Faster evolving Drosophila paralogs lose expression rate and ubiquity and accumulate more non-synonymous SNPs

Abstract

Background: Duplicated genes can indefinately persist in genomes if either both copies retain the original function due to dosage benefit (gene conservation), or one of the copies assumes a novel function (neofunctionalization), or both copies become required to perform the function previously accomplished by a single copy (subfunctionalization), or through a combination of these mechanisms. Different models of duplication retention imply different predictions about substitution rates in the coding portion of paralogs and about asymmetry of these rates.

Results: We analyse sequence evolution asymmetry in paralogs present in 12 Drosophila genomes using the nearest non-duplicated orthologous outgroup as a reference. Those paralogs present in D. melanogaster are analysed in conjunction with the asymmetry of expression rate and ubiquity and of segregating non-synonymous polymorphisms in the same paralogs. Paralogs accumulate substitutions, on average, faster than their nearest singleton orthologs. The distribution of paralogs' substitution rate asymmetry is overdispersed relative to that of orthologous clades, containing disproportionally more unusually symmetric and unusually asymmetric clades. We show that paralogs are more asymmetric in: a) clades orthologous to highly constrained singleton genes; b) genes with high expression level; c) genes with ubiquitous expression and d) non-tandem duplications. We further demonstrate that, in each asymmetrically evolving pair of paralogs, the faster evolving member of the pair tends to have lower average expression rate, lower expression uniformity and higher frequency of non-synonymous SNPs than its slower evolving counterpart.

Conclusions: Our findings are consistent with the hypothesis that many duplications in Drosophila are retained despite stabilising selection being more relaxed in one of the paralogs than in the other, suggesting a widespread unfinished pseudogenization. This phenomenon is likely to make detection of neo- and subfunctionalization signatures difficult, as these models of duplication retention also predict asymmetries in substitution rates and expression profiles.

Figure 1

Topologies of trees with duplications. A: Phylogeny topology used for the placement of substitutions on the paralogous branches. B and C: discarded topologies (duplication followed by a loss of one of paralogs and nested duplication). Dot: speciation event; star: duplication event; green: used duplications, red: discarded duplications.

Figure 2

Substitution rates in paralogs and their sister singletons. A: Scatterplot of K_a (log scale) in paralogous vs. their nearest sister outgroup orthologous singleton clades. Colours of circles and lines indicate three classes of clade age measured in K_s units estimated for the speciation event. Inset: Frequency of triplets in which paralogs evolve faster (K_{a, para} > K_{a, single}) than singletons (above 0 line) and vice versa (below 0 line). Filled portions of bars represent the frequency of triplets in which this difference is significant after Bonferroni adjustment (P_B < 0.05). B-D: Distribution of substitution rates (relK_a), normalized by the upper bound estimate of duplication age (B), absolute difference in polarity (|dPolarity|, C) and exchangeability (D) in paralogous (red) and singleton (blue) clades.

Figure 3

Asymmetry of substitution rates in paralogs. A: The distribution of K_a asymmetry in paralogous clade (red) relative to simulated asymmetry based on random placement of substitutions on the two clades (black). B: The portion of pairs of paralogs in which the observed asymmetry is either low (green) or significantly high (red, adjusted χ2 test P value <0.05 after Bonferroni correction). C. Substitution rates asymmetry of in paralogous clades vs. K_a values observed in the orthologous singleton clade, binned at 0.05 unit intervals on the logarithmic scale. Vertical bars are standard errors. Black lines: expected asymmetry in case of the shape parameter of gamma-distribution of substitution rate among amino acid sites 2 (top line) and 0.5 (bottom line). Linear regressions drawn through all points with different upper-bound duplication age estimate of duplication ages (colours as on Figure 2).

Figure 4

The effects of expression level (A), expression evenness (B) and chromosomal location (C) on K_a asymmetry of paralogous pairs in D. melanogaster, for which expression data are available [34]. Vertical bars are standard errors. Regression lines are drawn based on all data points. A: regression coefficient = 0.397, P < 0.0004. B: regression coefficient = −0.0995, P < 0.11. C: mean log-transformed K_a asymmetry in paralogs located on different chromosomal arms, same arms but over 5 kb apart, less than 5 kb apart in collinear orientation and less than 5 kb apart in inverted orientation. Categories labelled with different letters are significantly different by Tukey test (P < 0.05).

Figure 5

Coefficient of correlation over 26 larval and adult tissues (A) and signed difference of mean gene expression (B, black circles), expression CV (B, open circles) and SNPs K_a/K_s (C) ratio in pairs of paralogs for which D. melanogaster expression data are available[39], plotted against logarithm of K_a asymmetry in the same pair (binned by 0.5 of a log₁₀ unit). Asymmetry is directionalized so that negative values correspond to the faster evolving paralog having lower mean expression level or lower evenness of expression. Vertical bars are standard errors. Regression coefficients (R) and significance levels for regression lines (linear regression on K_a asymmetry, both dependent and independent variables log-transformed, fitted for data points in which Z² >1) are: A – R = −0.226; P < 1.20E-06; B - mean gene expression asymmetry: R = −0.442; P < 1.4E-08, gene expression evenness asymmetry: R = 0.356; P < 0.03; P < 0.004; C – R = 0.073; P < 0.002.