|
|
Comprehensive evaluation of the capacities of microbial cell factories
Nature Communications volume 16, Article number: 2869 (2025) Cite this article
Abstract
Systems metabolic engineering is facilitating the development of high-performing microbial cell factories for producing chemicals and materials. However, constructing an efficient microbial cell factory still requires exploring and selecting various host strains, as well as identifying the best-suited metabolic engineering strategies, which demand significant time, effort, and costs. Here, we comprehensively evaluate the capacities of various microbial cell factories and propose strategies for systems metabolic engineering steps, including host strain selection, metabolic pathway reconstruction, and metabolic flux optimization. We analyze the metabolic capacities of five representative industrial microorganisms as cell factories for the production of 235 different bio-based chemicals and suggest the most suitable host strain for the corresponding chemical production. To improve the innate metabolic capacity by constructing more efficient metabolic pathways, heterologous metabolic reactions, and cofactor exchanges are systematically analyzed. Additionally, we present metabolic engineering strategies, which include up- and down-regulation target reactions, for the improved production of chemicals. Altogether, this study will serve as a comprehensive resource for the systems metabolic engineering of microorganisms in the bio-based production of chemicals.
시스템 대사공학은
화학물질 및 소재 생산을 위한
고성능 미생물 세포공장 개발을 촉진하고 있다.
high-performing microbial cell factories
그러나
효율적인 미생물 세포공장 구축에는
여전히 다양한 숙주 균주 탐색 및 선별과 최적의 대사공학 전략 도출이 필요하며,
이는 상당한 시간, 노력 및 비용을 요구한다.
본 연구에서는
다양한 미생물 세포공장의 역량을 종합적으로 평가하고,
숙주 균주 선정, 대사 경로 재구성, 대사 유동 최적화 등
시스템 대사공학 단계별 전략을 제안한다.
235종의 다양한 바이오 기반 화학물질 생산을 위한 세포 공장으로 활용 가능한
5가지 대표 산업 미생물의 대사 능력을 분석하고,
해당 화학물질 생산에 가장 적합한 숙주 균주를 제안한다.
더 효율적인 대사 경로 구축을 통해
선천적 대사 능력을 향상시키기 위해
이종 대사 반응 및 보조인자 교환을 체계적으로 분석한다.
또한 화학물질 생산량 향상을 위한
대사 공학 전략(표적 반응의 상향 및 하향 조절 포함)을 제시한다.
종합적으로 본 연구는
생물 기반 화학물질 생산을 위한 미생물의 시스템 대사 공학에 대한 포괄적인 자료가 될 것입니다.
Similar content being viewed by others
Physiological limitations and opportunities in microbial metabolic engineering
Article 02 August 2021
Systems metabolic engineering of microorganisms for food and cosmetics production
Article 27 June 2023
A versatile microbial platform as a tunable whole-cell chemical sensor
Article Open access27 September 2024
Introduction
Systems metabolic engineering1,2, which integrates the strategies and tools of synthetic biology, systems biology, and evolutionary engineering with traditional metabolic engineering, allows more efficient development of microorganisms for the sustainable production of various chemicals, including bulk chemicals3,4,5,6, fine chemicals7,8, fuels9,10,11, polymers12,13,14, and natural products15,16,17,18,19 from renewable resources instead of fossil resources. Starting with project design, systems metabolic engineering aims to optimize host strain selection, metabolic pathway construction, and metabolic fluxes, while considering fermentation and downstream processes20. However, exploring the vast metabolic space, represented by the combinations of the metabolic networks of different host strains and strain engineering strategies, still demands significant time, effort, and costs.
서론
합성 생물학, 시스템 생물학, 진화 공학의 전략과
도구를 전통적인 대사 공학과 통합하여
벌크 화학물질3,4,5,6, 정밀 화학물질7,8, 연료9,10,11, 폴리머12,13,14, 천연물15,16,17,18,19 등의
다양한 화학 물질을 화석 자원이 아닌
재생 가능 자원에서 생산하기 위한 미생물을 보다 효율적으로 개발할 수 있게 합니다.
Systems metabolic engineering1,2,
which integrates the strategies and tools of
synthetic biology,
systems biology, and
evolutionary engineering with traditional metabolic engineering,
allows more efficient development of microorganisms for the sustainable production of various chemicals,
including bulk chemicals3,4,5,6, fine chemicals7,8, fuels9,10,11, polymers12,13,14,
and natural products15,16,17,18,19 from renewable resources instead of fossil resources
프로젝트 설계부터 시작하여,
시스템 대사공학은 발효 및 다운스트림 공정을 고려하면서20
숙주 균주 선택, 대사 경로 구축 및 대사 유동을 최적화하는 것을 목표로 한다.
그러나
서로 다른 숙주 균주의 대사 네트워크와 균주 공학 전략의 조합으로 대표되는
광대한 대사 공간을 탐구하는 것은
여전히 상당한 시간, 노력 및 비용을 요구한다.
Model microorganisms such as Escherichia coli and Saccharomyces cerevisiae have been the primary workhorses for metabolic engineering due to the availability of the most abundant knowledge on their genetic and metabolic characteristics and also the gene manipulation tools. However, selecting a host strain requires consideration of the most suitable metabolic characteristics for the production of a target chemical. These include the presence of a native biosynthetic pathway for the target chemical, or the potential to produce it effectively when a heterologous or new biosynthetic pathway is introduced, capacity to produce the target chemical, the safety of the microorganism, and the environmental conditions in which the microorganism can thrive21. Recent advancements in bioengineering tools, such as clustered regularly interspaced short palindromic repeats (CRISPR)22 and serine recombinase-assisted genome engineering (SAGE)23, have enabled the metabolic engineering of non-model organisms that naturally produce target chemicals more amenable. Obviously, performing metabolic engineering on a host strain that possesses the highest biosynthetic capacity toward the target product is a promising strategy, as the strain has a potential to more efficiently produce chemicals compared to the other strains with lower biosynthetic capacity. The production performance is defined by three key metrics: titer (the amount of product per volume), productivity (specific productivity, the rate of production per unit of biomass, or volumetric productivity, the rate of production per volume), and yield (the amount or mole of product per amount or mole of consumed substrate)24. Among these key metrics, yield determines the required raw material costs, significantly affecting the overall bioprocess costs. Thus, selecting a host strain with a biosynthetic pathway that maximizes the yield of chemical production is crucial.
Escherichia coli 및 Saccharomyces cerevisiae와 같은 모델 미생물은
유전적·대사적 특성에 대한 풍부한 지식과 유전자 조작 도구의 가용성으로 인해
대사공학의 주요 작업 도구로 활용되어 왔다.
그러나
표적 화합물 생산에 가장 적합한 대사적 특성을 고려하여
숙주 균주를 선정해야 한다.
여기에는
표적 화학 물질에 대한 고유한 생합성 경로의 존재, 또는
이종 또는 새로운 생합성 경로가 도입될 때
이를 효과적으로 생산할 수 있는 잠재력, 표적 화학 물질을 생산할 수 있는 능력,
미생물의 안전성,
그리고 미생물이 번성할 수 있는 환경 조건이 포함됩니다21.
집단적으로 규칙적으로 간격을 둔 짧은
팔린드롬 반복서열(CRISPR)22 및 세린 재조합 효소 보조 게놈 공학(SAGE)23과 같은
생물공학 도구의 최근 발전으로,
표적 화합물을 자연적으로 생산하는 비모델 생물체의 대사 공학이 더 용이해졌습니다.
목표 제품에 대해 가장 높은 생합성 능력을 가진 숙주 균주에 대사 공학을 수행하는 것은
다른 낮은 생합성 능력을 가진 균주에 비해 화합물을 더 효율적으로 생산할 잠재력이 있으므로
분명히 유망한 전략입니다.
생산 성능은 세 가지 핵심 지표로 정의됩니다:
역가(부피당 제품량),
생산성(특정 생산성,
단위 생물량당 생산 속도 또는 부피당 생산성, 부피당 생산 속도),
수율(소비된 기질의 양 또는 몰당 제품의 양 또는 몰)24.
이러한 핵심 지표 중 수율은
필요한 원료 비용을 결정하여 전체 생물공정 비용에 상당한 영향을 미칩니다.
따라서
화학 생산 수율을 극대화하는 생합성 경로를 가진 숙주 균주를 선택하는 것이 중요합니다.
Genome-scale metabolic models (GEMs), which represent gene-protein-reaction associations in organisms through mathematical models, have been used to analyze the biosynthetic capacities and engineering strategies for developing microbial cell factories20,25,26. For example, gene knockout targets for the improved production of l-valine in E. coli were identified at the systems level by performing in silico knockout simulations for each gene in the strain, which would otherwise require considerable time, effort, cost for real experiments27. GEM-based approaches have not only identified gene targets for engineering but also characterized strain variations28, constructed biosynthetic pathways toward desired chemicals29,30, analyzed metabolic resource allocations in host strains31, and predicted metabolic interactions between microbial communities32. Although GEMs have been utilized to optimize host strain selection, metabolic pathway construction, and metabolic fluxes, a comprehensive exploration of the processes at the systems level still demands significant effort.
유전체 규모 대사 모델(GEMs)은
생물체의 유전자-단백질-반응 연관성을 수학적 모델로 표현하며,
미생물 세포 공장 개발을 위한 생합성 능력 및 공학적 전략 분석에 활용되어 왔습니다20,25,26.
예를 들어,
E. coli에서 l-발린 생산량 향상을 위한 유전자 녹아웃 표적은
균주의 각 유전자에 대해 컴퓨터 시뮬레이션 녹아웃을 수행함으로써
시스템 수준에서 확인되었으며,
이는 실제 실험을 수행하는 데 상당한 시간, 노력 및 비용이 소요될 수 있는 작업이었다27.
GEM 기반 접근법은
공학화를 위한 유전자 표적 식별뿐만 아니라 균주 변이 특성 분석28,
숙주 균주의 대사 자원 배분 분석31,
미생물 군집 간 대사 상호작용 예측32 등에도 활용되었습니다.
GEM이 숙주 균주 선택, 대사 경로 구축, 대사 유동 최적화에 활용되어 왔음에도 불구하고,
시스템 수준에서의 과정에 대한 포괄적인 탐구는
여전히 상당한 노력이 필요합니다.
In this study, we aim to provide resources for host strain selection, metabolic pathway construction, and metabolic flux optimization. To support host strain selection, we provide the metabolic capacities for 235 chemicals that have been produced, even if only minimally, in representative industrial microorganisms by calculating the maximum theoretical yield (YT), the maximum production of the target chemical per given carbon source when resources are fully used for the target chemical production, and maximum achievable yield (YA), the maximum production of the target chemical per given carbon source, considering cell growth and maintenance. For further improvement of metabolic pathway reconstruction, we have also systematically analyzed the expansion of innate metabolic capacity through the addition of heterologous reactions and cofactor exchanges in native metabolic reactions, and rewiring of innate metabolism to improve target chemical production. Furthermore, metabolic engineering strategies, which include the target reactions to be up- and down-regulated, are suggested for the improved production of chemicals. To demonstrate the versatility and applicability of these resources, we selected various products, including amino acids (l-lysine and l-glutamate) and ornithine used as nutritional supplements; precursors for biopolymers (sebacic acid and putrescine); a bulk chemical (propan-1-ol); and a key precursor for various natural products (mevalonic acid) as case studies. The resources presented in this study can also be employed for analyzing the other 229 chemicals (Supplementary Data 1–23) and also for other chemicals not described here using similar approaches.
본 연구에서는
숙주 균주 선택, 대사 경로 구축 및 대사 유동 최적화를 위한 자원을 제공하는 것을
목표로 합니다.
호스트 균주 선정을 지원하기 위해,
우리는 대표적 산업 미생물에서 최소한이라도 생산된
235가지 화학 물질에 대한 대사 능력을 제공합니다.
이를 위해 최대 이론적 수율(YT),
즉 자원이 목표 화학 물질 생산에 완전히 사용될 때 주어진
탄소원당 목표 화학 물질의 최대 생산량과,
세포 성장 및 유지를 고려한 최대 달성 가능 수율(YA), 즉
주어진 탄소원당 목표 화학 물질의 최대 생산량을 계산합니다.
