All right. Thank you, everyone. Joining us for the Citizens JMP Life Sciences Conference. My name is Roy Buchanan, Biotech Research Analyst, and we're happy to have Gritstone bio with us. We have Andrew Allen. He's a Founder and CEO of Gritstone. So, Andrew, maybe to get started, thank you for joining us and maybe give us a bit of history of the company, kind of the foundations, the technology platforms, how you came up with the idea and got started, and maybe look quickly on your background as well. I know you have some of the history developing some of the first immunotherapies in cancer. Yeah. Thanks for having us, Roy. Good to be here. So if we get back into history, I trained as a physician with a sort of heavy immunology bias. So I have an MD, PhD from Oxford and London and then entered industry. And as you say, I ended up at Chiron in the early 2000s, which was the owner of high-dose interleukin-2, so the first immunotherapy which sort of resonated with my immunology interests. Chiron obviously was acquired by Novartis, and I stayed in the drug development side. And then, to be honest, I think like many people who'd worked on IL-2, you saw miracles with IL-2. You saw people with widely metastatic melanoma and kidney cancer being cured, literally cured, so that years later they're on no therapy, cured of disease, and they would carry these photographs of themselves with disseminated metastatic lesions in brain, skin, lung, liver. And here they are today, completely healthy. And so there was a time when the only thing any investor wanted to talk about was targeted therapy. And there were some academics and a few companies still working on immunotherapy. And they were widely regarded, I think, as somewhat crazy because who would be interested in immunotherapy when you can do elegant targeted drug therapy against genetic mutations? And the reason for the persistent interest and passion was this notion of real durable benefit. Obviously what we've learned over the years is that targeted therapies, while, of course, wonderful and really life-changing for many patients, unfortunately very, very rarely affect long-term disease control, and acquired resistance is the norm. I was involved in drug development of drugs like that. It was sobering that patients would expect to develop acquired resistance. Immunotherapy was the hope that we had of something that was more durable, but we just didn't understand enough about immunotherapy of cancer with high-dose IL-2. What did we know? We knew that IL-2 activated T cells, and that was about it. Didn't know much about types of T cells, knew nothing about the relevant antigens. So I stayed away from immunotherapy because we just didn't know enough to de-risk the program. But that changed at the end of 2014 when a team at Memorial, led by Tim Chan, published the first evidence that in humans responding to checkpoints, the targets of their T cells seemed to be mutated neoantigens. And obviously a deluge of data followed soon thereafter supporting that notion. And it was clear that there was an opportunity to build a company that would exploit this biological insight, this fundamental insight, and also exploit some of the new technologies around DNA sequencing. So what we were able to do when we started the company in 2015 was begin a program to say we will essentially develop a personalized cancer vaccine. We'll sequence tumors. We'll identify which of the mutations creates neoantigens. And then we'll make personalized vaccines containing those neoantigens. And this therapy should then bring the benefits of immunotherapy to the many patients with colder tumors who essentially derive no benefit from standard checkpoint inhibitor therapy. So that was the initial thesis, and it hasn't really changed. Obviously, we've been building the company. We started off having to develop two technologies. First of all, how do you identify which mutations create neoantigens? When we started, there was a way of doing it that was in the scientific literature. But almost immediately after we began the company, there was a paper from Genentech showing that it didn't work. And so that was kind of sobering. But of course, challenge is opportunity. And so then we built what we now refer to as our EDGE platform, which is one of the early applications of deep learning mathematics. And essentially what you can do is take human tumor samples. There's a technique referred to as immunopeptidomics. Essentially what you do is you take the HLA peptide complexes off of the surface of the tumor cells, and you identify the peptides. You identify at a molecular level the peptides that are displayed on the surface of the tumor cell, which often includes mutated peptides or so-called neoantigens. If you do that analysis on a single tumor sample, you'll get about up to, let's say, 10,000 different peptides off of that sample. You can sequence that sample, and you can start to ask the question, what are