Metabolic engineering of Corynebacterium glutamicum for de novo production of 3-hydroxycadaverine

Functionalization of amino acids and their derivatives opens up the possibility to produce novel compounds with additional functional groups, which can expand their application spectra. Hydroxylation of polyamide building blocks might allow crosslinking between the molecular chains by esteri ﬁ cation. Consequently, this can alter the functional properties of the resulting polymers. C. glutamicum represents a well ‐ known industrial workhorse and has been used extensively to produce lysine and lysine derivatives. These are used as building blocks for chemical and pharmaceutical applications. In this study, it was shown for the ﬁ rst time that C3 ‐ hydroxylated cadaverine can be produced de novo by a lysine overproducing C. glutamicum strain. The lysine hydroxylase from Flavobacterium johnsoniae is highly speci ﬁ c for its natural substrate lysine and, therefore, hydroxylation of lysine precedes decarboxylation of 4 ‐ hydroxylysine (4 ‐ HL) to 3 ‐ hydroxycadaverine (3 ‐ HC). For optimal precursor supply, various cultivation parameters were investigated identifying the iron concentration and pH as major effectors on 4 ‐ HL production, whereas the supply with the cosubstrate 2 ‐ oxoglutarate (2 ‐ OG) was suf ﬁ cient. Deletion of the gene coding for the lysine exporter LysE suggested that the exporter may also be involved in the export of the structurally similar 4 ‐ HL. With the optimised setting for 4 ‐ HL production, the pathway was extended towards 3 ‐ HC by decarboxylation. Three different genes coding for lysine/4 ‐ HL decarboxylases, LdcC and CadA from E. coli and DC Fj from F. johnsoniae, were expressed in the 4 ‐ HL producing strain and compared regarding 3 ‐ HC production. It was shown in a semi ‐ preparative biocatalysis that all three decarboxylases can accept 4 ‐ HL as substrate with varying ef ﬁ ciencies. In vivo, LdcC supported 3 ‐ HC production best with a ﬁ nal titer of 11 mM. To improve titers a fed ‐ batch cultivation in 1 L bioreactor scale was performed and the plasmid ‐ based overexpression of ldcC was induced after 24 h resulting in the highest titer of 8.6 g L ‐ 1 (74 mM) of 3 ‐ hydroxycadaverine reported up to now. Determination of the substrate specificity of different lysine decarboxylases Fermentative production of hydroxylated lysines via regiospecific C-H-hydroxylation using different KDOs Role of LysE for 4-hydroxylysine production


