BIOSTAT 830 winter 2006

1
BIOSTAT 830winter 2006

• Association studies

2
Outline

• Background.
• Comparison to linkage studies.
• Requirements and key components of WGA.
• Issues and cautions of using WGA.
• Case study of AMD.

3
Different study designs

• The goal is to map the genes responsible for
  common diseases. Two categories.
• Candidate gene studies.
• Association,
• Resequencing.
• Genome-wide studies.
• Linkage mapping,
• Whole genome association study.

4
Linkage studies

• Genome-wide, 1,000 microsatellite markers across
  genome.
• Very successful in mapping monogenic Mendelian
  diseases.
• hemochromatosis, cystic fibrosis, Fanconi
  anemia
• Using technique called positional cloning.
• Works well in locating highly penetrant variant
  that causes rare diseases.

5
Linkage study

• Linkage studies use individual families where
  members are affected and attempt to demonstrate
  linkage between the occurrence of the disease and
  genetic markers (creates associations within
  families, but not among unrelated people)

6
Background

• Linkage or pedigree analysis mapped disease genes
  to within 1cM at most.
• Further refinement in location using family
  studies is difficult because recombinations are
  rarely observed even within the large pedigrees.
• Boehnke, 1994.
• 1 Mb of DNA to search, daunting. need method that
  is able to narrow it.

7
Complex traits

• Phenotype is determined by the sum total of,
  and/or interactions between multiple genetic and
  environment factors.
• Studying inheritance of traits that show no clear
  Mendelian inheritance but cluster in families
• Usually a mix of genetic and environmental
  factors
• Several models
• Single gene modified by environment
• Several genes each with significant contributions
  (oligogenes)
• Many genes each making a small contribution
  (polygenes)
• Last two- multifactorial inheritance

8
Problems of complex diseases

• Low heritability of complex traits.
• Imprecise definition of phenotype.
• genetic heterogeneity
• Alleles at more than one locus can trigger a
  specific disease
• Microsatellite markers 10cM apart too sparse
• Reduced penetrance
• Individuals with predisposing genotype are
  unaffected.

9
Problems of complex diseases

• Phenocopy
• Disease is triggered by environmental factors in
  the absence of a predisposing genotype
• Gene-gene interaction.
• Gene-environment interaction.
• Inadequately powered study designs.
• Even if linkage study successful, candidate gene
  follow up studies are often needed to locate the
  causal locus. Linkage region typically 10cM (10
  Mb).

10
Association Studies
Cases
Controls
11
Association studies

• Based on populations and attempt to show an
  association between a particular allele and
  susceptibility to disease (a statistical
  statement about the co-occurrence of alleles or
  phenotypes).
• In simplest form, compare the frequency of
  alleles or genotypes of a particular variant
  between disease cases and controls.

12
Association studies

• Allele A is associated with disease D if people
  who have D also have A more (or less) often than
  would be predicted from individual frequencies of
  A and D in the population
• eg. HLA-DR4 is found in 36 of UK population, but
  in 78 of people with rheumatoid arthritis.

13
Association tests

• Observe 3 by 2 table
• cases controls
• AA nAA mAA
• Aa nAa mAa
• aa naa maa
• Likelihood ratio test
• Comparing different models no constraint
  dominant, recessive, multiplicative.
• Score test.

14
Other reasons for associations

• Direct causation
• Having allele A makes you susceptible to disease
  D (increases the likelihood)
• Expect to see same allele A associated with
  disease in any population (bypasses common
  ancestor)
• Natural selection
• people with disease may have a competitive
  advantage if they also carry allele A
• These are unlikely if the associated DNA is a
  variant in non-coding DNA.

15
Candidate gene association study

• Rely on knowledge of the correct gene(s), based
  on biological hypotheses, or located near the
  linkage region.
• Successes have been reported.
• Review papers Cardon et al. 2001 Tabor et al.
  2002 Hirschhorn et al. 2002.
• The main problem is this type of study is not
  comprehensive.

16
Genome-wide association study

• Survey most of the genome for causal genetic
  variants, therefore no need to guess the identify
  of the causal gene. Unbiased and fairly
  comprehensive.
• Future of genetic studies?
• http//www.ncbi.nlm.nih.gov/WGA/

17
Requirements for WGA

• Knowledge about common genetic variation and the
  ability to genotype a sufficiently comprehensive
  set of variants in a large patient sample.
• dbSNP contain gt10 million SNPs.
• Genotyping technology advanced and cost lowered.
• Now 0.01, feasible around 0.001 ? 500 for
  500K genotypes.
• Knowledge of LD patterns on a genome-wide scale.
• Provided by HapMap.

18
Choices of markers

• LD-based markers.
• Indirect approach, one marker can serve as a
  proxy for many others.
• Algorithm for this purpose
• LDselect Greedy Carlson et al. 2004
• FESTA Exhaustive (almost) Qin et al. 2006

19
Missense approach

• Focus on missense SNPs only.
• Aka, nonsynonymous mutations are types of point
  mutations where a nucleotide is changed which
  results in a different amino acid. This in turn
  can render the resulting protein nonfunctional.
• For example, in sickle-cell disease, the 17th
  nucleotide of the gene for the beta chain of
  hemoglobin fonund on chr 11 is erroneously
  changed from the codon GAG (for glutamic acid) to
  GTG (which codes valine), so the sixth amino acid
  is incorrectly substituted.
• http//en.wikipedia.org/wiki/Missense_mutation

20
Arguments-for

• More likely to have functional consequences.
• High proportion of missense mutations underlie
  Mendelian disorders.
• However, ascertainment bias, and implicitly
  biased in declaring association.
• Only need to survey 30,000 60,000 SNPs genome
  wide.
• Botstein and Risch 2003.

