The state of psychiatric genetics
The world congres of psychiatric genetics went beyond GWAS, and with that come new challenges.
I try to synthesise and cherry pick the content of various presentations I saw at world congress of psychiatric genetics (WCPG) I’ll cover what I feel where widely shared concerns, trends, both promising developments, and the challenges I think these might present us. This overview is based on talks, parts of talks, discussions in the hallways or during lunch, preprints people shared, slides people send me and in some cases conversations deep into the night (Montreal has one of the best cocktail bars I have ever ben too: Nhậu bar, in the basement of restaurant Ha). The topics I selected are a deeply personal set of impressions as I am obviously guided by my own interest when deciding what session to attend. It will therefore be somewhat light on biology as I am no expert on things like single cell sequencing, or imaging based neuroscience. This post is titled “The state of psychiatric genetics“ but really its okay to come away thinking it should have been called “navel gazing with Michel Nivard”, psychiatric genetics is a big tent with many perspectives on what is critical, or important, and that’s very healthy! Finally, when I highlight certain presentation or work, and then move on to discuss a statistical challenge I picked the presentation because I love the ideas/concepts and its done well, the challenge is a broader issue challenging all of us.
Before I departed for WCPG, I wondered whether we as a field, those who study the genetic epidemiology of psychiatric diagnoses, where about to be “the dog who caught the car”. Genome wide association studies, where all common genetic variance (variants carries by more than ~ 1% of us) are associated with a diagnoses are getting so large that we are about to find all we can find. I think the question on many minds was “okay, now what?’.
Global representation
A common criticism of the field, and complex trait genetics more broadly, is that genome wide association studies were almost entirely based on western European ancestry individuals, and this would limit generalisation of certain types of findings. This year it became clear that that criticism has been heard, and some have taken action to ensure both our studies, and crucially the scientific workforce in our field, will become more representative. Tens of thousands of research participants are enrolled in Asian, African and south American biobanks and research studies. Regional and continental consortia to foster regional collaboration were announced. These studies are often (co)led by local PI’s and where needed these are supported by training and capacity building. However, as noted by Cathryn Lewis, western visa laws mean many global collaborators can’t join us at these conferences!
So clearly there is work for us as a field to ensure people stand a chance at a visitors visum and perhaps we should as scientific societies weight local visa polices in conference location selection.
Cross disorder analysis
Based on very high genetic correlations between psychopathologies, many in psychiatric genetics are interested in analysis across diagnostic boundaries. Given work led by me and my collaborations (see HERE) I almost feel like I should offer you the reader a disclaimer: I am a totally on board with this kind of work. However, cross disorder analysis doesn’t always have to mean you directly study multiple disorders. This year we saw a lot of sessions on features or symptoms that occur across multiple disorders, features the classifications like HITOP (Hierarchical taxonomy of psychopathology) classify as lower level common features that tie across diagnoses and are potentially socially, psychological and biological more coherent targets of study than diagnostic categories.. We saw sessions and GWAS of features like sleep, fussy eating and avoidant or restrictive eating, hot flashes and impulsivity.
The work on Impulsivity presented by Travis Mallard (scholar page), is worth highlighting as a great example of making genetic analysis inform how we think about psychopathology. Impulsivity is a low level feature though to to be central to various externalising diagnoses and symptoms. The appeal of identifying a “lower” level feature is that it would be potentially closer to biology/neuroscience/psychology than the diagnosis. Mallard et al. set out to investigate how coherent impulsivity actually is. They performed GWAS of 8 measures of impulsivity, as measured across two validated questionnaires in 123.000 people, and analyse whether these measures share a common cause. They conclude that, superficially they do, either a 1 factor model or if you squint a 2 factor model explains a substantial portion of the genetic covariance between the impulsivity measures. They then use GenomicSEM to perform a factor analysis GWAS, finding that; while the factor model fits reasonably, very few SNPs actually influence the measures as you would expect given the factor model factor! Rather the hits they find (wont reveal these don’t want to pre-empt their work) quite a few genetic variants have opposite effects on various impulsivity measures, or only influence one or two specific impulsivity measures.
A common limitation of GWAS cross disorder work is that while we are measuring multiple related outcomes, rarely do peopel consider whether these occur in specific sequences, co-occure in the same eprson at the same time or simply share genetic causes without very frequently co-occuring. Analytic tools based on GWAS summary statistics, like GenomicSEM aren’t suited for those kinds of analyses (yet…), and cross sectional diagnostics data, or data that ignores age/cohort simply wont be sufficient. Joeri Meijsen (google scholar) presented work at this WCPG based on decades of Danish national register data, family analysis stratified across age and cohort very clearly showed secular changes in both heritability and genetic correlations between outcomes over time.