the genomic features that predict whether a particular peptide will make it to the cell surface, i.e., be seen as an antigen? You do that times 1,000. Now you've got millions of training data points, and you've got the genomic data, and you can use deep learning mathematics to figure out what are the genomic features that predict whether a particular mutation will or will not be processed and presented on the surface of that tumor cell. Published this in Nature Biotechnology in 2018. This is now the standard. So we were early in that game. Obviously, you've got some established IP now in that space, and everyone basically does this approach, and it works very well. So now when you sequence a patient's tumor, you can predict, or we can predict with over 80% accuracy, positive predictive value, which mutations will be neoantigens. So we've solved that problem. So you're halfway there. Now you've got to administer those neoantigens to humans to drive CD8 T cells. We're very focused on CD8 T cells because those are the killers. Those are the cells that actually kill tumor cells. Most vaccines are not very good at priming CD8s because they're a very, very potent killing force. Naturally, we have a lot of control mechanisms to ensure that we don't lightly generate CD8 T cells. Some vaccines, peptides would be the example, I'll throw out, just are not very good at priming CD8 T cells. It doesn't matter what adjuvants you use. Fundamentally, we're not able to generate good CD8 T cell responses with peptide vaccines. We chose vectors that were known to drive good CD8 responses, particularly an adenovirus vector. That's what we use for priming. Then we developed a self-amplifying mRNA as a boost vector. We use two different vectors. In the world of vaccinology, that is a way of driving the strongest immune response because your immune system sees the same antigens but presented in two different ways by two different vectors. It's referred to as heterologous prime-boost, and it drives a really strong CD8 response, which is what we're all after. We built that. We had to build our own manufacturing facility because doing all of this at scale for GMP was not something you could do easily through third parties. All of that led to our phase 1 study that began around 2018, 2019. We started to see that in patients, we were generating T cells that seemed to be able to kill tumor cells. We saw this in our phase 1/2 study in advanced disease. So about half the patients we treated, you could see their tumor cells being destroyed. And you could see that because their protein markers are going down, and their circulating tumor DNA is going down. And they seem to then have long survival, long PFS compared to the half of patients who didn't have those markers of tumor destruction. So in a non-randomized setting, you had clear proof of concept. We'd also shown on the journey that when we vaccinated patients, they generated T cells. You could see them in the blood. You could see those same T cells entering tumors. You could see them proliferating in tumors. So we were checking all the boxes. The biology was playing out exactly as you would expect. So all of that encouraging work led us to start a phase II study. It's been a tough sort of financing road because what we weren't showing was RECIST responses. When we started this, that's all any investor would care about because they were trained on targeted therapeutics. They're not focused on survival. They're focused on short-term RECIST responses, i.e., lesion shrinking, because that's what you look for with targeted therapeutics. Now, of course, this is a different approach. What we're trying to do and what we are doing is driving T cells into tumors where they proliferate. So the notion that I have to see quick lesion shrinkage in order for the therapy to be working is clearly wrongheaded, but it's a widely held belief. So showing good PFS data and ctDNA responses in the absence of RECIST response is something that was a little bit challenging for many people. Of course, at the end of the day, what we care about is survival. If we're on the road to improved survival, we keep going. We designed a randomized phase II study, and I'm sure we'll be talking about that. That's the history of the company. That's how we got here to this point we're at now, where we're generating data from this randomized phase II on this platform. Yeah. Great. So let's talk about that phase, the phase II, III. So the GRANITE, that's what you call the construct, I believe, that's going into based off the EDGE technology. And you went into microsatellite stable colorectal cancer, considered a cold tumor. So maybe just describe the rationale behind picking that indication. Why did you pick this cold tumor? Yeah. And maybe about the trial design. Sure. Checkpoints are very good in melanoma. A hot tumor. What we mean by hot is that the tumor is growing. It has a lot of mutations, which means it has a lot