Growth of E. coli on formate and methanol via the reductive glycine pathway

Engineering a biotechnological microorganism for growth on one-carbon intermediates, produced from the abiotic activation of CO2, is a key synthetic biology step towards the valorization of this greenhouse gas to commodity chemicals. Here we redesign the central carbon metabolism of the model bacterium Escherichia coli for growth on one-carbon compounds using the reductive glycine pathway. Sequential genomic introduction of the four metabolic modules of the synthetic pathway resulted in a strain capable of growth on formate and CO2 with a doubling time of ~70 h and growth yield of ~1.5 g cell dry weight (gCDW) per mol-formate. Short-term evolution decreased doubling time to less than 8 h and improved biomass yield to 2.3 gCDW per mol-formate. Growth on methanol and CO2 was achieved by further expression of a methanol dehydrogenase. Establishing synthetic formatotrophy and methylotrophy, as demonstrated here, paves the way for sustainable bioproduction rooted in CO2 and renewable energy. Redesigning the central carbon metabolism of Escherichia coli with the reductive glycine pathway enables growth on the one-carbon compounds formate and CO2, and the addition of methanol dehydrogenase further enables growth on methanol and CO2.

is very limited 8 . Aerobic cultivation, while associated with lower bioconversion efficiency, is generally much more flexible in terms of production capability. Despite considerable progress in developing better genetic tools for engineering natural aerobic formatotrophs and methylotrophs, their biotechnological application is still limited. This is in part due to unfavorable cultivation parameters (for example, cell concentration and growth rate) and low efficiency of the relevant metabolic pathways 9 . Adapting a biotechnologically relevant microorganism for growth on formate or methanol has therefore been a key goal of the synthetic biology community in the last decade [10][11][12][13][14][15][16][17][18][19][20][21] . However, so far, the success of these efforts has been limited. This could be partially explained by the complexity of the natural pathways-the Calvin cycle, the serine cycle and the ribulose monophosphate cycle 22 -the cyclic activity of which strongly overlaps with central metabolism and requires complex regulation of the fluxes that converge into and diverge away from the pathway.
Here we use a modular engineering approach to enable E. coli to grow on formate and methanol. Instead of attempting to engineer a cyclic pathway, we focus on the reductive glycine pathway (rGlyP), a linear route that directly assimilates formate and CO 2 into central metabolism. We divide the pathway into four modules and show how their sequential expression from the genome enables the bacterium to grow on formate. We then cultivate the engineered E. coli strain on formate for several generations and isolate a mutant with substantially higher growth rate and yield. We identify two genes, the overexpression of which explains the enhanced growth. Further expression of methanol dehydrogenase (MDH) enables E. coli to metabolize methanol to formate, thus supporting growth on this C 1 carbon source.
The rGlyP, as shown in Fig. 1, was designed to be the aerobic twin of the rAcCoAP (ref. 24 ). Both are linear routes with limited overlap with central metabolism, minimizing the need for regulatory optimization. Both pathways start with the ligation of formate and tetrahydrofolate (THF) and proceed via reduction into a C 1 -THF intermediate, which is then condensed, within an enzyme complex, with CO 2 to generate a two-carbon (C 2 ) compound (acetyl-CoA or glycine). The C 2 compound is finally condensed with another C 1 moiety to generate pyruvate as biomass precursor. Importantly, both the rAcCoAP and the rGlyP are characterized by a 'flat' thermodynamic profile 24,25 ; that is, both are mostly reversible such that the direction of the metabolic flux they carry is determined mainly by the concentrations of their substrates and products. This thermodynamic profile, although constraining the driving force of the pathway reactions 26 , indicates very high energetic efficiency, where no energetic input, for example, in the form of ATP hydrolysis, is wasted. Indeed, both pathways are associated with a very low ATP cost: only 1-2 ATP molecules are invested in the metabolism of formate to pyruvate 24 . Yet, unlike the rAcCoAP, the key enzymatic components of which are highly oxygen sensitive, the rGlyP can operate under aerobic conditions. Hence, the rGlyP represents the most efficient theoretical route-in terms of energy utilization, resource consumption and biomass yield-to assimilate formate in the presence of oxygen 24 .  A recent study suggests that the complete rGlyP might be naturally operating in a phosphite-oxidizing microbe 27 . Moreover, the key enzymatic conversion of the rGlyP, catalyzed by the glycine cleavage system (GCS), was shown to be fully reversible in many organisms [28][29][30] . Previous studies demonstrated that the GCS can support glycine and serine biosynthesis from formate in an engineered E. coli strain at elevated CO 2 concentration 31,32 . However, growth of the bacterium on formate (and CO 2 ) has not yet been demonstrated and remains an open challenge.
