De novo active sites for resurrected Precambrian enzymes

ARTICLE Received 12 Jan 2017 |Accepted 30 May 2017 |Published 18 Jul 2017 De novo active sites for resurrected Precambrian enzymes Valeria A. Risso1, Sergio Martinez-Rodriguez1, Adela M. Candel1, Dennis M. Kru¨ger2, David Pantoja-Uceda3, Mariano Ortega-Mun ˜oz4, Francisco Santoyo-Gonzalez4, Eric A. Gaucher5, Shina C.L. Kamerlin2, Marta Bruix3, Jose A. Gavira6& Jose M. Sanchez-Ruiz1 Protein engineering studies often suggest the emergence of completely new enzyme functionalities to be highly improbable. However, enzymes likely catalysed many different reactions already in the last universal common ancestor. Mechanisms for the emergence of completely new active sites must therefore either plausibly exist or at least have existed at the primordial protein stage. Here, we use resurrected Precambrian proteins as scaffolds for protein engineering and demonstrate that a new active site can be generated through a single hydrophobic-to-ionizable amino acid replacement that generates a partially buried group with perturbed physico-chemical properties. We provide experimental and computational evidence that conformational flexibility can assist the emergence and subsequent evolution of new active sites by improving substrate and transition-state binding, through the sampling of many potentially productive conformations. Our results suggest a mechanism for the emergence of primordial enzymes and highlight the potential of ancestral reconstruction as a tool for protein engineering. DOI: 10.1038/ncomms16113 OPEN 1Departamento de Quimica Fisica, Facultad de Ciencias University of Granada, 18071 Granada, Spain. 2Science for Life Laboratory, Department of Cell and Molecular Biology, Uppsala University, BMC Box 596, S-751 24 Uppsala, Sweden. 3Departamento de Quimica Fisica Biologica, Instituto de Quimica Fisica Rocasolano, CSIC, c/Serrano 119, 28006-Madrid, Spain. 4Departamento de Quimica Organica, Facultad de Ciencias University of Granada, 18071 Granada, Spain. 5School of Biology, School of Chemistry and Biochemistry, Parker H. Petit Institute for Bioengineering and Biosciences, Georgia Institute of Technology, Atlanta, Georgia 30322, USA. 6Laboratorio de Estudios Cristalograficos, Instituto Andaluz de Ciencias de la Tierra, CSIC-University of Granada Avenida de la Palmeras 4, Granada, 18100 Armilla, Spain. Correspondence and requests for materials should be addressed to S.C.L.K. (email: [email protected]) or to J.M.S.-R. (email: [email protected]). NATURE COMMUNICATIONS | 8:16113 | DOI: 10.1038/ncomms16113 | www.nature.com/naturecommunications 1 The generation of completely new active sites capable of enzyme catalysis is, arguably, one of the most fundamental unsolved problems in protein science. Rational design approaches to this problem have often used complex computational methods, have targeted simple model reactions and have typically led to low levels of catalysis1. These studies would seem to suggest, therefore, the unlikelihood of the emergence of completely new active sites in non-catalytic scaffolds. Certainly, most modern enzyme functions likely evolved from previously existing functionalities. On the other hand, most biochemical processes are extremely slow in the absence of enzymes2and specialized enzymes are likely to have catalysed many different reactions already in the last universal common ancestor3,4.It could be inferred from this that efficient mechanisms for the emergence and subsequent evolution of completely new enzyme functionalities must exist or, at least, that they must have existed at the primordial protein stage. However, little is known about such mechanisms. Buried and partially buried ionizable groups with perturbed properties often play essential catalytic roles in modern enzymes. Single hydrophobic-to-ionizable residue mutations that generate partially buried groups with perturbed properties may have plausibly provided a feasible route to the generation of completely new active sites. This notion has been previously proposed5,6 but never tested in practice. Conformational flexibility could have assisted the emergence of the new enzyme functionalities and its subsequent evolution by facilitating substrate and transition-state binding, through sampling a greater number of potentially productive conformations. Many years ago, Jensen proposed in a highly influential article7that primordial enzymes were capable of catalysing a diversity of reactions. It is conceivable that the conformational flexibility that is likely linked to such a broad generalist nature8–10 may have facilitated the emergence of new enzyme functionalities in the first place. Here, we explore and test these notions using resurrected Precambrian b-lactamases11,12 as scaffolds for the engineering of completely new active sites. To date, only a handful of systems have been studied using ancestral resurrection, and only a few of these resurrection efforts targeted ‘old’ (B3 billion years) phylogenetic nodes (Fig. 2 of ref. 13). Of these systems, resurrected Precambrian b-lactamases have been thoroughly characterized in terms of their structure, function and stability11. These putative ancestral proteins have been shown to be highly stable and able to efficiently degrade several antibiotics11. Previous9and current computational analyses, as well as new NMR relaxation studies reported here, support that this substrate promiscuity is linked to enhanced conformational flexibility. The broad substrate scope of resurrected Precambrian b-lactamases may reflect the wide variety of substances these enzymes had to hydrolyse11,12. We do not claim, therefore, that they are necessarily at Jensen’s ancestral generalist stage, although, strictly