Biosynthesis of Unnatural Cyclodipeptides through Genetic Code Expansion and Cyclodipeptide Synthase Evolution

Abstract

Genetic code expansion (GCE) technology has primarily been devoted to the introduction of noncanonical amino acids (ncAAs) into ribosomally synthesized proteins or peptides. Its potential for modifying non-ribosomal natural products remains unexplored. In this study, we introduce a novel strategy that integrates GCE with the directed evolution of cyclodipeptide synthase (CDPS) to engineer a new class of CDPSs capable of biosynthesizing cyclodipeptides containing ncAAs. Using this approach, we achieve the efficient incorporation of 4-azido-L-phenylalanine (AzF) into AlbC-derived cyclodipeptides, generating a diverse array of new-to-nature cyclodipeptides. Molecular dynamics (MD) simulations and binding free energy calculations provide insights into the potential catalytic mechanisms responsible for the enhanced recognition of ncAA-tRNAs by the engineered CDPS variants. Furthermore, we expand the substrate scope to additional ncAAs and extend this approach to other CDPSs, enabling the creation of an even broader range of novel compounds. Together, these findings reveal the significant potential of GCE technology to precisely engineer biomolecules beyond proteins, unlocking new structural and functional diversity.

Graphical Abstract

INTRODUCTION

Although proteins in nearly all organisms are built from 20 canonical amino acids that support essential biological functions, their side-chain diversity is limited, constraining the range of chemistries available for engineering new protein functions. The emergence and development of genetic code expansion (GCE) technology have made it possible to incorporate noncanonical amino acids (ncAAs) into proteins, endowing them with novel structures and functions.^1–6 The ncAA of interest is incorporated into proteins at a reassigned nonsense or frameshift codon by an orthogonal aminoacyl-tRNA synthetase (aaRS)/tRNA pair that functions independently of the host’s endogenous translational machinery. Using this strategy, novel ncAAs have been precisely incorporated into proteins in both prokaryotic and eukaryotic cells.^7–10 However, the current application of GCE has been limited to the generation of novel ribosomally synthesized polypeptide chains,^11–13 and its potential for biosynthesizing non-ribosomal biomolecules with unique structures and functions remains largely unexplored.

In nature, 2,5-diketopiperazines (2,5-DKPs) represent a broad family of natural products with diverse biological and pharmacological activities.¹⁴ The two peptide bonds within 2,5-DKPs are formed by nonribosomal peptide synthetases (NRPSs)¹⁵ and cyclodipeptide synthases (CDPSs).¹⁶ CDPSs, a unique class of enzymes, have the remarkable ability to utilize aminoacyl-tRNAs (aa-tRNAs) synthesized by aaRSs to generate 2,5-DKPs with diverse structures. One of the most striking features of CDPSs is their structural similarity to the catalytic domains of class I aaRSs, such as tyrosyl-tRNA synthetase (TyrRS) and tryptophanyl-tRNA synthetase (TrpRS), suggesting they evolved from a class I aaRS-like ancestor.^17,18 In 2018, 2,5-DKPs with limited noncanonical groups were biosynthesized by hijacking the poly-specificity of native E. coli aaRSs in auxotrophic strains expressing CDPSs.¹⁹ However, the structural constraints of native aaRS binding sites limit their substrate scope, and their intrinsic editing and proofreading functions further restrict their activity. To address these challenges, several promiscuous E. coli aaRS mutants have been developed to broaden the substrate range for recognizing ncAAs.²⁰ Among them, PheRS-A294G,²¹ which can accept multiple para-substituted phenylalanine and tyrosine analogs, was reported to enhance the diversity of 2,5-DKPs incorporating ncAAs.²² Therefore, the introduction of additional aaRS mutants may further expand the available substrate repertoire.

Here, we report the generation of unnatural 2,5-DKPs containing ncAAs with diverse structures by integrating an evolved E. coli aminoacyl-tRNA synthetase (EcRS)/tRNA system from GCE technology with novel CDPSs obtained through directed evolution (Figure 1). This engineered system exhibits high recognition for ncAA-charged tRNAs (ncAA-tRNAs) and produces unnatural 2,5-DKPs at significantly higher yields than the natural products synthesized by wild-type CDPSs. This work highlights the potential of combining GCE technology with directed evolution as a powerful strategy for synthesizing novel, non-ribosomally synthesized biomolecules.

Figure 1. Schematic overview of the CDPS evolution.

