Great. Thanks so much, everybody, for joining. My name is Julian Pino. I'm an associate on Paul Matteis's team with Stifel. And I'm joined here by Amit Etkin, Founder and CEO of Alto Neuroscience. Thanks for joining us, Amit.
My pleasure.
Great. I guess we can just go ahead and dive into it. Maybe first, you can just provide an overview of Alto, including the formation of the company, any information you want to provide on your precision psychiatry approach, and updates on your pipeline. That would be great.
Yep. As kind of broad background, right, the core idea behind Alto, and the core idea behind my work as a professor at Stanford before that, where I was on faculty and had a lab for a decade, is this idea that the way we've developed drugs historically in psychiatry has not worked for two fundamental reasons. One is we've not really known how different drugs target different brain processes and been able to find the right indications in patients because of that. Also, when we do our trials in all comer populations, truly the target population, the responsive population, is the minority. We're trying to treat the whole population, and you end up often getting null or very small effects.
What we have done at Alto is systematically put together a biomarker platform, ways to characterize patients, and doing it at scale, the kinds of things you could do in routine clinical care with some basic tools around things like EEG, electroencephalography, to understand brain activity or behavioral measures that index different aspects of brain function together with drugs that target different mechanisms in the brain for particular subpopulations. What that has meant in the process is that the way we go about developing drugs is understanding how to dose a drug based on a biomarker, finding a biomarker that enriches for bigger clinical effect, prospectively replicate that so that we know it's solid science, and then target the right larger scale efficacy study for the right populations.
We have now been doing that across four different assets in different populations: in depression, bipolar disorder, and schizophrenia, and have two in phase IIb trials that we will talk about, as well as two in earlier stage trials focused more on the biomarker effects of the drugs.
Excellent. Yeah, really appreciate that overview. There's definitely a lot of excitement about your approach, certainly from folks in the medical community. I guess just for one of those programs that's in phase IIb, ALTO-100 , you had an update last year that didn't quite go your way. I'm just curious if you can opine a little bit on what you believe led to the failure. How much in part was it execution versus driven by the drug not being active? What are your plans currently for that program?
Yep. ALTO-100 is a novel compound targeting a protein that's never been targeted before in therapeutic studies, but it has been of great interest with respect to understanding neuroplasticity. It was a drug that was identified based on its ability to enhance neurogenesis and neuroplasticity. The core premise here is that those patients with impairments, especially in neuroplasticity within the hippocampus, which is important for cognition and mood, might be those who benefit the most here. Part of the drug's action then is through modulation of plasticity pathways like BDNF, which has been implicated for a long time.
What we did in phase IIa, which is a study of over 200 patients, is found that those patients characterized by impairments in hippocampal plasticity in the form of impaired memory did better with a drug with respect to antidepressant effect, replicated that prospectively, and took that patient selection approach into phase IIb. We studied two populations, monotherapy and adjunctive use, under one kind of umbrella, and did not find statistical significance in that phase IIb overall. Actually, when you take those two populations apart, the monotherapy group did not distinguish between drug and placebo, but there was evidence of a clear effect of the sort we were expecting in the adjunctive group. To your question, the issue is, why was that? Why did we see an effect in the adjunctive group and not in the monotherapy group?
Is there something we did not understand about the biology, or is it an issue with execution? In this case, really a site-level execution issue that I think pervades the entire field. I think we have seen it again and again across readouts over the past few months. That was really an issue in this case of noncompliance, where our adjunctive patients were fully compliant with the drug, which backs up the evidence of a clinical effect there, whereas we saw only 56% compliance in the monotherapy arm. That itself was really heavily clustered at certain sites. There were certain professional research sites that are used in a lot of people's studies that brought in a fair number of patients and a high rate of noncompliance. That is a site-level execution issue. We have now taken that to heart.
We do not think that that is a molecule or a biomarker issue, in part because when you look at people who were compliant, we see a drug signal. If you look at the people who are on drug and compliant, you see a biomarker signal that all replicates what we expected before. That is why we have continued to develop the compound in bipolar depression. It also, I should say, is an adjunctive program, but really further strengthening even beyond that, our ability to control and quantify these site-level risk factors. That also led to our learnings that we carried into our ALTO-300 program now and really throughout to find and root out all of these systematic risks across the field, in part through our biomarker platform, in part through our ability to have a much closer view of what is going on in these sites in running trials ourselves.
We don't use CROs. We have an internal clinical operations team that works directly with research sites.
