NADPH-dependent Secondary Amine Organocatalysis hosted by a Nucleotide-binding Domain

The concept of organocatalysis has been applied to facilitate “new-to-nature” reaction modes via articial enzyme design. However, it remains challenging to recruit structurally complex natural molecules as synthetic reagents. Here, we have reported a generic design strategy that allows generation of a NADPH-dependent hybrid catalyst whose action is orchestrated by a secondary amine; this system recruits a reaction mode not commonly seen among enzymes, whilst involving an intricate cofactor that cannot be used by existing organocatalysts. A secondary amine organocatalytic motif was incorporated into protein scaffolds as an unnatural amino acid by expansion of the genetic code. When introduced into the multidrug binding protein LmrR, a hybrid catalyst accepting α,β-unsaturated carbonyl substrates for transfer hydrogenation was established but was conned to the much-simplied biomimetic benzyl dihydronicotinamide (BNAH). Conversely, dihydrofolate reductase (DHFR) contains a nucleotide binding domain and can be converted into a hybrid catalyst that favourably uses NADPH for reaction, thus highlighting the importance of choosing an appropriate scaffold. The DHFR-hosted system tolerates a range of aldehyde substrates and can be coupled with an enzymatic NADPH regeneration scheme. The presented engineering approach can be readily extended to other protein scaffolds for use of different natural molecules in non-natural reaction modes.


Introduction
Organocatalysis has tremendous potentials for its applications in arti cial enzyme design [1][2][3][4] . By hosting small organocatalytic motifs proteins can catalyse reactions with "new-to-nature" mechanisms, and various methods have been reported to generate protein-based organocatalysts [5][6][7][8][9][10][11][12][13][14][15] . However, a generic approach that recruits natural molecules as reagents is yet to be established. NAD(P)H, ATP, SAM, peptides and nucleic acids are some of the structurally complex, natural molecules that can be serve as substrates. Their use in arti cial enzyme reactions will facilitate modi cations of biomolecules via nonnatural reaction mechanisms and development of combined catalytic cascades, both of which underpins critical chemical and synthetic biology applications. Given the signi cant potentials, we have reported a design strategy that facilitates the creation of a hybrid catalyst which recruits an intricate natural cofactor NADPH in a non-natural reaction mode.
To the best of our knowledge there is currently no organocatalyst including secondary amines capable of recruiting NADPH as a synthetic reagent, though more reactive biomimetics such as benzyl dihydronicotinamide (BNAH) and Hanzstch ester have been employed as surrogates [16][17][18][19] . Contrarily, there are many natural enzymes utilizing NADPH for reactions, but none of them uses a secondary amine for substrate activation. Old yellow enzymes (OYE), a group of commonly used biocatalysts, activate substrates by non-covalent interactions and subsequently use avin derivatives as a medium for hydride transfer 20,21 , whereas NADPH-dependent ene-reductases which utilise metal ions for substrate activation have also been reported 22 . Arti cial metalloenzymes that use reactive transition metals to transfer hydride from NAD(P)H have also been developed 23,24 . Nevertheless, an NADPH-dependent catalyst that activates substrate by a secondary amine is needed, because its availability will help ll in the gap between the reaction repertories of bio-and organocatalysis. For instance, carbonyl substrates can be selectively activated by secondary amines within protein scaffolds through iminium or enamine catalysis, which can be further segregated into single-or two-electrons transfer [25][26][27] . NADPH, being readily recyclable through well-established regeneration schemes [28][29][30] , can be applied in organocatalytic transformations including transfer hydrogenation, dehalogenation 31 and deacetoxylation 32 . Hence, an arti cial enzyme design strategy that facilitates NADPH-dependent secondary amine organocatalysis needs to be developed, as such a catalytic system does not exist in nature.
