Founder Stories: Harlan Robins, Surprisingly Small Effective Size of the Human TCR Repertoire

So let me introduce the last speaker, and thank you all for staying for the last speaker. That's very nice. Holland Robbins, who comes from the Hutchinson Cancer Center, and he's going to tell us about direct assessment of TCR diversity and frequency.

Thanks. All right. I know a lot of people have to leave at 5:30, so I will make sure that I'm done by then. So we've heard a lot about cancer systems biology in various talks today, and essentially there's— we've probably gone through the whole set of different types of measurements that are involved in the system to the extent that we're able to to make these measurements at this point. However, the immune system has been sort of left out of this in a lot of ways. And part of the reason is that the— if you look at, say, genotypes, or let's start with microarray data. If you're looking at microarray data, you might have 50,000 different genes that you're looking at, or 30,000 different genes on on your array. However, each person has those same 30,000 different genes on the array. The immune system does something very different, which is that the T cells and B cells, which make up your adaptive immune system, they are predicted to come in 5 × 10^18— T cells alone, 5 × 10^18 different forms. Forms, and that's just an absolutely massive number that really can't be measured, okay, by any technology that we have. So, and not only that, there's supposed to be such a huge variation between people that actually studying the difference between someone's T-cell repertoire in one person versus another would seem relatively fruitless. However, I'm going to show you that that's not exactly true, and what I'd like to do is I'm not— I'm not actually going to be able to tell you much about how this applies to cancer right now, but what I'd like to do— and if you remember Annalisa's cloud picture with the bubbles, and they had metabolomics and proteomics and genotypes and everything else, I'd like to add an extra bubble, and that bubble would be your adaptive immune repertoire, okay? Now, but in order to do that, I first need to show you that that's a relevant measurement to do, and in order to do that, I need to show you that between any given people, there actually is commonality in our adaptive immune system. So that measuring one person and measuring another is— measuring one can tell you something about another, and therefore tell you something about the correlated with different forms of cancer. OK. We already know that the adaptive immune system plays a huge role in cancer. It does it in three different ways that I can think of. First is that— first is a relatively indirect role in controlling viruses, which have a strong rate of causing cancer. We know this. If you have a kidney transplant and you're on immunosuppressive drugs, your rate of getting certain types of skin cancer goes up 100-fold. So it's clear in many, many examples that the adaptive immune system is directly suppressing sets of viruses that cause cancer. That's number 1. Number 2 is that it's fairly well known in cancer immunology groups that cancer is actually controlled. Cancer expresses aberrant proteins. And those are seen by and attacked by our adaptive immune system. And so there's a direct control of cancer by our adaptive immune system. And then the third is, in sort of a practical way, is how do we use that information, and it's being used in the idea of both cancer vaccines and immunotherapy, which is that you actually harvest these T cells that are known to attack tumors, and you grow them up in a certain way and put them back in a person. And in certain cases, they've shown really strong efficacy in actually destroying the tumor. We're involved in a couple of different projects like that, but they're not at the point where I can tell you about them. So I'll just hopefully— what I'd like to do right now is convince you that T cell receptor repertoires belong as another bubble and should be added in the future to when you do cancer systems biology. This should be another set of measurements that has strong potential to give us viable information. In particular, one of the pieces of information that we're looking at is the idea that we can use the immune repertoire as a biomarker to determine either what stage of cancer someone has or whether or not someone actually has cancer. OK, so just a quick rundown on what I'm going to be studying, telling you about T cells in particular. So there's two types. Your adaptive immune system has the difficult job of— Its job is to recognize and rid the body of an unknown set of foreign pathogens that you've never been exposed to before. So its strategy to do this is that it creates a set of millions of pseudo-random receptors, puts them on the surface of different cells, either T cells or B cells, and when it sees the appropriate antigen, So basically, these receptors are supposed to cover what you'd call shape space. So any shape that comes in, one of these is going to be sufficiently good a binder to it that it can recognize and stick to its target, its pathogen. Then it starts off a cascade where you have three processes. One is that they clonally expand, the particular cell that has the