speaking, this possibility should not be ruled out. In any case, the available resurrected Precambrian b-lactamases11 should provide an adequate model with which to address the role of ancestral conformational flexibility in the emergence of new enzyme functionalities. More generally, we have recently reviewed14 several arguments and recent publications that support that promiscuity may be a common outcome of ancestral protein resurrection. Here, we use carefully selected systems that span a vast region of the sequence space of both ancestral and modern b-lactamases, and probe the enzymatic features that allow for the emergence of a non-natural activity in these enzymes, as well as why it only appears in a specific snapshot of evolutionary time. We demonstrate the role of conformational flexibility in allowing for the emergence of new enzymatic functions, as well as its importance in the subsequent evolvability of the enzyme. Finally, our data highlights the potential of ancestral reconstruction as a tool for protein engineering by providing far more powerful evolutionary starting points than can be obtained from modern enzymes. Results Selected model systems. In the present study, we have used the proteins encoded by the most probabilistic sequences at six Precambrian phylogenetic nodes in the evolution of class A b-lactamases (Fig. 1). The reconstructed sequences and the procedure used to obtain them have been described in detail in Supporting Information of ref. 11. These proteins display large sequence differences between themselves (Supplementary Table 1) and they are properly folded, highly stable, active and share the b-lactamase fold11. In addition, we have also used here proteins encoded by alternative sequences at the GNCA node (common ancestor of Gram-negative bacteria). As is custormary in the field, these alternative sequences were derived11,13 from a random weighted sampling of the posterior probability distribution. They differ from the most probabilistic sequence at 8–20 positions (Supplementary Tables 2 and 3). Finally, for comparison, we have also used 10 modern b-lactamases (Fig. 1) that have been considered in the literature to be archetypical examples of b-lactamases and are well characterized in terms of their structure and function15. These modern proteins provide a fair representation of the b-lactamases from the several bacterial taxa (Fig. 1) and they span modern b-lactamase sequence space to a substantial extent, as they show limited sequence identity between themselves (Supplementary Table 4). Note also their limited sequence identity with the putative ancestral b-lactamases studied here (Supplementary Table 5). Overall, we explore in this work a vast region of the sequence space of both ancestral and modern b-lactamases. We have selected the Kemp elimination reaction (Fig. 2) as our primary target for the reaction to be catalysed by the generated new active site, for several reasons. First, it provides a simple activated model for proton abstraction from carbon, which is a fundamental chemical process that underlies many biochemical reactions. Completely new active sites are expected to display low catalysis levels (high catalysis levels would be the outcome of subsequent evolution) and their emergence is best probed by activated substrates. Second, Kemp elimination is a non-natural reaction (unknown in biological organisms). No natural enzyme has evolved, therefore, to catalyse this reaction16 and studies on engineered Kemp eliminases are unlikely to be compromised by contamination from natural enzymes. Following from this, Kemp elimination has been often used as a benchmark for rational enzyme design1. Therefore, use of the same system allows for a quantitative comparison of the catalytic efficiencies of the designed constructs presented in this work to those generated in previous enzyme design studies. Generation of a new active site for Kemp elimination. The possibility that slow conformational changes (on the microseconds to seconds timescale) play roles in enzyme catalytic cycles has been proposed and explored17. Consequently, we used NMR relaxation determinations (see Methods section and Supplementary Methods) on the b-lactamase encoded by the most probabilistic sequence at the GNCA node (GNCA MP b-lactamase) to guide our design of a new active site. This putative ancestral protein displays a large number of residues with conformational contribution to the relaxation rates determined by NMR (see the comparison with the modern TEM-1 b-lactamase shown in Fig. 3). A large accumulation of ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16113 2NATURE COMMUNICATIONS | 8:16113 | DOI: 10.1038/ncomms16113 | www.nature.com/naturecommunications such residues is visually apparent in the region encompassing helix h1 (residues 26–41), helix h11 (residues 271–290) and the loops 225–229 and 252–257. Residue 229 within this region appears as a suitable target for the generation of a new active site through a hydrophobic-to-ionizable residue replacement. A substantially buried and highly conserved tryptophan residue is present at position 229 in both modern and reconstructed Precambrian b-lactamases. The indole side chain of tryptophan has a shape similar to that of the Kemp substrate. Precisely because of this shape congruence, replacement of W229 to a new residue with basic properties will not generate a new active site capable of catalysing the Kemp elimination unless, of course, conformational rearrangements occur in the enzyme to be able to avoid steric clashes between the substrate and the new residue. We found that a simple W229D replacement leads to substantial levels of Kemp elimination activity in the b-lactamases at all ancestral nodes