Excellent. No, I think you summed that up perfectly. This isn't a problem that's been only experienced by you guys. Like you said, it's certainly a challenge that's pervasive in the field today. I guess before we go into some of the details with ALTO-300 , can you just very quickly just verify or sort of remind folks, how did you exactly verify compliance in ALTO-100 ? What would you just say to folks that may be more skeptical of this precision biomarker approach now and what underscores your confidence?
Yeah. A number of factors, right? Just on the verifying compliance, it's literally just measuring drug in the blood as a direct measure of compliance. I should mention that the characteristic of these kind of professional patient populations is that they do, on IP accountability, on pill counting, show you that they are taking the pills, but clearly they're not. You can see we've done blood draws at the end of double-blind and at the end of open-label in that trial. The people who are negative at the end of double-blind are consistently negative at the end of open-label despite being supposedly on drug and popping out pills the whole time. That's clearly deceptive behavior. I think that that's a site-level risk in letting these folks in and to some degree a patient-level risk. The things that back up the precision thesis are multiple.
I mean, we have found and replicated a mechanistically related ALTO-100 memory biomarker. We'll talk more about 300 in a second, but that's an EEG biomarker we developed through a machine learning approach that we had validated through years of work that we have prospectively replicated. We've now shown is directly related to the drug's mechanism of action. This is a marker of neural variability. Our hypothesis was around it relating to low dopamine, dopamine important for determining signal-to-noise in the brain, and dopamine being a downstream effector of ALTO-300 's response. We've shown that not only by manipulating in rodents, one of the targets for ALTO-300 , the 5-HT2C receptor, which has effects on dopamine, we can create that biomarker.
We have posters coming up at Biological Psychiatry in April showing that even in humans, if you deplete humans of dopamine, which you can do through a dietary intervention, you also create a biomarker positive-like phenotype. Even though this is all data-driven, we now have been able to replicate it and tie it to mechanism. Moreover, we actually have posters as well at Biological showing that we can even predict placebo response. We have developed an EEG machine learning model and prospectively replicated it twice, showing that we can predict placebo response. I think that the ability to understand biology, to use it in the context of clinical development and ultimately clinical care, and understand things like drug signal and placebo signal using objective biological tools is there.
It's a matter of making sure that that connects with running a clinical trial in the kind of real-world clinical trial systems that we have to run trials in right now. That's really where our focus is. The data continually support the importance of precision in stratifying populations, the ability to replicate these findings, and the value of these findings in leading to entirely new insights mechanistically.
Excellent. No, definitely makes sense. Yeah, we look forward to those updates from you and your team shortly. I guess we can just go ahead and dive into ALTO-300 . This is an approved drug in the EU, right? This is agomelatine. Can you just walk us through the history with this drug again in terms of who developed it and again, why did it not get developed in the U.S.?
Yeah. So developed originally by Servier, got it approved in Europe and in Australia. Novartis had licensed it for development in the United States. They actually had positive phase III studies, but unfortunately, they'd split the doses. It was developed at 25 and 50. Turns out those two doses do not have any difference in clinical efficacy. Certainly, the view back then in the FDA was that even though they had a positive 25 milligram and a positive 50 milligram phase III, they did not have two at the same dose. That created the kinds of problems that would require you to do another phase III. They at that point were really pulling up roots with respect to CNS drug development and running out of composition of matter timeline. Clearly, there is efficacy signal there.
Moreover, that history of some positive trials, some negative trials is the classic all-comer pattern. I mean, we talk about Prozac having failed half of its trials, and yet it's a canonical drug in the space. That's what happens when you take unselected populations. There are clear efficacy signals there. Just then due to the kind of accidents of history and timing, that unfortunate outcome in terms of it not coming to market in the United States . is our opportunity here with a new lens to bring this drug forward in the U.S.
No, definitely makes sense. Yeah, I think efficacy for an all-comers population, like you talked about, is perhaps a little bit less controversial and more kind of about execution. Certainly, with your approach, seems like it would be able to boost POS. I guess there is a little bit more to discuss on safety. This is a drug that has had a sort of liver signal, I guess, at certain doses. Can you just remind folks of where that's been seen and where are you dosing relative to what's been investigated in the past? If you've seen anything on a blinded basis, if you could opine on that as well.