right receptor for that pathogen. Colonially expands and creates an army. There's a change in phenotype where for T cells they become killers and they kill the actual infected cells. And the third thing they do is they form immunological memory so that you have a standing army waiting in case you get that virus again, okay, or pathogen. And the adaptive immune system has two parts. It has B cells and T cells. The B cells, our job is to recognize particles that are basically floating in your system. And T cells have the job of recognizing when a cell has been infected, an intracellular pathogen. And that happens by the cell has a mechanism for chopping up fractions of every bit of proteins produced and presents them on the surface with these HLA markers, the HLA molecules. And T cells, so just, does this thing work? All right, so this is the HLA, and this is a little piece of that protein that was in the cell that's presented. And T cells come and bind to this thing, okay? And then that's where the cascade starts, and some of them turn into killer T cells and kill the cell down here that's infected. And this little part up here with the colors is called the CDR3 region. And that's the region that's the variable region that has all the different shapes, OK? And so how are these things made? Well, they're a heterodimer. They basically form a Y-shaped receptor with an alpha chain and a beta chain. This is the most common type of T cell. And basically, Unlike every other cell in your body, if I took a liver cell and a kidney cell from your body and compared the DNA, they would be almost identical. For the adaptive immune cells, if I look everywhere in your genome, they would be the same except for one region. And that region is the T cell receptor region, where the DNA itself actually rearranges. And what happens is you cut out one— it's called a segment— of one of one of 2 different Ds, and one of 13 different Js, and you stick them next to each other, so you get some combinatorial diversity by picking one of these different sets, okay? But it's not that much diversity. It's only on the order of 500 or 600. However, at the junctions where you put between the V and the D, you delete back a little bit of the V, some of the D, and some of the J, and then there's another protein called TDT which randomly inserts nucleotides in the middle. So you're really making a pseudo-random receptor. Now that's massive diversity, because each new insertion you put in, you get this exponential blowup in diversity. You get a power of 4 each time, okay. Most of the diversity is in the betas, and the reason is the betas rearrange first. Then when you have a functional beta, for each beta, it's predicted that about 20 different alphas are made. So you make on the order of between 5 and 10 million of these guys and then add another factor of 20 for each. Okay? All right. So the idea is how do you— what do you do with this? Well, you kind of— we need to sort of pick out a needle in a haystack. What we want to know is what is this sequence down here in the VDJ region? Well, this is only on average, you know, somewhere on the order of— 13 to 15 nucleotides long. Well, your— the human genome is 3 billion nucleotides long, so we have to pick out these 13 out of the 3 billion. And not only that, we need to know what V we had and what J we had. So we designed a strategy where we anchored— used PCR to PCR up this particular region. So we had to design specific primers to bind and then sequence through this particular region, leaving enough J and enough V to be able to recognize which one we had. And we wanted to do this in a very tight region so that we could use the super high-throughput sequencing technology and look at lots of them quantitatively at the same time. Okay, so let me just give you a scale of the problem that we started with. Well, the problem is that the dogma in the field was— and this is based on a theoretical prediction, which is that of these different betas, that there was about 5 × 10^11 possible betas, okay, that could be created, which led to 5 × 10^18 total predicted T cells in the population. And this was done on some model, and this has sort of been the standard in the field for a very long time. Well, you only have, in a normal human, you have about 10^12 total T cells in your blood. So if you're taking 10^12 out of a pool of 10^18 and you look at two different people, they should— less than 1 in a million should be the same between those two people. So you should have almost no overlap. Two people should have no overlapping sequences, okay? So is this really true? So if this is true, then this wouldn't be very useful in terms of systems— cancer systems biology. I wouldn't be talking about this if it was true, so I'm going to tell you it's not true, obviously. So what we did is we just took 7 different people. The reason 7 is because we were using the Illumina sequencing technology, and it has 8 lanes, and one of them is a control, leaving 7. I mean, we carefully designed this experiment knowing power calculations. Anyways, so 2 were HLA-identical sisters. We took their mother, and we took 4 unrelated people of very different ethnicities, which I'll show you in a second. The reason we had the two HLA identical sisters is that we wanted to just rule out what role HLA was playing to make sure that this wasn't— we don't know if it scaled up where HLA was playing a role that would change our results by a factor of 10 or 100, or if it just didn't make much of a difference. And what we did is then we looked at cell populations that were both naive and memory, as defined by surface markers. Okay, so here's the actual chart of who these people were. Here's their age distribution. There's a mix of males and females. The 3— these 2, 1 and 3 are the sisters. 