CVM Grand Rounds 9/17/2025

Transcript

• 00:00And our leadership of our
• 00:01of our CME series for
• 00:03rebranding
• 00:04us and moving us to
• 00:06this date.
• 00:07Last week's
• 00:09introductory session
• 00:10was really wonderful.
• 00:12And the fact that we
• 00:12were able to,
• 00:14kind of, share new ideas
• 00:16of how to continue to
• 00:17build on on this long
• 00:18tradition path of removing to
• 00:20this time and and and
• 00:22day of the week.
• 00:23So, moving forward, obviously, Wednesdays
• 00:26at noon will be our
• 00:27time, and and, and and
• 00:29there is a schedule that
• 00:30would incorporate
• 00:31not only a case based
• 00:33clinical conference led by a
• 00:35fellow and a a faculty
• 00:36mentor for this for the
• 00:38section at large, a research
• 00:39forum led, by doctor Bender,
• 00:43and,
• 00:44at least two visiting professors.
• 00:46One ex you know,
• 00:50one we have today that
• 00:51we'll I'll have Tara describe,
• 00:53and introduce in a moment.
• 00:55Before we get started, I
• 00:57just wanna remind folks that
• 00:59our CME process had not
• 01:01changed,
• 01:02so please
• 01:03use these five numbers to
• 01:06sign on for your CME
• 01:07credit. We'll be tracking this,
• 01:10just to give us a
• 01:11sense.
• 01:12And,
• 01:14also,
• 01:15please
• 01:17separately
• 01:18for the for the foreseeable
• 01:19future, sign up with our
• 01:21administrative team up front because
• 01:24as we have members of
• 01:25the community at large, the
• 01:27heart and vascular center as
• 01:28well as the section in
• 01:29the department,
• 01:30we are trying to just,
• 01:32right size
• 01:33the in room,
• 01:35ordering of food and that
• 01:36so there's two separate things.
• 01:38CME activity is electronic,
• 01:40but the sign up is
• 01:41really to give our team
• 01:42some, information around how to
• 01:44plan
• 01:45to make sure we're, good
• 01:47stewards of our finances and
• 01:48don't over order,
• 01:50which,
• 01:51unfortunately, is a kind of
• 01:52a tradition at Yale from
• 01:53what I can tell.
• 01:56And then I, before I
• 01:58get,
• 01:59maybe one other quick intro,
• 02:02just wanna,
• 02:03thank,
• 02:05all the members of our
• 02:07section who are viewing this,
• 02:10live, in
• 02:11groups,
• 02:13at so called watch parties.
• 02:15And hopefully, you got your
• 02:16lunch delivered to Greenwich, New
• 02:17London, and Bridgeport,
• 02:19and wherever you are,
• 02:21at the VA. And I
• 02:22think it's a a wonderful
• 02:24new
• 02:25tradition,
• 02:26so that, we all can
• 02:27come together at wherever we
• 02:28are,
• 02:29for this hour,
• 02:31of of fellowship,
• 02:33and, and collaboration.
• 02:35And so with that,
• 02:37I wanted to, surprisingly,
• 02:40introduce,
• 02:41our new chair, Wolfram Gosling,
• 02:42who's here with us. And
• 02:44everyone give me give him
• 02:45a a hand
• 02:46of applause.
• 02:49Wolfram,
• 02:51reached out this morning,
• 02:52to me and and,
• 02:54just alerted me that he
• 02:56used to be his lab,
• 02:57partner across the hall or
• 02:58somewhere, and your labs were
• 02:59near each other. And he
• 03:01was excited, and and, I,
• 03:03was excited he could join
• 03:04us. But he does not
• 03:05wanna take the thunder away
• 03:06from, Raj. And so, we
• 03:08will leave it for another
• 03:09day to have him more
• 03:10formally introduce himself and his
• 03:12goals
• 03:13to the, section at large.
• 03:14And and those of you
• 03:15who were,
• 03:16as I was inspired by
• 03:18his, town hall for departmental
• 03:20faculty,
• 03:21should certainly, if you weren't
• 03:22able to go look at
• 03:23those slides in that presentation,
• 03:25I think, will be very
• 03:26important. So,
• 03:28well, from my hope, I
• 03:29I didn't make put you
• 03:30too much on the spot,
• 03:31but,
• 03:32so, again, here's the CME
• 03:33activities. This is what we
• 03:34have coming up in the
• 03:35next month.
• 03:37Critical care conference,
• 03:39on the first,
• 03:42and then we have the
• 03:43we have a holiday, a
• 03:44Jewish holiday in the middle,
• 03:46there before our next one.
• 03:47And then Raj,
• 03:49Rajesh Vran Vranasen
• 03:51from, NYU,
• 03:52an imaging case conference, and
• 03:54then Mark Peltier, our new
• 03:55chief of cardiac surgery, is
• 03:56gonna give us grand rounds,
• 03:58towards the end of the
• 03:58month.
• 04:00We also just wanna make
• 04:01a quick announcement for our
• 04:03faculty.
• 04:05We have
• 04:07historically, had our faculty meetings
• 04:09at five o'clock on Wednesdays,
• 04:11which we will have today.
• 04:12Larry Young is on deck
• 04:14to give us a bit
• 04:15of a presentation. It'll be
• 04:16a short fact meeting, but
• 04:17we are going to move
• 04:19those,
• 04:20from here on end to
• 04:21a different day as we
• 04:23move their grand rounds so
• 04:24that there's not
• 04:25too much,
• 04:27time,
• 04:28in conferences,
• 04:29for people to attend.
• 04:31Is there a disclosure to
• 04:32accreditation?
• 04:33And, and, here are our
• 04:35leaders. And if he's that,
• 04:36I'd like to,
• 04:37introduce,
• 04:39Tara Kumar to tell us
• 04:40a little bit about,
• 04:42doctor Gupta and introduce,
• 04:44our speaker.
• 04:45Thanks. Thanks, sir.
• 04:50Good afternoon, everyone. Welcome,
• 04:52back to the academic year,
• 04:54and
• 04:55everyone looks, completely refreshed and
• 04:58good to be back. As
• 04:59someone who
• 05:00ran Grand Rounds sometime back,
• 05:02this is an incredible
• 05:04improvement, and it's so happy
• 05:06to see how all this
• 05:07is going. I do wanna
• 05:08give a special shout out
• 05:09to Joanne who just did
• 05:11an incredible job,
• 05:13with this visit. So thank
• 05:14you and, obviously, the rest
• 05:15of the team as well.
• 05:17So that it's a true
• 05:18honor and privilege to introduce
• 05:20my my friend, Raj Gupta,
• 05:23who's an associate professor of
• 05:24medicine at Harvard Medical School
• 05:26and a physician scientist at
• 05:27the Brigham and Women's Hospital,
• 05:29which
• 05:30is now mass general Brigham,
• 05:32of course.
• 05:34Raj,
• 05:35born and raised in Michigan,
• 05:37went to,
• 05:38Michigan for undergrad.
• 05:40He reminds us often that
• 05:42we're at the Michigan of
• 05:43the East here at Yale.
• 05:46He went to Penn for,
• 05:48for medical school and then
• 05:50Mass General for for residency
• 05:52and then Brigham for,
• 05:54for fellowship, and he's been
• 05:56sent there since then.
• 05:59Raj has devoted his career
• 06:00to understanding the genetic basis
• 06:02of vascular disease.
• 06:03He pioneered, he's pioneered single
• 06:05cell and CRISPR approaches that
• 06:07have fundamentally reshaped our understanding
• 06:10of cardiovascular
• 06:11biology.
• 06:12His work is, many of
• 06:13you know has been, published
• 06:15in the highest impact journals
• 06:17supported by multiple NIH grants,
• 06:19and his lab is already
• 06:21training
• 06:21the next generation of, of
• 06:23physician scientists.
• 06:25On a personal note, we've
• 06:27known,
• 06:28many of us have known
• 06:29Raj, since our days in
• 06:30in Boston. And even then,
• 06:32it was clear that he
• 06:34was single mindedly driven,
• 06:36by the joy of discovery.
• 06:38And since then and back
• 06:40then as well,
• 06:42chose to live a near
• 06:43monastic existence,
• 06:45in the pursuit of science.
• 06:47So none of us were,
• 06:49surprised when he's, emerging as
• 06:51a leader in cardiovascular
• 06:54medicine.
• 06:55Of course, his scientific,
• 06:56you know, record is is
• 06:58nothing short of extraordinary,
• 07:01but we I would be
• 07:02remiss if I, didn't mention,
• 07:04that,
• 07:06that on the squash court,
• 07:07the balance of power was
• 07:09very different between,
• 07:11our colleague, Nirard Desai, myself,
• 07:13and and doctor Gupta.
• 07:15And, we beat him really
• 07:17badly,
• 07:18consistently.
• 07:19So no matter how many
• 07:20nature papers he publishes,
• 07:23we will always have the
• 07:24upper hand,
• 07:25when it comes to that
• 07:26time in the squash board.
• 07:27So, Raj, with that,
• 07:29Trent,
• 07:30welcome.
• 07:35Well, thank you for that
• 07:37very unique introduction.
• 07:39A real a true pleasure,
• 07:40I think, to be introduced,
• 07:42by Tarek. I'd say, one
• 07:44of the greatest pleasures and
• 07:45distractions in my life is
• 07:46to be on a text
• 07:47chain with Nihar
• 07:49and Tariq, which,
• 07:50is a a blessing and
• 07:52a curse. But,
• 07:54but one one of the
• 07:55reasons I'm so excited to
• 07:56have been invited, by Tariq,
• 07:58and I'm really grateful to
• 07:59be here, is that, you
• 08:00know, it's it's always great
• 08:01to have friends from your
• 08:03training
• 08:04days who are your biggest
• 08:05cheerleaders and, you know, at
• 08:06this stage of our career
• 08:07to be invited to then
• 08:08present my work in this
• 08:09forum is is is really
• 08:10a highlight. I've been looking
• 08:11forward to this, you know,
• 08:12for about a year,
• 08:14in scheduling it.
• 08:15And, you know, the talk
• 08:16that I wanna give is
• 08:17is about using genetics that
• 08:19we do in a in
• 08:20a my very basic science
• 08:22lab. But I wanna make
• 08:23it, you know, that there's
• 08:24some clinical impact of the
• 08:25work we're doing, and I
• 08:26think that's the challenge that
• 08:28I hope to convince you
• 08:29that, you know, genetics is
• 08:30not just just a an
• 08:31in vitro system that we're
• 08:33we're trying to, tinker with,
• 08:35but there but there is
• 08:36something clinically actionable.
• 08:38And and if I can,
• 08:39you know, I I welcome
• 08:40your your comments on on
• 08:42whether or not I I
• 08:42prove that point. But in
• 08:44giving this talk, I I
• 08:45I tried to think what
• 08:46can I really tell you
• 08:47about genetics that you haven't
• 08:49already heard? Right? This Yale
• 08:51here with Rick Lifton is
• 08:52sort of the birthplace of
• 08:53both Mendelian
• 08:54and common varying genetics. And
• 08:56probably for, you know the
• 08:58last twenty five years you've
• 09:00heard about the promise of
• 09:01human genetics to change how
• 09:02we treat patients right. From
• 09:04you know just from Rick
• 09:05Lifton's career itself
• 09:08you know a solid ten
• 09:09years twenty years of discovery
• 09:11in the the Mendelian drivers
• 09:13of hypertension,
• 09:15and then in the GWAS
• 09:17studies for hypertension,
• 09:18and then he even did
• 09:20variant to function for one
• 09:21of the top loci for
• 09:22hypertension.
