Deepanwita
Banerjee
ab and
Aindrila
Mukhopadhyay
*ab
aBiological Systems and Engineering, Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, CA 94720, USA. E-mail: amukhopadhyay@lbl.gov
bJoint BioEnergy Institute, Emeryville, CA 94608, USA
First published on 26th November 2022
Metabolic engineering of microbial systems and conversion routes can provide robust platforms for the production of bulk commodities for food, materials and fuel targets. For products in this range, the maximum conversion of starting materials and stable phenotypes in a bioreactor are vital for an economically viable process. Strain engineering approaches to improve the conversion efficiency and reduce the phenotypic variability have witnessed significant development in the past decade. Herein, we review several of the main categories of these approaches including growth coupling, growth decoupling, regulatory control and use of non-metabolic cellular functions. We discuss these topics in the context of microbial host physiology and its impact in the selection of the most effective approach. We also discuss the importance of growth medium optimization and studies using bioreactors in delineating a bioproduction system that is most likely to provide stable conversion over a longer period.
Sustainability spotlightIn 2019, the renewable energy consumption increased but its share in the total energy consumption was 17.7%, which is only 1.6% higher than that in 2010. In 2020, governments spent $375 billion on subsidies for fossil fuels. Subsequently, in 2021, the fuel-related emissions were at their highest and eliminated the pandemic-related reduction seen in 2020. These challenges can be addressed through the faster scale-up of renewable fuels and commodity chemicals. The sustainable production of many previously petrochemically derived chemicals can be realized via microbial production using renewable starting materials, but for success, this requires the maximum conversion and balancing cultivation with final product formation in the process. This is in alignment with the UN Sustainable Development Goals including affordable and clean energy, responsible consumption and production, and climate action. |
A key parameter to consider in de-risking microbial production is that growth and bioproduction are linked. Although robust growth is required for efficient production, both growth and production also utilize the same pool of starting materials, creating a trade-off (Fig. 1A). Bioproduction mainly falls into three categories (Fig. 1B) that utilize a natural phenotype for accumulating a bioproduct. The first is secondary metabolism, rerouting a small amount of cellular intermediates to make a final product that has specialized use. Most engineering pathways (including heterologous pathways) fall in this category. The second is in the form of storage molecules such as polyhydroxyalkanoates (PHA), lipids, and fatty acids. The third is redox balancing by-product accumulation such as CO2, ethanol, and organic acids. Ethanol is one of the best bioconversion examples that works in exactly the desired conditions, i.e., high sugar load and low O2. Moving forward, mixed carbon co-feed and carbon (C) and nitrogen (N)-centric genome-scale metabolic representations hold immense potential in addressing the trade-off for the bulk production of commodity chemicals.
For the maximum conversion and ideal hosts, products, substrates, and scale agnostic bioconversion success of any type (Fig. 1C), the trade-off introduced through strain/host engineering can be addressed with the right combination of host and product, and thus these methods need to be more widely applied.
![]() | ||
Fig. 2 Product substrate pairing (PSP) approach for the growth-coupled microbial conversion of glucose (Glc) to indigoidine (Indg), adapted from Banerjee et al. |
Although growth-coupled strategies are promising, sometimes their direct implementation may result in auxotrophy. For example, OptGene-based S. cerevisiae strain engineering for the production of succinic acid resulted in glycine auxotrophy.13 Upon ALE the resulting strain had significant improvement in succinate titre (0.9 g L−1) and growth rates, but the yields were still low (0.05 g g−1 of glucose or 0.43 versus 0.69 g g−1 of biomass) due to the trade-off between growth and production. This trade-off due to competition for the same resources also resulted in significant dependence on available oxygen when growth-coupled strain engineering in E. coli was performed using elementary mode analysis for ethanol production.14 Recently, growth-coupled production examples have been comprehensively reviewed.15,16 Growth-coupled schemes are also being investigated for in vivo biohalogenation17 as well as protein production.18
A major challenge of decoupled microbial conversions that use constitutive production pathways is that during the growth phase, there may be a phenotypic drift due to suboptimal growth or toxicity and low-producing variants with fitness advantages may overtake the population in the bioconversion process. One approach to address this challenge is the native auto-induction method for bioconversion. Auto-induction is based on the lac operon regulatory function during diauxic growth in the presence of multiple carbon sources such as glucose, glycerol, and lactose.35 Although the control of these natural circuits is well characterized in E. coli, catabolite repression and preferential carbon source utilization in “rising star” hosts still need to be elucidated for bioproduction purposes. In this case, tighter control through synthetic feedback or feed-forward loops may help devise better solutions to overcome these trade-offs.
