were combined into composite Z-scores across three sub-categories: Activated Immune (CD8, IL6/JAK/STAT3, IFN-a, IFN-?, tertiary lymphoid structures), Suppressed Immune (CTLA4, PD-L2, TREG, tumor purity, MDSC), and DNA Damage/Repair (CIN70, DDR, HR deficiency). A Net Immune Activation score was derived as Activated minus Suppressed composite. Associations were tested using linear regression. IMRT subgroup analysis assessed for modality confounding.

Results: Median Decipher score was 0.63 (range 0.13-0.96). No grade 3+ toxicities occurred. 7 (7.7%) had no toxicity, 47 (51.6%) grade 1, 37 (40.7%) grade 2. Mean unique toxicity events was 3.4 (range 0-6). Treatment modality was not associated with toxicity events (p=0.996). Higher Decipher risk was associated with more toxicity events vs lower risk (3.74 vs 2.58, p=0.003). Among individual markers, tumor purity was the strongest predictor (R²=0.143, p=0.0002), followed by TREG (p=0.005), IL6/JAK/STAT3 (p=0.011), tertiary lymphoid structures (p=0.011), CD8 (p=0.018), and IFN-? (p=0.025). All activated immune markers predicted increased toxicity while suppressive markers were protective. No marker predicted grade =2 toxicity.

The Suppressed Immune composite was inversely associated with toxicity events (R²=0.102, p=0.002). The Activated Immune composite showed a positive association (R²=0.081, p=0.006). The Net Immune Activation score was significant (R²=0.101, p=0.002) and remained significant in IMRT-only subgroup analysis (R²=0.075, p=0.036). DNA Damage/Repair composite showed no association (R²<0.001, p=0.996).

Conclusion: Tumor immune microenvironment features from Decipher significantly predict the number of distinct toxicity types following prostate cancer radiation. Immune activation pathways including IL6/JAK/STAT3, IFN-?, and CD8 signaling predicted toxicity diversity while DNA repair pathways showed no association, implicating tumor microenvironment rather than radiosensitivity as the genomic determinant of treatment tolerability. This novel application of existing genomic data could enable personalized supportive care planning. Multi-institutional validation is warranted.