The genomes of all eukaryotes have undergone at least one round of whole-genome duplication (WGD) (see Glossary) during their evolutionary history [1,2]. Duplicate genomes can undergo massive gene loss and genomic rearrangements, leading to a diploidized state, as shown in yeast [3], Arabidopsis [4-7] and rice [8]. During evolution one copy of the gene duplicate can be lost by the accumulation of deleterious mutations [9]. Evidently, many duplicate genes are retained because the redundancy conferred by duplicate genes might facilitate species adaptation [1] and genetic robustness against null mutations [10]. Both copies can be retained if a higher dosage is advantageous [11], or the function of the duplicate can diverge from that of the ancestral gene by subfunctionalization [12] (such as tissue specificity). Alternatively, one gene duplicate can evolve to possess a novel function by neofunctionalization [1,13].

To evaluate the relative contributions of environmental and developmental factors to expression divergence between duplicate genes, we analyzed the Pearson correlation coefficient of expression (Box 1) between gene duplicates in developmental (R_dev) or environmental (R_env) processes using the same number of expression data sets: 63 in different developmental stages and 63 treatment and time-course combinations in roots and shoots, respectively. The distributions of expression correlation coefficients were compared using the Wilcoxon rank-sum test [20]. As expected, correlation coefficients of expression profiles between randomly chosen genes showed a normal distribution with mean zero, and there was no significant difference in expression variation among random gene pairs in all three conditions (data not shown). Interestingly, the expression divergence of duplicate genes under environmental stress is significantly greater than that under developmental process (Figure 1a, P ≤ 2.2 × 10⁻¹⁶). Furthermore, we analyzed the correlation coefficient difference (D) of each gene duplicate in developmental and environmental processes (D_i = R_dev,i – R_env,i, for the ith duplicate gene pair). The cumulative probability difference for the gene duplicates and random gene pairs between environmental and developmental processes was significantly different by either the Kolmogorov–Smirnov (KS) test [21] or the Wilcoxon rank sum test (see Figure S2 in Online Supplementary Material). Taken together, the data indicate that expression divergence between gene duplicates occurs faster in response to the environmental stresses than to the developmental changes.

Duplicate genes display greater levels of expression diversity than do random gene pairs in response to external and internal processes. However, the duplicate genes involved in developmental processes tend to be coregulated, whereas the duplicate genes involved in abiotic and biotic stresses tend to diverge in expression (Figure 2). There is experimental support for the above conclusion. For example, SEP1 (formerly AGL2) and SEP2 (formerly AGL4) are gene duplicates expressed at the flower developmental stage [23], and their expression patterns are correlated throughout plant development (Figure 2, right lower panel). The two genes have a redundant function in the floral organ identity, and single-gene knockout shows no developmental defect [24]. By contrast, cyclophilin gene duplicates, At2g21130 (CYP1) and At4g38740 (ROC1), are induced by abiotic and biotic stresses (Figure 2, left lower panel) [25,26]. Their expression levels differ greatly in response to various external stimuli, suggesting that the gene duplicate is involved in different regulatory networks. Although some extreme examples of duplicate genes that show expression similarity or divergence exist in large data sets, the experimental data collectively support the above notion.