Autotrophic organisms are able to fix CO₂ (occasionally CO as well) from the environment using various pathways: 1) the Calvin-Benson-Bassham (CBB) cycle, 2) the Wood-Ljungdahl pathway (WLP), 3) the reductive tricarboxylic acid (rTCA) cycle, 4) the 3-hydroxypropionate–4-hydroxybutyrate (3HP-4HB) cycle, 5) the dicarboxylate–4-hydroxybutyrate (DC–4HB) cycle, 6) the 3-Hydroxypropionate (3HP) bicycle and 7) the reductive glycine pathway (rGlyP) (Claassens et al., 2016; Sánchez-Andrea et al., 2020). Among the seven pathways, formate can serve as the feedstock in the WLP and rGlyP instead of CO₂ and H₂.

The CBB cycle is ubiquitous in photoautotrophic organisms including plants, algae, and cyanobacteria and in some chemoautotrophic bacteria. The CO₂ fixation in the CBB cycle depends on two key enzymes ribulose-1,5-bisphosphate carboxylase/oxygenase (RuBisCO) and phosphoribulokinase (PrkA), both of which are missing in heterotrophs (Figure 2A). To date, many studies have been reported on transplanting the CBB cycle into various heterotrophic microorganisms to fix CO₂. The introduction of RuBisCo and PrkA into E. coli (Parikh et al., 2006) and S. cerevisiae (Guadalupe-Medina et al., 2013) enabled carboxylation of C5 sugars, which also increased the yield of glucose-to-ethanol fermentation in S. cerevisiae. Recently, a completely functional CBB cycle has been established in E. coli (Gleizer et al., 2019) with the introduction of RuBisCo, PrkA, carbonic anhydrase (CA) and formate dehydrogenase (FDH). In the study, formate is used as the energy source (Figure 2B). Of note, in addition to metabolic engineering, the successful transition of E. coli from a heterotroph to an autotroph was also attributed to laboratory evolution by gradually reducing the xylose supply in chemostat. Similarly, the CBB cycle has also been introduced into Pichia pastoris, which converted the methylotrophic yeast to an autotrophic yeast using methanol as the reducing power (Gassler et al., 2020). Unlike the E. coli study, the success of the Pichia study was mainly achieved by metabolic engineering, with the overexpression of eight genes, including six pathway genes (RuBisCo; Prk; PGK1, phosphoglycerate kinase; TDH3, glyceraldehyde-3-phosphate dehydrogenase; TPI1, triosephosphate isomerase; TKL1, transketolase) and two chaperone proteins groEL/S from E. coli, and the deletion of three genes, alcohol oxidase (Aox1) and two dihydroxyacetone synthases (DAS1/2) (Figure 2C). The deletion of Aox1 was to reduce the formation rate of formaldehyde, which was still produced by Aox2, a less active oxidase than Aox1. DAS1/2 are responsible for the conversion of methanol to central metabolites (GAP and dihydroxyacetone, or DHA), the deletion of DAS1/2 prevented methanol assimilation into biomass and made methanol only an energy source to supply ATP and NADH (Gassler et al., 2020). In addition, the CBB enzymes were introduced into the peroxisome, and the resulting strain was more efficient on CO₂ fixation than that used cytosolic CBB pathway.

The WLP, also known as the reductive acetyl-coenzyme A (CoA) pathway, is the most efficient non-photosynthetic carbon fixation system. WLP requires only one ATP molecule per pyruvate. In contrast, the CBB cycle consumes seven ATP per pyruvate (Claassens et al., 2016). The key enzymes in the WLP are carbon monoxide dehydrogenase (CODH), formate dehydrogenase (FDH), acetyl-CoA synthase (ACS), pyruvate:ferredoxin oxidoreductase (PFOR) (Figure 3). CO₂ is fixed by FDH to formate in the methyl branch, while CO used by ACS is in the carbonyl branch (Figure 3). As two key enzymes PFOR and CODH are oxygen sensitive, the WLP only functions anaerobically and so do the microbes harnessing the WLP, e.g., acetogens. Such anaerobes have extremely good carbon conversion efficiency but very poor growth as the production of acetate from acetyl-CoA is the principal ATP generating step. Nevertheless, acetogens have been widely explored both academically and industrially on gas fermentation using syngas and industrial waste gas to produce various chemicals such as acetate, ethanol, and butyrate. The topic has been extensively reviewed previously (Liew et al., 2016). Although the WLP is energetically highly efficient, reconstructing WLP in heterologous hosts is very challenging. The first attempt to transplant the WLP in E. coli failed as CODH and corrinoid iron-sulphur-containing protein (CoFESP) did not function well. Another recent study also failed to demonstrate strain growth on CO and H₂ (Liew et al., 2016). One major difficulty is that E. coli and yeasts lack the proper intracellular conditions especially on the cofactor production (e.g., vitamin B12) and assembly of delicate metal centres.

