Design of Artificial Enzymes: Insights into Protein Scaffolds

The design of artificial enzymes has emerged as a promising tool for the generation of potent biocatalysts able to promote new‐to‐nature reactions with improved catalytic performances, providing a powerful platform for wide‐ranging applications and a better understanding of protein functions and structures. The selection of an appropriate protein scaffold plays a key role in the design process. This review aims to give a general overview of the most common protein scaffolds that can be exploited for the generation of artificial enzymes. Several examples are discussed and categorized according to the strategy used for the design of the artificial biocatalyst, namely the functionalization of natural enzymes, the creation of a new catalytic site in a protein scaffold bearing a wide hydrophobic pocket and de novo protein design. The review is concluded by a comparison of these different methods and by our perspective on the topic.


Introduction
Since their discovery in the late 1800s, enzymes have fascinated scientists due to their efficiency and selectivity in catalyzing a variety of energetically demanding reactions. Historically, they played a key role as green and sustainable alternatives to traditional chemical protocols, due to their unmatchable rateacceleration and selectivity, accompanied by mild reaction conditions and minimized waste production. [1,2] Despite this tremendous potential, enzymes' typical substrate-and reactionspecificity constitute the main pitfalls towards wider synthetic (and industrial) applications. Therefore, the urge to create biocatalysts able to promote new-to-nature reactions or displaying improved catalytic performances has never been more attractive. In this context, the recent progress in biotechnology and protein engineering has finally opened the possibility of developing artificial enzymes. These catalysts are originated by the functionalization of an appropriate biomolecular scaffold with an artificial moiety responsible for the newly inferred catalytic properties. It is evident that the design process of artificial enzymes constitutes a formidable challenge, as many factors including substrate binding, product dissociation and transition state geometry play a crucial role in the catalytic activity. [3] Nevertheless, considering the increasing understanding of enzymatic mechanisms and the key contribution of computational techniques and directed evolution, a large number of artificial enzymes have been created so far. [4][5][6][7] This review will focus on biomolecular scaffolds that are converted into artificial enzymes by the incorporation of an artificial moiety. For this reason, increasing the catalytic promiscuity of existing enzymes or their repurposing, [8][9][10] despite being useful strategies to unlock abiological reactivities, will not be addressed in this context. The most exploited techniques to generate these artificial biocatalysts involve the functionalization of an already existing enzyme with an artificial catalytic unit ( Figure 1A), and the creation of a novel catalytic domain in a scaffold bearing a hydrophobic pore wide enough to accommodate diversified substrates, to overcome the substrate-specificity typical of natural enzymes ( Figure 1B) and obtaining a more versatile biocatalyst. In both cases, new features such as a novel activity or new binding properties can be inferred into the resulting artificial enzyme. Considering the importance of metal cofactors for the functioning of several natural enzymes, it is not surprising that a key strategy consists of the coordination/conjugation of synthetic metal-complexes in protein scaffolds, developing artificial metalloenzymes (ArMs), whose properties will depend directly on the characteristics of the metal complex. [11][12][13] The protein-metal interaction can occur through the incorporation of metal-binding noncanonical amino acids, [4,14] supramolecular and dative anchoring, [15,16] covalent linkage (e. g. through Click chemistry) or via the metal replacement in natural metalloenzymes. [17,18] However, the ultimate goal in the artificial enzymes' development remains their de novo design from scratch, to originate ad hoc catalysts able to promote the desired transformation. [19][20][21] For this approach, an adequate knowledge of the reaction mechanism and transition state is necessary, to identify an optimal protein backbone bearing amino acid residues arranged in a favorable position to accommodate the reaction partners. [1,22] Alternatively, an already existing enzymatic activity (usually a cofactor) can be incorporated in a simple motif (e. g. a helix bundle) designed from first principles, originating the new catalyst. Utilizing tools such as the Rosetta design suite, several computationally designed metal-binding peptides and proteins were recently described. [23] These proteins frequently demonstrate sub-Ångstrom accuracy of the predicted tertiary and quaternary structure. Although the catalytic performances of the accessed de novo proteins are often lower than their natural counterparts, they can be effectively optimized by rational mutations or directed evolution, [24,25] two techniques that can be used to boost the performances of artificial enzymes accessed by any of the strategies depicted in Figure 1. [26,27] Nevertheless, considering that directed evolution and rational mutagenesis could be time and labor-consuming processes, computational methods and accurate protein structural information are still of great importance to guide engineering approaches. [28,29] Once suitable de novo scaffolds are identified, their catalytic properties can be expanded via the incorporation of an artificial moiety such as a non-canonical amino acid or a metal-coordinating unit, generating artificial enzymes ( Figure 1C).
In summary, artificial enzymes have significant potential as novel green alternatives to traditional catalytic processes. They can be accessed through different strategies, but by all routes the choice of an appropriate biomolecular scaffold in which to introduce the new properties is a crucial factor for a successful design. In this context, this review aims to present an overview of the most prominent biomolecular scaffolds used to this extent, with a focus on their versatility and catalytic applications. The selected examples will be disclosed and categorized according to the strategy employed for the generation of artificial enzymes (as illustrated in Figure 1). In particular, section 2 will deal with the functionalization of already existing enzymes, disclosing examples like POP (unique conformational dynamics), [30] heme proteins [31] or carbonic anhydrase. [32] The focus of section 3 will be the creation of a new catalytic site in wide hydrophobic cavities, such as streptavidin (tight supramolecular substrate binding), [33] serum albumins (hydrophobic binding pocket), [34] multidrug resistance regulators (inherently substrate promiscuous), [35] and β-barrel proteins (large internal void, often transmembrane proteins). [12] In section 4 a short overview of the application of de novo design to artificial enzymes generation will be presented.
The review will be then concluded with our perspective on the properties that novel biomolecular scaffolds should possess to serve as an excellent starting point in the design of versatile artificial enzymes.

Phytase
The phosphatase phytase is a 65 kDa enzyme which catalyzes the hydrolysis of phytate to inorganic phosphate and myoinositol-phosphates. The structure of phytase consists of 2 domains, a larger α/β domain of six-stranded β-sheets and a smaller domain of a long α-helix surrounded by further αhelices. [36] Sheldon et al. generated a semi-synthetic peroxidase by loading a vanadium ion into the active site of phytase from Aspergillus ficuum. The binding of the metal was ensured by several amino acid residues, including two histidines (H404, H496), two arginines (R360, R490) and a lysine (K353). The resulting ArM promoted the hydrogen peroxide-mediated enantioselective sulfoxidation of several sulfides, reaching 66 % ee and full conversion in the case of thioanisole. [37] Additional metallic oxanions such as selenate, molybdate, tungstate and perrhenate were incorporated into the enzyme. However, the produced catalysts displayed much lower activity, demonstrating that only the vanadate-phytase complex was effective. [38,39]

Heme-proteins
Heme-proteins are a wide class of polypeptides bearing a heme unit (i. e., a protoporphyrin ring coordinating an iron atom) as a prosthetic group. Due to the presence of the heme, these proteins display a variety of functionalities, including the transport or storage of oxygen in tissues, the binding and activation of oxygen for enzymatic reactions and the ability to act as electron carriers. [40] Prominent examples, such as cytochrome P450s, myoglobin (Mb) and Rma-cytC, proved to be suitable candidates for efficient catalysis. An easy way to unlock new catalytic properties consists in modifying the native heme cofactor within the biomolecular scaffold, for example replacing the iron atom with another metal, or modifying the porphyrinlike frame (Figure 2A). In this context, the Fasan group substituted the iron in the myoglobin-based catalyst H64V/ V68A with non-native metals such as rhodium, iridium, ruthenium, and cobalt. It was discovered that the activity and selectivity of the afforded ArMs were highly influenced by the alteration of the metal center and the environment of the firstsphere coordination. These modifications led to engineered myoglobins with an expanded reaction scope and tuned catalytic activities, able to promote different transformations such as cyclopropanation and S-H ( Figure 2B) and NÀ H insertion reactions. [41] The same research group also developed an ArM based on myoglobin, but with an iron-chlorin e6 complex substituting the typical protoporphyrin core of the heme cofactor. The mutant H64 V/V68 A showed outstanding cyclopropanation activity towards vinylarenes under aerobic conditions, with ee up to 99 % and around 7,000 TON ( Figure 2B). [42] Another interesting class of artificial hemeproteins derives from the conjugation of frame-modified porphyrin cofactors with xylanase (Xln in Figure 2B), exploited as a protein host.
Xylanases are hydrolases able to break the β-1,4 bonds of the main chain of xylan. These proteins bear a positive charge and a wide active site, ideal to accommodate metalloporphyrins. For example, Nakamura and co-workers developed heat-resistant oxygen-carrying heme-proteins combining xylanase B with a synthetic iron(II) porphyrin motif (FeP, Figure 2A). [43] The obtained artificial enzymes represented the first case of completely synthetic thermostable O 2 -carriers. Similarly, Ricoux et al. functionalized xylanase A with iron(III)-carboxy-and iron(III)-sulfoxy-substituted porphyrins (Fe(TpCPP) and Fe-(TpSPP), respectively, Figure 2A). These hybrid heme-proteins were tested in the H 2 O 2 -mediated oxidation of peroxidase cosubstrates (guaiacol, 2,6-dimethoxyphenol, and o-dianisidine) and in the stereoselective oxidation of thioanisole to the (S)sulfoxide (yield up to 85 %, ee: 40 %, Figure 2B). [44,45] Notably, the same TpCPP framework was complexed with Mn(III) and incorporated in xylanase10 A, originating an ArM able to catalyze the enantioselective epoxidation of p-methoxystyrene, in the presence of KHSO 5 (yield: 16 % and best ee: 80 %, Figure 2B). [46]