Introduction
Microbial enzymes are exceptionally attractive to be incorporated in chemical processes as they are sustainable and environmentally friendly (Sheldon and Woodley, 2018). One major advantage of these biocatalysts is their ability to catalyse interconversions of many functional groups with well-defined selectivity (chemo-, regio-, stereo-) ensuing reduced waste during synthesis and high atom economy. The recent developments in bioinformatics, synthetic biology, and computational biology enabled a deeper understanding of the enzymatic reactions and the establishment of effective biotransformation systems for industrial use (Bornscheuer et al., 2012;Sun et al., 2018). It is important to discover novel enzymes or unknown side activities of known enzymes to access new products, e.g. for the functionalization of non-activated C-H bonds, which are difficult for organic chemists (Bastard et al., 2018). One group of enzymes, which catalyse the formation of various C-heteroatom bonds, are the iron (II)/α-ketoglutarate dependent dioxygenases (KDO) (Hausinger, 2004;Wan et al., 2017), that e.g., hydroxylate C-H bonds (Dunham et al., 2018;Hausinger, 2004). These enzymes require the cofactor iron (II) and three substrates: molecular oxygen, 2-OG and a primary substrate. One oxygen atom is transferred to the primary substrate to yield the hydroxylated product, and the second oxygen is used for oxidation of 2-OG to succinate and carbon dioxide (Bastard et al., 2018;Martinez and Hausinger, 2015;Mitchell et al., 2017). Typical substrates for these metalloenzymes are amino acids, which are hydroxylated. KDOs are mainly involved in the biosynthesis of secondary metabolites and can be found in bacteria, where they hydroxylate the side chains of free amino acid (derivatives) or tether peptides in non-ribosomal peptide biosynthesis (Wu et al., 2016). The hydroxylated amino acids are highly attractive intermediates in pharmaceutical industries and find application as valuable chiral building blocks for fine chemistry (Jing et al., 2021;Ren and Fasan, 2021). KDOs that are active towards amino acids and their derivatives belong to the Clavaminate Synthase Like (CSL) family. They are highly substrate specific and exhibit high regio-and stereoselectivities (Bastard et al., 2018;Hara et al., 2017). Among the KDOs which act specifically on ʟ-lysine, the KDOs from Catenulispora acidiphila (Baud et al., 2014;Peters and Buller, 2019) and Kineococcus radiotolerans SRS30216 (Hara et al., 2017) yield ʟ-3-hydroxylysine, while the KDOs from Flavobacterium johnsoniae UW101 (Bastard et al., 2018;Baud et al., 2014) or Niastella koreensis (Wang et al., 2020) show different regioselectivity and form D-4-hydroxylysine. As the physiological function of hydroxylysine depends on the location of the hydroxyl group, the regio-and stereoselective synthesis is highly important. In nature, the most abundant isomer of hydroxylated lysine is D-5-hydroxylysine and it is found in a particular type of collagen peptide (Peters and Buller, 2019). It frequently occurs in the extracellular matrix of animal cells, where it stabilizes the collagen scaffold by subsequent O-glycosylation. Moreover, ʟ-lysine residues present in proteins can be hydroxylated as a posttranslational modification by lysyl hydroxylase (EC 1.14.11.4) (Turpeenniemi-Hujanen and Myllylä, 1984). 5-hydroxylysine (5-HL) can also be found in bacteria, like Staphylococcus aureus, where it is used as a cell-wall precursor instead of lysine (Smith et al., 1965). Moreover, regio-and stereoisomers are highly demanded as synthons for pharmaceutical agents. For example, C3-hydroxylation of ʟ-lysine provides the precursor for synthesis of the building block tambroline (X. , a precursor of the antibiotic tambromycin (Goering et al., 2016). In addition, ʟ-3-hydroxylysine (3-HL) is an intermediate in the synthesis of (-)-balanol in Verticillium balanoides, which is a potent protein kinase C inhibitor (Lampe et al., 1996). ʟand D-4-Hydroxylysine (4-HL) are also promising precursors for pharmaceutical agents, like functionalized piperidine-2-ones, which are highly versatile building blocks for the synthesis of many bioactive substances (Herbert et al., 2012). They can also be used for the synthesis of 4hydroxypipecolic acid, a constituent of certain cyclic peptide antibiotics; palinavir, a potent HIV protease inhibitor (Lamarre et al., 1997), or for the anti-cancer agent Glidobactin A (Amatuni and Renata, 2019).
Like lysine, one of the most industrially relevant amino acids, hydroxylysine could be used as a precursor for chemical synthesis. By decarboxylation of lysine, the C 5 -diamine polymer building block cadaverine, also known as 1,5-diaminopentane, can be obtained (Cheng et al., 2018). Using hydroxylysine as substrate would yield hydroxylated cadaverine, which could be incorporated as novel building block for polyamides with new properties. Additional hydroxyl groups are attractive in polymers, as they can undergo various reactions, e.g., esterification. Moreover, they are hydrophilic and can act as initiation sites for ring-opening polymerization of cyclic esters, thus enabling easy access to complex polymers (Gómez and Varela, 2007;Kakwere and Perrier, 2011). Orgueira et al. (2001) showed the production of a hydroxylated polymer derived from the building blocks cadaverine and D-2-hydroxyglutarate (Orgueira et al., 2001). Biomaterials, which are derived from polysaccharides showed promising applications in biomedical realms because of their good biocompatibility on the hydroxyl-enriched material surfaces (Yang and Yang, 2014). For the production of cadaverine several approaches have been established, including efficient whole-cell biocatalysis in E. coli (Kim et al., 2019). Furthermore, C. glutamicum has been engineered for in vivo production of cadaverine by overexpression of the genes coding for different pyridoxal phosphate (PLP) dependent decarboxylases from E. coli: CadA (Mimitsuka et al., 2007) and LdcC (Kind et al., 2011). The deletion of snaA coding for spermi(di)ne-N-acetyltransferase increased product titers by avoiding N-acetylation of cadaverine (Kind et al., 2010;Nguyen et al., 2015b). Further optimisation strategies included the application of synthetic promoter-based expression cassettes and integration of different ldcC variants in the genome yielding up to 125 g L -1 cadaverine Oh et al., 2015). C. glutamicum was further engineered to exploit starch, wheat sidestream concentrate hydrolysate (WSCH) and methanol as carbon sources for the production of cadaverine Leßmeier et al., 2015;Tateno et al., 2009).
In this study, a lysine overproducing C. glutamicum strain was chosen to extend the lysine biosynthesis pathway to hydroxylated lysine by overexpression of codon-optimised genes coding for KDOs with different regiospecificity to yield either 3-HL (KDO Kr encoded by Krad_3958 from K. radiotolerans) or 4-HL (KDO Fj encoded by Fjoh_3169 from F. johnsoniae). To improve production, the effects of cofactor and cosubstrate supply as well as substrate/product export and pH alterations were tested. Since hydroxylated lysines can be decarboxylated in vitro to hydroxylated cadaverines by different decarboxylases as demonstrated by Bastard et al. (2018), the substrate specificity of the lysine decarboxylases LdcC (Yamamoto et al., 1997) and CadA (Sabo et al., 1974) from E. coli and the predicted 4-hydroxylysine decarboxylase DC Fj (Fjoh_3171) (Bastard et al., 2018) were investigated with regard to produce 3hydroxycadaverine (3-HC) by decarboxylation of 4-HL (Fig. 1). Additionally, the inverse route was explored with decarboxylation of lysine to cadaverine prior to hydroxylation of cadaverine to yield 3-hydroxycadaverine.

Material and methods
Microorganisms and cultivation conditions C. glutamicum WT strains were cultivated in lysogeny broth (LB) (Bertani, 1951) supplemented with 25 μg mL −1 kanamycin. C. glutamicum GRLys1 derived strains were cultivated in brain heart infusion with 0.5 M sorbitol (BHIS), supplemented with 25 μg mL −1 kanamycin, 100 μg mL −1 spectinomycin or 5 μg mL −1 tetracycline, when appropriate. All bacterial strains and plasmids are listed in Tables 1  and 2. Growth experiments with C. glutamicum were performed in CGXII minimal medium at pH 7.0 (Eggeling and Bott, 2004) supplemented with 40 g L -1 glucose as sole carbon source and induced with 1 mM IPTG if necessary. The amount of iron and the pH were adjusted to 1.04 mM and 6.5, respectively, as indicated in the result section. Overnight cultures were harvested and washed twice in TN buffer (50 mM Tris-HCl, 50 mM NaCl, pH 6.3) before inoculation to an initial OD 600 of 1. The cultivations in the BioLector microcultivation system (m2p-labs, Baesweiler, Germany) were performed in 3.2 mL Flower-Plates at 30°C, 1100 rpm and a filling volume of 1000 μL.

