A Synthetic Acetyl-CoA Bi-cycle Synergizes the Wood-Ljungdahl Pathway for Efficient Carbon Conversion in Syngas Fermentation

The Wood-Ljungdahl pathway (WLP) is a natural carbon fixation pathway capable of converting one-carbon (C1) compounds (CO 2 , CO, formate) to two-carbon (C2) metabolite acetyl-CoA or coordinating with canonical glycolysis to convert sugar feedstocks to acetyl-CoA with high carbon yield. The catalytic inefficiency and engineering difficulty in key enzymes, however, limit the biosynthetic potential of this pathway. Here we design a synthetic acetyl-CoA bi-cycle to synergize the WLP for efficient C2 metabolite synthesis. This pathway produces an acetyl-CoA by fixation of two CO 2 equivalents via three functional modules acting in series: carbon fixation, gluconeogenesis, and non-oxidative glycolysis. We examine the pathway through comprehensive in silico thermodynamic and kinetic analyses. The prototypic pathway is implemented in a syngas-fermenting Clostridium ljungdahlii DSM 13528 by expressing a heterologous phosphoketolase and coordinating with native enzymes in the host acetogen. We demonstrated the effectiveness of this synthetic pathway in carbon conversion under various growth conditions, which complements the WLP for valorization of syngas as well as sugar feedstocks with high catalytic efficiency. This study underscores the reductive acetyl-CoA bi-cycle as a practical strategy to improve carbon conversion and redox homeostasis in the acetogenic host.