We envisioned that an NADPH-dependent organocatalytic system can be generated by incorporating a secondary amine motif into an appropriate nucleotide binding domain. Hence, in this work unnatural amino acids bearing a cyclic secondary amine were site-speci cally introduced into protein scaffolds by applying the technique of genetic code expansion 33 (Fig. 1). Both the multidrug-binding protein LmrR and dihydrofolate reductase (DHFR) were tested as templates and transformed into substrate-promiscuous hybrid catalysts for reduction of various α,β-unsaturated aldehydes. Whereas the LmrR-hosted organocatalytic system is con ned to BNAH as the hydride donor, the DHFR counterpart that contains a nucleotide binding domain favourably uses the natural cofactor NADPH for transfer hydrogenation.
Mechanistic analysis revealed that the DHFR scaffold facilitates non-rate-limiting delivery of the pro-R hydride to the iminium intermediate. To demonstrate the potential applications, the DHFR-hosted organocatalytic system was coupled to an enzymatic NADPH regeneration scheme.

Results
Genetic code expansion. Unnatural amino acids (Fig. 1a) containing a secondary amine were tested for incorporation and screened for organocatalytic activity. Amino acids 1, 2, and 3 contain D-proline, Lproline and L-thiazolidine-4-carboxylic acid (thioproline), respectively. Based on previous reports, 1 and 3 are substrates of the wild-type Methanosarcina bakeri pyrrolysyl-tRNA synthetase (MbPylRS) and its engineered variant ThzKRS, respectively 34,35 . Due to the structural similarity between 2 and 3, we envisaged that 2 could be a substrate of ThzKRS. Incorporation of unnatural amino acids was tested using the reporter protein sfGFP that bears a TAG codon for the 150th amino acid residue, and formation of the full-length sfGFP was con rmed by SDS-PAGE analysis (Fig. S1). Subsequently, unnatural amino acids were incorporated into the multidrug binding protein LmrR 36 . Neatly presented in previous arti cial enzyme design [37][38][39] , LmrR becomes catalytically competent after modifying its hydrophobic pocket, including the insertion of an unnatural amino acid. We therefore selected four previously reported residues, Val15, Asn19, Met89 and Phe93 (Fig. 1d), located in the hydrophobic pocket for substitution with 1, 2 or 3. SDS-PAGE analysis revealed successful recombinant production of the 12 LmrR variants in E. coli (Fig. S2). The dimeric nature of these variants was con rmed by size exclusion chromatography, with the exception of LmrR-Met89-2 whose oligomeric nature appears to be hampered by the modi cation (see SI, Pg S7-8). Furthermore, liquid chromatography-mass spectrometry (LC-MS) investigation indicated that variants containing amino acid 2 were unstable, and up to 35% hydrolysis of the L-prolyl group was observed (Fig. S4). This is likely due to the presence of proline iminopeptidase in E. coli which cleaves N-terminal proline in polypeptides. Such a truncation was particularly profound when 2 was at the position 89. In contrast, no detectable hydrolysis was observed in proteins carrying either Dproline 1 or L-thioproline 3.
Use of LmrR as the scaffold for transfer hydrogenation. The LmrR variants that possess a cyclic secondary amine were then tested for their ability to mediate transfer hydrogenation (Fig. 2a). The conversion of cinnamaldehyde 4 (1 eq.) to its reduced counterpart 5 served as the model reaction and tested with different hydride donors. BNAH was rst used, and the wild-type LmrR which contains seven lysine residues and a N-terminal amine yielded less than 3% conversion as detected by gas chromatography-mass spectrometry (GC-MS) analysis (Fig. 2b), indicating minimal catalytic activity.