• 09:23And based on really his
• 09:25contributions
• 09:26to the field, you know,
• 09:27the the cover of of
• 09:29basically our famous journals always
• 09:31promise that, you know, the
• 09:33human genome and the genome
• 09:34revolution was coming, that we
• 09:36would be able to really
• 09:37deliver on the clinical impact
• 09:39of these findings.
• 09:40And here we are, you
• 09:41know, twenty five years after
• 09:43the sequencing of the human
• 09:44genome. And what what we
• 09:46promised was the genomic era
• 09:47of medicine was that we
• 09:48would have precision medicine
• 09:50from low cost genetic sequencing,
• 09:52and we'd have targeted therapies,
• 09:54early detection, and reclassification
• 09:56reclassification of diseases into genetic
• 09:58subtypes. Right? That was the
• 10:00promise we made when when
• 10:01the genome was sequenced.
• 10:03And I would say that,
• 10:04you know, as as my
• 10:06my friends, Tarek and Nihar,
• 10:07often tell me, maybe we
• 10:08haven't delivered on that promise
• 10:10as much as we we
• 10:11had hoped by now.
• 10:13There are certainly success stories
• 10:15in the application of genetics
• 10:17to clinical care, and I
• 10:18tried to put a few
• 10:19of them here. Right? And
• 10:20and a lot of these
• 10:21have Yale roots,
• 10:23particularly eGFR inhibitors for non
• 10:25small cell lung cancer. Right?
• 10:27You know, was a targeted
• 10:28therapy in the cancer realm.
• 10:30Pharmacogenomics
• 10:31for variants that affect warfarin
• 10:33metabolism.
• 10:34And then gene therapies. Right?
• 10:36Genetic correction for certain Mendelian
• 10:38diseases.
• 10:39But and, you know, this
• 10:40is this is from the
• 10:41popular press.
• 10:43The amount of genomic data
• 10:45is just increasing exponentially. This
• 10:47is just ancestry dot com
• 10:48and twenty three and me,
• 10:50and it went from, you
• 10:51know, maybe a million people
• 10:52in twenty thirteen
• 10:54to twenty five million people,
• 10:56like, twenty five x the
• 10:57number of genomic data.
• 10:59And so you would think
• 11:00that, you know, the data
• 11:01is there, so then the
• 11:02impact must be there. And
• 11:04I would say that I
• 11:05will admit that the revolution
• 11:06has been slow. For coronary
• 11:08disease,
• 11:10we've been successful
• 11:11at identifying data that we
• 11:13can analyze. Right? My mentor,
• 11:15say, Katharason,
• 11:16led genotyping
• 11:18and and applied it to
• 11:19genomic biobanks all over the
• 11:21world. Lipid lowering drugs have
• 11:23been supported by genetic data,
• 11:25but we don't use genetic
• 11:27risk scores to guide therapy.
• 11:28I think, you know, that's
• 11:29one of the major questions
• 11:31I've gotten today
• 11:32is that, you know, are
• 11:33you are you are you
• 11:34using genomic risk scores at
• 11:35all? Have they had any
• 11:36impact? And,
• 11:39Eric Topol, right, made available
• 11:41this this,
• 11:43iPhone app, my gene rank,
• 11:44and this is my genomic
• 11:46risk score for coronary disease
• 11:47at the bottom right corner.
• 11:49So I'm in the ninety
• 11:50second point third percentile,
• 11:52which is humbling. Right? You
• 11:53know? So I'm I'm one
• 11:53of, you know, one of
• 11:54the, you know,
• 11:55highest genetic risk for coronary
• 11:57disease, And I don't do
• 11:59anything about that, and this
• 12:00is what I study every
• 12:01day. And so that's sort
• 12:03of the challenge that I
• 12:04I was excited to tackle
• 12:05when I started my lab.
• 12:06And certainly, you know, this
• 12:07slide is is certainly relevant
• 12:10is, you know, despite years
• 12:11of progress in coronary disease
• 12:13and heart failure,
• 12:14the last ten years have
• 12:15seen these curves reverse direction.
• 12:18So coronary artery disease increasing
• 12:20prevalence
• 12:21in just the last five
• 12:22years and heart failure certainly
• 12:24over the last fifteen years
• 12:26increasing prevalence.
• 12:27So the outline for what
• 12:29I wanna present today is
• 12:31really three stories that come
• 12:32from the work we're doing.
• 12:34The first is I wanna
• 12:35review the promise and perils
• 12:37of these genetic risk scores.
• 12:38They've been around for about
• 12:40ten years. You'll you'll read
• 12:41on Twitter or x that,
• 12:44people are now making these
• 12:45clinically available, and I'm certain
• 12:47Yale will be one of
• 12:47the places that's the first
• 12:49to adopt these things,
• 12:50but there are pros and
• 12:52cons. The second is that
• 12:53my lab as a basic
• 12:54science lab is trying to
• 12:55provide a mechanistic approach to
• 12:57these polygenic risk scores.
• 12:59I firmly believe that the
• 13:01future of genetic risk prediction
• 13:03is at the intersection of
• 13:04genetics and function.
• 13:06And then finally,
• 13:07my focus has been on
• 13:08one cell type that's endothelial
• 13:10cells, and I'll I'll show
• 13:11some of the data that
• 13:12supports that.
• 13:14So first, the promise and
• 13:15perils of genomic risk scores.
• 13:17So I focused on coronary
• 13:19artery disease, but genomic risk
• 13:20scores have developed for AFib,
• 13:22for breast cancer, for all
• 13:24kinds of cancer, and other
• 13:25cardiovascular diseases as well. For
• 13:27coronary disease, right, we've got
• 13:29the kind of the most
• 13:30robust clinical data. The genomic
• 13:32data for coronary disease is
• 13:34the strongest because every biobank,
• 13:36the number one cause of
• 13:37death, the number one diagnosis
• 13:38is coronary
• 13:39disease. So at this point,
• 13:40we have over a million
• 13:42cases and controls.
• 13:43We are able to explain
• 13:44about fifty to sixty percent
• 13:46of the heritability,
• 13:48through genetics, and
• 13:50forty percent of that heritability
• 13:51is captured by common variant
• 13:53association studies.
• 13:54And,
• 13:55you know, the the common
• 13:57variant piece is what's sort
• 13:58of interesting. I'll get back
• 13:59to this later,
• 14:00but
• 14:01common variants often individually have
• 14:04small effect sizes,
• 14:05whereas rare variants, which were
• 14:07the focus of Rick Lifton's
• 14:09major impact in hypertension,
• 14:11have are are are obviously
• 14:13rare, but have a bigger
• 14:14effect size. But when you
• 14:16analyze
• 14:17millions of cases, you can
• 14:18find the power to identify
• 14:20these common variants. And so
• 14:21this is sort of a
• 14:22summary of the last fifteen
• 14:24years of GWAS studies for
• 14:25coronary disease. I joke that
• 14:27GWAS studies are sort of
• 14:28the gift that keeps on
• 14:29giving because we just add
• 14:31another cohort to the meta
• 14:32analysis and lo and behold,
• 14:34it's another high impact paper.
• 14:35But the very first GWAS
• 14:37study for coronary disease was
• 14:38conducted in Europe. Europe. Just
• 14:39three thousand cases, five thousand
• 14:41controls,
• 14:42and they identified three loci
• 14:44associated with coronary disease. Most
• 14:46recently in in Nature Genetics,
• 14:48we published the million heart
• 14:49study. Right? A million cases
• 14:51and controls. We're at two
• 14:52hundred and forty loci. And
• 14:54pretty soon, we're gonna publish
• 14:55a meta analysis with the
• 14:56million veterans program. We'll be
• 14:58at five hundred loci. Right?
• 15:00The there's diminishing returns on
• 15:02more genetic,
• 15:03analysis. Right?
• 15:05What was really exciting about
• 15:07the the GWAS that we
• 15:09did is that when you
• 15:10look at these two hundred
• 15:11and forty one loci,
• 15:12over eighty percent are not
• 15:14associated with LDL cholesterol. Right?
• 15:16So for coronary disease, really
• 15:18the only preventive therapy we
• 15:19have is lipid lowering therapy.
• 15:21But of the two hundred
• 15:23and forty one loci, only
• 15:25about twenty percent are associated
• 15:26with LDL cholesterol at all.
• 15:28Eighty percent are not at
• 15:30all associated with lipid cholesterol.
• 15:32And on the left, you
• 15:33see the the lipid related
• 15:35loci, and those are very
• 15:36hypothesis validating. Right? They're they're
• 15:38genes that we all recognize
• 15:40in the lipid metabolism pathway,
• 15:41LDL receptor, APO c three,
• 15:43PCSK nine, HMG choroid ductase.
• 15:46On the right is sort
• 15:47of alphabet soup. Right? These
• 15:49are all genes that we
• 15:50don't really link to coronary
• 15:52disease or treat.
• 15:53And and just to sort
• 15:54of drive home that point,
• 15:55the number of lipid therapies
• 15:57we have for prevention
• 15:58are statins, ezetimibe, p c
• 16:00s k nine inhibitors, and
• 16:01this is now an expanding
• 16:02list.
• 16:03But non lipid therapies is
• 16:05sort of a graveyard
• 16:06of unsuccessful trials and unsuccessful
• 16:09therapies,
• 16:10each of which were very
• 16:11exciting at some point, but
• 16:13ultimately didn't amount to anything.
• 16:14So the motivation for, like,
• 16:16GWAS people like me was
• 16:17to to identify new drug
• 16:19targets. But that's been exceedingly
• 16:22slow as I'll I'll sort
• 16:23of get into, but we're
• 16:25very good at variant discovery,
• 16:27but function has only been
• 16:28identified for a handful of
• 16:29these variants. So a dozen
• 16:31of these two hundred and
• 16:32forty one are linked to
• 16:33any target gene or mechanism
• 16:35of action.
• 16:35And the reasons for that
• 16:36are that most of these
• 16:37variants are in noncoding DNA,
• 16:39so we don't know what
• 16:40gene they target or how
• 16:41they regulate it or in
• 16:43what cell type they're relevant.
• 16:44And that causal cell type
• 16:46for a complex disease is
• 16:47hard. Some diseases are driven
• 16:49by a single cell type,
• 16:50and the genetics is a
• 16:51little more straightforward.
• 16:53But for coronary disease, right,
• 16:54it could be any one
• 16:55of these cell types in
• 16:56the vascular wall, endothelial cells,
• 16:58vascular muscle cells, macrophages.
• 17:00And so, you know, when
• 17:02people take these individual variants
• 17:05from variant to function, it
• 17:06can be very powerful. And
• 17:08so here are some of
• 17:09my favorite examples, but,
• 17:11you know, the the most
• 17:12famous is b c l
• 17:13eleven a. So that was
• 17:14a single variant
• 17:16in the b c eleven
• 17:17a,
• 17:18gene in the promoter for
• 17:19that gene. That was associated
• 17:21with higher levels of hemoglobin
• 17:22f, and that is now
• 17:23a therapy
• 17:24that's being targeted for sickle
• 17:26cell disease. People have done
• 17:28this for all sort of
• 17:28colitis or obesity,
• 17:30but the power of these
• 17:32individual stories is now sort
• 17:33of they they take almost
• 17:35a decade to to to
• 17:37get through the experimental pipeline,
• 17:38and then they often don't
• 17:39go beyond that.