Yeah. I think this is actually more a sort of a tempest in a teapot than anything else. There's not really a liver issue. There's a liver monitoring requirement in the EU and in Australia. That's because at low rates, only the 50 milligram dose, but not the 25, leads to a slight elevation in LFTs. Remember, the drug's been on the market now for over 15 years. We've never seen this lead to liver failure. These are all reversible LFT elevations, which, by the way, sometimes reverse when you're still on drug. These are really mild effects in small populations, but have required in the label monitoring of LFTs. Arguably, this is something that Servier had done almost to themselves because there was never any screening for LFT elevations in the early days of the first set of trials.
What Novartis did in screening for LFTs and allowing, like we do, up to 2x upper limit of normal for entry is that they showed that at 25 milligrams, there's no difference in LFT elevation, never mind any long-term effects, which there aren't, right? Even transient LFT elevations, there's no effect at 25 milligrams. There's the same efficacy clinically as an antidepressant at 25 and 50. There's a slight elevation in rates at 50. That's why we chose 25, is that our goal here is to see if we can just eliminate the need to monitor at all, given that we know from 15 years of clinical experience, there's not any long-term risk anyway. It'd be nice to take that off the label.
You can see in the reviewers' comments from the Australian review of the drug that the reviewers felt that at 25, there is no need to monitor for LFTs. We, to your point, have also put as a stopping rule an elevation in LFTs in our phase IIb. Nobody has triggered that stopping rule. We saw no elevation of LFTs in our phase IIa and using the same standard cutoff that others have used, which is three times upper limit of normal. I mean, we all know, right? If you go out for a night of drinking or whatever, you can bump your LFTs. Those kinds of things are all transient. That really has not been an issue in these trials.
Excellent. That's great to hear and appreciate the summary there. Yeah, you mentioned the phase II. We'd just like to talk about, you talk about prospective validation of these biomarkers. Did you do that similarly with this program as you did with ALTO-100 ? Can you just talk a little bit more about this EEG biomarker? I think over the last several months, you've disclosed a little bit more information on what exactly it is that you're looking for in these patients. What's it actually testing, and what would you like to tell folks about that?
Yep. Just as a reminder for how we do the data science, as it were here, and it's really a data science-oriented approach, you take your phase IIA, in this case, a large, almost 240-person study. You split it into two populations. One population is your discovery population. That's where you find your biomarker. The second is your test or validation population, totally independent, locked and blinded and inaccessible to people determining the biomarker. You find your biomarker. You nominate it. This is the only EEG biomarker we tested. You then test prospectively in that validation or test set whether you can stratify. In stratifying by the biomarker, you get a difference in clinical outcome. That's exactly what we saw. You can then test the biomarker in other populations where we have EEG and other interventions. For example, standard of care treatments, placebo, and so forth.
We did all that to verify that it does not predict nonspecifically a higher response. That gets taken into your patient selection approach in the phase IIb, where we both include, by the way, people with the biomarker and those without. We enrich and power on the people with the biomarker, but by including people without, we also help maintain patient expectations. That was very successful in the 100 study with respect to keeping placebo responses low, which is an important part of the whole process, obviously. What we have learned since then is that this biomarker, which our machine learning approach had designed to be as simple as possible, so it is a single signal in the brain called sample entropy, which relates to neural variability, that that actually gives us mechanistic insight.
Higher neural variability, higher entropy is more biomarker positive-like, more patient-like, and leads to better response. That neural variability, we think, relates to low dopamine. As I mentioned before, we've now directly demonstrated in humans that when you deplete them of dopamine, you create that neural variability calculated in exactly the same way as we calculate the biomarker for our trials. In as analogous a method as possible in rodents, we can likewise create that biomarker, in this case, with two different drugs, both of which stimulate the 5-HT2C receptor, which decreases dopamine, whereas agomelatine ALTO-300 increases dopamine by blocking the 5-HT2C receptor. All of that gives us a nice mechanistic link. Also means, by the way, that we have identified a unique human biomarker for dopaminergic function in clinically relevant ways, and that itself we're building on in different ways.
It really shows the insights possible by taking a precision approach, by leveraging machine learning, and importantly, by protecting yourself from false discovery in the way that we do our training and discovery of models and testing and validating them.
Got it. Yeah, makes sense. That's a really, really helpful summary in describing that. I think it will really help folks better understand, like you said, the sort of data science-focused approach. Focusing on 300 just a little bit more, you recently had an interim analysis for that program that's ongoing right now. I think interim analyses in MDD for a mid to late-stage study are a little bit sort of out of the norm. I think it would be helpful maybe if you can explain why it is that you decided to do that, what are the outcomes of that, and how does it impact the program moving forward?
