Patrick Sullivan: Why care about psychiatric genetics?

0:17 paul thank you very much for the very kind introduction um in the united states if you go to a comedy club or stand-up comedy you never sit in the front row it's a thing people will call on you and ask you to say things i won't this time okay so why care about psychiatric genetics and the stuff that paul mentioned is exactly the case i think the one thing that'll be a theme in this talk is just how much cooperation internationally has actually made a difference and and certainly that sort of ethos is everywhere now so a couple years ago we needed some replication data i i sent five emails two days later i had replication data from many thousands of subjects that's how easy it is right now there's a culture of cooperation people work together you know nearly universally and that's the thing that's made progress really really start to take off

1:04 so what i hope to tell you about today is is really why you should care about psychiatric genomics it's been a bit of a sinking ship as paul had on a slide there's been some real problems there's been a lot of high profile difficulties but the too long didn't read aspect of this is that first there are several things that are ready for the clinical utility right now first and second i think a bunch of other things are coming fairly soon now in saying this i'm effectively predicting the future or trying to and if there's anything that's very difficult to do if you're a human being it the temptation of course is uh is unavoidable we all want to predict the future in one way shape or form and of course when it comes to technology people have done this a lot and you know clearly uh this is from 1951 i've yet to see my personal helicopter

1:51 or my personal jet pack and i don't know if you've been to los angeles lately but that definitely is not what it looks like so some of these things actually have come through this was somebody back in the 40s talking about a tv set you could hang on your wall okay that was that happened star trek communicator um you know most of us have cell phones you know not the same thing but it's pretty close um still waiting for the jet pack and of course you know we're a little worried about what ultimate intelligences are going to do to us at some point so the the us baseball scribe and zen master yogi berra has this famous quote it's tough to make predictions especially about the future and if you think about that really hard your brain might break but it's probably true

2:38 and medicine of course is is uh full of similar things where here we have uh i think this person was the personal position to queen victoria the abdomen the chest and the brain will be forever shut from the intrusion of the wise and humane surgeon not so much this code i really like by the president of ibm i think there's a world market for maybe five computers i'm going to bet there are people in this room who have five computers on their person right now and finally of course there's no reason why anyone would want a computer in their home and this was the a guy who was ahead of a computer company which of course doesn't exist anymore because i think they kind of missed this particular boat but i think computers are actually a sensible um uh sort of analogy for what we're talking about here because

3:24 if we think about how computers have changed in the course of the lives of many of us from 1980 through 2015 cindy and i got one of these in 1983 in fact so this was one of the first macs that came out 125k because who needs more than 105 25k of ram um it's a reasonable analogy in terms of how things progressed some things that people have talked about with computers have actually gone completely wrong some things in fact have have become completely right and a lot of things that were not anticipated at any point are in fact part of people's dailies lives right now and so this comes down to technology development and knowledge and experience with the tech and let's not forget amara amara's got something called the first law of of technology which is you come with a you have a level excitement and time

4:10 you have an idea enthusiasm it becomes over hyped it crashes to this disillusionment pit and eventually we start to move toward true progress i would argue we are there that we're that that whatever you know disillusionment and hype we've had in the past i think people are really starting to to really think hard about what this might do and how it might be important clinically so let's go over the technology relatively briefly um the first thing up here is the sort of the latest greatest from aluminum one of the big sequencing companies is called the hiseq x10 um to my knowledge there are four there are 10 machines and this is the thing you buy if you want to sequence the whole genome sequence populations it can do something like 18 000 genomes per year sci life lab split between here and

4:56 uppsala has one of these i think it's one of four in europe that does i don't know how many are in china but it's way more than that um this is another technology that's kind of popular it's sort of a bench top sequencer from iron torrent and the one of course which is just spectacular it doesn't work yet but it's kind of spectacular to think about is this usb one uh from um from oxford anapor and so eventually what people are and people actually trying to develop handheld sequencers where in theory you could actually take a blood sample from somebody in an icu with sepsis just put a drop in and figure out you know it's more or less real time what the what defending organism might be so again this is just like computers isn't it where we heard such great things are just around the corner and sometimes some version of that comes but often it doesn't

5:42 this technology is established ready to go this is um actually in labs right now this is still on the horizon now if we look at the the i think this graph is really kind of cool because the y-axis is a log scale and so this is the the cost to sequence one genome and when the human genome project was done around completed around the year 2000 um the cost to do one genome took years and it was like 100 million dollars um and as you can see there was a slow development with the same technology a sharp drop and again this is on a log scale so this is a huge decrease new technology um bit plateaued of it but there's been a recent drop as well so if you wanted to seek one's one genome right now the cost is a bit over one thousand dollars u.s

6:27 takes a couple of weeks it it's feasible to do a very large scale um and so this is with sequencing which is rereading the genome and then of course the other technology we have is genotyping which typically looks at a fixed set of genetic markers sequencing means you actually go back and for a person you redetermine every single base or as many of them as you can in that person from scratch this is one where you have basically a fixed set of markers and you genotype them like mcdonald's you know if you get the same thing over and over again you can do it cheaply and quickly if you actually do it at scale when you're trying to recreate everything it's more expensive and so the cost tiers have fallen as well so in about 15 years ago to do 500 odd markers would be about two dollars each it would take