Molecular biology methods
Isolation of genomic DNA and classical methods which include plasmid isolation, molecular cloning and heat-shock transformation of E. coli and electroporation of C. glutamicum were performed as described previously (Eikmanns et al., 1994;Simon et al., 1983). Fig. 1. Schematic overview of the predicted route towards 3-hydroxycadaverine from lysine. Enzymes are depicted next to their reaction. Heterologous proteins are boxed. Deletion of coding gene is marked by a red cross. DC: Decarboxylase from different organisms (lysine decarboxylase from E. coli MG1655 (LdcC, CadA), PLP-dependent decarboxylase from Flavobacterium johnsoniae UW101 (DC Fj )); KDO Fj : α-ketoglutarate dependent dioxygenase/lysine 4-hydroxylase from Flavobacterium johnsoniae UW101; SnaA: spermi(di)ne-N-acetyltransferase from C. glutamicum WT; LysE: lysine/4-HL exporter; CgmA: cadaverine/3-HC exporter, PTS: phosphotransferase system). Question marks display transport systems that may be involved for export of 4-HL and 3-HC, respectively, based on genetic evidence, but in the absence of biochemical evidence. Dashed lines represent multiple reactions. 2-OG: 2-oxoglutarate; SA: succinic acid; TCA: tricarboxylic acid cycle; AR: anaplerotic reaction. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) ALLin HiFi DNA Polymerase (HighQu, Kraichtal, Germany) was used to amplify DNA sequences with plasmid or genomic DNA as template. The oligonucleotides which were used as primers in this study are listed in Table 3. The genes cadA and ldcC were amplified from genomic DNA of E. coli MG1655 and Fjoh_3171 from Flavobacterium johnsoniae UW101 (DSMZ 2064). A standard ribosomal binding site (RBS 5 0 -GAAAGGAGGCCCTTCAG-3 0 ) was introduced via the respective forward primer. The genes Krad_3958 from Kineococcus radiotolerans SRS30216 and Fjoh_3169 from Flavobacterium johnsoniae UW101 were codon-optimised (Bio-Rad Laboratories, Hercules, FL, USA) and an optimised RBS was introduced (Salis, 2011). For the variant odhI T14A the vector pK19mobsacB-odhI T14A was used as a template (Nguyen et al., 2015a). For construction of the overexpression vectors, the respective plasmid was digested with BamHI and assembled with the amplified DNA fragments using Gibson Assembly (Gibson et al., 2009).

Quantification of amino acids, carbohydrates and organic acids by HPLC
The quantification of extracellular amino acids, their derivatives and carbohydrates in the cultivation medium was performed with a high-performance liquid chromatography system (1200 series, Agilent Technologies Deutschland GmbH, Böblingen, Germany). After centrifugation of 1 mL of cell cultures at 20238 g for 10 min the supernatant was stored at −20°C prior to analysis. Analysis of amino acids, diamines and ω-amino acids was performed by an automatic pre-column derivatization with ortho-phthaldialdehyde (OPA) and separated on a reversed phase HPLC using a a pre-column (LiChrospher 100 RP18 EC-5μ (40 × 4 mm), CS Chromatographie Service GmbH, Langerwehe, Germany) and a main column (LiChrospher 100 RP18 EC-5μ (125 × 4 mm), CS Chromatographie Service GmbH). A mobile phase of buffer A (0.25% (v/v) sodium acetate, pH 6.0) and buffer B (methanol) was used. The following gradient was applied: 0-1 min 20 % B (0.7 mL min −1 ), 1-11 min a linear gradient of B from 20% to 85% (1.2 mL min −1 ), 12-13 min a linear gradient of B from 85% to 20% (1.2 mL min −1 ). To quantify amino acids, diamines and ωamino acids, the standards of ornithine, lysine, 5-HL, putrescine, cadaverine, GABA and 5AVA were analysed in a range from 50 to 800 µM. The peak area was normalised to the internal standard of 100 µM asparagine (Schneider and Wendisch, 2010) by division of the respective standard peak area by the area of the internal standard.
This study This study This study  The area quotient was plotted against the concentration of the respective standard. The slope of the calibration curve was used to quantify the respective substance. As C3/C4-hydroxylated lysine and C2/C3hydroxylated cadaverine are not commercially available, 3-HL/4-HL/4,5-DHL and 2-HC/3-HC were quantified as equivalents. To determine 3-HL, 4-HL and 4,5-DHL concentrations, the slope of 5-HL standard calibration curve was used, and concentrations were given as 5-HL equivalents. To determine 2-HC and 3-HC, the slope of cadaverine standard calibration curve was used, and concentrations were given as cadaverine equivalents. Detection of the fluorescent derivatives was carried out with a fluorescence detector (FLD G1321A, 1200 series, Agilent Technologies, Waldbronn, Germany) with an excitation wavelength of 230 nm and an emission wavelength of 450 nm. Glucose and 2-oxoglutarate concentrations were measured with an anion exchange column (Aminex, 300 mm × 8 mm, 10 μm particle size, 25 Å pore diameter, CS Chromatographie Service GmbH, Langerwehe, Germany) under isocratic conditions (5 mM H 2 SO 4 ) at 60°C with a flow of 0.8 mL min −1 as described previously (Schneider et al., 2011). The substances were detected with a refractive index detector (RID G1362A, 1200 series, Agilent Technologies, Waldbronn, Germany).

