Deconvoluting the Directed Evolution Pathway of Engineered Acyltransferase LovD

FULL PAPER 1 Deconvoluting the directed evolution pathway of engineered acyltransferase LovD Guillermo García-Marquina,[a,b] Reyes Nuñez-Franco,[c] Francesca Peccati,[c] Yi Tang,[d] Gonzalo Jiménez-Osés*[c,e] and Fernando López-Gallego*[a,e] [a] G. García-Marquina, Prof. F.López-Gallego Center for cooperative Research in Biomaterials (CIC biomaGUNE) - Basque Research and Technology Alliance (BRTA) Heterogeneous biocatalysis laboratory Paseo de Miramón, 182, 20014 Donostia-San Sebastián, Spain Email: [email protected] / https://flg802.wixsite.com/flopezgallego [b] G. García-Marquina Universidad de La Rioja Departamento de Química Centro de Investigación en Síntesis Química, E-26006 Logroño, Spain Email: [email protected] [c] R. Nuñez-Franco, Dr. F. Peccati, Prof. G. Jiménez-Osés Center for Cooperative Research in Biosciences (CIC bioGUNE), Basque Research and Technology Alliance (BRTA) Computational Chemistry Laboratory Bizkaia Technology Park, Building 801A, 48160 Derio, Spain. Email: [email protected] [d] Prof. Y. Tang University of California–Los Angeles Department of Chemistry and Biochemistry Los Angeles, California, USA – Department of Chemical and Biomolecular Engineering, University of California–Los Angeles, Los Angeles, California, USA. [e] Prof. G. Jiménez-Osés, Prof. F. López-Gallego lkerbasque, Basque Foundation for Science, Plaza Euskadi 5, 48009 Bilbao, Spain. Email: [email protected] / https://www.cicbiogune.es/people/gjoses Supporting information for this article is given via a link at the end of the document.((Please delete this text if not appropriate)) Abstract: Pharmaceutical industry is progressively replacing the chemical synthesis of cholesterol-lowering agents by enzymatic processes. The directed evolution of acyltransferase LovD was a breakthrough in the synthesis of simvastatin, although little is known about how the in vitro evolution path raised up an engineered variant (LovD9) with excellent biocatalytic properties (high catalytic efficiency and stability under reaction conditions). In this study, we unveil how different mutations clusters scattered in LovD9 primary sequence specifically contribute to enhance both enzyme kinetics and stability. To that aim, simvastatin synthetic and hydrolytic activities, kinetic parameters and thermostability of several engineered variants were assessed. Through a rational combination of those clusters of mutations, we have generated the variant LovDBuCh2 whose catalytic efficiency is around 90% of than obtained with LovD9 but with 15 less mutations. Supported by molecular dynamics simulations, this work demonstrates the cumulative effect of mutations at both the active site and substrate entrance channel to enhance binding of the acyl donor and speed up the acyl transfer step from the acyl-enzyme complex to monacolin J acid, while simultaneously minimizing detrimental side-reaction pathways and substrate inhibition and increasing thermostability. Introduction Enzymes are biological catalysts that assist a huge variety of biochemical reactions. In the last decades, chemical industry has implemented enzymes in many successful synthetic schemes that outperform the traditional organic chemistry proceedings in terms of efficiency, selectivity and environmental compatibility [1]. Nature has designed enzymes with a very narrow substrate scope to guarantee their perfect orchestration in the cell metabolism, where thousands of reactions take place simultaneously. In industry, protein engineers have addressed enzyme substrate specificity by generating reams of enzyme variants capable of manufacturing a larger diversity of new molecules or even catalyzing fully abiological processes [2]. Among the different tools to engineer protein sequences, directed evolution is considered one of the most successful ones. This revolutionary technique is based on the generation of mutant libraries through iterative rounds of random mutagenesis that are further screened to ultimately select the best candidates towards a desired property (activity, selectivity, stability, etc.) [3]. Directed evolution [4] combined with rational computational design [5] has proven to be a winning strategy, particularly in tailoring tasks not yet exploited by natural selection [2d, 6]. One successful example that illustrates the potential of directed evolution in pharmaceutical manufacturing is the engineering of the acyltransferase LovD to develop an enzymatic process for the synthesis of simvastatin commercialized as Zocor®. This drug is one top-marketed cholesterol lowering-drug to treat cardiovascular diseases [7]. LovD from Aspergillus terreus catalyzes the conversion of the inactive precursor monacolin J acid (MJA) into the cholesterol-lowering agent lovastatin acid (LVA) [8]. During this process, the lovastatin diketide synthase LovF generates an -S-methylbutyrate side chain, which is regioselectively transferred by LovD to the C8 hydroxyl of MJA. Ser76 is deprotonated by Tyr188, which is stabilized by Lys79 forming a catalytic triad. The catalytic cycle consists in the acylation of the Ser76 with the acyl donor and followed by the acyl transfer step through the nucleophilic attack C8 of MJA (acyl acceptor) whose deprotonation is also aided by the catalytic triad [9] This catalytic reaction strongly depends on the protein-protein interaction between LovD and the acyl-carrier protein (ACP) domain of LovF. Unfortunately, the requirement for ACP makes FULL PAPER 2 the system more complex and less efficient for applied biocatalysis. Moreover, the low number of acyl donors accepted by wild-type LovD limits the chemical diversity of statins that can be manufactured with these enzymes. In previous research, the substrate promiscuity of LovD was assessed, demonstrating that this enzyme can accept small-molecule surrogate thioesters (i.e -dimethylbutyryl-S-methyl-mercaptopropanoate (DMB-SMMP)) to yield simvastatin acid (SVA) [8b, 10]. Unfortunately, wild-type LovD was 1300 times less catalytically efficient towards these acyl donor surrogates