Introduction
Syngas, a gas mixture consisting of carbon monoxide (CO), hydrogen (H2), and carbon dioxide (CO2) can be converted and upgraded by anaerobic acetogens 1 . As a standalone bioprocess or coupled with sugar conversion mixotrophically 2 , syngas fermentation provides an attractive strategy to valorize low-cost substrates, waste streams, and produce fuels as well as value-added chemicals. Compared to thermochemical processes, biological syngas conversion has numerous advantages including: a higher tolerance to impurities such as sulfur compounds in syngas, a wider range of usable H2, CO2, and CO mixtures, a lower operating-temperature and -pressure, and higher product yield and uniformity 3 . However, broad use of this bioprocess needs to overcome technical bottlenecks such as fundamental limits in cell metabolism and enzyme kinetics that could lead to catalytic inefficiency.
Syngas-fermenting bacteria utilize the Wood-Ljungdahl pathway (WLP), converting CO2/CO to acetyl coenzyme A (CoA), the key precursor of biomass components and fermentation products (acetate, ethanol, butanol, etc.). This carbon fixation pathway synthesizes acetyl-CoA (C2) through a CO-methylating acetyl-CoA synthase (ACS), which combines CO (C1) with CoA and a methyl group (C1, initially from CO2) to realize "C1 + C1 = C2". The apparent simplicity of this linear pathway, however, relies on the complex, interconnected enzymatic mechanisms that enable carbon fixation. ACS, for example, consists of three functionally diverse subunits that associate tightly in a complex with carbon monoxide dehydrogenase (CODH), and utilizes the product of the CODH reaction (CO) as its substrate in a kinetically coupled reaction linked to the generation of acetyl-CoA 4 . The complexity of the native C1/C2 biochemistry renders enzyme modification and pathway engineering a grand challenge. In addition, the WLP was identified with rate-limiting steps including formation of 10-formyl-tetrohydrofolate, a key intermediate in the methyl branch by the synthetase which occurs with a very low kcat value (1.4 s −1 for the enzyme from Moorella thermoacetica) 5 . To optimize syngas bioconversion, an engineering-amenable pathway with enzymes of superior catalytic kinetics is pursued. However, beyond the WLP, no such a natural pathway was found so far in syngas-fermenting bacteria capable of converting C1 compounds to acetyl-CoA.
In recent years, synthetic pathways have been proposed to reduce carbon loss during acetyl-CoA synthesis, including a non-oxidative glycolysis (NOG) 6 that can bypass the C3 decarboxylation step in the canonical glycolytic pathway. The NOG pathway converts two glyceraldehyde 3-phosphate (C3) into three molecules of acetyl-phosphate (C2) without carbon loss. Other carbon conservation solutions entail the use of native carbonfixing pathways, for example, reversed pyruvate:ferredoxin oxidoreductase (rPFOR) which is evolved in anaerobes for CO2 incorporation by converting acetyl-CoA to pyruvate (C3) 8 . However, whether and how these pathways could apply to syngas-fermenting bacteria for augmenting carbon fixation efficiency remain elusive.
Here we introduce a synthetic acetyl-CoA bi-cycle to circument the suboptimality of the WLP in C2 metabolites synthesis. The bicyclic pathway aims to synthesize an acetyl-CoA by fixation of two CO2 equivalents and to maximize the use of native enzymes in the host's metabolism. We first investigate the feasibility of the acetyl-CoA bi-cycle in silico and validate the functionality of essential pathway modules in a model syngas-fermenting chassis Clostridium ljungdahlii 7 . As a proof-of-concept, we engineer the prototypic acetyl-CoA bi-cycle in the bacteria and demonstrate the beneficial effects of coupling the bicycle with the WLP for efficient C2 metabolites synthesis.
Two of three generated acetyl-CoA can replenish the initial investment of the C2 metabolite. In terms of net stoichiometry, this pathway realizes C1 + C1= C2 (CO2 + CO2 = Acetyl-CoA) while bypassing ACS/CODH complex and other rate-limiting enzymes in the WLP. The key carbon-fixing enzyme in the pathway is rPFOR, which is native in many syngas-fermenting bacteria and can operate in the reductive direction 8,9 . Besides PFOR, the pathway contains 16 metabolic enzymes, 15 of which are native in acetogens as being found in C. ljungdahlii (Table S1). Coupled with a heterologous phosphoketolase that activates the NOG by converting sugar phosphates to acetyl phosphate (C2), the prototype of this bicyclic pathway may be built in vivo with minimal genetic modification of the host. In terms of energetic cost, the pathway can use reduced ferredoxin for CO2 fixation via rPFOR. Gluconeogenesis consumes NAD(P)H, and burns ATP to overcome thermodynamic barrier over the conversion of pyruvate to phosphoenolpyruvate.
Considering that gluconeogenesis is a native metabolic pathway and functions indispensably in homoacetogens for sugar synthesis from acetyl-CoA, no extra energetic expense will be exerted to the host when incorporating it into the bi-cycle. Acetogens can generate cellular reductants and ATP from syngas and various organic substrates. In a highly reductive syngas ambiance, acetogens produce reducing equivalents directly from H2 and CO through H2-uptake hydrogenase and CODH, respectively. Versatile phosphorylation mechanisms including an Rnf complex, a membrane ferredoxin: NAD oxidoreductase in C. ljungdahlii, contributes to H + -translocating ATP synthesis, hence balancing energetic demand 10 .
To expand substrate spectrum, functional variants of reductive acetyl-CoA bi-cycle are also designable. For instance, replacing rPFOR by a reversible pyruvate formate lyase (PFL), which is also native in C. ljungdahlii, could enable acetyl-CoA synthesis from formate (Fig.1b). In addition, conversion of methanol to acetyl-CoA could be feasible by incorporating two native enzymes in many acetogens: alcohol dehydrogenase (ADH) and aldehyde oxidoreductase (AOR) 11 that convert methanol to formate via formaldehyde.

