Variant to Function (V2F) Symposium: Keynote - Robert Plenge (2025)

Uh so it's my my great pleasure to uh introduce the second keynote speaker of today uh Robert Plenge who is the executive vice president and chief research officer at Bristol Myers Squib.

He came to Bristol my script from Seljian that was acquired by BMS and before that he was head of translational medicine and vice president at Merc. uh

he began his journey in academic medicine training for his MD and PhD on the west coast at UCSF and UCSD and then he came to the east coast to do a fellowship in clinical rheumatology and

he liked the climate so much that he ended up staying here I guess uh and this is also of course uh both clinical rheumatology and genetics of course and not just at the at B at the Bighamin

Women's Hospital but at the broad institute as well so I think your mentor here was David Alill is that correct I couldn't quite remember that piece and he also has uh brought brought this unique voice to our community through

his blog that I really recommend that you guys read called Plench Gen. If you

hadn't read it, if you haven't read it already, it's about uh impact in genetics and medicine uh and really has some interesting thoughts about how we should approach drug development uh through uh those types of approaches.

And I know that this is also what his presentation will be about. And a fun fact which for those of you who know me, I love fun facts. Uh I actually was had the good fortune of interacting with Robert when he was at the broad and he

when he transitioned to Merc gave me an opportunity to get my first industry collaboration and grant that actually launched my lab. It was my second grant

ever I think around 2012 or 13 and it was a real gamecher for me and I don't think I ever told you that but uh it was really great interacting with your team at Merc and uh the funding was fantastic. We actually did some really

fantastic. We actually did some really good work with with that money. So, it's

just an enormous pleasure to welcome you back. I've been excited to hear your

back. I've been excited to hear your talk and I love the title. So, can't

wait to see what you're going to tell us.

I I I did not know that. So, so thank you and I'm glad we were able to help.

So, um first of all, thanks for inviting me and uh for all of you who are still here, thanks for uh hanging in there. I

know it's been a been a long day. Um, so

I'm I'm going to start in the same way that uh Ben actually started, which is to thank a number of people who actually contributed to the talk. So Joe

Moranville, who's part of Bristol Myers Squib I've worked with for a number of years. He contributed a lot of the the

years. He contributed a lot of the the data that I'll show you today. We had a great pleasure to work with a company called Octant. Uh Shri Kosuri uh was a

called Octant. Uh Shri Kosuri uh was a posttock with uh with George Church. So

kind of starting in the beginning and the end of the day and they did a lot of the work on on saturation mutagenesis that I'll show you at the end of the talk. Uh and then lastly, I'm going to

talk. Uh and then lastly, I'm going to give you a number of examples that have come out of our Bristol Myers pipeline.

Um I have worked on the minority of them until recently. So there are a lot of

until recently. So there are a lot of people who have worked on these programs uh for a decade or more. And I'll be able to share some of the examples uh with you today. Um so this is a talk I haven't given before. So you're going to

have to tell me if this concept of a billion dollar bet actually works. I'm

going to start in the way that Molina actually started the day as well uh which is to frame the concept of some of the challenges that we fra face in drug discovery and development which is we

take we spent all of this time doing what what what VJ talked about in MSI 2 and trying to understand the genetics and the functions of a particular protein and we do it in animal models.

We make molecules. We take them into human clinical trials and the vast majority of those things just fail and they fail because the drugs don't don't work. Somehow we pick the wrong target.

work. Somehow we pick the wrong target.

So what you can actually see on the left hand side of this graph is is that simple pie chart of why most drugs fail.

Most drugs fail because the drugs don't work despite many many years of hard work leading up to making the medicine and testing it in in humans. And I think on the right hand side, this sort of

frames at a high high level uh one of the reasons that drugs don't work because we just don't have enough insight into human biology to make all of the right choices that are required

to identify a target, make a molecule, take it into the right patient population, test it, and to see this PKPD relationship, and ultimately get a drug approved. So hopefully as we can

drug approved. So hopefully as we can focus more and more not just on human model systems but causal human model systems, we'll begin to improve uh the

rates of efficacy in clinical trials. Um

so I think we've heard about this a bunch already today and I think I'm sure you're all very familiar with this is that human genetics is a very powerful tool uh to help predict the success of a

drug target in clinical trials. So Matt

Nelson um had a paper a number of years ago. Eric Minl expanded on this a couple

ago. Eric Minl expanded on this a couple of years ago. And basically the take-home here is if a drug if a drug is against a validated genetic target, it's at least a two to threefold increased

likelihood of probability of success in clinical development. Um we've done some

clinical development. Um we've done some work on this. Uh Joe Moranville and his team have done some work on this uh within within BMS. And one thing in particular is we've added on this kind of proteomr

approach which if you can actually have proteomic data link with genetic data that's also a good predictor of of drug success. But as I go through the talk

success. But as I go through the talk one of the things that I really want to emphasize is yes genetics is a powerful predictor of success for drugs. But