Modular engineering establishes growth on formate. To facilitate the establishment of formatotrophic growth, we divided the rGlyP into four metabolic modules ( Fig. 1 and Supplementary Fig. 1): (1) a C 1 module (C 1 M), consisting of formate-THF ligase, methenyl-THF cyclohydrolase and methylene-THF dehydrogenase, all from Methylobacterium extorquens 33 , together converting formate into methylene-THF; (2) a C 2 module (C 2 M), consisting of the endogenous enzymes of the GCS (GcvT, GcvH and GcvP), that condenses methylene-THF with CO 2 and ammonia to give glycine; (3) a three-carbon (C 3 ) module (C 3 M), consisting of serine hydroxymethyltransferase (SHMT) and serine deaminase, together condensing glycine with another methylene-THF to generate serine and finally pyruvate; and (4) an energy module (EM), which consists of formate dehydrogenase (FDH) from Pseudomonas sp. (strain 101) 34 , generating reducing power and energy from this C 1 feedstock.
Our strategy was to establish the activities of the different modules in consecutive steps, integrating subsequent modules and selecting for their combined activity. We started with an E. coli strain that is auxotrophic for serine, glycine and C 1 moieties (ΔserA Δkbl ΔltaE ΔaceA), where the first deletion abolishes native serine biosynthesis, the second and the third abolish threonine cleavage to glycine and the final deletion prevents the formation of glyoxylate that could potentially be aminated to glycine 32 . The combined activity of the C 1 M and the C 2 M, together with the native activity of SHMT, should enable the cell to metabolize formate into C 1 -THF, glycine and serine, relieving these auxotrophies (Fig. 2a).
Into the serine auxotroph strain, we introduced the enzymes of the C 1 M and the C 2 M, either on plasmids or in the genome (Supplementary Fig. 1). For genome integration of C 1 M, we combined all relevant enzymes into one operon, under the regulation of a strong constitutive promoter 35 , which was inserted into a genomic 'safe spot' , SS9 (ref. 36 ). In the case of the C 2 M, we replaced the native promoter of the GCS with a strong constitutive one (Supplementary Fig. 1), increasing transcript levels 20-50-fold ( Supplementary  Fig. 2). As expected, growth with formate was observed upon overexpression of both modules (Fig. 2b) and was dependent on high CO 2 concentration (10% in the headspace) which thermodynamically and kinetically supports the reductive activity of the GCS. While genomic integration of the enzymes of the C 1 M (gC 1 M) did not improve growth compared with plasmid expression (pC 1 M), replacing plasmid-borne expression of the enzymes of the C 2 M (pC 2 M) with genomic overexpression (gC 2 M) supported a higher growth rate (Fig. 2b).
Next, we aimed to establish formate as the primary carbon source, which requires high expression of the enzymes of the C 3 M to convert glycine into the central metabolism intermediate pyruvate (Fig. 2c). To enable formate assimilation to biomass, an energy source is required, which at this stage we chose to be acetate. The tricarboxylic acid (TCA) cycle can fully oxidize acetate to generate reducing power and energy, while the deletion of isocitrate lyase (ΔaceA) abolishes the activity of the glyoxylate shunt, thus preventing the cell from using acetate as a carbon source. Growth should therefore be dependent on formate assimilation via the rGlyP for biomass generation and acetate oxidation for the production of reducing power and energy (Fig. 2c).