7:13 months literally um the price when the first big chip came out was about this and then currently they're very inexpensive and the cost per genotype is very very small in fact we we do these for quality control and cell line experiments that's how relatively straightforward it is um to do uh to do uh you know 5000 people many labs can do 5000 of these per week and large labs can large labs can do even more than that now this was not supposed to pop up quite yet oh well so much for that if um if the genomic results that we that we use in schizophrenia were scaled to be one pixel um the actual the the raw data would be on the order of 1200 kilometers or approximately stockholm to vienna

7:58 so just to give you an idea there's this enormous big data funnel right where you have to handle every single step with impeccable care in order to have anything invalid at the end um but the scale is actually really quite huge so that's my my so the whole thing about technology could in fact be an entire one-hour thing in terms of what's happened what's developing let's just leave it like there and come back and ask and answer questions on this if people like uh in the q a now genetic studies for humans are typically not always but typically are variants on case control studies or are directly case control studies and a lot of it is very straightforward so um the typical thing that we do revolves around logistic regression which of course is the workhorse of epidemiology and clinical research you're trying to predict case versus control status

8:45 so the dependent variable is you know schizophrenia yes or no autism yes or no cardiovascular disease yes or no you have many issues as always in terms of how you define a case whether your controls are appropriate whether there's some bias in your ascertainment or selection um but and those are always as important for a genetic study as they are for anything else the key predictor the independent variable is one genetic marker so you've measured you've measured something and you're trying to see whether the frequency of that marker is similar or different in cases and controls um and the cool thing about genetic markers is that almost by definition exposure precedes the disease the development of the disease we're exposed to our genomes from the moment of conception onward i mean yes there are some models in which that assumption doesn't hold

9:31 but generally what i just said is true exposure precedes disease onset and so we're not trying to we don't play guesses about reverse causation next we typically include covariates particularly ancestry to avoid a common bias we have to take into account multiple comparisons you know a typical study might cast one million different tests and and you can't use 0.05 as a level of significance you can't use you know 10 to the minus 4 is your level of significance you have to adjust for the fact you've done a million comparisons um power of course is quite important as well especially since you have to have a very small type one air level we typically need very large sample sizes unless the effect that we eventually find is massive we've got a lot of experience in this right now there are a couple of instances where the effects are kind of

10:16 massive but generally the effects are really really subtle but this whole the whole idea of a genetic study is you basically take this model you run it once you run it a second time you run it a million times and again this is why we've got computers and since we're doing this over the whole genome this is a genomic study now going a level deeper on this this graph which i stole from a review from steve mccarroll and steve hyman shows the this is the current time going back in time 200 000 years and this shows you the population of of humans on the planet in billions and as you can see we were really really rare for a long time and then around 3000 years ago with agriculture the population started

11:01 to climb and in very very recent time there's been this huge increase in human population now so as it turns out if there's a genetic variant which is common it tends to be old so in other words if there's a particular genetic variant that that half the people in this room have by and large that variant will actually have its origin way way back in time which means it tends to be found pretty much everywhere maybe a little bit more maybe a little bit less sometimes not at all but usually if you see it in one human being and it's common you're going to find it many other human beings as well now on the other hand with this new expansion um if something if something happened very recently it's going to be pretty rare so typically rare mutations are new

11:46 and older mutations are common mutations are rare and the technology you use to go after these two different types of events depends so if you're looking for old variation genotyping with a fixed set of markers is the way to go that's a relatively effective thing to do it's cost efficient if you're looking for something which is is really quite rare maybe present only in one person or in that person and their immediate relatives they may be literally the only people in the in human history who've ever had that mutation and if you want to find that you simply can't use this method you have to use a sequencing method so again this is sort of the the thing that goes behind what we call common variant association studies looking for things that are old and but common and things that are rare variant associations where we actually

12:31 start to look at things that on the frequency spectrum become low frequency you know things you might find once in a large sample things that are private just to like a person and their immediate relatives and even things that are found um literally in one person ever a couple of examples of this this was a paper we did in schizophrenia a common variant study where we essentially looked at something like 8 million genetic markers in a large sample of 5000 cases from sweden and 6 243 controls and again when i mentioned coming up with all these replication samples that was the you know five emails in a couple of days in order to get that data so that's how that happened this is an example of looking for rare variation where we resequenced

13:16 redetermined the bases in the protein coding portions of the genome the exome in 2500 cases from sweden and 2500 controls and again this was uh this is an example of a common variant and a rare variant study okay so why should we care and my little computer here is to remind me to say that there actually are a couple of clinical uses right now in effect this is kind of like where computers were back in 1980 or something where it was very clear that were there were a bunch of uses which were immediately ready for prime time so for for instance you know you know the use of computers for managing payrolls for running the telephone for running um you know telecommunications systems for working for air traffic control that

14:02 was pretty straightforward you know pretty obvious running payrolls whatever the case might be so again there are some there are some uses which are actually extremely uh important and fairly straight up and a lot of this comes down to what is a clinically relevant or actionable sort of variation and typically the the ones that we're talking about are almost by definition really quite rare and typically have strong effects and this will be something that'll come up over and over again that in addition to rare things being new rare things tend to have a stronger effect common things tend to have a more subtle effect if you have something which has a very strong effect on a phenotype it has a strong effect on a human trait it really makes a big difference in terms of the development of the body