The simple reason why we did that is because of the learnings from ALTO-100 . We wanted to make sure that what we learned about site-level risk is something that we can address proactively here for 300 and really root out as much as we can and make sure that we're on the right path. Our goal is not to be doing interim analyses for different trials. Exactly as I said, it's not the common thing. It really is to implement as quickly as we could after we got the 100 results, those learnings. What we did here is a retrospective case review. We'd enrolled at that point 171 people, of which about three-quarters are positive for the biomarker and a quarter not. That's exactly what we targeted. All of them, by the way, are split one-to-one for drug versus placebo.
We looked at, blinded to any outcome, what are the factors, especially site-level issues, that we could identify? We took out data from four sites. That was 52 patients. We then did an interim analysis on the biomarker-positive population to ask the question of, in the way we'd powered and set up the study, how does the outcome look with respect to maintaining conditional power for the rest of the trial? That led to or could have led to three different outcomes: stop early for success, stop early for futility, both of which were unlikely as a partial sample, or continue with the potential for sample size re-estimation. It is that middle base case that happened as expected, slight upsizing in the sample to put ourselves in the best spot possible. The upsizing could have been much larger.
I think that speaks to the drug signal that was found. And now, prospectively, identifying all those patient and site-level risk factors and now seeing continually higher and higher quality patients come through because of that. I think that itself has been very rewarding. I think puts us in a good spot looking at a mid-2026 readout and really trying to root out these systematic issues that we see across the field and we saw directly in our 100 experience.
Excellent. No, I think that's really helpful. I think laying out, like you said, that this hit sort of like the base case of your expectations, I think that makes a lot of sense. I appreciate that updated guidance on the readout. I guess what gives you confidence you're going to be able to meet that guidance? I guess you never did you ever stop enrolling patients? Obviously, some sites are now excluded. What have you done to make sure that you're going to be able to meet that recent guidance that you issued?
Yeah. We paused briefly to be able to do the interim, but a number of factors feed into this. One is, of course, as we eliminate sites, we do bring other sites on. We're looking for different profiles for sites, so we don't sources of risk. Also, if you look just at our recruitment history, either for ALTO-100 or ALTO-300 , the numbers that we're using are pretty reasonable for the projection. That trial started in early Q3 2023 and enrolled 171 people by mid-December. If you just take that math and you actually make it even a little bit more conservative and project out, that puts you at mid-2026.
Excellent. Great. There is definitely a few other questions I would like to ask about this, but as we're coming up on time, we'd just like to hear about some of the other studies that you have ongoing. You have a couple of readouts that are expected this year. If you could just run through those sort of at a high level, that would be really helpful.
Yeah. ALTO-203 is an H3 inverse agonist readout first half. That is a POC study looking at single doses in patients with depression, essentially looking for two kinds of or a couple of kinds of outcomes. One is subjective on immediate change in symptoms, potentially reflecting subcortical dopamine effects. Another is the objective effects, for example, on cognition for that drug, on wearables as measures of sleep and activity. We know that H3s have been given in situations of excessive daytime sleepiness. That is coming out first half. Second half is a proof of concept trial that is a little bit later. It is 10 days of dosing in patients with schizophrenia with ALTO-101 , which is a PDE4 inhibitor. We have already identified target engagement with schizophrenia-relevant biomarkers in a dose-dependent way there.
We've also shown that by reformulating this drug as a transdermal patch, we largely eliminate the adverse events that are known for that entire class around nausea, vomiting, and diarrhea while increasing fairly substantially the total exposure. Here, we're looking at a crossover design in patients with cognitive impairment in schizophrenia to see if we improve the EEG signals that are target engagement markers and cognitive outcomes that are key for overall improvement in CIAS.
Excellent. Yeah, no, that patch technology is super, super cool and interesting. Yeah, we're looking forward to updates on those programs. Great. Thanks so much. I guess just really quickly before we wrap up, can you just remind folks of your cash runway and any other sort of final comments before we depart?
Yeah. So about $168 million at last report. That takes us through 2027 into 2028. Our last readout is at the end of 2026. So over a year past that last readout, which puts us in a good spot to execute and then to make the right next steps for the company.
Excellent. Great. We are up on time. Thank you so much, Amit, for joining us for what has been a really great discussion. Yeah, looking forward to staying in touch and for next time. Thanks, everybody else, for joining in and listening. We appreciate it.
Thank you.