Iterative rounds of directed evolution of CDPSs in E. coli through the combination of GCE, which used an engineered aaRS/tRNA pair to incorporate ncAAs.

RESULTS AND DISCUSSION

Activity and Specificity of Evolved EcTyrRS/tRNA from GCE.

The evaluation of 2,5-DKPs production catalyzed by various CDPSs is typically performed using E. coli as a host organism.^23,24 This suggests that most CDPSs are compatible with E. coli aa-tRNAs. To ensure that the generated ncAA-tRNAs could be recognized by CDPSs, we focused on an evolved E. coli tyrosyl-tRNA synthetase (EcTyrRS) developed in Saccharomyces cerevisiae,²⁵ which exhibits poly-specific activity toward diverse phenylalanine and tyrosine analogs.²⁶ Before expressing the evolved EcTyrRS and CDPSs in E. coli to produce ncAA-containing 2,5-DKPs via the ribosome-independent mechanism unique to CDPSs,²⁷ we first need to test the aminoacylation efficiency of various ncAAs by measuring GFP fluorescence intensity in eukaryotic cells.

As a first step, we evaluated the aminoacylation efficiency of different ncAAs. Thus, we can select functional ncAAs with good aminoacylation efficiency as starting points for biosynthesizing new CDPS products. We did this by measuring GFP fluorescence intensity in eukaryotic (HEK293T) cells (Figure 2a),^26,28 including 4-borono-L-phenylalanine (BorF, 1), 4-acetyl-L-phenylalanine (AcF, 2), O-methyl-L-tyrosine (OMeY, 3), 4-azidomethyl-L-phenylalanine (AzMeF, 4), 4-azido-L-phenylalanine (AzF, 5), 4-iodo-L-phenylalanine (IodF, 6), 4-nitro-L-phenylalanine (NitroF, 7), 4-cyano-L-phenylalanine (CyF, 8), 4-bromo-L-phenylalanine (BroF, 9), O-allyl-L-tyrosine methyl-ester (AllY, 10), 4-fluoro-L-phenylalanine methyl-ester (FluF, 11), 4-trifluoromethyl-L-phenylalanine methyl-ester (TrifF, 12), and O-propinyl-L-tyrosine methyl-ester (PropY, 13). To access the activities and specificities of this bioorthogonal aaRS/tRNA pair, we used pAcBac2.tR4-EcTyrRS/tRNA-EGFP-Y39TAG plasmid (Figure 2b), which encodes the poly-specific EcTyrRS/tRNA_CUA^Tyr pair as well as an EGFP gene harboring an amber permissive site (Tyr39).²⁹ We transfected pAcBac2.tR4-EcTyrRS/tRNA-EGFP-Y39TAG into HEK293T cells to monitor the expression of EGFP in the presence of 1 mM ncAA. The addition of these ncAAs resulted in the expression of full-length EGFP, whereas control experiments conducted in the absence of ncAAs demonstrated minimal GFP expression (Figure 2c and S1). These results demonstrate that the EcTyrRS/tRNA pair evolved in eukaryotes can specifically recognize a broad spectrum of ncAAs with distinct structures and functions.

Figure 2. Screening the efficiency of the EcTyrRS/tRNA pair toward 13 ncAAs in HEK293T cells and directed evolution of AlbC.

(a) Chemical structures of 13 ncAAs used in this study. (b) Plasmid design for the ncAA incorporation assay. (c) Fluorescence results of the ncAA incorporation assay in HEK293T cells. Data are presented as mean ± standard deviation (n = 3 independent samples). a.u. = arbitrary units. (d) Pipeline of the directed evolution strategy for AlbC in E. coli to enhance the efficiency and specificity for producing AzF-containing cyclodipeptide.

Synthesis of 2,5-DKPs with Evolved EcTyrRS/tRNA and Wild Type AlbC.

