Command Palette
Search for a command to run...
MAGNET: lithium in UNC13A-stratified ALS fails the confirmatory test
TRICALS halted the lithium arm of the MAGNET platform trial after a planned interim showed no survival benefit. The trial was the pre-registered confirmation of a UNC13A genotype-stratified post-hoc finding from 2017. The subgroup hypothesis did not hold.
Mechanism Risk Score
| Component | Points |
|---|---|
| Phase-weighted failure burden | 15.7 / 40 |
| Archetype severity | 8.5 / 25 |
| Temporal recency | 4.3 / 15 |
| Genetic evidence deficit | 10.3 / 15 |
| Programmatic saturation | 2.5 / 5 |
For GSK3B in Amyotrophic lateral sclerosis, UNC13A rs12608932 C/C carriers, the Mechanism Risk Score is 41/100 (yellow band). The score is a failure-burden index derived from Claidex post-mortems on this target–disease pair, not a probability of approval.
MRS 41/100 (YELLOW). 1 programs across GSK3B have been documented for GSK3B in Amyotrophic lateral sclerosis, UNC13A rs12608932 C/C carriers: 1 Phase 3, 0 Phase 2, 0 Phase 1 — of which 0 were efficacy failures, 0 safety, 1 biomarker, and 0 operational (enrollment, sponsor, or funding). The most informative failure on file is MAGNET: lithium in UNC13A-stratified ALS fails the confirmatory test. This score quantifies the documented failure burden; the Open Targets association score of 0.32 reflects weak genetic anchoring, compounding the documented failure record. The MRS is not a prediction of future trial outcomes — it is a structured summary of the empirical record, recomputed live from the Claidex claims table, and intended to flag mechanisms where any new program must explicitly resolve each prior failure mode before pursuit is justified.
This score does not predict whether the next trial will succeed. It flags how heavy the documented mechanistic failure record is before a new program is justified.
What was tried
The MAGNET platform trial (NCT06008249) ran a Phase 3, 2:1 randomised, double-blind comparison of lithium carbonate (target plasma 0.4 to 0.8 mmol/L) versus placebo in ALS patients carrying the UNC13A rs12608932 C/C genotype, TRICALS risk score between -6.0 and -2.0 [1]. Approximately one in six ALS patients carry this genotype. The primary endpoint was overall survival, defined as time to death or sustained respiratory insufficiency. Enrollment reached 88 of a target larger sample before the trial was terminated on a pre-planned interim look. The posted reason: "The planned interim analysis showed no survival benefit of taking the drug when compared with placebo."
The biological hypothesis
Lithium has at least three relevant mechanisms in motor neurons. It inhibits glycogen synthase kinase-3 beta (GSK3B), it depletes inositol via inhibition of inositol monophosphatase, and it induces autophagy. GSK3B activity is elevated in ALS post-mortem tissue, and lithium-induced autophagy clears mutant SOD1 and TDP-43 aggregates in animal models [4]. Open Targets reports a GSK3B-ALS association score of 0.316, with literature contribution 0.54 and clinical 0.49, weaker than the UNC13A-ALS association score of 0.505. The UNC13A connection is genetic and mechanistic: UNC13A encodes a presynaptic priming protein, the rs12608932 C-allele alters splicing in stressed motor neurons, and TDP-43 loss of function exposes a cryptic exon in UNC13A. The post-hoc finding by van Eijk and colleagues in 2017 [2] proposed that lithium might preferentially benefit UNC13A C/C homozygotes, with a pooled subgroup hazard ratio of 0.39 (95% CI 0.19 to 0.81) across three earlier negative trials.
What actually happened
Three prior unselected lithium ALS trials had already failed. The NEALS dose-finding trial in 2010 was halted for futility at the 11-month look (N=84) [5]. UKMND-LiCALS in 2013 enrolled 214 patients and reported a survival hazard ratio of 1.13 (95% CI 0.61 to 2.07), numerically worse than placebo [6]. The Italian dose-ranging effort was stopped early on the same futility logic. The 2017 meta-analysis identified a UNC13A subgroup signal that motivated a confirmatory protocol [3]. MAGNET was that confirmation. The pre-specified interim look showed no separation in survival between arms in 88 UNC13A C/C patients, and TRICALS terminated the lithium sub-study.
Failure mechanism, best guess
The most parsimonious read is that the original UNC13A subgroup result was a chance finding in a small post-hoc analysis, the standard fate of subgroup-discovery analyses. A pooled subgroup HR of 0.39 from 46 patients across three prior trials carries wide credible intervals once a realistic skeptical prior is applied. The Bayesian shrinkage of the 2017 estimate toward the null was always going to be severe given the prior failures in the broader population.