Carbonic anhydrase
The human carbonic anhydrase (hCA) is a ubiquitous metalloenzyme responsible for the hydration of carbon dioxide, which plays a crucial role in many biological processes. [47] The monomeric protein has a size of 29 kDa and possesses a tunnelshaped cavity 15 Å in width and 15 Å in depth at its mouth. At the bottom of the pocket, a Zn(II) ion is coordinated in a tetrahedral manner by three histidine residues H94, H96, and H119 and a molecule of water. [32,48] Arylsulfonamides have a high affinity towards the Zn(II) ion, leading to inhibition of the function of the hCA by blocking its active site. Taking advantage of this affinity, aryl-sulfonamide substituted ligands can be anchored within the hCA. Ward and co-workers incorporated IrCp* pianostool complexes within hCA, resulting in ATHases able to reduce imines. The best performance was observed with the piano-complex hCA-1 ( Figure 3) with an ee of 68 %. However, wild-type (WT) ATHase displayed serious substrate inhibition, therefore the substrate binding pocket of the best ATHase was enlarged by incorporating single mutations I91 A and K170 A. The mutant I91 A of hCA-1 exhibited no substrate inhibition, and higher catalytic rates were observed. [32,49] For better localization of the Ir-pianostool cofactor within the hCA scaffold, a dually anchored ATHase was developed by the same research group. The covalent linkage of the cofactor in the scaffold was guaranteed by introducing a cysteine residue at position 191, undergoing a nucleophilic addition with a p-nitropicolinamide group attached to the cofactor. Activity and selectivity of the ATHase were enhanced by 3 rounds of directed evolution, obtaining an artificial enzyme able to catalyze the reduction of harmaline with 97 % ee for the (R)-isomer and TONs > 350. [50] Moreover, Ward and co-workers managed to access an artificial metathase based on the complex hCA-2, incorporating an arylsulfonamide-ligated Hoveyda-Grubbs catalyst within the scaffold. Through chemo- genetic optimization of the complex hCA-2 (Figure 3), the mutant L198H was identified and was able to perform ringclosing metatheses under physiological conditions with TON of 28. [51] A different strategy involves the generation of artificial enzymes by replacing the zinc ion of the hCA with other metals, giving access to new catalytic activities. [52][53][54][55] Kazlauskas and Soumillion reported that zinc substitution with manganese led to a hCA that functions as a peroxidase, with the ability to perform the epoxidation of styrene. [52,55] Moreover, when replacing the Zn(II) ion with rhodium, a carbonic anhydrase exploitable for the styrene hydroformylation and olefins hydrogenation was accessed. [54]

Papain
Papain is a cysteine protease obtained from the latex of the papaya tree Carica papaya. The enzyme is 23 kDa in size and consists of 2 domains, (L and R domains), that are similar in size, but represent different conformations. Whereas the R domain displays an antiparallel β-sheet, the L domain portrays an αhelix. The active site can be found at the interface of the two domains, in a groove bearing a catalytic dyad formed by residues C25 and H159. [56,57] Papain is a suitable protein host that can be used for anchoring metal complexes to achieve hybrid catalysts. Salmain and co-workers reported maleimidesubstituted metal complexes that can be covalently linked to C25. They tested several maleimide derivatives in enzymatic and ESI-MS studies to check if the maleimides underwent alkylation with the C25 and therefore inhibit the catalytic activity. [58,59] By taking advantage of this strategy, a maleimidesubstituted (η 6 -arene)Ru(II) complex was incorporated into papain and tested in a Diels-Alder reaction of cyclopentadiene and acrolein. Even though the uncatalyzed reaction exhibited a conversion of 62 % in 250 min, the (η 6 -arene)Ru(II) anchored metalloenzyme achieved a conversion of 88 % with a TOF of 220 in 30 min. [60] Moreover, in the same research group, maleimide and chloroacetamide substituted ruthenium and rhodium complexes [(η 6 -arene)Ru(N^N)Cl] + and [(η 5 -Cp*)Rh-(N^N)Cl] + were conjugated to papain, again at C25. The resulting catalysts exhibited dehydrogenase activity and were able to promote the reduction of NAD(P) + into NAD(P)H. Rh(II)complexes showed greater performance than the Ru(II). [61] These (arene)Ru(II) and (Cp*)Rh(III) complexes carrying a dipyridylamine ligand were further examined in a transfer hydrogenation reaction of trifluoracetophenone. The resulting enzymes were all active, showing high conversion but low enantioselectivity and again Rh(III)-based biocatalysts exceeded metallopapain with a Ru(II)-complex. [62]

tHisF
tHisF is a subunit of the imidazole glycerol phosphate synthase that derives from the thermostable organism Thermotoga maritima and is crucial in the biosynthesis of histidine. The enzyme has a molecular weight of around 28 kDa, and its structure displays a TIM barrel with its classic eight-folded α,βmotif, where both ends are open and wide on the top and narrow on the bottom. [63][64][65] THisF possesses interesting properties making it a promising protein host for the assembly of biocatalysts. First, expression in E.coli is established and the purification can be performed on a large scale, with the advantage that cell lysis can be accessed through heat treatment, leading to the precipitation of most unwanted proteins, due to the thermostability of the enzyme. Moreover, the top TIM-barrel opening, where the catalytic site lies, is wide enough to constitute a suitable environment for introducing a metal complex. Reetz and co-workers originated suitable binding sites for Cu(II) ions within the tHisF by incorporating a 2-H-1-D motif next to favorable existing amino acid residues. [65] They envisioned that the 2-H-1-D motif could be promising for the binding of Cu(II), as the triad is known in Fe(II) metalloenzymes. For this reason, the aspartate residue at position 11 was chosen and additional histidine residues at positions 50 and 52 were introduced through site-directed mutagenesis. Moreover, C9 was exchanged for an alanine as it lies in close proximity to the potential metal complex and could lead to interference. The resulting ArM was employed to catalyze a Diels-Alder reaction and showed higher catalytic activity than the WT with an ee of 35 % for the desired endo product (endo/exo = 14 : 1). In order to eliminate further interference with the metal complex, four additional histidine residues (H84, H209, H228, H244) were exchanged to alanine and the resulting variant HHD-4xala was again tested in the Diels-Alder reaction, displaying an ee of 46 %. To support these findings, a control experiment was run with a mutant where the key D11 residue of the triad was mutated to an alanine. The ee decreased drastically (4 %) leading to more evidence that the motif D11/H50/H52 is indeed responsible for the coordination of Cu(II) ions. As a crystal structure was not provided, electron paramagnetic resonance spectroscopy (EPR) was performed to obtain further evidence that supported the findings. Hyperfine sublevel correlation (HYSCORE) spectroscopy exhibited that in the HHD-4xAla mutant the coordination of Cu(II) was occurring through a histidine moiety, but not in the case of the negative control NC-4xAla, which does not carry the two histidine residues at position 50 and 52. Therefore, the only histidine residues involved in the coordination of Cu(II) are H50 and H52 of the 2-His-1-Asp motif, demonstrating that the Diels-Alder reaction occurs in this binding site. [64][65][66] Moreover, Lewis and co-workers took advantage of strain-promoted azide-alkyne cycloaddition in order to anchor bicyclononyne-substituted metal complexes within tHisF. By incorporating p-azidophenylalanine within the protein host, the amino acid underwent an azide À alkyne cycloaddition with the alkyne-substituted metal complex, leading to an ArM able to promote several transformations such as SiÀ H and olefin insertions. [67] A few years later the same group applied this method in other protein scaffolds such as POP. [68]

