(WCPG): Oral Abstracts

ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] European Neuropsychopharmacology (2018) 000, 1–44

www.elsevier.com/locate/euroneuro

Abstracts of the 26th World Congress of Psychiatric Genetics

(WCPG): Oral Abstracts

Friday, October 12, 2018 Results: All PRS significantly predicted MDD status. When stratified by early and late onset, PRS for coronary artery disease, stroke, BMI, and type 2 diabetes also significantly predicted late onset MDD, with coronary artery disease, Oral Session: Depression BMI, and type 2 diabetes also predicting early onset MDD.

1:30 p.m. - 3:00 p.m. Significant genetic correlations were identified between MDD and the cardio-metabolic traits. I will also discuss find- 1 ings from age at onset stratified genetic correlations. GENETIC COMORBIDITY BETWEEN DEPRESSION AND Discussion: This study provides possible support for the vas- CARDIO-METABOLIC DISEASE, STRATIFIED BY AGE AT cular depression hypothesis by showing a genetic association ONSET between late onset MDD and cardio-metabolic traits.

∗ Saskia Hagenaars 1, , Jonathan Coleman 1, Disclosure: Nothing to disclose.

1 1 Helena Alexandra Gaspar , Karen Hodgson , The PGC MDD doi: 10.1016/j.euroneuro.2018.08.008 Consortium, METASTROKE Consortium 2, Gerome Breen 1,

Cathryn Lewis 1 2 1 King’s College London DOES HIGH BMI IN THE ABSENCE OF METABOLIC 2 University of Virginia CONSEQUENCES CAUSE DEPRESSION?

∗ Background: The association between major depressive dis- Jess Tyrrell 1, , Anwar Mulugeta 2, Andrew Wood 1, order (MDD) and cardio-metabolic disease is well estab- Ang Zhou 2, Robin Beaumont 1, Marcus Tuke 1, lished. MDD increases the risk of cardio-metabolic disease Samuel Jones 1, Katherine Ruth 1, Hanieh Yaghootkar 1, onset and mortality, but cardio-metabolic disease itself can Cathryn Lewis 3, Tim Frayling 1, Elina Hypponen 2 also increase risk of developing MDD. However, the underlying mechanisms of the association between MDD and cardio- 1 University of Exeter metabolic disease remain elusive. A possible mechanism 2 University of South Australia may be represented by cerebrovascular disease and its risk 3 King’s College London factors predisposing MDD in later life, according to the ‘vas- cular depression’ hypothesis. Background: Depression is more common in obese than non- Methods: Polygenic risk score (PRS) analysis was used to ex- obese individuals, but the causal relationship between obe- amine how coronary artery disease, stroke, and type 2 dia- sity and depression is complex and uncertain. Previous stud- betes predict MDD status, stratiﬁed by early and late age ies have used Mendelian randomisation to provide evidence at onset in the Psychiatric Genomics Consortium and UK that higher body mass index (BMI) causes depression but Biobank (N early onset MDD ∼ 16K, late onset MDD ∼16K, have not tested whether this relationship is driven by the controls ∼67K). Genome-wide association studies were per- metabolic consequences of BMI or differs between men and formed with MDD cases stratiﬁed by early and late age at women. onset, in order to calculate genetic correlations between Methods: We performed a Mendelian randomisation study cardio-metabolic traits and early and late age at onset for using 48 791 individuals with depression and 291 995 con- MDD. trols in the UK Biobank to test for causal effects of higher BMI on depression. We used two genetic instruments, both

0924-977X ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 2 Abstracts representing higher BMI, but one with and one without its (0.87, s.e. = 0.04) with Wray et al (2018). This provided a adverse metabolic consequences, in an attempt to “un- total of 807,553 individuals (246,363 cases and 561,190 con- couple” the psychological component of obesity from the trols) with which to conduct the meta-analysis and further metabolic consequences. We further tested causal relation- downstream analyses. ships in men and women separately. Results: The genome-wide meta-analysis of the three con- Results: Higher BMI was strongly associated with higher odds tributing cohorts revealed a total of 118 independent sig- of depression, especially in women. Mendelian randomisa- nificant (P < 5 × 10 −8) variants associated with depres- tion provided evidence that higher BMI partly causes de- sion. The estimate of the heritability on the liability scale pression, with a genetically determined 1 SD higher BMI was 0.089 (0.003) with genetic correlations identified for (4.9 kg/m2) associated with higher odds of depression in 33 other traits, including bipolar disorder, schizophrenia, all individuals (Odds ratio (OR) 1.18 [95%CI: 1.09-1.28], smoking and triglyceride levels. We identified 269 genes and P = 0.00007) and women only (OR: 1.24 [95%CI: 1.11-1.39], 14 gene-sets that were significantly associated with depres- P = 0.0001). Meta-analysis with 45 591 depression cases and sion, including gene-sets involved in synaptic transmission. 97 647 controls, from the Psychiatric Genomics Consortium Discussion: This research presents the results from a rela- (PGC) strengthened the statistical confidence of the find- tively well-powered genetic analysis of depression, by com- ings in all individuals. Using a metabolically favourable adi- bining previous studies of the condition. A marked increase posity genetic risk score, and meta-analysing data from UK in the number of associated variants and genes was ob- Biobank and PGC, a genetically determined 1 SD higher BMI served which will enable additional comprehension of the (4.9 kg/m2) was associated with higher odds of depression underlying biological aetiology. The gene-sets identified as in all individuals (OR: 1.26 [95%CI: 1.06-1.50], P = 0.010. significant typically relate to brain function and may ulti- Discussion: Higher BMI, with and without its adverse mately allow us to identify the molecular mechanisms un- metabolic consequences, is likely to have a causal role in derlying genetic predisposition to depression. determining the likelihood of an individual developing depression. Disclosure: Nothing to disclose. doi: 10.1016/j.euroneuro.2018.08.010 Disclosure: Nothing to disclose. doi: 10.1016/j.euroneuro.2018.08.009 4 DETERMINING THE RELATIONSHIP BETWEEN CANNABIS 3 USE AND MAJOR DEPRESSION IN UK BIOBANK GENOME-WIDE META-ANALYSIS OF DEPRESSION Karen Hodgson ∗, Saskia Hagenaars, Kirstin Purves, ∗ David Howard 1, , Mark J. Adams 1, Toni-Kim Clarke 1, Helena Alexandra Gaspar, Kylie Glanville,

Jonathan D. Hafferty 1, Jude Gibson 1, Jonathan Coleman, Marta Di Forti, Gerome Breen,

Jonathan R.I. Coleman 2, Ian J. Deary 1, 23andMe Research Paul O’Reilly, Cathryn Lewis Team, Major Depressive Disorder Working Group,

Daniel J. Smith 3, Patrick F. Sullivan 4, Naomi R. Wray 5, King’s College London

Gerome Breen 2, Cathryn M. Lewis 2, Andrew M. McIntosh 1 Background: Cannabis is the most widely used drug world-

1 University of Edinburgh wide. With many countries engaged in debates about the

2 King’s College London legalisation of cannabis, it is important to fully understand

3 University of Glasgow any potential relationships between drug use and mental

4 University of North Carolina at Chapel Hill health outcomes. Psychosis is often the focus of work in

5 University Of Queensland this field, but cannabis use has also been linked with both depression and self-harm behaviours (Hodgson et al 2017, Background: Depression is a heritable and polygenic condi- Agrawal et al 2017, Smolkina et al 2017). Here, we take tion that causes extensive periods of disability and increases advantage of successful genome-wide association studies the risk of suicide. Previous genetic studies have identified for both cannabis use (Stringer et al 2016, Pasman et al a number of common risk variants which have increased in 2018) and depression (Wray et al 2018), to examine the re- number in line with increasing sample sizes. The availability lationships between these prevalent phenotypes with much of summary statistics from these studies represents an op- greater statistical power. Specifically, we determine the portunity to conduct the largest genome-wide meta-analysis phenotypic and genetic relationship of cannabis use with study of depression to date. depression and self-harm, within the population-based UK Methods: We examined the effect of 8,098,588 genetic vari- Biobank. ants on depression by performing a meta-analysis of the Methods: In UK Biobank, the mental health questionnaire summary statistics from the Hyde et al. (2016) analysis includes items assessing lifetime history of cannabis use, of 23andMe, the Wray et al. (2018) analysis of the Major depression and self-harm. Focusing on unrelated individu- Depressive Disorder Working Group of the Psychiatric Ge- als of European ancestry with high quality genotypic data nomics Consortium data, and the Howard et al. (2018) anal- available (n = 126,291), n = 28,282 report a life-time history ysis of UK Biobank. We used the broad depression phenotype of cannabis use, n = 30,075 report depression and n = 5,520 from UK Biobank which shares a high genetic correlation report self-harm. ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 3

We examined the phenotypic relationships between these Background: Large, rare copy number variants (CNVs) are traits. Then, using LD score regression and polygenic risk associated with neurodevelopmental disorders but it is un- scoring with recently available summary statistics (Wray et clear whether such CNVs also contribute to the risk of de- al 2018, Stringer et al 2016), we tested the genetic relation- pression. A substantial proportion of the CNV enrichment ship between phenotypes. Finally using Mendelian Randomi- in schizophrenia is explained by CNVs associated with neu- sation, we test evidence for the causal direction of effects. rodevelopmental disorders. Depression shares genetic risk Results: Within UK Biobank, cannabis use is associated with schizophrenia and frequently occurs comorbid with with an increased likelihood of depression (OR = 1.64 95% neurodevelopmental disorders. We, therefore, aimed to ex- CI = 1.59-1.70, p = 1.19 × 10-213), and amongst those with amine the relationship between CNVs and depression in the depression, is associated with recurrence of depression UK Biobank sample. We hypothesised that (i) neurodevelop- (OR = 1.28, 95% CI = 1.22-1.35, p = 3.42 × 10-20). Cannabis mental CNVs are associated with increased rates of depres- use is also associated with an increased likelihood of self- sion and, (ii) after excluding CNVs relevant to the primary harm (OR = 2.85, 95% CI = 2.69-3.01, p = 3.46 × 10-304). hypothesis, there is a residual unexplained CNV burden. Using LD score, we observe significant genetic correla- Methods: We called CNVs using PennCNV-Affy in 455,913 in- tions between cannabis use and both depression (rg = 0.283, dividuals from UK Biobank who reported white British or SE = 0.055, p = 2.00 × 10-7) and self-harm (rg = 0.304, Irish ethnicity (37-73 years, 54% female) and annotated 53 SE = 0.075, p = 5.44 × 10-5). We will also present results CNVs associated with neurodevelopmental disorders. We de- using polygenic risk scores calculated from available sum- fined depression as a self-report of having received a de- mary statistics (Wray et al 2018, Stringer et al 2016) to pression diagnosis (24,575 cases, 5.87%) and then repeated predict both cannabis and depression-related phenotypes. our analyses in two alternative depression variables –self- Finally, we discuss the use of Mendelian Randomisation reported depression and antidepressant prescription at visit to probe the nature of the causal pathways linking these 1 (15,748 cases, 3.76%) and hospital discharge diagnosis of phenotypes. depression (11,395 cases, 2.94%). In addition, we used data Discussion: In this study, we observe phenotypic and genetic gathered from an online follow-up mental health question- associations between cannabis use with both depression and naire to further characterise the depression phenotype. self-harm. Stronger phenotypic relationships are observed Results: The group of 53 neurodevelopmental CNVs was as- for more severe phenotypes of recurrent depression and sociated with self-reported depression (OR 1.36, 95% CI 1.22 persistent cannabis use, echoing previous literature. – 1.51, p 1.61 × 10-8) and this result was consistent when We note that the observed effect sizes are larger than repeated in the two alternative depression variables. The those reported previously for psychosis-related phenotypes. association was partially explained by social deprivation, a For example, in a large phenotypic meta-analysis, Moore known risk factor for depression. There was a nominally sig- et al (2007) report OR = 1.41, 95% CI 1.20–1.65 between nificant association between carrier status of CNVs greater cannabis use and psychotic outcomes, whilst Verweij et al than or equal to 500kb and depression, but this result did (2017) calculate a genetic correlation of rg = 0.22 (SE = 0.07) not survive Bonferroni correction for 5 tests. Exploratory between cannabis use and schizophrenia. analyses found neurodevelopmental CNVs to have a stronger Given the strength of both phenotypic and genetic as- effect on risk of depression in females (OR 1.48, 95% CI 1.30 sociations for cannabis use with depression and self-harm, - 1.69, p 3.07 × 10-9) compared to males (OR 1.16, 95% CI together with the high prevalence of these traits, further 0.96 – 1.39, p 0.12) for the primary depression phenotype. research into understanding the causality of these associ- Discussion: We report the largest study of CNVs in depres- ations is important in the light of changing attitudes to sion to date and the first to robustly demonstrate associ- cannabis use. ation between CNVs and depression. Four CNVs were individually associated with depression at levels of significance Disclosure: Nothing to disclose. which survived Bonferroni correction for the 53 CNVs tested (1q21.1 duplication, PWS duplication, 16p13.11 deletion, doi: 10.1016/j.euroneuro.2018.08.011 16p11.2 duplication). In addition, our study identifies a link between neurodevelopmental CNVs and social deprivation, a potential target that may have a beneficial impact on 5 rates of depression in CNV carriers. Altogether, these find- THE ROLE OF NEURODEVELOPMENTAL COPY NUM- ings indicate that neurodevelopmental CNVs may offer im- BER VARIANTS IN DEPRESSION portant insights into the causes of and pathways to depression. ∗ Kimberley Kendall 1, , Elliott Rees 2,

Matthew Bracher-Smith 1, Lucy Riglin 3, Stanley Zammit 3, Disclosure: Nothing to disclose.

Michael O’Donovan 3, Michael Owen 1, Ian Jones 3, doi: 10.1016/j.euroneuro.2018.08.012 George Kirov 3, James T. R . Walters 3

1 MRC Centre for Neuropsychiatric Genetics and Genomics, Cardiff University

2 MRC Centre for Neuropsychiatric Genetics and Genomics, Institute of Psychological Medicine and Clinical Neuro- sciences, Cardiff University

3 Cardiff University ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 4 Abstracts

6 of brain cerebellar tissues, while weak preservation of mod- A MULTI-TISSUE NETWORK-BASED ANALYSIS OF RISK ules was observed between brain and whole blood. GENES FOR MAJOR DEPRESSION Discussion: Genes are known to interact with one another in highly organised networks. We show the network-based ∗ Zachary Gerring 1, , Eric Gamazon 2, 23andMe Research analysis of multi-tissue gene expression data and the sub-

Team, Eske Derks 3 sequent integration of GWAS summary statistics for MD provides a new and biologically meaningful approach for the

1 QIMR Berghofer Medical Research Institute interpretation of disease-associated loci. The identiﬁcation

2 Vanderbilt Genetics Institute, Vanderbilt University of networks and biological processes can be further inte-

3 QIMR Berghofer grated with drug repositioning strategies, and may thereby identify molecular targets for the development of diagnos-

Background: Major Depression (MD) is a highly disabling tic tests and precision treatment strategies for MD. mental health disorder that accounts for a sizable propor- Disclosure: Nothing to disclose. tion of the global burden of disease. MD has a multifacto- rial molecular background, driven in part by a highly poly- doi: 10.1016/j.euroneuro.2018.08.013 genic mode of inheritance. A recent genome-wide association study (GWAS) of 130,664 MD cases and 330,470 controls identified 44 single nucleotide polymorphism (SNP) loci Oral Session: Schizophrenia associated with the disorder (Wray et al., 2018). Detailed 1:30 p.m. - 3:00 p.m. functional studies showed SNP loci are enriched in multiple brain tissues, but not peripheral tissues, and contain SNPs 7 that regulate the expression of multiple genes with puta- SCHIZOPHRENIA POLYGENIC RISK SCORE ANALYSIS tive roles in neurite growth and synaptic plasticity. How- IN 22Q11.2 DELETION SYNDROME ever, the functional gene mechanisms underlying the statistical associations remain unknown. Pathogenic mechanisms Thomas Monfeuga 1, Michael P. Epstein 2, underlying MD risk are hypothesised to involve hundreds of Aaron M. Holleman 2, Isabelle Cleynen 3, Henry Johnston 2, genes across multiple loci that interact with each other Yingjie Zhao 4, Donna M. McDonald-McGinn 5, in large co-expression networks. Thus, the study of multi- Raquel E. Gur 6, Stephen T. Warren 2, Joris Vermeesch 3, tissue gene co-expression networks may identify functional Beverly S. Emanuel 5, Bernice E. Morrow 4, Anne S. Bassett 7, genes and processes in MD. ∗ Nigel Williams 1, Methods: Normalised RNA-Seq data for 13 brain tissues and whole blood were downloaded from the Genotype-Tissue 1 Cardiff University Expression (GTEx) study portal (version 7). MD GWAS sum- 2 Emory University mary statistics for 10,468,942 autosomal SNPs (Wray et al., 3 KU Leuven 2018) were provided by the Psychiatric Genomics Consor- 4 Albert Einstein College of Medicine tium (PGC). First, we built unsigned gene co-expression net- 5 Children’s Hospital of Philadelphia works for 13 brain tissues and whole blood using a weighted 6 University of Pennsylvania gene co-expression network analysis (WGCNA). Each tissue- 7 University of Toronto specific gene co-expression network was divided into modules (groups) of correlated genes using hierarchical clustering. Second, we characterised biological pathways and Background: It is now well established that individuals with processes in gene modules using g:Profiler. Third, we used 22q11.2DS are at an increased risk of psychosis, however as ∼ MAGMA (v1.06) to assign MD SNP associations to genes only 40% of adults with 22q11.2DS develop psychotic symp-

(risk genes) and subsequently test for the enrichment of toms, the increased risk is clearly not fully penetrant. Multi- risk genes within the tissue-speciﬁc gene modules. Finally, ple hypotheses have been proposed for this incomplete pen- we assessed the preservation (replication) of gene mod- etrance, including breakpoint heterogeneity, environmental ules across brain tissues and whole blood using the module contributions, and an effect of other genetic variants either

Preservation function of WGCNA. within the remaining allele and/or outside the 22q11.2 lo-

Results: Hierarchical clustering of tissue-specific gene co- cus. Polygenic Risk Score (PRS) analysis allows the risk con- expression networks formed highly organised gene modules ferred by common genetic variants associated with a disor- with enrichments in specific biological processes. Gene- der to be considered in aggregate. Studies of large cohorts based analyses identified 137 genes associated with MD of idiopathic schizophrenia have demonstrated that cases risk (P < 2.77 × 10-7). MD risk genes were enriched in have a significantly greater PRS composed of schizophre- a single module in 8 out of 13 brain tissues, with the nia risk alleles than unaffected controls. In this study we strongest association in anterior cingulate cortex (cor- used analysed schizophrenia PRS to investigate the effect rected P = 1.00 × 10-4). No enrichment of risk genes was of common genetic variants in the aetiology of psychosis in observed in brain cerebellar tissues or whole blood. MD 22q11.2DS patients. gene modules contained genes involved in synaptic sig- Methods: Whole-genome sequencing (WGS) and genome- nalling (P < 4.07 × 10-66) and nervous system develop- wide genotyping array data was provided by the 22q11.2DS ment (P = 1.58 × 10-51). Gene co-expression modules were Brain and Behaviour Consortium. WGS data was available strongly preserved across brain tissues, with the exception for 435 samples with known psychosis status (214 with psychosis, 221 without), of which 374 had also been genotyped ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 5 using Affymetrix arrays (178 with psychosis, 196 without). Methods: A total of 466 schizophrenia patients recruited We calculated PRS in both datasets with the software PRSice at the Centre for Addiction and Mental Health (CAMH) (v2), using the schizophrenia GWAS of the Psychiatric Ge- (Toronto, Canada) were sequenced using the Affymetrix SNP nomics Consortium 2 (PGC2) as the discovery sample. Re- Array 6. 0 assay. CNVs were called using two algorithms, in- gression analyses were performed to test if psychosis status cluding the Canary Software and PennCNV. Only overlapping could be discriminated by PRS, after correcting for gender CNV calls from both algorithms were used for further analy- and population stratification. sis to increase the confidence of each CNV called. Currently Results: Using the genotypes from WGS data only (n = 435), used clinical CNV size thresholds of 200kbp and 500kbp were the 22q11.2DS individuals with psychosis had a significantly used to filter all called deletions and duplications, respec- higher schizophrenia PRS than 22q11.2DS patients with no tively. All CNVs were individually assessed to characterize known history of psychosis (p < 10e-6, Nagelkerke’s R2 ∼ the presence of known, clinically well-characterized CNV 0.8 at SNP p-value threshold pT = 0.01. Conducting PRS anal- syndromes with the use of the UCSC Genome browser, DECI- ysis SNP array genotypes for the 374 samples generated PHER GRCh37, and GeneReviews® database. analogous results, confirming the robustness of the method. Results: A total of 861 deletions and 171 duplications were Discussion: This study has used the largest available sample called on 348 schizophrenia patients, which passed qual- of 22q11.2DS patients to demonstrate that 22q11.2DS pa- ity control and size threshold filtrations. Upon CNV analy- tients with psychosis have a significantly increased PRS com- sis, a total of 14 patients were identified with a total of posed of common schizophrenia risk alleles. This suggests 18 CNVs associated with previously recognized CNV syn- that genetic variants outside of the 22q11.2 deleted region dromes. A further 30 patients were identified to have one also have an important role in mediating the increased risk or more genes of interest associated with brain-related phe- to schizophrenia in 22q11.2DS. Current work aims at mod- notypes within 37 CNVs. Taken together, in our SCZ co- elling whether the schizophrenia liability in 22q11.2DS pa- hort, 4.02% (14/348) of patients were identified to have tients is due to the deletion and common variants additively. known pathogenic CNVs and 8.62% (30/348) have candi- A better understanding of the molecular mechanisms lead- date brain-related CNVs, for a combined total of 12.64% ing to psychiatric symptoms in this syndrome will help early (44/348) of SCZ patients with CNVs of interest. CNVs as- diagnostic and intervention. sociated with known syndromes include: del 1q21.1, del Presented on behalf of the Genomics Group, and the In- 16p11.2, del 22q11.2, del 2p16.3, del 5q35.3, del 15q11.1- ternational 22q11.2 Deletion Syndrome Brain Behavior Con- q12, del 16p13.3, del 16p13.11-p12.3, del 17p12, and dup sortium. Xp22.12. All of the detected CNVs within these syndrome regions are associated with one or more of the following neu- Disclosure: Nothing to disclose. rocognitive phenotypes: schizophrenia, autism, intellectual disability, developmental delay, brain anomalies, learning doi: 10.1016/j.euroneuro.2018.08.014 disabilities, behavioural abnormalities, and/or seizures Discussion: We observed a greater than expected number of 8 syndromic CNVs amongst the schizophrenia cohort (14/348,

PREVALENCE OF CLINICALLY WELL-ESTABLISHED 4.02%), particularly CNVs already hypothesized or known

CNV SYNDROMES WITHIN A TORONTO SCHIZOPHRENIA to associated with neurodevelopmental disorders (30/348, 8.62%). Screening for these rare genetic disorders could lead PATIENT POPULATION to better understanding of the pathophysiology of psychi- ∗ atric disorders, as well as the prevalence of these syndromic Venuja Sriretnakumar 1, , Brianna Barsanti-Innes 2, CNVs within various psychiatric population subtypes. Cor- Syed M. Wasim 3, Clement Zai 1, James L. Kennedy 1, rectly identifying syndromic CNVs within psychiatric popula- Joyce So 3 tions can improve patient prognosis to other non-psychiatric

systemic symptoms, and phenotypic characterization of CNV 1 Centre for Addiction and Mental Health carriers versus non-carriers will provide a baseline diagnosis 2 University of Toronto criteria for clinicians. Further analyses will be undertaken 3 University Health Network, Mount Sinai Hospital to deﬁne speciﬁc phenotypes associated with their respective CNVs to better characterize potential genetic effects

Background: Many rare genetic syndromes are known to on the phenotypic presentation of SCZ patients. phenotypically manifest with psychiatric symptoms that can be indistinguishable from primary psychiatric disorders. Disclosure: Nothing to disclose. While the majority of ongoing research in psychiatric genetics has been dedicated to the identiﬁcation and char- doi: 10.1016/j.euroneuro.2018.08.015 acterization of genes involved in primary psychiatric disorders, there has been little research to determine the extent to which rare genetic variants/syndromes contribute to the overall psychiatric disease load. In our study, we aim to investigate the prevalence and role of clinically well- characterized pathogenic copy number variant (CNV) syndromes within a schizophrenia patient population. ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 6 Abstracts

9 is obtained either directly for the clusters or as imputation IDENTICAL BY DESCENT SEGMENTS ASSOCIATES WITH reference. SCHIZOPHRENIA The validity of the method still needs further testing, and we approach this on various levels. ∗ Anders Rosengren 1, , Alicia Martin 2, Vivek Appadurai 3, We also detect IBD clusters highly associated with ADHD,

Thomas Als 4, iPSYCH-BROAD Working Group 1, ID, Bipolar, Anorexia, recurrent depression and Autism.

