It’s now been 3 months since the first site began recruiting for Ionis’s ION717 ASO trial in prion disease (NCT06153966). A total of 9 trial sites are now recruiting worldwide, and today I was brought to tears by the description in this news article of some of the first brave people who’ve volunteered for the trial. I want to take this moment to share some reflections on what this milestone means and what lies ahead.

this is the end of the beginning

In the beginning of the beginning, there was no hope of treating prion disease, because no one knew what prion disease was. A chronicle of all the early attempts to treat prion disease in the clinic is given by [Stewart 2008], particularly E-Table 7. In the early 1980s there were one-off case reports of patients being treated with the antiviral drugs interferon, acyclovir, and vidarabine. Prion disease isn’t caused by a virus, and there was no more hope of success in these efforts than there was for medieval alchemists trying to transmute lead into gold, not knowing that matter was made of atoms whose number of protons determined their chemical element.

Fast forward a couple of decades. The prion hypothesis was widely accepted. But even though we knew what prion disease was, there didn’t yet exist drug modalities that could be rationally designed to target the disease via a known mechanism, or methods to screen for active molecules. The only source of therapeutic ideas was serendipity. Pentosan polysulfate (PPS) and quinacrine are two examples that made it to clinical testing. Why do I say serendipity? Based on reported activity of polyanionic compounds against some viruses — very much the wrong intellectual starting point — researchers hypothesized polyanions could work against scrapie, and by sheer luck, happened to be right [Kimberlin & Walker 1983, Ehlers & Diringer 1984]. One such compound, pentosan polysulfate, came back into focus years later and proved to work even once prions were in the brain, although only with direct intracerebral infusion and only before symptoms [Doh-Ura 2004]. Separately, it was hypothesized (why was never stated) that compounds that antagonize the lysosome or lysosomal proteases could antagonize prions, and in cultured cells, a few including quinacrine did so [Doh-Ura 2000]. Why and how, we still don’t exactly know. Arguably, doxycycline falls in the serendipity bucket, except its preclinical evidence was even less strong. For PPS and quinacrine, there genuinely was some antiprion activity in some system but neither had shown efficacy in an animal disease model analogous to the human situation. And more importantly, the mechanism of action was either unclear or indirect such that it would be challenging to improve and optimize the compounds, establish a target engagement biomarker, and so on. They were the kind of drug candidates that academics would test in a small trial, but from which industry would run screaming.

Then we had an era of small molecule screening that ultimately yielded no drug candidates tested clinically, but yielded for the first time some really impressive results in prion-infected mice. The therapeutic hypothesis was to prevent conversion of PrPC to PrPSc, and multiple labs established moderately high-throughput phenotypic screens for finding compounds that could do this in cell culture. This relied on knowing what causes prion disease, and having established facile tools for studying it, such as PrP antibodies, infected N2a cells, and transgenic mice. It was much more than serendipity. But a weakness was that phenotypic screening is still fundamentally mechanism-agnostic — you want to prevent PrPC from misfolding into PrPSc, but you don’t know how. So when one after another, all the candidates — IND24, cpd-b, anle138b —all failed to translate from mouse prion strains in wild-type mice to human prion strains in humanized mice (references here), there was no path forward. Without knowing how the molecules worked in the first place, no one could hypothesize why they didn’t work against some prion strains, or envision how to modify them to do so. Despite all the progress, we were still “flying blind” in important senses.

Throughout all the above eras, there has been a background drumbeat of reports using antidepressants, antidyskinetics, antiepileptics, painkillers, and so on — the most studied were amantadine and flupirtine. No one could reasonably hypothesize, and to my knowledge no one ever did hypothesize, that such drugs would fundamentally change the course of prion disease. They weren’t intended to be disease-modifying, just symptom-managing. None has very strong clinical evidence for efficacy in prion disease in particular, but regardless of what the clinical evidence says, most neurologists I’ve met have their own favorite regimen of symptom-managing drugs they prescribe. Some may indeed “work”, meaning they may make a patient’s final days more comfortable; this would not be hard to believe, but nor would it be inspiring. Sadly, across neurodegenerative disorders in general, a large fraction of clinical trials are simply testing symptom-managing drugs [Mortberg 2022], and until recently, all drugs with FDA-approved labels specifically for neurodegenerative diseaes were in this category. None is a suitable vessel for our hopes.