The rTCA cycle, identical to the TCA cycle but in the reverse (reductive) direction (Figure 4), depends on four key enzymes to fix CO₂: α-ketoglutarate synthase (KGS) or α-ketoglutarate:ferredoxin oxidoreductase; isocitrate dehydrogenase (ICDH); phosphoenolpyruvate (PEP) carboxylase (PEPC); pyruvate synthase (PyrS). Among the four enzymes, KGS and PyrS are oxygen sensitive and require ferredoxin as the reducing cofactor. ICDH uses NAD(P)H as the reducing cofactor. PEPC, which prefers bicarbonate than CO₂ as the substrate, is also the key enzyme in C4 carbon fixation or the Hatch–Slack pathway, which contributes to reduce the wasteful process of photorespiration of RuBisCO in C4 plants (Paulus et al., 2013). As one of the most energy-efficient carbon fixation pathways, the rTCA cycle (requiring 1-2 ATP per pyruvate) has been proposed to produce chemicals and fuels from atmospheric CO₂ using microbial cells. This process is named the third-generation biorefineries (Liu et al., 2020). However, to date, few studies have applied the rTCA cycle in metabolic engineering for carbon fixation. In a recent study, rTCA cycle was used to recycle CO₂ in E. coli using KGS. The CO₂ production was reduced. Concurrently, formate production was observed, and the production of acetate and ethanol were increased (Chen C.-H. et al., 2020).

The 3HP-4HB cycle depends on two enzymes to fix CO₂, acetyl-CoA carboxylase (AcC) and propionyl-CoA carboxylase (PrC). AcC and PrC catalyse the addition of bicarbonate to acetyl-CoA and propionyl-CoA, respectively (Figure 5). The 3HP-4HB cycle has two variants of different energy efficiency. In the Crenarchaeota phylum, the 3HP-4HB cycle requires 9–10 ATP per pyruvate. In contrast, the 3HP-4HB cycle in the Thaumarchaeota pythlum requires only five ATP per pyruvate (Claassens et al., 2016). Like the rTCA cycle, the DC–4HB cycle also uses PyrS and PEPC to fix CO₂. As PyrS is oxygen sensitive, microbes such those in the order Thermoproteales using the DC–4HB cycle typically grow anaerobically (Hawkins et al., 2013). In the 3HP-4HB and DC-4HB cycles, two CO₂ molecules are fixed by PyrS to acetyl-CoA (C2) to produce succinyl-CoA (C4). Subsequently, succinyl-CoA is rearranged to acetoacetyl-CoA that is cleaved into two molecules of acetyl-CoA. In contrast, the acetyl-CoA is re-generated from citrate in the rTCA cycle (Figure 4). The 3HP bicycle is energetically expensive as it consumes seven ATP per pyruvate. The 3HP bicycle shares half the metabolic pathway (from malonlyl-CoA to succinyl-CoA) as that of the 3HP-4HB cycle and the two key enzymes for CO₂ fixation, AcC and PrC. However, the re-generation of acetyl-CoA in the 3HP bicycle has two routes: 1) from malyl-CoA that is produced from malate; 2) from citramalyl-CoA that is synthesized from glyoxylate and propionyl-CoA (Herter et al., 2002; Mattozzi et al., 2013). To date, the metabolic engineering applications of the 3HP-4HB cycle on carbon fixation are scarce. To our knowledge, the only example is that 3HP-4HB cycle genes are partially transplanted into the thermophilic host Pyrococcus furiosus to produce 3-hydroxypropionate from CO₂ and H₂ (Keller et al., 2013). The DC-4HB cycle has not been expressed in industrially relevant microbes for CO₂ fixation. Yet, the 3HP bicycle has been divided into four sub-pathways and individually expressed in E. coli (Mattozzi et al., 2013). Although all the sub-pathways were functional, none of them could support autotrophic growth, and the attempt to reconstruct the complete 3HP bicycle was not successful in E. coli (Claassens et al., 2016).