The enzymes of the C 3 M were either overexpressed on a plasmid (pC 3 M) or in the genome (gC 3 M) ( Supplementary Fig. 1); in the latter case, the native glyA and sdaA were deleted and a synthetic operon harboring both genes under the regulation of a strong constitutive promoter was introduced into another genomic 'safe spot' , SS7 (ref. 36 ). Overexpression of the enzymes of the C 3 M, within a strain that genomically expresses the enzymes of the C 1 M and the C 2 M, resulted in growth on formate and acetate (at 10% CO 2 ) (Fig. 2d). Genomic expression of C 3 M supported more robust growth compared with the C 3 M expressed from a plasmid. To confirm that the expression level of C 3 M does not constrain the growth rate, we tested a strain in which the expression of glyA and sdaA is controlled by a stronger ribosome binding site (RBS-A instead of RBS-C (ref. 35 )). We found this strain to grow rather poorly ( Supplementary Fig. 3), indicating that higher expression of these genes is deleterious.
Finally, we introduced the EM such that formate can serve as sole carbon and energy source (Fig. 2e). Overexpression of FDH on a plasmid ( Supplementary Fig. 1), in the strain carrying the genes of the C 1 M, C 2 M and C 3 M in the genome, enables growth on formate ( Supplementary Fig. 4). However, when we introduced FDH into yet another genomic 'safe spot' , SS10 (ref. 36 ), we failed to establish growth ( Supplementary Fig. 4), suggesting that the expression level of FDH was too low. We therefore tested a strain in which the genomic expression of FDH was controlled by a stronger ribosome binding site (RBS-A instead of RBS-C (ref. 35 ); Supplementary  Fig. 1). This strain, carrying no plasmid, was able to grow on formate as a sole carbon and energy source ( Fig. 2f and Supplementary Fig. 4).

Short-term evolution improves growth on formate.
To improve growth on formate we decided to conduct a short-term evolution experiment in fed-batch mode. We cultivated the engineered strain in test tubes, where formate was added every 3-6 d, increasing the concentration in the medium by 30 mM (Fig. 3a). Once cell turbidity reached an optical density at 600 nm (OD 600 ) of 0.4, we diluted the cells to OD 600 of 0.03-0.05 and started a new cycle of cultivation ( Fig. 3a shows six typical cycles).
Within 13 cultivation cycles (≤40 generations), growth rate on formate was substantially improved (Fig. 3a), with the doubling time dropping from 65-80 h in the first two cycles to less than 10 h in the last cycle (Fig. 3b). This growth rate is at least double that of a recently reported E. coli strain growing on formate via an engineered Calvin cycle 37 . The short-term evolution also improved the growth yield on formate, from ~1.5 gCDW per mol-formate in the first cycle to 2.3 ± 0.2 gCDW per mol-formate. This yield is similar to that of microorganisms growing autotrophically on formate via the Calvin cycle (3.2 ± 1.1 gCDW per mol-formate (ref. 38 )). The growth of the evolved bacterium on formate was directly coupled to a decrease in the concentration of the feedstock in the medium (Fig. 3c). Furthermore, as formatotrophy consumes protons (net oxidation and net assimilation both consume formic acid rather than formate), we observed a direct correlation between cell density and the pH of the medium ( Supplementary Fig. 5).
To better characterize growth on formate, we conducted growth experiments in 96-well plates, automatically measuring OD 600 every ~10 min. We found that maximal cell density increased monotonically with increasing formate concentration from 10 mM to 150 mM (Fig. 3d). Similarly, the doubling time decreased monotonically with increasing formate concentration: from 17 h with 10 mM formate to less than 8 h at formate concentrations higher than 100 mM (Fig. 3d). The cellular toxicity of formate, which is attributed to inhibition of respiratory proteins 39 and dissipation of the proton motive force 40 , probably explains the increased lag time at formate concentrations of 109 mM and 153 mM, and the failure to grow at higher concentrations.
Adaptive laboratory evolution usually requires hundreds of generations to improve the fitness of E. coli in a substantial way 41,42 . Our strain required less than 40 generations, presumably as the growth of the parent strain was so poor that a small number of mutations were sufficient to drastically improve fitness. To check whether this was indeed the case, we isolated multiple colonies of the evolved strain and sequenced their genomes. We found two mutations that occurred in all sequenced colonies ( Supplementary Fig. 6). The first was a single base-pair substitution in the 5' untranslated region (UTR) of the newly introduced FDH gene, which increased the level of transcript 2.5-fold ( Supplementary Fig. 7) and resulted in a 7.4-fold increase in formate oxidation activity in cell extract assays ( Supplementary Fig. 8). The second mutation was a single base-pair substitution in the promoter region of pntAB, which encodes for the membrane-bound transhydrogenase. This mutation increased transcript level by more than 13-fold ( Supplementary Fig. 7).  Genetic background: ∆serA ∆kbl ∆ltaE ∆aceA (∆glyA ∆sdaA) The beneficial effect of these two mutations is to be expected, as the first increases energy supply to the cell from formate and the second increases the availability of NAPDH, a key cofactor for the activity of the rGlyP (consumed by methylene-THF dehydrogenase), the supply of which apparently limits pathway activity.
