May 15, 2024
Participants
Introduction and Welcome
Oliver Ball: Hey! Welcome to those people who have joined the webinar on time! We'll just wait a minute for some more people to dial in, and we'll get started in just a second.
Yeah, I think we'll get going, and the others can join as and when. So, welcome everybody to the seventh of the Unbridled Excellence webinar series. I'm your host, Oliver Ball.
We set up this webinar series to share some of the insights and experience that Dark Horse has generated over the 10 years now that we've been operating. In that period we've had about 375 clients, all within cell and gene, but across all product modalities and all development stages. So there's quite a good level of experience that we wanted to use this webinar series to share across the industry a little bit.
Since starting the webinar series, one of the most popular requests for webinar topics has been on nonclinical development strategy. So the people have spoken, and today we are answering that call for a webinar on this topic.
So across this webinar, we're going to cover overall nonclinical strategy, deep dives into model selection, dose determination, safety study design and a few other things, too.
So I'm delighted to introduce today Nate and Sean, who are presenting the main topic. Nate is a senior principal and the head of nonclinical at Dark Horse. He's been with the practice now for 5 years, and most recently, before that was leading preclinical development at Serious Biotherapy, a public Bay Area biotech developing numerous cell therapy types. And earlier in his career did his PhD and postdoc at Stanford, in the Gary Steinberg group, where he focused on developing neural cell therapies for stroke.
At Dark Horse, Nate works mainly on nonclinical strategy pathways, including nonclinical testing, analytical development, early stage regulatory interactions like INTERACT and pre-INDs, as well as basically everything that is included within preclinical development.
Sean is a UK-based senior consultant who has specialist expertise in immunobiology, and in particular Gamma Delta T-cells, having previously done his PhD and postdoc at King's College, in the Adrian Hayday lab, and then going on to work at Gamma Delta Therapeutics before joining Dark Horse.
So just before we get started, I want to also mention the next webinar that we have coming up. So this will be on the increasingly popular topic of AAV product characterization. This has been a theme that we have published a few pieces on recently that you may already be aware of. And if you saw us at GCC last week, we had another panel discussion on this topic. So it's very much something that is front and center for people's minds at the moment developing AAV products.
So do sign up for that topic on June 26th. We're going to be having that webinar hosted by Jacob Sexton, one of our principal gene therapy consultants.
So just before I hand over to Nate and Sean for the main topic today, just a quick reminder that you can submit questions throughout the webinar today using the Q&A function in Zoom interface, and we will be having a Q&A session at the later stage of the webinar, where we'll be answering those questions. So don't be shy. Submit your questions there, and we'll try to get to them later in the webinar.
And finally a reminder that the webinar will be available to view on demand. So if you want to share it with your colleagues, or re-watch it in future, there will be that option.
So without further ado, I will hand over to Nate and Sean to get into the meat of the discussion today.
Overview of Nonclinical Development Strategy
Nathan Manley: Great. Thank you, Oli, and thank you, everyone for attending and taking time out of your busy schedules to join us for this webinar. Today, Sean and I will be discussing nonclinical development for cell and gene therapy products. Suffice it to say there is no one size fits all solution when it comes to nonclinical strategy for cell and gene based products. Rather, these kinds of biologics typically require de novo construction of a nonclinical package that can account for the product's biological complexity and how that may impact the target clinical population. Nonetheless, Sean and I have observed some common challenges and lessons learned both from our time at Dark Horse and having previously worked at cell and gene therapy companies which we'd like to share with you today.
So during this webinar, we will focus on 3 main topics in nonclinical development, namely, model selection, dose determination and design of pivotal nonclinical safety studies, after which, as indicated by Oli, there will then be time for some question and answer sessions. So please, as you're listening to this talk, if some questions pop into your mind, go ahead and place those into the Q&A panel, and we will try to get to those during the Q&A session.
So before we dig into our 3 focus topics, I would first like to set the stage a bit with an overview of nonclinical development strategy. Generally speaking, formal nonclinical development can be divided into 3 main stages, each of which we will touch on in today's webinar.
Stage 1, starting on the left, involves the selection and development of suitable nonclinical models to characterize product efficacy and safety which may include a combination of in vitro and in vivo model systems. Stage 1 also is when a candidate product will be verified to have therapeutic potential via generation of pilot proof of concept or POC efficacy data, which serves as a key indicator that further product development is warranted.
During Stage 2, moving along to the right, chosen nonclinical models are then used to perform dose finding studies both with respect to efficacy and safety. In addition, Stage 2 activities often include further refinement of things, such as study endpoints as well as gaining a deeper understanding of model variability and collection of pilot safety and biodistribution data.
Then, finally, information gathered during stages 1 and 2 then feed directly into Stage 3, which consists of the final pivotal nonclinical studies to enable first in human or FIH studies. Critically, the progression of nonclinical development must be properly aligned with ongoing CMC and regulatory activities to ensure that the resulting nonclinical data are maximally positioned for clinical translation. At the end of the presentation, we'll return to this concept of interdependency, and we'll highlight some of the key CMC and regulatory activities that should align with nonclinical development, which are also represented here on this slide.
So this is, of course, kind of an ideal scenario where there's this nice logical stage progression of nonclinical development. It doesn't always happen this way. And we're going to touch on a couple of scenarios where this may not be the case and what to potentially do about it, but where and when possible, this is certainly the way that we recommend one progresses through their stages of nonclinical development.
