KER-012 in pulmonary arterial hypertension: an activin ligand trap stopped for pericardial effusions

What was tried

Keros Therapeutics ran TROPOS (NCT05975905), a Phase 2, randomized, quadruple-blind, placebo-controlled study of KER-012 added to background pulmonary arterial hypertension therapy in adults with WHO Group 1 disease. KER-012 is a modified activin receptor type IIB ligand trap dosed subcutaneously every four weeks. The trial randomized 113 participants across three active doses and a placebo arm, with a 24-week treatment period feeding the primary endpoint of change in pulmonary vascular resistance, followed by a planned 72-week extension (NCT05975905). Enrollment ran from October 2023 to a primary completion date of March 2025. Keros then terminated the study. The registry reason reads: "The study was voluntarily terminated early, following a safety review and observation of pericardial effusions" (NCT05975905).

The biological hypothesis

Pulmonary arterial hypertension reflects a shift in transforming growth factor beta superfamily signaling. Signaling through BMPR2 falls, and activin and growth differentiation factor signaling through activin receptor type II rises, which drives pulmonary vascular remodeling (Thenappan 2018, Cuthbertson 2023). Sotatercept, an activin receptor type IIA Fc ligand trap, tested that model directly. In the PULSAR Phase 2 study and the STELLAR Phase 3 study it lowered pulmonary vascular resistance and improved six-minute walk distance, and it reached regulatory approval (Humbert 2021, Hoeper 2023, Souza 2023). KER-012 used the same rebalancing idea on an activin receptor type IIB backbone, built to sequester activin A and activin B. The receptor it is derived from is only weakly anchored to the disease by human genetics. The Open Targets overall association score for ACVR2B and pulmonary arterial hypertension is 0.0055, because the strongest genetic signal in this disease sits at BMPR2, not at the activin receptors.

What actually happened

KER-012 engaged its target. Mean change in pulmonary vascular resistance at week 24 was -15.3 dyn.s/cm5 at Dose A, -73.0 at Dose B, and -81.0 at Dose C, against -47.8 in the placebo arm (NCT05975905 posted results). The two higher doses moved the primary endpoint, so this was not a target engagement problem. The placebo change was large, and the lowest dose looked worse than placebo, which would have made the efficacy read noisy even without an early stop. Six-minute walk distance shifted only a few minutes in either direction across arms and showed no consistent dose pattern. The trial did not end on efficacy. It ended on safety. Serious adverse events concentrated at the top dose: 9 of 25 participants at Dose C, or 36.0 percent with a 95 percent Wilson interval of 20.2 to 55.5, against 5 of 39 on placebo, or 12.8 percent with an interval of 5.6 to 26.7 (NCT05975905 posted results). One death occurred in each of the two higher-dose arms and one in placebo. Pericardial effusion was the specific observation that prompted the safety review and the decision to stop.

Failure mechanism, best guess

The pattern points to an on-target liability of the ligand trap rather than a chemistry-specific accident. Pericardial and serosal fluid handling is sensitive to transforming growth factor beta superfamily tone, and broad sequestration of activin and related ligands can perturb it. The activin receptor type IIB backbone binds a wider ligand set than the activin receptor type IIA scaffold used by sotatercept, including myostatin and several growth differentiation factors, so KER-012 likely cast a broader net across tissues. Target engagement was clear from the pulmonary vascular resistance drop, which argues against an inert molecule and for a real pharmacologic effect that reached the pericardium. This is a safety signal failure. The efficacy signal was present, but the therapeutic window at the doses that moved hemodynamics was too narrow to continue.

How to prevent this next time

Two quantitative tools would have sharpened the decision earlier. First, a Bayesian safety-stopping rule on serious adverse events, run continuously rather than at fixed reviews. The observed Dose C rate of 9 of 25 gives a Beta(1,1)-updated posterior mean near 37 percent with a 95 percent Wilson interval of 20.2 to 55.5, already separated from the placebo interval, which a pre-specified boundary would have flagged before the full cohort completed. The general posterior form is

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

Second, a competitive landscape red-team analysis. Sotatercept had already de-risked the activin thesis on an activin receptor type IIA scaffold with a defined safety profile. A structured comparison of ligand selectivity between the type IIA and type IIB backbones would have priced in the wider ligand net of KER-012 as a serosal and fluid-retention risk, and set a tighter starting dose band. A number-needed-to-harm framing on pericardial effusion, anchored to the placebo rate, would have made the top-dose risk explicit before expansion.

The single highest leverage change would have been a pre-specified Bayesian serious-adverse-event boundary that compared each active dose against placebo in real time, so Dose C stopped enrolling at the first credible separation rather than at a scheduled review.

What this means for similar programs

The activin and bone morphogenetic protein axis in pulmonary arterial hypertension is now crowded and partly validated. Sotatercept set the efficacy bar and the safety expectations for a ligand trap in this disease (Hoeper 2023). Follow-on traps do not compete on whether the axis matters. They compete on ligand selectivity and therapeutic window. A backbone that broadens the captured ligand set may buy potency on hemodynamics at the cost of serosal and hematologic effects. Programs that copy the mechanism without copying the receptor selectivity inherit a different, and in this case worse, safety profile.

Open questions

Was the pericardial effusion dose-dependent in a way that leaves room for a lower-dose window, given that Dose C carried most of the serious event burden. Did the broader ligand profile of the type IIB backbone, rather than activin A and B capture specifically, drive the effusions. Would selective activin A neutralization, closer to the type IIA profile, separate the hemodynamic benefit from the serosal risk. The posted results give arm-level rates but not the per-participant timing that would settle whether earlier stopping was feasible.

Sources

1. • ClinicalTrials.gov, NCT05975905 (TROPOS), protocol and posted results: enrollment 113, PVR changes by arm, serious adverse event counts, why-stopped text. - Hoeper MM et al. Phase 3 Trial of Sotatercept for Treatment of Pulmonary Arterial Hypertension. N Engl J Med 2023.- Humbert M et al. Sotatercept for the Treatment of Pulmonary Arterial Hypertension (PULSAR). N Engl J Med 2021.- Souza R et al. Effects of sotatercept on haemodynamics and right heart function (STELLAR). Eur Respir J 2023.- Thenappan T et al. Pulmonary arterial hypertension: pathogenesis and clinical management. BMJ 2018.- Cuthbertson I et al. Cardiovascular signaling in pulmonary hypertension. Circ Res 2023.- Open Targets Platform: ACVR2B and pulmonary arterial hypertension overall association 0.0055. - Wilson score intervals and the Beta(1,1) posterior mean computed from posted-result counts. Available from: https://clinicaltrials.gov/study/NCT05975905.