21
Arguments-against

• Alleles underlie complex traits typically have
  modest impact, more likely to be non-coding
  regulatory variants.
• E.g., Thr17Ala coding polymorphism in CTLA4 is in
  strong LD with a non-coding regulatory variant,
  which is more strongly associated with autoimmune
  disease.
• Adding SNPs in evolutionarily conserved
  non-coding regions, close to WGA.

22
A convenience-based approach

• Select markers not based on LD or functional
  considerations. But convenience or logistic
  considerations. E.g., equal-spaced markers.
• Genome coverage varies, mostly inadequate. Least
  comprehensive is using 400-1000 markers used in
  linkage studies. One such marker only covers 20
  kb.
• Pre-selected set of markers, e.g., Illumina 300K
  and Affymetrix 500K.

23
Study design for WGA

• Staged design for cost-efficiency and less false
  positives.
• Andrew will address these issues in his guest
  lecture.

24
Control for multiple testing

• Multiple testing is a serious issue in WGA.
• Due to large number of markers and LD between
  them, independence assumption is strongly
  violated. Bonferroni correction is punitively
  conservative.
• Karen will tell us about other strategies.
• A good alternative for defining significance
  thresholds is permutation testing, which
  empirically assessing the probability of having
  observe equal or more extreme result by chance.
• Computationally intensive.

25
Permutation test
26
Cost-efficient association designsuse founder
populations.

• Are those that have been recently derived100 or
  fewer generations agofrom a limited pool of
  individuals. Such as Finnish people,
  French-Canadians and Ashkenazi Jews.
• Because of higher extended LD, less markers are
  needed to survey the entire genome.
• Rare variants on long shared haplotypes.
• More isolated populations will be more
  homogeneous and therefore might have the
  advantage of a more consistent environment.

27
Cost-efficient association designsuse pooled
samples

• Equal amount of DNA from multiple individuals are
  mixed into a single well before genotyping.
• Review paper Sham et al. Nat Rev Gen 2002.
• Reduce genotyping cost. Measure allele frequency,
  no genotypes.
• Require high-throughput, accurate determination
  of allele frequencies.
• Variants with pure recessive effects will be
  difficult to identify, since genotype frequency
  will be much more informative.
• To study multiple traits, a different pool must
  be made for each traits, which reduced
  efficiency.
• Gene-gene, gene-environmental interaction can not
  be studied without genotype data.

28
Population stratification

• Most noticeable source of systematic bias.
• The presence of multiple subgroups within a
  population that differ in disease prevalence,
  which leads to the over-representation of one or
  more subgroups in cases. When allele frequencies
  in different subgroups differ, false positive
  associations can ensure.
• Simpsons paradox.

29
Examples

• first half second half whole season
• Player A 5/10 (0.400) 25/100 (0.250) 30/110
  (0.264)
• Player B 36/100 (0.360) 2/10 (0.200) 38/110
  (0.336)
• subpop A subpop B total
• Case
• Control

Observe the association at the population level,
but not at the subpopulation level.
30
Correct for population stratification

• Detecting population substructure
• STRUCTURE Pritchard et al. 2000.
• Correcting for population stratification
• Genomic control Devlin and Roeder 1999.
• Mild stratification might exist in less admixed
  population which can be a problem when testing
  alleles with modest effects on diseases.

31
Solutions

• Precise assessment of population stratification
  is possible given the large number of markers
  typed in WGA.
• Match cases and controls based on their
  genotypes. Will lose some power since match is
  not likely to be one-on-one.
• An alternative is to use family-based samples.

32
Family based design

• Can control for the effects of shared
  environment.
• Allow combined analysis of linkage and
  association, family based association tests.
• family-based samples are difficult to collect.
• Sibship-based association studies are
  underpowered relative to case-control studies.
• Adding living parents may introduce an
  age-of-onset bias towards younger patients when
  study late-on-set diseases.

33
Gene-gene interaction

• Epistasis. Unfortunately, due to massive number
  of hypotheses, fully powered, unconstrained scans
  for epistasis that account for multiple
  hypothesis testing might not be possible in the
  near future.
• However, it is possible to detect the main
  effects. Hence an effective scan for epistasis
  involves searching for modest individual effects
  and then query for interactions among them.

34
Real casesAMD study

• Regarded the first real WGA.
• Klein et al. 2005 Zareparsi et al. 2005
• Age-related macular degeneration.
• 96 cases and 50 control, controls older than
  cases.
• genotyped by Affymetrix 100K chip.

35
Klein et al. study
36
Admixture mapping

• Admixture mapping is a method for localizing
  disease causing genetic variants that differ in
  frequency across populations.
• most advantageous to apply this approach to
  populations that have descended from a recent mix
  of two ancestral groups that have been
  geographically isolated for many tens of
  thousands of years for example, African
  Americans.
• The approach assumes that near a disease causing
  gene there will be enhanced ancestry from the
  population that has greater risk of getting the
  disease. Thus if one can calculate the ancestry
  along the genome for an admixed sample set, one
  could use that to identify disease causing gene
  variants.

Zhu et al. 2005, Reich et al. 2005
37
Homozygosity mapping

• Homozygosity mapping is a rapid means of mapping
  autosomal recessive genes in consanguineous
  families by identifying chromosomal regions that
  show homozygous IBD segments in pooled samples.

38
References

• Hirschhorn and Daly Nat Rev Gen 2005.
• Wang et al. Nat Rev Gen 2005.

39
Acknowledgement

• Terry Speed