Scientist Stories: Alex Marson, Inventor pioneering immune cell engineering

So our next speaker you've already heard briefly from, but it's Alex Marson. He's the director of the Gladstone UCSF Institute of Genomic Immunology and a professor at UCSF. After he got his PhD and did his medical training in Boston, he started his lab as a Sandler Faculty Fellow at UCSF in 2013. In addition to these positions, he's a scientific director of human health at the Innovative Genomics Institute, a member of the Parker Institute for Cancer Immunotherapy, and an investigator at the Chan Zuckerberg Biohub. And his research is focused on adapting CRISPR genome editing techniques to human immune cells in order to understand the genetic programs controlling immune cell function and to manipulate T cells to generate cell-based therapies. And he'll talk to us today about reprogramming T cells with CRISPR. So go ahead, Alex.

It's great to hear the talks this morning, and again, thank you all for joining. I'm excited to have a chance to tell you a little bit about what we're doing in my lab. To try to take advantage of gene engineering technologies to not only understand specific DNA sequences that control critical functions of cells in the human immune system, but also more and more to think about how can we look— take those lessons and actually start reprogramming human immune cells to have new functions that transform them from cells into cell therapies. For a range of human diseases. So before I launch into it, I— technical troubles today— before I launch into it, I want to show you my financial disclosures, which some of which are relevant for things I'll be talking about today. And we are Okay, so we are broadly inspired by successes in the world of cell therapy. These are cell therapies that are actually FDA approved and clinically being delivered that are genetically modified human immune cells. And where this is the farthest along is now in the, in the field of CAR T cells, or chimeric antigen receptor T cells, for certain types of cancer. The current CAR-T cells, patients' immune cells, are taken out of the— out of circulation, outside of the body, and then genetically modified. They're currently modified with viral vectors that insert genetic material into non-targeted sites in the genome of those T cells that have been taken out of the body, and that inserts a gene for an artificial receptor, a chimeric antigen receptor, that directs T cells against targets on certain types of cancer. And then these genetically modified T cells can be reinfused into a patient. And the results for certain cancers have been really remarkable. There are now a number of different responses for different blood cancers. We'd like to really be able to now make these much more powerful to create more durable and lasting responses for patients. With, with blood cancers, but also to start expanding these responses to diseases that have been harder. That includes solid tumors, where CAR-T cells still have largely been ineffective, and to other classes of human disease that we're starting to think about how do we engineer cell therapies for. And as we think about this, we're really inspired to think about the power of cell therapies much more broadly, by thinking that we can move beyond these viral vectors and take advantages of really powerful new gene engineering technologies that now allow us to pick very specific targeted sites in the genome and actually start rewriting DNA sequences to fix individual mutations in the case perhaps of rare disease, but also to think about what we can do to really broadly reprogram cells by removing genes from the genome that are limiting the function of cells, or by adding new DNA sequences to install a new gene, new gene program at a defined site in the genome of human immune cells. So over the past number of years, we have developed a technology, a platform that allows us to very efficiently rewrite DNA sequences in a non-viral way at targeted sites in human immune cells. This started a number of years ago through a collaboration with Jennifer Doudna, who is at Berkeley and also an affiliate member of the Genomic Immunology Institute here. We started using recombinant Cas9 protein loaded in vitro with a guide RNA to make a ribonucleoprotein, or RNP, and found that that RNP can be electroporated into human T cells to get very efficient gene disruption. We also found that we could co-electroporate a nonviral piece of DNA that has homology arms on either side of the cut site to get very high efficiency knock-in sequences, actually pasting in a defined sequence at that defined site in the genome. We have now made a number of technological advances in this, and most recently I have a bioRxiv paper up with some new tricks that really extend the efficiency of this. And this is now a platform that's broadly applicable to a number of different cells in the human immune system. That can be cultured. So we are applying this a number of different ways to march towards clinical applications. One of the most powerful things that we'd like to be able to paste in with a defined— at a defined site is our antigen receptors so that we can direct what a T cell will recognize. This can either be