Nucleophilic α-Functionalization of Benzyl Amines Using an Engineered Threonine Aldolase

Abstract

Chiral amines are ubiquitous in pharmaceuticals and agrochemicals, making their efficient and selective synthesis a significant synthetic challenge. Threonine aldolases synthesize chiral amines via stereoselective C–C bond formation, however, they are restricted to small amino acids as pro-nucleophiles, limiting their utility in chemical synthesis. Here, we report an engineered threonine aldolase capable of α-functionalizing benzylamines. The evolved enzyme has excellent catalytic efficiency and accepts a broad range of (heterocyclic)benzyl amines and structurally diverse aldehydes to yield single-enantiomers of 1,2-amino alcohols in high-yield and diastereoselectivity. Mechanistic and crystallographic studies provide a rationale for how these mutations enable this previously unknown function. Moreover, beneficial mutations can be transferred to a related pyridoxal-dependent protein, highlighting the generality of these insights.

Graphical Abstract

Chiral amines are found in nearly 40% of pharmaceuticals. Consequently, chemical methods that enable their rapid and selective synthesis are valuable (Figure 1A).¹ The most well-developed reactions for their preparation are asymmetric reductions which occur with excellent enantioselectivity but rarely increase the molecular complexity.^2-6 Reactions that prepare chiral amines via convergent C–C bond formation have the potential to accelerate the synthesis of bioactive molecules while providing access to untapped areas of chemical space.^7-10 However, current synthetic methods that use this approach often require cumbersome protection/deprotection steps, have limited substrate tolerance, and provide varying levels of enantioselectivity.^11-13

Figure 1. Functionalization of amines.

(A) Existing biocatalytic strategies for preparing 1,2-amino alcohols via redox construction by comparison to the skeletal construction strategy reported in this work. (B) Pronucleophile limitations to threonine aldolase. (C) Mechanism of pyridoxal-5-phosphate catalysis in threonine aldolases and limitations of existing threonine aldolases.

Enzymes are attractive catalysts for chiral amine synthesis because they can be engineered to provide unparalleled levels of enantioselectivity and offer retrosynthetic strategies distinct from small molecule catalysts.^14,15 Transaminases, amine dehydrogenases, and imine reductases catalyze formal reductive aminations and have been successfully applied to synthesizing pharmaceuticals.^16-22 Alternatively, amine lyases have been developed to couple amines with Michael acceptors to prepare α- and β-amino acids. While these strategies are highly enantioselective, they require the construction of skeletal C–C bonds prior to C–N bond formation, limiting the range of structures that can be prepared from commercial chemicals (Figure 1B).^23-27 Reactions that prepare chiral amines via intermolecular C–C bond formation can accelerate chemical synthesis by setting the amine stereocenter in a complexity-building reaction.

Pyridoxal-5-phosphate (PLP)-dependent threonine aldolases can synthesize enantiopure non-canonical amino acids via C–C bond-formation.^28-34 These enzymes provide nearly perfect control over the amine stereocenter but are reported to only accept glycine, alanine, and serine as pronucleophiles (Figure 1C).³⁵ While this limitation can be overcome using an elegant biomimetic PLP catalyst, it comes at the expense of catalytic efficiency and generality.^36,37 We envisioned engineering threonine aldolases to accept non-amino acids as pronucleophiles to access enantiopure 1,2-amino alcohols in a convergent C–C bond-forming reaction (Figure 1B).^38-47 This method would provide a means of synthesizing a wide variety of 1,2- amino alcohols from commercially available amines and aldehydes while avoiding the need for stoichiometric activating/protecting groups.

Threonine aldolases and transaldolases use the PLP cofactor to catalyze the aldol/retro-aldol reaction.^32,48 PLP reacts with amino acids to generate an aldimine, which serves as an electron sink to stabilize the negative charge at the α-amino position. This negative charge stabilization increases the acidity of the α-amino proton of glycine by approximately 18 orders of magnitude (Figure 1C).^49-51 Recognizing the importance of the carboxylic acid motif in helping to acidify the α-amino protons, we sought amines with functional groups that would have a similar acidifying effect. We identified 2-methyl pyridine as a promising target because its benzylic protons have a similar pKa to the α-protons of ethyl acetate (pKa (THF) of 2-methylpyridine ~32-34, pKa (DMSO) of ethyl acetate = 29.5).^51,52 Moreover, as heteroaromatics are ubiquitous in bioactive compounds, threonine aldolase variants engineered to accept these amines would provide access to useful products.

