S. tsukubaensis NRRL 18488 from the NRRL Culture Collection of the Agricultural Research Service (USA). All mutants are derivatives of S. tsukubaensis NRRL 18488. Escherichia coli Novablue cells from Novagen were used for standard cloning procedures, while E. coli Rosetta DE3 cells served as hosts for protein expression experiments. For plasmid transfer intergeneric conjugation between the non-methylating E. coli ET12567/pUZ8002 strain and S. tsukubaensis was used (Kieser et al., 2000). For overexpression experiments the integrative pRM4 vector containing the strong constitutive ermE promoter was used (Menges et al., 2007).

For heterologous expression experiments concerning the lysine cyclodeaminases, the pipA gene was amplified from genomic DNA from Streptomyces pristinaespiralis Pr 11 (Aventis Pharma) (Mast et al., 2011). As template for the amplification of the pip gene from Actinoplanes friuliensis the plasmid pRF37 from C. Müller et al., 2007 was used (Müller et al., 2007). S. tsukubaensis NRRL 18488 was used as a template for the amplification of the fkbL, fkbP and dapA.

The genomic DNA for the amplification of the feedback inhibition deregulated aspartate kinase gene lysCCg* from Corynebacterium glutamicum DM1730 (Seibold et al., 2006) was kindly provided by Jörn Kalinowski (University of Bielefeld, Germany).

All strains and plasmids used in this study are listed in the Supplementary Table S2. All oligonucleotides used in this work are listen in the Supplementary Table S3.

For generation of the S. tsukubaensis fkbL mutant, a 2.2 kb upstream and downstream regions of the fkbL gene were amplified from the genomic DNA of S. tsukubaensis NRRL 18488 with the primer pair fkbLupfw/fkbLuprev and fkbLdownfw/fkbLdownrev. The apramycin resistance cassette was amplified from the vector pSET152 using the primers aprneufw/aprneurev. Purified PCR fragments have been first cloned in the intermediate vector pDrive and afterwards in the pK18 vector. Novablue competent E. coli cells were then transformed with the recombinant plasmid pK18oriTΔfkbLaprR containing the inserted fkbL upstream and downstream fragments as well as the apramycin resistance cassette. The clones were selected for kanamycin and apramycin resistance and verified by sequencing. Afterwards, the pK18oriTΔfkbLaprR plasmid was transferred into the methylation-deficient E. coli ET12567/pUZ8002 and E. coli ET12567/pUB307 strain. After the biparental conjugation between the methylation-deficient E. coli strains and S. tsukubaensis NRRL 18488, the ex-conjugants were selected for the double recombination event at both flanking homologous sequences by demonstrating apramycin resistance and kanamycin sensitivity.

Spores and mycelia preparations were obtained from ISP4 (Difco, Sparks, MD, USA) agar plates. FK506 production by S. tsukubaensis NRRL 18488 was analyzed in the liquid media MG optimized by Martínez-Castro et al., 2013. FK506 production was also analyzed in R2YE medium (Kieser et al., 2000) modified by Temuujin et al., 2011. The composition of MGm medium was as follows: 50 g/l starch (Difco), 8.83 g/l glutamic acid, 2.5 mM KH2PO4/K2HPO4, 0.2 g/l MgSO4·7H2O, 1 mg/l CaCl2, 1 mg/l NaCl, 9 mg/l FeSO4 ·7H2O, 21 g/l MOPS, and 0.45 ml/l 10× trace elements, adjusted to pH 6.5. The composition of R2YE medium was as follows: 103 g/l saccharose, 0.25 g/l K2SO4, 10.12 g/l MgCl2*6H2O, 10 g/l glucose, 0.1 g/l casamino acids, 5 g/l yeast extract, 10 ml/l 0.5% K2HPO4, 80 ml/l 3.68% CaCl2*2H2O, 15 ml/l 20% L-prolin, 100 ml/l 5.73% TES, 2 ml/l trace elements, pH 7,2. For 10 ml solution (100×) of trace elements, the components are: 39.0 mg CuSO4 ·5H2O, 5.7 mg H3BO3, 3.7 mg (NH4)6Mo7O24·4H2O, 6.1 mg MnSO4·H2O, and 895.0 mg ZnSO4·7H2O. Fermentation was performed using a two-stage culture system. For the seed culture a mixture of 1:1 modified YEME (3 g/l yeast extract, 5 g/l Bacto peptone, 3 g/l malt extract, 10 g/l glucose and 340 g/l sucrose) and Tryptic Soy Broth (Difco) was inoculated with spores or mycelia from an ISP4 plate and cultivated for at least 3 days at 28°C and 220 rpm. For the main culture 100 ml of MGm medium were inoculated with the seed culture and further incubated at the same condition for 6 days.