done by pasting in the sequence for a CAR, as I described, or for— by actually directly replacing the cell's own T cell receptor, which is the endogenous mechanism by which it recognizes an antigen. As we've gotten higher and higher efficiency for this nonviral CRISPR-based reprogramming of cells, we are now in a position where we have an active march towards clinical applications at different stages of development for a number of different diseases. I just want to give you a flavor of these. These are broad partnerships in the Gladstone UCSF Institute of Genomic Immunology, working closely with other groups, and with the Living Therapeutics Initiative at UCSF. So we— there's now an FDA-approved BCMA CAR to treat multiple myeloma. We are now going to make this in a nonviral way that takes advantage of actually putting the CAR under the endogenous gene regulation of the TRAC locus, which Justin Ikem has shown previously is a major benefit for CAR function. So with our latest efficiencies, we're now able to get upwards of 40, even higher, percent knock-in of CARs with this nonviral process in a way that we're now— is compatible with clinical manufacturing. So we're working with the Multiple Myeloma Translational Initiative here at UCSF and with the Living Therapeutics to advance this towards a clinical trial for this nonviral CAR T cell. We're also working on TCRs with Hideo Okado at CIRM and others. We're working on a non-viral version of a TCR that recognizes an antigen found in certain types of pediatric brain cancer. And we're actively extending this beyond cancer. So we're working with the Weill Neurohub and a number of different groups to try to also make genetically engineered TCR cells that recognize viral peptides. And we're starting with sequences in the JC virus, which causes a horrible neuroinfectious disease in certain immunocompromised individuals. And we think that this is the tip of the iceberg of reprogramming the antigen specificity of different T cells to go after cancer, infectious disease, and we're also thinking broadly about autoimmune disease. But as we think about these, we know that antigen specificity isn't the whole answer of what we want to program into these cell therapies. And so let me— the rest of my talk will really focus on How do we improve cell therapies, especially starting with solid tumors? We know that even if we can make an antigen-specific T cell with a CAR or TCR, if that cell goes into a solid tumor, it will face a number of challenges that can prevent it from being effective in clearing the tumor. Now I've shown here just schematized a few of these. Once it gets into the tumor microenvironment, it'll encounter the challenge of chronic antigen exposure, checkpoints, which can cause exhaustion and other modes of dysfunction. There are suppressive cells that it'll encounter and suppressive soluble factors. So all of these need to be overcome. And so as we think about what idea— designing an ideal cell therapy, all of a sudden we start to think about, well, what's the target profile of a cell therapy that we'd like to have to be more effective? And so we've thought a lot about some of the features that we'd like an ideal cell therapy to have. We'd like it to proliferate when it finds its target. We'd like it to be resistant to the suppression. We'd like it to be durable. We'd like it to be able to sense low antigens. And we'd like it to really ultimately be effective at clearing tumors. And so as we think about these desirable target profile features, how do we go about designing these? And sometimes we can get hypotheses from the literature. But there's still so much we want to fundamentally discover about the genetic programs that control human immune cells that we can start engineering around, that we have been inspired by to think about the power of— sorry— to think about the power of forward genetics. Now, forward genetics goes back at least as far as Drosophila studies, but we're not interested in applying this to a model organism. We'd like to use the unbiased power of forward genetics to discover genetic modifications in a primary human cell type that is actually the basis for cellular immunotherapy. And so the idea of forward genetics is that you have to take advantage of mutations that are introduced into a population, and then you pull out the individual cells in this case that have a phenotype of interest. So how do we, how do we introduce a large number of mutations into cells? Well, CRISPR has given us the basis to do this in extraordinarily powerful ways. And I'm going to give you a little flavor in the rest of the talk about how we're using a combination of CRISPR-based functional genetic screens to pull out phenotypes of interest. We're using genome-wide knockout screens with CRISPR. We're also moving beyond that into things that we can add into T cells. So we're doing pooled knock-ins to see if— are there DNA cassettes that we can put in and see which ones drive phenotypes of interest. And what I'm really going to focus on today is a new technology, unpublished work, that now allows us to do genome-wide gain-of-function screens using CRISPR activation systems. And all these technologies allow us to create a population of