2 is their mother. And the rest have extremely diverse ethnicities. We have Japanese, Indian, and 2 African. And a huge difference in HLA types. And in our— so this is just what the raw data looks like. If you look at the— in the naive compartment, we were getting about between 5 and 6.5, 5 and 7 million reads, T cell receptor reads per sample. We now get closer to 20. The technology improves every 6 months. It more than doubles. And this was done 6 months ago. And of those, they fell into this number of unique sequences, okay? So, you know, as I was saying, we predicted that there's maybe 5 million different T-cell types in your blood, and we do that with a set of formulas that I'm not going to present today. But— and so we're getting, you know, maybe a tenth or something like that of the total sequences in your blood. But remember, we're also getting all the big ones, right? Anything that— the rare ones are the runs we're missing. OK? And that's going to be important, because if you're going to share T cells between any two people, it's going to be the common ones that are going to be relevant for us. OK? All right, so let's just look at the distribution. So first question is, if you— I plotted the V versus J just as a two-dimensional histogram. And we ordered this for— just according to its use. And what we realized is that they're not used evenly at all. And this has error bars on it, even though they're really hard to see. And the reason they're hard to see is that there's not much variation. So the interesting thing is that the V-J combinations that aren't used very much are not used very much in anybody. And the ones that are used a lot are used a lot in everybody. And the other interesting thing is that if you go over to memory, you actually get some significant differences. Which really means that, you know, this V here, this J is J26, this blue one. It doesn't— it's pretty rare in the naive compartment, but it blows up in the memory. And what that means is that there's probably some set of common— some common pathogen or set of common pathogens that have a preference for things with J26, okay, because they're blowing up much more relatively speaking. All right. So our sort of breakthrough on this came when we organized our distribution of sequences by number of insertions. Now remember I told you that the diversity was mainly driven by this number of insertions. The reason is, is that your combinatorial diversity isn't that high because you had 54 times 13 times 2, that's all you get, whereas if you, and your— the number of— the diversity you get from deletions isn't that high either because you can only delete the nucleotide that's there. Whereas when you put a new one in, an insertion, you get 4 different choices. So if you delete 5 nucleotides, there's 1 way to do that. If you insert 5 nucleotides, there's 4 to the 5th power ways to do that. So this is really driving the diversity. And the models are based— that model where I said we had 10 to the 18th total T cells, or 10 to the 11th beta chains. It was based on having 12 insertions in the T cell receptor beta chain. Well, what we just noticed, just no one really had the data to see if that was right. So we just looked at it, and here's a cumulative distribution plot. What this tells us is that 90% of the sequences, T cells in any given person's blood, has less than 12 insertions. So that's just a way overestimate. In fact, There's sequences way down here, even with 2 insertions, 10% or 15% of people— of your sequences in your blood have 2 or less insertions. So the diversity of sequences with 2 or less insertions is just way, way smaller than something way up here. So the point is, is that even though our chances of sharing— any 2 people sharing sequences with 12 insertions is almost zero, it's pretty good when you start looking at sequences with 1 or 2 insertions. And remember, we had, you know, we have 10 to the 12th cells in our blood, so we have a lot of these sequences even though it's 10%. That's 10% of 10 to the 12th is a very, very large number of sequences. And memory tracks about the same. Okay, so we just explicitly computed how many different ways you could make sequences with 0, 1, 2, et cetera insertions. And you find with low numbers of insertions, you get, let's say, 1 million, or with 0, you get 200,000. So here you're talking 10 to the 6th, which is very different than 5 times 10 to the 11th, which is what was predicted. OK, so what do we expect the overlap to be? Well, since we don't have that much time, and you probably— I don't know, probably not a huge math audience, if I had to guess. I'll just go quickly through this. We just estimated what this distribution and this distribution together would give us for a predicted number of overlaps between