Reading the China Question from the IP
An analysis of the structural-IP layer in the post-2020 CLDN18.2 patent record, read against the backdrop of recent China-originator deal flow and Representative Moolenaar's letter to Treasury Secretary Bessent about the COINS Act.
On May 21, 2026, Representative Moolenaar, as Chairman of the House Select Committee on the Chinese Communist Party, wrote to Treasury Secretary Bessent urging the Department to cover biotechnology as a “prohibited technology” under the Comprehensive Outbound Investment National Security (COINS) Act of 2025. The letter’s deal statistics are striking and worth noting in their own right. Chinese drug developers’ cross-border out-licensing transactions reached approximately $136 billion in 2025, with 48% of all global pharmaceutical licensing deals of $50 million or more inked with Chinese companies — up from 0% in 2020. The first quarter of 2026 alone produced $60 billion in Chinese out-licensing, nearly half of the total for all of 2025.
The recent activity that BD&L professionals are reading about — Positive Phase III outcome in gastric cancer announced on June 4 from the Innovent-Takeda alliance (commenced October, 2025 with $1.2 billion upfront, including a $100 million equity investment at a 20% premium, and potential total value of up to $11.4 billion); BMS-Hengrui at $15.2 billion in May, Pfizer-Innovent at $10.5 billion the same month — is the indisputable deal reality, regardless of how Treasury eventually resolves the COINS Act framing.
What Rep. Moolenaar’s letter does not address, and what the BD&L professional cannot avoid, is the structural-IP question underpinning each of these China-based transactions. When Takeda’s October 2025 deal with Innovent paid $1.2 billion upfront for a three-asset package — an outlicense of IBI343 (the CLDN18.2 ADC arcotatug tavatecan that in June became the first CLDN18.2 ADC to meet a Phase 3 primary endpoint), joint co-development of IBI363 (a PD-1/IL-2 α-bias bispecific), and an option on IBI3001 (an EGFR/B7H3 bispecific ADC) — the structural rigor of the underlying claims determines whether the IP layer that supports each asset is foundational or participation-level. When BMS transfers IP and know-how bidirectionally to Hengrui in a $15.2 billion partnership, the durability of the IP layer on each side determines whether the partnership terms are reasonable or generous. When Pfizer pays $10.5 billion for access to Innovent’s ADC portfolio, the structural rigor of the underlying claims determines whether the acquisition is foundational or participation-level. In each instance, the deal announcement gives BD&L a valuation signal; but the patent record provides the basis on which to evaluate whether that valuation signal aligns with what is actually being acquired.
The patent record, read at the level of structural rigor in the issued claims and the prosecution layer underneath them, is the only honest answer to the question Rep. Moolenaar’s letter implicitly raises to Treasury Secretary Bessent: What is American capital actually buying when it crosses into the Chinese biopharma sector, and how does that compare to what it is buying when it crosses into the European, Japanese, or domestic sectors instead?
Claudin 18.2 (CLDN18.2) is the target class where these questions converge most sharply right now, and the timing could not be more apt. The tight-junction protein expressed on gastric, gastroesophageal-junction, and pancreatic tumors became the foundation of Astellas’s zolbetuximab (approved 2024 as Vyloy in major markets) and is now the anchor of a competitive landscape spanning naked monoclonal antibodies, CAR-T programs (most prominently CARsgen’s CT041), bispecific T-cell engagers, and antibody-drug conjugates. On June 4, 2026 Innovent announced that its CLDN18.2 ADC arcotatug tavatecan (IBI343, Takeda code TAK-921) met its primary endpoint in the international Phase 3 G-HOPE-001 trial in advanced refractory gastric cancer, with the China NMPA accepting the NDA under priority review. The world’s first CLDN18.2 ADC submitted for regulatory review is a Chinese-originated asset, validated in a multi-center China-and-Japan Phase 3 readout, licensed for global ex-Greater-China rights to a Japanese pharma in a deal that closed six months earlier. That reality captures, in its single chain of relationships, the concern behind Rep. Moolenaar’s letter: Chinese-originator biopharma innovation, increasingly first-in-class, increasingly competitive on Phase 3 clinical data, and increasingly acquired by Japanese and American pharma at multi-billion-dollar deal valuations. There’s no disputing this is happening; the question is what is actually being acquired.
