Very nice paper! It's always a pleasure to read about a nice, well-powered experimental data set combined with solid analysis. With 60 individuals you are clearly well powered to detect most alternative splicing/transcription events as well as look at gene-gene correlations across individuals.
It's nice to see that you can detect the same widespread 3'UTR shortening in macrophage response to bacterial infection that we recently reported in response to LPS stimulation (http://www.nature.com/articles..., Figure 4e). Furthermore, we only had data at 6 hours after stimulation, but you clearly show that this effect is very stable and persists all the way from 2 hours to 24 hours post-infection. I also liked your follow up work on miRNAs and especially your hypothesis that 3'UTR shortening might be a mechanism for those genes to escape down-regulation by miRNA that are upregulated in response to infection. What would be good way to experimentally test that? Maybe use CRISPR to disrupt the proximal poly-adenylation site and look at transcript stability?
Another thing that you did not explicitly focus on in your analysis, but what seems to be quite clear from both of our data is that alternative promoter events (alternative first exons) seem to show larger changes in proportion (larger changes delta-psi) than alternative exons and 3'UTRs (Figure 2D in your paper and and Figure 4c in our paper). There might be some differences in power, namely alternative 3'UTR are longer and hence it's easier to detect smaller changes in proportion, but it seems that there is still an excess of very large effect sizes in alternative promoters.
Interestingly, in our analysis we detected only a very small number of alternative splicing events (retained introns and skipped exons) in LPS response (Figure 4d in our paper), but that could be due to multiple reasons (different time point, different stimulation, different annotations and analysis strategy).
A couple of technical comments as well:
1. The smoothed scatter plot on Figure 2D is a bit crowded and its hard to see the differences between TandemUTRs, SEs, and ALEs. Maybe using multiple small plots would help here?
2. You used MISO to detect significant alternative splicing events in response to infection. One of the limitations of MISO is that it does explicitly model replicates and hence you have to define arbitrary criteria to detect significant hits (BF > 5 in at least 10% of individuals). You do show that this does not lead to false positives (Figure 2A), but you might still be losing power. Have you considered using something like mmseq (https://github.com/eturro/mmse... that explicitly models replicates? In our paper we were able to successfully use it to analyse splicing events (in addition to full transcripts).
I hoe your paper gets accepted soon.