The functional importance of co-evolving residues in proteins

Abstract

Computational approaches for detecting co-evolution in proteins allow for the identification of protein–protein interaction networks in different organisms and the assignment of function to under-explored proteins. The detection of co-variation of amino acids within or between proteins, moreover, allows for the discovery of residue–residue contacts and highlights functional residues that can affect the binding affinity, catalytic activity, or substrate specificity of a protein. To explore the functional impact of co-evolutionary changes in proteins, a combined experimental and computational approach must be recruited. Here, we review recent studies that apply computational and experimental tools to obtain novel insight into the structure, function, and evolution of proteins. Specifically, we describe the application of co-evolutionary analysis for predicting high-resolution three-dimensional structures of proteins. In addition, we describe computational approaches followed by experimental analysis for identifying specificity-determining residues in proteins. Finally, we discuss studies addressing the importance of such residues in terms of the functional divergence of proteins, allowing proteins to evolve new functions while avoiding crosstalk with existing cellular pathways or forming reproductive barriers and hence promoting speciation.

Introduction

The term co-evolution refers to the coordinated changes that occur between species or molecules so that their productive interactions are maintained through evolution. In the past two decades, the development of a variety of computational tools has allowed for the identification of co-evolution of many proteins in several organisms [1–5]. The development of such approaches relied on the availability of large experimental databases describing many protein–protein interactions of several model organisms for testing and refining the computational algorithms employed in such studies [6–8]. Indeed, these computational approaches facilitated the prediction of protein–protein interaction networks in under-explored organisms as well as the assignment of function to previously unannotated proteins [2]. In addition, methods for the detection of co-evolution at the residue level, namely the detection of related changes in positions based on sequence alignment within or between proteins, using sequence co-variation analysis were developed. These approaches allow for the identification of co-evolving residues that are often found to be in close spatial proximity in the protein three-dimensional structure [9–13]. These methods also enable the identification of functional residues that can control the activity and specificity of proteins in their cellular environment [14–16]. The large number and variety of computational approaches and their applications for the study of protein co-evolution have been extensively reviewed elsewhere [17–20]. Here, we review studies describing the application of combined bioinformatics and experimental approaches to provide new insight into the structure–function relationships of proteins and their evolution.

Typical multiple sequence alignment of proteins often reveals dramatic variations in residue conservation, allowing for identification of residues that are completely conserved, partially conserved or non-conserved [16, 21]. The importance of completely conserved residues for the function of a protein has been investigated by biochemists and structural biologists for decades, using site-directed mutagenesis followed by in vitro assaying [22]. Such studies made tremendous contributions to our understanding of the basic mechanisms of protein function in the test tube (such as enzyme catalytic mechanisms and protein–protein interactions), revealing the importance of key conserved residues to the protein function. In contrast, partially conserved residues, those that are differentially conserved within particular subfamilies of proteins, can control the co-adaptation of proteins to their native cellular environments [16, 21]. These residues often diverged according to the phylogenetic tree and can fine-tune the activity of proteins, including their binding affinity or catalytic efficiency, their substrate specificity, and/or tolerance to unusual environmental conditions. Currently, it is extremely difficult to identify such residues due to the composition and characteristics of sequence changes in proteins that take place through natural evolution. Thus, advanced computational tools for the identification of these functional residues in proteins can guide biochemists and geneticists to experimentally examine the importance of such residues in reshaping protein activity and specificity. Such tight collaboration between computational and experimental biologists can reveal new structure–function relationships in proteins and identify major divergence in protein function during natural evolution.

Here, we review prominent approaches that combine bioinformatics with experimental tools to explore the functional implications of protein co-evolution. We focus on co-evolution at the residue level and describe recent approaches allowing the identification and examination of specificity-determining residues in proteins that are essential for the function of the protein in the cell. We describe how ancestral sequence reconstruction can be utilized to study the functional divergence of proteins on large evolutionary time scales. In addition, we discuss how co-adaptation of proteins can lead to the generation of new species in nature. Finally, we describe the application of co-evolutionary analysis for predicting the three-dimensional structure of proteins at atomic resolution.

