Maplight Therapeutics Inc Stock

Good day, and welcome to the top line results from the Phase II ZEPHYR study of ML-007C-MA in schizophrenia. [Operator Instructions] As a reminder, this call may be recorded.

I would now like to turn the call over to Chris Kroeger, Co-Founder and Chief Executive Officer of MapLight Therapeutics. Please go ahead.

Good morning, everyone, and thank you for joining us. I'm Chris Kroeger, Co-Founder and Chief Executive Officer of MapLight Therapeutics. Today, we're going to walk through top line data from our Phase II ZEPHYR trial evaluating ML-007C-MA in schizophrenia. I'm joined by our Chief Medical Officer, Dr. Erin Foff, who will take you through the study design and results in detail. Before we get started, I'd like to remind everyone that this presentation contains forward-looking statements that are subject to risks and uncertainties as described in our risk factors in our filings with the SEC.

Let me quickly walk you through today's agenda. I'll start with an overview of the ZEPHYR results, followed by a brief outline of the significant remaining unmet need in the treatment of schizophrenia and the ways in which we believe the data from the ZEPHYR study address that need. Dr. Foff will then walk through the study design and describe the efficacy and safety results in detail. After that, I'll come back to discuss the ZEPHYR results in the context of the competitive landscape and outline our future development considerations, followed by some closing remarks. We'll leave time at the end for Q&A.

We're very excited to report that ZEPHYR, which was designed as a pivotal trial intended to support registration, met its primary endpoint, demonstrating a substantial treatment effect that was supported by concordant improvements across multiple secondary efficacy measures. Importantly, on top of efficacy on PANSS, we also observed a robust and clinically meaningful improvement in cognitive performance on a prespecified secondary endpoint, something we believe could support the differentiated clinical profile, not only in schizophrenia, but also in our broader development programs in both Alzheimer's disease psychosis and Alzheimer's disease dementia.

ML-007C-MA showed a favorable safety and tolerability profile with low rates of all-cause discontinuations and low rates of moderate GI side effects. We believe this is a profile that will allow patients to take the drug and stay on therapy. ML-007 was designed for ease of use. There's no fasting requirement, and only a short one dose titration, aspects that we believe will support translatability of these results into real-world use. Taken together, we believe this data supports proceeding to an additional pivotal study to support registration, and planning and site identification are already underway ahead of our end of Phase II interactions with the FDA.

On Slide 5, I want to spend a moment on why these promising results matter. Unmet need in schizophrenia remains very high, affecting more than 20 million people globally, and patients remain underserved by current standard of care antipsychotics. Nearly 1/3 showed no meaningful response to treatment, approximately half experienced an inadequate response, and a significant majority discontinued oral therapies within 18 months, often due to lack of efficacy or burdensome side effects like extrapyramidal symptoms or weight gain.

As the first new mechanism approved in decades, Cobenfy represents an important advancement in the treatment of schizophrenia. However, significant obstacles to effective therapy remain. Cognition is one of the key symptom domains in schizophrenia. And while Cobenfy has shown some suggestion of cognitive improvement in its clinical studies, the potential for cognitive benefit remains unproven as those signals came from post hoc or exploratory pooled analyses. Additionally, many patients do not reach or remain at Cobenfy's target dose due to significant side effects, primarily GI side effects. If patients are unable to take a drug, they're unable to benefit from it. And tolerability is significantly worse in the real world relative to controlled clinical trials. We'll discuss why that may be later in the presentation. Importantly, we think the attributes of ML-007 may avoid these liabilities.

Moving to Slide 6. We believe the data from ZEPHYR demonstrate meaningful differentiation and address the key challenges that saddle existing muscarinics. At the 210/3 mg BID dose, we met our primary endpoint on the PANSS total score with an effect size of 0.37, well within range with currently approved therapies used to treat schizophrenia. In our prespecified completer analysis, where modeled assumptions for missing data are not utilized, the effect size was 0.50. We also saw separation on multiple concordant secondary endpoints, including Clinical Global Impression of Severity and positive symptoms.

This dose also demonstrated a robust improvement in cognitive performance. In participants with baseline cognitive impairment, we saw a substantial improvement versus placebo on our prespecified cognitive composite score with an effect size of 0.51. We looked at whether that cognitive change tracks with PANSS improvements and did not find a significant correlation, indicating the effect was not likely secondary to reduced psychotic symptoms. We believe this cognitive signal, consistent with ML-007's strong agonism of M1, is a critical differentiating feature not only for schizophrenia, but also for the treatment of ADP and AD dementia. And we believe the combination of a meaningful effect on PANSS along with a meaningful effect on cognitive performance represents a powerful, comprehensive overall efficacy profile that is unique.

When we set out to develop ML-007, we focused extensively on a precise PK match of the pro- and anticholinergic components of our drug candidate in order to minimize significant peripheral side effects while delivering efficacious exposures, and we believe the safety and tolerability profile in ZEPHYR rewarded those efforts. ML-007C-MA was generally well tolerated. There were no serious or drug-related severe adverse events. Discontinuation across the active arms was low at 19.9%, nearly half the rate demonstrated for Cobenfy in EMERGENT-3. Only two patients discontinued for GI adverse events in the BID arm, and only four participants reduced dose due to adverse events. The rates of moderate GI side effects were low, and there was no hepatic, metabolic, weight gain, or movement-related signal, and no urinary retention. And again, there was no fasting requirement or complex titration, which we believe will reduce barriers to adherence outside the controlled trial setting.

With that, I'll hand it over to Erin to walk through the study design and results in more detail.

Thank you, Chris, and good morning, everyone. I'll briefly take you through the ZEPHYR study design and then walk through the efficacy and safety results in more detail. Moving to Slide 8. ZEPHYR was designed and sized to serve as a registrational study, as Chris mentioned, and our engagements with the FDA thus far have signaled that intention. The study was quite large. We operated at 25 U.S. sites, and we randomized 307 participants equally to placebo, the twice-daily, and the once-daily active arms.

Dosing employed a single dose titration to both target doses, and participants had the option to dose reduce between Weeks 1 and 3 based on tolerability. The primary endpoint was change from baseline in the PANSS total score at Week 5. The key secondary endpoints are listed here in the order in which they appeared in the statistical hierarchy, and I'll come back to that concept later. As Chris has mentioned, we also had a prespecified secondary endpoint of change from baseline in the cognitive composite score of the Cogstate battery in the subgroup of participants who had cognitive impairment at baseline.

Before I get to the study results, I want to take a brief look at baseline characteristics on Slide 9. The 3 arms were generally well balanced in the study. Mean age was 40 to 42 across the arms, and most participants were male and Black or African-American, which is similar to historical inpatient studies. The baseline PANSS total score averaged 96 to 97 across the arms. So I won't dwell on the slide, but just wanted to establish that we started from a population that was generally well matched in baseline characteristics.

On Slide 10, we'll turn to the primary endpoint. As Chris has already mentioned, ZEPHYR met its goal of demonstrating efficacy of active treatment over placebo. Both doses yielded numerical separation beginning as early as Week 1, and the 210/3 mg BID dose showed a significant reduction in the PANSS total score versus placebo beginning at Week 3 and separating further at Week 5, and this is with multiplicity adjustments. In the modified intent-to-treat population, which is the standard population analyzed in schizophrenia trials, the LS mean change from baseline was 11.7 points for the BID dosing arm versus 7.2 for placebo, a difference of 4.5 points and an effect size of 0.37 with a statistically significant p-value.

Though it separated numerically, the effect in the 330/6 mg once-daily arm did not reach statistical significance on the primary endpoint. But notably, it did separate on several other endpoints, including the CGI-S, the PANSS positive Marder factor, and the PANSS positive subscale score. Importantly, this is our lower dose. Its total daily exposure is proportionately lower than the 420/6 mg associated with BID dosing. And I'll also note that the 210/3 mg BID dose is the same dose currently being evaluated in our ongoing VISTA study in Alzheimer's disease psychosis.

And before I move on, I just want to point out the very low placebo response rate in this study. So we made every effort to control placebo response in ZEPHYR, and throughout trial conduct, we detailed that approach on many occasions. And it's clear that those efforts worked very well across this large study. You can see the tight error bars on measures demonstrated here, and you'll see that again on several other slides. This has been a challenge in many studies in this space, and we think that our aggressive efforts, including repeated placebo response training, in-house site monitoring responsibilities, and strict data quality oversight produced a placebo response rate that's rarely observed in trials this size, and it indicates that we can operationalize this control in future studies.

As shown on Slide 11, we also conducted a prespecified completer analysis. And this excludes assumptions about missing data from the MMRM model and instead relies solely on observed data across the full assessment period. This is similar to analyses done with other agents to treat schizophrenia, some of which we'll show you later. In that analysis, the 210/3 mg BID effect size was larger at 0.50, a 6-point difference versus placebo. That larger difference was driven by additional improvement in the BID arm itself and not by a change in the placebo response of completers.