• 17:41So that's where polygenic risk
• 17:42scores come up is that
• 17:44instead of studying individual variants
• 17:45and their minuscule effects, why
• 17:47don't you just aggregate multiple
• 17:49risk variants and identify the
• 17:51highest risk parent patients for
• 17:53intervention? And that was always
• 17:54a hypothesis, but it was
• 17:55sort of underpowered until the
• 17:57GWAS data got big enough.
• 17:59And so this was from
• 18:00my colleague when, you know,
• 18:00I was a postdoc, Amit
• 18:02Cara, who was in Sake's
• 18:03lab, published this, like, beautiful
• 18:04paper showing that, you know,
• 18:06at the ends of the
• 18:06spectrum, so the the people
• 18:08at the highest polygenic risk
• 18:10percentiles
• 18:11for heart disease,
• 18:12AFib, diabetes, breast cancer. In
• 18:14each case, the people at
• 18:15each end of the spectrum
• 18:17were showing the greatest risk.
• 18:18Right? You could aggregate these
• 18:19variants and find the people
• 18:21at greatest risk. And in
• 18:22fact, the people at the
• 18:23at the in the top
• 18:25five percent
• 18:26had equivalent risk to those
• 18:27people who had Mendelian mutations.
• 18:29So an LDL receptor familial
• 18:31hypercholesterolemia
• 18:33patient had the identical level
• 18:35of risk as someone who
• 18:36just inherited
• 18:37three hundred small effect risk
• 18:39variants in LDL receptor pathways.
• 18:41More recently,
• 18:43they have improved this and
• 18:44show that there's even greater
• 18:46predictive capacity with these even
• 18:48improved polygenic risk scores.
• 18:50But
• 18:51at this point, you know,
• 18:52now five years after that
• 18:53paper, we we have this
• 18:55promise for polygenic risk scores
• 18:58that they could
• 18:59provide early detection of risk.
• 19:01They could inform treatment decision
• 19:03making, and they maybe could
• 19:05be relevant in drug development,
• 19:06right, for clinical trials. You
• 19:07could enrich patients using these
• 19:09scores.
• 19:10But the perils are well
• 19:11described too that there's limited
• 19:13predictive power.
• 19:15There's sometimes no benefit beyond
• 19:17nontraditional
• 19:18risk factors. Right? The additive
• 19:20power of these beyond just
• 19:21using age,
• 19:23race,
• 19:24sex, and, you know, with,
• 19:26hypertension and cholesterol numbers. Right?
• 19:28Just the Framingham risk score
• 19:30plus genetics. There's a minimal
• 19:32additive predictive value,
• 19:34and treatment interactions are unproven.
• 19:36And so,
• 19:38one of my mentors, Tommy
• 19:39Wang, right, published this editorial
• 19:41where he basically
• 19:42just took polygenic risk scores
• 19:44to task. He said genetic
• 19:45risk scores for cancer are
• 19:46unique
• 19:47since they identify
• 19:48specific mechanisms of disease. Right?
• 19:50Whereas for CAD, it's just
• 19:52a different set point for
• 19:53the current therapies. Right? You
• 19:55just are just treating people
• 19:56at different levels of the
• 19:57same therapy. I thought that
• 19:59was a a a pretty
• 20:00biting indictment of polygenic risk
• 20:01scores.
• 20:02And and, you know, and
• 20:03I'm excited about them. But
• 20:05his point was that for
• 20:06cancer, right, if you find
• 20:07a pathway,
• 20:08like a, you know, BRCA
• 20:10one pathway, then you would
• 20:11target therapies that are specific
• 20:14to that pathway.
• 20:15But for coronary disease, we
• 20:17haven't used these scores to
• 20:18identify biologically relevant pathways. And
• 20:20so I just repeat this
• 20:22slide. I'm super motivated to
• 20:24find the pathways that are
• 20:25relevant to someone who has
• 20:27a high polygenic risk. So
• 20:29I sort of took this
• 20:30as a challenge for my
• 20:31lab. Is that can we
• 20:33provide a better pathologic
• 20:35understanding of CAD risk variants
• 20:37to improve genetic risk prediction
• 20:39and then guide targeted therapies
• 20:41one day? That's, like, the
• 20:41ultimate goal of all this
• 20:43work.
• 20:44And so I I sort
• 20:45of come to my second
• 20:46approach is is how do
• 20:47we do this? And so
• 20:49I'd say that one of
• 20:50the more exciting things about
• 20:51genetics in the last ten
• 20:52years, and and I I
• 20:54I'm I'm sort of on
• 20:55my soapbox about this with
• 20:56every meeting I've I've had
• 20:58is that now functional genomics
• 21:00is can be highly systematized.
• 21:02So GWAS was a powerful
• 21:05technology because sequencing got cheap,
• 21:07and you could go to
• 21:08every biobank in the world
• 21:10and genotype everyone for twenty
• 21:12dollars a sample.
• 21:13But you couldn't study the
• 21:15biologic effect of thousands of
• 21:16variants at that price.
• 21:18But with pooled CRISPR technology
• 21:20and single cell RNA sequencing,
• 21:23you can basically do GWAS
• 21:24in a dish. You can
• 21:25study the effect of millions
• 21:27of variants
• 21:28individually in individual cells. And
• 21:30so that technology is called
• 21:32PerturbSeq. It was developed in
• 21:33a Viva Gev's lab in
• 21:35twenty sixteen.
• 21:36And I sometimes joke that
• 21:38at the Broad Institute where
• 21:39this was developed, there were
• 21:40two kinda hot technologies at
• 21:42the time. There was CRISPR
• 21:43editing and there were single
• 21:44cell RNA sequencing.
• 21:45So why not just combine
• 21:46them? Right? Just do them
• 21:47together. And that's sort of
• 21:48what this is, is you
• 21:49take a pool of cells,
• 21:50a million cells.
• 21:52You target them with a
• 21:53lentiviral libraries that introduces different
• 21:56CRISPR guide RNAs. Then you
• 21:57do single cell RNA sequencing
• 21:59of all that pool of
• 22:00cells.
• 22:01And in that sequencing, you
• 22:02not only find which CRISPR
• 22:04got into which cells, you're
• 22:05finding out which gene is
• 22:07knocked out, but you're finding
• 22:08the transcriptional effect of that
• 22:10knockdown. Right? And so people
• 22:12have been doing these pooled
• 22:13CRISPR screens for years for
• 22:15cancer therapy. They basically knock
• 22:17down every gene in the
• 22:18genome and then just see
• 22:19which cells grow.
• 22:20You know, they evade the
• 22:22the the therapy. So the
• 22:23first CRISPR screen was done
• 22:24for vemurafenib,
• 22:26which is a melanoma therapy.
• 22:27And so patients who get
• 22:29vemurafenib, there's a high level
• 22:30of resistance to vemurafenib.
• 22:32And,
• 22:33they did the first CRISPR
• 22:34screen to see which genes
• 22:36are mediating that resistance. So
• 22:38the cells that survive emirafenib
• 22:40are are mediating resistance.
• 22:41For coronary disease, we don't
• 22:43have a phenotype like that.
• 22:44Right? Like, for cancer, it's
• 22:46obviously which cells grow. For
• 22:47coronary disease, if we had
• 22:49to just prioritize,
• 22:55lamp post with this current
• 22:56screens we do. But with
• 22:57PerturbSeq,
• 22:58we're unbiased. We're just looking
• 23:00at the effects in in
• 23:02transcriptional
• 23:02space. And so,
• 23:04the Ragev Lab had applied
• 23:06this technology to fifty, sixty,
• 23:07seventy genes at a time,
• 23:09but we had thought, why
• 23:10don't we just do this
• 23:11for all the genes
• 23:13that are close to all
• 23:14the coronary artery disease GWAS
• 23:16loci?
• 23:17So when I started my
• 23:18lab, the first thing I
• 23:19did was made a library
• 23:20of all the genes that
• 23:21we wanted to study, the
• 23:23the two thousand genes we
• 23:24wanted to knock down. And
• 23:25so we built this library
• 23:27kind of by hand, and
• 23:28we took all genes within
• 23:29a megabase of every coronary
• 23:31artery disease GWAS locus, and
• 23:33that ended up being three
• 23:34thirteen hundred genes. Then we
• 23:36tried to include genes that
• 23:37were at other vascular relevant
• 23:39phenotypes, migraine headache, blood pressure
• 23:41control, blood clotting. That was
• 23:43another three hundred genes.
• 23:45Then we added four hundred
• 23:46genes and pathways we wanted
• 23:47overrepresented,
• 23:49like the VEGF pathway or
• 23:50the TGF beta pathway. Those
• 23:52turned out to be the
• 23:52wrong pathways to overrepresent, but,
• 23:54you know, so be it.
• 23:55And then we had three
• 23:56hundred negative control,
• 23:58genes from inflammatory bowel disease
• 24:00GWAS studies. A grand total
• 24:02of twenty three hundred genes.
• 24:03We had fifteen CRISPR guides
• 24:05per gene, thirty seven thousand
• 24:06six hundred and thirty seven
• 24:08guides. So my first month
• 24:09of being a PI,
• 24:11I and a technician hand
• 24:12cloned each of these guides
• 24:14into our lentiviral vector. Took
• 24:16us about a month of
• 24:16every day just like sort
• 24:17of, like,
• 24:19changing the bacterial media and
• 24:21things like that. But,
• 24:22we didn't know if this
• 24:23was gonna work. Right? We're
• 24:24gonna infect a million endothelial
• 24:26cells with thirty seven thousand
• 24:28guides and do single cell
• 24:29RNA sequencing on this big
• 24:30pool of cells, but it
• 24:31worked. Right? This is sort
• 24:32of a summary of the
• 24:33data we got. On the
• 24:34left, you see a cartoon
• 24:36of how we did the
• 24:36experiment.
• 24:37In the middle panel, each
• 24:39one of those dots on
• 24:40this UMAP plot is one
• 24:41of the cells that we
• 24:42perform single cell RNA sequencing
• 24:43on. And what you see
• 24:44is there's two hundred and
• 24:45fifteen thousand different dots, two
• 24:47hundred and fifteen thousand cells.
• 24:49Each one is targeted with
• 24:50a different guide RNA. We
• 24:51had about ninety cells per
• 24:53guide RNA. And so we
• 24:54have, you know, a grand
• 24:56total of of
• 24:57twenty thousand gene expression points
• 25:00for every single one of
• 25:00these single cells. So you
• 25:02see we created this matrix
• 25:03at the top right corner
• 25:05of cells per genes. And
• 25:07at this point, we can't
• 25:08analyze this data with an
• 25:09Excel spreadsheet. Right? This is
• 25:11a huge amount of data
• 25:12for a huge amount of
• 25:13genes. And so we had
• 25:14to think about how to
• 25:15analyze the data, and we
• 25:16went through many iterations. And
• 25:17with a great collaborator at
• 25:18Stanford, Jesse Engreits,
• 25:20we built a, we applied
• 25:22basically what's called topic modeling.