CAL-B
The Candida antarctica lipase B (CAL-B) is an esterase composed of an eight-stranded β-sheet surrounded by α-helices, while the active site is characterized by a SÀ H-D catalytic triad (Figure 4A). [69,70] Gebbnik and co-workers reported the anchoring of a Ph 4 CpRu(CO) 2 Cl catalyst to an immobilized CAL-B. In this way, a bifunctional catalyst was created, and tested on the sequential racemization of the (S)-1-phenylethanol, and acylation of the racemate to (R)-1-phenylethyl acetate, with outstanding ee of up to 99 % ( Figure 4B). [71] In another study of the same group, a Rh-NHC phosphonate complex (Rh-pNP in Figure 4A) was successfully incorporated into both lipases CAL-B and cutinase (see paragraph 2.8). These hybrid catalysts were tested in the hydrogenation of a mixture of acetophenone and methyl 2acetamidoacrylate. Due to the sterically hindered active site, Rh-pNP-CAL-B displayed high chemoselectivity and yields towards the reduction of the olefin, while the ketone remained untouched ( Figure 4B). [72] Another example, reported by Palomo and co-workers, involved the anchoring of a p-nitrophenylphosphonate palladium pincer complex (Pd-pincer in Figure 4A) to the catalytically active serine of CAL-B. The artificial enzyme was immobilized on various supports and tested in Heck crosscouplings. In particular, enzyme SP-CAL-B-C8-1 showed excel-lent enantioselectivities of 97 % in the reaction between iodobenzene and 2,3-dihydrofuran ( Figure 4B). [73]

Prolyl oligopeptidase
An interesting example of the introduction of new catalytic properties in an existing enzyme was reported by Lewis and coworkers, that modified the prolyl oligopeptidase (POP) into a novel artificial metalloenzyme (POP-ArMs). POP belongs to a family of serine peptidases and can specifically hydrolyze peptides composed of less than 30 amino acids. [74] The crystal structure of prolyl oligopeptidase (POP) was elucidated in 1991, demonstrating that POP is composed of a peptidase domain with a hydrolase fold and a seven-bladed -propeller domain. Structurally, POP displays a cylindrical conformation, with an approximate height of 60 Å and diameter of 50 Å. [75] The activity of the enzyme depends on the catalytic triad S554, D641 and H680, which is located at the interface of the two domains. To expand POP's reaction scope, Lewis and co-workers incorpo-rated the non-canonical amino acid p-azidophenylalanine at position S477, to introduce an anchor point for the conjugation with a bicyclononyne-substituted dirhodium complex, exploiting the azide À alkyne cycloaddition. [76] The accessed POP-ArM found application in the cyclopropanation of a wide range of styrenes and diazoester compounds. To further broaden the POP internal cavity to accommodate larger cofactors, four bulky residues in proximity of the active site (E104, F146, K199, and D202) were mutated to alanine. Moreover, by introducing a histidine residue in the active site (H328), a two-points cofactor binding was assured and resulted in higher enantioselectivity with up to 85 % ee and 61 % yields. A further optimization involved the introduction of phenylalanine residues at positions G99 and G594. Moreover, variant 3-VRVH, displaying 12 mutations, showed even better performances (92 % ee), whereby only 3 mutations contribute to substrate selectivity. After Lewis' group solved the crystal structure of POP, it emerged that only two of these mutations involved residues inside the active site, while the remaining 10 mutations were far away. These findings demonstrated the importance of engineering ArMs second-coordination sphere to improve their catalytic performances in abiological transformations. [30,76] Recently, the same research group developed a series of different photocatalysts by covalently and non-covalently conjugating Ru-(Bpy) 3 2 + cofactors into POP. Different combinations of the substituents on the bipyridine and the coordination position within the scaffold were tested to improve the binding and the luminescence properties of the ArMs. The covalently conjugated proteins were chosen to study two model photochemical transformations: the 5-exo-trig reductive cyclization of a dienone via single-electron-transfer (SET) and the [2 + 2] cycloaddition between a C-cinnamoyl imidazole and an electron-rich styrene. For both reactions, the ArMs gave improved yields and rate acceleration compared with the metal cofactor used alone, due to an entropic effect. As almost no preference towards one of the enantiomers was observed, it appears that the binding pocket of POP is not suitable to ensure stereoselectivity. [77]

Miscellaneous
In this paragraph, we collect less common examples of natural enzymes that were functionalized to generate artificial catalysts.
One of these cases is chymotrypsin, a serine protease consisting of two domains divided by a cleft, in which four subsides (S1-S4) can be found. Hirta et al. accessed an artificial enzyme by incorporating the L-phenylalanyl chloromethylketone-based inhibitor fused with an Hoveyda-Grubbs moiety into the hydrophobic pocket of chymotrypsin. The resulting metalloenzyme promoted a ring-closing metathesis of a watersoluble diolefin compound. [78] The Tm1459 cupin superfamily protein is a thermostable homodimeric Mn-binding protein deriving from the organism Thermotoga maritima. Itoh and coworkers converted the protein into an osmium peroxygenase by replacing the manganese. The catalyst showed efficient cisdihydroxylation performances on several alkenes with TON = 9100 in the case of 2-methoxy-6-vinyl-naphthalene. The substitution of the osmium not only gave access to new reactivities, but even increased thermostability of the protein up to a T m of 120°C. [79] Finally, the lipase cutinase from Fusarium solani pisi, is characterized by a key β-sheet motif with five parallel strands surrounded by α-helices with a SÀ HÀ D catalytic triad in the active site, which is accessible to hydrophilic and hydrophobic substrates. [80,81] The active serine of cutinase could be functionalized to link organometallic complexes, spanning from NCN-pincer-Pt, to ECE-pincer-Pt or -Pd, [82] to Grubbs complexes. [83] These ArMs were exploited in crystallographic studies to get more insight into the binding of the metal complex inside the protein scaffold [82] and to perform cross-metathesis of allylbenzene and ring-closing metathesis of N,N-tosyl diallylamine. [83] 3. Functionalization of Wide Hydrophobic Pores (without any Previous Activity)

