Genetics for Imagers How Geneticists Model Quantitative Phenotypes

Genetics for Imagers: How Geneticists Model Quantitative Phenotypes Nelson Freimer UCLA Center for Neurobehavioral Genetics

What makes a genetic association significant?

Outline • The problem of achieving validated findings in psychiatric genetics • Approaches to genetic mapping and statistical significance - linkage analysis (+ examples) - association analysis (+ examples

Psychiatric genetics: The brains of the family 10 July 2008 | Nature 454, 154 -157 (2008) Does the difficulty in finding the genes responsible for mental illness reflect the complexity of the genetics or the poor definitions of psychiatric disorders?

“The studies so far are statistically underpowered. We need bigger studies. ” — Jonathan Flint

“Geneticists know nothing about psychiatric disease. ” — Daniel Weinberger

WHAT IS THE PROBLEM? • Psychiatric disorders are highly heritable • No psychiatric susceptibility genes known • Studies so far are underpowered – Phenotypes are of uncertain validity – Samples are too small and markers too few – Signal to noise ratio is too low (etiological heterogeneity: genetic and non-genetic)

We are just too ignorant of the underlying neurobiology to make guesses about candidate genes. ” —Steven Hyman “

This is why geneticists have turned to genome wide mapping

Genome-wide mapping and allelic architecture

Effect Size Large Allelic architecture and genetic mapping approaches NOT FOUND TO DATE LINKAGE Family-based Case-control Small OR COPY NUMBER VARIANTS Rare (<1%) ASSOCIATION Common (>5%) Disease Gene Allele Frequency

Founder Disease Gene IBD Region Present-day affected individuals Shared IBD Region IBD= Identical By Descent

The Principle of Genetic Linkage If genes are located on different chromosomes they show independent assortment. compute this probability. However, genes on the same chromosome, especially if they are close to each other, tend to be passed onto their offspring in the same configuration as on the parental chromosomes.

Genetic markers: SNPs

Detecting Genetic Linkage: Linkage Analysis vs Association Analysis • Linkage Analysis – Using pedigree samples, search for regions of the genome where affected individuals share alleles more than you would expect • Association Analysis – Compare allele frequency distributions in cases and controls • For quantitative traits can apply similar principles

Linkage Analysis G, T T, T Association Analysis T, T G, T G, T T, T

When are two genetic loci significantly linked?

Stringent significance thresholds based on… • Low prior probability of linkage between any two loci – Considered when there were few markers • Multiple tests involved in genotyping studies – Considered after there were many markers • Both considerations yielded ~ same threshold: LOD score (log. base 10 of the likelihood ratio) >~ 3 (i. e. p < 10 -4)

• Prior probability of linkage between a given locus and a random genome location: 0. 02 • To obtain posterior probability of linkage of >0. 95 (i. e. <0. 05 false positive linkages), apply Bayes theorem: • Solving for the likelihood ratio Pr(Data | Linkage) / Pr(Data | No. Linkage)… – ratio must be >1, 000, i. e. LOD >3

Controlling for multiple testing in linkage • With complete genome marker sets, prior probability that some marker linked is 1 • ~500 fully informative, independent markers cover linkage in all regions of the genome • To control at 0. 05 level, the global hypothesis of no linkage anywhere in the genome: 0. 05/500 = 10 -4 for each test, i. e. LOD >3

Significance thresholds for linkage Lander and Kruglyak, 1996 • Suggestive linkage: a lod score or p value expected to occur once by chance in a whole genome scan. LOD >2. 2, p < 7. 4 x 10 -4 • Significant linkage: a lod score or p value expected to occur by chance 0. 05 times in a whole genome scan LOD >3. 6, p < 2. 2 x 10 -5 • Highly significant linkage: a lod score or p value expected to occur by chance 0. 001 times in a whole genome scan. LOD > 5. 4, p < 3 x 10 -7 • Confirmed linkage - a significant linkage observed in one study is confirmed by finding a lod score or p value expected to occur 0. 01 times by chance in a specific search of the candidate region.