The rGlyP, naturally used in the sulphate-reducing bacterium Desulfovibrio desulfuricans (Sánchez-Andrea et al., 2020), is another energy-efficient carbon fixation system (1-2 ATP per pyruvate), almost matching the WJP. Also, the rGlyP shares half of its pathway with the WJP, from CO₂/CO to 5,10-methylene-THF (Figure 6). Glycine cleavage/synthase system (GCS), a reversible four-component enzyme, is the key enzyme for CO₂ fixation and catalyses 5,10-methylene-THF, CO₂ and NH₃ to produce glycine. Glycine can be further assimilated to pyruvate and biomass via two main routes: 1) aerobically, by the serine deaminase pathway variant, which consumes two ATP per pyruvate; 2) anaerobically, by the glycine reductase pathway variant, which requires one ATP per pyruvate (Figure 6). The structural simplicity, energy efficiency, and the ability to operate both aerobically and anaerobically make rGlyP a very attractive route for metabolic engineering in both CO₂ fixation and formate assimilation. Recently, the rGlyP has been widely reconstructed in various microbes and tested in both aerobic and anaerobic conditions. In 2018, Arren Bar-Even group first assembled the synthetic rGlyP in E. coli to produce serine from formate and CO₂ (Yishai et al., 2018), and further engineered E. coli to grow on formate and CO₂ and achieved relatively fast growth with 8 h doubling time after ALE from the initial doubling time of ∼70 h (Kim et al., 2020). The core rGlyP has also been functionally expressed in S. cerevisiae to produce glycine from formate and CO₂ (Cruz et al., 2019). The rGlyP is also explored to replace the native Calvin cycle in Cupriavidus necator to grow on formate, achieved 2.6 g CDW/mole-formate (Claassens, 2021). Also, the GCS is successfully introduced into Clostridium pasteurianum for anaerobic utilization of formate and CO₂ (Hong et al., 2021).

The CO₂ fixation pathways have different energy efficiencies. Typically, the pathways operating in anaerobic conditions have higher ATP efficiency: the WLP and rGlyP require only 0.5 ATP per CO₂ fixed in anaerobic conditions. In contrast, those operating in aerobic and microaerobic conditions require more ATP: the CBB cycle, 3 ATP/CO₂; rTCA cycle, 1 ATP/CO₂; the 3HP-4HB cycle, 2 ATP/CO₂; the DC–4HB cycle, 1.5 ATP/CO₂; the 3HP bicycle, 1.67 ATP/CO₂. The ATP efficiency can better support the growth of anaerobes in which relatively less ATP is available than that in aerobes. In addition to ATP, the requirement of NAD(P)H equivalents is the same for various pathways at 2 NAD(P)H/CO₂, assuming the product is acetyl-CoA (if pyruvate were the product, it would require 1.67 NAD(P)H/CO₂) (Xiao et al., 2022). Also, different enzymes may use different reducing equivalents and electron donor. For example, CODH and PFOR in the WLP, as well as PyrS and KGS in the rTCA cycle, require ferredoxin as cofactors; other enzymes mostly use NAD(P)H as the reducing equivalents (Figures 2–6). Of note, ferredoxin (E ^′0 = –430 mV) has a higher energetic driving force than NAD(P)H (E ^′0 = –320 mV). This also contributes to the difference in energetic efficiency of various pathways.

The assimilation of methane in methanotrophs starts with its oxidation to methanol catalyzed by methane monooxygenases (MMOs). Two types of MMO are identified in methanotrophs: the intracellular soluble form (sMMO) present in several methanotrophs; and the more ubiquitous membrane-bound, particulate enzyme complex (pMMO) (Hakemian and Rosenzweig, 2007). sMMO has wider substrate specificity but can be inhibited by high copper concentration. Owing to the membrane association, pMMO has greater access to methane than sMMO and is proposed to oxidize methane more quickly (Ge et al., 2014). It is very challenging to express functional sMMOs or pMMOs in E. coli. Many attempts have made with limited success (Balasubramanian et al., 2010; Kim et al., 2019). A recent study developed a pMMO-mimetic catalytic protein by assembling the catalytic domains of pMMO on apoferritin as a biosynthetic scaffold. The pMMO-mimetic enzyme has a turnover number (0.084 s⁻¹) comparable to that of native pMMO and can be produced in E. coli with a high yield (Kim et al., 2019), paving the way for future metabolic engineering applications on methane.