Streptavidin
Over the last decades, many ArMs were created on the basis of the biotin-streptavidin technology, initiated in 1978 by Wilson and Whitesides. [30,84] Streptavidin is a tetrameric protein isolated from the bacterium Streptomyces avidinii. Each monomer is constituted by an antiparallel β-barrel originated by the folding of eight antiparallel β-strands. Each β-barrel accommodates a binding site for biotin and as a result, tetrameric streptavidin can simultaneously bind four biotin molecules. This is one of the strongest non-covalent interactions between a protein and a ligand reported in nature (K d � 10 À 14 M) and has found several applications in biotechnology. [85] The biotin-binding vestibule is formed by residues S112 and K121 of adjacent streptavidin monomers (SavA and SavB) and the affinity is ensured by strong non-covalent interactions between the active site residues and the bicyclic structure of biotin. For instance, van der Waals interactions and hydrophobic effects are ensured by tryptophan residues W79, W92, W108, and W120; residues D128, N23, S27, S45, and Y43 guarantee H-bond formation, and loop structures on the surface allow conformations changes between open and close state, where the closed conformation is further reinforced after biotin binding. [86] The Ward group has exploited this homo-tetrameric protein thoroughly in the last 20 years, creating various ArMs based on the streptavidin technology. [33,87,88] These ArMs found their application in 13 different reactions such as hydrogenation, [89] olefin metathesis, [90] alcohol oxidation, [91] asymmetric transfer hydrogenation (ATH) of enones, [92] imines [93,94] and ketones, [95] sulfoxidation. [96] This impressive versatility depends on the excellent affinity of biotin for streptavidin, but also on the other great features that the biomolecular scaffold possesses. Streptavidin is prone to comprehensive genetic alterations, facilitating the directed evolution process. Moreover, the protein can be expressed in fed-batch cultures in high quantities (up to 8 g/ L). [33,97] Optimization can occur with chemical and genetic alterations, either by varying the biotin ligand complex or incorporating mutations in the streptavidin scaffold by screening libraries of the protein with various biotinylated cofactors. In this way promising cofactors can be rapidly identified. [98] For the design of ArMs, streptavidin can be employed in different oligomeric states. Ward and co-workers incorporated biotinylated pianostool complexes into the tetrameric form, to examine the transfer hydrogenation of imines. Using the artificial enzyme incorporating [(η 5 -Cp*)Ir(Biot-p-L)Cl] at position Sav S112 A, the required secondary amine was obtained with 4000 TON and an ee of 96 % for the (R)-isomer ( Figure 5A). [94] They also found out that by increasing the ligand/protein ratio four times, the stereoselectivity decreased, because of the interaction among neighboring cofactors. This suggested that the secondary coordination sphere was modified at different cofactor-protein ratios. On the other hand, when investigating [(η 5 -Cp*)Ir(Biot-p-L)Cl] at position S112K, no impact on the stereoselectivity was observed. Kinetic and X-ray analysis has shown that all four binding positions of S112K can be used by the cofactor. This example showed that ArMs based on the tetrameric streptavidin scaffold possess some limitations. When incorporating point mutations, all four subunits will reflect these mutations, making it harder to implement genetic finetuning of the ArM. Moreover, a decrease in the catalytic performances could derive from the non-cooperative binding of the biotinylated cofactor to streptavidin. [99] To overcome these limitations, the same research group established a single-chain dimeric streptavidin (scdSav), allowing for the incorporation of mutations in the two adjacent streptavidin subunits independently. [100] 33 scdSav-based ArMs were tested in their ATHase activity towards various challenging substrates. It was found that the generated ArMs displayed higher selectivity and activity than ATHases based on homotetrameric streptavidin. However, non-cooperative binding of the cofactor to streptavidin remained still problematic, because even at ratios of � 1, part of the cofactor [(η 5 -Cp*)Ir(Biot-p-L)Cl] underwent ineffective binding, resulting in a reaction rate loss. Therefore, the monovalent scdSav variants scdSav(SARK)mv1 and scdSav(SARK)mv2 were engineered, eliminating the interaction between neighboring cofactors. Catalytic studies showed that the scdSav(SARK)mv2 variant produced the reduced amine with a TON > 17000 and an ee up to 96 %, outperforming the divalent scdSav ( Figure 5A). [100] A different approach was finally reported by Luk's research group. [15] Instead of creating ArMs, they developed proteinorganocatalyst hybrids. Taking inspiration from well-known iminium-based organocatalysts, like L-proline or cyclic imidazolidinones such as MacMillan's catalysts, they coupled the key catalytic secondary amine (proline, pyrrolidine or 4-imidazolidinone) with biotin through a spacer. The accessed biotinylated organocatalysts alone were tested in promoting the Michael addition of nitromethane to cinnamaldehyde, a reaction that involves as a key step the formation of an iminium intermediate between the secondary amine and the cinnamaldehyde. The first screening revealed a significant background reaction and lack of stereoselectivity. To verify if the anchoring on a protein host could have been beneficial for the reactivity, the biotinylated-organocatalysts were conjugated with tetrameric streptavidin and tested again on the model Michael addition. It emerged that the protein-conjugation improved the performances of the two pyrrolidine-based biotinylated-catalysts (Figure 5B), with the biomolecular scaffold that appears to be crucial to guarantee the stereoselectivity. Interestingly, the two generated artificial enzymes, that differ only in the configuration of the pyrrolidine stereocenter, afforded opposite enantiomers of the final product, even if with moderate yields (15-30 %). After conditions optimization, the (R)-pyrrolidine-Sav promoted the generation of the Michael adduct with 80 % yield and 91 : 9 er (in favor of the (S)-enantiomer). This example is a nice demonstration of the benefit of a chiral supramolecular complex with a simple organocatalytic motif, to improve activity and stereoselectivity.

Serum albumins
Albumin is a globular, un-glycosylated serum protein with a molecular weight of 65 kDa. This protein is the most abundant in mammals' blood and is composed of three homologous domains (I, II and III), each one containing two sub-domains, labeled A and B. Despite lacking a specific catalytic site, serum albumin is able to bind a broad range of hydrophobic molecules; [101] for this reason, since 1980, bovine and human serum albumins (BSA and HSA, respectively) have been identified as powerful biocatalysts, able to accelerate several organic reactions such as asymmetric oxidations and reductions, additions, condensations and eliminations. [102] This versatility can be indeed ascribed to the absence of a specific binding site, typical of the majority of natural enzymes. Both BSA and HSA appear to elicit their catalytic activity through their hydrophobic pockets, located on the subunit A of domain II, and in this environment the key residues are K222 in BSA and K199 in HSA. [103][104][105] Serum albumins are therefore good biocatalysts even in their native state and they promote a wide plethora of reactions (e. g. Biginelli, Morita-Baylis-Hillman, Kemp elimination, aldol condensation), in both aqueous and organic solvents. [34] However, their catalytic properties could be further expanded when complexed with transition metals ions, originating in this way interesting ArMs, of which general structures and reactivities are summarized in Figure 6.
For example, several studies reported the use of BSA-ArMs and HSA-ArMs to promote (enantioselective) sulfoxidation reactions ( Figure 6B). Mahammed and Gross prepared Fe(III)and Mn(III)-corrole complexes, that were then conjugated with HSA, obtaining catalysts able to promote the enantioselective oxidation of prochiral sulfides to the corresponding sulfoxides. [106] Thus, differently substituted thioanisoles were treated in the presence of the metal-corrole complex, serum albumin and hydrogen peroxides for 1.5 h at 24°C. It resulted that the Mn-corrole-SA artificial enzyme gave better yields (often > 90 %) and ee (> 60 %) compared with the corresponding iron complex and was also more stable. Moreover, its mode of action resembled that of some heme enzymes. A similar reactivity can be promoted by Schiff base complexes conjugates with BSA, as reported by Tang and collaborators. [107,108] In detail, they prepared a small library of artificial enzymes, exploiting the non-covalent conjugation of BSA with different metal-Schiff base complexes, characterized by the general structure M-L (where the metal could be Co, Mn, V, Fe or Cr, and the ligand could be 2-hydroxynaphthalen-1-naphthaldehyde and 3,4-diaminobenzenesulfonic acid). These ArMs were tested on the enantioselective sulfoxidation of differently substituted prochiral thioanisoles, performed in aqueous media and the presence of hydrogen peroxide as oxidant. From the preliminary results, it was already clear that the BSA-ML conjugates gave better yields and chemoselectivity compared with the simple ML complexes. Even the stereoselectivity, although still unsatisfactory, was improved (ee up to 32 % for BSA-ML, ee < 5 % for ML). These results suggested that the BSA hydrophobic binding pocket played a key role, favoring the collision of the substrates. After the optimization of reaction conditions (that can be influenced by pH, substrate, oxidant and catalyst concentrations), all of the five ArMs showed promising catalytic activity, with the best results obtained for the vanadium-based Schiff base-BSA conjugate (yield up to 100 %, ee up to 96 % and TOF up to 500 h À 1 ).
Other interesting transformations that can be unlocked by BSA-based ArM include alkenes hydroxylation and hydroformylation and Diels-Alder cycloadditions.
In 1983, Kokubo developed a 1 : 1 BSA-OsO 4 complex (BSA-2-phenylpropane-1,2-diolatodioxo-osmium) that could promote the cis-hydroxylation of alkenes ( Figure 6B), affording diols with ee up to 68 %. [109] Notably, similar OsO 4 -based artificial enzymes were prepared by Ward in 2011. Their work was focused more on streptavidin-OsO 4 complexes, but a BSA-OsO 4 was considered as a comparison and tested on the asymmetric hydroxylation of α-methylstyrene. Interestingly, even if the best results in terms of ee and TON were obtained for the streptavidin ArM (ee: 95 %, TON: 27), also the BSA-OsO 4 complex displayed interesting catalytic properties, favoring the formation of the opposite enantiomer, with decent ee (77 %), but lower TON of 4. [110] Furthermore, alkenes can undergo BSA ArM-mediated hydroformylation ( Figure 6B), as reported in Marchetti's work. [111][112][113] He accessed a new nanostructured, water-soluble biocatalyst, originated by the interaction between Rh(CO) 2 (acac) and HSA. This Rh(I)-BSA complex was tested on the biphasic hydroformylation of styrene and other olefins, performed in an autoclave, pressurized to 70-80 atm with CO/H 2 = 1, at 40-60°C. Considering the relatively high reaction temperature, it has been suggested from denaturation studies that the metal should stabilize the protein's secondary structure. Under these conditions, several olefins were hydroformylated, with high conversion and regioselectivity for the branched aldehyde.
Serum albumin was also the biomolecular scaffold of choice for the generation of artificial metalloenzymes able to promote Diels-Alder reactions, as reported by Reetz and Jiao (Figure 6B). [114] They took advantage of the strong non-covalent interaction between different serum albumins (human, porcine, rabbit, sheep and chicken egg) and a Cu(II)-phthalocyanine complex, to access novel biocatalysts able to promote the stereoselective cycloaddition between differently substituted azachalcones and cyclopentadiene. HSA proved to be the most efficient biomolecular scaffold in this study (even though also porcine and sheep serum albumins performed reasonably well), and in the best conditions afforded the desired product with endo/exo ratio of 96 : 4, enantioselectivity of 93 % for the major stereoisomer and with only 2 mol% of catalyst loading.
Finally, an interesting work by Xia et al. highlighted the possibility of originating peroxidase-like biocatalysts via biomineralization, to assemble platinum salt-resistant nanoparticles (PtNPs) in BSA protein scaffold. [115] This artificial metalloenzyme displayed the advantage of preventing the Pt nanocrystals agglomeration, which often occurs in colloidal solutions as the ionic force (given by additive salts) increases. The accessed BSA-PtNPs complexes were able to efficiently promote the H 2 O 2mediated oxidation of several chromogenic substrates, such as of 3,3',5,5'-tetramethylbenzidine, 2,2'-azino-bis(3-ethyl benzothiazoline-6-sulfonic acid) diammonium salt, o-phenylenediamine, 3,3'-diaminobenzidine, pyrogallol, 4-aminoantipyrine/ phenol, and 3-amino-9-ethylcarbazole. Notably, the peroxidase activity of the artificial enzyme was preserved even in a high ionic strength environment and the BSA-PtNPs complex proved to possess a higher affinity for hydrogen peroxide compared with the nanoparticles alone.