Alfonso Buil 5, Thomas Werge 6 Disclosure: Nothing to disclose. 1 Research Institute of Biological Psychiatry, PC Sct Hans, doi: 10.1016/j.euroneuro.2018.08.016 iPSYCH

2 Massachusetts General Hospital

3 Research Institute of Biological Psychiatry 10 4 iPSYCH, iSEQ, Aarhus University FIRST GENOME-WIDE ASSOCIATION STUDY OF 5 Institute of Biological Psychiatry SCHIZOPHRENIA IN AN INDIAN POPULATION REVEALS A 6 Institute of Biological Psychiatry, MHC Sct. Hans, Mental NOVEL SUSCEPTIBILITY LOCUS Health Services - Copenhagen

Sathish Periyasamy 1, Sujit John 2, Raman Padmavati 2, Background: Rare genetic variants are thought to con- Preeti Rajendren 2, Priyadarshini Thirunavukkarasu 2, tribute to the risk of mental disorders including Schizophre- Jacob Gratten 3, Elizabeth Holliday 4, Andrew Bakshi 5, nia. These are however difficult to detect. Traditional meth- Lynn Jorde 6, Matthew Brown 7, Naomi Wray 3, ods such as GWAS are now routinely run on even very large Rachel Suetani 8, Jean Giacomotto 8, Rangaswamy Thara 2, samples, but they may still lack the power to detect rare ∗ Bryan Mowry 8, variants, or may even lack the variants themselves as in these are only available as a Priori selected or subsequently 1 Queensland Brain Institute imputed into the data using some non exhaustive reference. 2 Schizophrenia Research Foundation In this study we use clusters of Identical By Descent 3 University of Queensland Segments. Leveraging only genotyped common SNPS to ap- 4 University of Newcastle proach rare variants. 5 Peter MacCallum Cancer Centre Methods: Our sample consist of the 70500 people of Euro- 6 University of Utah pean descent in the Danish iPSYCH Study. 7 Queensland University of Technology We use the refined IBD method of Beagle to detect all au- 8 Queensland Brain Institute, The University of Queensland tosomal pairwise IBD segment of minimum length 2cM. From these we form clusters consisting of 3 to hundreds of people all sharing one specific IBD segment using the Efficient Background: Polygenic variation predisposes to schizophre-

Multiple-IBD algorithm, EMI. We then apply the framework nia. Genome-wide association studies (GWAS) mostly in Eu- of fast-lmm and LEAP to do association on each individual ropean populations have identiﬁed over 100 schizophrenia- cluster. We associated clusters and Schizophrenia in a sub- associated loci. We report the ﬁrst schizophrenia GWAS in a set consisting of roughly 3000 cases and 20000 controls. We unique Indian population. account for their relatedness as a random effect and include Methods: Our sample of 3092 Tamil ancestry individuals age and gender as covariates. from Chennai, Tamil Nadu consists of schizophrenia fami-

Results: We identify 34 clusters that exceed our Bonferroni lies and unrelated cases and controls (1321 affected – 816 threshold of 1e-8 with an average length of 1.3cM and rang- case-control, 505 family; 1771 controls – 900 unrelated ing from 20 to 167 persons corresponding to MAFs of 0.1%- case-control, 871 -family). The GWAS was conducted using

0.01% Genome-wide Complex Trait Analysis. Polygenic risk scores,

The clusters cover 10 distinct genomic regions each cov- heritability analysis, gene-set analysis, runs of homozygos- ering from one to tens of genes. ity, copy number variant analysis, cross-trait genetic cor-

These include candidate genes such as NRXN3 and ITSN2, relation for SCZ between India and Europe, eQTL analyses, but also genes apparently not already implicated. SMR analysis, pathway analyses, and network connectivity-

The combined set of genes is enriched for Schizophrenia enrichment analysis were conducted to further investigate genes as listed by the GWAS Catalog. this dataset. We investigated the regulatory effects of our

Discussion: Here we do a direct test of individual IBD clus- top SNP, highlighting one gene for functional testing. ters and are able to detect clusters highly associated with Results: Our sample was characterized by a uni- > Schizophrenia. form ethnicity ( 97% Tamil), a degree of inbreeding = Trying to ﬁnemap any single causal variant within in the (F_HET mean 0.026 [-0.02 to 0.02]) and a homogeneous associated IBD clusters segment using the subset of samples schizophrenia phenotype with very low rates of alcohol and also Whole Exome Sequenced have so far been unsuccessful. illicit substance abuse. We observed a genome-wide signiﬁ-

Since no population rare variant of high cluster frequency cant association between schizophrenia and a chromosome = × has been shown to itself be associated with Schizophrenia in 8q locus (OR: 1.06; P 4.35 10-8), with support from the = × the cohort adjusted for multiple testing. Hopefully this will schizophrenia PGC2 (OR: 1.03; PGC P 7.56 10-4). eQTL possible as a more speciﬁc and detailed sample reference analysis revealed that this locus was regulating a single gene downstream of our top SNP. PGC2-generated genetic proﬁle scores predicted schizophrenia in our sample (maximum R2 ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 7

(observed scale) = 0.05; P = 1 × 10-37), as did the frequency The aim of this study was to quantitatively assess whether of runs of homozygosity (FROH) burden (P = 3.357 × 10-5). a number of experimentally-derived gene-sets based on For network connectivity-enrichment analysis of brain re- core genes would make a larger than expected contri- gions the putamen generated the highest enrichment score, bution to polygenic risk, as proposed by the omnigenic similar to what was observed in the PGC2 schizophrenia model. The main hypothesis of this study was that sets dataset. For the MHC locus we observed no evidence for based on schizophrenia core genes would be associated with replication (P < 0.05) in Tamil Nadu samples; for other PGC schizophrenia at a greater level than expected under an genome-wide significant loci, 31/97 PGC regions replicated equally distributed omnigenic effect. Random genic and (P-value < 0.05) in our TN dataset, with the chromosome non-genic sets, as well as two sets associated with cardiac 2q32.3 locus (rs59979824; ranked 66 in PGC) having the best disease and cancer, were selected as comparisons to the P-value (0.0007298) in our dataset. We investigated gene schizophrenia core gene sets. expression at our top locus and observed that only one gene Methods: Polygenic profile scoring was performed with SNPs showed a dose-response association with our index SNP. Pre- derived from six recently validated putative core gene- liminary zebrafish data further suggests that partial loss-of- sets centred on four schizophrenia genes identified either function of this gene leads to abnormal brain development. by rare variant or GWAS (TCF4, FMRP, miR-137, CHD8) in Discussion: This first schizophrenia GWAS in an Indian pop- the data available for secondary analysis from the PGC2 ulation –characterized by features which are advantageous schizophrenia case-control cohort (N = 29,125 cases and for genome-wide analyses - has identified a genome-wide 34,836 controls). A leave-one-out polygenic risk score anal- significant locus that has also attracted PGC2 support. Fur- ysis was implemented across the 39 different studies, and ther we provide evidence for genome-wide sharing of com- combination of p-values from independent ‘left out’ sets mon genetic variation for schizophrenia between Europe was performed using Stouffer’s z test. and India. Moreover, the frequency of runs of homozygos- Results: We observed a robust polygenic signal us- ity also predicted schizophrenia in our Indian population. ing SNPs from each of the TCF4 (p-value = 1.18 × 10- Larger Indian samples than those currently available will 46), FMRP (p-value = 1.66 × 10-33), miR-137 upregulated be required to replicate the associations identified here (p-value = 3.28 × 10-23) and CHD8 down-regulated (p- and to discover more population-specific variants associated value = 1.91 × 10-33) gene-sets. Additional analyses using with disease. randomly selected sets of genic and non-genic SNPs of the same size showed a consistent floor effect in amount of vari- Disclosure: Nothing to disclose. ance explained, more pronounced when using genic SNPs, which was also the case when non-schizophrenia specific doi: 10.1016/j.euroneuro.2018.08.017 sets were used. Discussion: Our findings indicated a significant association 11 between the gene-sets based on putative core genes and

GENE-SETS GOING FORWARD: THE ROLE OF GENE- schizophrenia. Additionally, we report the existence of a

SETS IN THE CONTEXT OF THE OMNIGENIC MODEL OF floor effect, linked with the underlying omnigenic model, which needs to be taken into consideration when interpret- SCHIZOPHRENIA ing the results of a polygenic score study on complex traits. ∗ Finally, the method developed allows a quantification of the Alexandros Rammos 1, , Lara A. Neira Gonzalez 2, contribution of specified core gene-sets as well as poten- Schizophrenia Working Group 3, Daniel R. Weinberger 4, tially identifying peripheral genes, and should be practically Kevin J. Mitchell 1, Kristin K. Nicodemus 2 applicable in the selection of sets of SNPs that yield the

greatest signal to noise ratio in the construction of a poly- 1 Trinity College Dublin genic score. 2 University of Edinburgh

3 Psychiatric Genomics Consortium Disclosure: Nothing to disclose. 4 Lieber Institute for Brain Development Reference

Background: The genetic architecture of schizophrenia in- Boyle, EA, et al., 2017. Cell 169 (7), 1177–1186. volves both rare and common variation in many genes. Re- cent theories developed by the Pritchard group (1) have sug- doi: 10.1016/j.euroneuro.2018.08.018 gested an even broader omnigenic background for the disorder. Within that context, genes can be split into core and peripheral genes. Core genes may be identiﬁed by rare variant or Genome-Wide Association studies and can be used as a basis to identify peripheral genes, which work in common networks with them. ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 8 Abstracts

12 from 57 to 34, the number of associations mapped to < = 5 COMPARATIVE GENETIC ARCHITECTURES OF variants increased from 8 to 15, the number of associations SCHIZOPHRENIA IN EAST ASIAN AND EUROPEAN POPU- mapped to a single variant with greater than 50% probability LATIONS increased from 16 to 20, and median size of the genomic regions the associations mapped decreased from 277 Kb to 111 ∗ Max Lam 1, , Chia-Yen Chen 2, Shengyin Qin 3, Pak Sham 4, Kb. Polygenic risk score modeling demonstrated that vari-

Nakao Iwata 5, Kyung Hong 6, Sibylle Schwab 7, Weihua Yue 8, ance explained across various p-value thresholds provides a

Ming Tsuang 9, Jianjun Liu 10, Xiancang Ma 11, René Kahn 12, proxy for the signal-to-noise ratio, which differs by training

Yongyong Shi 13, Hailiang Huang 2, PGC-EAS Schizophrenia population - relative to EUR training data, variants from the Working Group EAS training data with more permissive p-values improve EAS prediction accuracy, indicating that larger EAS studies

1 Institute of Mental Health will be needed to explain similar case/control variance as

2 Massachusetts General Hospital currently explained in EUR individuals. We dd not ﬁnd sig-

3 Collaborative Innovation Center, Jining Medical University niﬁcant EAS association in the MHC. The key SNP rs13194504 < 4 Brain and Cognitive Sciences, Centre for Genomic Sci- reported in EUR, has 1% MAF in EAS, and EUR-speciﬁc LD ences, University of Hong Kong patterns within MHC region might account for signals re-

5 Fujita Health University School of Medicine ported in previous reports.

6 Sungkyunkwan University School of Medicine, Samsung Discussion: The large-scale EAS sample allowed the empir-

Medical Center ical evaluation of congruence of the genetic architecture

7 Centre for Medical and Molecular Bioscience, Illawarra of schizophrenia between EAS and EUR. In spite of cross-

Health and Medical Research Institute, University of Wol- population genetic correlation indistinguishable from 1, we longong found that polygenic risk models trained in one population

8 National Clinical Research Center for Mental Disorders, have reduced performance in the other population due to

Ministry of Health, Peking University differential allele frequency distributions and LD-structure.

9 University of California, San Diego This highlights the importance of including all major ances-

10 Genome Institute of Singapore tral groups in genomic studies both as a strategy to improve

11 Clinical Research Center for Mental Disease of Shaanxi power to ﬁnd disease associations and to ensure the ﬁndings

Province, The First Afﬁliated Hospital of Xi’an Jiaotong have maximum relevance for all populations. University Disclosure: Nothing to disclose. 12 Icahn School of Medicine at Mount Sinai

13 The First Teaching Hospital of Xinjiang Medical University doi: 10.1016/j.euroneuro.2018.08.019

Background: Schizophrenia is a severe psychiatric disorder with a lifetime risk of approximately 1% worldwide. Oral Session: Neurogenetics & Ethics Most large-scale schizophrenia genetic studies have stud- 1:30 p.m. - 3:00 p.m. ied individuals of primarily European ancestry missing out on important biological insights. Here, we compile a sizable 13 schizophrenia sample of East Asian descent to further eluci- ENGAGING MENTAL HEALTH SERVICE USERS AND date the genetic architecture of the illness. THEIR FAMILIES IN THE GENETICS TESTING IN MENTAL Methods: 22,778 schizophrenia cases and 35,362 controls from samples from Singapore, Japan, Indonesia, Korea, HEALTH AND MENTAL HEALTH IN GENETIC TESTING

Hong Kong, Taiwan, and mainland China were compiled. ∗ European samples (56,418 cases and 78,818 controls) were David Crepaz-Keay then included for comparison. Fixed-effect inverse variance meta-analysis was used to combine GWAS summary statis- Mental Health Foundation tics first from the EAS samples and then with the EUR samples. Downstream analyses included genetic correlation, Background: Psychiatric genetic testing is often thought of partitioned heritability, gene-set, natural selection, trans- as an objective endeavour that increases our knowledge and ethnicity finemapping and polygenic risk score conducted to understanding. This presentation will challenge the notion draw further insights from the genome-wide data. of objective neutrality and explore the following themes: Results: 21 significant GWAS associations at 19 loci were • identified within the EAS samples. 15 of these associations Current understanding of genetic testing among mental have marked allele frequency deviations from EUR MAF. Ge- health service users and their families • netic correlations and gene set analysis were largely consis- Common questions raised about genetics and genetic tent with EUR based schizophrenia studies. Population dif- testing as it relates to mental health • ferentiation did not appear to drive the effects in these ob- Ethical and psychological impact of genetic testing and served loci. When meta-analyzed with EUR schizophrenia, ways of enabling people to make informed decisions pre 208 independent variants across 176 loci were identified, 53 and post genetic testing loci were novel. EUR-EAS trans-ethnicity finemapping improved identification of causal variants compared to EUR Methods: The presentation will explore the benefits and alone. The median size of the 95% credible set decreased risks associated with genetic testing from the perspectives of patients, families and other lay interests. ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 9

In order to explore these potential impacts of genetic was determined based on parental AD status weighted by testing, I will draw analogies with experiences of responses parental age. We conducted extensive functional annota- to existing diagnostic processes in psychiatry and how these tion of GWAS results, including positional, eQTL and chro- have changed alongside other social changes, using existing matin interaction mapping. We conducted pathway anal- literature and previous research. I will also draw on discus- yses, tissue enrichment and cell-type analyses as well as sions with service user academics and activists and family genetic correlation analyses. Mendelian randomization was supporters. performed to detect direct effects between AD and other Results: The positive opportunities offered for accurately traits. identifying and supporting people who are at elevated risk Results: Results from the AD-by-proxy status and clinical AD need to be balanced with the potential stigma, including showed strong genetic correlation (rg = 0.81). Genetic meta- self-stigma and discrimination, both direct and indirect ex- analysis identiﬁed 29 risk loci, of which 9 novel, and impli- perienced by people with a psychiatric diagnosis, their fam- cated 215 potential causative genes. Independent replica- ilies and broader communities. tion further supported these novel loci in AD. Associated Discussion: The presentation will consider the broader pub- genes are strongly expressed in immune-related tissues and lic health and public mental implications of genetic testing cell types (spleen, liver and microglia). Gene-set analyses and how information is related to testing is shared and used. indicate the genetic contribution of biological mechanisms This will include ethical issues around resource allocation involved in lipid-related processes and degradation of amy- and an exploration of the beneﬁts and risks of measures de- loid precursor proteins. We show strong genetic correlations signed to prevent conditions that may develop, where those with multiple health-related outcomes, and Mendelian ran- interventions have risks associated with them and hence the domisation results suggest a protective effect of cognitive importance of assessing people’s values and understanding ability on AD risk. how the might be applied to choices about genetic testing, Discussion: These results are a step forward in identifying data sharing and advanced decision-making. more of the genetic factors that contribute to AD risk and add novel insights into the neurobiology of AD to guide new Disclosure: Nothing to disclose. drug development. doi: 10.1016/j.euroneuro.2018.08.020 Disclosure: Nothing to disclose.

doi: 10.1016/j.euroneuro.2018.08.021 14 GENETIC META-ANALYSIS IDENTIFIES 9 NOVEL LOCI AND FUNCTIONAL PATHWAYS FOR ALZHEIMER’S DIS- 15 EASE RISK POSITIVE DOSE RESPONSE OF THE 1Q21.1 DISTAL CNV ON ICV THROUGH AN EFFECT ON CORTICAL ∗ Danielle Posthuma 1, , Iris Jansen 1, Jeanne Savage 1, SURFACE AREA

Kyoko Watanabe 1, Julien Bryois 2, Dylan M. Williams 2, ∗ Wiesje M. van der Flier 3, Richard Dobson 4, Lea K. Davis 5, Ida Sonderby 1, , Omar Gustafsson 2, Dennis van der Meer 3,

Hreinn Stefansson 6, Kari Stefansson 6, Nancy L. Pedersen 2, Nhat Trung Doan 4, Derrek Hibar 5, Lars Tjelta Westlye 4,

Stephan Ripke 7, Ole A. Andreassen 8, PGC-Alzheimer Srdjan Djurovic 6, Sébastien Jacquemont 7, Paul Thompson 8,

Workgroup Ole Andreassen 9, ENIGMA-CNV Working Group

1 VU Amsterdam 1 NORMENT, KG Jebsen Centre for Psychosis Research

2 Karolinska Institutet 2 deCODE genetics

3 VU University Medical Center 3 Norwegian Centre for Mental Disorders Research

4 King’s College London 4 NORMENT

5 Vanderbilt University Medical Center 5 Janssen Research & Development, LLC

6 deCODE Genetics/Amgen 6 Oslo University Hospital

7 Broad Institute of MIT and Harvard 7 University of Montreal

8 University of Oslo 8 Imaging Genetics Center, Keck School of Medicine, Univer- sity of Southern California

Background: Late onset Alzheimer’s disease (AD) is the 9 University of Oslo most common form of dementia with more than 35 million people affected worldwide, and no curative treatment Background: The 1q21.1 CNV predisposes to e.g. delayed available. AD is highly heritable and recent genome-wide development, schizophrenia, congenital heart defects and meta-analyses have identified over 20 genomic loci associ- obesity and there is an increased prevalence of macro- ated with AD. Here, we performed the largest genome-wide cephaly in duplications carriers and microcephaly in dele- association study of clinically diagnosed AD and AD-by-proxy tion carriers. Frequency of 1q21.1 in the UK biobank: 0.027 (71,880 AD cases, 383,378 controls) aiming to further char- % (deletion) and 0.044 % (duplication). acterize the genetic architecture of AD. Methods: Structural T1-MRI data from 16,078 subjects Methods: We meta-analysed genetically informative data from 34 research samples world-wide as well as data from AD cases (N = 24,087) and controls (N = 55,058) and from the UK biobank were analyzed (FreeSurfer) and 47,793 by-proxy cases/328,320 proxy controls. AD-by-proxy CNVs called (PennCNV). List of ENIGMA-CNV datasets: ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 10 Abstracts http://enigma.ini.usc.edu/ongoing/enigma-cnv/enigma- genetic variants associated with short-term verbal memory cnv- co- authors/ . Brain measures were normalized by an and learning in adults. inverse normalization of the residual of a linear regression Methods: We conducted meta-analyses for verbal memory correcting for age, age squared, gender and scanner. Dose immediate recall test scores and verbal learning scores in a response (deletion = 1, non-carrier = 2, duplication = 3) was discovery sample across 27 cohorts (N = 44,874) and 22 co- analyzed in a linear model on the normalized brain values. horts (N = 28,909), respectively. Respective sample sizes of Results below 0.005 (p = 0.05/10 regions) were considered the replication samples were N = 8763 and N = 3853. All par- significant. ticipants were adults of European descent and were free of Results: Subcortical volumes, cortical area and thick- dementia and stroke. ness and ICV of deletion (n = 25) and duplication carriers Results: For verbal memory immediate recall test scores, (n = 16) of the 1q21.1 CNV were compared to non-carriers we observed a genome-wide significant signal in an intron (n = 19,711). We found a positive dose-response effects of of CDH18 and for verbal learning, we observed a signal peak copy total surface area ( β= 1.72, P = 3.0 ∗10-9) and intracra- in the 3p21.1 region with a large LD block in/nearNT5DC2, nial volume ( β= 1.49, P = 1.1 ∗10-21). In addition, a small, STAB1, ITIH1, ITIH4, and PBRM1. For both phenotypes, we significant negative dose response on caudate was observed also observed a signal nearAPOC1. Associations of lead SNPs ( β= -0.45, P = 0.0045). in all regions replicated (p < .05). MAGMA gene-based anal- Discussion: These findings suggest that the mechanism be- yses implicated an additional nine genes for verbal learn- hind the head circumference change in 1q21.1 distal CNV ing (CALN1, TOMM40, SMIM4, NEK4, GNL3, AGXT2, MUSTN1, carriers is an increase in cortical surface area and may indi- GLT8D1, and ITIH3). Several SNPs in the 3p21.1 region are cate an effect on early development of the (dorsal) telen- eQTLs for POC1A, GNL3, GLYCTK, DUSP7, ITIH4, PPM1M, and cephalon. This may explain the various biological and neu- GLT8D1 in brain tissues in the GTeX or Braineac databases. S- rodevelopmental phenomena in 1q21.1 distal carriers and PrediXcan transcriptome-wide analyses showed that POC1A underlines the value of large-scale collaboration such as expression in the putamen was associated with lower verbal ENIGMA-CNV for studies of rare genetic variants implicated learning scores. Finally, we show with LDscore regression in brain pathology. a negative genetic correlation of schizophrenia and Type 2 Diabetes (T2D) to both memory traits; a negative genetic Disclosure: Nothing to disclose. correlation with anxiety and neuroticism for verbal immediate recall; and with Alzheimer’s disease (AD) for verbal doi: 10.1016/j.euroneuro.2018.08.022 learning. Discussion: We showed novel genomic regions in cadherin 18 (CDH18) and 3p21.1 that associated with verbal short- term memory and learning, respectively, and we replicated 16 a previously reported finding in APOC1. Cadherins are cru-

GENOME-WIDE META-ANALYSES REVEAL NOVEL LOCI cial in synaptic plasticity and the 3p21.1 region has been

FOR VERBAL SHORT-TERM MEMORY AND LEARNING AND implicated in psychiatric disorders in previous studies. Our SHOW GENETIC CORRELATION WITH SCHIZOPHRENIA, results may also partly explain phenotypic links between ALZHEIMER’S DISEASE, AND TYPE 2 DIABETES memory and psychiatric disorders, T2D, and AD.

∗ Jari Lahti 1, , Samuli Tuominen 1, Qiong Yang 2, Disclosure: Nothing to disclose.

Sarah Medland 3, Jodie Painter 3, Zachary Gerring 3, CHARGE doi: 10.1016/j.euroneuro.2018.08.023 Consortium Cognitive Working Group, Jan Bressler 4,

Thomas Mosley Jr. 5, Cornelia van Duijn 6, Lenore Launer 7,

Stephanie Debette 8, Ian Deary 9, Sudha Seshadri 2,

Katri Räikkönen 1 17 AN EPIGENOME-WIDE ASSOCIATION STUDY (EWAS)

1 University of Helsinki OF ALCOHOL CONSUMPTION REVEALS DNA METHYLA-

2 Boston University TION SIGNATURES OF ALCOHOL USE

3 QIMR Berghofer Medical Research Institute ∗ 4 University of Texas Toni Clarke 1, , Stewart W. Morris 2, Rosie Walker 1,

5 University of Mississippi Medical Center Heather Whalley 1, Kathryn Evans 1, Andrew McIntosh 1

6 Erasmus Medical Center University Medical Center

7 National Institutes of Health 1 University of Edinburgh

8 University of Bordeaux 2 Institute of Genetics and Molecular Medicine, University

9 University of Edinburgh of Edinburgh

Background: Understanding the biological basis of mem- Background: Genetic studies of alcohol consumption have ory functions may help combat neurodegenerative disor- revealed several loci that are robustly associated with al- ders. Despite moderate heritability in twin studies, con- cohol consumption. Despite this, the proportion of vari- sistent findings on the molecular genomic basis of memory ance in alcohol consumption that is explained by SNP ge- functions are scarce. This reflects low power in most of the netic effects is low. Recent epigenetic studies have shown studies. In a large genome-wide association study (GWAS) that around 12% of the variance in alcohol consumption can based on the CHARGE consortium we investigated common be explained by patterns of CpG methylation across the ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 11 genome; a remarkable advance on the 0.6% of variance ex- of age, gender and total brain volume. False discovery rate plained by SNP genetic effects using polygenic risk scores correction for multiple comparison was applied. Significant (McCartney et al, 2018). associations were subject to a mediation analysis to exam- Methods: We performed an EWAS study of alcohol consump- ine the potential effect of those over cognitive performance tion in 5101 individuals in Generation Scotland: the Scot- (i.e. scores in a Fluid Intelligence [FI] test). tish Family Health Study. Using the Infinum MethylationEPIC Results: In carriers of SZ-CNV, reductions in right thala- BeadChip 860,926 CpG probes were assessed for their asso- mus, pallidum, hippocampus and accumbens, and left puta- ciation with alcohol consumption in current drinkers. men were associated with schizophrenia CNVs (after cor- Results: Significant differential methylation was observed rection for multiple comparison). Four of the above FDR at several loci including the SLC7A11 gene. This encodes corrected brain areas showed a significant association with a sodium-independent cystine-glutamate antiporter protein FI and were included in the mediation analysis –i.e. thala- that is involved in regulating extra-synaptic receptors and mus, hippocampus and pallidum right and putamen left. The modulating glutamate signalling. Interestingly, SLC7A11 is results suggest a direct effect of SZ-CNVs on FI (B = -0.46, implicated in several neurodegenerative disorders and has p = 0.002) as well as an effect mediated by volume (B = -0.07, a mechanistic role in cocaine-withdrawal in animal models. p < 0.001) which accounted for almost 14% (p = 0.006) of the Discussion: Exposure to alcohol leaves an epigenetic signa- variance shared between SZ-CNV carrying status and FI. ture across the genome which can act as a biomarker of Discussion: SZ-CNVs are, in non-affected adults, associ- heavy consumption. Furthermore, studying these epigenetic ated with alterations in subcortical brain volumes in a man- marks may help to identify novel genes and pathways in- ner consistent with reduced hippocampus, thalamus, amyg- volved in the biological response to alcohol consumption. dala and accumbens being potentially pathogenic, although pleiotropy is also possible. The divergence of effects in Disclosure: Nothing to disclose. cases and in CNV carriers suggests enlarged pallidum is un- likely causally related to schizophrenia. Some, but not all, doi: 10.1016/j.euroneuro.2018.08.024 of the effects of CNVs on cognition in the general population are likely mediated through an influence on sub-cortical volumes. 18 THE EFFECT OF CARRYING SCHIZOPHRENIA PENE- Disclosure: Nothing to disclose. TRANT CNVS ON SUBCORTICAL BRAIN ANATOMY doi: 10.1016/j.euroneuro.2018.08.025 ∗ Xavier Caseras 1, , Anthony Warland 2, Kimberley Kendall 3,

Elliott Rees 4, George Kirov 5

1 Cardiff University School of Medicine Oral Session: Animal & Cell Based Models 2 MRC Centre for Neuropsychiatric Genetics and Genomics 1:30 p.m. - 3:00 p.m. 3 MRC Centre for Neuropsychiatric Genetics and Genomics, Cardiff University

4 MRC Centre for Neuropsychiatric Genetics and Genomics, 19

Institute of Psychological Medicine and Clinical Neuro- TRANSCRIPTOMIC ANALYSIS OF CORTICAL NEURONS sciences, Cardiff University DERIVED FROM PSYCHIATRIC PATIENTS CARRYING A

