Using Large-Scale Clinical Data for Discovery In Multiple Sclerosis And Epilepsy

April 30, 2021

ID6551

To CiteDCA Citation Guide

00:00It's my pleasure to now introduce our
00:03next speaker Doctor Chris Christie,
00:05office Doctor Kostakis,
00:07graduated from the Imperial College
00:09London and earned his PhD at the
00:12University of New South Wales in Sydney.
00:15After postdoctoral training at
00:16the Broad Institute and MTH,
00:18where he undertook some of the
00:21first Genome wide Association
00:22studies in autoimmune disease,
00:24he joined Yale in 2010.
00:26His laboratory uses genetics,
00:28genomics and epidemiological
00:29approaches to identify the
00:31biology underlying autoimmune.
00:32And neurological diseases
00:33dersum floor is yours,
00:35Doctor Kostakis.
00:39Thank you Nicole.
00:40I'm afraid I can't start my video
00:42so it's the tech people would like
00:44to start that, but it's fine.
00:48There we go. Hello everyone,
00:49I also have no disclosures and I
00:51would like to take the next few
00:53minutes to tell you a little bit
00:55about some of the things we've
00:57been thinking about in my lab.
01:02Specifically, how to use large
01:04scale data of both clinical and
01:07genetic data to make discoveries in
01:10diseases that we're interested in,
01:12specifically multiple sclerosis and epilepsy.
01:16And I chose these two projects
01:18to talk about because they are
01:20quite early on in their inception,
01:22so we're still not 100% sure
01:25what the story is,
01:26but I think it is instructive to
01:29look at what we can do with data.
01:33So the first story is multiple sclerosis.
01:36This is a large scale project
01:39out of the from the EU,
01:41primarily funded by one of
01:44the EU horizon programs.
01:45It's led by colleagues at the
01:48Carolyn Skirt Institute in Sweden,
01:50and it covers 10 countries,
01:52including site at Yale and a site at UCSF,
01:56and all the other partners are
02:00in in Europe and our main.
02:03Or our main focus is on multiple sclerosis,
02:08which is a predominantly autoimmune
02:10disease of the brain where the immune
02:14system basically decides that it does not
02:17like the myelin sheath around white matter.
02:21In neurons you get
02:23infiltration of immune cells.
02:25Stereotypically T cells that
02:27cause myelin stripping around.
02:30Blood vessels in the brain and
02:32you get these large lesions in
02:35the brain and progressively get
02:37more and more lesions overtime.
02:39And that leads to a relapsing remitting,
02:42usually mode of disease,
02:43where there is both physical and
02:45cognitive decline and eventually
02:47this becomes permanent and
02:49patients experience ongoing and
02:51progressive disability.
02:52It is a lifelong disease.
02:54There are disease modifying therapies,
02:56but there exists no cure and
02:58it is one of the more common
03:02neurological diseases out there.
03:04And like most such diseases,
03:06you can find some families where the
03:09disease appears to run in the families,
03:12but most cases are sporadic.
03:14It looks like familiar landmass
03:16is not a single gene form of Ms.
03:19It is exactly like the sporadic
03:21form it is polygenic.
03:23It is extremely complex.
03:24We have at least 200 loci mapped from large
03:28scale genome wide Association studies.
03:31We estimate there's probably another
03:33800 to 1000 out there in the genome,
03:36and a large effort now across the
03:39world is has been initiated to try
03:42and figure out what those genes do.
03:45But also to see how we can use
03:47some of this information and one
03:50of the problems has been that
03:52this disease is quite common,
03:54but it's not type 2 diabetes comma,
03:57so it's about one in 1000
03:59in European population,
04:00so it's fairly common,
04:01but no one really has a cohort of
04:0420 or 30,000 patients who have all
04:07been seen for a very long time in one
04:10clinic where data have been collected.
04:13In the same way,
04:15by the same people.
04:16And So what you have to do is Unite
04:19data across many centers.
04:22Often with differing practices
04:24with differing CHRS,
04:25or before that just paper records
04:29and try and put these data
04:32together in some meaningful way.
04:36So you can make large scale inferences
04:38and this goes back to what IRA
04:41said initially about how even as
04:43a biobank we need to be one of
04:45the network of biomax and this is
04:48very much what we've been trying
04:50to do in a disease focused way,
04:52and so this project has been
04:54aiming to do exactly that,
04:56and then from these large scale data,
04:58try and see if there are subsets
05:00of patients who seem to respond
05:02differently to therapy who seem to have
05:05different outcomes that is predictable.