Catalytic properties of the acetyl-CoA bi-cycle
Evaluating pathway properties for massive possible physiological scenarios is computationally feasible. We therefore examine the designed pathway in silico to understand its feasibility. We first modeled the Max-min driving force (MDF) 12,13 and sought to understand the pathway in a thermodynamic landscape. Even though this pathway has reversible reactions such as PFOR which could have high activity toward oxidative direction, optimization of metabolite concentrations in a physiological range enables all reactions thermodynamically favorable (ΔG' < 0). It indicates that the thermodynamic driving force can effectively drive the pathway toward the formation of acetyl-CoA (Fig.2). To evaluate the catalytic property of this pathway, we performed enzymatic protein cost analysis 12, 14 which takes both reaction thermodynamics and enzyme kinetics into account and calculates the theoretical lower bound of protein expense that affords the production of an acetyl-CoA unit. Interestingly, despite more reactions than in the WLP, the bi-cycle expends much lower protein cost than the WLP (1.5×10 4 g/(mol s -1 ) vs. 9.4×10 4 g/(mol s -1 )). This result highlights superior catalytic efficiency of the bicyclic pathway.
In addition, we used ensemble modeling to assess metabolic robustness 15,16 , which investigates a set of models with different kinetic parameters and perturbs them by varying maximum rate (Vmax), which is largely proportional to the expression levels of the enzyme 14, 17 . The synthetic pathway displays a high system stability when overexpressing every pathway enzyme individually by 10-fold, which only leads to less than 20% failure probability (Fig.3a). A moderate robustness is also exhibited when pathway enzymes are downregulated. In 15 of 18 enzymes, greater than 50% probability of success can still be achieved when knocking them down by an order of magnitude. Pathway stability is sensitive to the downregulation of phosphoketolase (pkt), fructose-1,6-bisphosphatase (fbp), and phosphoenolpyruvate carboxykinase (pepck) as the perturbation might drain their reaction products, which are the substrates of the next reaction, hence forming a kinetic trap. Complementing to robustness analysis, we also investigated the response of metabolic flux to perturbed enzyme level and showed that the pathway flux is tunable by altering certain enzyme expression levels (Fig.3b). For example, overexpression of pepck and fbp by 10-fold tends to change pathway fluxes by 1.5-fold, indicating there exist operable rate-determining steps in the pathway for flux control.
Next, we evaluated the biosynthetic potential of the acetyl-CoA bi-cycle in the context of metabolic network by using a genome-scale model for flux balance analysis (FBA) 18,19 .
C. ljungdahlii genome carries all genes in the acetyl-CoA bi-cycle except the phosphoketolase gene. We thus added this reaction into the stoichiometric model and tested maximal carbon yield (represented as the flux for ethanol and acetate production) as a function of the acetyl-CoA bi-cycle (represented by phosphoketolase flux) and the WLP (represented by ACS flux) (Fig.4). When fructose serves as the sole carbon source in heterotrophic mode, we observed a coordination of synthetic acetyl-CoA bi-cycle and native WLP to raise carbon yield of the C2 metabolites (Fig. 4b). Synergetic improvement is also observed when fructose and CO are provided in mixotrophic mode, indicating a potential impact of the bi-cycle to improve carbon efficiency under this feeding condition ( Fig. 4c). With a mixture of CO, CO2 and H2, the implementation of the bi-cycle appears to improve the carbon yield again (Fig.4a), when phosphoketolase flux is ranged between 0-10 mmol gDCW -1 h -1 . Note that a tightly constrained FBA model is used where the production and consumption of redox cofactors (e.g. ferredoxin and NAD(P)H) must be balanced. Under this reductive limitation in cellular level, the bi-cycle appears to improve the fluxes for C2 metabolites (acetate and ethanol) production, implying its potential value in metabolic engineering of acetogens.