Model Selection
So now, moving to our focus topics, let's begin with model selection. And beginning specifically with pharmacology model selection, 2 key parameters to consider in this case are, 1, the model's ability to recapitulate disease state or relevant injury state, and 2, the model's ability to maximally support product engraftment or integration depending on what the product is, and its subsequent persistence.
In the context of pharmacology studies, accurate modeling of disease or injury state should be the primary driver to maximize the likelihood that nonclinical efficacy data will translate to meaningful therapeutic benefit in the target clinical population. Some key questions relevant to modeling disease or injury state include: is there a gold standard model? If you can answer yes to that, then you have your model, and please use that. However, even if that is the case, there are questions you need to ask within that gold standard model or non-gold standard, whatever models are available out there that you are considering, such as how similar is the relevant anatomy and physiology of that model to humans.
If there are limitations, which there almost always are, are those limitations acceptable? We certainly know of scenarios or examples where that is indeed the case. Probably the most common one currently in the industry is the use of the NSG mouse tumor xenograft cancer model as an efficacy model for things like CAR-T or TCR products, etc. This is well accepted as an efficacy model for developing targeted cancer therapies, even though the NSG mouse is completely lacking an immune system that is arguably quite relevant to the function of these types of products, but is nonetheless supportive of the product's primary mechanism of action or MOA. And that's one of the real key questions here to always keep in mind. Even if there are limitations to that model, does it enable you to study and demonstrate your product's primary MOA?
Secondly, how good are the endpoints within a given model that you're considering, and how good should be based on things such as the relative consistency or variability of that model. How much noise are you going to have to deal with? How easy, therefore, will it be to detect a discernible treatment effect? What magnitude of effect will your product have to have for you to demonstrate efficacy? And critically, how relevant are the available endpoints within a given model to clinical outcome, or thought of another way, patient quality of life? As these will be critical for ultimately arguing that risk-benefit profile, and trying to move into first in human studies.
Now, if we move to the other side of the balance, which is perhaps not the primary focus of pharmacology model selection, but still nonetheless important, looking specifically at product engraftment or integration and persistence, some key questions to ask for guiding your selection include: what is my product's expected engraftment or integration profile in humans? And similarly, what is my product's expected persistence profile in humans? So a given model for pharmacology must support sufficient engraftment or integration to allow you to measure therapeutic impact, obviously, but from a persistence perspective also needs to be able to allow you to measure that therapeutic effect for a sufficient amount of time to demonstrate durable efficacy, and how we define durable efficacy completely depends on what your target indication is clinically and what your product is meant to do in that context.
In addition, it's critical to understand whether or not that model enables full functionality of your product. And within that, from a prioritization perspective, again, does it support activity of the primary MOA of your product? If there are secondary MOAs that are hypothesized or known for your product and if not supported by that model, can they potentially be supplemented with other data so that you could still potentially go forward with a given model choice?
When we're thinking about transgene containing products specifically, there also is the important question of homology for that given transgene. Will that transgene, once expressed, be functionally active within the model system? Can it bind a cognate ligand or receptor needed to carry out its function, and that can be looked at both initially by just straight up sequence comparison, but also needs to be explored functionally in a variety of ways.
And then in the context of in vivo gene therapies specifically, another critical question for pharmacology studies, and arguably safety studies, is are the correct cell types or tissue and organ types transduced or integrated by your gene therapy? We are well aware of some differences in the AAV field, for example, where transduction profiles in a mouse do not at all represent what you see in primates. And so there needs to be careful consideration there, of course, for in vivo gene therapies of whether a model system is appropriately representative of what cell types will be impacted or transduced by your product.
So now, what if there are multiple models to choose from? What do you do? What sort of consideration should you be thinking about? Sean is now going to walk us through this fairly common challenge as a scenario.
Sean O'Farrell: Thank you, Nate. So what we've got for you here in this presentation is a few "what if" scenarios and this is the first of these, and you'll see a few of these pop up later in the presentation as well. So staying in the context of pharmacology model selection, as Nate mentioned to you, you can be in a scenario, for instance, where there are several potentially suitable preclinical in vivo models to choose from.
You could imagine yourself or your company in a situation like this, which is entirely arbitrary, and not really based on a disease model in itself, but is more here for illustrative purposes. So what we have here is we have 4 hypothetical preclinical models, 2 in mice, one in zebrafish, and one in a rabbit, and we have assigned different scores to these, based on 5 preclinical model attributes that are hopefully, you will agree, quite important for these models.
So if I just take you through these one by one: recapitulation of human disease is a very important component of these models. So put very simply, how well is the human disease, the human target indication reflected in animals? For instance, things like tumor growth, injury, wound healing, things like that would be quite important to consider, and how well those are reflected. And you can see here that our models score quite differently across the board for that first attribute.
Moving down, the next key piece is the clinical relevance of endpoints, which usually you can translate quite well. But in this instance we've received sort of quite mediocre scoring across the board for these models with zebrafish performing quite poorly on this front. These scores are all out of 3, by the way, on a per row basis.
The next sort of endpoint consideration is actually the readout consistency of your endpoints. Is there going to be a high degree of variation in the data? Or are you going to get some quite tight data points? And you can see here again the board is split. So previously the rabbit scored quite well on recapitulation of human disease, but the endpoint readout consistency only got a 1 out of 3, whereas some of the mouse models have done better as well as the zebrafish.
The next point, then, is the actual permissiveness to use your human product. So can you actually make the human product you intend to make and put it in an animal model without having to add any tweaks, any changes? And again, a lot of these models score quite poorly in this arbitrary situation, whereas one mouse model scored incredibly well, which hadn't been scoring too well so far.
