When Phase 2b Wins and Phase 3 Fails: The Biomarker Stratification Crisis Hidden Inside Astegolimab’s COPD Data

Three respiratory biologics have now entered Phase 3 COPD trials with strong Phase 2b signals, and three times the confirmatory data has come back muddier than the earlier read. That pattern has a name: the Biomarker Stratification Trap — and Genentech’s astegolimab just walked straight into it.

The Lancet publication of ALIENTO and ARNASA this week is not simply a story about a drug that underperformed. Read the two trial results side by side and you see something structurally more unsettling: a Phase 2b trial enrolling 1,301 patients that hit its primary endpoint, paired with a Phase 3 trial enrolling 1,375 patients — a nearly identical population, the same drug, the same indication — that did not reach statistical significance on the annualized exacerbation rate at 52 weeks. The drug did not change. The disease did not change. What changed was almost certainly the patient mix, and that exposes a fault line running through how the field currently thinks about COPD trial design.

The regulatory pressure to drop eosinophil thresholds was understandable on its face.

The Signal That Moved, Then Disappeared

Astegolimab targets the ST2/IL-33 pathway — a mechanism that sits upstream of eosinophilic inflammation but does not map cleanly onto blood eosinophil count. Genentech’s press release from July 20, 2025 confirmed the ALIENTO Phase 2b result: the every-two-weeks dosing arm was associated with a lower annual rate of exacerbations versus placebo in patients with a history of frequent exacerbations, regardless of baseline blood eosinophil count. That “regardless of eosinophil count” framing was the design choice that defined both trials — and, in hindsight, may have defined their divergent fates.

ALIENTO enrolled current or former smokers aged 40 to 90 with moderate to very severe COPD and a history of frequent exacerbations. The trial ran as a randomized, double-blind, placebo-controlled, multi-center study — a clean design on paper. The Phase 2b signal was robust enough to advance directly into Phase 3 ARNASA, which enrolled 1,375 patients under a structurally comparable protocol. But ARNASA did not meet its primary endpoint. The annualized exacerbation rate reduction did not achieve statistical significance. Topline results were announced on July 20, 2025 — the same day as ALIENTO’s positive readout, a pairing that the Roche communications team managed deftly but that the clinical operations community should read more carefully.

What the Lancet data now surfaces — and what press releases predictably obscured — is that the absence of an eosinophil enrichment criterion likely introduced sufficient heterogeneity across the two trials to swamp the treatment effect in the larger confirmatory study. The ST2/IL-33 pathway drives both eosinophilic and non-eosinophilic inflammation, which is precisely why the “eosinophil-agnostic” design was intellectually appealing. But appealing biology and clean trial execution are different things.

The Precedent the FDA Already Set — and Then Complicated

This outcome does not arrive in a regulatory vacuum. The FDA’s Biomarker Qualification Program has formally recognized blood eosinophil count as a predictive biomarker intended to enrich clinical trial populations of high-risk COPD patients who have a history of exacerbations in the previous year. The agency’s position is that eosinophil count predicts responsiveness to pharmacological intervention — not just to eosinophil-directed therapies, but more broadly. That qualification existed before ALIENTO was designed. The decision to run an eosinophil-agnostic enrollment strategy was a deliberate wager against that framework.

The precedent from mepolizumab (GSK’s Nucala) is instructive here. In 2018, an FDA advisory committee declined to recommend Nucala for COPD, citing mixed evidence from earlier trials. The pivotal METREX trial, which enrolled 836 participants, was initially ruled insufficient to support general COPD treatment approval. It was only after GSK demonstrated efficacy specifically in the eosinophil-high subgroup — and pursued a label restricted to that population — that a regulatory path materialized. The lesson from mepolizumab was not subtle: in COPD biologics, eosinophil stratification is not optional enrichment, it is the mechanism by which the signal separates from the noise. Genentech appears to have decided the ST2/IL-33 biology was different enough to escape that logic. ARNASA’s result suggests it was not.

The deeper regulatory complication is that the FDA now faces a dataset that genuinely supports the ST2/IL-33 pathway as a therapeutic target — ALIENTO’s Phase 2b result is real and enrolled over 1,300 patients — while lacking a Phase 3 confirmation. That is not a trivial problem for a potential NDA. The agency has approved drugs under accelerated pathways with smaller datasets, but COPD exacerbation reduction requires robust endpoint achievement in at least one well-controlled confirmatory trial. Neither ALIENTO’s Phase 2b status nor ARNASA’s failed Phase 3 maps cleanly onto that requirement.