The productivity of an S. tsukubaensis strain was usually determined as production of FK506 per g biomass (cell dry weight). During the fermentation time of 6 days all 24 h 1 ml of bacterial culture was gathered, centrifuged, washed twice with destilated water, and dried by lyophillization. For FK506 determination, the broth samples were mixed with equal volume of ethylacetate (1:1), stirred for 10 min and subsequently centrifuged. The organic phase was analysed via an HP1090 M HPLC system equipped with a diode array detector and a thermostatic autosampler. A Zorbax Bonus RP column, 3 × 150, 5 μm (Agilent Technologies, Santa Clara, USA) constituted the stationary phase. The mobile phase system was applied with 0.1% phosphoric acid and 0.2% triethylamine as eluent A and acetonitrile with 1% tetrahydrofurane as eluent B. The flow rate was 850 μL/min, the column temperature was set to 60°C. The absorbance was monitored at a wavelength of 210 nm. Data sets were processed with the help of the Chemstation LC 3D, Rev. A.08.03 software from Agilent Technologies. Standards of pure FK506 (Antibioticos SA) and ascomycin (Sigma-Aldrich Chemie GmbH, Munich, Germany) were used as controls.

A 3D protein model of a target sequence was generated in comparative modelling by extrapolating experimental information from an evolutionary related protein structure that serves as a template using SWISS-MODEL (Waterhouse et al., 2018). Template search was based on the target sequence, which served as a query to search for evolutionary related amino acid sequences by BLASTP (Altschul et al., 1997) and for protein structures against the SWISS-MODEL template library. Templates were ranked according to expected quality of the resulting models and used to automatically generate 3D protein models by loop modelling. The quality of obtained models was estimated based on expected model accuracy, the QMEAN scoring function (Studer et al., 2020) as well as based on the Ramachandran plot (Ramachandran et al., 1963). 3D models of proteins were analyzed using the Swiss-PdbViewer software (Guex et al., 2009).

The pipAf gene was amplified using the above mentioned template from Müller et al., 2007. It was cloned in the expression vector pET30 Ek/lic containing an N-terminal His-tag with the help of the ligation independent Ek/lic system according to the protocol proposed by Novagen, UK. E.coli Rosetta DE3 was transformed with the resulting construct pET30/pip. 200 ml of Luria Bertami broth (LB) (Sambrook & Russell, 2006) with kanamycin (50 μg/ml) and chloramphenicol (34 μg/ml) was inoculated with 2 ml of an overnight culture and incubated at 37°C on a rotary shaker (180 rpm). Expression was induced with 1 mM IPTG after the culture broth has reached an OD578 of 0.4. The induced culture was further incubated for 18 h. The cells were harvested by centrifugation, resuspended in lysis buffer (20 mM Tris, 100 mM NaCl, 20 mM imidazol, protease inhibitor (Roche, Mannheim, Germany)) and lysed by 3-4 passages through Emulsiflex B15 (Avestin, Ottawa, Canada). Protein purification was either carried out by gravity flow Ni NTA Superflow columns (Iba) or FPLC using a His-TrapFF 1 ml column (GE Healthcare, Munich, Germany). Concentrated protein eluates were stored at 4°C.

Mutated Pip* variants were generated by site-directed mutagenesis. The procedure was based on the Stratagene quickchange protocol, which was optimized by Zheng et al. i (2004). For each mutation specific primers were designed. Mutations were introduced into the pET30/pip expression plasmid using PCR. Non-mutated template DNA was digested using the restriction enzyme DpnI. Subsequently, the competent E. coli Novablue strain was transformed. Plasmid-carrying clones were identified by selection for kanamycin and verified by PCR and sequencing.