14:48 the development of the brain the liability of cancer typically these things don't last that long they're selected against and so things that are common tend not to have at best a subtle effect and so it's much more difficult for um for selection to operate on them conversely things that are quite rare can have extremely strong effects and the things that were that may actually work in clinical scenarios right now or a subset of things that typically give you relative that they're they're involved in conditions that have relatively early onset that are relatively severe and or they have um clear inheritance patterns and families some give three examples the first psychiatric disorder that would fit this bill is intellectual disability um there was a moderately sized studies that

15:34 looked at this recently um that we really need a much larger more definitive study that may well be ongoing but if you actually take uh 50 children with intellectual disability and and throw a lot of genomic technologies at them one study found that uh 39 of the kids um had a de novo um single base pair change de novo in the sense that the child had with intellectual disability had it but neither parent did so it that de novo the new variant happened in the mate in the making of that child and so in terms of a one base pair change typically this is a base pair that happened in a critical exon of a critical protein involved in synaptic function or the development of the brain and basically that buggered things up

16:19 and led to intellectual disability another sort of variant is a copy number variant and again denoble copy number variance the affected child has it but neither parent does and in that instance this is not just a single base pair change this is actually the loss or the duplication of a region of the genome and so occasionally these things these things happen not infrequently we all have many of them mostly they don't matter occasionally they happen and they're quite large they can be as much as you know half million bases a million bases or more and they can influence many many genes there's many examples of these that are out there that are important clinically um and it's if you add up these two portions of the pie you come up with something in the order of 60 percent being explained by one of the other a

17:06 very small piece of the pie are our inherited variants and in this case recessives where where the child by definition got it from um a very unlucky event where two people each of whom had one of the variants married and and the child had uh had both and had a disorder as um as as part of that and so um again this is one example where i think in many places in the world um this technology is often used one of these one of these technologies is often used to screen children with intellectual disability um autism again is the second answer a second example and again we're looking at an early onset condition but the numbers here a little bit differently they're really a little bit different so um the de novo piece is much smaller um seven percent this is after you

17:51 remove cases people with mentally kids with mental retardation by the way so this is just non um kids with autism that do not have mental retardation and again a small fraction of kids have a de novo event in other words neither parent had it but they do a few of them have three percent have an inherited recessive fragile x syndrome rhett syndrome there's a there's a handful of disorders that have a pretty strong chance of of being something like autism or autistic like um if you actually have a classic sort of protein changing variation but again the vast majority of the pie is unknown you can change these by um by selecting people in certain ways in other words if you look at children with autism that have that sort

18:37 of a syndromic picture that they've got sort of signs and symptoms that are reminiscent of a classic mendelian disorder that of course will increase your selection as well but again um and in a lot of places in the world um children with autism or developmental delay actually do get these technologies applied to see whether or not they they might fit into one of these categories and the final one we'll talk about is alzheimer's and with alzheimer's um there's three genes in particular um amyloid precursor protein pre-stilling one president two there are many examples of families in the planet where they've got a protein changing mutation in these in one of these genes and that leads to a typically a dominant early onset form of alzheimer's there's a there's a rare protective

19:23 nordic variant that decode discovered and of course there's the classic and relatively famous um the swedish mutation in app uh as well and again typically a sequence sequencing technology is what used to get this but here obviously this is not an early onset disorder but it is in a relative sense individuals with this tend to have early onset of alzheimer's disease early onset dementia and again it tends to run in a dominant way people have looked at this in a lot of dominantly acting families but typically you have to have a relatively large family even to know and have a really good family history before you can get to this point okay back to our little uh 1984 mac again what's the point of this i mean you can argue that for each of the three disorders i've mentioned we can't do anything right there's no

20:09 way right now that we can fix intellectual disability alzheimer's there are some treatments for um for alzheimer's sorry so intellectuals building autism there are some treatments for alzheimer's but again there's nothing that that works super well the first thing i think is sort of a medicine 101 thing from the very first day of medicine medical school they teach us that you need to call things by the right name so i think there's certainly something important about having a correct diagnosis in addition especially for the childhood onset conditions the answer kind of matters for a parent if they're thinking of having another child so in other words if if two parents have inherited a recessive and that's why their child has intellectual disability that gives you a chance to actually screen the next child to see whether or not that child has inherited but gotten

20:54 unlucky again and inherited both copies that would be information that a lot of parents would would want to have i should mention there's actually a company called council in the united states where for about 300 u.s they will screen um both people who are thinking of having a child for something like 80 different mendelian disorders and ones that have strong effects just to make sure that there isn't like you know two people with recessives coming together um apparently the yield is very low but apparently something like three percent of all couples actually in the in the san francisco bay area actually get the service done so they actually have a fantastic business model just to sort of cater to people who are actually relatively well educated and somewhat nervous about having kids

21:41 um the other thing reason why i think this is important is you know especially a lot of parents wonder is this my fault and information about what exactly is happening and why can be something that can sort of can help the person um sort of understand you know to help often information helps this point i'm trying to make but there's something actually more practical here which is increasingly um it's important that we know why people have certain things first because typically that can get them connected with support and services so for here for example there's a couple of relatively rare conditions 16p11 is a common copy number variant associated with autism and schizophrenia um 22q11 is another one associated with autism and schizophrenia and other things and fragile x of course is a relatively