Ferritin
Ferritin is an iron-storage protein with a large hollow cavity around 80 Å, able to accumulate up to 4500 Fe(III) ions. The structure of ferritin is formed by 24 subunits (20 kDa each) divided into a heavy chain (H) and a light chain (L), displaying two types of channels, a polar threefold channel and a nonpolar fourfold channel. Iron enters the protein through the threefold channel inside the capsid, while molecular oxygen required for Fe(II) oxidation, enters through the fourfold channel. [116,117] Watanabe and Ueno incorporated [Rh(nbd)Cl] 2 complexes inside apo-ferritin, developing a hybrid catalyst able to promote the polymerization of phenylacetylene. [118] In a follow-up study, QM/MM studies were performed to obtain more insights into the reaction mechanism. It emerged that a new hydrophobic pocket, composed of the side chains of residues F50, K143 and L171 is the true active site in which the coordination and polymerization of phenylacetylene take place. [119] In the same research group dinuclear [Pd(allyl)Cl] 2 -complexes were also anchored to ferritin. The resulting artificial enzyme was able to catalyze Suzuki cross-coupling reactions. [120]

Nitrobindin
Another protein scaffold that has been widely used for the construction of metalloenzymes is the heme protein nitrobindin. This scaffold possesses a rigid 10-stranded β-barrel structure with a heme cofactor in its hydrophobic cavity. Several studies demonstrated that the removal of the heme cofactor originates a suitable environment for accommodating metal complexes, ultimately accessing novel artificial enzymes. [7,[121][122][123][124][125][126][127] Groups of Hayashi, Okuda, and Schwaneberg accomplished a biocatalyzed metathesis by incorporating cofactors such as rhodium or Grubbs À Hoveyda catalysts into the mutant called NB4 (M75L-H76L-Q96C-M148L-H158 L) of nitrobindin. Covalent linkage between the ligand and the scaffold was achieved by conjugation of the maleimide substituted complex with an engineered cysteine residue at position 96 (Figure 7A). [121,123,124,128] In detail, Okuda and co-workers reported the generation of a hybrid biocatalyst incorporating a Rh-complex in nitrobindin, able to promote the polymerization of phenylacetylene (Figure 7B). [124] Interestingly, mutagenesis was exploited to optimize the active site shape, favoring the selectivity for the transcompound. The same NB4 variant was exploited as a starting point for the design of artificial enzymes incorporating a Grubbs catalyst-like complex. Considering the steric hindrance of the complex, the cavity of NB4 was enlarged through further mutations, resulting in the variant NB11 (NB4 L75A-L158 A). [123] The accessed NB11-Grubbs enzyme was successfully employed to catalyze the ring-closing metathesis of 4,4-bis(hydroxymethyl)-1,6-heptadiene and the ring-opening metathesis polymerization of a 7-oxanorbornene derivative, accessing TON values > 100 and > 9000, respectively ( Figure 7B). The same authors managed to perform a biocatalyzed metathesis in whole cells, showcasing the first ever reported cell surface display-based whole-cell biohybrid catalyst (ArMt bugs). Therefore, the nitrobindin variant NB4 with a Rh catalyst was fused with an inactivated esterase autotransporter, to display the catalyst on the E. coli surface. Through this platform, they were able to perform the polymerization of phenylacetylene with higher efficiencies up to 39 × 10 6 TONs (per cell). [130] Recently, Hayashi and co-workers took advantage of this cell surface display system by constructing a biohybrid catalyst catalyzing the cycloaddition of acetophenone oxime with diphenylacetylene. For this purpose, they incorporated a (pentamethylcyclopentadienyl)rhodium(III) (Cp*Rh(III)) complex with latent catalytic activity into the cavity of nitrobindin, which is displayed on the E. coli surface. By addition of Ag(I) ions, the dithiophosphate ligands are dissociated and the biocatalyst activated, promoting the desired reaction. [131] Many different artificial cofactors could be incorporated into the nitrobindin scaffold, showing performance in transformations like CÀ C bond formation, peroxidase, [125] epoxidase [126,127] and hydrogenase activity. [132]

FhuA
The ferric hydroxamate uptake protein component A (FhuA) is a large channel protein known to be one of the largest β-barrels, with a molecular weight of 79 kDa. The protein consists of 22 antiparallel β-sheets and has a "cork" domain which functions as a plug regulating iron uptake from the outer membrane of E. coli. Okuda and co-workers engineered FhuA removing the "cork" domain, obtaining a passive mass transfer channel FhuA Δ1-159 wide enough to bind metal complexes up to 14 Å. The covalent binding was ensured by a K545C mutation, introducing a cysteine as a handle for maleimide-substituted complexes anchoring, analogously to nitrobindin ( Figure 7A). For example, in Okuda's group, the cysteine was exploited to conjugate Rubased Grubbs-Hoveyda catalytic moieties. Further engineering was performed introducing mutations N548V and E501F for improving accessibility of the cysteine residue. Additionally, a Tobacco Etch Virus (TEV) cleavage site was inserted, enabling MS analysis of the incorporated catalysts. The final variant FhuA Δ1-159_C545 V548_F501_tev was then anchored with a Grubbs-Hoveyda type catalyst via a maleimide linker (hybrid FhuA-Grubbs) and tested in ring-opening metathesis polymerization of a 7-oxanorbornene ( Figure 7B). The exploited catalyst displayed high conversion but lacked in terms of selectivity. [133] Interestingly, a FhuA-Grubbs catalyst was used by the same group to promote sequential one-pot reactions, in combination with a FhuAÀ Rh complex (depicted in Figure 7A). In detail, the FhuA-Grubbs complex was involved in the first step of the cascade reaction, namely the cross-metathesis of chloro-styrene, while the FhuAÀ Rh biocatalyst promoted the hydrogenation of the double bond. In this way, the chloro-styrene was converted into dichloro-diphenylethane with 87 % yield ( Figure 7B). The reaction was known to be one of the first examples in which two hybrid catalysts are employed in a one-pot reaction. [134] Followed by this success, further cascade reactions were examined, such as the synthesis of stilbene derivatives from cinnamic acids. In this case, the first step (the decarboxylation of the cinnamic acid) was catalyzed by the ferulic acid decarboxylase (FDC1), while the previously reported FhuA-Grubbs complex allowed the cross-metathesis of the accessed styrene. [135] In an additional study, Okuda investigated how the length of the maleimide linker affected the selectivity and activity of the Grubbs catalyst. They observed that shortening the linker from a 1,3-propanediyl to methylene resulted in higher turnover numbers and increased the cis selectivity. [136] However, FhuA ΔCVFtev can be exploited for other reactivities. In the same research group, copper(I) NHC and copper(II)terpyridyl complexes were conjugated via maleimide linkers with the protein scaffold. The accessed artificial enzymes (FhuA-NHC-Cu and FhuA-terPy-Cu) displayed high activity in promoting the Diels-Alder reaction between azachalcone and cyclopentadiene, with high endo selectivity ( Figure 6B). [137] The reaction scope of FhuA ΔCVFtev-based artificial enzymes was further expanded towards cyclotrimerization of phenylacetylene and alkene-alkyne coupling, promoted by η5 -cyclopentadienyl)cobalt(I) [138] and ruthenium complexes. [139]