any two people. And it gave us, if we use the original method of with the 5 × 10^11 beta sequences, we get, you know, an estimate of about 20 or something like that. Now if we use our, what the actual data, we're getting like 15,000. That's 15,000 identically shared T-cell receptor sequence beta chains between any two people. So now, as I said, we have the data, so we can see if this is right. And so here it is. So here's— this axis is the naive. So we go up here, and we see, well, this is the distribution. And the average is almost exactly that. It's about— it's a little over 14,000. Our sort of hypothesis seems to be proving right here, which is that we really do have a large overlap between any set of people, any two people. And here are the HLA identical sisters and their mother. Sorry, other way around. These are the sisters, and this is with one sister and the mother and the other sister and the mother. And they're a little bit higher, so it's playing some role, but it's not— we were thinking it could be way up here, but it's not. So that was, you know, obviously we need to do more people to get some hard and fast rule on whether or not HLA is playing a role, but this at least tells us they're not off scale, is what we were worried about. Okay, so then let's ask. Well, this suggests that these overlapping sequences have different numbers of insertions, so we can just check that. So the red is the overlapping sequences, and the bluish purpley thing is the non-overlapping sequences, and we just plot a number of insertions, and we find a significant, obvious difference between the overlapping guys and the non-overlapping guys. So it's exactly these— so, so far this model seems to be, you know, working out exactly as it should. Okay, so one thing that we've— so I should stop and say for a second that this really means that that T cells now become— T cell receptor sequences become viable, you know, sort of bubbles to put for doing systems biology, and in particular for doing diagnostics, which is that there's a large number of shared T cell receptors so that if a subset correlates with a particular disease, it can be used throughout the population as a biomarker. And I'll show you an example of that in one second. But right before we got that, this is something that we can also learn specific things about development. And so basically, when your T cell develops, it rearranges, as I said, that VDJ combination and expresses a receptor. Well, it does this in the thymus. The reason it's called a T cell is because of thymus. And B cell is bone marrow. And so since it develops in the thymus, the first thing you have to do when you create this T cell receptor is it goes through a set of positive and negative selection in the thymus. The positive selection is to make sure that it has some affinity to bind to the type of things that are presented by cells in your system, because if it doesn't bind to anything even related to what could possibly be presented, it's just useless. And then more importantly, I think, is the negative selection, which is that if it binds super strongly to something that's normally expressed in your body, like if it binds to an HLA-presented part of actin or something, you're going to be in big trouble because you're going to have an autoimmune disease. Your immune system is going to attack your own tissue. So part of that— However, one thing that I should say stupidly, it took me about 6 months to figure out, because we just were throwing this data out, which is that sometimes when a T cell rearranges, it— most of the time when a T cell rearranges, since it has variable numbers of insertions and deletions, it's actually out of frame. So 2/3 of the time, it's just out of frame. And then another fraction of the time, it has a stop code on it. So what happens when that happens? Well, oftentimes, the other allele, which we've been hearing a lot about today, the other allele rearranges. So in some subset of cells, sometimes the cell just dies, but in some fraction of cells, we have two alleles both rearranged, one totally dysfunctional and one that's the productive one that's the viable T cell. Well, we could compare the difference between the out-of-frame guys and the in-frame guys, the productive ones, and this should tell us a lot about the process of selection inside the thymus. So for instance, if we look here, this is the in-frame. These are the productive guys that showed up that got out of the thymus into the blood. These are the guys who we only see as remnants sitting as the other allele on T cells. This is the allele that's not functional. And we can see that there's some significant differences between their VJ usage, just their VJ usage. And we haven't talked about a bunch of, you know, the other properties. But you can see that some Vs and Js which were not used very much didn't, you know, clearly got negatively selected because they were made at pretty high levels and didn't get out of the thymus. Okay. This is to be explored since I just thought of this like 3 days ago, stupidly. I've been sitting with that data for like 6 months. Okay, so then just one quick comment on germline evolution. So one thing that we noticed here, what I've been telling you is that the sequences that we all share, the T cell receptor