5 Cardiff University T(1;11) TRANSLOCATION, AND A CORRESPONDING MOUSE MODEL Background: Recent imaging research from large consortia 1, ∗ 1 1 has highlighted the existence of brain anatomical differ- Shane O’Sullivan , Marion Bonneau , Philippe Gautier , 2 1 1 ences between schizophrenia patients and healthy controls. Yasmin Singh , Helen Torrance , Daniel L. McCarthney , 2 3 1 Patients have reduced hippocampus, thalamus, amygdala Hansjuergen Volkmer , Maarten Loos , Kathyrn L. Evans , 1 1 4 and accumbens, and enlarged lateral ventricles and pal- Colin A. Semple , Siddharthan Chandran , Michel Didier , 1 1 1 lidum. If volumetric abnormalities represent a genetically David J. Price , David J Porteous , J. Kirsty Millar driven risk factor for the disorder, non-affected carriers at higher genetic risk for schizophrenia should have similar dif- 1 The University of Edinburgh ferences relative to non-affected carriers of lower genetic 2 The University of Tübingen risk. In the present study we investigated the association 3 Sylics Amsterdam between rare schizophrenia-associated CNVs and volume of 4 Sanofi subcortical structures in ∼10,000 non-affected adult participants from UK Biobank. We also examine whether any Background: A chromosomal translocation in a Scottish potential associations found had an impact in participant’s pedigree is linked to high risk for schizophrenia and af- cognitive performance. fective disorders. The translocation disrupts the DISC1 and Methods: There were 9292 participants with no known DISC2 genes on chromosome 1, and fuses DISC1 with the pathogenic CNV and 49 carriers of at least one schizophre- disrupted DISC1FP1 gene on chromosome 11. It is likely nia associated CNV. Linear regression models were used to that one or more of these gene disruptions contributes to investigate the association between the volume of subcor- the mechanism by which the translocation increases dis- tical structures and CNVs after accounting for the effects ease risk. It is also possible that positional effects of the ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 12 Abstracts translocation adversely influence expression of neighbour- Methods: To demonstrate the feasibility of personalized ing genes. medicine for a rare ASD/NDD, we have generated and evalu- Methods: We have carried out RNASeq and quantitative RT- ated a mouse model using CRISPR/Cas9 genome editing. This PCR analysis of induced pluripotent stem cell (IPSC)-derived model recapitulates a previously reported case of autosomal cortical neurons from translocation family members, and of dominant mental retardation 10 (MRD10, OMIM: 614256) and brain cortex from a corresponding mouse model that reca- is characterized by an exon 2 in-frame deletion of the AMPA- pitulates the effects of the translocation upon DISC1 expres- receptor binding domain of Cacng2 (Cacng2 e2), a trans- sion. The data were subsequently evaluated for differential membrane AMPA receptor regulatory protein that mediates gene expression and gene ontology predictions, and com- rapid glutamatergic excitatory signaling facilitating recep- pared to independent models of psychiatric disorders. tor trafficking, gating, and developmental synaptic plastic- Results: The set of dysregulated genes in the human cortical ity. In Cacng2 e2/ e2 mice, mRNA is expressed at normal neurons is enriched for matches to genes whose expression levels in the cerebellum; however, no protein is detectable is altered by independent DISC1 mutations in IPSC-derived in cerebellar whole-cell lysate or post-synaptic fractions, human neurons. A substantially overlapping set of genes is suggesting that the mutant protein may be targeted to the dysregulated in both human cortical neurons and mutant lysosome for degradation. mouse cortex. Moreover, gene ontology and pathway anal- Results: Behaviourally, homozygous Cacng2 e2/ e2 ysis predict largely the same major dysregulated functions C57BL/6J mice present with severe motor ataxia. Heterozy- in both datasets, including excitatory synapse dysfunction. gous Cacng2 + / e2 males and females show significant Multiple putative schizophrenia genes identified through re- cognitive and behavioral phenotypes such as hyperactivity, cent large-scale genome-wide association and copy number decreased anxiety-related behaviour, decreased startle variant studies are dysregulated in the human cortical neu- response, deficits in reversal learning, and heightened male rons and mutant mouse cortex. A number of dysregulated aggression. Maternal care deficits in female Cacng2 + / e2 genes have been confirmed at the qPCR level, including the carriers resulted in a dramatic loss of fecundity ( < 40% pup historic schizophrenia risk gene NRG1, and the putative risk survival by postnatal day 5). Low survival rates in offspring genes DRD2, ERBB4 and PDE4B. of heterozygous females is consistent with this mutation Discussion: Altogether these observations indicate that i) being under strong negative selection. many of the gene expression changes in the human cortical Discussion: With the goal of reversing cognitive impair- neurons are due to disruption of DISC1, ii) DISC1 disruption ments associated with MRD10, we have initiated treatment may act as a trigger for common shared disease pathways studies using the ampakine compound CX717 at doses of 0.3 in schizophrenia, and iii) defective excitatory synapse func- and 3 mg/kg. Given the function of CX717 as a positive al- tion and plasticity due to the translocation may be a disease losteric modulator of AMPA receptor activity, this represents mechanism. a valuable opportunity to evaluate a potential treatment for MRD10. Disclosure: Nothing to disclose. Disclosure: Nothing to disclose. doi: 10.1016/j.euroneuro.2018.08.026 doi: 10.1016/j.euroneuro.2018.08.027

20 EVALUATING GENETIC CAUSATION AND PERSONALIZED 21 PHARMACOLOGICAL TREATMENT OF AN ULTRA-RARE ABERRANT NRXN1 α SPLICING ARISING FROM 2P16.3 DISEASE ASSOCIATED WITH DELETION OF CACNG2 HETEROZYGOUS DELETIONS IMPACTS NEURONAL FUNCTION

Morgan Kleiber 1, Timothy Chapman 1, M.L. Maile 1, ∗ O. Hong 1, R. Purcell 2, A. Lippa 2, Amanda Roberts 3, Erin Absherty 1, , Shijia Zhu 1, Esther Cheng 1, ∗ Jonathan Sebat 1, Alesia Antoine 1, Madeline Halpern 1, Gintaras Deikus 1,

Peter Gochman 2, Judith Rapoport 2, Gabriel Hoffman 1,

1 University of California, San Diego Nadejda Tsankova 1, John Crary 1, Robert Sebra 1,

2 RespireRx Pharmaceuticals, Inc. Gang Fang 1, Kristen Brennand 1

3 Scripps Research Institute

1 Icahn School of Medicine at Mount Sinai

Background: The characterization of genomic sequence 2 National Institute of Mental Health, National Institutes of variants in autism spectrum disorder (ASD) and neurode- Health velopmental disorders (NDD) studies has led to the identification of causal genes that may provide novel opportuni- Background: Intragenic heterozygous deletions in NRXN1, a ties for individualized treatment strategies. Effective treat- highly alternatively spliced presynaptic cell-adhesion pro- ments for most ASDs and NDDs have been elusive due to a tein essential for synaptic function, are strongly associated lack of understanding of how a genetic variant within a pa- with schizophrenia (SZ) and autism spectrum disorder (ASD). tient plays a causal role, as well as the establishment of Homozygous NRXN1 KO mice display behavioral deficits and an intervention that can demonstrably target the underly- electrophysiological deficits consistent with these neurode- ing molecular pathology and improve cognitive function in velopmental disorders. Moreover, synaptic deficits observed vivo. in the NRXN1 KO mice were recapitulated in human induced ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 13 pluripotent stem cell (hiPSC)-derived neurons with condi- Background: Genome-wide association studies (GWAS) of tional heterozygous NRXN1 α deletions. Single molecule long schizophrenia revealed hundreds of genes implicated in this read sequencing of mouse prefrontal cortex has identified disorder, and one approach to elucidate their relevance to more than 200 NRXN1 α isoforms alone. It remains unclear to neurobiology is functional genetics. We have previously re- what extent rodent NRXN1 alternative splicing patterns are ported that the top schizophrenia risk variants on chromo- conserved in humans and the mechanism by which specific some 10q24 (rs11191419 and ch10_104957618_I) exert cis- transcripts contribute to neuropsychiatric disease remains regulatory effects on the expression of the positional candi- unknown. date genes BORCS7, AS3MT, and NT5C2, in the adult and de- Methods: Here, we apply new methods to integrate long veloping human brain. More specifically, we observed that read and short read sequencing data, in order to not NT5C2 was the only gene for which both risk alleles were only catalogue but also quantify, NRXN1 α isoform reper- associated with concordant effects on gene expression, in toires in human post-mortem and hiPSC-derived neurons. which case its expression was reduced. The NT5C2 gene Using a rare cohort of human induced pluripotent stem (5’-nucleotidase cytosolic II, cN-II) has been previously im- cell (hiPSC) neurons from four individuals with heterozy- plicated in susceptibility to spastic paraplegia, intellectual gous NRXN1 deletions, we investigate the impact of patient- disability and Parkinson’s disease. It encodes an enzyme specific NRXN1 deletions on neuronal function using the that cleaves phosphate from purine nucleotides, although multi-electrode array. its function during neurodevelopment and distribution in Results: We have cataloged and quantified hundreds of the adult brain remains unknown. unique transcripts arising from alternative splicing of the Methods: The distribution of the NT5C2 protein in the neuropsychiatric risk gene NRXN1 in post-mortem adult human DLPFC and in cortical human neural progenitor dlPFC(154), fetal PFC (65) and hiPSC forebrain neurons cells (hNPCs) was determined using immunostaining. To (102), demonstrating for the first time that hiPSC derived probe the effects of a loss-of-function, the expression of neurons can, to a large extent, recapitulate the in vivo NT5C2 was transiently reduced in hNPCs, and in the central splice repertoire. We quantify the differential transcript ex- nervous system (CNS) of Drosophila melanogaster, using pression and identify novel transcripts in two 2p13.1 psy- RNA interference (RNAi). Gene Ontology (GO) analysis was chosis patients with 3’ deletions in NRXN1. Functionally, applied to the transcriptomic signature elicited by the NRXN1 deletion hiPSC neurons show decreased neuronal ac- knockdown in hNPCs, and motility behaviour was assessed tivity, which can be partially ameliorated by overexpression in the model organism. of a single differentially expressed NRXN1 α transcript Results: Immunostaining of NT5C2 revealed a ubiquitous ex- Discussion: We present hiPSCs neurons as a platform for pression of this protein in hNPCs, neurons and a fraction of gaining insight into how the typical transcriptional reper- glial cells in the DLPFC. The knockdown of NT5C2 in hN- toire is impacted by patient specific deletions in NRXN1, a PCs was associated with an over-activation of adenosine CNV associated with many neuropsychiatric disorders. Iden- monophosphate-activated protein kinase (AMPK) signalling, tification of the repertoire and abundance of NRXN1 α tran- which is partly involved in protein translation; and downreg- scriptional variants is important for understanding protein ulated GO terms involved in cytoskeleton remodelling and diversity at human synapses. Our objective is to dissect protein translation. Reduced expression of CG32549 (NT5C2 what role specific transcriptional variants may play in neu- homologue) in Drosophila melanogaster elicited a reduction ronal function how this contributes to neuropsychiatric dis- in motility behaviour. ease risk. Discussion: Our work describes the hitherto unknown pattern of NT5C2 expression in human neuronal cells and post- Disclosure: Nothing to disclose. mortem brain tissue. We implicate loss of NT5C2 function in the regulation of the cytoskeleton, protein translation, and doi: 10.1016/j.euroneuro.2018.08.028 AMPK signalling in hNPCs, and demonstrate that expression of CG32549 in the CNS is critical for motility behaviour in flies.

Disclosure: Nothing to disclose. 22

THE PSYCHIATRIC RISK GENE NT5C2 REGULATES PRO- doi: 10.1016/j.euroneuro.2018.08.029 TEIN TRANSLATION IN HUMAN NEURAL PROGENITOR CELLS, AND IS INVOLVED IN LOCOMOTOR BEHAVIOUR IN DROSOPHILA MELANOGASTER

∗ Rodrigo R.R. Duarte 1, , Nathaniel D. Bachtel 2,

Ioannis Eleftherianos 2, Douglas F. Nixon 3, Robin M. Murray 1,

Nicholas J. Bray 4, Timothy R. Powell 1,

Deepak P. Srivastava 1

1 King’s College London

2 George Washington University

3 Cornell University

4 Cardiff University ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 14 Abstracts

23 24 MODELING A GENETIC RISK FOR SCHIZOPHRENIA: RNA-SEQ ANALYSIS IN HIPSC-DERIVED NEURONS FROM PHENOTYPIC DIFFERENCES IN HUMAN NEURAL PRE- PATIENTS WITH SCHIZOPHRENIA CURSORS AND CEREBRAL ORGANOIDS FROM PATIENTS

WITH CHR16P13.11 MICRODUPLICATIONS Laura Stertz 1, Guangsheng Pei 1, Peilin Jia 1, Shenglan Li 1,

Henriette Raventos 2, Ying Liu 1, Zhongming Zhao 1, ∗ Mandy Johnstone ∗, Navneet Vasistha, Miruna Barbu, Consuelo Walss-Bass 1, O. Dando, Karen Burr, E. Christopher, S. Glen,

Eve C. Johnstone, Heather Whalley, Andrew McIntosh, 1 University of Texas Health Science Center at Houston

Stephen Lawrie, Siddharthan Chandran 2 University of Costa Rica

The University of Edinburgh Background: Human induced pluripotent stem cells (hiPSC) have provided a new way of studying schizophrenia (SZ), Background: Schizophrenia (SCZ) and other major men- by allowing the establishment of brain cellular models ac- tal illnesses including classic neurodevelopmental disorders counting for the patient genetic background. Here we con- are highly heritable. Large-scale studies have shown that ducted an exploratory RNA-sequencing profiling study of cell genomic variation, in the form of copy number variants lines derived from hiPSCs generated from lymphoblastoid (CNVs), accounts for a significant portion of risk. CNVs in the cell lines (LCL) of subjects from the population isolate of disrupted in schizophrenia 1 (DISC1)-interactor and nuclear the Central Valley of Costa Rica (CVCR) distribution factor E-homolog 1 gene (NDE1) on chromosome Methods: Five cell lines (LCL, hiPSC, NPC, cortical neu- 16q13.11, that lead to SCZ and neurodevelopmental disor- rons and astrocytes) derived from 6 healthy controls and ders are proposed to result in abnormal neuronal precursor 7 SZ patients from the CVCR were generated using standard cell (NPC) proliferation and differentiation. methodology. RNA from these cells was sequenced using Illu- Methods: We have tested this hypothesis by generating a mina HiSeqTM2500. Normalization and differential expres- platform of human iPSCs from patients with schizophrenia, sion (DE) analysis were performed using DESeq2 (|FC| > and other neurodevelopmental disorders, who are known 1.5 e BH corrected p-value < 0.05). Functional enrichment to have CNVs affecting NDE1. We have differentiated these analysis was performed using DAVID 6.8. iPSCs into NPCs in vitro and have undertaken comparative Results: hiPSC-derived neurons were responsible for 94.4% studies between mutant and control cell lines. In parallel of the variance seen on DE analyses. We found 454 genes dif- we have also studied the effects of NDE1 on developmental ferently expressed on neurons differentiated from SZ com- pathways in ‘cerebral organoids’; a three-dimensional tis- pared to HC. Noteworthy, one of these genes was ZNF804A sue culture of human iPSC that mimics early stages of human (FDR = 0.032), a strong candidate gene for schizophrenia cortical development. Studying neurodevelopmental disor- susceptibility, with solid evidence of association from GWAS. ders in three-dimensional in vitro cultures can teach us fun- Discussion: ZNF804A is a zinc finger protein has been shown damental aspects of the development of the human cortex, to regulate neurite outgrowth, dendritic spine maintenance that are beyond reach in current animal model systems. and activity-dependent structural plasticity. It is expressed Results: Human brain imaging of affected carriers of the broadly throughout the brain, especially in the develop- 16p13.11 microduplication showed reduced brain volume. ing hippocampus and the cortex, as well as in the adult IPSC-derived brain organoids from these patients were cerebellum. A great advantage of using iPS-derived cells is smaller and showed reduced neuronal progenitor cell prolif- that the effect of outside environmental influences, such as eration. Transcriptomic and proteomic data shows deficits in use of medications, is removed and only the effects of ge- key intracellular signaling pathways associated with prolif- netic composition, which is unchanged by transformation, eration which we have been able to rescue both genetically are left. The DE of ZNF804A only on hiPSC-neurons highlight and pharmacologically. the crucial role that this protein can have on SZ pathology Discussion: This is the first study demonstrating a bio- during neuronal development. Functional studies will help logically relevant, potentially ameliorable, signaling path- us further characterize this mechanism. way underlying chromosome 16p13.11 microduplication syndrome in patient-derived neuronal precursor cells and pro- Disclosure: Nothing to disclose. vides mechanistic insight into the underlying basis of this doi: 10.1016/j.euroneuro.2018.08.031 mutation.

Disclosure: Nothing to disclose. doi: 10.1016/j.euroneuro.2018.08.030 ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 15

Saturday, October 13, 2018 Discussion: Our analyses, using some of the largest imputation reference panels and case/control datasets available, suggest a possible and previously underappreciated role of rare variation for GAD and BPD. Furthermore, ex- Oral Session: Statistics tended GREML-LDMS models support a substantial inﬂuence 1:15 p.m. - 2:45 p.m. of shared environments of cohabiting individuals or close relatives in GAD diagnosis. We are currently examining these 25 same models in MDD, BPD, and SCZ, and will estimate ge- EXPLORING THE GENETIC ARCHITECTURE OF PSYCHI- netic correlations among disorders within LDMS-partitioned ATRIC DISORDERS USING PARTITIONED HERITABILITY marker bins. APPROACHES Disclosure: Nothing to disclose. ∗ 1, 1 1 Luke Evans , Richard Border , Alta du Pont , doi: 10.1016/j.euroneuro.2018.08.032 Naomi Friedman 1, Emma Johnson 2, Jian Yang 3,

Peter Visscher 3, Matthew Keller 4

26 1 Institute for Behavioral Genetics, University of Colorado

2 Washington University in St. Louis POLYGENIC LIABILITY AND DEPRESSION IN THE

3 University of Queensland IPSYCH2012 COHORT

4 University of Colorado Boulder ∗ Katherine Musliner 1, , Preben Bo Mortensen 2,

3 4 John McGrath , David Hougaard , Background: Complex trait genetic architecture has impor- 4 5 Jonas Bybjerg-Grauholm , Marie Bækvad-Hansen , tant implications for the understanding of psychiatric dis- 6 2 7 Carsten B. Pedersen , Marianne Pedersen , Ole Mors , orders and for the utility of marker-based genetic risk pre- 8 2 9 Merete Nordentoft , Anders Børglum , Thomas Werge , diction. Narrow-sense heritability (h2) quantifies the role of 10 11 genetic variation in disorder liability, and also is the upper Nis Suppli , Esben Agerbo limit of theoretical risk prediction accuracy from genetic 1 markers. Partitioned heritability approaches extend these National Centre for Register-based Research, Aarhus Uni- summaries to explore the relative contribution of variants versity 2 across the allele frequency spectrum or relevant annota- Aarhus University 3 tions, and can be used explore genetic correlations among QLD Brain Institute 4 traits. These approaches use imputed variants across the Statens Serum Institut 5 full allele frequency spectrum to estimate SNP-based h2 Danish Center for Neonatal Screening, Statens Serum In- (h2SNP) within partitioned bins. stitute 6 Methods: We used partitioned heritability approaches with Centre for Integrated Register-based Research National imputed data to explore the genetic architecture of four Centre for Register-based Research, Aarhus University 7 psychiatric disorders using the UK Biobank (imputed to the Aarhus University Hospital, Risskov 8 Haplotype Reference Consortium sequence data) and Psy- Mental Health Centre Copenhagen, University of Copen- chiatric Genomics Consortium (PGC; imputed to the TOPMed hagen 9 whole genome sequence reference panel) datasets. We ap- Institute of Biological Psychiatry, MHC Sct. Hans, Mental plied genomic relatedness matrix (GRM) restricted maxi- Health Services - Copenhagen 10 mum likelihood (GREML) with linkage disequilibrium (LD)- Psychiatric Center Copenhagen 11 and minor allele frequency (MAF)-stratified (LDMS) GRMs CIRRAU - Centre for Integrated Register-based Research, in unrelated individuals to estimate partitioned heritabil- National Centre for Register-based Research, Aarhus Uni- ity of major depression (MDD), generalized anxiety disorder versity (GAD), bipolar disorder (BPD), and schizophrenia (SCZ), particularly to estimate the contribution of rare variation, and Background: Although the usefulness of polygenic risk to explore genetic architecture. We extended the LDMS ap- scores as a measure of genetic liability for major depres- proach to incorporate close relatives and cohabitants along sive disorder (MDD) has been established, their capacity to with MAF and LD partitioning. predict who will develop depression in the general popula- Results: Using GREML-LDMS in unrelated individuals, we tion remains relatively unexplored. Our goals were to evalu- estimated total ( + /-SE) GAD h2SNP = 0.16 + /-0.02, MDD ate whether polygenic risk scores for MDD, bipolar disorder h2SNP = 0.12 + /-0.03, and BPD h2SNP = 0.21 + /-0.01. No- (BPD) and schizophrenia (SCZ) can predict depression in the tably, approximately 10% and 13% of the GAD and BPD h2SNP, general population, and explore whether these polygenic li- respectively, was due to variants with MAF < 0.01, though abilities are associated with heterogeneity in age at onset none of MDD h2SNP was due to rare variants. When cohabi- and severity at initial depression diagnosis. tation and close relatives were included in the analysis, over Methods: Data were obtained from the iPSYCH2012 sample, 30% of GAD phenotypic variance was attributed to factors a case-cohort sample comprised of a subcohort of 30,000 in- shared between close relatives (possibly even rarer genetic dividuals randomly sampled from the Danish population and or non-genetic) and 7.5% to cohabitation of individuals. all additional individuals diagnosed in a psychiatric hospital with five different psychiatric disorders, including depression (ICD-10 codes F32, F33). For this study we included all ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 16 Abstracts genotyped individuals from the subcohort with Danish an- distress and high levels of subjective wellbeing despite life- cestry (N = 20,158) along with all additional genotyped de- time trauma exposure. Although genome-wide association pression cases with Danish ancestry (N = 18,030). Severity at studies (GWAS) have now identified loci associated with depression diagnosis was assessed using the ICD-10 severity a range of psychiatric conditions, resilience has been un- specifier (mild, moderate, severe without psychotic symp- derstudied. Here, we present the first GWAS of psycho- toms, severe with psychotic symptoms), and treatment set- logical resilience as a multidimensional continuous trait ting (outpatient, inpatient, emergency). Age at onset (AAO) conditioned on trauma exposure, in 89,914 individuals of was operationalized as each individual’s age in years at European ancestry from the UK Biobank (UKB Application first depression diagnosis. Polygenic risk scores were trained #25163). using the most recent results from the PGC as discovery Methods: Participants self-reported their lifetime trauma datasets. Hazard of depression was estimated using Cox re- exposure (i.e., childhood trauma, partner violence, and gressions modified to accommodate the case-cohort design. other traumatic events) as well as current psychologi- Case-only analyses were conducted using linear and multi- cal symptoms (i.e., depression, anxiety, and PTSD) and nomial regressions. All models were adjusted for sex, birth subjective wellbeing (i.e., happiness, and meaning). Us- year and the first 4 principal components. ing confirmatory factor analysis, we tested a theoretically Results: In this nationally-representative sample of the Dan- based latent factor model of psychological health con- ish population, a one standard deviation increase in poly- sisting of symptom and wellbeing sub-factors, with good genic liability for MDD was associated with a 32% increase resulting fit: X2(3) = 810.06, p < .00001, RMSEA = .07, and in hazard for depression (p < .0001). The corresponding in- CFI/TLI = .99/.97. Notably, latent factor models capture creases associated with PRS-BPD and PRS-SCZ were 10% and common variance across traits and reduce measurement er- 12%, respectively (p < .0001 for both). Hazard of depression ror by design. For individuals who endorsed at least one was 2.65 times higher among individuals in the top decile lifetime trauma exposure (83%, N reported above), latent of PRS-MDD relative to individuals in the bottom decile. factor scores were regressed on lifetime trauma variables; PRS-MDD was not associated with differences in either AAO resilience was thus operationalized as residuals from this or severity, however among depression cases, PRS-BPD and model, such that higher resilience scores reflected greater PRS-SCZ showed small associations with earlier age at diag- levels of psychological health than predicted by one’s life- nosis (p = .013 for both), and PRS-BPD was associated with time trauma exposure. Using imputed dosage data, we con- increased odds of inpatient (OR = 1.05, 95% CI = [1.01-1.09]) ducted a GWAS of resilience using SNPTEST, and examined and emergency (1.04, [1.00-1.08]) treatment. heritability, genetic correlations, and gene-based results. Discussion: Polygenic liability for MDD trained using preva- Results: Resilience was modestly heritable (h2 = 6%). No lent samples of MDD cases predicts depression in the general genome-wide significant loci were identified, although population, suggesting that these scores are tapping in to an three independent suggestive SNPs (p < 1x10-6) were de- underlying liability for developing depression, not just risk tected. Interestingly, gene-based analyses in MAGMA identi- for maintaining the disorder. The fact that PRS-BPD and PRS- fied two significant genes (MEIS2, MYO1H) previously linked SCZ also predict depression, although to a lesser extent than to behavior and/or psychopharmacological response. Con- PRS-MDD, is consistent with prior results suggesting a degree sistent with its derivation across multiple domains of psy- of common genetic overlap among these disorders. Variation chological health, our resilience phenotype showed high but in polygenic loading for BPD and SCZ may contribute slightly not complete genetic correlations with subjective wellbeing to heterogeneity in clinical presentation at first diagnosis (r = 0.80), depressive symptoms (r = -0.76), and neuroticism among depression cases. (r = -0.70). Despite evidence that heritability of trauma- related phenotypes can be higher in females, sex-stratified Disclosure: Nothing to disclose. GWAS of resilience in females (N = 48,753) revealed similar heritability (h2 = 7%). doi: 10.1016/j.euroneuro.2018.08.033 Discussion: Resilience—defined as broad psychological health despite lifetime trauma—is partially heritable, genetically correlated but somewhat distinct from existing

27 GWAS phenotypes, and has gene-based signals relevant for GENOME-WIDE ASSOCIATION STUDY OF RESILIENCE psychological processes. However, no genome-wide signiﬁ- TO LIFETIME TRAUMA IN THE UK BIOBANK cant loci have yet been identiﬁed. Efforts to meta-analyze across further cohorts will clarify whether unique genome- 1, ∗ 2 1 1 Karmel Choi , Murray Stein , Chia-Yen Chen , Tian Ge , wide signals emerge for this resilience phenotype.

Erin Dunn 1, Kristen Nishimi 3, Karestan Koenen 3,

Jordan Smoller 4 Disclosure: Nothing to disclose.

doi: 10.1016/j.euroneuro.2018.08.034 1 Massachusetts General Hospital

2 University of California, San Diego

3 Harvard T. H . Chan School of Public Health

4 Massachusetts General Hospital, Harvard Medical School

Background: Not all individuals exposed to trauma expe- rience poor mental health. People who demonstrate resilience are expected to show low levels of psychological ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 17

28 Discussion: Our ﬁndings suggest that at least some genetic GENETIC ASSOCIATION ANALYSES IN DIVERSE AN- factors may not be shared across populations for a given CESTRY POPULATIONS complex trait. Thus, caution may be needed before general- ising ﬁndings from European ancestry to non-European pop- ∗ Karoline Kuchenbaecker 1, , Theresa Reiker 2, Arthur Gilly 3, ulations. The rising numbers of biobanks and cohorts with

Deepti Gurdasani 3, Lorraine Southam 3, Gershim Asiki 4, health record linkage provide a unique and cost-effective

Janet Seeley 4, Anatoli Kamali 4, George Dedoussis 5, opportunity to build a resource of diverse ancestry sam-

Manjinder Sandhu 3, Eleftheria Zeggini 3, Erin Dunn 6, ples from existing studies. Novel mixed model approaches

Cathryn Lewis 7 provide a powerful analysis tool that addresses population structure, relatedness and imbalance in the number of cases

1 University College London and controls.

2 Swiss Tropical and Public Health Institute

3 Wellcome Trust Sanger Institute Disclosure: Nothing to disclose. 4 Medical Research Council/Uganda Virus Research Institute doi: 10.1016/j.euroneuro.2018.08.035 (MRC/UVRI)

5 Harokopio University of Athens

6 Harvard Medical School 29 7 King’s College London GENETIC ASSOCIATIONS BETWEEN PSYCHIATRIC DIS- ORDER RISK AND PHENOTYPIC FACTORS IN THE UK Background: Most studies investigating the genetic basis of BIOBANK psychiatric disorders use data from individuals of European ancestry, raising questions about the genetic factors that ∗ Caitlin Carey 1, , Rebecca Shafee 2, Raymond Walters 3, shape disease risk in individuals with non-European ances- Duncan Palmer 4, Liam Abbott 1, Daniel Howrigan 4, try and whether genetic risk factors are the same across Claire Churchhouse 1, Benjamin Neale 4, Elise Robinson 4 populations. Our goal was to address these knowledge gaps by performing a trans-ethnic analysis and constructing what 1 Broad Institute will be the largest mega-analysis of individuals with and 2 Harvard Medical School without major depressive disorder (MDD). 3 Stanley Center for Psychiatric Research and Program in Methods: Our ﬁrst aim was to determine whether com- Medical and Population Genetics, Broad Institute of Har- plex traits generally showed consistent genetic associa-

vard and MIT tions across populations. We used imputed genotypes from 4 Massachusetts General Hospital genome-wide arrays and measures of blood biomarkers from samples from the UK (N = 9,961), two isolated Greek populations (N = 1,641 and N = 1,945), and Ugandan samples Background: Common polygenic variation strongly con-

(N = 4,778). Using linear mixed models, we tested associa- tributes to risk for psychiatric disorders such as attention tions between established lipid-associated SNPs identiﬁed deﬁcit/hyperactivity disorder (ADHD), autism spectrum dis- through European samples and polygenic scores based on order (ASD), and schizophrenia (SCZ). Every individual has these variants. Secondly, we created a database of MDD in some degree of polygenic risk for each of these conditions. samples with diverse or admixed ethnicity and established Biobanks with large-scale phenotypic and genomic data, an analysis pipeline to assess variant associations. such as the UK Biobank (UKBB), can thus be leveraged to

Results: Few of the established lipid loci displayed nomi- better understand the full spectrum of outcomes associated nally significant associations with their target biomarker in with that risk. Such analyses can clarify how genetic risk for the Greek and Ugandan samples. The polygenic scores were psychiatric disorders varies in phenotypic presentation and associated with highly consistent effects across the Euro- highlight clinically useful associations. pean populations (p < 3x10-4 in each case). For the Ugan- Methods: Beginning with more than 2500 traits measured in dan cohort, there were significant associations of the HDL the UKBB, we identified 350 traits with significant common-

(p = 4x10-26) and LDL scores (p = 4x10-20) but no evidence variant heritability in the white British UKBB subsample = of association for the triglyceride score (p = 0.14). This sug- (N 333,554) using LD Score Regression (LDSR). To improve gests that some traits might share fewer genetic factors interpretability, we next conducted factor analysis on these across populations. 350 traits in the same subsample and identiﬁed a 40-

For MDD, we identiﬁed 28 studies, totalling more than factor solution yielding cohesive and interpretable factors.