there are properties of human genetics which can actually be extended into other areas that we call causal human biology that also can be powerful predictors of success of of clinical

trials. And so we want to take those

trials. And so we want to take those principles in human genetics, apply them in other areas. So Khloe talked about uh some of the work in profiling immune cells. You know, if we can identify

cells. You know, if we can identify those pathogenic immune cells, for examples that are driving diseases like lupus and figure out how to selectively deplete those cells, that's causal human

biology. Potentially, that is another

biology. Potentially, that is another way to actually come up with new drug targets. Okay. So, we've tried to

targets. Okay. So, we've tried to distill some of these very complex topics for drug discovery and development into five principles. And

behind each of these principles, there's a tremendous amount of information. And

one of the nice things about going last in the day is you can kind of pull on the examples over the course of the day to kind of really emphasize the these concepts. So when I made the move from

concepts. So when I made the move from academics to industry, it was really because of that first principle. I

thought human genetics could change the way that we do do drug discovery development. If only we worked on the

development. If only we worked on the right targets and use human genetics to pick those targets, everything else would fall into place. I think the biggest education that I've had since joining industry is I realized that yes,

picking the right target is really important but there are a lot of other things that come after it uh that are really important as well and again VJ gave a great example of MSI 2 show that

how much time goes into understanding any individual talk any individual target you don't want to just understand in this case a genetic association but what are the details of the of the

biology of the particular target how do you want to modulate it. What are you going to measure once you get into the clinic? That's all captured by that

clinic? That's all captured by that single concept of of causal human biology. But it's a really really

biology. But it's a really really important starting point for drug discovery and development. So the second step then is to match a therapeutic modality to a molecular mechanism of

action. And Ben just gave you a great

action. And Ben just gave you a great example of how to actually do that in the form of of crisper cast 9. If you

understand a mutation that is causing the disease and it's expressed in the liver, you can figure out how to give that drug into the liver and correct it in the liver and make the mutant base go

back to a wild type base. That's a

perfect example of matching modality to mechanism. We also heard about

mechanism. We also heard about manipulating BCL11A and there it's actually not introducing something new but it's taking something out preventing the binding site of a

transcription factor that allows the repression of fetal hemoglobin. So by

removing that site that's matching molecular mechanism. You're basically

molecular mechanism. You're basically recapitulating human genetics to turn on in this case fetal hemoglobin that's shown to be a very protective and effective way to um help patients with

cickle cell anemia. But it's a third category that I've been spending more time on recently. So if you pick the right target, you match the right modality to a molecular mechanism of action, then you take that molecule into

humans for the first time. And this is when you really begin to get at dose.

How much of the drug is required?

Schedule. How often do you give it? Do

you give it once a day, twice a day, once a week? What's the frequency with which you give the drug? And then what's the relationship between pharmacocinetics, the level of drug in the body, and pharmacodnamics, what that

drug is doing to the target or the cells within the body. And then how does that relate to clinical efficacy? So we call this path to clinical proof of concept.

And as I'll talk about in just a few minutes, this is the point really for the first time that you know if a molecule a medicine is working as you

actually predicted. And so it's this

actually predicted. And so it's this really important flection point in a drug discovery journey. Then once you see clinical proof of concept, step number four is to accelerate clinical

development as many indications as as you can at as fast as you can because there's incredible competition uh to make not just the first drug but also the best drug and there are a lot of

people that are doing this across the entire world and being first and best is incredibly important. And then all along

incredibly important. And then all along the way what you want to do is the fifth principle uh which is to uh build out a complete uh compendium of information uh that will allow this drug to be used in

the most number of patients so the most number of patients can benefit. So

that's uh maximizing a market access. So

these are the five principles and so what I thought I would try to do today is to describe this in a way that uses poker as as a metaphor. So just before I

I start this m how many people actually play poker.

All right. So a decent number. So Texas

holdem. So this is going to be based upon the game Texas holdm. And I'll

describe the game to you. And again this is the first time I've actually used this metaphor. So you'll have to tell me

this metaphor. So you'll have to tell me if it actually works or not. So causal

human biology matching modality mechanism. If you play Texas holdm each

mechanism. If you play Texas holdm each person is given two cards. Those cards

are private to you. So I'm playing against Jesse. I'm playing against Ben.

against Jesse. I'm playing against Ben.

I shouldn't pick statisticians because they're really good at this stuff. But

but I I'm playing against them. I've got

my two cards. They've got their two cards. And I can see mine, but they

cards. And I can see mine, but they can't see my cards. And in this case, I've got two. I've got an ace and a queen. Two pretty good high cards. So

queen. Two pretty good high cards. So

this is the target and matching modality to mechanism. So I'm feeling pretty good

to mechanism. So I'm feeling pretty good about the hand. In this case, the hand that I'm taking from research into early development. Any ideas on what the

development. Any ideas on what the average cost of a drug to take from research into early development is about how much do you think that costs per drug taken into early development? Any

estimates? Anybody want to shout out a number?

100 million. That's actually right.

