All strains and plasmids used in this study are listed in Table 1. E. coli DH5α was used for routine cloning. The starting E. coli SIJ488 was derived from E. coli MG1655 carrying genome-integrated gene deletion machinery (Jensen et al., 2015). High-copy expression vector pTrc99a was used for the construction of the hybrid methanol assimilation pathway.

To construct plasmid pTrc99a-mdh-das for the hybrid methanol assimilation pathway, mdh gene encoding methanol dehydrogenase from Acinetobacter garner and das gene encoding dihydroxyacetone synthase from Pichia angusta were codon-optimized and synthesized with a consensus RBS (AAGAAGGAGATATAC) under the control of the Trc promoter and inserted into the restriction site of EcoRI/SmaI of pTrc99a. To construct plasmid pTrc99a-mdh-das-antigapA, a fragment containing high-performing MicF M7.4 Hfq binding site (Hoynes-O’Connor and Moon, 2016) with the target binding region (TBR) of gapA gene was inserted into plasmid pTrc99a-mdh-das under the control of the Trc promoter. TBR sequence was designed to be complementary to the initiation codon (AUG) and extended 27 nucleotides into the coding region of gapA gene. Gibson assembly was used for all plasmid construction following the standard procedure (Gibson et al., 2009).

Gene knockout in E. coli SIJ488 was based on the lambda Red recombineering system as described by Jensen (Jensen et al., 2015). All of the primers and synthesized gene sequences used in this study are listed in Supplementary Tables S2, S3.

Strains were cultivated in M9 minimal medium with 400 mM methanol, 1 g/L yeast extract, and 100 μg/mL ampicillin at 37°C and 200 rpm unless otherwise stated. The M9 minimal medium contains: Na₂HPO₄⋅12H₂O 17.1 g/L, KH₂PO₄ 3 g/L, NH₄Cl 1 g/L, NaCl 0.5 g/L, 2 mM MgSO₄⋅7H₂O, 0.1 mM CaCl₂, trace element solution 250 μL/L. The trace element solution contained FeCl₃⋅6H₂O 1.62 g/L, ZnCl₂ 0.13 g/L, CoCl₂⋅6H₂O 0.2 g/L, Na₂MO₄⋅2H₂O 0.2 g/L, CuCl₂⋅6H₂O 0.09 g/L, and H₃BO₃ 0.05 g/L. If necessary, an amino acid mix solution was additionally added, consisting of (per liter of final culture medium): L-arginine hydrochloride, 72.5 mg; L-cystine 24.0 mg; L-histidine hydrochloride 42 mg; L-isoleucine 52.4 mg; L-leucine 52.4 mg; L-lysine hydrochloride, 126.4 mg; L-methionine,15.1 mg; L-phenylalanine, 33 mg; L-threonine, 47.6 mg; L-tryptophan, 10.2 mg; L-tyrosine, 36 mg; L-valine 46.8 mg. The expression of the hybrid methanol assimilation pathway was induced by adding 0.1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) initially.

Adaptive laboratory evolution of strain X5 was performed by serial transfers with inoculation of 2% (v/v) cultures every 2–4 days. Medium for evolution was a mixture of M9 minimal medium and Hi-Def Azure (HDA, Teknova) (Bennett et al., 2020b; Chen et al., 2020) with 400 mM methanol, 0.1 mM IPTG, and 100 μg/mL ampicillin. Strain X5 was first cultivated in the 100% HDA medium at 37°C and 200 rpm. After being cultivated to the stationary phase, the cultures were used as seed cultures to inoculate fresh medium containing 75% HDA and 25% M9 minimal medium. From passage 2 to passage 5 (liquid transfer cycles), the ratio of HDA was further reduced from 75% to 30%. From passage 6 to passage 11, cell cultures were continuously transferred in 15% HDA and 85% M9 minimal medium. Starting from passage 12, cell growth was achieved in 5% HDA and 95% M9 minimal medium, and the subsequent evolution was continued with the same medium. At passage 28, a colony was isolated and termed Ev17 strain for further studies.