of neoantigens. Typically, the immune system recognizes that tumor and generates neoantigen reactive T cells. Those T cells infiltrate the tumor. These are typically CD8 T cells. One of the last defenses that tumors seem to have is to then express PD-L1, which shuts down T cells because it activates the negative signaling of PD-1 on the T cells. One of the strongest predictors of response to a checkpoint is the presence of preexisting intratumoral CD8 T cells. That's the so-called hot tumor. As I say, checkpoint inhibitors work well in that population. It doesn't really make a lot of sense for us to go into that population because they have the T cells. So the logic of what we do is that it is designed to work in cold tumors. So that's kind of the whole point, is that it is intended to work in this population. And most people have cold tumors. If you look at the big solid tumors that kill most people, it's lung, breast, colon, ovary, and prostate. Those are the big five. Checkpoint inhibitors work in a subset of lung and a small slice of breast, triple negative breast cancer. They're not approved outside of the rare MSI patient. They're not approved in prostate or ovarian or colorectal. So this is the area you want to be in where we have huge unmet need. And if we're right, then obviously this is transformative therapy. So big unmet need, big population, i.e., easy to run trials. And even better, several big companies, including Roche, Merck, and Bristol Myers Squibb, ran trials with their checkpoints in microsatellite stable colorectal cancer and showed there was no effect. So we could now come in with our immunotherapy, which is vaccine plus checkpoint. And we didn't have to do a checkpoint-only control arm. So we could do simple studies, two arms, chemo, chemo plus immunotherapy. FDA was accepting of that. And that was why this study makes all the sense in the world because it's a test of the biological thesis. It's a big population. And it has enormous signaling value, we believe, meaning that if we show activity and efficacy in a metastatic cold tumor like colorectal cancer, you're opening the door to essentially every other solid tumor. It should work just about everywhere. So massive opportunity. That's why we're in this indication. OK. Great. The FDA was OK with not having the checkpoint and the control because they really don't work. There's been multiple examples. You have a slide in. We have a slide. Yeah. Roche did a big study called the MODUL study. It was several hundred patients randomized. These were patients with newly diagnosed microsatellite stable metastatic colorectal cancer, i.e., exactly the same population we're in, front line. They showed that the addition of atezolizumab or Tecentriq, which is their PD-L1 antibody, as maintenance therapy in a randomized study added nothing to PFS or OS. There was really nothing, not even a small effect. There was just nothing. So FDA agreed that that was the control group, and therefore we did not need to repeat it. Physicians actually said, we think it's unethical to repeat it because the question has been asked and answered. We know that checkpoints have toxicity. When we know there is toxicity associated with the drug and we know that there's no benefit, it is no longer ethical to put patients onto that study. So physicians actually were pretty adamant they didn't want a control group because they wouldn't enroll patients to it. Happily, the agency agreed with that perspective. OK. Great. I'll point everyone to the company web page. There's slides with the phase II results. It's 41 slides, so a lot of data there. So we probably won't go through that. But what was most encouraging from the phase II results in your view going forward? Then you're going to have an update in 3Q. What can we expect there, and what's going to be the key lookout? These were preliminary data that we showed. Let's be clear. We will have mature data, which obviously will be much more robust in third quarter of this year. In the preliminary data, obviously safety was fine. No issues on the safety side. Let's be clear about that. Importantly, we were able to make product for every patient that was randomized to receive the vaccine. Manufacturing is not an issue here. The study is a randomized study, one-to-one randomization. We take patients with newly diagnosed metastatic microsatellite stable colorectal cancer. Induction therapy is pretty standard. It is FOLFOX, which is 5-FU, and oxaliplatin plus bevacizumab or Avastin. That's standard. A few young patients get FOLFOX plus irinotecan. Irinotecan is often held for second line, but you can combine it in the front line, and it has a slightly longer overall survival. But it's more toxic. So young people sometimes choose that regimen. So we allowed that as well, and that was a stratification factor. So the patients are randomized at the beginning of the study as they are diagnosed and starting their induction