An example of linkage to a quantitative neurobehavioral trait

Monoamine Neurotransmitters Norepinephrine and epinephrine Attention Blood pressure Histamine Dopamine Reward Serotonin Appetite, Mood Gastrointestinal motility Gastric acid release Immune response From David Krantz

Catecholamine Synthesis and Degradation

Genome wide linkage analysis of HVA in a vervet monkey pedigree

Vervet research colony pedigree

Heritability of Monoamine Metabolites in vervet monkeys

HVA level in Vervets on Chromosome 10

Linkage analysis in extended pedigrees may be powerful for structural MRI phenotypes

Brain MRIs in the VRC 357 Vervets scanned Mobile Siemens Symphony 1. 5 Tesla scanner

Genetic association analysis Linkage analysis is not very powerful for mapping complex traits (with many alleles of small effect)

Effect Size Large Disease gene discovery methods NOT FOUND TO DATE LINKAGE Family-based Case-control Small OR COPY NUMBER VARIANTS Rare (<1%) ASSOCIATION Common (>5%) Disease Gene Allele Frequency

Linkage Analysis G, T T, T Association Analysis T, T G, T G, T T, T

Significance thresholds for association Consider simple Bayesian argument: - Prior probability that a random gene associated with trait: ~1/30, 000, assuming 30, 000 genes/genome - Likelihood ratio should be > 550, 000 for association to be significant (posterior probability >0. 95) - With χ2 test, p< 2. 6 x 10 -7

A more complete evaluation of significance Posterior odds (for true association) = Prior odds x Power Significance • Strength of evidence depends on likely number of true associations and power to detect them • These depend on effect sizes and sample sizes • Less well-powered studies need more stringent thresholds to control false-positive rate See Wacholder et al. , J. National Cancer Institute 2004

Genome wide association thresholds • Controlling for multiple testing E. g. Bonferroni: 0. 05 x No. of SNPs x No. of traits E. g. For single trait with 106 SNPs, p < 5 x 10 -8 • However, more complicated… – SNPs are not all independent (LD) – LD varies across genome and populations – traits are not all independent • False discovery rate (FDR) increasingly used (proportion of false positives among all positives) …if 1 out of 20 hits are false not so bad

Evaluating association in neurobehavioral genetics studies

Monoamine Neurotransmitters Norepinephrine and epinephrine Attention Blood pressure Histamine Dopamine Reward Serotonin Appetite, Mood Gastrointestinal motility Gastric acid release Immune response From David Krantz

Serotonin Transporter Promoter Polymorphism Association Studies as of 2002 Phenotype P<. 05 P>. 05 Schizo. 2 7 BP/mood disorder 8 13 OCD 2 2 Personality traits 12 10 Drug response 3 0 Suicide 4 1 Anorexia 0 2 Late Onset Alzheimer’s 2 2 Smoking related 4 1 Alcohol related 5 2 Autism 2 2 Fibromyalgia 1 0 Panic disorder 0 3

Association of Anxiety-Related Traits with Polymorphism in the Serotonin Transporter Gene Regulatory Region Lesch et al. Science. 1996; 274(5292): 1527 -31. • Two samples (N = 221, N = 284) • Association with P ~ 0. 02

A more complete evaluation of significance Posterior odds (for true association) = Prior odds x Power Significance • Strength of evidence depends on likely number of true associations and power to detect them • These depend on effect sizes and sample sizes • Less well-powered studies need more stringent thresholds to control false-positive rate See Wacholder et al. , J. National Cancer Institute 2004

In large samples: No association of 5 HTTLPR with temperament Example from Northern Finland Birth Cohort, N ~ 4000

Influence of Life Stress on Depression: Moderation by a Polymorphism in the 5 -HTT Gene Caspi et al. Science 301: 386 – 389 2003

Interaction Between the Serotonin Transporter Gene (5 -HTTLPR), Stressful Life Events, and Risk of Depression: A Meta-analysis Risch et al. JAMA. 2009; 301(23): 2462 -2471.

Logistic Regression Analyses of Risk of Depression for 14 Studies Copyright restrictions may apply.

