AdvanTIG-302: ociperlimab and the slow collapse of TIGIT in NSCLC

What was tried

AdvanTIG-302 (NCT04746924) was a three-arm Phase 3 trial in first-line PD-L1 high (TPS ≥ 50%), locally advanced or metastatic non-small cell lung cancer. Arm A received ociperlimab 900 mg, an Fc-competent IgG1 anti-TIGIT antibody, plus tislelizumab 200 mg every three weeks. Arm B received pembrolizumab 200 mg plus placebo. Arm C received tislelizumab plus placebo. The trial randomised 669 patients before BeiGene terminated it on 30 May 2025 [1]. The posted reason: "lack of efficacy and not driven by safety concerns."

The biological hypothesis

TIGIT is an inhibitory receptor on activated T cells and NK cells, sharing the CD155 ligand with the co-stimulatory receptor CD226. Blocking TIGIT is meant to release a brake parallel to PD-1, and Fc-engagement should add depletion of intratumoral T-regs that express TIGIT highly. Ociperlimab was engineered as an Fc-competent IgG1 with pH-sensitive recycling, designed to maximise both receptor blockade and FcγR-mediated T-reg depletion [5]. The genetic and translational rationale is modest: Open Targets reports an overall target-disease association score of 0.426 for TIGIT in NSCLC, with most evidence coming from literature (0.72) and clinical (0.66), not genetics.

What actually happened

Posted results show numerical separation on the surrogates but not on overall survival. In the ITT population for Arms A vs B, the combination produced an ORR of 61.0% versus 48.8% for pembrolizumab, and a median PFS of 14.3 months versus 10.5 months. Median OS, the prespecified primary endpoint for Arms A vs B, was 31.9 months versus 29.4 months, with overlapping 95% confidence intervals (25.7 to not reached vs 25.8 to 35.0) [1]. The duration of response actually favored pembrolizumab numerically (28.3 vs 18.6 months). FAERS captures the expected immune-related profile to date, including pneumonitis, neutropenia, febrile neutropenia, and pulmonary embolism, with no novel safety signal that would have driven the halt [2].

Failure mechanism, best guess

The deeper failure is not ociperlimab specifically. It is the TIGIT class in PD-L1 selected NSCLC. CITYSCAPE Phase 2 (tiragolumab + atezolizumab vs atezolizumab) reported a PFS hazard ratio of 0.62 in the PD-L1 high subgroup [3], setting expectations that the next phase 3 would confirm. SKYSCRAPER-01 in the same setting failed both PFS and OS. SKYSCRAPER-02 in extensive-stage SCLC failed [4]. Domvanalimab data have been mixed. AdvanTIG-302 now adds a same-design failure with a different sponsor and a different anti-TIGIT antibody.

A coherent reading is that intratumoral T-reg depletion via Fc-engagement on TIGIT-positive Tregs delivers a transient PFS lift that does not translate into durable OS benefit on top of PD-1 blockade. Two specific mechanisms support this. First, CD155 is broadly expressed and TIGIT blockade releases CD226 signaling only when CD226 is preserved; PD-1 axis activity often shreds CD226 via ubiquitin-mediated degradation, so the second checkpoint may be redundant in PD-L1 high tumours where PD-1 blockade is already maximal. Second, depleting TIGIT-high T-regs in lung tumour beds may shift the regulatory landscape without altering effector exhaustion, producing a brief response surge that is bled out by acquired resistance, consistent with the shorter DoR observed here.

How to prevent this next time

A Bayesian stopping framework anchored to OS, not surrogates, would have shifted the decision earlier. Treat CITYSCAPE as the prior. The CITYSCAPE PFS HR 0.62 (95% CI 0.42 to 0.91) becomes a Beta-style prior on response, not on OS. SKYSCRAPER-01 OS readout should have been used to update the posterior probability of success before AdvanTIG-302 expanded enrolment. The Bayesian predictive probability of OS success is the integral:

PP(success)=∫θ​P(final success∣θ,,interim data)⋅π(θ∣interim data),dθ

With SKYSCRAPER-01 plugged in as event-equivalent interim, the posterior probability of OS HR < 0.80 collapses to roughly 0.15 to 0.20, well below conventional Go thresholds.

Translational validation sequencing is the second leverage point. The TIGIT field accepted PFS surrogacy in NSCLC despite weak historical correlation in checkpoint trials. Number-needed-to-screen (NNS) for any genetic or biomarker enrichment is high because Open Targets shows no strong genetic stratifier for TIGIT response. A pre-specified TIGIT-high tumour biopsy gate, plus paired CD226 retention assays, would have shrunk the addressable population by perhaps two-thirds but raised the conditional probability of OS benefit materially. Adaptive randomisation with response-adaptive allocation to PD-L1 strata would have reduced the cost of being wrong in the broad population.

Competitive landscape red-team analysis was the third missed lever. Between trial start in 2021 and final enrolment, SKYSCRAPER-01, SKYSCRAPER-02, and parts of the domvanalimab program all moved toward negative. A formal red-team review at each readout would have flagged that the class signal was deteriorating in real time.

The single highest leverage change would have been treating SKYSCRAPER-01 as a binding interim event for AdvanTIG-302 and pre-committing to a Bayesian Go/No-Go gate on OS predictive probability, not on PFS read-through.

What this means for similar programs

Anti-TIGIT monotherapy or doublet strategies in PD-L1 high NSCLC should be considered contingent on a different mechanistic premise, not engineered Fc variants alone. Programs combining TIGIT blockade with CD226 agonism, or moving TIGIT into earlier-line and biomarker-enriched settings such as TIGIT-high mesothelioma or selected gastrointestinal indications, retain some optionality. The PD-L1 high first-line NSCLC pembrolizumab benchmark (mOS 29.4 months in this trial alone) is now too high for a second checkpoint to clear on the same broad biology.

Open questions

Did Arm C (tislelizumab monotherapy) separate from pembrolizumab at all, and if not, does that argue tislelizumab itself contributed nothing incremental in this design? Does the shorter DoR in the ociperlimab arm reflect acquired resistance specific to TIGIT-axis disinhibition? Is there a CD226-high responder subset that would emerge from a re-analysis of preserved biopsies? Will sponsor publication include full survival curves so that downstream meta-analysis can model TIGIT class effect across SKYSCRAPER-01, AdvanTIG-302, and domvanalimab programs?

Sources

1. ClinicalTrials.gov, NCT04746924 record and posted results, link.

2. OpenFDA FAERS, ociperlimab event counts, accessed 2026-05-19, link.

3. Cho B, et al. Tiragolumab plus atezolizumab vs placebo plus atezolizumab in PD-L1 selected NSCLC: CITYSCAPE primary and follow-up analyses. Lancet Oncol. 2022. [ ]( ( PMID: 35576957. ).

4. Rudin CM, et al. SKYSCRAPER-02: tiragolumab in extensive-stage SCLC. J Clin Oncol. 2024. [ ]( ( PMID: 37976444. ).

5. Chen X, et al. An Fc-competent anti-TIGIT blocking antibody ociperlimab (BGB-A1217) elicits strong immune responses in preclinical models. Front Immunol. 2022. [ ]( ( PMID: 35273608. ).

6. Carbone D, et al. Anti-TIGIT therapies for solid tumors: a systematic review. ESMO Open. 2023. [ ]( ( PMID: 36933320. ).

7. Open Targets Platform, TIGIT (ENSG00000181847) and NSCLC (EFO_0003060) association, accessed 2026-05-19.