chemotherapy. Those randomized to receive vaccine, we make the product. And then as they stop induction, which is usually around month five, and that's necessitated usually by the onset of neuropathy from oxaliplatin, then you stop the oxaliplatin and irinotecan if they're on that. And you just continue with maintenance 5-FU and bevacizumab. So that's the standard of care. And the median progression-free survival is about 11 months for that regimen. So that's what we expected in the control arm. And in the test arm, we add on the immunotherapy as they enter the maintenance phase. And the question, obviously, for us was around the primary efficacy endpoint. PFS is the standard. We were a little nervous about it because in our phase II, we'd seen a couple of lung lesions get bigger in the short term on our therapy and then cavitate and shrink down. And that's a phenomenon typically referred to as pseudoprogression. And what it means, it seems, is that T cells are entering lesions and expanding. And so the lesions do physically get larger. So on a CT scan, the measurements get bigger. And using standard rules of tumor assessment, the so-called RECIST rules, that's called progressive disease. So they're mislabeled as doing badly when, in fact, they're doing well. No one's done this before, really. So we had no idea how big an issue that might be. But we were nervous about it. So we elected to make PFS our first secondary endpoint. In its stead as the primary endpoint, we said, let's use this direct metric of tumor destruction, which is ctDNA drop. We'd seen that in our phase I/II study in advanced disease. There's a lot of evidence that patients who do well, their ctDNA comes down over time. Whereas patients doing badly, their circulating tumor DNA rises. In other words, circulating tumor DNA acts as some kind of proxy for tumor mass. On therapy, going down is good. Going up is bad. That's the sort of simple message. So we needed to set some rules for ctDNA response for our trial. Now, we knew about the vaccine arm because we'd treated patients previously. We'd seen that drops were good. So we set the rule as a 30% drop at any time point. We were making an assumption for the control arm because there were no data about controls. What does chemo do to ctDNA? We know that it generally will make it go down. But for how long and to what extent, these were really unknown. We made an assumption that in the control arm, just getting chemo, ctDNA would drop during the induction phase. But by the time they moved to maintenance, we assumed that the ctDNA would be flat or even maybe starting to rise in the control arm. So we set ctDNA drop by at least 30% at any time point for just one time point as our definition of response. What we learned is that was a mistake because, in fact, the control arm, the ctDNA kept going down during the first three or four weeks of the maintenance phase. There was some kind of a delayed or extended effect of the induction chemotherapy. And so we saw several controls. The ctDNA carried on going down for just that first three or four weeks. Then it went up. But they were classified as responders by our definition. So what we saw in this preliminary data is that the response rate was broadly similar between the controls and the active arm. We think over time that will wash out. But the way we defined it, it didn't work. So there was an endpoint fail. And obviously, we own that. But more importantly, is like, is the drug actually working? And if you look at ctDNA over time, you see clear signals. And I won't sort of draw figures with words. But if you look at the data, you can clearly see that the ctDNA in the control arms tends to rise over time. Whereas in the vaccine arm, for many patients, it's either flat or going down over time. So ctDNA actually is doing what we wanted it to do but just not meeting the rules the way we set them up. The upside surprise from all of our data was around PFS. So we're not seeing pseudoprogression. So that's good. And that may be because all the lesions, really, that we're measuring are primarily liver lesions. And pseudoprogression doesn't seem to happen in the liver. So PFS does seem to be a robust endpoint. And to our surprise, and this obviously was a good surprise, we're already seeing separation of curves, even in this immature data set where most patients are still being censored, meaning they haven't yet achieved a progression event. We're seeing separation of the curves that starts at around six months, exactly what you'd expect from the preclinical data because we'd seen T cells forming. It takes about a month or so. So if they start immunotherapy at month five, you would expect that the T cells are killing tumors. And good things start to happen by month six. And that's exactly what we see on our curves. The PFS curve starts to separate at month six. Now, as I say, the data are heavily censored. But you can still interrogate the data and