sequences that we all share, are these sequences with very few insertions, okay? So the question would be— and they're playing— these are playing a clearly prominent role. Is this prominent role strictly by development or they actually have some extra function? Well, there's a hint that they might be doing more than just being developed sort of randomly, and that is that if you take a mouse and you knock out TDT, this this enzyme that inserts nucleotides, you create a mouse that has no insertions at all. However, that strain of mouse, it has no immunological deficiencies that people can see. I mean, they've— there's a couple studies where they make these mice with no TDT, they confirm that they don't have any insertions, and then they hit these things with every disease they can think of, and they don't do any worse than normal mouse who have the TDT. So one of two things, either we just don't need a diverse repertoire, it's maybe there for random cases where some new disease shows up that we've never seen before, or, and these particular sequences, these particular low insertion beta chains, T-cell receptor sequences are playing a particularly prominent role, and in fact, it's possible that they are, unlike the high insertion guys, and that is because if you don't have many insertions, your entire receptor is made up of things, of nucleotides that are encoded by the germline. So you actually— so natural selection can actually act on the set of receptors that have no insertions. So it's possible that these are truly special receptors. I'm not sure if that argument didn't come across quite clear. What I'm trying to say is that Great. Why don't we save that. If you have questions about it, we'll go into it in a minute. But what I'm trying to— what I'd like you to get out of this for now is that we're trying to separate the question, is there something special about the particular receptors that have no insertions, these ones that I'm talking about here? And so we need to get a hint about that. We have to look at— this gets into a standard problem that you find in epidemiology all the time, which is that you can have confounding effects, right, which is that sequences that show up a lot also have higher clonotype counts. They're— the number of sequences in your blood are also higher because they're made more often. So you— and sequences that are made more often are more likely to respond to a pathogen because there's just more of them. And so you can't just look at things like how often do these low insertion sequences get into memory. So you have to do some tricky arguments. And the tricky arguments is asking, okay, we already know that they're more likely to get into memory, but once they get into memory, we don't know anything about whether they're more likely to expand super high or not. That should be determined by whether or not these things are— have a particular affinity to the pathogen at hand. And so that's a question we had to ask, which is that in general, do you have a correlation between getting— between how big you expand in memory and whether you got into memory in the first place? And these charts have— you can see there's no correlation here. So we actually did this. We just mapped these sequences between naive and memory, and we find there's no correlation. So that implies that there's nothing special about these sequences getting into memory. However, these sequences expand like crazy in memory. So the point is that I'm confusing the hell out of everybody, I'm sure. But my point is that these sequences, there is some evidence that these sequences really do play a somewhat special role, and that we can see that, and that they're particularly designed to hit particular viruses. And maybe it would help if I give you some examples. So an example is Epstein-Barr virus. So the nice thing about Epstein-Barr virus is that everybody has it. And when I say everybody has it, my colleague who's a clinical immunologist who does transplants said he's seen, you know, maybe 1,000 patients who have been typed and every single one who was over the age of 10 had EBV. So of our 7 people we checked, and everybody actually does have EBV, as expected. And so what we did is we just combed the literature and said, what T cell receptor sequences have been shown to respond to EBV? And the way you do that experimentally is someone takes a particular HLA molecule in what's called a tetramer and then sticks on there a particular epitope from the virus itself and has a way of pulling out whatever sticks to it. And so what they do is they actually find the set of T cells that stick to the particular, you know, HLA antigen combination that are expressed by EBV. So these are functional T cells that in those particular people are hitting the EBV virus. And so some of them have been shown to show up what's called publicly. In other words, they found the same sequences in multiple people. So we wanted to see, are we finding the same sequences? Where are they? And what is their role? So in HLA-B8, which is extremely rare— it's a rare HLA type— there's this one sequence, this one EBV sequence, combined with a particular HLA-B8 that seems to show up everywhere. So do we see it in our sequences? Well, first, this is the part that other people haven't been able