And finally, there's the durability of therapeutic as well as disease effect in these models. Again, where there's some pretty different scores across the board, and then, if you were to find yourself in a situation where you wanted to sort of discern which model might be best, and you sort of add these scores together, and as was intended for this slide, they all score quite similarly. So you may be in a situation where there is no frontrunner model. They all have their strengths and weaknesses. So what do you do in that kind of situation?
So some of the next steps in the nonclinical path that you could consider for a situation like this would be twofold. First of all, you could conduct pilot nonclinical studies to evaluate the magnitude and consistency of the therapeutic effect of your drug product candidate. And this one's quite an interesting point, because what I would encourage you to think about is actually, if we look at this table right here, this is telling us one thing, but you might find that in your hands some of these parameters change a bit, right? Maybe you will get better readout consistency in your own hands. Maybe there is a possibility to have a more clinically relevant endpoint, etc. Maybe you have better durability of effect depending on the mouse house that you're in, for example. So these things can change in your own hands. We would always encourage you to consider that.
And, secondly, is considering the potential just to strengthen the risk-benefit argument by actually collecting data in multiple nonclinical models. So if you find yourself in a situation like this one, you may say, "Okay, they're all scoring quite similarly, maybe we could just run those 2 mouse models, for instance, side by side and have a more robust package that way." So there are methods to get around this. But this is a nice "what if" that we do encounter quite often with clients, so hopefully, that's at least some food for thought.
So if we move on to the next slide, please, Nate, we're going to now talk you through a "what if" scenario that's perhaps even more extreme than the one that we just showed you. It's one that's maybe pretty familiar with some of you: when there are no relevant in vivo models, or at least not ones that are rapidly identifiable. What do you do then? What are the considerations that you could maybe take on board?
So you may find yourself in this kind of situation, right? So, just going from top to bottom, you may be in a situation where it's very likely, for example, that you would have to go to clinic without much in vivo efficacy data. However, in our experience, the best recipe for regulatory success on this front is to actually robustly support that position by evaluating all your options. So what do we mean by this?
So if we go to the top left of this graphic and we see literature evaluation, that's typically where you probably start. And let's say, we look at the literature, and we do find, we go down this graph, and we do find maybe there's more than one model possible. We've decided this doesn't sound too bad. Let's do some feasibility pilot studies. But we find out in those studies that the primary mechanism of action is not supported, or there's no engraftment of your cell therapy. So you really are kind of stuck with no in vivo model then.
Conversely, if we go back to the top left and we do our due diligence, we research the literature, and we really find that there's no relevant models for drug product efficacy testing, we always end up in the same green square in the middle, here, and there's sort of 4 points that you then could consider that might help you put together a nonclinical line of argumentation.
So first of all would be literature evaluation and any pilot data that you may have, and what we mean by this is actually presenting that in perhaps a more concise way to really show that you've quote unquote "done your homework." I know this may not apply to all of you, but you may find yourself in a situation where there's actually data from similar products available which you could leverage to your advantage and to support your nonclinical position.
Thirdly, perhaps most importantly, you might say, is actually to have in vitro efficacy data to support your position, right? So if there's no possibility to have any in vivo data, a strong in vitro package is almost certainly required in our experience.
And fourthly, which actually will bring me to the second half of this slide is to consider homologous modeling. So what do we mean by that? If we consider ourselves in a position where there's no relevant in vivo model, and we maybe find ourselves in a position where we do need to do some in vivo work, we need to make some sort of animal equivalent product to test in vivo. And here are some of the things that in our experience can be considered when you find yourself in that situation.
So here we're really thinking about developing a surrogate animal derived product. So again, if the slide is being presented by myself, you will be put through the paradigm of CAR-Ts. Unfortunately, that's my background. So for this scenario here, let's imagine ourselves in a position where we're developing a CAR-T that targets a cancer testis antigen, like NY-ESO-1 on tumor cells. And we've really drilled in and we found there's no really relevant in vivo model to test this example product.
There's a number of variables that we can then consider in our nonclinical argumentation. So first of all, is actually the prevalence of the antigenic target in animals, so is it expressed at all? If not, is there maybe a similar antigen that could be used?
Secondly, is the degree to which the animal product can be made. And I know for some products that can be incredibly difficult, and that might be a show stopper at this point. But, for instance, with CAR-T, if you take a moment to think about it, it's not too bad, right? You could get immune cells from the periphery of the mouse, peripheral blood, or the spleen. You could change all the CAR components to their murine equivalents. Right? You could have murine CD8 transmembrane domain, murine CD28 costimulatory domain, murine CD3 zeta, etc. So that's not too bad.
Thirdly, perhaps most nuanced out of this entire list is the animal product homologue performance testing. So how well does that animal homologue or surrogate actually perform? Is it in any way similar to the human product? Hopefully, yes.
And then finally, just to show you the other side of the coin, and maybe to highlight to you that there's some advantages to homologous modeling, is the clinical representativeness of the tumor model in animals, because as we'll get into a little bit later, in NSG mice, the tumors grow sort of under the skin typically, whereas in homologous modeling situations you may find yourself in a position where you can actually recapitulate the tumors at the intended target site for your clinical application. So you could, for instance, have chemically induced skin carcinogenesis or the DSS colitis model.
So these are some of the things to think about in this "what if" situation. Probably a lot there for you to take in. But I will now pass you back to Nate, who's going to take you through some safety model considerations. So back to you, Nate.
