Apparent dependence of protein evolutionary rate on number of interactions is linked to biases in protein-protein interactions data sets

Abstract

Background: Several studies have suggested that proteins that interact with more partners evolve more slowly. The strength and validity of this association has been called into question. Here we investigate how biases in high-throughput protein-protein interaction studies could lead to a spurious correlation.

Results: We examined the correlation between evolutionary rate and the number of protein-protein interactions for sets of interactions determined by seven different high-throughput methods in Saccharomyces cerevisiae. Some methods have been shown to be biased towards counting more interactions for abundant proteins, a fact that could be important since abundant proteins are known to evolve more slowly. We show that the apparent tendency for interactive proteins to evolve more slowly varies directly with the bias towards counting more interactions for abundant proteins. Interactions studies with no bias show no correlation between evolutionary rate and the number of interactions, and the one study biased towards counting fewer interactions for abundant proteins actually suggests that interactive proteins evolve more rapidly. In all cases, controlling for protein abundance significantly decreases the observed correlation between interactions and evolutionary rate. Finally, we disprove the hypothesis that small data set size accounts for the failure of some interactions studies to show a correlation between evolutionary rate and the number of interactions.

Conclusions: The only correlation supported by a careful analysis of the data is between evolutionary rate and protein abundance. The reported correlation between evolutionary rate and protein-protein interactions cannot be separated from the biases of some protein-protein interactions studies to count more interactions for abundant proteins.

Figure 1

(A) shows the relationship between evolutionary rate and expression level as measured by gene microarrays [21]. (B) shows the relationship between expression level and the total number of interactions from all studies. (C) shows the relationship between evolutionary rate and the total number of interactions from all studies. Some outlying data points are not shown, but are included in the calculations of the correlations in Table 1.

Figure 2

The correlation between evolutionary rate and the number of interactions is directly related to the bias towards counting more interactions for abundant proteins, both when abundance is measured by (A) gene microarray expression levels and (B) CAI. Correlations are Kendall's rank correlation τ, and points are for all data sets listed in Table 1.

Figure 3

Controlling for abundance reduces the magnitude of the correlations between evolutionary rate and the number of interactions from those shown in Figure 2, and the remaining partial correlation still depends on the bias towards counting more interactions for abundant proteins, both when abundance is measured by (A) gene microarray expression levels and (B) CAI. The partial correlations are Kendall's partial τ, the correlation between interactions and abundance is Kendall's rank correlation τ, and points are for all data sets listed in Table 1.

Figure 4

The correlation between evolutionary rate and the number of interactions does not depend on the size of the interactions data set as it would in the absence of bias in the counting of interactions. (A) shows the correlation and data set sizes for all sets in Table 1. (B) shows how the mean and standard deviation of the correlation should depend on the data set size in the absence of experimental bias in the counting of interactions, based on sampling simulations of the mass spectrometry (green) and yeast two-hybrid (red) method of counting interactions.