The mechanisms underlying the emergence of novel functions in nature remain a mystery. Gene duplication is believed to be the primary source of new genes and functions and has consequently been credited with great evolutionary importance 
. Our knowledge on the importance of duplication in functional innovation is impressive, yet our ability to model the functional fate of duplicated genes is highly limited.
A number of studies have attempted to establish a causal link between gene duplication and the emergence of major evolutionary innovations. For example, most Angiosperms have undergone at least one genome duplication (polyploidy) 
in the Creataceous era, contemporary with the explosion of plant metabolic and physiological diversity 
. This diversity resulted from the expansion of protein families by gene duplication, including pepsin- and subtilisin-like proteases 
, metacaspases 
, regulatory genes 
and developmentally important MADS-Box genes 
. In animals, although much rarer, gene duplications have preceeded the appearence of key developmental features and were concomitant with major events of species diversification 
. It is tempting to establish a link between gene duplication and biological complexity, but the mechanisms underlying the persistence of genes in duplicate and determining their functional fate remain largely obscure.
Population genetics theory predicts that most duplicated genes return to single copies “shortly” after duplication because an entirely redundant duplicate will fix deleterious mutations and undergo decay and erosion after duplication 
. Following this prediction, genes will persist in duplicate in the genome if: (i) gene duplication, hence redundancy, endows organisms with mutational robustness 
; (ii) there is selection for increased gene dosage 
; or (iii) gene duplicates have diverged functionally through the partitioning of the ancestral gene functions 
, thereby generating entirely new functions 
, or have diverged in their expression profiles 
. Because gene dosage is immediately unbalanced after duplication, other factors or mechanisms should counterbalance such a constraint to mediate the persistence of genes in duplicate 
. These mechanisms remain hitherto a major question in molecular evolution 
The mode of gene duplication (WGD or SSD) has been proposed to have a key role in the fate of duplicated genes 
(although see 
for the role of species ecology in the functional fate of duplicates), with WGDs being more likely to persist than SSDs, as the former does not upset the stoichiometric balance in the cell 
. Long-term survival of WGDs in the genome can offer opportunities to generate novel functions, albeit this is constrained by gene dosage balance. Therefore, whether genes and their products resulting from both WGD and SSD are subject to the same evolutionary constraints and have similar potential to generate novel functions is unclear.
Typically, events of functional divergence between duplicated genes can be inferred using evolutionary parameters, assuming that when the protein sequences of duplicates are more divergent so are their functions 
. However, determining whether two copies of a duplicated gene have identical, similar or different functions requires the concerted and careful examination of the function of each gene product. While this approach is useful at a single gene level, genome-scale analyses of functional divergence between gene duplicates are unfeasible on a gene-by-gene basis. Alternatively, high-throughput methods, such as genetic interactions screening 
, yeast two-hybrid screening 
and analysis of protein complexes by mass spectrometry 
provide substantial information that can aid in testing the roles of WGD and SSD in innovation.
Using such high-throughput information, several authors have contributed to the understanding of the role of the modes of gene duplication in the functional divergence of duplicated genes. For example, Wagner analyzed the number of shared interactions between duplicated genes in a network as a crude measure of their functional overlap 
. Analysis of various types of networks on a large scale led Conant and Wolfe 
to the observation of asymmetry and partitioning of genetic interactions (sub-functionalization) between the daughters of genes after WGD in the yeast S. cerevisiae
. The different contribution of WGD and SSD to functional divergence was also pointed out in another study using information on protein interactions 
. Finally, Hakes and colleagues 
used protein interactions and Gene Ontology (GO) annotations as proxies for protein function to show that functional divergence between SSDs is greater than between WGDs, WGDs produce less deleterious effects when deleted and WGDs are usually part of the same protein complexes.
Recently, Costanzo and colleagues 
have constructed a functional map that includes the genetic interaction profiles (epistasis) for approximately 75% of the genes in S. cerevisiae
. Two genes are considered to interact when the phenotypic effect of a variant of one gene is aggravated (synergistic or negative epistasis) or alleviated (antagonistic or positive epistasis) by variation in the second gene 
. In the extreme, these combinations can lead to synthetic lethality in which mutation of a single gene, although having little or no effect on the cell in isolation, results in cell death when combined with a mutation in a second gene 
. These interaction profiles provide a means to identify functional relationships between duplicated genes. Accordingly, VanderSluis and colleagues 
used genetic interaction profiles to demonstrate that duplicated genes can be functionally redundant, show subtle functional differences, their persistence depends on their dosage and gene copies can show asymmetry in their interaction profiles. Moreover, Jiang and colleagues 
unearthed the role of gene duplication in the evolution of genetic interaction networks and in mediating functional diversification of the interaction partners of a duplicate.
Despite their insightful findings, a model that describes the contribution of the mode of gene duplication to innovation is lacking. More precisely, the different propensities of WGDs and SSDs to generate novel functions that depart from the ancestral ones remain to be inferred.
We used the genetic interaction dataset of Costanzo and colleagues and a large-scale evolution experiment across which we examined mutational dynamics in duplicated genes formed by SSD and WGD. Exhaustive analysis of interaction profiles and genome-wide mutational dynamics allowed us to distinguish the role of WGD and SSD in the functional specialization of S. cerevisiae genes and shed light on the complexity of the dynamics of evolution by gene duplication. In particular, we show that: (a) SSDs establish more functions and have stronger epistatic effects in the cell than WGDs; (b) SSD is often followed by neo-functionalization while sub-functionalization is likely to follow WGD and (c) we propose and test a model that explain the role of the mechanism of duplication in the functional fate of duplicates.