Gene circuits for synthetic addiction have been successfully developed in model organisms but can be context specific. Metabolite-responsive transcription factor-based biosensors have been the most successful thus far and hold immense potential but their dynamic range may be growth dependent and vary under different media conditions and growth environments. Recently, it was reported that the dynamic range of aTc-TetR and IPTG-LacI sensors has positive correlations with the cell growth rate, whereas the FA-FadR biosensor has a negative correlation with the cell growth rates when tested for several carbon sources in minimal media condition, confirming the trade-off between the dynamic range and growth condition.43 However, optimization is challenging when using a synthetic feed, given that only narrow sweet spots exist. These gene circuits are still not scalable across products and formats due to several challenges including their narrow dynamic range, linearity, and signal to noise ratio. A successful alternative was the design of genetically stable circuits or “landing pads” in the E. coli genome, where the expression level is high,44 and creation of insulators using the NOR gate logic. Recently, this has been extended to S. cerevisiae and Bacteroides thetaiotaomicron.45 Computational workflows are also being developed for multi-objective optimization for the trade-offs associated with biosensor development.46,47 Another challenge that remains is the enrichment of producers versus escapers or non-producers, as reported for baker's yeast.48 Recent examples of biosensor-based dynamic regulation have been reviewed elsewhere.49
Recently, optogenetic tools have also been utilized for the production of various value-added products such as lactic acid, isobutanol, and shikimic acid. Zhao et al.50 produced 8.49 g L−1 of isobutanol in S. cerevisiae using OptoEXP- and OptoINVRT-based control of EL222 transcriptional activator for metabolic switching to the production phase. Subsequently, a further optimized version, i.e., OptoAMP-based control of LDH, was reported for the production of lactic acid 6 g L−1 in S. cerevisiae even at low light intensities.51 In another study, optogenetic tools were developed using the TetR system together with the tobacco etch virus protease (TEVp) for the production of shikimic acid in E. coli at 35 g L−1 using glucose minimal media.52 The major bottlenecks with optogenetic tools include the limited number of photo-switchable proteins, restricted implementation in popular industrial hosts, insufficient and heterogeneous lighting at high cell density in large-scale bioreactors and cost associated with specifically designed industrial-scale light-bioreactors (up to 5000–10000 L) for the production of commodity chemicals. Another major challenge is the metabolic burden associated with the cellular resource allocation to build the multiple proteins and cofactors (FMN, NADH, and NADPH) required for the functioning of these optogenetic systems. Examples of various optogenetic tools have been recently reviewed elsewhere.53
Multi-substrate-based synthetic circuits for metabolic control of bioconversion processes have been reported54 but are still to be routinely implemented for production control. These synthetic genetic control systems should at least include substrate-induced transitions from the growth to production phase (decoupled production). Thus far, some progress in this direction has been reported for S. cerevisiae55 and E. coli24 but there has not been much progress in growth-coupled bioproduction strategies to the best of our knowledge. An aspect to consider in dynamic control circuits is the maintenance of the phenotype in a variable or heterogeneous environment. Specifically, for use in larger-scale production, the circuit must retain a high signal to noise ratio and have minimal interference from any crosstalk. Given that it is challenging to predict these issues a priori, implementation and examination of these systems in larger-scale fed-batch mode with industrially relevant feed sources are necessary.