대사 경로 재구축의 추가적 개선을 위해,
이종 반응 추가 및 고유 대사 반응 내 보조인자 교환을 통한 선천적 대사 능력 확장,
그리고 목표 화학물질 생산 향상을 위한 선천적 대사 재구성을 체계적으로 분석하였습니다.
또한, 생산성 향상을 위한
대사공학적 전략(상향/하향 조절 대상 반응 포함)을 제안합니다.
이러한 자원의 다용도성과 적용 가능성을 입증하기 위해,
영양 보충제로 사용되는 아미노산(l-라이신 및 l-글루타메이트)과
오르니틴, 바이오폴리머 전구체(세바산 및 푸트레신),
대량 화학물질(프로판-1-올),
다양한 천연물의 핵심 전구체(메발론산) 등 다양한 제품을 사례 연구로 선정하였습니다.
본 연구에서 제시된 자료는
다른 229가지 화학 물질(보충 자료 1–23)을 분석하는 데에도 활용될 수 있으며,
유사한 접근법을 사용하여 본 연구에서 다루지 않은 다른 화학 물질에도 적용될 수 있습니다.
Results
Selection of a suitable host strain having the high metabolic capacity
Bacillus subtilis, Corynebacterium glutamicum, E. coli, Pseudomonas putida, and S. cerevisiae are the five most frequently employed and preferred microbial strains in industrial biomanufacturing and academic research. Here, we analyzed the metabolic capacities - the potential of metabolic networks to produce chemicals - of five representative industrial host strains for the production of 235 chemicals (Supplementary Figs. 1–7). To calculate metabolic capacity, two types of yields of chemical production are used: maximum theoretical yield (YT) and maximum achievable yield (YA). Ignoring metabolic fluxes toward cell growth and maintenance makes the YT to be determined solely by the stoichiometry of reactions in the given metabolic network. However, unlike chemical processes, bioprocesses require resources and energy for the generation and maintenance of cells, which serve as biocatalysts, making it impossible to achieve the YT. To more realistically describe the metabolic capacity of strains for chemical production, we calculated YA, which accounts for non-growth-associated maintenance energy (NGAM) and setting the lower bound of the specific growth rate to 10% of the maximum biomass production rate to ensure minimum growth requirements, as suggested by Monk et al. (Supplementary Note 1)33.
높은 대사 능력을 가진 적합한 숙주 균주 선택
Bacillus subtilis,
Corynebacterium glutamicum,
E. coli,
Pseudomonas putida,
S. cerevisiae는
산업적 생물 제조 및 학술 연구에서 가장 빈번하게 사용되고 선호되는
5가지 미생물 균주입니다.
본 연구에서는
235개 화학물질 생산을 위한 5가지 대표 산업용 숙주 균주의
대사 능력(대사 네트워크가 화학물질을 생산할 수 있는 잠재력)을 분석하였다(보충 그림 1–7).
대사 능력 계산에는
두 가지 유형의 화학물질 생산 수율인 최대 이론 수율(YT)과 최
대 달성 가능 수율(YA)이 사용되었다.
세포 성장 및 유지에 필요한 대사 유동을 무시하면 YT는 주어진 대사 네트워크 내 반응의 화학량론에 의해 단독으로 결정된다. 그러나 화학 공정과 달리 생물공정은 생물학적 촉매 역할을 하는 세포의 생성 및 유지에 자원과 에너지가 필요하므로 YT를 달성하는 것은 불가능하다. 화학 생산을 위한 균주의 대사 능력을 보다 현실적으로 기술하기 위해, 우리는 YA를 계산하였다. 이는 비성장 관련 유지 에너지(NGAM)를 고려하고, 최소 성장 요구 사항을 보장하기 위해 특정 성장률의 하한을 최대 생물량 생산률의 10%로 설정하는 것으로, Monk 등(33)이 제안한 바와 같다. (보충 노트 1)33.
We calculated both YT and YA for 235 chemicals when produced in five microorganisms using nine key carbon sources (i.e., l-arabinose, d-fructose, d-galactose, d-glucose, d-xylose, glycerol, sucrose, formate, and methanol) under different aeration conditions (aerobic, microaerobic, and anaerobic conditions) (Supplementary Data 1–5). To calculate the yields of chemical production, we constructed GEMs that incorporate the biosynthetic pathways for each chemical, using metabolic reactions that have been previously reported to function properly for target chemical production. For the construction of this GEM, we selected 235 target chemicals from a metabolic map previously compiled34. We organized all metabolic reactions associated with these target chemicals into mass- and charge-balanced equations using the Rhea database35. For reactions not found in the Rhea database, we manually constructed the corresponding equations. Overall, we developed 272 metabolic pathways leading to the biosynthesis of 235 chemicals, including multiple pathways for a single target chemical when available. We constructed a separate GEM for each chemical biosynthesis pathway in each host, resulting in a total of 1360 GEMs. Out of these, 1092 GEMs were supplemented with heterologous reactions not present in the host strain’s GEM to establish functional biosynthetic pathways. The remaining 268 GEMs utilized native biosynthetic pathways for the production of the target chemicals. For more than 80% of the target chemicals, fewer than five heterologous reactions were required to construct biosynthetic pathways in the host strains, with percentages of 88.24%, 84.56%, 88.97%, 85.29%, and 90.81% for B. subtilis, C. glutamicum, E. coli, P. putida, and S. cerevisiae, respectively (Supplementary Fig. 8). These results indicate that the majority of bio-based chemicals can be synthesized with minimal expansion of metabolic networks. Furthermore, the length of biosynthetic pathways exhibited a weak negative correlation with maximum yields (Spearman correlations of −0.3005 and –0.3032 for YT and YA under aerobic conditions with d-glucose as the carbon source, respectively; p-values of 8.991e-30 and 2.601e-30 for YT and YA, respectively. n = 1360 for both cases), suggesting that maximum yields should be analyzed at the systems level for more comprehensive insights.
우리는 9가지 주요 탄소원
(즉, l-아라비노스, d-프럭토스, d-갈락토스, d-글루코스, d-자일로스, 글리세롤, 수크로스, 포르메이트, 메탄올)을 사용하여
생산할 때의 YT와 YA를 계산하였다(보충 자료 1–5).
화학 생산 수율을 계산하기 위해,
우리는 각 화학 물질의 생합성 경로를 포함하는 GEM(대사 모델)을 구축했습니다.
이 모델은
표적 화학 물질 생산에 대해 이전에 제대로 기능하는 것으로 보고된
대사 반응을 사용했습니다.
이 GEM 구축을 위해,
우리는 이전에 작성된 대사 지도에서 235개의 표적 화학 물질을 선정했습니다34.
이러한 표적 화합물과 관련된 모든 대사 반응을
Rhea 데이터베이스35를 사용하여
질량 및 전하 균형 방정식으로 구성했습니다.
Rhea 데이터베이스에서 찾을 수 없는 반응의 경우,
해당 방정식을 수동으로 구축했습니다.
전체적으로,
단일 표적 화합물에 대해 가능한 경우 여러 경로를 포함하여
235개 화합물의 생합성으로 이어지는 272개의 대사 경로를 개발했습니다.
각 호스트에서 각 화학물질 생합성 경로에 대해
별도의 GEM을 구축하여 총 1360개의 GEM을 생성했습니다.
이 중 1092개의 GEM은 기능적 생합성 경로를 확립하기 위해
호스트 균주의 GEM에 존재하지 않는 이종 반응을 보충했습니다.
나머지 268개의 GEM은
표적 화학물질 생산을 위해 고유 생합성 경로를 활용했습니다.
표적 화학물질의 80% 이상에서
숙주 균주의 생합성 경로 구축에 필요한 이종 반응은 5개 미만이었으며,
그 비율은 B. subtilis, C. glutamicum, E. coli, P. putida, S. cerevisiae에서
각각 88.24%, 84.56%, 88.97%, 85.29%, 90.81%로 각각 나타났습니다(보충 그림 8).
이러한 결과는
대다수의 바이오 기반 화학물질이
대사 네트워크의 최소한의 확장으로 합성될 수 있음을 시사한다.
또한 생합성 경로의 길이는 최대 수율과 약한 음의 상관관계를 보였다(호기성 조건에서 d-글루코스를 탄소원으로 사용할 때 YT와 YA의 스피어먼 상관계수는 각각 −0.3005와 –0.3032; p값은 YT와 YA 각각 8.991e-30 및 2.601e-30). n = 1360 (양측 모두). 이는 최대 수율을 보다 포괄적으로 이해하기 위해 시스템 수준에서 분석해야 함을 시사한다.
Based on metabolic capacities, it is possible to identify the most potent strain for producing a specific chemical. To explore the variability in host performance across chemicals, we performed hierarchical clustering of host ranks based on maximum yields (Supplementary Figs. 9 and 10). Under aerobic conditions with d-glucose as the carbon source, the clustering shows that while most chemicals achieve their highest yields in S. cerevisiae, a few chemicals display clear host-specific superiority (e.g., pimelic acid in B. subtilis; see Discussion for other selection criteria). Notably, these chemicals do not group according to conventional biosynthetic pathways or chemical categories, highlighting the necessity of evaluating each chemical individually rather than applying a universal rule. For instance, the metabolic capacities of host strains for producing l-lysine, an essential amino acid used in animal feed and as a human nutritional supplement, were compared under aerobic conditions with d-glucose as the sole carbon source. Among the strains, S. cerevisiae showed the highest yield (YT) of 0.8571 mol/mol d-glucose, followed by B. subtilis (0.8214 mol/mol d-glucose), C. glutamicum (0.8098 mol/mol d-glucose), E. coli (0.7985 mol/mol d-glucose), and P. putida (0.7680 mol/mol d-glucose). Except for S. cerevisiae, which synthesizes l-lysine via the l-2-aminoadipate pathway, the other strains utilize the diaminopimelate pathway, albeit with differing metabolic capacities (Supplementary Fig. 11). While metabolic capacities are crucial for selecting host strains for chemical production, other factors, such as actual in vivo metabolic fluxes toward the target chemical and chemical tolerance, play important roles in industrial applications. For example, C. glutamicum is widely utilized as an industrial strain for l-glutamate production due to its high metabolic fluxes in the l-glutamate biosynthetic pathway, capability for high cell density cultivation, and the GRAS status of its products. However, C. glutamicum has lower YT and YA under aerobic conditions with d-glucose as the sole carbon source (1.0000 mol/mol d-glucose and 0.9290 mol/mol d-glucose, respectively) compared to the other four strains (B. subtilis: 1.1729 mol/mol d-glucose, 0.9225 mol/mol d-glucose; E. coli: 1.1917 mol/mol d-glucose, 1.0652 mol/mol d-glucose; P. putida: 1.1915 mol/mol d-glucose, 1.0708 mol/mol d-glucose; S. cerevisiae: 1.2000 mol/mol d-glucose, 1.0868 mol/mol d-glucose). It should be noted that maximum yields represent upper-bound estimates based on in silico modeling and do not capture factors such as enzyme kinetics, product tolerance, and GRAS status of strains/products. Thus, our resource is intended to serve as an initial screening tool that provides a quantitative basis for narrowing down candidate strains. Although C. glutamicum is successfully used in industry, strains with higher maximum yields suggest a higher capacity for chemical production. Therefore, if strains with higher maximum yields are fully explored and engineered, they could potentially be developed into even more efficient cell factories. This also highlights that optimizing metabolic fluxes, a key objective of metabolic engineering, remains crucial for strain development. To further elucidate host-specific differences in maximum yields, we calculated the coefficient of variation (CV) for both YT and YA across the host strains for each target chemical. Our analysis shows that the mean CV for YT is 0.3684, whereas for YA it is 0.5172. A one-sided Wilcoxon signed-rank test confirmed that the CV for YA is significantly higher than that for YT (p-value of 0.000, n = 6292), indicating that YA captures additional host strain-specific variability. These results suggest that YA provides more discriminatory power for identifying hosts with a higher latent capacity for chemical production, which is not evident from YT alone. Importantly, even small differences in maximum yield can result in meaningful improvements in large-scale production, emphasizing the value of these metrics in guiding host selection for metabolic engineering.
대사 능력에 기반하여 특정 화학물질 생산에 가장 적합한 균주를 식별할 수 있다.