To confirm that the two mutations suffice to support the improved growth on formate, we used multiplex automated genome engineering (MAGE) 43 to introduce these mutations into a nonevolved strain. We found that while the parent strain could hardly grow in 96-well plates, the strain in which the two mutations were present displayed a growth profile almost identical to that of the evolved strain ( Supplementary Fig. 9). We therefore concluded that overexpression of FDH and PntAB was sufficient to enable the observed improved growth on formate. By further optimizing cultivation conditions, we found that addition of 100 mM sodium bicarbonate to the medium enabled the evolved strain, as well as the reconstructed strain, to grow at higher formate concentrations, tolerating even 300 mM ( Supplementary Fig. 10). The increased tolerance to formate might be attributed to a higher buffer capacity of the medium containing bicarbonate, possibly decreasing fluctuations in local pH due to formate consumption.
Carbon labeling sheds light on cellular fluxes. To confirm that growth on formate indeed proceeds via the rGlyP, we performed carbon-labeling experiments. We fed the cultures with 13 C-formate/ 12 CO 2 , 12 C-formate/ 13 CO 2 and 13 C-formate/ 13 CO 2 , and measured the labeling pattern of proteinogenic amino acids using liquid chromatography-mass spectrometry. We focused on seven amino acids-glycine, serine, alanine, valine, proline, threonine and histidine-that either directly relate to the activity of the rGlyP or Experiments were conducted at 10% CO 2 . Plate reader experiments were performed in triplicates, which displayed identical growth curves (±5%), and hence were averaged. The experiment (in triplicates) was repeated three times, which showed highly similar growth behavior. DTs are shown in the figure.
DTs were considerably shorter in the plate reader than in a test tube as the measurements were more accurate (taken every 10 min rather than once per day) and the conditions were different (for example, more stable cultivation environment in the plate reader).
originate from different parts of central metabolism, thus providing an indication of key metabolic fluxes. As shown in Fig. 4, the amino-acid labeling confirms the activity of the rGlyP. Specifically, feeding 13 C-formate/ 12 CO 2 resulted in singly labeled glycine and doubly labeled serine and pyruvate (as indicated by the labeling of alanine). As valine-derived from two pyruvate molecules, one of which loses its carboxylic acid carbonis mostly quadruply labeled, we deduce that pyruvate is labeled in its two noncarboxylic carbons, as predicted for growth via the rGlyP ( Supplementary Fig. 11). Conversely, feeding 12 C-formate/ 13 CO 2 resulted, as expected, in singly labeled glycine, serine and pyruvate. As valine is also singly labeled in this condition, we deduce that pyruvate is labeled in its carboxylic carbon, again confirming the activity of the rGlyP (Supplementary Fig. 11). On feeding 13 C-formate/ 13 CO 2 , all seven amino acids were nearly completely labeled, where the overall fraction of labeled carbon (marked in italics above the bars in Fig. 4) is 97-98%, as expected by feeding with 99% 13 C-labeled formate and 99% 13 C-labeled CO 2 .
The labeling of threonine (derived from oxaloacetate) and proline (derived from 2-ketoglutarate) sheds light on the flux via the anaplerotic reactions and the TCA cycle. Specifically, if cyclic flux via the TCA cycle predominates over anaplerotic flux, threonine and proline would be expected to be almost fully labeled on feeding with 13 C-formate and almost fully unlabeled when feeding with 13 CO 2 ( Supplementary Fig. 11). Conversely, if anaplerotic flux and noncyclic flux predominate over the cyclic flux, then threonine would be expected to be mostly doubly labeled on either 13 C-formate or 13 CO 2 , and proline would be expected to be mostly quadruply labeled on 13 C-formate and singly labeled on 13 CO 2 ( Supplementary Fig. 11). The results shown in Fig. 4 are thus consistent with high anaplerotic flux and low cyclic flux. This indicates that the cell obtains sufficient reducing power and energy from formate oxidation via FDH, and hence does not wastefully oxidize the assimilated carbons within pyruvate and acetyl-CoA (that is, investing cellular resources for C 1 assimilation, only to completely oxidize the assimilated product).