Determination of substrate specificity via biotransformation
Pellets from C. glutamicum strains WT (pVWEx1), WT (pVWEx1-ldcC), WT (pVWEx1-cadA), WT (pVWEx1-DC Fj ), WT (pVWEx4) and WT (pVWEx4-KDO Fj ) were obtained from cultivations in 50 mL CGXII minimal medium supplemented with 20 g L -1 glucose, 1 mM IPTG and 25 µg mL -1 kanamycin. The pellets were washed twice in 20 mL 50 mM HEPES buffer (pH 7.5) and centrifuged for 10 min at 4000 rpm and 4°C and resuspended in 2 mL of 50 mM HEPES buffer (pH 7.5). Cells were disrupted by sonication (cycle. 0.5, amplitude of 55%) on ice for 9 min. To remove cells debris, centrifugation was performed for 1 h at 20238 g and 4°C. The crude extract was used for the biocatalytic reaction. Protein concentrations were determined with the Bradford assay kit (Bio-Rad Laboratories, Hercules, CA, United States) using BSA (bovine serum albumin) as standard.
The determination of the biocatalytic activity of the hydroxylase was performed as described before (Baud et al., 2017(Baud et al., , 2014. 1.5 mL reaction mix contained 50 mM HEPES buffer (pH 7.5), 15 mM 2oxoglutarate, 2.5 mM ascorbate, 1 mM ammonium iron (II) sulfate and 10 mM of the respective substrate (lysine, 5-HL, ornithine, putrescine, cadaverine, GABA, 5AVA). The reaction was started by addition of 2 mg mL −1 protein. The reaction mixture was stirred at 30°C in Duetz microcultivation plates (Kuhner Shaker GmbH, Herzogenrath, Germany) with culture volumes of 1.5 mL at 300 rpm in an Ecotron ET25-TA-RC (Infors HT, Einsbach, Germany) and plate sandwich covers for low evaporation (1.2 mm hole diameter) were used. Samples were taken after 24 h. Proteins from the crude extract were precipitated with trichloroacetic acid (TCA). 50 µL 100% (w/v) TCA were added to 500 µL of sample and incubated for 30 min on ice (Koontz, 2014). After centrifugation at 20238 g for 10 min, the supernatant was neutralized with NaOH and stored at − 20°C prior to analysis. The product formation was determined by HPLC.
To identify the substrate specificity of the different decarboxylases the following assay was performed: 1.5 mL reaction mix contained 50 mM HEPES buffer (pH 7.5), 1 mM PLP, 1 mM DTT and 10 mM of the respective substrate (lysine, 5-HL, 4-HL from fermentation broth) (Baud et al., 2017). The reaction was started by addition of 1 mg mL −1 protein. Incubation of the reaction mix, sampling, sample treatment and analysis were performed as described for the KDO assay.

Fermentative production of 3-HC
A baffled bioreactor with a total volume of 3.6 L was used for scaleup experiments (KLF, Bioengineering AG, Switzerland). Two sixbladed Rushton turbines were placed on the stirrer axis with a distance from the bottom of the reactor of 6 and 12 cm. The stirrer to reactor diameter ratio was 0.39. The dissolved oxygen concentration in the batch phase was kept at a minimum of 30% by stepwise increases of the stirrer rate. A constant airflow of 0.75 L min −1 was maintained from the bottom through a sparger, corresponding to an aeration rate of 0.75 vvm. The pH was kept constant at 6.50 ± 0.05 by automatic addition of phosphoric acid (10% (v/v)) and ammonia (25 % (w/v)). The temperature was maintained at 30°C. To prevent foaming 0.6 mL L -1 of the antifoam agent AF204 (Sigma Aldrich, Darmstadt, Germany) was added, and a mechanical foam breaker was present to serve as an additional foam control. The fermentation was performed with a head space overpressure of 0.5 bar. The initial working volume of 1 L was inoculated to an OD 600 of 1.5 from a shake flask pre-culture in CGXII minimal medium (pH 6.5) supplemented with 40 g L -1 glucose and 1.5 mM FeSO 4 . Samples were collected by an autosampler and cooled down to 4°C until further use. The feed consisted of 400 g L -1 glucose, 0.75 g L -1 MgSO 4 ·7H 2 O and 1.5 mM FeSO 4 (ρ = 1 .15 kg m −3 ). The feed was applied at 1.2 mL min −1 as long as the pO 2 surpassed 60%. If the feed-pump was constantly active for more than 1 min, the feed was halted to prevent overfeeding. The plasmid-based overexpression of ldcC was induced with 1 mM IPTG after 24 h.

Determination of the substrate spectra of the lysine-4-hydroxylase KDO Fj
Hydroxylases are known for their high substrate specificity. Therefore, it is essential to test if non-natural substrates like cadaverine can be accepted. It was shown before that the lysine-4-hydroxylase KDO Fj can accept 5-HL and convert it to 4,5-dihydroxylysine (4,5-DHL) with a low efficiency of 15 % conversion next to its natural substrate lysine (65 % conversion) (Baud et al., 2014), but not ornithine. In this study, the diamines cadaverine and putrescine as well as the ω-amino acids γ- Table 3 Oligonucleotides used as primers in this study (RBS in bold, overlaps in italic).

Primer
Sequence ( Prell et al. Current Research in Biotechnology 4 (2022) 32-46 aminobutyrate (GABA) and 5-aminovalerate (5AVA) were tested as well. Indeed, it could be confirmed that 79% of the added lysine was converted to 4-hydroxylysine using crude extract of C. glutamicum WT (pVWEx4-KDO Fj ) within 24 h (Table 4). Moreover, 4,5-DHL could be produced from 5-HL. However, none of the other substrates was hydroxylated (Table 4, Figure S1, Table S1). Thus, as cadaverine was not hydroxylated by KDO Fj , conversion of lysine to 3-HC had to occur in a reaction sequence with lysine hydroxylation occuring before 4-HL decarboxylation.
Determination of the substrate specificity of different lysine decarboxylases The capability of different PLP-dependent decarboxylases to decarboxylate lysine and hydroxylated lysines was tested. It was described before that the decarboxylase from Flavobacterium johnsoniae DC Fj decarboxylates 4-HL with high efficiency, but not 5-HL (Baud et al., 2017). Crude extract of C. glutamicum WT (pVWEx1-DC Fj ) converted 4-HL (18 % of 10 mM), whereas neither 5-HL nor lysine were decarboxylated within 24 h (Table 5, Figure S2, Table S1). By contrast, CadA and LdcC from E. coli fully converted lysine and 4-HL within 24 h (Table 5) yielding cadaverine and 3-HC, respectively. Only about 50% of 10 mM 5-HL was converted to 2-hydroxycadaverine by LdcC and CadA (Table 5).