Yeah. 100 million.

Wow. This is this is a great crowd. Yes.

About hund00 million uh to make that medicine and take it into humans for for the first time. So you have to be very very careful about how many of these medicines you you take in into early

development. All right. So then you

development. All right. So then you start to do your early development study. This is where you're doing dose

study. This is where you're doing dose and schedule. You're giving it to the

and schedule. You're giving it to the right patient populations. And this is when you see a really big inflection point. So at the start of a clinical

point. So at the start of a clinical trial until the end, it's about a 10% probability of success. At this point, you go from about a 10% probability of success. If you do the right trial,

success. If you do the right trial, you're going to about a 20, 30, 40, potentially even a 50% probability of success after what's called, at least in

poker, the flop. So now you get three cards. So Jesse and Ben, there are three

cards. So Jesse and Ben, there are three cards that we all can play together.

I've got my two private cards, they have their two private cards, and then we have three in the middle. This is the flop. I'm feeling pretty good because I

flop. I'm feeling pretty good because I now have a pair of queens in addition to ace high. This is a pretty good hand. So

ace high. This is a pretty good hand. So

basically what I'm trying to say here, I hope this metaphor is working, is that in early development, I'm feeling really good at this point. I've picked the right target. I've actually matched a

right target. I've actually matched a modality to a molecular mechanism of action and I've done this early clinical development study and it looks really positive. But now I need to complete the

positive. But now I need to complete the full hand. We do accelerated full

full hand. We do accelerated full development and we start to see in this case an ace. Now I got a pair of aces, two queens, two pair. That's a good hand. But at the end of the day, it's

hand. But at the end of the day, it's about having a great label the FDA can approve. And in this case, a queen,

approve. And in this case, a queen, that's a full house. That hand wins almost every single time. So basically,

you need all of these together in order to have a winning hand, in which case this is a an approved drug. I mentioned

about hund00 million um to take a drug into humans for the first time, and it's over two two3 billion to ultimately get a drug approved. But it's not the cost of the successes, it's the cost of the

failures that really drives up that price. So this is why human genetics can

price. So this is why human genetics can help us with these billion dollar bets.

So what I'm going to do now is I'm going to go through a few examples of of programs in our pipeline. And I'm going to go through them relatively quickly, but the points are really going to emphasize everything that I've just talked about in the last few minutes.

Picking the right targets, matching modality mechanism, seeing that early clinical data, and ultimately doing what you can to accelerate full development.

Now, and just one other example here, you could have a competitor that goes after the exact same targets. They pick

the right target, they make the right molecule, but they stumble in early clinical development. They don't design

clinical development. They don't design the right early development trial. They

don't have the right biomarkers. They

don't have VJ on their team to really understand details of of the mechanism and how to understand if this drug is working the way that it should. And so,

ultimately, uh, they're they're not successful. So again, it's more than

successful. So again, it's more than just picking the right targets and making the right molecules. Uh there's a lot that happens in early development as well. Okay. So the first story that I'm

well. Okay. So the first story that I'm going to tell you very quickly is a story that I worked on when I was back in academics. And this was understanding

in academics. And this was understanding the genetic basis of a molecule called tick 2. And what this is showing is the

tick 2. And what this is showing is the causal human biology for this is very compelling. There are at least two or

compelling. There are at least two or three different common alals that are loss of function and protect from risk of multiple autoimmune diseases including psoriasis, lupus and

rheumatoid arthritis. Uh there was a

rheumatoid arthritis. Uh there was a study that was published by Lars Fuger and colleagues Khali Dendrew that showed that one copy leads to about a 10 to 20% reduction in tick 2 signaling but two

copies that is in the homozygous state leads to about a 70% reduction in tick 2 signaling. So it's loss of function at

signaling. So it's loss of function at about a 70% level that protects from monmunity. And then this was one of the

monmunity. And then this was one of the first studies that actually did a few was to look for few waswas as a way to predict adverse drug events. And what we saw with those same variants is they did

not increase risk of infection or other blood disgraases. So it seemed like not

blood disgraases. So it seemed like not only was a good target based upon efficacy, but it was a potentially a safe target as well. But it was really hard to make a therapeutic molecule

against this target. on the left hand side is the regulatory domain. That's

where that that triangle is. There are a lot of other molecules like tick 2, Jack one, Jack 2, Jack 3 that have a lot of homology in that orthosteric domain. So

every single time a company tried to make a binder against the orthosteric park pocket, they hit Jack one, Jack 2 and Jack 3, and that was associated with

side effects. So what BMS chemists did,

side effects. So what BMS chemists did, what BMS biologists did is they designed a phenotypic screen and they uncovered this alossteric pocket. Um actually I got it backwards. The active site is

where most of the homology is. The

regulatory domain where the triangle is that's where we identify the BMS scientists a drug could bind to this alossteric pocket but it could lock the active domain in the inactive state. And

so BMS discovered a small molecule that really rec recapitulated human genetics in a very faithful way. And then in the first in human clinical trial in this

case uh looking at an interferon signature in psoriasis you can see about a 70% reduction. So really precisely phenopying the human genetics. So about

70% inhibition for human genetics and about 70% inhibition uh for the drug. So

there's really nice phenopy of the drug on this particular phenotype. Um and so so tick 2 is now approved in psoriasis and we have ongoing trials and other autoimmune diseases specifically lupus

and and chogrin. So really if you go through the metaphor you know you have full house this is a successful drug and an approved drug and we have other ongoing trials as well. Okay. So now I

want to go to that second principle matching modality to mechanism. And to

me again if I go back to where I was about 15 or 20 years ago in academics I was really focused on the target and then I came to appreciate how much there

is now in our therapeutic armamentarium.

