Confused about scanone and scantwo with different results

[Forest Hunt]

I have an F1 population with ~250 grape individuals which I am testing for the presence of color causing anthocyanin compounds. I have classified 28 different anthocyanin compounds in this population. 

During my first round of scanone and scantwo testing, a MYB factor on LG2 was blocking out any other loci that may be involved with the control of the anthocyanin chemical pathway. To block out the noise from LG2, I was advised to split my population into two subgroups, Dark and Light, which I did with hierarchical clustering. 

The scanone results from this round of testing gave me more promising results and little to no noise from LG2. I had a very promising pattern of a loci on  LG3 showing up in 5 of my 28 phenotypes in the Dark subgroup as being above the significance threshold (for scanone, I consistently have used a 5% significance threshold [summary(perm1.anth15.2p,alpha=c(0.05))]). 

When running these same 5 phenotypes in the Dark subgroup through preliminary  scantwo testing, I received much more promising results than my first pass (I was able to get more than just the loci on LG2, with y ~ Q1, which was the majority of my results for the scantwo with the entire population) but weirdly I did not see the previously identified LG3 from the scanone results. 

I am absolutely stuck and unsure how to proceed. I am receiving advice from multiple sources to not bother with scantwo with the divided population, but I my results from scanone with the divided population is barely enough evidence to base a paper around. I am also unsure why the LG3 loci has disappeared in my scantwo analysis. My initial hope for this project was to be able to place relevant loci on parts of the anthocyanin pathway as potential triggers for those differences in anthocyanin distribution. But my results have not given me anything to work off of there. 

Any and all advice on how to proceed is welcome.

-Forest Hunt