Nathan Manley: Thanks, Sean. Yes. So turning our attention now to safety model selection, now that we've successfully selected our pharmacology model, in this case for safety models, the balance is somewhat shifted regarding disease state versus product engraftment or persistence. For safety modeling, ability to support maximal product engraftment or integration, and persistence is key, while disease state is not necessarily a must have for all nonclinical safety studies.
There are instances where it is still necessary which are generally captured by asking the following questions, so 1: is disease state required to support the expected degree of engraftment or incorporation of my product? And just as an example of where this is relevant is when you are thinking about developing a say, neural progenitor cell based product that will be implanted into the central nervous system. It's fairly well established in the literature that those cells don't survive too well when you put them in an uninjured, naive brain. They effectively need some sort of post-injury or disease state niche to enable maximal engraftment, and not only that, but potentially to promote some degree of desired differentiation, if that's part of their therapeutic mechanism. And so in that case, disease or injury state is very important for enabling the expected degree of engraftment of that product.
Secondly, as a key question, are there any theoretical safety concerns directly linked to or driven by your product's MOA? And 2 examples here: one, first going back to the NPC or the neural progenitor cell example, we do know of some instances in the literature where post-CNS injury plasticity or disease state plasticity within that host microenvironment can actually drive ectopic tissue formation by progenitor cells such as NPCs, which would be a safety finding of potential concern that you would not pick up if you were studying those cells in a non-disease or uninjured, naive host environment.
Second example, that is probably much more well known within the industry is within CAR-T cells specifically, and their potential to induce cytokine release syndrome which can really only happen upon their stimulation by target antigen, and so putting them into a naive animal where there is no antigen dependent stimulation would not give you any sense of their potential for cytokine release syndrome, or what kind of cytokine repertoire and magnitude of release they produce upon detection of that target antigen.
So, moving back over to the engraftment and integration/persistence side of the balance, which again, here on the safety side, is really our key driver for model selection. Key questions to guide selection include: what is the best system to support engraftment or incorporation of my product? Where, again, the goal here is maximum amount, and the reason for that is because it then therefore maximizes one's ability to detect treatment related toxicities. If you have maximized the ability for your product to be present and persist within your model system?
And then, secondly, critically, what is the expected persistence of my product in humans? Can my chosen safety model accurately reflect that? And so, for example, on the far end of the spectrum for products that are expected to be permanently engrafting or integrating in humans, you're going to need a model that can support a long in-life duration for your safety study. And so you need to consider that accordingly.
Okay, so a last topic within model selection before we move to our second topic of the webinar is, what about large animal studies? When do I need to consider whether my program might require one or more large animal studies as part of nonclinical development?
In the case of cell and gene therapy products there are generally 3 main reasons that large animal studies are required which include, first, dosing specifically, that a clinically relevant dose from a scaling perspective is not achievable in a small animal system. And a fairly common example of this is when you're talking about a cell plus scaffold based product that is going to be administered directly onto the surface of a tissue or organ. And the relevant surface area of a small animal cannot proportionally scale to the equivalent size that you would use in humans. So that would be one consideration that might warrant the need for large animal studies.
Secondly, is delivery of your product. Almost always, if you are planning to implement a novel clinical administration device that has never before been used in humans for that purpose, you will likely need to do a large animal study to demonstrate feasibility and provide some degree of safety assurance of that novel device.
Thirdly, what if there are just known important differences in relevant or target anatomy or physiology between small animals and humans that will limit the utility of that small animal data? And there are various examples of this out there. Some of which include use of rabbits for studying ocular disorders, because the architecture of the rabbit eye has more similarities to humans than rodents, although even more true in the case of NHPs, and pigs for various things. They're often used for cardiac studies and also even for studies related to the spinal cord.
Now, one other important consideration in the context of large animal studies is specifically for in vivo gene therapy products. There may also be a need for large animal studies driven by one or both of the following: 1 being transgene biology which we touched on previously. If there's really just no or insufficient homology of that transgene to small animal systems, that may be something that needs to then be addressed via large animal studies, unless you're going to go the route of homologous modeling, as described by Sean.
Or, secondly, and I think perhaps most commonly, issues with cell and tissue targeting. And this comes up in the case, for example, of novel engineered viral capsids for AAV programs or lentiviral programs with novel surface engineered proteins on them that will alter their tropism in what will be expected to be a new kind of distribution or targeting profile that typically will warrant a need for large animal studies as well.
So these are all considerations that can help guide an understanding of whether that will be required or not. Ultimately, as will kind of be a reoccurring theme throughout, and perhaps forever in gene therapy, it depends. It depends specifically on your product and your target indication and your development strategy and always, always, always will have to also require buy-in from regulators to really understand necessity or not.
Dose Determination
Okay. So now, with key concepts of model selection covered, we're ready to move into our second stage, Stage 2 of nonclinical development and second webinar topic for which I'll turn it back over to Sean.
Sean O'Farrell: Thank you, Nate. So we now want to talk to you about dose determination, specifically enabling clinical dosing based on the evidence that you may have collected during your nonclinical studies.
So there's kind of 3 parts to this story to the dose determination piece. So the first message we'd like to try and get across to you really, is that nonclinical data can really help justify a product's clinical dosing strategy and indeed overall clinical strategy. So a dosing strategy should be well supported by using multiple lines of evidence. As hopefully most of you are aware, this includes, in addition to nonclinical pharmacology and safety data, a few other things.