22:27 rare um disorder and typically you know these sort of thing you know with within with the internet with social media with facebook et cetera et cetera there's a lot of places where people can get connected and a lot of parents have told me they find that profoundly helpful and actually sort of knowing what to do and occasionally they can actually find you know the right service the right person to go to the right doctor who actually who's expert in the condition or sort of knowledge about which medicines to try for certain things can actually be quite helpful but i think even more to the point though is that is there's the emerging therapeutics um so for instance if for this paper was in nature where they actually came up with a really interesting at least in mouse uh potential treatment for um angelman syndrome angelman syndrome

23:13 is really quite rare it has sort of mental retardation maybe even an autistic-like presentation associated with it if a child with id has angelman and you don't know and if a treatment becomes available you won't know what to target that person for it i've been talking with some people from roche one of their big models right now is trying to figure out specific treatments for these rare disorders again if if we know the correct diagnosis perhaps in five years from now there will be a therapy that might um become available um this is one of my favorite stories right now this is deborah levy from harvard um and there's a pair of relatives second-degree relatives one of whom has schizophrenia one of the other of whom has bipolar disorder they both have a

23:58 duplication of this gene glycine decarboxylase this is a gene that gets rid of the amino acid glycine under double-blind conditions um the administration of oral glycine actually makes their symptoms better so it so the other way to get this if we can actually come up with molecular diagnostic it might in fact be possible to get very targeted and directed therapeutics for individuals so that's why i think we should care now so the first part of my talk was talking about some of the clinical uses currently now the next part is more um the potential and so again you know we're getting off in yogi berra space here right where i'm talking about things that might happen and there's lots of different ways this might play out um and we don't know but i think it's important that we work on it and that's

24:44 going to be a big role of my of the group that i'm trying to establish here let's talk about schizophrenia so firstly schizophrenia is i think many people here know has massive morbidity mortality personal familial societal costs so just to sort of nail that you know the life expectancy of people um tournament in sweden i think it's like 10 or 12 years less than everyone else there was a very good paper on this recently um in the u.s the cost per lifetime is something in the order of 1.4 million dollars all told so again this is a hugely you know and that's the one part of this we can actually put a number on um the the cost to an individual to their family of course is incalculable um we've done a lot of different things over the last 100 years to try to understand schizophrenia and largely there's been a few insights

25:30 but largely we haven't really had things which have tended to replicate or to be clear and consistent over time and i think for a number of reasons we have to keep in mind that we're probably dealing with a subtle process so this is not something which typically hits a child at birth or early on typically people develop more or less normally for a couple of decades of life and then they develop a relapsing remitting condition even when people are really severely ill typically they still have preserved sensorium preserved consciousness a mostly preserved cognition and certainly preserved on basic neurological function and so i think by definition we're not talking about it like as in a typical mendelian disorder that affects the cns once you get onset typically you get a very quick relatively rapid relatively

26:16 direct progression to seizures comb and death so we have something which almost by definition has to be subtle now in terms of why genetics this has been a relatively strong ideological clue a lot of the basic work that is established is in fact has come from scandinavian a lot of it's come from sweden um and and again i think we're looking at the the data are consistent with um genetics of having a probabilistic rather than a deterministic role in other words it may increase the probability you you become ill rather than being a genetic variation which determines that you get ill i think that's a critical thing again we're looking at a subtle process the reason i say that is the risk of schizophrenia if if a person has uh if you have a first-degree relative schizophrenia we determine in sweden the risk to the risk to the people in

27:03 your immediate family is like fold increased but since schizophrenia has a prevalence around one percent a ten-fold increase means that only about ten percent of the relatives might have it again it's a significant difference it's an important difference but by no means is it deterministic and then the other one that the other fact i like too is the risk to co-twins so if one monozygotic twin has schizophrenia on average the risk to the co-twin is only 50 50. so again on average what's going on there of course is there's more there's more to this than genes are involved we've done a lot of looking over over the years um and i've yet to hear of a convincing um single gene uh report for schizophrenia and again all this is consistent with a subtle process

27:50 as paul indicated we did a lot of genetics but we were hampered there's sort of this catch-22 you you're in if you know what the true genetic architecture is you can design the correct study but since we don't know the the the architecture we can't design the study and so that's a very difficult situation to be in not knowing the genetic architecture not knowing the impact of diagnostic heterogeneity we don't have any good tests for schizophrenia there's no pathology which will tell us yes this person has schizophrenia or how much of it they have we deal with sort of clinical interview which is you know reliable certainly but the validity is not known and we can't unknowable in fact until we actually have a better idea as to what the uh as to what the ideological processes are

28:35 um in addition in genetics we had incomplete genotyping a lot of the samples that are out there were actually really really small a couple hundred people at most when in fact that's off by a couple of zeros and uh there certainly were indications that some of these studies had multiple different types of bias so as paul mentioned in 2007 we started something called the psychiatric genomics consortium you can go there to find out more about the pgc we're open source and so you can actually download all the results of all the studies we've done all the public ones and as soon as things get uh get published they're up there within a day or so one of the things we were thinking about when we started this was that we'd all read the history of this field and we had known that every single time a new technology came along we had gotten the sample size

29:22 requirements wrong and we assumed right from the start that the sample sizes we were going to need were going to be far larger than any one study could any one group could get hence cooperation was essential so we made something that was devoted to being democratic participatory open and rapid just to really try to maximize your chance of progress and again i'd point this out as sort of a an interesting sort of um thing like computer we actually tried to rethink how we did science effectively you know how can we do this in a way that's different from what we've done in the past you know competing labs you know people at war how can we actually come together and perhaps achieve more as a group than we than we could individually so cooperative collaborative comprehensive blah blah but i think that the cool thing about this is that a lot of the

