Clara Moreau “Impact of genetic heterogeneity & pleiotropy in psychiatry on brain functional connectivity”

March 09, 2023

Information

ID: 9633
To Cite: DCA Citation Guide

Download Transcript

00:06So I will present our work on the
00:08impact of genetic heterogeneity
00:10and pleiotropy in psychiatry and
00:13brain functional connectivity.
00:15So a bit of genetics.
00:18And so psychiatric condition
00:19as you might know,
00:20I've been mainly studied through
00:22a top down approach starting
00:24from the behavioral diagnosis,
00:26moving down to brain on the phenotype,
00:28cellular component, trying to
00:30identify genetic risk factors etcetera.
00:33Such approaches are revealed a lot
00:35of heterogeneity at any any scales
00:37from the phenome to the genomes
00:39with a lot of comorbidities.
00:42At the clinical level,
00:43it's almost impossible to find a patient
00:45with autism without an additional.
00:48Diagnosis.
00:48And at the imaging level we were
00:51not really able to find clear
00:54reproducible biomarker and when
00:56we found something it was not
00:58really specific to that condition.
01:00So it could not,
01:01it was not really a biomarker.
01:03This result at the imaging level
01:05where mirror at the genetic level
01:08with these two phenomena called
01:10polygenic city and player trophy.
01:12So I'm going to dive into this.
01:15Things during my talk.
01:17So when I say polygenic city,
01:19I meant thousands of genes
01:22involved all over the genome.
01:24So for example in this recent
01:27review from labrum and colleague,
01:30they reported more than 1500 high
01:33confidence genes associated with
01:35neurodevelopmental disorders.
01:38And on top of that they identify six
01:41stars more than 6000 candidate ND genes.
01:44These genes as you can see, we're located.
01:46All over the genome.
01:47So it's like in the brain,
01:49you don't have the region of autism
01:51and you don't have the chromosome
01:53or the low thigh of autism.
01:55And to add complexity to that we have
01:58this second mechanism called pleiotropy.
02:00Pleiotropy is when the same genetic
02:02mutation confer risk for several
02:04condition at the same times.
02:06So if we take the example
02:09of autism and schizophrenia,
02:10so you can see that in autism you
02:13have rare and common variant involved,
02:15so CVS.
02:16And single nucleotide polymorphism
02:18it's also the case for schizophrenia.
02:21But as you can see in this Venn diagram,
02:23you have an overlapping part where
02:26genetic variant covers for both
02:28condition at the same time and
02:30some of them also confer risk
02:32for ADHD and other condition.
02:34OK,
02:34so these two concepts can be
02:36summarized by this sentence,
02:37one for all,
02:38all for one,
02:39where one for all stand for pleiotropy
02:42when one genetic risk factor confer risk
02:45for several condition and all for ones,
02:48then for polygenic city when
02:50several genetic risk factor confer
02:52risk for the same condition.
02:55So OK,
02:56so I'm going to start with player trophy.
02:59Play Toby has been mainly studied at the
03:02genetic level using genetic correlation,
03:04snip base so common variant based.
03:07So for example if you want to
03:09study the genetic link between
03:11schizophrenia and autism so
03:13you can use GWAS summaries,
03:15you can use that GWAS summaries
03:18from schizophrenia and statistics
03:21summaries for schizophrenia and
03:24for ASD you just correlate this.
03:27These tables and you obtained
03:29a degree of correlation between
03:31these two condition which is reported here.
03:34So for example between schizophrenia and ASD
03:37you have a correlation around these 0.330%.
03:40So the Psychiatric Genomics Consortium
03:43did that for this eight condition
03:46using the eight largest GWAS available
03:49into 2019 and they reported 8 three
03:52group of anti related disorder.
03:55So Alien said. Developmental disorder,
03:59mood and psychotic disorder and
04:02disorder with compulsive paviors.
04:04Similar results have been obtained
04:07at the transcriptomic level,
04:09so by this study published
04:11by Gundalian colleague.
04:13So in the X axis you have the
04:15previous result published at the
04:16genetic level and in the Y axis.
04:19That's a new result from this
04:21study published at the obtained
04:23at the gene expression level.
04:25And you can see that you have a good
04:28concordance between observation at
04:30the genetic and transcriptomic level.
04:33And here we wanted to know if such a
04:36pleiotropy mechanism could also be
04:39observed at the connectomics level.
04:42So for this purpose do we collect it?
04:44I mean we didn't collect,
04:45we used publicly available datasets.