Computational and experimental analysis of functionally important residues

As described above, many protein families contain differentially conserved residues that are important for protein function and adaptation to the complex environment within a cell. These sub-family-conserved residues are not necessarily located in close spatial proximity with one another and thus can be distinct from co-evolving pairs of residues that are in direct interaction. Such residues can significantly affect protein-binding specificity, catalytic activity and allosteric communication. In recent years, several computational methods have been developed to detect such specificity-determining positions (SDPs) in many proteins in an automated manner [15, 16, 23]. These methods are based on different algorithms that combine sequence information and phylogenetic and structural analysis in order to achieve high prediction accuracy of functional sites in proteins (for recent reviews describing many of these computational approaches, see [17, 24, 25]). The detection of SDPs in a target protein sequence can guide site-directed mutagenesis experiments followed by activity assays to identify residues that are directly involved in protein function. One prominent method for the detection of functionally important residues based on sequence conservation is the Evolutionary Trace (ET) method developed by Lichtarge and co-workers [15]. The ET computational algorithm utilizes a phylogenetic tree generated from a multiple sequence alignment of the protein to identify and score conservation in different branches of the tree. This approach allows for the ranking of each amino acid in a protein in a systematic manner according to its relative evolutionary divergence points. High-ranked residues that diverged early in evolution can make an important contribution to the structure and function of that protein [26, 27]. In contrast, low-ranked residues usually have little influence on protein function. Mapping high-ranked residues on the protein structure reveals that such amino acids tend to cluster on the protein surface, allowing for identification of functional regions of the protein [27, 28].

The ET approach can be employed for the identification of functional sites in proteins that are distinct from the active site. For example, Baameur et al. [29] managed to uncover several uncharacterized functional sites of the G protein-coupled receptor kinases 5 and 6 (GRK-5 and -6) that are important for allosteric activation of these proteins. Although previous studies identified specific GRK residues involved in GRK localization and recognition of G protein-coupled receptors (GPCRs) [30–32], none of these studies succeeded in showing the activation mechanism of GRKs by GPCRs. Using ET analysis, Baameur et al. identified three helices in the RH domain of GRK-5 and GRK-6 and experimentally demonstrated their direct effects on GPCR phosphorylation. The ET approach was also employed in an examination of GPCR specificity, in which the focus was on the identification of specificity determinants in bioamine GPCRs [33]. Interestingly, the D2 dopamine and 5-HT2A serotonin receptors exhibit similar functions and present nearly identical binding site structures, yet can still efficiently discriminate between their endogenous and non-endogenous ligands. Rodriguez et al. [33] addressed the question of how these receptors achieve high specificity, using a comparative ET approach followed by a swapping of key residues between the two receptors. The authors identified residues that affect the relative affinities of the ligands as well as other distinct residues that affect ligand-induced receptor activation. Their results highlighted two types of specificity-determining residues in bioamine receptors, with one type of residues controlling the affinity of the ligand and the other dictating the efficacy of GPCR activation by triggering ligand-induced conformational rearrangement of the receptor. This study thus demonstrates the power of the ET approach in identifying key functional residues in GPCRs and illustrates the plasticity of GPCR signaling, enabling a functional switch by introducing only few mutations.

In another case, Adikesavan et al. [34] utilized ET analysis to explore the biological roles of the RecA protein. RecA plays a crucial role in maintaining bacterial genome stability and accordingly participates in a wide variety of processes, including recombination, induction of DNA damage response, and error-prone DNA synthesis [35]. Still, despite extensive studies, the interaction sites of RecA that are essential for promoting these activities remain unknown. Using the ET approach, Adikesavan et al. successfully confirmed the biological role of known functional sites in RecA and identified previously uncharacterized functional surfaces of the protein. Specifically, they identified residues essential for promoting LexA protease cleavage but which are not crucial for RecA DNA damage response activity. These results show that the ET approach can reveal new structure–function relationships and identify distinct RecA functional sites specific for either recombination or the DNA damage response. ET analysis was also utilized for the detection of co-evolution between transcription factors (TFs) and their DNA-binding sequence and was exploited for the rational redesign of TF specificity. Raviscioni et al. [36] performed a large-scale ET analysis of more than 20 TFs together with their respective DNA-binding sites. These authors successfully validated their analysis by swapping top ET-residues between two LRH-1 transcription factors in order to redesign DNA-binding specificity. Overall, this large-scale analysis provided significant evidence for the existence of correlated evolution between a TF and its target DNA molecule and demonstrated that such correlative changes are traceable by the ET method and sufficient for the rational design of binding specificity.

