Transcript Is Child Psychiatry Ready for Personalized Medicine?

Betwixt and Between;
Common and Rare Genetic
Variants in Human Disease
Peter Szatmari MD
Offord Centre for Child Studies
McMaster University
McMaster Children’s Hospital
Financial Disclosure






The Canadian Institutes of Health
Research
Autism Speaks
Sinneave Family Foundation
Ontario Research Fund
Royalties from Guildford Press
No other sources of funding (stocks,
industry, Big Pharma etc)
Objectives





What have we learned about the genetic
architecture of ASD;
Focus on explanatory power of common and rare
variants
Copy Number Variants as examples of rare risk
factors
Neither story provides much explanatory power
So we are “betwixt and between”; what does the
future hold? WGS?
What is Genetic
Epidemiology?



The study of inherited factors in disease
Combination of epidemiology and statistical
genetics
Uses a variety of study designs to meet its
objectives
Steps in Genetic Epidemiology







Is the disorder familial?- family studies
Is the familiality due to genetic factors?-twin and
adoption studies
Can candidate genes be identified?
Can chromosomal susceptibility regions be
identified?-GW linkage and association studies
Exome and Whole genome sequencing?
A disease can be genetic without being inherited
The history of autism genetics thru these steps
Autism spectrum disorders
An heterogeneous ‘spectrum’ disorder involving deficits in 3 domains of function
Social
communication
deficits
0.6 % to 1% prevalence
4 to 1 sex ratio, more females with severe ID
Changing epidemiology; more non-autism ASD
Strict autism
Spectrum
Changing epidemiology; less frequent ID
Increasing prevalence due to better case finding
Diagnostic substitution occurring
medical comorbidities
25-40%
6
Family Studies




RR to sibs; 5% but based on old data
collected retrospectively
Stoppage rules; when taken into account, sib
RR increases to 10%
Baby sibs studies; RR now 19%
Intermediate phenotypes in another 20%
Twin Studies


Twin studies; traits in general population and in
diagnosed twins
Older studies of ASD twins;



Hallmeyer et al 2011; MZ vs DZ concordance



MZ vs DZ=.65 vs .05
Heritability >90%
Males .58 vs.21
Females .60 vs. 27
Greater role for shared environmental factors (55%)
than genetic (37%)
The Genetic Architecture of
ASD


Some single gene disorders; TS, FraX, NF,
etc (5%)
Chromosomal abnormalities spread
throughout the genome (5%)
Kelleher III R.J and Bear M.F (2008) Cell 135, October 31, 2008
391 cytogenetically-visible breakpoints in autism
Source: http://projects.tcag.ca/autism/
1
2
3
4
5
6
7
8
9
10
11
12
13
Breakpoints
Translocation (n=126)
Deletion (n=128)
Inversion (n=37)
Duplication (n=100)
14
15
16
17
18
19
20
21
22
X
Y
What About the Other 90%?




Little family history of autism, low risk to sibs
and twins
Like other genetically “complex” disorders
such as CVD, epilepsy, obesity, diabetes, etc
Except that effect on fertility is greater
Two models of genetic complexity


Common disease-common variant
Common disease-rare variant
The Common disease-common
variant model; finding genes

Candidate gene studies

Genome wide linkage

GWAS
Common Disease-Common
Variant model



Non-syndromic, non-Mendelian ASD is a
common disease, therefore it might be
caused by common genetic variants
Polygenic multifactorial model; each gene
has a small to moderate effect size
Many different variants with an additive effect
The London Underground; de Vries Nature Medicine 15 (8) August
2009
Candidate Gene Studies





ASD considered to be “caused” by
neurotransmitters; 5HT, dopamine, NE
Focus on genes associated with regulating
those proteins
Hundreds of positive results
Hundreds of non-replications
Small sample sizes, multiple testing of
different alleles, marker density, population
stratification etc
Linkage Studies

Common variants of moderate to large effect
size

Genetic (locus) homogeneity

Focus on affected sib pairs and nonparametric models
Linkage; Parametric Methods






Based on non-independent segregation of genetic
markers and disease alleles
Developed for Mendelian disorders
“Log of the odds” of linkage vs no linkage (>3.0 is
significant)
Need dense families
Accurate classification is essential
Must specify a genetic model (gene frequency,
mode of transmission, penetrance)
Non Parametric methods





Degree of allele sharing among affected
relatives, most commonly sibs
Sibs share 0,1 and 2 alleles at 25%, 50% and
25%
Is there distortion in allele sharing?
Model free, less vulnerable to
misclassification
Major challenge is power; esp when there is
genetic (locus) heterogeneity!
Common Disorder/Common
Variant Linkage Studies in ASD




Many genome wide linkage studies using
affected sib pairs (using non-parametric
methods)
Each with sample size 50 to 400
Many significant linkage peaks but few are
replicable
Conclusion; disorder is so heterogeneous
and effect of common variant so small we
need very large sample sizes
Autism Genome Project