So number one is, for instance, your understanding of your product's mechanism of action. Secondly, is the expected magnitude of effect needed to impact patient outcome. Now I know that's quite a big sentence there on its own. So if we take a moment to examine that, a great example that's quite relevant is, for instance, CD19 CAR-T. So, for instance, if we're trying to treat a lymphoid tumor, these tumors can be large. They're disseminated throughout the body. You might need quite a high dose of CAR-Ts in order to eliminate these pretty large tumors that have been refractory to multiple lines of treatment.
Conversely, as hopefully, some of you are aware, a lot of these CD19 CAR-Ts are being tested in autoimmune diseases where perhaps the magnitude of effect needed is lower because you're not looking to eliminate large tumors. You're actually looking to eliminate a small number of autoreactive B-cells that are producing autoantibodies, I should say. So that's a bit more of a dive into that second point.
Thirdly, and again, this relates to a previous slide of mine, is any clinical dosing experience with similar products can really help. Again, I'm aware that doesn't really apply necessarily to everyone.
So let's dive into the nonclinical piece. So in terms of pharmacology data to support a clinical dosing strategy, the objective really would be to support the potential benefit of the starting dose as well as informing your proposed dose escalation, which is quite typical, for instance, for cell therapy products. And the strategy would be to conduct dose finding efficacy studies to identify the minimum and the optimal therapeutic doses.
We can't just do pharmacology on its own. Unfortunately there is safety as well. So in terms of supporting clinical safety, when you go back to your nonclinical studies, you could, for instance, look at supporting the safety of the highest planned clinical dose and the strategy to support that would, for instance, include conducting dose finding safety studies that, for instance, let's say, bracket the clinical doses, sort of capture them, but also include a maximum feasible dose for the safety model.
So we've mentioned to you here that nonclinical data can help justify a dosing strategy. It can actually do a lot more than that. Besides enabling clinical dose levels, nonclinical studies should ideally mirror as well as inform some key clinical dosing parameters, such as the route of administration. Right? So for CAR-T, or for most immune cell therapy products, route, you know, intravenous. There is the delivery device, if that applies, there's the formulation of the drug product as well, in-use stability. So how long are those cells still functional for once they're thawed, for instance, and the dosing frequency.
So nonclinical studies, while they do tell you a lot about how your product works, how safe it is, there is always that final piece to tie it all together where you're actually using those data to support the clinical route that you're setting out in a regulatory submission.
So let's dive into dose determination a little more. So the big question you might be asking yourself at this stage is, when should dose finding studies be performed? And really, we believe that this is most commonly and most effectively done during Stage 2 of nonclinical development, and this should be in sync with CMC.
If we just go back to that timeline that Nate showed you earlier, there is that middle stage that we've just highlighted here for you, which is dose finding and model refinement where you may end up getting some data, some beautiful data hopefully, like the data here below, where you've got 3 different dose levels in blue, all with different degrees of effects. And you can kind of pick your optimum dose. Maybe that middle blue one, for example.
So let's consider 2 things. Let's firstly, consider what you need to go into dose finding, and secondly, what you might get out of it. So in terms of what goes into it, the key preceding activities. And as Nate mentioned to you earlier, there is this balance between CMC and nonclinical, and the cross talk and trying to achieve the milestones together so that it's most efficient. And again, this applies here.
So there's multiple nonclinical things that can be done to inform dose finding study, things like efficacy and safety models being selected, pilot data with your chosen endpoints, locking of the route of administration and the dosing frequency. And the CMC parallel activities that were usually really beneficial to have done by this point would include things like at least a path to the phase 1 process lock. Some candidate assays for identity, purity, and maybe even potency, at least being done regularly or being considered. And finally, the intended drug product formulation.
Now let's say you've ticked all of those off, or most of them. You do your dose finding. What do you get out of it? The key outputs would be things like minimum and optimal efficacy doses, efficacy endpoints confirmed. Maybe you'll get some statistical powering for your pivotal study. That would be quite nice, right? So maybe you don't have to use as many NSG mice as you thought, for example. Maximum tolerated dose in the safety model. And maybe, should you be able to do it, some pilot biodistribution data.
So the main point out of all of this, what we really like to get across to you is that if you do some heavy lifting in Stage 2, your Stage 3, even though that in itself is still a heavy lift, because it's the IND enabling study. Hopefully, your Stage 3 is as low risk as possible. So you can really pick your model that you want. You can go in. You can kind of almost predict what data you get. And hopefully, that puts you firmly on the path towards your first in human trial.
So finally, before I hand back to Nate, we just want to explore the dose finding piece a little bit more and really considering using preclinical/nonclinical data to justify a human dose. And again, because it's me, there's going to be the CAR-T paradigm. So what we're looking at here is looking at this through the lens of a solid tumor CAR-T. And really, considering that actually, the nonclinical dose levels can vary quite greatly on a sort of per kilogram of body weight in mouse versus human, quite greatly. However, this usually is not a problem so long as robust argumentation to support that difference can be put forward.
If we just consider for one moment, I know a lot of you will be very aware of this already, but just to get everyone on the same page. If we compare the NSG mouse to the human, there's some pretty strong differences, and there's a little bit of overlap as well. So if you look at disease induction, and where that disease happens, in the NSG mouse, we're growing human tumor cells in the lab. We're injecting them subcutaneously, growing things like liver tumor cells under the skin, which is an artificial location, whereas in humans, the tumors are actually developing at the target organ, and unfortunately metastasizing as well to distal sites in some instances.
