by Dennis Crouch
In consecutive weeks, the Federal Circuit has reversed two jury verdicts totaling over $80 million in pharmaceutical patent disputes, holding in both cases that the specifications failed (as a matter of law) to satisfy §112(a)’s disclosure requirements. Seagen Inc. v. Daiichi Sankyo Company, Ltd., Nos. 2023-2424 (Fed. Cir. Dec. 2, 2025); Duke University v. Sandoz Inc., No. 2024-1078 (Fed. Cir. Nov. 18, 2025).
In my view, these cases represent a significant doctrinal development. I see these as revealing three interconnected trends:
- The Federal Circuit’s increasing willingness to treat patent specifications as legal texts subject to judicial interpretation of the text rather than factual determinations of its meaning;
- The practical (re)convergence of written description and enablement into a unified adequacy inquiry; and
- The court’s growing confidence in overturning jury verdicts on §112(a) grounds as a matter of law.
Seagen involved U.S. Patent 10,808,039, covering antibody-drug conjugates (ADCs) used in cancer treatment. The idea here is to combine a cell-targeting antibody with a cell-killing cytotoxic, connected by a “linker” protein. All three components are important, including the linker protein because it determines whether and how the cytotoxic drug is released. Strong binding reduces efficacy and weak binding exposes healthy cells to the toxic compounds. The ‘039 patent claims ADCs containing a specific tetrapeptide linker unit limited to glycine and phenylalanine amino acids, a subgenus encompassing 81 possible species (reflecting three options for each of four amino acid positions or 3x3x3x3).
Seagen filed the application in July 2019 but claimed priority to a November 2004 application. The timing mattered because Daiichi had invented its blockbuster drug Enhertu in 2011 and publicly disclosed its structure (which uses a Gly-Gly-Phe-Gly tetrapeptide falling within the claimed subgenus) in December 2015. The basic question in the case boils down to whether the 2004 application provides adequate support for the claimed sub-genus. A jury in the Eastern District of Texas sided with the patentee – finding the claims valid and willfully infringed – awarding >$40m in damages. Judge Lourie, writing for a unanimous panel, reversed on both written description and enablement grounds.
In parallel proceedings, Daiichi and AstraZeneca petitioned for post-grant review of the same claims, which the PTAB granted. In January 2024, the Board issued a final written decision holding the asserted claims unpatentable on the same written description and enablement grounds. Seagen appealed that decision as well. In a companion opinion issued the same day, the Federal Circuit dismissed Seagen’s appeal of the PTAB decision as moot given its reversal of the district court judgment.
Duke v. Sandoz presented a parallel situation. The ‘270 patent claimed methods of treating hair loss using prostaglandin F analogs with specific structural features at two positions: an amide group at R1 and phenyl at Z. Allergan markets Latisse (containing bimatoprost) under this patent. When Sandoz was sued for its version, the generic maker stipulated to infringement but challenged validity. The specification describes billions of possible compounds through branching categorical definitions at multiple variable positions. The asserted claims cover a few thousand of those compounds (depending on how one counts). But, no working example within the claim’s scope was disclosed.
A Colorado jury sided with the patentee – finding that Sandoz failed to prove invalidity by clear and convincing evidence and awarding Allergan $39 million. The appeal focused on whether there were sufficient “blaze marks” to single-out the specific sub-genus in a claim. Oral arguments included dueling metaphors associated with forests and trees. Duke’s counsel characterized the specification as more of a “backyard” rather than a “forest” since the specification (as interpreted by POSITA) includes limited choices from constrained menus. Sandoz countered that the specification presents not 13 trees but “13 forests,” each defined so broadly as to encompass countless variations. Judge Stark, writing for a panel sided with the patent challenger and reversed as a matter of law for lack of written description.
The "blaze marks" metaphor comes from forest navigation. When hiking through dense woods, trail markers painted on or cut into trees guide travelers to their destination. In patent law, the metaphor asks whether a specification provides sufficient guidance for skilled artisans to navigate from a broad disclosure to a specific claimed invention. A patent cannot "disclose a forest in the original application, and then later pick a tree out of the forest" as the invention. Purdue Pharma L.P. v. Faulding Inc., 230 F.3d 1320 (Fed. Cir. 2000); In re Ruschig, 379 F.2d 990 (CCPA 1967)
Both decisions treat the patent specification’s structural logic as determinative, largely independent of expert testimony about what skilled artisans would understand. This approach extends Markman v. Westview Instruments, Inc., 517 U.S. 370 (1996), which held that claim construction presents a question of law for courts rather than juries. As I discussed in a recent post, the Federal Circuit has been steadily expanding this principle beyond claim construction into other specification-based inquiries. See Dennis Crouch, Smartrend and the Stretching of Markman: When Specification Parsing Becomes a Question of Law, Patently-O (Nov. 18, 2025) (same issue with respect to doctrine of equivalents). The Seagen and Duke decisions represent another step in this progression.
In Duke, Judge Stark’s analysis focused on the specification’s branching structure. The court examined the patent’s categorical definitions for the R1 position, which lists 13 options, and found that characterizing this as “13 options” was misleading because most options are themselves large categories requiring additional embedded choices. The court explained that selecting C(O)NHR3 from among the 13 options would be “merely the first decision an artisan would need to make before having to make other decisions just to determine what to place at C1.” This structural analysis proceeded largely without reference to expert testimony. The court quoted its earlier decision in Regents of the University of Minnesota v. Gilead Sciences, Inc., 61 F.4th 1350, 1357 (Fed. Cir. 2023): “Following [such a] maze-like path, each step providing multiple alternative paths, is not a written description of what might have been described if each of the optional steps had been set forth as the only option.”