studied here (Supplementary Table 6), with the only exception of the evolutionary recent ENCA node (Fig. 1). By substantial activity, we mean that the observed levels of Kemp elimination activity were clearly distinguishable (and much higher in most cases) than the background (enzyme-free) levels (Supplementary Fig. 1). In contrast, the W229D variants of all 10 modern b-lactamases studied led to levels of Kemp elimination activity that, even at protein concentrations of about 20mM, could be barely distinguished from the background levels (Supplementary Fig. 1). It is important to note that the W229D variants of all 10 modern b-lactamases studied did show antibiotic degradation activity (linked to the natural active site) at nM concentrations (Supplementary Fig. 2). Therefore, their lack of Kemp elimination activity cannot be attributed to the ‘disruptive’ W229D mutation preventing their folding, which is anyhow a problem that is less likely to arise with the highly stable ancestral b-lactamases. It is worth noting here that enhanced stability is a common outcome of Precambrian protein resurrection11,18–20, likely linked to the thermophilic nature of early life. A wide diversity of experimental results confirm (or are consistent with) the main features of the designed approach used. Specifically, X-ray crystallography in the presence of 5(6)- nitrobenzotriazole, a known transition-state analogue of the Kemp elimination reaction (Fig. 4) and inhibition by this analogue (Supplementary Fig. 3) support that Kemp elimination does occur at the site generated by the W229D replacement. The catalytic role of the aspartate at position 229 is further confirmed by mutational studies: the W229G variant shows negligible activity and the nearby aspartate at position 228 does not have a catalytic role, as the replacement of D228 with A does not substantially impair the activity (Supplementary Table 6). The catalytic role of the aspartate at position 229 is also confirmed by the pH dependence of the catalysis, which is consistent with the raised pKvalue expected for an aspartate residue in a hydrophobic environment (Fig. 5). Following from this, 3D-structure determination in the presence of FCA Actinobacteria S. aureus (blaZ) M. fortuitum (blaF) Firmicutes S. albus (SABLA) B. licheniformis (BEPEN) PNCA GPBCA Pseudomonas P. aeuriginosa (Per-1 and PSE-4) GNCA 8 proteins ENCA Enterobacteria E. cloacae (Nmc-A) P. vulgaris (Bla-B) E. coli (TEM-1 and Toho-1) 32 Time before present (billion years) 1 Today AFCA Figure 1 | Ancestral and modern b-lactamases used as scaffolds in this work. Schematic representation of the phylogenetic tree for class A b-lactamases11. The ancestral nodes studied in this work correspond to the common ancestors (CA) of Firmicutes (FCA), Actinobacteria and Firmicutes (AFCA), Enterobacteria (ENCA), Gammaproteobacteria (GPBCA), various Gram-negative bacteria (GNCA) and various Grampositive and Gram-negative bacteria (PNCA). The proteins encoded by the most probabilistic sequences at these nodes11 were prepared and used as scaffolds for engineering. In addition, seven alternative reconstructions at the GNCA node (Supplementary Tables 2 and 3) were also prepared; these are labelled GNCA1to GNCA7, with the most probabilistic sequence at the node being labelled GNCA MP .b-lactamases from 10 modern organisms are also studied in this work and these organisms are shown at the right. N O O2N H B O2N B H N O δ+ δ– CN O– O2N BH N N N O2N H N H NH2 COOH a bc d ‡ pK = 17.0 N pK = 6.3 O O2NO e OH O2N +HO O OH2 + Figure 2 | Reactions and compounds studied in this work. (a) The Kemp elimination of 5-nitrobenzioxazole (5-nitro-benzo[d]isoxazole). A schematic transition state structure is shown here; (b) tryptophan; (c) 5(6)-nitrobenzotriazole (a transition-sate analogue) and (d) indole (the tryptophan side chain). Finally, (e) the hydrolysis of p-nitrophenyl acetate is also shown here. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16113 ARTICLE NATURE COMMUNICATIONS | 8:16113 | DOI: 10.1038/ncomms16113 | www.nature.com/naturecommunications 3 the transition-state analogue (Fig. 4) confirms the role of conformational flexibility in the generation of a new function, as transitionstate binding is shown to rely on conformational rearrangements. Specifically, the bound transition-state analogue is displaced with respect to the position originally occupied by the tryptophan 229, as required by the presence of an aspartate residue at position 229 in the active variants. Such a displacement is made possible by a shift of the h11 a-helix (and the concomitant shift of the substantially solventexposed h1 a-helix). We note also that the catalytic efficiency for the Kemp elimination in the engineered ancestral proteins correlates with the transition-state analogue binding constants derived from inhibition experiments (Supplementary Fig. 4), supporting that catalysis is linked to transition-state stabilization. Finally, catalysis of Kemp elimination is enhanced by an additional F290W amino replacement in the neighbourhood of position 229, which likely stabilizes the transition state through a face-to-edge interaction with the new tryptophan residue (Fig. 4). Comparison with previous rational designs. Kemp elimination is a non-natural reaction that is unknown to biological organisms. No enzyme is, therefore, expected to have evolved to catalyse this a b c d e h1 h11 W229 D229 TS analogue h1 h11 W229 D229 TS analogue h1 h11 W229 L225 P226 I250 D229 I259 P252 TS analogue F290 V287 V286 V261 V48 R256 L225 P226I250 D229 I259 P252 TS analogue W290 V287 V286 V261 V48 R256 P254 P254 Figure 4 | 3D-structures of the de novo Kemp eliminases. (a,b) Shown here are the structures of the W229D variant of