04:49For more than 1000 subjects with
04:51psychiatric condition as schizophrenia,
04:53autism, bipolar disorder, ADHD,
04:55we also used local Montreal data data set
04:59and we used UK Biobank as general population.
05:02We preprocess the data using yak
05:04software developed by Pierre Bellec.
05:06We persuaded the brain into 12
05:09and 12 network and 1664 region.
05:12We perform sequence analysis and a Pearson
05:16correlation as well as permutation test.
05:19To it as the significance of the
05:21correlation and we obtained this list
05:23of value represented in this diagram.
05:26So for example,
05:28we have schizophrenia versus bipolar disorder
05:32that are correlated around like 450%.
05:3750% ASD versus schizophrenia,
05:39so we have a correlation around
05:4230% and neuroticism,
05:44neuroticism versus freedom diligence
05:46where anticorrelated at like minus 30.
05:49So we compare this value with this
05:52bunch of nature genetics publication
05:54that have been that have studied,
05:57we have studied genetic correlation.
06:00So, uh, here in the exact X axis,
06:03you have our result at the connectivity
06:05level and in the Y axis you have
06:07the result at the genetic level.
06:09So as you can see,
06:10we have a good concordance.
06:12I was quite surprised about that.
06:15We don't have any correlation with
06:17our control condition, which was IBD,
06:19inflammatory bowel disease.
06:21So if we go back to our example
06:23about ASD and schizophrenia,
06:25we have a correlation at the brain level
06:28around 0.3 and it was quite similar.
06:30To what was obtained at the genetic level.
06:33So yeah,
06:34it was quite impressive and then we
06:37compared also with what has been
06:39published at the transcriptomic level,
06:42but in that case we only have we
06:44only had six pairs of condition to
06:48compare but still it was quite good.
06:51So I guess you versus schizophrenia
06:53you can see like we are still at
06:5630 at the brain level and they
06:58were around like 40-5 at the.
07:01Transcriptomic gene expression level.
07:03So what we have seen in this first
07:06part is the stability of the degree of
07:08overlap of player trophy across the scales
07:11of observation which in a way validate
07:15our biological overlap across these
07:17clinical diagnosis and also highlight
07:19the need to move from our case control
07:22diagnostic first approach to more like
07:25dimensional approach such as our dog as
07:28third explain yesterday introduced more
07:31than me yesterday it also looks like.
07:33Functional connectivity seems to be a
07:35good on the phenotype to study psychiatry.
07:38And before moving to Polygen City,
07:42we also tested some variable at the
07:45anatomical level and we focused
07:48on early onset anorexia,
07:50which is a type of anorexia with
07:52an onset before puberty, puberty.
07:54So we characterize the impact of early,
07:58early onset anorexia on cortical sickness
08:01and subcortical volume and as you can see.
08:04So we use local data from Parisian hospital,
08:07so 100 subject,
08:08around 100 subject with early
08:11onset anorexia and 100 control.
08:131/4. And we reported a reduced.
08:20Thickness reduction in parietal area and
08:23occipital area as well as post central
08:26gyrus and the reduced volume of the dynamics.
08:29But so we compare this.
08:33Since brain map of alteration with
08:35with what has been reported by the
08:37Enigma Consortium for OCD, ASD, ADHD,
08:40OC, T, so we obtain this correlation,
08:45so 49% with OCD,
08:47minus 20 with ASD,
08:49minus 39 with ADHD and minus 25 with obesity.
08:54We then as you might expect compare that with
08:58genetic correlation previously published.
09:01So for example, OK.
09:03You're gonna try to move my face?
09:05Yeah,
09:06for example.
09:08You see,
09:08you can see in there babe in like
09:10the previous paper I presented.
09:14And struggling like a medium
09:17link between anorexia and OCD.
09:20So it was a correlation around 50 and we
09:22had like similar degree of correlation.
09:25They didn't have any correlation with ASD.
09:27We have a negative correlation.
09:29We had also negative correlation
09:31with ADHD as they did report and a
09:34negative correlation with obesity.
09:36So it really seems that there is some
09:40similarity between genetic results and
09:42anatomical and functional connectivity.
09:44Result. As.
09:50As a parent as a note,
09:52I was thinking that I was wondering
09:56why we found this parietal.
09:59Region with sickness reduction.
10:02So I was comparing,
10:05I was thinking about developmental process.