Substitution of Met89 with 1, 2 or 3 showed marginal improvement (~ 5% conversion) 40 . Low conversion (< 15%) was observed in variants incorporated with 2, likely due to the hydrolysis of the isopeptide bond. Substitution of thioproline 3 at the position 19 and D-proline 1 at the position 15 resulted in a detectable amount of product conversion (> 20%). Notably, the replacement of Phe93 with 1 (LmrR-Phe93-1) was most promising giving 58% yield which is at least 19-fold higher than that of the wild-type enzyme. Kinetic analysis of LmrR-Phe93-1 displayed enzyme-like behaviour, showing a saturation curve between reaction rate constants and BNAH (Fig. 2c) with a catalytic e ciency k cat /K M of 0.86 M − 1 ·s − 1 .
When NADPH was tested as the hydride source for organocatalysis by LmrR-Phe93-1, no conversion was detected. Whilst disappointing, this observation aligns with the literatures. NADH was found to be a poor hydride donor for small organocatalytic secondary amines, and organic solvents was used for these reactions 16,41 . Whereas 1-3 are water soluble, transfer hydrogenation with NADPH by these secondary amines in aqueous buffer was not observed. This is also the case for the protein-hosted secondary amines presented here and previously 11 . It is likely that the dinucleotide hinders the dihydronicotinamide motif from approaching the intermediate for reactions. Hence, a protein scaffold that can place the reacting centres in proximity is desired.
Use of DHFR as the scaffold for transfer hydrogenation. Seeking an NAPDH-dependent organocatalytic system, we extended to examine a template that contains a domain naturally evolved for binding nucleotide-containing cofactors 42 . Escherichia coli dihydrofolate reductase (DHFR) is a well-investigated enzyme that catalyses the step of hydride transfer from the C4 pro-R hydride of NADPH to C6 of dihydrofolate with concomitant protonation at the N5 position in the one-carbon metabolism for thymidine biosynthesis 43 . DHFR contains a Rossmann fold, which is composed of a multi-stranded β sheet packed by two α helices on each side and binds nucleotide-containing cofactor NADPH with an association rate up to 10 5 M − 1 ·s − 1 44,45 . Additionally, it contains a mobile M20 loop (residue 9-23) which closes the active site upon binding NADPH generating a shielded environment surrounding that nicotinamide motif 46,47 . These features present DHFR as an ideal scaffold for transfer hydrogenation by NADPH through introduction of a cyclic secondary amine organocatalyst. To this end, Ala7, Phe31 and Ser49 that are in proximity to the nicotinamide motif of the NADPH binding site were individually replaced with 1 (Fig. 1e). The resulting DHFR variants were veri ed by SDS-PAGE (Fig. S3) and mass spectrometry (Fig. S5).
Activity tests revealed that DHFR variants incorporated with 1 could facilitate NADPH-dependent transfer hydrogenation (Fig. 2a). For the wild-type DHFR, the reaction product 5 was undetectable by GC-MS analysis. Replacement of Ser49 with 1 led to considerable protein precipitation with no detectable products. Contrarily, replacement of Ala7 with 1 could effectively catalyse the reaction as evident by signi cant product conversion (Fig. 2d). Though BNAH can be used as the hydride source, the extent of product conversion was noticeably higher when utilising NADPH. With 0.1 equivalent (10 mol%) of DHFR-Ala7-1, 90% of starting material 4 was converted into product 5 in aqueous buffer at pH 7.0 using 2 equivalents of NADPH after 18 hours (versus 20% conversion with BNAH under the same conditions). DHFR-Ala7-1 also demonstrated enzyme-like kinetic behaviour showing a saturation curve between the rate constants and cofactor concentration with catalytic e ciency (k cat /K M ) reaching 5.77 M − 1 ·s − 1 (Fig. 2e). It is noteworthy that, while DHFR-Ala7-1 uses the structurally complex NADPH for reaction, its catalytic e ciency (k cat /K M ) is ~ 5-fold higher than that of LmrR-Phe93-1 measured with BNAH.