After the methane oxidation step, its pathway may merge with the methanol assimilation pathway. Firstly, methanol is oxidized into formaldehyde by methanol dehydrogenase (MDH) in prokaryotes or by alcohol oxidase (AOX) in yeasts (Pfeifenschneider et al., 2017) (Figure 7). In methylotrophic bacteria such as Bacillus methanolicus and Methylobacterium extorquens, MDH oxidizes methanol using pyrroloquinoline quinone (PQQ) or nicotinamide adenine dinucleotide (NAD+) as dependent electron acceptors (Pfeifenschneider et al., 2017). In methylotrophic yeasts such as Candida boidinii, Pichia pastoris and Hansenula polymorpha, methanol is oxidized by AOX with molecular oxygen as an electron acceptor (Yurimoto and Sakai, 2019).

Formaldehyde is either oxidized to CO₂ for NADH production or assimilated into biomass. The breakdown of formaldehyde to CO₂ is catalysed by the two enzymes formaldehyde dehydrogenase (FADH) and formate dehydrogenase (FDH) (Figure 7). The dissimilation process not only supplies NADH and also serves as an important way to detoxify formaldehyde. In methanotrophs, formaldehyde is assimilated by mainly two routes: the ribulose monophosphate (RuMP) cycle and the serine cycle (Gęsicka et al., 2021). In methylotrophs, in addition to the two routes, additional pathways include: the Calvin cycle in the soil bacterium Cupriavidus necator and the dihydroxyacetone (DHA) cycle, also known as xylulose monophosphate (XuMP) pathway, in methylotrophic yeasts. In addition, as we previously discussed, the WLP and rGlyP can also be used to assimilate methanol and formate (Kim et al., 2020) (Figures 3, 6).

The RuMP cycle has several variants with different energetic efficiency as described and compared previously (Cotton et al., 2020). Nevertheless, all the RuMP cycle variants share the common enzymes 3-hexulose-6-phosphate synthase (HPS) and 6-phosphate-3-hexuloisomerase (PHI) (Figure 7). In different prokaryotes, F6P is channelled into various metabolic pathways to regenerate Ru5P, including glycolysis, non-oxidative pentose phosphate pathway (PPP), the Entner–Doudoroff (ED) pathway or the pathway via sedoheptulose 7-phosphate (Cotton et al., 2020). In the DHA (XuMP) cycle, formaldehyde and xylulose 5-phosphate (Xu5P; C5 compound) are catalysed by dihydroxyacetone synthase (DAS) to generate GAP and DHA, which are further channelled back to regenerate Xu5P (Figure 7). Different from bacteria, the methylotrophic yeasts express the methanol assimilation enzymes in the organelle peroxisome. The compartmentalization in peroxisomes can isolate the toxic intermediates (e.g., formaldehyde) from the rest of cells and is potentially more efficient than prokaryote systems.

The RuMP cycle shares most enzymes with sugar metabolism in sugar heterotrophs such as E. coli, except for three enzymes: MDH, HPS, PHI. However, simple expression of the three enzymes is insufficient to convert E. coli to a methylotroph growing solely on methanol, although methanol could be assimilated into the central metabolism with the supplementation of xylose (Müller et al., 2015), glucose (Bennett et al., 2018) or threonine (Gonzalez et al., 2018) as co-substrate. The critical issue is how to elegantly balance the generation and consumption of formaldehyde and preventing its accumulation intracellularly. The accumulation of formaldehyde induces DNA-protein crosslinking that leads to cell death (Chen F. Y.-H. et al., 2020). Recently, this problem has been solved by rational pathway design and ALE. Kinetic modelling identified that high activities of phosphofructokinase (Pfk) and glyceraldehyde 3-phosphate dehydrogenase (Gapdh) may destabilize the metabolic system by diverting the flux away from the RuMP cycle. After the deletion of the two genes, the E. coli strain carrying the RuMP cycle genes was evolved to grow solely on methanol and higher growth rate was achieved with a doubling time of 8 h (Chen F. Y.-H. et al., 2020). The growth rate of the evolved E. coli strain is comparable to the E. coli using the rGlyP as we discussed previously (Kim et al., 2020). Also, Corynebacterium glutamicum is explored for methanol assimilation (Zhang B. et al., 2020).