To explore if the ncAA-tRNAs generated using the evolved EcTyrRS/tRNA_CUA^Tyr pair can be used to biosynthesize 2,5-DKPs, we used the first-reported CDPS, AlbC, which is known for the production of cyclo-(L-Phe–L-Leu) (cFL, 14, Figure 2d).²⁷ To incorporate ncAAs and produce the corresponding cyclodipeptides in E. coli, we constructed two plasmids, pUltra-EcTyrRS/tRNA and pET22b-T5-AlbC, which together encode the evolved EcTyrRS/tRNA_CUA^Tyr pair as well as wild type AlbC (AlbC-WT). These plasmids were transformed into E. coli BL21 (DE3) cells. In the absence of ncAAs, liquid chromatography-mass spectrometry (LC-MS) analysis revealed four major peaks: cFL (14), cyclo-(L-Phe–L-Phe) (cFF, 15), cyclo-(L-Phe–L-Tyr) (cFY, 16), and cyclo-(L-Phe–L-Met) (cFM, 17), with retention times and mass spectrometry (MS) profiles identical to those of chemically synthesized standards (Figure S2). Next, we explored the use of AzF (5) as a 2,5-DKP substrate, which features a “clickable” azido group. In the presence of 5, we successfully identified a new compound with m/z of 336 [M + H]⁺, which corresponds to cyclo-(L-AzF–L-Phe) (cAzFF, 19, Figure 2d) through a comparison with a synthesized standard (Figure S3). The yield of cAzFF was relatively low compared to the native products synthesized by AlbC-WT (Figure S4), and no other products incorporating AzF were clearly detected. This suggests that although AlbC-WT can utilize AzF-tRNA as a substrate, its activity and specificity are relatively low. We hypothesize that the directed evolution of AlbC could improve both its catalytic efficiency and specificity against the AzF-tRNA substrate.

Directed Evolution of AlbC for Improving Activity toward AzF-tRNA.

AlbC contains two binding pockets, P1 and P2, each responsible for recognizing specific amino acid motifs on two aa-tRNAs (Figure 3a). P1 exhibits higher specificity for Phe-tRNA, while P2 is more flexible, accommodating a wider range of aa-tRNAs.^30,31 Given the structural similarity between phenylalanine and ncAAs recognized by the evolved EcTyrRS, we first focused on evolving the P1 pocket to enhance its recognition of ncAAs. Based on the crystal structure of AlbC,³² eight key amino acid residues involved in Phe-tRNA recognition (Leu33, Gly35, Val65, Val67, Leu119, Leu185, Phe186, and Leu200; Figure 3a and 3b) were subjected to site-saturation mutagenesis (SSM) using NNK randomization (N = any nucleotide, K = G or T) in the first-round evolution. To evaluate the activity of each library member, we transformed the pET22b-T5-AlbC variant plasmids expressing the corresponding mutants into the E. coli BL21(DE3) cells containing the pUltra-EcTyrRS/tRNA plasmid encoding the evolved EcTyrRS/tRNA_CUA^Tyr pair. After overnight expression in the presence of AzF, the cyclodipeptides in the supernatant were analyzed by LC-MS. After screening approximately 800 colonies to achieve 95% coverage of the NNK library at eight positions (100 colonies for each position), we identified the L200G variant (AlbC-G), which resulted in the production of four new products with AzF (Figure S4). Interestingly, Sauguet et al. previously reported that the L200N variant of AlbC shifted production from cFL to cYL, suggesting that Leu200 plays a crucial role in determining the specificity of the first aminoacyl-tRNA binding.³⁰ Although the yield of four AzF-containing products remained relatively low, the production of natural compounds 14–17 was nearly abolished using the AlbC-G variant, indicating a significant enhancement in recognition for AzF-tRNA with the L200G mutation (Figure 3c). Based on LC-MS analysis and comparison with authentic standards (Figure S3), three of the newly identified compounds were cyclo-(L-AzF–L-Leu) (cAzFL, 18, Figure 2d), cyclo-(L-AzF–L-Tyr) (cAzFY, 20, Figure 2d) and cyclo-(L-AzF–L-Met) (cAzFM, 21, Figure 2d). Most excitingly, we observed a product with m/z 376 [M + H]+, corresponding to cyclo-(L-AzF–L-AzF) (cAzFAzF, 22, Figure 2d and S5). The formation of cAzFAzF is likely due to the substrate promiscuity of the P2 binding site. Notably, no reports exist on CDPS-generated cyclodipeptides incorporating ncAAs on both sides. With an AlbC variant producing more 2,5-DKPs containing AzF, we next aim to develop a strategy to further enhance its efficiency in AzF incorporation. The yields of different products are used here as a proxy of enzyme efficiency.

Figure 3. Directed evolution of AlbC for improved and specific recognition toward AzF-tRNA.