다양한 화학물질에 걸친 숙주 성능의 변동성을 탐색하기 위해,
최대 수율을 기반으로 숙주 순위를 계층적 클러스터링 분석했습니다(보충 그림 9 및 10).
d-글루코스를 탄소원으로 하는 호기성 조건에서,
대부분의 화학물질은 S. cerevisiae에서 최고 수율을 달성하는 반면,
일부 화학물질은 명확한 숙주 특이적 우월성을 보였습니다(예: B. subtilis의 피멜산; 기타 선정 기준은 논의 참조).
주목할 점은
이러한 화학물질들이 기존의 생합성 경로나 화학적 범주에 따라 그룹화되지 않는다는 것으로,
보편적인 규칙을 적용하기보다는 각 화학물질을 개별적으로 평가할 필요성을 강조한다.
예를 들어, 동물 사료 및 인간 영양 보충제로 사용되는
필수 아미노산인 l-라이신 생산을 위한 숙주 균주의 대사 능력을,
d-글루코스를 유일한 탄소원으로 하는 호기성 조건에서 비교하였다.
주변 균주 중
S. cerevisiae가 0.8571 mol/mol d-glucose의 최고 수율(YT)을 보였으며,
다음으로 B. subtilis (0.8214 mol/mol d-glucose), C. glutamicum (0.8098 mol/mol d-glucose),
E. coli (0.7985 mol/mol d-글루코스), 그리고 P. putida (0.7680 mol/mol d-글루코스) 순이었다.
l-2-아미노아디페이트 경로를 통해 l-라이신을 합성하는 S. cerevisiae를 제외하고,
다른 균주들은 다이아미노피멜레이트 경로를 이용하지만
대사 능력은 서로 달랐다(보충 그림 11).
대사 능력은 화학 생산을 위한 숙주 균주 선택에 중요하지만,
목표 화합물로의 실제 생체 내 대사 유동 및 화학적 내성과 같은 다른 요소들도
산업적 응용에서 중요한 역할을 한다.
예를 들어, C. glutamicum은 l-글루타메이트 생합성 경로의 높은 대사 유동, 높은 세포 밀도 배양 능력, 그리고 그 산물의 GRAS(일반적으로 안전하다고 인정되는 물질) 지위 덕분에 l-글루타메이트 생산을 위한 산업적 균주로 널리 이용된다. 그러나 C. glutamicum은 호기성 조건에서 d-글루코스를 유일한 탄소원으로 사용할 때 다른 네 균주(B. subtilis: 1.1729 mol/mol d-글루코스, 0.9225 mol/mol d-글루코스; E. coli: 1.1917 mol/mol d-글루코스, 1.0652 mol/mol d-글루코스; P. putida: 1.1915 mol/mol d-글루코스, 1.0708 mol/mol d-글루코스; S. cerevisiae: 1.2000 mol/mol d-글루코스, 1.0868 mol/mol d-글루코스). 최대 수율은 컴퓨터 모델링을 기반으로 한 상한선 추정치이며, 효소 동역학, 제품 내성, 균주/제품의 GRAS(일반적으로 안전하다고 인정되는) 상태 등의 요소를 반영하지 않음을 유의해야 합니다. 따라서 본 자료는 후보 균주를 선별하는 정량적 근거를 제공하는 초기 스크리닝 도구로 활용될 수 있습니다. C. glutamicum이 산업 현장에서 성공적으로 사용되고 있음에도 불구하고, 더 높은 최대 수율을 보이는 균주는 화학 물질 생산 능력이 더 우수함을 시사합니다. 따라서 최대 수율이 높은 균주를 충분히 탐색하고 공학적으로 개량한다면, 더 효율적인 세포 공장으로 개발될 잠재력이 있습니다. 이는 또한 대사 공학의 핵심 목표인 대사 유동 최적화가 균주 개발에 여전히 중요함을 강조합니다. 최대 수율의 숙주 특이적 차이를 더 명확히 밝히기 위해, 각 표적 화학물질에 대해 숙주 균주 전반에 걸친 YT와 YA의 변동계수(CV)를 계산했습니다. 분석 결과 YT의 평균 CV는 0.3684인 반면 YA는 0.5172로 나타났다. 일측 윌콕슨 부호순위 검정 결과 YA의 CV가 YT보다 유의하게 높은 것으로 확인되었다(p-값 0.000, n=6292). 이는 YA가 추가적인 숙주 균주 특이적 변동성을 포착함을 시사한다. 이러한 결과는 YA가 화학 물질 생산 잠재력이 더 높은 숙주를 식별하는 데 더 큰 판별력을 제공함을 시사하며, 이는 YT만으로는 명확히 드러나지 않습니다. 중요한 점은 최대 수율의 작은 차이조차도 대규모 생산에서 의미 있는 개선으로 이어질 수 있다는 것으로, 이는 대사 공학을 위한 숙주 선택을 안내하는 데 있어 이러한 지표의 가치를 강조합니다.
The emission of greenhouse gases accelerates global warming, spurring the development of technologies to reduce one-carbon gases (e.g., carbon dioxide, carbon monoxide, and methane). Consequently, one-carbon compounds have emerged as promising carbon sources for chemical production. Although the development of microbial cell factories for producing value-added chemicals exclusively from one-carbon compounds is still in its early stages, we also analyzed the potential to convert one-carbon compounds (i.e., methanol, carbon dioxide, and formate) into chemicals, aiming to provide a basis for future research. For instance, in the production of sebacic acid, a precursor of nylon-6,10, using methanol, the YT and YA were higher in E. coli (0.1091 and 0.0969 mol/mol methanol) and P. putida (0.1082 and 0.0970 mol/mol methanol), followed by S. cerevisiae (0.1000 and 0.0900 mol/mol methanol), B. subtilis (0.0944 and 0.0778 mol/mol methanol), and C. glutamicum (0.0667 and 0.0600 mol/mol methanol) when methanol was used as the sole carbon source through the ribulose monophosphate (RuMP) cycle. Maximum yields varied depending on the one-carbon compounds and their assimilation pathways. In E. coli, when using carbon dioxide as the carbon source and formate as both a carbon source and reducing power, a strain utilizing the reductive glycine cleavage (rGly) pathway showed higher maximum yields (YT and YA of 0.0321 mol/mol formate and 0.0270 mol/mol formate, respectively) than a strain using the Calvin-Benson-Bassham (CBB) cycle (YT and YA of 0.0243 mol/mol formate and 0.0204 mol/mol formate, respectively) (Fig. 1). Methanol exhibits a more negative standard enthalpy of combustion (−638.2 kJ/mol) compared to carbon dioxide (0 kJ/mol) and formate (−211.5 kJ/mol)36. This lower enthalpy of combustion indicates a higher intrinsic energy content, which provides increased availability of reducing power and ATP during metabolism. As a result, strains employing the RuMP cycle achieve higher maximum yields for sebacic acid production compared to those using the CBB cycle and the rGly pathway. Taking into account the toxicity of intermediate metabolites, the catalytic efficiency of enzymes in the one-carbon assimilation pathway, and other economic factors including storage and transport costs of the sources, these maximum yields of each pathway can help identify the most viable assimilation pathway and carbon sources.
온실가스 배출은 지구 온난화를 가속화하여
일탄소 가스(이산화탄소, 일산화탄소, 메탄 등) 감축 기술 개발을 촉진하고 있습니다.
이에 따라 일탄소 화합물은
화학 생산을 위한 유망한 탄소원으로 부상했습니다.
일탄소 화합물만을 이용해 부가가치 화학 물질을 생산하는 미생물 세포 공장 개발은 아직 초기 단계이지만, 향후 연구의 토대를 마련하기 위해 일탄소 화합물(메탄올, 이산화탄소, 포르메이트)을 화학 물질로 전환할 가능성도 분석했습니다. 예를 들어, 나일론-6,10의 전구체인 세바산 생산 시 메탄올을 사용했을 때, YT 및 YA 값은 E. coli(0.1091 및 0.0969 mol/mol 메탄올), P. putida(0.1082 및 0.0970 mol/mol 메탄올)에서 가장 높았으며, 그 다음으로 S. cerevisiae(0.1000 및 0.0900 mol/mol 메탄올), B. subtilis (0.0944 및 0.0778 mol/mol 메탄올), 그리고 C. glutamicum (0.0667 및 0.0600 mol/mol 메탄올) 순으로 높았다. 최대 수율은 일탄소 화합물과 그 동화 경로에 따라 달라졌다. E. coli에서 이산화탄소를 탄소원으로, 포르메이트를 탄소원이자 환원력으로 사용할 때, 환원성 글리신 분해(rGly) 경로를 이용하는 균주는 칼빈-벤슨-배셤(CBB) 경로를 이용하는 균주보다 더 높은 최대 수율 (각각 포름산 0.0321 mol/mol 및 0.0270 mol/mol)을 보였으며, 칼빈-벤슨-배셤(CBB) 사이클을 이용하는 균주(각각 포름산 0.0243 mol/mol 및 0.0204 mol/mol)보다 높았다(그림 1). 메탄올은 이산화탄소(0 kJ/mol) 및 포르메이트(-211.5 kJ/mol)에 비해 더 음의 표준 연소 엔탈피(-638.2 kJ/mol)를 나타낸다36. 이 낮은 연소 엔탈피는 더 높은 고유 에너지 함량을 의미하며, 이는 대사 과정에서 환원력과 ATP의 가용성을 증가시킨다.
결과적으로,
RuMP 사이클을 이용하는 균주는
CBB 사이클 및 rGly 경로를 사용하는 균주에 비해 세바산 생산 최대 수율이 더 높다.
중간 대사 산물의 독성, 일탄소 동화 경로 내 효소의 촉매 효율, 원료의 저장 및 운송 비용을 포함한
기타 경제적 요인을 고려할 때,
각 경로의 이러한 최대 수율은
가장 실행 가능한 동화 경로와 탄소원을 식별하는 데 도움이 될 수 있다.
Fig. 1: Comparison of one-carbon assimilation pathways for sebacic acid production.
a Methanol assimilation via RuMP cycle for sebacic acid production. b Formate and CO2 assimilation via CBB cycle for sebacic acid production. c Formate and CO2 assimilation via rGly pathway for sebacic acid production. Metabolic fluxes normalized by carbon source uptake rate to achieve YT and YA are shown in blue and red boxes, respectively.
Overall, this comprehensive analysis of metabolic capacities for bio-based chemical production highlights which microbial cell factories offer the most efficient biosynthetic pathways under targeted bioprocess conditions, providing a valuable resource for systems metabolic engineering.
그림 1: 세바산 생산을 위한 1탄소 동화 경로 비교.
a 세바산 생산을 위한 RuMP 사이클을 통한 메탄올 동화. b 세바산 생산을 위한 CBB 사이클을 통한 포르메이트 및 CO2 동화. c 세바산 생산을 위한 rGly 경로를 통한 포르메이트 및 CO2 동화. YT 및 YA 달성을 위한 탄소원 흡수율로 정규화된 대사 유동은 각각 파란색 및 빨간색 상자에 표시됨.
종합적으로, 바이오 기반 화학 물질 생산을 위한 대사 능력에 대한 이 포괄적인 분석은 목표 생물공정 조건 하에서 가장 효율적인 생합성 경로를 제공하는 미생물 세포 공장을 강조함으로써 시스템 대사 공학에 유용한 자원을 제공합니다.
Improving the innate metabolic capacity for chemical production
Metabolic engineering enhances the chemical production abilities of microbial cell factories by optimizing cellular characteristics within their metabolic capacity or by improving their innate metabolic capacity37. We systematically analyzed metabolic reactions expected to improve the metabolic capacity of a host strain for target chemical production. To improve the innate metabolic capacity of a host strain, we performed two approaches: expanding the native metabolic network by introducing heterologous reactions and replacing cofactors used in the native metabolic network to non-native cofactors.
To expand the native metabolic network, we explored heterologous metabolic reactions that could build more efficient biosynthetic pathways when coordinated with native metabolic reactions. These heterologous reactions were collected by constructing a universal model, an assembly of all metabolic reactions. The universal model was curated from the universal model provided by the BiGG database38 and contains 3814 metabolites and 6846 metabolic reactions. We identified yield-improving heterologous reactions if the YT of a target chemical was improved by at least 1% when these reactions were added to the GEM of a production strain. The simulation was performed for all 1360 constructed GEMs (representing 1360 different microorganisms), limiting the number of added heterologous reactions to three. We also provide the source code for the simulation to analyze more diverse conditions of interest to researchers.