Fermentative production of hydroxylated lysines via regiospecific C-Hhydroxylation using different KDOs
Lysine biosynthesis in C. glutamicum was extended by regiospecific hydroxylases (KDOs) to produce hydroxylated lysines fermentatively. C. glutamicum strain GSL was chosen as base strain, as this lysine overproducer has been successfully engineered to produce lysine-derived compounds Prell et al., 2021). By addition to growth medium, 5-HL was shown neither to be toxic to C. glutamicum GSL nor to be catabolized (data not shown). Genes encoding two KDOs showing different regioselectivity were tested. To obtain 3-HL a codon-optimised version of Krad_3958 (KDO Kr ) from K. radiotolerans was heterologously overexpressed in GSL, whereas for 4-HL production a codon-optimised version of Fjoh_3169 (KDO Fj ) was used. In shake flask cultivation, 3 ± 1 mM 3-HL was accumulated by GSL (pVWEx4-KDO Kr ) with concomoitant production of 35 ± 3 mM lysine (Fig. 2B). In contrast, overexpression of KDO Fj led to a higher 4-HL concentration (23 ± 1 mM) and a lower lysine (15 ± 2 mM) concentration (Fig. 2B). Biomass formation of both producers was comparable to the empty vector carrying strain GSL (pVWEx4), whereas the growth rate was significantly reduced in the 4-HL producer (µ= 0.2 0 ± 0.00 h −1 ) compared to the control strain (µ= 0.25 ± 0.02 h −1 )and the 3-HL producer ( Fig. 2A). Since 4-HL production led to higher titers than 3-HL production, production of 4-HL was further optimised and the strains GSL (pVWEx4) and GSL (pVWEx4-KDO Fj ) were named Lys1 and HLys1, respectively.

Role of LysE for 4-hydroxylysine production
Transport engineering is a promising approach for optimising producer strains (Krämer, 2002;. For example, overexpression of the gene coding for the lysine exporter LysE accelerated lysine production (Gunji and Yasueda, 2006), while its deletion improved production of the lysine-derived ʟ-pipecolic acid by minimizing loss of lysine as precursor of L-pipecolic acid . Strains unable to export lysine while oversynthesizing it suffer from growth perturbation (Pérez-García et al., 2017) as up to 1 M lysine may accumulate intracellularly as consequence of such metabolic imbalance (Vrljic et al., 1996). Growth perturbation may be overcome by conversion of lysine to another compound that is exported independently of LysE, e.g. conversion of lysine to ʟpipecolic acid . To test if the deletion of lysE might affect 4-HL production, strain GSLE2 which lacks lysE was used. As expected, the absence of LysE severely perturbed growth of the lysine producer GSLE2 (pVWEx4) (pEKEx3) (=Lys2) ( Table 6). Growth of the 4-HL producer GSLE2 (pVWEx4-KDO Fj ) (pEKEx3) (=HLys2) was severely perturbed under the tested conditions as well (Table 6). When both strains were complemented by plasmid-based overexpression of lysE, (strains named Lys3 and HLys3) growth, accumulation of lysine (16 ± 1 mM) as well as of 4-HL (2 ± 0 mM) in the supernatant were restored (Table 6). Thus, unlike conversion of lysine to L-pipecolic acid in the absence of LysE (Pérez-García et al., 2017), conversion of lysine to 4-HL did not restore growth in the absence of LysE (Table 6). One may speculate that LysE is not only exporting lysine, arginine and citrulline out of the C. glutamicum cell (Lubitz et al., 2016), but may also be involved in export of 4-HL.

Effect of increased supply of 2-oxoglutarate as cosubstrate on 4-HL production
It was demonstrated before that sufficient supply of the cosubstrate 2-OG is crucial to facilitate hydroxylation of amino acids catalysed by KDOs. By dynamic modulation of the 2-oxoglutate dehydrogenase complex (ODHC) in C. glutamicum production of 4-hydroxyisoleucine (C.  and trans-4-hydroxyproline (Long et al., 2020) were improved. In this study, the impact of extracellularly added 2-OG on 4-HL production in HLys1 was investigated first. Increased biomass formation and a decreased growth rate were only observed at 60 mM 2-OG (Fig. 3A). Notably, 2-OG concentrations remained stable when 30 mM or less 2-OG were added, but 2-OG appeared to be catabolised at 45 mM and 60 mM (Fig. 3A). The addition of 30 mM 2-OG or more only resulted in a minor increase from 20 ± 0 mM up to 22 ± 0 mM 4-HL. Interestingly, the lysine concentrations gradually decreased from 14 ± 0 mM to 7 ± 1 mM with increasing 2-OG concentrations up to 60 mM. Therefore, addition of 2-OG hardly affected 4-HL titers but was beneficial as it decreased production of lysine as major by-product (Fig. 3B).

Effect of increased supply of iron as cofactor on 4-HL production
The influence of the KDO cofactor iron (II) was investigated since it was shown before that elevated concentrations of iron (II) in the cultivation medium improved production of ʟ-2-hydroxyglutarate by C. glutamicum significantly which involved hydroxylation of glutarate . Therefore, up to 2.5 mM iron sulfate were added to the CGXII minimal medium, which already contained 37 µM Fe 2+ . Addition of iron sulfate to the growth medium slowed growth of strains Lys1 and HLys1 ( Fig. 4A; K i = 1.9 mM for Lys1 and K i = 1.3 mM for HLys1). Lysine production by Lys1 was increased by addition of 0.5 mM or more iron sulfate (Fig. 4B). Of note, gradually increasing the addition of iron sulfate to the culture broth of HLys1 reduced lysine accumulation and increased 4-HL production proportionally (Fig. 4D). When 1.0 mM Fe 2+ were added, HLys1 produced 29 ± 1 mM 4-HL, but only 6 ± 0 mM lysine in comparison to standard cultivation conditions (20 ± 0 mM 4-HL, 14 ± 0 mM lysine) (Fig. 4D). Since at 2.5 mM Fe 2+ only slightly more 4-HL titer (32 ± 1 mM) were produced, but the growth rate dropped to 0.07 ± 0.00 h −1 , addition of 1.0 mM Fe 2+ was chosen for the following experiments since the growth rate was still 0.11 ± 0.00 h −1 .

