Why is the correlation between gene importance and gene evolutionary rate so weak?

Abstract

One of the few commonly believed principles of molecular evolution is that functionally more important genes (or DNA sequences) evolve more slowly than less important ones. This principle is widely used by molecular biologists in daily practice. However, recent genomic analysis of a diverse array of organisms found only weak, negative correlations between the evolutionary rate of a gene and its functional importance, typically measured under a single benign lab condition. A frequently suggested cause of the above finding is that gene importance determined in the lab differs from that in an organism's natural environment. Here, we test this hypothesis in yeast using gene importance values experimentally determined in 418 lab conditions or computationally predicted for 10,000 nutritional conditions. In no single condition or combination of conditions did we find a much stronger negative correlation, which is explainable by our subsequent finding that always-essential (enzyme) genes do not evolve significantly more slowly than sometimes-essential or always-nonessential ones. Furthermore, we verified that functional density, approximated by the fraction of amino acid sites within protein domains, is uncorrelated with gene importance. Thus, neither the lab-nature mismatch nor a potentially biased among-gene distribution of functional density explains the observed weakness of the correlation between gene importance and evolutionary rate. We conclude that the weakness is factual, rather than artifactual. In addition to being weakened by population genetic reasons, the correlation is likely to have been further weakened by the presence of multiple nontrivial rate determinants that are independent from gene importance. These findings notwithstanding, we show that the principle of slower evolution of more important genes does have some predictive power when genes with vastly different evolutionary rates are compared, explaining why the principle can be practically useful despite the weakness of the correlation.

Figure 1. Frequency distributions of Spearman's rank correlation coefficient ρ between gene importance (i.e., fitness reduction upon gene deletion) and evolutionary rate across many conditions.

Gene importance is measured by experiments in 418 lab conditions (panels A–C), predicted by FBA for enzyme genes in 10⁴ simulated nutritional conditions (D–F), or predicted by MOMA for enzyme genes in the same 10⁴ conditions (G–I). Gene evolutionary rate is measured by nonsynonymous substitution rate d _N (A, D, G), nonsynonymous/synonymous rate ratio d _N/d _S (B, E, H), or propensity for gene loss PGL (C, F, I). The yellow arrow in each panel indicates the observed correlation using gene importance values experimentally determined in the YPD medium and the red arrow indicates the strongest correlation across the conditions examined. The numbers of genes used are 3999 for panels A–C, 478 for panels D, E, G, and H, and 546 for panels F and I. The gene number is lower than 546 for panels D, E, G, and H, because some S. cerevisiae genes do not have orthologs in S. bayanus. The yellow arrow is on the left-hand side of the red arrow in panels G, H, and I, because, under all simulated conditions, MOMA-predicted fitness values have weaker correlations with the evolutionary rates than that observed under YPD.

Figure 2. Always-essential enzyme genes do not evolve significantly slower than sometimes-essential and always-nonessential ones, regardless of the measure of the evolutionary rate.

Error bars show one standard error. P-values are from Mann-Whitney U test between groups of genes. The numbers of genes used are 478 for panels A and B and 546 for panel C.

Figure 3. Relationship between the importance (β) and functional density (α) of genes.

Gene importance is measured by the experimentally determined fitness reduction upon gene deletion in YPD. Functional density is measured by the proportion of amino acid sites within functional domains predicted by (A) the ProSite algorithm or (B) InterProScan. In InterProScan, a site is considered a domain site when predicted by at least two algorithms. A total of 5936 yeast genes are used in this analysis.

Figure 4. Predictability of the principle of slower evolution of more important genes.

(A) Predictions of relative gene importance are more likely to be correct when the difference in evolutionary rate between the two genes under comparison increases. Rank difference shows the minimal fraction of genes in the genome whose ranks in d _N are between the two genes under comparison. Gene importance is measured by the amount of fitness reduction caused by the deletion of the gene under YPD. For each rank difference criterion, 100,000 random pairs of genes satisfying the criterion are used to estimate the prediction accuracy. (B) Extremely conserved genes (measured by d _N) are more likely to be essential. For the 418 lab conditions, the average proportion of essential genes among the 418 lab conditions and its standard error are shown.