The cell concentration was determined at an optical density of 600 nm (OD₆₀₀). The percent increase of biomass caused by methanol was calculated as Eq. 1 (Gonzalez et al., 2018). f i n a l   O D 600 w i t h   m e t h a n o l −   f i n a l   O D 600 w i t h o u t   m e t h a n o l f i n a l   O D 600 w i t h o u t   m e t h a n o l − i n i t i a l   O D 600   o f   c u l t u r e * 100 % (1)

Quantification of methanol concentration was carried out by using High-performance liquid chromatography (HPLC) equipped with an Aminex HPX-87H Column (300 × 7.8 mm) using 5 mM sulfuric acid as the mobile phase with a flow rate of 0.8 mL/min at 65°C. The consumption of methanol has been adjusted by subtracting the evaporation of methanol in the medium without cells. The concentration of formaldehyde was quantified by the Nash reaction (Woolston et al., 2018). 125 μL of cells supernatant was mixed with 125 μL Nash reagent (5 M ammonium acetate, 50 mM acetylacetone). The mixtures were incubated at 37°C for 1 h and measured at 412 nm. Formaldehyde standard solution needs to be prepared fresh daily and the standard curve was in the range from 0 to 100 μM.

To carry out the ¹³C-methanol isotopic analysis for intracellular metabolites and amino acids, cells were cultured in M9 minimal medium with 400 mM ¹³C-methanol and 1 g/L yeast extract for 48 h and prepared according to the protocols described by Long and Antoniewicz (Long and Antoniewicz, 2019). Samples were injected into the UHPLC-Q-Orbitrap liquid chromatography-mass spectrometry system (Thermo Fisher Scientific, United States ). Intracellular metabolites and amino acids were separated on Waters BEH Amide Column (2.1 × 100 mm; 1.7 µm) at 35°C with solvent A (acetonitrile with 0.1% formic acid) and solvent B (deionized water with 0.1% formic acid and 10 mM ammonium acetate) as the mobile phase with a flow rate of 0.3 ml/min. The elution gradient was (% B): 0–5 min at 0%; 5–6 min at 0%–25%; 6–15 min at 25%; 15–16 min at 25%–50%; 16–25 min at 50%; 25–26 min, 0%; 26–30 min, 0%. Mass spectrometry was configured with heated electrospray ionization (HESI) and operated in both positive and negative ion modes in full MS scan mode. Isotopologue fractions (IFs) and average carbon labeling were calculated as Eqs 2, 3 (Keller et al., 2020). ¹³C metabolic tracer analysis was obtained from three biological replicate measurements. I s o t o p o l o g u e   f r a c t i o n s = m i ∑ j = 0 n m j (2) A v e r a g e   c a r b o n   l a b e l i n g = ∑ j = 0 n m j n ∗ ∑ j = 0 n m j (3) Here, m i and m j are the abundance of a mass isotopologue, n is the number of carbon atoms of the metabolite.

DNA re-sequencing and mRNA sequencing were carried out on the Illumina HiSeq instrument by Azenta (Suzhou, China). Cells were cultivated for 48 h in M9 minimal medium with 400 mM methanol and 1 g/L yeast extract. For each sample, 200 μg genomic DNA or 1 μg total RNA was used for library preparation by standard protocols. For transcriptomics analysis, HTSeq (v0.6.1p1) was employed to estimate gene expression levels from the pair-end clean data. Differential expression analysis was by the DESeq2 Bioconductor package which is a model based on the negative binomial distribution. p-value of genes were settled <.05 and log   2 ⁡ f o l d   c h a n g e s ≥ 1.5 to detect differential expressed genes. For whole-genome sequencing, BWA (V0.7.17) was used to map clean data to reference genome. The raw datasets can be found in NCBI Bioproject database (accession number: PRJNA896773).