In the late 90s basically 90% of drugs were small molecules. There were a handful of more conventional auto uh more conventional antibodies. Fast

forward today you've got and this is just the BMS portfolio. You've got a bunch of different ways to modulate therapeutic targets. And there are a lot

therapeutic targets. And there are a lot on this list that we just don't pursue, including many of the the genetic based approaches that you've heard about over the course of the day. But what I'll do

is I'll walk through a couple of stories starting with small molecules, but I'll touch on on a few of these. Um, so here is a genetic target TLR7. Uh, there's

very strong human genetics as shown on the far left. So TLR7 gain of function mutations lead to a lupus-like phenotype suggesting that inhibiting TLR7 could be

therapeutically beneficial. There are

therapeutically beneficial. There are some challenges in actually inhibiting this target and that you had to get it a small molecule to actually inhibit this within the context of the endoome and on

the far right is that first in human clinical study where we also saw a change in an interferon signature uh for a drug called aphimatorin. Um this drug is now uh has just finished phase two

and is entering into phase three clinical trials. Uh and this is some of

clinical trials. Uh and this is some of the clinical data uh that shows a very nice separation in a subset of patients uh with lupus that are called cutaneous lupus. So CLE and what you're actually

lupus. So CLE and what you're actually seeing in the graph in the bottom row in the purple is a nice difference between a clinical score in patients with cutaneous lupus compared to those same

patients who received a placebo. Um I

want to spend just a few minutes on a different therapeutic modality target of protein degradation. This is a way to

protein degradation. This is a way to actually not just inhibit a a target of interest but to degrade that that target of interest. Um so the idea here is

of interest. Um so the idea here is there are a lot of uh validated uh targets a lot of human disease associated targets but most of them are not amendable to more conventional small

and large molecules. In the case of certain types of uh of proteins such as transcription factors or those with scaffolding functions, there's no enzyatic function to inhibit. You have

to actually degrade or get rid of the target, which is often difficult to do.

Um, and there's also benefit uh in terms of using uh protein degraders uh to overcome resistance of uh of small molecule uh kynise inhibitors as well.

Um so this is a story that actually builds nicely on the story that VJ told in terms of um being able to induce fetal hemoglobin. Um what I'm showing

fetal hemoglobin. Um what I'm showing you on this slide is that we actually have identified two um uh one uh two two targets that are that influence fetal

hemoglobin production. One of those two

hemoglobin production. One of those two is a genetically validated target. It

goes by the name Pokemon, also known as ZBTV 7A. The other target is Whiz. And

ZBTV 7A. The other target is Whiz. And

what we found is that we can identify a molecular glue molecule um that can bind to an E3 ligase on one part. So it's

basically one molecule that's a therapeutic molecule binds to an E3 ligase and then it recruits these two transcription factors to a single complex. It shuttles them to the

complex. It shuttles them to the proteosome and leads to degradation. So

we can basically degrade these two transcription factors turning on fetal hemoglobin in our pre-clinical models and we are um in clinical trials today

and I think by by the end of the year early next year uh we should know the extent to which this turns on a fetal hemoglobin. Um so if it does work it has

hemoglobin. Um so if it does work it has the potential to be an oral small molecule pill that achieves the same level of efficacy as cascade which is an

Xvivo uh gene editing uh approach.

Uh here's another example where the target BCL6 is a well-known um uh ankcoenic driver. Here the genetic data

ankcoenic driver. Here the genetic data is not from germline genetic data but it's from sematic cell genetic data. So

there's a lot of great sematic cell genetic data in oncology. Again, you

heard about it from VJ in terms of a clonal amount of poesis, but many of these encogenic drivers can actually lead to to uh to cancers and BCL6 is a known encogenic driver. Um but like many

transcription factors and other targets, it's just very difficult to drug with a typical kinase inhibitor. So what

scientists at BMS did is they created what's called a heterobifunctional degrader. So one arm binds to BCL6, the

degrader. So one arm binds to BCL6, the other arm binds to an E3 liase called cereablon. It brings the complex

cereablon. It brings the complex together leads to BCL6 degradation. So

very strong human biology in this case from from sematic cell genetics. We've

got a really interesting uh degrader that's heterofunctional by degrader. And

this is also a program that's advancing in clinical development and we'll know probably within the next year if we're seeing the PKPD relationship that we expect. Okay, just a few more quick

