Wyeth Patented Any Compound That Binds Cysteine 773. It Never Said What Dose Wouldn't Poison the Patient.
By Prashant Kotian, PhD Researcher (Chemistry), ICT Mumbai
Non-small cell lung cancer driven by an overactive EGFR receptor responds well to drugs like gefitinib and erlotinib, until it doesn't. Both are "reversible" inhibitors: they sit on the receptor through a non-covalent bond, and non-covalent bonds let go. Tumors that develop resistance to gefitinib or erlotinib have been a known problem in oncology for two decades, and the proposed fix has always been the same one organic chemists reach for whenever a non-covalent interaction isn't holding: replace it with a covalent one.
Wyeth's two patents at issue in this case, US 10,603,314 and US 10,596,162, tried to claim that fix at the broadest possible level. The claims covered a method of treating gefitinib/erlotinib-resistant NSCLC by administering "a unit dosage of an irreversible EGFR inhibitor that covalently binds to cysteine 773 in the ligand-binding pocket of EGFR." Not a compound. Not a class of related compounds sharing a scaffold. Any compound, so long as it forms that one covalent bond to that one cysteine residue.
The specification backing that claim disclosed exactly three compounds - EKB-569, HKI-357, and HKI-272 - all tested only in vitro. No patient was ever dosed with any of them in the patent's own data. What the specification offered instead of real dosing examples was a "general" range (0.5 to 1000 mg/kg) and a "projected" total daily dose (1 to 1000 mg, preferably 2 to 500 mg) - numbers explicitly hedged as estimates, not measurements.
That gap between test-tube potency and a survivable human dose turned out to be the whole case. At trial, AstraZeneca put in unrebutted evidence that if you actually dosed patients within the ranges the patent itself disclosed, two of its own three named compounds - HKI-272 and EKB-569 - would exceed the maximum tolerated dose in humans, in one case by roughly fivefold. Wyeth's own expert agreed. So did a co-inventor, who testified that the concentrations in the patent's in vitro data were higher than anything you could actually give a patient, and that Wyeth's own scientists said as much in a 2008 publication - after the patent had already been filed, using data the specification never had.
A jury didn't see it that way the first time around, siding with Wyeth and awarding $107.5 million after finding AstraZeneca's Tagrisso (osimertinib) - itself a real, marketed irreversible covalent EGFR inhibitor - induced infringement. The district court threw that verdict out on judgment as a matter of law, and on July 9, 2026 the Federal Circuit (Lourie, joined by Linn and Hughes) affirmed: the patents are invalid for lack of enablement under Section 112(a).
The reasoning tracks the same logic the Supreme Court used in Amgen v. Sanofi three years earlier for antibody genus claims: the broader you claim, the more your specification has to actually teach. Wyeth claimed every covalent EGFR-cysteine-773 binder, at every dosage a "skilled artisan" might calculate. What it disclosed was in vitro data for three compounds and a range with no stated method for translating potency into a dose a patient could survive. The court didn't require Wyeth to run an FDA-grade clinical trial inside its patent - it required the patent to at least show a skilled chemist how to get from test tube to patient without guessing. Handing a chemist a range that includes toxic doses and calling the rest "the judgment of the practitioner" is not that.
There's a lesson here that has nothing to do with cancer specifically. A functionally-defined genus - "any compound that does X" - is exactly the kind of claim that reads broadest on paper and holds up worst in court, because breadth is a promise the specification has to keep. Every patent drafted this way, in any therapeutic area, is carrying the same risk this one just lost on.
What we do
See if this applies to one of your matters
We run genus/species (Markush) claim coverage analysis, Section 3(d) compliance screening, and prior-art structural triage on real chemical structures — not keyword search. If a case like this one is live for you, we'll run a free sample on a compound of your choosing.