20,000 affected and 300,000 unaffected individuals with The factors included a wide range of physical (e.g., pain, non-European ancestry to be analysed through the Psychi- asthma), behavioral (e.g., neuroticism, sleep), and cogni- atric Genomics Consortium MDD working group. Our pipeline tive outcomes (e.g., intelligence/education). We then per- is based on a single mixed model analysis rather than strat- formed a GWAS of each factor, controlling for age, sex and ifying analyses by ancestry. Mixed models can account for principal components. To evaluate the relationship of these population structure and the presence of relatives. How- factors with psychiatric disorders, we estimated the genetic ever, standard implementation for genetic studies are de- correlation of each with ADHD, ASD, and SCZ using LDSR signed for continuous outcomes. Therefore, we use logis- with the latest Psychiatric Genomic Consortium GWAS for tic mixed model implementations. Results can be combined each disorder. across studies through trans-ethnic meta-regression. Results: All three psychiatric disorders were moderately genetically associated with poorer overall mental (e.g., life ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 18 Abstracts dissatisfaction factor: rg = .30 to.39) and physical (e.g., gen- causal effect of CT on MDD), and (5) no interaction & CT- eral health factor: rg = -.18 to -.53) health, as well as with h2l = 50% with genetic correlation with MDD. On these sim- certain behavioral factors (e.g., sleep issues: rg = .22 to ulated data, we performed PRS analyses (logistic regression .56). In contrast, divergent profiles of genetic correlation of MDD on PRSxCT) and GRM-based analyses (cross-product across disorders were observed for factors such as alcohol Haseman-Elston regression to estimate (1) the genetic cor- use (ADHD only rg = .57), weight (ADHD rg = .37, vs. SCZ rg = - relation between MDD in CT = 1 and MDD in CT = 0, and (2) .12), and wearing glasses (ADHD rg = -.27, vs. ASD rg = .28; the difference in magnitude of the liability-scale heritably all ps < 1e-9). of MDD in CT = 1 compared to CT = 0). Discussion: It has been clearly established that the com- Results: The PRS- and GRM-based analyses provided non- mon polygenic influences on risk for neuropsychiatric dis- biased results for scenarios 1-4. For scenario 5, a slight bias orders partially overlap. While polygenic risk for each dis- was introduced when simulating a relatively strong genetic order is defined by association to its clinical features (e.g. correlation between MDD and CT of 0.53 (MDD-h2l of 0.31 in behavioral problems), we highlight how this polygenic risk CT = 1 and 0.33 in CT = 0). can also differ substantially in their biomedical and cogni- Discussion: Contrary to our hypothesis, these simulations tive correlates. These results suggest that the shared and suggest that polygenic GxE analyses generally provide re- unshared polygenic influences on risk for psychiatric disor- liable results. However, GxE results should be interpreted ders can be used to distinguish elements of the disorders’ with caution when a heritable environmental factor is stud- underlying biology. ied with a strong genetic correlation with the disorder. A limitation of our work is that we cannot rule out potential Disclosure: Nothing to disclose. bias from other genetic architectures or polygenic methods than those studied here. doi: 10.1016/j.euroneuro.2018.08.036

Disclosure: Nothing to disclose.

30 doi: 10.1016/j.euroneuro.2018.08.037 ENVIRONMENTAL FACTORS ARE OFTEN HERITABLE: DOES THIS BIAS POLYGENIC GENE-BY-ENVIRONMENT Oral Session: Autism Disorders INTERACTION ANALYSES? 1:15 p.m. - 2:45 p.m. ∗ Wouter J. Peyrot 1, , Matthew Keller 2, Wouter van

Rheenen 3, Naomi Wray 4, Brenda Penninx 1 31 DISCOVERY AND CHARACTERIZATION OF 102 GENES

1 VU University Medical Center ASSOCIATED WITH AUTISM FROM EXOME SEQUENCING

2 University of Colorado Boulder OF 37,269 INDIVIDUALS

3 UMC Utrecht Brain Center Rudolf Magnus ∗ 4 University Of Queensland Frederick Satterstrom 1, , Jack Kosmicki 2, Jiebiao Wang 3,

Ryan Collins 2, Silvia de Rubeis 4, Michael Breen 4, 2 5 6 Background: In the era of genome-wide genotype data, re- Sherif Gerges , Anders Børglum , Joseph Buxbaum , 7 8 3 search in gene-by-environment interaction (GxE) has shifted David Cutler , Bernie Devlin , Kathryn Roeder , 9 2 2 its focus from candidate genes to polygenic methods, such Stephan Sanders , Michael Talkowski , Mark Daly as analyses of polygenic risk scores (PRS) and the genetic relatedness matrix (GRM). This progress acknowledges the 1 Broad Institute highly polygenic nature of psychiatric disorders, but also in- 2 Center for Genomic Medicine, Massachusetts General Hos- troduces new challenges. In particular, most environmen- pital, Stanley Center for Psychiatric Research, Broad Insti- tal factors tested in GxE analyses have a genetic compo- tute of Harvard and MIT, Harvard Medical School nent, such as smoking, cannabis use and even exposure to 3 Carnegie Mellon University childhood trauma (CT). We hypothesize that commonly used 4 Seaver Autism Center for Research and Treatment, Icahn polygenic GxE analyses may be biased when analyzing heri- School of Medicine at Mount Sinai table environmental factors. 5 iPSYCH, The Lundbeck Foundation Initiative for Integra- Methods: Polygenic data were simulated using population tive Psychiatric Research, iSEQ, Centre for Integrative Se- parameters representative of Major Depressive Disorder quencing, Aarhus University

(MDD), i.e., heritability of 35% and lifetime risk 15%. In ad- 6 Seaver Autism Center for Research and Treatment, Fried- dition, we simulated CT for the same individuals, assuming man Brain Institute, & Mindich Child Health and Develop- a cumulative CT risk of 20%, and an odds ratio for MDD of ment Institute, Icahn School of Medicine at Mount Sinai

2.5 for exposed individuals (CT = 1) compared to unexposed 7 Emory University School of Medicine individuals (CT = 0). We simulated 5 scenarios: (1) no inter- 8 University of Pittsburgh School of Medicine action & CT not heritable (CT-h2l = 0), (2) different direc- 9 UCSF Weill Institute for Neurosciences tion of genetic effect on MDD in CT = 1 compare to CT = 0 & CT-h2l = 0, (3) larger genetic effects on MDD in CT = 1 com- Background: The genetic architecture of autism spectrum pared to CT = 0 & CT-h2l = 0, (4) no interaction & CT-h2l = 50% disorders (ASDs) involves the interplay of common and rare with no genetic correlation with MDD (other than via the variation, with hundreds of genes conferring risk to the dis- ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 19 order. Identifying the genes associated with ASD and under- 32 standing their effects in the context of ASD’s heterogeneous MOST EXOME SEQUENCING IDENTIFIED AUTISM SPEC- phenotypic presentation have both been longstanding goals. TRUM DISORDER GENES CONFER GREATER RISK TO Previous studies have focused almost exclusively on de novo INTELLECTUAL DISABILITY / DEVELOPMENTAL DELAY (newly arising) variation for gene discovery and have not in- THAN ASD corporated other sources of variation or utilized genic con- ∗ straint metrics. Jack Kosmicki 1, , Liu He 2, Jeffrey Barrett 2, Mark Daly 3 Methods: On behalf of the Autism Sequencing Consortium, we present the largest exome sequencing study in ASD to 1 Harvard University date, with 37,269 samples from 31 sampling sources across 2 Wellcome Trust Sanger Institute large sequencing studies supported by SFARI, NHGRI, and 3 Analytic and Translational Genetics Unit NIMH, including the iPSYCH study in Denmark. To identify ASD-associated genes, we use a Bayesian framework that in- Background: Studies of de novo variation, which is iden- corporates both de novo and case-control variation and also tified by exome sequencing parent-offspring families, have leverages gene (pLI) and regional (MPC) constraint metrics. successfully identified over 100 genes associated with neu- Results: We discover 26 genome-wide significant genes and rodevelopmental disorders, most notably intellectual dis- 102 genes (FDR < 0.1) associated with ASD. Thirteen of the ability / developmental delay (ID/DD) or autism spectrum 102 genes overlap 12 large recurrent copy number variants, disorders (ASDs). Mutations in the genes first discovered in indicating that these genes may contribute to the autism as- de novo studies of ASD were specifically enriched in those sociations in 1p36.3, 2p15, 2q37.3, 11q13, 15q11.2, 15q24, ASD individuals with comorbid ID/DD, and these genes were and 16p11.2. The 102 genes are also enriched in both gene commonly assumed to affect both traits. As such, some crit- expression regulators expressed primarily during the pre- icized these studies for failing to separate ID/DD genes from natal/fetal period and neuronal communication genes ex- those genes that confer risk to ASD without ID/DD. Through pressed during later postnatal development. a meta-analysis of de novo variants from 3981 probands Meta-analyzing our results with published de novo vari- ascertained for ASD (1179 with comorbid ID/DD) and 5264 ants from 5264 intellectual disability / developmental de- probands ascertained for ID/DD (503 with comorbid ASD) we lay (ID/DD) trios indicates that 47 of the 102 ASD-associated show that previously associated ASD genes are more often genes are more strongly associated with ID/DD than ASD, mutated in individuals with ID/DD who do not have ASD. as evidenced by a higher rate of de novo variants in ascer- Methods: We meta-analyzed of 16,454 published de novo tained ID/DD individuals than in ascertained ASD individu- variants from 9,245 families. We leveraged detailed pheno- als. Of the remaining 55 genes discovered, 52 are preferen- typing on 85.9% of the probands to increase power to both tial towards ASD, and 3 are equally associated with both. We identify associated genes and calculate the degree of risk demonstrate that comorbid ASD-ID/DD genes are markedly each gene confers for different disorders. different from ASD-preferential genes in terms of degree Results: To identify associated genes, we compared the of negative selection, phenotypic presentation, and expres- number of observed de novo missense and protein- sion from single-cell RNA sequencing in multiple cell types. truncating variants (PTVs) to the number expected from a Among ASD probands, individuals with variants in comorbid null mutational model across three sets: ASD + ID/DD, ASD ASD-ID/DD genes on average walk 2.6 months later (P = 6e-4) only, and ID/DD only. This yielded a total of 107 significant and have an IQ 11.7 points lower (P = 5e-04) than those with genes. The contribution of evidence towards each gene’s a variant in ASD-preferential genes. association is overwhelmingly provided by ID/DD compared Discussion: Our data constitute the most comprehensive to ASD (OR = 4.9; P < 9e-58), with only 10 of the 107 genes description of the genetic architecture of ASD to date. The possessing more de novo missense and PTVs in ASD cases 102 ASD-associated genes are a substantial increase from than expected by chance; the remaining 97 genes are asso- the 65 (at an equal FDR of 0.1) identified in the previous ciated with ID/DD. Of the 10 ASD-associated genes, only 3, effort from our consortium. With the division of the 102 ANK2, DSCAM, and CHD8, are significantly more strongly as- genes into those that are preferential for comorbid ID/DD sociated with ASD than ID/DD, with the de novo variants in and those that are not, this work lends evidence that differ- DSCAM and ANK2 observed solely in ASD probands. De novo ent biological processes underlie ASD-associated genes and missense and PTVs in the remaining 7 ASD-associated genes the phenotypic presentation of ASD probands. are observed 2.9 times more in 4789 ID/DD cases who do not have ASD (P < 7e-7). This same finding extends to prior Disclosure: Nothing to disclose. published ASD gene lists (OR = 1.8; P < 0.002), indicating that doi: 10.1016/j.euroneuro.2018.08.038 most ASD-associated genes from previous de novo studies are more strongly associated with ID/DD. Furthermore, ascertained ASD cases harboring an ID/DD de novo variant have an IQ 20 points lower, begin walking 4 months later, and are 5 times more likely to have seizures than the rest of the ASD population. Discussion: Studying de novo variation has been extremely successful for identifying disease-associated genes, with ASD being arguably the absgship trait for this study design. However, the presence of comorbid ID/DD in ASD probands ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 20 Abstracts with associated de novo variants begged the question as to Nucleotide Polymorphism (SNP)-heritability (SNP-h2) for so- how much risk each gene confers to the different disorders. cial traits was estimated with linkage disequilibrium score Leveraging detailed phenotyping in a meta-analysis of 3981 (LDSC) regression, based on genome-wide quasi Poisson re- ASD and 5264 ID/DD individuals, we identified 107 disease- gression analyses modelling right skewed data. Genome- associated genes, of which only 3 are more strongly asso- wide summary statistics for ASD and ADHD were obtained ciated with ASD than ID/DD. Furthermore, genes previously through the PGC/iPSYCH consortium. The association be- associated with ASD are more often observed in ID/DD cases tween polygenic risk for disorder and social traits was as- without comorbid ASD; ASD cases with de novo variants in sessed with polygenic scoring (PGS, p-threshold < 0.1), also ID/DD genes are phenotypically more similar to ID/DD cases using a quasi Poisson regression framework. Estimates were than the rest of the ASD population. Therefore, researchers combined with random-effects meta-regression. should be cautious with using published lists of ASD genes Results: We identified opposite developmental trends in for follow-up functional studies - especially if the goal is to LDSC-h2 estimates for mother-reported LPS and PP re- learn about the genetic etiology of core ASD features in the spectively (trait x age interaction, p = 0.0002). While LPS absence of cognitive impairment - namely social and behav- SNP-h2 decreased with progressing age, with the high- ioral impairments. est estimate at age 4 years (SNP-h2 = 0.18(SE = 0.08)), PP SNP-h2 increased, with the highest estimate at age of 17 Disclosure: Nothing to disclose. years (SNP-h2 = 0.43(SE = 0.12)). Teacher-rated LDSC-h2 was strongest at 11 years, both for LPS (0.21(SE = 0.09)) and PP doi: 10.1016/j.euroneuro.2018.08.039 (0.15(SE = 0.10)), although the data was too sparse to model developmental patterns. Using a meta-regression framework, we identified a positive association between ASD

33 polygenic risk and PP, irrespective of age and rater (beta-

EXPLORING THE DEVELOPMENTAL GENETIC ARCHI- PGS = 0.05 (SE = 0.01), p = 0.0002), but not for LPS. Similarly, TECTURE OF SOCIAL BEHAVIOUR: EVIDENCE FOR we observed evidence for a developmentally stable posi- GENETIC OVERLAP WITH ASD AND ADHD tive genetic overlap between ADHD polygenic risk and social behaviour. This was observed for mother-rated (beta- 1 2 3 Fenja Schlag , Jan Buitelaar , Jakob Grove , PGS = 0.05 (SE = 0.01), p < 10-4) and teacher-rated (beta- 1 1 3 Ellen Verhoef , Chin Yang Shapland , Ditte Demontis , PGS = 0.11(SE = 0.02) p < 10-4) PP, as well as teacher-rated 4 5 6 George Davey Smith , Dheeraj Rai , Simon E. Fisher , LPS (beta-PGS = 0.07 (SE = 0.02), p < 10-4), but not mother- 3 7, ∗ Anders Børglum , Beate St Pourcain rated LPS. Discussion: Social traits have complex genetic architectures

1 Max Planck Institute for Psycholinguistics that follow trait-speciﬁc developmental patterns, but show

2 Radboud University genetic overlap with both ASD and ADHD that is develop-

3 Aarhus University mentally stable. However, the strength of genetic links with

4 University of Bristol psychiatric disorder can differ by rater and/or type of social

5 School of Social and Community Medicine, University of behaviour. Bristol

6 Max Planck Institute for Psycholinguistics, Donders Insti- Disclosure: Nothing to disclose. tute for Brain, Cognition and Behaviour, Radboud Univer- doi: 10.1016/j.euroneuro.2018.08.040 sity

7 Max Planck Institute for Psycholinguistics; MRC Integrative Epidemiology Unit (MRC IEU), University of Bristol 34 NOVEL INSIGHT INTO THE AETIOLOGY OF AUTISM Background: The acquisition of social skills represents an important developmental milestone that most children mas- SPECTRUM DISORDER GAINED BY INTEGRATING FUNC- ter effortlessly, while developmental delays and deviations TIONAL DATA WITH GENOME-WIDE ASSOCIATION in social behaviour often have long-reaching effects on later SUMMARY STATISTICS mental health and wellbeing. The spectrum of social abili- ∗ ties is phenotypically complex. For example, prosocial be- Oliver Pain , Andrew Pocklington, Richard Anney haviours capture positive interactions, while socially dis- ruptive behaviours may lead to problematic peer relations. Cardiff University The developmental genetic architecture of social behaviour and its genetic overlap with psychiatric disorders is, how- Background: Autism spectrum disorders (ASD) are a group ever, largely unknown. Here, we investigate age-, trait- and of neurodevelopmental disorders affecting 1% of individu- reporter-specific changes in genetic factors underlying so- als with core features typically occurring in early childhood. cial skills in the general population and study shared ge- ASD are highly heritable, with common genetic variation ex- netic influences with Autism Spectrum Disorders (ASD) and plaining ∼18% of the variance between cases and controls. Attention-Deficit/Hyperactivity Disorder (ADHD). Genome-wide association studies (GWAS) have identified Methods: Tw o social traits capturing low prosociality (LPS) several loci associated with ASD. However, associated loci and peer problems (PP) were repeatedly assessed in par- are often difficult to interpret functionally. To gain insight ticipants of a UK birth cohort (ALSPAC; N ≤6174, 4 to 17 into the functional consequence of ASD-associated loci, and years), using either mother- and/or teacher-report. Single identify novel ASD-associated genes, we integrated gene ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 21 expression data with the summary statistics of the largest 35 ASD GWASs to date. The results were then used to highlight OPPOSITE GENETIC EFFECTS FOR POLYGENIC ASD gene-sets dysregulated in ASD, and to estimate the variance RISK SHARED WITH AND INDEPENDENT OF ADHD: in ASD attributable to differential gene expression. EVIDENCE FOR A CANCELLING-OUT HYPOTHESIS Methods: We performed a transcriptome-wide associa- STUDYING GENETIC OVERLAP WITH LANGUAGE AND tion study (TWAS) of ASD by integrating 14 gene ex- LITERACY pression datasets including adult brain, blood and adi- = ∗ pose tissues with the largest ASD GWAS (Ncases 18,381, Ellen Verhoef 1, , Jakob Grove 2, Chin Yang Shapland 1, = χ Ncont. 27,969) (Grove et al. 2017, bioR iv). Analyses Ditte Demontis 3, Stephen Burgess 4, iPSYCH-Broad-PGC ASD were performed using the software FUSION. 14,943 inde- Consortium 5, George Davey Smith 5, Dheeraj Rai 6, pendent genes were tested for differential expression in Simon E. Fisher 7, Anders D. Børglum 3, Beate St Pourcain 8 ASD. Gene-set enrichment analysis of the TWAS results was performed using 135 brain relevant gene-sets, as well as 1 Max Planck Institute for Psycholinguistics 1,329 canonical pathway gene-sets. Stratified linkage dis- 2 The Lundbeck Foundation Initiative for Integrative Psychi- equilibrium score regression (S-LDSC) was used to estimate atric Research, iPSYCH, Centre for Integrative Sequencing, the proportion of ASD SNP-heritability that can attributed iSEQ, Aarhus University, Aarhus, Denmark; Department of to differential gene expression. Biomedicine - Human Genetics, Aarhus University Results: TWAS identified 12 genes associated with ASD at 3 The Lundbeck Foundation Initiative for Integrative Psychi- transcriptome-wide significance, of which 11 did not over- atric Research, iPSYCH, Centre for Integrative Sequencing, lap with genome-wide significant loci identified by the iSEQ, Aarhus University GWAS. Several of the associated genes have been impli- 4 University of Cambridge cated in ASD, including PTGDR (prostaglandin D2 recep- 5 University of Bristol, School of Social and Community tor). However, a 1Mb region on chromosome 17 contained Medicine 7 transcriptome-wide significant genes, highlights the need 6 Bristol Medical School, University of Bristol for further analysis to infer causal associations. Gene-set 7 Max Planck Institute for Psycholinguistics, Nijmegen, Don- enrichment analysis identified several canonical pathways ders Institute for Brain, Cognition and Behaviour, Radboud as enriched, the two most significant being the amyloid and University, Nijmegen telomere end packaging pathways. The proportion of ASD- 8 Max Planck Institute for Psycholinguistics, University of SNP-heritability explained by gene-expression was strongly Bristol, Donders Institute for Brain, Cognition and Be- dictated by the size of the sample used to derive the gene- haviour, Radboud University, Nijmegen expression SNP-weights, with the CommonMind Consortium prefrontal cortex SNP-weights explaining the most variance Background: Autism Spectrum Disorders (ASD) share ge- in ASD (12% of SNP-heritability). netic liability with complex cognitive skills, such as verbal Discussion: This study has identified several differentially intelligence quotient (IQ) scores, and thus genetic overlap expressed genes in ASD, supporting previous findings and with highly related literacy- and language-related abilities highlighting novel genetic associations. PTDGR has been (LRAs) is expected. However, recent evidence suggests that previously reported as differentially expressed in ASD by such genetic overlap can differ by ASD subgroup. Especially two studies using observed gene-expression data, support- ASD with comorbid Attention-Deficit/Hyperactivity Disor- ing the validity of the TWAS approach in our study. Gene- der (ADHD) symptoms may genetically correlate more with set enrichment highlighted the importance of amyloid and impaired than enhanced cognitive performance. Moreover, telomere-related pathways congruent with previous find- multi-trait scores such as IQ may conceal association pat- ings of increased beta-amyloid levels in individuals with terns for individual subskills. Here, we study genetic over- ASD and shorter telomere length in families with high risk lap between ASD and a wide range of LRAs. Specifically, we of ASD. Partitioned heritability estimates further demon- disentangle polygenic links into ASD genetic effects shared strated the utility of the TWAS approach for understanding with and independent of ADHD using methodologies robust ASD, and that this approach will become more powerful as to collider bias. SNP-weights derived from larger samples become available. Methods: Thirteen LRAs capturing oral comprehension, reading, spelling, phonological awareness/memory and ver- Disclosure: Nothing to disclose. bal IQ were studied in 7 to 13-year-old children from a doi: 10.1016/j.euroneuro.2018.08.041 UK birth cohort (ALSPAC; N ≤5,919; SNP-h2 > 0.25). Genome- wide summary statistics for 1) ASD excluding individuals comorbid for ADHD (cases:10,321; controls: 22,664), and 2) ADHD (cases:14,584; controls:22,492) were obtained through the iPSYCH consortium. To disentangle polygenic relationships between ASD and LRAs, ASD genetic effects on LRAs were modelled as effects shared with and independent of ADHD using multivariable regressions (MVR), analogous to Mendelian Randomization (MR). ASD instruments were selected by applying conservative (p < 5x10 ^ -8), typical MR (p < 0.0015) and common polygenic-scoring (PGS, p < 0.05) thresholds. ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 22 Abstracts

Results: Studying PGS-comparable instruments and with- however, and which cell types express one or more ASD risk out implying causality, we identified positive genetic as- genes, at what stage of neurodevelopment, and in which sociations between ASD risk and all LRAs, conditional on cellular compartment are unanswered questions. ADHD, passing an experiment-wide multiple testing thresh- Methods: To establish the neurodevelopmental trajectory old (p < 0.007). In contrast, polygenic ASD risk shared with of ASD risk genes during brain development, we have per- ADHD showed an inverse association, based on the same formed single cell transcriptome analysis (scRNA-seq) of set of ASD instruments. This pattern was observed for stan- embryonic cortices harvested at stages spanning corticoge- dardised reading and spelling instruments, verbal IQ and nesis. We use the developing mouse brain as a model due oral comprehension, but not phonological memory, phone- to a number of advantages: we can investigate in paral- mic awareness and study-specific LRAs. The strongest ef- lel a multitude of cell types analogous to those found in fects were present for reading speed at age 9 with a 0.02 the human developing brain, ASD risk genes are largely con- Z-score increase (SE = 0.003, P = 6x10 ^ -10) and a 0.038 de- served between mouse and human species, and robust ge- crease (SE = 0.007, P = 2x10 ^ -8) in performance per log odds netic models exist for follow-up studies of cell-types with in ASD and ADHD liability respectively (latter capturing specific ASD risk gene expression. Dissociated, viable sam- shared ASD/ADHD effects). Using typical MR instruments, ples were individually loaded onto the 10X Chromium Con- this genetic overlap was attenuated, but still present for troller for single-cell encapsulation and barcoding. Gene several reading and spelling abilities, and completely abol- expression values were imputed using the program MAGIC ished when analysed with conservative ASD instruments. (Markov Affinity-based Graphical Imputation of Cells) to re- Discussion: Subthreshold genetic variants capturing ASD li- cover dropout values, and we used PHATE (Potential Heat- ability may show opposite polygenic effects on reading, diffusion Affinity-based Trajectory Embedding) to visualize spelling and reasoning for ASD with and without ADHD the latent trajectory structure. A principle curve was fit symptoms, although implicating the same genomic regions. to the main trajectory of variation, followed by correlation Polygenic ASD risk was associated with increased reading, analysis of ASD risk gene expression. spelling and reasoning performance conditional on ADHD. Results: The first two components of variation capture the However, the same genomic regions were also associated trajectory of corticogenesis, with progenitor cells and ma- with lower reading, spelling and reasoning performance via ture neurons separating along the principle curve. This re- ADHD, as shared ASD/ADHD effects. This has implications lationship is present within each time point and persists for the presence of null effects in ASD collections with large when all times points are merged. ASD risk genes parti- proportions of patients comorbid for ADHD and detectable tion into subgroups based on correlation of expression with genetic correlations between clinical ASD and ADHD. marker genes specific for cortical regions and cell types. Immunohistochemistry of selected ASD risk genes in sec- Disclosure: Nothing to disclose. tioned mouse cortices confirms the developmental pattern and compartmentalization of expression found in our single doi: 10.1016/j.euroneuro.2018.08.042 cell analysis. Discussion: These investigations will yield insight into the role of cell types in the development of ASD. Cell types

36 showing co-expression of multiple ASD risk genes will be

SINGLE CELL RNA SEQUENCING OF EMBRYONIC MOUSE further assessed using mouse lines genetically engineered CORTEX REVEALS OVERLAPPING AND DISTINCT PAT- for conditional labelling to identify cell type specific ASD TERNS OF ASD RISK GENE EXPRESSION risk gene target networks. ASD risk gene mutant mouse lines will be used to assess the effect of loss of function alleles on ∗ Rebecca Muhle , Kristina Yim, Martina Krenzer, cell types with high ASD risk gene expression. By deciphering Guillermina Hill-Teran, Kathleen Malison, James Noonan how individual ASD risk genes impact brain development, we can better understand how individual risk genes may con- Yale University tribute to risk for ASD in order to spark new investigations for novel therapeutics. Background: Autism Spectrum Disorder (ASD) is a lifelong and often debilitating neurodevelopmental condition. Ge- Disclosure: Nothing to disclose. netic risk factors for ASD exhibit locus heterogeneity, and doi: 10.1016/j.euroneuro.2018.08.043 researchers have identified dozens of genes associated with ASD risk at genome-wide significance. Deciphering how these genes might contribute to ASD risk is still a major task. While many ASD risk genes encode proteins with known neuronal functions, other ASD risk genes encode proteins that regulate the expression of other genes. Regulatory ASD risk genes may act as ASD risk gene coordinators, regulating the expression of other ASD risk genes during neurodevelopment. Consistent with this, we previously showed in a model of early brain development that decreased levels of the ASD risk-associated regulatory gene CHD8 dysregulates the expression of other ASD risk genes. The developing brain has a complex and tightly orchestrated composition of cell types, ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 23

Oral Session: Affective Disorders Discussion: This study provides evidence that the differen- 1:15 p.m. - 2:45 p.m. tial polygenetic liabilities between BD, SCZ and MDD partly account for different clinical presentations of BD. 37 Disclosure: Nothing to disclose. ASSOCIATION OF THE POLYGENIC RISK SCORE FOR MAJOR PSYCHIATRIC DISORDERS WITH CLINICAL FEA- doi: 10.1016/j.euroneuro.2018.08.044 TURES IN BIPOLAR DISORDER AND CONTROLS

∗ Jie Song 1, , Lu Yi 1, Sarah Bergen 1, Erik Smedler 2,

Mikael Landen 2 38 WHOLE EXOME SEQUENCING OF MULTIPLEX BIPOLAR

1 Karolinska Institutet DISORDER FAMILIES AND FOLLOW-UP RESEQUENCING

2 University of Gothenburg IMPLICATE RARE VARIANTS IN CELL ADHESION GENES CONTRIBUTING TO DISEASE ETIOLOGY Background: Despite the overlap in genetic etiology and ∗ clinical presentations, bipolar disorder (BD), schizophrenia Anna Maaser 1, , Fabian Streit 2, Kerstin U. Ludwig 3,

(SCZ) and major depressive disorder (MDD) display distin- Anna C. Koller 3, Franziska Degenhardt 3, Holger Thiele 4, guishing characteristics. Whether genetic factors contribute José Guzman Parra 5, Fabio Rivas 5, Fermín Mayoral 5, to these clinical features remains to be investigated. Here, Stefan Herms 6, Per Hoffmann 7, Sven Cichon 7, we test whether polygenic risk scores (PRSs) of BD, SCZ and Marcella Rietschel 8, Markus M. Nöthen 9,

MDD associate with clinical presentations of BD and a con- Andreas J. Forstner 6 trol sample.