On Slide 12, on the Clinical Global Impression of Severity, we saw a clinically meaningful and statistically significant multiplicity-adjusted improvement versus placebo at Week 5 with a very robust effect size of 0.48. And the reason I mentioned the multiplicity adjustment here is that this did hit in the hierarchy of the statistical plan. And the CGI-S has the potential to be in the label. Several other antipsychotics do have the Clinical Global Impression of Severity in the label, but that does require statistical separation in the hierarchy, which has been demonstrated. Numerical separation again begins early, and you can see that, that continues throughout the treatment period.

On Slide 13, I want to focus on PANSS positive symptoms. So our key secondary endpoint was change from baseline in the positive Marder factor score at Week 5. But we similarly prespecified a look at the positive subscale score, which was the key secondary endpoint in the EMERGENT trial. We again see early and continued separation from placebo, significant by both measures, with an even greater effect with the positive subscale. Here, the effect size was 0.56. As a reminder, the effect size observed on this endpoint in EMERGENT-2 was 0.59.

Next, shown on Slide 14 is an analysis that we were really eager to see and we were particularly encouraged by. So this is the prespecified cognitive composite score of the Cogstate battery. So in the participants with cognitive impairment at baseline, the twice-daily dosing arm demonstrated a robust improvement versus placebo with an effect size of 0.51 and a p-value of 0.041. We looked closely at whether this cognitive improvement was simply tracking with PANSS improvement in this exact population, and we did not see a significant relationship between the two. As you can see, the R values are low and the p-values are nonsignificant with really shallow slopes. And this apparent independence matters, because it suggests that the cognitive signal is not simply a downstream effect of reduced psychotic symptoms.

On Slide 15, I want to spend a moment on why we think this cognitive finding matters beyond ZEPHYR itself. It was observed on a prespecified secondary endpoint, as was mentioned, and it appears independent of PANSS change. It's also consistent with the established role of M1 receptor activation in learning and memory, the strong M1 agonism of ML-007C-MA, and the effect that we observed on cognition in preclinical animals. And together, this provides a compelling mechanistic rationale for this effect. It also matters because cognitive impairment is estimated to affect more than 80% of patients with schizophrenia, and it's a major determinant of long-term functional outcome.

Despite decades of research, no therapy has been approved to treat cognitive impairment in schizophrenia, which speaks both to the unmet need and to the difficulty of demonstrating meaningful cognitive benefit. Sedation and dopamine blockade for many standard of care antipsychotics may, if anything, worsen cognition. Importantly, to have a labeled indication for cognition in this space, one has to conduct a specific study in patients with baseline controlled psychotic symptoms. However, we do believe that this will be meaningful to clinicians, and we may decide to pursue such an indication in the future.

This positive effect on cognition also increases our confidence that this biology may translate across neuropsychiatric populations, including in Alzheimer's disease psychosis, where we know off-label antipsychotic use has been associated with accelerated cognitive decline. Thus, it's particularly relevant to VISTA, our ongoing study evaluating ML-007C-MA in Alzheimer's disease psychosis, and in any study with ML-007C-MA assessing cognition in AD. As noted, VISTA utilizes the same BID dose that demonstrated cognitive benefit in ZEPHYR, and we'd expect to use the same dose in additional Alzheimer's disease indications.

Stepping back, Slide 16 summarizes the effect of twice-daily dosing across the endpoints that I just walked through. And I think this is a really important slide to put into context all of the key secondary and secondary endpoints that we've detailed here and some that we haven't yet talked about. All of these measures favor ML-007C-MA active treatment. And in fact, the only one that doesn't reach statistical significance is the PANSS Marder negative factor. I haven't yet talked about the CGI-C and the PGI-C, but both here did also demonstrate improvement with active treatment with effect sizes of 0.47 and 0.38, and these represent both the Clinical Global Impression of Change and also the Patient Global Impression of Change.

Notably, we also assessed 2 additional important outcome measures. First, participants on ML-007C-MA twice daily were 2.4x more likely to demonstrate a meaningful 30% response on the PANSS total score by Week 5, a threshold commonly employed to represent clinical meaningfulness. We also assessed how likely participants were to have improved enough to be prepared to leave the hospital, which could have meaningful economic and quality of life ramifications. Those on active treatment were 2.8x more likely to be considered ready for discharge from the inpatient setting.

Turning to safety and tolerability on Slide 17, which was a key objective for this program from the outset. ML-007C-MA was generally well tolerated across both doses. Adverse events were primarily procholinergic in nature and mostly mild. There were no serious or drug-related severe treatment-emergent adverse events with either dose. The one severe TEAE observed, a case of pneumonia, was assessed as unrelated to study drug. And the single serious adverse event of worsening schizophrenia occurred in the placebo arm, which helped confirm that we enrolled the right population. We'll spend more time on the next slide on the details in the second section here. But in general, the GI profile indicated low rates of moderate events and low rates of GI-related discontinuations, and we'll discuss later why we think that will translate into the real world.

As anticipated, given our slightly more procholinergic profile, we did not see a signal of urinary retention and had only a single moderate urinary event of urinary tract infection in each of the active and placebo arms. Also, as expected, we saw no metabolic, hepatic, or EPS motor signal with mean weight, glucose, lipid, liver enzyme, and movement-related parameters comparable to placebo. Heart rate increases were small and consistent with those observed and noted in the fesoterodine label, and we did not see evidence of significant blood pressure elevation.

Moving to Slide 18. As noted by Chris, all-cause discontinuation in the actively treated participants in the study was quite low, 19.9%. I do want to mention that we interrogated each of these discontinuations extensively to ensure we were not missing discontinuations due to indolent adverse events. Discontinuation due to a treatment-emergent adverse event was 7.1% in the twice-daily dosing arm. Discontinuation specifically due to a GI TEAE occurred in only 2 participants.

Dose reductions were also infrequently employed. Only 4 participants in the BID dosing arm versus 2 in placebo reduced dose due to TEAEs, and inability to reach the target dose occurred in only 1% of participants. This single participant who failed to reach the target dose was actually removed from the study following titration due to abnormalities in pre-dose baseline labs that were not treatment emergent. Importantly, the large majority of TEAEs were mild with only the single unrelated severe TEAE I've already mentioned and no serious adverse events with active treatment.

Looking at the specific GI events occurring in more than 2% of the active arm and occurring at a rate greater than that of placebo, all were mild to moderate in severity and most were mild. Anticholinergic GI events such as dyspepsia, constipation, and GERD were less frequent than procholinergic events, but similarly mostly mild. To give that context, mild was defined as those adverse events that were easily tolerated with no significant limitations in activities, such as eating and exercising, and moderate events were those that caused discomfort significant enough to limit normal activities and potentially require intervention. We did not have a blood pressure signal. Only a single participant had an adverse event of hypertension that was considered unrelated. They had entered the study with uncontrolled hypertension. And the incidence of orthostasis was similar in the placebo and active arms.

On Slide 19, I want to spend a moment on published data to provide context for tolerability, because we think it's a really meaningful part of the story. Data from the pivotal EMERGENT-2 and EMERGENT-3 studies for the only approved class member of Cobenfy show all-cause discontinuation ranging from 25% to 37%. Discontinuations due to adverse events were similar at 7.1% and 6.4%, and discontinuations due to GI treatment-emergent adverse events were 3.2%, generally higher than what we observed in ZEPHYR.

Critically, in both EMERGENT studies, 7% to 8% of participants had severe adverse events, which, of course, are quite clinically meaningful, and overall rates of moderate-to-severe events were higher, 34% to 36%, than the 26% observed in our trial despite similar study sizes and site numbers. Importantly, we don't believe that the inpatient experience with Cobenfy has particularly matched the outpatient experience. In their open-label extension studies, the discontinuation rates were 51% to 78%, and early data on prescriptions indicate that at 1 year, over 80% of those prescribed drug were no longer taking it.

A separately published real-world analysis of 90 patients suggests GI tolerability challenges with Cobenfy may be more pronounced outside the controlled inpatient trial setting. In that analysis, nausea rates were more than 3x those observed in patients, and vomiting rates were more than 2x those in the pivotal trial. Over 80% of patients did not reach the highest dose, and roughly 25% remained on the titration dose. Importantly, 55% of patients required medication to manage GI symptoms, many of them prophylactically, and the majority did not achieve full relief with medication for these GI side effects.

Reasons for the disconnect here are not really fully understood. It's plausible to us that it's related to exposure differences secondary to the fasted state. In the inpatient setting, the requirement for fasting can be closely adhered to. However, in an outpatient setting, where patients may not be able to fully comply with the requirement to fast 1 hour before or 2 hours following dosing, the impact on exposure may complicate tolerability. When taken with food, trospium levels plummet by greater than 85% and xanomeline exposures tend to increase. This may exacerbate the mismatch between the two components leading to impaired tolerability.

Conversely, ML-007C-MA is recommended to be taken with food because it is a gastro-retentive pill, but the ratio of the components remains within target range in the fasted state, suggesting that tolerability in the inpatient setting will not be altered significantly by outpatient adherence to food requirements.

So with that, I'm going to turn it back to Chris to talk about the context of these results and what's next for the program.