than towards its natural ACP-presented thioesters [10-11]. As a first optimization attempt, LovD was engineered through directed evolution to improve the SVA synthase activity, achieving up to 11-fold increase in the catalytic efficiency for the biosynthesis of SVA [9]. By combining directed evolution with computational analysis, a second evolution attempt was rather more successful. After nine rounds of evolution and selection, a new variant with 29 scattered mutations (LovD9) emerged for the efficient synthesis of SVA. LovD9 exhibited a catalytic efficiency three orders of magnitude higher than the wild-type enzyme [12]. Beyond the Michaelis-Menten kinetic parameters and the productivity of the free variant, other process relevant parameters remained uncharacterized. Protein stability under drastic operation conditions (i.e. elevated temperatures, use of organic co-solvents, non-physiological pH, etc.), selectivity and product/substrate inhibition are important reaction parameters which need to be assessed for evaluating the operational productivity of any biocatalyst aimed at industrial exploitation [13]. In SVA biosynthesis, LovD9 catalyzes this reaction through a ping-pong acylation mechanism where MJA acts as a competitive inhibitor of the acyl-enzyme complex formation [10].. Furthermore, LovD also presents a side thioesterase activity under aqueous conditions that competes with the acyltransferase activity, hydrolyzing the acyl-enzyme complex and yielding 2,2-dimethylbutanoic acid (DMB) as a dead-end pathway. Hence, the thioesterase activity precludes the acylation of MJA thus hampering the production of SVA. This thioesterase activity has been evaluated neither for the wild-type LovD nor its engineered variants. Finally, LovD also presents an esterase activity able to hydrolyze SVA back to MJA and the corresponding acid (DMB), diminishing the final yield of the target product (Scheme 1). This catalytic mechanism resembles the kinetically controlled synthesis of esters and amides catalyzed by esterases [14] and amidases [15], in which there are two main side unwanted reactions affecting product yield, i.e. the hydrolysis of the activated acyl-donor and the hydrolysis of the final product. This mechanistic perspective encouraged us to study how different mutations in LovD9 contribute to relevant features such as substrate inhibition, side hydrolytic activities and thermal stability which define the operational productivity of the engineered biocatalyst. As previously reported [16], partial deconvolution can be used to access important information regarding additive, cooperative or antagonistic mutational effects. In this work, we have generated three new variants, dividing the 29 mutations of LovD9 into three main clusters: residues located at the substrate entrance channel (LovD-Ch), at buried positions near the active site cavity (LovD-Bu), and a combination of both (LovDBuCh1,2) (Figure 1.a). We have performed kinetic studies on these three variants and quantified their thioesterase and esterase activities, comparing them to those measured for highly active variants LovD7 and LovD9. These data have provided a sequence landscape where we can weigh the contribution of the specific amino acid clusters to the enzyme improvement during the artificial evolution path. Results and Discussion Rational design and functional characterization of new cluster variants of LovD Depending on the distribution of the 29 mutations across the LovD9 3D structure we arbitrarily classified all mutations in three groups; one group including eight buried mutations near the active site (I35L, A178L, N191G, L192I, G275S, L361M, V370I and A383V), a second group including six mutations at the substrate entrance channel (M157V, S172N, L174F, Q241M, S256T and A261H) and a third group encompassing the remaining mutations located at the protein surface (I4N, A9V, K26E, R28S, N43R, D96R, S109C, A123P, S164G, A247S, R250K, Q297G, L335M, N391S and H404K). During the evolution path, the first four rounds increased the acyltransferase activity 89 times by mostly introducing buried and channel mutations (5 out of 7 total mutations in LovD4), whereas the last five rounds increased the enzyme activity 11 times through enriching the protein mainly with surface mutations (13 out of 22 total mutations from LovD4 to LovD9) [12]. According to these previous results, we hypothesized that both buried and substrate entrance channel mutation clusters are the major contributors to the activity enhancement during the in vitro evolution. To demonstrate this hypothesis, we have generated three new variants containing each mutation cluster: buried cluster variant (LovD-Bu), channel cluster variant (LovD-Ch), as well as a combined variant that contains both buried and channel clusters (LovD-BuCh1). LovD-Bu, LovD-Ch and LovD-BuCh1 contain Scheme 1. Catalytic reactions involved in simvastatin acid (SVA) synthesis catalyzed by LovD. DMB-SMMP:  -dimethylbutyryl-S-methylmercaptopropanoate. SMMP: S-methyl-mercaptopropionic acid. MJA: Monacolin J acid. SVA: Simvastatin acid. DMB: 2,2-dimethylbutanoic acid. ACP: acyl carrier protein. PPN: 4’-phosphopantetheine FULL PAPER 3 27%, 21% and 48% of the 29 mutations introduced in LovD9, respectively (Figure 1.a, Table S1). To understand the specific contribution of such clusters to the enhancement of the catalytic efficiency along the evolution path, we extracted the initial rate and the product yields from the timereaction courses (Figure S1) performed with pure LovD1-9, LovD-Bu, LovD-Ch and LovD-BuCh1 variants. The reaction time-courses provide an accurate picture of the initial synthetic rates (v0) of the kinetically controlled acylation process and the temporal stability of the maximum product yield. Analyzing these two parameters under the same reaction conditions, we identified the specific contributions of each cluster to the enzyme productivity (Figure 1.b). Expectedly, the v0 of the reaction is increased along the evolution rounds. The largest increase in v0 occurs within the first rounds: while wild-type LovD exhibits a v0 = 0.56 nmol mg-1 