The simulation identified candidate reactions that resulted in carbon- or energy-efficient biosynthetic pathways for the chemicals. For example, phosphoketolase, an enzyme that converts xylulose 5-phosphate (or fructose 5-phosphate) into acetyl phosphate and glyceraldehyde 3-phosphate (or d-erythrose 4-phosphate), was most frequently predicted to improve YT of chemicals across host strains. This enzyme facilitates a non-oxidative glycolysis pathway that conserves all carbons from the consumed sugar to acetyl-CoA (Fig. 2a). It has been demonstrated that this non-oxidative glycolysis pathway enhances carbon assimilation and improves the innate metabolic capacity of E. coli for acetyl-CoA derived chemicals including acetate39 and mevalonic acid production40, as predicted in this study. While several heterologous reactions demonstrated conserved improvements in maximum yields across host strains, hierarchical clustering of the heterologous reactions showed that the sets of target reactions differ among host strains (Supplementary Fig. 12). For instance, meso-diaminopimelate dehydrogenase was commonly predicted for l-lysine-derived chemicals (i.e., 7-hydroxyheptanoic acid, pimelic acid, cadaverine, l-2-ammoniohexano-6-lactam, 3,6-diammoniohexanoate, l-lysine, glutaric acid, 5-oxopentanoaic acid, 5-ammoniopentanamide, 5-aminopentanoic acid, and valerolactam) under aerobic conditions using d-glucose as the sole carbon source in E. coli (Supplementary Fig. 13). It should be noted that detailed engineering strategies for introducing heterologous reactions vary with the chemical of interest. For example, ornithine transacetylase, which catalyzes the conversion of l-glutamate to ornithine, was predicted to be a candidate enzyme for improving YT of ornithine production in E. coli, increasing it from 0.9817 mol/mol d-glucose to 1.0142 mol/mol d-glucose under aerobic conditions—a 3.31% increase. Ornithine transacetylase replaces two enzymatic steps—specifically, N-acetylglutamate synthase and acetylornithine deacetylase—in the ornithine biosynthetic pathway (Fig. 2b). Flux-sum analysis, which calculates the sum of the incoming and outgoing fluxes from a metabolite (see Methods), was performed to identify major differences in flux distribution between the two pathways (Fig. 2d). Acetate, acetyl phosphate, and acetyl-CoA showed the largest decreases in flux-sum from the native ornithine biosynthesis pathway to the pathway with the introduced ornithine transacetylase. The native pathway requires acetyl-CoA for N-acetylglutamate synthase and produces acetate via acetylornithine deacetylase. To supply the required acetyl-CoA, the metabolic network must convert acetate back to acetyl-CoA, consuming an ATP via acetate kinase. We performed flux variability analysis (FVA) and confirmed that introducing ornithine transacetylase reduces the metabolic fluxes of acetate kinase and phosphotransacetylase (Fig. 2d). This reduction decreases the energy required to convert acetate into acetyl-CoA. The yield of ornithine production in the pathway with the newly introduced ornithine transacetylase is consequently increased (Fig. 2d). Although C. glutamicum and S. cerevisiae natively possess ornithine transacetylase, experimental studies have demonstrated that further overexpression of ornithine transacetylase-encoding genes (i.e., argJ)—thereby reinforcing the flux through this pathway—can enhance ornithine production in microorganisms41,42.
화학 물질 생산을 위한 선천적 대사 능력 향상
대사 공학은
미생물 세포 공장의 대사 능력 범위 내에서 세포 특성을 최적화하거나
선천적 대사 능력을 향상시켜 화학 물질 생산 능력을 강화한다37.
우리는
표적 화학 물질 생산을 위한 숙주 균주의 대사 능력을 향상시킬 것으로 예상되는 대사 반응을
체계적으로 분석했다.
숙주 균주의 선천적 대사 능력을 향상시키기 위해
두 가지 접근법을 수행했다:
이종 반응을 도입하여 선천적 대사 네트워크를 확장하고,
선천적 대사 네트워크에서 사용되는 보조인자를
비선천적 보조인자로 대체하는 것이다.
내재적 대사 네트워크 확장을 위해, 내재적 대사 반응과 협응력을 발휘할 때 더 효율적인 생합성 경로를 구축할 수 있는 이종 대사 반응을 탐색했다. 이러한 이종 반응들은 모든 대사 반응을 집합한 보편적 모델을 구축하여 수집했다. 보편적 모델은 BiGG 데이터베이스38에서 제공된 보편적 모델을 큐레이션하여 제작되었으며, 3814개의 대사산물과 6846개의 대사 반응을 포함한다. 생산 균주의 GEM에 이 반응들을 추가했을 때 대상 화합물의 YT가 최소 1% 이상 개선되면 수율 향상 이종 반응으로 식별했습니다. 시뮬레이션은 구축된 1360개 GEM(1360종 미생물 대표) 모두에 대해 수행되었으며, 추가되는 이종 반응 수는 3개로 제한했습니다. 연구자들이 관심 있는 더 다양한 조건을 분석할 수 있도록 시뮬레이션 소스 코드도 제공합니다.
시뮬레이션은 해당 화합물에 대해 탄소 또는 에너지 효율적인 생합성 경로를 생성하는 후보 반응을 식별했습니다. 예를 들어, 자일룰로스 5-인산(또는 프럭토스 5-인산)을 아세틸 인산과 글리세랄데히드 3-인산(또는 d-에리스로스 4-인산)으로 전환하는 효소인 포스포케톨라아제는 다양한 숙주 균주에서 화합물의 YT를 개선할 가능성이 가장 높은 것으로 예측되었습니다. 이 효소는 소비된 당으로부터 아세틸-CoA까지 모든 탄소를 보존하는 비산화성 당분해 경로를 촉진한다(그림 2a). 이 비산화성 당분해 경로는 탄소 동화 작용을 강화하고, 아세테이트39 및 메발론산 생산40을 포함한 아세틸-CoA 유래 화합물에 대한 대장균의 선천적 대사 능력을 향상시키는 것으로 입증되었다. 본 연구에서 예측된 바와 같습니다. 여러 이종 반응이 숙주 균주 전반에 걸쳐 최대 수율에서 보존된 개선을 보였지만, 이종 반응의 계층적 클러스터링은 표적 반응 집합이 숙주 균주마다 다르다는 것을 보여주었습니다(보충 그림 12). 예를 들어, 메소-디아미노피멜산 탈수소효소는 l-라이신 유래 화합물(즉, 7-하이드록시헵탄산, 피멜산, 카다베린, l-2-암모니오헥산-6-락탐, 3,6-디아미노헥산산, l-라이신, 글루타르산, 5-옥소펜타노산, 5-암모니오펜타나미드, 5-아미노펜타노산, 발레롤락탐)에 대해 공기 조건에서 E. coli 내 d-글루코스를 유일한 탄소원으로 사용할 때 공통적으로 예측되었다(보충 그림 13). 이종 반응 도입을 위한 세부적인 공학적 전략은 관심 화합물에 따라 달라진다는 점을 유의해야 한다. 예를 들어, l-글루타메이트를 오르니틴으로 전환하는 오르니틴 트랜스아세틸라제는 대장균(E. coli) 내 오르니틴 생산의 YT(효소당 생산량)를 0.9817 mol/mol d-글루코즈에서 1.0142 mol/mol d-글루코즈로 증가시켜 3.31% 향상시키는 후보 효소로 예측되었다. 오르니틴 트랜스아세틸라제는 두 가지 효소 단계를 대체하는데, 구체적으로 N-아미노글루타메이트의 아세틸기 전-글루코스로부터의 YT를 1.0142 mol/mol d-글루코스로 증가시킬 수 있는 후보 효소로 예측되었으며, 이는 3.31% 증가에 해당한다. 오르니틴 트랜스아세틸라제는 오르니틴 생합성 경로에서 두 가지 효소 단계, 즉 N-아세틸글루타메이트 신타제와 아세틸오르니틴 디아세틸라제를 대체한다(그림 2b). 두 경로 간 유동 분포의 주요 차이를 확인하기 위해 대사물질의 유입 및 유출 유동 합계를 계산하는 유동합계 분석(방법 참조)을 수행하였다(그림 2d). 아세테이트, 아세틸인산염, 아세틸-CoA는 기존 오르니틴 생합성 경로에서 오르니틴 트랜스아세틸라제가 도입된 경로로 전환 시 유동합계가 가장 크게 감소하였다. 원래 경로는 N-아세틸글루타메이트 합성효소에 아세틸-CoA를 필요로 하며, 아세틸오르니틴 탈아세틸효소를 통해 아세테이트를 생성한다. 필요한 아세틸-CoA를 공급하기 위해 대사 네트워크는 아세테이트를 아세틸-CoA로 다시 전환해야 하며, 이 과정에서 아세테이트 키나아제를 통해 ATP를 소비한다. 우리는 플럭스 가변성 분석(FVA)을 수행하여 오르니틴 트랜스아세틸레이스 도입이 아세테이트 키나아제와 포스포트랜스아세틸레이스의 대사 플럭스를 감소시킨다는 점을 확인했습니다(그림 2d). 이 감소는 아세테이트를 아세틸-CoA로 전환하는 데 필요한 에너지를 줄입니다. 결과적으로 새로 도입된 오르니틴 트랜스아세틸레이스를 포함한 경로에서의 오르니틴 생산 수율이 증가합니다(그림 2d). 비록 C. glutamicum과 S. cerevisiae가 원래 오르니틴 트랜스아세틸레이스를 보유하고 있지만, 실험 연구에 따르면 오르니틴 트랜스아세틸레이스를 암호화하는 유전자(즉, argJ)를 추가로 과발현하여 이 경로의 유동을 강화하면 미생물에서 오르니틴 생산을 향상시킬 수 있음이 입증되었습니다41,42.
Fig. 2: Improved YT of chemical production via introduction of heterologous reactions.
a Non-oxidative glycolysis for improving YT of acetate production. b Use of ornithine transacetylase for improving YT of ornithine production. c Use of NADH-dependent homoserine dehydrogenase, homoserine deaminase, and pyruvate carboxylase for improving YT of propan-1-ol production. Black arrows indicate native metabolic pathways, while blue arrows indicate metabolic pathways with heterologous reactions. Dotted arrows show bypassed reactions due to the addition of heterologous reactions. The added heterologous reactions are shaded in cyan. d Flux-sum differences of metabolites between the native pathway and the pathway with introduced ornithine transacetylase for YT calculation. Metabolites are sorted in descending order of flux-sum decrease. e Flux-sum differences of metabolites between the native pathway and the pathway with introduced NADH-dependent homoserine dehydrogenase, homoserine deaminase, and pyruvate carboxylase for YT calculation. Metabolites are sorted in descending order of flux-sum increase. Flux values in parentheses are the lower and upper bounds of the FVA results when maximizing the target chemical production flux. FVA results for native pathways and pathways with introduced heterologous reactions are in black and blue, respectively. Abbreviations for metabolites and reactions are available in Supplementary Data 24. Source data are provided as a Source Data file.
그림 2: 이종 반응 도입을 통한 화학 물질 생산의 개선된 YT.
a 아세테이트 생산 YT 향상을 위한 비산화성 당분해. b 오르니틴 생산 YT 향상을 위한 오르니틴 트랜스아세틸라제 활용. c 프로판-1-올 생산 YT 향상을 위한 NADH 의존성 호모세린 탈수소효소, 호모세린 탈아미노효소 및 피루브산 카복실라제 활용. 검은색 화살표는 원생 대사 경로를, 파란색 화살표는 이종 반응이 도입된 대사 경로를 나타냅니다. 점선 화살표는 이종 반응 추가로 우회된 반응을 표시합니다. 추가된 이종 반응은 청록색으로 음영 처리되었습니다. d YT 계산을 위한 원생 경로와 오르니틴 트랜스아세틸레이즈가 도입된 경로 간 대사체 플럭스 합 차이. 대사산물은 플럭스 합 감소량 내림차순으로 정렬됨. e YT 계산을 위해 도입된 NADH 의존성 호모세린 탈수소효소, 호모세린 탈아미노효소 및 피루브산 카복실라아제를 포함한 경로와 원생 경로 간 대사산물 플럭스 합 차이.