Engineered growth of E. coli on methanol. Next, we aimed to use the rGlyP for methanol assimilation. A single enzyme, MDH, can convert methanol to formaldehyde, which can be oxidized to formate by the endogenous glutathione system 44 (Fig. 5a). The expression of MDH can thus be regarded as the introduction of another module-a methanol module-that serves to metabolize methanol to formate, while providing the cells with reducing power (Fig. 5b). We tested NAD-dependent MDHs from several organisms: Bacillus stearothermophilus (BsMDH) 19 , Corynebacterium glutamicum (CgMDH) 45 and Cupriavidus necator N-1 (CnMDH, wild type mdh2) 46 , as well as two MDHs from Bacillus methanolicus (BmMDH2 and BmMDH3) 10,47 and an improved variant (BmMDH2*, carrying Q5L A363L modifications) 47 . These MDH variants were expressed on plasmids in three genetic backgrounds: the parent strain (gC 1 M gC 2 M gC 3 M gEM), the evolved strain and the parent strain to which the mutation within the promoter of pntAB was introduced via MAGE. Overexpression of BsMDH supported growth on 600 mM methanol, which was most efficient in the latter strain (Fig. 5c) and somewhat poorer in the other strains (Fig. 5d). The other MDH variants failed to support growth (Fig. 5d, final OD 600 not higher than inoculation, as indicated by the brown dashed line).
To confirm that growth on methanol indeed depends on formaldehyde oxidation via the glutathione system, we deleted the endogenous gene encoding for S-(hydroxymethyl)glutathione dehydrogenase (∆frmA) in the above strains. We found that this deletion completely abolished growth on methanol (Fig. 5d), confirming the essentiality of the glutathione system to the observed growth. Moreover, overexpression of NAD-dependent formaldehyde dehydrogenase from Pseudomonas putida (PpFADH), as demonstrated in a previous study 12 , or from Pseudomonas aeruginosa (PaFADH (ref. 48 )) did not improve growth on methanol (Fig. 5d), indicating that the endogenous glutathione system is sufficiently fast and that the rate-limiting step lies in methanol oxidation.
To confirm that growth on methanol indeed proceeds via the rGlyP, we performed a carbon-labeling experiment. We fed the cultures with 13 C-methanol/ 12 CO 2 and measured the labeling pattern of the proteinogenic amino acids glycine, serine, alanine, valine, proline, threonine and histidine. The labeling pattern we measured (Fig. 5e) was essentially identical to that observed with 13 C-formate/ 12 CO 2 (Fig. 4), confirming that growth on methanol takes place via the synthetic route.
Notably, the growth rate on methanol was considerably lower than that on formate-doubling time of 54 ± 5.5 h. This can be attributed to the slow rate of methanol oxidation. The observed biomass yield was 4.2 ± 0.17 gCDW per mol-methanol, considerably lower than that of microorganisms naturally growing on methanol (7.2 ± 1.2 gCDW per mol-methanol via the Calvin cycle, 12 ± 1.6 gCDW per mol-methanol via the serine cycle and 15.6 ± 2.7 gCDW per mol-methanol via the ribulose monophosphate cycle 38 ). We speculate that the low yield is also related to the slow rate of methanol oxidation: a low growth rate increases the proportional consumption of energy for cell maintenance, thus lowering biomass yield. Addition of 100 mM sodium bicarbonate increased the final OD 600 , but the growth parameters did not improve: doubling time of 55 ± 1 h and biomass yield of 4.2 ± 0.1 gCDW per mol-methanol ( Supplementary Fig. 12, also showing methanol consumption during growth).