min-1, LovD4 shows a v0 = 87.5 nmol mg-1 min-1. This means that only 9 mutations were sufficient to enhance v0 around 150 times. From this intermediate variant to the most evolved one (LovD9), 20 more mutations increased the initial rate just 2.4 times (LovD9, v0 = 212 nmol mg-1 min-1). These data follow the same trend described in the original directed evolution study [12]. The key role of those 9 initial mutations in LovD productivity is also supported by the fact that all evolved variants from LovD4 reached 100% simvastatin yield in 24 hours. During the first 4 rounds of mutation, half of the so-called substrate entrance channel and buried mutations were introduced, reflecting the importance of these positions at improving both the formation of the acyl-enzyme complex with the acyl donor (DMB-SMMP) and the subsequent acyl transfer to MJA. Noteworthy, the LovD-Bu variant (8 mutations) reaches a 95% product yield in 24 hours with v0 = 29.3 nmol mg-1 min-1, an activity that falls between those of LovD3 (7 mutations) and LovD4 (9 mutations) which share only 4 buried mutations (L361M, A178L, N191S and G275S) with the LovD-Bu (Table S1). Therefore, buried mutations clearly contribute to increase the synthetic capacity towards the formation of SVA starting from artificial acyl donors, but these mutations by themselves are not enough to pair the catalytic activity observed for the most evolved variants. On the contrary, the LovD-Ch variant containing the six substrate entrance channel mutations found in LovD9 only reaches an activity similar to that of LovD2, revealing that these mutations contribute to the total LovD9 activity to a lower extent than the buried ones when taken separately. When we combined buried and substrate entrance channel mutations, the resulting variant (LovD-BuCh1) reaches 100% yield in 4 hours with v0 similar to that measured for LovD7, but with 9 fewer mutations. Merging both engineered clusters gave rise to one variant 7 and 18 times more active than LovD-Bu and LovD-Ch, respectively. This finding evidences a cumulative effect between these two mutation clusters that enhances the enzyme performance towards SVA biosynthesis. Starting from the LovD-BuCh1 variant, we created a new variant with an additional single mutation involving position 261. We hypothesized that this position plays an important role in activity, since it was mutated to Val in the 7th round and to His in the 8th round. To our delight, the simple replacement of A261V by A261H mutation increased v0 by 20%. This single amino acid substitution achieved roughly the same activity enhancement as the one obtained during the evolution from LovD7 to LovD8 that required 3 new mutations. Hence, the cumulative clustering of entrance channel and buried mutations including the A261H replacement gave rise to a new variant (LovD-BuCh2) as efficient as LovD8 (with 11 additional mutations mainly located at external positions). Kinetic characterization of LovD variants As reported before [11-12], wild-type LovD performs the acyl transfer reaction through a ping pong bi-bi substituted-enzyme mechanism where the acyl acceptor (MJA) acts as a competitive inhibitor. Contextualizing the industrial production of SVA, substrate inhibition jeopardises the process productivity under the high substrate concentrations often demanded by largescale manufacturing batch processes. In this work, we provide additional information on this key feature. Figure 1. (a) Arbitrarily identified clusters in LovD9. Buried mutations are coloured in gold, channel mutations are coloured in blue, surface mutations are coloured in green and catalytic triad residues are coloured in pink. (b) Initial velocities (v 0 , in circles) and yield after 24 h (in bars) for simvastatin acid (SVA) production catalyzed by LovD evolved variants and the new cluster variants produced in this work. Reaction conditions: 5 mM monacolin J acid (MJA), 2mM  -dimethylbutyryl-S-methylmercaptopropionate (DMB-SMMP) in 50 mM HEPES, 10 mM MgCl 2 (pH 8) and 10% dimethyl sulfoxide (DMSO) (see Experimental Section ). LovD LovD1 LovD2 LovD3 LovD4 LovD5 LovD6 LovD7 LovD8 LovD9 LovD-Ch LovD-Bu LovD-BuCh1 LovD-BuCh2 0 20 40 60 80 100 Yield at 24 h v0 Yield at 24h (%) 0 50 100 150 200 250 v0 (nmol mg-1 min-1) a b FULL PAPER 4 Although MJA negligibly inhibits wild-type LovD when using DMB-SMMP as acyl donor [11], we studied substrate inhibition for two of the most productive LovD engineered variants (LovD7 and LovD9) together with our new cluster variants, complementing the kinetic parameters determined in the original evolution study [12]. To that aim, we performed steady-state kinetics for the synthesis of SVA under fixed DMB-SMMP concentration (2 mM) and varying the concentration of the acyl acceptor (MJA). When we inspected the Michaelis-Menten (MM) curves we observed that wild-type LovD and LovD-Ch did not presented inhibition at high MJA concentrations, while a strong substrate inhibition for LovDBu and some of the most evolved variants at MJA concentrations higher than 3 mM was detected (Figure S2). To assess the impact of substrate inhibition on the catalytic efficiency of each variant, we fitted the MM plots to the classical kinetic MM model for LovD and LovD-Ch (Equation 1, see Experimental Section) and to the classical kinetic substrate inhibition model for LovD-Bu, LovD-BuCh1, LovD-BuCh2, LovD7 and LovD9 [17] (Equation 2, see Experimental Section). In Table 1, the LovD-Ch variant shows a kcat/KM of 3.5 M-1 s-1, which improves around 4 times the turnover number of wild-type LovD, mostly due to a 4-times lowering of KM. LovD-Bu variant resulted in a kcat/KM 19 times higher than wild-type, but the Ki of 1.83 mM suggests that substrate inhibition emerges when internal mutations near the active site are introduced. When the two mutation clusters are combined into LovD-BuCh1, kcat/KM increases 12 and 55 times with respect to LovD-Bu and LovDCh variants, respectively, and Ki is higher than in LovD-Bu. These kinetic data suggest that entrance channel mutations mitigate the inhibition capability fostered by buried mutations, increasing