The Duke court also examined the specification’s preferences and found they pointed away from the claimed invention. Five of the 13 R1 options are identified as “preferred” or “more preferred,” but C(O)NHR3 (the claimed amide) is not among them. The court was left with the basic problem that – although the claimed subgenus was disclosed, it was just thrown out there in the midst of many millions of other subgenera and without any guidance suggesting anything particular about the ones claimed.
In Seagen, Judge Lourie applied the same analytical approach. The 2004 application describes 39 different amino acids that can constitute each position in the peptide unit. When different isomers are included, there are 83 unique options for each amino acid position. For tetrapeptides alone, over 47 million different species (83^4) are encompassed. The claimed 81-member Gly/Phe-only tetrapeptide subgenus represents an infinitesimal fraction of this disclosure. The court found “no adequate blaze marks that would lead a skilled artisan to the 81-member Gly/Phe-only tetrapeptide subgenus or any species within it.”
The principle parallels established doctrine on numeric ranges. A specification disclosing a temperature range of 55-212°F does not provide written description support for a later-amended claim to 103-105°F unless the original disclosure includes some indication directing skilled artisans to that particular subrange. Just as one cannot pluck an undisclosed numeric subrange from a broad continuum, one cannot extract an undisclosed chemical subgenus from a vast combinatorial space, regardless of whether the subrange or subgenus technically falls within the original disclosure’s scope.
But, both cases the patentee presented expert testimony presenting factual evidence that POSITA would have received sufficient direction from the specification – in one case making the claims something of a “straightforward leap.” But the court rejected this – writing that any leap is a leap too far.
[T]hat testimony dooms Seagen’s case. That which one must leap to is obviously not there. . . It in effect admits that a skilled artisan would not understand the inventors of the 2004 application to have “possession” of GFFG.
Here, the court was able to disregard the entire expert testimony. And then in both cases use their own de novo interpretation of the specification to find no written description.
Convergence of Written Description and Enablement
The Seagen decision addressed both written description and enablement, and the parallel analyses reveal how the doctrines have converged for genus claims in unpredictable arts. On written description, the court found the 2004 application’s broad disclosure of over 47 million tetrapeptides failed to demonstrate possession of the claimed 81-member subgenus. On enablement, the court applied the Supreme Court’s reasoning in Amgen Inc. v. Sanofi, 598 U.S. 594 (2023), finding that the functional limitation requiring intracellular cleavage meant skilled artisans would need to conduct assays for each ADC to determine whether it meets the claim limitations. This amounts to impermissible “trial-and-error discovery.”
In Amgen Inc. v. Sanofi, 598 U.S. 594 (2023), the Supreme Court unanimously held that patents claiming entire genera of antibodies by function required enablement of the entire genus. The specification disclosed 26 antibodies and two methods for finding others through screening. The Court held this insufficient, characterizing the methods as "little more than two research assignments" requiring trial-and-error discovery. Justice Gorsuch emphasized: "The more one claims, the more one must enable."
The rhetorical convergence between written description and enablement appeared explicitly in Duke. Judge Stark’s opinion deployed language traditionally associated with enablement analysis: the specification leaves skilled artisans “forced to engage in ‘painstaking experimentation’ to see what works,” provides “little more than two research assignments,” and constitutes “a ‘laundry list’ disclosure of every possible moiety for every possible position.” This language tracks Amgen‘s enablement analysis but was made as part of a written description decision.
I am now having some amount of difficulty seeing any light between these doctrines – especially as the court has moved more strongly to the position of effectively treating both as questions of law that can be decided based upon the court’s reading of the specification and claims.
Professors Jonathan Masur and Lisa Larrimore Ouellette recently explored these “disclosure puzzles” and questioned whether maintaining separate written description and enablement requirements serves any remaining purpose when both now require the same structural proof for genus claims in unpredictable arts. See Jonathan S. Masur & Lisa Larrimore Ouellette, Disclosure Puzzles in Patent Law, 92 U. Chi. L. Rev. 6 (2025). The Seagen and Duke decisions suggest that at least for pharmaceutical genus claims, the doctrines have effectively merged into a unified §112(a) adequacy inquiry asking whether the specification provides sufficient structural information to allow skilled artisans reliable access to the full scope of what is claimed.
Overturning Jury Verdicts as a Matter of Law
The final issue that I want to note follows above, but it continues to grind my gears: the Federal Circuit’s continued confidence in reversing jury verdicts on issues of fact.
Written description presents a question of fact: whether the specification discloses enough that a person of ordinary skill would conclude the inventor possessed the full scope of the claimed invention. Factual questions ordinarily belong to juries, particularly when expert testimony conflicts. Both Seagen and Duke involved conflicting expert testimony and jury verdicts for the patentee, yet both panels concluded that “no reasonable jury” could have found in the patentee’s favor.
The burden on appeal in these cases was “doubly high.” As Judge Stark noted in Duke because of the high clear-and-convincing evidence standard of invalidity. In other words, the JMOL cases require the court to decide that the defendant proved its factual case so well that any reasonable jury would find clear and convincing evidence of invalidity.
What made it easier in these cases is that the court has determined that all interpretations and understandings of the specification are under its purview as questions of law. Once the specification’s structure, branching paths, preferences, and blaze marks are characterized as matters of legal interpretation rather than factual understanding, the jury’s role shrinks to near irrelevance. The specification either provides adequate guidance or it does not, and that determination belongs to judges parsing the document much as they parse claim language under Markman. Expert testimony about the meaning of the specification becomes merely informative. This approach effectively converts the written description inquiry from a factual question about possession into a legal question about documentary sufficiency — about what the specification says instead of what it means. In my view, this represents a significant extension of Markman principles beyond those outlined by the Supreme Court.