• 25:24And this was a algorithm
• 25:25that was developed by Twitter
• 25:26when they used to share,
• 25:28all their algorithms and computer
• 25:29science, pipelines. But if Twitter
• 25:32was gonna say that today,
• 25:33Yale University is trending, they
• 25:35don't just count the number
• 25:36of times someone mentions Yale
• 25:37University in tweets. They have
• 25:39a a topic
• 25:41of co expressed words
• 25:43with the word Yale. And
• 25:44that might be New Haven,
• 25:46that might be Connecticut, that
• 25:48might be Harvard, that might
• 25:49be, you know, Columbia, that
• 25:50might be, you know, oh
• 25:51oh, you know, Yale New
• 25:52Haven Hospital. That's that coexpress
• 25:54set, and and that's how
• 25:55they tell you what topics
• 25:56are trending. So we applied
• 25:58that same algorithm to our
• 25:59data. We wanted to look
• 26:00at coregulated
• 26:01gene networks.
• 26:03And the amount of data
• 26:04we had was incredible, and
• 26:05I could spend the next
• 26:07forty minutes, and I'd love
• 26:08to, going through each one
• 26:10of those discoveries that we
• 26:11made using this data, but
• 26:12I won't. I I I'll
• 26:14really resist. I'll just focus
• 26:15on the three kind of
• 26:16really clinically relevant and what
• 26:18I thought were the most
• 26:19interesting
• 26:20stories that came from this
• 26:21data. The first is that
• 26:23we can find the regulators
• 26:24and coregulated gene networks in
• 26:26an unbiased way from this
• 26:27sort of experiment.
• 26:29The second is that central
• 26:30genes for coronary disease are
• 26:32shared between multiple vascular diseases,
• 26:35and that's been sort of
• 26:36the motivation for my clinic
• 26:37now, and I'll get I'll
• 26:38get into that. Right? We
• 26:39often talk about ourselves as
• 26:40a cardiovascular division,
• 26:42but a lot of vascular
• 26:43patients see neurosurgeons
• 26:45for some reason. And so,
• 26:46you know, can we sort
• 26:48of,
• 26:49provide insights into other diseases
• 26:51that just aren't happen to
• 26:52be next to the heart?
• 26:54And then finally, I made
• 26:55a
• 26:56a a independent
• 26:57EC driven risk pathway
• 27:00to target with maybe precision
• 27:01therapies and see if it
• 27:03predicted risk in patient populations
• 27:05uniquely.
• 27:05And so those are the
• 27:06three stories I'll share with
• 27:08the last twenty minutes here.
• 27:09And so the first,
• 27:11is the just the raw
• 27:12data from our screen. So
• 27:14for a cardiology audience, I
• 27:15like to use this example.
• 27:16This is HMG coereoiductase.
• 27:18Right? The target of statins.
• 27:19So let's say you didn't
• 27:20know what this gene did.
• 27:22Even though I'm studying endothelial
• 27:24cells, this isn't even in
• 27:25the liver,
• 27:26our data would tell you
• 27:27that this is a major
• 27:28regulator of cholesterol biosynthesis. So
• 27:30in our data, you see
• 27:31that we had a hundred
• 27:32and sixty cells that got
• 27:34a CRISPR knockdown for the
• 27:35gene HMG cholera reductase, and
• 27:37we had six thousand cells
• 27:38that got a negative control
• 27:40guide. And so we got
• 27:40about fifty percent knockdown of
• 27:42that gene. Then when we
• 27:43look at the expression data,
• 27:45the effect of HMG co
• 27:46reductase
• 27:48knockdown,
• 27:48you see that there's up
• 27:49regulation of every gene in
• 27:50the cholesterol biosynthesis pathway. So,
• 27:50yes, HMG co reductase is
• 27:50a regulator
• 27:51of biosynthesis pathway. So, yes,
• 27:53HMG choroid ductase is a
• 27:55regulator of cholesterol biosynthesis. No
• 27:57surprise to anyone in this
• 27:58audience. And that was one
• 28:00of our topics. Right? You'll
• 28:01see here that topic nineteen
• 28:03kinda halfway down the left
• 28:04side of the the slide
• 28:06is the cholesterol biosynthesis
• 28:08pathway. In total, we had
• 28:10fifty of these coregulated networks
• 28:11that were identified from topic
• 28:13modeling.
• 28:14Thirty seven of them have
• 28:15nothing to do with endothelial
• 28:17cells. They're sort of ubiquitous
• 28:18pathways that would have been
• 28:19found in any cell type.
• 28:20Type. Right? Ribosomal biology or
• 28:22cell cycle control.
• 28:24But in the box are
• 28:25thirteen endothelial cell specific pathways
• 28:28that I'm sure are near
• 28:29and dear to someone like
• 28:29Dan Greif's heart here, like
• 28:31EndoMT
• 28:32or, angiogenesis here. And these
• 28:35pathways, I think, are the
• 28:37disease relevant pathways. So we
• 28:39wanted to see which of
• 28:40these showed enrichment for GWAS
• 28:42risk. And, again, we applied,
• 28:43you know, sort of a
• 28:44series of statistical tests to
• 28:45see which of these pathways
• 28:47were enriched for GWAS risk
• 28:49loci.
• 28:50And every test we applied,
• 28:52it was basically one pathway
• 28:54rose to the top, and
• 28:55that's what I'm showing you
• 28:56here. And I didn't know
• 28:57the name of this pathway
• 28:58when we found it from
• 28:59our data, but it's called
• 29:00the CCM
• 29:01pathway, the cerebral cavernous malformation
• 29:04complex pathway.
• 29:05It's named for Mendelian disorder
• 29:07that I'm gonna talk about
• 29:08in a second.
• 29:09But you'll see that there's
• 29:10genes upstream of KLF two
• 29:12and genes downstream of
• 29:14two. And in total, forty
• 29:16one of the GWAS hits
• 29:18for coronary disease are in
• 29:19this pathway.
• 29:20So forty seven of the
• 29:22GWAS hits for coronary disease
• 29:24are in the LDL pathway.
• 29:25Right? Something that we unquestionably
• 29:27say causes coronary disease. And
• 29:29here, just about the same
• 29:30number are in this CCM
• 29:32pathway.
• 29:33And it's regulated by three
• 29:36genes at the center of
• 29:37the CCM complex.
• 29:39And genes that are upstream
• 29:41of KLF two, so in
• 29:42the the top part of
• 29:43the slide, when you knock
• 29:44them down, they have a
• 29:45protective effect on coronary disease,
• 29:48and genes downstream of KLF
• 29:49two, when you knock them
• 29:50down, have the opposite effect.
• 29:51So this is, like, you
• 29:52know, kind of from our
• 29:53data, you're able to sort
• 29:54of see the directionality of
• 29:56the effect on disease.
• 29:57So then I had to
• 29:58read a lot of papers
• 29:59that came from Yale,
• 30:01on what is this pathway.
• 30:02And it turns out it
• 30:03was sort of discovered because
• 30:05when you have loss of
• 30:06function mutations in these genes,
• 30:08it drives a cerebral vascular
• 30:10disease called cerebral cavernous malformations.
• 30:12Extremely rare. One in a
• 30:13hundred thousand, one in five
• 30:15hundred thousand patients have it.
• 30:16But when they have large
• 30:18deletions in the CCM two
• 30:19gene, for example, they get
• 30:21these, these black spots on
• 30:23their MRI, which are blood
• 30:25blisters essentially. They're venous malformations.
• 30:27They're outpouchings
• 30:28of venous endothelial cells.
• 30:31But what we're finding from
• 30:32GWAS is that common mutations
• 30:34in the same pathway
• 30:36are somehow protective. So eight
• 30:38percent of Europeans
• 30:40have protective mutations
• 30:42in this pathway that lower
• 30:44their risk of coronary disease.
• 30:46And so we weren't sure,
• 30:47you know, do we have
• 30:47the directionality wrong? Is this
• 30:49really real? So we may
• 30:50knock out mice. Right? Something
• 30:52that I hadn't really done
• 30:53a lot of before,
• 30:54but we we we felt
• 30:55like, you know, I'd, sort
• 30:56of resisted going into, like,
• 30:58a biologic model system, but
• 30:59we really had to do
• 31:00it. And you see that
• 31:01here when you take a
• 31:02heterozygous CCM mouse, a homozygous
• 31:04knockout is lethal. Right? They
• 31:05don't make a vascular system.
• 31:07It's very involved in vascular
• 31:08development. But in heterozygous mice,
• 31:10in both male and female
• 31:11mice, there's a trend toward
• 31:13lower,
• 31:14atherosclerosis.
• 31:15And it's significant in the
• 31:16female mice who get more
• 31:17athero in general in this
• 31:19model.
• 31:20And then what was really
• 31:22interesting to us is we
• 31:23then looked at, like, well,
• 31:25this pathway hit on the
• 31:26left had been described
• 31:27in the literature.
• 31:29But could we find new
• 31:30regulators of this pathway? And
• 31:31were those new regulators
• 31:33also associated with coronary artery
• 31:35disease risk? And so what
• 31:36I'm showing you is that
• 31:37these are the forty one
• 31:39genes,
• 31:40that are, GWAS loci. These
• 31:42are CAD, GWAS loci. These
• 31:44are this is part of
• 31:45that GWAS pathway that we
• 31:46think are endothelial cell acting.
• 31:48And when you knock down
• 31:49CCM two, it has this
• 31:50sort of regulatory effect on
• 31:52all forty one of those
• 31:53genes. But there was a
• 31:54second gene that even had
• 31:55a bigger effect on these
• 31:56genes. It was called TLNRD
• 31:58one, and there were no
• 31:59papers on this gene. Right?
• 32:00So you you look it
• 32:01up in PubMed, not a
• 32:02single paper.
• 32:04And what we found is
• 32:06that this gene what you
• 32:07see on the left is
• 32:08that it's the strongest regulator
• 32:10of the CAD associated pathways.
• 32:12CCM two was number two,
• 32:13and HMG choroid ductase was
• 32:15number five. And it also
• 32:16showed the greatest correlation with
• 32:18CCM two knockdown. And all
• 32:20the other genes in black
• 32:21on this other, waterfall plot,
• 32:23all those other genes are
• 32:24known members of the CCM
• 32:26complex. T one r d
• 32:27one was sort of never
• 32:28discovered
• 32:28despite, you know, twenty years
• 32:30of research on t one
• 32:31r d one. So we
• 32:32then, you know, now getting
• 32:33into the basic biochemistry, we
• 32:35went to alpha fold. Right?
• 32:36Alpha fold had just come
• 32:37out and you could,
• 32:39make a crystal structure of
• 32:40a complex of proteins.
• 32:42And we we put two
• 32:43one r d one in,
• 32:44which is the orange protein
• 32:45at the bottom, and CCM
• 32:47two is the green protein.
• 32:48An alpha fold predicted that
• 32:50two one r d one
• 32:50would directly bind to this
• 32:52complex. So,
• 32:53and what what you see
• 32:54on the left is that
• 32:55it it was alpha fold
• 32:56was correctly predicting
• 32:58the CCM two crit one
• 32:59interaction,
• 33:00which is a known interaction
• 33:02for CCM two, and then
• 33:03we found this novel interaction.
• 33:05And then we confirmed it
• 33:06with immunoprecipitation
• 33:07experiments in this bottom right
• 33:08hand corner.
• 33:11And that's the only Western
• 33:12blot I will show during
• 33:13the rest of this talk.
• 33:16But from a functional biology
• 33:18standpoint, you know, we we
• 33:19now find this physical interaction
• 33:20between these two proteins, and
• 33:22we're seeing that they both
• 33:23have similar effects on how
• 33:25endothelial cells behave. So in
• 33:27the middle panel here, you
• 33:28see that when you put
• 33:29flow on endothelial cells, control
• 33:31endothelial cells, they all align
• 33:32in the same direction. They
• 33:34they develop this sort of
• 33:35protective phenotype.
