> When one is performing a study of association of alleles with a
> particular trait, is there any point in studying microsatellites and frequencies
> of different microsat alleles which may have an association with the trait.
>
I am assuming that in such an association study you are dealing with the
situation where you have mapped a disease gene to a particular chromsomal
region, and you are looking for an association which may suggest that such a
marker is very close to the gene. In such a case the empirical data says that
yes, it is very useful to look for associations between such markers and the
disease locus. We reported such an association with Werner's syndrome (AJHG
1994, 55:356-364), and a quick medline search with keywords "linkage
disequilibrium" and "microsatellite" pulls up about 60 such references since
1990, most of which (although not all) report or use linkage disequilibrium with
microsatellites.
>> I thought that microsats would evolve too quickly to be of use, but I'm open
> to contradiction. References particularly welcome.
>
The mutation rate does add some noise to the data which complicates analysis
of extended haplotypes, but the rate is not so high that such associations
cannot be seen. In fact, for markers with many alleles, because the "average"
disease mutation ends up on a haplotype with an allele which is not too high
in frequency, the subsequent enrichment of that allele in the case population
is particularly easy to detect. It is, however, important to use a method of
analysis which can both cope with small cell sizes and which does not require
the "pooling" of alleles. We have results which indicate that pooling of
alleles (so that you can perform a chi square test) can give very misleading
results, either way - false positive or false negative. Fisher exact works
when there aren't too many alleles at the marker (it is computationally
intensive, however). The Monte Carlo Markov chain method of Guo & Thompson
(Biometrics, June 1992) works very well for larger numbers of alleles, and is
computationally much faster.
Ellen Wijsman
wijsman at u.washington.edu