Well-defined media with minimal nutrient supplementation are ideal from a techno-economic perspective; however, the use of resources coming from poorly utilized but abundant feedstocks is also necessary, both aiming towards a carbon-negative bioconversion process. Furthermore, successful bioconversion for commodity chemicals is measured by titer, rates, and yields (TRY), which is challenging to quantify in rich media or in undefined media such as plant biomass-derived feedstocks. These factors play a role in the use of growth-pairing strategies. To date, growth-coupled strategies have been examined for only a single C source. Mixed C-sources have been shown with DC strategies but it is necessary to further engineer strategies that account for more than one carbon source to address the growth productivity trade-off. An example towards advanced bioconversion systems using carbon negative feedstocks reported that a cofeed of CO2 and glucose enhanced acetogenesis in Moorella thermoacetica, whereas gluconate and acetate cofeed enhanced fatty acid production in Yarrowia lipolytica under fed-batch condition.56 Another example involved DC engineering of a heterologous Weimberg pathway bypass in the itaconate producer E. coli Δicd strain, which led to a high titer of 20 g L−1 itaconate and ∼65% of maximum theoretical yield (glycerol) using a cofeed of glycerol and xylose.57 Given that GC strategies are hard wired to the carbon source, it will be interesting to see how the production phenotype performs when multiple carbon sources or a nutrient-rich medium such as plant biomass-derived hydrolysate is used as the carbon stream for bioconversion to commodity chemicals using such engineered strains. Carbon sources that are incompatible with growth-coupling engineering may reduce the TRY, whereas compatible carbon sources may have a synergistic effect on the product yield (for example, growth coupling with glucose is synergistic with galactose but may be incompatible with aromatic substrates).12 The GC algorithm used in the PSP pipeline is customizable for co-feed substrate utilization but it has not been experimentally implemented thus far. To truly understand the growth versus production Pareto front, accurate measurements of TRY are required, but remain a challenge to obtain routinely. In the case of DC strategies, glucose is a well-studied and preferred carbon source, whereas the bioconversion medium components such as nitrogen, oxygen, phosphate, temperature and other limitations may have far more complex relationships to have tight transitions between the two phases and may not scale successfully to industry relevant bioreactor cultivation conditions. For example, biotin limitation under the fed-batch regime induced glutamate secretion as a by-product in C. glutamicum.28 In another extreme example, only 4 out of 16 phosphate-regulated promoters tested in E. coli scaled successfully to the bioreactor condition (ugpB, yibD, phoA, and phoB promoters).58
Finally, instead of implementing GC with multiple substrates in the same strain, a viable near-future alternative is a one-pot synthetic microbial community of GC-engineered strains for the division of labor for the bioconversion of specific substrates efficiently. A similar division of labor has been used to reduce the metabolic burden in a P. putida and S. cerevisiae consortium to produce 295.7 mg L−1 mcl-PHA titer.59 Recently, a synthetic microbial consortium of P. putida and E. coli was reported, where substrate utilization and production were decoupled in the two strains, which resulted in 1.32 g L−1 of mcl-PHA from 20 g L−1 of a glucose–xylose mixture (1:
1).60 Computational dynamic modeling has also been used to study the trade-off between productivity and efficiency of substrate utilization in a synthetic consortium of E. coli strains, with one producing a heterologous protein together with a second E. coli strain engineered to scavenge acetate.61
Nitrogen is another major constraint that is usually overlooked in various engineering strategies. The optimization of the C/N ratio in bioprocess optimization is required to identify the best bioconversion cultivation condition62–64 or supplementation of large amounts of nitrogen sources (e.g., 1 g L−1 or higher supplementation in production medium during scale-up). Firstly, nitrogen has been extensively used for metabolic engineering purposes to address the growth versus production trade-off of storage metabolites that are triggered by N starvation such as PHA production in P. putida strains.65,66 Secondly, research is shifting towards “greener” non-conventional carbon feedstocks containing aromatics, furfurals, etc., which follow different catabolic routes and regulatory mechanisms to generate energy and maintain cellular biomass (not the conventional glycolysis → TCA → oxidative phosphorylation). Aromatic catabolism is very different from traditional sugar glycolytic pathways. Conventional knowledge is that when glucose, fatty acids, or some amino acids are utilized as a carbon source under aerobic conditions, the C/N ratio sensing metabolic node is alpha-ketoglutarate (AKG).67 However, this may not be true for aromatics or other carbon sources. C and N-centric context-specific genome-scale metabolic models (GSM) and 13C- and 15N-labeled metabolic flux analysis (MFA) under a range of different conditions will advance the GSM closer to C- and N-relevant experimental conditions. Further, as is often the case with the carbon source, a certain N source can also be more suitable for a host or a conversion process, and a given medium formulation may not hold true if the host, pathway or culture format is changed for the same bioconversion system.63,68
Given that the final products span a greater range of targets, we now encounter products that are metabolized or degraded by the host microbe. Product degradation or catabolism also indirectly related to growth-production pairing and medium amendments has been used to prevent product catabolism. For example, in isoprenol production in P. putida, it was found that the natural catabolism of isoprenol occurs only after the consumption of glucose (Xi et al.,82 in their review). In general, metabolically versatile hosts such as P. putida have catabolic pathways for many final products or precursors. This can also be addressed via the deletion of the catabolic route when known.69,70 However, to the best of our knowledge, there are not enough systematic studies on the effect of media components on highly engineered strains that are tailored for bioconversion on a large scale.