The post-2020 USPTO patent record for CLDN18.2 captures 266 biopharma filings across the four primary modality buckets, including substantial participation from Chinese-originator filers. Read against the six-component KC scoring framework that anchored the GLP-1 backtest, the cohort is 39.1% Chinese-originator — by a substantial margin the highest Chinese-originator share among the target classes the rDNA.ai series has analyzed to date. That cohort-level share, though, is not the analytical observation worth keeping. The cohort is the entire field; what matters for BD&L diligence is what the structural-IP layer reveals as foundational within the field, and whether the Chinese-originator share at the foundational layer matches the share at the cohort level. It does not. The cohort is 39.1% Chinese-originator. The top quartile is 33.3% Chinese-originator. The top decile is 25.9%. The top ten is 10%. The decrease is strongly monotonic — Chinese-originator filings concentrate progressively further from the structural-IP layer that the framework reads as foundational, with each tightening of the analytical lens.
The composition of the top decile makes the pattern concrete. Allogene Therapeutics holds the tournament-calibrated rank #1 (a constitutively-active chimeric cytokine receptor CAR-T platform patent, ensemble 4.225 with a 26-0 pairwise sweep) and rank #2 (a protease-activating CD45-gate CAR co-assigned with Pfizer). Astellas holds rank #3 (the post-Ganymed drug-conjugate composition that anchors the zolbetuximab lifecycle-extension layer) and three further top-decile filings in the naked-mAb and combination-therapy buckets. Regeneron sits at rank #4 with an engineered T-cell receptor patent. Suzhou Transcenta — the highest-ranked Chinese-originator filing in the cohort — sits at calibrated rank #8, having moved from rank #1 in the pre-tournament ensemble after losing nine of its twenty-six head-to-head pairwise comparisons. Shenzhen Frontiergate’s dendritic-cell-activating CAR-T sits at rank #5, the only Chinese-originator filing in the top ten. The remaining Chinese-originator presence in the top decile is concentrated in the ADC bucket — Beijing Xuanyi at ranks #18 and #24, Shanghai TTM-Bio at #19, Sichuan Kelun-Biotech at #20, Biotheus at #10, Harbour Biomed at #25 — clustered in the middle and lower half of the calibrated rankings rather than at the top. Innovent’s four CLDN18.2-specific patents (an anti-CLDN18.2/CD3 bispecific, a naked-mAb composition, and two CAR-T constructs) sit at cohort ranks 44, 50, 162, and 188 — all in the cohort’s middle bands, none in the top decile.
The Innovent placement is worth dwelling on alongside the IBI343 Phase 3 result, because together they illustrate what the structural-IP layer can and cannot tell the BD&L professional. The structural-IP mechanism reads Innovent’s CLDN18.2 patent estate as competent middle-band IP — well-drafted, structurally sound, but not foundational in the sense the framework defines (claim craftsmanship, structural anchoring, dependent-fallback hierarchy, prior-art neighborhood, Amgen v. Sanofi enablement, multi-jurisdictional breadth). The Phase 3 G-HOPE-001 result, by contrast, is clinical validation of a specific asset — IBI343 — against the actual disease endpoint that determines commercial value. This is a similar distinction to what the GLP-1 back test demonstrated: a structural IP reading can’t predict an indication shift that results in commercially significant clinical outcomes.
The deal timing matters here. Takeda’s $1.2 billion upfront commitment in October 2025 valued a three-asset package in which IBI343 was already in Phase 3 enrollment in gastric cancer (G-HOPE-001), had received Breakthrough Designation from China’s NMPA for gastric and Fast Track Designation from the U.S. FDA for pancreatic cancer, and had completed a global Phase 1/2 in previously-treated pancreatic ductal adenocarcinoma. Eight months later, the Phase 3 primary endpoint hit and the China NDA was accepted under priority review. The deal economics responded to clinical de-risking that was substantially in place at signing; the Phase 3 result was an expected reduction of remaining clinical risk, not a discovery of new asset value. The structural-IP framework, by contrast, reads the IP estate underneath the asset — middle-band cohort ranking, competent rather than foundational — as a separate question that the deal economics do not answer. The BD&L professional reading the IP layer would not have predicted IBI343’s Phase 3 success from the IP alone, and the framework does not claim to. What the framework would have said, accurately, is that the IP estate underneath IBI343 is structurally sound but not the cohort’s foundational layer — which is a different question than whether the asset works clinically, and a useful diligence input that the deal economics alone do not provide.