Co-evolution in signaling pathways and protein–protein interaction networks

Signal transduction pathways require a specific recognition between proteins to enable the faithful transfer of information within the cell. In bacteria, two-component signal transduction systems are central for sensing and responding to changes in the environment [37]. These systems comprise receptor histidine kinase that responds to environmental signals by phosphoryl group transfer to cognate regulators to initiate a signal transduction cascade. Bacteria encode dozens to hundreds of two-component systems, allowing for responses to a diverse range of environmental signals by modulating the level of gene expression [38]. Due to the high similarity shared by different two-component systems co-existing in bacteria, it is not clear how these systems achieve the desired specificity for translating a variety of environmental signals into the required cellular response [39]. Indeed, high specificity in these systems is essential to avoid harmful crosstalk between different signal transduction pathways. In recent years, it was shown that histidine kinases and their cognate response regulators exhibit tight co-evolution, leading to specific molecular recognition and phosphorylation activity [39]. Such co-evolution enables efficient insulation of signaling pathways from other co-existing pathways in the same bacteria, leading to faithful transmission of environmental signals. Thus, co-evolutionary changes in proteins can allow the generation of new signaling pathways following gene duplication. In a pioneering study, Laub and co-workers relied on bioinformatics followed by biochemical and genetic analyses to identify and examine residues that co-evolved in a histidine kinase and its response regulator to ensure high specificity [40]. The authors utilized sequence alignments of ~1,300 kinase and response regulator pairs to enable the detection of co-variations between pairs of residues in these proteins. They specifically focused on identifying co-evolving residues located at the interface of the kinase and regulator that can dictate the specificity of the kinase–regulator interaction [40]. Subsequent structural examination of the complex between the kinase and its cognate regulator confirmed that these specificity-determining residues are indeed located at the interface between the proteins [41]. To validate that the predicted specificity-determining residues actually serve such role, Laub and co-workers generated and examined mutants of the EnvZ kinase containing the specificity-determining residues of the RstB kinase. They found that these mutations led to a complete switch in EnvZ specificity such that the mutated EnvZ exhibited no activity toward its natural OmpR regulator, yet showed wild-type levels of activity toward the RstR regulator that is the natural substrate of the RstB kinase (Fig. 1a). In addition, these authors found that only three specificity-determining mutations were sufficient to switch EnvZ specificity toward RstR and that analogous substitutions in EnvZ, identified on the basis of the sequences of other kinases, leads to the recognition of other regulators. In vivo examination of these mutants confirmed that the switch in EnvZ specificity also takes place in living bacteria, leading to novel patterns of gene expression [40]. Further dissection of the contribution of individual mutations to the EnvZ-specificity switch has shown that these mutations do not contribute in an additive manner but rather act in a concerted manner that is dependent on the protein sequence background [42]. Such analysis indicates that a smooth transition in kinase specificity that avoids the generation of inactive intermediates is possible but is probably subject to high selective pressure. Recently, the specificity-determining mutations in kinases were analyzed on a large scale and found to be adaptive, allowing the prevention of cross-talk between paralogous signaling pathways following gene duplication events [43]. In addition, Armitage and co-workers performed structural analysis to rewire a two-component signaling pathway by residue replacements in the response regulator, allowing it to be phosphorylated by a non-cognate kinase [44]. In summary, these studies highlight the importance of co-evolution in reshaping molecular recognition in the bacterial two-component signal transduction system to achieve high in vivo specificity.

Fig. 1.

Bioinformatics and experimental analysis of specificity-determining residues in the bacterial two-component signal transduction pathway (a) and in the fungal proliferating cell nuclear antigen (PCNA)–partner interaction network (b). a In the bacterial two-component signal transduction system, three specificity-determining residues in the kinase EnvZ are sufficient to rewire the system, such that EnvZ kinase containing the specificity-determining residues of RstB (top) can specifically recognize RstA but not the cognate OmpR regulator. Similarly, the EnvZ kinase containing the specificity-determining residues of CpxA (bottom) can specifically recognize CpxR but not the cognate OmpR regulator [40]. Model structures are derived from the Spo0B kinase and Spo0F response regulator (pdb code 1F51) that exhibit high sequence similarity to EnvZ and the response regulators OmpR, RstA, and CpxR [82]. b In PCNA–partner interaction network, specificity-determining residues located in the inter-domain connecting loop of PCNA (red or blue loop) enable recognition of cognate and closely related partners but prevent any binding to distantly related partners. This divergence in the mode of recognition leads to hybrid network incompatibility that can promote and fix speciation in yeast [45]. The inter-domain connecting loop of PCNA is highlighted in red or blue on the PCNA structure (PDB: 3K4X) [83]. The models of the structures were generated using the UCSF chimera program