Disease progression/Case analysis
Most psychaitric genetics GWAS focus entirely on diagnosis. A logical next step is to figure out, initially observationally, what the genetic variants you find in GWAS do in affected people. there were a number of presentations on outcome sin large cohorts of thousands of diagnosed individuals, measured in detail. Some of these follow up studies focus on large datasets (10.000 affected individuals or even more… ) others, like Jet Termorshuizen (google scholar) possibly anticipating even bigger GWAS, zoom into qualitative interviews with tens of people with eating disorders and very high or low genetic risk scores, and the physicians who interact or treat these people.
A very cool and specific specific example came up in a symposium on Genetics around the world, and of which there is a preprint available, is the effort to use physician labeling and natural language processing models retrieve longitudinal symptoms and signs of severe psychopathology from intake, discharge and progress notes in EHR data. The ability to accurately track the presence of features like for example delusions, suicidal ideation and hallucinations within patients trough time will potentially enable research on diagnostic switching, co-occurence of diagnosis and features within a person and trough time. There were similar themed talks on bipolar subtypes within cases within up to 20.000 individuals diagnosed with bipolar disorder (cant find speaker name in my notes), where the features like onset, psychosis, rapid cycling where studied, and genetically correlated within people with a bipolar disorder diagnosis. In the eating disorders sessions there was a presentation by Zeynep Yilmaz (website) where signs of comorbidity in ~7000 people with a eating disorder diagnoses where studied.
Not unlike the theme that emerged in the context of cross disorder analysis, there appears to be great and legitimate interest in studying progression, mechanism, symptoms within people with a diagnosis. Very clever analysis of qualitative data collected in electronic health records seems to further enable the study of temporal progressions, co-occurence of, and switching between diagnoses and symptoms within a person over time. The big challenge in studying a population affected with a specific diagnosis is that you implicitly always condition on disorder risk, a form of index event bias. Index event bias, as I have outlined on twitter yesterday, can cause restriction of range and collider bias and those can really mess with your results (Click the tweet to read the thread if you want to know more).
As I outline in the twitter tread, conditioning on disorder very likely results in collider bias or range restriction, both of which will influence results downstream analyses of features in only cases. Most researchers at WCPG seemed aware and cautious of some of these risk, or tried to pick analyses to avoid them, but it is pitfall we must be weary of. There are plenty statistical and analytical solutions and I am confident their implementation will unlock very insightful analyses of disease process, mechanism and progression beyond mere diagnostic labels. I think its worth considering targeted retraining. as we move beyond our bread and butter the case/control GWAS. As we do so, we inevitably veer into epidemiology, and we may label it “genetic epidemiology” or “molecular epidemiology” but it is fundamentally epidemiology and requires us to offer our students and trainees appropriate training and knowledge.
Rare variation
One old answer to that question of “what’s after GWAS” has always been “studying rare variation”, while GWAS points to regions in the genome that associate with disease, very many regions in fact, all of which harbour common variants with small effects, which per allele raise the risk of schizophrenia with at most ~10%. That sounds like a lot but since only 1% of people develop schizophrenia, this means carries would have a 1.1% risk (I am simplifying the math her committing many statistics crimes). These GWAS end up pointing to hundreds of regions of the genome, near genes expressed in the brain and in neurons. However, the multitude of GWAS “hits”, in other words that fact that traits are incredibly polygenic (caused by many genes) poses a problem to biology. Biology becomes incredibly hard if you have very many potentially weak causal targets to study in labs, in zebrafish, cells etc.
there is solid evidence that rare genetic variation related to severe psychiatric outcomes will be found in far fewer, more specific, regions of the genome, which would easy biological analysis. At this WCPG we saw preliminary presentations of rare variant analysis for Bipolar disorder (presented by Calwing Liau on behalf of a huge consortium), and continued efforts for Schizophrenia (didn’t note the speaker, wish I could give them a shoutout). These analyses involve well over 50.000 exome sequenced cases, and identify genes in which the burden of ultra rare protein truncating variants has huge effects on psychiatric outcomes (odds ratio for carriers in specific significant genes all the way up to ~20) , or or the burden of slightly less damaging but equally rare missense variant have substantial effects (odds ratio for carriers in specific significant genes ~5). Its not the case that rare variants have bigger effects in general, rather variants with rare deleterious effects never become common, due to selection against them. So the vast majority of ultra rare variants is entirely benign, we have to focus on and filter the deleterious variants using biological knowledge. Improved filtering of these deleterious variants from among benign variants based on evolutionary constrained, knowledge of protein structure, and machine learning models of the expected impact on the protein (alpha fold misssense) means the authors narrow in on very probably causal missense mutations.
Trough talking to people that know better then me, it becomes clear that this work will eventually lead to sets of 20-300 individuals with the same psychiatric diagnosis and a disruption to the same gene. When eventually several hundred carriers are identified per affected gene/diagnosis combination, these individuals, and their phenotypes/life-course could provide sufficient grounds for clustering phenotypic features influenced by the gene (prodromal symptoms, signs, neuroscience, biochemical measures etc). There is a group of researchers who, informed by preliminary evidence and solid theory, are making a big bet on finding key genes related to psychiatric outcomes in sufficiently large groups. Their bet is these rare large effect disruptions may not explain a big portion of individual differences between people (they are to rare for that) but they likely will bring us way closer to actionable biology. Off course conditioning on a single gene/psychiatric diagnosis is also a form of case only analysis, and so there is still this serious risk of range restriction and collider bias discussed before.