対照的に、NSGマウスおよびヒトにおけるCAR-Tの投与経路は静脈内投与です。これは好都合な共通点と言えます。そして最後に、もう一つの大きな違いは、レシピエントの免疫状態です。NSGマウスは免疫不全状態にあり、例えばCAR-Tの有効性は、CAR-Tが腫瘍細胞を溶解する能力に完全に依存していると言えます。一方、ヒトではリンパ球除去後に一定期間を経て免疫能が回復するため、エフェクター細胞が追加で活性化され、有効性が増強される可能性があります。
このように考慮すべき違いがいくつかあります。ここで実際のデータを見てみましょう。残念ながらこの資料には多くは含まれていませんが、現在フェーズ1、2、または3の試験が進んでいるいくつかの製品について、非臨床試験での投与量とヒトでの投与量の倍率差を示します。ここでも固形がんのパラダイムに沿って、肝細胞がん、消化器がん、卵巣がん、膵臓がんを取り上げました。
ご覧の通り、体重あたりの投与量は大きく異なります。グレーで示した肝細胞がんの例では、初回ヒト投与量はマウスへの投与量より約30倍低いですが、最大投与量はかなり近い値になっています。対照的に、消化器がんではマウスとヒトで約100倍の差があります。卵巣がんおよび膵臓がんのモデルではさらに顕著で、初回投与量とマウスの最大投与量との間に2,000倍の差があります。
投与量が大きく異なる可能性があるという先ほどの指摘は、まさにその通りです。しかし、規制当局への申請において、その論拠を提示する方法はあります。非臨床試験の次のステップを検討する際、私たちは常に非臨床データの論理的妥当性を重視することを推奨しています。パッケージの強みは何か、有望なデータは何か、といった点です。例えば、恣意的ではあっても高い安全域に焦点を当てることです。赤いグループの例で言えば、「非常に高用量を使用する必要があるが、その安全性も確認されている」という両面から説明できます。
また、非臨床モデルシステムの有効性における限界を認めることも重要です。さらに、2点目として、なぜその製品がヒトでより効果的に作用する可能性があるのかを検討します。さらに踏み込むのであれば、インビトロ（試験管内）データを活用することも考えられます。CAR-Tの文脈では、実際にスポンサーがこれを行っているのを見たことがあります。例えば、低いエフェクター対ターゲット比で腫瘍細胞の溶解が見られれば、「インビトロで1対1の比率でも50%の腫瘍溶解が得られているため、ヒトでもより良い効果が期待できる」といった主張が可能になります。
これらが検討すべき事項です。繰り返しになりますが、万能な解決策はありません。しかし、細胞免疫療法製品を開発する際には、こうした状況に直面するのが一般的です。以上が投与量設定の根拠に関する説明でした。ここからはネイトに交代し、安全性について、そしてプレゼンテーションの締めくくりまでを担当してもらいます。
ピボタル安全性試験のデザイン
ネイサン・マンリー： ありがとうございます、ショーン。さて、非臨床開発は正式にステージ3へ移行し、初回ヒト投与（FIH）に向けた最終的な非臨床安全性試験を実施する準備が整いました。ピボタル非臨床安全性試験の主な目的は、臨床での投与戦略が安全であることを証明することです。ショーンが先ほど説明したように、適切な投与経路に基づいた外挿法を用い、計画されている最高臨床用量に対して安全域を確保することが理想的です。
次に、提案する臨床用医薬品のリリース試験（規格試験）を裏付けるデータを提供することです。そのためには、安全性試験に使用する試験品が、意図するリリース規格を満たしていることが理想的です。特に遺伝子編集製品の場合、確認されたオフターゲット編集や転座イベントがあれば、選択した試験品においてそれらが代表的なレベルで存在している必要があります。
そして3つ目の主要な目的は、治療に関連する急性または長期的な毒性を特定することです。その具体的な評価項目は、製品の生物学的特性や、潜在的な懸念を示唆する可能性のあるパイロット安全性試験の結果に基づいて決定する必要があります。この結果は、患者の安全性モニタリングや、必要に応じた対応計画の策定に役立てられます。
ピボタル安全性試験のデザインにおける具体的な属性について説明します。一つずつ見ていきましょう。まず、試験の全体的な整合性を実行面とデータ面の両方から確保するため、ピボタル試験はGLP規制に準拠して実施されるべきです。モデルについては、製品の支持が第一であり、疾患モデルは二の次であると先ほど述べた通りです。
試験品の選択は、先ほど触れたように、リリース試験を含むプロセスおよび製品特性の両面において、臨床製品を代表するものである必要があります。投与量は、選択したモデルにおける最大耐用量または最大実施可能量を含めるのが一般的であり、これにより開始用量を設定し、計画されている最高臨床用量に対して安全域を確保できることが理想です。コホートサイズは種によって大きく異なります。典型的な数値はありますが、規制当局や保健当局の考え方は、非臨床試験に必要な動物数を削減・最適化する方向へシフトしており、状況は変化しています。
中間サンプリングのタイミングは、パイロット試験の持続性データに基づき、製品レベルがピークに達する時期とその後の変化を特定し、適切な動態推移を構築する必要があります。最後に、Q&Aセクションで詳しく触れるかもしれませんが、非臨床ピボタル安全性試験の実施期間についてです。これは製品のヒト体内での予想持続期間や、腫瘍原性のような長期的な安全性の懸念があるかどうかによって大きく異なります。詳細については、後ほど改めてお話しします。
ピボタル安全性試験は、非臨床開発全体の中で最も費用がかかり、期間も長くなるイベントの一つであるため、スポンサーが効率化を図ることは一般的です。実際に、ここに挙げた各デザイン属性において、効率を最大化しコストを削減する成功事例も見てきました。ただし、効率化や削減の試みは、科学的に強固な裏付けが必要であり、規制当局の合意を得てリスクを低減しておくことが理想的です。