The obtained PipAf was concentrated by Amicon Ultra spin concentrators and further purified by gel filtration on a Superdex 200 run with 150 mM NaCl and 25 mM HEPES pH 7.4. The main peak corresponding to a dimer was pooled and concentrated up to absorption at 280 nm of 2.5 equal to a protein concentration of 9.5 mg/ml. The absorption at 340 nm thereof varied from 0.4 to 0.8 depending on the batch. For crystallization the concentrate was supplemented with 1 mM NAD+ and diluted to 7.3 mg/ml. The same solution was also used for co-crystallization with the substrate but contained additionally lysine to a final concentration of 140 mM. The crystallization drops contained 450 nl of protein and 300 nl of crystallization solution and were set up in sitting drop microtiterplates with a Tecan Freedom Evo pipetting robot. First crystals grew after few hours at 4°C in several conditions. Lysine-Cocrystallization was optimized with JCSG + suite screen (Qiagen, Hilden, Germany) condition 42 (20% PEG8000, 0.2 M MgCl2, 0.1 M Tris pH 8.5) and 10% glycerol added before freezing. From the setups without lysine crystals were obtained from condition 36 of Morpheus screen (Molecular dimensions) which were of sufficient quality.

The crystals were mounted in loops and flash-frozen in liquid nitrogen for storage and measured at the Swiss Light Source (SLS, Villingen, Germany). Data reduction was carried out with the XDS package. For phase determination with PHASER the OCD structures were truncated and used as search model. The structure was built in interactive cycles with coot and for refinement; simulated annealing was included in the very first refinement run. The representation of the solved protein structure was carried out using the free graphics 3D software PyMOL (The PyMOL Molecular Graphics System, Version 2.0 Schrödinger, LLC).

Overnight cultures of expression strains E. coli Rosetta 2 (pET/pipAf) and E. coli Rosetta 2 (pET/pipAf*) and the control E. coli Rosetta 2 were incubated overnight at 37°C in 10 ml LB medium with the corresponding kanamycin and chloramphenicol concentrations. Subsequently, 100 ml of LBKM/CM of the main culture was supplemented with 1 µL of the corresponding pre-culture and incubated up to an OD578 of 0.6 at 37°C. After reaching the OD578 of 0.6, the expression of the PipAf/PipAf* proteins was induced by adding 1 mM IPTG with the subsequent incubation at 28°C for further 20 h. At the next day, the main cultures were supplemented with 0.5 ml of the glycerol stock solution (50%) as well as 0.5 ml lysine stock solution (50%) and incubated for 6 h at 32°C. Afterwards, the cultures were centrifuged for 10 min at 5,000 rpm. To determine the formation of pipecolic acid, 5 µL of supernatant from each culture was spotted at a silica gel plate. The mobile phase, which consisted of a mixture of 3:1:1 N-butanol:acetic acid:water, was used for the separation of substances. As controls the pure lysine and pipecolic acid (5 µg each) were used. After the running front had reached a height of approx. 10 cm, the silica gel plate was stained with 0.5% ninhydrin solution and analyzed. Quantification of pipecolic acid on TLC was performed using the open-source software ImageJ: percentage of each peak was related to the total area of all peaks and represents the relative amount of the pipecolate based on digital imaging.

L-lysine is an essential amino acid, used in nutrition, as supplementary and nutraceutical, which production has been established in C. glutamicum. Also, lysine is an important product that serves as a precursor for pipecolic acid (piperidine-2 carboxylic acid), which is a non-proteinogenic amino acid widely used in all phylogenetic domains of life and which is a building block of biologically active substances such as streptogramin pristinamycin (Mast et al., 2011) as well as macrocyclic immunomodulators such as rapamycin and FK506 (Turlo et al., 2012). Enhancing the L-lysine precursor pool should result in a positive effect on secondary metabolite production. To proof this hypothesis classical media supplementation experiments with various amino acids have been performed showing that particularly exogenous L-lysine feeding lead to a significant boost of rapamycin (Cheng et al., 1995) respectively FK506 production (Huang et al., 2013; Martínez-Castro et al., 2013), consequently making the L-lysine primary biosynthesis pathway a putative target for genetic engineering. For instance, enrichment of the semi-defined medium for FK506 production in S. tsukubaensis with 2.5 g/l L-lysine enhanced the production by approximately 30% (Martínez-Castro et al., 2013). In order to test, whether additional lysine can further improve the FK506 yield, we extended this experiment by increasing the exogenous lysine end concentration beyond 2.5 g/l (17 mM). However, at higher exogenous lysine concentration (50 mM) the FK506 production in the wild type was clearly inhibited (Supplementary Figure S1), what can be attributed to feedback inhibition.