대사산물은 플럭스 합 증가량 내림차순으로 정렬됨. 괄호 안 플럭스 값은 목표 화학물질 생산 플럭스 극대화 시 FVA 결과의 하한 및 상한임. 원생 경로와 이종 반응 도입 경로의 FVA 결과는 각각 검정색과 파란색으로 표시됨. 대사산물 및 반응 약어는 보충 자료 24에서 확인 가능함. 원본 데이터는 Source Data 파일로 제공됨.
Exploring heterologous reactions also suggests biosynthetic pathways that offer advantages in redox balance. For propan-1-ol production, NADH-dependent homoserine dehydrogenase, homoserine deaminase, and pyruvate carboxylase were predicted to improve the YT in E. coli from 0.7059 mol/mol d-glucose to 1.0909 mol/mol d-glucose under anaerobic conditions—a 54.55% increase (Fig. 2c). The propan-1-ol biosynthetic pathway without these heterologous reactions requires an ATP and two NADPH from glyceraldehyde 3-phosphate. Furthermore, producing the intermediate l-aspartate from l-glutamate and oxaloacetate indirectly requires NADPH, necessitating a sufficient NADPH supply for propan-1-ol production. The introduced enzymes reduce the dependency on the NADPH pool. The native NADPH-dependent homoserine dehydrogenase consumes one NADPH, while the introduced NADH-dependent homoserine dehydrogenase consumes an NADH to convert l-aspartate 4-semialdehyde to l-homoserine, thus reducing the total NADPH requirement for propan-1-ol production. Additionally, homoserine deaminase bypasses the ATP-requiring native pathway to convert l-homoserine to 2-oxobutanoate, and pyruvate carboxylase converts pyruvate to oxaloacetate, generating ATP in the reaction. Therefore, the heterologous reactions make the metabolic network more ATP efficient, redirecting metabolic fluxes from acetate and formate production pathways to the pentose phosphate pathway (Supplementary Fig. 14). The addition of heterologous reactions increased the flux-sum of metabolites in the pentose phosphate pathway, demonstrating that the engineered metabolic network can redirect metabolic fluxes toward the pentose phosphate pathway to meet the high NADPH demand rather than producing ATP from fermentation pathways (Fig. 2e).
We further assessed how the number of added heterologous reactions affects increases in maximum yields. Among the 784,774 heterologous reaction targets that improve YT of chemicals, 57,045 involved a single reaction, 191,901 involved two reactions, and 535,828 involved three reactions. The distribution of the percentage increases in YT differed significantly across these groups (Kruskal-Wallis H-test; p-value of 0.000, n = 784,774) with median increases of 6.296%, 8.800%, and 10.37% for the addition of one, two, and three reactions, respectively (Supplementary Fig. 15). However, introducing multiple heterologous reactions should be carefully optimized to balance yield improvements with the potential metabolic burden on the host.
Similarly, we identified metabolically efficient biosynthetic pathways for chemical production in industrial microorganisms (Supplementary Data 6–10). However, the predicted heterologous reaction sets should be carefully selected for validation experiments. Although all metabolic reactions used in the analysis are mass- and charge-balanced, flux balance analysis (FBA) often fails to exclude thermodynamically infeasible solutions due to insufficient knowledge of the reaction directions. To address this issue, we employed loopless FBA to reduce thermodynamically infeasible solutions43. Thermodynamics-based FBA, which utilizes metabolite concentration ranges and Gibbs free energy of reactions to predict feasible flux profiles, could further reduce infeasible solutions. However, incomplete knowledge of genome-scale metabolome data and reaction thermodynamics hinders the application of such thermodynamics-based network analysis. Additionally, the search space for non-native metabolic reactions is constrained by the reactions in the universal model. Therefore, further improvement of the universal model in both quality and quantity is necessary for better construction of metabolically efficient biosynthetic pathways. Community efforts to elucidate genome-scale metabolome, kinetome, and other relevant omics data will facilitate the construction of efficient biosynthetic pathways for chemical production.
Next, we analyzed how to improve metabolic capacity by modifying the use of cofactors in native metabolic reactions. Addressing the burden of cofactor usage is crucial for the biosynthesis of chemicals, particularly for those that are highly oxidized or reduced. To alleviate this burden in biosynthetic pathways, metabolic engineering studies have employed strategies such as changing enzymes44 or engineering the cofactor specificities45. For example, GEMs have been utilized to identify cofactors for exchange in constructing microbial cell factories46,47. Furthermore, the development of computational designs48 or adaptive evolution-based approaches49 for altering enzyme specificity makes cofactor swapping a more viable strategy for metabolic engineering. To propose potential strategies for improving metabolic capacity through changes in cofactor usage in metabolic reactions, we analyzed the effect of cofactor exchanges in the 1360 constructed GEMs to improve the YT of target chemicals. Metabolic reactions involving NADPH or NADH were considered candidates for cofactor exchange. Cofactor exchanges were identified if the YT of the target chemical improved when NADP and NADPH in the candidate reactions were exchanged to NAD and NADH, or vice versa (Supplementary Data 11–13).
For example, under aerobic conditions using d-glucose as the sole carbon source in E. coli, 80.6% (29/36) of the biosynthetic pathways for chemicals derived from acetyl-CoA showed improved YT’s when the cofactor (NADH) of the glyceraldehyde 3-phosphate dehydrogenase was exchanged for NADPH. This cofactor exchange enables the metabolic network to produce NADPH from glycolysis, reducing the high demand for NADPH in the production of acetyl-CoA-derived fatty acids and isoprenoids. Mevalonic acid, a key precursor in the mevalonate pathway that produces isoprenoids, showed improved YT from 0.8000 mol/mol d-glucose to 0.8229 mol/mol d-glucose when the cofactor (NAD) of the glyceraldehyde 3-phosphate dehydrogenase was swapped for NADP. Introducing NADPH-producing reaction reduces metabolic fluxes toward NADP transdehydrogenase and NADH dehydrogenase, providing NADPH required for 3-hydroxy-3-mehtylglutaryl-CoA reductase from the glyceraldehyde 3-phosphate dehydrogenase (Fig. 3a). Similarly, 61.2% (30/49) of the 2-oxoglutarate-derived chemicals showed improved YTs when the cofactor (NADPH) of the glutamate dehydrogenase was exchanged for NADH, under aerobic conditions using d-glucose as the sole carbon source. Putrescine, a four-carbon diamine used in manufacturing engineering plastics, showed improved YT from 0.9907 mol/mol d-glucose to 1.0221 mol/mol d-glucose when the cofactor (NADPH) of the glutamate dehydrogenase was exchanged for NADH (Fig. 3b). Exchanging the required NADPH for NADH in glutamate dehydrogenase allows the metabolic network to provide cofactors from glycolysis, without diverting additional metabolic fluxes toward NADPH-producing pathways. Experimental studies have demonstrated that employing enzymes with different cofactor specificities can substantially improve chemical production. For example, engineering glyceraldehyde 3-phosphate dehydrogenase to favor NADP+ increased l-lysine production in C. glutamicum50, and overexpression of an NADP+-dependent glyceraldehyde 3-phosphate dehydrogenase gene enhanced 3-hydroxypropionic acid production in E. coli51. Furthermore, introducing an NADH-dependent aspartate-semialdehyde dehydrogenase improved l-homoserine production in E. coli52. These experimental findings support our simulation predictions on cofactor exchange strategies.
이종 반응 탐색은 또한 산화환원 균형 측면에서 이점을 제공하는 생합성 경로를 제시합니다. 프로판-1-올 생산의 경우, NADH 의존성 호모세린 탈수소효소, 호모세린 탈아미노효소 및 피루브산 카복실라아제가 혐기 조건에서 E. coli의 YT를 d-글루코스 몰당 0.7059 mol/mol에서 1.0909 mol/mol로 54.55% 증가시킬 것으로 예측되었습니다(그림 2c)이러한 이종 반응이 없는 프로판-1-올 생합성 경로는 글리세랄데히드 3-인산으로부터 ATP 1분자와 NADPH 2분자를 필요로 한다. 또한 중간체인 l-아스파르트산을 l-글루타메이트와 옥살로아세테이트로부터 생산하는 과정은 간접적으로 NADPH를 필요로 하므로, 프로판-1-올 생산을 위해 충분한 NADPH 공급이 필수적이다. 도입된 효소들은 NADPH 풀에 대한 의존도를 감소시킨다. 원래의 NADPH 의존성 호모세린 탈수소효소는 NADPH 하나를 소비하는 반면, 도입된 NADH 의존성 호모세린 탈수소효소는 l-아스파르트산 4-세미알데히드를 l-호모세린으로 전환하는 데 NADH 하나를 소비하므로, 프로판-1-올 생산에 필요한 총 NADPH 요구량을 감소시킵니다. 또한, 호모세린 탈아미노효소는 l-호모세린을 2-옥소부탄산으로 전환하는 데 ATP가 필요한 기존 경로를 우회하며, 피루브산 카복실라제는 피루브산을 옥살아세트산으로 전환하면서 반응 과정에서 ATP를 생성합니다. 따라서 이종 반응들은 대사 네트워크의 ATP 효율성을 높여, 아세테이트 및 포르메이트 생산 경로에서 펜토스 인산 경로로 대사 유동을 재분배합니다(보충 그림 14).. 이종 반응의 추가로 오탄당 인산 경로의 대사체 유동량 합계가 증가하여, 공학적으로 설계된 대사 네트워크가 발효 경로에서 ATP를 생산하기보다 높은 NADPH 수요를 충족시키기 위해 대사 유동량을 오탄당 인산 경로로 재분배할 수 있음을 입증했습니다(그림 2e).
우리는 추가된 이종 반응의 수가 최대 수율 증가에 미치는 영향을 추가로 평가했습니다. 화학물질의 YT를 향상시키는 784,774개의 이종 반응 대상 중 57,045개는 단일 반응, 191,901개는 두 반응, 535,828개는 세 반응을 포함했다. YT 증가율 분포는 이들 그룹 간 유의미한 차이를 보였다(Kruskal-Wallis H-검정; p값 0.000, n = 784,774)이며, 각각 1개, 2개, 3개 반응 추가 시 중앙값 증가율은 6.296%, 8.800%, 10.37%였습니다(보충 그림 15). 그러나 다중 이종 반응 도입은 수율 향상과 숙주에 대한 잠재적 대사 부담 사이의 균형을 맞추기 위해 신중하게 최적화되어야 합니다.
마찬가지로, 산업 미생물에서 화학 물질 생산을 위한 대사 효율적인 생합성 경로를 확인했습니다(보충 자료 6–10). 그러나 예측된 이종 반응 세트는 검증 실험을 위해 신중하게 선택해야 합니다. 분석에 사용된 모든 대사 반응은 질량 및 전하 균형을 이루지만, 플럭스 균형 분석(FBA)은 반응 방향에 대한 지식 부족으로 열역학적으로 불가능한 해법을 배제하지 못하는 경우가 많습니다. 이 문제를 해결하기 위해 우리는 열역학적으로 불가능한 해법을 줄이기 위해 루프리스 FBA를 활용했습니다43. 대사체 농도 범위와 반응의 깁스 자유 에너지를 활용하여 가능한 플럭스 프로파일을 예측하는 열역학 기반 FBA는 불가능한 해법을 더욱 줄일 수 있습니다. 그러나 게놈 규모 대사체 데이터와 반응 열역학에 대한 불완전한 지식은 이러한 열역학 기반 네트워크 분석의 적용을 방해합니다. 또한 비고유 대사 반응에 대한 탐색 공간은 보편적 모델의 반응들에 의해 제한됩니다. 따라서 대사 효율이 높은 생합성 경로를 더 잘 구축하기 위해서는 보편적 모델의 질적·양적 개선이 필요합니다. 게놈 규모 대사체, 키네토메 및 기타 관련 오믹스 데이터를 규명하기 위한 공동체적 노력은 화학 생산을 위한 효율적인 생합성 경로 구축을 촉진할 것입니다.