Thanks, Erin. Moving to Slide 21. In order to understand how ML-007 could fit into the current treatment landscape, it's helpful to view it in the context of currently approved and commonly used medications in the space. This chart shows standardized effect sizes, Cohen's d, for PANSS total score across approved antipsychotics, alongside our own modified intent-to-treat and completer results for the 210/3 mg BID dose. These figures come from different studies, different populations, different points in time, and they're not head-to-head comparisons. But with those caveats, the use of effect sizes does allow for some interpretation of relative efficacy results.

With that framing, our effect size of 0.37 falls within the range reported for currently approved antipsychotics, which has historically spanned roughly 0.26 to 0.56. As you can see, many highly commercially successful drugs were approved with very similar effect sizes. Importantly, our completer effect size of 0.50 is at the higher end of the range among those with published completer analyses shown here in yellow. Data from the completer analysis of EMERGENT-2 and 3 suggests the effect size of 0.48 and 0.63, indicating that our effect is comparable when missing data assumptions are not utilized. Notably, we believe prescribing physicians are focused on whether a drug is approved and works and is tolerable enough to be taken by their patients, and they are not terribly focused on effect size or PANSS deltas. Consequently, we do not believe the effect sizes from clinical studies have meaningful impact on commercial uptake.

On Slide 22, I want to walk through potential differentiation for ML-007C-MA within the competitive landscape. Again, I stress that these are qualitative assessments based on publicly available information and our own internal analysis and not head-to-head data, and that comparisons across different study designs warrant appropriate caution. With that said, ML-007C-MA and Cobenfy share similar mechanism, an M1/M4 agonist combined with a peripheral antagonist, and both, along with atypical antipsychotics, have demonstrated substantial efficacy results in large randomized controlled trials.

On cognition, we observed robust improvement on a prespecified secondary endpoint. While xanomeline has demonstrated promise in this sphere, as mentioned, these analyses have been post hoc or from exploratory pooled data sets, and the magnitude of effect has not yet been fully assessed. The M4-only approaches, direclidine and emraclidine, lack M1 activity and thus are not expected to confer cognitive benefit, and standard of care antipsychotics may contribute to cognitive impairment secondary to sedative and dopaminergic effects.

As we've discussed, our tolerability profile reflects significantly lower rates of clinically meaningful side effects. And since ML-007 has no fasting requirement and uses a simple one dose titration, we believe this tolerability profile will translate to real-world use. It's critically important in the schizophrenia population, and patients can only benefit from drugs they are actually able to tolerate. Lastly, like Cobenfy, we are advancing BID dosing into subsequent studies, including in ADP, where Cobenfy is dosed 3 times a day. However, we continue to explore an alternative once-daily dosing frequency.

Moving to Slide 23. Often, there's concern following a Phase II data readout that those Phase II results may not translate to a larger registrational study. ZEPHYR was designed and sized from the outset to support registration and is, in fact, larger than both EMERGENT-2 and EMERGENT-3, the two Phase III studies for Cobenfy. Following our end of Phase II meeting with FDA, we plan to run 2 additional confirmatory pivotal studies. ZEPHYR-2 will essentially replicate ZEPHYR, evaluating the 210/3 mg BID arm versus placebo in a similarly sized study at U.S. sites. Together with ZEPHYR, we intend for it to support our initial NDA submission. ZEPHYR-3 will again evaluate the 210/3 mg BID dose with the potential to evaluate alternative dosing regimens, including a potential alternative once-daily regimen.

Our exposure response analysis is ongoing, but preliminary data suggests that overall exposures in both arms were meaningfully lower than exposures in our Phase I studies. The once-a-day arm did separate numerically on PANSS and delivered separation from placebo on several of the secondary endpoints. So we continue to believe that there is a path forward. Once we've analyzed the full data set and have better insight into the exposure response, we may elect to explore alternative once-daily options in ZEPHYR-3.

Separately, our VISTA study is ongoing in Alzheimer's disease psychosis and is also designed to support registration. It remains on track with top line results expected in the second half of 2027. We're encouraged by the prespecified cognitive signal we saw in ZEPHYR, particularly because the VISTA study employs the same BID dosing regimen that achieved the safety and efficacy results in this trial. Beyond ADP, we're also evaluating the drug candidate potential in additional indications, including mild-to-moderate Alzheimer's disease, for which we also believe a positive signal bodes well.

In summary, ZEPHYR met the primary endpoint on PANSS total score with an even stronger effect in our completer analysis and hit on multiple concordant secondary endpoints. Importantly, ML-007 demonstrated a robust effect on our prespecified cognition endpoint that appears independent of its antipsychotic effect. And our precision matching of the two components of the drug candidate produced a very attractive safety and tolerability profile that we believe will translate well into real-world use.

Taken together, we believe this data supports advancing ML-007 into a confirmatory pivotal study and continuing our investments in indication expansion, leveraging the cognitive signal we observed.

Before we move to Q&A, I want to briefly zoom out on our broader pipeline, which we believe reflects the versatility of our circuit-based discovery platform. Beyond ML-007C-MA in schizophrenia and Alzheimer's disease psychosis, we're advancing ML-004, a 5-HT1B/1D agonist to treat irritability and autism spectrum disorder. Following our recent disclosure of the encouraging results from our IRIS study, we plan to engage the FDA in an end of Phase II meeting. ML-009 is a GPR52 positive allosteric modulator we're developing for hyperactivity and impulsivity and is expected to complete IND-enabling studies in 2027. ML-055 is a next-generation M1/M4 agonist for neuropsychiatric disorders, and it's on track for preclinical candidate nomination in 2026. Lastly, ML-021, an M4 antagonist for Parkinson's disease, which we expect to reach candidate selection in 2027.

With that, we'll open the line for questions. Operator, please go ahead.

[Operator Instructions] Our first question comes from Marc Goodman with Leerink.

Chris, can you talk about the lower drug exposures that you saw and just what you thought happened there and how you're thinking about that? And then also, can you comment on the negative symptoms data? There didn't seem to be anything there. I'm just curious why you thought that was the case.

Thanks, Marc. Appreciate the question. So with respect to our overall exposures, we -- again, the exposure response analysis is ongoing, so it's very preliminary, but we did see lower exposures in both arms of the study compared to what we saw in our Phase I study. So we don't really have a complete explanation as to why that is. Our drug is -- the pill is a gastro-retentive pill, so we're looking at a number of things that may affect gastric retention like proximity of dosing to the meal and also kind of time of day of dosing, because GI transit can slow in the evening. But again, as we get that full data set and are able to perform the full analysis, then we'll be able to make some clear conclusions about why the exposure was a little bit lower. And maybe I'll ask Erin to answer the question on the negative symptoms.

Sure. So what we didn't really talk about deeply was that the negative symptoms did move in the right direction. We did have numerical separation. We didn't hit statistical separation. Those are historically some of the hardest symptoms to move in a short study like 5 weeks. Actually, in our study, the LS mean difference in the BID arm is actually greater than we've seen in one of the two EMERGENT studies. As you know, those were differential responses across the Cobenfy program. So these are distinct cluster of symptoms. They change independently from positive and negative symptoms. So we'll need to dig in a bit deeper, like Chris said, even exposure response for the efficacy in the negative domain to understand that data more deeply.

Our next question comes from Paul Matteis with Stifel.

On this point around exposure and predicted exposure, I guess, what's your confidence level that the exposure levels you're predicting for ADP will be correct? And how would you think about the risk that just other factors of concomitant medications in this population could potentially confound that? And then as it relates to a subsequent schizophrenia study, based on this effect size here, how might you think about powering that study conservatively? And are you thinking that that's likely a two-arm study at this point?

Yes. Thanks, Paul. Appreciate the question. Yes. So with respect to exposures in the VISTA study, I think we do expect the exposures are likely to be similar. We haven't seen any exposure differences between elderly and younger folks in our Phase I studies. And so I don't think we're expecting anything to be different there. And again, I think we believe the effects that we've seen in this study actually are quite clinically meaningful and robust. So we expect that will carry through in the VISTA study. And importantly, having hit on cognition, we think that, I think, bodes quite well for that study. And remind me, what was the second part of the question?

It was about powering for the next schizophrenia study.

Right. So with the design of that study still ongoing, so we don't have any final numbers of that. But again, I think it will be of a similar size to this study. It will be a two-arm study. So given this study was powered for an effect size of 0.50, we had an effect size slightly lower than that. But given that it will only be a two-arm study, I still think it will be generally in the same ballpark in terms of sizing.

Our next question comes from Andrew Tsai with Jefferies.

I appreciate all the analyses for a successful schizophrenia study. So given this kind of data set you've generated now, can you maybe summarize all the various pieces of evidence that gives you the confidence ML-007 can be superior to Cobenfy on efficacy in ADP, especially in the context in this hypothetical scenario where Cobenfy failed, why could you still be better? And then secondly, should we expect an outcome from the end of Phase II meeting sometime second half 2026? Or could it spill into 2027?