The Suzhou Transcenta rank-1-to-rank-8 movement is worth dwelling on because it illustrates what the tournament-calibration layer is doing. The ensemble-mean ranking, which is the cohort-wide structural score before pairwise comparison, surfaced Suzhou Transcenta as the cohort’s single highest-scoring filing — a Chinese-originator ADC composition reading as more structurally rigorous than any other patent in the field. The pairwise tournament refined that read by directly comparing the patent against each of the other 26 top-decile filings; Suzhou Transcenta won 17 and lost 9, which is a top-decile result but not a single-foundational-filing result. The patent is foundational-quality IP; it is not the cohort’s single foundational IP. The methodology working as designed reveals the distinction between “strong filing in a competitive field” and “single dominant filing across the competitive field,” and the BD&L professional valuing any specific CLDN18.2 transaction needs to know which of those two results applies to the IP being acquired.
What the structural read does not say is that Chinese-originator IP is broadly weaker than non-Chinese-originator IP. The Suzhou Transcenta ADC, the Shenzhen Frontiergate dendritic-cell CAR-T, and Biotheus’s bispecific CD3 fusion at rank #10 are real top-decile filings on their own structural merits, and Innovent’s middle-band IP estate is the foundation on which a clinically-validated first-in-class ADC has been built. What the structural read does say is that the foundational-IP layer in this specific target class — post-2020 CLDN18.2, with zolbetuximab approved in 2024 and a maturing post-approval competitive frontier — sits predominantly with Western originators and Allogene’s engineered-cell-therapy platform, with specific Chinese-originator exceptions that the framework recognizes wherever the structural rigor exists. The 39.1% cohort share and the 25.9% top-decile share are both accurate; but they describe different questions. Cohort share answers “who is filing in the field”; top-decile share answers “whose filings the structural-IP framework reads as foundational.” The first question is what the deal statistics in Rep. Moolenaar’s letter to Treasury measure; the second question is what the BD&L professional needs answered before valuing a transaction.
The six-component KC scoring framework, the tournament-calibrated top-decile rankings for CLDN18.2 (266 patents), the per-patent sub-dimension breakdowns, the cohort-category-primary classification across the four modality buckets, and the methodology documentation are written up in full at rDNA.ai, along with the underlying scoring results. The cohort captures the 2020-forward post-Phase-III-readout competitive landscape, which is the period most BD&L diligence on CLDN18.2 is now valued against; the pre-2020 Ganymed/IMAB362 foundational priority chain predates the discovery horizon and is acknowledged in the methodology section as a known constraint.
What the structural-IP layer can do for the BD&L professional reading Rep. Moolenaar’s letter to Treasury and the recent China-originator deal flow is keep the professional from confusing the field with the foundation, and from confusing clinical validation with foundational IP. The Innovent IBI343 Phase 3 result that landed on June 4 is a clinical validation of an asset built on competent middle-band CLDN18.2 IP, licensed to a Japanese partner at multi-billion-dollar deal economics. None of those facts are inconsistent. The 39.1% Chinese-originator cohort share in CLDN18.2 is an accurate description of who has been filing in the field over the past six years. The 25.9% top-decile share, and the 10% top-ten share, are accurate descriptions of how the structural-IP framework reads the foundational layer within that field. Both numbers are honest; they answer different questions. The BD&L professional valuing the next CLDN18.2-related alliance, partnership, or acquisition needs both answers, and needs to recognize that the structural-IP layer at the foundation of one target class is not the structural-IP layer at the foundation of another. Reading the China question from the IP rather than from the deal flow reveals where the pattern holds, where it breaks, and where the analytical center sits within each target’s specific competitive landscape — which is the diligence layer the professional needs to bring to the deal negotiation.
This is the seventh article in the rDNA.ai biopharma BD&L series.