S. cerevisiae, a non-methylotrophic yeast, has also been studied to assimilate methanol. A recent study proved that the S. cerevisiae CEN. PK strain has the native methanol assimilation capability using ¹³C tracer-analysis (Espinosa et al., 2020). ALE and sequencing experiments further pinpointed an uncharacterized transcriptional regulator Ygr067cp that supports improved methanol assimilation. It was found that the deletion of the alcohol dehydrogenase 2 (adh2) or the acetyl-CoA synthetase gene (acs1) reduced methanol assimilation. However, the key enzymes for methanol assimilation in S. cerevisiae are still unknown. In another study, the transplantation of the DHA cycle genes from P. pastoris supported S. cerevisiae growth on methanol (Dai et al., 2017). Also, another non-methylotrophic yeast Yarrowia lipolytica has been recently engineered to assimilate methanol by introducing the DHA and RuMP cycle genes, as well as by using laboratory evolution (Wang et al., 2021).

It is straightforward to use wildtype microbes and to harness natural assimilation pathways. However, natural assimilation pathways are not always the best choice as they can be rather complex involving too many enzymes (e.g., 3-HP bicycle requires 18 enzymes), kinetically inefficient due to inefficient/unspecific enzymes (e.g., RuBisCO catalyses side reactions with O2 that under atmospheric conditions), have tight intrinsic regulations (e.g., NADH/NAD⁺ regulation) or contain special cofactors (e.g., the WLP) that are hard to reconstitute in industrial workhorse strains. Therefore, it is also attractive to design synthetic routes using novel enzymes and novel combinations of well-studied enzymes. Well-designed synthetic routes can be superior to natural ones in: 1) circumventing the limitations of natural pathways, for example, the use of oxygen-tolerant enzymes makes rGlyP also functional in aerobic conditions (Bang and Lee, 2018; Bang et al., 2020); 2) simplifying the metabolic pathway using less enzymes (Xiao et al., 2022); 3) bypassing the host metabolism and regulations using orthogonal designs (Chou et al., 2021); 4) achieving higher efficiency (Schwander et al., 2016; Cai et al., 2021) or making the pathway more thermodynamically favourable (Siegel et al., 2015).

Formolase, a computationally designed enzyme that catalyses formaldehyde to produce DHA, supports a synthetic pathway that assimilates formate to DHAP and further into central metabolism (Siegel et al., 2015). Another interesting enzyme is 2-hydroxyacyl-CoA lyase (HACL) that catalyses the ligation of carbonyl-containing molecules of different chain lengths with formyl-CoA to produce C1-elongated 2-hydroxyacyl-CoAs (Chou et al., 2019). HACL enables the bioconversion C1 feedstock to C2 or longer products, such as glycolate and 2-hydroxyisobutyrate. Using HACL, the same team further developed the formyl-CoA elongation (FORCE) pathways, which can use various C1 feedstocks (CO₂, CO, formate, formaldehyde, methanol, and methane) to produce multi-carbon products including glycolate, ethylene glycol, ethanol and glycerate (Figure 8A). The FORCE system also demonstrated the potential in converting E. coli to synthetic methylotrophy in two-strain co-culture system (Chou et al., 2021). In the co-culture system, the first E. coli strain converts the C1 feedstock formate or methanol to the C2 product glycolate, which is further used as the feedstock to support the growth of the second E. coli strain (Figure 8B).