genetically modified cells that are effectively barcoded by different DNA sequences. And once we have a a population of genetically modified cells, we can actually pull out the cells that have a phenotype of interest. We can look for the modifications that make the cells more abundant in a central situation, but we can also look for markers of interest, and we can, we can actually also even couple this with single-cell technologies, which you've heard about earlier today. So just to give you a flavor of where this is going, and where, and where we already are, we started this by seeing could we use the power of genome-wide CRISPR screens, which have been widely used in cell lines and animal models, to actually do this in primary human T cells. And early on, we had a big problem of delivering CRISPR effectively through viral vectors into human T cells. So we developed— Eric Shifrut, who was a postdoc in my lab and now has his own group in Israel, And Giulia Carnevale, who now is a member of the Genomic Immunology Institute, who you'll hear from later, they developed a system to do a genome-wide screen where you put libraries of lentiviral vectors into cells that carry the guide RNAs. And then because we had trouble getting Cas9 in with virus, they actually electroporated the Cas9 protein to do very efficient gene editing. And then when you had this population of cells, you could screen for any phenotype of interest. And they've systematically dissected gene modifications that make cells proliferate more when they get restimulated. And as we've— as they've continued to push on this, this has actually led to discovery of a new gene target that fundamentally advances our ability to do cell therapies. So Julia will tell you more about this today, but in collaboration with Justin Ikewun's lab, we've found a gene target through these screens, ROSA2, that can be knocked out that dramatically enhances the ability of CAR-T cells to clear tumors from preclinical models. And Julia will tell you about this more later. So I think that this really has, has got me excited about the power to do high-throughput screens to discover gene modifications that will make cell therapies more powerful. And now we're really thinking about, can we go beyond what we can take out of a gene, out of a cell? What can we add into a cell? That will enhance its function. And this started with work led by Theo Roth, who was an MD-PhD student in my lab, who is now at Stanford working with Anshu Sadpathy, another member of the genomic immunology community. Theo, as he got better and better at introducing gene programs into targeted sites with CRISPR, thought about, well, maybe you could introduce the antigen receptor, either a TCR or CAR, plus an extra gene, where that extra gene may have a beneficial function for promoting a cell state of interest in cells, in cell therapies. And rather than guess, again, we'd like to be able to actually test this in an unbiased way, to make many different DNA constructs that might have beneficial effects, and knock them all into a cell of interest, and then effectively race them against each other in vitro or in vivo, or analyze them all with single-cell sequencing to see which of those programs is most productive in moving cells to a state that is compatible with effective cell therapies. So just to very briefly introduce what Theo did in this work, Theo started making a library of candidate constructs, either genes that could be overexpressed in the TCR alpha locus and made constitutively expressed in T cells, or even thinking about synthetic receptors that don't actually exist in nature. So we started thinking about these so-called switch receptors, where you, you make an artificial fusion where you take what would otherwise be an inhibitory receptor, keep the extracellular domain that recognizes this otherwise inhibitory ligand, and start fusing it to an activating intracellular domain. And this switch receptor then should respond to something that would normally suppress T cells in the microenvironment, but instead would now make them even more strongly activated. So Theo designed a candidate library of 36 different members, including controls, natural genes, and synthetic receptors, and he knocked them into the TCR locus, along with a defined antigen specificity. So each cell got a different knock-in, and we could track which knock-in went into which cell in this population based on a barcode that was put in along with them. So just to give you a flavor of how powerful this was, we were actually— Theo was actually able to do this in an in vivo study to see which of these knock-in constructs made the cells accumulate more in a tumor microenvironment. So these are human immune cells knocked in, each one got the same T cell receptor, but a different gene added along with it. This is the T cell receptor we used, recognized an NYESO cancer antigen, and then we put into an immune deficient mouse an NYESO-positive tumor. And then Theo compared these— this input population of cells to the cells that actually infiltrated and accumulated in the tumor, and looked at which barcodes led to preferential accumulation in the tumor, and started finding genes, including these artificial receptors