Completed on 30 Mar 2016 by Arjun Raj. Sourced from http://biorxiv.org/content/early/2016/03/19/044719.
Login to endorse this review.
Author response is in blue.
Interesting paper! The Mendelian Randomization approach is quite clever.
Had a couple thoughts:
I like the idea of using SNPs to infer expression level and then look for changes in transcription, and I agree that it removes the "shared confounder" issue. However, I'm not sure it necessarily proves the case for a trans interaction. Let's say lincRNA A causes mRNA B (which is closeby, hence in cis) to change expression, which in turns causes distal mRNA C to change expression. Then, the mechanism is in cis, but the effect is in trans, right? I get that the cis changes didn't show up, but it is possible that the effects were below the detection threshold. Am I missing something?
For those of us, umm, naive to the "naive" approach, the description was rather opaque. Is there a more plain English description of how the naive approach works and thus how it differs from Mendelian Randomization?
Anyway, cool work!
Thanks for the feedback, Arjun! Both great points.
You are absolutely right on point one. There is no way to definitively say that the lincRNA is *the* mechanism of distal regulation (as you say, there could be local effects, or, possibly, pleiotropic effects of the SNP locally as well). We (and others) recognize this as a failing of MR so far (see this great paper on assumptions in MR, http://www.ncbi.nlm.nih.gov/pm... Figure 2), and we are working on methods to handle this assumption directly. We will include this caveat clearly in the manuscript. That said, the mRNA controls, and the "naive analysis" both hint at the fact that this "alternative mechanism" may not exist, but these alternative tests are not conclusive, and experimental validation here is difficult for reasons mentioned in the paper.
I will definitely be more clear on what the naive approach is. In plain English: We tested every SNP that was a (top) cis-eQTL for lincRNA and mRNA for association with all mRNA genes on different chromosomes than the cis-eQTL across the four tissues. We did this same test permuting the sample order to calibrate our Bayes factor threshold via FDR. We found a huge number of trans effects for the cis-pc-eQTLs, and almost no trans effects for the cis-linc-eQTLs. It's naive in that we filter the "discovery SNP set" to be only the known top cis-eQTLs, but other than that it is the usual trans-eQTL analysis.
Thanks again for your thoughtful comments -- we will revise the paper accordingly.