GNCA MP b-lactamase and the W229D/F290W variant of GNCA4b-lactamase, respectively. In both cases, the structures with 5(6)-nitrobenzotriazole, a transition-state analogue, bound at the de novo active site (orange-coloured structures with analogue in red) are superimposed with the 3D-structures of the corresponding GNCA MP and GNCA4 backgrounds (light-blue coloured structures with W229 in darker blue). The shift of the h1 and h11 a-helices is apparent in a, but not in b.The reason for this is that the h1 and h11 helices are already displaced in the background GNCA4variant, as shown by the superposition of the two backgrounds in c(GNCA MP is shown in grey and GNCA4is shown in light green). (d,e) blow-ups of the new active site region in the structures of the W229D variant of GNCA MP b-lactamase (d) and the W229D/F290W variant of GNCA4b-lactamase (e) with the bound transition-state analogue. h11 h11 h1 h1 ab W229 W229 Figure 3 | NMR relaxation studies on modern and ancestral b-lactamases. The 3D-structures of (a) TEM-1 b-lactamase (PDB 1BTL) and (b) the b-lactamase encoded by the most probabilistic sequence at the GNCA node11 (PDB 4B88) are displayed. The residues for which the relaxation rates cannot be explained without including a conformational exchange contribution (ref. 31 and this work) are highlighted in red. The h1 and h11 a-helices are labelled. The tryptophan residue at position 229 is highlighted in blue. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16113 4NATURE COMMUNICATIONS | 8:16113 | DOI: 10.1038/ncomms16113 | www.nature.com/naturecommunications reaction and, consequently, Kemp elimination activity is not expected among natural proteins. Indeed, Hilvert and coworkers21 found no detectable Kemp eliminase activity in calmodulin, barnase, lysozyme, trypsin, chymotrypsinogen and chymotrypsin. Furthermore, Tawfik and coworkers16 screened the ASKA library (B4,300 clones) for natural and non-natural activities and found only two enzymes that could catalyze Kemp elimination, although these appeared to be promiscuous activities. On the other hand, serum albumins can catalyse Kemp elimination22 with a turnover number that is about 102s1 at neutral pH, but that approaches 10 s1at pH B10. However, serum albumins, even in the absence of bound metals, can catalyse a remarkable diversity of non-natural reactions23, due to hydrophobic pockets, which can bind different substrates and to the presence of lysine residues with catalytic properties within those pockets. The capability of albumins to catalyse the Kemp elimination reaction (among several other non-natural reactions) suggests that the scarcity of natural enzyme catalysts for such a simple reaction is simply due to the lack of a selective pressure to generate Kemp elimination activity during evolution. On the other hand, rational design efforts to engineer artificial Kemp eliminases have resulted in limited success1, despite the simplicity of the targeted reaction. Indeed, comparison with these previous Kemp eliminase designs (Fig. 6) provides further evidence that the catalytic efficiencies (k cat / K M values) we have generated in the ancestral b-lactamases are indeed substantial and consequential. The lower end of the variation range spanned by the W229D and W229D/F290W variants of the 12 successful ancestral backgrounds used is actually similar or clearly above the results previously reported using minimalist design24,25. The upper end of the range is above the results previously reported using more complex design approaches26,27 (iterative design and design based on Rosetta), which also involved large numbers of mutations to reach those efficiencies, and it is less than two orders of magnitude below the catalytic efficiency for the best Kemp eliminase reported for to date28 (which is the outcome of 17 rounds of directed evolution from an iterative design background). Regarding the turnover numbers, the k cat values obtained for the W229D/F290W variants of the ancestral b-lactamase scaffolds are up to approximately seven orders of magnitude above the rate of the uncatalysed reaction (Fig. 6). For comparison, the best artificial Kemp eliminase reported to date28 displayed a approximately nine orders of magnitude enhancement, but, as noted above, this was obtained as a result of 17 rounds of directed evolution on a designed background with already substantial activity. Furthermore, the upper end of the range spanned by our k cat values is clearly above the reported k cat values for complex designs that required much larger numbers of mutations (Fig. 6). Michaelis plots for the single W229D variants are linear and do not allow k cat values to be calculated. However, lower limit estimates of k cat can indeed be derived from suitable analysis of linear Michaelis plots (see Supplementary Fig. 5 for details). These estimated lower limit values for the single W229D variants are shown with open circles in Fig. 6. Remarkably, they are similar to the actual k cat determined for the double W229D/ F290W variants. Therefore, the k cat enhancement of up to seven orders of magnitude over the uncatalysed reaction is actually produced by the single W229D mutation, while F290W mostly favors substrate and transition-state binding. 