Phase I




Affymetrix 10k SNP genotype data
Linkage analysis in 1146 multiplex autism
families
Initial scan for CNV
Phase II




Illumina 1M SNP genotype data
High-resolution scan for de novo and inherited
CNV
Genome-wide association analysis
Molecular studies of candidate loci
Linkage Peaks Stratified by
Sex
Problems with Linkage for
Complex Disorders




Very sensitive to locus heterogeneity
Low power for loci of small to moderate
effects
Very sensitive to misclassification of
phenotype
Turn to GWAS; much greater power than
linkage for alleles of small effect
Genome Wide Association Studies
(GWAS)

1 Million genetic markers (SNP’s are biallelic
markers)

Which markers in which genes are more
common in children with ASD than expected?
Trio based or case-control

Are those markers located in genes (or in LD
with genes) that are expressed in brain?
GWAS






Very successful if MAF>5%
500 SNP’s (genetic markers) associated with
many common diseases
Eg Type 2 diabetes; 5000 cases and 5000
controls
18 SNP’s associated with type 2 diabetes
(OR=1.09 to 1.37)
Explain 6% of the heritability
Actual causal variant not discovered
GWAS

Wang et al (2009); cadherin genes at 5p14

Ma et al (2009); also at 5p14 but only in
secondary analysis

Weiss et al (2009) 5p15 at SEMA5A

Anney et al (2010) MACROD2
All Ancestry − Autism Dx − Additive Model
MACROD2
All Ancestry − ASD Dx − Additive Model
MACROD2
Bottom Line of GWAS?





One SNP barely reaches GWS
No subtype or ASD quantitative trait reaches
GWS (especially if correct for multiple testing)
None of the other results can be replicated
But beware of the “Winner’s Curse”!
GWAS very sensitive to allele frequency and
allelic heterogeneity
Power curves
2.0
1.8
1.6
Odds
Ratio
Power
1.4
1.2
Risk allele frequency
Largest sample evaluated in Stage 1
N = 1385 ASD subjects
1.6- 2.0
1.4
Odds
Ratio
Power
1.2
Risk allele frequency
The Argument for the Common
Variant Model






We should be studying more “familial cases”
We should be using intermediate phenotypes,
quantitative traits
We should be looking at gene X gene, gene X
environment interactions
We should be looking at parent of origin effects
We should ignore p-values and instead rank order
SNP’s
All true, next generation of GWAS
The Argument Against the
Common Variant Model





ASD is associated with reduced fertility
New variants must arise de novo that are risk
factors to keep prevalence stable
If they are new they are rare
Each person carries on average 175 de novo
mutations, deletions, duplications that are
mostly benign
If a deleterious variant occurs in a brain
expressed gene? Might cause ASD
Is ASD a Common Disease/Rare
Variant?



ASD a disorder with reduced fertility
De novo mechanisms of causation (like a
spontaneous mutation)
These will necessarily be rare until they
diffuse thru the population
What is a Rare Event?





Frequency of risk factor<1%
Variation in DNA sequence that affects
protein coding
SNP; biallelic marker (by itself or in LD with a
DNA sequence)
Structural variant; chromosomal abnormality
(ie a CNV, insertions, duplications,
translocations etc)
But they might have a big effect size
Slide courtesy of Dr. C. Marshall
The Boston Underground; de Vries Nature Medicine 15(8) August 2009
What are Copy Number
Variants (CNV’s)?

Variations in DNA segments >1kb

Deletions, insertions, duplications, others

Rare or common; inherited from parents or arise de
novo?

If CNV overlaps a gene expressed in brain, AND it
disrupts the function of that gene, it could lead to
ASD
Copy Number Variation (CNV)
Deletion
Duplication
“CNV refers to DNA segments for which copy number
differences have been observed in the comparison of two or
more genomes”
Slide courtesy of Dr. C. Marshall
Lee and Scherer, Expert Reviews in Mol. Med. 2010
Slide courtesy of Julie Cohen, ScM, CGC, Kennedy Krieger Institute
Copy Number Variations (CNVs)
•
•
•
We all have them!
Most of them do not
harm us
Most of them we
inherited from our
parents
Rare Variants in ASD

What is the evidence that rare variants, as
measured by CNV’s, play a role in ASD?

