GI Oncology Daily Digest

Daraxonrasib is the oral RAS(ON) multi-selective tri-complex inhibitor whose phase 1/2 activity in previously treated RAS-mutant metastatic PDAC justified the randomised phase III RASolute 302 trial. This paper asks the question that matters next: when it stops working, why? The investigators sequenced more than 800 genes in paired pretreatment and end-of-treatment circulating tumour DNA from 44 patients. Treatment-emergent alterations in the RAS signalling pathway appeared in 26 of 44 patients (59%), dominated by mutant KRAS amplification in 16 of 44 (36%), with additional alterations in RTK (4/44, 9%), MAPK (11/44, 25%) and PI3K (4/44, 9%) pathways. The most striking finding is a negative one: no acquired secondary KRAS mutations were observed, a resistance profile categorically different from the mutant-selective KRAS G12C(OFF) inhibitors, where on-target secondary mutations are the recurring story. Human and murine preclinical models reproduced the same mechanisms — mutant KRAS and MYC amplification, RTK upregulation — and combining daraxonrasib with DNA damage response agents, RTK inhibitors, or the mutant-selective RAS(ON) G12D inhibitor zoldonrasib averted resistance in those models.

Cholangiocarcinoma has a poor prognosis largely because it progresses asymptomatically and there is no established screening strategy, so the practical question is not how to screen everyone but how to decide whom to screen at all. This group trained machine-learning models on 487,495 UK Biobank participants, 649 of whom developed CCA during follow-up, using English data (80%) for five-fold cross-validation and withheld Scottish, Welsh and Newcastle data (20%) for testing. Iterative ablation reduced more than 150 candidate features across demographics, lifestyle, health records, blood parameters, genomics and metabolomics down to models built on just five and ten routinely available clinical parameters, with biliary-disease-associated health records and gamma-glutamyltransferase carrying most of the signal. External validation spanned the Penn Medicine Biobank (n=2,638), All of Us (n=330,433), the Japanese JMDC claims database (n=8,425,522) and TriNetX (n=728,886), yielding AUROCs of 0.71, 0.77, 0.796 and 0.80 respectively. In All of Us, the Youden-optimised threshold gave a number needed to screen of 79, and a group-level TriNetX analysis produced hazard ratios up to 82.5 (95% CI 26.4-257.96). Performance declined as the interval between assessment and diagnosis lengthened, and separate intrahepatic and extrahepatic models did not improve on the pooled version.