(a) Structure and binding pocket of AlbC. (b) Flowchart of the directed evolution strategy for AlbC to enhance substrate recognition specificity toward AzF-tRNA. (c) Yields of products from AlbC-WT and its variants toward AzF-tRNA and other aa-tRNAs. Fold changes in compound production were calculated by setting the cFF concentration (quantified using a standard curve) produced by AlbC-WT without feeding as 1.0. Fold changes for the variants were then normalized to their respective protein expression levels to obtain the final normalized fold change. Data are presented as mean ± standard deviation (n = 3 independent biological replicates). (d) HPLC chromatogram showing cyclodipeptide peaks produced by AlbC-GR (UV=280 nm). (e) Yields of cyclodipeptides from large-scale AzF feeding using AlbC-GR.

To enhance the yield of AzF-containing 2,5-DKPs, we performed SSM on remaining 14 residues within the P1 and P2 binding pockets (Leu33, Gly35, Val65, Val67, Leu119, Met152, Ala155, Val156, Asn159, Leu185, Phe186, Ile204, Thr206, and Pro207; Figure 3a and 3b) together with L200G mutation that was identified critical for AzF-tRNA recognition (Figure 3c). After screening approximately 1,400 colonies, we identified the variant AlbC-L200G/I204R (AlbC-GR), which produces cAzFL at a higher yield than cFL produced by AlbC-WT (Figure 3c). Moreover, AlbC-GR exhibited a remarkable 9-fold increase in cAzFL yield compared to AlbC-G. Notably, AlbC-GR also exhibited the ability to synthesize unnatural 2,5-DKPs 14–17, albeit at a very low yield. Interestingly, Ile204, located within the P2 pocket, highlights the critical role of P2 pocket residues in CDPSs for product accumulation. This effect is reminiscent of the double mutant T82V/Y196F in the P1 pocket, which exhibited enhanced product formation compared to the wild-type CDPS-Np.³³ In this second-round screening, we also generated other variants, including AlbC-V65F/L200G (AlbC-FG), AlbC-V67A/L200G (AlbC-AG), AlbC-F186I/L200G (AlbC-IG), and AlbC-F186M/L200G (AlbC-MG), which exhibit higher yields of cAzFM than cAzFL (Figure S6). This implies that mutations at Val65, Val67, and Phe186 within pocket P1 can alter the recognition of the second substrate, thereby facilitating the incorporation of Met instead of Leu. This observation is consistent with a recent report that F186L mutation within AlbC can alter the substrate specificity for the second substrate, resulting in the production of cFV, a compound that is not detectable in the AlbC-WT.³⁴ Collectively, these data highlight the critical role of residue Phe186 in recognizing and altering the second substrate. Additionally, mutating Val156 in the P2 pocket to Phe, Lys, Arg, or Tyr led to increased production of both compounds, cAzFL and cAzFF (Figure S6). Notably, AlbC-V156R/L200G (AlbC-RG) enhanced the yield of 2,5-DKPs containing AzF compared to L200G alone (Figure S6).

To further enhance the yield of AzF-containing products based on AlbC-GR, we screened 1,300 colonies using SSM focusing on the binding pockets (Leu33, Gly35, Val65, Val67, Leu119, Met152, Ala155, Val156, Asn159, Leu185, Phe186, Ile204, Thr206, and Pro207). However, no significant improvements were observed. Consequently, we decided to explore additional mutagenesis of His203, Asp205, Thr206, and Pro207, as previous observations indicated that mutations in neighboring residues exhibited cooperative effects.^35,36 Starting from AlbC-GR, screening 400 colonies revealed that the addition of D205I mutation (AlbC-GRI) led to a two-fold increase in the yield of compound cAzFL compared to that of AlbC-GR (Figure 3c). In parallel, we also utilized error-prone PCR to create a random mutagenesis library of AlbC-GR. The evaluation of activities of about 600 colonies revealed that the AlbC-LGR variant, with the P161L, L200G, and I204R mutations, yielded twice as much cAzFL as AlbC-GR (Figure 3c). In the final round of evolution, we combined these mutations to generate the most active variant, AlbC-LGRI, which produced 23 times more cAzFL than AlbC-G (Figure 3c). The yields of other products (19-22) in AlbC-LGRI also showed a significant increase compared to those with AlbC-WT. Notably, AlbC-LGRI produces approximately five times more cAzFL than its natural product counterpart, cFL, biosynthesized by AlbC-WT. This highlights that AlbC-LGRI produces AzF-containing products at higher yields than AlbC-WT produces its natural products. To reduce the yield of 14-17, we introduced the V67A and V156R mutations identified in the second round of screening into AlbC-LGRI, generating the variants AlbC-ALGRI and AlbC-RLGRI. While both variants demonstrated an enhanced ratio of products containing AzF, their yields were lower compared to AlbC-LGRI (Figure S7). AlbC-RLGRI predominantly produced compound cAzFL, suggesting that the P1 and P2 binding pockets may have become more constricted compared to the promiscuous AlbC-WT. In addition, the exclusive production of a single product, compound 18, simplifies the separation and purification processes, representing a significant advancement for the practical application of this platform.