expect. Okay, just a few more quick stories before I go on to the saturation mutagenesis part of the talk. Uh and

that is biootherrapeutics. Uh and here again another story of causal human biology. In this case it's T-regs and

biology. In this case it's T-regs and cancer. There's no really strong human

cancer. There's no really strong human genetics to implicate CCR8 positive T-regs in cancer. But through single cell sequencing, and you've heard a lot

about that today, what we've been able to do is to identify that a subset of T-regs uh that are identified in patients with lung cancer and other types of cancers are a negative

predictor uh for for outcome. that is

the presence of these highly um uh um suppressive inflam or highly suppressive CC positive T-regs are uh a negative predictor of response to checkpoint

inhibitors and so we predict is by binding to CCR8 on the surface of cells and degrading um uh the basically

getting rid of these CCR8 positive T-regs will be able to allow anti- PD1 and other checkpoint inhibitors uh to work more effectively. This is also a

program that is in in first in human studies now and we should know more in the next year or two whether we've achieved clinical proof of concept. All

right, I'm going to go through one more example and do this very quickly because we talked about small molecules and large molecules but we're also in an age where we can actually engineer living

cells uh to be therapeutics themselves and this is in the form of of autotogus carti. Um and then to me this is one of

carti. Um and then to me this is one of the most exciting areas of science and biology right now. Uh because I used to practice clinical rheatology. Patients

would come in young women childbearing age with lupus. They would they would give them a very strong immunosuppressive which could get their disease under control and inevitably they would ask well when do I come off

of these medicines and the answer was almost always never. So imagine being, you know, a young healthy person suddenly getting this new diagnosis and being told you're going to be on

imunosuppressants for the rest of your life is pretty devastating. So this

concept of being able to reset immune memory to deplete the auto reactive cells to get rid of them completely and allowing a new immune system to reconstitute and basically that new

reconstituted immune system is healthy.

This is what we're able to do with auttogus cartis. for these carti cells.

auttogus cartis. for these carti cells.

Um, but by the way, just on the bottom left, there wasn't a whole lot of genetic data that implicated C19.

Specifically, it was a human pharmarmacology study where an investigator uh, uh, George Shet actually took CD19 CARTT, which is approved for lymphoma, tried it in a

very small number of patients with lupus, and sure enough, it completely reset their immune system, and they've been off all of imunosuppressive drugs for now for several years. um we've now

treated multiple patients uh dozens of patients with not just lupus but other autoimmune diseases and we've been able to recapitulate this finding as well. So

causing biology from human pharmarmacology matching modality mechanism it's an autotogus engineered cell that actually kills the CD19

lineage of B cells and we now have clinical proof of concept and we're entering into registrational trials. Um,

so these are the examples that sort of show how the sort of framework of these five principles work in action and how we're using them all the time within our portfolio today. Okay, so I have about

portfolio today. Okay, so I have about 10 minutes left and I'm going to go through uh how we're trying to expand our understanding of individual variants

uh to basically build out this toolkit even further. Um, so when I moved from

even further. Um, so when I moved from academic to industry, this was kind of my simplistic view of how I thought we could use human genetics to identify new targets. Not just identifying a single

targets. Not just identifying a single alil at a particular gene, but with sequencing, we'd find more and more alals. And we could say, all right, each

alals. And we could say, all right, each one of those X's that's meant to be a um that's disease associated. Along the Y ais is a human phenotypes. Imagine high

LDL cholesterol, high risk of disease, um, as examples, you know, high body mass index. And if you could find alals

mass index. And if you could find alals that were associated with a human phenotype that was an appropriate surrogate for drug efficacy, and you could understand the function, gain of function or loss of function, you could

begin to estimate these dose response curves, an efficacy curve, and also know if there was an adequate therapeutic window. Um the challenge has been and

window. Um the challenge has been and we've been talking about this all day is we can find that sort of x-axis pretty easily using bioank. So we can look quantitatively at whether an alil is

associated with high or low cholesterol.

What level of risk um is contributed in terms of risk of autoimmune diseases.

But what's incredibly hard is getting that variant uh to function. Uh and um just a really quick story when I just started my posttos Pierre I don't know if any of you know ick um but we had

just done one of the first like a thou h 100 thousand snip chips for math metrics on the hatmap cell lines and we could actually identify 30% of common variance and that was a number that totally blew

my mind and I remember I saying to me and we were sitting over there in one Kendall Square up in one of the conference rooms he's like can you imagine if we could just annotate every

single variant not just by its where it is in the genome but by its function.

