On Recursion's SYCAMORE Trial Readout
Disclosure: I joined Recursion in 2015 when the team was fewer than 10 people. I spent nearly 8 years building the company’s platform, technology and culture, including 4 years serving on the company’s executive team, and am still a significant shareholder despite leaving Recursion over a year ago. I also advise close to a dozen biotech startups and have invested in a handful of others. Despite my best efforts to be objective and embrace nuance (which is one of the main points of this post), I am human and therefore undoubtedly biased. Just be aware.
I never intended this substack to contain frequent writings about Recursion, but the amount of news recently feels like it necessitates it. Tuesday morning Recursion shared a press release that briefly summarized the results of its Phase 2 clinical trial (SYCAMORE) for Cerebral Cavernous Malformation (CCM). This was met with mixed reaction: strong negativity on the outside, incredible excitement on the inside. I figure it’s worth a quick write-up to explain what I see as going on here.
Human beings are extremely reductionist. It simplifies processing for our brains that are already tasked with handling so much constant information. We seek to simplify and binarize things that are in reality quite nuanced. This happens politically, ethically, and scientifically. But the world is not so simple - it’s complex. It’s not so black and white - it’s full of gray. And whenever we are so confident that we understand and know the truth, we need to take a step back and check our biases and assumptions, because odds are we are missing something important.
So what are people missing here? What assumptions are they making based on incomplete or faulty context? Let’s dive into what we know.
What We Know
Recursion just completed its first Phase 2 clinical study (as a company) and equally importantly, the first Phase 2 clinical study ever for the indication of Cerebral Cavernous Malformation. Recursion (excluding what will become true with the Exscientia tie-up) has primarily been operating on indications that put it as either first-in-disease or first-in-class. Which is a hard path to take. No prior clinical efficacy studies in CCM means that neither Recursion nor the FDA were sure what should even be measured to determine efficacy. Which is why the study itself was a “signal-finding” study. The FDA specifically requested the primary endpoint to be safety and tolerability, rather than efficacy, because it was unclear how to appropriately measure efficacy in such a study as it had never been done before. Instead, the secondary endpoints were exploratory measures of efficacy - a grab bag, if you will, to see if anything moves in any significant way during a relatively short period of time in the context of a disease that people fight for their entire lives.
So what did Recursion report?
That REC-994 demonstrated requisite safety and tolerability.
That MRI-based efficacy endpoints showed a time-dependent trend toward reduced lesion volume and hemosiderin ring size in patients in the highest dosage arm (400mg) as compared to placebo.
No improvements in patient or physician reported outcomes (PROs) at the 12 month time point.
Safety and Tolerability
The data showed great safety and tolerability for both tested doses; the 200mg and 400mg dosage arms showed no difference from placebo in terms of frequency and severity of adverse events. Which is fantastic - prerequisite, really - for a disease that will need to be treated for a lifetime. But that’s not what people really care about here, nor what Recursion really cares about here (because safety is just table stakes, at this point). If it’s safe and shows no difference from placebo in terms of tolerability, it’s possible it’s not even appropriately active. So people want to see efficacy. They want to see not just that it’s safe, but that it works.
Objective MRI Measures
Two of the most objective measures included in the grab-bag of potential efficacy endpoints were lesion volume and hemosiderin ring size (a deposit present around the lesions). Here, Recursion demonstrated a “trend towards reduced lesion volume and hemosiderin ring size in patients at the highest dose (400mg) as compared to placebo”. The language is specific, but not overhyped. Recursion could not conclude nor say that there was statistically significant improvement here because it was a signal-finding study that “was not powered to demonstrate statistical significance.” But it shared that the trend was not just towards decreased growth of lesion volume or even stabilization of lesion volume, but a reduction. Patients that had growing lesions growing presumably for years saw them reduced by the 400mg dosage. I’m no clinician, but I’m not sure that was entirely expected or even hoped for. If I remember correctly, the in vivo studies tested REC-994 somewhat prophylactically, where the animal subjects were dosed right from the beginning, showing a reduction in lesion growth rate, but not a reduction in the lesion volumes themselves. To me, this report is encouraging - exciting even. But it’ll be significantly more exciting when I can see the data for myself and draw my own conclusions. Hopefully that will be released soon, and the peer-reviewed journal publication process won’t prevent data access for long.
