3140 - RNA-Sequencing-Derived Radiosensitivity Index for Genomic-Adjusted Radiotherapy

05:00pm - 06:00pm PT

Hall F

Screen: 9

POSTER

Presenter(s)

Steven Tau, PhD - Dartmouth College, Hanover, NH

S. Tau¹, S. A. Eschrich², J. G. Scott³, J. F. Torres-Roca⁴, and D. T. Bergman^1,5; ¹Geisel School of Medicine at Dartmouth, Hanover, NH, ²Department of Biostatistics and Bioinformatics, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL, ³Department of Radiation Oncology, Taussig Cancer Center, Cleveland Clinic, Cleveland, OH, ⁴Department of Radiation Oncology, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL, ⁵Department of Radiation Oncology and Applied Sciences, Dartmouth-Hitchcock Medical Center, Lebanon, NH

Purpose/Objective(s):

The Radiosensitivity Index (RSI), a 10-gene rank-based signature developed using microarray data, has shown promise in predicting clinical outcomes and adjusting radiotherapy (RT) dose. As molecular pathology shifts to next-generation sequencing (NGS)–based transcriptomics, it remains unclear whether RSI from whole-transcriptome RNA-sequencing (RSI_seq) reproduces microarray-based RSI (RSI_MA). We hypothesized that RSI_seq would closely correlate with RSI_MAin matched clinical samples and have similar utility in genomic RT dose adjustment.

Materials/Methods:

Matched microarray and RNA-seq gene expression data were obtained from three TCGA cancer cohorts: glioblastoma (GBM, n=150), lung squamous cell carcinoma (LUSC, n=130), and ovarian serous cystadenocarcinoma (OV, n=298). RSI_seq and RSI_MA were calculated using the same 10-gene rank-based model. Platform agreement was assessed via Spearman correlation, linear regression, and mean difference. To evaluate RSI_seq clinically, we analyzed 146 primary RT-treated head and neck (HNC) tumors (57% oral cavity, 53.4% stage IV, 70.6% HPV-) from TCGA, calculated genomic-adjusted RT dose (GARD) and assessed association with OS.

Results:

Across 578 matched TCGA samples, global rank concordance between platforms was strong (median Spearman ? >0.8). RSI_seq was strongly correlated with RSI_MA (R²=0.58), and the mean difference showed a negligible but significant bias (-0.026, p = 9.5 × 10?⁵). 62% of samples differed by less than 0.15 RSI. Discrepancies were driven by variability in high-coefficient genes (ABL1).

In the HNC cohort, patients with GARD >34.5 derived from RSI_seq had a lower risk of death compared to those with GARD <34.5 (HR=0.50, p=0.0084). This association remained significant in patients receiving EQD2 =60 Gy (n=120, HR=0.46, p=0.011). Additionally, patients in the top 15% of GARD (>79.6) did not exhibit significantly different OS compared to those with GARD >34.5, <79.6 (HR=1.04, 95% CI: 0.44–2.47, p=0.936). In a continuous model, each 1-unit increase in GARD was associated with a 1% reduction in the hazard of death (HR = 0.99, 95% CI: 0.9809 – 0.9995, p=0.039).

Conclusion:

RSI_seq showed high concordance to RSI_MA, supporting its clinical integration in an increasingly NGS-driven landscape. RSI_seq-GARD was prognostic of OS as both a dichotomous and continuous variable in an RT-treated, majority oral cavity HNC cohort, identifying a threshold beyond which dose escalation provides no additional benefit. These findings support RSI_seq-GARD as a genomic-adjusted radiotherapy signal comparable to RSI_MA-GARD.

Limitations include sparse demographic and treatment data in TCGA and evaluation of only HNC. Validation of these results is required in additional clinical cohorts. Refining RSI coefficients for high-variability genes may further improve cross-platform concordance. Expanding radiosensitivity biomarker investigations using modern multi-omics will be important for advancing genomic precision RT.