In order to avoid this feedback inhibition we intended to adjust the lysine biosynthetic pathway in S. tsukubaensis NRRL 18488 by overcoming the inhibition of the aspartate kinase and redirecting the carbon flux towards the lysine branch. We therefore chose the dihydropicolinate synthase (DapASt) and the aspartate kinase (AskSt) to be processed to direct the precursor flux towards lysine and hence towards elevated levels of the building block pipecolic acid.

The FK506 production in all strains grown in the MG production medium was analyzed by HPLC, whereby samples were taken after every 24 h and used for the biomass estimation and FK506 analytic. We analyzed the production after 6 days of Streptomyces tsukubaensis fermentation for two reasons. On the one hand, this strain is in the late stationary phase and the maximum biomass accumulation was achieved. On the other hand, many reports describe production of FK506 and other secondary metabolites in Actonibacteria in the late stationary phase as well (after 6–8 days of fermentation) (Müller et al., 2007; Jung et al., 2009; Goranovic et al., 2010; Mo et al., 2011; Chen et al., 2012).

In the WT strain the FK506 production in the MGm medium remained almost unchanged after 2 days of production (Supplementary Figure S2). Interestingly, although the WT demonstrated in our experiments higher initial FK506 production in the R2YE medium, recombinant strains showed strongly reduced FK506 titer in this medium. Similar results were obtained for industrially used M1 and M2 medium as well as for the ISPz for FK506 production. Since in our experiments only MGm medium delivered stable production levels, we performed all fermentations in this medium.

Overexpression of both enzymes DapASt and AskSt in the wild type producer S. tsukubaensis NRRL18488 individually did not result in a significant increase of FK506 yield (Figure 3). Considering the feedback regulation of the aspartate kinase as the major bottleneck in lysine biosynthesis, it is not surprising that overexpression of the dihydropicolinate synthase (DapASt) alone does not lead to a profound effect on FK506 production (∼7%). The overexpression of AskSt alone demonstrated no significant effect on FK506 production (∼8%) (Figure 3), since this enzyme is subject to strict regulation by a final product inhibition lysine and threonine.

The primary strategy to relieve this bottleneck was constructing a feedback insensitive aspartate kinase (AskSt*) using the data of Sahm et al., 1996 available for the aspartate kinase from the actinomycete C. glutamicum. For this purpose a single amino acid exchange in the wild type Ask (from Ser301 to Tyr301) was introduced via site-directed mutagenesis. The resulting AskSt* was overexpressed in S. tsukubaensis. The growth of strains was monitored in parallel to the FK506 production measurement (Supplementary Figures S2, S3). The biomass dry weight estimation revealed almost identical growth profile of the tested strains in comparison to the WT. The overexpression of AskSt* in S. tsukubaensis lead to a slight enhancement of approximately 17% of FK506 production (Figure 3).

Therefore, a second strategy involving the heterologous expression of a gene encoding a deregulated aspartate kinase (AskCg* = LysCCg*) (Seibold et al., 2006) from the high-level lysine producer strain C. glutamicum ATCC 13032 has been used. In our approach, the askCg* gene was additionaly introduced into the genome of S. tsukubaensis on the integrative plasmid pRM4 under the control of the constitutively expressed ermEp promoter. This attempt resulted in a significant FK506 yield improvement of about 46% (Figure 3) delivering a more beneficial outcome than supplementation of the production medium with lysine only reaching 30% improvement (Figure 3). Combining overexpression of both enzymes, the deregulated aspartate kinase (LysCCg*) together with the dihydrodipicolinate synthase (DapASt), consistently shifted the carbon flux toward L-lysine formation and resulted in a considerable upgrade of FK506 production of about 73% (Figure 3). This demonstrates the advantage of genetic engineering in contrast to exogenous feeding approaches. The overexpression of both enzymes resulted in an overproduction of FK506, presumably from the increase of lysine availability. However, the increase in the lysine intracellular pool might not be a consequence of a redirection of the carbon flux, but a more efficient pathway. In cephamycin producing S. clavuligerus strain a positive correlation between deregulated aspartate kinases and the improvement of antibiotic production was also previously described (Özcengiz et al., 2010).