that manipulate TGF-beta sequences, that actually lead to preferential accumulation. So now these TGF-beta switch receptors that couple an active— this inhibitory TGF-beta receptor to activating intracellular domains, seemed to lead to tumor microenvironment accumulation. And we tested that one at a time and actually took this TGF-beta 4-1BB construct that Theo had generated, put it in along with an antigen receptor into a mouse, and the results were really striking. Compared to the control mice that don't get this 4-1BB receptor with the TCR, the TGF-beta 4-1BB switch receptor leads to complete responses in a number of these mice at this dose, which was not otherwise achievable at this dose. So this has got us really interested in thinking about how do we now scale this up? How do we discover even more rapidly what can be overexpressed in the cell to make it more effective at doing what we want to do in a number of different disease contexts? So we'd like to be able to discover systematically every gene that could be overexpressed, either on its own or put into a synthetic circuit, which you'll hear more about from Cole Roybal, or engineered into synthetic receptors, these types of fusion receptors. And there is actually a very powerful technology to do genome-wide screens that would allow for gain of function. But this has really only been done in cell lines. And we have had enormous problems in using this technology in primary cells. So this is CRISPR activation. This is a technology that uses a dead Cas9. So this is now— Cas9 is not cutting and doing gene editing. Cas9 is actually being used to recruit transcriptional activators, in this case a number of different transcriptional activators that are fused to the dCas9 itself and to the guide RNA to be— to turn on genes of interest. There's another version of this called CRISPR interference, where the dead Cas9 actually brings a transcriptional functional repressor. So we'd like to be able to use these modalities to turn genes on and off. But the challenge has been that unlike Cas9, where we could deliver a transient pulse of Cas9 protein and get permanent genetic modifications, these technologies need to be sustainably introduced into T cells, which means that in really, in order to use them in practice, we need to be able to get stable transduction with viral vectors. Has been a big challenge because these are big constructs, and the efficiency has been really low. So with our previous methods, our transduction at getting big constructs into primary cells, you can see, has been quite low. But Zach Steinhardt and Ralph Schmidt, who are two really amazing postdocs, were undeterred, and they set out to see, could they actually adapt this technology to deliver it efficiently into primary human cells? And what Ralph did initially was really see, could he just go through systematically and start testing a number of different variables to see what could be optimized to get higher efficiency? And it turned out that it was a number of different things, but taken together, all of a sudden, when Zach and Ralph started achieving very high viral transduction, that was really a step function from what was previously achievable. And with that, we could actually start introducing this CRISPR activating system into primary human cells. So this just gives you a flavor of putting in this SAM system that was developed at the Broad into human T cells. Now all of a sudden, compared to the control cells with this control guide RNA, which don't express these receptors measured by flow cytometry, you can actually get robust induction across multiple different guide RNAs using this CRISPR activation system. And with that in hand, Zach and Ralph went— were off to the races and said, let's not do this one at a time, but let's see, can we use CRISPR activation to introduce— to activate a different gene from every gene in the genome in a different cell? And so they were able to now use a combination of this CRISPR activation virus along with a separate virus that puts in the guide RNAs and introduced them both into cells, and they were able to do genome-wide libraries to introduce gain-of-function perturbations in a population of primary human cells. Once you have that, they could, in, in theory, they could test any different phenotype of interest to see the gain of function. But we wanted to ask a question, a complex question of what are the genes in the genome that can be tuned to change how cells will respond to stimulation? We know that as cells respond to stimulation, CD4s and CD8s, depending on how they're stimulated and the state of the cells, they will turn on different cytokines, which will have a big effect on their effector functions. So Zach and Ralph looked at the CD4 cells to look at what are all the genes that could— would affect the extent to which IL-2 is produced. And in CD8s, what are all these genes that would affect the extent to which interferon gamma is produced? So now they're actually looking at endogenous cytokine production. These are not reporters. They're actually doing intracellular staining and sorting for cells that have different levels of these cytokines produced after perturbations. And I have to say, these results