to do so far, which is, does it show up in the people who don't have B8, okay? Which was 6 out of our 7 donors. Well, we find in the memory compartment, zero. We don't find any copies of this thing. And in the naives, we only found one copy at a low level. So it's never responding to this, there's no, particular response with this T cell. Now we go down to the person that did have B8, and we found one in this person's memory compartment, but we found 42,000 copies of it. It's 1% of this person's entire T cell repertoire, is this one particular T cell that responds to EBV. And then, okay, so that was sort of a— and when I say could we use this as a biomarker, Well, in this case, I would say absolutely yes. I could have told you with just looking at this person's repertoire that this person had EBV, okay? They had this particular sequence at 42,000 copies. They clearly had EBV. I could also have told you they were B8, that they had HLA-B8 without HLA typing them. But now we go down over here and we— there's more studies with HLA-A2 because 50% of Caucasians have A2. It's by far the most common HLA type. So they did the same thing with HLA-A2 with a different epitope, also from EBV. And we looked. And we had 5 donors who— 2 of our 5 donors had HLA-A2. The 5 without, we didn't find any copies of any of the— oh, sorry. We used a study that had found 10 public T-cell receptors. So these are exact amino acid sequences. We found zero of these in any of the five donors, but then we went down to the two donors who do have HLA-A2, and we found five. So I mean, I haven't done the statistics yet, but I mean, it's kind of overwhelming here that there's a signal. All right, so that kind of connects it back to this is viable as a biomarker. So then we went to do a quick study on diabetes. We're actually funded to do a much bigger study now, but this was just our pilot study. We just did 3 cases and 3 matched controls, and we also had 7 non-matched controls. And we did the same thing, we sorted into naive and memory, and then we sequenced both naive and memory for all 13 donors. And just to start out with, we noticed that the average copy number of the CD4 effector cells were higher in the people with than the people without. And in terms of comparing sequences, we're still in the process of doing that, so I don't want to present the data yet, but we're clearly seeing sequences that are in the type 1 diabetes population that are in their memory compartment that aren't in the memory compartment of the people without type 1 diabetes. So it looks like this is a method to find viable biomarkers. Obviously, we need more than our 3 cases to confirm anything interesting. And we're doing that. So I can— next time, I'll invite myself back, Arnie. Next time, I'll tell you about that. So just to sum up real quick is that we now have this ability to sequence millions of T-cell receptor sequences per sample. We learned that the effective size of the repertoire is far smaller than predicted, and that this implies a huge overlap and also public T cells. And just, we have a bunch of projects underway. I told you about some of the cancer projects with, we're doing a couple different immunotherapy projects plus some projects related to looking for biomarkers. And then just one other thing that might or might not be of interest to anybody if they happen to be interested in T cells or B cells, we actually just spun off a company so that because just to open it up and we sort of created this technology so if anybody wants to use it, we just made it really easy to use. And also the way we did it this way is for my own research, it's way cheaper to do it through the company than through the Hutch, which was killing us. But anyways, if you're interested, we have a website. You can just go down there and contact us and we can do this for you. And then just to acknowledge a few people, The labs of Hootie Warren, who's a clinical immunologist, and Chris Carlson, who's an expert in sequencing. It was kind of a collaboration between all three of us. A lot of the work was done by Santosh in my lab. And then we've had some— Jerry Nepom is a big guy in type 1 diabetes. And Stan Riddell is a clinical immunologist in our group. And yeah, that's it. Questions? Yeah. [INAUDIBLE] Oh, can we find the D? Yeah, so the D ranges— so basically, the reason I didn't talk a lot about the D is that you have the D gets always mapped into the NDN region, it's called, where the insertions and the D— because a significant fraction of the D is usually deleted. So the Ds get all squished up. Yes, so we have all our data divided into a set of data where we can see the Ds, and we know which one it is. —and then a set of data where there's so little D left that we can't tell which of the two Ds it is. [INAUDIBLE] About over 3/4, we can identify which D it is. So it's mostly insertions on the start of the D region? Well, the D regions are— there's two D regions. One's 12 and one's 16 bases long. So even if you delete— they all have some deletions. So you— How many did you know we could use? Two. Yeah, for humans. So it's pretty easy to— usually it's pretty reasonable to identify. They really have to be down to 2 nucleotides or something before we have a hard time, or 3 nucleotides before we have a hard time.

