I'm studying the population genetics of an intestinal nematode (raccoon roundworm) using microsatellites. An analysis using GenePop shows significant LD in 20 out of 28 possible comparisons between loci (!). Is this a reflection of how highly partitioned the worm populations are between hosts, or an indication that the loci are somehow problematic vis-a-vis making population genetic inferences?
More broadly, what does finding (or not finding) LD tell us in the context of a PopGen analysis? Seems like checking for LD is almost a ritual feature in the MM of articles, but I'm unclear on its purpose. As a newcomer to the field, any insights would be welcomed. --Thanks
LD results from a non-random association among alleles at 2 or more loci. This may be a result of physical linkage: the alleles are on the same chromosome, perhaps very close together, or in some cases, 2 loci considered to be different are actually the same (b/c two differing primer sets were constructed that amplify the same locus). Assuming physical linkage is not the cause, LD can arise through drift, selection, and/or gene flow. Popgen programs like STRUCTURE assume that data is in Linkage equilibrium, and loci that are in LD can lead to the program overstating popgen structure. See: http://www.nature.com/hdy/journal/v99/n4/full/6801010a.html
Dear Doug,
Have you used a bonferroni correction (for multiple comparisons)? Using that correction of the critical P value may decreased the number of loci in LD.
Josselin
Another factor to consider is that LD can arise due to the presence of selfing or inbreeding within a population as the lack of random mating prevents the breakdown of linkage. The potential for substructured is definitely another possible cause of linkage as alleles may not be shared between the multiple subpopulations even while random assortment is occurring within the subpopulations.
If you have access to a copy then I would recommend having a look at 'Principles of Population Genetics' by Hartl & Clark, it's one of the best introductory text books for population genetics.
Thank you all for the suggestions. Josselin (and anyone else), can you recommend a popgen program that uses Bonferroni correction in its LD calculations? I find no mention of it in the GenePop manual, or in Arlequin, or in FSTAT...
Thanks
Doug,
I use a website (SISA) to calculate alpha with a bonferroni correction.
In your case, I understand that you have 8 microstaellite markers (because you mentionne 28 possible comparisons) and thus each marker is involved in 7 comparisons. The alpha using the bonferroni adjustment move from 0.05 to 0.0071429.
Here is the link for the website I used :
http://www.quantitativeskills.com/sisa/calculations/bonfer.php?Alpha=0.05&N=7&Corr=0.00&Df=00
Josselin
Doug
Bonferroni adjustment is known to over-correct (i.e., you will lose some truly significant associations). it is widely accepted that the FDR (false discovery rate) performs much better than Bonferroni. You can search for it within the R statistical package (for example, the function "p.adjust").
Most importantly, I agree with Craig that population structure is the most likely reason for detecting extensive LD patterns. Nei and Li (1973) demonstrated that if selection maintains allele frequencies differences at several loci in different populations, or if drift caused populations to differentiate, then gene flow would result in generalized LD (note the.gene flow part of the statement). Indeed, you can observe the same pattern if you mix differentiated populations.
Hope this helps
Pablo
Ryan, if one detects significant LD, does this call into question conclusions based on STRUCTURE results? I have very unequivocal results from STRUCTURE indicating two clusters, and this jibes with the geography of my sampling area: the two subpopulations are on either side of a river. And I have a plausible explanation for why the river could be acting as a barrier to gene flow. But this all rests on the STRUCTURE results, hence my concern.
My finding of significant LD, if considered on its own, would be in keeping with the ecology of this host-parasite system--i.e. it's probably the result of inbreeding--but I'm curious if this can be reconciled with the STRUCTURE results (?)
Dough
In spite of what the publication linked by Ryan said, in my experience Structure is rather "strong" (sorry, I miss the English word) against model violations. You can see my results with two oak species, each showing a large numbers of significant LD between loci, in the Supplementary files of my last publication (which is available in my ResearchGate page).
Definitively, I wouldn't be worried by your LD results, if the 2 groups structure fits to the population genetics of your system..
Regarding the paper cited by Ryan, there is a better way to infer the true number of clusters than just looking at the LnP(K) results.
http://onlinelibrary.wiley.com/doi/10.1111/j.1365-294X.2005.02553.x/abstract
Unfortunately, this method is not appropriate to differentiate between 1 or 2 groups.
Dough
I just realized you might have estimated LD in the total population. Please, estimate LD within populations (otherwise you are in the mixing of populations scenario, which leads to extensive LD)
Pablo
Doug,
If you are concerned about basing all of your results on Structure, you could try running a PCA (which doesn't assume HWE). If the results coincide, it would add more strength to your conclusions.
Erin
Dough,
you can adjust bonferroni correction in R with the function p.adjust ( ).
p.adjust(p, method = p.adjust.methods)
where p is a vector with the p-values (you obtained this p-values from genepop rigth? )
Example: with four p-values
p.adjust(c(0.05,0.05,0.1, 0.001),"bonferroni")
The corrected p-values will be: [1] 0.200 0.200 0.400 0.004
Carlos
Following Pablo G. Goicoechea you also may perform a fdr correction.
p.adjust(c(0.05,0.05,0.1, 0.001),"fdr")
[1] 0.06666667 0.06666667 0.10000000 0.00400000
In addition to last commen of Erin L Koen you can explore Spatial principal component analysis (sPCA). You can use the package adegenet.
http://www.nature.com/hdy/journal/v101/n1/full/hdy200834a.html
In addition to the causes Ryan Walter mentioned, genetic drift also generates LD. A recent reduction in population size can generate LD between widely separated loci. Details at: http://www.genetics.org/content/197/4/1329.short
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts