Fig 1.
Engineered pathways used in this study and their co-factor requirements.
Eight pathways that produce butanol (dark blue circle) and butanol precursors (light blue circles) were selected and introduced into the Escherichia coli Core Model to yield the stoichiometric models used in this study. These pathways are based on variations of the so called ‘Core Pathway’ (module A, grey), which is redox dependent and ATP neutral. By combining these modules, 8 unique pathways with varying demands for ATP and redox are possible. (B) Co-factor requirements of all pathways introduced into the E.coli Core model to simulate butanol and butanol precursor production, and the aerobic (black) and anaerobic (red) carbon yields are shown as a percentage of glucose carbon influx after target production maximization. Co-factor requirements are calculated as the sum of stoichiometric coefficients in all reactions starting from acetyl-CoA through to the final target molecule. Negative ATP/NAD(P)H coefficients represent co-factor demand, which refers to the consumption of a particular co-factor by the introduced pathway, indicating ATP/NAD(P)H going into the reaction. Co-factor surplus, alternatively, is used to describe any co-factor being produced or released by a pathway. NAD(P)H surplus is indicated as positive NAD(P)H released by the pathway (subsequently from NAD(P) going into the reaction). CP–Core Pathway; ACP–acyl carrier protein; AtoB–acetyl-CoA acetyltransferase; AdhE2 –aldehyde alcohol dehydrogenase; NphT7 –acetoacetyl-CoA synthase; TPC–acyl-ACP thioesterase; CAR–carboxylic acid reductase.
Table 1.
Summary of key features of the modified E.coli models used in this study.
We have indicated model names, alongside their introduced reactions, target chemical, corresponding objective function (as per reaction ID), total number of model reactions and metabolites, and also the ATP and NAD(P)H pathway coefficients, calculated as the sum of reaction stoichiometry coefficients of all introduced reactions from acetyl-CoA to the final target. CP—Core Pathway.
Fig 2.
Toy illustration of ATP and NAD(P)H reactions and reaction categories accounted for by the CBA protocol.
(A) All reactions in the E.coli Core Model that directly contribute to the intracellular levels of ATP and NAD(P)H pools (blue or yellow circle, accordingly). Arrows pointing inwards on the left display reactions leading to ATP or NAD(P)H build-up (i.e. co-factor production), while arrows pointing outwards on the right show reactions that drain the co-factor pools (i.e. co-factor consumption). The thickness of the arrows represent the varying fluxes of these reactions. The CBA protocol identifies all co-factor related reactions producing and/or consuming ATP or NAD(P)H, it records their fluxes and distributes them across five core categories: (1) co-factor production, (2) biomass production, (3) waste release, (4) cellular maintenance and (4) target production (this category is target product specific). (B) Theoretical example of how the classification of ATP reactions is handled by the CBA protocol. Co-factor fluxes (here illustrated by the varying arrow thickness) are dependent on the co-factor stoichiometric coefficient and flux calculated by FBA. ATP production accounts for all reactions that generate a positive ATP flux. The ATP waste category accounts for both ATP produced during acetate production, but also ATP consumed in ATP-hydrolysing reactions (also known as ‘ATP burning’ reactions), such as ATPM and ADK1. ATP biomass includes the ATP flux consumed during biomass formation. The ATP target category is pathway-specific, accounts for only those synthetic reactions introduced into the stoichiometric model, and will lead to a positive or negative flux according to whether the synthetic pathway leads to the formation or drain of intracellular ATP, respectively. If the synthetic pathway is ATP-neutral, the net value for this category will be zero. ATP maintenance includes any ATP consumed in additional metabolic activities and not considered in the aforementioned categories. (C) Theoretical example of how the classification of redox reactions is handled by the CBA protocol, similarly to (B). The NAD(P)H waste category also accounts for reactions GND, PDH, AKGDH, ICDHyr, which produce NAD(P)H but simultaneously release CO2, and reactions such as LDH_D and ADHEr that consume NAD(P)H and release fermentation products. For categories including both positive and negative co-factor fluxes, the net is calculated for that category. Figure design inspired by [38].
Fig 3.
CBA-derived network co-factor usage profiles.
After FBA optimization, the COBRA-based CBA protocol classifies ATP and NAD(P)H-related reactions according to whether these co-factors were consumed or produced during biomass, waste, target production or cellular maintenance. All models were initially unconstrained and simulated under both aerobic and anaerobic conditions. (B) ATP and NAD(P)H profiles under aerobic conditions; (C) ATP and NAD(P)H profiles under anaerobic conditions.
Fig 4.
Identification and removal of futile cycles.