PRO Measures
Several of the secondary endpoints in the study were various patient or physician reported outcome measures (Cerebral Cavernous Malformation Health Index, Modified Rankin Scale and the SymptoMScreen Score). These can be tricky. They are less objective in many ways, and there are many indications where medicines have been approved without PROs showing a meaningful difference (and these are typically not preferred endpoints, but used in indications like CCM where it is unclear what you should even be measuring in a clinical study). To be clear, Recursion reported no improvement (not even a trend) in PROs for either dosage arm relative to placebo. It’s possible that it may simply take more time to notice a difference in such a chronic condition. Perhaps the lesions must shrink significantly before seeing a reduction in observable symptomatic events in patients. At this point, we don’t know. And that is the point. There is just so much we don’t know about how to measure this disease in a clinical efficacy study, and that was one of the biggest purposes of the SYCAMORE trial - to figure out what it is we should even be looking at.
The Future
So what happens next? I’m going to break this down into two parts:
What’s next for Recursion?
What’s next for TechBio?
What’s Next for Recursion?
I was at Recursion HQ this week for their all-hands meeting where they internally shared the clinical trial readout news with employees (Chris asked me to come and share a few words about the early days of building the platform amidst skepticism and challenges). The sentiment there was immense excitement and gratitude. In the minds of Chris, Najat, Tina and the rest of the Recursion leadership team, it felt like this was a clear win - maybe not a home run, but a solid hit. What are they going to do next? They are certainly excited for the coming clinical readouts (for NF2, FAP, AXIN1/APC, CDiff, etc.) and new clinical entrants (in fibrosis and advanced HR-proficient cancers). But they are also excited to move REC-994 into the next phase, after they meet with the FDA shortly to discuss their plans. Maybe some critics out there will call this naive, or claim they can’t accept a “loss” and move on. I can tell you that the excitement I saw was genuine, and I’ll wait to see the data myself before casting judgment.
What’s Next for TechBio?
Some are claiming that TechBio is dead. They say this proves that AI for drug discovery is a failure, and we should stick with the tried and true (but continually worsening) approaches to drug discovery and development. Hopefully the above will convince you that this one readout is far more nuanced than gets captured on Twitter or by retail traders who drive so much price action while the institutional investors sit tight. But if that context doesn’t convince you that TechBio is not “dead”, hopefully this will: there are just so many different technology-based innovations happening in drug discovery now, and Recursion’s approach primarily represents one. It’s an approach that I think is sound (I’m biased - I significantly helped build it) and I think Recursion is one of the best positioned companies to successfully execute its meta-experiment, testing its approach (see here for my reasoning why). But it’s far from alone. There are other brilliant companies out there doing incredible work in this space that are wildly different from Recursion’s approach, to the point that they are highly independent experiments.
Enveda’s scaled metabolomics approach to natural product drug discovery is ingenious. Terray’s work in next-generational DEL-backed machine learning for small molecule optimization is brilliant. Noetik’s world-model simulators of patient biology bring multi-modal modeling for therapeutic discovery to a whole new level. Monte Rosa’s molecular glue degrader platform takes a wildly innovative approach to targeting proteins of interest, backed by impressive protein surface-based machine learning and proteomics data generation. Even much smaller companies you’ve probably never heard of (like Forcyte and MelliCell) are finding clever ways to generate and analyze meaningful functional data of a form unavailable to traditional biotech approaches. And this doesn’t cover those companies not directly engaged in drug discovery but that are enabling greater experimental and scientific efficiencies through the technologies and services they offer (such as Nomic, Molecule.One and Kaleidoscope).
So is Recursion’s trial readout a referendum on TechBio? Absolutely not. I think we’re still in the early innings of tech-enabled drug discovery, and it’s too early to draw any conclusions about its impact. I, for one, am excited to see the next decade play out with all of this innovation flooding into the space.