다음으로, 기존 대사 반응에서 보조인자 사용 방식을 수정하여 대사 능력을 향상시키는 방법을 분석했습니다. 보조인자 사용 부담을 해결하는 것은 화학물질 생합성, 특히 고도로 산화되거나 환원되는 물질의 생합성에 매우 중요합니다. 생합성 경로에서 이러한 부담을 완화하기 위해 대사공학 연구에서는 효소 변경44이나 보조인자 특이성 공학화45와 같은 전략을 활용해 왔습니다. 예를 들어, GEMs는 미생물 세포 공장 구축 시 교환할 보조인자를 식별하는 데 활용되었습니다46,47. 또한 효소 특이성 변경을 위한 계산적 설계48 또는 적응 진화 기반 접근법49의 발전은 보조인자 교환을 대사 공학에서 더 실행 가능한 전략으로 만듭니다. 대사 반응에서 보조인자 사용 변화를 통한 대사 능력 향상 잠재 전략을 제안하기 위해, 1360개의 구축된 GEM에서 보조인자 교환이 표적 화합물의 YT 향상에 미치는 영향을 분석하였다. NADPH 또는 NADH가 관여하는 대사 반응은 보조인자 교환 후보로 고려되었다 . 후보 반응에서 NADP와 NADPH를 NAD와 NADH로, 또는 그 반대로 교환했을 때 목표 화합물의 YT가 개선되면 보조인자 교환이 확인되었다(보충 자료 11–13).
예를 들어, E. coli에서 d-글루코스를 유일한 탄소원으로 사용하는 호기성 조건에서, 글리세랄데하이드 3-인산 탈수소효소의 보조인자(NADH)를 NADPH로 교환했을 때 아세틸-CoA 유래 화학물질의 생합성 경로 중 80.6%(36개 중 29개)에서 YT가 개선되었습니다. 이러한 보조인자 교환을 통해 대사 네트워크는 당분해에서 NADPH를 생산할 수 있게 되어, 아세틸-CoA 유래 지방산 및 이소프렌오이드 생산에 필요한 NADPH의 높은 수요를 줄일 수 있습니다. 이소프렌을 생성하는 메발론산 경로의 핵심 전구체인 메발론산은 글리세롤알데하이드 3-인산 탈수소효소의 보조인자(NAD)를 NADP로 교체했을 때 YT가 0.8000 mol/mol d-글루코스에서 0.8229 mol/mol d-글루코스로 향상되었습니다. NADPH 생성 반응을 도입하면 NADP 트랜스데하이드로게나제와 NADH 탈수소효소로의 대사 흐름이 감소하여 글리세르알데하이드 3-인산 탈수소효소로부터 3-하이드록시-3-메틸글루타릴-CoA 환원효소에 필요한 NADPH를 공급합니다(그림 3a). 마찬가지로, 2-옥소글루타레이트 유래 화합물의 61.2%(30/49)는 글루타메이트 탈수소효소의 보조인자(NADPH)를 NADH로 교체했을 때, d-글루코스를 유일한 탄소원으로 사용하는 호기성 조건 하에서 YT가 개선되었습니다.공학용 플라스틱 제조에 사용되는 4탄소 디아민인 푸트레신은 글루타메이트 탈수소효소의 보조인자(NADPH)를 NADH로 교체했을 때 YT가 d-글루코스 0.9907 mol/mol에서 1.0221 mol/mol로 향상되었습니다(그림 3b). 글루타메이트 탈수소효소에서 필요한 NADPH를 NADH로 교체함으로써, 대사 네트워크는 NADPH 생성 경로로 추가 대사 유동을 전환하지 않고도 당분해로부터 보조인자를 공급할 수 있게 된다. 실험 연구에 따르면 서로 다른 보조인자 특이성을 가진 효소를 활용하면 화합물 생산을 크게 향상시킬 수 있음이 입증되었다. 예를 들어, 글리세랄데하이드-3-인산 탈수소효소를 NADP+ 선호형으로 개량함으로써 C. glutamicum에서 l-라이신 생산량이 증가하였으며50, NADP+ 의존성 글리세랄데하이드-3-인산 탈수소효소 유전자의 과발현은 E. coli에서 3-하이드록시프로피온산 생산을 증진시켰다51. 또한, NADH 의존성 아스파르트산 세미알데하이드 탈수소효소를 도입함으로써 E. coli52에서 l-호모세린 생산량이 향상되었다. 이러한 실험적 결과는 보조인자 교환 전략에 대한 우리의 시뮬레이션 예측을 뒷받침한다.
Fig. 3: Improved YT of chemical production via cofactor exchange.
a Improvement of YT of mevalonic acid production by exchanging the native cofactor (NAD) of glyceraldehyde 3-phosphate dehydrogenase with NADP. b Improvement of YT of putrescine production by exchanging the native cofactor (NADPH) of glyceraldehyde 3-phosphate dehydrogenase with NADH. Blue arrows represent cofactor swapped reactions. c Flux-sum differences of metabolites between the native pathway and the pathway with cofactor-swapped glyceraldehyde 3-phosphate dehydrogenase for YT calculation. Metabolites are sorted in descending order of flux-sum decrease. d Flux-sum differences of metabolites between the native pathway and the pathway with cofactor-swapped glutamate dehydrogenase for YT calculation. Metabolites are sorted in descending order of flux-sum decrease. Flux values in parentheses are the lower and upper bounds of the FVA results when maximizing the target chemical production flux. FVA results for native pathways and pathways with cofactor swapped reactions are in black and blue, respectively. Abbreviations of metabolites and reactions are available in Supplementary Data 24. Source data are provided as a Source Data file.
These cofactor preferences are largely conserved across different carbon sources but exhibit distinct patterns under varying aeration conditions (Supplementary Fig. 16). For instance, in E. coli under aerobic conditions, the predicted cofactor exchange targets predominantly favor NADPH for isoprenoids (e.g., bisabolene, carotene, farnesene, farnesol, farnesyl diphosphate, geraniol, geranyl diphosphate, geranylgeraniol, geranylgeranyl diphosphate, limonene, myrcene, pinene, sabinene, and santalene) and for aromatic compounds (e.g., 2-phenylethanol, 4-amino-l-phenylalanine, 4-aminocinnamic acid, 4-aminophenyl ethylamine, 4-hydroxyphenyl acetaldehyde, 4-hydroxyphenylacetate, and resveratrol). In contrast, under microaerobic and anaerobic conditions, both NADH- and NADPH-dependent reactions are predicted with similar frequencies for these chemical groups. These findings suggest that while the inherent cofactor demands of biosynthetic pathways remain relatively consistent, the optimal cofactor utilization strategy is modulated by other factors (e.g., oxygen availability), reflecting shifts in redox balance and energy requirements. Consequently, tailoring cofactor exchange to the specific condition may be important for optimizing production performance. Although engineering or designing enzymes with specific cofactor preferences is challenging, recent advances in machine learning are expected to facilitate the enzyme engineering and design53,54,55. In this context, we present metabolic engineering strategies that employ cofactor swapping for the production of 235 bio-based chemicals (Supplementary Data 11–13).
While our in silico simulations were designed to predict engineering strategies (i.e., the addition of heterologous reactions and cofactor exchanges) to improve YT of chemical production, we also found these predicted strategies enhanced YA of chemical production (Supplementary Note 2). This indicates that the resource presented in this study have practical applicability, potentially leading to more efficient microbial production processes.
Rewiring metabolic fluxes toward target chemicals
Although a high theoretical yield of a biosynthetic pathway can indicate its maximum potential efficiency, the actual metabolic flux toward the target chemical does not necessarily correlate with the theoretical yield. Therefore, constructing an effective microbial cell factory requires rewiring metabolic fluxes toward the target chemical, rather than solely relying on biosynthetic pathways with high theoretical yields. To suggest metabolic rewiring strategies for chemicals, we conducted flux variability scanning based on enforced objective function (FVSEOF)56,57 and iBridge56,57 to predict metabolic reactions to be up-regulated or down-regulated to enhance the production fluxes of target chemicals. Initially, FVSEOF analysis was performed for each chemical production using each GEM of the host strains (Supplementary Data 14–18). A metabolic reaction was considered an up-regulation candidate if its flux showed a positive Pearson correlation with the target chemical production flux (see Online Methods). To provide general engineering strategies for chemical production, we analyzed target reaction profiles, which are sets of metabolic reactions predicted to be up-regulation targets, for 272 metabolic pathways leading to the biosynthesis of 235 chemicals (Fig. 4). Hierarchical clustering of these reaction profiles suggests that target chemicals within the same cluster share common engineering strategies. For instance, isoprenoids and metabolites in the isoprenoid biosynthetic pathways formed a distinct cluster (cluster A in Fig. 4). In this cluster, metabolic reactions within the isopentenyl diphosphate biosynthetic pathways, as well as those in the oxidative pentose phosphate pathway (glucose 6-phosphate dehydrogenase, 6-phosphogluconolactonase, and phosphogluconate dehydrogenase) and lower glycolysis (glyceraldehyde-3-phosphate dehydrogenase, enolase, and pyruvate dehydrogenase), were predicted to be up-regulation targets for 20 biosynthetic pathways in cluster A. This highlights the importance of enhancing metabolic fluxes toward key precursor production (acetyl-CoA and isopentenyl diphosphate) and NADPH production to meet the high cofactor demand for isoprenoid production.
Fig. 4: Metabolic pathways for bio-based chemicals with cluster groups predicted by metabolic reaction profiles using FVSEOF.
Metabolic reaction profiles, representing the types of predicted metabolic reactions to be up-regulated for improved chemical production, were analyzed across 272 metabolic pathways which lead to the production of 235 chemicals. Pairwise similarities of metabolic reaction profiles were calculated using cosine similarity. Chemicals with similar metabolic reaction profiles were grouped into distinct clusters (Supplementary Fig. 18), and each chemical was colored according to the cluster it belongs to. When multiple pathways for a chemical were available and their clusters differ, both clusters are denoted. The simulation was performed for E. coli under aerobic conditions with d-glucose as the sole carbon source. Metabolic pathways for chemicals in each cluster are available in Supplementary Figs. 19–28.
Metabolic flux optimization strategies can vary depending on the target chemicals, even if they share common precursors. For example, aromatic compounds derived from d-erythrose 4-phosphate were grouped into two clusters, cluster I and J (Supplementary Fig. 17). While the chorismate biosynthetic pathway was predicted to be an up-regulation target for both clusters, the oxidative pentose phosphate pathway was exclusively predicted for 30 biosynthetic pathways in cluster I, and the TCA cycle (citrate synthase, aconitase, 2-oxoglutarate dehydrogenase, succinate dehydrogenase, and fumarase) was exclusively predicted for 13 biosynthetic pathways in cluster J. This suggests that, despite sharing the same precursor, the engineering strategies for different chemicals can vary considerably, emphasizing the need for tailored metabolic engineering approaches to optimize metabolic fluxes for each target chemical.
Similarly, iBridge analysis was conducted for each chemical production in each GEM of the host strains to predict reactions to be regulated, providing metabolic engineering strategies complementary to those from FVSEOF (Supplementary Data 19–23). It is important to note that these two algorithms capture different aspects of metabolic networks. Specifically, FVSEOF identifies reactions whose fluxes exhibit a strong positive correlation with the target production flux, thereby robustly predicting up-regulation targets. In contrast, iBridge leverages a metabolite-centric analysis based on flux covariances, enabling the identification of both up- and down-regulation targets that may be missed by FVSEOF. For example, hierarchical clustering of reaction profiles from iBridge showed that pyruvate kinase, not predicted by FVSEOF, was predicted as an up-regulation target for isoprenoids and metabolites in the isoprenoid biosynthetic pathways in cluster A (Fig. 5). As another example, aromatic compounds derived from d-erythrose 4-phosphate formed a distinct clade, cluster B (Fig. 5). In this cluster, 3-deoxy-d-arabino-heptulosonate 7-phosphate synthetase and transaldolase were predicted as up-regulation targets, where 3-deoxy-d-arabino-heptulosonate 7-phosphate synthetase was also predicted by FVSEOF while transaldolase was unique to iBridge. iBridge also provided down-regulation targets not predicted by FVSEOF. For instance, phosphoglycerate dehydrogenase was predicted as a down-regulation target for cluster D, which includes pyruvate-derived chemicals (e.g., l-leucine, l-valine, isobutanol, isovalerate) and acetyl-CoA-derived chemicals (e.g., butyrate, hexanoate, octanoate, propan-2-ol). Similarly, phosphoenolpyruvate carboxylase was predicted as a down-regulation target for cluster E, which includes pyruvate-derived chemicals (e.g., acetoin, 2,3-butaneiol, propionate, isopentane) (Fig. 5). These differences highlight that while some engineering targets were commonly predicted by both approaches, each algorithm captures distinct metabolic features. Consequently, the complementary use of FVSEOF and iBridge enriches our resource by providing a broader set of potential engineering strategies tailored to specific chemicals (Supplementary Data 14–23).
