3.2 The chain of psychiatric genetic research
3.2.5 What are the risk-conferring variants of the genes and what is the mechanism
of disease?
3.2.5.1 Association studies
Crossing over during meiosis shuffles the parental genes so that the chromosomes we receive from our fathers and mothers are not identical to any of their original chromosomes. Through the generations,
genes are constantly shifting from one chromosome to another. As a result, we should expect no association between alleles of genes on the same chromosome. For example, assume that locus 1 can have allele a or A and locus 2 can have allele b or B.
If the two loci are on the same chromosome then the probability that any chromosome contains the pair Ab, P(Ab), should be equal to the probability of A, P(A), times the probability of b, P(b). That is, P(Ab)
=P(A)×P(b). Similarly, P(AB)=P(A)×P(b), and so on. Put simply, if we know that a chromosome contains allele A at locus 1, this tells us nothing about the probability of locus 2 containing allele B or b.
This random distribution of alleles at different loci on the same chromosome is only partially true. It is an empirical fact that some loci are associated with one another so that P(Ab)=P(A)×P(b) [39]. For example, it may be that chromosomes with allele A at locus 1 are more likely to have allele b at locus 2 than we would expect by chance (i.e. than we would expect from the frequency of allele b in the population). Now, assume that locus 1 is a disease locus and that A is a dominant pathogenic allele.
Also assume that locus 2 is a DNA marker locus. If the two loci are associated as indicated above, then people with the disease should be more likely to have marker allele b than people without the disease.
This nonrandom association of alleles at different loci is called linkage disequilibrium. One cause of linkage disequilibrium is the fact that the reshuffling of genes on chromosomes depends on genetic dis- tance. If two genes are very close to one another, then they will rarely be separated by crossing over and will usually be transmitted together. Thus, due to very close linkage, the alleles at two loci will tend to be transmitted together. We say ‘tend to’ because eventually crossing over will separate them.
Fortunately, the reshuffling of linked genes can take many thousands of years. This means that, the- oretically, we should be able to detect associations between diseases and DNA markers if the marker locus isvery close to the disease locus. Compared with a linkage study, the design and analysis of an association study is straightforward. We do not require pedigrees with multiple ill members. Sam- ples of unrelated patients and controls will suffice (though family-based association study designs exist, and have their advantages). Instead of a complex
linkage analysis, all we need do is compare the rates of marker alleles (or genotypes) in patients and con- trols with standard statistical tests [39].
Genes within a linked region are candidates for involvement in the phenotype based on their chro- mosomal location or position (i.e. they are ‘positional candidate genes’). Within a linked region or even in the absence of linkage evidence, a gene may also be a candidate if there is some compelling reason to sus- pect that the gene influences risk for a given disorder.
Association of candidate genes can be evaluated in an independent sample of cases with the disorder and matched control subjects (i.e. in a ‘case–control’
study), or in small family units, where the transmis- sion of variant and normal forms of the gene from parents to offspring can be monitored.
In a case–control association study, we simply count the number of each type of allele of a gene, that is found in cases and compare these counts with the allele distribution seen in the control group;
this process can also be performed for genotypes.
A statistical test is then used to determine if the distribution of alleles observed among cases differs from that seen among controls. If it is different, then we have found evidence for a genetic association with the disorder, where the allele that is over-represented in the group of cases is considered the risk allele.
The degree of over-representation of the risk allele in cases relative to controls can be used to derive an odds ratio, which gives a numeric indication of the probability of an affected individual possessing the allele compared the probability of an unaffected individual possessing the allele. Association studies can be performed for alleles or genotypes. In addi- tion, a disorder can be tested for association with a haplotype, which is a pattern of alleles across several markers on the same chromosome (for a description of the International HapMap Project, which is ded- icated to cataloguing the haplotype structure of the entire human genome, see http://hapmap.ncbi.nlm .nih.gov/. If linkage disequilibrium, or unusually tight linkage, occurs between the markers in a haplotype, they will typically be inherited together, as no recombination will occur between them.