So what do you think— one of the reasons why we're all so impressed with the immune system is its combinatorics were enormous because it has to handle anything it has never seen before. And then the first time you take a good look at it, not taking advantage of its combinatorics at all. So what do you think is going on? Do you think that it's just, it is responding to an environment which is boring in Seattle?

No, I mean, this is, so when I said, so our ethnicities, although we actually obviously drew the blood of everybody in Seattle, when I said African, these are people who had been in America for, you know, really grew up in Africa, and our Indian person really grew up in India. So it's not necessarily— but what do I think? I think that probably that it's sort of there for a rainy day, which is that population-wide we do end up having a reasonable amount of diversity. There's a much larger set of commonality. And then there's— so I think basically what we call our adaptive immune system has a part that's basically innate. And that the adaptive part isn't that often used. It's used in cases where new things come in that we haven't been exposed to. I think it's probably way more efficient. I mean, we get the flu every year, and we've been doing that for many years.

But what about age? What about what? The ages. I mean, were the ages varied a lot too?

I guess you had a mother and two daughters, right? Yeah, so we didn't really get the group. We're doing a study right now. On aging. And so basically, the literature says that until you're about age 50, you're— it's not— between about age 20 and 50, there is no real noticeable difference between diversity in your immune system. At 50, it changes a little bit, but mostly changes between relative ratios of memory to naive. And then by age 70 or 80— not till you're actually probably close to 70 75, 80 does the actual number of T cells start dropping. So the reason that, for instance, 70-year-olds have a harder time responding to vaccines is supposed to be because they have much less naive T cells and many more memory T cells that aren't going to respond to a new— one more question here. Two more. Naive T cells are after they've been through the thymus or before? After. After. [INAUDIBLE] So yeah, so that's a really interesting project. We are actually doing that in mouse this week. Reason mouse— it's actually not— I mean, I would have thought when I started this it would be impossible to get human thymus, but it's actually not. Because whenever they do— if you get open heart surgery and your thymus happens to be in the way, they just take it out. So we actually have a set of samples, human thymic samples that we have on the agenda to do, and that'll be done within the next few months or 6 months at least. So, but it'll be a really interesting question. But then you get into some other issues, which are you have to separate out which cells you think are pre- and post-selection, and there are dogma in the field of which are which, but, you know, I think what we're learning is that the dogma isn't necessarily quite right, so we're going to get stuck stuck at some point. But it's going to be a really cool experiment. You have one more question here? Yeah. So if I was a parasite, I guess I would want to have antigens that would only be recognized by this really rare recombination event, or the blue ones, the nearest row. Yeah, well, I mean, I think— so it's probably true that new T cells, new viruses, or new pathogens probably start out— it's probably easier to get into the immune system if you're completely masked from are from the repertoire that we have, or is common. But probably— and viruses evolve so fast that they probably could stay— avoid our immune system in a lot of cases. But it's not in a viral particle, for instance, interest to kill its host. So I think there's this weird interplay between how the human immune system evolves and how viruses evolve. There's a lot of cases in the innate immune system where this has been really well studied. But in the adaptive immune system, it hasn't really been well studied yet. But I would guess that there's an evolution in the germline that as we get new viruses, you start to create new T cell receptors that take over the population. And the viruses actually benefit from this because they can learn to live within the host as opposed to killing their host. But that's just pure guesswork. It's an interesting field, and there's lots of people in the innate immune system that are studying. Harmeet Malik, if you are interested, he's studying at the Hutch. He's got some really cool data on the innate immune system.

OK, let me wind up by thanking all of the speakers and the audience for a very, very good day. Thank you.