Yes. So on the second question, we can't really answer that at this time. Once we've had our engagement with FDA and know when that's going to happen, we'll guide to that a little bit further down the line. With respect to the first question, I'll start and Erin, feel free to jump in. I think when we look at the overall efficacy of ML-007, it's really the totality of the package that we see that gives us real confidence and it is really the basis for assessing whether we think the drug is working in this patient population.

So that's the combination of efficacy on psychosis, which is the PANSS score as well as all the other key secondary and secondary endpoints, all of which move in the same direction and all of which move in a meaningful direction that, combine with a meaningful and statistically significant improvement in cognition on a prespecified endpoint combine with tolerability. So again, patients can't benefit from a drug that they're not able to take and remain on. And so that tolerability piece is so critically important. If you can have a better tolerated drug, which we think we have demonstrated here, patients will be able to remain on the drug, take the drug consistently. And that's really kind of the key element that physicians and prescribers care about.

And I think in terms of the robustness of the effect, if we look across all of the key secondary symptoms, I mean, you can see we had a strong effect on the PANSS score. That is a single data point, an important data point, the one that's used for approval in the space, we hit on that, and with an effect size right in the range of existing atypical antipsychotics. When we look at the completer analysis, obviously, it's quite strong and well within the range of Cobenfy. But beyond that, the other important secondaries like the Clinical Global Impression of Severity had a very strong effect size.

So the clinicians' impression of how well the patient is doing had a strong effect and well within the range of Cobenfy. The positive symptom score had a very strong effect essentially equal to that demonstrated by Cobenfy. And then lastly, the 30% responder rate, the odds ratio there, again, was in line with that for Cobenfy. So all of those things really give us a lot of confidence that the total efficacy picture here is quite strong.

And maybe I'll just jump in and sort of talk a little bit about how we think this translates into ADP, and then specifically, I think, Andrew, your question about what if the ADEPT studies don't read out positively. So specifically, as it pertains to the ADP population, when we look at the subscale scores that moved the most in this study, they are the ones that are most translatable to the NPI-C that's being assessed in the ADP trial. So that's number one. I think that gives us a lot of confidence that the needle will be moved on those meaningful psychosis domains in the Alzheimer's disease psychosis population.

But then specifically regarding why would we still believe we might move if ADEPT doesn't hit, there are significant design differences between our ADP trial and the ADEPT trials, setting aside the relapse prevention trial design, which is a completely different design. In the acute studies, the meaningful differences there that we think may impact and impact possible positive study readout are the fact that the ADEPT studies, the acute studies are longer in duration. They are 12- to 14-week studies. We are mindful of the fact that in natural history studies in Alzheimer's disease psychosis, there's some waxing and waning with some improvement right around that time point. So it's a difficult time point to assess differences from placebo.

Conversely -- and that's really a necessary part of their design, because there is a 5-week titration to target dose there. We can stick with a really truly acute study. We have a 7-week study because our titration is only a week in the VISTA trial. And so we think that, that's a better time point to assess change for acute psychosis. And then lastly, everybody in the VISTA study will be titrated to target dose. They can drop down if they need to based on tolerability, but there really is only the target dose and that single drop-down dose, which we believe will confer efficacy as well.

And in the ADEPT studies, there are multiple steps in the titration. Patients can stay on the lower doses, and it's really only the higher dose that has known efficacy in the population based on the Lilly study. So we still have reason to believe, if ADEPT does not read out, that VISTA will demonstrate efficacy with the same dose that was used here to demonstrate efficacy.

Our next question comes from Sean Laaman with Morgan Stanley.

This is Mike on for Sean. In the context of the cognitive improvement you've observed with ML-007 on Cogstate, do you think the weaker absolute PANSS scores in ZEPHYR relative to EMERGENT and direclidine? Does that support the view that M1 is favoring cognition, M4 agonism is for psychosis? And then I have a follow-up.

Yes. Thanks, Mike. I can start and Erin can follow up. I mean, again, I kind of reiterate that while our PANSS score effect size may be a little bit smaller than the effect size demonstrated for Cobenfy, these are different studies at different points in time, different populations, et cetera. So they're not head-to-head comparisons. It's not easy to do those cross-trial comparisons. That said, we've got a strong effect size. And that's supported by multiple other measures that I just mentioned, including importantly, that 30% responder rate, which more than 1/3 of the patients kind of hit that 30% response rate. So I think all important and telling elements of the efficacy picture.

Clearly, we hit on cognition and we hit on it robustly. And I do think that's a consequence of M1 activity. We have a stronger M1 agonist that is Cobenfy based on both in vitro and in vivo preclinical animal model data, where we showed very significant improvement and improvement in cognition superior to Cobenfy. So we do believe that, that's kind of part of the profile. But we also have a strong M4 agonist. And so that's why we think we're able to deliver across the full suite of symptoms.

That's very helpful. And then maybe just a follow-up question on the dose response with the lower 210/3 BID arm outperforming the high dose once daily. I guess it's too early to say whether it's an actual inverted dose response, but are there any data on Cmax or AUC that would help to explain the incrementally lower efficacy observed at the high dose?

Yes. Thanks for the opportunity to clarify. So it is very much not an inverse dose response. So the 210/3 BID dose has a greater AUC than the 330 once a day dose because 330 is once a day. 330 has a higher Cmax, but we believe efficacy is driven by AUC, whereas adverse events are largely driven by Cmax. So it is, in fact, dose proportional in terms of the effect that we saw.

Our next question comes from François Brisebois with LifeSci Capital.

Just a couple ones, and sorry if this was mentioned. I was just wondering if you can touch on the efficacy side, a little bit of that delta with the completers and how that kind of works there? And then just maybe touch on the efficacy side of a potential functional unblinding in this trial. And then on the tolerability front, if you could just kind of help us understand the differences in severity here seen versus Cobenfy and why it's possible? Maybe -- you mentioned fasting requirements, maybe it's just possible to think that this trial was a lot more like real-world setting than a Cobenfy type trial.

Erin, do you want to respond to that one?

Sure. So in terms of the completer analysis, maybe I'll back up and just talk about the standard analysis first to put that in context. So in a standard mITT, that's the population everybody uses in the schizophrenia trial, when participants leave the trial early for any reason, then the MMRM model has to make assumptions about the course of that particular patient's response at 5 weeks, which is what the primary endpoint is. So it modeled missing data. A completer analysis, on the other hand, doesn't need to do that. So instead, it looks only at observed data at Week 5 and says, in the participants in whom you have all of the actual data accumulated, what is the response? And this can really go either way for you.

Completer analysis do not automatically make things better or worse. They just tell the story of the people for whom you have all of the observed data. And so when we look, for example, at the EMERGENT-2 and 3 trials, the completer analyses there went actually in opposite direction for those two trials. So in one of the two trials, the completer analysis slightly bumped up the effect size and in the other, it bumped it down. And that puts our completer analysis really stacked out in the middle with an absolute delta PANSS that is right in line with them. So that's helpful. And it's helpful for a clinician, because you can say, if your patient is taking this drug 4, 5 weeks, this is likely to be their response.

In think your second question was about functional unblinding. So when we powered the study, we actually thought that because coming in, we believed we have a better tolerability profile, which is now actually borne out to be the case, that it is possible that we would have less functional unblinding in the trial, which is good for the science, but can certainly impact the observed changes in scores because essentially, people cannot unsee projectile vomiting or severe constipation or urinary retention, and it has the potential to impact their scoring. And so we do believe there's a possibility that because our rates of moderate, severe or worse AEs were low, that it's likely that functional unblinding was less of an issue for us than it may have been in other trials.

In terms of translatability into the real world, I think you're hitting on exactly what we think to be the case. We asked in the trial that people take the drug with food, but it wasn't an absolute requirement. If people weren't hungry, they didn't have to eat, they weren't forced to eat. And there was no absolute or minimum requirement on the amount of food that had to be taken. So as we mentioned, during the presentation, what we do know is that if our drug is gastro-retentive, so eating helps retain it in the stomach, which will lead to higher exposures overall. But if people are fasted, then the ratios between the anticholinergic and the procholinergic remain within target. So we don't think that there's going to be a meaningful difference in tolerability between what was seen inpatient and outpatient.

And furthermore, I'm sorry about the long-winded response, I will say, having a single dose titration also means that in the real world where people may interrupt dosing, they may miss a few doses, and we're sort of not sure how that's translating for Cobenfy, when people want to go back on drug, do they have to retitrate, are they retitrating, we do know that you can go right to 210 with minimal difficulty, but certainly being able to take a single dose to the target dose is quite easy with a single dose titration.

Okay. That's super helpful. And then just quickly on the -- if I can jump in, if anyone was to worry that it seems like BID might have been better here than QD on the efficacy side, why the confidence in the ADP program that BID might be sufficient versus 3 times a day? Is there a correlation there that could be misunderstood?

Yes. I mean, there is an exposure with the BID. I mean, just to take a step back with respect to Cobenfy, the reason why that they've gone to TID is because of tolerability. So they lowered the dose and then spread it out over the day, presumably to address tolerability issues. Since this drug was well tolerated and has demonstrated efficacy across the full spectrum of endpoints, including cognition, I think we have a lot of confidence at this dose, which is the dose that's being used in the VISTA study that, that will deliver similar results.

