QTL detected in single-QTL scan disappears cim

[Maunveri]

Hello,

I am new to QTL mapping and R/qtl.

I am trying to find QTL explaining resistance to a disease in a large hexaploid genome (21 chrs). My mapping population consists of doubled haploid progeny from a biparental cross (75 individuals). The corresponding genetic map was already published (7328 markers) and I have generated phenotype data for 2 different phenotypes at three different time points. 

When I perform a single-QTL scan with scanone and the "em" method, I get (consistently for all 6 phenotypes) two clear and significant (alpha = 0.05 from 1000 permutations) peaks on "5B" and "7B" (maximum LOD score across all phenos around 15 and 7, respectively). The rest of the LOD curve is relatively smooth. I also tried the non-parametric model option since my phenos are generally right skewed, which yields almost the exact same peak positions (+/- 2 cM).

Both QTL are from the same parental genome, and since I observed some lines with a more severe phenotype than my susceptible parent, I chose to continue to look for minor or interacting QTL that could stem from the susceptible parent.

I next tried composite interval mapping using the cim function, and to my surprise it only considered 5B as the first covariate and then chose a different locus than 7B as a second covariate. The peak for 7B had disappeared. 

If I manually repeat this with scanone+addcovar (5B as a covariate), 7B also disappears.

If I use only 7B as a covariate, I get 5B, and smaller peaks on 5D and 6D.

To investigate this further, I performed a 2-QTL scan using the scantwo function, where I see a faint LOD "tail" for 7B and a strong one for 5B as well as to minor QTL interacting (epistasis) on 5D and 6D (determined as significant with 1000 permutations in scantwo; although bear in mind my phenos are non-normal which is not supported by scantwo).

When I use all 4 QTL in a multiple QTL model, irrespective of which combination / "formula", the only QTL ever seen as significant in the "drop one QTL at a time ANOVA table" is the one on 5B, and 7B doesn't really seem to aid in explaining phenotypic variance (around 10 % seems to be explained by 5B and another 4-5% by 5D-6D). The LOD score over the null model is always low (around 2) and the best Pvalue(F) was for only considering 5B alone (0.007).

This all leads me to ask myself whether the QTL on 7B is actually "legit". I saw that the marker density at this region is not the highest, however repeating scanone with the multiple imputation method, which should be less sensitive to sparse genotype information, gives the same results as for em.

From your experience, what could be a reason for a QTL being detected in a single-QTL scan and disappearing in multiple QTL scans (except maybe scantwo)? Is this an artifact, or could there be a biological reason?

Thanks in advance,

Marie

[Luzie Ursula Wingen]

Dear Marie,

I assume I have an answer for your question. It might have to do with the species / variety you are working with.

You haven't told us, but I assume you are conducting QTL mapping in wheat (hexaploid, 21 chromosomes, chromosome names 5B, 7B).

Moreover, you are saying that you are using a  previously published biparental cross - most likely a cross between modern wheat varieties.

Several European varieties are known to carry a chromosome translocation between 5B and 7B.

If you cross a variety with translocation to a variety without some 'weird' stuff happens to these two chromosomes in the progeny, but the chromosomal behaviour is not covered by genetic mapping programmes.

This leads to genetic maps that are not full representations of reality.

If you now use these maps in QTL mapping, a QTL on 5B will also show up to some extent on 7B - due to the unusual or additional linkage in about 50% of the progeny.

I predict that this is what you are describing here. If you happen to map in Avalon x Cadenza, I know that one of them carries the translocation but not the other one.

I hope this is helpful

Best wishes

Luzie

--
You received this message because you are subscribed to the Google Groups "R/qtl discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rqtl-disc+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/rqtl-disc/e42874d8-25a0-4b0d-af45-f10b2d617826n%40googlegroups.com.

[Karl Broman]

Yes, I'm also suspecting association between genotypes in the two regions.

You can use the function geno.crosstab() to investigate such association.

karl

[Maunveri]

Dear Luzie,

thanks, this was extremely helpful - I am indeed working with wheat and with the population Avalon x Cadenza! Hence you could correctly predict part of my experiment just by a description of the problems I am encountering :,D. Having this information, everything makes more sense now.

I will also test out the geno.crosstab() function as Karl suggested. Perhaps that way I can learn how to correctly identify such a problem the next time I may encounter it.

Thanks again and cheers,

Marie

[Karl Broman]

The genotype associations may also show up in a plot of pairwise recombination fractions using est.rf() and then plotRF(). 

With so many markers, you may want to plot just a few chromosomes at a time.

karl