Psychiatric genetics in the context of family, environment and society.
Genetic effects on mental health do not exist in a vacuum, rather they play out in various environmental contexts. Several researchers highlighted work that placed genetic effects on psychiatric diagnoses and symptoms in the context of family, community and society. Specifically genes in parents, trough their infuence on the familial, social, economic environment might influence outcomes in their kids. But, genes in kids might also influence health outcomes in their parents and due to the un-even distribution of care specifically mothers. Ziada Ayorech (faculty website) who stepped in for a speaker that couldn’t make it outlined 3-4 papers that collectively set out the rational and value of her own work on maternal mental heath in the context of the MoBa study. The MoBa study studies the health of a huge Norwegian birth cohort (±110.000 kids) at ±30 timepoints from early pregnancy to adolescence, but crucially also considers the mental health of the mother both during pregnancy, and as the kids grow up (they also genotyped and survey the father albeit less frequently). But as she outlined the MoBa study goes way further than that, data can selectively be tied into national registers on medication, hospitalisation, income, school performance, housing, neighbourhood, social and physical mobility etc etc. Finally rhe people working the data seem to be going at it with a plan and with solid theory, which is very exciting. There where talks by Beate St Pourcain (scholar) on low prosocial behaviour, and peer or social problem in children and adolescents, as well as a talk by Andrea Allegrini (google scholar) on the relation between polygenic scores for adult psychopathology and hierarchical models of childhood pathologies. Andrea’s work revealed gene environment correlation where parental genotypes influenced their kids behavior, or their rating of their kids behavior (subtle difference thats hard to parse) over and above the effect of the parental genotype inherited by the child These presentations clearly dug into important developmental, social and contextual aspects in which we should place the genetics of psychiatric disorder, and did so in well reasoned ways,. RThe presentations also highlighted how even with full register access on hundreds of thousands of Norwegians, and measures of child development, we are scrambling to find comprehensive ways to measure the relevant environments. SNP heritabilities of childhood measures are fairly low, which to my mind likely reflects the noise in their measurement across instruments, raters, cohorts, age, place etc. To an extend having access to very rich nationalregister data is a huge gain in this respect, but I feel to get close to a true measure of the social and familial environment we’d need to rconsider whether we can bring int type of extensive analysis of admission, progress and release notes written by mental health professionals in EHR records, we saw mentioned elsewhere at WCPG.
There is a key second way in which psychiatric genetics in enbedded in society and culture, Robbee Wedow a sociologists with experience at the Broad institute (talk about rare intersections…) spoke as discussant in a session on SES and Mental health and send us home with the following abstract from Bruce Link’s 2008 paper on the social shaping of population health:
“When biomedical knowledge and technology create the capacity for humans toavoid disease and circumvent early death, socioeconomic factors become more, not less important for population health. The transformation of disease causation from cruel fate, accident, and bad luck to circumstances that are under some degree of human control facilitates a powerful social shaping of disease and death. When humans have control, it is their policies, their knowledge, and their behaviors that shape the consequences of biomedical accomplishments, and thereby extant patterns of disease and death. Absent a robust societal investment in social epidemiology, population health will reside below its optimal level and the maldistribution of health-enhancing innovations will continue to create health disparities.” (Bruce Link 2008, paper here)
Robbee’s discussion, and the Link paper really struck a chord with me. We talk big about translational medicine, from molecule to mind and then to medicine. But the nature of mental health is such that severely delayed or unaffordable care is that translation into healthier outcomes is only halfway done if we ever get to better psychiatric medication, better psychiatric epidemiology, and psychiatric nosology trough psychiatric genetics. The (local) organisers for WCPG2023 seemed keenly aware of some of these issues when the choose their theme: “Putting people first in psychiatric genetics“, lets make sure we collectively follow trough.
closing thought…
Conferences, by their very nature, are on a yearly cadence, maybe once every two years for those who minimise travel, and therefore its only natural we present and discuss our day to day progress. We are understandaly consumed with the immediate future, the tactics of science, but we are at a point in out field where we should probably talk strategy, long term strategy. Obviously there was a lot of that happening in specific sessions and presentation, but the nature of conferences and big consortia like the PGC is that we are far more likely to present important incremental progress. I would love for people to devote serious time at future WCPGs zooming out from their day to day, and explicitly discuss their long terms goals and visions, obviously backed up by rigorous theory and build on solid empirical support.
Thanks for your summary of WCPG! Looks like lots of interesting work. I'm most interested in the (ultra)rare variant work.
Is there a resources where the psychiatric SNP-h2s approaching saturation are summarized? I've seen Dr. Appie's prior summary figures but wondered if there there's a good review of this progress that you recommend or other resources.