Our next question comes from Sumant Kulkarni with Canaccord Genuity.

First, do you have any specific quantitative comments you can make on how the PANSS scores that you eventually saw in ZEPHYR compared versus your expectations heading into the data?

Erin, do you want to take that one?

Sure. Yes. We sort of backed into a PANSS delta. When we sized ZEPHYR, we actually sized it up to the effect size of 0.50. And we had taken that deliberate sort of downgrade from what had been reported in the EMERGENT trials, because we thought we would likely suffer for some expectation bias. And again, as we talked about earlier, potentially less functional unblinding. If you back that out with assumptions that we made at the time about the actual standard deviations, it would have translated into about a 7- to 8-point change in the PANSS total score. So with that being said, maybe I'll just reiterate something Chris has said and say it a little bit differently. We just don't think that the PANSS total score is telling the entire story even of efficacy on the psychotic symptoms.

So despite the fact that ultimately, that 4.5 point difference was not quite what we had planned for in the beginning, we actually have a stronger effect on the PANSS positive subscale than our original projections would have planned for. And this readiness to discharge measure, which is a really important measure of actual clinical functional outcome and the 30% response rate, our numbers needed to treat to achieve those two outcomes are quite low, 5 and 6, respectively. And on the response rate are actually lower than the most commonly cited meta-analysis for antipsychotics. So ultimately, what that's telling us on the actual clinical impression of patient response of the psychotic symptoms and of the schizophrenia symptoms that there's an efficacy story here that's not being fully covered in just the PANSS total score.

Got it. So given that comment, do you see, what you've seen so far on safety and efficacy versus Cobenfy, as enough for a relatively small company like MapLight and your resources to play the commercial game in what is quite a competitive market in schizophrenia versus larger players? Or do you think the eventual difference maker for this product and the company could be Alzheimer's disease psychosis based again on what you've seen so far?

I think, again, what's going to drive the commercial potential of the drug is whether clinicians like and prescribe the drug, and what can drive that is whether patients are experiencing the benefit of the drug and they're able to take the drug. So I think ultimately, we have a very differentiated profile, much safer, easier to take, easier to stay on. We deliver improvements in symptoms, both on the psychotic symptoms and on cognition. So I think a very well-differentiated profile that will do well relative to Cobenfy.

And if I can just ask, going forward, given that we're short on time, we want to get to everybody, if we can, just everybody limit it to one question going forward, please.

Our next question comes from Joseph Thome with TD Cowen.

Maybe can you go into a little bit more detail on the timing of some of the AEs that you were seeing in the study? And as the ADP study does have a little bit of a longer titration period, do you think the tolerability will be improved in this population? And would you consider extending the titration period for schizophrenia in either ZEPHYR-2 or ZEPHYR-3?

Erin, do you want to take that one?

Yes, I'm sorry. My sound went out just briefly. So this is a question about timing of adverse events. Is that correct?

That's right.

Okay. Great. Sorry about that. Yes. So very similarly to the rest of the agents in the space, most of the adverse events that were experienced did occur early in the trial within the first 1 to 1.5 weeks. And in fact, when we looked very carefully at that, what we learned was that if you made it through the first week of treatment without having nausea and vomiting or other GI events, you are actually very unlikely to develop them later on. So I think that's obvious. And also very similarly, in many patients, we do see tachyphylaxis and resolution over time.

In terms of whether or not a longer titration may help and how that may translate to ADP, I think that, that's reasonable. We took the 1-week titration in ADP, not because we saw a difference in our Phase I trials between shorter and longer titrations in elderly, but because we thought clinicians would be more comfortable with a longer titration there, and it is a more frail population. And it's early days, and we're blinded in that trial, but we're really encouraged by the overall tolerability that we're seeing there.

So we may potentially explore options for at least some flexibility in the titration. I don't know that we'll do that. We're still thinking through designs of the next ZEPHYR study. Right now, we're planning on just replicating what we're doing here. But certainly, with one trial intended to be pivotal under our belt already, presumably no matter what we did, the ability to titrate with just a single dose will remain part of the allowable dosing strategy.

Our next question comes from Ami Fadia with Needham.

I wanted to better understand how you're thinking about the exposure, the target exposure in VISTA, keeping in mind the ability for a onetime step down to 105/1.5 mg dose. And given that in ZEPHYR, you saw slightly lower exposures, how are you thinking about the overall exposure that you'll be able to achieve in VISTA?

Erin, do you want to speak to that one?

Yes. So a couple of things. So we do still think, based on our overall average, the response that we're seeing here, and I think the efficacy across all of the domains, not focusing entirely on that PANSS total, that the 105/1.5 mg dose should still confirm efficacy. And I say this again because some of our early look at the specific domains that cross translate between schizophrenia and ADP have some of the strongest response that we saw in the sub-analysis of the data. So we think that will translate quite well into ability to see efficacy even with a lower dose.

That being said, we had very low use of the drop-down dose here in the schizophrenia trial. I think we mentioned it was only four patients who dropped down due to tolerability, and that is the protocol-driven mandate is that drop-down should be based only on tolerability. And again, so far in the blinded data for VISTA, we are seeing similarly very low use of that drop-down dose. So at this point, not only do we think it will be effective, but we are also not seeing a meaningful use of that dose. So we're encouraged by the fact that most participants are making use of the higher dose.

Thank you. That concludes our question-and-answer session. I'd like to turn the call back over to Chris Kroeger for closing remarks.

Yes. Thank you. And first off, apologies for those questions we weren't able to get to due to time. We're happy to connect with folks online and take your questions. But I also just want to thank everyone for taking the time today and for all of your questions. Appreciate it.

Thank you for your participation. This does conclude the program. You may now disconnect. Good day.

Good day, and welcome to the top line results from Phase II IRIS study for ML-004 in Autism Spectrum Disorder. [Operator Instructions] Please note this call is being recorded. I would now like to turn the call over to Chris Kroeger, Co-Founder and Chief Executive Officer at MapLight Therapeutics. Please go ahead.

Good morning, everyone, and welcome. I'm Chris Kroeger, Co-Founder and Chief Executive Officer of MapLight Therapeutics. Thank you for joining us today. With me is Erin Foff, our Chief Medical Officer, who will be walking you through the Phase II data.

Before we begin, I'd like to remind you that today's webcast contains forward-looking statements within the meaning of the Private Securities Litigation Reform Act of 1995. Such statements represent management's judgment as of today and may involve risks and uncertainties that could cause actual results to differ materially from those expressed and/or implied by those statements. For more information on such risks and uncertainties, please refer to our filings with the Securities and Exchange Commission, which are available from the SEC or on our corporate website. Any forward-looking statements represent our views as of today, June 22, 2026. This presentation only speaks of the date hereof, and we don't undertake any duty to update.

For today's call, I'll begin with a high-level overview, then Erin will take us through the Phase II IRIS study results in detail. I'll follow that with some discussion of the unmet medical need and commercial opportunity, and close with brief summary remarks before we open to Q&A.

Let's start with a high-level summary of the ML-004 program and the efficacy and safety top line results from the IRIS study. ML-004 is an immediate-release, extended-release reformulation of zolmitriptan, a 5-HT1B/1D agonist. We originally developed this approach based on optogenetic data demonstrating that activation of 5-HT1B receptors in the nucleus accumbens reproduced the prosocial effects of MDMA. As we evaluated zolmitriptan in animals, we found that it showed robust activity in preclinical models of both sociability and aggression. The effects on aggression are hypothesized to be the result of suppression of striatal D1 medium spiny neurons.

Based on these findings, we designed the IRIS trial as an exploratory signal finding Phase II study. And that framing matters when interpreting these results. It was not a pivotal registrational trial, but rather was designed to generate clinically coherent, biologically plausible signals to guide the next stage of development. We designed the trial knowing we have strong preclinical data on both sociability and irritability, and we wanted to explore both in IRIS. IRIS did not meet its primary endpoint, a caregiver reported measure of social communication deficits in autism spectrum disorder. Social communication in autism is an indication with no approved pharmacologic therapies and no established treatment-sensitive outcome measures nor a successful regulatory track record.

The study did, however, demonstrate a compelling effect on irritability in ASD, and we believe the data supports the path forward in that indication. In the prespecified, preplanned adolescent subgroup with moderate to severe baseline irritability, defined as an ABIC score of 16 or greater, ML-004 demonstrated clinically meaningful improvement on both the caregiver reported ABC-I with an effect size of 1.33 and a nominal p-value of 0.013 and the clinician-rated CGI-I irritability scale with an effect size of 1.08 and a nominal p-value of 0.036. Those are 2 separate measures assessed by different raters showing consistent results.

Treatment effects were greater among adolescents with higher levels of baseline irritability, an important pattern that Erin will elaborate on in more detail in a few slides. On safety, ML-004 was generally well tolerated. There were no severe or serious adverse events in the active treatment arm. All of the serious and severe adverse events were in the placebo arm. There were no extrapyramidal events and the rates of somnolence and adolescents were low, mean weight gain was lower with ML-004 than with placebo. All of these are important elements of potential differentiation from the standard of care atypical antipsychotics.