A hybrid methanol assimilation pathway with prokaryotic NAD-dependent Mdh and eukaryotic dihydroxyacetone synthase (Das) can convert methanol and Xu5P into glyceraldehyde 3-phosphate (GAP) and dihydroxyacetone (DHA) with the generation of NADH. DHA can be further converted into dihydroxyacetone phosphate (DHAP) by the endogenous DHA kinase of E. coli (Figure 1). Previously, De Simone et al. showed that a strain expressing mdh from A. garner and das from P. angusta exhibited a higher growth rate in minimal medium with methanol and xylose than that with xylose alone (De Simone et al., 2020). In that case, Xu5P was mainly derived from xylose, thus, it was not clear whether the cyclic XuMP pathway would be functional in sugar-free conditions. Thus, we first constructed the same methylotrophic module (mdh from A. garner and das from P. angusta) in plasmid pTrc99a and transformed the resulting plasmid pTrc99a-mdh-das into E. coli SIJ488 to generate strain X1. When cultured in M9 minimal medium with methanol and 1 g/L yeast extract, strain X1 showed 10.90 ± 0.01% higher final biomass concentration compared to that without methanol, indicating the hybrid methanol assimilation module was functional (Figure 2A). To enhance methanol assimilation, glutathione-dependent formaldehyde dehydrogenase gene (frmA) and S-formylglutathione hydrolase gene (frmB) were knocked out to reduce formaldehyde dissimilation into CO₂, generating strain X2. The methanol consumption by strain X2 and the increase of finial biomass concentration by methanol were similar to those of strain X1 (Figure 2B). When cultured with methanol, strains X1 and X2 accumulated 35.66 ± 2.43 µM and 68.78 ± 1.10 µM formaldehyde in 12 h (Figure 2C), suggesting that mdh from A. garner was successfully expressed in E. coli. Blocking the formaldehyde dissimilation pathway in strain X2 significantly increased formaldehyde accumulation but did not improve cell growth and methanol consumption (Table 2), suggesting an imbalance between formaldehyde formation and consumption.

Previous studies showed that the recombinant E. coli with Mdh and RuMP pathway cannot efficiently synthesize several amino acids (Bennett et al., 2020a). We hypothesized that the biosynthesis of some endogenous amino acids may also be limited for stains X1 and X2, which may affect cell growth and methanol consumption. When supplemented with an extra amino acid mix solution, strains X1 and X2 showed 52.1 ± 0.04% and 54.4 ± 0.05% increases of final biomass concentration by methanol respectively (Figures 2D, E). Moreover, the consumption of methanol by strains X1 and X2 was increased to 2.02 ± 0.01 g/L and 1.41 ± 0.22 g/L, which were both over 2-fold higher than those without amino acid mix solution (Figures 2D, E). The significant increase of methanol consumption and cell growth benefit by methanol confirmed that some amino acids or their intermediates cannot be efficiently synthesized in strains X1 and X2. It should be noted that the consumption of methanol by strain X2 was still lower than X1 even with the addition of amino acids mix solution. We hypothesized that it was due to the cyclic XuMP pathway was not active and the regeneration of Xu5P was not sufficient for formaldehyde assimilation. Thus, we decided to further improve the efficiency of the cyclic XuMP pathway by rational optimization.

To increase the efficiency of methanol assimilation, we tried to divert the metabolic flux toward the cyclic XuMP pathway. GAP and fructose-6-phosphate (F6P) are two key metabolic nodes linking the active XuMP cycle with the glycolysis pathway and pentose phosphate pathway. Balancing the supply and consumption of GAP and F6P is highly important for the functional XuMP cycle as well as for Xu5P regeneration and biosynthesis. Phosphofructokinase is a key enzyme catalyzing the irreversible phosphorylation of F6P to fructose-1,6-bisphosphate, diverting the intracellular pool of fructose-6-phosphate to the glycolysis pathway (Figure 1). Previous studies showed that the high activity of phosphofructokinase in E. coli tend to destabilize the cyclic RuMP system or CO₂ assimilation cycle (Antonovsky et al., 2016; Gleizer et al., 2019). Thus, we first tried to reduce the diversion of flux away from the XuMP cycle by knocking out pfkAB genes encoding 6-phosphofructokinase of strain X2. The resulting strain X3 did not show improved methanol consumption or cell growth on methanol compared to strain X2 (Figure 3A), indicating that deleting pfkAB genes alone is not sufficient to increase methanol assimilation. However, the accumulation of formaldehyde by strain X3 was significantly reduced compared to strain X2 (Figure 3D), suggesting an improved balance of formaldehyde formation and consumption.

