In this talk, Dr. Crystal Mackall reflects on her position as a hypothesis‑driven biologist presenting at a big‑data symposium and uses the opportunity to illustrate how engineered cell therapies—both T‑cell receptor (TCR)‑engineered cells and chimeric antigen receptor (CAR) T cells—are developed, administered, and evaluated. She outlines the manufacturing workflow (apheresis, viral engineering, infusion) and contrasts the two modalities: CAR T cells achieve rapid, massive expansion but often contract quickly, especially with CD28 co‑stimulatory domains, whereas engineered TCR cells show slower, more durable tumor shrinkage that correlates with long‑term persistence. Persistence is linked to a phenotypic shift from effector memory to stem‑cell memory (CD8) and central memory (CD4) subsets. Using high‑throughput endogenous TCR sequencing as a barcode, her lab quantifies clonotype diversity, showing that persisting cells retain the diversity of the original apheresis product and are distributed across multiple memory pools, indicating both regenerative capacity and effector potential. She then describes how CAR T cells frequently enter an exhaustion trajectory, identified through high‑dimensional CyTOF and X‑shift force‑directed analysis, revealing distinct marker patterns (e.g., IL‑7R, CD45RA vs. CD39, 2B4). From these data she derives exhaustion and stemness scores and proposes engineering strategies to make CAR T cells more exhaustion‑resistant. Throughout, she emphasizes the need to combine deep, systems‑level data with hypothesis‑driven immunology rather than relying on “big data” alone, coining the term “deep data” to describe interactive, high‑resolution analyses that can uncover biologically meaningful mechanisms.
Chapters
I'm going to be honest. When I was invited to speak at the Big Data Symposium, my feeling was, oh, I don't really belong there. I don't do big data. I'm a biologist. I do hypothesis-driven research. I think of big data as being taking thousands of individuals or thousands of samples and using algorithms to find maybe patterns that the human mind couldn't come up with on their own, sort of artificial intelligence. So I'm going to share with you the way we try to advance the field and the a couple vignettes regarding adoptive cell therapy. And you'll see that it starts out hypothesis-driven, but maybe we do venture into the field, the area of big data. So as Sean alluded to, there are a variety of types of cancer cell therapies. The one that I work on is engineered cell therapy. And basically, as you can see in the very simplified figure there, this involves taking a patient who has cancer, harvesting lymphocytes from that patient through a process called apheresis. You get somewhere on the order of 10 to the 9th T cells. And then you put those cells into a vessel, at which time you engineer them, usually with a virus. And so of course every cell is not being engineered. It's a random process. And now the cells come back out with a receptor of some kind that enhances the ability of the cell to find the tumor and therefore work its magic. You then infuse a dose of these cells, which again is only a subset of what was engineered. And the two main types of engineered T-cell therapy are those that use the native T-cell receptor. Sometimes we tinker with it to give it a higher affinity. But of course, the T-cell receptor is what, you know, we have evolved for many years to be able to recognize intracellular pathogens, viral pathogens. And so the TCR recognizes peptides presented in the context of MHC, and it requires co-stimulation in order to get sufficient enhancement. And when you administer engineered T-cell receptors, you're really dependent on the host to give that co-stimulation. And there's a novel approach to using engineered cells, and that's chimeric antigen receptors. They're called chimeras because you use the signaling domain of the T-cell receptor with the antibody or antigen-binding domain of an antibody, and so a chimera. One of the points about T CARs is that because we've embedded the co-stimulatory domains in there, on a cell-per-cell basis, you get more bang for your buck. We administer many less CAR T cells to patients than engineered TCRs. And this is an example of what these CAR T cells can do just in terms of sheer numbers after infusion into a patient. This is a 16-year-old who had refractory leukemia. And on the bottom, you can see the CD19-expressing cells. The red are the leukemia cells, the green are the normal B cells. And then we're using flow cytometry to measure the CAR T cells circulating in the blood of this patient. And you can see that, of course, they aren't there at baseline. These are non-natural receptors. But after infusion, within really only 10 days, you see this dramatic expansion of this product such that it is comprising, in this case, 79% of the entire T cell pool of this patient. We find these cells not only in the bloodstream and the bone marrow and the CSF, really wherever we look. So we really believe that these things have taken over, at that point in time, a large component of the T cell repertoire. And then as fast as they come, we often see them disappear. And this disappearance has been an issue that my lab has been focused on like a laser since I arrived at Stanford a little over 2 years ago. This doesn't happen all the time, and it varies with CAR-T. Vectors. This one happens to incorporate a CD28 co-stimulatory domain. And they typically disappear when you use that. Now we think this is a problem. Because we think if you're going to cure cancer, many of the cancers that we need to use these things for, you need a longer hit than just 28 days. And this is an example of the type of response you can see when engineered T cells don't disappear. When they persist. And here we're using an engineered T cell receptor targeting a molecule called New Yorkiso-1. This is expressed in this tumor that this young adult had, circled there in red, pressing up against his heart, resistant to all other— certainly to all chemotherapies. You see also he had a metastatic lesion in the lung. And we infused these engineered TCR-expressing T cells. And you see at 1 month, it doesn't look like that much has happened. As opposed to the leukemia, which was gone at 1 month. And at 2 months, you start to see a little more shrinkage. And at 4 and a half months, a little more. And this patient had continued shrinkage out to about 9 months. And if you look at our cohort of patients— and this is where I get a little shy around the big data. We're typically talking about small numbers of patients. This is a cohort of only 12 patients. But half of those patients responded. And the question is, well, what can you learn from 12 patients? My goodness, all the heterogeneity. In the personalization? Can we really learn anything? Well, you know, we look for clues. And for me, one of the major clues here was that when they respond, they respond over time. So