30:08 people in this are clinicians all of us get it we know why this is so important and and a lot of us have sort of agreed explicitly to put the good of the many ahead of the good of an individual group um the the pgc structure is this um there's a coordinating committee which has about 15 people on it um there are nine different disease working groups um adhd anorexia et cetera et cetera and including schizophrenia we also have some cross-cutting groups that that uh that look at uh sort of connections across these disorders a copy number variant group a stat analysis group um and i'll talk about site chip in a second um in terms of things we've accomplished um we are as paul mentioned the largest consortium in the history of psychiatry

30:53 it's a complete pain in the butt to do the author lists for our papers by the way one of my colleagues spent i don't know 12 hours typing names into a website and then he found out he didn't have to afterwards it was horrible we laughed um it's i think we're clearly the so i think with the largest consortium history of psychiatry largest biological experiment um we are a mega analysis consortium we actually instead of sharing results we actually share the individual the raw data and that gives us a lot of flexibility and a lot of speed that other consortium don't have but it was a decision we made from the very start and we've been fortunate to make some good contributions currently have genomic data from 172 000 people actually in analysis right now so these have been uploaded run through the same quality control imputation

31:40 and analysis pipeline and are being analyzed we're currently adding another 100 000 subjects and the mad goal that we have is to see if we can get a hundred thousand cases for each of the nine disorders so a million people um this is not as bad as it sounds so major depression is actually one of these disorders and we're going to be there later this year mainly because studies like uk biobank we've convinced to ask sensible questions about depression which means we'll be able to use the half million people that they have and they'll probably have 60 000 cases along those lines um and to get more cases for other disorders is complicated um cindy bullock who's my partner in crime genetics and family life as well as intellectual stuff um is live tweeting this right now and i

32:26 hope she notes that has embarrassed her um she she leads the anorexia group for the pgc and she's been done a great job with using social media and large national register studies like here in sweden for instance um to actually come up with relatively large sample sizes so they're going to go from having something on the order of you know a couple of studies of a thousand people a thousand cases to one monster study which we hope will have on the order of 20 000 cases before too long and maybe even more if things work out well and again that's that's by sort of creatively rethinking the process and and trying to find ways to do it a lot of the cases she gets in the states by the way are from social media curiously the cases from social media often are more severe than the cases that we that she gets

33:11 from clinical sources so it's not like doing this actually kill you know hurts the study in fact it probably helps it taps into a pool of people that you might not get otherwise okay so let's talk about how genetic data can be used and again the little mac guy is up there just to sort of remind us that i guess i could have used the little stupid windows paper clip man but that would have been a worse one would it anyways the little mac is that remind us uh that this is really kind of off the page so this is not necessarily established but i think this is something which is a major idea or major intellectual thrust that people are working on so there are three ways you can do this the first let's talk about etiology and the logic here is that psychiatric drug discovery has almost always been

33:59 through serendipity somebody develops a pre-med for surgery and notice that it calms patients they try people with schizophrenia and lo and behold it's an antipsychotic um one of the first antidepressants was developed as a tuberculosis medicine and they noticed that people in sanitaria who were getting this they started becoming less depressed the first dentist and a depressant okay this is not a rational process that's driven by uh by biology that's driven rationally and so the whole idea first is that we want to see if we can use the genetic findings to illuminate biology or really do not understand much for sure if we get this if this gets us into the pathways into the readouts then there's a way to do high throughput drug discovery and high throughput evaluation of whether combinations of

34:45 medicines might actually be better than single drugs the one of the my pharmacology colleagues tell me that one of the main concepts they have right now is the notion of a magic shotgun rather than a magic bullet so in terms of where we are right now in terms of the genomic findings three slides real fast so in terms of the exome sequencing stuff there's been a couple of studies published three small too large the two large ones are here this one was from sweden with christina holtmann and others and basically together as well separately no single genes pop up they're certainly indications of pathways the pathways which appear to to occur more likely than chance have to do with calcium signaling um the post synaptic density activity-related cytoskeleton complex

35:30 nmda receptors and genes whose messenger rnas bind to the fragile x mental irritation protein to my knowledge we've yet to see mendelian disorder i've heard of a couple of things that might be in the review process but i haven't seen them directly and in analysis there's 10 000 cases and 1800 trios markedly more than here and i can tell you in confidence based upon sort of pretty good analyses the same general pattern reported in this paper holds next copy number variation again this is change in the gene dosage over sometimes a single gene sometimes 20 genes sometimes more there's 10 rare but potent examples of this these are things that are quite rare in the population you know three and three and a thousand

36:15 cases might have one of these but they increase the risk for schizophrenia markedly by fourfold or maybe even as much as twentyfold the problem with these guys is that they often point to multiple genes not just one and if they increase risk for schizophrenia they probably do something for autism epilepsy intellectual disability um and or developmental delay um the pgc has a they've done a very elegant analysis of i think it's actually yeah 30 000 cases and again the same pattern pretty much holds genome-wide association again is a way to look for common variation i won't say anything to these two people leaving because i said i wasn't going to abuse people in the front row so