The essentiality of FkbL for FK506 production was tested by the construction of an fkbL deletion mutant in S. tsukubaensis through the replacement of fkbL by an apramycin resistance gene, which resulted in a tacrolimus null-mutant. Subsequently, the fkbL mutant was successfully complemented with fkbL. The FK506 production in the mutant and the recombinant strains grown in the MG production medium was analyzed by HPLC. The loss of FK506 production in the ΔfkbL mutant of S. tsukubaensis could also be reconstituted by exogenous addition of pipecolate, which again proved the indispensability of this building block for FK506 biosynthesis (Supplementary Figures S2, S3).

In order to test whether overexpression of the essential fkbL gene increases FK506 production, the plasmid pRM4/fkbL including the fkbL gene under control of the constitutive ermE* promoter was introduced into S. tsukubaensis WT. The resulting construct revealed a yield increase of 47%. (Figure 4).

Pipecolate is incorporated into the growing FK506 backbone by the NRPS FkbP, which could constitute a limiting factor for FK506 production. Therefore, the effect of overexpression of fkbP was analyzed. However, only a slight yield improvement (11%) was observed when fkbP alone was overexpressed. However, combination of overexpression of both genes fkbL and fkbP led to a final increase of ∼55% (Figure 4). These results were acquired from production assays, which were carried out in flask cultivation with the chemically defined medium. Similar observations were reported from fermentation studies with S. tsukubaensis D852, another FK506 producer (Huang et al., 2013). For this study, genes fkbO, fkbL, fkbP, fkbM and fkbD were introduced into the parental strain (each as a single construct) using the integrative E. coli–Streptomyces vector pIB139 containing the ermE* promoter (PermE*) and overexpressed.

Since apparently the production of pipecolate is a highly relevant criteria for efficient FK506 production, this step of the biosynthesis was analyzed in more detail. Pipecolic acid is a widely distributed constituent of secondary metabolites like for example the streptogramin antibiotic pristinamycin I from Streptomyces pristinaespiralis (Cocito, 1979; Mast et al., 2011) and the lipopeptide friulimicin B from Actinoplanes friuliensis (Aretz et al., 2000; Müller et al., 2007). In these microorganisms the biosynthesis of the pipecolate building block is also catalyzed by a lysine cyclodeaminase encoded by pipASp in S. pristinaespiralis (Mast et al., 2011) and by pipAf in A. friuliensis (Müller et al., 2007). The products of both genes, PipASr and PipAf, show a protein identity of 57% and 50% to the wild type FkbL, respectively. In order to test, whether the heterologous lysine cyclodeaminases can replace the endogenous enzyme, each of the lysine cyclodeaminase genes was expressed heterologously in the ΔfkbL mutant under the control of the constitutive ermE* promoter. Indeed PipASp as well as PipAf were able to restore FK506 production demonstrating that both enzymes are able to form pipecolic acid needed for FK506 biosynthesis (Supplementary Figures S2, S3). In order to compare the potential of the PipASp and PipAf for the improvement of FK506 we overexpressed both pipASp and pipAf in S. tsukubaensis wild type under the control of the constitutive ermE* promoter. An improvement of FK506 production of 56% in case of pipASp and 65% in case of pipAf overexpression was observed (Figure 4).

Regarding the significant sequence homology of PipASp and PipAf to FkbL, it was rather unexpected to see a substantial difference in FK506 enhancement. So far, there was no evidence that the PipAf protein might be more efficient than the native FkbL in converting lysine into pipecolic acid. However, the results in FK506 enhancement were hinting towards structural differences and features of the binding pocket of PipAf that might allow it to be more efficient in lysine binding compared to PipASp and FkbL. To understand the basis for this effects, we aimed to analyze and compare the structures of the three lysine cyclodeaminases PipASp, PipAf and FkbL, to optimize subsequently the effectiveness of the FK506 production. Resolving of the LCD structure should provide a base for precise enzyme classification, understanding its properties, and future enzyme engineering.