Based on this data, we believe there is a clear development opportunity with ML-004 for the treatment of irritability and autism. Current standard of care for the treatment of irritability in ASD are aripiprazole and Risperdal, 2 atypical antipsychotics, and these are the only 2 approved therapies for this indication. The effect sizes we have demonstrated are at or above those shown in clinical trials for these medications and the tolerability of ML-004 suggests a potentially differentiated profile. No weight gain relative to placebo, no extrapyramidal events and low rates of somnolence and adolescents. These are side effects that matter clinically, particularly for long-term use in this patient population.

The regulatory pathway for irritability has been established with aripiprazole and Risperdal approved using the ABC-I scale as their primary endpoint, a scale that we also used in IRIS. We have a clear understanding of the design requirements, the patient population and the operational considerations for a subsequent study. Following our full data review, we expect to request an end of Phase II meeting with the FDA to establish the subsequent development path.

From a commercial perspective, we believe the total addressable market for irritability and autism is large and the side effect profiles of the approved therapies create substantial challenges for patients and families, particularly in the adolescent population, resulting in a very high unmet medical need. Before Erin walks through the clinical data, I want to spend a moment on the preclinical science underlying the irritability signal we are reporting today.

We knew before we designed the IRIS trial that improvement on irritability was a strong possibility based on what was demonstrated in animal models. The biological rationale begins with serotonin in the 5-HT1B receptor. Decreased serotonin or loss of 5-HT1B receptor function has been linked to aggression in both mice and humans across multiple studies. Conversely, increased serotonin or activation of 5-HT1B receptors inhibits the brain circuits associated with aggressive behavior.

This slide shows the resident intruder aggression assay, a standard preclinical paradigm for measuring aggressive behavior in mice. The data show that zolmitriptan, the active pharmaceutical ingredient of ML-004, reduces aggression at clinically relevant exposures with an effect greater than that of risperidone, 1 of the 2 approved therapies to treat irritability in children with autism spectrum disorder. These findings establish the direct preclinical rationale for evaluating irritability-related behaviors in the IRIS study. The clinical results you will hear from Erin are consistent with what this preclinical data predicts.

And now I'll turn it over to Erin to walk through the Phase II results in detail.

Thank you, Chris. I appreciate the opportunity to walk us through the study top line. IRIS was a multicenter, randomized, double-blind, placebo-controlled trial evaluating the efficacy, safety and tolerability of ML-004. 161 participants were randomized 1:1 to either ML-004 or placebo, resulting in 102 randomized adolescents aged 12 to 17 and 59 adults aged 18 to 45.

ML-004 was administered as a once-daily oral bilayer immediate-release extended-release tablet with a flexible dosing paradigm. The target maintenance dose was 48 or 72 milligrams based on the weight and baseline contraceptive use, and there was a 24-milligram option that was permitted based on tolerability. The primary endpoint, as Chris mentioned, was change in the ABI-SC domain score from double-blind baseline to the end of maintenance dosing.

Key secondary endpoints included the CGI-I global score, change from baseline in the ABI-C and change in the ABC-I score from baseline to end of maintenance dosing in participants who entered the study with more than moderate irritability at baseline.

I want to briefly underscore the design context, building on what Chris has already said. IRIS was explicitly an exploratory signal finding study. The primary outcome measure, the ABI-SC, had not demonstrated treatment response in prior clinical studies with other agents. The dosing regimen was untested. And importantly, success in this study was defined not by a single statistically significant result, but by whether the findings were internally consistent, independently corroborated and grounded in a plausible biologic mechanism. That consideration is important as we go on to discuss the data.

This slide summarizes the results across all primary and key secondary endpoints. Starting with the primary endpoint. As noted, ML-004 did not achieve statistical separation from placebo on the ABI-SC Social Communication scale, either in the overall population nor across the age subgroups. The CGI-I global score and the ABIC clinician-rated scales measuring more global autism features also did not reach statistical separation.

The results that really stand out are on the 2 irritability measures. On the care partner reported ABC-I in the total population with a baseline score greater than or equal to 16, which represents the key secondary endpoint, we observed a large effect size favoring treatment, although this didn't reach nominal statistical significance. The same impact was observed on the clinician-rated CGI-I irritability, also clinically meaningful, but again, not nominally statistically significant. However, in the prespecified adolescent subgroup with significant baseline irritability, what we observed was more profound despite the small n. Both on the ABC-I and on the CGI-I irritability scale, this was observed, and I'll go into more detail on the numbers on the following slides. Importantly, across both of these measures, treatment effects were greater in participants with higher levels of baseline irritability, and I'll show you that pattern shortly.

This slide shows the time course of the change from baseline in the prespecified adolescent subgroup with clinically significant baseline irritability. A few things to note. Numerical improvement over placebo emerged early. And by visit 7, which represents approximately 7 weeks of maintenance dosing and is roughly equivalent to the time point used in pivotal trials in this indication, there was a mean change from placebo of 7.2 points and an effect size of 1. This effect was maintained and by the end of the 12-week maintenance period, visit 9, the LS Mean Difference from placebo was 9.6 points with an effect size of 1.33 and a nominally statistically significant p-value.

The increasing separation over time is biologically consistent. This isn't a flash separation seen early in treatment, but rather a signal that has grown across the maintenance period with increasing duration of treatment, and that pattern adds to the coherence of the finding.

This slide is one of the most important in the deck. It shows the irritability results side by side across both measures, the ABC-I, which again is caregiver reported and the CGI-I irritability, which is clinician rated in both the total population and the prespecified adolescent subgroup. In the total population with an ABC-I greater than or equal to 16 at baseline, again, representing the key secondary endpoint. On the ABC-I scale, the effect size observed was 0.64 with a p-value of 0.131. The CGI-I irritability analysis resulted in an effect size of 0.61 and a p-value of 0.153.

Now neither of these results reached nominal statistical significance, but the effect sizes are meaningful in a sample of this size. However, in the prespecified adolescent subgroup of 20 participants, the LS Mean Difference on the ABC-I scale is a substantial 9.6 points. The effect size now observed is 1.33, and there is a nominally statistically significant p-value of 0.013. Similarly, the CGI-I irritability showed an effect size of 1.08 with a nominally statistically significant p-value of 0.036. Two instruments, two rater types, one consistent conclusion. That internal consistency across these 2 assessments is a meaningful indicator of true drug effect.

This slide shows the treatment effect stratified by increasing baseline irritability severity in the adolescent population from an analysis of all adolescents who entered the study with a non-zero baseline ABC-I to those with the most significant baseline irritability corresponding to a baseline score equal to or greater than 18. Importantly, the non-zero group includes 91 of the 102 enrolled adolescents. And with this increased sample size, despite inclusion of some participants with minimal baseline symptoms, we still see a meaningful effect size of 0.4 and a near nominally statistically significant P.

Importantly, the pattern we observe and which is shown here is consistent across both the ABC-I and the CGI-I irritability measures. As baseline severity increases, the treatment effect increases. This is evidenced by increasing effect size and thus a leftward shift of the forest plot, culminating in effect sizes on the ABC-I of greater than 1. The CGI-I mirrors this pattern, reaching an effect size of 1.08 in the adolescents who entered the double-blind period with ABC-I scores indicating moderate to severe irritability. This kind of severity response is biologically plausible.

Participants with the highest baseline burden have the most room to improve and may be the population where the drug's mechanism is most relevant. It also gives us clear pilot data that we can leverage in the design of future studies.

Turning now to safety. The overall profile of ML-004 was generally favorable, particularly in adolescents. The majority of treatment-emergent adverse events were mild, transient and resolved without need for dose reduction. There were no severe or serious adverse events in the ML-004 arm. All severe TEAEs and the single SAE occurred in the placebo arm. There were no deaths or life-threatening events.

The most frequently reported adverse events overall were headache, nausea, somnolence, vomiting, fatigue and dizziness. Adolescents experienced fewer TEAEs overall than adults as shown here. Two adolescents representing 3.9% of the enrolled population discontinued secondary to a treatment-emergent adverse event compared to 0 in the placebo arm. Three specific observations are really worth highlighting here for the adolescent population.

Regarding weight gain, the mean weight gain in the ML-004 treated adolescents was 1.1 kilograms versus 1.9 kilograms in the placebo arm. No adverse events related to weight gain were observed with ML-004. Regarding somnolence, the placebo-adjusted value was low at 11.8% in the adolescents. This is very relevant given the importance of school function and daily performance in this population. As expected, based on the mechanism of action, no extrapyramidal events were observed. These are the safety dimensions that are most clinically meaningful in the context of chronic treatment in adolescents.

Notably, events that are classically associated and common with commercial zolmitriptan in the treatment of migraine such as jaw or chest tightening and paresthesias were rare with only one case each, and there were no cardiovascular, cerebrovascular or vasoactive safety signal.