37:01 so this was uh the paper that we published last summer and this was where my my friend typed in 12 hours worth of names um needlessly we pulled together data from about 37 000 cases well over 100 000 controls um there were like something like 52 studies that were 52 samples contributed to this to this thing so clearly it was a worldwide effort tons of people and the end we ended up with 128 independent snips in 108 genomic regions a few of the regions actually had two this is something called a manhattan plot because the manhattan plot is what you sort of imagine manhattan looking like from across the east river at night so you sort of see these lovely peaks corresponding to sort of imaginative skyscrapers this is a lovely plot we were so pleased when we saw this

37:46 for the first time because many other times we had these basically you know holland plots you know we sort of totally flat maybe a little bit of windmill and that was it but nothing nothing but notable um we went at this really hard the first thing we said when we saw this result was some version which i won't repeat of what did we do wrong and so we actually went back at this for about a year and a half large team of people attacked it from every angle we could think of and in the end the results held up and we published and some of the things that came out of this were really kind of cool so first on the list is the dopamine 2 receptor popped up as having a nearby genome-wide significant association dopamine this receptor of course is where most effective antipsychotics actually bind

38:33 and so it's been a major player in the field the nmda receptor story is is a known one there's a lot of biology in this i like this one in particular the there's a copy number variant on chromosome 16p and nicoquetsanas did a zebrafish cool zebrafish study where he determined that that was in fact the driver it was the dosage of that gene which was driving it and we have a genome-wide significant association like 2475 bases away so very very close hypothesis is is that if you change the dose of this by 50 or 150 percent it has a strong effect on phenotype if you change it a little bit with the regulatory variant it has a more mild effect the calcium channel story is still there as i mentioned um i'll go over this later if we have time

39:18 um but the thing that's that's real interesting is these processes among others actually come up in multiple different types of studies neuronal calcium signaling postsynaptic density of the art complex nmda receptors and genes who interact with fmrp and again this stuff pops up with an exome study from cmv studies and gelos the fact that different types of studies of different types of variation tend to implicate the same pathways i think is is a lovely bit of convergence that is is i think increases the chance that these are true findings now in terms of um uh what we learned about the architecture i think for interest i'm going to skip this slide the site chip is something we're working on right now and this is something that has been led by the pgc it covers common variation

40:05 as well as rare variation and and snips that we put on it um it's cheap it's like it's 100 bucks i think at most it might even be 80 if you get it right we're currently genotyping 100 000 samples and we're hoping that for schizophrenia the number of cases will get up to around 62 000 later this year so the published study has 36 000 we're hoping to get up to 62 000 cases this year now the next potential use of genomic data has to do with therapeutics in particular the prediction of adverse drug reactions or treatment response i think the prediction of treatment response is quite hard but i'd like to talk about adverse drug reaction i think this is actually kind of a key output of precision medicine there are some examples where you have genetic effects on common measurable things like warfarin dosing but i think it's real sweet spot might

40:53 in fact be for things that are rare and very serious so stephen johnson syndrome is effectively when you become allergic to your skin and it sloughs off that's obviously a bad thing to have drug-induced liver injury where some people who are genetically unlucky take a typical antibiotic and they kill their liver and have to get a liver transplant a lot of these things curiously actually point end up being part of the same gene hlab there's a story about the anticonvulsant carmazepine and stevens-johnson syndrome in asia where this particular variant of hlab has a huge odds ratio in fact they've actually started to screen for it in taiwan now and the incidence of stephens johnson has gone really really low um there's a lamotrigine also binds there but it's a different variant with a strong odds ratio

41:38 a back of your does as well and so basically what happens is you get off target binding of a drug to the pocket of hlab it gets presented as an antigen and it triggers an auto antigen response that's the general idea and we found for clozapine um the for people a case control study of granulocytosis people's white count really really dropped with exposure to clozapine versus controls that the same gene popped up and we actually found through molecular modeling that it probably also fit in the pocket in a similar way um unfortunately the the connection here between uh a grand wasn't anywhere near as large enough to be you know something that we'd want for a diagnostic test this was really interesting and helps us with biology but it doesn't help us to

42:25 determine which patients should get this drug and which patients who can take it safely and who are the ones who might be risky um and then the final point about the use of genetic data has to do with stratification so it's possible to compute for each person the number of schizophrenia risk alleles common variant alleles that they've inherited this is something which is very easy to do right now and so effectively we could for each of you with genetic data we could figure out how many variants you had on average um compared to other people and so for instance what we see here is this is the curve in a large sample for controls this is the curve for cases um the difference between the the means is enormously significant it's about six standard deviations if

43:12 this were height it would be on the order of like four or five centimeters and so this is a typical example where we've got a shift in the mean but it's not to the point of being diagnostic or near diagnostic it's a probabilistic risk factor rather than the deterministic one but the thing that's kind of interesting is if you if you drill down you can say okay what's the risk of being in the second decile versus the first so you take these scores you divide them into tens and what's the risk of being the third versus the first etc and you see that you get this this really interesting relationship where in fact if you're in the top decile your risk of schizophrenia is like 15 fold elevated over people in the lowest decile and this is something we found in two samples one from sweden one from the u.s and the danes had a looser definition why which is why you don't see as clear of a relationship

43:59 but a couple of things become apparent so if you actually look at people in the lowest decile we have a whole bunch of cases who are down here in the lowest decimal why do these people have schizophrenia they inherited a very low risk of the illness and yet they did not develop the clinical syndrome conversely up here we have about 25 of the top decile or controls these are individuals who inherited you know a lot of schizophrenia alleles and yet do not have the illness what's up with that i know there are five people who are in here who actually have converted over to that group so maybe it's just a delay in some respects but this also allows us some interesting uh possibilities for epidemiology can we look at what protective factors these individuals might have sorry what