say, well, quantify that PFS curve separation. And right now, the separation has a ratio as 0.82. So 0.8, as you know, is sort of mild efficacy, as it were, modest efficacy. But these are very immature data. So the question we asked was, is there a population that has mature data where we could maybe get a glimpse of what the future can hold for us? And actually, there is, of course. There are patients who do worse. And they, therefore, have progression events faster. And they give you more mature data more quickly. So the question is, how can I define, in a very fair and even-handed way, a group of bad actors who are going to do poorly? And can I look at the randomized data in that subset? And you can. And the way we did it was to use baseline ctDNA. So again, circulating tumor DNA is a proxy for tumor mass. Having more of it is bad. This is a very strong prognostic factor. We took our data. We obtained baseline ctDNA on a large number of patients, not yet all. We had data from many. You could take the control group. There was a midpoint. We just split around that midpoint. Anybody above the median, we called high risk. The figure is actually 2%. That's the actual statistic for ctDNA. Below 2% was low risk. If you then aggregated the patients and just split them according to that 2% cutoff, lo and behold, nearly all of the progression events were in the high risk group. It's an incredibly efficient sorting mechanism. It's a very strong prognostic factor. In that high risk group, you can now see, okay, the data look reasonably mature there. What does the efficacy look like if I now take the high risk group and split it into vaccine and control? And now we've got a separation of the curves that was much more dramatic with a hazard ratio of 0.52, which is a very strong effect. And we moved the threshold around just to do some sensitivity analyses. And the effects were pretty consistently with a hazard ratio better than 0.6. So the exact numbers don't really matter. The point is, in more mature data, the treatment effects strengthened significantly. In the low risk group, we just have no data. And that's an important point. Some people misinterpret this. It doesn't mean that the product doesn't work in low risk. It just means we have no data yet in low risk because the events accrue more slowly. So wait. By Q3 of this year, we should have enough events in the low risk group to actually get mature data across the overall population, which is the expected outcome. Now, of course, the right question is, is there something special about this high risk group that means you're seeing data that's unrepresentative? And the truth there, we don't, of course, know yet. Truth will emerge. But one supposition actually is that the high risk group will do least well on vaccine because most people believe that vaccines work best in low volume disease, less immunologically complex disease, where you have time for the vaccine to do its thing. That's the zeitgeist that's out there. That's why Moderna and Merck are focused on adjuvant disease. Well, if that's all true, this high risk population is the worst group to look at. And yet, we seem to see a strong effect there. So if that's all true, the treatment effect in the low risk group should be stronger. If anything, we have informative censoring that works against the vaccine, not in favor of it, against it. So that's good. And that suggests that we just need to be patient. And we potentially have, then, in Q3, a game-changing result with strong effects observed in this overall population of metastatic colorectal cancer. That's a big deal for the field. And obviously, we hope that that will then enable our share price to respond accordingly so that we can raise capital and move into phase III. So of course, our goal now is to then get to the agency and have the end of phase II conversation, agree on a phase III endpoint, and then prep ourselves for phase III study start next year. OK. I'll ask a question. It may not be an answer. But the 3Q readout, it's not event-driven, right? You can decide. Are you inclined to wait till later in the quarter to get more mature data or early in the quarter just to get the data earlier and then go to the agency? And you expect to meet with the agency this year. Is that something that can happen? Yeah. Time is our friend here in terms of data. Time is our enemy in terms of cash and runway. So we have to balance those two, as you always do in biotech, or at least we've always had to. So we're not going to cut on July 1, right? We're not going to cut on the first day of Q3 yet. I will tell you that. We need to wait longer because I think more mature data is helpful to us. So waiting. Yes, we then do expect to go to the agency because an end of phase II meeting is usually on a type B 60-day clock. Well, it looks like we're at time. And I didn't get to ask you about CORAL or SLATE and the other programs. But I'll point everybody to the earnings call.