Post date: Mar 24, 2015 11:5:27 PM
As a first pass, I examined patterns of allele frequency change (really differences in allele frequencies between extant populations), to ask whether (i) similar patterns held across lines, and (ii) whether allele frequencies reverted in the reversion lines. R code is below.
## combine allele frequencies into one file
nl<-51371
np<-14
P<-matrix(NA,nrow=nl,ncol=np)
p<-read.table("p_L1-F91.txt",header=F,sep=",")
P[,1]<-p[,1]
p<-read.table("p_L1-F100.txt",header=F,sep=",")
P[,2]<-p[,1]
p<-read.table("p_L1R-F35.txt",header=F,sep=",")
P[,3]<-p[,1]
p<-read.table("p_L1R-F46.txt",header=F,sep=",")
P[,4]<-p[,1]
p<-read.table("p_L2-F78.txt",header=F,sep=",")
P[,5]<-p[,1]
p<-read.table("p_L2-F87.txt",header=F,sep=",")
P[,6]<-p[,1]
p<-read.table("p_L2R-F35.txt",header=F,sep=",")
P[,7]<-p[,1]
p<-read.table("p_L2R-F45.txt",header=F,sep=",")
P[,8]<-p[,1]
p<-read.table("p_L3-F76.txt",header=F,sep=",")
P[,9]<-p[,1]
p<-read.table("p_L3-F85.txt",header=F,sep=",")
P[,10]<-p[,1]
p<-read.table("p_L3R-F35.txt",header=F,sep=",")
P[,11]<-p[,1]
p<-read.table("p_L3R-F45.txt",header=F,sep=",")
P[,12]<-p[,1]
p<-read.table("p_M-13.txt",header=F,sep=",")
P[,13]<-p[,1]
p<-read.table("p_M-14.txt",header=F,sep=",")
P[,14]<-p[,1]
colnames(P)<-c("L1F91","L1F100","L1RF35","L1RF46","L2F78","L2F87","L2RF35","L2RF45",
"L3F76","L3F85","L3RF35","L3RF45","M13","M14")
# lentil vs mung
dpL1<-P[,2] - P[,14]
dpL2<-P[,6] - P[,14]
dpL3<-P[,10] - P[,14]
# lentil vs. reversion
dpL1R<-P[,4] - P[,2]
dpL2R<-P[,8] - P[,6]
dpL3R<-P[,12] - P[,10]
Overall there were positive correlations in allele frequency change in the three lentil lines relative to the mung line:
> cor(dpL1,dpL2)
[1] 0.442137
> cor(dpL1,dpL3)
[1] 0.5176451
> cor(dpL2,dpL3)
[1] 0.5369097
And negative correlations between allele frequency change on lentil vs. change during reversion on mung:
> cor(dpL1,dpL1R)
[1] -0.570391
> cor(dpL2,dpL2R)
[1] -0.5004933
> cor(dpL3,dpL3R)
[1] -0.5136938
These numbers are too big to reflect selection alone, and suggest that inherent constraints on allele frequency 'change' are driving much of this (i.e. allele frequencies can only increase if they start out low, etc.). Thus, I think it is time for some null simulations. One final thing first, there isn't much correlation in change during reversion (relative to lentil) across replicates (which is perhaps the one case where inherent constraints are less pronounced):
> cor(dpL1R,dpL2R)
[1] -0.007898195
> cor(dpL1R,dpL3R)
[1] 0.08090099
> cor(dpL2R,dpL3R)
[1] 0.0958562