We then tried to further increase the supply of GAP to support the functional XuMP cycle in strain X3. For the aerobic cultivation of E. coli on sugars, GAP is mainly diverted to the lower glycolysis pathway and then mainly consumed via TCA cycle. Natural methylotrophs such as Methylobacillus flagellatus and B. methanolicus often contain an incomplete or less active TCA cycle which may be beneficial for retaining a high flux to formaldehyde assimilation pathway (e.g., RuMP pathway) and reducing NADH generation (Müller et al., 2015). NADH is a kinetic inhibitor of methanol dehydrogenase and high generation of NADH via the TCA cycle may disturb the redox balance and reduce methanol assimilation (Meyer et al., 2018). Reduced metabolic flux toward the TCA cycle was also observed for synthetic methylotrophs of E. coli (Meyer et al., 2018; Chen et al., 2020). Thus, we knocked out the sucA gene encoding E1 subunit of the α-ketoglutarate dehydrogenase to down-regulate the TCA cycle of strain X3, generating strain X4. Strain X4 showed a lower final cell density than strain X3, however, methanol consumption by strain X4 was increased (Figure 3B). Especially, the percent increase of biomass by methanol was increased to 24.00 ± 0.01% (Table 2), indicating that deletion of sucA gene to reduce the TCA cycle activity was beneficial for methanol assimilation. To further increase methanol assimilation, we set out to adjust the flux distribution at GAP node. Previous studies showed that the high activity of glyceraldehyde 3-phosphate dehydrogenase would strongly divert the metabolic flux to the glycolysis pathway, destabilizing the cyclic formaldehyde assimilation pathway (Meyer et al., 2018; Chen et al., 2020). To reduce the activity of glyceraldehyde 3-phosphate dehydrogenase, we introduced the antisense RNA to inhibit the expression of gapA gene of strain X4, giving strain X5. With this modification, methanol consumption by strain X5 was increased to 1.01 ± 0.26 g/L, which was over 2-fold higher than strain X4 (Figure 3C). Moreover, the percent increase of biomass by methanol was increased to 31.6 ± 0.01% (Table 2). Although strain X5 had the highest methanol consumption rate among all of the engineered strains, the low accumulation of formaldehyde during the cultivation indicated an increased formaldehyde consumption via the assimilation pathway (Figure 3D). Thus, reducing the activity of glyceraldehyde 3-phosphate dehydrogenase was also highly important for the functional cyclic XuMP pathway.

Although strain X5 showed improved methanol utilization, the final cell density was still lower than strain X1. To promote cell growth of strain X5 on methanol, ALE was performed. Strain X5 was first cultivated in Hi-Def Azure (HDA) medium, a semi-minimal medium containing amino acids (Chen et al., 2020). The percent of HDA was gradually reduced and replaced by M9 minimal medium during the evolution. After 12 passages, the culture can grow on methanol with 5% HDA while strain X5 cannot grow on methanol with less than 15% HDA (Supplementary Figure S1). After 28 passages, a best growing single colony in the medium with 5% HDA was isolated and termed Ev17. When cultured in M9 minimal medium with methanol and 1 g/L yeast extract, the final cell density (OD₆₀₀) of strain Ev17 reached 0.71 ± 0.04 and the percent increase of biomass by methanol achieved 90.9 ± 0.01%, which was 1.8-fold higher than that of strain X5 (Figure 4). Moreover, strain Ev17 consumed 1.62 ± 0.23 g/L methanol, which was 60.4% higher than strain X5. Especially, when cultured with methanol and 0.1 g/L yeast extract, strain Ev17 achieved OD₆₀₀ 0.65 ± 0.016 and consumed 0.66 ± 0.11 g/L methanol while other unevolved strains only showed marginal methanol consumption and cell growth (Figure 5), indicating that methanol assimilation was significantly improved by ALE.

