CENTER FOR HUMAN GENETIC RESEARCH The Burden of Mental Illness Imaging :! Decoding ! More disability than any other medical disorders:

Costs: $317.6 billion Lethal consequences: Jordan W. Smoller, MD, ScD • Life expectancy of patients with chronic mental illness is shortened by an average of 25 years Department of and Center for Human Genetic Research • Suicide: 4th leading cause of death--ahead of diabetes, stroke and Massachusetts General Hospital chronic lung disease

Genetic Epidemiology of Selected Disorders The Usual Suspects Disorder Familial Relative Risk Heritability Autism 25-50 60-90% Schizophrenia 10 85% Bipolar Disorder 7-10 85% ADHD 2-6 77% Alcohol/Drug 3-8 55% Eating Disorders 10 55% OCD 4-10 30-50% Anxiety Disorders 5 40% Depression 3 40%

Serotonin Dopamine Norepinephrine GABA Neuropeptide Genes Genes Genes Genes Genes

1 A New Era in

Low prior probability of association and low power Before 2008: virtually no specific genetic risk factors identified for common psychiatric disorders Simulation of COMT candidate gene studies: - 10 polymorphisms, 500 cases and 500 controls - 97% of simulated case-control analyses produce association at 0.05 level even though Confirmed Psychiatric Risk Variants data randomly generated 250

200

150

100

50

Psychological Science 2012 0 3 independent samples 1996-2003 2004-2007 2008-2011 2012 2013-2014 Combined N ~10,000

The Book of Phenotypes

N = 36,989 cases

2 Psychiatric Disorders: The View Psychiatry’s DSM: From Genetic Epidemiology Specific Phobias 600 The Phenotype Problem Bulimia Social Phobia Obsessive - PTSD 450 Compulsive Panic Disorder Generalized Disorder 300 Alcoholism Anxiety Disorder Major Tourette Depression Syndrome Bipolar 150 Disorder Schizoaffective 1952 1968 1980 1987 1994 2013 Disorder ADHD 0 I II III III-R IV 5 Schizophrenia Autism Spectrum

From DSM to DNA: N > 60,000 Shared Genetic Risk Factors

CACNA1C CCDC68 NT5C2 ITIH3+

Autism .16

Schizophrenia

.68 Schizophrenia Autism ADHD Genetic Depression Bipolar .43 Bipolar Correlations

.47

Depression

.37 CNNM2 AS3MT+ PCGEM1 TCF4 ADHD

3 Common Findings in of But is it Heritable? Twin Studies Psychopathology STRUCTURAL

Disorder Structural Functional Phenotype Heritability Mean cortical thickness 65%-82% • éearly TBV • Atypical connectivity of multiple networks Autism Spectrum • êcorpus callosum volume Anterior cortical thickness 80%-100% • Delayed cortical surface • Disturbed function in attention and inhibitory Corpus callosum white matter density 80% area development networks Total cortical unmyelinated white matter volume 85% ADHD • Frontal cortical thinning • Reduced reward sensitivity • Widespread êWM integrity • Atypical DMN and PFC/Nac connectivity Intracranial volume 71%-91% • Ventricular enlargement • éDMN connectivity; Amygdala volume 49-83% • Accelerated cortical GM loss • êfronto-parietal connectivity; Lateral ventricle volume 31%-92% Schizophrenia • Widespread êWM integrity • altered DLPFC activation (working memory tasks) Caudate volume 90% Accumbens 49% • Widespread êWM integrity • Altered cortico-limbic connectivity Bipolar Disorder • édeep WM hyperintensities Corpus callosum volume 85%-92% • êhippocampal volume • éamygdala & dACC reactivity to negative stimuli Hippocampal volume 77-79% • êrACC activation to emotional stimuli Depression Thalamus volume 74-88% • êventral striatal activation to positive stimuli White matter integrity (various regions) 40-80% Anxiety • Altered amygdala volume • éamygdala, insula, and dACC reactivity to • Reduced amygdala-PFC threat; FUNCTIONAL Disorders WM integrity Working memory task-related activation ~40%-65% • éamygdala volume • éamygdala reactivity to threat; Resting state functional local connectivity 46-89% PTSD • êhippocampal volume • Altered ACC activation to trauma stimuli Resting state functional global connectivity 37-62%

Two General Approaches Two General Approaches

• Discovery: Identify loci that influence • Discovery: Identify loci that influence neuroimaging endophenotypes neuroimaging endophenotypes • Characterization: Characterize neural • Characterization: Characterize neural expression of established genetic risk expression of established genetic risk factors factors

4 Endophenotypes • Neurocognitive and Neural Phenotypes are Ø linked to major neuropsychiatric disorders Ø highly heritable Ø more direct expression of gene effects?