Then there was the development of therapeutic monoclonal antibodies against PrP, culminating in the treatment of 6 patients with PRN100. Here, at last, was a rationally designed therapeutic targeted at the root cause of disease, PrP. This was a vastly better idea than anything that came before it. Yet still, the PRN100 program never “had POC”, in pharma parlance. Meaning, there were no proof-of-concept (POC) experiments in animal models to suggest the approach would work. To be sure, PRN100’s mouse analog, ICSM18, had been studied in prion diseased mice, but while it worked in mice against prions in the periphery (outside the brain), patients have prions in their brains, and ICSM18 never worked against prions already in the brain [White 2003]. Moreover, PRN100 had no potential for a target engagement marker — it binds PrP, but it is not known to change PrP in any way that we have the ability to measure, which makes it hard to troubleshoot a negative result. Did not enough drug get into the right compartment of the brain? Did it not manage to bind PrP? Was binding PrP not enough? We still don’t know the answer to these questions.

Ionis’s drug candidate has the missing pieces. Finally, here is a drug rationally designed to target the root cause of disease, PrP, via a known and measurable mechanism — lowering the anount of PrP in the brain. We can measure how much PrP has been lowered in the brains of preclinical animal models, and in CSF of living patients, and ultimately probably in brain autopsy tissue from patients who pass away. This powers us with some ability to predict how much needs to be dosed, combined with an ability to get feedback on whether enough has been dosed. If ION717 fails because it can’t be dosed high enough to effectively lower PrP without toxicity, that’s something we’ll be able to learn from the trial. Moreover, there’s solid evidence from animal models that this approach can work. From both genetic models (het knockouts, conditional systems) and ASO treatment in animal models, we know that PrP lowering is helpful, and we know a fair amount about how the efficacy of PrP lowering depends upon the amount of lowering achieved, the disease stage at which lowering is started, the prion strain or subtype involved, the endpoint to which animals are followed, and so on. This is the first time that a drug that worked in relevant prion disease animal models in relevant treatment paradigms is being tested in humans. While there’s still an enormous leap in translating from animals to humans, we’re leaping from a place logs higher than we ever stood before.

Finally we know what we’re doing. We’re not flying blind. The beginning is over.

this could fail for so many reasons

While there are unprecedented reasons for optimism, the fact remains that statistically, failure is always the likely outcome in clinical trials [Wong 2019]. Here are just a few of the reasons why the ION717 trial could fail.

Toxicology. Because PrP appears to be dispensible, on-target toxicity is somewhat less of a concern than in many Phase I trials. But off-target toxicity is a prominent possibility, and we still know too little about mechanisms of ASO toxicity in the brain, including that we do not know why the huntingtin and C9orf72 ASOs had bad outcomes in trials. To be sure, Ionis will have done a tremendous amount of toxicology studies in animals to de-risk their drug candidate. But to some extent, so does every company for every drug — FDA requires it — and yet industry-wide, safety issues are still a major contributor to early phase clinical trial failures [Waring 2015]. In the words of a wise scientist who I respect:

You don’t know anything about safety in humans until you test safety in humans.

— Fyodor Urnov

Danaher Genomic Medicines Summit, December 7, 2022

Biodistribution. ION717 is delivered intrathecally, so it relies on diffusion from CSF into brain tissue to get where it needs to be. It’s not perfect. In cynomolgus macaques, whose brain is just 5% the size of ours, ASOs have great target engagement in cortex and only minimal target engagement in the putamen, one of the deepest brain regions [Jafar-nejad & Powers 2021]. Data are only now beginning to emerge about how broadly ASOs distribute in the human brain; it appears that ASOs can penetrate reasonably deep through layers of the cortex [Korobeynikov 2022], but how well they reach deep brain regions is still an unknown. Prion disease is a whole brain disease. From everything we know, even in the worst case where we knock PrP down a lot in the cortex and not at all in deeper structures, that’s probably better than nothing. But there would still be jeopardy as to how much better than nothing, and whether it is enough to power a trial.