4.0 ab c 3.5 3.0 2.5 2.0 log10 (kcat KM –1/M–1 s–1) log10 (KM/M–1)log10 (kcat/s–1) kcat= 12± 2 s–1 kcat/Km =5497± 600 M–1 s–1 pKa= 6.97±0.13 pKa= 6.71±0.08 pKa= 6.14±0.17 pKa= 6.23±0.09 1.5 –2.5 1.0 0.5 0.0 –3.0 –3.5 6.0 7.0 8.0 pH 9.0 6.0 7.0 8.0 pH 9.0 6.0 7.0 8.0 pH 9.0 Figure 5 | pH dependence of the de novo Kemp eliminases. (a) Plot of catalytic efficiency (k cat /K M ) versus pH. (b) Plot of turnover number (k cat )versuspH. (c) Plot of the Michaelis constant (K M ) versus pH. The data shown are derived from the fitting of the Michaelis–Menten equation to the experimental profiles of rate versus substrate concentration. Error bars stand for the s.d.’s of the fitting parameters as provided by the fitting program used (Igor Pro 6.37).The colour of the data points refers to the b-lactamase variant studied, specifically: GNCA MP -W229D (green); GNCA3-W229D (blue); GNCA4-W229D/F290W (red). The continuous lines in (a,b) represent the best fits of equation (1) to the experimental data. The pKvalues determined from such fits are shown alongside the lines, together with the standard errors derived from the fittings. The high pH limiting values for our best Kemp eliminase, GNCA4-W229D/ F290W, are shown as dashed lines. K M values for this variant (shown in c) appear to be essentially pH-independent. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16113 ARTICLE NATURE COMMUNICATIONS | 8:16113 | DOI: 10.1038/ncomms16113 | www.nature.com/naturecommunications 5 Finally, it must be noted that catalysis of Kemp elimination by carboxylic acids is strongly accelerated in aprotic solvents. The acetate ion in acetonitrile is in fact an excellent catalyst of the Kemp elimination with a reported second order rate constant of 2,800M1s1, a value that has been used as a metric to judge the catalytic efficiency of artificial Kemp eliminases1. Unlike previous rationally designed Kemp eliminases (Fig. 6a), our best eliminase displays a maximum catalytic efficiency (B5,500M1s1, Fig. 5a) that exceeds the acetate in acetonitrile level. Additional activities of the engineered ancestral enzymes.We used our best eliminase (the W229D/F290W variant of the alternative GNCA4reconstruction at the GNCA node), to test whether the introduction of the new active site affects the catalysis at the ‘old’ natural active site (catalytic residue S70, located at about 23 Å from position 229). b-Lactamases encoded by reconstructed sequences corresponding to the GNCA node have been previously shown11 to be able to degrade a variety of antibiotics, including penicillin and third-generation antibiotics with efficiencies similar to that of a modern average enzyme (by contrast, the modern TEM-1 b-lactamase is a specialist enzyme that displays high catalytic efficiency with penicillin and a substantially lower efficiency with third-generation antibiotics). We therefore determined the Michaelis–Menten parameters for the degradation of a penicillin antibiotic (benzylpenicillin) and one third-generation antibiotic (cefotaxime) catalysed by GNCA4-W229D/F290W (Fig. 7). We found levels of catalysis for antibiotic degradation similar to those previously reported for the GNCA MP b-lactamase (Supplementary Table 7). We also used our best eliminase to explore the potential promiscuity at the generated new active site. Our results show that the GNCA4-W229D/F290W b-lactamase does in fact also catalyse the hydrolysis of p-nitrophenyl acetate (Figs 2 and 7 and Supplementary Fig. 6), a substrate commonly used to assess esterase activity16. However, lactam hydrolysis and ester hydrolysis are chemically similar and the natural (antibiotic degradation) active site could therefore in principle contribute to the observed esterase activity. Nevertheless, our results (Fig. 7) indicate a minor natural-site contribution that only becomes apparent when the de novo active site is saturated at the higher substrate concentrations. This is specifically shown by the fact that the esterase activity of GNCA4-W229D/F290W is inhibited by 5(6)-nitrobenzotriazole, a transition-state analogue of the Kemp elimination reaction (Supplementary Fig. 3). In addition, replacing the catalytic serine at the antibiotic degradation active site with alanine does not significantly impair the rate of p-nitrophenyl acetate hydrolysis (Fig. 7), except at the higher substrate concentrations at which the de novo active site is saturated. Similarly, blocking the antibiotic degradation active site through the irreversible reaction with clavulanic acid does not significantly impair the rate of p-nitrophenyl acetate hydrolysis (Fig. 7), except at the higher substrate concentrations at which the de novo active site is saturated. Finally, saturating concentrations of benzylpenicillin do not significantly impair the rate of p-nitrophenyl acetate hydrolysis (Supplementary Fig. 7), except a 105 104 103 102 Catalytic efficiency (M–1 s–1) kcat (s–1) 101 100 10–1 10–2 10–3 103 102 101 100 10–1 10–2 10–3 10–4 10–5 10–6 12 Best Kemp eliminase to date W229D/F290W W229D/F290W W229D W229D Minimalist designs on ancestral scaffolds Minimalist designs on GNCA ancestral backgrounds d es i g n I terat i ve d es i g n Rosetta design Previous minimalist designs Rosetta design Best Kemp eliminase to date Uncatalysed reaction (17 rounds of directed evolution from an iterative design background) (17 rounds of directed evolution from an iterative design background) 345 Number of mutations 10 15 20 b 12 5 Number of mutations 10 1520 Figure 6 | Comparing our Kemp eliminases with previous rational designs. (a) The catalytic efficiencies (k cat /K M values) of the single (W229D) and double (W229D/F290W) mutant variants of the ancestral b-lactamases studied here are shown in red. The red square data point represents the high pH value for GNCA4-W229D/F290W (Fig. 4a). The values for the ‘previous minimalist designs’ are taken from Korendovych et al.24 (large orange data point at number of mutations value of unity) and Merski and Shoichet25 (small orange data points). In both cases, the design is based on a single mutation, but, in the latter study25, additional mutations had to be introduced for mainly stabilization purposes (the values shown correspond to the variants displayed in Fig. 3 