Fig. 5: Metabolic pathways for bio-based chemicals with cluster groups predicted by metabolic reaction profiles using iBridge.
Metabolic reaction profiles, representing the types of predicted metabolic reactions to be up- and down-regulated for improved chemical production, were analyzed across 272 metabolic pathways which lead to the production of 235 chemicals. Pairwise similarities of metabolic reaction profiles were calculated using cosine similarity. Chemicals with similar metabolic reaction profiles were grouped into distinct clusters (Supplementary Fig. 29), and each chemical was colored according to the cluster it belongs to. When multiple pathways for a chemical were available and their clusters differ, both clusters are denoted. The hierarchical clustering heatmap of metabolic reaction profiles and metabolic pathways for chemicals in each cluster are available in Supplementary Figs. 30–35.
Discussion
Planning a metabolic engineering project necessitates an extensive search through the entire decision-making process, including the selection of target chemicals, host strains, and pathways to be engineered. Finding an optimal strategy for the project is challenging without systematically exploring the vast metabolic space. To aid the initiation of a metabolic engineering project, we provide a comprehensive evaluation of the capabilities of microbial cell factories. The maximum yield of a chemical indicates how efficiently a biosynthesis pathway can transform a carbon source into the target chemical, thereby guiding metabolic engineers in selecting the most optimal host strain by comparing the maximum yields across different strains.
In this study, we calculated the YT and YA of 235 bio-based chemicals in five representative host strains for metabolic engineering (i.e., B. subtilis, C. glutamicum, E. coli, P. putida, and S. cerevisiae). It should be noted that the development of genome engineering tools for non-model organisms increasingly enables the use of non-traditional strains for metabolic engineering58,59,60. Therefore, expanding the analysis of this study to encompass all available genome data could inspire the use of less-explored organisms for metabolic engineering as well.
Conventional constraint-based modeling approaches (e.g., FBA) do not account for gene expression, regulatory network, or allocation of macromolecules within an organism61,62,63. Although our study is limited to calculating the maximum yields of bio-based chemicals using FBA, integrating multi-scale mechanisms will allow for a more accurate calculation of the biotechnologically achievable maximum yields of these chemicals. Moreover, since we neglected the target chemical-specific transport reactions in our analysis, further characterization and inclusion of these exporters will enable more accurate yield calculations. Recent advancements in metabolic modeling have highlighted the importance of integrating enzyme kinetics and proteome constraints to understand the metabolism of microorganisms. Although proteome-integrated GEMs offer valuable insights under specific experimental conditions, their applicability is limited when exploring diverse environmental and substrate conditions. To assess the potential impact of such constraints on maximum yields, we compared the maximum yields obtained from iML1515, a GEM of E. coli, and its enzyme-constrained counterpart, eciML1515 (Supplementary Note 1)64. While the magnitude of maximum yields from the enzyme-constrained model can vary, the overall trends and distributions were statistically indistinguishable. These findings demonstrate that while enzyme constraints could provide a more realistic value of maximum yields, the general trends across conditions remain consistent. Thus, the current GEMs provide robust results for exploring metabolic capacities over a wide range of conditions, although further incorporation of enzymatic information continues to be a valuable tool when more precise, condition-specific yield calculations are required.
We also proposed engineering strategies to enhance the innate metabolic capacities of microbial cell factories or to rewire their metabolism toward target chemical production. To identify heterologous reactions for introduction into a microbial cell factory, we utilized and curated a universal model that accounts for all reported metabolic reactions. It is evident that the search of heterologous reactions is limited by the quality and quantity of the universal model. As most currently available GEMs have been reconstructed using highly curated reference GEMs or established reaction databases, the universal model is constrained by limited knowledge of biological reactions rather than reflecting the extensive metabolic space of nature. Developing GEM reconstruction pipelines that directly extract specific metabolic reactions from genomes would enable the exploration of more diverse and plausible metabolic engineering strategies. Altogether, the resources showcase 42,976 cases detailing the capacities of host strains for 235 bio-based chemicals under different aeration conditions using different carbon sources (5440 cases for B. subtilis, 9792 cases for C. glutamicum, 11,424 cases for E. coli, 3264 cases for P. putida, and 13,056 cases for S. cerevisiae), alongside 1,925,500 cases detailing engineering strategies (784,774 heterologous reaction targets + 32,867 cofactor exchange targets + 613,863 targets identified by FVSEOF + 493,996 targets identified by iBridge). These resources provided in this study will be useful for selecting a host strain, improving innate metabolic capacity by constructing more efficient metabolic pathways through the introduction of heterologous metabolic reactions and cofactor exchanges, and identifying target reactions for up- and down-regulation to enhance the bio-based production of chemicals.
Selecting the production strain needs to consider various factors, such as growth rate, maximum achievable or optimal/desirable cell concentration, culture condition (e.g., temperature, pH, nutritional requirement, and medium cost), ease of product purification, GRAS status, and others, in addition to the maximum theoretical and achievable yields presented in this study. While the presented maximum yields alone do not capture all dynamic aspects such as growth kinetics and process-specific conditions, they provide a valuable approximation for assessing the inherent metabolic capacity of different strains. As such, this resource serves as an essential reference for narrowing down candidate strains for further experimental validation. Moreover, when combined with additional criteria that reflect the conditions of interest, such as the high productivity achieved in fed-batch fermentations driven by rapid cell growth, the resource can guide strain selection and further cell factory design. Although our approach does not offer a complete solution, it will play an essential role in advancing towards the development of high-performing microbial cell factories.
토론
대사공학 프로젝트를 계획하려면
표적 화학물질, 숙주 균주, 공학화할 경로 선택을 포함한
전체 의사결정 과정을 광범위하게 탐색해야 합니다.
광대한 대사 공간을 체계적으로 탐색하지 않고서는
프로젝트에 대한 최적의 전략을 찾는 것이 어렵습니다.
대사공학 프로젝트의 시작을 돕기 위해,
우리는 미생물 세포 공장의 능력을 종합적으로 평가합니다.
화학물질의 최대 수율은
생합성 경로가 탄원료를 표적 화학물질로 얼마나 효율적으로 전환할 수 있는지를 나타내며,
이를 통해 대사공학자들이 서로 다른 균주 간의 최대 수율을 비교하여
가장 최적의 숙주 균주를 선택하도록 안내합니다.
본 연구에서는
대사공학에 대표적으로 사용되는 5가지 숙주 균주
(즉, B. subtilis, C. glutamicum, E. coli, P. putida, S. cerevisiae)에서
235종의 바이오 기반 화학물질에 대한 YT 및 YA를 계산했습니다.
비모델 생물체에 대한 게놈 공학 도구 개발이 점차 진전되면서
대사공학에 비전통적 균주를 활용할 수 있는 가능성이 높아지고 있다는
따라서
본 연구의 분석을 이용 가능한 모든 게놈 데이터로 확장하면,
덜 탐구된 생물체들도 대사공학에 활용할 수 있는 계기가 될 수 있다.
기존의 제약 기반 모델링 접근법(예: FBA)은
유전자 발현, 조절 네트워크, 또는 생물체 내 고분자 물질의 배분을 고려하지 않는다61,62,63.
본 연구는
FBA를 이용한 바이오 기반 화학물질의 최대 수율 계산에 국한되지만,
다중 규모 메커니즘을 통합하면 이
러한 화학물질의 생물공학적으로 달성 가능한 최대 수율을 보다 정확하게 계산할 수 있을 것이다.
또한 분석 과정에서 대상 화학물질 특이적 수송 반응을 고려하지 않았으므로,
이러한 수출체(exporter)의 추가 특성화 및 포함을 통해 수율 계산의 정확도를 높일 수 있을 것이다.
최근 대사 모델링의 발전은
미생물 대사를 이해하기 위해 효소 동역학 및 단백질체 제약 조건을 통합하는 것의
중요성을 부각시켰다.
단백질체 통합 GEM(대사체 기반 모델)은
특정 실험 조건 하에서 유용한 통찰력을 제공하지만,
다양한 환경 및 기질 조건을 탐구할 때는 적용성이 제한된다.
이러한 제약이 최대 수율에 미치는 잠재적 영향을 평가하기 위해,
우리는 대장균(E. coli)의 GEM인 iML1515와 그 효소 제약 모델인 eciML1515(보충 노트 1)64에서 얻은
최대 수율을 비교했습니다.
효소 제약 모델의 최대 수율 규모는 달라질 수 있지만,
전반적인 경향과 분포는 통계적으로 구분할 수 없었습니다.
이러한 결과는
효소 제약이 최대 수율의 보다 현실적인 값을 제공할 수 있지만,
조건 전반에 걸친 일반적인 경향성은 일관되게 유지됨을 보여줍니다.
따라서
현재의 GEM은 광범위한 조건에서 대사 능력을 탐구하는 데
견고한 결과를 제공하지만,
보다 정밀한 조건 특이적 수율 계산이 필요할 때는 효
소 정보의 추가 통합이 여전히 유용한 도구임을 시사합니다.
또한 미생물 세포 공장의 선천적 대사 능력을 향상시키거나 목표 화학물질 생산을 위해 대사를 재구성하는 공학적 전략을 제안하였다. 미생물 세포 공장에 도입할 이종 반응을 식별하기 위해, 보고된 모든 대사 반응을 고려한 범용 모델을 활용하고 정제하였다. 이종 반응 탐색은 범용 모델의 질과 양에 의해 제한받는다는 점이 명백하다.
현재 이용 가능한 대부분의 GEM(대사공정모델)은
고도로 정제된 참조 GEM이나 확립된 반응 데이터베이스를 활용해 재구성되었기에,
이 보편적 모델은 자연의 광범위한 대사 공간을 반영하기보다는
제한된 생물학적 반응 지식에 얽매여 있습니다.
게놈에서 특정 대사 반응을 직접 추출하는 GEM 재구성 파이프라인을 개발한다면
더 다양하고 실현 가능한 대사공학 전략을 탐구할 수 있을 것입니다.
종합적으로,
이 자료들은 다양한 통기 조건과 탄소원을 사용한 235종의 바이오 기반 화학물질에 대한
숙주 균주의 생산 능력을 상세히 기술한 42,976건의 사례
(B. subtilis 5,440건, C. glutamicum 9,792건, E. coli 11,424건, P. putida 3,264건, S. cerevisiae 13,056건)과 함께,
1,925,500건의 공학적 전략 사례(784,774건의 이종 반응 표적 + 32,867건의 보조인자 교환 표적 + FVSEOF로 식별된 613,863건의 표적 + iBridge로 식별된 493,996건의 표적)를 상세히 기술하고 있습니다.
본 연구에서 제공된 이러한 자원들은
숙주 균주 선택, 이종 대사 반응 및 보조인자 교환 도입을 통한
효율적인 대사 경로 구축으로
선천적 대사 능력 향상, 그리고
생물 기반 화학물질 생산 증진을 위한 상향/하향 조절 대상 반응 식별에 유용할 것이다.
생산 균주 선정 시 본 연구에서 제시된 최대 이론적 및 달성 가능 수율 외에도
성장 속도, 최대 달성 가능 또는
최적/바람직한 세포 농도, 배양 조건(예: 온도, pH, 영양 요구량, 배지 비용),
제품 정제 용이성, GRAS 상태 등
다양한 요소를 고려해야 합니다.