LmrR and other multidrug resistance regulators
A biomolecular scaffold which finds great application in the design of artificial enzymes is LmrR, a multidrug resistance regulator from Lactococcus lactis. This dimeric protein is composed of a β-winged helix-turn-helix domain, with a Cterminal helix involved in the dimerization. The two subunits create a large, flat-shaped hydrophobic pore that in the natural LmrR serves as a multi-drug binding site. In particular, the tryptophan residues at positions W96 and W96', located in the Figure 7. A) General structure of nitrobindin-and FhuA-based artificial metalloenzymes (adapted from PDB: 3WJB [124] and 1BY3 [129] ), displaying the covalent binding of maleimide-functionalized cofactors with C96 and C545. B) Representative examples of the reactivity of these artificial enzymes. proximity of the pocket, appear to be crucial for binding planar aromatic molecules. [140] With its great versatility, LmrR is able to accommodate multiple metal complexes, therefore binding different substrates and promoting various reactivities. These features make LmrR's hydrophobic pore an outstanding starting point in the design of artificial enzymes. In this context, Bos and co-workers created promising ArMs from LmrR, exploiting introduced cysteine residues within the biomolecular scaffold for the covalent linkage of Cu-phenanthroline and Cu-bipyridine complexes. The accessed artificial enzymes were successfully applied to Diels-Alder reaction and conjugate addition of water to �,β-unsaturated carbonyl compounds. [141,142] In the following years, two other strategies for the introduction of novel catalytic activity in LmrR found wider applications: the use of metalbinding non-canonical amino acids (incorporated into the biomolecular scaffolds through the genetic code expansion), and the supramolecular assembly of metal-aromatic ligand complexes within the two coordinating W96 residues of LmrR ( Figure 8A). [4,35,[143][144][145] Roelfes and co-workers introduced the metal binding amino acid (2,2-bipyridin-5yl)alanine (BpyA) at various positions into the LmrR scaffold, taking advantage of amber stop codon suppression methodology. [146] The resulting ArMs were able to perform Friedel-Crafts alkylation of indoles with α,β-unsaturated 2-acylimidazoles, showing the best results when incorporated at position 89 (LmrR_M89BpyA), with ee up to 80 % ( Figure 8B). By introducing an additional F93 W mutation in the active site, the ee could be improved to 83 % and the conversion to 94 %. [4] Encouraged by these results, they investigated more challenging reactions such as the enantioselective hydration of enones, accessing chiral β-hydroxy ketones ( Figure 8B). Via quantum mechanical calculations, molecular dynamics simulations and protein-ligand docking, 3 variants of LmrR_M89BpyA with glutamate mutations were identified, that could perform the desired reaction. For the mutant V15E, an improvement in enantioselectivity and yield towards the βhydroxyketone product was observed. Compared with the parental designer enzyme, LmrR_M89BpyA_V15E displayed a threefold higher catalytic efficiency. [143] However, other metalbinding amino acids such as (8-hydroxyquinolin-3-yl)alanine (HQA), showed great potential in the design of ArMs. The two mutants LmrR_V15HQA and LmrR_M89HQA displayed not only a good affinity for Cu(II) ions but also for Zn(II) and Rh(III) metal ions. The resulting LmrR_HQA variants coupled with Cu(II) promoted Friedel-Crafts alkylation and water addition reactions ( Figure 8B), whereas ArMs complexed with Zn(II) ions possessed hydrolase activity. [14] Taking advantage of the two central tryptophan residues within LmrR, Bos et al. managed to assemble a Cu(II) complex within LmrR hydrophobic pocket generating an ArM able to catalyze enantioselective Friedel-Crafts alkylation of indoles with ee up to 94 %. For the mutant LmrR_W96A�Cu(II)-phenanthroline moderate conversions and ee < 5 % were observed, demonstrating again the importance of central tryptophan residues for binding of hydrophobic substrates ( Figure 8B). [147] Interestingly, this supramolecular approach was not limited to copper complexes, but also heme groups such as hemin could be coordinated into LmrR. [145] Moreover, the incorporation of non-canonical amino acids in LmrR was exploited by Roelfes' research group to introduce not only a coordinating handle but directly a new catalytic site. In fact, in 2018 Drienovská et al. obtained new metal-free artificial enzymes, bearing an aniline moiety, deriving from the incorporation of the non-canonical amino acid p-azidophenyl alanine, that was eventually reduced in situ to the corresponding amine (pAF). [148] This organocatalytic-like moiety, once embedded into LmrR's hydrophobic pore, acted as a promising catalyst in promoting the formation of aromatic hydrazones ( Figure 8B). Intriguingly, the key W96 and W96' residues played a fundamental role in interacting with the aromatic substrates (4-nitro-or 4-hydroxybenzaldehyde and 4-hydrazino-7-nitro-2,1,3-benzoxadiazole), while the primary amine of pAF was crucial in promoting the hydrazone formation, in particular when replacing V15. The accessed LmrR_V15pAF variant was the most efficient enzyme developed in this study and was also successfully applied to the synthesis of the oxime deriving from 4-nitrobenzaldheyde and O-(7-nitrobenzo-[2,1,3-d]-oxadiazol-4yl)hydroxylamine. More recently, the reaction scope of pAF- functionalized LmrR was expanded towards Friedel-Crafts alkylation of indoles with enals [149] and tandem Friedel-Craftsalkylation/enantioselective protonation of 2-methylindole and methacrolein. [150] Encouraged by these findings, other multidrug resistance regulators (MDRs) were explored by the same research group. In particular, CgmR, RamR and QacR were considered. These proteins belong to the Tet family and similarly to LmrR, are homodimers with a large hydrophobic pore involved in drug recognition. [151][152][153] BpyA was incorporated at four various positions into these MDRs, obtaining artificial enzymes able to perform Friedel-Crafts alkylations with low to moderate enantioselectivity. Out of all the mutants prepared, reactions performed with QacR_Y123BpyA showed outstanding enantioselectivity up to 94 %. [154] Furthermore, as these scaffolds have the optimal environment restraining cationic and aromatic molecules, they were assembled with Cu(II) ions, without a ligand anchoring the metal. The generated ArMs were tested in the enantioselective vinylogous Friedel-Crafts alkylation of α,βunsaturated imidazoles and showed moderate performance with enantioselectivity up to 75 % for the Cu(II)-QacR complex. [155] Moreover, these MDRs were tested in the Friedel-Crafts/enantioselective protonation (F-C/EP) of indoles. It emerged that QacR_C72A_C141S (alone and with the addition of Cu(II) ions), showed moderate yields and enantioselectivity of 75-83 % at lower pH. On the basis of computational studies, it was found that E120 functions as a catalytic residue and therefore is responsible for the donation of a proton to the reaction intermediate. [156]

Human steroid carrier protein SL
Another interesting candidate for the design of artificial enzymes is the human steroid carrier protein SL (SCP-2L). This protein presents a cylindrically shaped tunnel (18 Å in length, 10 Å diameter), that originates a suitable environment to accommodate a wide range of linear aliphatic substrates. [157][158][159] Using the SCP-2L as starting point, the Kamer group introduced rhodium-phosphine complexes into the hydrophobic tunnel, accessing a catalyst able to perform the hydroformylation of linear alkenes. Covalent linkage of the metal complex with SCP was also in this case ensured by bioconjugation between maleimide containing ligands and cysteine residues, that were introduced at the end of the hydrophobic tunnel at positions V83 and A100, leading to mutants SCP-2LV83C and SCP-2L A100C. [158] To further improve the catalytic performances and obtain more in-depth knowledge of the active site of the hydroformylase, biophysical studies and site-directed mutagenesis were performed. Therefore, the impact of protein modifications on four methionines (M1, M80, M105 and M112) in proximity to the active site was investigated using various NMR techniques to probe the structure. It emerged that engineering the methionine M1 for both variants SCP-2LV83C and SCP-2L A100C led to higher selectivity and catalytic activity. [160] Recently, Jarvis and co-workers investigated through sequence-based mutant selection and rational design which mutations in SCP-2LV83C and SCP-2 L A100C could lead to higher structural stability and optimized catalytic rates in the hydroformylation of 1-octene. In this way, the activity of the artificial hydroformylase could be improved (five-fold better TON and a selectivity of 80 %) for linear aldehydes. [161]