Adaption of extracellular pH for improved precursor supply and 4-HL production
For KDO Fj various pH values ranging from 6.0 (Hara et al., 2017) to 7.5 (Baud et al., 2014) were described as optimal condition for the enzymatic reaction of the purified protein. Although pH homeostasis of C. glutamicum is effective in this pH range (Jakob et al., 2007), the influence of the external pH on lysine and 4-HL production was investigated. Growth of HLys1 was affected stronger at pH 6.5 than that of Lys1 (Fig. 5A, C). Lysine production by Lys1 was decreased from pH 6.5 (65 ± 5 mM) to pH 7.5 (47 ± 3 mM), while lysine production by HLys1 was affected less by the medium pH (Fig. 5B, D). Notably, 4-HL production was maximal at pH 6.5 with a titer of 54 ± 4 mM and only 9 ± 1 mM lysine accumulating as a byproduct (Fig. 5D). Although slow growth was observed at pH 6.5, this medium pH supported the highest 4-HL titers with the least accumulation of lysine as by-product. Thus, pH 6.5 was chosen as standard for the following experiments.
Hydroxylation of lysine to 4-HL and subsequent decarboxylation of 3-HC competes with lysine decarboxylation and export of cadaverine as by-product. Thus, cgmA coding for the cadaverine exporter was Fig. 3. Effect of 2-oxoglutarate on 4-HL production. Effect of extracellularly added 2-OG on (A) OD 600 (black squares), and maximum growth rates (black traingles) and (B) production of lysine (blue circles), 4-HL (turquoise squares), glutamate (yellow diamonds) and 2-OG (brown triangles) by HLys1. (C) Growth and (D) production of the lysine producer Lys4 (blue squares) and the 4-HL producer HLys4 (turquoise triangles) in comparison to the strains Lys5 (dark brown empty squares) and HLys5 (brown empty triangles) overexpressing odhI T14A .The strains were cultivated in the Biolector microcultivation system in CGXII minimal medium supplemented with 40 g L -1 glucose and 1 mM IPTG. For the cultivation of HLys1 (A, B) increasing 2-oxoglutarate concentrations (0, 15, 30, 45, 60 mM) were added. Supernatants were analysed after 48 h. Values and error bars represent means and standard deviations (n = 3 cultivations). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) deleted. As consequence, export of cadaverine was almost completely abolished. Moreover, 3-HC was no longer exported in HCad1ΔcgmA and only lysine (46 ± 2 mM) and 4-HL (17 ± 2 mM) were detected in the supernatant (Fig. 6B). These results indicated that CgmA might also be the exporter for 3-HC. Thus, deletion of cgmA could not be used to avoid accumulation of cadaverine during 3-HC production.

Fermentative production of 3-HC in reactor scale
A 1 L bioreactor cultivation was performed to test if 3-HC production is stable and if operation in the fed-batch mode provides a means to reduce cadaverine formation as by-product. During the batch phase, the aeration rate was kept at 0.75 vvm since this supported 4-HL production using KDO Fj best (data not shown). After 21.5 h the feed started and 1 mM IPTG was added to induce the expression of ldcC after 24 h. The cells grew in the batch phase with a growth rate of 0.13 h −1 to an OD 600 of 29 (Fig. 7). In the feed phase, a maximum OD 600 of 114 was reached after 42 h. After that, the rDOS steadily rose up to 85%, and no more feed was applied. After 44.5 h nitrogen was added manually using the base pump. The decrease in the rDOS led to the addition of more feed solution indicating that nitrogen might be the limiting factor. Even though more feed was applied, no cell growth and further product accumulation could be observed. After 42 h, 74 mM 3-HC (8.8 g L -1 , corresponding to 11.4 g L -1 when normalized to the initial volume of 1 L) was produced with a volumetric productivity of 1.55 g L -1 h −1 (corresponding to 2.0 g L -1 h −1 when normalized to the initial volume of 1 L) and a product yield on biomass of 0.31 g per g CDW (Fig. 7). The product yield on substrate was low (0.07 g g −1 ) as cadaverine accumulated as main by-product (390 mM; 39.8 g L -1 ) besides 4-HL (25 mM; 4.1 g L -1 ). Taken together, compared to the microscale cultivation (11 mM; 1.3 g L -1 , 0.44 g L -1 h −1 , 0.03 g g −1 , Fig. 6), the bioreactor fed-batch cultivation enabled a 7- Fig. 4. Effect of iron (II) on (A,C) OD 600 (black squares), maximum growth rate (black triangles) and (B,D) lysine (blue circles)/4-HL (turquoise squares) production in Lys1 and HLys1. The strains were cultivated in the Biolector microcultivation system in CGXII minimal medium supplemented with 40 g L -1 glucose, 1 mM IPTG and increasing iron (II) concentrations (0.04, 0.54, 1.04, 1.54, 2.04, 2.54 mM). Supernatants were analysed after 48 h. Values and error bars represent means and standard deviations (n = 3 cultivations). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) fold higher 3-HC titer, a 3.5-fold higher volumetric productivity, and more than 2-fold higher product yield on substrate.