Discussion
This study demonstrates synthetic formatotrophy and methylotrophy. We show that rational design alone can suffice to achieve such a goal, but that short-term evolution can provide useful fine-tuning to improve growth characteristics. Further improvement of growth on C-formate, 12 Fig. 4 | Labeling pattern of proteinogenic amino acids confirms the activity of the rGlyP. as elaborated in Supplementary Fig. 11, the labeling pattern is consistent with the assimilation of formate and CO 2 via the synthetic pathway, and indicates low cyclic flux via the TCa cycle. Numbers written in italics above the bars correspond to the overall fraction of labeled carbons.
formate and methanol can be achieved via long-term evolution or via the introduction of metabolic routes that bypass limiting reactions. For example, replacing NAD-dependent MDH with methanol oxidase might reduce biomass yield (as this enzyme dissipates reducing power) but could support a much higher growth rate as it replaces a thermodynamically and kinetically limited reaction with a favorable and fast one. We recently used computational analysis to compare different C 1 assimilation pathways according to the biomass and product yields they are expected to support on formate and methanol 49 . For formate assimilation, we found that the rGlyP has the potential to outperform its natural and synthetic counterparts in terms of both biomass and product yields. With regard to methanol assimilation, the ribulose monophosphate cycle supports the highest biomass yield. However, this pathway is outperformed by the rGlyP for the production of the key metabolic precursors acetyl-CoA and pyruvate. This is attributed to the overflow of reducing power in the ribulose monophosphate cycle, while the rGlyP pathway uses CO 2 as an electron sink 49 . Overall, the rGlyP seems to be the most flexible C 1 assimilation pathway, with the potential to support the highest yields of acetyl-CoA and pyruvate using either formate or methanol as feedstocks 49 . However, reaching the full potential of the rGlyP would require considerable growth optimization via rational design and adaptive laboratory evolution.
The C 1 assimilating strains can be further engineered for the production of value-added chemicals. Especially interesting are chemicals that can be derived directly from the rGlyP intermediates or product, and can thus be produced with high yield and productivity. For example, lactate and isobutanol, both of which are derived from pyruvate, should be produced with high yield. Similarly, cysteine, which is derived from serine, a key pathway intermediate, might be an ideal product. Coupling the abiotic synthesis of formate and methanol with their microbial conversion to chemicals of interest will enable an integrated process for the valorization of CO 2 into renewable commodities.

Online content
Any Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review  Supplementary Fig. 6), enabled growth on methanol within test tubes. Experiments were conducted at 10% CO 2 . Dots correspond to triplicate experiments (using biologically independent strains) and the line to their average. d, Comparison of growth on methanol (shown are final cell densities) with different expressed enzymes and on different genetic backgrounds. We tested NaD-dependent MDH from several organisms: B. stearothermophilus (BsMDH), C. glutamicum (CgMDH) and C. necator N-1 (CnMDH), as well as two MDHs from B. methanolicus (BmMDH2 and BmMDH3) and an improved variant (BmMDH2*, carrying Q5L a363L modifications). We further tested formaldehyde dehydrogenases from P. putida (PpFaDH) and P. aeruginosa (PaFaDH). Experiments were conducted in (biologically independent) duplicates; dots show the measured OD 600 values and bars correspond to the averages. N, not determined. e, Labeling pattern of proteinogenic amino acids on feeding with 13 C-methanol/ 12 -CO 2 is identical to that with 13 C-formate/ 12 -CO 2 (Fig. 4), confirming the activity of the rGlyP. Numbers written in italics above the bars correspond to the overall fraction of labeled carbons.

Articles
Nature ChemiCal Biology methods Chemicals and reagents. Primers were synthesized by Integrated DNA Technologies. PCR reactions were carried out either using Phusion High-Fidelity DNA Polymerase or Dream Taq. Restrictions and ligations were performed using FastDigest enzymes and T4 DNA ligase, respectively, all purchased from Thermo Fisher Scientific. Glycine, sodium formate, sodium formate-13 C and methanol-13 C were ordered from Sigma-Aldrich. 13 CO 2 was ordered from Cambridge Isotope Laboratories.
Genome engineering. Gene knockouts were introduced in MG1655 by P1 phage transduction 51 . Single-gene knockout mutants from the National BioResource Project (National Institute of Genetics, Japan) 52 were used as donors of specific mutations. For the recycling of selection marker (as multiple gene deletions and integrations were required) all of the antibiotic cassettes integrated into the genome were flanked by flippase recognition target (FRT) sites. Cells were transformed with a flippase recombinase helper plasmid (FLPe, replicating at 30 °C, Gene Bridges), which carries a gene encoding FLP which recombines at the FRT sites and removes the antibiotic cassette. Elevated temperature (37 °C) was subsequently used to cure the cell from the FLPe plasmid.