this is showing you now the pattern of shrinkage. You can see within the first month or two, the responders are already starting to decline. But they don't hit their peak decline until several months out. And again, So this is a signal to me that if the T-cell receptor or T-cell engineered therapy is going to work, it needs to be able to persist. So let's start looking at the cells in this patient population and talk about how well did they persist. And you can see this is a very different pattern than what we saw with the chimeric antigen receptor T-cells. We have persistence now for CD8 cells out to 28 months. And the question is, what makes these cells so different? And so we have ways of characterizing cells into different subsets. And what you can see is that the subset that was infused largely would be categorized as effector memory. That's the MP, the manufactured product. But very soon, 1 month after, 2 months after, 3 months after, the persistent cells become a completely different looking type of a cell, so-called stem cell memory cell. There are the pie charts. Look at the little orange sliver in the manufactured product. It's only minorly represented, and yet after infusion, the little orange sliver becomes the dominant pool. And so for CD8 cells, we see these stem cell memory cells, and for CD4 cells, it's largely the central memory cells shown in pink. Now the question is, where did these cells come from? Are they a few clones that are able to expand, or are there large numbers of cells that are able to contribute to this pool? Pool. So now we're using high-throughput sequencing, not of the engineered TCR because those are all identical, but of the native endogenous TCR that every single one of these T cells expresses. And we use it as a barcode to be able to identify how much diversity there is in this expanding and long-term persisting pool. And so using high-throughput sequencing, next-generation sequencing, to sequence now the endogenous TCRs, we want to ask how clonally diverse is the repertoire. And we know that effector cells, TEMs, are naturally more clonal and less diverse than these other subsets—stem cell memory, naive, central memory. And remarkably, the persistent cells in this patient that are measured here are remarkably diverse. These aren't a few cells that made it through. Many of the cells that were infused, they show the same level of repertoire diversity as the baseline apheresis. So we are really getting a very efficient persistence. Now we begin to ask questions about the long-lived clones. These are the ones we're the most interested in. So we identify those clone or clonotypes, those T-cell receptor beta sequences that we find at 6 months, and we start to ask questions about where were those clonotypes back at the time of apheresis or in the manufactured product. And we saw some surprises here. One of the surprises is that we didn't find them in the naive T-cell pool. We thought that naive cells might be the most predisposed for persistence, but in fact they were less likely than a random clone to be in the naive pool. Yet they were enhanced in a variety of other pools, including stem cell memory and central memory. The same in the manufactured product, and we see these same relationships after infusion until we get to month 6. When you can see the regenerative pool, the stem cell memory pool, which is simply giving us the regeneration, starts to decline. And this is now associated with the decline in the numbers. But the other point to make is that these regenerative pools are not the only pools that the clonotypes are present in. They're always also present in the effector pools. And so what it begins to give us a picture is that if you want a long-term T-cell pool that's capable of fighting cancer, it has to do two things. It has to be able to regenerate itself, and it has to be able to give rise to an effector. Now, here's the ancestry.com experiment, I call it, where we take the 11 most plentiful clones, the clonotypes, and we now are sorting these subsets and then doing the TCR-BV sequencing. And we're looking at that clonotype and asking, where was it distributed throughout all of these time periods? And every time we see that we see them in the regenerative pools, but we also see them in the effector pools. These cells are very promiscuous, the clonotypes. They're not just sticking in one pool, they're able to diversify. So let's just briefly get back to the CAR T cells. What's wrong with our CAR T cells? Well, they're not regenerating because they're going down this pathway that we call T cell exhaustion. And this is using a novel approach to look at complex relationships between a variety of parameters, in this case cell surface markers on CyTOF, that was developed by Gary Nolan's lab. And it's called X-shift analysis, an algorithm that allows you to generate these force-directed plots that show the change as T cells move down the pathway from start to exhaustion. And so blue are the starting populations. We have an in vitro model for exhaustion. Red are the ending populations.. It occurs rapidly over 4 days. And what we're now beginning to see is using these high-dimensional analyses, we can look at different trajectories of the way these T cells evolve. And Zena Good really developed this approach of these force-directed plots to look at T cell trajectories. And she'll be presenting a poster here today on this. And just a final slide here. Now these trajectories kind of look like mishmash. But an immunologist wants to see its old friends in there. And our old friends are things like IL-7 receptor and CD45RA. And yes, indeed, they show up at the end of the trajectory that's associated with health and a non-exhausting T cell, whereas these problematic markers, CD39 and 2b4, show up on the other trajectory. And so what we can come up with now is an exhaustion versus a non-exhaustion score or a stemness score. And of course, What we're going to be doing going forward is looking at these scores in various subsets for the CAR T-cell populations and engineering these to become more resistant to exhaustion to see whether we can alter the trajectory. So thoughts from a biologist regarding deep data. I'm not a big data person. Progress in cancer immunotherapy relies on our ability to understand the biology of the immune responses. And what the— I think one of the important One of the important points is that in immunology, we already have existing deep knowledge and understanding. So we have to work from hypothesis-driven questions. But at the same time, we also want to use the state-of-the-art technologies that allow us to ask unbiased questions. So the magic is when we line these cells— line these approaches up to be able to really advance the field. And so what I think we're using is not big data necessarily, but what we call deep data, interactive data, systems biology to interrogate these data sets to really understand the biology. Thank you very much.