Discussion
In this study, C. glutamicum was engineered to produce 4-HL and 3-HC by fermentation. Lysine biosynthesis was extended first by lysine-4-hydroxylase to yield 4-HL and second by a decarboxylase (E. coli LdcC > E. coli CadA > DC Fj from F. johnsoniae) for 3-HC production.
To reduce formation of cadaverine as by-product, two-phasic fedbatch cultivation in which KDO Fj was expressed constitutively, whereas expression of ldcC was induced when the feed was started, resulted in production of 8.8 g L -1 3-HC (Fig. 7).
The observation that decarboxylation of cadaverine and 4-HL are comparable in vitro (Table 5), while more cadaverine, but much less 3-HC accumulated in vivo (Fig. 6) points to the conclusion that lysine hydroxylation is the bottleneck of 3-HC production. This may be explained by cadaverine being derived from lysine by direct decarboxylation, whereas 3-HC formation depends on two consecutive reactions: lysine hydroxylase followed by decarboxylase. Thus, there is a need to identify more efficient lysine-4-hydroxylases. The lysine-4hydroxylase KDO Fj accepts lysine and with lower affinities 3-HL and 5-HL with low efficiency as substrates in vitro, but not ornithine (Baud et al., 2014;Hara et al., 2017). Hydroxylation of simple (di)amines such as cadaverine is not possible, likely because a conserved arginine residue interacts with the carboxylate group of lysine via a salt bridge (Baud et al., 2014). The α-amino group of lysine is directly involved in the lid closure via an H-bond with Asn232, which may be relevant for substrate binding (Strieker et al., 2007), i.e. the lid is in its opened-form in the absence of substrates, while it is in closed form when lysine is bound (Bastard et al., 2018). Our finding that Fig. 5. Effect of pH on lysine and 4-HL production. Comparison of OD 600 (black squares), maximum growth rate (black triangles) and production of lysine (blue circles) and 4-HL (turquoise squares) in Lys1 (A, B) and HLys1 (C, D). The strains Lys1 and HLys1 were cultivated in the Biolector microcultivation system in CGXII minimal medium at different pH (6.5, 7.0, 7.5, 8.0) supplemented with 40 g L -1 glucose, 1.04 mM FeSO 4 and 1 mM IPTG. Supernatants were analysed after 48 h. Values and error bars represent means and standard deviations (n = 3 cultivations). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) cadaverine is a major by-product of 3-HC production is commensurate with the inability of KDO Fj to hydroxylate cadaverine to 3-HC. Identification and use of KDO enzymes hydroxylating also cadaverine may help increas 3-HC production.
Hydroxylation by KDOs is associated with loss of carbon as carbon dioxide, which negatively impacts the carbon efficiency of the process in addition to the loss of carbon dioxide by decarboxylation of 4-HL. The use of enzymes, which are capable to perform C-Hhydroxylation without using 2-OG as cosubstrate is desirable, such as monooxygenases (EC 1.14.16), which use pteridines as cosubstrates to regiospecifically hydroxylate aromatic amino acids (Pey et al., 2006;Zhang et al., 2006Zhang et al., , 2006. Comparable to the KDOs, they are highly specific for their natural and closely related substrates. The tyrosine 3-monooxygenase from Homo sapiens can accept next to tyrosine with lower efficiency structurally related amino acids like phenylalanine and tryptophan (Roberts and Fitzpatrick, 2013). Alternatively, NAD(P)H/FADH 2 -dependent monooxygenases (EC 1.14.13.) use reduction equivalents as cofactors, and form the N-hydroxylated product and water. One example is the L-lysine N6-monooxygenase, which converts lysine to N6-hydroxy-L-lysine. (Dick et al., 2002). Fig. 7. 3-HC production by C. glutamicum HCad1 operated in fed-batch mode. HCad1 was cultivated in 1 L CGXII minimal medium in fed-batch mode over 48 h, containing 40 g L -1 glucose from the batch medium and 95 g L -1 glucose from the feeding solution. (A) OD 600 is shown in black squares, glucose concentration (g L -1 ) is plotted as pink empty triangles, feed solution (mL) is depicted as pink line, the relative dissolved oxygen saturation (rDOS, %) is indicated in light blue and the stirrer frequency (rpm) is shown as black line. (B) 3-HC production is indicated in light blue diamonds (mM), OD 600 is shown in black squares, lysine in dark blue circles (mM), 4-HL in turquoise squares (mM), and cadaverine concentration (mM) in grey triangles. Cultivation was performed at 30°C, 0.75 vvm and a constant pH 6.5 regulated with 10 % (v/v) H 3 PO 4 and 25% (w/v) ammonia. The pO2 was kept above 30% by a stepwise increase in stirrer rate. An overpressure of 0.5 bar was applied. 0.6 mL L -1 of antifoam agent AF204 (Sigma Aldrich, Taufkirchen, Germany) was added to the medium manually before inoculation. Plasmidbased overexpression of ldcC was induced after 24 h with 1 mM IPTG. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) KDO enzymes require iron and increasing the iron concentration in the growth medium improved production of hydroxylated glutarate involving CsiD from P. putida  and, as described here, of hydroxylated lysine involving KDO Fj (Fig. 2). Increasing the medium iron concentration may positively affect iron cluster containing TCA cycle enzymes such as aconitase or succinate dehydrogenase although the standard iron concentration in CGXII medium is sufficient for growth and does not trigger stress response in C. glutamicum (Wennerhold et al., 2005). KDO enzymes depend on 2-OG as cosubstrate. However, increasing provision of 2-OG is difficult since it is part of the TCA cycle. Replenishing the TCA cycle by anaplerotic reactions could help to provide more 2-OG for the KDO enzyme as has been shown for increased supply of oxaloacetate for lysine production and 2-OG for glutamate production by C. glutamicum (Peters-Wendisch et al., 2001).