Correlated mutations in specificity-determining residues were also observed in complex protein–protein interaction networks. Recently, Zamir et al. [45] showed that the proliferating cell nuclear antigen (PCNA)–partner interaction network, responsible for mediating DNA replication and repair in all eukaryotes, exhibits dramatic co-evolution. This work allowed for the identification of a major transition in PCNA–partner interaction specificity during the course of fungal evolution due to correlated mutations in PCNA and its partners. This transition in specificity is characterized by a lack of binding of PCNA from one species to partners from distantly related species. Systematic bioinformatics and experimental analysis of PCNA–partner interaction networks from diverse fungal species showed that these networks diverged early during fungi evolution into two distinct groups due to tight co-evolution (Fig. 1b). The authors found that such dramatic co-evolution has resulted in an incompatibility of hybrid PCNA–partner interaction networks in Saccharomyces cerevisiae, highlighting that co-evolution can form functional barriers between species that can promote and fix speciation.

Intramolecular co-evolution identified by ancestral protein reconstruction

The functional divergence of proteins through evolution can be experimentally examined by the swapping of specificity-determining residues between orthologous proteins from extant species, as described above. This approach can be successful in cases in which these specificity-determining mutations do not significantly affect the folding and/or stability of the protein. However, in many cases, mutations that lead to a novel protein function compromise the stability of the protein and need to be compensated for by stabilizing mutations accumulated during natural evolution [46–49]. In these cases, direct swapping of functional residues between extant proteins would compromise the folding and stability of the protein, leading to its inactivation. To allow for identification of stabilizing and functional mutations that are responsible for the evolvability and functional divergence of proteins, ancestral protein prediction methods followed by experimental examination can be used [50]. Current ancestral prediction methods can infer the maximum likelihood sequence at any lineage of the phylogenetic tree with the highest probability of generating all sequences of extant proteins [51]. Ancestral genes can be synthesized, cloned, and expressed in heterologous systems in order to experimentally characterize their activity and specificity, relative to modern proteins. This approach allows for the introduction of key mutations onto the ancestral background, based on modern protein sequences, permitting dissection of their contributions to the functional divergence of a protein family. In addition, this approach can be utilized to follow the fate of duplicated genes and their functional specialization in extant organisms [52]. Recently, the prediction of ancestral sequences was successfully exploited to guide the generation of mutant libraries for directed evolution applications [53–55].

A prominent example of utilizing ancestral predictions for the study of natural divergence in protein function is provided by the opsins [56]. Opsins are a family of G protein-coupled receptors that absorb light in the visual system of vertebrates [57]. In all opsins, an 11-cis retinal chromophore is covalently bound to a key lysine residue to enable light absorption. Interestingly, different opsins that bind the same retinal can detect a wide range of light spectra with a maximal absorption wavelength (λ _max) ranging from 360 nm in the UV range to 560 nm in the red region [58]. The molecular basis for spectral tuning of different opsins has been extensively investigated over the past two decades, albeit without conclusive results. These studies employed site-directed mutagenesis to swap key amino acids between opsins that absorb light at different wavelength so as to understand how the environment of a given opsin protein tunes the retinal absorption wavelength. These studies showed that mutations can have completely different effects, depending on the opsin sequence background, suggesting the existence of strong coupling between key functional amino acids in opsins and those elsewhere in the protein sequence [59, 60]. Yokoyama and co-workers utilized the predicted ancestral protein of the family of red and green pigments to examine the functional divergence of this family of proteins (Fig. 2). The authors introduced 14 combinations of five functional mutations, previously suggested to control opsin absorption wavelength, into the same ancestral background and examined the absorption wavelength of each variant [61]. They found that different functional mutations generated on the background of the ancestral protein led to shifts in the opsin absorption maxima that were completely consistent with the expected shift in absorption predicted from the extant opsins (Fig. 2). Their results can explain the absorption wavelength of all modern opsins belonging to this family, highlighting the importance of ancestor-based predictions for understanding functional divergence in this important family of proteins [61].

Fig. 2.