Methods: This study included participants from Swedish 1 Institute of Human Genetics, University of Bonn, School of Bipolar Collection Study (SWEBIC) that have been genotyped Medicine & University Hospital Bonn and interviewed regarding their diagnoses, clinical charac- 2 Central Institute of Mental Health, University Medical teristics, medical intervention, and outcomes. Clinical fea- Center Mannheim/University of Heidelberg tures of BD, including full inter-episodic remission, global 3 Institute of Human Genetics, University of Bonn assessment of functioning (GAF), severity in symptoms dur- 4 Cologne Center for Genomics, University of Cologne ing follow-up, psychotic features during mood episodes, al- 5 University General Hospital of Málaga, Biomedical Re- cohol and drug abuse, and hospitalization due to suicide at- search Institute of Málaga IBIMA tempt/deliberate self-harm were collected from telephone 6 Institute of Human Genetics, University of Bonn, School of interviews, the bipolar quality assurance registry and the Medicine, University Hospital Bonn, Human Genomics Re- National Patient Registry. The final sample consists of 5099 search Group, University of Basel patients with BD and 3331 controls. Standardized PRSs of 7 Institute of Human Genetics, University of Bonn, School of BD, SCZ and MDD were computed using summary statistics Medicine & University Hospital Bonn, University of Bonn, from the genome-wide association studies of SCZ, BD and University of Basel, Institute of Neuroscience and Medicine MDD working group from the Psychiatric Genomics Consor- (INM-1), Research Center Jülich, Germany tium. Linear, logistic and ordinal logistic regression models 8 Central Institute of Mental Health, University Medical were used to estimate associations between PRS and clini- Center Mannheim/University of Heidelberg cal outcomes among the participants, with adjustment for 9 Institute of Human Genetics, University of Bonn sex, age, the first six population principal components and genotyping platforms. With respect to the outcomes sui- Background: Bipolar disorder (BD) is a complex psychiatric cide, drug and alcohol abuse, the effects of PRS were evalu- disorder affecting around 1% of the general population. The ated both in the combined cohort and in cases and controls disease is characterized by recurrent episodes of mania and separately. depression and has a high heritability of around 70%. Since Results: PRS of BD and PRS of SCZ/MDD showed associations the cumulative impact of common variants with small effect with clinical symptoms of BD in opposite directions. Higher may only explain 25-38% of the phenotypic variance for BD, BD-PRS was associated with full inter-episodic remission, rare variants of high penetrance were suggested to confer higher GAF and less severe symptoms, while higher PRSs of risk for BD. SCZ and MDD showed associations with lower rate of inter- Methods: In the present study, whole exome sequencing episode remission, lower GAF, and more severe symptoms. (WES) was performed on 226 individuals of 68 large multi- PRS of SCZ was significantly associated with psychotic symp- ply affected families of European origin. We selected two to toms. PRSs of BD, SCZ and MDD were positively associated five affected individuals with BD or recurrent major depres- with hospitalization due to suicide attempt/self-harm, but sion from each family. WES was executed on the Illumina the associations did not hold after adjustment of BD status. HiSeq2500 platform. For data analysis, we used the Var- Finally, PRS of MDD associated with diagnoses of alcohol and bank pipeline (varbank.ccg.uni-koeln.de). We filtered for drug abuse regardless of adjustment for BD status. rare (minor allele frequency < 0.1%), nonsynonymous, potentially functional and segregating variants. Subsequently, we applied the Residual Variation Intolerance Score and performed an enrichment analysis with the genes that were ranked among the 20% most intolerant genes in the genome. ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 24 Abstracts

For follow up analyses, we prioritized candidate genes prefrontal cortex from the CommonMind Consortium, to test that were either implicated in at least two unrelated fami- whether predicted gene expression in the brain is associated lies or previously reported in NGS studies or GWAS of BD. In with bipolar disorder. Work is currently underway to incor- addition, we enclosed genes that were predominantly driv- porate additional BD cohorts from population biobanks to ing the significant pathways in our gene enrichment anal- this GWAS and perform genomic imputation of all cohorts to ysis. Using the single molecule molecular inversion probes the Haplotype Reference Consortium panel. (smMIPs) technology targeted resequencing of the 42 most Results: Preliminary results from this GWAS indicate 48 in- promising candidate genes is currently performed on the dependent genomic loci associated with BD (P < 5e-8). Of HiSeq2500 in an independent case/control cohort. these loci, 24 were significant in the most recent BD GWAS Results: In total, WES identified 1214 rare, segregating and conducted by the PGC, replicating regions including the potentially functional variants implicating 1122 different CACNA1C, ANK3, TRANK1 and SCN2A genes. Tw o additional genes. A significant enrichment for a total of 18 pathways loci were previously reported in other GWAS on BD, includ- was observed (e.g. neuron projection, post-synapse and cell ing the MAD1L1 locus. The 22 novel regions from this study adhesion; p < 0.001). The 42 prioritized candidate genes in- encompass genes encoding ion channels or neurotransmitter clude several cadherin genes (e.g. CDH7 and CDH23). Inter- receptors such as KCNN3, CACNB2 and GRINA, and for the estingly, CDH23 encodes a calcium-dependent cell adhesion first time identifies the major histocompatibility complex protein that may play a role in neurogenesis and that was (MHC) as genome-wide significant for bipolar disorder. TWAS previously implicated in risk for BD and other major psychi- found 7 genes in the novel loci where BD-associated SNPs atric disorders. significantly influence gene expression in the prefrontal cor- Discussion: Our preliminary results strongly suggest that tex, four of which are in the MHC region. Genetic correla- rare and highly-penetrant variants in neuronal and cell ad- tions were observed between BD and adult-onset psychiatric hesion genes contribute to BD etiology. The results of our disorders, as well as behavioural traits such as risk taking follow-up resequencing of a large case/control sample will (rG = 0.31, P = 2e-21) and smoking (rG = 0.20, P = 6e-13). provide further statistical evidence for an involvement of Discussion: This is the largest GWAS on bipolar disorder to particular genes and pathways. date and has thus far found 48 loci reaching genome-wide significance. Future work will examine the custom content Disclosure: Nothing to disclose. of the Psychchip array, including low frequency variants, for association with BD. Bioinformatics approaches which lever- doi: 10.1016/j.euroneuro.2018.08.045 age multi-omics data will be used to dissect the biological mechanisms underlying associated loci and prioritise causal genes and variants. 39 GENOME-WIDE ASSOCIATION STUDY IDENTIFIES NOVEL Disclosure: Nothing to disclose. LOCI ASSOCIATED WITH BIPOLAR DISORDER doi: 10.1016/j.euroneuro.2018.08.046 ∗ Niamh Mullins 1, , Gio Pan 2, Manuel Mattheisen 3,

Vassily Trubetskoy 2, Stephan Ripke 2, Laura Sloofman 1,

Alexander Charney 1, Bipolar Disorder Psychchip

Investigators, Bipolar Disorder Working Group 4, Eli Stahl 1 40 COMMON SNPS AND RARE CNVS CONFER RISK FOR

1 Icahn School of Medicine at Mount Sinai DIFFERENT CLASSES OF PSYCHIATRIC SYMPTOMS

2 Charite ∗ 3 Aarhus University Alexander Charney , Pamela Sklar

4 Psychiatric Genomics Consortium Icahn School of Medicine at Mount Sinai Background: Bipolar disorder (BD) is a common and severe mood disorder with a high heritability. Thirty genomic Background: Genetic studies of bipolar disorder (BD) over loci have previously been associated with BD in a large in- the past decade have shown that risk is conferred through ternationally collaborative genome-wide association study many common single nucleotide polymorphisms (SNPs) of (GWAS) conducted by the Psychiatric Genomics Consortium relatively weak effects, while evidence for a role of rare (PGC). Here, we extend this work, by adding over 29,000 ad- copy number variants (CNVs) has been found in some stud- ditional samples genotyped on the Psychchip array to iden- ies but not others. The putative lack of CNV burden stands tify novel associations with BD. as a point of contrast between BD and schizophrenia (SCZ), a Methods: We have performed GWAS meta-analysis of 48 co- clinically similar disorder with which BD has a very high SNP- horts to date, totaling 35,398 BD cases and 60,842 con- based genetic correlation. Here, we perform a genome-wide trols of European descent. Samples include 11,297 cases and CNV study in a cohort of 6,353 BD cases and 8,656 controls, 17,805 controls genotyped on the custom Psychchip array a case sample size approximately 2.5-fold larger than ear- and 1,758 cases, 5,937 controls from a recent GWAS in the lier work. In addition to our primary analyses of CNV burden iPSYCH cohort. LD Score regression was used to investigate in BD compared to controls, we integrate CNV, SNP and clin- the genetic correlation between BD and a range of human ical data from each subject in order to assess if different diseases and traits. A transcriptome-wide association study classes of genetic variation confer risk to different classes (TWAS) was conducted using data on gene expression in the of psychiatric symptoms. ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 25

Methods: For our primary analyses we tested CNV burden 2 NCRR-National Centre for Register-Based Research, in BD cases compared to controls at various size and fre- Aarhus University quency thresholds and followed-up all previous reported as- 3 University of Queensland sociations of CNV burden in BD. Next, we tested if com- 4 University of North Carolina at Chapel Hill mon SNPs and rare CNVs contribute to different psychiatric 5 CIRRAU - Centre for Integrated Register-based Research, symptoms within BD using CNV burden and polygenic risk National Centre for Register-based Research, Aarhus Uni- scores (PRS) derived from previous genome-wide association versity studies of schizophrenia (SCZ) and educational attainment 6 Aarhus University

(EDU), stratifying cases according to history of psychosis, 7 NORMENT, University of Oslo and Institute of Biological age of onset, and severity of clinical course. Psychiatry, St. Hans Hospital

Results: Primary analyses of CNV burden in 6,353 BD cases 8 Mental Health Centre Copenhagen, University of Copen- compared to 8.656 controls found no significant differences. hagen In cases diagnosed with the bipolar subtype of schizoaffective disorder (SAB), an increased rate of large ( > 500KB), Background: The postpartum period is a particularly vul- rare ( < 1% of sample) CNVs was observed relative to both nerable time for psychiatric disorders, with greater heri- controls (p-value = 0.001) and bipolar I disorder (BD I) cases tability than major depression (MD) outside of the perina- (p-value = 0.0002). By definition, all SAB cases have a his- tal period. How genetic liability of postpartum psychiatric tory of psychosis. To determine if CNV burden influenced disorders varies by other significant risk factors is unknown. risk for psychotic symptoms, we stratified BD I cases into a A single study has considered genetic risk scores (GRS) and group with and a group without psychosis. CNV burden and risk of postpartum depression and found that GRS for bipolar SCZ PRS were significantly higher in SAB compared to either disorder (BD) was more strongly associated with postpartum BD I group. In contrast, CNV burden did not differ between depression than major depression outside the postpartum BD I groups stratified by psychosis, while SCZ PRS was higher period. We aimed to: 1) estimate associations of GRS for in the group with psychosis compared to the group without. MD, BD, and schizophrenia (SCZ) with postpartum psychi- EDU PRS did not differ between SAB and either BD I group, atric disorders, and 2) examine differences by prior psychi- or between the BD I groups. We next stratified cases into atric history. those with childhood onset ( < 12 years old) and average on- Methods: We conducted a case-control study based on Dan- set (18-24 years old). Lower EDU PRS was seen in childhood ish population-based registers. Association analyses were compared to average onset cases in both SAB and BD I, while conducted between MD, BD, and SCZ GRS and case-control the SCZ PRS was not associated with age of onset. In SAB a status of a postpartum psychiatric disorder. The study con- higher CNV burden was seen in childhood compared to av- sisted of all women in the iPSYCH2012 cohort who had given erage onset cases, while in both SAB and BD I the childhood birth before December 31, 2015 (n = 8,931). Cases were onset cases with symptoms continuous throughout the life women with a diagnosed psychiatric disorder or a filled psy- course had higher CNV burden compared to those with an chotropic prescription within one year after delivery (n = episodic course. 6,516 cases, 2,334 controls). Genome-wide data were ob- Discussion: In the largest study of CNVs in BD to date, we tained from neonatal biobanks. Information on case status, do not find evidence of CNV burden in 6,353 BD compared family psychiatric history, personal psychiatric history, and to 8,656 controls. However, we show that a more nuanced other demographic factors were obtained from national reg- picture emerges when BD cases are stratified according to isters. Results from meta-analyses of MD, BD, and SCZ from clinical phenomenology. By integrating CNV and SNP data the Psychiatric Genomics Consortium including 23andMe and with high-dimensional clinical data we begin to dissect the excluding individuals in the iPSYCH2012 cohort were applied contribution of different classes of genetic variation to dif- to calculate genetic risk scores. Results are presented for ferent symptom profiles in BD. SNPs with P < 0.05. Results: Previously identified risk factors including primi- Disclosure: Nothing to disclose. parity and family history of psychiatric disorders were con- doi: 10.1016/j.euroneuro.2018.08.047 firmed. Parental psychiatric history was associated with postpartum psychiatric disorders among women with previous psychiatric history (OR, 1.14; 95% CI 1.02 – 1.28) but not without psychiatric history (OR, 1.08; 95% CI: 0.81 – 1.43). GRS for MD was associated with an increased risk of post-

41 partum psychiatric disorders in both women with (OR, 1.44;

GENETIC RISK SCORES AND POSTPARTUM PSYCHIATRIC 95% CI: 1.19 – 1.74, per 10-decile increase) and without (OR, DISORDERS 1.88; 95% CI: 1.26 – 2.81, per 10-decile increase) psychiatric history. GRS for BD and SCZ were associated with risk of psy- 1, ∗ 2 3 4 Anna Bauer , Xiaoqin Liu , Enda Byrne , Patrick Sullivan , chiatric disorders at any time point, but not with postpar- 3 5 6 Naomi Wray , Esben Agerbo , Mette Nyegaard , tum psychiatric disorders in our study. 2 2 2 Katja Ingstrup , Benedicte Johannsen , Merete Mægbæk , Discussion: Our cases predominantly consisted of depres- 7 8 Yunpeng Wang , Merete Nordentoft , sion diagnoses and women prescribed antidepressants, thus, 6 2 Preben Bo Mortensen , Trine Munk-Olsen , it is encouraging that MD GRS predicts psychiatric episodes 4 Samantha Meltzer-Brody in our target sample of postpartum women. We did not ﬁnd an expected association between GRS BD and postpartum

1 UNC School of Medicine ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 26 Abstracts psychiatric disorders; however, this may be due to case se- P < 0.01). Similar findings were obtained when examining lection and lower power of the BD GRS discovery sample. schizophrenia rare and common variant studies. In con- In conclusion, genetic liability for major depression was as- trast, we have observed no such risk relationship for burden- sociated with postpartum psychiatric disorders, suggesting positive language genes in neurodevelopmental conditions GRS can provide additional information about risk not en- (such as ADHD or autism), suggesting that the relationship compassed solely in simple measures of family history. may be specific to adult-onset conditions (like bipolar disorder and schizophrenia). Disclosure: Nothing to disclose. Discussion: Together, these findings suggest an intimate relationship between the genetic potentiators of human lan- doi: 10.1016/j.euroneuro.2018.08.048 guage and the risk mechanisms for major mental illness.

Disclosure: Nothing to disclose. 42

THE COST OF ADVANCED LANGUAGE ABILITY: IN- doi: 10.1016/j.euroneuro.2018.08.049 CREASED RISK FOR MENTAL ILLNESS?

Oral Session: Schizophrenia Tanner Koomar 1, Bruce Tomblin 1, James Potash 2, ∗ Jacob Michaelson 1, 1:15 p.m. - 2:45 p.m.

1 University of Iowa 43

2 Johns Hopkins University RARE COPY NUMBER VARIATIONS ARE ASSOCIATED WITH POORER COGNITION IN SCHIZOPHRENIA Background: Emerging results from our whole genome se- 1, ∗ 2 3 quencing studies of language suggest that so-called damag- Leon Hubbard , Elliott Rees , Derek Morris , 1 1 4 ing variation in specific brain genes and pathways actually Amy Lynham , Sophie Legge , Denise Harold , 1 4 4 leads to increased language ability. These same genes were Stanley Zammit , Aiden Corvin , Michael Gill , 1 5 6 found to be enriched (P = 0.003; OR = 1.7) for genomic inter- Michael O’Donovan , Michael Owen , Gary Donohoe , 1 1 1 actions with human accelerated regions (HARs), which are George Kirov , Andrew Pocklington , James T. R . Walters conserved non-coding elements showing human-specific accelerated substitution rates. Taken together, these findings 1 Cardiff University suggest there may be human-specific selective pressures in 2 MRC Centre for Neuropsychiatric Genetics and Genomics, favor of hypomorphic alleles in these pathways, resulting in Institute of Psychological Medicine and Clinical Neuro- overall improved abilities in language and social communi- sciences, Cardiff University cation. It remains unclear, however, what brain and behav- 3 Neuroimaging and Cognitive Genomics (NICOG) Centre, ior trade-offs may be incurred as a result of the accumu- School of Psychology, National University of Ireland Galway lation of these hypomorphic alleles. Such a tradeoff might 4 Trinity College Dublin manifest in individuals with superior language and social 5 MRC Centre for Neuropsychiatric Genetics and Genomics, communication skills who also have a corresponding eleva- Cardiff University tion in risk for mental illness. Previous reports in the litera- 6 National University of Ireland Galway ture have noted these patterns in performance artists such as actors and comedians. Background: Rare copy number variations (CNVs) are asso- Methods: In light of these observations, we sought to di- ciated with increased risk of neuropsychiatric disorders and rectly test the hypothesis that genes and pathways that con- lower cognitive ability in healthy individuals. Their contri- tribute positively to language ability with increasing genetic butions to cognitive functioning in people with schizophre- burden ("burden-positive", N = 472 genes from our pathway nia have not been clearly defined. We compared cognitive burden model presented elsewhere at this meeting) show a ability in schizophrenia patients with and without known corresponding excess of risk for neuropsychiatric conditions, schizophrenia CNVs. such as bipolar disorder and schizophrenia. Methods: 479 participants with schizophrenia recruited in Results: When examining the results of a study of rare vari- the UK were used in a within-case cross-sectional design ation in bipolar disorder, we found that predicted language in the Cardiff Cognition in Schizophrenia sample (Cardif- ability (based on genetic factors) was significantly and pos- fCOGS). We sought replication in an independent Irish sam- itively correlated with affected status (logistic regression ple of 519 participants with schizophrenia, schizoaffective P = 0.003; Z = 2.93). Further examination revealed that this disorder or schizophreniform psychosis. We investigated the trend was disproportionately driven by the presence of relationship between schizophrenia CNV status and general damaging variation (singleton variants with CADD > 15) in cognitive ability, measured using the MATRICS composite z- our burden-positive language genes, lending support to the score and estimated premorbid IQ in CardiffCOGS using lin- "trade off" hypothesis (logistic regression P = 0.004; Z = 2.85). ear regression. In the Irish sample, general cognitive ability Examination of summary statistics from genome-wide as- was measured using full-scale IQ. sociation studies (GWAS) of bipolar disorder further sup- Results: In CardiffCOGS, individuals with a schizophre- ported burden-positive language genes having significantly nia CNV (N = 8) performed over one standard deviation elevated risk for bipolar disorder (P = 0.003; OR = 4.17 at a below cases without a schizophrenia CNV (N = 471) on gene-wise median bipolar risk threshold of OR > 1.12 and the MATRICS composite z-score ( β= -1.16., 95%CI = -2.06 ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 27 to -0.25, p = 0.011). In the Irish sample individuals with Cochran’s (IVW) and Rücker’s (MR Egger) Q tests. All analy- schizophrenia CNVs (N = 7) performed 12 full-scale IQ points ses were performed using the TwoSampleMR R package. lower than cases without a schizophrenia CNV (n = 512) Results: The IVW MR method provided evidence of a ca- ( β= -0.96, 95%CI = -1.82 to -0.11, p = 0.027), replicating sual effect of genetically instrumented neuroticism on risk the direction and magnitude of effect observed in Cardif- of schizophrenia (p = 0.018). This causal association was fCOGS. Meta-analysis of the two samples showed those with also evident when using the weighted median (p = 0.002) schizophrenia CNVs (n = 15) performed approximately one and weighted mode (p = 0.018) approaches. However, the standard deviation below those with no schizophrenia CNV MR Egger intercept provided strong evidence of presence (n = 983) ( β= -1.06, 95%CI = -1.67 to -0.44, p = 8x10-4). We of horizontal pleiotropy (p < 0.001) and Q tests provided also observed schizophrenia CNV carriers in CardiffCOGS strong evidence of presence of heterogeneity in the effect were also impaired on a measure of estimated premorbid estimates (p < 0.001). There was also evidence of a causal IQ (NART; 9 IQ points difference, β= -1.01, 95%CI = -1.71 to effect of schizophrenia on neuroticism (IVW p = 0.001), -0.32, p = 0.004). however evidence was weaker when using the weighted me- Discussion: In the largest combined schizophrenia sample dian (p = 0.148) and weighted mode (p = 0.150) approaches. to date with cognition data, we have demonstrated that in- The I2GX statistic indicated potential violation of the NOME dividuals with the disorder who are carriers of schizophrenia assumption and Q tests again provided strong evidence of CNVs display clinically important levels of cognitive impair- presence of heterogeneity in the effect estimates (p < ment. 0.001). Discussion: Assuming certain MR assumptions are met, our Disclosure: Nothing to disclose. results provide evidence of a causal effect of neuroticism on schizophrenia, and also of schizophrenia risk on neu- doi: 10.1016/j.euroneuro.2018.08.050 roticism. However, MR Egger analyses provided evidence of horizontal pleiotropy and Q tests provided strong evidence of presence of heterogeneity. A potential implication of our evidence supporting a causal effect of neuroticism on

44 schizophrenia is that a greater focus on treatment of anx-

A TWO-SAMPLE MENDELIAN RANDOMIZATION STUDY iety in high-risk subjects could reduce risk of transition to ASSESSING A CAUSAL ASSOCIATION BETWEEN NEU- psychosis, however further work is needed, potentially us- ROTICISM AND SCHIZOPHRENIA ing biologically relevant sub-sets of anxiety speciﬁc SNPs, to establish whether the associations between anxiety and 1, ∗ 1 Hannah Jones , George Davey Smith , schizophrenia is causal.