Medium optimization is an integral part of process optimization for any bioconversion process.71–73 From a synthetic biology perspective, metabolic pathways and host engineering research must also incorporate the knowledge of the medium components on the metabolism being engineered. It is understood that during scale-up from the lab to industrial level, medium and process optimization plays a significant role in research and development. However, even at the lab-scale design, requirements of media formulations can be incorporated into strain design. In the case of strains devoid of hierarchical substrate uptake and utilization (using modification in master regulators, e.g., delta crc), accessory machinery (e.g., EM42 strain) allows rewiring the host metabolism as well as cell physiology, and thus the trade-off is now substantially altered relative to the basal strain. The C/N ratio for the new strain designs will be different, and therefore requires reintegration in the form of a medium optimization module. However, despite the essentiality of medium optimization, it remains ad hoc in the literature73 and few examples of high throughput data-driven medium optimization exist.74,75 One can anticipate that with improvements in automation and data-driven approaches and the ability to examine configurations in high throughput (e.g., via microfluidics76), this aspect will also witness a lot of improvements. The consilience of metabolic and host engineering approaches with the optimization of production medium will enable ideal growth-production ratios that are stable (and thus predictable) across scales.
Product | TRYa | Host | Media | Reference | |
---|---|---|---|---|---|
a TRY – titer (T), rate (R) and yield (Y). b Commodity chemical has been commercialized. c AA – amino acid; CDW – cell dry weight; Gly – glycine; His – histidine; MAD – modified AD7; MOPS – 3-(N-morpholino)propanesulfonic acid; PHA – polyhydroxyalkanoate; SD – synthetic defined; Thr – threonine; TSB – tryptic soy broth; YE – yeast extract; YPD – yeast extract peptone dextrose. | |||||
Growth coupled | |||||
1 | Lactic acidb | 1.75 g L−1 | E. coli | M9 glucose | Fong et al., 2005 (ref. 3) |
2 | 1,4 BDOb | 18 g L−1 | E. coli | M9 glucose | Yim et al., 2011 (ref. 4) |
3 | 1-Butanolb | 30 g L−1, 70% yield, 0.18 g L−1 h−1 | E. coli | Glucose | Shen et al., 2011 (ref. 5) |
4 | 1-Butanolb | 29.9 mg L−1 | Synechococcus sp. | BG-11 | Lan and Liao, 2012 (ref. 6) |
5 | 2-Methyl-1-butanol, isobutanol | 171 mg L−1 and 181 mg L−1 | Synechococcus sp. | MAD media | Purdy et al., 2022 (ref. 7) |
6 | 3-Hydroxypropionic acidb | 463 mg L−1 | S. cerevisiae | Glucose | Chen et al., 2014 (ref. 8) |
7 | Itaconateb | 32 g L−1, 0.68 mol mol−1 and 0.45 g L−1 h−1 | E. coli | Glucose + glutamic acid | Harder et al., 2016 (ref. 9) |
8 | Indigoidine | 26 g L−1 | P. putida | M9 glucose | Banerjee et al., 2020 (ref. 12) |
9 | Succinic acidb | 0.9 g L−1, 0.05 g g−1 | S. cerevisiae | Glucose + Gly/Thr | Otero et al., 2013 (ref. 13) |
10 | Ethanolb | 0.44 g g−1, 0.39 g L−1 h−1 | E. coli | Glycerol | Trinh and Srienc, 2009 (ref. 14) |
![]() |
|||||
Growth decoupled | |||||
1 | Propionic acid | 34.62 g L−1 | P. jensenii | Glycerol | Zhuge et al., 2014 (ref. 19) |
2 | Alpha-ketoglutarate, succinate | 11 g L−1 | C. glutamicum | Xylose | Tenhaef et al., 2021 (ref. 20) |
3 | Protocatechuate | 9.51 g L−1 | C. glutamicum | Glucose then xylose | Labib et al., 2021 (ref. 21) |
4 | Vanillin | 900 mg L−1 | E. coli | Ferulic acid | Lo et al., 2016 (ref. 22) |