Reaction modes used by LmrR-Phe93-1 and DHFR-Ala7-1. Similar to many other secondary amine organocatalysts 48 , formation of iminium ion is critical towards carbonyl substrate activation, and thus its transient existence during the catalysis by LmrR-Phe93-1 and DHFR-Ala7-1 was examined for mechanistic insights. The two variants were incubated with the starting material 4, treated with NaCNBH 3 and subjected to high resolution mass spectrometric analysis, as previously described (see SI) 3 . The LmrR-Phe93-1 sample predominantly yielded one protein species with an observed mass matching the calculated molecular weight of the reduced covalent protein-substrate intermediate (Fig. S6A). Similarly, a substantial portion of the DHFR-Ala7-1 variant also yielded the reduced covalent intermediate under the same condition, but hydrolysis of the prolyl-group was also observed upon treatment with NaCNBH 3 (Fig.   S6B). Treatment of these protein species with chymotrypsin revealed digested peptides, whose elemental composition identi ed by liquid chromatography tandem mass spectrometry (LC-MS/MS) within 1 ppm accuracy corresponds to the trapped intermediate ( Fig. S7 and S8). In contrast, the wild-type LmrR and DHFR that cannot catalyse the transfer hydrogenation did not show any evidence of the iminium intermediate formation. Accordingly, these results support a non-natural LUMO-lowering reaction mechanism 48 , in which the secondary amine activates the starting material through iminium ion formation for transfer hydrogenation from the corresponding hydride donor (Fig. 3).
Contrary to previous work which relied on a LmrR-hosted aniline for catalysis 3 , no lysine modi cation was detected in our protein scaffolds after NaCNBH 3 treatment. The chemoselectivity observed in this work implies the signi cant reactivity of the secondary amine, enabling iminium intermediate formation with relatively inert carbonyl substrate under aqueous conditions. Furthermore, in LmrR the aniline organocatalysis was found to be optimal at position 15 3 , whereas we obtained higher conversions with the secondary amine organocatalyst at position 93, indicating that activity test at different positions is essential during protein-based catalyst design.
Prochirality and kinetic insights of the transfer hydrogenation reaction catalysed by DHFR-Ala7-1. When [4R-2 H]-NADPH (NADPD) was used in the catalysis by DHFR-Ala7-1, only one product isotopologue with the deuteride located at the C β position was identi ed by GC-MS (Fig. S9). This observation agrees with previous crystal and biochemical analysis which illustrated that the wild-type DHFR catalyses the transfer of the C4 pro-R hydride of NADPH 49,50 . Furthermore, this protein-hosted organocatalyst is regioselective favouring 1,4-over 1,2-addition, the latter was observed in the streptavidin-hosted secondary amine organocatalytic system as a side reaction 10 . Given its stereoselectivity, primary kinetic isotope effect (KIE) of hydride transfer was assessed. By comparing the turnover rate constants under saturating conditions between the reactions that use NADPH and NADPD, a kinetic isotope effect (KIE) of 1.10 ± 0.16 (Fig. S10) was obtained, implying that the step of hydride transfer is not rate-limiting. Other chemical step(s) such as iminium ion formation can be rate-limiting in this multi-step catalytic cycle 51 . Alternatively, the physical step of releasing the oxidised cofactor from the enzyme can dictate the rate of catalytic turnover, as observed in the natural reaction catalysed by the wild-type enzyme 44,47 .
Substrate Scope. With the active systems in hand, substrate promiscuity of LmrR-Phe93-1 and DHFR-Ala7-1 were investigated. Aromatic and ketone analogues of cinnamaldehyde were evaluated as alternative substrates. Both systems were able to accept all the aldehydes tested for the selective reduction. With respect to per mol% of catalyst, the conversions of aldehydes to the corresponding reduced products by DHFR-Ala7-1 was generally higher than those by LmrR-Phe93-1 (Fig. 4, Table S6 and S7). On the other hand, the ketone substrate showed no turnover in either system. Hence, two proteinbased organocatalytic systems, with one preferentially using BNAH and the other NADPH, that have substrate promiscuity for aldehydes have been developed.