Another example of a synthetic CO₂ fixation pathway is designed by metabolic retrosynthesis. The synthetic pathway is named the crotonyl–coenzyme A (CoA)/ethylmalonyl-CoA/hydroxybutyryl-CoA (CETCH) cycle, comprising a reaction network of 17 enzymes from nine different organisms (Schwander et al., 2016). Using the more efficient enoyl-CoA carboxylases/reductases (ECRs), the CETCH cycle is about 2–4 times more efficient than that of RuBisCO. In a very recent study, a minimized synthetic CO₂ fixation cycle was designed and optimized with only four enzymes (Xiao et al., 2022). The synthetic cycle, named the POAP cycle, consists of pyruvate carboxylase, oxaloacetate acetylhydrolase, acetate-CoA ligase, and pyruvate synthase (Figure 8C). According to the study (Xiao et al., 2022), the POAP cycle fixes two CO₂ to produce oxalate using two ATP and one NAD(P)H, which is energetically more efficient than the seven natural and one synthetic CO₂ fixation pathways, the CBB, 3HP-4HB, rTCA, DC-4HB, 3HP and CETCH cycle, rGlyP and the WLP. However, POAP has yet to be demonstrated in microbes, and the in vivo reconstitution might face some challenges, e.g., the conversion from pyruvate and CO₂ to oxaloacetate can be outcompeted by other native pathways.

As compared to C1 feedstocks, the assimilation of C2 feedstocks by microorganisms is considerably easier and more efficient (Ma et al., 2022). Industrial model microbes are naturally, or with minimal engineering, capable of using both acetate and ethanol efficiently. Therefore, the current focus of C2 feedstocks is not on how to produce biomass but on how to better use them to produce valuable products in high yields and productivities.

Acetate primarily enters microbial cells by passive diffusion in the form of undissociated molecules (HAc). In addition, some bacteria (e.g., E. coli and Corynebacterium glutamicum) also use transporters for the uptake of dissociated acetate molecules (Ac⁻) (Gimenez et al., 2003; Sá-Pessoa et al., 2013). The assimilation of acetate requires only one or two enzymes. In E. coli and C. glutamicum, the acetate is assimilated by two routes: 1) ACS, which converts acetate to acetyl-CoA; 2) AckA and Ptak, which together catalyse acetate to acetyl-phosphate and further to produce acetyl-CoA (Enjalbert et al., 2017). Both routes require one ATP per acetyl-CoA. In some bacteria such as Pseudomonas sp. and acetic acid bacteria, the acetate assimilation is catalysed by succinylCoA:acetate CoA-transferase (SCACT). SCACT coverts acetate to acetyl-CoA using succinyl-CoA as CoA donor (Mullins et al., 2008). In S. cerevisiae, ACS1 is responsible for the assimilation of acetate and ethanol. ACS1 is expressed in peroxisome together with the glyoxylate shunt enzymes (e.g., citrate synthase, malate synthase) (Berg and Steensma, 1995). As described in Table 1, acetyl-CoA is a precursor that leads to various value-added bioproducts. As such, acetate has been used to produce many high-value products including small organic acids, alcohols, amino acids, terpenoids and lipids in several model microbes (Kiefer et al., 2020; Kim et al., 2021). For example, E. coli has been engineered to produce succinic acid (Huang et al., 2018), mevalonic acid (Xu X. et al., 2017) and β-caryophyllene (Yang and Nie, 2016) with relatively good yields. The oleaginous yeasts Rhodosporidium toruloides and Y. lipolytica were explored to produce lipids solely from acetate (Huang et al., 2016; Xu J. et al., 2017). Trichosporon cutaneum AS 2.571 was reported to have >2× cell mass and >3× lipid productivity of Y. lipolytica AS 2.1398 using acetate as the sole carbon source. R. toruloides was found poor in using acetate (Xu J. et al., 2017), however, our recent work showed that some R. toruloides strains have similar cell mass productivity to T. cutaneum AS 2.571. With ALE approach, we have increased the growth rate of R. toruloides on acetate by >30% as compared to the wild type, also shortened the lag phase from >20 h in wildtype to ∼10 h (Manuscript in preparation). Using metabolic engineering strategies that we developed (Zhang et al., 2018; Zhang and Too, 2020), we are currently engineering E. coli to produce terpenoids from acetate.