05:07And that might maybe mechanistic because
05:10the problem is like most complex diseases,
05:13Ms is extremely heterogeneous at diagnosis.
05:16There is effectively no prognosis that
05:18one can give to a patient 'cause they may
05:22be severely disabled within five years,
05:25or they may be just fine 20
05:28years down the line,
05:30it's very hard to tell anything
05:32to tell a patient anything,
05:35and that is a major issue.
05:38And So what we've been doing is we have
05:41been warehousing both clinical and
05:43genetic data across these collections.
05:46And what I'm showing you so far
05:49is the progress we've done.
05:52We've made with the genetic data,
05:54which is about 45,000 Ms patients across
05:5810 centers and 26,000 controls to date.
06:01And this by itself has been
06:03a fairly major nightmare,
06:05not least of which has been the paperwork
06:08becauses the GDP are the privacy law
06:10that is come into effect in Europe,
06:13has really done a number of this on
06:16this and we've had like a major.
06:19It took us a year and a half to unwind
06:23the legal implications of that,
06:26but these are real issues that
06:28will have to be faced when we think
06:32about federations of biobanks,
06:34or of case control cohorts across places,
06:37and we're also trying to Unite clinical data.
06:41We have about 60,000 patients worth of
06:44clinical data with different amounts
06:46of data for different patients.
06:49And we are still trying to resolve those,
06:51and ultimately what we want to be
06:54able to do is to build predictors
06:56of outcomes which we have captured
06:58in the clinical data using both.
07:01Other data that we have on the data
07:03on the patients and the genetic data.
07:07Just the genetic data,
07:09which is a fairly standard platform.
07:13Gina type this.
07:14The vast majority of this is
07:16genotyping relevant sequencing.
07:17There are different platforms
07:19on which one can genotype,
07:21but they're fairly standard.
07:22It's a fairly homogeneous data type.
07:24It has taken us about a year to put
07:27these data together because there
07:29is a pretty significant amount of
07:31work involved in actually Q seeing
07:34and processing data,
07:35and so just that has been
07:37a nontrivial challenge.
07:39We have now overcome this.
07:40We now have this unified collection.
07:43Unlike most case control cohorts where
07:45we do genome wide Association studies,
07:47we actually have deeper information
07:49rather than just whether someone
07:51is a case or a control,
07:52and we're now trying to put these
07:55data together so this the next couple
07:57of years I think are going to be
08:00very exciting here as we try and
08:02figure out if there are predictors
08:04for both outcomes and treatment
08:07outcomes in treatment
08:08responses in these patients.
08:10What we have so far in the clinical data,
08:13I will show you there very briefly.
08:16These are all sorts of lifestyle and clinical
08:19data that seem to segregate patients.
08:21This is a principle components analysis
08:23of our entire phenotype matrix,
08:25and you can see that there are.
08:30Phenotypes seem to correlate with age.
08:32In the top left you can see that the
08:36dominant trend in our patients is
08:38actually age and that kind of makes sense.
08:42It's a progressive disease.
08:43It's a lifelong disease.
08:45Older individuals tend to have more symptoms,
08:48and you can definitely see things like that,
08:51but that's an important confounder as well.
08:54Age is an important aspect of disease
08:57that we often don't talk about.
09:00We see more interesting things
09:02if you look at that second panel
09:05from from the left on the top.
09:08There's a correlation with
09:10natural UV exposure,
09:11'cause it turns out the vitamin D is actually
09:13an important component of Ms Pathology.
09:16It is a risk factor.
09:17It appears to be causal in ways
09:19that we don't really understand.
09:22But there are lifestyle exposures
09:23like that as well and they are
09:26definitely coming out of the the data.
09:28We also see smoking behaviors.
09:30Gender is an important component and so on.
09:32So as we start pulling all of
09:34these clinical data together,
09:36we were getting patterns even
09:37from very simple.
09:40Views of data. This is like a very
09:44naive exploratory way to look at data,
09:46but we're seeing patterns even in that way,
09:49just for the remainder of the time.
09:52I'd like to switch for a second and tell
09:55you about a different project that is
09:58still quite similar in flavor I think,
10:01which is about epilepsy.
10:03Identifying predictors of
10:04psychiatric disease, and epilepsy.
10:05This is funded by the NINDS and it
10:08is a collaboration between Yale,
10:10Arhus University,
10:11Helsinki University and the.
10:13Rodents chew epilepsy is a.
10:18Basically it disease where
10:20of seizures in the brain.
10:22It is abnormal electrical activity.
10:26That is often repeated.
10:27You see 2 broad types of seizures on EG.