締めくくりの前に、「もしも」のシナリオをもう一つ提示させてください。それは、学術機関で初期のプロセス材料を用いて非GLP環境下で実施され、記録や文書に不備がある試験を、ピボタル試験として利用したい場合です。試験品の製造プロセスには、後に自動化や管理が強化された手作業が含まれており、一部の工程では異なるグレードの原材料が使用されていました。また、試験自体も、本格的なGLP試験で通常見られるような網羅的な安全性評価項目が含まれていませんでした。
このようなケースは時折発生します。これを決定的な非臨床安全性試験として使用することは可能でしょうか？可能性はあります。受け入れられる可能性を判断するための緩和戦略としては、提示された非臨床パッケージ全体の強みに焦点を当てる機会があるかどうかが挙げられます。安全性データが不足している場合、有効性データがどれほど強力かを確認します。これらはリスクとベネフィットの計算において、どちらも重要な側面だからです。
可能な限り、医薬品の分析試験データを提供して、初期試験に使用された材料が依然として十分に代表性があることを主張し、インビトロ安全性試験で補完します。インビトロ試験は安価で実施しやすく、開発のかなり後期段階でも、完全に代表性のある試験品を用いて実施できるためです。また、関連する場合は、類似製品の公開されている非臨床データや臨床データを活用します。これらすべてが、主張を補強するために利用できます。
ただし、これは保証されるものではありません。そのため、慎重に検討し、最終的な規制当局への申請前に早期の協議を通じてリスクを低減しておくことが理想的です。
最後に、生体内分布（バイオディストリビューション）についてです。これまで触れてきませんでしたが、従来の低分子医薬品開発に携わってきた方にとって、これは細胞・遺伝子治療版の薬物動態学にあたるものです。ほとんどの細胞・遺伝子治療製品において、非臨床データパッケージの不可欠な要素となります。
モデルの選択にあたっては、製品の生物学的特性を理解することが不可欠です。安全性試験と同様に、その種が製品の生着、統合、および持続性を最大限にサポートできるかを確認する必要があります。また、疾患の状態が製品の拡散に与える潜在的な影響についても考慮しなければなりません。
理想的には、ステージ2のパイロット試験において、決定的な生体内分布試験に含めるべき主要な組織を特定しておくことが望ましく、これは用量設定試験の期間中、あるいはその一部として実施するのが理想的です。これにより、製品の分布に対する用量関連の影響を調査できるだけでなく、パイロット試験で何らかの遺伝子改修やタグ付けシステムを利用することで、パイロットデータを収集するためのプロセスをより効率化・簡素化できる可能性があります。
決定的な生体内分布試験に移行する際は、代表的な試験品を使用する必要があります。つまり、タグは取り除かなければなりません。また、定量的で感度の高い検出手法が必要です。具体的には、その手法が目的に適っており、完全に定量的であり、かつ稀な事象を検出できる能力があることを証明しなければなりません。このアプローチには、qPCRベースのプラットフォームが非常によく用いられます。
CMCおよび規制関連業務との統合
まとめとして、寄せられている質問に移る前に、非臨床開発パスに関する全体的な見解を振り返ります。私たちは、非臨床開発を3つのステージに分け、それぞれのステージで理想的に行われるべき活動を概説してきました。冒頭や随所で触れたように、非臨床側で進行している内容と整合させるべき重要なCMCおよび規制上のマイルストーンが存在します。
例えばCMCの観点から見ると、ステージ1に移行する際には、POC試験に使用する材料を生成できる初期段階の候補プロセスが必要です。しかし、ステージ2に移行して用量を決定し、医薬品の製剤を精製する段階に入る前には、最終的に使用するものに近いベースラインプロセスを確立しておくべきでしょう。そして理想的には、最終段階に向けて、パイロット作業を行う時間を確保しつつ、製造プロセスを固定し、分析試験計画を整えることで、ステージ3に供給する材料を生成できるようにしておく必要があります。
規制の観点から考えると、まずFDAとのINTERACTミーティングや、MHRAやEMAとの科学的助言ミーティングなどを通じて、モデル選択やPOCデータの考え方を早期に提示することができます。次に、pre-INDやpre-CTAといった2回目の早期エンゲージメントを行い、ステージ1および2の全データを提示し、重要な非臨床試験のデザインを実施前に検証します。これらは、IND、CTA、あるいはその他の形式を問わず、最終的な初回ヒト投与（FIH）申請を行う前に行われます。
以上でプレゼンテーションを終了し、Q&Aセッションに移りたいと思います。ありがとうございました。すでにいくつか質問が届いています。ショーン、最初の質問をどれにするか決めていただけますか。
Q&Aセッション
ショーン・オファレル： 承知しました。本日はお忙しい中、ご参加いただきありがとうございます。チャットを通じて素晴らしい質問を多数いただいています。そのほとんど、あるいはすべてに回答できればと思います。では、テンポよく進めていきましょう。一言で答えるような質問ではないことを願っています（笑）。
最初の質問です。「遺伝子編集治療薬では、どのような追加データやエンドポイントを評価する必要がありますか？」
