Extensive Decoupling of Metabolic Genes in Cancer

http://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1004176

Ed Reznik and Chris Sander

It is commonplace in gene expression analysis to look for differential expression between two conditions. However, the goal of this study was to take such analysis to second-order, by identifying pairs of metabolic genes which are co-expressed in different manners between normal and tumour tissue. This is called "differential co-expression".

To do this, the authors calculated the Spearman's rank correlation coefficient, across samples, for pairs of genes (r_ij, abbreviated to r). Genes which do not have a statistically significant correlation, after Bonferonni multiple hypothesis testing, have their correlation coefficients set to zero. This yields a table of significant correlation coefficients, for both normal (r_N) and tumour tissue (r_T).

To compare r_N and r_T requires care. Taking their difference as a means of comparison is insufficient since distances between small correlation coefficients (e.g. r_N = 0.1 and r_T = -0.1) are less significant than distances between large correlation coefficients (e.g. r_N = 0.99, r_T = 0.79). This is because the variance of a sample correlation coefficient approaches zero, as the true value of a correlation coefficient for a population (rho_N and rho_T respectively) approaches one. The authors used the Fisher r to z transformation, to transform r_N and r_T into a single quantity which is normally distributed with mean zero and variance one. This allows the pair (r_N,r_T) to be converted into a p-value, under the null hypothesis that the population correlation coefficients obey rho_N = rho_T (using a Gaussian assumption). These p-values can then be Bonferroni corrected once more, to generate differential co-expression statistics.

Of particular mitochondrial interest was the authors' investigation of differential co-expression across seven cancer studies, to find gene pairs which were consistently differentially co-expressed. The authors found approximately 50 metabolic genes, from a library of 1789, which were differentially co-expressed in at least 3/7 studies. Amongst these genes were components of ATP synthase (ATP5F1 and ATP5L), a subunit of complex IV (COX7B) and complex I (NDUFV2). They particularly focused on coexpression of ATP5F1 and ATP5L to find that the expression of these genes is almost precisely equal in normal tissue, but is substantially noiser in tumour samples, suggesting a stoichiometric imbalance in cancer. Further experiments are required to evaluate their hypothesis.