The Evidence Layer for Scientific AI

Evidence for causality starts from CRISPR/Cas9 loss-of-function and rescue: in human B-ALL cell lines (Nalm6, 697), a NUDIX-targeted CRISPR screen and subsequent NUDT5 knockout identify NUDT5 as a strong thiopurine-response regulator, and re-expression of NUDT5 restores thiopurine sensitivity. Mechanistically, NUDT5 loss depletes thioguanine-derived nucleotide pools (e.g., TIMP and DNA-TG) measured by targeted LC-MS/MS and perturbs purine metabolism consistent with impaired salvage via HPRT activity (with isotope tracing suggesting salvage reduction more than de novo collapse). NUDT5 knockout/rescue and metabolite depletion

How to implement the experiment (scientifically, evidence-aligned)

• Perturbation design: create isogenic NUDT5 loss lines in patient-derived leukemia cells (or model cells supporting the mechanistic linkage) and include NUDT5 rescue to distinguish on-target effects from editing artifacts. NUDT5 rescue confirms specificity
• Causality readouts: under thioguanine/mercaptopurine exposure, quantify thiopurine metabolite pools (TIMP, TGTP, DNA-TG) with targeted LC-MS/MS and couple this to broader targeted metabolomics to map pathway position. Targeted LC-MS/MS metabolite pools
• Clinical bridge: test whether germline NUDT5 variants correlate with thiopurine pharmacodynamics such as dose intensity and NUDT5 expression (rs55713253 shows both in the reported cohort). rs55713253 dose-intensity association

What remains uncertain / critical control logic

• Patient-derived generalization: the mechanistic causality is demonstrated in human B-ALL cell lines; effects may be subtype-dependent in primary leukemia. Generalizability limits acknowledged by authors
• Mechanism completeness: the reported link involves impaired salvage/HPRT activity, but the exact molecular intermediary path from NUDT5 to HPRT function is not fully resolved. Mechanistic gap (NUDT5→HPRT)