Exchange of E. coli native promoter with a synthetic one was performed by using PCR-mediated λ-Red recombination method. The synthetic promoter fused with FRT-flanked kanamycin resistance gene was cloned into the pZ vector and the DNA fragment was obtained by PCR amplification with primers containing 50-base-pair homology for recombination. Recombinant E. coli MG1655 harboring λ-Red recombinase (pRed/ET, Gene Bridges) was cultivated at 30 °C, and the expression of λ-Red recombinase was induced by the addition of 10 mM l-arabinose. Electro-competent cells were prepared by washing three times with double-distilled water (ddH 2 O). The PCR product was introduced into E. coli expressing the λ-Red recombinase via electroporation. Mutants with exchanged promoter occurred via homologous recombination, were selected on the Luria Bertani (LB) agar plate containing 50 μg ml −1 kanamycin and were subsequently screened by colony PCR.
To enable genomic overexpression from a synthetic operon, conjugationbased genetic recombination methods was adapted as previously described 35 . The synthetic operons were digested with BcuI and NotI, and ligated by T4 ligase into pDM4 (with oriR6K) genome integration vector previously digested with the same enzyme. This vector has two 600 base-pair homology regions compatible with the target spot, chloramphenicol resistance gene (camR), a levansucrase gene (sacB) and the conjugation gene traJI for the transfer of the plasmid. The resulting ligation products were used to transform chemically competent E. coli ST18 strains. Positive clones growing on chloramphenicol medium supplemented with 5-aminolevulinic acid (50 μg ml −1 ) were identified by colony PCR, and the confirmed recombinant ST18 strain was used as donor strain for the conjugation. Chloramphenicol-resisting recipient E. coli strains were screened as positive strains for the first round of recombination. Subsequently, sucrose counter selection and kanamycin resistance tests were carried out to isolate recombinant E. coli strains with the correct synthetic operon integration into the chromosome. All constructs were verified via PCR and sequencing.
Introducing point mutations on the genome-to establish the mutation shown in Supplementary Fig. 6-was achieved by using MAGE 43,53 . A single colony of desired strain(s) transformed with pORTMAGE 53 (Addgene catalog no. 72680) was incubated in LB medium supplemented with 100 mg l −1 ampicillin at 30 °C in a shaking incubator. To start the MAGE cycle, overnight cultures were diluted by 100 times in the same medium and cultivated to an OD 600 of 0.4-0.5. Then, 1 ml of each culture was transferred to sterile microcentrifuge tubes, and transferred to the 42 °C thermomixer (Thermomixer C, Eppendorf) to express λ-Red genes by heat shock for 15 min at 1,000 r.p.m. After induction, cells were quickly chilled on ice for at least 15 min, and then made electrocompetent by washing three times with ice-cold ddH 2  Electroporation was done on a Gene Pulser XCell (Bio-Rad) set to 1.8 kV, 25 μF capacitance and 200 Ω resistance for a 1-mm gap cuvette. Immediately after electroporation, 1 ml of LB was added to the cuvette and the electroporation mixes in LB were transferred to sterile culture tubes and cultured with shaking at 30 °C, 240 r.p.m. for 1 h to allow for recovery. After recovery, 2 ml of LB medium supplemented with ampicillin was added and then further incubated in the same conditions. When the culture reached an OD 600 of 0.4-0.5, cells were either subjected to additional MAGE cycles or analyzed for genotype via PCR and sequencing. We performed eight consecutive MAGE cycles before analyzing the genotype to identify strains carrying the required mutations.
All strains used are shown in Supplementary Table 1.