Genomewide association analysis

Progress in identifying gene variants for common traits Cholesterol Obesity Myocardial infarction QT interval Atrial Fibrilliation Type 2 Diabetes Prostate cancer Breast cancer Colon cancer height PPAR IBD 5 NOD 2 Age Related Macular Degeneration Crohns Disease Type 1 Diabetes Systemic Lupus Erythematosus Asthma Restless leg syndrome Gallstone disease Multiple sclerosis Rheumatoid arthritis NOS 1 AP Glaucoma IFIH 1 CTLA 4 CD 25 IRF 5 PCSK 9 KCNJ 11 PTPN 22 CFH 2000 2001 2002 2003 2004 2005 PCSK 9 CFB/C 2 LOC 3877 15 8 q 24 IL 23 R TCF 7 L 2 2006 CDKN 2 B/ A 8 q 24 #2 8 q 24 #3 8 q 24 #4 8 q 24 #5 8 q 24 #6 ATG 16 L 1 5 p 13 10 q 21 IRGM NKX 2 -3 IL 12 B 3 p 21 1 q 24 PTPN 2 TCF 2 CDKN 2 B/ A IGF 2 BP 2 CDKAL 1 HHEX SLC 30 A 8 Slide from David Altshuler MEIS 1 HMGA 2 LBXCOR GDF 5 UQCC 1 BTBD 9 HMPG JAZF 1 C 3 8 q 24 CDC 123 ORMDL 3 ADAMTS 4 q 25 9 TCF 2 THADA GCKR WSF 1 FTO LOXL 1 C 12 orf 30 IL 7 R ERBB 3 TRAF 1/C KIAA 035 5 STAT 4 0 CD 226 ABCG 8 16 p 13 GALNT 2 PTPN 2 PSRC 1 SH 2 B 3 NCAN FGFR 2 TBL 2 TNRC 9 TRIB 1 MAP 3 K 1 KCTD 10 LSP 1 ANGLPT 3 8 q 24 GRIN 3 A 2007 51

HDL Association at 16 q 22. 1

HDL Association near LIPC

Progress in identifying gene variants for common traits Cholesterol Obesity Myocardial infarction QT interval Atrial Fibrilliation Type 2 Diabetes Prostate cancer Breast cancer Colon cancer height PPAR IBD 5 NOD 2 Age Related Macular Degeneration Crohns Disease Type 1 Diabetes Systemic Lupus Erythematosus Asthma Restless leg syndrome Gallstone disease Multiple sclerosis Rheumatoid arthritis NOS 1 AP Glaucoma IFIH 1 CTLA 4 CD 25 IRF 5 PCSK 9 KCNJ 11 PTPN 22 CFH 2000 2001 2002 2003 2004 2005 PCSK 9 CFB/C 2 LOC 3877 15 8 q 24 IL 23 R TCF 7 L 2 2006 CDKN 2 B/ A 8 q 24 #2 8 q 24 #3 8 q 24 #4 8 q 24 #5 8 q 24 #6 ATG 16 L 1 5 p 13 10 q 21 IRGM NKX 2 -3 IL 12 B 3 p 21 1 q 24 PTPN 2 TCF 2 CDKN 2 B/ A IGF 2 BP 2 CDKAL 1 HHEX SLC 30 A 8 Slide from David Altshuler MEIS 1 HMGA 2 LBXCOR GDF 5 UQCC 1 BTBD 9 HMPG JAZF 1 C 3 8 q 24 CDC 123 ORMDL 3 ADAMTS 4 q 25 9 TCF 2 THADA GCKR WSF 1 FTO LOXL 1 C 12 orf 30 IL 7 R ERBB 3 TRAF 1/C KIAA 035 5 STAT 4 0 CD 226 ABCG 8 16 p 13 GALNT 2 PTPN 2 PSRC 1 SH 2 B 3 NCAN FGFR 2 TBL 2 TNRC 9 TRIB 1 MAP 3 K 1 KCTD 10 LSP 1 ANGLPT 3 8 q 24 GRIN 3 A 2007 55

A success story in neuropsychiatry

Genome Wide association in narcolepsy in Japan (222 cases vs 389 controls) -log 10 (P value) 8 HLA 6 4 2 Chr 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 From Emmanuel Mignot

J. Hallmayer et al. Nature Genetics 41, 708 - 711 (2009) Narcolepsy is strongly associated with the T-cell receptor alpha locus 2000 cases in GWAS + ~2000 cases in replication ~

Strong genome-wide evidence

Known genes and environment explain little of trait variance

Sequencing: the currently unexplored middle of the allelic spectrum

Whole genome sequencing is coming soon… But we don’t have very good models for it yet

Summary • The allelic spectrum of complex traits determines the appropriate genetic mapping approach • Genetic linkage and association studies require stringent statistical thresholds • Single candidate gene studies have very low probability of being true positives • Genome-wide linkage and association studies are beginning to bear fruit for neurobehavioral traits • Whole-genome sequencing is just around the corner