What This Forces Sponsors and Trial Designers to Confront

The conventional wisdom in COPD drug development is that broader enrollment criteria produce more generalizable results and larger commercial labels. Biomarker restriction feels like leaving patients on the table. That assumption deserves direct interrogation.

Consider what ARNASA’s failure actually cost: 1,375 patients, a multi-year global trial, and a regulatory package that now cannot support an unqualified NDA submission. If the trial had been designed with an eosinophil enrichment stratum — even a pre-specified secondary analysis population — the dataset would be far more actionable today. The FDA qualified eosinophil count as a biomarker for exactly this reason: not to restrict who can access therapies post-approval, but to ensure that confirmatory trials can isolate the signal cleanly enough to reach regulatory thresholds. Broad enrollment is a commercial strategy. Regulatory success requires a different kind of discipline.

For CROs and trial operations teams, the ALIENTO/ARNASA divergence points to a specific protocol design failure mode that is becoming more common as sponsors pursue novel pathway targets in heterogeneous populations. When a Phase 2b uses “regardless of biomarker” language as a hypothesis — not a convenience — the Phase 3 must either pre-specify a biomarker-defined primary population or commit to sample sizes large enough to detect treatment effects across subpopulations independently. ARNASA’s 1,375-patient enrollment was not substantially larger than ALIENTO’s 1,301. If the treatment effect exists predominantly in a biomarker-defined subgroup, and that subgroup represents perhaps 40 to 60 percent of enrolled patients, the statistical power calculation needs to reflect that reality from the protocol inception — not from a post-hoc subgroup analysis presented after the primary endpoint fails.

Technology vendors building eCOA and RTSM platforms for respiratory trials face a more immediate implication. Adaptive enrichment designs — where interim analyses can shift enrollment criteria toward responding subpopulations — require infrastructure that most current IRT systems can support mechanically, but that most sponsors are not yet deploying at protocol inception. The astegolimab experience gives the adaptive enrichment design a specific, named clinical argument: had ARNASA included a pre-specified adaptive enrollment criterion tied to baseline eosinophil count, the trial might have concentrated statistical power in the subpopulation most likely to show an ST2/IL-33 treatment effect without sacrificing the broader enrollment intent.

The Lancet data also carries an implication for the COPD biologics competitive landscape that has not yet been priced into the broader market narrative. Dupilumab (Sanofi/Regeneron), which secured FDA approval for COPD with an eosinophil threshold of 300 cells per microliter in 2024, made exactly the opposite design decision: it embraced biomarker restriction and built its regulatory package around it. The triple therapy inhaler market, currently valued in the multi-billion dollar range annually, represents the commercial baseline that any COPD biologic must displace. Astegolimab’s path to that market now runs through a question Genentech has not yet publicly answered: does the company run a third trial with biomarker stratification built into the primary endpoint, or does it pursue a narrower label application based on the ALIENTO Phase 2b data alone?

Twelve months from now, watch for one of two things to happen first. Either Genentech announces a redesigned Phase 3 with pre-specified eosinophil enrichment strata — which would validate the biomarker stratification framework the FDA has been quietly reinforcing since 2018 — or the ARNASA dataset becomes a case study in FDA guidance on adaptive enrichment designs, cited in a future draft guidance on respiratory biologics that finally names the operational standard the ALIENTO/ARNASA divergence has been pointing toward all along. The ST2/IL-33 pathway works. The question was never the biology. The question was always whether the trial was designed to prove it.

References

1. The Lancet — “Safety and efficacy of astegolimab for COPD with frequent exacerbations regardless of baseline blood eosinophil counts (ALIENTO and ARNASA)”
2. Genentech — “Genentech Provides Update on Astegolimab Phase 2b/3 COPD Trial Results” (July 20, 2025)
3. FDA Biomarker Qualification Program — Blood Eosinophil Count as Predictive Biomarker for COPD Trial Enrichment
4. MedPage Today — “FDA AdCom Declines to Recommend Mepolizumab for COPD; GSK Eosinophil Subgroup Pathway”
5. PMC — “Astegolimab Phase 1 Clinical Studies: Safety, Tolerability, and Pharmacokinetics”
6. Patient Care Online — “Roche’s Investigational Anti-IL-33/ST2 Antibody for COPD Misses Phase 3 Primary Endpoint”