A second possibility is target engagement. Lithium plasma 0.4 to 0.8 mmol/L is the standard range, but central nervous system exposure relative to the GSK3B IC50 at the motor neuron compartment is uncertain. If the prior subgroup signal reflected differential brain exposure rather than UNC13A biology, the confirmatory trial used the same dose and would not separate. A third possibility is that UNC13A-driven cryptic splicing in ALS is downstream of TDP-43 mislocalisation, and that lithium-induced autophagy cannot reverse established splicing pathology. The trial was event-driven. With 88 patients and a halt at interim, the conditional power to detect even a moderate survival benefit was low, consistent with the futility stop.
How to prevent this next time
Apply Bayesian historical base-rate adjustment to subgroup-discovery findings. The pre-MAGNET prior on lithium efficacy in ALS, integrated over the three prior negative trials, places mass close to HR 1. A Beta(2, 8) prior on probability of HR < 0.8 yields a 20% chance before any subgroup data. Updating with the 2017 subgroup result lifts this only to roughly 35 to 45% under a properly skeptical model. The posterior predictive probability of MAGNET success was modest from the outset:
Power calculation discipline is the second leverage point. For an absolute survival improvement of 15% at 18 months in a UNC13A C/C population with baseline event rate ~ 45%, detecting a hazard ratio of 0.65 with 80% power and two-sided alpha 0.05 requires roughly 230 events and 350 to 400 enrolled patients given the slow accrual of a 16% genotype-restricted cohort. The interim look at 88 enrolled patients was correctly powered for futility, not for benefit detection.
Biomarker enrichment strategy should have layered. UNC13A C/C alone may not be sufficient; pairing genotype with plasma neurofilament light (NfL) progression rate, or with cryptic UNC13A splicing measured in CSF, would have produced a denser-effect enrichment. The number needed to screen for the UNC13A C/C plus high-NfL combination is high, perhaps 12 to 15 per enrolled patient, but the conditional treatment effect would be larger if the biological hypothesis were correct, making the futility look more decisive.
The single highest leverage change would have been pre-registering a biomarker-stratified two-stage adaptive design, with an early Bayesian futility gate at 25 to 30 enrolled patients and a graduated stop rule tied to the posterior probability of HR < 0.8, rather than a single fixed interim look on an underpowered cohort.
What this means for similar programs
Three falsified lithium ALS trials plus one falsified UNC13A-stratified confirmation should close this hypothesis at the dose tested. Future GSK3B-focused ALS programs should consider brain-penetrant selective inhibitors rather than lithium, with target engagement confirmed in CSF and motor neuron compartments before efficacy testing. UNC13A remains a legitimate biological target: the cryptic exon splicing mechanism described in 2022 makes UNC13A-directed antisense oligonucleotides and splicing modulators a more direct path than salt-based GSK3B inhibition. Subgroup-discovery findings in small ALS trials should be treated as hypothesis-generating only, requiring orthogonal mechanistic validation before phase 3 confirmation.
Open questions
Will TRICALS publish full MAGNET interim data with biomarker covariates so others can probe whether NfL trajectory or cryptic UNC13A splicing modulate response? Did achieved lithium plasma levels match the target range in C/C carriers, and is there pharmacogenetic interaction between UNC13A status and lithium pharmacokinetics? Should the lithium arm be re-attempted with plasma levels titrated upward, or is the dose ceiling already at the safety boundary in this population? What does this negative result imply for other ALS repurposing efforts resting on post-hoc subgroup signals?
Sources
ClinicalTrials.gov, NCT06008249 record (MAGNET platform trial, lithium sub-study), link.
van Eijk RPA, et al. Meta-analysis of pharmacogenetic interactions in amyotrophic lateral sclerosis clinical trials. Neurology. 2017. [ ]( ( PMID: 28978660. ).
van Eijk RPA, et al. Lithium carbonate in ALS patients homozygous for the C-allele at rs12608932 in UNC13A: protocol for a confirmatory trial. Trials. 2022. [ ]( ( PMID: 36471413. ).
Brown CA, Lally C, et al. UNC13A in ALS: from genetic association to therapeutic target. J Neurol Neurosurg Psychiatry. 2023. [ ]( ( PMID: 36737245. ).
Aggarwal SP, et al. Lithium carbonate in amyotrophic lateral sclerosis: lack of efficacy in a dose-finding trial. Neurology. 2010. [ ]( ( PMID: 20702794. ).
UKMND-LiCALS Study Group. Lithium lacks effect on survival in ALS: a phase IIb randomised sequential trial. J Neurol Neurosurg Psychiatry. 2012. [ ]( ( PMID: 22378918. ).
Open Targets Platform, GSK3B (ENSG00000082701) and UNC13A (ENSG00000130477) associations with ALS (MONDO_0004976), accessed 2026-05-19.
Related failure claims
Linked claims sharing target, indication, or failure mechanism.