Michael O’Donovan 2, Michael Owen 2, James Walters 2,

Stanley Zammit 2 Disclosure: Nothing to disclose.

doi: 10.1016/j.euroneuro.2018.08.051 1 University of Bristol

2 Cardiff University

45 Background: Anxiety is a prominent feature of schizophrenia, present in the prodromal phase of the illness. There is INTERROGATING THE EVOLUTIONARY PARADOX OF strong evidence that the personality trait neuroticism, an SCHIZOPHRENIA: A NOVEL EVIDENCE FROM NEAN- underlying factor strongly associated with anxiety, is genet- DERTHALS AND DENISOVANS ically correlated with schizophrenia, implying that neuroti- 1, ∗ 1 2 cism and schizophrenia share genetic risk factors in com- Chenxing Liu , Christos Pantelis , Ian Paul Everall , 3 mon. However, a genetic correlation may also suggest a pos- Chad Bousman sible role of neuroticism in the pathogenesis of schizophrenia. We therefore performed a Mendelian randomization 1 The University of Melbourne

(MR) analysis using publicly available data to investigate 2 Institute of Psychiatry, Psychology and Neuroscience, the potential casual association between neuroticism and King’s College London schizophrenia. 3 University of Calgary Methods: We performed bi-directional two-sample MR between neuroticism and schizophrenia using the most re- Background: Schizophrenia is a psychiatric disorder with a cent publicly available summary-level genome-wide data. worldwide prevalence of 1%. The high heritability and re- Independent, genome-wide significant (p ≤ 5e-8) single nu- duced fertility among schizophrenia patients have raised an cleotide polymorphisms (SNPs) associated with neuroticism evolutionary paradox: why has negative selection not elimi- (91 SNPs) and schizophrenia (88 SNPs) were combined us- nated schizophrenia associated alleles during the evolution ing an inverse-variance-weighted (IVW) multiplicative ran- process? dom effects approach. Impact of potential MR assumption Methods: To address this question, we examined evolu- violations were explored using weighted median, weighted tionary markers, known as modern-human-specific sites mode and MR Egger methods. The MR assumption of no mea- and archaic-human-specific sites, using existing genome- surement error in the SNP exposure associations (NOME as- wide association study data from 34,241 individuals with sumption) was assessed using FGX (IVW) and I2GX (MR Egger) schizophrenia and 45,604 healthy controls included in the statistics and presence of heterogeneity was assessed using Psychiatric Genomics Consortium. ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 28 Abstracts

Results: By testing the distribution of schizophrenia SNPs tive for schizophrenia; ii) SNPs demonstrating paradoxical with risk and protective effects in the human-specific sites, effects between schizophrenia and education, where alle- we observed a positive selection of protective alleles for les confer risk to schizophrenia but also advantage to ed- schizophrenia in modern humans relative to archaic humans ucation. “Concordant” variants were more likely to impli- (Neanderthal and Denisovans). Such findings indicated that cate early neurodevelopment pathways such as CHD8, cell risk alleles of schizophrenia have been gradually removed adhesion, cell projection, and cell metabolic processes. By from the modern human genome due to negative selection contrast, “paradoxical” variants were associated with adult pressure. neuronal function pathways such as voltage-gated ion chan- Discussion: Our study, for the first time, provides experi- nel signaling, synaptic transmission, and postsynaptic activ- mental evidence supporting the role of negative selection in ity. Genetic correlation analyses suggest that the paradox- eliminating risk alleles for schizophrenia. The present find- ical subset of variants may be elucidating a disease con- ings expand on the notion of schizophrenia as a ‘by product’ tinuum that implicates BMI, ADHD, cardiovascular disease, of obtaining high-level cognition, contribute to our under- mood and bipolar disorder. standing of the genetic origins of schizophrenia, and should Discussion: Pleiotropic analysis successfully identified more assist future research aimed at unraveling the genetic ar- than 100 novel loci underlying brain function, many of chitecture of the disorder. which are in the process of validation by larger, more recent single-phenotype GWAS. Using a subset-based tech- Disclosure: Nothing to disclose. nique, we were able to isolate subsets of genomic variants that could not have been otherwise investigated using doi: 10.1016/j.euroneuro.2018.08.052 standard inverse-variance based meta-analysis or genome- wide based multi-trait analysis. Leveraging the joint genetic correlations of cognition, education, and schizophrenia, we

46 were able to identify two broad biological mechanisms – USING PLEIOTROPY TO DISSECT FUNCTIONAL PATH- early neurodevelopmental and adulthood synaptic pruning WAYS IN COGNITION, EDUCATION, AND SCHIZOPHRENIA pathways that contribute not only to the etiopathogenesis of schizophrenia, but broader biological dimensions that are 1, ∗ 2 Max Lam , Todd Lencz implicated in both general health and psychiatric illness. Further work is necessary to further identify speciﬁc aspects

1 Institute of Mental Health of how these biological mechanisms could be implicated in

2 Zucker Hillside Hospital the disease process.

Background: Multiple studies have demonstrated an inverse Disclosure: Nothing to disclose. correlation between the genetic architectures of cognition doi: 10.1016/j.euroneuro.2018.08.053 and schizophrenia (Rg ∼.20), and a positive genetic correlation between education and cognition (Rg ∼ .70). Paradox- ically, however, a modest (though consistent) positive association between schizophrenia and educational has also 47 been reported. Here, we leverage prior results of multiple TRANSCRIPTOME-WIDE ASSOCIATION ANALYSIS OF GWASs on cognition, education, and schizophrenia to investigate this phenomenon, in an attempt to parse biological SCHIZOPHRENIA IMPLICATES PROCESSES INVOLVED mechanisms that might shed light on the etiopathogenesis IN SYNAPTIC DEVELOPMENT AND PLASTICITY, AND of schizophrenia. IMPAIRED LONG-TERM POTENTIATION Methods: GWAS summary statistics for cognition 1, ∗ 1 1 (N = 107,207), education (N = 328.917), and schizophre- Lynsey Hall , Christopher Medway , Antonio Pardiñas , 2 2 1 nia (N = 77,096) were meta-analyzed using a technique Elliott Rees , Valentina Escott-Price , Andrew Pocklington , 1 1 2 known as Analysis based on SubSets (ASSET), which allows James Walters , Peter Holmans , Michael Owen , 1 pleiotropic signals across the entire genome to be pooled Michael O’Donovan using a two-tailed z-test approach. Genomic loci were consolidated within subsets representing overlaps between 1 Cardiff University the three phenotypes. These subsets were subjected to 2 MRC Centre for Neuropsychiatric Genetics & Genomics, further downstream analyses including S-Predixcan brain Cardiff University tissue expression profile analysis (GTex and CommonMind); a series of MAGMA competitive and GENE2FUNC hyper- Background: Schizophrenia is a complex highly herita- geometric pathway analyses; and LD-score regressions for ble disorder. Genome-wide association studies have iden- genetic correlations with a range of other phenotypic traits. tified multiple loci that influence the risk of develop- Results: ASSET analysis revealed 300 lead SNPs across 235 ing schizophrenia, although the causal variants driving loci met the genome-wide significance threshold of p < 5e-8. these associations and their impacts on specific genes are Of these, 103 loci were novel when compared with the ini- largely unknown. Here we seek to link genetic findings to tial input GWAS. We focused downstream analyses on com- changes in gene expression in human brain by perform- paring two key subsets of variants: i) SNPs which showed ing a transcriptome-wide association study in which we effect concordance, where alleles conferred advantages to integrate the largest published genome-wide association cognitive performances, and education while being protec- dataset (40,675 schizophrenia cases and 64,643 controls) of ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 29 schizophrenia with publicly available post-mortem expres- reduced fecundity of patients, this rate is stable across pop- sion data from the dorsolateral prefrontal cortex (DLPFC). ulation groups separated by geography and time. To ex- Methods: TWAS was performed using FUSION soft- plain this stable occurrence of the disorder, the ‘Evolution- ware, using genome-wide summary statistics from the ary hypothesis of schizophrenia’ has been gaining ground. PGC2 + CLOZUK schizophrenia meta-analysis and DLPFC ex- The most well-known version of the hypothesis was proposi- pression weights from the Common Mind Consortium for tioned by TJ Crow in 1998. The key assertion made by Crow genes with significantly non-zero cis-heritable expression. A was the emergence of schizophrenia as a by-product of hu- competitive gene-set enrichment analysis, where gene-set man evolution via language. membership is a linear predictor of absolute TWAS z-score, In the past two decades, the emergence of genomic tech- was performed using MAGMA. This analysis was carried out nologies and resources has made it possible to test the for two different gene-sets; i) 10 significant gene-sets re- evolutionary aspect of the hypothesis. There is a growing ported in the CLOZUK + PGC2 manuscript and ii) 6677 data- body of evidence from the field of genomics that supports driven gene-sets. the notion that human evolution may have played a role in Results: We identify a significant correlation between schizophrenia. schizophrenia risk and expression at 89 genes in DLPFC, Developments within the last five years have allowed re- including 42 genes not identified in earlier similar stud- searchers to trace the evolution of epigenomes as well, giv- ies of this tissue. Genes whose expression correlate with ing an unprecedented window on gene-environment (GxE) schizophrenia were enriched for those involved in CNS interactions of the past several thousand to millions of synaptic transmission and long-term potentiation. 33 sig- years. As part of my PhD, I undertook a body of work that nificant TWAS genes, corresponding to 28 distinct loci, fell looks at the evolutionary question of schizophrenia from this outside established schizophrenia risk loci as defined by new field of evolutionary epigenetics. the CLOZUK + PGC2 meta-analysis. Given that TWAS might Methods: We first investigate whether human-specific dif- have greater power than GWAS to detect disease-associated ferentially methylated regions (DMRs), determined in com- genes when multiple variants influence the phenotype parison to Neanderthals and Denisovans (Gokhman et al, through changes in expression, we anticipate that among 2014) are enriched for schizophrenia markers. These methy- these loci will be true schizophrenia associated genes. For lated regions represent at least 750,000 years of evolu- gene-set analysis, of the 10 significant CLOZUK + PGC2 gene- tion since the last common ancestor diverged from Ne- sets, 2 were significantly enriched for TWAS signal; abnor- anderthals and Denisovans. We utilized summary statis- mal long-term potentiation (p = 6.03E-05) and genes that are tics from genome-wide association studies (GWAS) of intolerant to loss-of-function mutations (loss-of-function in- schizophrenia and 12 other phenotypes and tested for en- tolerance; p = 1.73E-04). In the broader data driven analysis richment of association in human, neanderthal & denisovan of 6159 gene sets, significant enrichment was found for 5 DMRs using a polygenic enrichment testing pipeline (Schork gene sets related to nervous system development, abnormal et al, 2013). This was followed up by investigating primate- synaptic transmission, abnormal long-term potentiation, re- methylated regions that represent at least 13MYA of epige- duced long term potentiation, and calcium-dependent cell- nomic evolution. Finally, we investigate whether human- cell adhesion, with all but abnormal long term potentiation specific methylated regions, as defined in the first study are remaining significant after conditional analysis. amenable to methylation variation in actual patients with Discussion: Previous genetic studies have implicated schizophrenia using a candidate evidence-based approach. post-synaptic glutamatergic and gabaergic processes in Results: We find evidence that recent evolution denoted by schizophrenia; here we extend this to include molecules human-specific methylated regions tracing ∼750,000 years that regulate presynaptic transmitter release. We identify of methylation development are enriched for schizophrenia large numbers of specific candidate genes to which we as- markers (Banerjee et al, 2018). Primate methylation mark- sign predicted directions of effect in terms of expression ers are not enriched for schizophrenia variants with the level, facilitating downstream experimental studies geared exception of the extended Major-Histocompatibility Region towards a better mechanistic understanding of schizophre- (MHC) region (Paper II, Schizophrenia Research, under re- nia pathogenesis. view). Finally, we find evidence of methylation disruption in brain samples of patients with schizophrenia in regions that Disclosure: Nothing to disclose. underwent human-specific methylation evolution (Paper III, manuscript). doi: 10.1016/j.euroneuro.2018.08.054 Discussion: Our results provide support that recent evolution, denoted by methylation changes since the divergence of the common ancestor of humans, Neanderthals

48 and Denisovans, may have played a role in susceptibility to AN EVOLUTIONARY EPIGENETICS APPROACH TO schizophrenia at a group-level. SCHIZOPHRENIA Disclosure: Nothing to disclose. Niladri Banerjee ∗, Tatiana Polushina, Stephanie Le Hellard doi: 10.1016/j.euroneuro.2018.08.055

University of Bergen

Background: Schizophrenia is a psychotic disorder with an estimated lifetime prevalence of ∼1% worldwide. Despite ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 30 Abstracts

Sunday, October 14, 2018 rophysiological deﬁcits in schizophrenia patients. Together with additional genetics clues, CaV3.3 may represent a novel target of schizophrenia where modifying thalamocor- tical connection could be a novel therapeutic strategy for Oral Session: Schizophrenia treating schizophrenia. 12:30 p.m. - 2:00 p.m. Disclosure: Nothing to disclose. 49 doi: 10.1016/j.euroneuro.2018.08.056 DYSFUNCTION OF CACNA1I IMPAIRED SLEEP SPIN- DLES DURING NREM

Ayan Ghoshal 1, david Uygun 2, lingling Yang 1, 50

Violeta Lopez-Huerta 1, Mario Arias Garcia 1, Zhanyan Fu 1, DE NOVO LOF AND DAMAGING MISSENSE VARIANTS

James McNally 2, Qiangge Zhang 3, Xiaohong Mao 4, ASSOCIATED WITH SCHIZOPHRENIA ARE CONCEN-

Thomas Nicholson 4, Guoping Feng 3, Robert Strecker 2, TRATED IN LOF INTOLERANT GENES ∗ Shaun Purcell 5, Qian Pan 1, ∗ Elliott Rees 1, , Jun Han 2, Madeleine Dufﬁeld 2, 3 4 5 6 1 Broad Institute of Harvard and MIT Sarah Tosato , Celso Arango , Micha Gawlik , Jim van Os , 7 2 2 Harvard Medical School Andrew Pocklington , Valentina Escott-Price , 7 7 7 3 MIT George Kirov , Peter Holmans , James Walters , 2 7 4 Novartis Institute of Biomedical Research Michael Owen , Michael O’Donovan

5 Icahn School of Medicine at Mount Sinai

1 MRC Centre for Neuropsychiatric Genetics and Genomics, Background: Schizophrenia affects 1% of the population Institute of Psychological Medicine and Clinical Neuro- yet has no effective treatments. Antipsychotic medications sciences, Cardiff University 2 can interrupt positive symptoms such as hallucinations in a MRC Centre for Neuropsychiatric Genetics & Genomics, subset of individuals, but they have little impact on neg- Cardiff University 3 ative or cognitive symptoms. Novel and more effective University of Verona 4 treatments for schizophrenia patients depend on emerg- Universitario Gregorio Marañón 5 ing clues from the recent psychiatric genetics that are ex- University of Wuerzburg 6 pected to converge on neurobiological pathways. CACNA1I is University of Utrecht 7 a schizophrenia risk gene identified in GWAS. It encodes Cardiff University the pore-forming human CaV3.3 α1 subunit, a subtype of voltage-gated calcium channel that contributes to T-type Background: Exome sequencing studies have shown de novo currents. CaV3.3 expresses abundantly in Thalamic Reticu- coding variants to be important risk factors for neuro- lar Nucleus (TRN) and mediate rebound bursting in TRN that psychiatric disorders, such as intellectual disability (ID), plays a key role in sleep spindle generation. Previously, we autism spectrum disorder (ASD), and schizophrenia. The risk reported biophysical and biochemical deficits of a missense from de novo LoF variants for ASD and ID has been refined variant R1346H of hCaV3.3, a de novo mutation discovered to those not observed as standing variation in large pop- in a schizophrenia patient, when expressed in human cell ulation reference exomes and occurring in LoF intolerant lines. genes. A similar non-significant trend has been observed Methods: We have generated CaV3.3-knockout and CaV3.3- for schizophrenia, although current trio samples are smaller R1305H knock-in animals using CRISPR-cas9 genome editing than in studies of ID and ASD. Here we analysed de novo and studied the neurobiological phenotypes of these mice. variants in a new schizophrenia sample (n = 624) and com- We studied the whole-cell patch clamp of TRN neurons, as bined it with published data (total n = 1,763) to refine the well as TRN excitability using ex vivo brain slice recording. risk for schizophrenia from de novo coding variants to spe- Subsequently, we utilized in vivo EEG recording of naturally cific genes. wake/sleep animals to investigate the impact of CaV3.3 on Methods: We performed exome sequencing of 675 sleep spindles. new proband-parent schizophrenia trios using Illumina Results: We showed that R1346H is associated with lower HiSeq4000 instruments. A validation experiment was per- CaV3.3 membrane-bound levels, together with lower whole formed through Sanger sequencing 150 putative de novo cell T-type current density in TRN neurons. Reduction in variants. To refine de novo risk variants, we followed pub- CaV3.3 function was also accompanied by reduced rebound lished methods by comparing the rate of de novo variants burst firing of TRN neurons. This reduction in TRN func- not observed/observed in ExAC and occurring in LoF intol- tion was reflected in a significant reduction in NREM specific erant/LoF tolerant genes between schizophrenia and unaf- spindle oscillations in both the KO and R1346H mice, de- fected siblings from published ASD studies. The rate of syn- tected using a novel automated wavelet algorithm, without onymous de novo variants was used as a negative control. alternation in the overall sleep architecture. Single Gene and gene-set enrichment tests compared the Discussion: As sleep spindle density in schizophrenia pa- observed number of de novo variants to that expected based tients have been shown to be reduced, our data suggest on per-gene mutation rates. a potential causal link between CACNA1I and certain neu- Results: After stringent sample and variant quality control, we observed 650 coding de novo variants in 624 trios, corre- ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 31 sponding to a coding de novo rate (1.04 per proband) consis- AHC is a rare disease starting typically before the age of 18 tent with the literature. A high validation rate was observed months and the clinical presentation is characterized by re- for both SNVs (0.95) and indels (0.94). The rate of synony- peated episodes of hemiplegia that alternately affects one mous de novo variants was not significantly different be- side of the body. A few genes are known to be associated tween schizophrenia and controls for any test. Although the with AHC and these guided our genetic exploration of the rate of LoF de novo variants (both observed and unobserved two patients using targeted sequencing of the ATP1A3 gene. in ExAC) did not differ between schizophrenia and controls Then, we replicated our findings in an independent cohort in LoF tolerant genes, a significant excess of schizophrenia of 17 trios of unrelated COS cases recruited in the NIMH. We variants not observed in ExAC was found in LoF intolerant performed whole exome sequencing (WES) and specifically genes (risk ratio (RR) = 1.86, P = 0.00053). The rate of dam- investigated this gene and its known brain-expressed inter- aging missense de novo variants (defined as variants with actors. Only variants in exonic positions, with a frequency MPC scores > 2) not observed in ExAC in LoF intolerant genes < 0.01 in the 1000 Genome project and ExAC database, iden- was also increased in schizophrenia (RR = 2.13, P = 0.0032). tified as possibly damaging by at-least three algorithms, and In an analysis of 134 central nervous system gene sets, FMRP in a phylogenetically-conserved position were retained. targets were associated with nonsynonymous de novos after Results: In the cases with comorbid AHC, two distinct correction for multiple testing. No single gene remained sig- pathogenic de novo variants were identified in the ATP1A3 nificant after correction for multiple testing. gene. In the replication cohort, we identified a third case Discussion: Our new exome sequencing sample substan- with a possibly damaging missense variant in the same tially increases the number of de novo variants studied in exon of ATP1A3. Three other cases presenting missense schizophrenia, thus increasing our power to identify risk variants predicted to be deleterious in genes from the variants. Following previous observations in studies of ASD FXYD family (FXYD1, FXYD6, and FXYD6-FXYD2 readthrough) and ID, we find significant evidence that schizophrenia de were observed. ATP1A3 encodes the α-subunit of a neuron- novo variants not observed as standing variation in ExAC are specific ATP-dependent transmembrane sodium-potassium enriched in LoF intolerant genes. We also provide novel ev- pump. FXYD genes encode proteins that modulate the ATP- idence that damaging missense de novo variants in these dependent pump function. Their patterns of brain expres- genes confer similar levels of risk as LoF de novo variants. sion are compatible with the age of onset. However, no novel genes were associated with schizophre- Discussion: Our report is the first to identify variants in the nia after correction for multiple testing, suggesting even same pathway for COS. It illustrates the interest of explor- larger samples are required for novel gene discovery. ing medical comorbidities and stratifying a complex condition according to the age of onset for the identification of Disclosure: Nothing to disclose. deleterious missense variants. ATP1A3 is a replicated gene in rare neuropediatric diseases and we extended the phe- doi: 10.1016/j.euroneuro.2018.08.057 notype to a pure psychiatric presentation. The association with rare variants in FXYD gene family is novel. Finally, our study highlights the interest of DNA sequencing in psychiatry

51 and opens the way to genetic counseling. IDENTIFICATION OF NEW GENES ASSOCIATED WITH CHILDHOOD-ONSET SCHIZOPHRENIA: ATP1A3 AND THE Disclosure: Nothing to disclose. FXYD GENE FAMILY doi: 10.1016/j.euroneuro.2018.08.058 ∗ Boris Chaumette 1, , Vladimir Ferraﬁat 2,

Amirthagowri Ambalavanan 1, Alice Goldenberg 2,

Alexandre Dionne-Laporte 1, Dan Spiegelman 1, 52

Patrick Dion 1, Priscille Gerardin 2, Claudine Laurent 3, SCHIZOPHRENIA AND URBAN LIVING: A STUDY ON

David Cohen 3, Judith Rapoport 4, Guy Rouleau 1 HALF A MILLION PEOPLE FROM THREE COUNTRIES

∗ 1 McGill University Lucia Colodro Conde 1, , Baptiste Couvy-Duchesne 2,

2 CHU Rouen John Whitﬁeld 1, Fabian Streit 3, Loic Yengo 2,

3 Hopital Pitié Salpêtrière Maciej Trzaskowski 2, Eveline de Zeeuw 4, Michel Nivard 4,

4 NIMH David Whiteman 1, Dorret Boomsma 4, Jian Yang 2,

Marcella Rietschel 3, John McGrath 5, Sarah E. Medland 1, 6 Background: Childhood-onset schizophrenia (COS) is a se- Nick Martin vere form of schizophrenia deﬁned as onset before age of

13. This juvenile form of schizophrenia has similar disease 1 QIMR Berghofer Medical Research Institute manifestations to adult-onset schizophrenia with which it 2 University of Queensland shares a number of genetic risk factors. It is a rare disease 3 Central Institute of Mental Health with a prevalence estimated to be of 0.03%. The rate of co- 4 VU University morbidity of other neurodevelopmental disorders and medi- 5 QLD Brain Institute cal conditions are higher than in forms of schizophrenia with 6 Queensland Institute of Medical Research a later onset. Methods: We identiﬁed two unrelated cases diagnosed with Background: Schizophrenia is more prevalent in urban than both COS and alternating hemiplegia of childhood (AHC). in rural areas. This association has been observed in differ- ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 32 Abstracts ent countries and remains signiﬁcant after controlling for Disclosure: Nothing to disclose. possible confounders (OR [95% CI] = 1.72 [1.53, 1.92] ac- doi: 10.1016/j.euroneuro.2018.08.059 cording to meta-analysis). A large body of research has been generated on this topic after the classic study from Faris and Dunham in 1931. The stress of urban living has been proposed as a risk (and causal) factor for the disease.

We have investigated an alternative (but not incompat- 53 ible) explanation: that people with higher genetic risk for CONTROLLING TYPE 1 ERROR RATE IN CNV GENE- schizophrenia tend to live in more urbanized areas due to SET ENRICHMENT TESTS selective migration in either past or current generations. 1, ∗ 2 1 Methods: We used data of genotyped adults individuals from Andrew Pocklington , Elliott Rees , George Kirov , 1 1 2 four independent non clinical cohorts. Analyses performed James Walters , Peter Holmans , Michael Owen , 1 in the discovery cohort (QIMR, Australia, N = 15,544) were Michael O’Donovan replicated in participants from the UKB (United Kingdom,

N = 456,426), the NTR (The Netherlands, N = 16,434) and 1 Cardiff University

QSKIN (Australia, N = 15,726). 2 MRC Centre for Neuropsychiatric Genetics and Genomics, We calculated polygenic risk scores (PRS) of schizophre- Cardiff University nia for our participants, using the summary statistics of the GWAS on schizophrenia published in 2014 by the Psychiatric Background: Gene-set enrichment analysis of case-control Genomics Consortium. We fit linear mixed models with pop- CNV data has been instrumental in identifying association ulation density of the area of residence as the outcome and between the genetic perturbation of synaptic function and the PRS as a predictor, controlling for demographic variables schizophrenia risk (Kirov et al. 2012, Szatkiewicz et al. and ancestry principal components. 2014, Pocklington et al. 2015, Marshall et al. 2017). The re- Using multi-instrument Mendelian Randomisation (MR) we gression models used in these analyses belong to the class tested the hypothesis that having a higher propensity to of self-contained tests, which are known to inabste gene-set schizophrenia causes a person to live in a denser and less enrichment statistics in common variant analyses (de Leeuw remote area. Although with reduced statistical power, we et al. 2015). More recent evidence indicates that inabstion also tested the reverse causation hypothesis that population is also present in CNV gene-set enrichment statistics (Singh density induces the onset of schizophrenia. et al. 2017). Here we evaluate the extent to which inabsted Results: Our polygenic risk score analysis showed that PRS type 1 error rates confound synapse gene-set association in for schizophrenia predicted population density in our dis- schizophrenia case-control CNV analyses. covery and replication cohorts, so a higher genetic risk for Methods: Analyses were carried out on the combined ISC, the disease was associated with a denser area of residence. MGS and CLOZUK CNV dataset of (Rees et al. 2016). A stan- This association remained significant after controlling for dard logistic regression-based enrichment test (Pocklington socio-economic status (SES) of the area of residence. We et al. 2015) was performed on randomly generated gene- also tested whether PRS for schizophrenia predicted SES of sets to evaluate the rate of type 1 errors in this dataset. We the area of residence. Results were only significant in the then investigated methods for overcoming p-value inabstion UKB, so a higher genetic risk would be associated with liv- based upon comparison with random gene-sets or through ing in more deprived neighbourhoods. fitting a generalised least-squares regression model. These Our MR results suggested that schizophrenia could be a were used to generate unbiased p-values for synaptic gene- causal factor for living in denser areas. These results were sets previously associated with schizophrenia via CNVs. only significant in the UKB, our largest cohort. Results are Results: The p-values for randomly generated, biologically suggestive of a reverse positive causal relationship between uninformative gene-sets diverged strongly from the ex- population density and schizophrenia, with an effect size pected null distribution, showing clear evidence of inab- five times higher than the effect observed for schizophrenia stion. A number of synaptic gene-sets previously reported causing living in more populated areas (0.20 vs. 0.04). to be enriched in CNVs from individuals with schizophrenia Discussion: Our study investigates the association between survive correction for inabstion of the test statistic. genetic risk for schizophrenia and characteristics of where Discussion: First noted in gene-set enrichment analyses people live with the aim of increasing our knowledge of why of common polymorphisms (de Leeuw et al. 2015), sus- this disease is more prevalent in cities. We used data on ceptibility to p-value inabstion is an inherent weakness of where people live collected as part of four studies, from self-contained tests that must be taken into account when three countries (Australia, UK, Netherlands) for a total num- analysing other classes of variant. In the case of CNVs, inab- ber of 504,130 participants. stion is driven by rare (MAF < 0.1%) and highly multi-genic Our results show that the distribution of the genetic risk CNVs (Singh et al, 2017), a class which contains all robustly for the disorder is not uniform and concentrates in more associated loci. Attempting to ameliorate inabstion by re- populated, providing empirical evidence that the increased moving these loci prior to analysis (equivalent to removing schizophrenia prevalence in urbanized areas is not only due all genome-wide significant loci from a GWAS) is not a sat- to the environmental stressors of the city but also on the isfactory solution. While a self-contained test can in princi- genetic risk for the disease. Altogether, our results support ple be converted to a competitive test through comparison the selective migration hypothesis. to random gene-sets, this is typically a time-intensive pro- cedure and alternative methods capable of accounting for correlation between genes are preferable. Enrichment of ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 33 risk variants in synapse-related gene-sets remains one of the serve as a roadmap for inferring the biology of disease in most consistent genetic findings in schizophrenia to date. other disorders.