C. ljungdahlii carries functional modules of the acetyl-CoA bi-cycle
Prior to engineering a prototypic acetyl-CoA bi-cycle in C. ljungdahlii, we first examined the activities of the native carbon fixation module and the gluconeogenesis module since these genes are present in the C. ljungdahlii genome. We designed an isotope labeling experiment in which 13 C-acetate ( 13 C in place for both methyl and carboxyl carbons) was used as a tracer and the 13 C-fingerprints of target metabolites were recorded (Fig.5).
When CO/CO2 (4:1) was provided in the headspace of clostridial culture, 13 C-acetate was assimilated into the central carbon metabolism (CCM) via activated forms, e.g., acetyl-CoA or acetyl phosphate. Their bioconversion was further tracked by positional 13 Cpatterns of proteinogenic amino acids that are produced from the CCM. Specifically, two carbon (C2-3) labeled alanine was observed. This labeling pattern is in line with the activity of carbon fixation module in which rPFOR incorporates a 13 C-acetyl-CoA with an unlabeled CO2 for the synthesis of pyruvate, the precursor of alanine. In addition, 13 Cpyruvate propagated its labeled carbons to oxaloacetate (OAA) which was reflected by 13 C-pattern of aspartate consistently. The conversion of OAA to PEP was also detected by the labeling of PEP in the corresponding positions (PEP2-3), consistent with the activity of PEP carboxykinase, a key reaction in gluconeogenesis. Moreover, we also detected 13 C-labeling on corresponding positions in histidine and tyrosine. These two amino acids are synthesized from sugar phosphates (erythrose-4-phosphate and pentose 5-phosphate, respectively). The observed labeling indicates that C. ljungdahlii has a complete suite of enzymes to support gluconeogenesis in the conversion of C2 metabolites to sugar phosphates. The activities of separate modules were also examined under heterotrophic condition (Fig.S3) when clostridial cells grew on fructose.
Interestingly, we detected the carbon fixation activity by rPFOR but no equivalent gluconeogenesis activity, suggesting that the gluconeogenesis module is optional when cells are supplied with sufficient sugar substrates.

Overexpression of heterologous phosphoketolase enabled a complete acetyl-CoA bi-cycle in the host
We further expressed a heterologous phosphoketolase in C. ljungdahlii to enable a NOG module and complete the acetyl-CoA bi-cycle. We selected a phosphoketolase gene from C. acetobutylicum and tested its enzymatic activity by expressing it in E. coli, followed by enzyme purification and in vitro assay (Fig.S4). The gene was then cloned into a plasmid (pMTL82151) where the sequence is placed behind the non-coding sequence harboring the native promoter for pta gene (encoding phosphotransacetylase) and the plasmid was transformed into wild type C. ljungdahlii. The activity of phosphoketolase in cell lysate was determined to be 1.12 ± 0.08 μmol min -1 mg -1 . We further performed quantitative proteomic analysis of the new strain (named as acb, standing for Acetyl-CoA Bi-cycle) that was constructed. All enzymes essential in the bi-cycle including heterologously expressed phosphoketolase were detected in the proteome, accounting for 8.7% of metabolism proteome (Fig.6). As a comparison, enzymatic proteins for the WL pathway account for 24.2% of the proteome in metabolism category. These results indicated the acb strain expressed all enzymes in the acetyl-CoA bi-cycle and can serve as a prototype cell model for investigating this synthetic pathway.