Miscellaneous
As in section 2.8, here we report few examples of less-common biomolecular scaffolds bearing wide hydrophobic pockets suitable for the creation of new catalytic sites.
The heat shock protein from Methanococcus jannaschii (MjHSP) is a spherical hollow complex of 16.5 kDa and consists of 24 subunits with a large cavity of a diameter of 65 Å. The protein comes with a broad pH tolerance (4-11 range) and is heat stable up to 70°C. [162] Hilvert and co-workers managed to bind a bromoacetamide-ruthenium complex with a reactive cysteine (41C) within MjHSP cavity. The resulting biocatalyst was exploited in the ring-closing metathesis of N,N-diallyl-4toluenesulfonamide with turnover numbers of � 25. [163] A similar strategy was envisaged by DiStefano and collaborators, who transformed the adipocyte lipid binding protein (ALBP) into an artificial metalloenzyme via C117-mediated covalent anchoring of Cu(II) complexes into the host cavity. Using an iodoacetamido-1,10-phenanthroline cofactor, a hybrid catalyst able to perform hydrolysis of racemic amino acid esters with up to 86 % ee was successfully accessed. [164] In a follow-up study, the enantioselectivity was improved to 94 % by anchoring the complex at a different position (C72) in the scaffold. [165] Neocarzinostatin belongs to the enediyne chromoprotein family and consists of 133 residues. The protein is secreted from Streptomyces macromomyceticus during defense response and shows cytotoxic and antibiotic properties. Ricoux and coworkers equipped the neocarzinostatin (NCS) 3.24 variant with an iron(III)-tetrarylporphyrin-testosterone cofactor and exploited the obtained biocatalyst for the sulfoxidation of thioanisole, which resulted in low enantioselectivity (13 %). Molecular modeling studies revealed that, although the complex is well embedded into the protein cavity, the metal ion is exposed to solvent, leading to the low enantioselectivity. [166] Within the same research group, the expansion of NCS 3.24 reaction scope was pursued. By introducing a copper-phenanthroline-testosterone complex into the protein scaffold, the resulting biocatalyst was able to perform Diels-Alder cyclization of cyclopentadiene with 2-azachalcone. [167]

De novo Enzymes
The examples reported in the previous two sections demonstrate the significant progress in the development of artificial enzymes able to promote abiological transformations, primarily by functionalizing already existing enzymes or wide hydrophobic pores lacking an inherent biocatalytic activity. However, one of the long-standing objectives in biocatalysis is the de novo design of enzymes from scratch. The significant progress of computational protein design and prediction tools over the past decades has opened the possibility of designing novel protein backbones and integrating catalytic sites precisely into re-designed protein scaffolds. [3,22,23] The potential outcomes of this approach are clearly alluring and include the possibility of accessing the in silico design of enzymes capable of catalyzing specific new-to-nature transformations. In addition, the accessed de novo proteins could be used as a starting point for introducing an artificial moiety, to generate artificial enzymes with a further expanded reaction scope.
One of the most common approaches for the computational design of de novo enzymes is to integrate precisely arranged catalytic sites into 'matched' inert protein targets. Essential for this method is a comprehensive knowledge of the catalytic mechanism of the target reaction. On this basis, a preliminary active site, called 'theozyme', is predicted quantummechanically, to stabilize the transition state (TS). The theozyme must be docked into a protein backbone, which is further optimized by designing amino acid sequences able to improve the TS stabilization and its complementarity for the designed pocket ( Figure 9A). Finally, the top-ranking designs can be experimentally accessed and validated. [22,23,[168][169][170][171] The main drawback of this strategy is the low catalytic activity often displayed by the resulting enzyme. [170] This depends mainly by the lack of robust methods for predicting the stabilization of the TS and the folding of the final protein, as well as an unsatisfactory understanding of the structure/activity relationships and the difficulty of predicting solvent effects. [172] However, the ongoing improvements in design and prediction algorithms (such as the Rosetta software) are progressively improving the reliability of the de novo strategy. [173] Moreover, the performances of de novo enzymes could be boosted by direct evolution. [174] An alternative approach consists in integrating already existing enzymatic activities (usually a cofactor) in a simple motif, such as a 4-α-helix bundle designed from first principles (often called maquettes) or a TIM barrel, without homology for natural sequences. In this way, the properties of the de novo enzymes will be imprinted by the cofactor and the catalytic performances are often improved, without the need for direct evolution ( Figure 9B).
As already mentioned, this review focuses on the role of biomolecular scaffolds for the generation of artificial enzymes, through the strategies depicted in Chapters 2 and 3. Despite being aware of the importance of de novo enzymes in this field, we don't aim to give a comprehensive overview of examples, also considering the relevant number of reviews already published on this topic. [19,22,23,[175][176][177][178] Therefore, in section 4.1 only a few representative examples will be disclosed, dealing specifically with cases in which an artificial catalytic moiety has been introduced in a de novo protein.

