zimberelimab

Zimberelimab is Arcus Biosciences' anti-PD-1 antibody, licensed from Chinese biotech Gloria/WuXi (originator code GLS-010) and approved in China as Yu Qi Shu for classical Hodgkin lymphoma [1]. In the US and Europe it remains investigational. Its strategic purpose was never to compete head-to-head with Keytruda (Merck's $29.5B pembrolizumab franchise in 2024) [2] or Opdivo (nivolumab, ~$9B). It exists as the backbone PD-1 that lets Arcus test its downstream assets, primarily domvanalimab (anti-TIGIT), quemliclustat (CD73 inhibitor), and etrumadenant (A2A antagonist), in combination without paying royalties to Merck or BMS. The lead Phase 3 was STAR-221 (NCT05568095), which tested domvanalimab + zimberelimab + chemotherapy against nivolumab + chemotherapy in first-line HER2-negative upper gastrointestinal (gastric, gastroesophageal junction [GEJ], and esophageal) adenocarcinoma [3]. STAR-221 was discontinued for futility and presented negative results at the 2026 ESMO GI Congress: median overall survival 14.0 versus 13.4 months, with no benefit in the PD-L1-high subgroup either [4]. The companion NSCLC Phase 3, STAR-121 (NCT05502237), was discontinued for futility in April 2026 [5]. Gilead's initial 2020 investment was approximately $375M ($175M upfront collaboration payment plus a $200M equity purchase), with the ten-year deal envisioning several billion in potential milestones [6]. The whole thesis hinged on whether TIGIT plus PD-1 blockade could beat PD-1 alone in randomized Phase 3, and as of mid-2026 the answer for this asset pair is no. Zimberelimab as a standalone commercial asset is a small story; the combination bet that mattered has now failed.

Zimberelimab was approved in China in August 2021 as Yu Qi Shu (originally GLS-010, from Gloria Biosciences and WuXi Biologics) for relapsed/refractory classical Hodgkin lymphoma [1]. Arcus licensed ex-China rights in 2020 and has not filed for US or EU approval as monotherapy. The FDA/EMA route ran through combination trials, where zimberelimab traveled as the PD-1 partner to domvanalimab and other Arcus molecules. No FDA breakthrough or fast-track designation has been disclosed for zimberelimab itself; any regulatory pathway would have been driven by the doublet or triplet regimen it supports. That pathway has now narrowed dramatically. STAR-221 (NCT05568095) in first-line HER2-negative gastric/GEJ/esophageal adenocarcinoma missed its primary endpoint of overall survival and was discontinued for futility on recommendation of the independent data monitoring committee; results were presented at the 2026 ESMO GI Congress [4]. STAR-121 (NCT05502237) in first-line metastatic NSCLC with no actionable gene alterations was similarly discontinued for futility, announced April 21, 2026 [5]. Prior Phase 2 randomized data in NSCLC came from ARC-7, which showed domvanalimab + zimberelimab reduced risk of progression or death by ~45% versus zimberelimab alone in PD-L1-high 1L NSCLC [7]. That Phase 2 signal did not replicate in the STAR-121 Phase 3. Randomized supporting data also read out in biliary tract cancer (Phase 2, immunotherapy-refractory) [8] and pancreatic adenocarcinoma in combination with quemliclustat [9]. Yu Qi Shu revenue in China is not separately disclosed by Gloria/WuXi and appears modest against Keytruda's global scale [2]; no ex-US commercial launch of zimberelimab as monotherapy has been pursued by Arcus.

PD-1 is a receptor on T cells that acts like a brake pedal. When tumor cells (or infected cells) display the ligand PD-L1, PD-1 engagement tells the T cell to stand down. Tumors exploit this by overexpressing PD-L1, hiding from immune attack. An anti-PD-1 antibody like zimberelimab binds PD-1 and physically blocks the ligand from docking, taking the foot off the brake so T cells can recognize and kill tumor cells. The mechanism is as validated as any target in oncology. Pembrolizumab (Keytruda) generated $29.5B in 2024 revenue for Merck [2], and nivolumab (Opdivo) roughly $9B for BMS. Six PD-(L)1 antibodies are FDA-approved across dozens of indications. Zimberelimab is a fully human IgG4 antibody engineered along class-standard lines. The interesting question was never whether PD-1 blockade works but whether adding domvanalimab, which blocks TIGIT (a second inhibitory receptor T cells upregulate when PD-1 is engaged), delivers additive benefit. The rationale is straightforward: release one brake, T cells compensate by pressing another; block both, and you get a deeper anti-tumor response. Domvanalimab is Fc-silent, a deliberate design choice. The Fc region is the tail of an antibody that recruits immune cells and can trigger antibody-dependent cellular cytotoxicity; engineering the Fc to be silent prevents depletion of TIGIT-expressing regulatory T cells and effector T cells the drug is meant to reactivate. Roche's tiragolumab, which failed in SKYSCRAPER-01, is Fc-competent; the Fc-silent hypothesis was central to the argument that domvanalimab would succeed where tiragolumab failed [10]. In practice, several anti-TIGIT antibodies including domvanalimab have now failed to deliver on that hypothesis in randomized Phase 3 trials, so the mechanistic story looks weaker than it did in 2022.

