Genomic features of biliary tract cancers in paired tissues and ctDNA and an integrated molecular classification to predict immunotherapy outcomes.

e16190 Background: Recently, the first-line immunotherapy with immune checkpoint inhibitors (ICIs) significantly improves the prognosis of advanced biliary tract cancers (BTCs). However, the treatment benefit from ICIs is still very limited. It is necessary to further decipher molecular characteristics of BTCs and identify the advantageous population for immunotherapy. Methods: Next Generation Sequencing with a targeted panel were performed for 235 pairs of tissue and ctDNA matched samples from patients with advanced BTCs. An integrated molecular classification based on genomic mutations detected in both tissue and ctDNA was constructed in a multi-institutional subset. Results: Between ctDNA and tissues, the mutation frequencies were significant different in DNMT3A, TP53, ATRX, and KRAS. Specifically, more DNMT3A mutations were detected in ctDNA than in tissues. Compared with the wild-type, DNMT3A mutations were accompanied by significantly more mutations in genes involving in DNA repair. Cases with DNMT3A mutations had significantly higher TMB and MSI incidence than those with the wild-type. Nonetheless, DNMT3A mutations were associated with a poor prognosis of immunotherapy. We further analyzed the impact of TMB on prognosis, and found that both the tissue-based TMB (tTMB) and blood (ctDNA)-based TMB (bTMB) were not related to OS. Then, the number of different variations between tTMB and bTMB was defined as the differential TMB (dTMB). Interestingly, patients with high dTMB had significantly inferior OS than those with low dTMB (p = 0.034). For immunotherapy, high dTMB was associated with a significantly worse ORR than that for low dTMB (84.6% vs. 43.8%, p = 0.024). Furthermore, using the LASSO Cox regression model, mutated genes that significantly impacted prognosis were selected to construct a genetic signature based on tissues and ctDNA, respectively. Most genes were different in these two signatures. However, risk scoring based on any of these two signatures can predict prognosis. Using both the tissue- and ctDNA-based risk scoring, we constructed an integrated molecular classification. Patients were divided into low, moderate, and high risk subtypes, with significantly different prognosis in both the overall population and the immunotherapy population. Finally, the ORR of immunotherapy decreased with the increase of subtype risk, which were 100%, 76.9%, and 38.5% for the low, moderate, and high risk subtypes, respectively. Conclusions: DNMT3A mutation was an indicator of genomic instability in advanced BTCs. The dTMB, a novel definition of TMB, was associated with clinical outcomes. The integrated molecular classification using both the tissue- and ctDNA-based risk scoring can predict immunotherapy outcomes. These results may contribute to the development of precision therapy in advanced BTCs.