Beyond differential expression: the quest for causal mutations and effector molecules

Abstract

High throughput gene expression technologies are a popular choice for researchers seeking molecular or systems-level explanations of biological phenomena. Nevertheless, there has been a groundswell of opinion that these approaches have not lived up to the hype because the interpretation of the data has lagged behind its generation. In our view a major problem has been an over-reliance on isolated lists of differentially expressed (DE) genes which - by simply comparing genes to themselves - have the pitfall of taking molecular information out of context. Numerous scientists have emphasised the need for better context. This can be achieved through holistic measurements of differential connectivity in addition to, or in replacement, of DE. However, many scientists continue to use isolated lists of DE genes as the major source of input data for common readily available analytical tools. Focussing this opinion article on our own research in skeletal muscle, we outline our resolutions to these problems - particularly a universally powerful way of quantifying differential connectivity. With a well designed experiment, it is now possible to use gene expression to identify causal mutations and the other major effector molecules with whom they cooperate, irrespective of whether they themselves are DE. We explain why, for various reasons, no other currently available experimental techniques or quantitative analyses are capable of reaching these conclusions.

Figure 1

Frequency histogram of bovine skeletal muscle transcripts (blue circles). The overall distribution is bimodal (simulated by the red and yellow circles), and a relatively small number of highly abundant transcripts encoding muscle structural subunits, ribosomal proteins and mitochondrial proteins dominate. MSTN sits in the nexus between the two distributions, possessing an average abundance of ~7.

Figure 2

Needle in a numerical haystack. Despite being the causal effector molecule, MSTN is neither DE nor abundant when comparing MSTN mutant cattle versus MSTN wildtype cattle. Here DE is computed by subtracting the average expression in the Wagyu from the average in the Piedmontese, across the 10 time points. Figure from PLoS Computational Biology.

Figure 3

Comparing theMSTNmuscle building pathway in the two breeds. The yellow (no differential expression), red (upregulated in Wagyu) and green (upregulated in Piedmontese) colours were generated within Cytoscape, with the bright red and bright green representing the outermost bounds of the extreme DE molecules across the whole transcriptome at day 135 post conception. The colour bar beneath the molecular pathway is intended as a schematic guide only. ACVR2B was not reported on the array, but was included in the diagram for visual completeness of the pathway.

Figure 4

Co-expression, co-differential expression and differential co-expression. As illustrated for the Piedmontese across development, MYOD1 and MYOG are strongly positively co-expressed whereas USP13 and CCNB2 are strongly negatively co-expressed (A). In comparing Piedmontese and Wagyu, PRSS2 and KLK12 are co-ordinately or co-differentially expressed in addition to being co-expressed (B). The differential co-expression arrangement between MSTN and MYL2 is large (+1.1), despite the co-expression strength being relatively modest in the Piedmontese (+0.76) and Wagyu (−0.34) breeds treated separately (C).

Figure 5

MSTNis highly differentially co-expressed with many of the abundant, highly differentially expressed genes - mutant breed on the left, wildtype breed on the right. For example, MSTN has a differential co-expression of 1.1 (+0.76 - - 0.34) with MYL2 (Panel A). RIF accumulates these differential co-expressions for all the DE genes (85 in this instance), weighted by their abundance. The size of the bubble representing the various DE genes corresponds to the combination of the extent of DE and average abundance. An alternative measure of differential connectivity is given in Panel B, where the number of significant co-expressions possessed by MSTN in the two breeds is contrasted. MSTN does not get prioritised by this alternative approach.