Imagine how much more powerful our genetic association studies would actually be. So not just looking at

actually be. So not just looking at linkage to singleian and the static is it associated or not but you could add a coefficient that tells you something about function. And I remember thinking

about function. And I remember thinking wow is that's a great idea that sounds really hard. And sure enough fast

really hard. And sure enough fast forward to where we are today. I think

we're getting po we're getting to a state where we should be able to incorporate these types of weights into our association study. So I'm going to tell you now a story that's going to build on the tick 2 story that I told

you about earlier. And so the question here is can we take a known validated genetic target that has an orthosteric pocket and an alossteric pocket. Can we

mutate every single amino acid uh at this particular protein and can we uncover the known alossteric and the orthosteric sites where these drugs

actually bind and can we learn new things about um this particular molecule that could potentially help with large scale association study.

So um just just quickly in terms of what we know about the genetic variation across tick 2 and the upper graph is the clinbar variants. So there are three

clinbar variants. So there are three variants rare variants that can when when tick 2 is knocked out completely causes primary immune deficiency. The

next graph below is the nomad uh variants uh and there are several hundred uh variants and that's shown at the top part of the bar graph and above a particular frequency spectrum is

what's shown below. And if you blow this out and look at the fre frequently spectrum what you can see on the far right are the four common variants.

Three of these are associated with protection from autoimmunity. But if you take a threshold of 10 to the minus 4th or less, they're about 447 misses snips

from the UK bioank that are within this particular category. Um these are all

particular category. Um these are all below you. 01 in terms of the minor

below you. 01 in terms of the minor little frequency threshold. We know

nothing about the function. But can we use saturation immunogenesis to assign function to each of these 447 variants?

So this is where we worked with Octant um to do a deep mutational scan where again we mutated every single amino acid by every other possible amino acid across the entire protein. And we did

this with two different assays. The

first assay that I'll tell you quickly about is an interferon signal signaling assay. Uh what you can actually see uh

assay. Uh what you can actually see uh is there's a cassette that's inserted that has different tick 2 amino acid changes that codes for uh the uh tick 2

that will bind to the interferon signaling receptor. It'll signal down

signaling receptor. It'll signal down through and there's an interferon signaling response element which can then be read out in the form of a barcode. So about 11% of amino acids

barcode. So about 11% of amino acids when mutated by this assay could lead to changes in interferon signaling. uh and

you can see uh in particular this line uh where a stop variant was introduced but you can also begin to see domains and I'll describe these domains in just

a minute. Um the second assay that we

a minute. Um the second assay that we did was a protein abundance assay. to a

simpler assay and I'm going to come back to this because I think this has some application as we think about scaling this approach up but a protein abundance assay where we also introduce um every

single amino acid variant into tick 2 and then we measure we can sort by GFP also with a barcode and just measure relative protein abundance. So about 20%

of amino acids uh when mutated lead to a signal in this particular assay and what I think you can appreciate is there's a lot of overlap uh between uh the two different assays. And when you look at

different assays. And when you look at where where the um alossteric site and the is located, this is the name of the drug dravisitib. You can identify the

drug dravisitib. You can identify the alossteric binding site with this functional assay and you can also identify very clearly the catalytic site. Uh and at least in the protein

site. Uh and at least in the protein abundance assay, you can identify the disease associated P104 A variant. Um

although it's not quite as clear I know it's difficult to see it's not quite as as clear that that variant has an effect on on interfere on signaling.

So there are other positive controls that gave us confidence that the assays were working including uh different annotations by alpha missense some literature associated loss of function

variants and then three variants that were reported as as pathogenic. Okay.

And I'm going to come back to the three starred sites because these identify potentially novel aloseric sites. So um

this identifies again um a number of of variants that could be loss of function or gain of function by protein expression along the the y-axis or interferon signaling along the x-axis.

As I mentioned P104A which is one of the disease associated varants comes up with a good signal by protein expression but but not by uh interferon signaling. Um

and I think with these types of assays we can begin to model um where these mutations are occurring on the two domains of tick 2 the the the the regulatory domain and then the uh the

the kynise domain. We also refer to these as the the catalytic and non-catalytic domains. Um and you can

non-catalytic domains. Um and you can see again the active site alsteric site and the location of P104A.

Um but there are these variants that look like they have an effect on interferon alpha but no effect on protein abundance. um could this be a

protein abundance. um could this be a potential way to uncover novel alossteric sites? Um so what we did is

alossteric sites? Um so what we did is this filtering mechanism to look for variants that affected interferon alpha signaling along the top uh whether there was evidence of destabilization.

If not, could this be potentially a catalytic alesteric site? And you can see here we're just annotating a few of these potential sites, which actually

two of these sites are right at this kyna pseudocinase interface as shown by the blue stars. Um, and there are and this is a fulllength view of the protein. um where you can actually see

protein. um where you can actually see in in uh yellow here is the kind of this ribbon of where the receptor itself is uh is uh is is going through uh the

protein and the protein is binding around it. Uh and there are potential

around it. Uh and there are potential sites uh not just on the where the protein binds or interacts with the plasma membrane but also where tick 2 is engaging with the the interferon

receptor.