So with that, I'm going to turn it back over to Chris to discuss the unmet medical need.

Thanks, Erin. The unmet need in autism-associated irritability remains quite high. Irritability and aggression affects 20% to 35% of the 830,000 adolescents with autism spectrum disorder in the United States. Roughly 1 in 6 adolescents with ASD is being treated with an antipsychotic. As we've mentioned, there are only 2 FDA-approved therapies for this indication, and both of these drugs are associated with significant side effects that impact adherence and compliance and carry warnings and precautions specific to the adolescent patient population. There are multiple challenges with the use of atypical antipsychotics to treat irritability.

A significant proportion of autism patients with irritability don't respond to atypicals. So there's clearly a need for alternative therapeutics with non-dopaminergic mechanisms. Additionally, as we've mentioned already, the atypical antipsychotics are associated with very problematic side effects that often impede long-term use of the drugs. This, combined with the stigma of using an antipsychotic in patients without psychosis, leads to hesitancy to utilize this class amongst caregivers.

The side effect profile for the 2 drugs approved for the treatment of irritability is substantial and is shown here in more detail with label warnings denoted. Both contain warnings for weight gain with a significant proportion of patients with weight gain greater than 7% of total body weight. Both are associated with very high rates of sedation-related adverse events, which can profoundly impact school attendance and performance. Both have warnings for very concerning extrapyramidal symptoms and movement symptoms. Risperdal, the atypical with the most robust effect on irritability is associated with hyperprolactinemia and gynecomastia. Both have warnings for orthostatic hypertension and both are associated with a state of GI-related side effects.

ML-004 has the potential to address the significant unmet need in this indication. The efficacy signal from the IRIS study was at or above that shown with Risperdal and aripiprazole on the ABC-I scale, while ML-004 avoids their multiple significant safety and tolerability liabilities. No meaningful weight gain, no EPS and low rates of somnolence and sedation. As such, we believe that ML-004 represents a potentially attractive alternative to atypical antipsychotics for patients and caregivers. Importantly, the ABC-I scale on which ML-004 demonstrated meaningful benefit in our study has served as the primary efficacy endpoint supporting prior approvals for this indication. Additionally, clinical development in the space has historically been reasonably efficient with approval in studies enrolling on the order of 100 patients.

In closing, let me provide a brief summary before we move to Q&A. ML-004 demonstrated clinically meaningful improvements in irritability with effect sizes on par with current standard of care therapies across 2 independent scales, the ABC-I caregiver scale and the CGI-I clinician-rated scale of irritability. ML-004 was generally well tolerated with no drug-related severe or serious adverse events and importantly, was devoid of extrapyramidal symptoms and the significant weight gain associated with atypical antipsychotics and demonstrated meaningfully lower rates of fatigue and somnolence. Taken together, we believe these data support potential development of ML-004 for irritability associated with autism spectrum disorder, an indication with high unmet need and a clearly established regulatory path.

IRIS would not have been possible without the participants and families who chose to be part of this research and the investigators who brought rigor and dedication to the study. We are grateful for their contributions and hopeful that those contributions will ultimately make a difference for the autism community.

Before we open to Q&A, I want to briefly note that IRIS is one part of our broader pipeline. Our lead program, ML-007C-MA, and M1/M4 muscarinic agonist co-formulated with a peripherate-acting anticholinergic remains on track with top line results from our Phase II ZEPHyR trial in schizophrenia expected by mid-August 2026, while results for our Phase II VISTA trial in Alzheimer's disease psychosis are expected in the second half of 2027. As mentioned previously, ML-004 will move toward an end of Phase II FDA engagement.

We'll now open the call for questions.

[Operator Instructions] Our first question comes from Paul Matteis with Stifel.

This is Julian on for Paul. A few from us. Can you just talk a little bit about the threshold of 16 on the ABC-I and why that's important that you prespecified for? Is that something that's common in the field for the approved drugs and thinking about how you plan on designing your Phase III?

And then one more question on this study. If there are other prespecified analyses on secondary endpoints, can you just describe where this ranked sort of the irritability score on the caregiver assessment, among others, did you have especially high hopes here? And then one really quick one on the schizophrenia program, too. Can you just talk a little bit about previous comments as it relates to your ability to show less frequent or milder AEs versus COBENFY in the Phase II? And I'm just curious if you've seen anything in the blinded data that's notable or if it's tracking as expected?

Erin, maybe you can address the first 2 questions, and then I can address the ZEPHyR study. I can jump in until Erin is able to join. So with respect to 16 as the primary endpoint, so 16, we prespecified that 16 cutoff because that really is commonly perceived to be kind of the cutoff for moderate irritability in autism. So that's why we chose the 16. We also showed the cut at 18. 18 is the cut for more severe irritability that was used in the approval studies for risperidone and aripiprazole. So that's the reason why we showed both of those cuts of the data. I'm not remembering what the second question was. The...

Yes. The second question was just like I think you may have shown on the slide that there were some other prespecified analyses on secondary endpoints. I'm just curious like if this was one that you had specifically high hopes for or sort of where it ranked on your -- among your scientific hypotheses for the study?

Yes. So just the primary was social communication deficit. The 2 other secondaries were really kind of broad scales that kind of holistically evaluate autism and then irritability was the third. So I think we have been clear from the very beginning with this program that irritability was an important secondary.

I think we've always couched this as a study to evaluate both sociability and irritability. We have very strong preclinical data as we walk through for irritability and aggression. And so I think our view is that this study didn't uncover an irritability signal. It really confirmed an irritability signal. So this was always a very important secondary for us.

I think we had always viewed the 2 potential outcomes here. One was an improvement in sociability, a large market really affects the entirety of the autism patient population. So a bigger commercial opportunity, larger unmet need. But no demonstrated regulatory path to approval. Nothing has been approved for social communication deficit. And there is not a scale measure that has been definitively shown to demonstrate movement on social communication. Social communication is obviously also a much more complex kind of circuit biology than is aggression and irritability.

So on the irritability side, our sense was very large unmet medical need given the challenges with the existing therapies, but a smaller opportunity than the broader social communication mark. It also has the benefit of a clear regulatory path since there are precedent approved drugs with the ABC-I and typically kind of a more efficient development path going forward since the studies for approval, precedent studies have really been typically been on the order of 100 patients. So very much a key secondary for us really as we design the study. So not something that we looked at retrospectively at all.

And then turning to the ZEPHyR study, we're still blinded to the data, so we can't really say much there. I think what we can say is we had an attractive safety and tolerability profile in the Phase I. We're not seeing anything that's inconsistent with that in the blinded data so far. And that our -- if we look at all-cause early termination rates, those are at or below our expectations going into the study. But again, blinded.

Our next question comes from Andrew Tsai with Jefferies.

This is John on for Andrew. So as far as next steps, you'll talk to the FDA soon. When could we possibly expect an outcome? And then is your goal of running kind of like a Phase IIb next similar to ZEPHyR or VISTA or possibly even running just pivotal Phase III studies?

Yes. Thanks for the question. So I think first things first, we need to complete the comprehensive data set review. So again, this is just top line data. We're still waiting for the full data set, efficacy data, safety data, subgroup analysis, exposure response, et cetera. So once we have the totality of that data, we'll request an end of Phase II meeting with the FDA. And really, the outcome of that, we hope, will be kind of clear direction on population endpoints, ultimately kind of study size and design and the evidentiary requirements.

And once we have all those, we'll kind of make a decision about next steps with respect to definitively sizing and powering a study, which I think would likely be a registrational study, but again, will be dependent on the outcome of both the data and the regulatory interaction. And then obviously, once we know what the study will need to look like, we can assess the risk-adjusted probability of success of that study against the full potential commercial opportunity here and balance that against the opportunity costs of spend on the rest of the portfolio.

Our next question comes from Sean Laaman with Morgan Stanley.

So how should we reconcile the failed primary endpoint in social communication with the strong irritability signal? Is it your interpretation that 004 simply does not have meaningful activity on social communication or that the ABI social communication endpoint was not sensitive enough in the Phase II setting?

Yes. Thanks for the question, Sean. I don't know if Erin is back on, she could answer the kind of the endpoint question. Let me know if you are, Erin, but I can...

I am.

Okay. So why don't I -- I'll kind of address maybe the mechanistic side of things first. And then you can address the endpoint side of things. So from a mechanistic perspective, we started out with the concept that modulating the dorsal-raphe to nucleus accumbens circuitry would be important for modulating sociability. That was clearly true in animal models, but making the jump from animal models to humans is always complex, but particularly complex here given the kind of the numerous various circuits that interrelate to ultimately drive social reward and social interaction.

The effects of 5-HT1B on aggression really are different. We think -- we hypothesize that it's probably subset of D1 medium spiny neurons that are affected that drive that reduction in aggression. So we do think there's a reason to believe that there are different circuits that are being affected and that sociability is clearly a very complex set of interactions and maybe ultimately quite difficult to impact in humans. And then maybe, Erin, you can speak to the endpoint.