Disclosure: Nothing to disclose. Disclosure: Nothing to disclose. doi: 10.1016/j.euroneuro.2018.08.060 doi: 10.1016/j.euroneuro.2018.08.061

54 Oral Session: Functional Genomics EXOME SEQUENCING OF 23,851 CASES IMPLICATES 12:30 p.m. - 2:00 p.m. NOVEL RISK GENES AND PROVIDES INSIGHTS INTO THE GENETIC ARCHITECTURE OF SCHIZOPHRENIA 55 EXPRESSION QUANTITATIVE TRAIT LOCI IN THE DE- Tarjinder Singh ∗, Benjamin Neale, Mark Daly, SCHEMA VELOPING HUMAN BRAIN AND THEIR ENRICHMENT IN Consortium NEUROPSYCHIATRIC DISORDERS

Massachusetts General Hospital Heath O’Brien 1, Eilis Hannon 2, Matthew Hill 3,

Carolina Toste 1, Matthew Robertson 1, Joanne Morgan 1, 4 4 5 Background: Schizophrenia, a debilitating psychiatric dis- Gemma McLaughlin , Cathryn Lewis , Leonard Schalkwyk , 1 3 3 order, has a substantial genetic component with common Antonio Pardinas , Michael Owen , Michael O’Donovan , 6 1, ∗ intergenic and rare coding variants contributing to risk. De- Jonathan Mill , Nick Bray spite the discovery of hundreds of common risk loci, only a handful of associations have resulted in validated func- 1 Cardiff University School of Medicine tional variants that pinpoint novel biology underlying dis- 2 University of Exeter Medical School ease pathogenesis. This central challenge is shared with 3 Cardiff University most complex polygenic disorders. To address this short- 4 King’s College London coming, sequencing studies of rare coding variants can com- 5 University of Essex plement existing approaches by pinpointing likely causal 6 University of Exeter genes overlapping common risk loci, and complete the allelic spectrum in disease genes. However, success to this Background: Genetic influences on gene expression in the end has been hampered by power limitations. human fetal brain plausibly impact upon a variety of postna- Methods: The Schizophrenia Exome Sequencing Meta- tal brain-related traits, including susceptibility to neuropsy- Analysis (SCHEMA) Consortium is a global effort to analyze chiatric disorders. However, to date, there have been no whole-exome and genome sequencing data to advance gene studies that have mapped genome-wide expression quanti- discovery. We have sequenced 23,851 cases and 50,996 con- tative trait loci (eQTL) specifically in the human prenatal trols, which include individuals of European, Latin Amer- brain. ican, East Asian, Ashkenazi Jewish, and African American Methods: We performed strand-specific, whole transcrip- ancestry. We performed extensive quality control steps on tome sequencing of total RNA derived from brain tissue from all data jointly, with consideration of coverage differences 120 human fetuses aged 12-19 post-conception weeks, de- between capture technologies. We similarly processed an riving expression measures for 144,448 Ensembl transcripts, additional 56,100 non-psychiatric samples from gnomAD for annotated to 28,875 genes. Genomic DNA from each sample use as population controls. Our expanded data set consists was genotyped for approximately 710,000 single nucleotide of 23,851 cases and over 100,000 controls, one of the largest polymorphisms (SNPs), followed by genotype imputation us- sequencing analyses to date. ing the Haplotype Reference Consortium r1.1 panel. Cis- Results: We first implicate protein-truncating variants eQTL were identified by linear regression of allele dosage (PTVs) in two novel genes, TRIO and HERC1, as conferring against gene expression measures, adjusted for PEER fac- substantial risk for schizophrenia, and replicate a known tors and other covariates, using FastQTL. association in SETD1A. After adding external controls, we Results: We identified high confidence cis-eQTL for > 1300 identify additional novel genes at exome-wide significance, genes and > 3000 individual transcripts (FDR < 0.05). Fe- including NMDA receptor subunit GRIN2A, a target of psy- tal brain eQTL were found to be enriched among risk vari- choactive drugs. We discuss signs of convergence with com- ants for attention deficit hyperactivity disorder, schizophre- mon schizophrenia loci and risk genes for neurodevelopmen- nia and bipolar disorder. We further identified changes tal disorders, showing an allelic series in GRIN2A and an in gene expression within the prenatal brain that poten- association of de novo mutations in SETD1A and TRIO to tially mediate risk for neuropsychiatric traits, including in- broader neurodevelopmental phenotypes. After excluding creased expression of C4A in association with genetic risk novel risk genes, schizophrenia cases still carry a substan- for schizophrenia, increased expression of LRRC57 in associ- tial excess of rare PTVs, suggesting that more remain to be ation with genetic risk for bipolar disorder and altered ex- discovered. Finally, we present an online browser that dis- pression of multiple genes within the chromosome 17q21 in- plays variant- and gene-based results. version in association with variants influencing the person- Discussion: In summary, analyses of whole exomes comple- ality trait of neuroticism. ment those of common variants in expanding our under- Discussion: We have mapped eQTL operating in the human standing of schizophrenia, and the combined approach can fetal brain, providing evidence that these confer risk to ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 34 Abstracts certain neuropsychiatric disorders, and identifying gene ex- Discussion: We identified several methylation sites and pression changes in the prenatal brain that could mediate genes that may provide important leads for future func- susceptibility to these conditions. tional studies to get insight in the mechanisms through which genetic variants influence ADHD risk at an early devel- Disclosure: Nothing to disclose. opmental stage. We are going to follow up on these results by analyzing the currently largest ADHD GWA meta-analysis doi: 10.1016/j.euroneuro.2018.08.062 on ADHD diagnosis and symptoms including 149,290 individuals to increase statistical power. In addition, we will analyze a large dataset of blood samples to increase statistical power as well, as it has been shown that eQTLs and mQTLs 56 of brain and blood highly correlate. With these datasets INTEGRATIVE ANALYSIS OF GENOME-WIDE AS- we expect to identify additional DNA methylation sites and

SOCIATION STUDY RESULTS OF ATTENTION- genes associated with ADHD. DEFICIT/HYPERACTIVITY DISORDER (ADHD) AND HUMAN FETAL BRAIN METHYLATION DATA REVEALS Disclosure: Nothing to disclose. NOVEL GENES ASSOCIATED WITH ADHD doi: 10.1016/j.euroneuro.2018.08.063 ∗ Anke Hammerschlag 1, , Enda Byrne 2, ADHD Group 3,

Meike Bartels 1, Naomi Wray 2, Christel M. Middeldorp 2 57

1 Vrije Universiteit Amsterdam METHYLOMIC BIOMARKERS OF SCHIZOPHRENIA AND

2 University of Queensland ANTIPSYCHOTIC MEDICATION EXPOSURE

3 Psychiatric Genomics Consortium

Eilis Hannon 1, Emma Dempster 2, Georgina Mansell 2, 3 4 5 Background: Attention-Deficit/Hyperactivity Disorder Leo Schalkwyk , Robin Murray , Andrew McQuillin , 6 7 8 (ADHD) is a neurodevelopmental psychiatric disorder af- Kaarina Kowalec , David St. Clair , Derek Morris , 9 10 11 fecting 5% of children and 2.5% of adults. ADHD is highly Patrick Sullivan , Michael O’Donovan , James MacCabe , 12 2, ∗ heritable and recently the first genome-wide associated CRESTAR Consortium, David Collier , Jonathan Mill genetic variants have been identified. The next step is to identify the functional implications of the identified 1 University of Exeter Medical School variants. As methylation status is associated with gene 2 University of Exeter expression, methylation is one of the possible regulatory 3 University of Essex mechanisms through which genetic variants can exert their 4 Institute of Psychiatry effects. 5 University College London

Methods: We apply the analytical paradigm SMR (summary 6 Karolinska Institutet data–based Mendelian randomization) and HEIDI (hetero- 7 University of Aberdeen geneity in dependent instruments) to integrate summary- 8 Neuroimaging and Cognitive Genomics (NICOG) Centre, level data from genomic, transcriptomic and methylomic School of Psychology, National University of Ireland Galway studies to detect DNA methylation sites that are associated 9 University of North Carolina at Chapel Hill with gene expression and ADHD through shared genetic ef- 10 Cardiff University fects (i.e., pleiotropy). Because ADHD is thought to be a 11 King’s College London neurodevelopmental disorder, we investigate methylation 12 Eli Lilly and Company quantitative trait loci (mQTLs) measured in human fetal brain tissue. In addition, we investigate mQTLs and expres- Background: There is growing interest in the role of de- sion quantitative trait loci (eQTLs) measured in adult human velopmentally regulated epigenetic variation in the molec- brain to increase statistical power as these datasets are sev- ular etiology of schizophrenia, with studies of disease- eral magnitudes larger. We integrate this data with genomic discordant monozygotic twins, clinical sample cohorts and data from the currently largest ADHD genome-wide associa- post-mortem brain tissue identifying methylomic variation tion study (GWAS) on ADHD diagnosis including 55,374 indi- associated with disease. Leveraging on the considerable in- viduals. vestment in genome-wide association studies (GWAS), we Results: Our preliminary results reveal several DNA methy- are examining genome-wide patterns of DNA methylation lation sites in fetal and adult human brain samples that show across multiple cohorts with the aim of undertaking an associations with ADHD through pleiotropy at shared genetic integrated genetic-epigenetic approach to schizophrenia. variants. These sites influence the expression levels of mul- Although effective in many patients, approximately one- tiple genes, including PTPRF. This gene is within the chromo- third of schizophrenia cases are resistant to commonly pre- some 1 locus identified by the ADHD GWAS including many scribed antipsychotic medications. To date, the atypical an- genes, and our analysis suggests that this might be the gene tipsychotic drug clozapine is the only evidence-based treat- of interest at the locus. In addition, we found associations ment for these individuals, although its use is often asso- between gene expression levels measured in adult human ciated with severe side-effects. Clozapine is known to in- brain and ADHD through pleiotropy at shared genetic vari- fluence chromatin remodelling and has previously been as- ants, including TIE1 and FOLH1 that have not been reported sociated with global hypomethylation in the leukocytes of in the ADHD GWAS, likely because of the lack of power. schizophrenic patients. In addition to identifying peripheral ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 35 blood methylomic signatures of schizophrenia, this study sion levels and MAGMA for pathway analysis. Interactions aimed to identify epigenetic variation associated with an- between drugs and targets (or genes) were mined from tipsychotic medication exposure. various databases: ChEMBL v23, DGIdb, PubChem, Ki DB, Methods: DNA methylation was profiled in whole blood sam- DSigDB, and PHAROS. We present results for 3 pheno- ples from i) multiple schizophrenia case-control cohorts us- types: schizophrenia (PMID 25056061), inabsmmatory bowel ing the Illumina 450K HumanMethylation array (3,172 cases diseases (PMID 26192919), and Alzheimer’s disease (PMID and 2,453 controls) and ii) a cohort of chronic and first- 24162737). episode schizophrenia patients (n = 159) who were com- Results: CACNA1C is the top schizophrenia-associated gene pared to matched samples prescribed alternative medi- in the top drugs network. S-PrediXcan predicts negative cations (n = 439). Following stringent quality control, an regulation for CACNA1C in the cerebellar hemisphere (zs- epigenome-wide association study was performed to i) iden- core = -4.5) and substantia nigra (z-score = -4.9). Within tify differentially methylated positions (DMPs) associated the antipsychotics-only network, top genes are CACNA1I, with schizophrenia and ii) to compare schizophrenia pa- CHRM4, DRD2, ABCB1 and HTR5A. CACNA1I is connected to tients prescribed clozapine to those prescribed alternative penfluridol, a T-type calcium channel blocker. For inabsm- medications. matory bowel diseases, gene hubs are JAK2, GPX1, IL10, Results: Widespread schizophrenia-associated DMPs were STAT3 and TNF. JAK2, IL10, STAT3 and TNF are implicated identified in our EWAS of schizophrenia, with a meta- in anti-inabsmmatory cytokine pathways. JAK2 is signifi- analysis across cohorts identifying robust methylomic sig- cantly upregulated in esophageal mucosa (z-score = + 5.2), natures of disease. Of note, we identified 27 DMPs asso- and GPX1 in the terminal ileum (z-score = + 5.2). GPX1 en- ciated with clozapine exposure (P < 1x10-7); 26 of these codes glutathione peroxidase 1, an antioxidant enzyme. The 27 sites were associated with hypermethylation in pa- first drug hit connected to GPX1 is tocopherol (vitamin E). tients prescribed clozapine, significantly more than ex- For Alzheimer’s disease, the top gene is GPX4. Top drugs pected by chance (P = 4.17x10-7). These DMPs were sub- for Alzheimer’s disease are ACE inhibitors (not significant). sequently tested in two replication cohorts (n = 268; 289) Interestingly, tocopherol is also in the top drug hits. with a highly significant enrichment of consistent directions Discussion: With Drug Targetor networks, we found drug of effect (P = 0.000311 and P = 4.17x10-7). Finally, we show suggestions consistent with previous knowledge: anti- that at many sites where differential DNA methylation is as- inabsmmatory or antioxidant drugs for inabsmmatory bowel sociated with schizophrenia there is a considerable overlap disease, calcium blockers and D2 antagonists for schizophre- with medication effects. nia, and ACE inhibitors for Alzheimer’s disease. Although re- Discussion: We identify multiple differentially methylated sults highly depend on the GWAS study sample size, Drug positions associated with schizophrenia, which overlap Targetor might help to better understand drug mechanisms those associated with clozapine exposure. Given the dif- using genetic evidence. ficulties in establishing causal associations in epigenome- wide association studies, these data demonstrate the utility Disclosure: Nothing to disclose. of pharmacoepigenetic studies to aid the interpretation of doi: 10.1016/j.euroneuro.2018.08.065 schizophrenia case control studies.

Disclosure: Nothing to disclose. 59 doi: 10.1016/j.euroneuro.2018.08.064 GENOME-EDITING OF THE RERE SUPER-ENHANCER IN HUMAN NEURAL PRECURSOR CELLS ALTERS EXPRES- SION OF NEURODEVELOPMENTAL AND SCHIZOPHRENIA 58 RISK GENES DRUG TARGETOR: AN ONLINE TOOL TO VISUALIZE ∗ GENETICS-DRIVEN DRUG-TARGET NETWORKS Cathy Barr 1, , Yu Feng 1, Karen Wigg 1,

Maria Carol Marchetto 2, Fred Gage 2 Helena Alexandra Gaspar ∗, Gerome Breen

1 The Krembil Research Institute

King’s College London 2 Salk Institute for Biological Sciences

Background: Genome-wide association studies (GWASs) Background: The majority of associated markers for com- have shown their ability to identify drug targets, and statis- plex genetic traits reside in gene regulatory regions, partic- tical techniques have been developed to impute expression ularly enhancers and super-enhancers. Enhancers can reside levels in different tissues. This heterogeneous data can be megabases from the gene they regulate (target gene) and used to prioritize drugs and targets. We present Drug Targe- their targets are often not the nearest gene. Thus, the as- tor v1.0 (drugtargetor.com) a web application that draws sumption that the gene nearest a GWAS-significant marker on both fields of chemistry and genetics to visualize inter- will be the risk gene will in many cases be incorrect. actions between drugs and genes. Methods: To identify the target genes of enhancers with Methods: Drug Targetor orders drugs and genes by GWAS GWAS significant markers, we analyzed our Capture-HiC association, and connects drugs to genes by type of inter- data selecting enhancers for functional studies using action. GWAS summary statistics were collected for over CRISPR/Cas9 in human neural precursor cells (hNPCs) de- 500 phenotypes. S-PrediXcan was used to impute expres- rived from embryonic stem cells. The impact on expression ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 36 Abstracts was measured by digital droplet PCR and RNA-seq in the treatments, targeted to those patients who will gain most edited versus the mock-transfected cells. benefit. The SH-SY5Y neuroblastoma cell line shows reduced Results: We selected the super-enhancer spanning the 3’ proliferation in media supplemented with 8.5mM lithium end of the RERE gene for study, the site of GWAS signif- chloride. We have leveraged this phenotype to carry out a icant SNPs for schizophrenia and major depression. Fur- functional genetic screen for mutants with altered prolifer- ther, RERE is a co-repressor/co-activator involved in retinoic ation rates when treated with lithium. We hypothesised that acid signaling and the key gene in 1p36 deletion syn- such mutations would occur in genes mediating the cellular drome, a developmental disorder with autism spectrum actions of lithium, with potential relevance to the variation symptoms. Capture-HiC data indicate interactions of the in therapeutic response observed in patients. super-enhancer with RERE, PARK7 (Parkinsons 7, protects Methods: We used a ‘gene trap’ vector transfection neurons from oxidative stress and regulates dopamine neu- methodology to create a library of over 10,000 different rotransmission) and PER3 (Period 3). These 3 genes are heterozygous loss-of-function mutant SH-SY5Y cells. This li- transcription co-regulators or transcription factors. Us- brary was cultured in parallel for 12 cell doublings in the ing CRISPR/Cas9, we deleted a 2kb region of the super- presence or absence of lithium. High-throughput sequencing enhancer in hNPCs and analyzed the transcriptome by RNA- was applied to both cell populations to quantify longitudinal seq. We identified 107 genes that were differentially ex- changes in individual mutant allele frequency (via altered pressed, including 14 regulated by retinoic acid. Impor- proliferation rate) that were specific to lithium selection tantly, 3 of these are located in independent GWAS regions pressure - that is, alleles not in Hardy-Weinberg equilibrium. for schizophrenia including RAI1. RAI1 was previously im- Subsequent mutation of key genes by CRISPR and analysis plicated in psychiatric and cognitive disorders prior to the by immunofluorescence enabled validation and characteri- psychiatric GWAS finding because this gene resides in the lo- sation of these genes. cus for the developmental syndromes, Smith–Magenis (hap- Results: Thirty mutated genes showed significant allele fre- loinsufficiency) and Potocki-Lupski (duplications). Smith– quency changes, and a some of these cognate ‘lithium re- Magenis syndrome is characterized by intellectual disabil- sponse genes’ have been subsequently validated by CRISPR- ity, behavioral abnormalities, hyperactivity, speech delay mediated mutation of cells. The identified genes include and circadian abnormalities and Potocki-Lupski syndrome is those associated with the known GSK3B/B-catenin lithium characterized by autism and intellectual disability. response pathway (BTRC, PALMD, CDH4), and those with Discussion: Capture-HiC provides important new leads in links to psychiatric disorders (EGR1, PTPRE, FAT4). Surpris- pinpointing the target genes of enhancer-mediated regu- ingly, a large number of secreted protein genes were identi- lation emanating from the GWAS findings and functional fied (CCL2, NPY, SST, SPARCL1, IGFBP3, IL32, and SEMA3A) studies confirm altered expression of interacting genes. which, given the nature of the screen, formally requires The finding of altered expression of genes in independent them to act in a local, autocrine fashion. The identifica- genome regions is an important new lead in understanding tion of mutations in extracellular/perineuronal net proteins the regulation of psychiatric disorder risk genes and syn- (VCAN, TNC, COL3A1) offers a potential solution - they are dromic developmental disorders. We are currently differen- able to create and maintain a microenvironment of se- tiating the CRISPR/Cas9 edited NPCs to neurons to examine creted proteins close to the cell surface and its receptors. the impact of the edits on differentiation and neuronal phe- VCAN is a close ortholog of NCAN, one of the top GWAS notypes. Further, to understand the role of this region in ’hits’ for bipolar disorder. Moreover, VCAN and the screen’s brain development, we have created clonal edited ES cell strongest lithium resistance mutant gene, NGS2/HMP19, are lines for the creation of cerebral organoids. co-localised in the endosomal/lysosomal pathway –a pathway vital for receptor, matrix and membrane flux, and one Disclosure: Nothing to disclose. that is known to be inhibited by lithium. Discussion: Our functional screen approach has revealed doi: 10.1016/j.euroneuro.2018.08.066 causative cellular processes in lithium response/action and we believe their role in pharmacological response in the CNS can be justifiably inferred. Importantly, within the gene set 60 there is a clear overlap with known lithium actions, and es- A FUNCTIONAL GENETIC SCREEN FOR CELLULAR tablished psychiatric risk factors –as would be hoped from

MECHANISMS OF LITHIUM RESPONSE IDENTIFIES AN any therapeutic agent. The discovery that secretion and

INTERPLAY BETWEEN PROTEIN SECRETION, EXTRA- trafﬁcking pathways are over-represented within the mu- CELLULAR MATRIX, AND ENDOSOMAL/LYSOSOMAL tant gene set deﬁnes novel targets for future experimenta- PATHWAYS tion as well as selective drug design in the context of bipolar pathology and treatment. ∗ Benjamin Pickard 1, , David Beattie 1, Emma Craib 1,

Adil Abbasi 1, Chloe Littlejohn 1, Runqi Wang 1, David Sims 2 Disclosure: Nothing to disclose.

doi: 10.1016/j.euroneuro.2018.08.067 1 University of Strathclyde Glasgow

2 University of Oxford

Background: Revealing the pathways and processes through which the mood stabiliser lithium generates a therapeutic response will enable the design of safer and more speciﬁc ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 37

Oral Session: Heritable Neuropsychiatric constructing coherent subdimensions that can be used as Traits targets of investigation in future studies. 12:30 p.m. - 2:00 p.m. Disclosure: Nothing to disclose.

61 doi: 10.1016/j.euroneuro.2018.08.068 ITEM-LEVEL STUDY OF NEUROTICISM REVEALS GE- NETIC HETEROGENEITY

∗ Mats Nagel 1, , Kyoko Watanabe 2, Sven Stringer 2, Sophie van 62 der Sluis 2, Danielle Posthuma 3 SIGNIFICANT SHARED HERITABILITY UNDERLIES SUICIDE ATTEMPT AND CLINICALLY PREDICTED PROBABILITY 1 VU Medical Centre OF ATTEMPTING SUICIDE 2 VU University

3 Center for Neurogenomics and Cognitive Research, Neuro- ∗ Douglas Ruderfer 1, , Colin Walsh 1, Matthew Aguirre 2, science Campus Amsterdam, VU University Amsterdam, VU Yosuke Tanigawa 2, Jessica Ribeiro 3, Joseph Franklin 3, University Medical Center Manuel Rivas 2

Background: Genome-wide association studies (GWAS) of 1 Vanderbilt University Medical Center psychological traits, like neuroticism and depression, are 2 Stanford University School of Medicine typically conducted on composite scores (e.g. sums of items 3 Florida State University or symptoms, case-control status). However, the items or symptoms collectively operationalizing one trait can be very diverse in nature, which would affect the statistical power Background: Suicide accounts for nearly 800,000 deaths per to detect genetic variants. The aim of the current study year worldwide with rates of both deaths and attempts ris- was to test the genetic homogeneity of items used to mea- ing. Family studies have estimated substantial heritability sure the neuroticism personality trait, and to possibly iden- of suicidal behavior; however, collecting the sample sizes tify genetically homogenous subdimensions of neuroticism, necessary for successful genetic studies has remained a which may provide a better neurobiological understanding challenge. of this trait. Methods: We utilized two different approaches in indepen-

Methods: We conducted large-scale GWAS on 12 neuroticism dent datasets to characterize the contribution of common items in two UK biobank subsamples (ﬁrst and second re- genetic variation to suicide attempt. The ﬁrst is a patient lease; total N > 366,726). We then computed inter-item ge- reported suicide attempt phenotype from genotyped sam- netic correlations within the two subsamples to assess het- ples in the UK Biobank (337,199 participants, 2,433 cases). erogeneity, and subsequently meta-analyzed both subsam- The second leveraged electronic health record (EHR) data ples to maximize statistical power for the discovery of ge- from the Vanderbilt University Medical Center (VUMC, 2.8 netic variants. million patients, 3,250 cases) and machine learning to de-

Results: The item-level GWAS in both subsamples showed rive probabilities of attempting suicide in 24,546 genotyped substantial variation in genetic signal between items. Inter- patients. item genetic correlations within the two subsamples ranged Results: We identified significant and comparable heri- from 0.38 to 0.91, revealing genetic heterogeneity in the tability estimates of suicide attempt from both the pa- = full set of items. Meta-analyses identified 255 genome- tient reported phenotype in the UK Biobank (h2SNP 0.035, = wide significant independent genomic regions, of which p 7.12x10-4) and the clinically predicted phenotype from = = 138 were item-specific. Using hierarchical clustering anal- VUMC (h2SNP 0.046, p 1.51x10-2). A significant genetic ysis, we identified two clusters of four genetically homoge- overlap was demonstrated between the two measures of neous items, labeled depressed affect and worry. Extensive suicide attempt in these independent samples through poly- = = follow-up analyses including genetic correlations to exter- genic risk score analysis (t 4.02, p 5.75x10-5) and ge- = = = nal traits, functional annotation and Mendelian randomiza- netic correlation (rg 1.073, SE 0.36, p 0.003). Finally, tion analyses, support genetic differences between items we show significant but incomplete genetic correlation of = and confirm the genetic distinctness of the clusters. In ad- suicide attempt with insomnia (rg 0.34 - 0.81) as well as = dition, the validity of the depressed affect cluster was sup- several psychiatric disorders (rg 0.26 - 0.79). ported by comparing the genetic signal to that of a not pre- Discussion: This work demonstrates the contribution of viously published depression meta-analysis. common genetic variation to suicide attempt. It points to a

Discussion: Our study demonstrates that the items used to genetic underpinning to clinically predicted risk of attempt- measure neuroticism are genetically heterogeneous. These ing suicide that is similar to the genetic proﬁle from a pa-

ﬁndings suggest that studying the genetic heterogeneity of tient reported outcome. Lastly, it presents an approach for items or symptoms used as indicators of other complex psy- using EHR data and clinical prediction to generate quanti- chological traits, like depression, is desirable. In conclusion, tative measures from binary phenotypes that can improve we argue that item- and symptom-level analyses are a use- power for genetic studies. ful strategy for studying genetic homogeneity, as well as for Disclosure: Nothing to disclose.

doi: 10.1016/j.euroneuro.2018.08.069 ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 38 Abstracts

63 64 THE EFFECT, DETECTION AND ADJUSTMENT OF SAM- GENOME-WIDE ASSOCIATION STUDY OF SCHOOL PLE OVERLAP IN PRS STUDIES GRADES INFORMS COGNITIVE GENETIC ARCHITECTURE OF SIX MAJOR PSYCHIATRIC DISORDERS ∗ Shing Wan Choi 1, , Jonathan Coleman 1, Kylie Glanville 1, ∗ Pak Sham 2, Paul O’Reilly 1 Veera Rajagopal 1, , Andrea Ganna 2,

Jonathan R.I. Coleman 3, Esben Agerbo 4, Jakob Grove 5,

1 King’s College London Thomas D. Als 5, Henriette T. Horsdal 4, Liselotte Petersen 4,

2 The University of Hong Kong iPSYCH-Broad Consortium 6, Gerome Breen 3,

Anders Børglum 5, Ditte Demontis 5 Background: Sample overlap is known to cause test statistic inabstion in polygenic risk score (PRS) studies, but the ex- 1 Aarhus University tent of the problem has not been well characterised. There 2 Center for Genomic Medicine, Massachusetts General Hos- are also presently no methods to detect or correct for the pital, Broad Institute of Harvard and MIT bias introduced by sample overlap when the overlap is par- 3 Social, Genetic and Developmental Psychiatry Centre, In- tial and the base genotypes are unavailable. Here, we pro- stitute of Psychiatry, Psychology and Neuroscience, King’s pose a novel approach to detect sample overlap between College London, National Institute for Health Research the base and target data using summary statistics, and a Biomedical Research Centre, South London and Maudsley method for correcting for the bias in the PRS results. National Health Service Trust

Methods: We first produced base and target real data, and 4 National Center for Register Based Research, Aarhus Uni- a permuted null trait, by stratifying the UK Biobank sam- versity, The Lundbeck Foundation Initiative for Integrative ples. Next, we included a varying degree of base samples in Psychiatric Research, iPSYCH, Center for Integrative Se- the target sample to generate overlap. PRS analyses were quencing then performed using the base Genome-Wide Association 5 Aarhus University, The Lundbeck Foundation Initiative for Study (GWAS) results and the target genotype data and the Integrative Psychiatric Research, iPSYCH, Center for Inte- results assessed for goodness-of-fit. In an attempt to con- grative Sequencing trol for sample overlap when detected, we contrast the LD 6 The Lundbeck Foundation Initiative for Integrative Psychi- score regression (LDSC) intercept derived from the separate atric Research, iPSYCH, Center for Integrative Sequencing, base and target GWAS summary statistics with that from the Aarhus University, Broad Institute of Harvard and MIT meta-analysed GWAS results. Results: We first present results characterizing the extent Background: Education attainment (EA) correlates strongly of the problem of sample overlap in PRS studies, finding, with psychiatric disorders—both phenotypically and geneti- for example, that 10% of overlapped sample can lead to a cally. The extent to which the genetic architecture of EA in 0.1 increase in the R2. Next, we demonstrate that we can individuals with psychiatric disorders differs from the ge- detect potentially inabsting levels of sample overlap when netic architecture of EA in the general population is not the pooled GWAS summary has an LDSC intercept inabstion known. To address this, we performed a GWAS of school higher than the sum of the LDSC intercept generated from grades in individuals with psychiatric disorders in the iPSYCH individual GWAS. By exploiting the LDSC intercept across in- cohort, the first of its kind. dividual and pooled GWAS, we can infer the level of sample Methods: We conducted GWAS of ninth level grades given at overlap between the base and target samples. Additionally, Danish and mathematics exit-examinations in the municipal for target data with large sample sizes ( > 10k), an un-biased schools in Denmark. We studied ∼35000 individuals: atten- coefficient of determination (R2) for the PRS can be calcu- tion deficit hyperactivity disorder (ADHD, N = 4298), autism lated. spectrum disorder (ASD, N = 3466), schizophrenia (SCZ, Discussion: Even a small degree of sample overlap can re- N = 1074), major depressive disorder (MDD, N = 10,140), sult in false-positive findings in PRS studies, especially if the bipolar disorder (BPD, N = 607), anorexia nervosa (AN, target data is small. We have developed an algorithm to de- N = 1798) and controls (N = 11,273, i.e., without any of the tect sample overlap between the base and target data, as above six diagnoses). School grades and psychiatric diag- well as a method based on LDSC for adjusting the bias intro- noses were extracted from Danish national registers. We duced in terms of R2, and demonstrate their performance analyzed ∼8M variants using a linear regression adjusted and utility using UK Biobank real data. for age, sex, genotyping waves and ten principal components. We analyzed all samples together (with psychiatric Disclosure: Nothing to disclose. diagnoses as a covariate) and each psychiatric group, as well ∼ doi: 10.1016/j.euroneuro.2018.08.070 as controls, separately. In addition, we analyzed 500,000 UK Biobank (UKBB) individuals to (a) replicate genome- wide significant (GWS) loci (b) to measure the variance explained by genome-wide polygenic scores (GPS) trained on our GWASs and (c) to study how GPS for school grades correlate with GPS for psychiatric disorders in the general population. For (a) and (b) the analyzed traits were college completion and verbal and numerical reasoning (VNR) scores. ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 39