Coupling of the NADPH-dependent organocatalysis with an enzymatic regeneration scheme. Although it is structurally complex, NADPH is commonly used in biocatalysis for chemical manufacturing through coupling of cofactor regeneration schemes which excel at their speci city, e ciency and ease of handling 30 . Hence, we coupled the organocatalysis by DHFR-Ala7-1 with an enzymatic regeneration scheme enabling sub-stoichiometric use of NADPH (Fig. 5a). Importantly, this experiment proves the concept that secondary amine organocatalysis, amongst other chemical catalytic systems 52 , can be driven by a well-established biocatalysis tool. The conversion of starting material 4 to product 5 by DHFR-Ala7-1 was performed in the presence of the reaction by glucose 6-phosphate dehydrogenase (G6PDH), a commercially available enzyme that oxidises glucose 6-phosphate whilst reducing NADP + to regenerate NADPH 29 . The coupling between the organocatalytic and enzymatic reactions was assessed by the product conversion and the total turnover number, which is de ned as the number of moles of product formed per mole of cofactor used (Table S8) 29 . The product conversion could reach up to 90% when the coupling reaction was included, whereas under the same reaction conditions the conversion was barely detectable (< 1%) in the absence of G6PDH (Fig. 5b). Furthermore, the total turnover number was measured up to 10460 reaching the detection limit of our experimental approach (Fig. 5c). A value of this magnitude is arguably economically viable on an industrial scale 28,30,53 , highlighting the signi cant potential of the presented NADPH-dependent organocatalytic system for synthetic applications. Such work complements with the recent progress made in using alternative hydride donors in enzymatic reactions 54 , and together they will eventually enhance the merger of chemical and biocatalysis.

Conclusion
A NADPH-dependent secondary amine organocatalytic system enabled by use of a nucleotide binding domain has been established. Notably, NADPH can be now used as a synthetic reagent by an organocatalytic motif, whereas aldehyde substrates activated by a secondary amine for transfer hydrogenation is a "new-to-nature" enzyme reaction mode. Though the same substrate activation mode was used between the LmrR and DHFR variants, only the latter can use the natural cofactor for reactions, underlining the importance of the protein scaffold and the design strategy. Furthermore, this work complements with previous reports, where p-aminophenylalanine and N-methyl histidine were introduced into protein scaffolds for organocatalytic transformations 3,4,55,56 . In both of these cases a protein scaffold was chosen and then optimised for reactions involving relatively small reagents (< 200 Da), while in this work a sizeable structurally complex natural metabolite was used and laboratory evolution has yet to be applied. Lastly, our protein engineering strategy is most likely transferrable, facilitating the development of non-natural reaction modes that recruit natural metabolites as reagents.

Methods
All methods reagents and characterisation can be found in the supplementary material. T.L.W. performed the experimental work. All authors designed and planned the work. Initial draft was written by L.Y.P.L. All authors wrote the nal article. All authors have given approval to the nal version of the manuscript.     Proposed catalytic cycle for the organocatalytic transfer hydrogenation reaction by DHFR-Ala7-1. The unnatural amino acid forms an iminium ion with the α,β-unsaturated carbonyl substrate, and hydride transfer occurs from the pro-R position of NADPH to Cβ of the iminium intermediate. Substrate scope analysis of the organocatalytic proteins LmrR-Phe93-1 and DHFR-Ala7-1. The LmrR or DHFR variant was incubated with the indicated α,β-unsaturated carbonyl compound and 5 eq of their respective hydride donors (BNAH and NADPH, respectively). Conversion (%) per mol% was determined by comparison of the integrals of the protons arising from the aldehyde protons of the saturated substrate to the unsaturated product as revealed by 1H NMR spectroscopy (see SI). For the ketone, no change in the substrate could be observed. Each reaction was performed in triplicate and the mean yield (± standard deviation) reported.