We began our studies with 2-(aminomethyl)pyridine (1) as the model amine donor and 3-phenylpropylaldehyde (2) as the acceptor. We evaluated a small library of LTAs and found that several gave the desired product in 13 to 20% yield with high enzyme loadings and selected the threonine aldolase from Thermatoga maritima (TmLTA) as the parent for directed evolution because it formed the product with >19:1 diastereoselectivity (Table S1).⁵³ Based on the crystallographic structure information from the PDB database (PDB: 1LW5), we targeted several residues lining the active site for site saturation mutagenesis (N308, R316, L306, R140, Y30, H125, N123, and R171). We found that mutation of asparagine 308 to glutamic acid (N308E) increased the yield from 9 to 18% at 0.1 mol % loading, without erosion of the enantioselectivity and diastereoselectivity (>99:1 er and >99:1 dr). The next round targeted residues on the loop over the quinonoid intermediate (W86, Y87, R122, A309, and V310). Mutating tyrosine 87 to alanine (Y87A) further increased the yield to 45% yield (>99:1 er and >99:1 dr). Mutating arginine 122 to glycine (TmLTA-N308E-Y87A-R122G) increased the yield to 76%. Finally, mutating the neighboring proline (P121) to threonine or aspartic acid (P121T or P121D) increased the yield to 79% (P121T) and 83% (P121D), with >99:1 er, and >99:1 dr in both cases. With the two optimal variants in hand (TmLTA^EAGT and TmLTA^EAGD), we selected TmLTA^EAGT and tested its performance when used as cell-free lysate (CFL) and found that enzyme loading can be decreased to 13 wt% with no change in the enantioselectivity and diastereoselectivity (Figure 2C and Table S11). The final variant maintains its ability to accept glycine as a pronucleophile, indicating that carboxylate binding motif is still present. Subjection of the amino alcohol product to the enzyme results in minimal retro-aldol reactivity, potentially accounting for the higher diastereoselectivities than what is typically observed with threonine aldolases.

Figure 2. Reaction development and protein engineering.

(A) Model reaction and evolutionary campaign. Reaction condition: 2-(aminomethyl)pyridine (0.025 mmol, 5 equiv., 62.5 mM), aldehyde (0.005 mmol, 1 equiv., 12.5 mM), 0.1 mol% TmLTA, CHES (400 uL), rt. Yield and dr determined via LC-MS relative to an internal standard (mandelic acid). Enantiomeric ratio (er) was determined by LC-MS after derivatization with D/L-Marfey’s reagents.⁵⁸ (B) Targeted active site residues. (C) Cell-free lysate reactivity of the final mutant TmLTA^EAGT.

With the optimal variants, we explored the scope and limitations of various heterocyclic amines. Pyridines with Cl (4, 7, 9, 10), F (6 and 11), and Me (5 and 8) substituents at 3’-6’ positions were well-compatible and gave 21-69% yield and near-perfect stereoselectivities (>99:1 er and >99:1 dr). 3-Cl and F substituent exhibited decreased reactivity and showed modest reactivity when inversing the limiting reagent to amine, likely due to the dual electronic and steric effect from F and Cl. This variant will also accept glycinamide 12, an unreactive amine with the wild-type TmLTA, while amide has a higher pKa of 35 (DMSO)⁵¹ compared to carboxylate (pKa 29.5), indicating that our engineering campaign opted for mutants that are active toward inert substrate. To our surprise, the final mutant TmLTA^EAGD retained the native activity (Figure S3), indicating that protein engineering turns this enzyme from a specialist into a generalist and broadens its synthetic utility.

Pleasingly, we found that various heterocycles (13-17) displayed different reactivities. Pyrimidine (13) gave a 60% yield and perfect selectivity at 0.1 mol% enzyme loading. We found that the first-round variant TmLTA-N308E gave a doubled yield of pyrimidine (14) without erosion of the stereochemistry. Thioazole (15), imidazole (16), and benzimidazole (17) were both compatible with our condition and exhibited excellent stereoselectivities.