were really— floored me with how beautiful and clean the results are with these genetic screens with CRISPR activation. So what I'm just going to walk you through are the results of genome-wide screens for all the gain-of-functions that modulate levels of IL-2 or interferon gamma production. So the first thing I'll show you is that these were done in two separate human blood donors. So even though these were from different individual humans, The results were remarkably reproducible across these genome-wide experiments. You can see this linear relationship shown for both of these screens. And the results make sense. So among the top hits, if you activate IL-2, you see you get more IL-2 production. If you get more— if you activate interferon gamma directly, that's the top hit for interferon gamma high cells. But we also discovered many other regulators that when you overexpress them lead to more cytokine production or less. And these were reproducible not only between human blood donors, but were strongly reproducible among multiple different guide RNAs, suggesting that these are really real effects. So you can see here IL-2 is a strong hit in the IL-2, interferon gamma, and then a number of hits that when they're overexpressed, as expected, CD28 or Vav strongly affected cytokine production, as well as overexpression of negative regulators that would dampen these cytokine production. Zach and Ralph didn't stop there. They actually did the same screens, but instead of with gain of function, they actually did it also with CRISPR interference. And again, the results were highly reproducible across human blood donors. And now the hits come up, IL-2 over interference lowers IL-2 as expected, and interferon gamma lowers interferon gamma. And we start to really see complementary maps of genes that can be upregulated or downregulated to tune how the cytokine— what cytokines will be produced as the B cells are responding. Now, one of the things that became clear very quickly is that CRISPR activation and CRISPR interference are really providing useful and complementary maps. So just to show one insight that, that Jack and Ralph have had is that if you look at collectively at all the hits from the CRISPR interference screens, The hits, as you'd expect, tend to be well expressed in the cells that we're screening in. Now, that's because you will only find a phenotype with interference if that cell is— if that gene is already on. Now, in contrast, that is not true for CRISPR activation. Many of the hits, the range of expression is much bigger, and many of the hits are actually expressed at very, very low levels in the cells that we're screening. So we're actually able to detect gene function independent of the context where these genes are normally activated, normally used. And that allows us to start putting together these things to see, to look at whole pathways where we can see what comes up as a hit with CRISPR interference or CRISPR activation. So if we look at the NF-kappaB hits across the interferon gamma screen, for example, we see that a number of hits only come up in CRISPR activation, and others only come up in CRISPR interference. So a number of intracellular components, for example, they seem to be critical and non-redundant and expressed well at basal levels in these cells. And these come up as CRISPR interference hits. Getting rid of these blocks this ability of cells to make interferon gamma. In contrast, there's a whole number of cell surface receptors, including many of these TNF family super receptor— superfamily receptors, that only come up in CRISPR activation. And we think that many of these may not be expressed at sufficient levels under these cells, but we can actually pick up that these are actually sufficient to turn on interferon gamma, independent of the context where they may otherwise be used. And this is not just an arbitrary list. These actually include many things that point us to potential immunotherapy targets, which you'll recognize here, including OX40 and 4-1BB and CD27, CD47, which are being actively developed for a number of different types of immunotherapies. So now we're discovering receptors that can be activated to drive cells to states of interest. So what are those states? In the remaining few minutes, I just want to talk to you about how this technology for CRISPR activation can now be coupled to unbiased single-cell measurements of cell state to really tell us not only what is regulating an individual cytokine of interest, but what are the— what's the overall state that we're promoting by putting in this gene of interest into a T cell. So what Zach and Ralph have done is now really adapt this CRISPR activation perturbation system in primary human T cells to make it compatible with 10x droplet-based single-cell sequencing, so that we, for each one of these perturbed cells, we can not only look at the overall transcriptome, but we can look at, we can measure which guide RNA is actually in that cell, so we can ask which, when a gene of interest is overexpressed, what's the effect on overall cell state? Zach and Ralph picked out 70 different genes, including controls and hits from their screens, things that generally promoted cytokine production as