• 33:36Those are happier quiescent endothelial
• 33:38cells when they're aligned to
• 33:39the direction of flow, and
• 33:41this is worked on by
• 33:42Martin Schwartz here at Yale.
• 33:43But when we knock out
• 33:44these two genes, they sort
• 33:45of have this similar effect.
• 33:46They cause this endothelial cell
• 33:48disarray,
• 33:50and they cause these accentuated
• 33:51stress fibers in the endothelial
• 33:52cells, but knockdown of these
• 33:54genes is protective.
• 33:55So we don't know why
• 33:57that is, but when we
• 33:57look at barrier function, knockdown
• 33:59results in increased barrier function
• 34:01in these endothelial cells. So
• 34:03what we think is happening
• 34:04is that these cells are
• 34:05resistant to flow,
• 34:07which normally is bad. Right?
• 34:08Responding to flow is good.
• 34:10When you exercise, the reason
• 34:12you vasodilate
• 34:13after exercise is you've raised
• 34:14your blood vascular
• 34:16cells, and then you you
• 34:17develop you express nitric oxide
• 34:19causing vasodilation. And I always
• 34:21like to think that, like,
• 34:22you've hit a reasonable,
• 34:24amount of exercise when you
• 34:25see that vasodilation.
• 34:26As people have hypertension for
• 34:28fifty years, they lose that
• 34:29response. So people who chronically
• 34:32hypertensive
• 34:33no longer vasodilate to exercise,
• 34:36and so they lose that
• 34:37protective endothelial effect.
• 34:39But if they have these
• 34:40mutations in their endothelial cells,
• 34:42they are resistant to flow.
• 34:44So they they might not
• 34:45vasodilate as much when they're
• 34:47younger, but they also don't
• 34:49lose the vasodilatory
• 34:50capacity that they have when
• 34:52they've been exposed to hypertension
• 34:53for fifty years.
• 34:54And so that's sort of
• 34:55our working hypothesis of the
• 34:57mechanism is that most of
• 34:58us in this room have
• 34:59atherosclerotic
• 35:00endothelium. Right? We're prone to
• 35:02all the bad things that
• 35:03happen in our endothelial cells,
• 35:06but a few of us
• 35:07have these genetically atheroprotective
• 35:09endothelial cells. Endothelial cells. We
• 35:10have these common variants in
• 35:12these two genes that then
• 35:13regulate forty one other risk
• 35:15genes, and we have those
• 35:16people have decreased vascular inflammation,
• 35:19increased nitric oxide production regardless
• 35:21of the flow conditions their
• 35:23cells are in. And so
• 35:24that, you know, that's a
• 35:25a mechanism that we don't
• 35:26target.
• 35:27You know you know, could
• 35:29we make drugs that sort
• 35:30of mimic the effect of
• 35:31exercise
• 35:32in the endothelium?
• 35:34Yes. And that's something that
• 35:35drug companies have tried to
• 35:37do, but, you know, maybe
• 35:38with human genetics, we can
• 35:39find the true nodal biology
• 35:42regulators of that pathway.
• 35:44So so from that, I
• 35:45wanna shift into what I
• 35:47sort of started the the
• 35:48talk on, which is
• 35:50endothelial cell risk scores. And
• 35:52can we identify a new
• 35:53risk pathway from this data?
• 35:56One of the criticisms of
• 35:57GWAS is that
• 35:58you you just sort of
• 35:59find an un,
• 36:01unassociated
• 36:02list of genes. Right? They
• 36:03have nothing to do with
• 36:04each other.
• 36:05And if you did it
• 36:06for every disease, so the
• 36:07GWAS hits for schizophrenia and
• 36:09the GWAS hits for coronary
• 36:10disease overlap for some reason.
• 36:12So does that mean we're
• 36:13just, you know, identifying the
• 36:14same sort of genes that
• 36:16tolerate genetic variation?
• 36:18But I'd like to say
• 36:19that when you look at
• 36:20the pathways in a
• 36:21single cell type, maybe you'll
• 36:23find,
• 36:25actual biologically relevant information. And
• 36:27so that's what we tried
• 36:28to do with endothelial cells.
• 36:30So we created two different
• 36:32genetic risk scores for for
• 36:34from the GWAS data that
• 36:35we had. We created an
• 36:36endothelial cell genetic risk score
• 36:38with those, thirty five SNPs.
• 36:40There were forty one, but
• 36:41thirty five were sort of
• 36:42reproducibly genotyped.
• 36:44That's from our perturbseq data.
• 36:45Right? The the the big,
• 36:47you know, very expensive experiment
• 36:48that took my lab four
• 36:49years. And then we took
• 36:51forty seven forty six SNPs
• 36:52that are associated with LDL
• 36:54cholesterol. And we went to
• 36:55the UK Biobank, and you
• 36:56see that if you create
• 36:57different categories of patients,
• 37:00these two scores are additive.
• 37:02That the people with the
• 37:03highest number of coronary disease
• 37:05events after twelve years of
• 37:06follow-up,
• 37:07They have bad endothelial cells
• 37:08and they have bad lipid.
• 37:10And and,
• 37:11you know, as independent risk
• 37:13factors,
• 37:14this is sort of convincing
• 37:15in these statin naive patients.
• 37:17But with LDL cholesterol,
• 37:19right, you can just measure
• 37:21LDL. You don't need a
• 37:22genetic risk score for LDL
• 37:24cholesterol. You can just measure
• 37:25the serum levels.
• 37:26But for endothelial cell dysfunction,
• 37:28there is no measurable factor.
• 37:30Maybe hypertension, but well, I'll
• 37:32show you that, you know,
• 37:33perhaps not hypertension. Maybe this
• 37:35is an independent,
• 37:36pathway. But, certainly, this endothelial
• 37:39risk score does not correlate
• 37:40at all with LDL cholesterol
• 37:41as you see here.
• 37:43And then when we look
• 37:44at UK Biobank,
• 37:46sort of, you know, the
• 37:47the the sort of classic
• 37:48table one,
• 37:49Our endothelial cell risk score
• 37:51doesn't really correlate with much.
• 37:53There is sort of a
• 37:54nominal correlation with hypertension,
• 37:57and a nominal correlation with,
• 38:00kidney function.
• 38:01But when you actually look
• 38:02at the numbers, it's pretty
• 38:03minimal. So the people with
• 38:04the best endothelial score, the
• 38:06lowest twenty percent, five point
• 38:08four percent of them have
• 38:10of hypertension,
• 38:11and the people with the
• 38:12worst,
• 38:12endothelial risk score, six point
• 38:14two percent have hypertension. Right?
• 38:16You know, I wouldn't say
• 38:17that that's really driving the
• 38:18biological effect of this score.
• 38:21And same thing with, the
• 38:23the eGFR.
• 38:24Right? It's it's though it's
• 38:25nominally significant, it's really a
• 38:27minimal direction of effect. And
• 38:28and this this is sort
• 38:29of the list of snips
• 38:30that we included in the
• 38:31score.
• 38:32But then when you again,
• 38:33in UK Biobank sort of
• 38:35create different levels of score,
• 38:37there is an EC risk
• 38:38score and a lipid risk
• 38:39score. The people far and
• 38:40away with the most events
• 38:42have a bad EC risk
• 38:43score and a bad lipid
• 38:44risk score.
• 38:45And then the second highest
• 38:48risk group is is bad
• 38:49EC risk score, but low
• 38:51lipid risk score. Right? So
• 38:52the EC risk score is,
• 38:53like, the most predictive
• 38:55of poor outcomes. And some
• 38:56of that might be because
• 38:57statins exist in this world.
• 38:59Right? When you have a
• 39:00bad lipid risk score, it's
• 39:01treatable, whereas the endothelial risk
• 39:02score is not. Right? So
• 39:03these are this is a
• 39:04unique group of patients.
• 39:06But we wanted to see
• 39:07if there was, like, a
• 39:07treatment interaction that was unique.
• 39:09So we went to the
• 39:09JUPITER trial. Right? And and
• 39:11all this genomic data is
• 39:12sort of housed, you know,
• 39:13paid for by drug companies,
• 39:15housed at the Brigham, and
• 39:16not touched at all. Right?
• 39:17No one studies these, genetic
• 39:19risk scores and true clinical
• 39:20trial data. So we thought
• 39:21it was a unique opportunity.
• 39:23And so we took our
• 39:23endothelial risk score, and we
• 39:25went to the JUPITER trial,
• 39:26which was rosuvastatin
• 39:27randomized to people who had
• 39:29never had a coronary artery
• 39:30event. So primary prevention of
• 39:32coronary artery disease.
• 39:34And what you see is
• 39:35this remarkable treatment effect. So
• 39:38the hazard ratio is point
• 39:39two eight. There's,
• 39:41a absolute risk reduction of
• 39:42two two point two
• 39:43percent, and there's this, gradient
• 39:46that the people with low
• 39:47endothelial risk scores,
• 39:48don't
• 39:49have much benefit from being
• 39:51treated with with rosuvastatin,
• 39:53but the people with really
• 39:54bad endothelial cells
• 39:56drive the strongest benefit from
• 39:57being treated with rosuvastatin.
• 39:59And when you compare that
• 40:00to, like, the other risk
• 40:01scores you can make, so
• 40:02you could take all the
• 40:03other variants that aren't in
• 40:04the EC pathway.
• 40:06There's obviously benefit, but there's
• 40:08benefit at every level of
• 40:09that risk score.
• 40:11And the top group doesn't
• 40:13drive the same degree of
• 40:15benefit in this study.
• 40:17And then if you take
• 40:18the lipid risk score and
• 40:19now this is the opposite
• 40:20of what we expected to
• 40:21see. We thought a lipid
• 40:22lowering drug would work best
• 40:24on people with high genetically
• 40:26driven lipids.
• 40:28And, again,
• 40:29every level of the lipid
• 40:30risk score, they derive the
• 40:31sort of same benefit to
• 40:33rosuvastatin.
• 40:34And even in the top
• 40:35category,
• 40:36this score actually performed the
• 40:37worst
• 40:38that,
• 40:39people with high,
• 40:41genetically driven high LDL have
• 40:43the the worst response
• 40:45to,
• 40:46to rosuvastatin compared to other
• 40:48genomic risk scores that were
• 40:50independent pathways.
• 40:51So sort of unexpected result.
• 40:53We weren't sure if it
• 40:54was true. So we went
• 40:55to the Fourier trial, right,
• 40:56the PCSK9 trial that our
• 40:58own knee hard to say
• 40:58or your own knee hard
• 40:59to say, I should say,
• 41:00unfortunately.
• 41:01Your own knee hard to
• 41:02say was one of the
• 41:02leaders of. And,
• 41:04obviously, you know, an orthogonal
• 41:06treatment for LDL cholesterol
• 41:09and,
• 41:09now secondary prevention trial, much
• 41:11sicker patients,
• 41:13who have have a higher
• 41:15event rate. And all these
• 41:16patients are also on a
• 41:17statin therapy. So that also
• 41:19kinda confounds the effect of
• 41:20lipids a bit, but, you
• 41:21know, we thought it was
• 41:22a a unique population to
• 41:23study. And, again, genetics are
• 41:25available in this population. And
• 41:26so,
• 41:28again, you see the the
• 41:29the effect sizes are smaller.
• 41:31They're, like, a bit attenuated,
• 41:33in each risk score. But
• 41:35on the left, you see
• 41:36the EC risk score and
• 41:37really no benefit in the
• 41:39low and middle risk categories,
• 41:40you know, very modest benefit
• 41:42with, nonsignificant
• 41:43hazard ratios.