The evolved variant is tolerant of various aldehydes. Formaldehyde (18) provided a 30% yield and nearly perfect enantioselectivity. Racemic substrates bearing α-stereocenters (20 and 26) were reactive and afforded products with low levels of diastereoselectivity (2.3:1 and 2.2:1), suggesting a modest kinetic resolution of the aldehyde. Notably, medicinally interesting azetidine and aza- spiro motifs were accepted (46% isolated yield of 21 and 48% yield of 22). 2-hydroxytetrahy-drofuran(23) was successfully converted to the 1,4-diol, while α-phenyl substituted substrates (24 and 25) gave lower yields. Interestingly, this enzyme can also accept an unprotected nucleotide analog (27), providing the product in 26% yield. This product type is challenging to access using other synthetic methods, highlighting the unique chemical structures available using this method. The 1,2-amino alcohol products can be easily deoxygenated, converted to the enantiopure aziridine, or cyclized to the oxazolidine-2-one using standard synthetic sequences (Figure S5). Aldehydes are prone to autooxidation when stored under air. We found that alcohols can be oxidized in situ to the corresponding aldehyde using FAD-dependent Choline Oxidase (AcCO₆), the corresponding one-pot/one-step cascade produces product in 65% yield with 15:1 dr and >99:1 er (Figure S6).⁵⁴

UV-vis experiments were conducted to elucidate the details of the reaction mechanism. The quinonoid is a persistent intermediate when using amino acids as pronucleophiles, ensuring that C–C bond formation occurs once the aldehyde is bound. We hypothesized that the evolved variant's improved performance was partly due to the increased concentration of the heterocyclic quinonoid species. When 2-aminomethyl pyridine 1 added to wild-type TmLTA, we observed a weak absorption (λ_max = 567 nm) that we attribute to the quinonoid (Figure 4A). This feature is significantly red-shifted compared to the quinonoid derived from glycine or alanine (λ _max = 497 nm), presumably because of the increased conjugation of the aminomethylpyridine quinonoid. When the same experiment was conducted with TmLTA-N308E, the intensity of this feature increased by 10-fold. This trend continues across the evolutionary series with TmLTA-N308E-Y87A-R122G, affording a 19-fold increase in intensity compared to the wild-type protein. The increase in intensity supports our hypothesis that the mutations increase the concentration of the nucleophilic quinonoid in the reaction. The intensity of the quinonoid decreases over one hour, resulting in a new absorption feature at 330 nm, which we attribute to pyridoxamine formation and 2-pyridine carboxaldehyde, indicating that transamination occurs slowly in the absence of the aldehyde coupling partner (Figure S2).

Figure 4. Applications and mechanism.

(A) Determining quinonoid formation across the evolutionary series using UV-Vis studies: 1 (25 mM), variant (0.05 mM) in CHES (100 mM, pH 8.9). Orange: wild-type. Blue: TmLTA-N308E. Grey: TmLTA-N308E-Y87A. Red: TmLTA-N308E-Y87A-R122G. Green: TmLTA-N308E-Y87A-R122G-P121T. (B) Comparing protein crystal structures of the glycine and 2-aminomethyl pyridine quinonoid in TmLTA (PDB: 1LW5) and TmLTAEAGD (PDB: 9E97), respectively. For omit map, see Figure S8. (C) Expanding the substrate tolerance with R316X mutants (D) Transferring beneficial mutations to UstD^2.0.

Next, we conducted protein crystallography experiments to determine how the mutations introduced throughout the engineering campaign altered the protein structure. We grew crystals of TmLTA^EAGD and soaked them with 2-aminomethyl pyridine 1, which resulted in a color change from yellow to magenta, and subjected the resulting crystals to X-ray diffraction. The resulting structure (PDB: 9E97) was similar to the parent (PDB: 1LW5) with an RMSD of 0.283Å with the PLP cofactor located in the exact location and orientation in the protein active site. When examining the location of side chains, we noticed that arginine 316 (R316) has moved toward N308E in the structure by nearly 1.6 angstroms because of the introduction of a hydrogen bonding contact. This change moves the side chains in the active site, making room for the more sterically demanding pyridine. A similar hydrogen bonding contact is observed in the crystal structure of the internal aldimine (PDB: 9E9J)(Figure S7). Interestingly, the pyridine does not seem to form additional hydrogen bonding contacts within the active site. Additionally, the Y87A mutation increases the size of the active site, allowing for better accessibility for the more sterically demanding amine (Figure 4B).

Based on the hypothesis that movement of the R316 side chain is essential for activity on non-amino acids pronucle at this position would enable reactivity with benzylamine, an unreactive substrate with the wild-type TmLTA and TmLTA^EAGD. We were pleased that TmLTA^EAGD-R316A afforded product 28 in 6% yield, while TmLTA^EAGD-R316W formed product in 15% yield. We hypothesize that the improved performance with tryptophan is due to the π-π interaction between arene and indole. This variant is effective for other electron-poor benzylamines (29 and 30) and unlocks a new family of amines that can used by threonine aldolases (Figure 4C).