To characterize the products of AlbC variants in detail, we selected AlbC-GR for large-scale production because it produced a relatively high proportion of AzF-containing compounds. Since 2,5-DKPs are commonly found in culture supernatants, we employed a standard ethyl acetate extraction method to concentrate these compounds significantly. We then used preparative HPLC (Prep-HPLC) to isolate four major metabolites: 18, 19, 20, and 22 (Figure 3d and 3e). Product 18 exhibited identical NMR data to chemosynthetic cAzFL (Figure S13). Product 22 was conclusively identified as cAzFAzF, a symmetrical compound, based on its LC-MS data and NMR spectra (Figure S5 and S39–S40). The presence of two azido groups in compound 22, functionalized for bioorthogonal chemistry, expands the potential applications of these variants in chemical biology. The isolation of four distinct compounds in a single experiment demonstrates that evolving promiscuous CDPSs, in combination with GCE, can generate multiple ncAA-incorporated cyclodipeptides simultaneously. This approach offers a powerful platform for expanding the structural diversity of non-natural 2,5-DKPs.

The Molecular Basis of AlbC Variants for Recognizing AzF-tRNA.

Cyclodipeptide formation mediated by CDPSs involves a characteristic ping-pong catalytic mechanism. Initially, the first aa-tRNA substrate binds to the P1 pocket, transferring its aminoacyl group to a conserved serine residue (e.g., Ser37 in AlbC), resulting in an aminoacyl-enzyme intermediate. This intermediate is hypothesized to interact with the aminoacyl moiety of a second aa-tRNA substrate, forming a dipeptidyl-enzyme intermediate.³⁷ Subsequently, this intermediate undergoes intramolecular cyclization, producing the final cyclodipeptide product (Figure S14). Consequently, the dipeptidyl-enzyme intermediate has become a key focus for researchers investigating the molecular basis of various CDPSs and their derivatives. Using evolutionary insights, we applied computational modeling of the dipeptidyl-enzyme intermediate to illustrate the underlying mechanisms for the enhanced efficiency of AlbC variants toward the noncanonical substrates.

We modeled AlbC variants using the crystal structure of AlbC in complex with its intermediate (PDB: 4Q24)³⁸ as a template. In the substrate-binding pocket, the benzene ring of cFL is positioned near the critical residue L200 within the P1 pocket (Figure 4a). Upon mutating L200 to Gly in our initial directed evolution, the P1 binding pocket expanded, accommodating AzF and facilitating the formation of cAzFL (Figure 4b). Subsequent mutation of Leu204 to Arg showed an enhanced capability to form hydrogen bonds with dipeptidyl enzyme intermediates, stabilizing the complex structure (Figure 4c). In addition, residues Asn159, Arg160, and Asp163 from the α6-α7 loop, together with Asp205 from the β6−α8 loop, are previously reported to be critical for facilitating interactions with aa-tRNA substrates.³⁹ Based on the increased yield observed after introducing the D205I and P161L mutations (Figure 4d), we hypothesize that these modifications enhance the binding affinity between the α6-α7 and β6-α8 loops of AlbC and AzF-tRNA.

Figure 4. Molecular basis of AlbC variants in recognizing AzF-tRNA.

a–d Structure and binding pocket of AlbC variants with cyclodipeptide intermediates. (a) AlbC-WT with cFL. (b) AlbC-G with cAzFL. (c) AlbC-GR with cAzFL. (d) AlbC-LGRI with cAzFL. (e) Binding free energy of AlbC variants with cAzFL intermediates. (f) Binding free energy of AlbC-WT and its variants with different cyclodipeptide intermediates. Relative ΔG_binding values are ranked from low to high as “+++” (lowest, most favorable), “++”, “+”, and “−” (highest, least favorable).