10:31You see either a generalized seizure
10:34pattern that takes up a large portion
10:37of a hemisphere or the entire brain,
10:40or you see very focal abnormal electrical
10:44activity in one area of the brain.
10:48Again,
10:49it is a common neurological disease
10:52about one and 26 people in the
10:55US have a diagnosis of epilepsy.
10:58It is a complex disease.
11:00There exist certain single gene forms
11:03of it that explain about 14% of cases,
11:07but the other 8586% is this more common
11:11complex form again polygenic many genes.
11:14Heritable,
11:14but not simply heritable.
11:19And what we've been doing has
11:21been working with some colleagues
11:23at Arhus University in Denmark.
11:26Like many of the Nordic countries,
11:28Denmark has an integrated.
11:31Our health care system for which records
11:35are completely available for research,
11:37so the population of Denmark
11:40is about 5,000,000 people.
11:42There are records roughly for
11:44about 2,000,000 people who have
11:46interactions with the hospital system.
11:48We we tend to limit this to people
11:52who've had interactions recently.
11:55By which I mean after 1981.
11:59Becauses people born after 1981
12:01also have blood spots stored
12:04in the Staten Serum Institute
12:06from which we can extract DNA.
12:09So you can do population level
12:11genetics based on the hospital
12:13registers across the entire population.
12:16And so we limited this to this
12:19and one of the things that we
12:23observed about four years ago now.
12:26Is that if you look at individuals
12:30with a diagnosis of epilepsy,
12:34you find a strong overrepresentation
12:38of mental illness diagnosis.
12:41In that population.
12:42So if you look right at the top,
12:45there's about 1.3 million people who
12:47do not have a diagnosis of epilepsy,
12:50and there are reference.
12:51And there's about 10 and a half
12:54thousand people who do have a
12:56diagnosis of epilepsy and they have
12:59somewhere between 1.4 and 1.6 fold.
13:01Higher rates of psychiatric
13:03illness diagnosis.
13:03These are all diagnosis
13:05from hospital registers.
13:06They're not necessarily strong,
13:08strongly followed by individual physician,
13:10so this is not a cohort, these are.
13:13Medical records and that's worth.
13:16Highlighting, however.
13:19Psychiatric illness is itself genetic.
13:21Again, it is complex.
13:23There have been many.
13:25Genetic studies of that and so is epilepsy.
13:28And So what we are trying to
13:30figure out is if we can ask.
13:33Does the epilepsy cause psychiatric illness,
13:36or are these both either independent
13:39effects or both effects of a
13:43shared underlying pathology?
13:45That's an interesting question,
13:47because we think we can then develop.
13:49Predictors for given.
13:54That you have a diagnosis of epilepsy.
13:56What is the probability that
13:59you actually develop?
14:00Psychiatric illness post epilepsy.
14:02It's not even we can see in the data
14:07that not everyone is at equal risk,
14:10but we do not yet understand
14:12who is a higher risk,
14:14and so we're taking everything
14:16from school records which exist
14:18in separate registers which can be
14:20cross referenced to the hospital
14:23registers to genetic profiles.
14:25Two prescription reimbursements to see
14:28whether people have refractory disease,
14:31or whether they've cycled through
14:34many antiepileptic medications.
14:37And we're trying to build these predictors,
14:39becauses it seems rather important
14:41to know who is at substantial
14:43additional risk given a diagnosis
14:46of epilepsy relative to the others.
14:48I will just finish with a small
14:51little vignettes of almost accidental
14:53findings that we've seen again looking
14:55in these registers because we have
14:58an entire population to look at.
15:02We observe that. People with epilepsy
15:07are more likely to have a mother with
15:11epilepsy than a father with epilepsy.
15:15For reasons that we do not understand,
15:18there exists this maternal effect.
15:19If you just see in red,
15:21there's about a 1 1/2 fold.
15:24Enrichment of.
15:26Maternal epilepsy than paternal epilepsy.
15:29We're not sure if this is genetic
15:31or if something else is going on.
15:33This is a very unexpected finding.
15:36It's been reported at least once before in.
15:40In much smaller cohorts,
15:41and there's been a long standing dispute
15:45in the field about whether this is true,
15:48and across about 1.75 million
15:50people in Denmark weekend.
15:52Unequivocally see there is a sort of
15:55fairly meaningful increase in this risk and
15:57quite well underlies this maternal effect.
15:59We don't know, but it ties into our
16:02interest in *** bias in disease,
16:04and we're looking forward
16:06to following this up.
16:07So I want to just leave that there.
16:10Ask and I'll hand back to Nicole,
16:12who I think will introduce the
16:14next speaker or handle questions.