Duplicated genes frequently experience an initial increase in their rate of evolution and nonsynonymous substitutions relative to synonymous substitutions. Moreover, recent analyses of young gene duplicates in several eukaryotic genomes indicate that paralogs exhibit asymmetric rates of sequence divergence in the evolutionary period soon after duplication [11, 16, 20–24]. Together, these observations indicate that initial relaxation of selection, or adaptive evolution, after duplication is limited to one of the paralogs, and that the slower-evolving paralog is more constrained by its ancestral function [11, 22]. The majority of past studies did not distinguish between the ancestral and derived copies within a gene-duplicate pair, which in turn has precluded an unambiguous assessment of which copy is under stringent versus relaxed selective constraints.
There is some evidence that derived paralogs evolve faster than their counterparts residing at ancestral locations. In their study of evolutionarily young rodent gene duplicates, Cusack and Wolfe  assigned ancestral versus derived states to paralogs and demonstrated that genomic relocation of one paralog by retrotransposition engenders rate asymmetry in the sequence evolution of paralogs, commonly manifested as an accelerated rate of sequence evolution in the relocated paralog. Likewise, in bacterial genomes, the majority of paralogs that appear to have moved away from their ancestral gene neighborhood evolved faster than static paralogs . Furthermore, a study of gene duplicates in four mammalian genomes determined that signatures of positive selection were more frequent in the derived copies than genes at their ancestral locations .
In this study, we analysed the rate of evolution in yeast paralogs for which an ancestral versus derived status could be assigned by analyzing synteny as manifested in gene-neighborhood conservation. There was significantly greater gene-neighborhood conservation in ohnologs relative to SSD pairs. Although ohnologs originated from an ancient polyploidization event and rampant genome-wide deletions have since restored functional normal ploidy in these Saccharomyces species [26, 27], it is noteworthy that this extensive gene-neighborhood conservation has persisted. There is no difference in the extent of gene-neighborhood conservation in the upstream and downstream regions of the paralogs for both populations of duplicates (ohnologs and SSD), suggesting, on average, equal rates of preservation/loss of upstream and downstream neighboring genes.
The majority of gene duplicates with low sequence divergence in S. cerevisiae stem from an ancient WGD event rather than segmental duplications. Subsequent to the WGD event, there has been extensive loss of genetic material with an estimated 10% of the original ohnologs remaining . Deletions of genetic material within a WGD-derived homology block have the potential to remove or rearrange regulatory sequences for the remaining genes in the block. Therefore, the DNA sequence of a paralog associated with more extensive gene-neighborhood conservation (i.e. local synteny) might be under stronger purifying selection than a paralog residing in regions that have endured more gene loss and rearrangements. While it is problematic to assign ancestral versus derived states to gene duplicates originating from WGD events, we reasoned that a paralog within an ohnolog pair could be characterized as being ancestral-like or derived-like based on the extent of gene-neighborhood conservation it shared with a single-copy ortholog in an outgroup genome. We then sought to test the hypothesis that ancestral-like gene-copies within ohnolog pairs are more likely to maintain ancestral gene function and therefore exhibit lower rates of sequence evolution. In contrast, gene-copies displaying a reduction in the extent of local synteny relative to the ortholog may be predisposed to accelerated rates of sequence evolution and the resultant fates of neofunctionalization or nonfunctionalization. However, we find no evidence of an association between rate asymmetry in ohnologs and local gene-neighborhood conservation. In other words, for ohnologs, a decline in local gene-neighborhood conservation (derived-like) does not engender accelerated rates of sequence evolution either at the nucleotide or amino acid level. This is in contrast to a study of vertebrate genomes that found a significant correlation between synteny preservation and sequence conservation . We speculate that the greater number of regulatory sites in vertebrate genomes might engender greater sensitivity to syntenic changes relative to yeast. However, ohnologs in yeast do exhibit a strong significant relationship between rate asymmetry and CAI such that the faster-evolving paralogs have lower CAI. The rate asymmetry in ohnologs also seems to be to some degree caused by relaxation of selection for codon usage in one copy.
Among the SSD pairs in our sample, it is the derived copy that evolves faster on average, both at the nucleotide and the amino acid level. This lends credence to Ohno's original hypothesis that duplication enables redundancy, enabling one copy to explore new evolutionary space by accumulating mutations . It is likely that segmental duplications frequently do not capture the full repertoire of regulatory sequences  associated with the ancestral genes and/or result in the insertion of the derived copy into a region of the genome with different chromatin structure and potentially under the influence of different regulatory elements. Under these conditions, mutations that interfere with the ancestral gene's original function would still be selected against, whereas the derived copy could be under relaxed or positive selection. For SSD pairs, the rate asymmetry at the nucleotide level is likely due to a regime of relaxed selective constraints as there is a significant reduction in the CAI of the derived paralogs within SSD pairs. The CAI compares the codon usage of a gene to codon usage in highly expressed genes; hence, the reduction in the CAI values of derived paralogs suggests that selection for optimal codon usage has been relaxed in the derived copy. Puzzlingly, we failed to detect any correlation between nucleotide sequence asymmetry of SSD paralogs and changes in their CAI values. This may stem from limited power given the small sample size of available SSD duplicates in the yeast genome.
If the rate asymmetry in paralogs is largely a consequence of relaxation of selection in the derived paralog, it should also be manifested as different levels of expression among the two copies. Previous work has shown that the evolutionary rate in yeast is strongly influenced by gene expression [29, 30]. In both the yeast ohnologs and SSD pairs studied here, mRNA abundance is correlated with the rate of evolution. Moreover, within SSD pairs, it is the derived paralogs that have lowered mRNA abundance relative to the ancestral loci. Both the CAI and mRNA abundance suggest that selective constraints on gene expression is a significant driver of evolutionary rate asymmetry in paralogs.