Ancestral protein prediction for the study of the molecular basis of spectral tuning by contemporary opsins. The absorption spectra of red and green pigments can be explained by five key residues that change between the different opsins (the “five site” rule, [61]). The effect of single amino acid replacements and multiple replacements on the absorption maxima (λ _max) of opsins was studied using ancestral mammalian opsin [61]. Introducing the Y277F mutation, the S180A and T285A double mutations and the S180A, H197Y, and A308S triple mutation led to significant shifts in the λ _max of the opsins that match the expected shifts in λ _max that are derived based on the “five site” rule. The mutations were mapped on the crystal structure of ligand-free native opsin from bovine retinal rod cells (PDB: 3CAP) [84]. The models of the structures were generated using the UCSF chimera program

Ancestral prediction was also extensively employed to explain divergent evolution in mineralocorticoid (MR) and glucocorticoid receptors (GR). Thornton and colleagues relied on this approach, together with biochemical and structural characterization, to reveal the molecular basis for the divergence of the MR and GR receptors [62–64]. They found that GR specificity evolved over a 40-million-year period by the accumulation of 37 amino acid changes between an ancestor possessing both MR and GR specificity to a more modern ancestor possessing only GR specificity. Detailed analysis of the contribution of these residues to the functional divergence of the receptors showed that five mutations are essential for a narrowing of specificity from that of MR/GR to that of GR alone. Interestingly, three of the mutations that enhanced the specificity of the ancestor protein also destabilized the protein and can be tolerated only due to the accumulation of two permissive mutations in previous ancestral lineages. These permissive mutations that accumulated solely in the GR branch explain why modern MR receptors cannot tolerate the insertion of specificity-determining residues inducing the MR to GR transition in specificity. Overall, these studies highlight the cooperative effects of co-evolving mutations and the importance of the protein sequence background in enabling accommodation of new functional mutations.

Co-evolution as a driving force for speciation

Speciation, the process that leads to the generation of new species from a common ancestor, is one of the fundamental processes in biology [65, 66]. Speciation requires reproductive isolation between species such that hybrid species are either non-viable or sterile. About 80 years ago, Dobzhansky and Muller (DM) proposed a general mechanism for reproductive isolation by genetic incompatibility [65, 66]. This model suggests that genetic incompatibility evolved as a consequence of non-functional interactions between genetic loci that had diverged in two different species. According to this model, a pair of genes that diverged in two populations is sufficient to trigger the incipient stages of speciation. Interestingly, the DM model claims that such non-compatible pairs can initially evolve in two isolated populations by independent changes without a decrease in cell fitness. This initial stage of divergence can be then followed by reciprocal changes in the protein partner, leading to intra-species co-evolution between the protein pairs (Fig. 3). For many years, such gene pairs were elusive and studies on speciation revealed more global mechanisms of speciation, including chromosomal rearrangement and divergence of DNA sequences. Only in the past decade have DM pairs been identified in several different organisms through extensive genetic screens [67]. Prominent examples are DM pairs identified in closely related yeast species due to nuclear and mitochondrial gene incompatibility [68]. In a systematic study, Leu and coworkers constructed hybrid lines of Saccharomyces cerevisiae (Sc) and Saccharomyces bayanus (Sb) in which one or two chromosomes were derived from Sb and the rest from Sc [69]. Following extensive screening, this approach revealed incompatibility between the Sb-Aep2 nuclear protein and Sc mitochondria. Specifically, it was shown that Sb-Aep2 is unable to regulate the translation of the Sc-OLI1 mRNA, probably due to a failure of the Sb-Aep2 to recognize the 5′-untranslated region (5′-UTR) of Sc-OLI1. Further support for co-evolution between the Aep2 protein and the 5′-UTR of OLI1 comes from comparison of the protein sequences and the OLI1 5′UTR in the two species. The authors suggested that AEP2 and OLI1 co-evolved in S. bayanus during its adaptation to growth on non-fermentable carbons sources and that this adaptation plays an important role in promoting speciation. Further genetic screens for the identification of genes leading to nuclear-mitochondrial incompatibility between three closely related yeast species led to the identification of three more speciation genes [70]. These and additional studies in several other model organisms, including Drosophila and mice, have detected pairs that co-evolved independently in two closely related species, leading to speciation (for review, see [67]). These findings demonstrate that co-evolution between proteins can be a strong driving force in promoting and fixing speciation.

Fig. 3.