10:07So I was comparing our map of sickness
10:10alteration with what has been published
10:12by Richard Bethlehem and colleague
10:14in this nature paper last year.
10:16And I was really surprised to find
10:2073% of similarity between our result
10:23in anorexia and their that's not
10:26like a case controlled map that
10:28just the regional maturation.
10:30Timeline here but, and in our case,
10:34our patient are all the same age,
10:36like they're all not the same
10:38age but between 7 and 13,
10:39so it was quite surprising.
10:42So we're wondering if this person had
10:44genus and superior parietal area might
10:47be metabolically more costly region.
10:52Yeah. It's just like an open question.
10:55Now. I'm going to move to the
10:57second part about polygenic city.
11:00So as I told you, we have like one
11:02more than 1000 genetic risk factor
11:04conferring risk prophetic condition.
11:07So we have two possible scenario,
11:09I mean simplistic scenario a scenario.
11:13So if you have 101,080 crisp factor you
11:17could end up with 1000 form of autism.
11:21The alternative scenario is you might
11:23have some convergence from the genome
11:26to the phenome to at the end obtain a
11:30limited number of ASD subtype let's say.
11:35So we wanted to test that.
11:38So first.
11:41Just an explanation about how to classify
11:46really easily or genetic variant.
11:49You can simply classify them
11:51based on their frequency.
11:52So for example a single nucleotide
11:55polymorphism are a common variant
11:58and they are detected by GWAS.
12:00We use them to compute polygenic risk score,
12:04polygenic score, they have a low,
12:06they have a high frequency
12:08and a low effect size, white,
12:10red variant are detected.
12:11Through all the exam sequencing or
12:14or CGH area and they have a higher
12:17effect size on the lower frequency.
12:20So here we in particular focused
12:21on CMV which is a type of rare
12:24variants that are present in 10
12:26to 15% of children seen in the in
12:29the neurodevelopmental clinics.
12:30So because the frequency before
12:33were for general population.
12:35CNN's are deletion or duplication
12:37of the DNA segment that typically
12:40encompassed several genes.
12:41So for example,
12:42if we take the 16 P 1.2 CNV,
12:45which is one of the highest
12:47risk factor for autism,
12:48you can see that you you have
12:51the deletion of 2019 on the
12:53short arm of the chromosome 16.
12:56And so as I said,
12:57the deletion is associated with
13:00autism but also with obesity
13:02macrocephaly over connectivity.
13:05While the duplication at the same
13:07locus is associated with schizophrenia,
13:10anorexia.
13:12Micro microcephaly and underconnectivity.
13:16So it's quite unflagging.
13:18We have several examples like
13:21that in the across the genomes.
13:23So we collected the envies data
13:26from a different hospital as much as
13:30possible and so from Montreal hospital,
13:34Lozanne Hospital Safari Consortium,
13:37UCLA, Cardiff University and UK Biobank.
13:41We also study common variants using
13:44polygenic score which is simply the sum.
13:47Of an independent effect of each snips.
13:51So we computed PRS polygenic
13:53score for autism, schizophrenia,
13:54bipolar cross disorder,
13:56IQ and control players.
13:59So we first characterize their
14:02impact on connectivity and so here
14:05we simply so they are represented in
14:07this brain maps and here we simply
14:10represented the their effect and ranked
14:12their effect size and connectivity.
14:14And 1st what you can see is that polygenic
14:17score has a really tiny effect on
14:21connectivity which makes sense if you
14:24think that's based on common variant
14:26and that's a sum of of independent.
14:29Effect of each snip that might
14:31have different effects,
14:32so you can just cancel your signal.
14:34So Pierce and Pierce seems not
14:37really ready to be used for now.
14:41That's our observation.
14:43So we were looking for converging pattern.
14:46Of connectivity across our genetic risk.
14:51And as you can see here,
14:52I represented the 12 functional
14:54networks that we studied.
14:56And as you can see first it looks
14:59like it's we don't have the
15:01perfect convergence pattern at all.
15:03All our networks are altered
15:06by some genetic risk factors.
15:08So there is not no clear converging pattern,
15:12but we don't have a lot of genetic
15:14risk factor as well.
15:16But still it looks like we have a
15:18recurrent over connectivity pattern
15:19of basal ganglia. That I miss.
15:23Network and the Motor Network as well
15:25as the recurrent underconnectivity
15:27pattern of the lambic and auditory
15:30posterior insular network networks.
15:32So we correlated our genetic risk factor.