Selected examples of de novo enzymes functionalized with artificial moieties
A representative case of the application of the 'theozyme' approach for the generation of an artificial enzyme was reported by Mills et al. in 2013. [179] Interestingly, this is one of the first examples in which computational methods were exploited to design de novo proteins incorporating non-canonical amino acids. In fact, in this work a new metalloprotein bearing the non-canonical amino acid BpyA as an artificial metal-coordinating unit was created. The initial designs were oriented by the idea of originating an active site able to promote the oxidative ring opening of catechol-like substrates. On the basis of the involved TS, possible theozymes were postulated, using the crystal structure of a Bpy-iron-3,6-di-tertbutylcatechol complex as a starting point. Once identified the key elements of the theozymes (namely the Bpy unit, dopamine as a catechol-containing substrate and additional metal-coordinating residues, like histidine or tyrosine), RosettaMatch was exploited to identify suitable protein backbones to allow the recapitulation of the required geometry. After that, RosettaDesign was used to insert additional interaction to stabilize the TS and the identified proteins were expressed, incorporating the BpyA through the stop-codon-suppression method. Crystal structure elucidation of the accessed proteins revealed an unexpected extrusion of the Bpy outside the active site. Therefore, a second-round design was performed, to increase the Bpy constrain in the octahedral geometry required to bind divalent cations. To this extent, Bpy was placed on an element of secondary structure, new coordinating glutamic and aspartic acid residues were added, and tyrosine (often favoring trivalent cations binding) was removed. On the basis of the new theozyme, the design process was repeated and this time a suitable candidate was identified, showing a good accordance with the predicted structure and the ability of binding several divalent cations, such as Co(II), Zn(II), Fe(II) and Ni(II). The accessed metal binding site could be a useful starting point for the design of Bpy-dependent enzymes.
Interestingly, another non-canonical amino acid, N-Mehistidine (N-MeÀ H), was embedded in a de novo enzyme (BH32), Figure 9. A) Schematic representation of the theozyme approach, in which a preliminary active site is created to stabilize the transition state; then, the theozyme is docked in a suitable protein backbone, bearing further stabilizing amino acids and the key artificial moiety; B) functionalization of a simple de novo protein with an artificial moiety, to generate de novo artificial enzymes. Figure created using Biorender.com. previously developed to promote Morita-Baylis-Hillman reactions, [180] to infer new catalytic properties as an ester hydrolase. [181] The rationale behind this approach was the knowledge that the activity of hydrolases is often dependent on nucleophilic residues such as histidine, serine or cysteine, and is often limited by the formation of stable acyl-enzymes intermediates. To overcome this pitfall, enzyme BH32, bearing a key H23, was used as a starting point for the remodeling process. Firstly, H23 was replaced with N-MeÀ H, to originate a more reactive acyl-imidazolinium intermediate. This modification led to an increase in the rate of fluorescein 2-phenylacetate hydrolysis. The catalytic activity was further boosted via direct evolution, leading to an optimized artificial enzyme, 9000-fold more efficient than free N-MeÀ H and 2800-fold more efficient than dimethylaminopyridine, a commonly used organocatalyst. Directed evolution was also exploited to construct artificial catalysts able to promote stereoselective hydrolysis of fluorescein derivatives.
The other common strategy to design de novo metalloproteins is based on the functionalization of simple de novo proteins with artificial moieties. For example, maquettes are simple, non-native, 4-helices bundles in which (artificial) cofactors can be embedded. In this way, several de novo redox proteins have originated, anchoring hemes, flavins, Zn-tetrapyrroles cofactors and iron-sulfur clusters, which infer the properties of the final protein, despite the maquette structural simplicity. [182][183][184] 4-helical bundle-metal complexes were also chosen as privileged de novo scaffolds for the development of Due-Ferri and related artificial enzymes by Lombardo et al. [24] This class of metalloproteins were designed as minimalistic versions of the active site of natural diiron enzymes, that are well known to allow a wide range of chemical transformations, in which the reduced di-Fe(II) binds O 2 , generating a di-Fe(III)-peroxi intermediate, that can react with several substrates. A first D 2 symmetrical 4-helical bundle was designed from first principles and then modified into a C2 homodimeric helix-loop-helix motif, containing a single glutamic acid residue in helix 1 and a Glu4-His2 motif in helix 2. These residues are responsible for the coordination of the metal centers and dictated the final geometry of the helical bundle, that is then inserted into the second coordination shell by hydrogen bonding. The folded protein conformation was finally stabilized adding well-packed hydrophobic side chains at the remaining core positions, the intrahelical loop and the side chains at the surface positions. Experimentally, the design sequence folded in a dimeric 4helices bundle, whose crystal structure matched indeed the predicted one. Furthermore, versions composed by a single chain or by four disconnected helices were generated. The accessed Due-Ferri proteins displayed a marked oxidase activity and were tested on the oxidation of 4-aminophenol and catechol and for the oxygenation of p-anisidine. Moreover, the same design approach, combined with Rosetta, was exploited to create a protein binding a synthetic porphyrin, as well as a 4helices bundle incorporating an unnatural tetranuclear metal cluster, composed by four Zn(II) and four carboxylate oxygens, arranged in a distorted cube shape.
De novo 4-helices bundles were the scaffold of choice also for the generation of artificial metallo-β-lactamases, as reported by Song and Tezcan. [185] In this work, cytochrome cb562 (a 4helices bundle without any structural or functional homology with hydrolytic enzymes) was self-assembled in the presence of zinc, originating a tetrameric de novo protein (Zn8:AB34), in which each bundle coordinates two Zn(II) ions, for a total of eight in the tetramer. Interestingly, only four of the Zn ions are catalytically active, while the other half plays only a structural role. Zn8:AB34 resulted to be efficient in promoting the hydrolysis of ampicillin and is functional in E. coli periplasm to enable the bacterium to survive the treatment with this antibiotic.
A different helix-turn-helix motif was envisaged by Hilvert and co-workers, to build Zn(II)-based hydrolases. [186] In detail, the design of this class of artificial enzymes was based on a homodimeric peptide (46 amino acids per monomer), bearing two interfacial sites, each one composed of a histidine triad coordinating a Zn(II) ion. The dimer displayed a certain ester hydrolytic activity, that was improved by connecting the adjacent N and C termini of the dimer with a short linker and de-symmetrizing the active site by removing some of the metal-coordinating residues. In this way, a new artificial enzyme (called MID1sc, 97 amino acids) able to coordinate only one Zn(II) was developed. Its catalytic activity towards the stereoselective hydrolysis of the racemic fluorogenic ester leading to (R)-or (S)-2-phenylpropionate and coumarin was boosted via several rounds of direct evolution. The optimized candidate (called MID1sc10) was 10000-fold more active than MID1sc and displayed a marked preference for the hydrolysis of the (S)precursor. Crystallization studies confirmed the helical-bundle fold of the protein, even if with a higher crossover angle of the two helix-turn-helix fragments than in canonical four-helices bundles, allowing for the binding of more hindered substrates. Notably, direct evolution was applied also to MID1sc, leading to a Zn(II)-mediated hetero-Diels-Alderase, able to promote the abiological cycloaddition of an azachalcone and 3-vinylindole, with high TON and > 99 % selectivity towards the endoadduct. [25] Pecoraro and co-workers obtained artificial metalloenzymes starting from α 3 D, a de novo single-stranded antiparallel threehelices bundle. This protein was firstly functionalized with three cysteine residues, to originate a coordinating environment for heavy metals, such as Hg(II), Cd(II) and Pb(II). [187] The same approach was also exploited to bind zinc and copper ions. In detail, α 3 D was equipped with a histidine triad, mimicking the active site of carbonic anhydrase (section 2.3). In this way, upon coordination with Zn(II), a novel artificial metalloenzyme was identified and proved to be able to catalyze CO 2 hydration, even if with lower efficiency than the natural enzyme. [188] Furthermore, four different binding sites, called H 3 (3 histidines), H 4 (4 histidines), H 3 D (3 histidines in the same plane + 1 aspartic acid) and H 2 DH (2 histidines and 1 aspartic acid in the same plane + an additional histidine), were introduced in an elongated version of α 3 D, named GRα 3 D. The four accessed constructs were able to coordinate copper ions and behaved like mimics of copper-based superoxide dismutases, although with a lower reaction rate than the natural enzyme. [189] The final example we decided to highlight in this section was reported by Zeymer and co-workers and deals with the generation of a novel artificial metalloprotein, originated by the fusion of two de novo proteins. [190] In detail, previously computationally designed TIM barrels, [191,192] lacking native activity, were identified as a suitable starting point for the generation of artificial metalloenzymes, considering their wide cavity. To build a specific binding site for large trivalent cations, such as lanthanides, a C2-symmetric de novo 'lid' was fused on the top of the C2-symmetric half-TIM barrel, aligning the symmetry axes, using RosettaRemodel. The tool was exploited also to design suitable sequences for the loops connecting the two domains. To create a metal-coordinating environment, different combinations of glutamic acid and histidine residues were arranged in a C4 symmetric fashion at the top of the TIM barrel. The most successful design resulted from the combination of a C2 symmetric ferrodoxin 'lid' with DeNovoTIM15 dimer and presented two glutamic acids at positions 31 and 154. After the two-fold symmetric assembly, a heat-and co-solvent-stable protein, characterized by four coordinating D-residues and displaying an internal cavity of 420 Å, was constructed.
To test the binding affinity of some lanthanides, a N6 W mutation was introduced, to equip the protein with a tryptophan acting as an 'antenna' in proximity of the active site, allowing the monitoring of the lanthanide luminescence. The artificial protein was mixed with lanthanide salts, and by the extent of the tryptophan-enhanced lanthanide(III) luminescence (λ ex = 280 nm) it was possible to quantify the binding. In particular, Tb(III), Eu(III) and Yb(III) displayed ultra-high affinity for the protein, with a binding constant of 10 À 18 M, while Ce(III) interaction was 10-fold weaker. Considering the possibility of easily engineering the protein cavity and the strong Lewisacidity of lanthanides, it is evident that this study paves the road towards the development of novel artificial metalloenzymes, exploitable for hydrolytic and CÀ C forming reactions.

Discussion and Perspectives
This review aims to give a general overview of the most common biomolecular scaffolds in the design of artificial enzymes, including examples from the functionalization of a natural enzyme with artificial moieties, the introduction of a new catalytic site in wide hydrophobic pockets of proteins lacking biocatalytic properties, and the de novo design of novel enzymes from scratch. Over the years, more and more biomolecular scaffolds could have been converted into artificial enzymes, but an important question of course is: what do we consider an excellent starting point for this design process? Clearly, the decision of choosing a protein scaffold is influenced by benefits and limitations of each strategy that need to be taken into account. For instance, when functionalizing a natural enzyme, a predefined active site is already available, and this provides a good starting point for the required modifications. Herein, enzymes containing cofactors which are replaced by similar artificial moieties showed to have high potential. However, the presence of a specific active site from the beginning of the design process limits the functionalities that can be introduced and the accessible reactivities. Starting from a protein scaffold without any predefined active site, the incorporation of various new functionalities at different positions could be easily attempted, due to the wider environment available in the hydrophobic pore. In this way, it is possible to access versatile artificial enzymes, able to promote numerous different reactivities, therefore overcoming the substrate-specificity typical of natural enzymes. On the other hand, finding such protein scaffolds and deciding where to introduce the active site requires a deeper insight into the structure of the scaffold. Starting with a binding pocket that's either too wide or small in size influences the incorporation of new functionalities and/or engineering of the protein cavity. Designing artificial enzymes via de novo design from scratch presents undoubtedly the advantage to develop artificial enzymes tailored ad hoc on the basis of the transition state of the reaction of interest. This approach is particularly useful when an enzyme for a specific reaction is required, while is rather contra-productive when searching for a versatile scaffold that can be applied for the incorporation of different functionalities and various reactions.
In the same field, the incorporation of cofactors in maquettes or TIM barrels designed from first principles, offers the advantage of inferring an already existing enzymatic activity in a structurally simple protein backbone. Even though the current computational methods are not yet potent enough to afford de novo artificial enzymes with performances comparable to natural counterparts, the development of new computational tools and software is progressively bridging this gap.
In summary, we believe that there is not a unique, optimal method for the preparation of artificial enzymes, but a combination of the previously described strategies can be successfully employed, depending on the final purpose of the artificial enzyme itself. As already mentioned, rational design and directed evolution emerged as fundamental tools to improve the performance of artificial biocatalysts. Moreover, we anticipate that the application of new approaches such as machine learning to enzyme engineering will open thrilling opportunities towards the identification of novel protein scaffolds for the development of new exciting biocatalysts. [193]