the overall catalytic efficiency of the enzyme. kcat/KM and Ki of LovD-BuCh2 are very similar to those measured for LovD7, in agreement with the initial rate values obtained from reaction time-courses. Remarkably, the increase in the catalytic efficiency of LovD-BuCh2 accompanies an increase in Ki, proving so that the V261H mutation benefits the catalysis by reducing substrate inhibition. Finally, LovD9 variant shows both the highest kcat/KM and Ki of all variants herein studied. Therefore, these kinetic data support that the last evolution rounds (from LovD7 and LovD9) minimize MJA inhibition to decrease the Michaelis constant (KM). Evaluation of thioesterase activity as a competing reaction for the acyltransferase capacity of LovD variants As DMB-SMMP is a surrogate substrate for the acyl transfer reaction, the catalytic efficiency of the native ACP-dependent LovD to accept the acyl group from such artificial substrate is very low. The evolution path towards an efficient simvastatin synthase must necessarily improve the DMB-SMMP binding affinity to efficiently set up the acyl-enzyme complex that will be further attacked by MJA to ultimately yield SVA. Nevertheless, water competes with MJA for attacking the acyl-enzyme complex, driving the enzyme back to its initial state and releasing 2,2-dimethylbutanoic acid (DMB) as an undesired byproduct (Scheme 1). This side reaction is fatal to attain high conversion yields of SVA because it consumes one of the starting materials, derailing the process towards unwanted products. According to this, LovD acts as a promiscuous enzyme with two competitive catalytic capabilities: thioesterase (unproductive hydrolysis of DMB-SMMP) and acyltransferase (productive acyl transfer from DMB-SMMP to MJA). To assess which catalytic capability prefers each LovD variant, we developed a colorimetric method based on the detection of methyl 3-mercaptopropanoate released during the first reaction step with and without the acyl acceptor (MJA) in the reaction media. Measurements in the absence of MJA reflects the thioesterase activity (TE), where only water attacks the acylenzyme complex, while those performed in the presence of MJA report a competition between the thioesterase and the acyltransferase (AcT) activities, thus a competition between water and MJA to attack the acyl-enzyme complex. A similar methodology has been recently exploited for the identification of promiscuous acyltransferase activity in esterases [18]. The (AcT:TE) ratio measures which activity dominates over the other: low ratios indicate that hydrolysis of the acyl-enzyme complex is faster than acyl-transfer to MJA, while higher ratios prove the prevalence of acyl transfer reaction. Figure 2.a shows that from evolution round 2 to 6, the thiosterase activity scales with the SVA productivity, suggesting that the acyltransferase activity of LovD evolved in parallel to its thioesterase activity. The LovD scaffold likely evolved to better allocate DMB-SMMP as an acyl donor until the sixth evolution round. This adaptation to artificial evolutionary pressure is particularly apparent in round 4, where the substrate entrance channel mutations M157V and S172N are introduced in LovD4 variant. These two mutations seem to underpin the productive binding of DMB-SMMP into the active site to form the acylenzyme complex. Once enough affinity towards DMB-SMMP was achieved, further evolution rounds slightly decreased the thioesterase activity but still increased further the SVA synthetic rate. Remarkably, Figure 2.b shows an inflexion point in the AcT:TE activity ratio after LovD4 where the evolution path apparently starts to gradually reduce the unproductive hydrolysis of the acyl-enzyme complex, while kept increasing the capacity Table 1. Kinetic parameters of LovD variants varying the concentration of monacolin J acid (MJA) (0.25; 0.5; 1; 2; 3; 5; 7 and 10 mM) at 2 mM - dimethylbutyryl-S-methyl-mercaptopropanoate (DMB-SMMP). kcat is the turnover number, KM is the Michalis constant and Ki is the inhibition constant. Mutant kcat (s-1) KM (mM) kcat /KM (M-1s-1) Ki (mM) LovD 0.004 (± 0.002) 4.2 (± 1.2) 0.9 (± 0.3) - LovD-Ch 0.004 (± 0.001) 1.4 (± 0.7) 3.5 (± 1.1) - LovD-Bu 0.24 (± 0.15) 14.6 (± 1.6) 17 (± 6) 1.8 (± 0.3) LovDBuCh1 0.54 (± 0.15) 3.2 (± 1.3) 196 (± 57) 2.7 (± 1.3) LovDBuCh2 0.35 (± 0.07) 0.99 (± 0.27) 400 (± 173) 4.1 (± 1.2) LovD7 0.29 (± 0.01) 0.9 (± 0.1) 319 (± 13) 5.7 (± 1.2) LovD9 0.29 (± 0.01) 0.6 (± 0.1) 451 (± 103) 8.4 (± 1.5) FULL PAPER 5 of that complex to transfer the acyl group to MJA. We identify another inflexion point after the 7th evolution round, where the AcT:TE activity ratio reaches its maximum value that declines steadily afterwards. It is remarkable that the most active variant (LovD9) shows an AcT:TE activity ratio only slightly higher than the wild-type. The deconvoluted channel variant LovD-Ch exhibited even lower thioesterase activity and AcT:TE ratio than wild-type LovD. On the contrary, the LovD-Bu variant was as good thioesterase as the native enzyme, but it was the only variant herein studied whose acyltransferase activity significantly outperforms the thioesterase one (AcT:TE = 1.6), so the enzyme prefers MJA as a nucleophile rather than water. These data clearly indicate that both buried and channel residues by themselves play no role in the formation of the acyl-enzyme complex, but buried mutations strongly contribute to improve the acyltransferase capability of LovD variants. These insights are supported by the fact that LovD-Bu variant exhibits 4 times higher SVA synthetic rate than LovD-Ch one. When both buried and channel clusters were merged, LovD-BuCh1 outperforms the thioesterase activity of the variants harboring each cluster separately, exhibiting an AcT:TE ratio roughly 2 times lower than LovD-Bu. Hence, we suggest that the similar SVA synthetic rates (v0) measured for both deconvoluted LovD-BuCh1 and evolved LovD7 rely on the combination of these channel and buried clusters that additively enhance the kinetics of both first (acyl-enzyme complex formation) and second (acyl-transfer) steps. Comparing LovDBuCh1 