The assimilation of ethanol has two main routes: 1) alcohol dehydrogenase and acetaldehyde dehydrogenase (acetylating) that converts ethanol to acetaldehyde further to acetyl-CoA; 2) the acetate route in eukaryotes such as S. cerevisiae, ethanol is first converted to acetate via acetaldehyde, and acetate is assimilated to acetyl-CoA. The first route is found mainly in bacteria such as Clostridium acetobutylicum and E. coli. In S. cerevisiae, ethanol is oxidized by the alcohol dehydrogenase (adh2 or adh4) into acetaldehyde, which is further converted to acetate by aldehyde dehydrogenase (ald4 and ald5). The conversion of ethanol to acetate generates two NADH, which can be used for ATP regeneration. Therefore, higher theoretical yields are expected from ethanol than from acetate for reduced products. For example, the theoretical production yield of poly(3-hydroxybutyrate) (PHB) from ethanol, acetate and glucose are 0.935, 0.51 and 0.478, respectively (Sun et al., 2020). However, the assimilation of ethanol is a heat generating and oxygen-intensive process that may severely increase the production cost (Westfall et al., 2012). The acetate route genes were introduced into E. coli to assimilate ethanol via acetate as the intermediate (Cao et al., 2020). In addition, E. coli has been engineered to grow efficiently on ethanol by optimizing the two enzymes acetaldehyde dehydrogenase and alcohol dehydrogenase (acetylating). The ethanol-assimilating E. coli strain was further engineered to produce PHB or prenol from ethanol (Liang et al., 2021). In addition, ethanol has been used as sole or co-substrate with glucose to produce artemisinin precursor in S. cerevisiae (Westfall et al., 2012) and to produce up to 138 g/L of biomass in Candida brassicae (Mori et al., 1979). Although ethanol has a good potential, to date, ethanol has not been well explored as feedstock in metabolic engineering.

The selection of NGFs is crucial to establish efficient, rapid, and industry-compatible bioprocesses for the bioproduction of value-added chemicals. This requires a holistic view to fulfil the requirements of microorganisms (e.g., metabolic pathways, cofactors and growth conditions) and NGF-specific process design. Firstly, the continuous availability of NGF must be ensured and all the feedstocks chosen here fulfil this requirement. Indirect supply costs such as transport and purification must be evaluated, in which liquid NGFs are superior to gaseous NGFs (CO₂, CO, and CH₄). The safety profiles of the NGFs are directly related to this cost, as properties such as toxicity and flammability are key cost drivers, as summarized in Table 1. Next, special attention must be paid to the physical and chemical properties of the NGF, such as the aggregate state (gaseous or liquid) and water solubility, as these critically restrict mass transfer. Comparing next-generation gaseous and liquid feedstocks, their solubility differs by several orders of magnitude (Table 1). For comparison, the traditional feedstock glucose (909 g/L at 25°C) has a similar solubility to next-generation liquid feedstocks. This issue can be approached from different angles. Elevated pressure fermentation at 5–10 bar was identified as a straightforward but underexplored approach to increase mass transfer (Hecke et al., 2019). Beside the continuous stirred tank reactor that is typically used for liquid feedstocks, different fermentation systems have been tested such as bubble column, gas lift reactor, loop reactor, trickle bed reactor, membrane reactor and moving bed biofilm reactor amongst others which are discussed in detail elsewhere (Stoll et al., 2020). Another strategy is to convert gaseous feedstocks to liquid feedstocks using chemical (syngas to methanol), biological (gas fermentation to acetate) or hybrid approaches (microbial electrosynthesis), which are discussed in our previous section.

Another important factor is possible toxic effects of NGFs on the microorganism as well as possible adverse impacts of potential impurities (e.g., flue gas contains CO₂ but may also contain sulfur dioxide). Excessive NGF concentrations and the presence of significant amounts of trace gases can slow down or even stop the growth of microorganisms, resulting in low productivities. This issue can be overcome by strain engineering (e.g., development of detoxification pathways) or by bioprocess development. Particularly, appropriate feeding strategies and, if necessary, NGF purification process can be optimised to minimize/remove those toxic impurities. Exemplarily for syngas, depending on the gasification material and conditions used, impurities such as H₂O, N₂, particulates, alkali compounds (e.g., KOH, KCl), tars (organic hydrocarbon compounds >78 g/mol), nitrogen compounds (e.g., ammonia, hydrogen cyanide, and NO_x), sulfur compounds (e.g., H₂S), carbon-oxygen-sufur compounds (COS), halogen compounds (e.g., HCl), and heavy metals (e.g., Zn, Pb, Cd) amongst others can be found (Martin and Wolfgang, 2009). As gas fermentation is anaerobic, even oxygen is considered a gas impurity and has to be removed beforehand either by catalytic (Yan et al., 2013) or biologic processes (Mohr et al., 2019). An example highlights the importance of feedstock pretreatment. In 2014, in a commercial setting to produce ethanol from syngas, the plant was inoperable due to undetected hydrogen cyanide. The plant went into operation after the removal of the impurity from the feedstock was achieved by an additional treatment step (https://www.biofuelsdigest.com/bdigest/2014/09/05/on-the-mend-why-ineos-bio-isnt-reporting-much-ethanol-production). In addition to feedstock toxicity, inhibitory effects of products should also be evaluated for all fermentation systems.