To test our hypothesis, rigid protein-tRNA docking was conducted between AlbC-LGRI and AzF-tRNA. The results indicated that residues Leu161, Arg204, and Ile205 of AlbC-LGRI were positioned at the interface between AlbC-LGRI and AzF-tRNA (Figure S15). Notably, AlbC-LGRI and AzF-tRNA form hydrogen bonds between Arg204-guanine27 and Ile205-guanine29, suggesting a potentially stable protein-tRNA interaction. Furthermore, the importance of this interaction is underscored by the product profiles observed when different tRNA synthetases were used. Compared to the evolved EcTyrRS, PheRS-A294G led to distinct product distributions across AlbC variants (Figure S16). While AlbC-WT produced negligible amounts of AzF-containing compounds, the tested variants AlbC-GRI, AlbC-LGR, and AlbC-LGRI successfully generated AzF-derived products, with AlbC-LGR showing the highest fold change (Figure S16). These results highlight the critical influence of tRNA recognition on both product composition and overall yield.

Molecular Dynamics (MD) Simulations of AlbC Variants.

To better understand the binding mechanisms of different substrates, we separately docked cAzFL, cAzFAzF, and the native cFL into the binding pockets of AlbC and its variants, followed by 50 ns MD simulations.⁴⁰ The stability of AlbC and its variants was evaluated by the root mean square deviation (RMSD) analysis, while the binding stability of cyclodipeptide intermediates was assessed by their RMSD relative to AlbC variants. Over 50 ns, RMSD measurements revealed stable interactions between AlbC variants and cyclodipeptide intermediates (Figure S17). Notably, AlbC variants engineered for AzF reached equilibrium early, suggesting stable complex formation. Similarly, the rapid stabilization of RMSD in AlbC-WT with cFL, AlbC-GR with cFL, and AlbC-GR with cAzFAzF highlights their efficiency in synthesizing diverse cyclodipeptides (Figure S17). In contrast, the RMSD trajectory for AlbC-WT with cAzFAzF showed significant fluctuations, indicating the instability of the corresponding complex. Additionally, we assessed the root mean square fluctuation (RMSF) of the amino acid backbone of AlbC and its variants. Higher RMSF values indicate more flexibility, while lower values suggest greater stability. During complex formation with cAzFL, AlbC-LGRI exhibited a lower RMSF value, indicating greater stability than AlbC-G and AlbC-GR (Figure S18) and suggesting that AlbC-LGRI more effectively maintains amino acid stability. In interactions between AlbC with cFL, AlbC-GR with cAzFL, and AlbC-GR with cAzFAzF, the RMSF values showed minimal fluctuations, hinting at the stability of their amino acid backbones.

Furthermore, the binding free energy of AlbC and its variants with different intermediates was calculated and closely aligned with our experimental findings. Notably, the AlbC-LGRI variant exhibited a lower binding free energy against cAzFL than that of AlbC-GR and AlbC-G (Figure 4e and Table S5), which is correlated with the cAzFL production yields of these mutants. Consistent with the binding free energy calculations (Table S5), the product yield of AlbC-GR follows the order: cAzFL > cAzFAzF > cFL. In addition, AlbC-WT exhibits stronger binding affinity toward cFL but significantly weaker binding affinity with cAzFAzF (Figure 4f and Table S5). This finding aligns with the observation that AlbC-WT cannot biosynthesize unnatural 2,5-DKPs containing two AzFs. It should be noted that free energy calculations obtained from docking and MD simulations are used here as a proxy for catalytic activity but may not capture the full range of factors influencing enzyme performance.

Expanding the Substrate Scopes of AlbC Variants with Other ncAAs.