Um now one other way that we actually use this data is to actually look at where the drug dravisit binds and can we actually map um residues that are b

interacting between the the the gene the protein tick 2 uh and the known inhibitor at the alossteric site. Uh and

the quick version of this story is yes we can as we actually go through and use different concentrations of the drug. So

lower concentration of the drug you really find only those sites that are direct directly interacting uh with the drug itself and at higher concentrations

of the drug um you identify um other other sites as well. So it provides a nice kind of functional complement uh to

structural data. Okay. So we started by

structural data. Okay. So we started by saying if we could only just annotate uh the human um phenotype data by by function and what I've told you at least

for tick 2 we've been able to use deep mutational scan to begin to assign functions to each of those 447 variants and so now I want to go back to basically the question that I had

proposed 20 or so years ago is if we begin to incorporate this information into our genetic association studies you know can we actually have more wellpowered powered rare variant

associations. And so we took these um uh

associations. And so we took these um uh these these variants uh and we annotated uh whether not just by common and rare but by level of tick to 2 um expression.

And we used a couple of traditional burden test methods that are shown here without any waiting and it didn't find an association with the 300 or 35,000

cases from the UK biioank compared to 308 uh uh controls. uh but we began to incorporate the output from the um the DMS variant scores into a model, we

began to see more statistically significant associations with this genetic burden test. So again suggesting that as our methods of assigning function to each of the variants become

more and more powerful, we should be able to improve power of rare variant association studies.

Um you know I will say in listening to the talks today sort of occurs to me that if we can do this more and more uh not just with these types of functional data but with the predictive data and the modeling data that has been talked

about today hopefully we'll be able to do a better job of basically incorporating these types of weights into our association studies. And in

fact, here are a few newer methods that don't use um uh functional data directly uh but do appear to have uh stronger and more wellpowered um statistical results

uh using the same uh data set. Okay, so

just two more slides and then I'm going to stop. Uh I want to then sort of talk

to stop. Uh I want to then sort of talk about genetic associations and how this is used. We're using this information to

is used. We're using this information to guide indication selection. So what I'm showing you on this particular chart is three v are three variants within tick

2. And you can see for psoriasis, RA and

2. And you can see for psoriasis, RA and lupus, these three variants are protective at a very high level of statistical significance. Inflammatory

statistical significance. Inflammatory bowel disease has just a different pattern here where one of the three variants looks like it's a risk variant.

And this is something that we haven't really able been able to figure out. Um

but it does look like it has a very different pattern in inflammatory bowel disease. And at least the clinical

disease. And at least the clinical trials to date have been successful in psoriasis. They have not been successful

psoriasis. They have not been successful in inflammatory bowel disease. That is

so tick 2 did not work in treating patients with inflammatory bowel disease. And as I've mentioned, studies

disease. And as I've mentioned, studies are ongoing with lupus. So we think and hope that it'll work in this particular indication. Um but we'll know more by

indication. Um but we'll know more by the end of the year. Um but if you actually sort of plot out uh the the the different variants uh in a slightly

different way um and look at um the opposing effects across the proteome again you see in some cases the variants act in the same direction in other in

other instances they act in in opposite direction. Okay so I'm going to stop

direction. Okay so I'm going to stop here but just to conclude I know I went through things pretty quickly. Most

drugs fail because of lack of efficacy.

Hopefully human genetics can help. I

talked about placing billion-dollar bets using these principles to think about making the right decisions not just for the target and the modality but how do we actually do those early clinical

studies and ultimately how do we do full development for drug approvals. Um

genetic dose response curves um can hopefully be used to define uh exciting targets and these alyic series and the deep mutational scans in tick 2 um have

the ability to further build out these genetic dose response curves. And so I guess you know the question that I would leave you all with is what what should be done next? Is there a way for us to

do saturation mutenesis across a larger proportion of the proteium or are the computational methods sufficiently uh strong at this point in time that we don't need to necessarily do these types

of of functional studies. Um so there's a lot I think um we could uh we could certainly uh discuss. Um before I do, I want to let you know that um the paper on deep mutational scan should be posted

on bioarchchive uh this week. Uh here

are the authors and again a special shout out to the scientist at at Octant who did really all of the saturation mutagenesis work but working closely with Joe Moranville and his team uh to

work through the associations that I described to you. So thank you very much.

>> Thank you. Any questions?

Lovely talk, Robert.

And I'm very fond of it sick, too. Uh,

so I I feel like there's like a butt coming here.

>> Ben, the question ask is that the question I'm gonna ask is um about the tick 2 functional essay.

>> Yeah. I think, you know, to me a lot of this turns around the quality of the gene specific assays and how well we think they're actually recapitulating the function that may be perturbed by

the variance that we're introducing.

Right? So, you got to really trust the assay part.

>> And so, I guess the question is how many genes do you think we have that caliber of assay for? Is that assay capturing all of the functions for kick 2? And how

do you think about approaching it in that way versus I think a lot of what we've heard about today is you know transcrytoics and general cellular properties but they're maybe not specific about the function of the protein itself.