44:44 what sort of epidemiological risk factors these people have low genetic risk yet have the disorder or what protective effects these people have so there's some interesting possibilities there but one of the big things one of the big uses of genomic data is to try to get at new targets this empty pipeline is meant to remind us that in many respects the pipeline is dead there's nothing in it i think it's got a little bit better but for most the last five years many drug companies in fact have not been actually working actively on cns drugs and that's a huge difference than maybe 10 years ago when in fact if you looked at the bestseller lists you know often many of the top 25 were actually psychoactive drugs antidepressants and or any psychotics okay so let's talk about assumptions in

45:31 terms of how this might work so first i think the one thing we've learned is you have to be really really strict about genetic evidence it's very easy most genes given that something like 10 of all genes have a protein that's expressed in the synapse it's very easy and about half of all genes are brain expressed it's pretty easy to sort of convince yourself to make a story for a gene if the genetic finding is equivalent a lot of genes are interesting in the cns we have to be really really rigorous about the genetic evidence or else this is going to be just a complete mess that's one of the things we view as being very important um especially in terms of uh what we're seeing from the combined genomic results it certainly appears that more schizophrenia is due to um regulatory variants than protein coding

46:16 variants so it looks like it's more disease of regulation of the amount of proteins rather than the amount of the than actual change in and then the primary structure of a protein in addition you know again we the technology that we may need is may not be something that we have right now and again the optimal approach of how we actually uh try to understand the impact of the emerging genomic findings on the development function of the brain is simply not known right now so i think i think we actually do need multiple integrated approaches and the question arises how can we do it relatively quickly and so what we proposed is effectively trying to take the the model that the pgc has about the model of cooperation about just really working together in a concerted way to try and make maximal

47:02 progress we've taken that to the the wet lab sort of approach this too and so we've managed to put together a whole bunch of people who are um you know really good scientists who are also really interested in working in a collegial way these people are good eggs they're fun to go out and drink beer with and that really is kind of one of the key things quite frankly because if if that's the case then you can do science together um and so we've managed to pull together a team of people who are pharmacology from genomics nico's a developmental neurobiologist who works in zebrafish pamela is a psychiatrist who works a lot in ohmic study other sorts of homicide studies in brain etc and these individuals have all agreed to

47:47 come together to work in this we even have a sort of a complicated workflow as to how this is going to work and this is something that we're in the process of trying to actually operationalize as part of my work here i'll go through this in a set so the first step of course is target selection so it's it's difficult not all things we find in genomics are going to be actually things we could work on even if we wanted to some are extremely difficult to work on some are straightforward and again this is the sort of the crazy scale of data that come into this to the the one pixel that we want to do but the first thing of course is we need to gather all the relevant results process them in a very systematic way database them connect them to other findings so that's something that we have

48:33 we've done already to be honest and again this is something we're going to update many times given that there's a lot of active work happening right now um and the first the things that are easiest are in many ways exxon variation that pops up um and so here what we want the reason why it's relatively straightforward is if the gene for instance has a lot more variance in kids with autism protein changing variants in kids with autism then controls that becomes something relatively straightforward to model in many respects you need to do some work to make sure that you're not getting fooled there's some complexities where that can be a problem um but typically the great thing about it is that it connects directly to a gene and you have a direction of association of the effect in other words it

49:18 the the the variants occur in the gene and they lead to the loss of function of that protein and that's something that we can take and model in multiple systems um we don't have anything like this yet for schizophrenia but there certainly are a good handful of things for autism and larger studies are in progress and we've got some ideas about how we can actually get more discovery there as well the thing oops this was supposed to be different actually go back so one of the things that i think is is pretty important at least in one level is to try to determine what proteins touch connect each other and this is the whole protein protein interaction piece um this is a slide from one or one of our colleagues mark vidal who's a dana farber at harvard and mark has sort of shown that um the protein protein interaction

50:04 data out here um are highly biased and highly focused on a handful of targets so typically these are things involved in ubiquination ubiquinone ubiquitination as well as cell cycle control because those are the ones that have been studied really heavily p53 for instance is a big one here whereas a lot of you can actually find that a lot of genes that are important in the central nervous system haven't been studied at all and so what we propose to do is actually use two different sort of complementary technologies to look at protein protein interactions that are relevant in neural tissue i don't think i have time to go through the details but each of these can be done in a relatively high throughput fashion so one one technology can handle a hundred different genes per year the other can handle several hundred and i think that information will allow us

50:51 to sort of take a protein that we think is involved and expand out to find out what it connects to the guys who do this are very good there's a lot of biases but they're pretty good at figuring out and cross-checking assumptions the next thing that the other technology that we can bring to bear is something called high content screening this typically happens in a 96 or 48 well format where you have a nest of neurons that that can be either from a person they can be from a mouse they can be from a standardized human neural stem cell population differentiated in neurons and then we can manipulate them in many different ways we can not genes we can we can remove genes in singular in combination we can increase the expression of genes and it's possible to using automated