and LovD-BuCh2, the latter favors both the thioester hydrolysis and the SVA synthesis. Moreover, the AcT:TE ratio of LovD-BuCh2 was lower than the one observed for LovD-BuCh1. These results suggest that a His in position 261 contributes to form the acyl-enzyme complex using the DMB-SMMP as acyl donor. Altogether, these results support that laboratory evolution experiments can sequentially optimize several features (acyl donor intake, acyl-enzyme complex stability, acyl transfer rate and product hydrolysis) of the complex reaction mechanisms. In this particular case, the in vitro evolution of LovD reshapes the acyl transferase active site to adapt an unnatural acyl group (2,2-dimethylbutiryl) donated by an abiological acyl donor (DMBSMMP), forming an acyl-enzyme complex that is preferentially attacked by the enzyme natural acyl acceptor (MJA). Evaluation of detrimental esterase activity catalyzed by the LovD variants In a batch enzymatic synthesis, product hydrolysis arising from the reversible nature of this reaction limits SVA productivity, which obligates to add excesses of MJA to shift the equilibrium towards the synthetic direction and thus maximize the SVA titers. As SVA hydrolysis has already been reported when using wildtype LovD [11], we characterized this unwanted reversible activity for all the variants harvested in the directed evolution campaign as well as for the original cluster variants herein developed. Figure 3 shows the correlation between the SVA hydrolytic and synthetic activities for all the mutants. The SVA hydrolysis evolution profile follows a similar trend to that one observed for the AcT:TE ratio (Figure 2.b). Figure 3 Figure 3 . Correlation between simvastatin acid (SVA) hydrolysis yield and SVA synthetic rate (v 0 ) of laboratory evolved and deconvoluted cluster variants. SVA hydrolysis is determined as the residual SVA concentration after 24 h of reaction, being 2 mM the initial SVA concentration. V 0 is determined through the time-course reactions shown in Figure S3 . Green circles represent the native and engineered variants obtained through the laboratory evolution experiment (LovD and LovD19). Half-filled blue circles represent the separated cluster variants (LovDBu and LovD-Ch). Full-filled blue circles represent the variants with the two clusters (LovD-BuCh1-2). Dash green line is not a fitting, but only guides the eye of the reader along the evolution path. Figure 2. (a) Specific thioesterase activity versus simvastatin acid (SVA) synthetic rate (v 0 ). (b) Acyltransferase/thioesterase activity ratio (AcT:TE) versus SVA synthetic rate (v 0 ). Green circles represent the native and engineered variants obtained through the laboratory evolution experiment (LovD and LovD1-9). Half-filled blue circles represent the separated cluster variants (LovD-Bu and LovD-Ch). Fullfilled blue circles represent the variants with the two clusters (LovDBuCh1-2). Dash green line is not a fitting, but only guides the eye of the reader along the evolution path. a b FULL PAPER 6 shows how the hydrolytic activity increases until the 4th evolution round as LovD4 hydrolyzed 7 times more SVA than the wild-type enzyme. After the 5th round, the SVA hydrolysis diminished until round 7, and then slightly increased again during rounds 8 and 9. The higher hydrolytic capacity of LovD9 is likely compensated by its lower product inhibition profile, which might be one of the main factors contributing to its superior SVA productivity. Along with the improvement of the acyl-enzyme complex formation supported by the enhancement of the thioesterase activity, the first 4 evolution rounds also make LovD variants able to more efficiently hydrolyze SVA, reinforcing the important role of water accessibility to the active site. On the contrary, the second half of the evolution pathway diminishes the hydrolytic reaction in favor of the synthetic one. When the SVA hydrolytic activity was assessed for both LovD-Bu and LovD-Ch, both deconvoluted variants presented higher hydrolase activity than the wild-type enzyme but lower than any of the other evolved mutants. As expected, the combination of buried and channel mutations in LovD-BuCh1-2 variants increased both synthetic and hydrolytic SVA activities. Once again, the evolution path reshaped the substrate channel and the binding pocket to productively accommodate both acyl donor and acceptor to synthetize SVA. However, that molecular re-shaping also facilitates the binding of the acetylated product in the absence of the acyl acceptor, explaining the enhancement of the hydrolytic activity during the first part of the evolution path. Assessing the role of cluster mutations in active site integrity and acyl acceptor binding through molecular dynamics. In our previous report [12], we rationalized the gain in catalytic activity throughout laboratory evolution of different LovD variants, by analyzing the structural integrity of the active site (i.e. the Ser76-Lys79-Tyr188 catalytic triad) using molecular dynamics (MD) simulations. We proposed that wild-type and the early evolved variants present a catalytic Tyr188 able to experience wider motions, as reflected by conformational changes, that disrupted the catalytic triad thus hampering the catalytic activity in the absence of the acyl carrier protein partner LovF. This feature was elusive to x-ray crystallography, which showed a high degree of similarity of the catalytic triad among different LovD variants despite their very different activities. Herein, we further analyze the ability of different variants in their acylenzyme complex forms to transfer the α-dimethylbutyryl group to the acyl acceptor MJA. To this aim, we generated models for the selected evolved and cluster Ser76-acylated mutants with MJA bound to the active site as in available crystallographic structures (Figure 4.a) [9]. These models were further subjected to microsecond MD simulations under physiological conditions (see Experimental Section and Figures S3-S12). By monitoring the distance between the reactive fragments of acylated Ser76 and MJA (carbonyl and hydroxyl groups, respectively) throughout