Since the evolved EcTyrRS/tRNA pair showed effective activities against a broad spectrum of ncAAs, we next expand the substrate scope of AlbC variants. We screened the activities of AlbC-WT and its variants, including AlbC-G, AlbC-GR, AlbC-GRI, AlbC-LGR, and AlbC-LGRI, toward ncAAs 1-4 and 6-13 (Figure 2a) using expression in deep 96-well plates. After ncAA feeding and expression, we performed LC-MS analysis to compare the extracted ion chromatogram (EIC) peak of the corresponding cyclodipeptides. By comparing samples with and without ncAA feeding, we minimized interference from cellular metabolites in the mass spectrum signal. Next, to illustrate the substrate screening results, we generated a heatmap based on the EIC peak areas of the corresponding cyclodipeptides detected for each introduced ncAA (Figure 5b and S9–11). As shown in Figure 5b, AlbC-WT failed to produce detectable products when feeding 6, 10, 12, and 13 as substrates (Figure 2a). Conversely, its variants AlbC-LGR and AlbC-LGRI could recognize these four ncAAs and efficiently generated products including cyclo-(L-IodF–L-Leu) (cIodFL, 28, Figure 5a), cyclo-(L-AllY–L-Leu) (cAllYL, 29, Figure 5a), cyclo-(L-TriFF–L-Leu) (cTriFFL, 34, Figure 5a), and cyclo-(L-PropY–L-Leu) (cPropYL, 35, Figure 5a). Additionally, the AlbC-GRI and AlbC-LGRI variants led to increased production of cyclo-(L-FluF–L-Leu) (cFluFL, 30, Figure 5a), cyclo-(L-FluF–L-Phe) (cFluFF, 31, Figure 5a), cyclo-(L-FluF–L-Tyr) (cFluFY, 32, Figure 5a), and cyclo-(L-FluF–L-Met) (cFluFM, 33, Figure 5a). Among all tested variants, AlbC-LGRI exhibited higher yields of 2,5-DKPs containing ncAAs with diverse functional groups (Figure 5b). In particular, OMeY (3, Figure 2a) was efficiently incorporated by the AlbC-GRI, AlbC-LGR, and AlbC-LGRI variants, leading to the abundant production of cyclo-(L-OMeY–L-Leu) (cOMeYL, 23, Figure 5a and 5b). Additionally, all variants were capable of producing cyclo-(L-OMeTyr–L-OMeTyr) (cOMeYOMeY, 27, Figure 5a and 5b), a cyclodipeptide containing two ncAAs. AlbC-GRI yielded the highest amount of this unique product, while AlbC-GR produced a higher proportion of 2,5-DKPs containing ncAAs (Figure S11). To characterize products 23 and 27, we conducted a large-scale fermentation of E. coli expressing AlbC-GR and fed with OMeY (Figure 5c). After extraction with ethyl acetate and purification by Prep-HPLC, we successfully obtained enough compounds and characterized their structures using NMR analysis (Figure S41–44). Overall, these findings demonstrate that AlbC variants enhance the efficiency of ncAA incorporation by modulating their recognition toward ncAA-tRNA, positioning them as promising enzymatic tools for diversifying new-to-nature compounds. The formation of cOMeYOMeY and cAzFAzF further suggests that, under conditions where Phe competes for both P1 and P2, engineering these pockets enables AlbC to accommodate two ncAAs simultaneously.

Figure 5. Expanding the substrate scope of AlbC variants and other CDPSs.

(a) Chemical structures of cyclodipeptide products generated from feeding with various ncAAs. (b) Heatmap summarizing the production of cyclodipeptides by AlbC-WT and its evolved variants with selected ncAAs. EIC peak areas were normalized to the corresponding protein levels, reflecting relative product yields and substrate tolerance of each variant. The data highlight enhanced incorporation of specific ncAAs by several AlbC variants. (c) HPLC trace of AlbC-GR after large-scale fermentation with compound 3 (OMeY). (d) Expression of Rv2275 with and without AzF supplementation. (e) Expression of RmCDPS with AzF or OMeY supplementation. (f) Chemical structures of ncAA-containing cyclodipeptides produced by Rv2275 and RmCDPS.

Exploration of Other CDPSs for the Biosynthesis of ncAA-Containing 2,5-DKPs.

To broaden the incorporation of ncAAs into cyclodipeptide products using evolved EcTyrRS/tRNA and promiscuous CDPSs, we selected four enzymes including Rv2275,⁴¹ Rgry-CDPS,²³ MmCDPS (WP_019864150.1),⁴² and RmCDPS (WP_010598044.1)²³ to explore their compatibility with evolved EcTyrRS/tRNA. They are known to produce multiple compounds containing at least one Phe or Tyr (Table S4). After co-expressing each enzyme with evolved EcTyrRS/tRNA and separately supplementing the cultures with OMeY (3) and AzF (5), we successfully observed the production of corresponding products for Rv2275 and RmCDPS. To be specific, supplementing 5 in the medium with Rv2275 led to the appearance of a new UV chromatogram peak, matching the mass data and retention time of 20 (cAzFY, Figure 5d). Similarly, introducing 3 to the versatile RmCDPS resulted in two distinct UV peaks with m/z 275 [M+H]⁺ and m/z 277 [M+H]⁺ (Figure 5e and S12), corresponding to the putative compounds cyclo-(L-OMeY–L-Pro) (cOMeYP, 36, Figure 5f) and cyclo-(L-OMeY–L-Val) (cOMeYV, 37, Figure 5f). When 5 was used as a substrate, LC-MS analysis identified three major products (Figure 5e and S12), including cyclo-(L-AzF–L-Ala) (cAzFA, 38, Figure 5f), cyclo-(L-AzF–L-Pro) (cAzFP, 39, Figure 5f) and cyclo-(L-AzF–L-Val) (cAzFV, 40, Figure 5f), consistent with previous reports showing that RmCDPS predominantly generates cYA, cYP, and cYV.⁴² These findings suggest that combining evolved aaRS with promiscuous CDPSs has the potential to produce a diverse array of “unnatural” natural products with unique functional groups, including those absent in naturally occurring compounds.