Transport engineering, e.g., by overexpression of the endogenous thrE gene improved threonine production in C. glutamicum (Simic et al., 2002), while overexpression of lysE increased arginine production in Corynebacterium crenatum (Xu et al., 2013). It is known that CgmA is involved in cadaverine export and cgmA can be overexpressed for improved cadaverine production (Kind et al., 2011;Nguyen et al., 2015a). As it was demonstrated that deletion of cgmA also abolished 3-HC accumulation (Fig. 6B), CgmA likely functions as export system for 3-HC. In the case of 4-HL and lysine, indirect evidence may indicate that LysE is involved in 4-HL export. In the absence of LysE, high intracellular lysine concentrations (e.g. due to dipeptide feeding or lysine overproduction) drastically perturb growth of C. glutamicum (Vrljic et al., 1996), which can be avoided by conversion of lysine, e.g., to pipecolic acid, which is exported independently of LysE . Conversion of lysine to 4-HL did not abolish the growth perturbation observed in the 4-HL producer HLys2 that lacks LysE (Table 6). Overexpression of lysE may lead to loss of lysine as precursor of 4-HL, thus, decreasing 4-HL production. As alternative, lysine re-uptake into the cell by overexpression of the gene coding for the lysine importer LysI (Seep-Feldhaus et al., 1991) may improve 4-HL production under the hitherto unknown condition that LysI does not accept 4-HL.
Different metabolic engineering strategies were applied to increase provision of 2-OG as cosubstrate of KDOs (Jing et al., 2021). Smirnov et al. (2010) blocked the TCA cycle in E. coli by several deletions to avoid conversion from 2-OG to succinate and to couple KDO activity, which provides succinate from 2-OG besides hydroxylating the primary substrate (in this case isoleucine dioxygenase was used) (Smirnov et al., 2010). In a comparable approach, a tunable circuit for dynamic attenuation of ODHC activity was adopted to enhance the flux towards 2-OG and consequently trans-4-hydroxyproline production (Long et al., 2020). In this study, addition of extracellular 2-OG decreased the lysine titer but did not increase the 4-HL titer. Increasing repression of ODHC by OdhI T14A led to 4.4-fold more accumulation of 2-OG in Lys5, but at the expense of lysine production as its biosynthesis is derived from oxalacetate, an intermediate of the TCA cycle (Georgi et al., 2005;Schrumpf et al., 1991). Moreover, glutamate accumulated, while 4-HL was not improved. Possibly, withdrawal of oxaloacetate for lysine production is more important than provision of 2-OG for KDO.
Activities of KDOs respond to the intracellular pH. Lowering the external pH of the cultivation media to 6.5 resulted in increased production of lysine and 4-HL, while perturbing growth. C. glutamicum maintains pH homeostasis in a medium pH range from 6.0 to 9.0 (Täuber et al., 2021). Pleiotropic effects of changing the medium pH (i.a., iron starvation, protein folding, and stabilization) (Martín-Galiano et al., 2005) preclude interpretation of these effects of 4-HL production.
The inducible and constitutive lysine decarboxylases, CadA (Kanjee et al., 2011;Sabo et al., 1974) and LdcC (Yamamoto et al., 1997) from E. coli were tested with regard to decarboxylation of 4-and 5-hydroxylysine as alternative substrates. CadA is highly efficient, but the enzyme has its optimum at acidic pH (pH 5.0-6.0) and is rapidly inactivated at higher pH (>8.0) (Lemonnier and Lane, 1998) and inhibited at high concentrations of lysine  or cadaverine (Sabo et al., 1974). Therefore, for in vivo production LdcC proved to be superior as it exhibits a broader pH range (Kind et al., 2010) and is hardly inhibited by its substrate (Shin et al., 2018). The third decarboxylase DC Fj tested in this study was investigated before as the genes coding for the lysine-4-hydroxylase KDO Fj (Fjoh_3169) and the PLPdependent decarboxylase (Fjoh_3171) were found co-localized in the genome of Flavobacterium johnsoniae and as it decarboxylated 4-HL to 3-HC (Baud et al., 2017). While the main advantage of DC Fj is its substrate preference for 4-HL (rather than lysine), its activity in vivo was low. Further metabolic engineering for optimised expression including optimisation of the ribosome binding site, using different expression systems, and adaption of the codon usage to C. glutamicum, as the sequence is rather TA-rich (63 %), might be the key to making use of this substrate preference.
The process was successfully scaled up and the titer increased 7fold to 74 mM 3-HC. As KDOs are oxygen-dependent enzymes (Martinez and Hausinger, 2015), the process may have benefitted from better aeration. It was shown before, that the aeration rate of the process had a major effect on the production of L-2-hydroxyglutarate by C. glutamicum using the KDO glutarate dioxygenase CsiD . It remains to be studied if 3-HC can be separated from cadaverine efficiently in downstream processing. This and other challenges remain to be solved before the proof-of-concept of 3-HC production can be transferred to industrial application.

Funding
This research was funded in part by the European Regional Development Fund (ERDF) and the Ministry of Economic Affairs, Innovation, Digitalization and Energy of the State of North Rhine-Westphalia by grant "Cluster Industrial Biotechnology (CLIB) Kompetenzzentrum Biotechnologie (CKB)" (34.EFRE-0300095/1703FI04). Florian Meyer was funded by BMBF (KaroTec; grant number: 03VP09460). Fernando Pérez-García was funded by the Norwegian University of Science and Technology (NTNU). Support for the Article Processing Charge by the Deutsche Forschungsgemeinschaft and the Open Access Publication Fund of Bielefeld University is acknowledged. The funding bodies had no role in the design of the study or the collection, analysis, or interpretation of data or in writing the manuscript.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Bielefeld, for technical assistance and kind advice. Additionally, we thank Daniel Krüger and Arno Krieger for technical assistance.