Tiragolumab in resected NSCLC: an adjuvant trial closed by the class-wide TIGIT collapse

What was tried

Roche ran a Phase 3, randomized, double-blind study (NCT06267001, protocol GO45006) of tiragolumab plus atezolizumab versus placebo plus atezolizumab in participants with completely resected stage IIB, IIIA, or select IIIB, PD-L1-positive non-small cell lung cancer who had received adjuvant platinum-based chemotherapy (ClinicalTrials.gov). The trial enrolled 56 participants (actual) and randomized them to atezolizumab with tiragolumab or atezolizumab with placebo (ClinicalTrials.gov). Roche terminated the study with the posted reason that the decision was driven by results from the preceding tiragolumab study SKYSCRAPER-06 and by strategic priorities, and not by safety risks or quality concerns for any molecule or combination in the trial (ClinicalTrials.gov). The actual primary completion date was 2025-12-16 (ClinicalTrials.gov).

The biological hypothesis

TIGIT is an inhibitory receptor on T cells and NK cells that competes with the activating receptor CD226 for the ligands CD155 and CD112. Blocking TIGIT was designed to relieve that inhibition and, in combination with PD-L1 blockade, to restore antitumor T cell and NK cell activity beyond what PD-1 axis inhibition achieves alone (J Immunother Cancer 2020). Mechanistic work supported a real role for the axis, including evidence that anti-TIGIT improves PD-L1 blockade through effects on myeloid cells and regulatory T cells (Nature 2024). The randomized Phase 2 CITYSCAPE study in PD-L1-positive non-small cell lung cancer reported an improvement in response rate and progression-free survival for tiragolumab added to atezolizumab, which is the signal that launched the Phase 3 program (Lancet Oncol 2022). The adjuvant GO45006 study extended that bet into resected, earlier-stage disease where checkpoint blockade already has a foothold.

What actually happened

The adjuvant study never read out on its own. Its termination followed the first-line metastatic study SKYSCRAPER-06, which enrolled 524 patients with non-squamous non-small cell lung cancer and tested tiragolumab plus atezolizumab against a pembrolizumab and chemotherapy comparator (ESMO 2025). At the primary analysis the tiragolumab regimen showed reduced efficacy, with a progression-free survival hazard ratio of 1.27 and an overall survival hazard ratio of 1.33, both favoring the control arm, and no new safety signal (ESMO 2025). That result, where the experimental arm did worse than standard care, removed the rationale for the adjuvant study and for the broader program. Roche presented four failed pivotal tiragolumab studies at ESMO 2025 across lung, liver, esophageal, and head and neck cancer, and discontinued tiragolumab in July 2025 after a development effort that spanned roughly 5,000 patients (ESMO 2025).

Failure mechanism, best guess

The TIGIT axis is real, but the size and conditionality of the clinical benefit were overestimated from a single positive Phase 2. The Open Targets association between TIGIT and non-small cell lung carcinoma is 0.4128, carried by a clinical evidence score of 0.66 with no germline genetic support (Open Targets Platform v25), which means the case rested on trial-derived and literature signals rather than human genetics tying TIGIT to lung cancer risk. CITYSCAPE was a small randomized Phase 2 whose effect estimate carried a wide confidence interval, and the SKYSCRAPER program scaled that estimate into pivotal trials before the mechanism's context dependence was pinned down (Lancet Oncol 2022). The reversal in SKYSCRAPER-06, where the tiragolumab arm underperformed an active comparator, points to a benefit that was either absent or confined to a narrow setting that first-line and adjuvant designs did not capture. The best reading is an efficacy and translational failure, a class-level miss in which a redundant co-inhibitory axis did not add to PD-L1 blockade in unselected populations.

How to prevent this next time

Two quantitative levers would have slowed the scale-up. First, a Bayesian predictive probability of success carried forward from CITYSCAPE would have made the uncertainty explicit before four pivotal trials launched. The predictive probability integrates the chance of a positive pivotal readout over the posterior implied by the Phase 2 data:

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

A small Phase 2 with a wide posterior on the hazard ratio yields a predictive probability far below the confidence that a four-trial, 5,000-patient commitment implies, which argues for one confirmatory trial with an interim futility analysis before parallel expansion. Second, a competitive landscape red-team analysis would have stress-tested the comparator choice, because SKYSCRAPER-06 was scored against a pembrolizumab and chemotherapy standard that had itself improved, raising the bar the combination had to clear. Pairing that with a historical base-rate adjustment, where second-checkpoint combinations added to PD-1 axis blockade have a low confirmation rate in unselected lung cancer, would have set a sober prior. The single highest leverage change would have been gating pivotal expansion on a single confirmatory trial with a prespecified Bayesian futility interim, rather than launching four pivotal studies in parallel on a Phase 2 effect estimate with a wide confidence interval.

What this means for similar programs

The anti-TIGIT class is the clearest recent case of a co-inhibitory receptor that validated in mechanism and small Phase 2 yet failed at pivotal scale. The lesson applies to other CD226-axis and second-wave checkpoint programs, including LAG-3 outside its approved setting and TIM-3, where the same redundancy and comparator pressures hold. With this termination the Claidex failure graph logs a second Phase 3 efficacy failure for TIGIT alongside the earlier ociperlimab AdvanTIG-302 entry, and the Mechanism Risk Score is 41 of 100 (yellow), with a phase burden of 15.7 and an archetype severity of 9.8 (Claidex MRS). The score reflects a per-indication snapshot rather than a cumulative sum across failures, so the second miss reinforces the record without moving the band. A new TIGIT program now has to explain why its setting or biomarker differs from the ones that failed.

Open questions

Is there a CD155-high or CD226-intact subgroup in which TIGIT blockade adds measurable benefit, and can it be identified before a pivotal trial. Did the choice of an improving active comparator, rather than the mechanism itself, drive the SKYSCRAPER-06 reversal. Would an adjuvant setting with minimal residual disease selection have changed the calculus, a question GO45006 was built to ask but closed before answering. The discontinuation ends the tiragolumab record before these are resolved.

Sources

1. • ClinicalTrials.gov, NCT06267001 (terminated, n=56, primary completion 2025-12-16, reason cites SKYSCRAPER-06 and strategic priorities): https://clinicaltrials.gov/study/NCT06267001 - SKYSCRAPER-06 results (524 patients, PFS HR 1.27 and OS HR 1.33 favoring control, four failed pivotal studies, discontinuation July 2025), ESMO 2025 coverage: https://www.onclive.com/view/tiragolumab-plus-atezolizumab-chemo-misses-survival-end-points-in-metastatic-nonsquamous-nsclc - Roche scraps TIGIT after roughly 5,000 patients, ApexOnco/Oncology Pipeline 2025: https://www.oncologypipeline.com/apexonco/5000-patients-later-roche-scraps-its-tigit - CITYSCAPE: tiragolumab plus atezolizumab in PD-L1-selected NSCLC (Phase 2), Lancet Oncol 2022.- Anti-TIGIT antibody improves PD-L1 blockade through myeloid and Treg cells, Nature 2024.- TIGIT in cancer immunotherapy, J Immunother Cancer 2020.- SKYSCRAPER-02: tiragolumab with atezolizumab and chemotherapy in small-cell lung cancer, J Clin Oncol 2024.- Open Targets Platform v25, TIGIT (ENSG00000181847) to non-small cell lung carcinoma (EFO_0003060), overall association 0.4128: https://platform.opentargets.org/target/ENSG00000181847.