ネイサン・マンリー： 難しい質問ですが、お答えしましょう。遺伝子編集治療薬に関する追加の検討事項ですね。当然ながら、オフターゲット解析、転座解析、そして製品全体の遺伝毒性リスクを理解するために構築しなければならない分析パイプラインがあります。これらは完全にインビトロでの作業ですが、インビトロでの自律細胞増殖アッセイ（サイトカインまたは増殖因子除去アッセイとも呼ばれます）が必要かどうかといった、他の非臨床試験の検討事項にも影響します。遺伝子編集にはゲノム不安定性を引き起こす可能性があるため、遺伝子編集製品にはこれが一般的に求められます。また、これは重要な安全性試験に必要な期間やエンドポイントにも影響します。
遺伝子編集製品は、生存期間がやや長くなる可能性がありますが、これは製品の他の側面を考慮に入れる必要があります。NK細胞のように短命なものなのか、それとも永久的に生着させることを意図しているのか、あるいはその中間なのか。これらを総合的に判断して、必要な生存期間を検討する必要があります。しかし、その最後には、遺伝子編集製品の場合、通常は組織病理学による腫瘍原性評価が必要になることはほぼ間違いありません。
簡潔に言えば以上の通りです。他にも考慮すべき点はあります。これについてフォローアップの質問があれば詳細をお答えできますが、先に進めましょう。ショーン、次はどうしましょうか。
ショーン・オファレル： 承知しました。では、非常に良い質問が来ています。「重要な安全性試験として、毒性試験を組み合わせた試験が使用されることはどの程度ありますか？」これについては、ネイサン、間違いなく「よくある」ことですよね。
ネイサン・マンリー： その通りです。これは、スポンサーが非臨床プログラム全体を効率化しようとする良い例です。モデルの選択が適切であれば、それが可能です。両方のケースで同じモデルを使用できるなら、1つの試験にまとめることができます。また、用量の検討事項が適合する場合も同様ですが、必ずしもそうとは限りません。有効性を実証し、臨床開始用量への橋渡しをするための低用量を使用する場合と、安全性観点からの最大耐用量を使用する場合があるからです。それでも1つの試験として実施することは可能ですが、複数の用量設定が必要になるかもしれません。また、単独試験として実施しない限り、どこかに生体内分布試験も組み込む必要があります。モデルや用量の検討事項、そしてすべての項目を網羅するために必要なサンプリングが適合する限り、試験を組み合わせることはよくあります。
ショーン・オファレル： 素晴らしいですね。では、引き続きインビボ（生体内）試験に関する質問に移ります。「素晴らしい概要でした」というフィードバックをいただいています。ありがとうございます。NSGマウスを用いたピボタル非臨床安全性試験において、雌雄両方のマウスを使用する理由はありますか？CAR-T分野において、性別がCAR-Tの有効性や毒性に影響を与えるという臨床データはあまり耳にしません。非臨床試験では、この点をどのように考慮すべきでしょうか？通常は、雌雄同数で検討するものですよね？
ネイサン・マンリー： はい、その通りです。ヒトの治療において性差が予想されない場合であっても、男性と女性の両方を治療対象とするのであれば、原則としてピボタル安全性試験にはモデル動物の雌雄両方を含める必要があります。
ショーン・オファレル： ありがとうございます。オリ、あと2分ですが、もう1、2問質問を受ける時間はありますか？
オリバー・ボール： あと1問だけ、ショーン！あと1問でお願いします。
ショーン・オファレル： あと1問ですね。わかりました。では、インビボ遺伝子治療製品について、ベクターが新規である場合（例えば、新規のシュードタイプ・レンチウイルスなど）、初回ヒト投与量の設定はどのように進めるべきでしょうか？非ヒト霊長類は腫瘍を移植できないため疾患モデルではありませんが、そのモデルにおけるNOEL（無毒性量）を基準にするのは適切でしょうか？かなり専門的な質問ですが、ネイト、何か考えはありますか？
ネイサン・マンリー： そうですね。その場合は、投与量の正当性を裏付けるために複数の論理を組み合わせる必要があるでしょう。霊長類における無有害作用量（NOAEL）や無毒性量（NOEL）がその一部となることには同意します。また、ショーンが触れたように、作用機序の理解や、製品の働きに関するインビトロ（試験管内）データ、そして既存の類似製品との関連性なども考慮すべきです。つまり、投与量の正当性については複合的な根拠が必要になります。おっしゃる通り、疾患モデルではないにせよ、霊長類のデータはその重要な一部となります。
ショーン・オファレル： わかりました。時間のようですので、このあたりで締めくくりたいと思います。
ネイサン・マンリー： そうですね。
閉会の辞
オリバー・ボール： ありがとうございます。回答しきれなかったご質問がある方は、メールでお送りいただければ、個別に対応させていただきます。本日ご参加いただき、熱心に質問をお寄せくださった皆様に感謝いたします。私自身、今日学んだことは非常に多かったです。皆様にとっても有益な時間であったことを願っています。先ほど申し上げた通り、本ウェビナーはオンデマンドで視聴可能です。終了後にアクセス方法の詳細をお送りします。
最後に、素晴らしいウェビナーを企画してくれたネイトとショーンに感謝の意を表します。また次回のウェビナーでお会いできることを楽しみにしています。本日はご参加いただき、誠にありがとうございました。
ネイサン・マンリー： 最後にこちらのスライドをご覧ください。本日お話しした内容以外にも、Dark Horseの非臨床開発チームは、戦略立案から技術的な試験デザイン、監督、さらには規制当局への各種申請に至るまで、幅広くサポートを提供しております。ご関心をお持ちいただけましたら、ぜひお気軽にお問い合わせください。皆様のプログラムの推進を全力で支援させていただきます。
オリバー・ボール： 皆様、ありがとうございました。
ショーン・オファレル： 皆様、ありがとうございました。