CONCLUSIONS

Natural products with a cyclodipeptide skeleton often exhibit diverse biological and pharmacological activities, making them promising candidates for drug development. In vivo systems utilizing E. coli auxotrophic strains have been explored for producing cyclodipeptides containing ncAAs.¹⁹ However, insufficient levels of ncAA-tRNAs can result in low ncAA loading efficiencies by CDPSs. The structural constraints of native aaRS and CDPS binding sites restrict the range of ncAAs that can be incorporated into 2,5-DKPs. Our study tackles this challenge by leveraging engineered aaRSs evolved using GCE technology and CDPSs optimized through directed evolution. This approach provides a robust platform for the biosynthesis of novel non-ribosomally synthesized natural products containing noncanonical building blocks.

To achieve this, we evolved AlbC in the presence of the evolved EcTyrRS/tRNA from GCE technology to improve their recognition toward ncAA-tRNA. This led to the development of the variant AlbC-LGRI, which showed enhanced efficiency for incorporating the “clickable” amino acid AzF. Notably, this AlbC mutant produces significantly higher yields of AzF-containing products compared to the natural products generated by native AlbC. AlbC-LGRI also demonstrated improved recognition of other ncAAs, enabling the production of a wider range of cyclodipeptides containing Phe or Tyr analogs. Although the evolution of individual P1 and P2 binding pockets in CDPSs has been reported,^34,43–45 evolving both to incorporate two ncAAs is rare. The AlbC mutants successfully produced cAzFAzF containing two AzF moieties. Additionally, using OMeTyr as a substrate led to the production of cOMeYOMeY, suggesting that 2,5-DKPs entirely composed of ncAAs can be biosynthesized. Lastly, structural analysis, MD simulations, and free energy calculations have provided insights into the mechanisms underlying the substrate recognition of the AlbC variants.

GCE technology has been predominantly used to incorporate ncAAs into ribosomally-synthesized proteins or peptides in vivo.^46–51 However, its application in generating non-ribosomal biomolecules with noncanonical building blocks has been rare. Cyclodipeptides can be enzymatically synthesized by two distinct enzyme families: NRPSs and CDPSs. Unlike the complex multimodular structures of NRPSs,⁵² CDPSs offer a more accessible platform for synthesizing unnatural cyclodipeptides. Our study expands the substrate scope of CDPSs to recognize ncAA-tRNA by integrating poly-specific EcTyrRS/tRNA generated through GCE technology and employing novel CDPS variants obtained via directed evolution. This strategy not only accelerates the production of diverse cyclodipeptides containing ncAAs but also develops a robust toolkit for the biocombinatorial synthesis of diverse 2,5-DKPs. By incorporating more promiscuous CDPSs and leveraging evolved aaRSs from E. coli, such as EcLeuRS,⁵³ EcTrpRS,⁵⁴ and an additional evolved EcTyrRS,⁵⁵ we can produce cyclodipeptides with a broader range of functional groups. Furthermore, combining CDPSs with a variety of tailoring modifications holds significant potential for creating diverse 2,5-DKP libraries for drug discovery. In addition, aa-tRNA participates in the biosynthesis of various natural products beyond cyclodipeptides, including LanB-like dehydratases.⁵⁶ Thus, the growing number of aminoacyl-tRNA-dependent enzymes highlights this approach as a promising strategy for synthesizing diverse unnatural products.^57–59 This study presents a powerful strategy for employing GCE technology to introduce a broader range of chemical modifications into diverse natural products beyond ribosomally-synthesized polypeptides.

Supplementary Material

The Supporting Information is available free of charge on the ACS Publications website. General materials and detailed experimental procedures, supplementary tables and figures, and NMR spectra (PDF)