the simulations, we uncovered the ability of the intramolecular interactions optimized by the engineering process to keep the substrate in a productive configuration. For wild-type LovD, such distance fluctuates between 2.7-6.2 Å for the first 600 ns and finally MJA abandons the initially modeled binding mode. On the contrary, such distance is maintained at 3.3±0.2 Å along the whole simulation of the very efficient variant LovD6, reflecting the superior ability of highly evolved mutants to bind MJA in a productive orientation (Figure 4.b). Also, and reproducing the trend observed in our previous report [12], the catalytic Tyr188 of the native enzyme undergoes a conformational transition at the early stages of the simulation (ca. 130 ns) which obliterates its necessary catalytic Figure 4 . Molecular dynamics simulations of LovD variants ( a ) Active site residues (in blue), catalytic triad (in pink; acyl group in purple) and substrate (in orange) for modelled acyl-enzyme complexes of LovD bound to monacolin J acid (MJA). ( b , d ) Distance between the reactive carbonyl carbon of acylated Ser76 and hydroxyl group of MJA and ( c , e ) between the reactive hydroxyl groups of MJA and Tyr188 along the molecular dynamics (MD) simulations. ( f ) Surface representation of the residues constituting the substrate entrance channel (in blue) of LovD, together with bound MJA (in orange sticks). ( g , h ) Computed surface area of the main tunnel cluster calculated along the MD simulations. FULL PAPER 7 contacts with both the acylated donor and acceptor (Figure 4.c). Noteworthy, we observed a similar trend in the MD of the cluster variant featuring only the channel mutations (LovD-Ch). On the contrary, cluster variants LovD-Bu and LovD-BuCh1 comprising mutations in buried regions of the protein, maintained the starting, productive MJA binding mode and catalytic contacts (Figure 4.d-e). Therefore, MD studies support the notion that the improvement of the acyl transfer step to synthetize SVA mainly relies on buried mutations rather on channel ones. Furthermore, molecular dynamics on the apo protein unveiled another structural feature elusive to crystallography, namely the size and depth of the substrate entrance channel to the active site (Figure 4.f-h, S13 and S14). The ACP-dependent wild-type LovD variant showed a much wider channel that remains open most of the simulation time, being prone to accommodate the 4′-phosphopantetheine (PPN) acylating group present in its natural acylating partner; LovF [12]. However, highly evolved mutants showed a much narrower channel which barely opened during the simulations (17% and 20% of the simulation time for LovD6 and LovD9, respectively). This trend of narrowing the substrate entrance channel is observed also for the cluster variants, particularly for LovD-Bu, supporting how the laboratory evolution drastically alters the structure and dynamics of a protein scaffold [19]. Besides eliminating its need for LovF-ACP to render the catalytic site organization through mutations, substantially narrowing the substrate entrance channel likely mitigates exposure of both the acyl-enzyme complex and bound SVA to water, thus reducing detrimental hydrolytic activities and favoring product formation as supported by the experimental data shown above (Figures 2 and 3). Thermal stability of LovD variants Thermal stability of enzymes is a hallmark that guarantees the robustness of one biocatalyst for its biotechnological applications. Normally, highly thermostable enzymes operate for longer times under drastic conditions, reducing the enzyme replacement costs of the process. Reaction time-courses (Figure S1) demonstrate that mutations at the protein surface have a less significant impact in enhancing the catalytic properties of LovD for the synthesis of SVA. Despite of being significantly far from the substrate entrance channel and the active site, those external mutations might play a role in the overall stability of the enzyme as seen for other engineered biocatalysts [20]. Figure 5.a represents the melting temperature (Tm) of all the directed evolution and cluster designed variants of LovD, revealing how the evolution process improves the thermodynamic stability of the enzymes against temperature. A maximum Tm (52 ºC) was achieved after the 6th evolution round. All the deconvoluted variants with the channel, buried and the combination of both mutations showed lower Tm than LovD6-9 variants. These data confirm the relevance of surface mutations, mostly introduced in the second half of the evolution process. Figure 5.b shows the thermal inactivation first-order kinetics at 37 °C for wild-type LovD, an intermediate (LovD4) two of the most evolved variants (LovD7 and LovD9) and the four deconvoluted variants herein developed. These inactivation courses were fitted to Equation 3 to determine their inactivation constants and half-lives (Table S2). The complete inactivation of LovD occurred after 30 minutes. Consequently, the native enzyme was 40 times less stable than the most evolved LovD9 variant. The 9 mutations introduced in LovD4 significantly stabilized the protein scaffold and the most evolved LovD7 and LovD9 seemed to have reached the same stability limit. Cluster variants LovD-BuCh1 and LovD-BuCh2 exhibited respectively a half-life of around 3.3 and 1.3 times shorter than the most evolved variants. Therefore, the trend observed in the kinetic thermal deactivation (Figure 5.b) agrees with the data extracted from the thermodynamic inactivation against the temperature (Figure 5.a). Hence, we postulate that the external positions introduced along the evolution pathway have an important role in increasing enzyme stability. From these data, it is apparent that the increasingly harsher conditions employed during laboratory evolution [12] imposed a selective pressure towards stability, which in turn contributes to the higher productivity of the most evolved variants. Figure 5. a. Melting temperature (T m ) of different LovD variants determined through thermal-shift assays. b. Thermal inactivation timecourses for selected LovD variants measured at increasing incubation times at 37 °C. Residual activity (%) at different times means the enzyme activity at time = t divided by the enzyme activity at t = 0 (before thermal incubation) and expressed as percentage. 