Sure. Yes. I mean regarding the question as to whether or not this was a failed endpoint for the ABI-SC or failed effect on social communication, I think the jury is still a bit out on that. The ABI-SC is in essence, although not novel, it's an unproven endpoint. There really is no gold standard in social communication and nothing that has shown treatment response in the past. It was developed as a scale with face validity and internal validation in concert with the FDA. So we had high hopes for that, but of course, is not reported out positively in other trials that were started contemporaneously with ours.

That being said, we had a number of other social communication endpoints in the trial, more traditional measures that include those outputs, such as the SRS, the Vineland. We are still analyzing that data. And then we did have an objective measure of a social preference in eye tracking that was an exploratory measure, and we are still sort of waiting through that data as well in addition to some concordance analyses that will help us understand if there is any underlying signal there. So I think the sort of short then answer to what has been a long-winded explanation is just that we don't yet know whether or not this was a failed endpoint or whether there is some signal of effect.

Got you. And can I just squeeze in what were the absolute ABC-I changes in each arm?

Let me get that exact number for you. I might need to come back to you, Sean, on that shortly after another call.

Our next question comes from Marc Goodman with Leerink.

This is Alyssa on for Marc. Just a few from us. Based on the results, how are you thinking about the optimal patient population for a future study? Specifically, do you think you consider restricting enrollment in adolescents with the higher baseline irritability? Or do you still see the rationale for a broader patient population in either age or irritability baseline scores?

And then secondly, on the flexible dosing design, can you provide more detail on the actual dose distribution in the study? And how many patients or how many participants escalated or remained on the 24 mg option versus the weight-based option?

Erin, do you want to address those?

Yes. So regarding selecting the population, I think the data is telling us that the signal is strongest in adolescent. So I think it's reasonable for us to be on a path to consider restricting the population to adolescents in the next study. And of course, we'll dig a bit deeper into the study data to ensure that, that's the right path.

In terms of what is the appropriate baseline severity, the registrational studies in this field have always used 18 as a cutoff to sort of indicate that moderate to severe irritability. But of course, the approvals for the indication don't restrict to just moderate to severe. The indication approval is simply for the treatment of irritability and autism. We are seeing an effect even in the participants with less severe irritability, and there is a clinical acceptance that 13 and above represents clinically significant irritability.

So I think it will involve a deeper dive of the exposure response analysis in the study for us to fully land on what the appropriate baseline irritability score is and consider some of the operational considerations of difficulty or ease of enrolling participants into the trial. The older antipsychotics came into a field in which there wasn't approved therapies. And so now restricting to a much more severe population could make it more difficult to enroll because those patients are exactly the ones that are potentially on baseline therapy even if it comes with liabilities. I'm sorry, what was the second question again? I'm just...

Dose, dose response rate.

Dose response rate. Okay. Great. We haven't released the specific numbers on that. What I can tell you is that the vast majority of participants were able to escalate to their target dose and the target dose was based on their baseline weight and whether or not they were on oral contraceptive. And most of them were able to escalate to those doses and remain on those doses. So we did have a small proportion of participants that did make use of drop-down doses, but the maintenance dose was the most commonly used dose and most participants were able to tolerate that.

Our next question comes from Ash Verma with UBS.

This is So Youn on for Ash. Our question is, I know there are different programs and molecules, but do you think there's any read across to your other programs such as the efficacy or safety in the ongoing 007 schizophrenia trial?

I'm sorry, I think maybe could you repeat the question?

Sure. So although there are different molecules, do you think there's any read across from this result to your other programs such as the 007 schizophrenia trial?

No. I don't think there's any read across different molecule, different mechanism, different trial.

Our next question comes from Joseph Thome with TD Cowen.

Maybe for a future study, how do you see the need to maybe show non-inferiority or maybe even superiority against risperidone or aripiprazole? Or do you just plan to do a placebo control? Just thinking about what the FDA might want or what you think might be helpful from a commercial perspective? And then maybe second, why specifically do you believe that the impact is seen in the adolescent age group? Is it a time range where irritability is highest in autism? Or are these symptoms maybe just easier to capture on scales at this age? Any thoughts around that would be helpful.

Yes, I can tackle the noninferiority and then the adolescents. So the -- certainly, there's no regulatory expectation and will not be for a comparative trial or a non-inferiority trial. The FDA is well aware of the liabilities of the current study, but just in terms of regulatory guidance for demonstration of efficacy and safety for a drug, it will just be a simple placebo-controlled trial for registrational purposes. Whether or not we decide to pursue any sort of comparative work in Phase IV or switch work, for example, we certainly can discuss the program development.

In terms of why the adolescents, we actually came into the study assuming both on social communication and on adolescents that there may be a strong -- and on irritability that there may be a stronger signal in adolescents. There's actually some biological underpinning. There may be more impact of serotonergic modulation in adolescents. But beyond that, there's just a very practical consideration in a trial like this. So adolescents are surrounded by parents, by teachers. The ability to observe change in these participants is just heightened given the number of reporters and eyes on these patients.

Adults with ASD, particularly in this trial were largely living independently. Most were not in group homes. So there is just a potential for less of an ability to get accurate reporting of symptom change, particularly in people for whom the social burden and reporting might be impaired.

Maybe I can just -- I'll also just add on to that. The end was obviously small. There was a little bit of a bias in the adult patient population with patients with an ABC-I greater than 16, which is the cutoff. There was an imbalance with kind of lower rates in the active arm than in placebo. That's number one.

And in terms of biologic rationale that Erin had mentioned, there is preclinical animal model data suggesting that younger adolescent mice respond to 5-HT1B reduction in aggression, whereas older mice don't. So there is some preclinical suggestion underpinning the adolescent response rather than the adult response in addition to the reporter bias and the baseline entry demographics.

Our next question comes from Sumaira Zamurrad with Canaccord.

This is Sumaira on for Sumant. Just a couple from us. The first one is when you think of -- when you look at the efficacy and safety comparison with other and current antipsychotics, what do you think about placement for 004 as first line or second line eventually? And sorry if you may have answered this already, but how do you mechanistically think about or explain a reduction in irritability, but it's not translating as well into social communication? Will the core positioning be solely based on ASD irritability for 004?

Yes. So with respect to the first question, I think we need to still need to do a fair bit of work to fully understand that. Clearly, what we have -- what we see so far from the data is a robust effect size at or above that shown with the atypical antipsychotic and a safety profile that appears to be dramatically better. So I think clear differentiation and clear potential to be first-line therapy. Obviously, the existing therapies are generic.

And so I think there'll be -- it will take a little bit of time, probably there'll be step edits required at entry into the market, but assume over time, I think the obvious benefit of -- potential benefit of a drug like this would move it to first line. Sorry, can you -- what was the other question, if you could repeat it for me, please?

Yes, sure. We were just asking how do you mechanistically think or explain a reduction in irritability, but it's not translating into improvement in social communications, just considering that the neural networks for both processes are connected? Or will the core positioning for 004 be solely based on ASD irritability?

Yes. So we -- I do think -- obviously, we need to get the full data set and fully understand all the concordance of all the other measures that Erin had mentioned. And I think once we have all that, we can kind of make a decision. I think the primary thrust of our future clinical development will likely be an autism irritability. We may explore some sociability endpoints going forward. But clearly, the strong signal here is in the irritability and the clear regulatory path and the size of the clinical trials make that a likely attractive development path going forward.

With respect to hitting on irritability and not seeing an effect on sociability, again, I think it comes to the -- down to the complexity, right? We have a circuit rationale for social communication deficit, which is modulation of the dorsal raphe-nucleus accumbens circuit that clearly was sufficient in multiple different animal models, but social interaction is a very complex behavior involving both social reward and feedback. And so the complexity of that may just be difficult to modulate in a meaningful way in a heterogeneous population in humans.

We do think that the circuitry is different for aggression. It really is a subset of D1 medium spiny neurons within the striatum that respond to serotonin to reduce aggression. So I think at least that's the hypothesis. So I do think we think there's a different mechanistic rationale for the 2 effects.

We will now take our last question from Ami Fadia with Needham & Company.

This is Poorna on for Ami. Could you elaborate on what the response rate on the ABC-I and CGI-I for the adolescent subset population was? And could you provide additional color on the discontinuations seen in the adolescents? What were the AEs which resulted in those discontinuations?

Erin, do you want to take that?

Sure. So I'm sorry, the first part of the question was cut off for me. Could you repeat that first part?

It was really responder analysis.

Responder analysis. Yes, we are still in the process of doing our full responder analysis and our exposure response analysis. So stay tuned for that. In terms of the age of the discontinuations of the 2 participants, I don't have that handy exactly at this moment.

Sorry, I didn't ask the age. I was asking what the AEs were, which resulted in those discontinuations.

The AEs, thank you. Sorry, the sound was a bit cut off. I believe one of them was secondary to headache and one was secondary to somnolence.

Thank you. This concludes the question-and-answer session. I'd like to turn the call back over to Chris Kroeger for closing remarks.

Yes. I just want to thank everybody for joining us today and appreciate everyone's questions.

Thank you for your participation. You may now disconnect. Everyone, have a great day.