Affected

Unaffected

Genes/ Biological Brain Behavior/ Psychiatric Proteins Processes Systems Cognition Illness

Reduced anterior cingulate cortex - amygdala connectivity in “S” carriers Pezawas et al. Nature Neurosci, 2005

Bias: Winner’s Curse, Selective Reporting, Data-dredging, etc

Median power: Neuroimaging studies: 8% Animal studies: 31%

• Modest but significant effect • Larger when only published studies considered Smaller sample size è more significant foci • Significant heterogeneity of effects across studies • Variance explained ~1% compared to ~30% in original studies • No study was adequately powered to detect this effect

2-4 fold excess of significant findings

5 Genomewide Approaches How to Achieve Power? • Consortia and meta-analysis • Do not rely on prior genetic hypotheses • Example: ENIGMA • Proven success for complex traits – 50 cohorts worldwide, 125 institutions • But… – N ~30,000 – Typical effect sizes: allelic OR < 1.2 – Harmonized image analysis and QC – Multiple testing requires stringent – http://enigma.ini.usc.edu correction (typically p < 5 x 10-8) – Most imaging genetic studies vastly underpowered

Implications

• Common genetic variants underlying brain structural phenotypes can be found • Neural “endophenotypes” may not be less complex than disorder • Usual suspects nowhere to be seen

• iHippocampal volume in MDD, PTSD,BPD, SCZ

• ENIGMA consortium (N = 7794; replication including CHARGE ~10K): • Hippocampal volume 12q24.22; P=2.56x10-17 • Intracranial volume: 12q13.3, P=3.45x10-11

6 Another Challenge: What’s the Phenotype? • Prior imaging/disease studies provide leads Endophenotype ranking value: • But endophenotype concept implies these relationships must be genetically Generate genetic covariance between mediated disorder and endophenotype using: • But, the search space could be huge • Square root of disorder heritability • Square root of endophenotype heritability • How to select high-yield brain • Genetic correlation between them

phenotypes? Derived from pedigree or twin data • Need screening methods

Genomic Complex Trait Analysis Massively Expedited Genomewide (GCTA) Heritability Analysis (MEGHA) • Estimate heritability due to common variants directly from genotypes (“SNP-chip heritability”) • Problem: massive number of potential • Measure genetic similarity of unrelated phenotypes individuals and its linear relationship to • GCTA unfeasible for screening phenotype similarity • MEGHA (Ge et al.) uses same GRM but • Genetic relationship matrix (GRM) computes efficient variance component • Estimate heritability using residual maximum score test (kernel machine methods) likelihood analysis from a linear mixed model • Larger sample sizes reduce standard error • Can compute permuted p values • Rank putative endophenotypes Poster Talk (Group 1): Ge et al. Fast Heritability Analysis Using Genome-Wide Data via Kernel Machines

7 Two General Approaches

• Discovery: Identify loci that influence neuroimaging endophenotypes • Characterization: Characterize neural expression of established genetic risk factors Affected

Unaffected

Genes/ Biological Brain Systems Behavior/ Psychiatric Proteins Processes Cognition Illness

Neuroimaging Dissection of Established Risk Variants Polygene Risk Scores • For polygenic phenotypes, many loci of modest effect will not reach GWS • Can capture polygenic effects in single score • Derive in discovery sample and apply in test sample

PRS = Σj=1 xj * log(ORij),

where xj = number of risk alleles (0, 1, or 2) ith individual carries at the jth SNP; ORij is the allelic odds ratio for jth SNP for individual I

8 Genetic Risk Profiles Predict MGH Brain Genomics Superstruct Schizophrenia With Randy Buckner, Josh Roffman, MGH Psychiatry, MIT, McLean Rapid Acquisition Imaging on Multiple Matched Scanners Structural Functional

Web-based cognitive and behavioral tests

GWAS/Exome Analysis

> 4000 participants PGC-SCZ, Nature, 2014 White Matter Functional Connectivity

Dimensions of Social Cognition

Positive correlation with amygdala volume

Inverse association with rACC thickness Emotion perception

Strongest in those with high negative affectivity

Holmes, Lee et al. J Neurosci, in press

9 Polygenic Score for Depression Summary is Associated with ACC thickness • Psychiatric disorders are complex and highly polygenic • Completed GWAS of 470 individuals • A growing catalogue of risk variants – but disorders are • Calculated polygene score using data from Psychiatric GWAS Consortium (N = 61,220) clinical constructs with fuzzy boundaries and mechanisms • Question: Are polygenic influences on MDD associated with rACC from geneèbrainèillness are poorly understood thickness • Imaging genetics offers crucial tool for addressing this • Two approaches: – Discovery: need large samples – Characterization • Directions and Gaps: – Increasing Power: larger samples, consortia (e.g. PGC-ENIGMA) – Methods for addressing high dimensionality

Holmes, Lee et al. J Neurosci, 2012

Acknowledgements

165 scientists from 68 institutions in 19 countries

Workgroup Chair(s) PGC Cross-Disorder Genetics Group: Jordan Smoller, Ken Kendler, Nick Craddock, Shaun Purcell, Ben Neale, Mike Neale, Pat Sullivan, Marcella Rietschel, John Nurnberger, Roy SCZ Mick O’Donovan Perlis, Thomas Schulze, Anita Thapar, Susan Santangelo

PGC Cross-Disorder Genetics BPD Pamela Sklar/John Kelsoe Analyses: Stephan Ripke, Ben Neale, Phil Lee, Ken Kendler, Shaun Purcell, Mark Daly

MDD Patrick Sullivan SNP-based genetic correlation analyses: S. Hong Lee, Naomi Wray ASD Mark Daly/Bernie Devlin ADHD Steve Faraone Brain Genomic Superstruct: Randy FundingAnalysis Mark Daly Buckner, Josh Roffman, Phil Lee, Avram Holmes, Marisa Hollingshead, Mert NationalCross-Disorder Institute of MentalJordan Health Smoller/Ken U01 MH085520 Kendler Stanley Center for Psychiatric Research, Broad Sebancu,, others SpecialCNV thanks to: Jonathan Sebat Thomas Lehner

10