51:36 new microscopy and analysis techniques to actually come up with a fairly solid list of 40 odd parameters per nest per manipulation and again to do this in a matter of weeks um there's also a very in-depth synaptic approach where we work with a synaptic biologist mathias verha in amsterdam um and he's spent a lot of time and money to come up with a way to look at single neurons and then to characterize them with great depth so for instance they can do all the synaptic physiology they can actually they can actually track individual packets of information through the cells they can look at time course they can do a lot of different things as well and again these can be manipulated as well so to come back to the pipeline so

52:24 the first thing we want to do of course is to sort of understand whether there's any patient heterogeneity that might be involved in an exxon finding so we're looking at one specific gene that came up you know clearly with very very strong evidence um the two we reply the two proteomic techniques to understand what it binds to and how the variant changes binding um there's a there's a sort of a computer science bioinformatic approach here and then uh the point is we'd have about 30 people sitting down looking at the results and thinking okay what's the next direction if it's something where we think it's worth going further high content screening and actually doing single neuronal assays are possible and if that's the case we actually have uh some ways where we can actually screen things at relatively high throughput and then there's there's similar

53:10 pipelines to this for um for copy number variation the one that i like in particular are single g and c and b's so here you've got um a situation where cases have more um copy number variants that hit one gene than do controls so again you've connected to a specific gene and you have a dosage effect you know that cases have half the number of copies then controls you need to do some validation work etc but typically this is probably the most actionable class of variance we have for schizophrenia there's at least three of these right now for schizophrenia we have a fourth in the publication pipeline that we like a lot and i think it's true but again this is something that my neuroscience colleagues can actually model they can change the doses of a gene in many different ways in many different systems

53:56 for the very large cmvs we need a driver um if some of these won't be doable 22q11 probably has three drivers 16p11 looks like it has one and that might be something where we could focus on that gene to see what's going on and we're doing some work with nico catsinus on this right now um and then the one that's actually tricky is gwas because typically the the signal is subtle and it's relatively broad um and so the connection to the g to a gene and the direction of effect isn't clear however it certainly is possible to make this connection um i've got a folder where i keep track of these there's several dozen right now of of where people have done the follow-up work and managed to clearly connect experimentally a gwas finding to a gene and with the

54:41 direction of association it's more difficult it's not just a simple lookup it's not just an assumption that a loss of function means something you actually have to do additional work to figure it out um but one of the questions we have is how do you evaluate that at high throughput and here are three associations with schizophrenia that i just wanted to talk through um they're all genome-wide significant but as you can see the the um it's quite complicated so here we've got the one of the calcium components the the l-type um voltage-gated calcium channel the neuronal type um cac and a1c and all of these are genome-wide significant even some over here and they all lie in intron 3. and the question is of course how do you connect this to this gene or maybe a different gene here's another example where we have a

55:28 big area of genome-wide significance but as you can see there are many many genes in the region and here's a third example where we've got an area of genome-wide significance but there's these are nothing there's actually no gene within a couple hundred thousand bases either way and the question is how do we actually understand what these are doing to change the risk for schizophrenia and the answer is is i think you actually have to bring in different data but it's possible to create a very logical workflow as to what you might do i mean the first thing we need to do is to do the dna sequencing to make sure that you've assembled that region correctly that there isn't something funny going on there because occasionally there is um second you need to do some sequencing in brain to see whether there's um any any clear any unknown transcripts that are coming off that region maybe there is a gene there it's just

56:14 not annotated next um it's uh we wanted to see whether it's actually differentially expressed an eqtl study is a way to look at genetic variation and how it affects the amount of transcription and so there's a bunch of large-scale data sets that looked at this increasing numbers in brain as well um attack seek is a way to look for open chromatin um and of course looking at histone marks whether the snip actually has to do you know whether it actually changes um the the repressor enhancer etc in the usual way um and then finally um i think something that's really important is that there's a lot of these looping interactions especially in brain where you like you have an intergenic genome-wide significant association and there's no gene there but it has a regulatory effect on a gene that's 300

57:00 000 bases away we're seeing increasing examples of this what we need is we actually need a set of data where this has been done very carefully relatively large sample sizes and neural tissues fortunately we've done this in a pilot work in my group um and there's actually a bunch of large scale efforts that are happening right now and we're part of something called psych encode where we're like the encode project where we're doing this but unlike encode which focused nearly exclusively on cancer cell lines and just regular cell lines we're doing it all in brain and so there's an open source effort to do this pamela square is one of the principles of the common mind consortium where they're actually doing a lot of sequencing and brains of people's schizophrenia versus controls as well as a lot of the encoding type stuff as well

57:47 okay so that really comes down to this right there which is the functional genomic data so i think it's possible to take a geos association overlay other examples and come up with a hypothesis as to what's happening it's also then possible to to design a wet bench experiment to prove it to see whether or not that hypothesis is correct and that's the that's the the next step that i think we need to do okay so final slide i wanted to thank all kinds of people to be honest um i've got two labs now and i owe them a huge uh debt of gratitude because i learned stuff from them every day um i started coming over here uh back around just before 9 11 actually to do it with nancy pedersen expanded out to work with paul paul lichtenstein um christina holtmann

58:33 patrick magnuson sarah bergen are all people we've done a lot of work with and i certainly owe them a great debt because i wouldn't be standing here if they hadn't been so kind to me over the years um the site genomics consortium i'd like to thank the cast of hundreds um the swedish research council has been extremely kind as well to allow me to come here and then we've had many many funders who've really made critical contributions along the way and i thank you for your attention and i think we have time for questions you