One of biggest challenges in using NGFs for microbial fermentation is the slow cell growth and low maximal cell density (biomass). This is due to: 1) the cytotoxicity and low energy content of the C1/C2 substrates (Figure 1B); 2) limited supply of amino acids, nucleotide sugar and acetyl-CoA that are the precursors for the biosynthetic machinery (such as ribosomes and enzymes), cell wall, cytoskeleton and membranes; 3) formaldehyde and acetaldehyde, the minor metabolic intermediate of methanol and ethanol respectively, are highly reactive and cause cell cycle arrest which is triggered by the modification or cross-linking nucleic acids and proteins (Chen F. Y.-H. et al., 2020). To reduce the cytotoxicity of C1/C2 feedstock, substrate loading has to be maintained low, making it more unfavorable for gluconeogenesis (the reverse reactions of glycolysis) which makes UDP-glucose, certain amino acids such as S, H, F, Y, W, A, V, L, I and acetyl-CoA (Figure 9) (pathway for C1 metabolism is not discussed here as it can be found in many review papers). Thus, supplementation of medium with yeast extract, certain amino acids and vitamins has great impact on cell growth and final metabolite yields (Kaushik et al., 2020). Alternatively, such nutrients may be supplied by co-fermentation with conventional feedstocks although NGF substrate repression may have to be addressed, e.g., by using creA mutants in eukaryotes (Assis et al., 2021); cAMP-independent catabolite repressor (CcpA) in bacteria (Deutscher, 2008) or replacing the phosphotransferase system with the ATP-dependent sugar permease/kinase system (Zhang et al., 2015). The third option is to enhance the expression or activity of certain enzymes that are crucial for the biosynthesis of amino acids, sugars and nucleotides (Figure 9). In addition, for the assimilation of the highly oxidized feedstocks (CO₂ and CO), energy source should be carefully chosen to make it industrially compatible, which can be light, chemical electron donors (e.g., H₂, methanol, glucose), renewable electricity or their combinations.

Apart from medium compositions and reducing power, environmental factors such as medium pH and oxygen level are critical for microbial fermentation using NGFs. pH is particularly important for acetate and formate (Table 1). Increasing medium pH from 6 to 8–9 leads a drastic improvement in cell biomass growth and metabolite (lipids) production in Y. lipolytica. Importantly, this allows high acetate loading and yields much higher cell mass and lipid titres. Dry cell mass reached 37 g/L in small scale fed-batch fermentation when maintained at pH of 8 with 70 g/L initial acetic acid loading (Gao et al., 2020). Another important abiotic factor is oxygen. As we discussed previously, anaerobic fermentation has advantages for industrial bioproduction of certain types of products and carbon fixation. However, anaerobic conditions are incompatible with some production pathways or autotrophic systems that produce or require oxygen. Lastly, temperature control is critical for the fermentation of methanol and ethanol, which generates considerable heat and requires expensive cooling (Table 1). A solution to this problem is diverting reducing power towards anaplerotic metabolism (Schrader et al., 2009).

A one-fermenter-and-one-strain setup is the simplest but not always the best. System integration (e.g., multi-step bioprocess and microbial consortia) may drastically increase productivity and reducing production cost. Instead of directly using C1 gas to produce lipids, a two-step biosynthetic system may be more efficient than one-step system. A brilliant demonstration is the production of lipid by Y. lipolytica using dilute acetic acid (3%) derived from a gaseous biosynthetic system by acetogenic bacterium. A maximum lipid titre of 115 g/L was obtained, with substrate conversion rate of 0.16 g/g and productivity of 0.8 g/L/h (Xu J. et al., 2017). Co-cultures (Du et al., 2020) are a promising strategy which allows concurrent exploration of different genetic setups, thereby splitting pathways in two compartments. The anaerobic, non-photosynthetic mixotrophy was found to be a promising approach to increase product yield and decrease loss of CO₂ (Claassens et al., 2016).