The acetyl-CoA bi-cycle improved carbon utilization efficiency and/or redox homeostasis in C. ljungdahlii
We thus examined the growth and productivity of the acb strain under various growth conditions. The acb strain exhibited 20% higher heterotrophic growth rate than the wild type (4.9×10 -2 ± 3.4×10 -3 vs. 4.0×10 -2 ± 2.8×10 -4 ) when fructose was supplied as the carbon source (Fig. 7b). Correspondingly the fructose consumption rate, acetate, and ethanol production rate of the acb strain increased by 31%, 24%, 52%, respectively, compared to the parental strain (Fig.7b). These results indicate that heterologously expressed phosphoketolase improved sugar conversion rate in C. ljungdahlii. We then investigated the impact of this genetic engineering on carbon conversion efficiency. The acb strain exhibited 84% carbon molar yield for converting fructose to acetate and ethanol, comparable to the value in wild type C. ljungdahlii (86%, Fig.S6) and exceeding theoretical maximum of the canonical glycolytic pathway (67%). To distinguish the contribution of NOG and WLP, both of which increase carbon yield over glycolysis, we utilized a WLP-mutant and transformed the pMTL82151 plasmid carrying phosphoketolase into it. This strain (Δacs) has a site-directed mutagenesis in the ACS which leads to inactivation of this key WLP enzyme. Deficient ACS lowered the carbon molar yield from 86% (as in WT) to nearly 71% in Δacs, consistent with the role of WLP in improving heterotrophic carbon yield. We then investigated the effect of heterologous phosphoketolase expression in Δacs background strain (named as acbΔacs). We observed no obvious increase in carbon yield but interestingly, significant improvement of ethanol to acetate ratio (1.60 ± 0.04 vs. 0.80 ± 0.09, Fig.S7). Since ethanol is a more reduced C2 product than acetate, we hypothesize that this is an outcome of redox homeostasis in Δacs, and the NOG module elevates the proportion of reduced product as a trade-off of total carbon yield. To test this hypothesis, we leveraged a genome scale C. ljungdahlii FBA model with ACS knocked out, phosphoketolase knocked in, and all redox reactions balanced. Indeed, ethanol/acetate ratio is monotonically increasing with phosphoketolase flux, given a certain production rate for ethanol and acetate (represented by contour in Fig.S8). Our experimental results as well as in silico analysis demonstrate a new role of the NOG module in improving sugar conversion rate and redox homeostasis.
We next tested the performance of acetyl-CoA bi-cycle under mixotrophic conditions. The acb strain exhibited 21% higher mixotrophic growth rate than the wild type (2.3×10 -2 ± 4.8×10 -4 vs. 1.9×10 -2 ± 2.5×10 -3 ) when fructose and gaseous substrates (CO/CO2) were available in the culture system (Fig.7c). Consistently the rates for consuming fructose and producing C2 products (acetate and ethanol) increased by 56%, 39%, respectively. The headspace pressure also decreased more quickly than the control culture, indicating a faster consumption rate of gaseous substrates.
Without fructose supplied, both acb and wild type strains can grow autotrophically using CO and CO2 as the carbon sources. Under this condition, the acb strain displayed similar C2 production rate (acetate and ethanol) with the parental strain while the ethanol to acetate molar ratio is lower (1.2 ± 0.2 vs. 1.7 ± 0.2). This result is likely consistent with the global redox balancing required in anaerobic cells, as more electrons are consumed for CO2 fixation in acetyl-CoA bi-cycle than the WLP. Nevertheless, the acb strain grew much faster than the wild type C. ljungdahlii (specific growth rate: 2.4×10 -2 ± 5.8×10 -4 vs.
Their compatibility with the host's metabolic system is uncertain. The canonical Calvin-Benson-Bassham (CBB) cycle is another option. However, the CBB cycle synthesizes triose (C3) not the acetyl-CoA (C2) as the final product. Additional step converting C3 to C2 will lead to carbon loss.
In terms of net stoichiometry of carbon fixation, the acetyl-CoA bi-cycle consumes more ATP and reducing equivalents than the WLP. The bi-cycle might seem a luxurious pathway as three ATP and eight electrons are required for generating an acetyl-CoA from CO2. Nevertheless, the acetyl-CoA bi-cycle is closely coupled with the CCM, the intermediates from which provides numerous essential precursors to support biosynthesis and cell growth. More importantly, we consider that the design of a synthetic metabolism may not only reflect the energy cost in the pathway but also the amount of enzymatic protein required to sustain pathway flux. We used state-of-the-art computational methods for analyzing pathways in terms of thermodynamics and kinetics and showed that the bi-cycle requires 6-fold less enzymatic protein to achieve the same acetyl-CoA conversion rate as the WLP (Fig.2). Given that enzyme synthesis, assembly and maintenance are among the most expensive chemical processes in a living cell, recruiting acetyl-CoA bi-cycle in synergy with the WLP will endow host cells with additional plasticity to rationalize the allocation of energetic currencies between protein and metabolite level. The acetyl-CoA bi-cycle thus lays a foundation to foster global acetogenic metabolism, while it highlights protein cost as a critical criterion for pathway design.
In addition to protein cost, this study collectively illustrates the bicyclic pathway in terms of thermodynamic feasibility, metabolic stability, biosynthetic potential within a genomescale metabolic network, in vivo module functionalities and actual improvement on acetogenesis. Based on all these results, we posit that coupling the acetyl-CoA bi-cycle with the native WLP is a practical strategy to improve carbon conversion in acetogenic organisms.

Materials and Methods
In silico analysis All computational methods and resources are available by contacting the lead contact W.X..

Pathway thermodynamics and enzyme protein cost analysis
Thermodynamics and enzyme protein cost analysis were applied to assess the feasibility of a native or designed metabolic pathway as well as corresponding protein burden.