• 41:45But in the highest risk
• 41:46category, there was the strongest
• 41:47absolute risk reduction of four
• 41:49point four percent. It was
• 41:51significant.
• 41:52Whereas in the non EC
• 41:53risk score,
• 41:55again, very minimal improvement,
• 41:57no gradient of benefit.
• 41:59And even more disappointingly, in
• 42:01the lipid risk score, it
• 42:02was it was a null
• 42:03association. There was no gradient
• 42:05of benefit.
• 42:06So we found this intriguing.
• 42:07Right? It was the opposite
• 42:08of what we wanted this
• 42:09show. I we thought it
• 42:10was a very simple paper
• 42:11to write that, you know,
• 42:12the EC risk score is
• 42:14completely independent of LDL biology.
• 42:18And, the the EC risk
• 42:19score should have its own
• 42:20therapies.
• 42:21And what we're sort of
• 42:22saying now, which I'll sort
• 42:23of dig into a little
• 42:24more, is that, well, the
• 42:25EC risk score is an
• 42:26independent re
• 42:27predictor of risk, but but
• 42:29it predicts the people who
• 42:30benefit the most from LDL
• 42:31lowering therapy. Right? And so,
• 42:33you know, that we have
• 42:34to sort of dig into
• 42:35the mechanism behind that a
• 42:36bit. And so we went
• 42:37back to UK Biobank, and
• 42:39what you see is each
• 42:40one of these are,
• 42:42you know, tens of thousands
• 42:43of patients who had different
• 42:45baseline LDL values in the
• 42:47UK Biobank.
• 42:48And so,
• 42:49people who had a LDL
• 42:51greater than one ninety
• 42:53for on the x axis
• 42:55is every level of polygenic
• 42:56risk score in the population.
• 42:58So,
• 42:59when when
• 43:00and this is just the
• 43:01endothelial cell polygenic risk score.
• 43:03And so people with very
• 43:04high LDL, there's this huge
• 43:06gradient of benefit. Right? They
• 43:07have the highest risk derived
• 43:10from bad endothelial cell function.
• 43:12And,
• 43:14but when people have normal
• 43:15or or even low LDL,
• 43:17like the blue line,
• 43:19there's actually almost no benefit.
• 43:21I mean, there's no gradient
• 43:22of risk from the endothelial
• 43:23risk score. So, again, I'm
• 43:24I'm sort of saying this
• 43:25point over again that the
• 43:27the interaction
• 43:28between endothelial cell function and
• 43:30either genetically low LDL or
• 43:32pharmacologically low LDL seems to
• 43:34be showing up in UK
• 43:35Biobank, the JUPITER trial, and
• 43:37the Fourier trial.
• 43:39And, again, this is just
• 43:40another way of showing this
• 43:41sort of interaction. So in
• 43:43UK Biobank,
• 43:45on the x axis is
• 43:45the baseline LDL. And what
• 43:47you see on the left
• 43:48side is people in UK
• 43:49Biobank who have a baseline
• 43:50LDL of fifty.
• 43:52There's no hazard from having
• 43:54a bad endothelial cell risk
• 43:55score.
• 43:56But on the right side
• 43:57are people with baseline LDLs
• 43:59of one ninety, and they
• 44:00have the strongest hazard.
• 44:02If you compare that to
• 44:04an LDL risk score,
• 44:06their risk is the same
• 44:07no matter what their baseline
• 44:08LDL is. Right? So,
• 44:10the the
• 44:13the the increased risk from
• 44:14having bad LDL
• 44:16is present at every level
• 44:17of of baseline LDL, a
• 44:19very, very sort of different,
• 44:20biologic effect. And so why
• 44:22could this be? Right? What
• 44:23what is the mechanism behind
• 44:24this? Is this real? Does
• 44:25this make any sense? And
• 44:26this is where we started
• 44:27looking at Bill Sessa's paper
• 44:28from Yale
• 44:29that, you know, he has
• 44:30this line in a lot
• 44:31of his papers that the
• 44:32transport of LDL cholesterol into
• 44:34the subendothelial
• 44:35space is the fuel for
• 44:36the fire that drives atherosclerosis.
• 44:38Right?
• 44:39LDL is high throughout your
• 44:41body,
• 44:42but only in,
• 44:44certain spots do you develop
• 44:46plaques. And, you know, we
• 44:47we've all had theories on
• 44:48that. It's turbulent flow in
• 44:50some places. It's vascular injury.
• 44:53But it seems to be
• 44:54that there's a strong strong
• 44:56correlate,
• 44:56strong interaction between LDL
• 44:59and endothelial cell function.
• 45:01And the people where LDL
• 45:03is the most toxic are
• 45:05the people with bad endothelial
• 45:06cells driven by their poor
• 45:07genetics. And so that's sort
• 45:09of the summary of our
• 45:10paper.
• 45:11Now we're trying to study
• 45:12this sort of process of
• 45:13LDL uptake using phenotypic screens.
• 45:16And I I've sort of
• 45:17talked about this with some
• 45:17of the scientists who I've
• 45:18met with today. And and
• 45:20and, you know, I'm happy
• 45:21to share this data, but
• 45:22we've run these, like, genome
• 45:23wide CRISPR screens. And that's
• 45:25what I when I say
• 45:25that, like, functional biology can
• 45:27now be done at scale
• 45:28just like genetics was ten
• 45:30years ago. This is what
• 45:31I mean. Right? You can
• 45:32study the effect of every
• 45:33single gene,
• 45:35on LDL uptake. On the
• 45:36left is endothelial cells. Every
• 45:38gene involved in,
• 45:40LDL uptake and endothelial cells.
• 45:41And these are LDL receptor
• 45:43knockout endothelial cells. So we're
• 45:44working with Bill Sessa and
• 45:46hopefully Carlos Fernandez,
• 45:48to look at what mechanisms
• 45:50of disease are driving this.
• 45:51And then on the right,
• 45:52because Wolfram's here, basically, I
• 45:54I included our our work
• 45:55of of knocking out every
• 45:57gene in the genome in
• 45:58hepatocytes. And you, again, you
• 46:00rediscover, like, the entire cholesterol
• 46:03biosynthesis pathway. You rediscover the
• 46:05triglyceride synthesis pathway,
• 46:07but then you find new
• 46:08hits and new regulators of
• 46:09these pathways with these sort
• 46:10of pool genomic approaches.
• 46:12So I find this very
• 46:13exciting. And, again, I'm sort
• 46:15of saying, I've I'm I'm
• 46:16taking my mechanistic slide that
• 46:18I showed you,
• 46:19fifteen minutes ago, and now
• 46:20putting on that the people
• 46:22with atherosclerotic endothelium
• 46:25tend to have high EC
• 46:26risk scores, and people with
• 46:28low EC risk scores, I
• 46:29would say, have genetically athero
• 46:30protective endothelium.
• 46:32And it's really a function
• 46:33of bear you know, the
• 46:34the mechanisms that are driving
• 46:35this are things that we
• 46:36wanna study over the next
• 46:38ten years in my lab.
• 46:40And so with that, I'll
• 46:41summarize sort of the approach
• 46:42we take and then, you
• 46:43know, leave time for questions.
• 46:45But I'm very excited about,
• 46:46like, the role of human
• 46:47genetics in clinical work, and
• 46:49I I fully admit that
• 46:50we've maybe
• 46:52overpromised
• 46:52and under under delivered on
• 46:54in some ways in certainly
• 46:56in cardiovascular disease. Right? We
• 46:57don't use clinical genetics as
• 46:59much as we had hoped
• 47:00by now.
• 47:01But I think that there's
• 47:03been two revolutions. The first
• 47:04is that you can
• 47:06identify a lot of novel
• 47:07variants from genome association studies.
• 47:09And so my lab likes
• 47:10to do these. We collaborate
• 47:12on all these big international
• 47:13studies
• 47:14to do GWAS for coronary
• 47:15disease and hypertension.
• 47:17But we're also informed by
• 47:19these rare diseases, thoracic or
• 47:20aortic disease and cerebral cavernous
• 47:23malformation. Those the the genes
• 47:25that drive those diseases
• 47:26overlap
• 47:27with the genes that drive
• 47:29the common diseases.
• 47:30And so my clinic now
• 47:31is almost entirely patients with
• 47:33cerebral cavernous malformations.
• 47:35And they often come see
• 47:36me. We'll talk for, like,
• 47:37forty minutes. I'll take, like,
• 47:38an eight generation family history,
• 47:40and then they'll leave and
• 47:42they'll say, and you're a
• 47:43neurosurgeon too? Because, like, that's
• 47:44who they're used to seeing.
• 47:45Right? And I'll have to
• 47:46apologize and say, no. No.
• 47:46No. I can't do anything
• 47:47about the lesions you have.
• 47:48But
• 47:49but mechanistically,
• 47:51we'll one day be able
• 47:52to predict who in your
• 47:53family is gonna have these
• 47:54lesions.
• 47:55And so,
• 47:56but I do think as
• 47:57cardiologists, right, the shared risk
• 47:58between seemingly unrelated vascular diseases
• 48:01is, like, a really powerful
• 48:02genetic tool and a powerful
• 48:04clinical tool that's very exciting
• 48:06to focus on. The second
• 48:07thing is, like, I think
• 48:08that every variant that's discovered
• 48:10should be run through one
• 48:11of these high throughput functional
• 48:12genomic screens. They're cheaper, they're
• 48:15accessible, and they're very fun
• 48:16to use. Right? You instead
• 48:18of picking an individual variant
• 48:19and hoping it has an
• 48:20effect, you study the effect
• 48:22of every variant in the
• 48:23genome and only study the
• 48:24ones that have the strongest
• 48:26effect.
• 48:26We never would have picked
• 48:28to study the CCM pathway
• 48:29if we were just randomly
• 48:30picking variants because those genes
• 48:32are ubiquitously
• 48:33expressed in every cell type.
• 48:34There's nothing about them that's
• 48:35endothelial cell specific,
• 48:37but the diseases associated with
• 48:39those variants are only in
• 48:40endothelial cells, interestingly enough.
• 48:43And what we hope is
• 48:44that instead of, like, identifying
• 48:46specific variants that are worth
• 48:47studying, we'll find causal genes
• 48:49and risk pathways for vascular
• 48:51disease. And that's sort of
• 48:52the the the the output
• 48:54of our work so far.
• 48:55And so I've I've sort
• 48:56of gone over all these
• 48:58these things that, you know,
• 48:59we find co regulated programs.
• 49:00We find shared mechanisms
• 49:02of disease. I think this
• 49:03endothelial cell risk score identifies
• 49:05unique mechanisms of risk,
• 49:07and it identifies the patients
• 49:08with the strongest response to
• 49:10lipid lowering therapies. And just
• 49:11to preempt a question from
• 49:13a student clinician in the
• 49:14audience. Right? So then would
• 49:15you say that really the
• 49:17endothelial cell,
• 49:18risk score just says we
• 49:19should start statins earlier. Right?
• 49:22And, yes, in this population,
• 49:23these are the people who
• 49:24you might start statins in
• 49:25when they were when they're
• 49:26in their twenties or thirties.
• 49:27Right? Because they're the ones
• 49:28who are driving greatest benefit
• 49:30from statins, and these are
• 49:31the patients who have the
• 49:32highest toxicity from high LDL.