5 | 1,2-Indandiol | 262 mg L−1 | Rhodococcus sp. | Glucose | Stafford et al., 2002 (ref. 23) |
6 | Polyhydroxybutyrate | 1.4 g L−1 | E. coli | Rich and minimal media glucose | Bothfeld et al., 2017 (ref. 24) |
7 | Fatty alcohol | 150 mg L−1, 0.03 g g−1 | E. coli | Glucose | Chubukov et al., 2017 (ref. 25) |
8 | Linalool | 10.9 g L−1, 5.1% yield | Pantoea ananatis | Glucose + YE | Nitta et al., 2021 (ref. 26) |
9 | Isobutanolb | 0.53 g L−1, 0.31 C-mol C-mol−1 | C. glutamicum | Hemicellulose fraction + YE | Lange et al., 2018 (ref. 27) |
10 | Proline | 142.4 g L−1, 2.9 g L−1 h−1, 0.31 g g−1 | C. glutamicum | Rich TSB media | Liu et al., 2022 (ref. 28) |
11 | Itaconateb | 47 g L−1, 0.86 g L−1 h−1 | E. coli | Glucose | Harder et al., 2018 (ref. 29) |
12 | Gamma-aminobutyric acid | 614 g L−1, 40.94 g L−1 h−1 | E. coli | Glucose + YE | Ke et al., 2016 (ref. 30) |
13 | Alpha-hydroxy ketones | 62–84% yield | E. coli | Pyruvate and benzaldehyde | Liang et al., 2020 (ref. 31) |
14 | Caprolactone | 126 mM, 0.78 mol mol−1 | E. coli | Cyclohexanol | Xiong et al., 2021 (ref. 32) |
![]() |
|||||
Feedback loops | |||||
1 | Vanillin | 900 mg L−1 | E. coli | Ferulic acid | Lo et al., 2016 (ref. 22) |
2 | Naringenin | 523.7 mg L−1 | E. coli | MOPS glucose, glycerol, potassium acetate, palmitic acid, stearic acid | Zhou et al., 2021 (ref. 38) |
3 | Gamma-aminobutyric acid | 45.6 g L−1, 0.4 g g−1 | C. glutamicum | Glycerol | Wei et al., 2022 (ref. 39) |
4 | Salicylic acid | 520 mg L−1 | E. coli | M9 glucose + glycerol | Dinh and Prather, 2019 (ref. 40) |
5 | Itaconateb | 1.4 g L−1 | P. putida | Deconstructed lignin | Elmore et al., 2021 (ref. 41) |
6 | Lycopene | 150.9 mg L−1 | E. coli | Glucose + YE + peptone | Li et al., 2020 (ref. 42) |
7 | Insulated genetic landing pads with logic gate circuits | — | E. coli | M9 glucose + thiamine and casamino acids | Park et al., 2020 (ref. 44) |
8 | Insulated genetic landing pads with logic gate circuits | — | E. coli, S. cerevisiae, B. thetaiotaomicron | — | Jones et al., 2022 (ref. 45) |
9 | N-Acetylglucosamine | 0.2 g L−1 | S. cerevisiae | Glucose + SD-Leu + yeast nitrogen base without AA | Lee et al., 2021 (ref. 48) |
10 | Isobutanol | 8.49 g L−1, 53.5 mg g−1 | S. cerevisiae | YPD or SC-dropout medium with 2% glucose and branched AA | Zhao et al., 2018 (ref. 50) |
11 | Lactic acid | 6 g L−1 | S. cerevisiae | YPD or SC-His with 2% glucose | Zhao et al., 2021 (ref. 51) |
12 | Shikimic acid | 35 g L−1, 0.43 g g−1, 0.48 g L−1 h−1 | E. coli | Glucose minimal media | Komera et al., 2022 (ref. 52) |
13 | Nerolidol | 4 g L−1, 3.8–4.5% yield | S. cerevisiae | Glucose, sucrose, glucose + ethanol | Peng et al., 2017 (ref. 55) |
![]() |
|||||
Media optimization | |||||
1 | Fatty acids | 38% energy yield | M. thermoacetica and Y. lipolytica | CO2, glucose and gluconate, acetate | Park et al., 2019 (ref. 56) |
2 | Itaconate | 20 g L−1 | E. coli | Glycerol and xylose | Lu et al., 2021 (ref. 57) |
3 | — | — | E. coli | Rich and minimal media | Moreb et al., 2020 (ref. 58) |
4 | PHA | 295.7 mg L−1, 19.3% yield per CDW | P. putida and S. cerevisiae | Xylose and octanoate | Wei et al., 2022 (ref. 59) |
5 | PHA | 1.32 g L−1 | P. putida and E. coli | Glucose or xylose minimal media | Zhu et al., 2021 (ref. 60) |
4 | PHA | 48.5% yield per CDW | P. putida | E2 minimal media with glucose, fructose, glycerol, octanoate or decanoate | Huijberts et al., 1992 (ref. 65) |
5 | PHA | 41.5 g L−1, 67% yield per CDW, 0.83 g L−1h− 1 | P. putida | Glucose minimal media | Poblete-Castro et al., 2014 (ref. 66) |
This journal is © The Royal Society of Chemistry 2023 |