Metabolic robustness analysis
Pathway stability was evaluated using an ensemble modeling approach combined with the continuation method 17 . First, an ensemble of models were generated with randomly sampled kinetic parameters from the feasible spaces, meanwhile subjecting to the same flux distribution of reference state 15,16 . Then the continuation method was used to simulate the system response to enzyme expression perturbations by integrating the differential equations 14, 17 :

Visualization of proteomic data
Proteomic data was visualized with an online illustration tool Proteomaps which shows the quantitative composition of proteomes according to protein function 27 . Details of description and usage of the tool can be found at: https://www.proteomaps.net/index.html.

Experimental analysis
Strains, media, and chemicals C. ljungdahlii DSM 13528 was purchased from the Leibniz Institute DSMZ (Braunschweig, Germany). The C. ljungdahlii ACS deficient strain OTA1 was a gift from Dr. Amy M Grunden at NC State University 28 . The YTF rich medium consisting of 10 g L -1 Bacto yeast extract, 16 g L -1 Bacto tryptone, 4 g L -1 NaCl, 5 g L -1 fructose and 0.5 g L -1 cysteine-HCl and the defined PETC medium were utilized for growing C. ljungdahlii. Cell growth was monitored at 600nm with a DU 800 spectrophotometer (Beckman-Coulter, Brea, CA). All chemical reagents used in growth studies were purchased from Sigma-Aldrich, except Bacto yeast extract and tryptone which were purchased from Becton Dickinson.

Phosphoketolase expression in E. coli, purification, and activity assay
The phosphoketolase (CA_C1343) gene from C. acetobutylicum ATCC 824 was synthesized from Genscript and cloned into plasmid pET28a. The successful transformants were confirmed by DNA sequencing. E. coli (DE3) containing the expression plasmid for N-terminal His6-tagged protein was grown on LB medium to an OD of 0.8 at 37°C, induced by 0.2 mM IPTG and harvested after overnight shaking at room temperature. Protein purification was performed according to a nickel-nitrilotriacetic (Ni-NTA) agarose minicolumn protocol 29 . The purified protein was run on a sodium dodecyl sulfate-polyacrylamide gel to monitor its size and purity. Phosphoketolase activity was measured using enzyme-linked assay 30 . Briefly, the assay included 50 mM Tri-HCl pH7.5, 5 mM MgCl2, 5 mM K3PO4, 1 mM ADP, 2 mM glucose, 0.2 mM NADP, 0.5 U glucose kinase, 0.5 U glucose-6-phosphate 1-dehydrogenase, 1 mM thiamine pyrophosphate (TPP), 1U acetate kinase, 10 mM of ribose 5-phosphate, 2U ribulose 5phosphate 3-epimerase, 2U ribose-5-phosphate isomerase and 2 µg purified phosphoketolase. The formation of NADPH was recorded at 340 nm by plate reader.

Construction of acb strains in C. ljungdahlii
Plasmid pMTL82151 was from Chain Biotech (Nottingham, UK) and used to generate the constructs transformed into C. ljungdahlii. The C. ljungdahlii pta promoter was amplified from gDNA and linked with the phosphoketolase to drive expression. To construct the plasmid, pMTL82151 was linearized with SmaI and ligated with the pta promoter and phosphoketolase gene using Gibson assembly from New England Biolabs (Ipswich, MA).
Transformation was based on previously reported protocols 31  Phosphoketolase activity assay was adapted from previously reported protocols 29,32 .
Briefly, cell free extracts of the wild-type and transformed phosphoketolase strain were added to buffer containing 25 mM xylulose 5-phosphate to generate acetyl-P, which was then converted to acetate by adding 1 μl of 1 M MgCl2, 1 μl of 30 mM ADP, and 0.2 U of acetate kinase to 75 μl of the assay mixture and incubating at 30°C for 30 minutes. The acetate was then determined enzymatically with the Acetic Acid Assay Kit (Megazyme, Bray, Ireland), using an assay mixture without xylulose 5-phosphate as a control.