Finally, we were interested in determining whether the lessons learned from this engineering campaign could be used to expand the scope of other PLP-dependent enzymes. In examining the protein engineering campaign, the first mutation (N308E) plays a vital role in increasing the activity with heteroaromatic amines as indicated by our UV-Vis study (Figure 4A). We hypothesize that incorporation of this mutation into other fold type 1 PLP-dependent enzymes would enable activity with heterocyclic amines. We selected UstD,^55,56 a PLP-dependent enzyme with less than 15% sequence identity as TmLTA, which was initially reported to decarboxylate the side chain of L-aspartate in a cyclic peptide via a nucleophilic enamine intermediate. Buller and coworkers engineered a more robust version of this protein to catalyze a decarboxylative aldol reaction with a broad array of aldehydes to produceγ-hydroxyl amino acids (UstD^2.0). We first examined the UstD^2.0 reactivity 2-aminomethyl pyridine 1 for an aldol reaction and observed no product formation, making it an ideal candidate to test our hypothesis. After conducting structure-based protein sequence alignment using MUSTANG,⁵⁷ an online web server, we identified T388 as the homologous position to N308. The ophiles, we examined the impact of mutations at this position for other types of amines. We were curious whether mutations introduction of the glutamic acid mutation T388E resulted in a variant that afforded the aldol product 3 a 15% yield, with >99:1 er and 99:1 dr. The addition of the analogous Y87A mutation at position M393 in UstD^2.0 (T388E-M393A) formed the product in 12% yield with no change in the enantio- and diastereoselectivity (Figure 4D). This result highlights the opportunity to use insights from the threonine aldolase engineering campaign to rapidly prepare variants of other fold type I PLP-dependent enzymes with initial activity on non-amino acid substrates.

In conclusion, we demonstrated that threonine aldolase can accommodate versatile non-native amines as its substrates to access valuable heterocyclic amino alcohols with excellent enantioselectivity and diastereoselectivity. Variants of TmLTA were identified that enable these proteins to accept a wide range of benzylic amines, providing rapid access to a diverse array of 1,2-amino alcohols. Preliminary studies indicate that the beneficial mutations can be transferred to structurally related enzymes, suggesting that a similar approach can be used to expand the synthetic utility of other PLP-dependent enzymes.

Supplementary Material

Experimental procedures, characterization data, NMR spectra, and HPLC traces. This material is available free of charge via the Internet at http://pubs.acs.org.

Figure 3. Reaction scope.

Amine (0.025 mmol, 5 equiv., 62.5 mM), aldehyde (0.005 mmol, 1 equiv., 12.5 mM), 0.1-0.3 mol% TmLTA^EAGD, CHES (400 uL), rt. Unless otherwise mentioned, the reaction was carried out with 0.1 mol% TmLTA^EAGD at analytic scale. 10 and 11: amine (1 equiv.), aldehyde (2 equiv.), 0.3 mol% TmLTA^EAGD, CHES (400 uL), rt. 12: amine (1 equiv.), aldehyde (2 equiv.), 0.5 mol% TmLTA^EAGD, CHES (400 uL), rt. 14: amine (1 equiv.), aldehyde (2 equiv.), 1 mol% N308E mutant. 15: amine (1 equiv.), aldehyde (2 equiv.), 0.2 mol% TmLTA^EAGD. 16: 1 mol% enzyme in KPi (pH 8, 100 mM), isolated yield at 0.5 mmol scale. 18, 20, 24-26: Amine (1 equiv.), aldehyde (2 equiv.), 0.3 mol% TmLTA^EAGD; 19, 21-23, and 27: Amine (5 equiv.), aldehyde (1 equiv.), 0.1-1 mol% TmLTA^EAGD. Specifically, 13, 21, 22, and 27 (0.1 mol%, 0.1 mol%, 0.1 mol% and 1 mol% enzyme), 19 and 23 (0.3 mol% and 0.5 mol% enzyme) were isolated with TmLTA^EAGT cell-free lysate at 1 mmol scale. For analytic scale, yields were determined via LC-MS relative to an internal standard (mandelic acid). For preparative scale, isolated yields were reported. The dr of isolated products were also confirmed by NMR. The dr and er of all the compounds were determined by LC-MS after derivatization with D/L-Marfey’s reagents. ^aAnalytical yield. ^bIsolated yield. All the isolated products were purified by reverse-phase HPLC and isolated as ammonium formate salts.