ctDNA as a complement to gene-expression–based site prediction to guide therapy in cancer of unknown primary: Results from Fudan CUP-001.

Abstract

e15049 Background: Cancer of unknown primary (CUP) remains a clinical challenge with poor outcomes under empiric chemotherapy. Fudan CUP-001 trial showed site specific therapy (SST) guided by tissue origin gene expression profiling can improve progression free survival (PFS), while CUPSICO trial showed molecularly guided therapy based on comprehensive genomic profiling also provides PFS benefit. We hypothesized that the combined therapeutic approach would yield superior therapeutic efficacy. Methods: We analyzed 170 CUP patients enrolled in the Fudan CUP-001 trial with available ctDNA results, 129 of whom underwent 90-gene expression testing. Somatic alterations were profiled, and actionable variants were classified per Standards and Guidelines for the Interpretation and Reporting of Sequence Variants in Cancer. Actionable variants were reclassified after integrating the predicted tissue of origin with ctDNA results. Prognostic factors were assessed using Cox regression models, and potential prognostic models were developed and validated for predictive performance. Results: Among 170 patients (median age 59 years), 148 (87.1%) harbored ≥1 genomic alteration, most commonly TP53 (51%), followed by ARID1A , APC , RB1 , PIK3CA , NFE2L 2 , and KRAS . Forty-eight patients (28.2%) carried actionable alterations (13 tier 1, 35 tier 2); 22.9% harbored > 1 targetable mutation. Integration with tissue prediction reclassified 10 patients from tier 2 and 2 from tier 4 into tier 1, providing treatment options to higher-level targeted therapies (e.g., BRCA1/2 , PTEN , KRAS , MSH6 , ALK , ERBB2 , MET , EGFR ). SST improved PFS (HR 0.66, 95% CI 0.47–0.93) and OS (HR 0.62, 95% CI 0.42–0.92). Independent adverse factors included elevated LDH, bone metastasis, male gender, and CDKN2A or KRAS mutation. The Cox proportional hazards model achieved the most robust prognostic performance, with a C-index of 0.751 and time-dependent AUCs > 0.70 at 1-, 2-, and 3-year follow-up. Conclusions: Integration of ctDNA with tissue origin prediction can escalate high level targeted treatment and may provide a practical framework for precision oncology in CUP, warranting further exploration in randomized clinical trials.