The crystal structure of the LCD PipASp from S. pristinaespiralis was reported and deposited in the protein data base PDB as 5QZJ (Ying & Chen, 2016). Also, the biochemical activity of PipASp has previously been studied (Gatto et al., 2006; Tsotsou & Barbirato, 2007; Byun et al., 2015). The crystal structures of Pip from A. friuliensis as well as FkbL from S. tsukubaensis remained unknown so far. In order to compare the structural composition of PipASp, PipAf and FkbL, first homology modelling of PipAf and FkbL was performed. The structures of PipAf and FkbL were generated based on the template with the best query cover and known crystal structure–PipASp.

Although the PipASp structure has been deposited in the PDB database, its properties and key amino acids in the binding pocket have not been described. PipASp belongs to the µ-crystallin family and is related to ornithine cyclodeaminases (OCD). It is annotated as lysine decarboxylase WP_037775551.1, alongside with FkbL as WP_006350828.1 and other proteins including AF235504.1 in S. hygroscopicus (Wu et al., 2000) and AGP59511.1 in S. rapamycinicus (Baranasic et al., 2013). In contrast, PipAf is annotated in the NCBI database as OCD WP_023362365.1.

In order to identify possible key amino acids in the catalytic center of PipASp, we compared it with the structurally solved OCD from Pseudomonas putida PpOCD (Goodman at al., 2004) with PDB accession code 1X7D, using the Swiss-Model Structure Comparision tool and Swiss PDB-Viewer Magic Fit tool. PipASp consists, similar to described OCD, of a homodimeric fold whose subunits comprise two functional regions: a substrate-binding domain and a Rossmann fold that interacts with a dinucleotide positioned for re-hydride transfer (Supplementary Figure S4). In the OCD of P. putida, oligomerization results in a 14-stranded, closed β-barrel and each subunit contributes residues as shown in the Supplementary Figure S4) and Table 1. In the PipASp structure the β-barrel is similarly structured (Supplementary Figure S4, and Supplementary Figure S5; Table 1).

Based on the crystal structures of PpOCD and PipASp, the model structures of other LCDs PipAf and FkbL were generated through homology modelling. The structures of PipASp, PipAf and FkbL were overlaid; and the identified amino acid residues of PipASp were used to find the corresponding key amino acids in the catalytic center of PipAf and FkbL. Comparisons with PipASp demonstrated that in the lysine cyclodeaminase PipAf from A. friuliensis (Figure 5) the β-barrel is similarly but not identically structured (Supplementary Figure S5; Table 1). In the lysine cyclodeaminase FkbL from S. tsukubaensis the composition β-barrel differs from PipAf, but is almost identical to PipASp (Supplementary Figure S6; Table 1).

Our studies of the crystal structure of the lysine cyclodeaminase PipASp from S. pristinaespiralis as well as model structures of FkbL from S. tsukubaensis and PipAf from A. friuliensis revealed structural differences between PipAf and PipASt/FkbL. Key amino acids in PipAf and FkbL models were identified showing that PipAf features different amino acid residues in the catalytic domain (Figure 5; Table 1).

Homology modelling demonstrated that the composition of PipAf differs from PipASp and FkbL. In order to allow the engineering and specific design of PipAf for optimized FK506 production, we aimed to solve its crystal structure. PipAf was overexpressed as a His-tagged protein in E. coli BL21 and purified for further analysis. A monomer of PipAf consists of 339 residues and the purified enzyme appeared as a ∼36 kDa protein in SDS-page. Its oligomeric state from size exclusion chromatography was found to be dimeric as observed for many members of the µ-crystallin/OCD family, e.g. PpOCD (Goodman et al., 2004). PipAf crystals grew within few hours and diffracted up to 1.3 Å resulting in excellent crystallographic statistics (Supplementary Table S1). The enzyme consists of a homodimeric fold (Figure 6) whose subunits include two domains that function in the L-lysine and NAD+/H cofactor binding (substrate binding domain) as well as in oligomerisation of subunits. One molecule NADH binds per monomer via a canonical Rossman fold, consistent with previous reports for the structurally characterized PpOCD (Goodman et al., 2004; Dzurov et al., 2015).

