Remember, a clean plate doesn't mean the meal was safe — it means you didn't check the kitchen. Go crack some eggs.
👀 The Read
Rat: Clean. Monkey: Clean. Humans: Liver Injury.
A spotless tox package, wide safety margins, and a liver signal only one species could see.
📋 THE RECEIPT
Drug: TAK-994 (oral orexin-2 receptor agonist) — Takeda
Indication: narcolepsy
What happened: withdrawn from Phase 2 for severe drug-induced liver injury
The package said: rat clean, NHP clean, in vitro margins >100×
The miss: a reactive metabolite the standard studies were never built to see
Takeda had a promising oral orexin agonist for narcolepsy. The liver looked like a non-issue — every standard tox readout said so. Then Phase 2 patients developed severe drug-induced liver injury, and the drug was pulled.
What the package showed. Rat safety studies: clean liver. Non-human primate: clean. The in vitro DILI panel: wide margins to every known intrinsic mechanism — cytotoxicity and mitochondrial toxicity over 100× Cmax/IC50, bile-salt-efflux inhibition over 20×. On paper, a spotless hepatic profile.
What happened in humans. Severe DILI in Phase 2. Program withdrawn.
What the post-mortem found. The mechanism was CYP450 induction generating a reactive, covalent-binding metabolite. And the hepatic necrosis only appeared in mice — a species not in the standard package — and only once Takeda went looking, during the clinical trials.
A clean tox package isn't a clean liver. It's a clean tox package.
The smoking gun:
❌ Wide margins from the wrong assays. The in vitro panel tested the parent compound against known intrinsic DILI mechanisms. The killer was a metabolite formed by induction — something a parent-compound assay will never surface, no matter how big the margin. A 100× margin against the wrong mechanism is 0× protection.
❌ A two-species package that couldn't see it. Rat and NHP were clean; mouse wasn't. The toxicity was species-specific and metabolism-driven — exactly the liability standard species selection misses when metabolism isn't driving the choice.
Two clean species, huge in vitro margins, and a drug that still destroyed livers in Phase 2.
So how does a hepatotoxin walk through a full tox package untouched?
→ DMPK and tox run in parallel, instead of METID choosing the tox species and assays
→ "wide safety margins" trusted as an answer, instead of interrogated for which mechanism they actually cover
→ the standard package treated as sufficient by default — the checklist, not the molecule
These are the exact traps I watch small-molecule founders walk into while trusting a clean-looking package. If this is resonating, keep reading ⏬
🔧 The Move
Take five minutes on your lead candidate:
Pull your hepatic safety data and ask one question — did these assays test the metabolites, or only the parent compound?
If your reactive-metabolite and enzyme-induction liabilities were assessed on the parent, or waved off by "wide margins," you haven't cleared the liver. You've cleared the wrong compound.
📍 Reply with one line — did your DILI work test the parent only, or the metabolites too? I'll tell you what that leaves exposed. That's what I do.
👀 Wish you understood nonclinical better?
Check out my course, $499, use WELCOME50 for $50 off:
