All chemicals and solvents were purchased from Sigma-Aldrich unless otherwise stated. ¹³C labeled substrates (purity 99% ¹³C-atom) were purchased from Eurisotop, Omicron or Sigma-Aldrich. DNA plasmid isolation and DNA extraction from agarose gels were carried out using kits purchased from New England Biolabs. Genomic DNA of Clostridium acetobutylicum (ATCC824) and Bifidobacterium adolescentis (ATCC15703) was obtained from the German Collection of Microorganisms and Cell Cultures GmbH (DSMZ, Germany). DNA sanger sequencing was carried out by Microsynth AG (Balgach, Switzerland) or Eurofins SAS (Ebersberg, Germany).

All strains and plasmids used in this study are listed in Table 1 and Table 2.

Liquid lysogeny broth (LB) medium (10 g L⁻¹ tryptone, 5 g L⁻¹ yeast extract and 10 g L⁻¹ NaCl) was used for cloning procedures and strain maintenance, protein production and cell recovery from glycerol stocks (25% v/v) kept at –80°C. LB agar plates were prepared by the additional supplementation of 15 g L⁻¹ agar-agar. In vivo experiments were unless otherwise stated performed in M9 mineral medium (Walther et al., 2017) which contained 18 g L⁻¹ Na₂HPO₄·12H₂O, 3 g L⁻¹ KH₂PO₄, 0.5 g L⁻¹ NaCl, 2 g L⁻¹ NH₄Cl, 0.5 g L⁻¹ MgSO₄·7H₂O, 0.015 g L⁻¹ CaCl₂·2H₂O, 0.010 g L⁻¹ FeCl₃, 0.012 g L⁻¹ thiamine HCl, 0.4 mg L⁻¹ NaEDTA·2H₂O, 1.8 mg L⁻¹ CoCl₂·6H₂O, 1.8 mg L⁻¹ ZnCl₂SO₄·7H₂O, 0.4 mg L⁻¹ Na₂MoO₄·2H₂O, 0.1 mg L⁻¹ H₃BO₃, 1.2 mg L⁻¹ MnSO₄·H₂O, 1.2 mg L⁻¹ CuCl₂·2H₂O. This medium was buffered by adding 100 mM 3-(N-morpholino)-propanesulfonic acid (MOPS), adjusted to pH 7 by KOH and filter sterilized. The mineral medium was further supplemented with glycerol as main carbon source (for concentrations, see main text). Whenever required, antibiotics were added at the following concentrations: ampicillin, 100 mg L⁻¹; chloramphenicol, 35 mg L⁻¹; kanamycin sulphate, 50 mg L⁻¹; spectinomycin dihydrochloride pentahydrate, 100 mg L⁻¹.

In vitro enzyme studies were conducted using pET-28(a)+ vectors (Novagen) for gene expression. E. coli DH5α (New England Biolabs) was used for construction and storage of pET-28(a)+ derived plasmids. Wild-type genes fsaA, kdsD and rpe were amplified from purified genomic DNA extracted from E. coli MG1655, whereas xfp gene encoding the Pkt was amplified from genomic DNA of C. acetobutylicum ATCC824. The used primers are listed in Supplementary Table S2. After digestion of the resulting DNA fragments by suitable restriction enzymes, T4 DNA ligase (Biolabs) was used to clone target genes into the corresponding sites of pET-28(a)+ vector system. Remaining template DNA was digested by DpnI and the resulting products were transformed into competent cells.

N-His-tagged enzymes were expressed from E. coli BL21 (DE3) bearing pET-28a (+) derived plasmids. After an overnight pre-culture in 10 mL LB medium supplemented with kanamycin (37°C, 220 rpm), cells were inoculated at an optical density at 600 nm (OD₆₀₀) of 0.2 in 500 mL shake flasks containing 50 mL LB medium supplemented with kanamycin (37 °C, 220 rpm). To induce protein expression, 1 mM isopropyl-β-D-thiogalactopyranoside (IPTG) was added when an OD₆₀₀ of ∼0.6–0.8 was reached. FsaA, Rpe and KdsD enzymes were expressed for 16 h at 20°C, 220 rpm, whilst expression of Pkt from Clostridium acetobutylicum (Ca-Pkt) was carried out at 16°C for 30 h, 220 rpm. Afterwards, cell suspensions were centrifuged (at 4,000 g for 10 min, 4°C), the supernatants were discarded and pellets were stored at -20°C until further use.

For protein purification, cell pellets were first thawed on ice and resuspended in 0.75 mL lysis buffer (10 mM HEPES, 300 mM NaCl, pH 7.5). Cells were broken open by three successive rounds of sonication (sonication interval: 10 s, power output: 40%, Topas, UDS 751) and cellular debris was separated from the supernatant at 13,000 g for 10 min, 4°C. The supernatant was then transferred to a new reaction tube to which 75 µL of streptomycin sulfate solution (15 mg mL⁻¹) was added. After centrifugation at 13,000 g for 10 min (4°C), the resulting supernatant free of DNA was retained for subsequent steps of protein purification by immobilized metal affinity chromatography (IMAC). A volume of 0.75 mL of TALON^® Metal Affinity Resin (Cytiva, USA) was washed twice with 5 volumes of deionized water before being incubated with protein extract on a rotating wheel for 30 min at room temperature. After centrifugation (700 g for 5 min, 4°C), the supernatant was discarded and resin was washed twice with 2 mL of lysis buffer. His-tagged enzymes were then eluted by applying 0.5 mL elution buffer (10 mM HEPES, 300 mM NaCl, 250 mM imidazole, pH 7.5). The eluted fraction was then loaded on an Amicon Ultra-0.5 centrifugal filter unit (pore size, 10 kDa; Merck Millipore, USA) and after a round of centrifugation (14,000 g, 15 min), the buffer was exchanged by applying 0.45 mL lysis buffer and eventually recovering (1,000 g, 2 min) the protein fraction in 0.3 mL of this buffer. Concentration of purified proteins was determined by a Bradford assay (ROTI®Quant, Roth), using bovine serum albumin for preparing a standard calibration curve (0–100 μg mL⁻¹).

Prior to demonstrating the in vitro functioning of the synthetic pathway, functional enzyme expression was confirmed by measuring activities on corresponding natural substrates as described in Supplementary Information.

Functioning of the whole enzyme cascade was tested in a 1.5 mL reaction tube incubated on a rotary shaker (37°C, 260 rpm), and monitored by offline measurement of formed AcP. The reaction mixture (640 µL) contained 50 mM HEPES (pH 7, adjusted with KOH), 50 mM KH₂PO₄ (pH 7, adjusted with KOH), 5 mM MgCl₂, 1 mM (D)-GA3P, 1 mM thiamine pyrophosphate (TPP). The total protein concentration in the final reaction mix was set to 250 μg mL⁻¹. In order to maximize pathway flux, enzyme concentrations were adjusted as followed: FsaA, 60 μg mL⁻¹; KdsD, 60 μg mL⁻¹; Rpe, 50 μg mL⁻¹; Ca-Pkt, 80 μg mL⁻¹. Reactions were started by the addition of 10 mM GA. To ascertain pathway function, various controls were performed by replacing either (D)-GA3P, Ca-Pkt, or FsaA by deionized water. To estimate the specific decay kinetics of the unstable AcP, an additional reaction without enzymes and GA but containing 2 mM AcP was included in the analysis. Quantification of AcP was performed via its derivatization to iron-acetyl-hydroxamate-complex by the hydroxamate assay (Lipmann and Tuttle 1945). For each time point, 80 µL of sample were transferred to a 96-well plate and the reaction was stopped by the addition of 60 µL of 2 M hydroxylamine solution. After 10 min, 40 µL trichloroacetic acid solution (15% w/v), 40 µL 4 M HCl and 40 µL FeCl₃·6H₂O solution (5% w/v in 0.1 M HCl) were sequentially added. The derivatized product was immediately measured spectrophotometrically at 505 nm. For calibration, freshly prepared lithium AcP standard solutions (0–14 mM) were included in reaction mixtures without GA substrate or enzymes.

All vectors used and constructed in this study (Table 2) are based on the pZ expression system (Expressys). Backbone vectors used for plasmid construction were pZA23, pZS13 and pZS33, and target genes were cloned into respective multiple cloning site (MCS) module downstream of a P_A1lacO-1 promoter using NEBuilder^® HiFi DNA assembly reaction kit (New England Biolabs). Briefly, genes were amplified by PCR from corresponding genomic DNAs (fsaA, kdsD, pta, rpe, talB: E. coli MG1655; Ca-xfp: Clostridium acetobutylicum ATCC824; Ba-xfp: Bifidobacterium adolescentis ATCC15703) using forward and reverse primer pairs listed in Supplementary Table S2. Primers contained 15- to 20-bp homology arms for subsequent in vitro homologous recombination. The forward primer additionally inserted a unique synthetic ribosomal binding site (RBS) in front of gene of interest, which was designed with RBS calculator (Salis et al., 2009). Backbone vectors were linearized by PCR followed by digestion with DpnI. After purification, linearized plasmid and PCR-amplified gene(s) were assembled by using the reaction mix provided in the kit. Resulting plasmids were transformed into NEB^® stable cells (New England Biolabs) and verified by DNA sequencing to carry correctly assembled operons. Point mutation in talB gene (F178Y) was introduced by PCR using primers listed in Supplementary Table S2.

All strains used and constructed in this study are listed in Table 1 . E. coli MG1655 ΔyqhD ΔaldA lacI ^q Spc ^R served as reference strain for demonstrating in vivo conversion of C₂ compounds EG and GA. Disruption of yqhD and aldA genes was achieved by P1vir phage transduction method (Malke, 1993) using single gene deletion mutants from the Keio collection (Baba et al., 2006). The procedure introduces at target locus a cassette (FRT-kan-FRT) composed of kanamycin resistance gene flanked by flippase (Flp) recognition sites. After selection and isolation of antibiotic-resistant clones, gene deletions were confirmed by diagnostic PCR (primer: yqhD, TW216, TW217; aldA, TW218, TW219;) using DreamTaq polymerase (ThermoFisher Scientific). The kanamycin resistance gene was excised using Flp-recombinase encoded from pCP20 helper plasmid (Cherepanov and Wackernagel, 1995) before a new round of chromosomal modification was initiated. Finally, the lacI ^q locus containing a fused spectinomycin resistance cassette was transduced into producer strains by P1vir phage method using a phage lysate of the donor strain E. coli C600Zi strain (Expressys).

For chromosomal overexpression of endogenous kdsD and rpe genes, native chromosomal 5′-UTR of each gene was replaced by the insulated constitutive promoter proD (Davis et al., 2011) together with a RBS sequence (AGGAGGTATTAT) via PCR-mediated λ-Red recombination (Datsenko and Wanner, 2000). The FRT-kan-FRT cassette fused to proD promoter sequence (FRT-kan-FRT-proD) was amplified by PCR (primers: kdsD, TW1553 and TW1554; rpe, TW 1555 and TW1556; Supplementary Table S2) from the genomic DNA of CF220 strain (kindly provided by Prof. J. M. François from Toulouse Biotechnology Institute, INSA Toulouse, France). The PCR product contained in addition 50 bp flanking sequences which were homologous to the target genomic locus. Recipient E. coli MG1655 derived strains expressing λ-Red recombination genes from pKD46 helper plasmid were transformed with ∼400 ng of gel-purified product. After an overnight incubation under non-selective conditions, cell suspensions were plated on LB agar plates supplemented with kanamycin. Successful cassette insertion was confirmed by diagnostic PCR and DNA sequencing of region encompassing promoter and gene of interest. Kanamycin resistance gene was excised as described above. Plasmids were transformed into target E. coli strains using standard protocols (Chung et al., 1989).

All cell cultivation was performed in an orbital shaker (Ecotron, Infors) set at a shaking frequency of 220 rpm. Pre-cultures were started from frozen glycerol stocks and grown at 37°C in 10 mL of LB medium supplemented with appropriate antibiotics until an OD₆₀₀ of ∼2 was reached.

For EG assimilation studies, cells from first pre-culture were diluted by 100-fold in 25 mL of M9 medium containing 220 mM glycerol and appropriate antibiotics. After overnight growth (at 37°C), cells were harvested by centrifugation (4,500 g for 10 min, 4°C), washed once with sterile water and used to inoculate at a starting OD₆₀₀ of 0.2 100 mL shake flasks containing 5 mL of main culture composed of M9 medium supplemented with glycerol (concentrations indicated wherein appropriate in main text) and appropriate antibiotics. EG was added to the medium from the beginning at a final concentration of 500 mM. IPTG (1 mM) was added when cells reached an OD₆₀₀ of 0.6 (∼2.5 h after inoculation), except for MVA production from EG where the optimal induction time was determined to be 2 h after inoculation. Main cultures were carried out at 37°C for acetate production or at 30°C (Xiong et al., 2014) when aiming at MVA biosynthesis. The culture medium used in the experiments to convert EG to MVA contained 100 mM CaCO₃ as an additional buffer.

In experiments investing GA assimilation, cells from a first pre-culture were used to inoculate main culture (50 mL of 90% v/v M9 medium with 220 mM glycerol and 10% v/v LB in 500 mL shake flasks) with an OD₆₀₀ of 0.2. Cells were cultivated at 30°C and when an OD₆₀₀ of 0.6 was reached, IPTG (1 mM) was added to induce expression of genes under control of P_A1lacO-1 promoter. Cells were harvested by centrifugation (4,500 g for 10 min, 4°C) at an OD₆₀₀ equal to 1.5, washed twice with sterile water, and resuspended at an OD₆₀₀ of ∼3.5 in 5 mL of M9 medium containing co-substrates as specified in the main text, 1 mM IPTG and 10 mM ¹³C₂-GA. Samples for extracellular and intracellular metabolites were withdrawn regularly and treated as described in next subsections.

EG dehydrogenase activity was measured in strains of E. coli MG1655 ΔyqhD ΔaldA lacI ^q Spc ^R expressing FucO^Mut using a protocol adapted from Boronat et al., 1983. Briefly, cells grown in LB medium were used to inoculate main culture (50 mL of 90% v/v M9 medium with 110 mM glucose and 10% LB in 500 mL shake flasks) with an initial OD of 0.2. Cells were cultivated at 37°C (220 rpm) and 1 mM IPTG was added when an OD₆₀₀ of 0.6 was reached. Biomass was harvested by centrifugation (4,500 g for 10 min, 4°C) at an OD₆₀₀ equal to 1.8, the supernatants were discarded and cell pellets were stored at -20°C until further use. To prepare cell extracts, pellets were thawed on ice, suspended in 25 mM sodium glycine buffer (pH 9.5 by KOH), broken open by three successive rounds of sonication (sonication interval: 10 s, power output: 40%, Topas, UDS 751) and cellular debris was separated from the supernatant at 13,000 g for 10 min, 4°C. Clear supernatants were used in different dilution for further analysis and total protein concentration was determined by a Bradford assay as described above (Section 2.5).

Spectrophotometric assay for EG dehydrogenase activity was performed at 37 °C following NADH formation at 340 nm (ε_{340 nm} = 6.22 mM⁻¹cm⁻¹). Oxidation activity was measured in an assay mixture that contained 50 mM EG, 100 mM sodium glycine buffer (pH 9.5 by KOH) and 0.5 mM NAD.

At appropriate time intervals, 0.5 mL of cell culture were withdrawn and filtered (polyamide filter, 0.2 µm pore size, Sartorius Stedim) followed by the addition of 1 mL of room temperature deionized water to wash retained cells. Filter and cells were immediately transferred to a 10 mL glass tube containing 5 mL of 75% v/v hot ethanol (80°C). After vortexing for 10 s, glass tubes were incubated for 3 min at 80°C and chilled on ice for an additional 10 min period. The filter was then removed and suspension was centrifuged (13,000 g, 5 min, room temperature) to remove impurities and cell debris. The clear supernatant was stored overnight at -20°C. On the next morning, liquid was removed from the samples at 45°C for 4 h using a benchtop vacuum concentrator (CentriVap Concentrator System, Labconco, USA). Dried metabolites were suspended in 250 µL deionized water and analyzed directly by LC/MS.

Quantification of extracellular and intracellular metabolites was achieved by high performance liquid chromatography (UltiMate 3,000 system, Dionex, USA), equipped with an autosampler (Dionex, Sunnyvale, USA) keeping samples at 6°C. It was coupled to an UV/vis detector (Dionex, Sunnyvale, USA) measuring at 210 nm and a RI detector (ERC RefractoMax 520, Knauer, Germany). Measurements of ¹³C incorporation was performed by coupling liquid chromatography (Vanquish™, ThermoFisher™, USA) with a mass spectrometer (Q Exactive™ Focus orbitrap, ThermoFisher™, USA).

For analysis of extracellular metabolites, all withdrawn samples were centrifuged (2 min at 13,000 rpm) and syringe-filtered (0.2 µm), and the resulting supernatant was stored at −20°C until analysis. Cell-free supernatant was separated for the analysis of extracellular metabolites using Rezex™ ROA-Organic Acid H⁺ column (300 × 7.8 mm, 8%, Phenomenex) protected by a SecurityGuard™ Carbo H⁺ pre-column (4 × 3 mm, Phenomenex). For detection of glycerol, EG, acetate and MVA analytes were eluted with 2.5 mM H₂SO₄ at a constant flow rate of 0.5 mL min⁻¹ and a column temperature of 65°C was adjusted. GA, glycolate and d-threose analyses were performed at a column temperature of 35°C using 0.5 mM H₂SO₄ as eluent. For LC/MS measurement of ¹³C enrichments in acetate and MVA 0.1% formic acid was used as mobile phase and an isocratic flow of 0.4 mL min⁻¹ was adjusted.

For LC/MS-analysis of intracellular metabolites all samples and standards were prepared in 10 mM ammonium acetate (pH 9.2) in a matrix of 60% v/v acetonitrile and 40% v/v water. Separation was achieved by liquid chromatography using a SeQuant^® ZIC®-pHILIC column (5 µm polymer 150 × 2.1 mm) with a flow rate of 0.15 mL min⁻¹. For an optimal efficiency of separation, a gradient of A (5% ACN, 10 mM ammonium acetate, pH 9.2 adjusted by NH₄OH) and B (90% acetonitrile, 10 mM ammonium acetate, pH 9.2 by NH₄OH) was used. The gradient was 0 min, 95% B; 2 min, 95% B; 3 min, 89.4% B; 5 min, 89.4% B; 6 min, 83.8% B; 7 min, 83.8% B; 8 min, 78.2% B; 9 min, 78.2% B; 10 min, 55.9% B; 12 min, 55.9% B; 13 min, 27.9% B; 16 min, 27.9% B; 18 min, 0% B; 23 min, 0% B; 24 min, 95% B; 30 min, 95% B. The temperature of the autosampler was kept at 6°C, injection volume was 5 µL and oven temperature was kept at 25°C. Instrumental settings according to the electrospray ionization were optimized for a flow rate of 0.15 mL min⁻¹. Final parameters were adjusted as follows: sheath gas flow rate 32 (arbitrary units), auxiliary gas flow rate 8 (arbitrary units), sweep gas flow rate 0 (arbitrary units), spray voltage -3.5 kV, capillary temperature 250°C and auxiliary gas temperature 200°C. Monoisotopic mass (listed in Supplementary Table S4) and retention time were used to identify metabolites. Retention times of all shown analytes were measured by injecting unlabelled standards. Peak areas were corrected for the contribution of all naturally abundant isotopes using the software IsoCor (Millard et al., 2012).

The aim of this work was the in vivo implementation of the synthetic metabolic pathway for the whole-cell bioconversion of EG into AcCoA. Implementation of a synthetic metabolic pathway in vivo is much more challenging than its assembly with purified enzymes, since competing side reactions can convert intermediates of the pathway into undesired by-products, and because the activity of individual enzymes cannot be easily changed by simply adding more or less protein. An adequate activity of the enzymes of the target pathway must therefore be achieved, which requires both sufficiently high expression of the corresponding genes in the producer strain and a high specificity of the enzymes with respect to their substrates and products. To meet such endeavour, we introduced changes in selection of pathway enzymes compared to the Yang study (Yang et al., 2019) that first demonstrated the in vitro function of the Ara5P-dependent GAA pathway by converting GA into AcP. Specifically, fructose 6-phosphate aldolase (FsaA) from E. coli was selected to catalyze the cross aldol addition of GA to GA3P. Unlike the TalB^F178Y mutant used by Yang et al., 2019, the FsaA enzyme does not form the byproduct D-ribose 5-phosphate (Schürmann and Sprenger, 2001), which may be deviated into purine de novo synthesis. Furthermore, we used the Pkt from Clostridium acetobutylicum (Ca-Pkt) instead of the Pkt from P. stutzeri (Ps-Pkt). Ca-Pkt has a high preference for Xylu5P over F6P (Bergman et al., 2016) similar to Ps-Pkt. In addition, it was well expressed in E. coli ((Servinsky et al., 2012) and own data, not shown). For converting Ara5P into Xylu5P we wanted to recruit the native E. coli enzymes Ara5P isomerase (KdsD, Meredith and Woodard 2003) and d-ribulose 5-phosphate 3-epimerase (Rpe, Lyngstadaas et al., 1995) as Yang et al., 2019 (Figure 1). In vitro tests were then performed to verify that the synthetic reaction sequence was still functional despite the replacement of two enzymes. Upon incubation of his-tag purified enzymes with 10 mM GA and 1 mM GA3P, a maximum titer of 1.81 ± 0.02 mM AcP was detected, thereby confirming that the pathway is functional and that GA3P can be regenerated (Figure 2).

The synthetic reaction sequence is mainly composed of enzymatic activities which are endogenous to E. coli and for this reason, we investigated the requirements of overexpressing individual enzymes for in vivo pathway operation in this organism. We chose E. coli MG1655 ΔyqhD ΔaldA lacI ^q mutant strain (EC0) as host strain, in which genes encoding the major endogenous GA reductase YqhD (Alkim et al., 2015) and GA dehydrogenase AldA (Boronat et al., 1983; Cam et al., 2016) were deleted to support assimilation of GA via the synthetic pathway.

Genes encoding the pathway enzymes were expressed from low and medium copy plasmids in different combinations and 55 mM of glycerol (as a growth supporting carbon source) together with 500 mM uniformly labeled ¹³C₂ EG were fed to the resulting producer strains. Secreted fully labeled (M+2) acetate was used to confirm conversion of ¹³C₂ EG via the Ara5P-dependent GAA pathway (Figure 3A). Glycerol was added as the co-substrate to support growth because E. coli produces much less acetate when growing on glycerol than when utilizing glucose (Martínez-Gómez et al., 2012). A comparatively high EG concentration was applied because of the unfavorable thermodynamics of the initial NAD-dependent EG oxidation step (Δ_rG'° = 23.7 ± 3.9 kJ mol⁻¹; a thermodynamic analysis of the entire pathway is shown in Supplementary Figure S1, Noor et al., 2014). As a consequence, it was not possible to reliable quantify the comparatively low EG consumption in these experiments.

It was previously shown that E. coli does not naturally express EG dehydrogenase enzymes at sufficient levels even in the presence of this substrate (Pandit et al., 2021). Therefore, the oxygen-insensitive EG dehydrogenase variant FucO^I6L:L7V from E. coli (Pandit et al., 2021, herein denoted FucO^Mut) was expressed from pZA23 plasmid in host EC0 strain yielding strain EG1. Adequate expression of the FucO^Mut enzyme (0.25 µmol/min/mg_prot) was verified by an enzymatic assay and acetate production by strain EG1 was monitored. In the presence of EG, strain EG1 showed slightly inhibited growth and significantly increased acetate production compared to reference cultivation without EG (Supplementary Figure S3A). However, LC/MS analysis revealed that the mere expression of EG dehydrogenase did not suffice to secrete fully labeled (M+2) acetate, thus, confirming that this isotopologue cannot be formed by unexpected alternate pathways present in E. coli. To understand whether the enhanced secretion of unlabeled acetate was caused by the presence of EG or by the intracellular accumulation of GA, a strain (EG0) carrying an empty plasmid (pZA23_MCS) was cultured under the same conditions. In the presence of EG, strain EG0 accumulated much less acetate than strain EG1 which argues for GA being responsible for increased acetate secretion (shown in Supplementary Figure S2). When the Ca-Pkt enzyme was expressed in addition to FucO^Mut (strain EG2), trace amounts of M+2 labeled acetate could be detected which may have been produced through the desired pathway or by direct conversion of GA to AcP catalyzed by a background activity of the Pkt enzyme (Yang et al., 2019). We continued our analysis by constructing strain EG3 which overexpressed the Ara5P aldolase FsaA in addition to the EG dehydrogenase and Xylu5P Pkt enzymes. Much to our surprise, this strain did not produce M+2 acetate at detectable levels which indicated that the pathway was not operational despite the fact that KdsD and Rpe are constitutively expressed in E. coli (Lyngstadaas et al., 1998; Meredith and Woodard 2005).

To test whether pathway function was hampered by insufficient expression of KdsD and/or Rpe, strains EG4, EG5 and EG6 were constructed in which either KdsD (EG4), Rpe (EG5) or both enzymes (EG6) were overexpressed in addition to FucO^Mut, FsaA and Ca-pkt. Strain EG4 showed production of M+2 acetate in significant quantities (Figure 3B). The fraction of fully labeled acetate increased directly after induction of protein synthesis (2.4 h) and reached a maximum after 11 h (see Figure 4). Even though the M+2 fraction accounted only for 2.6% of the total amount of acetate formed by strain EG4, this result clearly demonstrated in vivo functioning of the synthetic pathway. In contrast to strain EG4, only trace amounts of M+2 acetate were detected during the cultivation of EG5 confirming that native expression of KdsD is not sufficient to ensure flux via the synthetic pathway. Simultaneous overexpression of KdsD and Rpe in strain EG6 overexpressing all enzymes from EG to AcP did not produce more M+2 acetate than strain EG4, indicating that activity of chromosomally expressed Rpe did not limit pathway flux under these conditions. Finally, we investigated the role of FsaA, and constructed strain EG7, which harbored plasmids encoding genes for all enzymes with exception of FsaA. Again no formation of M+2 acetate was observed for this strain. In summary, these results indicate that all component enzymes of the Ara5P-dependent GAA pathway except for Rpe need to be overexpressed in E. coli to convert the C₂ alcohol EG into the intermediate AcP and further into acetate.

While fixation of EG via the synthetic metabolic pathway was confirmed, acetate formation on EG was found to be extremely low (max. 0.096 mM acetate in presence of 500 mM EG; strain EG4). We attributed this observation to the thermodynamic hurdle in the NAD-dependent EG oxidation to GA (compare Section 3.2), which hampers carbon flux into the pathway. Therefore, we used the product of EG oxidation, i.e., GA, as the substrate for further investigations. Previously performed work confirms that uptake of this aldehyde is possible without additional expression of an importer (Frazão et al., 2018; Lu et al., 2019). AcCoA could not be reliably measured by our LC/MS method. To demonstrate that the pathway works up to the level of this central precursor, we installed a heterologous pathway to produce the stable isoprenoid-precursor MVA, which is exclusively derived from AcCoA (Xiong et al., 2014). The fraction of labeled carbon in extracellular MVA could therefore be used as a proxy for estimating the contribution of the synthetic pathway to the total biosynthesis of AcCoA. Xiong et al., 2014 previously demonstrated that production of MVA was possible in E. coli when AcCoA acetyltransferase from E. coli (AtoB), and heterologous acetoacetyl-coenzyme A thiolase and MVA synthase activities from Lactobacillus casei (Lc-MvaS and Lc-MvaE, respectively) were (over)expressed. Accordingly, a plasmid harboring the MVA pathway genes (pMEV-7, Xiong et al., 2014) was transformed into the strain which operated the Ara5P-dependent GAA pathway starting from GA.

Strain GA1, which was derived from the E. coli host strain EC0 (E. coli MG1655 ΔyqhD ΔaldA lacI ^q ) and expressed the MVA pathway as well as all genes constituting the Ara5P-dependent GAA pathway except for EG dehydrogenase, was selected as the starting point of our investigations. Anticipating an increase in flux when using GA instead of EG as the substrate and in an attempt to prevent any potential limitation of carbon flux through the pathway, we decided to use a strain which also overexpressed Rpe even though its overexpression did not yield any improvements when producing acetate from EG.

E. coli did not grow at elevated GA concentrations. Since an initial concentration of 10 mM GA was previously shown to be suitable for studying the in vivo function of a GA-dependent pathway in E. coli (Lu et al., 2019; Walther and Frazao, 2022), we chose this concentration to investigate MVA production. In these bioconversion experiments, cells were induced by 1 mM IPTG in the pre-culture, harvested and resuspended in fresh M9 medium containing 1 mM IPTG and 10 mM of fully labeled ¹³C₂-GA adjusting to an OD₆₀₀ of 3.5. The fraction of GA-derived carbon in extracellular metabolites was quantified using LC/MS analysis. When GA was the only carbon source, no MVA was detected in the supernatant after 24 h of incubation. However, detection of 0.21 mM fully labeled (M+2) acetate indicated that conversion of GA to the intermediate AcP via the synthetic pathway was functional (Table 3). The conversion of AcCoA to MVA requires two NADPH molecules which cannot be provided if GA is used as the substrate and no carbon is channeled away from the pathway into NADPH-yielding reactions. Therefore, we investigated whether the addition of a co-substrate could increase AcCoA and MVA production through the synthetic pathway and supplemented the cultures with unlabeled glucose (5 mM) or glycerol (10 mM) at an equal C-mole basis. Both in the presence of either glucose and glycerol, again fully labeled acetate accumulated in the first 2 h of incubation, but was consumed after co-substrate depletion (Supplementary Figure S4). The maximum concentrations of M+2 acetate for glucose and glycerol were 0.12 ± 0.08 mM and 0.46 ± 0.17 mM, respectively. Addition of either carbon source enabled the formation of MVA. When glycerol was used as the co-substrate, the fraction of GA-derived carbon in MVA amounted to 12.42%, which was significantly higher than during utilization of glucose (5.44%). Although this result showed that the assimilation of GA made a significant contribution to the biosynthesis of AcCoA and MVA, it is important to note that MVA was only produced in small quantities (0.28 mM), which accounted only for ∼ 1% [C-mol] of the consumed GA and ∼ 5% [C-mol] of the consumed glycerol. Major fermentative by-products were EG and d-threose (Table 3). Thus, reduction of GA to EG by unspecific GA reductases (Alkim et al., 2015), and the Fsa-catalyzed threose-yielding homo-aldol addition of GA (Garrabou et al., 2009) were the major reactions which drained GA-derived carbon into metabolic by-products. To further investigate why the balance of GA-derived carbon did not close (recovery was only ∼50%, Table 3), we incubated 10 mM of GA in glycerol-containing M9 medium without cells and measured the GA concentration after 24 h. We found that approximately 30% of the initially added GA was consumed in spontaneous abiotic reactions yielding unidentified products. However, even if loss through abiotic reactions would be considered in the balance, a significant portion of GA-derived carbon (∼20%) still could not be recovered.

The fact that the fate of at least 20% of the GA-carbon could not be explained by spontaneous degradation or conversion to extracellular products indicated the presence of additional metabolic side reactions which have diverted carbon away from the target pathway. To verify that the presence of labeled carbon in AcCoA and MVA could indeed be explained by the operation of the synthetic pathway, and to identify competing metabolic routes (see Figure 5A) which may have contributed to consumption of GA-derived carbon, we analyzed the distribution of isotopologues in metabolites of central metabolism. Samples of intracellular metabolites were taken at 1 h after incubation with fully labeled ¹³C₂-GA and unlabeled glycerol (data shown in Figure 5C). At this time point, when cells were not limited for neither glycerol nor GA (see Figure 5B), we found more than 80% of the measured P5P pool (which includes Ara5P, Ribu5P, Xylu5P and D-ribose 5-phosphate/Ribo5P) to be labeled in two carbon atoms (M+2). In addition, most of the intracellular AcP (∼90%) was fully labeled (M+2), whereas the GA-accepting molecule GA3P and the glycolytic metabolites phosphoenolpyruvate and pyruvate contained very little to no GA-derived carbon (Figure 5C). Together, these findings indicated that the intended synthetic reaction sequence was functional in vivo and that it was the only source of GA-derived carbon in AcCoA and MVA.

The Ara5P-dependent GAA pathway shares the intermediates Xylu5P and Ribu5P with the natural central carbon metabolism of E. coli (Figure 5A). To evaluate whether the apparent loss of GA-derived carbon (compare Section 3.3) could be explained by the consumption of these compounds through competing native pathways, the isotopologue distribution of adjacent metabolites was investigated. Indeed, we could detect a ∼10% (M + 2) fraction in the fructose 6-phosphate (F6P) pool, which argued for a transketolase-dependent carbon transition between Xylu5P and F6P. However, since there was no labeled carbon in neither glucose 6-phosphate nor GA3P, we concluded that this route did not contribute significantly to the overall carbon balance. Similarly, the absence of ¹³C-label in sedoheptulose 7-phosphate and tryptophan indicated that transaldolase did not divert carbon away from the synthetic pathway (Figures 5A, C). The presence of a ∼10% M + 2 fraction in both adenosine monophosphate and cytosine monophosphate showed that purine and pyrimidine de novo synthesis drained Ribu5P from the target pathway. But since the biomass concentration did not increase much during the experiment, it is likely that carbon consumption through these metabolic routes was also negligible. While these results suggested that carbon loss in the pathway section between GA and AcCoA was small, presence of (M+2) isotopologue fractions in 2-oxoglutarate (7.6%) and malate (4.1%) showed that a significant portion of AcCoA and therefore GA-derived carbon was channeled through the TCA cycle, thus being lost in the form of CO₂.

Taken together, we observed efficient incorporation of GA-derived carbon along the proposed pathway with high (over 80%) proportions of M+2 isotopologues in all intermediates. From the near absence of labeled carbon in pyruvate and the labeling pattern of MVA (Figure 5C) it can be concluded that 13.55% of the intracellular AcCoA were produced through the synthetic pathway during the first hour of incubation. However, a large fraction of the AcCoA was not converted to the target product MVA but most likely entered the TCA cycle as witnessed by the low overall MVA yield and the presence of significant fractions of GA-derived carbon in TCA cycle metabolites.

With the aim of achieving a higher metabolic flux via the Ara5P-dependent GAA metabolic pathway, various optimization approaches were pursued starting from reference strain GA1 (E. coli ΔyqhD ΔaldA lacI ^q + pZA23_fsaA_Ca-pkt + pZS33_kdsD_rpe + pMEV-7, Table 3 and Figure 5). The impact of different genetic modifications was quantified by comparing the relative amount of labeled carbon found in the extracellular product MVA after 24 h of incubation with 10 mM ¹³C₂ GA and 10 mM glycerol (Figure 6).

Strain GA1 expressed the Ara5P aldolase FsaA, which was chosen on the basis of its high specificity for the desired product. However, Yang et al., 2019 used an alternative aldolase TalB^F178Y (Rale et al., 2011) to demonstrate the in vitro functioning of the Ara5P-dependent GAA pathway. To check whether the use of TalB^F178Y could improve the in vivo conversion of GA, we replaced the FsaA encoding gene on the medium-copy number plasmid by either the wildtype TalB or the mutant TalB^F178Y gene resulting in strain GA2 and GA3, respectively. While GA1 showed an incorporation of 12.42% GA-derived carbon after 24 h, it was found that in strains GA2 and GA3 only 1.56% and 4.28% of the MVA was derived from GA (Figure 6). This observation confirmed FsaA Ara5P aldolase as a superior enzyme in the in vivo context.

In the next step, we investigated whether the replacement of the Ca-Pkt by the Pkt from Bifidobacterium adolescentis (Ba-Pkt) could improve pathway function. The Ba-Pkt is a well-characterized enzyme and has already been used in E. coli to construct a synthetic non-oxidative glycolysis (Bogorad et al., 2013; Lin et al., 2018). Strain GA4 carried a pZA23 plasmid which harbored Ba-pkt instead of the Ca-pkt gene. The MVA produced by GA4 contained less GA-derived carbon (9.46%) than the GA1 reference strain, demonstrating that Ca-Pkt is the better pathway enzyme, presumably, because Ba-Pkt has a lower specificity for Xylu5P (Bergman et al., 2016).

We next set out to investigate the impact of overexpressing KdsD and/or Rpe on flux through the synthetic pathway. To this end, the corresponding genes were overexpressed from a plasmid and/or the chromosome by replacing the natural promoter of the two genes by the synthetic constitutive strong promoter ProD (Davis et al., 2011). As expected from the results of the EG-to-acetate conversion experiments, only a very small fraction of MVA (2.2%) could be derived from GA when neither of the two enzymes was overexpressed in strain GA5. In contrast, the simultaneous plasmid- and chromosomally-based overexpression of both genes (strain GA9) resulted in a¹³C-enrichment of MVA (15.49%) which was increased by 24.7% compared to that observed for strain GA1 (12.42%, which overexpresses Rpe and KdsD from plasmid alone).

Based on the analysis of intracellular metabolites (strain GA1, Figure 5C), a strong discrepancy between labeling in the communicating metabolite pools AcCoA/MVA and AcP was detected. Possible explanations for a much lower ¹³C content in AcCoA/MVA are a low flux via the Ara5P-dependent GAA metabolic pathway or an insufficient phosphate acetyltransferase (Pta) activity. To test whether this enzyme activity was a bottleneck, we extended the genetic repertoire of strain GA1 by overexpressing Pta from the low-copy number pZS33 plasmid in addition to KdsD and Rpe. The resulting strain GA10 produced MVA containing a fraction of 16.12% GA-derived carbon which corresponded to an increase of 28.8% compared to the reference strain GA1. This result showed that endogenous expression of Pta enzyme had a limiting effect on the flux between AcP and AcCoA, which seems surprising given that this enzyme is highly active and reported to be constitutively expressed (Campos-Bermudez et al., 2010; Martínez-Gómez et al., 2012; Enjalbert et al., 2017). Finally, we tried to combine the positive effect observed when overexpressing Pta, KdsD and Rpe from the pZS33 plasmid with boosting the expression of KdsD and Rpe from the chromosome. However, the corresponding strain GA11 exhibited a strong growth phenotype (not shown) and a severe drop of the GA-derived carbon fraction in MVA (11.94%, Figure 6). Taken together, these results indicate that the synthetic pathway can make a significant contribution to the biosynthesis of AcCoA. To further enhance its performance, the expression of the constituent genes needs to be carefully tuned and a better understanding of the apparent limiting role of Pta is required.

The metabolic pathway was designed to enable the carbon-conserving conversion of EG to AcCoA-derived products. The assimilation of EG through the pathway has already been demonstrated by detection of EG-derived acetate. Since the plasmid-borne overexpression of the native enzyme Pta proved advantageous for MVA production from GA, the metabolic setting of strain GA10 (which is E. coli ΔyqhD ΔaldA lacI ^q + pZA23_fsaA_Ca-pkt + pZS33_kdsD_rpe_pta + pMEV-7) was extended for the bioconversion of EG into MVA by the additional expression of the EG dehydrogenase FucO^Mut (strain EG8 = E. coli ΔyqhD ΔaldA lacI ^q + pZA23_fucO^Mut_fsaA_Ca-pkt + pZS33_kdsD_rpe_pta + pMEV-7). To investigate the ability of the engineered strain to convert EG into MVA, it was cultured in the presence of 500 mM ¹³C₂-EG. An increased concentration (250 mM) of the growth-supporting co-substrate glycerol was added to extend growth and production phase. As mentioned above, by using a high EG concentration it was possible to overcome the thermodynamically unfavorable initial EG oxidation step, but EG uptake remained too small for a reliable quantification. Corresponding growth and production kinetics are shown in Supplementary Figure S5A. No accumulation of GA nor threose was detected during the whole cultivation. Sampling after complete glycerol consumption (46 h) showed a total MVA concentration of 10.80 ± 0.65 mM. LC/MS analysis revealed a relative ¹³C enrichment in MVA of 0.82% in the final sample and a maximum of only 1.24% (Supplementary Figure S5B) after 22 h of cultivation. However, these results indicating that the synthetic pathway was functional albeit at currently extremely small yields.