STAR-221 (NCT05568095) was a randomized, open-label Phase 3 trial in untreated, locally advanced unresectable or metastatic HER2-negative gastric, gastroesophageal junction (GEJ), or esophageal adenocarcinoma [3]. Patients were randomized 1:1 to domvanalimab + zimberelimab + FOLFOX (or CAPOX) versus nivolumab + the same chemotherapy backbone. Choosing nivolumab + chemo as the active comparator, rather than pembrolizumab + chemo, reflected CheckMate-649 as the established regimen in this indication for PD-L1 CPS ≥5 patients (CPS, or Combined Positive Score, is a scoring system that counts PD-L1-staining tumor and immune cells relative to total tumor cells; higher CPS predicts stronger PD-1 monotherapy benefit). Pembrolizumab + chemo (KEYNOTE-859) is also approved in 1L gastric/GEJ, so the market includes at least two entrenched PD-1-chemo regimens, and STAR-221's win would have had to beat one and imply superiority over the other. The primary endpoint was overall survival, with progression-free survival, objective response rate, and safety as secondaries. Enrollment target was around 970 patients and the trial enrolled all comers regardless of PD-L1 status. That all-comer design was aggressive and, in retrospect, part of the problem: the triplet did not beat nivolumab + chemo in the intention-to-treat population (median OS 14.0 vs 13.4 months) or in the PD-L1-high subgroup (15.2 vs 17.2 months, numerically worse) [4]. The companion Phase 3 STAR-121 (NCT05502237) in 1L metastatic NSCLC with no actionable gene alterations had dual primary endpoints of PFS and OS versus pembrolizumab + chemo, with a monotherapy zimberelimab + chemo arm as well; that trial was likewise discontinued for futility [5]. The Phase 2 ARC-7 study in PD-L1-high NSCLC that supported STAR-121's design showed a 45% PFS risk reduction for the doublet versus zimberelimab monotherapy [7], a signal that did not translate to Phase 3.

Our model estimates a 33% chance this drug is eventually approved. It starts from the historical base rate for Phase 3 drugs in this area (about 48%), then adjusts using ten facts about the trial and sponsor. What moves the number most: it is helped by its light or open-label blinding, larger-than-typical enrollment for this phase, and the sponsor's strong record of getting drugs approved; it is held back by weak or limited earlier-phase results. The other facts land near average for this stage, so they leave the estimate roughly where the base rate put it.

Efficacy risk is no longer speculative for the lead programs - it materialized. STAR-221 and STAR-121 both failed to beat active PD-1 comparators [4][5]. Two hypotheses that had competed to explain why prior anti-TIGIT antibodies failed [10] now look more like the same story: (1) TIGIT-expressing exhausted T cells may not be functionally reversible with antibody blockade at clinically tolerable doses, and (2) Fc engineering matters - Fc-competent versus Fc-silent antibodies engage different immune cells, and the field bet that Fc-silent (domvanalimab) would succeed where Fc-competent (tiragolumab) failed. That bet has not paid off in Phase 3. Safety risk was moderate and remained so: PD-1 blockade carries known immune-related adverse events (colitis, pneumonitis, hepatitis, endocrinopathies), and TIGIT blockade proved additive but not synergistic on toxicity. STAR-221 reported a safety profile comparable to nivolumab + chemo with no new signals [4]. Execution risk was low - enrollment ran on schedule and comparators were on-label. Commercial risk is now moot for the failed Phase 3 indications; the 1L gastric/GEJ market remains split between nivolumab + chemo (CheckMate-649) and pembrolizumab + chemo (KEYNOTE-859), with pembrolizumab likely to expand market share as it loses composition-of-matter exclusivity in 2028 and payer economics shift. Residual risk for Arcus/Gilead is that other Arcus combination trials read through negatively as well, further deprioritizing the ten-year partnership economics.

The event that would have defined this pipeline node has happened, and the answer was no. STAR-221 missed its primary endpoint in first-line upper GI adenocarcinoma [4] and STAR-121 was discontinued for futility in first-line NSCLC [5]. Zimberelimab remains an interesting backbone PD-1 for Arcus's other combination programs (quemliclustat in pancreatic [9], biliary tract [8], resectable mMRd gastric), but the TIGIT combination thesis that made this asset pair valuable to Gilead is materially damaged. Watch for two things going forward. First, whether Arcus and Gilead disclose subgroup analyses of STAR-221 that identify any PD-L1 stratum or molecular subgroup where the triplet showed hazard ratios <0.80 - that would keep a narrower registration path alive but the ESMO GI presentation showed no such signal [4]. Second, whether the Arcus-Gilead partnership is restructured; Gilead's total committed capital (equity, upfront, and milestones) exceeded $1.5B across the ten-year deal, and the failure of both flagship Phase 3s puts pressure on that economics [6]. Arcus (RCUS) has repriced significantly on the readouts. For BioCosm's purposes, this node should be reclassified from active-Phase-3-catalyst to historical-failed-readout with residual optionality in earlier-line combinations. The PoS displayed by the base_rate model is stale; treat it as approximately 10-15% for the residual program value, not 43.6%.