15:37We'll connect wall brain connectivity
15:39pattern and perform a PCA to try to
15:43understand which connection where
15:45responsible of our correlation
15:47across our genetic risk.
15:49And we found that the first PC was
15:51explaining 25% of the variance.
15:54And was mainly driven by connection
15:56between between the thalamus and basal
15:59ganglia here and the somatomotor networks.
16:03And if you look at the tamic.
16:07Connectivity profile across our
16:08genetic risk factors and traits.
16:11You can see recurrent over connectivity
16:14pattern with like Motorola auditory
16:17and visual area.
16:18We also found the even if the effect
16:22size for peers is really small,
16:25we didn't find it for G factor which
16:28was like just a control condition.
16:31So to to explain,
16:35to introduce our theoretical
16:37framework to summarize this result.
16:40So if we come back to our thousands
16:42of genetic risk factors.
16:43So on the bottom part of my slide,
16:49you can try to.
16:53Group them based on their role
16:55at the cellular level.
16:57So for example,
16:58we already know that a bunch of
17:00genes involved in autism such as
17:03Cheng free or FMRP highly involved
17:06at the synaptic level.
17:07So you can for example just like
17:10rank your genes based on what they
17:12do on the at the synapse or if
17:15they are involved in like neuronal
17:18migration or like pruning or etcetera.
17:21So that is.
17:22The hypothesis is that this limited
17:25number of cellular mechanism could
17:28altered a limited number of brain profile.
17:32That could be connectivity,
17:33but that might be anatomy,
17:34that might be eggs, that might be I don't.
17:38I don't know what.
17:39And so in our case,
17:41we found this converging pattern
17:43of the talamo motor connectivity
17:46profile. And so the hypothesis
17:48is that this pattern could alter
17:52some dimension that are impaired
17:55across different diagnosis.
17:56So that's not specific to autism,
17:59it's not specific to schizophrenias,
18:00not specific to CD's that could
18:03just like altered sensory motor
18:05function that are known to be
18:08altered across different diagnosis.
18:10So that's really going into this
18:13demand dimensional approach
18:14such as the one proposed by.
18:16Our dog.
18:19That's definitely really reductionist,
18:21but that's just a way to try to simplify
18:25what we are trying to do. Um, so.
18:32Just to finish we so we are a bit stuck
18:36here because we are going to struggle
18:40continuing this approach because we
18:44cannot connect this genetic variant are
18:47really rare if we use common variant,
18:50they are quite atherogenesis for now so.
18:55It will be quite complicated.
18:57So geneticists for now try to
19:00delineate general mechanism of
19:02genetic risk factor of on cognition.
19:06So they are testing several genetic
19:10score to try to predict the effect
19:13of any deletion on cognition
19:16and on IQ and autism risk.
19:19And they reported that the probability of
19:21being loss of function and relevant was
19:23the best predictor of the effect of our.
19:26Genetic risk on IQ and autism risk so
19:30they developed this this model and.
19:34Ohh, we have now clinical application.
19:36So if you go on this website,
19:39so CNV prediction,
19:41if you enter your genetic coordinate
19:44and you for example you have the
19:48deletion of the 16 P 11.2 so you would
19:51like 16 the coordinate and deletion
19:54and you get the estimated score of
19:57the impact of your mutation on IQ.
19:59So we wanted to to apply that
20:03on connectivity.
20:04So we and it was great because it
20:06allows us to go beyond this case control
20:09like one CNV other time analysis.
20:12So like that we can include any
20:16CNV carriers in our data set not
20:19only the recurrent one.
20:22So we simply annotated genes encompassed
20:25into each CNV using this genetic
20:27score and we could just like we just
20:31looked at the linear effect of this.
20:34Critics score on connectivity.
20:37By doing that, we've we found that.
20:41This genetic score was.
20:45Was impacting mostly the thalamus.
20:48As you can see,
20:50you have 30% of the connectivity
20:52profile of the of the connection of the
20:55dynamics that were impacted by this.
20:57Genetic score,
20:58constraint score and if you look at
21:00the dynamic again connectivity profile,
21:02it was again over connected with
21:05motor and auditory area etcetera.
21:07It was not only the thalamus as you can
21:10see it's like a complicated profile,
21:12but it was,
21:13it's just like a way to summarize
21:17the information.
21:18So to conclude,
21:19we have seen that pleiotropy and
21:22polygenic city do impact our ability to
21:25detect brain connectivity alteration.
21:27In their names like treat manifestation.