Results: Gender and psychiatric diagnoses were strongly as- 65 sociated with grades: females performed better; AN scored GENOME-WIDE ANALYSIS OF INSOMNIA AND SLEEP- the highest; ADHD scored the lowest. First, we performed RELATED TRAITS IN OVER 1 MILLION INDIVIDUALS a GWAS of Danish and math (DM), totaled to a single IDENTIFIES NOVEL GENES AND PATHWAYS score per individual. We identified three GWS loci (3p21.31, ∗ 5p13.2 and 6q16.1), two were GWS in the latest EA-GWAS Philip Jansen 1, , Kyoko Watanabe 1, Sven Stringer 1, from SSGAC consortium. The SNP-heritability (h2) was 0.26 Nathan Skene 2, Julien Bryois 2, Ana Muñoz-Manchado 2, = (SE 0.01). Genetic correlation with the previous EA-GWAS Joyce Tung 3, David Hinds 3, Vladimir Vacic 3, = was 0.91 (SE 0.02). GPS for EA explained up to 7.5% vari- Patrick Sullivan 4, Tinca Polderman 1, August Smit 1, ance (adjusted R2) in DM. In the analyses stratified by psy- Jens Hjerling-Leffler 2, Eus van Someren 5, chiatric disorders, we found one MDD-specific GWS locus Danielle Posthuma 1 (7p15.2). Maximum variance explained by EA-GPS differed > between groups: controls and BPD showed 10% variance. 1 VU University Amsterdam

Second, we performed a GWAS of Danish and math sep- 2 Karolinska Institutet arately. Since Danish and math are correlated (70%), we 3 22andMe, Inc. aimed to isolate the genetic signals specific to Danish and 4 University of North Carolina at Chapel Hill math. Hence, we did Danish-GWAS with math as a covariate 5 VU GGZ InGeest/Netherlands Institute Neuroscience (DadjM) and math-GWAS with Danish as a covariate (MadjD). The h2 was ∼0.18 in all four. We observed, however, dif- Background: Insomnia, i.e. problems falling or staying ferences in (a) GWS loci and (b) genetic correlations with asleep, is highly common in the general population, with psychiatric disorders. These differences were stronger in strong negative consequences on mental and physical DadjM and MadjD. We identified two MadjD-specific GWS health, and lack of effective treatments. Moreover, sleep- loci (6p22.1 and 11q23.2). In the UKBB individuals, we found problems are highly prevalent in psychiatric disorders, sug- a strong negative correlation of GPS for all psychiatric dis- gesting a possible shared aetiology between sleep-problems orders with GPS for math, MadjD and IQ. On the other hand, and these disorders. Despite a substantial role of genetic we found a strong positive correlation (except ADHD) with factors in the aetiology of insomnia, the number of iden- GPS for EA, DM, Danish and DadjM. Phenotypes followed tified risk loci, genes and associated neurobiological path- the same trend: college completion correlated with Danish- ways remains limited. GPSs and VNR correlated with math-GPSs. Methods: Here, we perform large-scale genome-wide meta- Discussion: Our findings showed that a substantial portion analysis of insomnia in over 1 million individuals by com- of the genetic variants influencing EA phenotypes (EA, DM, bining genome-wide association (GWAS) results in the UK Danish and DadjM) shared with psychiatric disorders, are not Biobank (N = 386,533) and 23andMe, Inc. (N = 944,477), lead- cognitive—but, rather, we speculate, behavioral. ing to a total sample of 1,331,010 individuals. We perform extensive functional annotation of the genome-wide results Disclosure: Nothing to disclose. using the FUMA online annotation platform. We integrate doi: 10.1016/j.euroneuro.2018.08.071 data from tissue-expression and novel single-cell gene expression to find pathways, and cell- and tissue types related to insomnia. In addition, we analyse the genetics of six sleep-related phenotypes in UK Biobank that are known to correlate with insomnia, including sleep duration, chrono- type, snoring, narcolepsy, daytime napping and ease of get- ting up. Results: We identified 202 independent genome-wide significant loci in insomnia, including two loci on the X chromosome. Gene-based association test and gene mapping in FUMA, including positional mapping, eQTL and chromatin interactions, identified 679 genes, many of which have been previously linked to psychiatric disorders. We observed sex- specific effects, and identify genes significantly related to insomnia in males or females only. Gene-set analysis showed that these genes were highly enriched in gene-sets related to neuronal function and axonal growth. Analysis of gene-expression data showed that the identified genes were highly expressed in basal ganglia of the brain, and single-cell RNA sequencing identified several neuronal cell types, including hypothalamic neurons, known to regulate circadian rhythm, and pyramidal neurons in the claustrum, a brain structure suggested to play a role in consciousness. Genetic correlations indicated a strong genetic overlap between our GWAS results and depression, anxiety and schizophrenia, and several metabolic traits. Using Mendelian Randomiza- ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 40 Abstracts tion, we showed that insomnia has a causal effect on obe- with available external genetic resources of individuals sity and cardiovascular disease, and bidirectional effects with psychiatric diagnoses is ongoing. between insomnia and psychiatric disorders. In addition, Discussion: Genetic risk factors in our resource of individu- we find 362 loci for six sleep-related phenotypes, of which als who died by suicide may be associated with co-occurring many show overlap with insomnia. psychiatric illness, possibly reflecting extreme lethal forms Discussion: The wealth of results from this largest GWAS these illnesses. Alternatively, genetic risk for suicide may study of insomnia to date in over 1 million individuals impli- span multiple diagnostic categories. In this case, focusing cates novel genes, gene-sets and cell-types associated with on death by suicide independent of diagnosis may allow for insomnia, and provide novel targets for future research in the detection of a more generalizable genetic risk for sui- sleep disorders. We show clear genetic overlap with a wide cide that can ultimately aid prevention efforts. variety of psychiatric disorders. In addition, our results show how large-scale biobanks and big data are rapidly leading to Disclosure: Janssen Research & Development LLC - Re- new discoveries and transform the field of psychiatric ge- search, Self netics. doi: 10.1016/j.euroneuro.2018.08.073

Disclosure: Nothing to disclose. doi: 10.1016/j.euroneuro.2018.08.072 Oral Session: ADHD 12:30 p.m. - 2:00 p.m.

67 66 MULTIVARIATE GWA META-ANALYSIS IN OVER 500K ANALYSES OF DISEASE-ASSOCIATED AND LIKELY FUNC- OBSERVATIONS ON AGGRESSION AND ADHD SYMPTOMS TIONAL VARIANTS FROM PSYCHARRAY IMPLICATE GENES INVOLVED IN RISK FOR COMPLETED SUICIDE Michel Nivard ∗, Dorret Boomsma

∗ Emily DiBlasi 1, , Qingqin Li 2, John Anderson 1, VU University

W. Brandon Callor 3, Erik Christensen 3, Leslie Jerominski 1, 1 1 1 Rob Sargent , Douglas Gray , Nicola Camp , Background: We present the results of a multivariate 1 4 1 Andrey Shabalin , Anna Docherty , Hilary Coon genome-wide association (GWA) study of the developmental genetic etiology of aggression (AGG) and attention-

1 University of Utah School of Medicine deﬁcit/hyperactivity disorder (ADHD). The project involves

2 Janssen Research & Development LLC a collaboration in over 20 international cohorts from Eu-

3 Utah State Ofﬁce of the Medical Examiner rope, Australia, New-Zealand and the USA. The cohorts

4 University of Utah School of Medicine, Virginia Institute are characterized by repeated measures of Aggression and for Psychiatric & Behavioral Genetics ADHD symptoms at different ages and assessment by multiple informants and multiple instruments. In total, the meta- Background: Almost 800,000 people a year die by suicide analysis included ∼526,000 observations from over 200 GWA worldwide, and it is the second leading cause of death studies. among people under 40. This global health crisis is of crit- Methods: First, a series of univariate GWA studies was per- ical importance. Heritability of completed suicide is esti- formed for every available combination of age, informant mated at 50%, suggesting that genetics plays a significant and instrument within each cohort. This resulted in 1 to role in this extreme phenotype. To find genetic risk fac- 36 analyses per cohort, with sample sizes ranging between tors for suicide, The Utah Suicide Research Study (USRS) has 309 and 10,812. Next, results were pooled into age-by-rater collected a unique genetic resource of > 5,500 population- combinations (e.g. mother-rated aged 3-5, teacher-rated 8- ascertained DNA samples from completed suicides. This ge- 11, etc.) that resulted in an excess of 10,000 independent netic resource is obtained through a collaboration with the observations, and then meta-analyzed. Genetic correlations Utah State Office of the Medical Examiner. between the age-by-rater combinations, both within and Methods: Using cases from the USRS, we genotyped ∼1300 across AGG and ADHD, were estimated with LD Score Re- individuals with the Illumina Infinium PsychArray platform. gression. Finally, we performed a meta-regression analysis To complement traditional common variant association across all GWA studies, correcting for the fact that repeat- study approaches, we prioritized 13,000 PsychArray non- edly measured subjects were included in the analyses. We synonymous variants with documented disease associations then estimated the genetic correlation between our meta- and putative functional consequences based on SIFT and analysis and several somatic and psychiatric traits. Polyphen annotations. Single-variant and gene-based tests Results: We obtained an average SNP-heritability of 6.3% were performed using Utah completed suicide cases and Eu- and 8.3% for AGG and ADHD, respectively, across the age- ropean ancestry controls. by-rater meta-analyses. Within phenotype, the highest av- Results: Preliminary results provide support for APH1B: erage genetic correlations were seen between maternal rat- rs745918508 (p = 5.04E-29) and SUCLA2: rs121908538 ings and self-reports: 0.49 (AGG) and 0.85 (ADHD). Genetic (p = 3.14E-12) as potential variants elevating risk for com- correlations between mother and teacher ratings was 0.39 pleted suicide. To explore the specificity of genetic risk for AGG and 0.74 for ADHD. Interestingly, the genetic corre- factors with completed suicide, comparisons of this dataset lation between teacher ratings and self-report approached ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 41 zero for both traits. Averaged across raters and age, the ge- recently it was unexpectedly found that ADHD and autism netic correlation across AGG and ADHD was 0.75. have similar significant excesses of constrained rare protein We estimate several significant correlations between ei- truncating variants compared to controls, and that these ther AGG and/or ADHD and cognitive/psychiatric/health variants occur in similar sets of genes. outcomes. Most notably were the genetic correlations with Methods: Here we present results from extended analyses childhood IQ (-0.61), age at first birth (-0.58), depressive of the role of ultra-rare variants in ADHD. The study is based symptoms (0.51), smoking initiation (0.46) and parental age on whole-exome sequencing of three cohorts of ADHD cases at death (-0.36), indicating the presence of a relation be- and controls: 1) a Danish cohort with samples identified in tween AGG/ADHD and poor (health) outcomes. Follow up the Danish Newborn Screening Biobank (DNSB) comprising analysis subset the relation between childhood ADHD symp- ∼4,400 cases and ∼4,700 controls (DNSB1 sample), 2) a sam- toms and poor health outcomes may be mediated by educa- ple consisting of ∼1,100 clinically ascertained cases from tional attainment and social economic status. Germany and the Netherlands and ∼1,700 controls with Ger- Discussion: The present work is, to our knowledge, the man and Dutch ancestry (clinical sample), and finally 3) an first repeated measures multi rater genome wide asso- additional ∼4,400 cases and ∼5,300 controls from the DNSB ciation study of psychiatric phenotype. Our works helps (DNSB2 sample), which we have just received and will be further the understanding of prodromal, and subclinical included in our analyses, bringing our total sample to 9,900 psychiatric symptoms during development. SAAGY (Study cases and 11,700 controls. of Aggression and ADHD trait Genetics in Youth) includes Results: Preliminary results from meta-analyses of the multiple international cohorts (ALSPAC, CATSS, FinnTwin12, DNSB1 and the clinical samples demonstrate a signif- GenR, GINI/LISA, INMA, NFBC 1966/1986, NTR, QIMR, RAINE, icant overrepresentation of deleterious ultra-rare vari- TCHAD, TEDS, Dunedin, MOBA, MTFS, TRAILS, Young Finns ants (dURVs) in cases compared to controls (OR = 1.23; Study, ABCD, MUSP, BREATHE, E-Risk, Add Health Michigan P = 8.21x10-10) in evolutionary constrained genes intoler- State University, Understanding Society). ant to loss of function (LoF) variation (pLI > 0.9). For comparison, counts of ultra-rare synonymous variants in highly Disclosure: Nothing to disclose. constrained genes showed no difference between cases and controls (OR = 0.99, P = 0.28). When restricting to dURVs in doi: 10.1016/j.euroneuro.2018.08.074 constrained genes highly expressed in the brain, the odds ratio increased even further (OR = 1.51; P = 5,67x10-7), supporting the idea that dURVs play a considerable role in ADHD

68 risk, especially when located in genes intolerant to LoF vari- THE ROLE OF DELETERIOUS ULTRA-RARE VARIANTS IN ation. ADHD RISK Discussion: We will present updated results from analyses of the impact of dURVs on ADHD risk based on meta-analyses 1 ,∗ 2 1 Ditte Demontis , F. Kyle Satterstrom , Jinjie Duan , of DNSB1 + 2 and the clinical samples. We will especially fo- 1 3 Francesco Lescai , Søren Dinesen Østergaard , cus on the burden of these variants in ADHD in evolution- 4 5 1 Klaus-Peter Lesch , Thomas Werge , Preben Bo Mortensen , arily constrained genes as well as in speciﬁc gene sets rele- 1 6 7 Simon Glerup , Barbara Franke , David Hougaard , vant to ADHD, such as “synaptic genes”, “neurite outgrowth 8 9 10 Andreas Reif , Mark Daly , Benjamin Neale , genes”, “genes bound by the fragile X mental retardation 1 Anders Børglum gene FMRP” and “genes with high brain expression”. Fur- thermore, we will evaluate whether ADHD risk genes iden-

1 Aarhus University tiﬁed based on common variant analyses also carry an in-

2 Broad Institute creased burden of dURVs, and we will conduct gene-based

3 Aarhus University Hospital association analyses to identify the genes most frequently

4 University of Würzburg hit by dURVs in ADHD cases compared to controls.

5 Institute of Biological Psychiatry, MHC Sct. Hans, Mental Health Services - Copenhagen Disclosure: Nothing to disclose.

6 Radboud University Medical Center, Donders Institute doi: 10.1016/j.euroneuro.2018.08.075 7 Statens Serum Institut

8 University Hospital Frankfurt

9 Analytic and Translational Genetics Unit

10 Massachusetts General Hospital 69 Background: Attention-deﬁcit/hyperactivity disorder ASSOCIATION OF AGGRESSION AND ADHD SUBSCALES (ADHD) is a highly heritable childhood behavioural disorder IN CHILDREN AND ADULTS affecting 3-6% of school-age children, and has a heritabil- ∗ ity of around 0.76. The SNP heritability, estimating the Dorret Boomsma 1, , Conor Dolan 1, Meike Bartels 1, amount of risk attributed to common genetic variation, Michel Nivard 1, Toos van Beijsterveldt 1, Tom Claassen 2, has been estimated to be 0.22. The difference between Jeffrey Glennon 3 SNP heritability and total heritability indicates that the genetic risk component of ADHD also includes the effects 1 VU University of rare variants. The role of ultra-rare deleterious variants 2 Radboud University Nijmegen has been established for schizophrenia and autism, and 3 Radboud University, Nijmegen Medical Centre ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 42 Abstracts

Background: Based on Bayesian machine learning analysis 2 Vall d’Hebron Research Institute (VHIR), Universitat performed in the MATRICS (Multidisciplinary Approaches to Autònoma de Barcelona

Translational Research In Conduct Syndromes) consortium, 3 Institut de Recerca Hospital University Vall d’Hebron Re- in clinical adolescent ADHD and population cohorts, dif- search Institute ferential associations between aggression and hyperactivity 4 Vall d’Hebron Research Institute (VHIR) and between aggression and inattention were suggested. 5 Aarhus University

Methods: We aimed to replicate these ﬁndings in the AC- 6 Radboud University Medical Center, Donders Institute

TION (Aggression in Children: Unraveling gene-environment 7 Medical Genetics Research Center, SUNY Upstate Medical interplay to inform Treatment and InterventiON strategies) University consortium, using the database from the large population- 8 University of Bergen based Netherlands Tw i n Register. Analyses were done for 9 University Hospital of Frankfurt data obtained during in childhood (age 7-16 years) and 10 UFRGS adulthood, employing both cross-sectional and longitudi- 11 Hospital Universitari Vall d’Hebron nal regression analyses. In children and in adults, outcome 12 Hospital Universitari Vall d’Hebron, Centro de Investi- and predictor variables were assessed by comparable instru- gación Biomédica en Red de Salud Mental (CIBERSAM), Insti- ments. Aggression was assessed by the Achenbach System of tuto de Salud Carlos III, Mental Health and Addiction, Vall Empirically Based Assessment (ASEBA) age-appropriate in- d’Hebron Research Institute (VHIR), Universitat Autònoma ventories, namely the Child behavior Check List (CBCL), and de Barcelona the Youth or the Adult Self Report (YSR / ASR). Hyperactivity and inattention were assessed by the Conners’ Parent Rat- Background: Attention-deficit/hyperactivity disorder ing Scale-Revised: Short version (CPRS-R:S) and the Conners’ (ADHD) is a highly heritable neurodevelopmental disorder Adult ADHD Rating Scales (CAARS). The data were analyzed whose impairing symptoms persist into adulthood in around in children and adult by linear regression and multivariate 65% of the diagnosed children. Being able to predict ADHD genetic structural equation modeling. persistence in childhood based on genetic load could be a Results: Based on linear regression analyses in which hyper- first step towards prevention of ADHD persistence, which activity and inattention predicted aggression, we observed is a key clinical concern. The aim of the present study different results in children and in adults. In children, hy- was to compare the genetic background of children and peractivity was a stronger predictor of aggression than inat- adult ADHD patients and to elucidate whether a subset of tention. However, in adults, inattention which tends to show children with ADHD exists which is more genetically similar stronger persistence into adulthood than hyperactivity, was to adult patients, as well as identifying new loci associated the stronger predictor. Multivariate genetic structural equa- with ADHD. tion modeling in twin families confirmed that in children hy- Methods: Genomic data from the International Multi-centre peractivity was the stronger predictor for aggression, with persistent ADHD CollaboraTion (IMpACT) and the Lundbeck the predictive power mainly due to the genetic associations. Foundation Initiative for Integrative Psychiatric Research In adults, predictive power of Inattention also was mainly (iPSYCH) were used to conduct a genome-wide associa- due to genetic associations. tion (GWA) meta-analysis of 6,619 adult ADHD patients and Discussion: We obtained empirical evidence in indepen- 15,976 controls. A meta-analysis of all childhood ADHD dent samples for differential associations between aggres- GWAS data available at the time, within the Psychiatric Ge- sion and hyperactivity and between aggression and inatten- nomics Consortium (PGC) and iPSYCH, was also conducted. tion. We estimated the genetic correlation between children and adult ADHD and also between adult ADHD and other traits Disclosure: Nothing to disclose. using LD-score regression. Moreover, we examined whether doi: 10.1016/j.euroneuro.2018.08.076 a subgroup of children was more genetically similar to adult patients, who could therefore be hypothesized to be at risk of persistent ADHD, using the well-powered BUHMBOX method. In addition, we ran a combined children and adult 70 meta-analysis of 17,236 patients and 32,513 controls. GENETIC INFLUENCES CONTRIBUTING TO ATTENTION- Results: The top hits of the adult meta-analysis (P < 2.92E-

DEFICIT/HYPERACTIVITY DISORDER ACROSS THE LIFES- 07) included genes previously associated with neuronal mi-

PAN: EVIDENCE FROM GENOME-WIDE ASSOCIATION gration. We observed significant genetic correlation be- STUDIES tween adult ADHD and major depressive disorder (P = 8.01E- 05), neuroticism (P = 5.73E-05), risk taking (P = 2.57E-16), 1, ∗ 2 Paula Rovira , María Soler Artigas , years of schooling (P = 1.76E-26) and intelligence (P = 6.91E- 3 4 Cristina Sanchez-Mora , Iris Garcia-Martínez , 13), among others. A strong correlation between genetic 4 5 5 Mireia Pagerols , Ditte Demontis , Anders Børglum , variants contributing to adult and childhood ADHD (rg = 0.81 6 7 8 Barbara Franke , Stephen V. Faraone , Jan Haavik , (S.E. = 0.08); P = 5.10E-21) was found and no evidence for 9 10 11 Andreas Reif , Claiton Bau , Miguel Casas , genetically different subgroups of children with ADHD was 12 4 Josep Antoni Ramos-Quiroga , Marta Ribasés detected. In the combined meta-analysis of adults and children, we identified three new genome-wide significant loci,

1 Vall d’Hebron Research Institute (VHIR), Universitat present near genes related to neuronal migration and to Autònoma de Barcelona, Mental Health Hospital Universi- monoamine and neurotrophin neurotransmission. tari Vall d’Hebron ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] Abstracts 43

Discussion: The results support the hypothesis of a shared Discussion: These results are in line with the temporal rela- genetic background between children and adult ADHD and tionship between ADHD and future cannabis use, reinforce point to new genetic variants associated with the disor- the need to consider substance misuse in the context of der. Prospective studies exploring the genetic background ADHD in clinical intervention, and highlight the need for fu- of children with ADHD will allow us to elucidate the genetic ture genetic studies to provide insight into the shared bio- impact on the persistence and remittance of ADHD symp- logical mechanisms underlying both conditions. toms. Disclosure: Nothing to disclose. Disclosure: Nothing to disclose. doi: 10.1016/j.euroneuro.2018.08.078 doi: 10.1016/j.euroneuro.2018.08.077

72 71 GENOMIC SEM: A POWERFUL FRAMEWORK TO IN- ATTENTION-DEFICIT/HYPERACTIVITY DISORDER AND TERROGATE THE CAUSES AND CONSEQUENCES OF LIFETIME CANNABIS USE: GENETIC OVERLAP AND PSYCHIATRIC DISEASE CAUSALITY ∗ Michel Nivard 1, , Andrew Grotzinger 2, Mijke Rhemtulla 3, ∗ María Soler Artigas 1, , Cristina Sánchez-Mora 2, Ronald de Vlaming 1, Stuart Ritchie 4, Travis Mallard 2,

Paula Rovira 2, Vanesa Richarte 2, Iris García-Martínez 2, W David Hill 4, Hill Fung Ip 1, Andrew McIntosh 4,

Mireia Pagerols 2, ADHD Group 3, International Cannabis Ian J. Deary 4, Dorret Boomsma 1, Philipp Koellinger 1,

Consortium, Miguel Casas 2, Josep Antoni Ramos-Quiroga 2, Paige Harden 2, Elliot Tucker-Drob 2

Marta Ribasés 2

1 VU University

1 Vall d’Hebron Research Institute (VHIR), Universitat 2 The University of Texas at Austin

Autònoma de Barcelona 3 University of California, Davis

2 Hospital Universitari Vall d’Hebron 4 University of Edinburgh

3 Psychiatric Genomics Consortium Background: Methods which leverage GWAS summary statis- Background: Attention-deficit/hyperactivity disorder tics to estimate genetic correlations between pairwise com- (ADHD) is a severely impairing neurodevelopmental disor- binations of traits have produced ’atlases’ of genetic ar- der where comorbid conditions play a key role in symptom chitecture. Genetic atlases reveal pervasive pleiotropy, and progression, disorder course and outcome. ADHD is associ- genome-wide significant loci are often shared across phe- ated with a significantly increased risk for substance use, notypes. Especially psychiatric disorders, personality and abuse and dependence. ADHD and cannabis use are partly behavioral phenotypes exhibit strong genetic correlations. determined by genetic factors; the heritability of ADHD We introduce genomic structural equation modeling (Ge- is estimated at 70-80% and of cannabis use initiation at nomic SEM), a multivariate method for analyzing the joint 40-48%. In this study we aimed to gain insights into the genetic architectures of complex traits. Using formal meth- genetic overlap and causal relationship of these two traits. ods for modeling covariance structure, Genomic SEM syn- Methods: We used summary statistics from the largest thesizes genetic correlations and SNP-heritabilities inferred available meta-analyses of genome-wide association stud- from GWAS summary statistics of individual traits from sam- ies (GWAS) of ADHD (n = 53 293) and lifetime cannabis use ples with varying and unknown degrees of overlap. (n = 32 330) to run LD score regression and estimate their Methods: We use genomic structural equation modeling to genetic correlation, undertake a cross-trait analysis to iden- study whether the relationship between attention deficit tify new loci and use a two-sample Mendelian randomization hyperactivity disorder (ADHD) and unhealthy or risky behav- approach to infer a causal relationship between these traits. iors (smoking behaviors, drinking, speeding, eating related Results: We estimated a genetic correlation of r2 = 0.29 behaviors) is causal and whether the causal relation is medi- (P = 1.63x10-5) and identified four new genome-wide sig- ated by educational attainment. (EA) and income. Our anal- nificant loci: two in a single variant association analy- ysis is based on the latest GWAS and our technique is robust sis (rs145108385, P = 3.30x10-8 and rs4259397, P = 4.52x10- for the fact that these GWAS have considerable sample over- 8) and two in a gene-based association analysis (WDPCP, lap. We further explore whether ADHD liability is a common P = 9.67x10-7 and ZNF251, P = 1.62x10-6). We found support cause which contribute between the strong genetic rela- that ADHD is causal for lifetime cannabis use, with an odds tionship between EA and mortality. To test the causal direc- ratio of 7.9 for cannabis use in individuals with ADHD in com- tions implied in these models we introduce genetic instru- parison to individuals without ADHD (95% CI (3.72, 15.51), ments (both individual SNPs, and Inferred effects of gene P = 5.88x10-5). expression from SMR/TWAS). ARTICLE IN PRESS JID: NEUPSY [m6+; October 3, 2018;8:33 ] 44 Abstracts

Results: Our results imply a causal effect of ADHD on un- Discussion: The ability to leverage genetics to identify mod- healthy or risky behaviors and that the relationship is me- erators on the causal path from psychiatric disease to ad- diated by EA, suggesting an educational intervention may verse outcome and physical health can enable the develop- reduce the impact of ADHD on physical health. We con- ment of preventive treatments and program which dampen trast and compare our ﬁndings with those based on pairwise the adverse effects which follow from ADHD. and multivariable Mendelian randomization (MR), randomized trials, and natural experiments. Results obtained with Disclosure: Nothing to disclose. genomic SEM are largely consistent with results obtained us- doi: 10.1016/j.euroneuro.2018.08.079 ing MR, randomized trials and natural experiments.