To further evaluate the impact of rational design and ALE on methanol assimilation, ¹³C-labeling analysis was performed with strains X1, X5, and Ev17. We cultured strains with ¹³C-methanol and 1 g/L yeast extract and measured the intracellular metabolites and amino acids using liquid chromatography-mass spectrometry (Figures 6, 7). Introducing the hybrid methanol assimilation module alone in strain X1 resulted in 6.87 ± 0.48% ¹³C-labeling of 2-phosphoglycerate/3-phosphoglycerate (2PG/3PG) and low but significant ¹³C-labeling of several important metabolites in glycolysis, TCA cycle, and pentose phosphate pathway (Figure 6). However, most of these labeled metabolites contained only one ¹³C atom (M1). Significant improvement of ¹³C-labeling of all measured metabolites was observed in both strains X5 and Ev17. For example, the average ¹³C-labeling of 2PG/3PG in strains X5 and Ev17 was increased by 49.1% and 138.6% compared with strain X1. Especially, the increase of ¹³C-labeling of Ru5P/Xu5P in strain X5 demonstrated the improved XuMP cycle by rational engineering. Strain X5 also exhibited the labeling of several metabolites with more than one ¹³C atom, such as 2PG/3PG, glucose-6-phosphate (G6P) and citric acid, which were caused by XuMP cycle running more than one time. All of the measured metabolites in strain Ev17 showed higher average ¹³C-labeling and had more than one ¹³C atom, confirming the significant improvement of methanol assimilation via ALE.

To confirm that methanol was involved in biosynthetic pathways, ¹³C-labeling analysis of intracellular amino acids was also carried out (Figure 7). Compared to strain X1, the significant increase in average labeling of glutamic acid (7.58 ± 0.50%), L-tyrosine (2.66 ± 0.61%) and L-serine (2.23 ± 0.47%) was detected in strain X5. Strain Ev17 displayed significant ¹³C-methanol incorporation into glutamic acid (11.50 ± 0.70%), aspartic acid (8.77 ± 0.86%), tyrosine (5.06 ± 0.92%), serine (4.61 ± 1.57%), and glutamine (4.57 ± 0.77%). These improvements supported that our strategies enhanced the biosynthesis of amino acids from methanol.

To analyze the change of global genes expression, transcriptomics analysis of strains X1, X5, and Ev17 was further performed (Figure 8). Comparing with strain X1, most of the genes in TCA cycle, including maldh, gltA, sucCD, sdhA, were down-regulated in strain X5, indicating that deletion of sucA successfully reduced the activity of TCA cycle. Owing to the introduction of antisense RNA, the expression of gapA and pyk (encoding pyruvate kinase) in strain X5 were also down-regulated, demonstrating that reducing the activities of glycolysis and TCA cycle is beneficial for the higher methanol assimilation (Figure 8A). Most of genes related to DHA and GAP assimilation, such as fbaAB (encoding fructose-bisphosphate aldolase), fbp (encoding fructose-bisphosphatase), and the dhaLMK operon (encoding dihydroxyacetone kinase), were significantly up-regulated in strain X5 and Ev17. However, the fructose-6-phosphate aldolase (FSA) pathway encoded by fsaAB operon in all three strains was not transcriptionally activated (Figure 8B). These results indicated that Xu5P regeneration was successfully enhanced mainly assimilated via the dihydroxyacetone kinase pathway which is consistent with a previous study by (Peiro et al., 2019). Besides, the up-regulation of talAB operon (encoding transaldolase), tktAB operon (encoding transketolase) in strain X5 and Ev17 also contributed to the Xu5P regeneration (Figure 8B).

We further performed the Whole Genome Sequencing (WGS) analysis of stain Ev17. No mutations were found in the plasmid and all mutations in the genome were listed in Supplementary Table S4. Several mutations on the transcriptional regulators such as narL, gntR, and iclR were identified. The mutations of global transcriptional factors may play important roles in methanol metabolism and cell growth. In addition, mutations on gltA and aceK genes were also identified, which probably alter the activity of glyoxylate pathway (due to the knockout of sucA gene). Mutations on genes related to amino acids biosynthesis, such as glnE (encoding glutamine synthetase), sdaB (encoding serine deaminase), and asnB (encoding asparaginase) were also identified. The function of these mutations and their relations to methanol assimilation should be clarified in the future.