VRG50635 – Lessons Learned in ALS

In a recent blog we took issue with a Nature article about ‘Lessons learned’ after a TIGIT inhibitor failed in late Phase 3 development.  We felt that the article provided few insights, mainly platitudes.

This week we saw a different kind of post-mortem.  It was about VRG50635, a PIKfyve inhibitor for ALS which failed in a Phase 1b study earlier this year.  In the world of ALS, few things work.  Let me rephrase: nothing really works.  Even the approved drugs like tofersen, edaravone and riluzole provide very little to no benefit. 

When dealing with a disease like ALS, we can only applaud those who try to tackle it scientifically and those who invest money on a putative new target.  It takes courage to engage in ALS exploratory research.  Failure is the norm at this stage in drug development, and even more so for ALS. [1]

That said, let’s have a look at VRG50635, a compound from Verge Genomics that came with a good rationale: the science for targeting PIKfyve was compelling, the compound itself was designed for CNS penetration.  The next steps were meant to prove target engagement and proof of clinical efficacy, i.e., clinical improvement of signs & symptoms in ALS patients.

Sadly, when the company finished the Phase 1b PoC study in 54 ALS patients, there was not a hint of clinical efficacy.  Notwithstanding, VRG50635 had a ‘biologic effect’ on at least 2 measures considered relevant, NfL and GFAS.  Totally unanticipated and contrary to expectations, treatment with VRG50635 produced a sustained increase in plasma NfL (p<0.001) and GFAP levels.  What are we to make of all this?

We would have concluded that NfL is not a particularly useful biomarker for ALS disease progression, and that PIKfyve is not central to ALS pathogenesis.  Full stop.

Verge’s former CEO, a neuroscientist and entrepreneur, begs to differ.   She published an explanation why VRG50635 was doomed.[2]  In her words, they were ‘in the right neighborhood’ but ‘not the right house’.  While we can agree that they ‘missed the house’ we stand unconvinced about  ‘the neighborhood’. 

Admittedly, we are at a disadvantage as full study data have not yet been published; all we have are Top Line results from a short communication.[3]  

Here are Verge’s learnings as proffered (with our comments in BLUE):

1. NfL is one of the field’s most relied-upon efficacy biomarkers
   Accumulating evidence indicates that NfL may be an epiphenomenon.  It was an unhelpful and misleading clinical  surrogate in the Amplyx and several other ALS trials.1  FDA never elevated it to biomarker status, and Ad Board members reviewing ALX0035 were divided in their opinion and quite unsure what to make of it.  Everyone agrees that real biomarkers should be tied to clinical outcome.  This data does not (yet) exist for NfL or for GPAP.
2. Severe / older / more advanced patients don’t respond with plasma NfL changes like early ALS cases
   This statement falls into the bucket of ‘hypotheses’.  It is impossible to conclude this given the small size cohorts in Verge Phase 1b study.  A mere conjecture at this time – to be confirmed!
3. VRG50538 was conceived by an AI platform predicting PIKfyve as a new target specific for ALS biology
   Using AI for target selection sounds like a great idea but given the results, should we still trust an AI-driven target selection algorithm?  There are a few other companies pursuing PIKfyve for ALS.  Apilimod clinical data are sparse and hence uninterpretable.[1]  Why does Verge Labs (the successor of Verge Genomics) believe that an ‘improved’ version of their AI approach has a better chance of success?  How many times does one have to fail before we acknowledge that AI-based targeting may not a panacea?
4. Study failure was a result of patient heterogeneity which needs to be controlled better
   Hmm, that sounds like good advice, but very generic if not outright trivial.  Why didn’t AI-modelling figure that out upfront?  How does Verge Lab plan to address this learning in the next go-around? 
5. Study points out difficulty of bridging dead tissue research to living people disease
   Is this is a new learning, really?
6. Hard to predict how biomarkers relate to intervention and clinical features
   Well, yes, right, of course, good point…

As we have seen, PoC (or lack thereof) can be established early or late for ALS drugs in development.  In the case of VRG50635, a Phase 1b study was sufficient to allow early decision-making.  Good so!  Well done! 

Let’s keep an eye on Verge Labs.

ABBREVIATIONS
NfL          neurofilament light
GFAP      glial filament alkaline phosphatase
PoC         proof-of-concept
PIKfyve  phosphoinositide kinase with FYVE zinc finger domain

REFERENCES
[1] Huang C.  Innovative therapies under clinical development for ALS treatment. Part 2: biologics and natural Products.  Expert Opinion on Investigational Drugs 2026, DOI: 10.1080/13543784.2026.2729512
[2] Zhang A.  What We Learned From The Trial of Our First AI-Discovered Drug.  In the right neighborhood, but not yet the right house.  BioPharma May 27, 2026
[3] Cadavid D.  Top Line Results of the Phase 1b, Open-label, Multiple Ascending Dose Proof of Concept Study of the PIKfyve Inhibitor VRG50635 in Familial and Sporadic Amyotrophic Lateral Sclerosis.  (P11-7.013) | Neurology 2026
[4] Babu S.  Apilimod dimesylate in C9orf72 amyotrophic lateral sclerosis: a randomized phase 2a clinical trial.  Brain 147; 2998, 2024