(A) examples of ATP futile cycles identified in this study–pairs of cycling reactions in which ATP is consumed through one reaction and the original metabolites are recycled through the pair reaction. (B) ATP-burning and high-flux futile cycles were identified by directly comparing the engineered strain and the wild-type flux distributions. (C) The identified ATP-burning reaction or futile cycle was constrained by limiting the upper bound to the maximal flux observed for the equivalent reaction in the wild type. (D) After optimization, the flux distributions of the wild type and engineered system were compared and the next high-flux futile cycle would be detected and constrained accordingly (as per C). Steps (C) and (D) were repeated until no more futile cycles were detected.
Fig 5.
CBA-derived co-factor usage profiles after manual curation of the models and comparison between CBA-derived estimates and estimates calculated using the method developed by Dugar et al.
The engineered models were manually constrained to minimize high-flux ATP futile cycles as described in the text, and led to the above (A) curated ATP and NAD(P)H CBA profiles under aerobic conditions and (B) ATP and NAD(P)H CBA profiles under anaerobic conditions. (C) Comparison of bioproduction carbon yields determined using the calculations developed by Dugar et al and FBA. uDugar–unadjusted Dugar-derived estimates; uFBA–unconstrained FBA (preliminary FBA results prior to applying any constraints); aDugar–Dugar-derived estimates after adjusting values according to ATP and NAD(P)H imbalances; cFBA–curated FBA yield estimates, obtained after manually constraining the flux distributions.
Table 2.
Maximum yield (YE and YEa), pathway yield (YP), adjusted pathway yields (YP,G and YP,G,X) and pathway efficiency (η) of all butanol and butanol precursor pathways calculated using the stoichiometric and energetic calculations proposed by Dugar et al. [5].
Maximum yield (YE) is the maximum amount of product that can be produced from the substrate. YEa indicates the maximum yield in mol product/mol substrate. Pathway yield, YP, is pathway-specific and calculated from pathway stoichiometry. YP,G is the adjusted pathway yield once any excess redox is depleted using an electron sink (i.e. glycerol). YP,G,X is the adjusted pathway yield after any excess ATP is diverted towards biomass formation. η is the ratio between YP,G,X and YE.
Fig 6.
MOMA compared to MFA-derived estimates, carbon yield efficiencies and CBA co-factor profile comparison across unconstrained, manually curated and experimentally constrained solutions.
(A) Flux ranges calculated with MOMA (green) and Metabolic Flux Analysis (orange stripes). MOMA ranges were estimated using the wild type solution as a reference and sequentially implementing the single-gene knockouts studied by Long et al. (2019) [46], with biomass formation as the objective function. MFA ranges were extracted from a pre-existing dataset (Long et al., 2019), using a Python algorithm to select the minimal and maximal flux ranges.(B) Carbon yields of butanol and butanol precursor models, compared across all approaches evaluated in this study: unconstrained pFBA (labelled ‘FBA’); manually curated pFBA solutions with minimized high-flux futile cycling (labelled ‘cFBA’); experimentally-constrained solutions using MFA-derived flux data (labelled ‘mFBA’); experimentally-constrained solutions using MFA-derived flux data with further capping in co-factor cycling reactions (labelled ‘cmCBA’) (C) ATP (blue) and NAD(P)H (yellow) CBA-derived cofactor usage profiles compared across all approaches evaluated in this study (labels identical as previously).
Fig 7.
butanol carbon yield (%) and biomass production rates (mmol gDW-1hr-1) of engineered E.coli strains in response to changes in ATP and NADH demands.
Each model represents a unique pathway variant for butanol production, which has been manually curated and optimized for the selected objective under aerobic conditions. (A) BuOH-0, comprised of route AtoB + AdhE2; (B) BuOH-1, including reactions NphT7 + AdhE2; (C) tpcBuOH, made up of AtoB + TPC7; (D) fasBuOH, comprising reactions NphT7 + TPC7.
Fig 8.
Components of Cofactor Balance Assessment (CBA) pipeline and summarised workflow.
A stoichiometric model, flux distribution and a list of target reactions are required to call the CBA function in the python environment. Stoichiometric models contain reaction information, such as whether they consume or produce ATP and NAD(P). We used the E.coli Core stoichiometric model and the COBRApy package, and selected reactions were implemented to build the path to novel products. CBA classifies reactions in the model according to whether they are involved in the consumption or production of NAD(P)/ATP, assigns them a cofactor balance score, and groups them into categories as represented above. Finally, the total balance per category is calculated the total sum of flux and adjusted to provide a final value for each category. The result is a profile displaying the fraction of the total cofactor produced involved in maintenance, biomass, target and waste production.