Simple comparison of “global burden” of brain
expressed CNVs or previously implicated
CNVs in ASD vs controls
Autism Genome Project

Collaboration of 13 research groups

Pooling of families (1500 families)


Common genotyping (1M SNP’s) and
clinical measures (ADI/ADOS) for all affected
sib pairs
Funded by Autism Speaks, CIHR, Genome
Canada, UK MRC, HRB (Ireland)
Global burden for rare CNVs in cases vs. controls
3 measures:
• CNV rate
• Estimated size
• CNV location and # of genes affected
*
PLINK v. 1.07, genome-wide P values, one-sided tests, 100,000 permutations
*Pcorr, controlled for global case-control differences, logistic regression
48
CNV burden in known ASD and/or ID genes
n=46
n=127
n=103
Enrichment of genic-CNVs in known ASD and ID loci
(1.69 fold, P= 3.4 x 10-4)
Genes in which CNV’s have
been replicated









Neuroligin 3 and 4
Neurexin
Shank2 and Shank3
Contactin associated protein 2
PTCHD1
Large region on chromosome 16p11
New ones reported each week!
Each one seen in <1% of cases
Range of effects; linked in common networks,
Walsh C.A., Morrow E.M. and Rubenstein J.L (2008) Cell 135, October 31, 2008
Functional Enrichment Gene-set Map for ASD
ASD and ID risk genes may be linked in a connected pathway
Familial segregation - examples
5444
G / --
829 kb dup
64 Kb del
2 adjacent 17q25.3 de novo CNVs
de novo del
17q25.3: SLC16A3, CSNK1D
de novo dup
17q25.3, 829Kb, 37 genes
5298
5290
-- / T
T/G
-- / G
121kb del
121kb del
121kb del
791 kb dup
de novo CNV
dup 8p23.3, 791kb, disrupts DLGAP2
maternal Xp22.11 del in males
DDX53/ PTCHD1AS
(non-coding RNA for PTCHD1)
maternal missense mutation
Xp21.3, IL1RAPL1 (A117S, 349G>T)
1/325 cases; 0/250 controls
*In red if there is previous evidence suggesting gene involvement in ASD or ID
54
MM0088 – MPX family. Proband has 676 kb de novo loss at 16p11.2
SK0102 – SPX family. Proband has 432 kb de novo gain at 16p11.2
SK0019 – SPX family. Proband has 676 kb de novo loss at 16p11.2
MM0088
676 kb loss
SK0102
432 kb gain
SK0019
676 kb loss
III. What does a de novo change mean in a complex disord
MPX #62346:
SPX #HSC0215:
De novo 1.2 Mb deletion at 3p25.1,
De novo 1 Mb deletion at 1p21.3
3.4 Mb deletion at 5p15, t(5;7)(p15;p13) Inherited t(19;21)(p13.q22.1)
t(19;21)
t(19;21)
PDD
AD
del 1p21.3
t(19;21)
AD
del 3p25.1
del 5p15
Prefer multiple lines of evidence supporting locus involvem
MM0160/MM1470-72 [SHANK1 deletion]
?
?
MM0160_007
lymphocyte
64kb del
11
5
MM1470_002
blood
no del
MM0160_001
blood
64kb del
?
11
MM1470_004
saliva
64kb del
Asperger
12
7c
6
MM0160_005
blood
64kb del
MM0160_006
blood
no del
13
MM1470_005 MM1470_003
saliva
blood
no del
64kb del
c3
2
MM0160_002
blood
no del
4c
MM1472_002
blood
no del
c
14
MM1472_003
blood
no del
MM0160_008
lymphocyte
no loss
8
10
c
9
MM0160_003 MM0160_004
no del (from old DNA)
blood
blood
64kb del
no del
MM1471_002
refused blood
collection
15
MM1471_003
saliva
no del
16
CNV’s in ASD





More de novo CNV’s in genes implicated in ASD
and ID, OR=1.69; 7% of cases vs 4% of controls
Population attributal risk is 3%
Discovered functional networks of genes
In ASD, a shift from neurotransmitters to synaptic
genes
Same CNV’s seen in ID, epilepsy, ADHD,
schizophrenia, BAD (?)
Next Generation of Studies






Search for rare inherited variants thru linkage
CNV’s smaller than 1KB
More complicated structural rearrangements
Whole exome and whole genome sequencing
Current efforts at WGS in ASD identifying
variants in another 10-15%?
Rare mutations common in unaffected
controls as well
Challenges




Annotation of functional significance of
variants
Determination of “causation” when risk factor
is rare and disorder is multifactorial
Are the health benefits of identifying rare
genetic variants worth the cost? Diagnostics
and therapeutics?
Heterogeneity is the main obstacle
Recent findings from WGS




Rare variants are common; due to populaiton
overgrowth and weak purifying selection
Most SNV in the genome are rare
>90% of SNVs detected to be functionally
relevant were rare
But it will take huge sample sizes to detect
the majority of rare variants involved in
disease mechanisms.
A final twist!
A final twist!
A final twist!
Conclusions
Data a mix of many
genetic subgroups
Conclusions
Draw a sample
May get lucky and catch
lots of the "orange" type
Conclusions
Draw a new sample
= reshuffling the mix
And now the disorder
looks green
conclusions





ASD is a complex genetic disorder with more
complexity than previously imagined
Many rare, de novo, variants account for an
increasing proportion of cases
Low hanging fruit in ASD genetics
Common vs rare variant models an
oversimplification
Many unanswered questions remain