A 7-stranded β-sheet domain forms on the one side an extended dimer interface of the PipAf LCD, on the other side it contributes most of the residues of the substrate binding pocket (Figure 6). The large interface between two monomer subunits with a surface of 2,996 Å² confers the stability of the dimer with -45 kcal/mol as calculated by the PISA server. Interestingly, both subunits contribute alternating hydrophobic residues (Phe49, Trp61, Pro63 and Tyr98) to the dimer interface and lock it similar to zipper-folds.

The model of PipAf was compared with the obtained crystal structure of PipAf. The analysis of amino acid residues of interest in the catalytic site of the enzyme revealed that residues Pro63 and Asp233 could be confirmed via crystallographic studies to be involved into the key functions of the catalytic domain of the enzyme.

In order to enhance the properties of PipAf we were aiming to engineer the enzyme by site-directed mutagenesis. In order to identify relevant amino acids involved in substrate binding, PpOCD crystal structure was compared with the obtained PipAf crystal structure. The substrate binding domain contributes conserved residues essential for conversion of L-lysine to L-pipecolate and directs them closely to NAD+. Like in PpOCD, series of the acidic and basic residues interacting with lysine are highly conserved in PipAf (Arg46, Glu60, Met62, Lys74, Tyr78, Ile91, Arg118, Asp233). The domain is composed of a varied α/ β fold involving a 7-stranded β-sheet packed against three α helices. Lysine binds in the active site running parallel to the nicotinamide ring. Tyrosine 78 (Tyr78) and isoleucine 91 (Ile91) in the PipAf structure substitute smaller, but also hydrophobic residues glycine and valine in PpOCD respectively. Such residues should allow the lysine binding (Figures 6, 7).

Our attempts of PipAf engineering require in-depth understanding of the enzyme-dinucleotide binding kinetics. The catalytic mechanism of PipAf includes the dinucleotide cofactor binding at first, followed by substrate binding in an ordered reaction mechanism (Supplementary Figure S7). All reported dinucleotide folds, also in PpOCD, bind the pyrophosphoryl moiety of the dinucleotide via a glycine-rich, 30 to 35 amino-acids long loop comprising of a hydrophobic core, a negatively charged residue at the C-terminus and a positively charged residue at the N-terminus (Kleiger & Eisenberg, 2002; Goodman et al., 2004). A glycine-rich phosphate-binding sequence GxGxxG/A/S (where x is any residue) was found in all members of the µ-cristalline family, including PpOCD (Goodman et al., 2004). Whereas most phosphate-binding motifs have the sequence GxGxxG/A, OCDs exhibit the sequence GxGxxS (Goodman et al., 2004). This explains why previous studies dismissed the existence of canonical dinucleotide binding sequences in OCDs (Kim et al., 1992). A precedent for use of the GxGxxS sequence was established in studies of dihydropyrimidine dehydrogenase, which uses the latter motif in FAD+ binding (Dobritzsch et al., 2001). The C-terminal part of PipAf lacks the coil-α6 segment reported to contribute some residues to NAD+/H recognition in OCD and thereby distinguish LCD from the rest of the μ-crystallin family (Goodman et al., 2004).

After purification by affinity chromatography and size exclusion chromatography PipAf exhibited an absorption at 340 nm corresponding to at least one molecule of NADH per dimer, if calculated with the extinction coefficient for NADH in aqueous solution. Crystallization was always performed with 1 mM NAD+, but each PipAf seems to contain NADH as cofactor as indicated by the non-planarity nicotinamide ring, which was clearly visible due to the high resolution. In analogy to the proposed mechanism of OCD the catalytic cycle of PipAf should start with a NAD+ oxidizing the αalpha-amino group of lysine. The resulting imine group is substituted by the epsilon-amino group of lysine and the circular imine is finally reduced to pipicolic acid, thereby regenerating the NAD+ cofactor. OCD also contained NADH, therefore we would like to propose in this study an alternative 2-step mechanism for cyclodeamination in which a direct substitution of the alpha-amino group by the hydride ion from NADH takes place which in turn is substituted by the epsilon-amino group, based on findings described previously by Min et al., 2018.