제시된 최대 수율만으로는
성장 동역학 및 공정별 조건과 같은 모든 동적 측면을 포착하지는 못하지만,
다양한 균주의 고유한 대사 능력을 평가하는 데 유용한 근사치를 제공합니다.
따라서 본 자료는
추가 실험적 검증을 위한 후보 균주를 선별하는 데
필수적인 참고 자료 역할을 합니다.
또한, 빠른 세포 성장에 의해 주도되는 공급배치 발효에서 달성된 높은 생산성과 같이
관심 조건을 반영하는 추가 기준과 결합할 경우,
이 자료는 균주 선택 및 추가적인 세포 공장 설계를 안내할 수 있습니다.
우리의 접근법이 완벽한 해결책을 제공하지는 않지만,
고성능 미생물 세포 공장 개발을 향한 진전에 필수적인 역할을 할 것입니다.
Methods
Model construction
GEMs containing metabolic pathways for bio-based chemicals were constructed by introducing metabolic reactions of the biosynthetic pathways into previously reported GEMs of representative industrial hosts: B. subtilis iYO84465, C. glutamicum iCW77366, E. coli iML151567, P. putida iJN146368, and S. cerevisiae Yeast869. To obtain basic information about the heterologous reactions to introduce, the metabolic reactions, which were compiled in a previously published metabolic map34, were obtained from the Rhea database35. The mass- and charge-balances of the reactions were manually curated. For 235 bio-based chemicals in the metabolic map, overall reactions and pathways toward the biosynthesis of the chemical from key metabolites (i.e., 2-oxoglutarate, 3-phospho-d-glycerate, acetyl-CoA, d-erythrose 4-phosphate, d-glucose 6-phosphate, d-glyceraldehyde 3-phosphate, oxaloacetate, and pyruvate) were identified. Based on the identified biosynthetic pathways, GEMs were constructed by introducing required heterologous reactions for each chemical production into the aforementioned five template GEMs. All reactions and metabolites were curated to have consistent BiGG IDs when possible38. To calculate the metabolic capacity for each target chemical production, an exchange reaction, which exports the target chemical from the cytosol to extracellular space, was added in the GEMs. If multiple pathways have been reported for a single target chemical production, a GEM was constructed separately for each biosynthetic pathway. For the construction GEMs with one-carbon assimilation pathways, the following reactions are added if not included in the original GEMs: alcohol dehydrogenase (methanol), hexulose-6-phosphate synthase, and phosphohexulose isomerase for RuMP cycle; formate dehydrogenase (NAD), formate dehydrogenase (NADP), formate-tetrahydrofolate ligase, and glycine cleavage system (bidirectional) for rGly pathway; ribulose-bisphosphate carboxylase, phosphoribulokinase and formate dehydrogenase (NAD) for CBB cycle.
Calculation of maximum theoretical yields and maximum achievable yields
To analyze the metabolic capacity for target chemical production, YT and YA of each bio-based chemical in the five representative industrial host strains (i.e., B. subtilis, C. glutamicum, E. coli, P. putida, and S. cerevisiae) were calculated. Using the default minimal media composition of each GEM constructed in this study, YT and YA were calculated using FBA70. Each of nine carbon substrates (i.e., l-arabinose, d-fructose, d-galactose, d-glucose, d-xylose, glycerol, sucrose, formate, and methanol) was used as a single carbon source in the simulation. To simulate other carbon sources, we scaled the carbon source uptake rate proportionally based on the number of carbons in the carbon source relative to d-glucose, the default carbon source in the GEMs. For anaerobic and microaerobic conditions, the oxygen uptake rates were constrained to be less than or equal to 0 mmol gDCW-1 h-1 and 0.5 mmol gDCW-1 h-1, respectively71. The biomass production equation, NGAM requirement, and reaction directions of Yeast8 were curated to enable simulation under anaerobic and microaerobic conditions69,72. The flux of the target exchange reaction, which extracts the target chemical from the cytosol, was maximized as an objective function using loopless FBA simulation. The exchange reaction flux was divided by the carbon source uptake rate to calculate the yields. Because YT does not account for the requirements of NGAM, the lower bound for NGAM was set to zero. On the other hand, to calculate YA the lower bound for NGAM was set to the default value in each GEM. Additionally, 10% of the maximum biomass production rate was used as the lower bound of biomass production to account for the minimum cell growth of the microbial cell factories when calculating YA. As iCW773 has no requirement for NGAM, 3.2 mmol gDCW-1 h-1 was used as the lower bound for ATP maintenance requirement reactions in C. glutamicum GEMs, according to a previous C. glutamicum GEM73. For bio-based chemicals with multiple biosynthetic pathways, the pathway with the highest yield value was used for comparison of the maximum yields between carbon sources or host strains. Yield calculations were performed for carbon sources that a host strain can metabolize in native pathways. Since B. subtilis subsp. subtilis str. 168 and P. putida KT2440 are aerobic strains, the yields of iYO844 and iJN1463 derivative GEMs were calculated only for aerobic conditions. One-carbon assimilation pathways were also analyzed only for aerobic conditions. Construction of the linear programming problem and solving the problem were performed using cobrapy package74 with Gurobi Optimizer (Gurobi Optimization Inc., Houston TX) in Python environment (Python Software Foundation, Delaware, United States).
Flux-sum analysis and flux variability analysis
Flux distributions were calculated using parsimonious FBA before conducting the flux-sum analysis75. Flux-sum analysis was performed by summing the incoming or outgoing fluxes of each metabolite in a GEM of interest76. Differences between the flux-sum of each metabolite in the flux distributions of the native pathway and the pathway with introduced heterologous reactions were calculated. FVA was performed using loopless solutions77,78.
Universal model construction
The metabolic capacity of a strain is constrained by its overall metabolic reactions. To systematically identify candidate reactions that could improve metabolic capacity for target chemical production, a universal model was constructed. This universal model, derived from the BiGG database38, was curated to contain only mass- and charge-balanced reactions. Reactions in BiGG universal model that involved metabolites without annotation for mass and charge annotations were removed. Annotation data were retrieved from the CarveMe universal model and the five template microorganism GEMs used in this study79. Reactions from the CarveMe universal model and the template microorganism GEMs were added to the universal model if the reactions had the necessary annotation data. Finally, reactions involving metabolites in compartments other than the cytosol, periplasm, and extracellular space were removed, resulting in a universal model comprising 3814 metabolites and 6846 reactions. The universal model was manually curated to correct incorrect reaction directions.
Identification of yield-improving heterologous reactions for target chemical production
Heterologous reactions that improve maximum yields of target chemical production were identified. For a GEM that produces a given target chemical, the following procedures were performed sequentially. To analyze the improvement of YT, NGAM was neglected (i.e., the lower bound of ATPM reaction was set to zero). Among metabolic reactions in the universal model that are not in the target GEM, those occurring in the cytosol were added. The target GEM was then used to solve mixed-integer linear programming (MILP), which identifies metabolic reactions required for improving YT.
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
(13)
(14)
(15)
In the above MILP, the meaning of the variables are as follows: , a flux of a metabolic reaction; , a binary variable whether the corresponding reaction from the universal model should be active or not; , a coefficient of metabolite which participates in metabolic reaction ; subscript ub, upper bound of a metabolic reaction; subscript lb, lower bound of a metabolic reaction; , a minimum threshold which ensures the activity of the candidate metabolic reaction; , a limit of the number of metabolic reactions to be added in the target GEM; , a set of metabolites in the target GEM; , a set of metabolic reactions in the target GEM; , a set of metabolic reactions from the universal model; , the flux of target chemical producing reaction when the YT was achieved. The constraint (1) makes mass balance be validated. Constraints (2) and (3) are for the innate constraints for metabolic reactions such as the thermodynamic, the limit of carbon substrate uptake. Constraints (4-13) identify candidate reactions from the universal model. If a binary variable is on, the corresponding metabolic reaction should be added to the target GEM to improve the YT. Here we used 0.001 as a value for . Constraint (14) limits the number of reactions to be added to the target GEM. We set the number of reactions to be added to three in this study. Constraint (15) ensures that the added metabolic reaction should make the target production flux be improved, which subsequently makes improved YT. To analyze improved YT, was sequentially decreased from 1.5-fold of the at YT to the native at YT in steps of 0.01-fold. Reactions from the universal model that formed infeasible cycles in the target GEMs were manually curated. The simulation was performed until no additional reaction sets were predicted.
Identification of yield-improving cofactor exchange for target chemical production
The identification of cofactor exchange was also performed for all of the constructed chemical-producing models. The default setting for the GEM simulation is the same as that used for identifying heterologous reactions. NADH (or NADPH) participating reactions were duplicated with the cofactor changed to NADPH (or NADH). The target GEM was used to solve a MILP which identifies sets of metabolic reactions that improve the YT when the participating cofactors are exchanged.
(16)
(17)
(18)
(19)
(20)
(21)
(22)
(23)
(24)
In the above MILP, the meaning of the variables are as follow: , a flux of a metabolic reaction; , a binary variable whether the corresponding reaction should be active or not; , a binary variable for the original cofactor participating reaction; , a binary variable for the cofactor exchanged reaction; , a coefficient of metabolite which participates in metabolic reaction ; subscript ub, upper bound of a metabolic reaction; subscript lb, lower bound of a metabolic reaction; , the upper limit for the number of cofactor swapping; , a set of metabolites in the target GEM; , a set of metabolic reactions in the target GEM; , a set of metabolic reactions which NADH or NADPH participate in; , the flux of target chemical producing reaction when the YT was achieved. The constraint (16) makes mass balance be validated. Constraints (17) and (18) are for the innate constraints for metabolic reactions such as the thermodynamic, the limit of carbon substrate uptake. Constraints (19), (20), (21), and (22) restrict only a single reaction to being active between the original metabolic reaction and the cofactor exchanged reaction. If a binary variable is on, the cofactor of the corresponding metabolic reaction should be exchanged to improve the theoretical yield. Constraint (23) ensures that the cofactor exchange reaction should make the target production rate be improved, which subsequently makes improved YT. Constraint (24) limits the number of cofactor exchanges. We limited the number of reactions to be exchanged to one in this study.
In silico simulation for identification of up- and down-regulation targets
To achieve a high production yield from a bioprocess, the flux toward the target chemical production should be sufficiently high. To enhance the flux of target chemical producing reaction, iBridge analysis was performed for all constructed GEMs57. iBridge analysis was performed to predict up- and down-regulation target reactions to enhance the production flux of a target chemical. First, parsimonious FBA was used to calculate a reference metabolic flux distribution. Based on the reference flux distribution, ten metabolic flux distributions were calculated with linear minimization of metabolic adjustment, varying the target chemical production flux from zero to its maximum. The covariance between each intracellular reaction flux and the target chemical production flux was calculated from these ten metabolic flux distributions. Metabolites were annotated as positive or negative based on whether the sum of covariances (SoCs) of their outgoing reactions was positive or negative, respectively. Bridge reactions that convert negative metabolites to positive metabolites were identified, and their scores were calculated by subtracting the SoC of the negative metabolites from that of the positive metabolites participating in the reaction, followed by min-max normalization. The reactions with scores equal to or greater than 0.5 are regarded as up-regulation targets, while reactions with scores equal to or smaller than −0.5 are regarded as down-regulation targets57. FVSEOF analysis was also performed for all constructed GEMs under aerobic conditions56. FVSEOF analyzes variations in fluxes in response to the enforced target chemical production rate. For FVSEOF analysis, ten metabolic flux distributions were calculated varying the target chemical production flux from zero to its maximum. From each metabolic flux distribution, FVA was performed for each metabolic reaction. To select reactions that should be up-regulated, the Pearson correlation between the enforced target production rate and the minimal flux of an analyzed reaction was used as a criterion. If the Pearson correlation was positive, the reaction was selected for up-regulation. Metabolic reactions that have corresponding gene information were analyzed using iBridge and FVSEOF.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
Source data are provided with this paper and also available from Figshare: https://doi.org/10.6084/m9.figshare.27874275 (ref. 80). Source data are provided with this paper.
Code availability
The computational pipeline for constructing the resources is available at https://github.com/kaistsystemsbiology/MEResource (ref. 81). The source code for FVSEOF simulation is available at https://github.com/kaistsystemsbiology/FVSEOF (ref. 82).
References
|
|