0,0 0,5 1,0 1,5 2,0 2,5 3,0 3,5 4,0 0 20 40 60 80 100 Relative activity (%) Time (h) LovD LovD4 LovD7 LovD9 LovD-Bu LovD-Ch LovD-BuCh1 LovD-BuCh2 LovD LovD1 LovD2 LovD3 LovD4 LovD5 LovD6 LovD7 LovD8 LovD9 LovD-Ch LovD-Bu LovD-BuCh1 LovD-BuCh2 30 35 40 45 50 55 60 Tm (º C) a b FULL PAPER 8 This thermostability trend could be qualitatively reproduced using Rosetta protein structure prediction software [21] (see Experimental Section, Figure 6 and Figure S15). Figure 6.a shows calculated Rosetta scores (i.e. total system energy, TSE) for selected directed evolution and cluster LovD variants. The lowest (i.e. most negative) scores related to that calculated for wild-type LovD (ΔTSE) provide a rough estimation of the relative thermostability of each enzyme (Figure 6.b) [22]. Of note, the Pearson correlation coefficient between the calculated ΔTSE and experimental Tm values is remarkably good for directed evolution variants LovD to LovD9 (ρ = -0.93; strong negative correlation) but decreases when the cluster mutants are included (ρ = -0.77; weak negative correlation). Together with these energy-related values, the gradually decreasing rootmeansquare deviations (rmsd) with respect to LovD9’s crystallographic structure, suggest that laboratory evolution gradually optimized thermostability while sculpting the protein backbone for improved simvastatin productivity. Remarkably, modeled structures showed a high flexibility at the loops located near the substrate entrance channel; these loops become more rigid at the late stages of evolution, particularly in LovD9 (Figure 6.c). This observation, together with the narrower tunnel calculated for evolved and cluster variants with respect to LovD (Figures 4.f, S3 and S4), suggest that evolution concurrently increased catalytic activity and thermostabilized the protein backbone, at least in part, by partially locking flexible loops in a productive conformation [23]. Such evolutionary strategy (or better said, outcome) likely mimics the effect of protein-protein interactions occurring between native, wild-type LovD and its acylating partner protein LovF [12]. Conclusion In this study, we have elucidated the intricate factors governing the reactivity of several variants of acyltransferase LovD engineered by directed evolution. Through deconvoluting the most evolved variant LovD9 in two mutational clusters that involve residues at the entrance channel and binding site of the substrates, we shined light on how the evolution path gradually improves the complex functionality of this acyl transferase. The functional analysis of the directed evolution pathway unveiled how the mutations inserted during the first rounds of the directed evolution of LovD [12] enhanced the productive intake of the acyl donor surrogate (DMB-SMMP), generating an acyl-enzyme intermediate more efficiently. Dissecting the 29 mutations of LovD9 into three clusters, we found that mutations at buried positions largely favor the acyl transfer vs. acyl-enzyme complex hydrolysis, while mutations at the substrate entrance channel mitigate the inevitable product hydrolysis. Remarkably, the combination of these two clusters had a cumulative effect as supported by the high productivity that LovD-BuCh1-2 exhibit towards the synthesis of SVA. Furthermore, the different functional properties exhibited by LovD-BuCh1 and 2 suggest that the polarity of the solventexposed position 261 at the substrate entrance channel enhances the enzyme acylation. The SVA productivity and thermostability observed for LovD-BuCh2 were comparable to the most evolved variants but with only half of the mutations. These insights demonstrate the dominant but cumulative role of buried and substrate entrance channel alterations in the Figure 6. (a). Distribution of Rosetta scores (i.e. total system energy, TSE) calculated for model full-atom structures of directed evolution and cluster LovD variants (2000 decoys for each mutant). The root-mean-square deviation (rmsd) with respect to the Rosetta-minimized crystallographic structure of LovD9. (b) Relative total system energies (ΔTSE) calculated from Rosetta scores for directed evolution (blue) and cluster (red) LovD variants, versus their experimental melting temperatures (T m ). The Pearson correlation coefficient (ρ) is calculated using all data points. (c) Ribbon representation of the 200 lowest-energy decoys calculated for wild-type LovD (left) and LovD9 (right). Highly flexible loops, most of them being part of the substrate entrance channel, are coloured in pink (residues 1-11), cyan, (residues 104-128), green (residues 147-179) and orange (residues 243264). FULL PAPER 9 enhancement of both enzyme activity and thermostability. Therefore, the mutational cluster deconvolution emerges as a valid approach to comprehend the specific contribution of mutations inserted during the laboratory evolution paths as well as to find new minimalist variant that reduce the number of mutations without worsening the functional properties of the most evolved variants. We foresee that this molecular exercise supported by experimentally and computational data has built the basis for future designs of new variants involving alternative mutational pathways. Experimental Section Materials. All LovD genes were synthetized and cloned into expression vector pET28b by GeneScript Gene Synthesis service (Piscataway, NJ, USA). The sequences are provided in the supporting information (Table S1). The genotype of the bacteria strains used for molecular biology and expression purposes are described in Table S3. Substrates MJA and DMB-SMMP were synthetized in our labs. Simvastatin hydroxyl acid ammonium salt 98% was purchased from Toronto Research Chemicals (Toronto, Canada). 