13
C labeling of proteinogenic amino acids. For stationary isotope tracing of proteinogenic amino acids, cells were cultured in 4 ml of M9 medium supplemented with either labeled or unlabeled carbon sources; that is, 13 C-formate, 13 C-methanol and/or 13 CO 2 . A 6-l vacuum desiccator (Lab Companion) was used for cultures grown in 13 CO 2 , where the original gas was expelled by vacuum pump followed by refilling with 90% air and 10% 13 CO 2 . The cell was collected by centrifugation for 3 min at 18,407g when the stationary growth phase was reached. Biomass was hydrolyzed by incubation with 1 ml of 6N hydrochloric acid for a duration of 24 h at 95 °C. Samples were dried via heating at 95 °C and re-dissolved in 1 ml of ddH 2 O. Hydrolyzed amino acids were separated using ultra-performance liquid chromatography (Acquity, Waters) using a C18-reversed-phase column (Waters) as previously described 57 . Mass spectra were acquired using an Exactive mass spectrometer (Thermo Fisher). Data analysis was performed using Xcalibur (Thermo Fisher). Before analysis, amino-acid standards (Sigma-Aldrich) were analyzed under the same conditions to determine typical retention times.

Dry weight analysis.
To determine the dry cell weight of E. coli grown on formate or methanol, precultures were inoculated at a final OD 600 of 0.01 into fresh M9 medium containing either formate (30 mM) or methanol (600 mM) in 125-ml Enzymes and chemical assays. Absorbance changes for all assays were monitored in a BioTek Epoch 2 plate reader. We confirmed working at the measurement linear range in all assays. Results represent averages of at least three cell preparations.
To determine the activity of FDH, 1.5 ml of OD 600 1.0 cell culture grown in M9 minimal medium and supplemented with glucose and formate from glass test tubes was washed twice with 9 g l −1 sodium chloride. Cells were lysed by adding CelLytic Reagent (Sigma) and allowed to sit for 20 min at room temperature. After cell disruption, cellular debris was removed by centrifugation (18,407g, 4 °C, 10 min) and the supernatant was used for crude assays without further purification. FDH assay was performed in the presence of 10 mM 2-mercaptoenthanol, 100 mM sodium formate, 200 mM sodium phosphate buffer pH 7.0 and 2 mM NAD + in a total volume of 200 μl at 37 °C (ref. 58 ). The increase in NADH concentration resulting from formate oxidation was monitored at 340 nm. Protein concentration was measured using the Bradford Reagent (Sigma) with BSA as a standard. Formate and methanol in the culture were quantified by a colorimetric assay using a formate assay kit (Sigma-Aldrich) and a methanol assay kit (BioVision), respectively. All samples were diluted to ensure the readings were within the standard curve range according to the manufacturer's instructions.
Quantitative PCR. Total RNA was extracted from 1 ml of overnight culture at an OD 600 of 0.5 using the RNeasy Mini Kit (Qiagen), and following the protocol of the supplier. All RNA samples were treated with DNase I (Sigma-Aldrich) to remove any residual DNA. First-strand complementary DNA was synthesized using a qScript cDNA Synthesis kit following the manufacturer instructions (Quanta Biosciences), and 1 μg of total RNA was used as a template in a 20-μl reaction volume. Quantitative PCR with reverse transcription was performed using a Maxima SYBR Green qPCR Master Mix (Thermo Fisher Scientific) supplemented with 5 μM primers and 5 μl of cDNA template, which was diluted up to 200 μl after synthesis. The primers used for quantitative PCR were: 5´-GCC AAT CTG CAA CAG TGC TC-3´ (pntA_forward), 5´-TTT TTG GCT GGA TGG CAA GC-3´ (pntA reverse), 5´-CGT GAC GAA TAC CTG ATC GTT-3´ (fdh forward), 5´-GGT AGC GTT ACC TTT AGA GTA AGA GTG-3´ (fdh reverse). PCR was performed in 96-well optical reaction plates (Thermo Fisher Scientific) as follows: 10 min at 50 °C, 5 min at 95 °C, 40 cycles of 10 s at 95 °C and 30 s at 60 °C, and finally 1 min at 95 °C. The specificity of the reactions and the amplicon identities were verified by melting curve analysis. Reaction mixtures without cDNA were used as a negative control. Data were evaluated using the delta-delta Ct method 59 and with correction for the PCR efficiency, which was determined based on the slope of standard curves. Normalization of gene expression levels was carried on using the rrsA gene 60 , and eventually the fold-differences in the transcript levels and mean standard error were calculated as described before 59 .
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Complete information on the experimental setup as well as detailed results are available from the corresponding author upon reasonable request.

Code availability
MATLAB code used for the analysis of the experiments is available from the corresponding author upon request.