This concept is particularly useful for family-based association studies. In family-based studies, we can use analogous statistics to determine if any difference from the expected equal inheritance of
risk and normal alleles of a gene (or haplotypes within or across several genes) is detected in affected probands who could have received either allele from their parent. In a family-based study, the odds ratio estimates the haplotype relative risk, which represents the increase in the probability of the affected offspring receiving the risk allele (which is presumed to be on the same haplotype as the marker allele) relative to the normal allele.
If the odds ratio, relative risk or other effect size attributed to a polymorphism is large enough to attain statistical significance, there are four possi- ble explanations for the result: (i) there is a true association with a causative risk allele; (ii) the asso- ciated polymorphism is in linkage disequilibrium (i.e.
is in close proximity and usually co-inherited) with the causal variant; (iii) there is some confounding factor that introduces a systematic bias (e.g. popula- tion stratification, or background genetic differences between case and control groups) or (iv) the result is due to chance or random error. A disadvantage of association studies is that the DNA marker must either be in the disease gene itself or very close to it. This is in contrast with the linkage method which can detect linkage over relatively large distances.
However, unlike linkage analysis, association analy- sis can detect genes having only a small effect on the susceptibility to illness.
Candidate gene association studies are limited by the method used to choose candidates. For a gene to become a positional candidate gene, it must map to a chromosomal region that has been observed to show linkage to the disorder. However, genes with a small but reliable effect on risk may not generate a linkage signal, and thus may never come under study.
Selecting genes for association analysis based on their theoretical involvement in the disease process is risky as well. Since our understanding of the biological basis of most psychiatric disorders is far from comprehensive, the pursuit of candidate genes typically progresses incrementally through genes that are expressed within systems widely implicated in the disorder. This is clearly not an optimal process, as the prior probability of selecting the right candidate gene (out of∼25 000 human genes) and the right polymorphism (out of more than 10 000 000 in the human genome) for analysis is remote. The recent advancement of laboratory and statistical methods
for genome-wide association analysis should allow for a more unbiased examination of association patterns throughout the genome and help resolve this dilemma in coming years [56].
Another limitation of association studies is that they are notoriously difficult to replicate, perhaps owing to their propensity towards false-positive results [57]. The problem of false positive results is exacerbated by the fact that close linkage is not the only cause of disease-marker associations. As discussed above, the frequencies of DNA marker alleles may vary among ethnic groups, cohorts of different ages or other isolated segments of a population. Thus, if case and control groups are not drawn from the same populations and carefully matched on all relevant factors, spurious differences in allele frequencies between groups will emerge due to the population admixture alone [58]. Because it may be difficult to find patient and control groups that are suitably matched for ancestry, genomic control methods have been advocated to account for any imprecision in matching. These methods genotype ancestry-informative markers (i.e. those that differ in frequency across ancestral groups) in addition to those of hypothesised importance in the study, and use the frequencies of these markers in cases and controls to derive an adjustment factor that can be applied to the results pertaining to the hypothesised risk locus.
Several investigators have developed tests of linkage disequilibrium that use the parents of ill individuals as controls [59–63], which also circumvents the problems associated with ancestral mis-matching between cases and controls. The transmission disequilibrium test (TDT) uses families having at least one affected offspring and one parent who is heterozygous for the DNA marker to be tested [61]. The TDT compares the number of times heterozygous parents transmit the associated marker allele to affected offspring with the number of times they transmit the other marker allele. If these probabilities differ from what is expected by chance, then we can conclude that linkage disequilibrium exists. Although the TDT solves the problem of ethnicity matching, it still faces the problem of false positives and must be cautiously interpreted in the absence of a credible candidate gene.
Family-based association tests (FBATs) have been developed as extensions to the TDT model, whereby parents or siblings of patients are used as controls.
Since each parent transmits only one allele to a child, the allele, that is not transmitted to the child is used as the control allele. The statistical test involves compar- isons of the transmitted versus the non-transmitted allele. Because both alleles come from the same par- ent, there are no differences in ethnicity.