>> There there was a lot of calibrating the interferon assay. There was a second

interferon assay. There was a second assay that actually never worked measuring 23 signaling that we never you know were able to sort of validate with the positive controls. P4A is a known functional variant and it didn't come up

in the interferon assay. I mean I kind of intrigued by this protein abundance assay. So maybe that is something that

assay. So maybe that is something that could be done um at scale and is sufficiently agnostic to sort of nuances of biology. So that's the one that I

of biology. So that's the one that I sort of wonder is that something that we could actually scale up and make relationships to circulating um pro proteins as well.

>> Yeah, Brad.

>> Yeah, this is me. Oh yeah, nice talk.

Remember David Alcheler talking about the arrow of time so many taught me about that 15 years ago or something like that. It's a beautiful application.

like that. It's a beautiful application.

One the one exception was the CCR8 or the getting rid of the T-reg populations. And there's so much data

populations. And there's so much data now on changes in population of cells and micro environment cancer and other diseases. That's not quite causal

diseases. That's not quite causal though, right? It's just associated. So

though, right? It's just associated. So

how do you you're you're clearly using that in one of your as a premise for one one of your studies.

>> Yeah. How widely are you using that and how do you get at causality in those cases?

>> Yeah. So I think I think in that case so you're totally right. I mean you're basically looking at an association the presence of these T-regs in patients who don't respond to checkpoint blockade.

That's a that's a correlation. There's

not a causal relation. We try to actually look more longitudinally to understand you know what do we know about the phenotype of those T-Rexs? You

know when do they infiltrate? Um but at the end of the day it's more of a correlative relationship than than than the causal relationship in that specific case. So I think you know I we often say

case. So I think you know I we often say you know causal human biology yes no and it's far more I think of a gradient and I would say something like tick 2 very strong that's probably a you know three

to fourfold increased probability of success. CCR8 I'd probably put it as a

success. CCR8 I'd probably put it as a you know you know 50 you know 100% maybe a 50% increased probability of success but not the same level as I think some of the genetics that we actually talked

about but one right here uh very cool talk and um on the tick 2 screen I was wondering about the

in the case where you don't have changes in abundance but you do see effect in activity and how much of those changes were uh due to disruption or creation of

denovo phosphorilation sites altering you know protein activity which is an old you know a way that a lot of these kynases are actually regulated by and how much of those actually exist naturally out there and you know just

yeah it's a great question I'm not totally sure um so I I I don't know I mean the way that we kind of frame this is maybe these are novel um alossteric sites but I think that that particular

biology is not entirely clear to me so I think that's going to take more work to sort of sort out. Thank you.

>> All right, a question in the or question in the back there.

>> Uh thank you that was really interesting. I was wondering you know

interesting. I was wondering you know when you when we talk about how genetics informs drug discovery my impression is it's mostly on the sense of like one variant going to one trait but as your example in the last few slides showed we

have a lot more much richer phenotyping of individuals now like and now we have the opportunity to look at variant plyotrophy do you have any comments on whether there's something that we what can we do with plyotrophy now that we

have this data like you think it will be useful >> I think I think plyotropy can be used for a couple things one it can be used to guide indications as a way to say it works here, it's going to work in the

next two or three indications as well.

But I think plyotropy can also be used to begin to predict you know adverse drug events and so plyotropy can work both ways. It can help you with

both ways. It can help you with phenotypes that are good surrogates for efficacy but also help you for good phenotypes that are surrogates for for toxicity.

>> And I know we're over time so maybe we can we can stop here and head over to >> maybe one last I think there's already a mic out in the audience. Thank you uh Robert to to my question is related to

your last comment about what's next should we do the deep mutational scan maybe more systematically so while you're talking about it that's really remind me in oncology field I came from

oncology field right that when we discover all the somatic mutations in patient and indicating maybe certain genes is uncleco gene and having a deep

map where every single gene are knocked out and functional effect measured across thousands of cells lines where we can correlate. So that's really a deep

can correlate. So that's really a deep kind of knowledge base we use to build that kind of a functional relationship and confidence of a genes a causal gene.

So in some way I would think if we can have this kind of systematic functional data that will really help us to interpret the genetic data in a much more objective way.

>> Yeah.

>> I don't know if it's feasible but I think that would be great to have.

>> Yeah. I think I think the more I mean we talk about how we train and build these models. I think we often train them off

models. I think we often train them off the data sets that we actually have. Um

you know I sort of imagine a world in which we actually um you know try to do far more like say mutational scans to look at protein abundance to uncover alesteric sites and maybe then that

could help us feed you know the models that we actually use to identify um alsteric these regulatory domains in a way that's it's kind of difficult to to do today. Um but it does go back to some

do today. Um but it does go back to some of the assays as Brad mentioned. So I

think in oncology for example um you know for for cell cycle growth and and basically things that you know accelerate growth or decelerate growth those are some assays that I think are amendable to this type of approach.

>> Thank you.

>> Great. Thank you so much for uh the presentation Robert. Uh please join me

presentation Robert. Uh please join me in giving him another round of applause.