Potency. In our mouse models, we could knock down PrP by about 50% with ASOs, and this was enough to confer a pretty large survival benefit, even at symptomatic stages. But 50% is not a cure, and it’s probably not as big a benefit as you’d get from 70% or from 90% knockdown. We learned from dose-response studies that any PrP lowering is better than none, but there’s an interaction term between tox, biodistribution, target engagement, and power. If, by bad luck, the ASO is both less potent than hoped, and has enough safety issues to prevent dosing high enough, and distributes less well than hoped, then at some point your target engagement gets kind of weak. If so, this could lead a Phase I target engagement endpoint to be underpowered for lowering CSF PrP, or more likely, lead a Phase II efficacy study to be underpowered for showing survival benefit.

Time to effect. Again speaking from experience with mouse models, PrP-lowering ASOs took 3 or 4 weeks to yield a detectable benefit. ASOs take effect at the RNA level within a few days, but then you’re waiting for PrP to degrade according to its natural half-life which we think is about 5 days, and then you’re still waiting for downstream neuropathological changes to alleviate. So NfL didn’t go down in treated mice until 3 weeks post-dose, and survival curves did not start to diverge until closer to 4 weeks post-dose, which means that when we treated at very late disease timepoints, we benefitted some, but not all, of the mice. Experimental mice have very tightly distributed disease timelines, while humans are more variable, having different genotypes and different disease subtypes. Ionis’s posting says an eligibility criterion is “Early-stage prion disease at the time of Screening”, and you can bet they are trying hard to make sure the trial is full of people still early enough in the disease to realistically benefit. But this is the first time anyone’s doing this (earlier, investigator-initiated trials in prion disease had few or no filters on disease stage at time of enrollment), so there’s certainly a risk that patients progress too quickly, before the drug can really kick in.

Trial design & power. Even for the best possible drug molecule, fate is determined in large part by a whole host of decisions about clinical trial design. These decisions range from smart data-powered decisions to educated guesses to total shots in the dark. What disease subtypes do you include, in what proportions — how many rapid subtype patients can you afford to take, versus slow subtype patients? If we want “early-stage” patients, does that mean an MRC score of 18 or of 13? Do we follow people until they’ve dropped to a 10, or a 5, or until they enter hospice or require life support? If it’s a crossover trial, how long do people stay on placebo before crossing over to drug? How many patients do we need, and in what ratio do we assign them to trial arms? And on and on and on. When the middling results first came out for tofersen Phase III, a lot of people pointed not to any properties of the drug, but rather to the short randomization period (6 months, where most ALS trials are 18 months) as a possible reason for failure. In the end, tofersen got Accelerated Approval, but only after many months of open-label data accumulated, perhaps consistent with the trial duration being a major factor. No company wants to be both cruel and wasteful by randomizing patients for too long, but they all worry that if they go too short, the trial fails and the drug never benefits anyone. Success depends on all, or nearly all, of these parameters being calibrated just right. A big reason why pharma companies all like to copy each other is that no one wants to guess first, and guess wrong. Ionis is brave for wading in first, and deserves our accolades.

All of the above problems interact with each other. If the drug molecule is absolutely incredible, then even a badly designed trial could succeed. If the potency is excellent, then you don’t need to dose as high, and tox is less of a concern. If the trial design is perfect, you’ll have enough statistical power for even a so-so drug candidate.

Notice that some of these risks are specific to the ASO, and some are potentially broader. Tox, potency, and biodistribution are to some extent properties of ION717 in particular, and to some extent may be shared across ASOs as a class. Time to effect is inherent to the mechanism of action — lowering PrP RNA — and could to some extent be true of other PrP-lowering drugs, though there may also be an interaction term with potency. Trial design & power are not specific to the drug at all; they’re properties of our disease, and more specifically, of how new we all are at designing trials in this disease.

Once you appreciate that all these different factors contribute to the risk of failure, you realize that any one trial failure won’t be the end of the road. ASOs are bigger than just ION717. PrP lowering is bigger than just ASOs. I really hope this trial succeeds. If it fails, we will still learn things, a lot of things, even though we may not learn all of the things. It will still have been the right decision to try. And we will try again.