of Merski and Shoichet25). The values obtained using a design approach that involved Rosetta are taken from Ro ¨thlisberger et al.26. Only the values for the eight designs that led to substantial Kemp eliminase activity are shown (59 designs were actually tested by Ro ¨thlisberger et al.26). Iterative design values are taken from Privett et al.27. Here, ‘iterative’ means that an original design with very low activity was improved on the basis of 3D-structural information and molecular dynamics simulations to achieve substantial levels of catalysis. The efficiency of the best Kemp eliminase reported to date28 is also shown here for comparison. (b) Comparison of the Kemp eliminase turnover numbers (k cat values) obtained in this work with those reported for previous rational designs. The meaning of the symbols is the same as in a. The values shown as open symbols for the W229D variants of the ancestral backgrounds are actually lower limit estimates calculated as described in Supplementary Fig. 5. We also show here the reaction rates corresponding to the best artificial Kemp eliminase reported to date28 and to the uncatalysed reaction for comparison. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16113 6NATURE COMMUNICATIONS | 8:16113 | DOI: 10.1038/ncomms16113 | www.nature.com/naturecommunications at the higher substrate concentrations at which the de novo active site is saturated. The catalytic efficiency for the hydrolysis of p-nitrophenyl acetate at the new active site, k cat /K M ¼11.7±1.1M1s1,is similar to the best values reported in the literature for rationally designed esterases29. Furthermore, the turnover number at the new active site, k cat ¼7.6103±2104s1, indicates a three orders of magnitude enhancement over the rate of the nonenzymatic reaction30 (k uncat ¼2.5106s1; Supplementary Fig. 6). These are remarkable results, in particular since, we are dealing here with a new active site generated on the basis of a minimalist approach that did not target esterase activity. Computational modelling of engineered ancestral enzymes.To further explore and confirm the role of conformational flexibility in the emergence of new enzyme functions, we have complemented our experimental work with molecular dynamics (MD) simulations of the wild-type and mutant forms of the modern TEM-1 and Bacillus licheniformis (BL) b-lactamases, as well as the ancestral ENCA, GNCA MP , GNCA4and PNCA b-lactamases (Fig. 1). We chose these particular proteins both due to the availability of X-ray structures (see the Methods section) and because they span the range of Kemp elimination activities obtained in this work upon new active site generation. In particular, the modern TEM-1 and BL b-lactamases, as well as the comparatively recent ENCA b-lactamase (Fig. 1), display negligible activity upon the W229 mutation, while this mutation confers substantial Kemp elimination activity to the PNCA, GNCA MP and GNCA backgrounds following the order PNCAoGNCA MP oGNCA4. Furthermore, the double W229/ F290W variant of the GNCA4scaffold is the most active Kemp eliminase reported in this work with a rate enhancement of about seven orders of magnitude over the rate of the uncatalysed reaction. As described below, our MD simulations support that conformational flexibility contributes substantially to these activity trends. We have used the root mean square fluctuation (RMSF) of each amino acid in our simulations as a measure of the overall flexibility of the system. We examined first the GNCA MP and TEM-1 b-lactamases, that is, the two proteins for which NMR relaxation data are available (Fig. 3) from this and previous31 work, respectively. As explained in the second section of the Results, the region of the ancestral scaffold structure targeted for new active site generation is characterized by a large number of residues with conformational exchange contributions to the relaxation rates. Clearly, complete agreement between the results of the MD simulations and the conformational exchange contributions to the NMR relaxation rates is not to be 3 a b Kemp elimination Ester hydrolysis v/[E]0 (s–1) v/[E]0 (s–1) ×10–3 10 5 0 2 1 0 25 Antibiotic degradation Benzylpenicillin (BZ) Cefotaxime (CTX) Antibiotic degradation 50 40 30 20 10 0 20 15 10 5 0 0 0 100 200 300 400 500 050 100 150 200 200 400 [Substrate] (μM) [Substrate] (μM) [Substrate] (μM) [Substrate] (mM) 600 800 pH 7.0 pH 8.2 1,000 01234567 Figure 7 | Various activities of the W229D/F290W variant of the GNCA4 scaffold. (a) Activities linked to the new active site. (b) Activities linked to the natural preexisting active site. Plots of rate versus substrate concentration are shown in (a,b), with the continuous lines representing the best fits based on the Michaelis–Menten equation. Ester hydrolysis data correspond to p-nitrophenyl acetate hydrolysis and were obtained with the GNCA4-W229D/F290W variant (closed data points) and also with this variant modified to block any esterase activity at the natural, antibiotic degradation site. This was achieved by mutating the catalytic S70 to alanine (open squares) or by inactivation by clavulanic acid (open triangles). Catalytic parameters for ester hydrolysis discussed in the Main Text are derived from the Michaelis–Menten fit to the modified variants. The equation used to fit to the data of the unmodified variant includes an additional linear term to account for esterase activity at the antibiotic degradation site. The esterase catalytic efficiency at the natural active site is, however, found to be B20-fold smaller than that at the new active site. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16113 ARTICLE NATURE COMMUNICATIONS | 8:16113 | DOI: 10.1038/ncomms16113 | www.nature.com/naturecommunications 7 expected, mainly because of the different timescales involved. That is, inclusion of exchange terms to explain relaxation rates reveals dynamic processes in the micro to milliseconds range32, while shorter timescales are typically probed by MD simulations33. Nevertheless, we find a clear correspondence between the calculated C a RMSF values and the number of residues with conformational exchange contribution to NMR relaxation at the region of the new active site (Supplementary Fig. 8). It is therefore likely that the approximately microsecond conformational transitions at the region of the new active site in the ancestral scaffold are reflected in our 200 ns ( ¼0.2 ms) MD simulations, which actually approach the lower end of the conformational exchange timescale range. The capability of our MD simulations to capture flexibility features that are relevant for new active site generation is further supported by the general congruence with the catalytic properties of several modern and ancestral scaffolds. In the case of the wildtype enzymes, a clear increase can be seen in the flexibility of residues 252 through to the h11 helix upon moving from TEM-1 and BL to the GNCA MP to GNCA4b-lactamases, which are the residues that are primarily involved in forming the cavity creating the de novo active site (Fig. 8). The flexibility of this region appears indeed to follow the general activity trend described above. Also, the largest overall changes in flexibility can be seen in helices h1 and h11, as well as the loops covering the de novo active site, in agreement with both the crystallographic data (Fig. 4) as well as with the NMR relaxation data shown in Fig. 3. The mobility of different regions of these enzymes has also been highlighted in Fig. 9. The function-generating mutation W229D appears to increase the flexibility, which is the likely outcome of the introduction of a charged residue in a hydrophobic environment6. Still, the effect is much more pronounced in the ancestral b-lactamases, while the modern b-lactamases show only a comparatively small increase in flexibility (Fig. 8). We subsequently performed MD simulations in the presence of the transition-state analogue 5(6)-nitrobenzotriazole (TSA) to explore the impact of transition-state binding on the conformational dynamics, as well as how well each system could accommodate the transition state in its new active site. The observed TSA binding features do correlate with the activity trends and with the flexibility trends defined by the RMSF data. The W229D variants of the ancestral scaffolds that give rise to substantial Kemp elimination activity (PNCA, GNCA MP and GNCA4) can accept the TSA in the MD simulations, although the TSA dissociates from the less active W229D variant of the PNCA scaffold after about 50ns (Fig. 10). By contrast, efficient TSA binding is not observed with the W229D variants that show very low Kemp elimination activity. Specifically, we could not obtain a stable TSA complex with the W229D variant of ENCA b-lactamase, while the TSA flies out of the cavity in the W229D variants of the modern TEM-1 and BL b-lactamases within about 20 ns (Fig. 10). We have also observed the same pattern (low flexibility in the region of the de novo active site and inability to retain the TSA bound) for the b-lactamases from Enterobacter cloacae (Nmc-A) and Proteus vulgaris (Bla-B) (Supplementary Fig. 9). This pattern, therefore, is likely to be a general feature of modern b-lactamases from Enterobacteria (Fig. 1). Finally, there are two apparent exceptions to the flexibility/ activity trend discussed above. First, ENCA b-lactamase develops substantial flexibility in the new active site region upon the W229D mutation, even though this protein shows negligible Kemp elimination activity. However, D229 is partially deprotonated at neutral pH and flexibility upon introduction of a charged group in a hydrophobic environment reflects, to some extent, the conformational fluctuations required to allow some water penetration and the consequent stabilization of the partially buried charge6. Such a specific kind of flexibility might not be relevant to the understanding of transition-state binding. Indeed, there should not be significant water penetration when the transition state is bound to the new active site, and the negative charge of the catalytic aspartate is actually stabilized by the interaction with the proton being abstracted. Second, the GNCA4 scaffold appears to be more rigid (in the new active site region) than the GNCA MP scaffold, despite the fact that GNCA4leads to higher levels of Kemp elimination activity upon the W229D mutation. However, 3D-structure determination shows that the GNCA4is actually more preorganized for transition-state binding than the GNCA MP scaffold (Fig. 4). This can again ultimately be viewed as a reflection of flexibility at the GNCA 3.0 a b RMSF (Å) RMSF ( Å ) 2.0 1.0 BL BL-W229D TEM1-M182T/W229D ENCA-W229D GNCAMP-W229D GNCA4-W229D/F290W PNCA-W229D TEM1 ENCA GNCAMP GNCA4 PNCA 0.0 3.0 Residue number 2.0 1.0 0.0 050 h1 52–62 86–116 150–160 173–178 194–208 215–229 252–266 266–277 h11 100 150 200 250 300 Figure 8 | C a -atom root mean square fluctuations of all simulated variants. Shown here are the root mean square fluctuations (RMSF, Å) of all C a atoms during our simulations of (a) wild-type and (b) variant forms of the TEM-1 (blue), Bacillus licheniformis (purple), GNCA MP (red), GNCA4 (orange), ENCA (dark green) and PNCA (light green) b-lactamases. In the case of the variant forms, the simulations were performed both in the absence of the transitionstate analogue (TSA) or in complex with the TSA, 5(6)-nitrobenzotriazole. The simulations shown in (b) were all performed in the absence of the TSA, and a corresponding comparison of the C a RMSF values of the ancestral enzymes both with and without the TSA present are shown in Supplementary Fig. 29. All values shown here are averages over the last 60 ns of three independent trajectories (180 ns total simulation time), which were obtained as described in the Methods section. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16113 8NATURE COMMUNICATIONS | 8:16113 | DOI: 10.1038/ncomms16113 | www.nature.com/naturecommunications node, as mutations removed from the active site can easily shift the conformational equilibria at this node. Clearly, the specific mutations present in the GNCA4scaffold have led to the population of a conformation in the W229D variant with a cavity where the substrate can easily form a reactive conformation, and with better architecture for effective transition-state stabilization. Discussion Modern proteins can perform an enormous diversity of molecular tasks, often with high efficiency and specificity. Most of these modern functions evolved from previously existing functionalities. Yet, it is inescapable to assume that the emergence of completely new functions has also occurred, at least at some early stages in protein evolution. For instance, it has been recently estimated from the number of unique enzymes and the number of domain superfamilies34 that at least 87% of all enzyme functions have evolved from another function or from ancestors with generic functionalities. This would leave a fraction of 13% of enzyme functions as plausible candidates for having emerged through the generation of new active sites. While the 13% figure may be an overestimation, as some cases of homology between superfamilies may be missed by sequence comparison35, it does however suggest that the generation of new active sites may not actually be an exceedingly rare event and thus should garner more serious consideration. This is supported by reports of alternate-site promiscuous enzymes that display catalysis of a secondary reaction at a site other than the active site of the natural catalytic process36. Furthermore, most fundamental biochemical processes are extremely slow in the absence of enzymes2and seemingly specialized enzymes are likely to have catalysed many different reactions already in the last universal common ancestor3,4. It may be reasonably inferred that efficient mechanisms for the emergence and subsequent evolution of new enzyme functionalities must exist or, at least, that they must have existed at the primordial protein stage. Little is known, however about these mechanisms. Recent work29 has shown that introducing a catalytic group in the hydrophobic cavity of the C-terminal domain of calmodulin can generate significant levels of catalysis for simple reactions. The simple mechanism for the emergence of new enzyme functions demonstrated here does not require the recruitment of a preexisting hydrophobic cavity. A hydrophobic-to-ionizable amino acid replacement generates a buried (or partially buried) residue with the perturbed properties that are useful in catalysis and conformational flexibility assists the generation of a completely new active site by facilitating substrate and transition-state binding. We have shown that, when using resurrected Precambrian b-lactamases as scaffolds for protein engineering, a minimalist design approach based on this mechanism leads to levels of catalysis for the Kemp elimination reaction of up to approximately seven orders of magnitude above the rate of the uncatalysed reaction, as well as to significant ester hydrolysis activity. The role played by conformational flexibility in the generation of the new function is clearly apparent in the X-ray crystallographic structures of our designed enzymes and supported by the extensive computational simulations reported, as well as by experimental binding studies (Supplementary Fig. 10). The process of inferring ancestral sequences inherently generates uncertainly. It is therefore customary in the field to test phenotypic robustness by studying not only the protein encoded by the most probabilistic sequence at a given node, but also the proteins encoded by alternative reconstructions derived from random sampling from the posterior probability distribution at the node. We have previously used this approach to demonstrate phenotypic robustness for the antibiotic degradation activity at the natural active site of the GNCA b-lactamase11. Remarkably, however, the new active site behaves differently and engineered b-lactamases encoded by several alternative reconstructions at the GNCA node (common ancestor of W229 D229 PNCA GNCA4 GNCAMP ENCA TEM-1 BL 2.0 0.4 Figure 9 | Tertiary structures of key b-lactamases coloured by RMSF. Shown here are the tertiary structures of the (left) wild-type and (right) variant forms of the TEM-1, ENCA, Bacillus licheniformis, GNCA MP , GNCA4 and PNCA b-lactamases, coloured by the calculated root mean square fluctuations (RMSF) of their C a atoms, based on the values shown in Fig. 8. The colour scale at the bottom of this figure shows the calculated RMSF, in Å. This figure is presented to allow for a visual comparison of the relative mobilities of the different enzymes relative to each other, and how this maps on to their tertiary structures. Note here that the Bacillus licheniformis b-lactamase has unstructured Cand N-terminal segments. The structures shown here are the most representative structures obtained from our molecular dynamics simulations, obtained from clustering analysis as described in the Methods section. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms16113 ARTICLE NATURE COMMUNICATIONS | 8:16113 | DOI: 10.1038/ncomms16113 | www.nature.com/naturecommunications 9