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

2. What makes a genetic association significant?

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

4. Psychiatric genetics: The brains of the family
10 July 2008 | Nature 454, (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?

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

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

7. Psychiatric disorders are highly heritable
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)

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

9. This is why geneticists have turned to genome wide mapping

10. Genome-wide mapping and allelic architecture

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

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

13. 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.

14. Genetic markers: SNPs

15. Detecting Genetic Linkage: Linkage Analysis vs Association 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

16. Linkage Analysis Association Analysis G,T T,T T,T G,T G,T T,T G,T G,T

18. When are two genetic loci significantly linked?

19. 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)

20. 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 | NoLinkage)…
ratio must be >1,000, i.e. LOD >3

21. 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

22. 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 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.

23. An example of linkage to a quantitative neurobehavioral trait

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

25. Catecholamine Synthesis and Degradation

26. Genome wide linkage analysis of HVA in a vervet monkey pedigree

27. Vervet research colony pedigree

28. Heritability of Monoamine Metabolites in vervet monkeys

30. HVA level in Vervets on Chromosome 10

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

32. Mobile Siemens Symphony
Brain MRIs in the VRC
357 Vervets scanned
Mobile Siemens Symphony
1.5 Tesla scanner

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

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

36. Linkage Analysis Association Analysis G,T T,T T,T G,T G,T T,T G,T G,T

37. 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

38. A more complete evaluation of significance
Posterior odds = Prior odds x Power
(for true association) 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

39. 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 x10-8
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

40. Evaluating association in neurobehavioral genetics studies

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

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

44. Lesch et al. Science. 1996;274(5292):1527-31.
Association of Anxiety-Related Traits with Polymorphism in the Serotonin Transporter Gene Regulatory Region
Lesch et al. Science. 1996;274(5292):
Two samples (N = 221, N = 284)
Association with P ~ 0.02

45. A more complete evaluation of significance
Posterior odds = Prior odds x Power
(for true association) 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

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

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

48. 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):

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

50. Genomewide association analysis

51. Progress in identifying gene variants for common traits
MEIS1
LBXCOR1
BTBD9 C3
8q24
ORMDL3
4q25
TCF2
GCKR
FTO
C12orf30
ERBB3
KIAA0350
CD226
16p13
PTPN2
SH2B3
FGFR2
TNRC9
MAP3K1
LSP1
HMGA2
GDF5-UQCC
HMPG
JAZF1
CDC123
ADAMTS9
THADA
WSF1
LOXL1
IL7R
TRAF1/C5
STAT4
ABCG8
GALNT2
PSRC1
NCAN
TBL2
TRIB1
KCTD10
ANGLPT3
GRIN3A
CDKN2B/A
8q24 #2
8q24 #3
8q24 #4
8q24 #5
8q24 #6
ATG16L1
5p13
10q21
IRGM NKX2-3
IL12B
3p21
1q24
PTPN2
TCF2
IGF2BP2
CDKAL1
HHEX
SLC30A8
Cholesterol
Obesity
Myocardial infarction
QT interval
Atrial Fibrilliation
Type 2 Diabetes
Prostate cancer
Breast cancer
Colon cancer
height
Age Related Macular Degeneration
Crohns Disease
Type 1 Diabetes
Systemic Lupus Erythematosus
Asthma
Restless leg syndrome
Gallstone disease
Multiple sclerosis
Rheumatoid arthritis
Glaucoma
NOS1AP
IFIH1
PCSK9
CFB/C2
LOC387715
8q24
IL23R
TCF7L2
CD25
IRF5
PCSK9
CFH
2001
IBD5
NOD2
2000
PPAR
2002
CTLA4
KCNJ11
2003
2004
PTPN22
2005
2006
2007
Slide from David Altshuler

52. HDL Association at 16q22.1

53. HDL Association near LIPC

55. Progress in identifying gene variants for common traits
MEIS1
LBXCOR1
BTBD9 C3
8q24
ORMDL3
4q25
TCF2
GCKR
FTO
C12orf30
ERBB3
KIAA0350
CD226
16p13
PTPN2
SH2B3
FGFR2
TNRC9
MAP3K1
LSP1
HMGA2
GDF5-UQCC
HMPG
JAZF1
CDC123
ADAMTS9
THADA
WSF1
LOXL1
IL7R
TRAF1/C5
STAT4
ABCG8
GALNT2
PSRC1
NCAN
TBL2
TRIB1
KCTD10
ANGLPT3
GRIN3A
CDKN2B/A
8q24 #2
8q24 #3
8q24 #4
8q24 #5
8q24 #6
ATG16L1
5p13
10q21
IRGM NKX2-3
IL12B
3p21
1q24
PTPN2
TCF2
IGF2BP2
CDKAL1
HHEX
SLC30A8
Cholesterol
Obesity
Myocardial infarction
QT interval
Atrial Fibrilliation
Type 2 Diabetes
Prostate cancer
Breast cancer
Colon cancer
height
Age Related Macular Degeneration
Crohns Disease
Type 1 Diabetes
Systemic Lupus Erythematosus
Asthma
Restless leg syndrome
Gallstone disease
Multiple sclerosis
Rheumatoid arthritis
Glaucoma
NOS1AP
IFIH1
PCSK9
CFB/C2
LOC387715
8q24
IL23R
TCF7L2
CD25
IRF5
PCSK9
CFH
2001
IBD5
NOD2
2000
PPAR
2002
CTLA4
KCNJ11
2003
2004
PTPN22
2005
2006
2007
Slide from David Altshuler

56. A success story in neuropsychiatry

57. Genome Wide association in narcolepsy in Japan (222 cases vs 389 controls)
HLA 6
-log10 (P value) 4 2
Chr
From Emmanuel Mignot

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

59. Strong genome-wide evidence

60. Known genes and environment explain little of trait variance60

61. Sequencing: the currently unexplored middle of the allelic spectrum

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

63. 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