• 49:35But also, the truth is
• 49:37that as cardiologists, right, we
• 49:38only start treating patients when
• 49:39they've had their first heart
• 49:40attack. Right? Prevention is never
• 49:42gonna catch up to secondary
• 49:43prevention.
• 49:44And so once you're treating
• 49:46a patient who's already had
• 49:47a heart attack, you sort
• 49:48of have to throw the
• 49:49kitchen sink at them. And
• 49:50that would be the motivation
• 49:51for having targeted therapies for
• 49:54this pathway as well. And
• 49:55then I'll I'll put in
• 49:56my sort of my my
• 49:58my plug that I think
• 49:59precision medicine will ultimately be,
• 50:02you know, at that intersection
• 50:03of of genetics and function.
• 50:05And so with that, I'll
• 50:06thank, you know, the the
• 50:08great members of my lab
• 50:09who've done all this work.
• 50:10I have wonderful collaborators.
• 50:12And,
• 50:13and I'll just end by
• 50:14saying that, you know you
• 50:15know, presenting here is, you
• 50:16know, been a real pleasure
• 50:17in front of a great
• 50:18audience, and, I'm excited to
• 50:20answer any questions you all
• 50:21have.
• 50:32Is there any questions in
• 50:33the audience?
• 50:37Have that back, and then
• 50:39I'll come back to
• 50:41John.
• 50:44Raj, that was
• 50:46Raj, that was awesome.
• 50:49I guess,
• 50:50a first question is if
• 50:52your polygenic risk score personally
• 50:54was higher for a different
• 50:55disease, would you be studying
• 50:56a different disease?
• 50:59Right. Yeah. Yeah. Exactly. I
• 51:00I got lucky that it
• 51:01was because I was already
• 51:02studying. Right? I guess you
• 51:03could say that.
• 51:04Yeah. No. I I definitely
• 51:06would probably that, yeah, I've
• 51:07been studying something else.
• 51:09So
• 51:10I guess I have a
• 51:11a number of questions, but
• 51:12I'll I'll stick with this.
• 51:13So,
• 51:15you know, it it's fascinating
• 51:17that the endothelial
• 51:18risk score, how it's interrelating
• 51:21with,
• 51:22LDL cholesterol with high cholesterol.
• 51:25What about with, other therapies
• 51:27for atherosclerosis? What about antihypertension?
• 51:29What about aspirin?
• 51:31Does that also predict them?
• 51:33Yeah. So sadly. Right? Like,
• 51:35we have access to the
• 51:36LDL data because it it
• 51:37was, you know, done as
• 51:39clinical trials.
• 51:40And therapies for aspirin and
• 51:42hypertension didn't genotype right at
• 51:44the time. So we don't
• 51:45have genomic information on that.
• 51:47I would predict
• 51:48that it would be even
• 51:49stronger for antihypertensive
• 51:51therapies, or for even aspirin,
• 51:54or, you know, maybe even
• 51:55in anti inflammatory therapies would
• 51:57have would have a a
• 51:58a a correlation as well.
• 52:01But,
• 52:02we don't have genomic information.
• 52:03We're we tried to get
• 52:05access to CANTOS, and Novartis
• 52:06did not allow it. Right?
• 52:08Because they were still doing
• 52:09the clinical trial for lung
• 52:10cancer.
• 52:11And, you know, they didn't
• 52:12wanna, like, kinda change the
• 52:14the story that was coming
• 52:15out for those anti l
• 52:16one beta therapies. But, it's
• 52:18just a lack of data.
• 52:19The reason we haven't studied
• 52:20that. And then doing small
• 52:22follow-up. So do you think
• 52:23the the effect then,
• 52:26based on the LDL level
• 52:27of the statin is independent
• 52:29then of its anti cholesterol?
• 52:31Do you think it's more
• 52:32of an anti inflammatory effect?
• 52:34Or Right. Yeah. Yeah. So,
• 52:36right. Like, one of the
• 52:37effects of statins is that
• 52:38it raises KLF two, which
• 52:40is that forty that the
• 52:41forty one genes are in
• 52:42that KLF two regulatory pathway.
• 52:44And so maybe there's a
• 52:45direct effect of the stat
• 52:47that's protective. But then we
• 52:48looked at PCSK nine inhibitors,
• 52:49which don't have that Klf
• 52:50two effect, and it was
• 52:51the almost the same association.
• 52:53And so I think that
• 52:55it's probably minimally
• 52:59the, the the direct effect
• 53:00of the drug, and it's
• 53:01more the toxicity of LDL
• 53:04in patients with dysfunctional endothelial
• 53:06cells. That that it's just
• 53:08that bad endothelial cells predisposed
• 53:10to plaque formation.
• 53:11And when we think about
• 53:12it, like, you know, what
• 53:14is, who are the people
• 53:15who are resilient to coronary
• 53:17disease? Right? The patients you
• 53:18have who smoked for fifty
• 53:19years or had hypertension
• 53:21or, you know, even your
• 53:22familial hypercholesterols
• 53:24FH patients who never get
• 53:26coronary disease. There are some
• 53:27people like that. They must
• 53:28have some endothelial resilience.
• 53:30And so I think this
• 53:31this polygenic risk score would
• 53:32probably capture some component of
• 53:34that.
• 53:39That was an awesome talk.
• 53:42And,
• 53:44as a vascular biologist, I
• 53:45like your endothelial cell centric
• 53:48approach to risk scores. I
• 53:49I think that's terrific.
• 53:52But,
• 53:53one thing I would say
• 53:54is
• 53:54I remember about fifteen years
• 53:56ago, maybe even a little
• 53:57bit more, Joe Lascalso put
• 53:59a slide up, at a
• 54:02small scientific meeting, and it
• 54:03was one and you've probably
• 54:05seen him do this. It
• 54:06was one of
• 54:08those very complicated
• 54:09system circuitry
• 54:11slides with
• 54:12four hundred arrows going in
• 54:14lots of different directions.
• 54:16And he said,
• 54:17a human being is not
• 54:18an inbred mass.
• 54:20So you all are knocking
• 54:22out genes in inbred animals,
• 54:24and there's so many gene
• 54:26gene interactions and gene gene
• 54:28modifiers.
• 54:29And so I guess, you
• 54:31know, I'm trying not to
• 54:32make this too long a
• 54:33question, but it gets to
• 54:35the point of even your
• 54:37risk scores based on a
• 54:39set of genes, right, that
• 54:40you define, which is beautiful
• 54:42work and incredible.
• 54:44But are we ever gonna
• 54:45get there until we are
• 54:47able to have this incredibly
• 54:47complicated scorecard where everything goes
• 54:47into
• 54:49the
• 54:59variants that are identified are
• 55:01noncoding
• 55:02regions.
• 55:03A lot of those noncoding
• 55:04regions you know better than
• 55:05I do
• 55:06are influential
• 55:08in a lot of different
• 55:09ways. I mean, interchromosomal
• 55:11interactions,
• 55:13epigenetics
• 55:14based on sequence,
• 55:15noncoding RNAs, etcetera, etcetera.
• 55:18So
• 55:19how do you think about
• 55:20Yeah. The noncoding
• 55:22sequence, and what about this
• 55:24very complicated
• 55:25Yeah. Process? Right. Right. Yeah.
• 55:27So I completely agree that,
• 55:28like, you know,
• 55:29this is all,
• 55:30you know, a reductionist
• 55:32sort of lens on a
• 55:33systems problem.
• 55:35And I'm telling you, like,
• 55:37one,
• 55:38conclusion from our data is
• 55:40that, oh, there's this direct
• 55:41interaction between LDL and endothelial
• 55:43cells. But an alternative theory
• 55:45is that, you know, these
• 55:46are just this is just
• 55:46a complex system. And if
• 55:48it's perturbed in any way,
• 55:49you're gonna see additive effects.
• 55:51And I'm just looking at
• 55:52two different
• 55:53parts of the system diagram.
• 55:55And, it would be the
• 55:56same if I had an,
• 55:58inflammation risk score, and it'd
• 55:59be the same if I
• 56:00had a vascular muscle risk
• 56:01score. And so we we'd
• 56:02have to investigate that. So
• 56:04I I think we're trying
• 56:05to build those things, and
• 56:06and, you know, the right
• 56:07set of controls don't exist
• 56:08to really answer your question.
• 56:10As far as noncoding. Right?
• 56:12So,
• 56:13you know, we're trying to
• 56:14actually identify the causal genes
• 56:16using this perturb seek method.
• 56:18But your point is that,
• 56:19you know, that's still assuming
• 56:21that it's just one gene,
• 56:22one effect for risk variant,
• 56:24but maybe the risk variants
• 56:25have different functions in different
• 56:27stages of disease.
• 56:28Maybe they have different functions
• 56:29in different cells at each
• 56:30stage of disease and that
• 56:31we cannot model in a
• 56:33dish. So I've I've sort
• 56:34of told people that, you
• 56:35know, we just got this
• 56:36PPG to do in vivo
• 56:38for TRP seq where we
• 56:39can take a a CRISPRi
• 56:40mouse,
• 56:41inject the same guide RNA
• 56:43library, and then do single
• 56:44cell RNA sequencing of the
• 56:45different cell types that got
• 56:47the the guides. And then,
• 56:48you know, the criticism of
• 56:49the work I've done is
• 56:50that it's just some cells
• 56:51in a dish, but now
• 56:53we can really do this,
• 56:54you know, in a systems
• 56:56system.
• 56:56Yes.
• 56:58That was a great talk.
• 56:59And my question is in
• 57:01some
• 57:02in some degree
• 57:03relates to Jeff's question also.
• 57:06So you
• 57:07are
• 57:08we talked about,
• 57:09implications for therapy.
• 57:12And I was looking at
• 57:13the graphs and prevalence of
• 57:14disease that we were talking
• 57:15about over a ten year
• 57:16period.
• 57:17What's the predictive positive predictive
• 57:19value of this this polygenic
• 57:21risk scores, whether it's lipid
• 57:24or endothelial
• 57:25because we're going to apply
• 57:26them to individual patients, not
• 57:28to a population.
• 57:30Right. Right. Yeah. So it's
• 57:31small. Right? So the the
• 57:33positive predictive value of, like
• 57:34so someone at the top,
• 57:37the top range of polygenic
• 57:38risk score has about two
• 57:41to five fold higher risk.
• 57:42Okay? And then it it
• 57:44so the positive predictive value
• 57:45is dependent on which population
• 57:47you're studying in. And if
• 57:48you're it's a twenty year
• 57:49old, right, two to two
• 57:50to five fold high risk
• 57:51is just a positive predictive
• 57:53value of maybe two two
• 57:54percent. Right? Because most of
• 57:55those people aren't gonna get
• 57:56coronary disease. If you apply
• 57:58to a secondary prevention population
• 57:59in, like, the four year
• 58:00trial, the positive predictive value
• 58:02is much higher.
• 58:03But, yeah, the incremental that
• 58:05that value is small, but
• 58:06it it was I I
• 58:08think the the take home
• 58:09for me was that there
• 58:10was a treatment interaction, and
• 58:11that was only seen with
• 58:12one pathway, but not others.
• 58:15And so if we could
• 58:16refine the treatment interaction, then
• 58:18we could personalize treatment decisions.
• 58:20But in just predicting who
• 58:21gets disease or not,
• 58:23you know, we have better
• 58:24metrics for that. Age,
• 58:26serum cholesterol, hypertension,
• 58:28you know, the the Framingham
• 58:29risk score is, you know,
• 58:30much better at that than
• 58:31these risk scores.