21:29So it might be complicated to
21:32identify clear biomarker of our
21:34current clinical diagnosis.
21:36We still found a clinical contribution
21:39of the thalamus and the somatomotor.
21:42But OK,
21:43that could be associated with
21:46with sensory impairment observed
21:49across different diagnosis?
21:52So this mutation also do impact
21:54the rest of our body,
21:55it's not only the brain.
21:57So we might need to include data
21:59from other organs in future study,
22:02but we need to pursue this study to
22:05have access to more like imaging genetic
22:08database such as exams like the represented.
22:11I think it's the one of the only
22:14database with imaging genetic
22:16data in psychiatric conditions.
22:18So we have UK biobank etcetera,
22:20but HCP.
22:21And it's general population,
22:23so we really need like a.
22:26A data set enriched in this
22:31rare genetic variance.
22:33And we also need to use like
22:37cognitive scores that are
22:39not specific to a diagnosis.
22:41So for the only score that
22:43we have in common across our
22:45different diagnosis is IQ so far.
22:47So it's quite complicated to to.
22:50And IQ depend off on age.
22:53So it's quite complicated to compare
22:55our diagnosis based on our our
22:57continuous dimension because we don't
22:59have that much to compare for example,
23:02you have specific score.
23:04For each diagnosis as you might know,
23:06so to characterize social impairment,
23:09language impairment like attention
23:11process etcetera, but it's not,
23:14it's specific to the to ADHD
23:16or to ASD or to schizophrenia.
23:18So it's complicated to compare this.
23:24Diagnose you?
23:25So I want to thank my previous.
23:30Lab so Sebastian Jackman,
23:32diabetic Toma Burgeon at Paris,
23:35and Richard Alarm and my new
23:37lab at USC with Paul Thompson.
23:46Meet up sometime. Very nice.
23:52Theoretical model.
23:58I wonder. So often this is the model
24:00of unfortunately right and often when
24:03we're thinking about opportunity,
24:05you want to like the patients into stuff,
24:08right, so that we can.
24:11The way I'm interpreting this and
24:13that you would essentially get if
24:15you did that in a in a robust way,
24:18you would essentially get a
24:20different answer at each other.
24:21And so that makes it complicated because
24:25what level kind of maps on to treatment?
24:27You know, how do we validate?
24:29If we often we we post from one level
24:31and then we use another level to kind
24:34of try to validate differences and
24:36this seems to me more likely to be.
24:41Maybe, but I wonder what it means
24:43for the search for something.
24:48Yeah, definitely I I was also thinking
24:52instead of going like bottom up,
24:55going top down like people
24:57are currently going top down,
24:59I think we should do like complementary,
25:01we should do it like we should do it both
25:04way because both way will be informative.
25:08And we could like do sub types and try to see
25:11if they are enriched in genetic mutation.
25:14But for that we need this imaging data,
25:16this data set with imaging
25:19genetics in psychiatry.
25:20But yeah, I totally agree with that.
25:23Which is great reductionist,
25:26but it's a way of summarizing. Sure.
25:31One question regarding.
25:34All of these analysis obviously
25:36based on people being patients,
25:39so being actually diagnosed seeking
25:41help and also that's one to put in.
25:47More related to resilience.
25:50In particular, whether you actually
25:53become over the negotiation.
25:55Plus having for pology,
25:58but not by the well enough bottled up so
26:01that you don't end up being a patient.
26:04Um, so. That would be what
26:08you would see the resilient.
26:12He makes you impatient.
26:14Somebody who just says it. A yeah.
26:26OK.
26:30So I'm trying to think.
26:36Yeah, it could be the case, but it will be a.
26:43I mean resilience is really,
26:45definitely really interesting
26:46to study right now. For example,
26:48we are trying to study in like general
26:50population like in UK Biobank.
26:52We have a study right now trying to
26:55investigate carriers of our genetic risk
26:58factor who did not get any diagnosis.
27:01So why is it the case?
27:02Is it's like the rest of the genome,
27:03is it the environment like?
27:07And I think it's like really
27:09a hot topic right now,
27:11but I don't really know if it's the
27:16player Tropic then I it really looks
27:19like the majority of the games are
27:21played topic only some of them have
27:25high penetrance and the specificity.
27:27Most of them have like are
27:29expressed at different places and
27:31conference for several diagnosis.
27:33So quite so it's this we will
27:36definitely not be able to do any.
27:39Gene editing or anything in psychiatry,
27:41because it's not specific.
27:46Thank you very much.