Nonrandom divergence of gene expression following gene and genome duplications in the flowering plant Arabidopsis thaliana

Abstract

Background: Genome analyses have revealed that gene duplication in plants is rampant. Furthermore, many of the duplicated genes seem to have been created through ancient genome-wide duplication events. Recently, we have shown that gene loss is strikingly different for large- and small-scale duplication events and highly biased towards the functional class to which a gene belongs. Here, we study the expression divergence of genes that were created during large- and small-scale gene duplication events by means of microarray data and investigate both the influence of the origin (mode of duplication) and the function of the duplicated genes on expression divergence.

Results: Duplicates that have been created by large-scale duplication events and that can still be found in duplicated segments have expression patterns that are more correlated than those that were created by small-scale duplications or those that no longer lie in duplicated segments. Moreover, the former tend to have highly redundant or overlapping expression patterns and are mostly expressed in the same tissues, while the latter show asymmetric divergence. In addition, a strong bias in divergence of gene expression was observed towards gene function and the biological process genes are involved in.

Conclusion: By using microarray expression data for Arabidopsis thaliana, we show that the mode of duplication, the function of the genes involved, and the time since duplication play important roles in the divergence of gene expression and, therefore, in the functional divergence of genes after duplication.

Figure 1

The duplicated genes of Arabidopsis thaliana were divided into six different subclasses according to the time and mode of duplication (see Materials and methods for details).

Figure 2

Histograms of the Spearman correlation coefficients for anchor points (black) and non-anchor points (grey) for both (a) 3R genes and (b) 1R/2R genes. A Mann-Whitney U test was used to test whether both distributions are significantly different from each other. Mean correlation coefficients: 0.40 for 3R anchor points; 0.32 for 3R non-anchor points; 0.28 for 1R/2R anchor points; and 0.11 for 1R/2R non-anchor points.

Figure 3

Smoothed color density representations of the scatterplots of the (a,b) absolute and (c,d) relative numbers of tissues in which the genes of a duplicated gene pair are expressed, for both (a,c) 3R anchor points and (b,d) non-anchor points. From (a,c) we can conclude that many anchor point genes are both expressed in a high number of tissues, and that many of these tissues are actually identical. On the other hand, (b,d) show that non-anchor point genes frequently show asymmetric divergence because many genes are expressed in a high number of tissues, while their duplicate is not. The plots were made using the 'smoothScatter' function, implemented in the R package 'prada' [69], by binning the data (in 100 bins) in both directions. The intensity of blue represents the amount of points in the bin, as depicted in the legend.

Figure 4

Hypothetical example showing possible scenarios for tissue-specific expression of two duplicates. A black box depicts expression in a particular tissue, whereas a white box represents no expression in that particular tissue. Following duplication of a gene that is expressed in six different tissues, the two copies can (a) both remain expressed in all six tissues (redundancy), (b) diverge asymmetrically, where one gene is expressed in only a small subset of the tissues, while its duplicate remains expressed in the original six tissues, or (c) diverge symmetrically, where tissue-specific expression is complementarily lost between both duplicates. The absolute number of tissues in which a gene is expressed is six for both duplicates in (a) and for the second duplicate in (b), one for the first duplicate in (b) and three for both duplicates in (c). The total number of tissues in which the pair is expressed is 6 in all three cases. The relative number is the fraction of the previous two, and is 1 for the two genes in (a) and for the second duplicate in (b), 0.17 for the first duplicate in (b) and 0.5 for both duplicates in (c).

Figure 5

Scatter plots of the correlation coefficient in function of the K_S value of the gene pairs belonging to different functional classes. The full black line represents the local regression (locfit) line fitted to the data of that particular class, together with its 95% confidence interval (dashed line). (a-b) Gene pairs that have diverged quickly after birth have an intercept of the regression line with the y-axis close to zero; (c-d) whereas slow divergence is reflected by an intercept with the y-axis close to one and a steep slope. (e-f) A more average situation can be observed for most classes. Data of the following classes are displayed: (a) signal transduction; (b) response to external stimuli; (c) macromolecule biosynthesis; (d) structural molecule activity; (e) nucleotide binding; (f) regulation of biological process. Plots of other functional classes of genes can be found in Additional data file 3.