In order to optimize the one-step LCD-mediated synthesis of pipecolic acid through metabolic engineering we aimed to carry out site-directed mutagenesis based on our structural studies. The mutations included the exchange of amino acid residues in the substrate binding domain, potentially leading to enhanced substrate binding capacities, including binding pocket residues: Val58 by Leu58, Val58 by Ala58, Glu60 by Ala60, Glu60 by Gln60, Glu60 by Leu60 (Figure 8, red; Table 1). The replacement of Val58 by Leu58 or Ala58 would lead to the strengthening or attenuation of the formation of a bond to the lysine side chain, which could be beneficial for deprotonation. The replacement Glu60 by Ala60 or Leu60 might open the entrance to the binding pocket a little more for the substrate. The replacement Glu60 by Gln60 might result in the generation of another hydrogen bond. Another strategy was the exchange of amino acid residues in the ammonia leaving group side chain residue Asp233 by Asn233 (Figure 8, red; Table 1). Such substitution might lead to the generation of another hydrogen bond as well. Especially interesting should be the exchange of the lysine binding residue Ile91 by Val91. Ile at this place corresponds to the Val in FkbL. IIle binds hydrophobically the middle part of lysine, unlike Val. In order to avoid the reduction in activity due to this fact, Ile91 was replaced through Val. (Figure 8, light green).

The selected amino acids were exchanged by site-specific mutagenesis in the pET30/pip expression plasmid. Using specifically designed primers with mutated bases, the mutagenesis PCR was carried out. The degradation of the non-mutated template DNA was performed with the restriction enzyme DpnI. Afterwards, the competent E. coli Novablue strain was transformed. Plasmid-carrying clones were identified by selection for kanamycin and verifyed by sequencing.

Mutated versions of the pipAf gene (pip*Ile91Val, pip*Val58Leu, pip*Val58Ala) were cloned into the expression vector and introduced into E. coli Rosetta 2 (DE3) pLysS. The product of the lysine cyclodeaminase reaction was detected via thin layer chromatography (TLC). In order to investigate and compare the activity of the individual mutants of the PipAf enzyme, a qualitative detection system for pipecolate production was established. The native pip gene from A. friuliensis served as reference. Since the host E. coli Rosetta 2 (DE3) pLysS has no lysine cyclodeaminase, the strain is not able to synthesize pipecolic acid and was therefore used as a negative control.

TLC revealed that lysine could only be converted into pipecolic acid by the E.coli Rosetta strains carrying the plasmids pET30/pip*Ile91Val, pET30/pip*Val58Leu, pET30/pip*Val58Ala and the native pET30/pip with the native pipAf gene (Figure 9). In the mutated Pip variants Pip*Glu60Ala, Pip*Glu60Gln and Pip*Glu60Leu only a minimal pipecolic acid production was detected. In the case of the Pip* variant Pip*Asp233Asn, no pipecolic acid formation was observed. In the case of the Pip* variants Pip*Ile91Val and Pip*Val58Leu, the highest pipecolic acid formation was determined (Figure 9).

According to the crystal structure of PipAf, Val91 does not hydrophobically bind the middle part of lysine, which should lead to enhanced enzyme activity. Thus, the mutated variant of Pip*Ile91Val was selected for the introduction into the recombinant host. The plasmid carrying the mutation pET30/pip*Ile91Val was generated by site-directed mutagenesis through the introduction of the point mutation using specific primers. Ile in contrast to Val should bind the middle part of the lysine hydrophobically, because of the different behaviour of their side-chains. S. tsukubaensis with the previously introduced dapASt and lysCCg genes was used as host. The specific FK506 production was quantified using RP-HPLC and represented in mg/g cells and in percentage. While the S. tsukubaensis WT could achieve a FK506 production of 1.7 mg/g cells, the S. tsukubaensis strain with the introduced mutated variant of PipAf produced of 3.4 mg/g, which is about 2-fold higher compared to the WT strain (Figure 10). Again, lysine supplementation did not result in an FK506 yield increase in the engineered strain S. tsukubaensis pET30/pip*Ile91Val, which correlates with studies showing that the availability of nutrients like lysine can have an inhibitory effect on antibiotic production (van Wezel and McDowall, 2011).