well as things that dampen cytokine production when they were overexpressed. And they did a CRISPR-A perturb-seq now in primary human T cells to look at with a single-cell transcriptome readout, what is the effect of overexpressing each of these targets or the controls. Now, you've seen UMAPs. So this is just to give you a measurement of the cells that were actually captured from these single-cell experiments. This was done in 2 different human blood donors again. And we used Jimmy— this was a close collaboration with Jimmy Yi's lab. And we used a method developed by Jimmy to deconvolute which donor each cell was from based on its genetic variants found in those cells. You can see that we were able to assign individual cells to one donor or the other in most cases. And we were very pleased that in this experiment, donor differences did not drive the cell states. You can actually see a very high degree of mixing of these cell states independent of blood donor. So that's great. Now, in contrast, we were able in a bulk population of stimulated T cells to very well resolve CD4 versus CD8 cells. So you can actually see that we're able to separate CD4 and CD8 cells and ask individually each of these perturbations, what do they have, what effects do they have on both CD4 and CD8 cells? And then we're able to start looking into this, into the cells in this UMAP and say, what cytokines are they producing? So just to give you a flavor of this, if we look at this heat map of which cells are producing interferon gamma, You see that they tend to cluster in a few different places here and here. IL-2 was captured less well at the single-cell level, but you can still see IL-2 cells that cluster together. And then with this map of all the cell states in the population, what we really want to know is which perturbation is driving cells selectively to different states that can be measured on this map. So this is density maps of looking at different populations of guide RNA perturbed cells. So the control cells are evenly distributed across the cell state map, as you'd expect. And then if we look at the cells with all the perturbations, they start to cluster in different states. And we can actually look at one perturbation at a time in this pooled experiment and start to really make sense of it. So overexpressing MAP4K1, which was a negative regulator, dampened activation and moved cells to this state that corresponded with overall low levels of cytokine production. And then we were actually able to see individual regulators. So here were a couple of canonical regulators overexpressing GATA3, for example, moves the cells to a strong Th2 state. Tbet moves cells to a Th1 state or interferon gamma-producing CD8 state. And then we could also find this for less well-characterized regulators. Do it is overall single-cell measurements of cell states. And for the aficionados, what Zach was really able to do was then to be able to make an unbiased— a relatively unbiased map of naming the different cell states that come up in this map and systematically map these different clusters to which guide RNAs are moving them to these different cell states. So among these different 15 clusters, we were able to name many of them, and they become recognizable. For example, proliferative cells, or Th1 cells, Th2 cells, and ask which guide RNAs are found in each of these clusters. And you can actually see that here, that there are certain guide RNAs that are moving cells to define cell states. So this really all of a sudden goes all the way from finding at the genome-wide level regulators of cytokine production all the way to be able to pair individual perturbations up with very detailed maps of cell state. And as we march forward, we're now thinking about how do we move this and really define the states that will best characterize the most effective cell therapies for cancer and for other diseases and map the gain-of-function and loss-of-function perturbations to directly program cells into the states that will make them best able to treat patients for a number of different diseases. So I will stop here. Just to summarize broadly, what I've tried to argue is that we are really now have a suite of forward genetic technologies that allow us to perturb genetic elements throughout the genome—gain of function, loss of function, knock-in. Everything I talked about today was coding sequences, but we're now very actively moving this beyond into non-coding sequences, This is applicable across different cell types. This is going to tell us and is already telling us fundamental ways that the genome programs the behavior of human cells. It's pointing to new drug targets, but it's also giving us the basic map of constituent elements that we can use to start programming cell therapies. And everything we learn can— has a very direct path to actually being the basis for the roadmap to design next-generation cell therapies. For cancer, autoimmunity, infectious disease, and others in collaboration with the genomic immunology community that you're going to continue to hear about from today. So with that, let me thank the people who've done this work, thank a big network of collaborators and funders, and thank you again for joining us today.