2,2’-Dithiodipyridine 100% and Amicon Ultra-0.5 centrifugal filter units (10 kDa) were purchased from Sigma-Aldrich (St.Louis, IL, USA). Agarose microbeads with cobalt chelates (AG-Co2+) (50-150 m diameter) were purchased from Agarose Bead Technologies (Madrid, Spain). Polypropylene (12 x 32 mm, 300 L volume) vials were purchased from Waters (Milford, Massachusetts, USA). MicroWell 96well microplates were purchased from Thermo Fisher Scientific (Waltham, Massachusetts, USA). QuickChange site directed mutagenesis. The lovD gene mutation A261H was introduced with the QuickChange site-directed mutagenesis strategies using the plasmid pET28b lovD-BuCh1 A261V as template and performed according to the protocol described in Table S4. Chemically competent E. coli DH5 strain cells were transformed with DNA preparation. The plasmid DNA was isolated with the NucleoSpin Plasmid Miniprep Kit (Macherey-Nagel) and the expected mutations were confirmed by DNA sequencing by STAB vida (Caparica, Portugal). Expression of LovD variants. The plasmids encoding the His-tagged LovD variants were transformed into chemically competent E. coli strains DH5 and BL21 through heat shock [24] for plasmid propagation and recombinant protein expression, respectively. Single colonies of E. coli containing a plasmid encoding a variant of LovD were inoculated into 3 mL of Lysogeny broth (LB) medium containing 30 g/mL of kanamycin. Cells were grown overnight at 37 °C with shaking at 250 rpm. The culture was diluted 1:50 into 50 mL of LB medium containing 30 g/mL of kanamycin and the culture was grown until OD600nm reached 0.6-0.8. At that point, the protein expression was induced by adding isopropyl-Dthiogalactoside (IPTG) to a final concentration of 0.1 mM and the culture was incubated overnight at 21 °C with shaking at 250 rpm. Cells were collected by centrifugation (2057 xg for 30 min at 4 °C) and cell pellet was resuspended with 6 mL of 50 mM Tris-HCl buffer (pH 8.5). Cells were lysed by sonication on ice and cell debris was removed by centrifugation (12857 xg for 30 min). Purification of LovD variants. All the LovD His-tagged proteins were purified through immobilized metal affinity chromatography (IMAC) using agarose-based resin functionalized with cobalt chelates (AG-Co2+) in bulk. First, the 1 mL of cell lysate was incubated with 100 mg of resin in purification buffer (50 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), 10 mM MgCl2 at pH 8) for 1 hour at 4 ºC, then the bound enzymes were washed with the same purification buffer and eluted with 200 mM imidazole in 50 mM HEPES 10 mM MgCl2 at pH 8. Protein concentrations were qualitatively assessed by sodium dodecyl sulphatepolyacrylamide gel electrophoresis (SDS-PAGE) (Figure S16) and quantitatively determined by Bradford protein assay using bovine serum albumin as a standard [25]. Time course reaction of LovD variants. Time-course assays of SVA production for different LovD variants, were performed at 5 mM MJA, 2mM DMB-SMMP in 50 mM HEPES, 10 mM MgCl2 (pH 8) and 10% dimethyl sulfoxide (DMSO) to facilitate the solubilization of DMB-SMMP. Reactions were triggered with 1 M of each LovD variant, and samples were withdrawn at different time points by passing them through tangential ultrafiltration unit (Amicon Ultra centrifugal filters, 10 kDa). Samples were analyzed by Ultra Performance Liquid Chromatography (UPLC) (Waters 2690) equipped with a PDA detector using a ACQUITY UPLC® BEH C18 1.7µm (2.1 x 50 mm) Waters column coupled to a LCT XE time-of-flight mass spectrometry detector with electrospray ionization source (ESI). Analytes were eluted with an isocratic mobile phase composed of 52 % (v/v) of acetonitrile in water (0.1 % (v/v) formic acid) during 15 min at flow rate of 0.3 ml min-1 (Figure S1, S17). The source parameters of the mass spectrometer were: capillary voltage 1000 V, cone voltage 50 V, cone gas 50 L/h, desolvation gas 600 L/h, mass range 100-1000 m/z. Source temperature was set at 120 °C. Kinetic characterisation of LovD variants. To determine the kinetic parameters of the different LovD variants, the concentration of MJA was varied from 0 to 10 mM, whereas the DMB-SMMP concentration was fixed at 2 mM. The assays were performed at the same conditions as described in the previous section. Reactions catalyzed by LovD7, LovD9, LovD-BuCh1 and LovD-BuCh2 were triggered with 1 M of enzyme and stopped after 30 min, 1 h and 2 h, respectively. Reactions catalyzed by LovD-Bu and LovD-Ch were triggered with 2 M and 4 M, respectively, and stopped after 1 h and 2 h for LovD-Bu and at 1 h, 16 h and 18 h for LovD-Ch. Data were plot and fitted to the substrate inhibition model with the Equation 1 (Table 1, Figure S2): 𝑣 =𝑣𝑚𝑎𝑥 [𝑆] 𝐾𝑀+[𝑆] (1) 𝑣 = 𝑣𝑚𝑎𝑥 [𝑆] 𝐾𝑀+[𝑆](1+[𝑆] 𝐾𝑖) (2) in which vmax is the maximum velocity, KM is the Michaelis constant and Ki is the inhibition constant. Thioesterase spectrophotometric assay. Reactions to characterise thioesterase activity were performed in 96-well plates for 30 min at 37 °C using 3 M enzyme, 0 or 3 mM MJA, 2 mM DMB-SMMP and 2 mM 2,2- ‘dithiodipyridine (2-DTDP) in 50 mM HEPES 10 mM MgCl2 (pH 8) 10 % DMSO. The thiol formed after the enzymatically driven thiolysis of DMBSMMP spontaneously reacts with 2-DTDP, generating 2-thiopyridone that absorbs at 323 nm with an absorption coefficient of 7600 M-1 cm-1 [26]. Esterase activity characterization. The ability of the LovD variants to hydrolyze SVA yielding MJA was monitored by UPLC as described above. The reactions were triggered by adding 2 M enzyme to a reaction mixture 2 mM SVA in 50 mM HEPES 10 mM MgCl2 (pH 8) 10 % DMSO and stopped after 24 h passing through tangential ultrafiltration (Amicon Ultra 0.5 mL centrifugal filters, 10 kDa) and quantified with the same UPLC method described previously. Thermal shift assay. To calculate the melting temperature (Tm) of LovD variants, 4 g of the protein was incubated with a 50X stock solution of SYPRO Orange Dye in a final volume of 25 l in a 96-well plate of qPCR with a ramp from 25 °C to 95 °C in increments of 0.5 °C for 10 seconds. Thermal inactivation assays. To determine the thermostability of some LovD variants, 3 M each variant was diluted in 50 mM HEPES 10 mM MgCl2 (pH 8) and further incubated at 37 °C. The residual activity was