Envudeucitinib in SLE: LUMUS primary failure and the interferon subgroup

TYK2 and interferon biology

Envudeucitinib inhibits TYK2, with development in SLE aimed at interferon-driven disease. On September 1, Alumis reported that LUMUS failed its primary and secondary endpoints overall, while describing favorable prespecified IFNGS-high findings. This is a failed overall efficacy test with a follow-up hypothesis, not a positive broad-population trial. [1] [2]

Analytical interpretation: a baseline signature can be prognostic, predictive, or both. Higher placebo response in signature-low patients does not by itself prove that the marker predicts treatment benefit. The relevant evidence is a treatment-by-marker interaction, supported by assay reproducibility and prospective replication using an unchanged cutoff.

Pharmacodynamics and dose selection

The company reported dose-dependent suppression of the interferon pathway. Analytical interpretation: pharmacodynamic suppression verifies pathway inhibition but cannot establish the optimal clinical dose after a negative primary result. Greater inhibition could improve the relevant biology while simultaneously exposing patients to unnecessary immunomodulation. [1] [2]

Dose selection for a subsequent trial should jointly consider exposure, response, tolerability and the biomarker subgroup. Choosing the arm with the best result on one endpoint after inspecting many outcomes can overfit this dataset. A coherent dose–response pattern across clinically related outcomes is more persuasive than isolated nominal significance.

Randomized LUMUS results

LUMUS enrolled 408 autoantibody-positive patients receiving background care, with four equal planned allocation groups and week-48 BICLA primary assessment. The presentation’s arm denominators total 405; this difference is not assigned to withdrawal without a disposition table. Reported response estimates use imputation rather than simple observed responder fractions. [2]

Analytical interpretation: every BICLA confidence interval for the treatment difference includes zero. Thus, the data allow clinically different effects in both directions and do not support a precise benefit estimate. Nominal secondary p values cannot bypass failure of the primary testing sequence. Nor should an imputed percentage be multiplied by N to manufacture an integer responder count.

IFNGS-high and the limits of subgroup rescue

In IFNGS-high, BICLA was 28.6% on placebo (N=63), 52.6% at 40 mg BID (N=60), 49.5% at 20 mg BID (N=62), and 40.7% at 20 mg QD (N=64). These findings motivate a biomarker-enriched trial, but the retrieved presentation does not supply a definitive interaction test. [1] [2] [3]

Analytical interpretation: prespecification reduces one source of selection bias, but does not grant an independent error budget to every subgroup. A credible next study would lock the assay, threshold, steroid taper and primary endpoint before enrollment. Enrollment enrichment also changes the population to which any positive result could apply.

Comparisons with anifrolumab or deucravacitinib across separate trials are not evidence of relative efficacy. Background medication, organ involvement, timing, geography and signature assays differ. Similar response percentages can conceal different placebo behavior and very different absolute benefit.

Analytical judgment: BICLA requires a pattern of improvement without unacceptable worsening in other domains; it is not identical to SRI-4. Different sensitivities to baseline organ activity can produce divergence between the measures without making either fraudulent or irrelevant. The correct interpretation is endpoint-specific: the prespecified primary question failed, while other measures may support hypotheses that need new testing.

An enriched phase 3 study should also prospectively address the negative or weak pattern in signature-low patients rather than simply remove them from the narrative. Exclusion can be scientifically justified when a marker predicts benefit, but it changes the intended indication. Regulators will need a usable assay and a clinically defensible cutoff, not only a retrospective separation in one dataset.

The subgroup hypothesis is stronger if the clinical interaction tracks exposure and pathway suppression, but these correlations must be tested without using the outcome to redefine the biomarker. Holding definitions fixed is what turns a plausible explanation of failure into a prospectively falsifiable development plan.

Safety and next clinical decisions

Alumis reported no new safety signals and plans regulatory discussions about an enriched phase 3 approach. Numerical arm-level adverse-event and discontinuation tables were not provided in the brief release. Absence of a new signal is not equivalent to established long-term safety in the intended enriched population. [1] [3] [4]

Analytical judgment: the near-term clinical decision is whether a reproducible treatment interaction and acceptable safety justify another randomized program. Psoriasis progress cannot retrospectively validate SLE efficacy. Any future protocol should make steroid exposure, rescue therapy, infections and missing visits auditable so that clinical response is not confounded with differences in background care.

The prospective study must preserve a clear distinction between a biomarker-selected indication and an all-comer indication; success in the former would not retroactively reverse the latter’s failed test.

Sources / References

1. Alumis · LUMUS topline release · September 1, 2026
2. Alumis · LUMUS results presentation · September 1, 2026
3. Alumis · September 1, 2026 Form 8-K
4. Alumis · Q2 2026 report