The contribution of co-evolution of a pair of proteins to speciation in accordance with the two-locus genetic model of reproductive isolation proposed by Dobzhansky and Muller [65, 66]. For simplicity, haploid populations of fungi are shown as an example of the speciation process. When the ancestral population of a haploid with a genotype AB split into two, A evolves to a° in one population while B evolves to b⁺ in the other (step I). Further accumulation of correlated mutations in the two independent populations can take place due to co-evolution between the genes, such that A evolves to a⁺ in one population while B evolves to b° in the other (step II). Co-evolving residues are often found in close spatial proximity to allow maintaining the interaction between the proteins. Upon mating between the haploids, the diploid hybrid is formed (step III). Following sporulation, partial hybrid sterility is caused by incompatibility between a⁺ and b° and a° and b⁺, leading to 50 % non-viable spores (step IV). During this process, hybrid sterility evolves between species due to neutral drift or adaptation of separated populations without ever appearing within a species. Thus, neither independent population needs to cross an adaptive valley with an accompanying loss of fitness upon accumulation of these mutations

Predicting protein structure based on evolutionary coupling between residues

The prediction of protein structure from its primary amino acid sequence has been a major challenge for biochemists and molecular biologists for the last three decades. Three-dimensional models of target proteins of unknown structure were mainly obtained based on the similarity of a target protein to another protein with a solved three-dimensional structure [71]. However, accurate structural models of a protein without any structural information on a homologue are much harder to achieve. Approaches for the de novo prediction of protein structure, such as Rosetta, are based on dividing the target protein sequence into short fragments, followed by the assembly of the structures of these fragments using empirical energy functions [72]. These approaches are computationally demanding and proved successful mainly in the structural predictions of small proteins. In recent years, significant advances in the prediction of protein structure without any prior structural information have been realized through the use of evolutionary information [73]. Large sequence alignments of a target protein enable the identification of co-variations in pairs of amino acids that can form direct intra-molecular interactions. The minimal requirement for such co-variation analysis is sequence alignment containing hundreds of different sequences [73] which are not available for all protein families. The information on coupling between pairs of amino acids within the target protein is then translated to distance constraints to predict the three-dimensional structure of proteins with high accuracy [73]. Several groups have recently shown that this approach provides sufficient residue contact information to enable the prediction of protein structures with atomic resolution [74–78]. While such global approaches for structure prediction of proteins based on evolutionary information only recently became available, basic approaches for the detection of amino acid contacts within proteins had already been developed two decades ago [10, 79, 80]. However, the major advantage of the current approach for predicting protein structure is its ability to filter out false positive interactions within proteins. This approach treats correlated pairs of residues as being statistically dependent on other pairs, yielding high scores only for those pairs of residues that are likely to be involved in true interactions. The practical application of this approach for the structural prediction of proteins is based on three major steps, beginning with the identification and alignment of a large number of evolutionary related sequences, followed by the calculation of the covariance matrix to provide scores for all possible residue pair relationships across evolution. Finally, the distances between pairs of residues with high covariance scores are constrained using standard distance geometry algorithms to calculate the three-dimensional structure of the target protein [73]. Currently, algorithms for calculating protein structure from distance constrains are used extensively when solving structures from NMR experimental data [81].

Recently, this global statistical approach was successfully used by three different groups to predict the structure of a large number of globular and membrane proteins with high accuracy. Marks et al. developed a global approach termed EVfold to predict the structure of 15 diverse globular proteins of up to 220 amino acids with an overall accuracy of 2.8–5.1 Å root mean square (RMS) deviation of the main cα chain [75]. In a more recent work, the same approach was used to predict the structure of 25 membrane proteins with up to 487 residues from diverse protein families, including important G protein-coupled receptors and transporters [74]. Two other groups used similar global statistical approaches, termed FILM3 [76] and DCAfold [77], to predict the structures of membrane proteins and bacterial protein domains with high accuracy. Such approaches are extremely promising for a deepening of our knowledge of the structure–function relationships of proteins.

Concluding remarks

In summary, we have described how computational and experimental approaches can be effectively combined to provide new insight into the structure, function and evolution of proteins. We have reviewed several studies showing the usage of computational tools to predict functional residues in proteins, followed by experimental examination of these predictions. These studies facilitated the discovery of specificity-determining residues in proteins that allowed the divergence of protein function throughout the course of evolution. These studies also highlighted the importance of specificity-determining residues for the adaptation of proteins to the complex cellular environment. In addition, we described the usage of ancestral sequence reconstruction in order to follow the accumulation of key mutations during natural evolution. Overall, we believe that the combination of experimental and computational approaches can be highly useful for deepening our understanding of the function of proteins in the test tube and in the cell.