• 58:33So, Raju, very nice talk,
• 58:35and,
• 58:35I have two questions.
• 58:37One is that, actually, you
• 58:38showed very nicely, actually, that,
• 58:41for example, IRX three Mhmm.
• 58:43Is actually a gene for
• 58:44obesity, while actually the variant
• 58:45is in the FTO gene.
• 58:47Right? So means that, actually,
• 58:49the
• 58:50locals doesn't mean gene itself.
• 58:52In a per terpsi, you
• 58:53actually went after the
• 58:55gene within the loci, not
• 58:56knowing that each is this
• 58:57variant actually regulates Sure. Or
• 58:59not. So I was wondering
• 59:01actually it would be actually
• 59:02based anything, maybe the next
• 59:03step to go rather than
• 59:04actually in per terpsi.
• 59:06And then the second question,
• 59:07when you're cleaning, when you're
• 59:08so calm, you actually always
• 59:10think about, you know, cascade
• 59:11screening, you know, etcetera.
• 59:14While the polygenic,
• 59:15you know, is gonna be
• 59:16distributed, it's not gonna be
• 59:17actually segregating
• 59:19perfectly. Right. And, also, often,
• 59:21you have actually patients to
• 59:22actually have families.
• 59:23Mhmm. So have you looked
• 59:25at to see actually
• 59:26how many how actually effective
• 59:28is this use use of
• 59:29this,
• 59:30kind of of yeah. Yeah.
• 59:31Yeah. Great. So, like, yeah.
• 59:33Great. Very basic question and
• 59:34very clinical questions. I'll try
• 59:36to try to switch switch
• 59:37gears as quickly as you
• 59:38can. So the first is
• 59:40the the the
• 59:41the true way that we
• 59:43generate the these risk of
• 59:45variance. Right? So, yes, we
• 59:47perturbacy targeted the genes.
• 59:49And then the way we
• 59:50look for enrichment, the way
• 59:51we found the SNPs is
• 59:53we we have this computational
• 59:54pipeline that I did not
• 59:55bore you with, but it's
• 59:56called variant to gene to
• 59:57program.
• 59:58And so we actually start
• 60:00with the snips,
• 01:00:01make a list of candidate
• 01:00:02genes,
• 01:00:03like, so every gene that
• 01:00:04could be regulated. And we
• 01:00:05use computational,
• 01:00:07algorithms for that, activity by
• 01:00:09contact, which was developed by
• 01:00:10Jesse Angrance's lab or,
• 01:00:12EQTLs.
• 01:00:13We we have a list
• 01:00:13of candidate genes. Right. And
• 01:00:15then we look for the
• 01:00:16pathways where those candidate genes
• 01:00:17are enriched.
• 01:00:18And so we are
• 01:00:20going back to the variant,
• 01:00:21the non coding variant.
• 01:00:23And but you so so
• 01:00:24so only seventy percent of
• 01:00:26the time is our causal
• 01:00:27gene the closest gene. Right.
• 01:00:28So that so so so
• 01:00:29do account for that. Would
• 01:00:31base editing be a better
• 01:00:32way? Theoretically,
• 01:00:34but you would have to
• 01:00:35sequence a lot more cells
• 01:00:36because the effect sizes are
• 01:00:38so small. So, you know,
• 01:00:39we got away with only
• 01:00:40ninety cells per gene, and
• 01:00:41it still cost us eighty
• 01:00:42thousand dollars to do this
• 01:00:43experiment.
• 01:00:44If we were gonna do
• 01:00:45base editing, this experiment would
• 01:00:47be impossible to perform.
• 01:00:49The clinical question k. This
• 01:00:51is, like, a few shifts.
• 01:00:52So, you know, are we
• 01:00:54able to actually apply these,
• 01:00:55you know, since they don't
• 01:00:56segregate? Like, do they have
• 01:00:57any clinical predictive value?
• 01:01:00So
• 01:01:01I'm a little biased. I
• 01:01:02don't think for coronary disease,
• 01:01:03the polygenic risk score is
• 01:01:05that useful. I think this
• 01:01:07possibility that
• 01:01:08we will one day have
• 01:01:09targeted therapies or it could
• 01:01:10motivate the generation of an
• 01:01:12endothelial cell specific therapy, that's
• 01:01:14where the excitement is.
• 01:01:16But for other vascular diseases,
• 01:01:18kind of building on this
• 01:01:19framework, I do think this
• 01:01:20is actually super interesting. So
• 01:01:21I see a lot of
• 01:01:22people with thoracic aortic aneurysm,
• 01:01:24most of whom don't have
• 01:01:25a Mendelian cause, but the
• 01:01:27polygenic risk score
• 01:01:29is kind of providing that
• 01:01:30they have, like, just the
• 01:01:31polygenic risk of, like, dysfunctional
• 01:01:33and extracellular matrix. And so
• 01:01:35it gives them some sort
• 01:01:37of, you know, peace of
• 01:01:37mind that they don't have
• 01:01:39a Mendelian variant that they're
• 01:01:40passing down in their family,
• 01:01:41but they do have genetic
• 01:01:42risk as they suspect it.
• 01:01:44And so for some of
• 01:01:45those diseases, I think the
• 01:01:46benefit is more. But for
• 01:01:47coronary disease,
• 01:01:48I think our other risk
• 01:01:49scores are are adequate.
• 01:01:53We only have a few
• 01:01:54minutes. I I want,
• 01:01:56John to ask perhaps the
• 01:01:57last question. I'll have a
• 01:01:59follow-up. I can just cut
• 01:02:00to this. Go. That's fine.
• 01:02:01Sure. Mhmm. That's really good.
• 01:02:03Thank you.
• 01:02:04The sequencing is in blood
• 01:02:06endothelial cells.
• 01:02:08Sorry. The perturbed sequencing?
• 01:02:11No. And for the patient,
• 01:02:12both the.
• 01:02:13Right. Yeah. All blood. Exactly.
• 01:02:15So, you know, it's
• 01:02:16true for.
• 01:02:17Yes. But how does somatic
• 01:02:19No. Right. And how does
• 01:02:21that play into your polygenic
• 01:02:23risk score? Yeah. Completely unknown.
• 01:02:25Right? So,
• 01:02:26we're really only powered to
• 01:02:28find the effects of germline
• 01:02:29mutations that are easier.
• 01:02:31But somatic variation, I think,
• 01:02:33drives disease as people have
• 01:02:34shown with CHIP. But, even
• 01:02:37like like, what you know?
• 01:02:38So CHIP is one example
• 01:02:39of somatic variation in the
• 01:02:40blood. But what about somatic
• 01:02:42variation in the blood vessel?
• 01:02:43We'll never be able to
• 01:02:44explain that. And I'll say
• 01:02:45that, you know, this pathway
• 01:02:46that I've implicated, the CCM
• 01:02:48pathway,
• 01:02:49it's very well known
• 01:02:50that the reason people form
• 01:02:52these lesions is they have
• 01:02:53one germline mutation,
• 01:02:55and then they have a
• 01:02:55second somatic variant. And maybe
• 01:02:57that's the same with coronary
• 01:02:58disease. Right? It's, you know,
• 01:02:59the the sort of a
• 01:03:00combination of the two. And
• 01:03:02so as a patient. Well
• 01:03:03Right. Exactly.
• 01:03:05Right. Right. So so, you
• 01:03:06know, maybe one day, right,
• 01:03:08you know, the the the
• 01:03:09direct sequencing for somatic variation
• 01:03:11and enough coronary plaques will
• 01:03:13answer that question, but we
• 01:03:14haven't even touched that. Right?
• 01:03:16So first of all, thank
• 01:03:18you so much for this
• 01:03:19wonderful talk. I had a
• 01:03:20very brief kinda observation question,
• 01:03:22maybe more into the application
• 01:03:24perspective. So I found,
• 01:03:27your,
• 01:03:28you sharing the data from
• 01:03:30Fourier and Jupyter really
• 01:03:32quite quite interesting. I think
• 01:03:33one of the challenges you've
• 01:03:34identified is the importance of
• 01:03:35developing new targets for
• 01:03:37new new mechanisms that we
• 01:03:39are even unaware. So
• 01:03:41have you tried to turn
• 01:03:42this on the on on
• 01:03:43its head a little bit?
• 01:03:44Because arguably,
• 01:03:46it would have made,
• 01:03:48Nihar's career much shorter if
• 01:03:50he hadn't enrolled eleven thousand
• 01:03:52patients. But perhaps
• 01:03:54by defining
• 01:03:55this in this, you know,
• 01:03:57this EC score, you could
• 01:03:59have identified a pathway to
• 01:04:01either fail quickly or succeed
• 01:04:03rapidly,
• 01:04:04with regards to targeting a
• 01:04:06population risk. And and I'm
• 01:04:07curious if you've been thinking
• 01:04:09about applying this even though
• 01:04:11it's not driving
• 01:04:12a new target
• 01:04:14to a mechanism of evaluating
• 01:04:16therapeutics quicker. Right. Right. Yeah.
• 01:04:18So that is actually kinda
• 01:04:19one of the exciting applications
• 01:04:21of this is that
• 01:04:22could you just enrich
• 01:04:24your your clinical trial population
• 01:04:26for people at the extreme
• 01:04:27of a polygenic risk score?
• 01:04:28Then instead of enrolling eleven
• 01:04:29thousand, you enroll four thousand,
• 01:04:31and it makes cardiovascular trials
• 01:04:33much cheaper. So, you know,
• 01:04:34maybe we would have more
• 01:04:35therapies in the space.
• 01:04:37And I think one thing
• 01:04:38we were excited about is
• 01:04:40that, normally, for a lipid
• 01:04:41lowering therapy, you would have
• 01:04:42used a lipid risk score.
• 01:04:44But to your system's biology
• 01:04:46point that, you know, maybe
• 01:04:47you have to use orthogonal
• 01:04:48risk scores or for different
• 01:04:49therapies, you have to use
• 01:04:50different risk scores to really
• 01:04:51find which one truly enriches
• 01:04:54for the highest risk patient
• 01:04:55population.
• 01:04:56I would argue that future
• 01:04:58lipid lowering therapy should enrich
• 01:04:59for a bad EC function
• 01:05:01and not bad LDL function.
• 01:05:02But it might be also,
• 01:05:06specific to each therapy. And
• 01:05:07until we really understand the
• 01:05:08mechanism, I don't know if
• 01:05:09I can convince a drug
• 01:05:10company to put all their
• 01:05:11eggs in this basket. But
• 01:05:13if we knew the mechanism
• 01:05:14and we, you know, we
• 01:05:15could convince them that, yes,
• 01:05:16this makes sense, you know,
• 01:05:17I think that's where I
• 01:05:18just could go. Well, I
• 01:05:20I think it'd be a
• 01:05:20wonderful project for a certain
• 01:05:22someone to do to evaluate
• 01:05:24how much money could have
• 01:05:25been saved if that was
• 01:05:27applied. Right. And use that
• 01:05:29in, in discussions with our
• 01:05:30with our colleagues in in
• 01:05:31industry. This was wonderful. Thank
• 01:05:33you so much on on
• 01:05:34behalf of all else.
• 01:05:40As everyone's leaving, I just
• 01:05:41wanna remind everyone that tomorrow
• 01:05:43morning, we have medicine,
• 01:05:44grand rounds, and our own
• 01:05:46John Forrest, will be speaking.
• 01:05:48So do your very best
• 01:05:50to attend and