Recent years have seen several advances in the usage of bioprocessing to produce a wide range of chemicals (Erickson et al., 2012). Microorganisms can produce diverse and complex products from simple carbon sources. Nevertheless, there are many challenges in designing economically feasible bioprocesses. The advancements in synthetic biology have enabled the metabolic engineering of organisms to improve yield and productivity (Yadav et al., 2019). Various computational strain design algorithms have been developed to identify the genetic manipulations required to over-produce a single product (Burgard et al., 2003; Rocha et al., 2008; Yang et al., 2011). Despite the increase in yield achieved through such rational strain design, the bioprocesses are unable to compete with the traditional chemical processes in many cases (Cai and Bennett, 2011). This is due to two main reasons: 1) the cost of raw materials and 2) the maximum yield achievable for a given product in a given organism and environment is limited by the number of genetic manipulations that can be successfully implemented in a single strain (Silva et al., 2012). The former issue can be reduced by using agricultural waste as feedstock instead of a synthetic nutrient medium. The latter can be overcome by co-producing multiple products in the same bioprocess (da Silva et al., 2014).

Co-production equips us to exploit the system in a better fashion and produce more valuable products from the same raw materials. A high-value, low-volume chemical can be co-produced with a low-value, high-volume product in order to increase the economic feasibility, as in the case of riboflavin and butanol, respectively (Cai and Bennett, 2011). Co-production is essential when a cocktail of metabolites need to be produced together, rather than a single metabolite, as in the case of biofuels and fatty acids (Xin et al., 2018). A mixture of different alcohols or fatty acids of varying chain length need to be co-optimized in such cases. It can also balance carbon metabolism, as in the case of uridine, and acetoin (Fan et al., 2018). High carbon inflow towards uridine causes excess production of acetate, which hampers the growth of the organism. Conversion of acetate to acetoin prevents over-acidification of the nutrient medium and thereby improves growth and uridine production. There are many studies that have successfully achieved co-production of a variety of products with/without genetic manipulation of the organisms. Polyhydroxyalkanoates are a common class of metabolites that are co-produced with other metabolites (Li et al., 2017; Kumar and Kim, 2018; Yadav et al., 2021). Butanol and hydrogen have been co-produced in Clostridium beijerinckii (Zhang et al., 2021), and ethanol and xylitol have been co-produced in Candida tropicalis (Raj and Krishnan, 2020; de Souza Queiroz et al., 2021). The carbon source, nutrient medium, pH etc., are optimized in such cases to improve the yield of metabolites. Metabolic engineering can further expand the number of products that are co-produced and also improve their yield significantly. Multiple metabolites like ethanol, isopropanol, butanol and 2,3- butanediol have been co-produced by optimizing the acetone-butanol-ethanol (ABE) fermentation pathway in Clostridium acetobutylicum (Collas et al., 2012). Nisin and 3-phenyllactic acid, two antimicrobial agents, have been co-produced in Lactococcus lactis through genetic manipulation (Julien-Laferrière et al., 2016). Non-native metabolites can also be co-produced with other metabolites, as in the case of butanol and riboflavin, by engineering the heterologous pathway in C. acetobutylicum (Cai and Bennett, 2011).

Although many strain design algorithms have been successfully employed for metabolically engineering organisms to optimize a single product (Pharkya et al., 2003; Kumelj et al., 2019), few studies have applied it for co-production. The studies listed above only use existing literature and readily apparent deletion targets to achieve co-production. This limits the robustness of the bioprocesses that are designed. There is a lack of algorithms that can be easily applied to study co-production. In this study, we have extended the Flux Scanning based on Enforced Objective Flux (FSEOF) (Choi et al., 2010) algorithm to study co-production. Further, while deletion targets can be obtained for metabolites independently using existing algorithms like OptKnock (Burgard et al., 2003), OptGene (Rocha et al., 2008), there are very few algorithms that can identify amplification targets (Ranganathan et al., 2010). In order to identify amplification targets in addition to knock-out targets for co-optimizing a set of metabolites, we propose a new methodology, co-FSEOF, adapting the FSEOF algorithm. Co-FSEOF has a simple computational framework that can be easily modified, and it also provides the entire set of potential intervention strategies in a single run while many algorithms are sequential, returning one intervention target per run. The utility of the potential intervention strategies obtained was further assessed using Flux Variability Analysis (FVA). We applied co-FSEOF to evaluate all possible pairs of secretory metabolites in Escherichia coli and Saccharomyces cerevisiae. The different pairs of metabolites that can be co-produced through a single reaction deletion or amplification were obtained. This analysis helps us choose favorable pairs of metabolites for which higher-order intervention strategies can be obtained. We have demonstrated this by identifying the amplification targets, knock-out targets and mixed intervention strategies of size up to three to co-optimize the production of isobutanol and succinic acid in S. cerevisiae. Higher-order intervention strategies were able to achieve better yield with very little reduction in growth rate. Overall, our analyses provide an overall picture of the biosynthetic capabilities of an organism, particularly highlighting key interdependencies in metabolism.

Metabolic engineering strategies for the co-production of all pairs of secreted metabolites in E. coli and S. cerevisiae were obtained using co-FSEOF as described in Section 2.4. We identified the intervention strategies required to optimize the co-production of metabolites in both aerobic and anaerobic conditions. Anaerobic conditions favor the co-production of more pairs of metabolites when compared to aerobic conditions. The intervention strategy for each pair of metabolites is scored as in Section 2.4. The best intervention strategy can be chosen using the o v e r a l l   s c o r e . In cases where one metabolite might be favored over the others, the individual scores, S c o r e i can be used to choose the best intervention strategies. Some of the intervention strategies obtained have been successfully verified through experimental studies in literature. This shows the credibility of the intervention strategies obtained. We discuss a few of the industrially significant metabolites and their intervention strategies, along with supporting literature. We also propose many other intervention strategies, which form a ready short-list for experimental validation. We were able to identify other hitherto unexplored intervention strategies, which may be better alternatives to those in existing literature, further demonstrating the utility of the algorithm.

Chemical processes based on fossil fuels are cheaper when compared to bioprocesses, which leads to reluctance in the adoption of sustainable bioprocesses in industries. To improve the economic feasibility of a bioprocess, we can optimize the process variables and/or genetically engineer the microbes (Dzurendova et al., 2020). Even then, in some cases, the bioprocess might be less lucrative when compared to their chemical counterparts (Zhang et al., 2021). In such cases, we can co-produce multiple metabolites to improve the economic feasibility and efficiency of a bioprocess. For example, in the case of biofuels and fatty acids, we need to design processes that can support the production of multiple metabolites of similar nature (Xin et al., 2018). Co-production also ensures better utilization of microbial capabilities, and better balance in the carbon metabolism (Fan et al., 2018). While there are multiple computational tools and algorithms to identify intervention strategies for a single product, there is a lack of readily appliable algorithms for co-production. As a result, almost no co-production study in existing literature was found to use computational algorithms to aid rational strain design. All of the studies rely on previous findings or readily apparent strategies to achieve co-production. This limits the intervention strategies designed.

In this study, we present co-FSEOF, by adapting the effective FSEOF algorithm to study the co-optimization of a set of metabolites. FSEOF is a well-established constraint-based modelling algorithm, which has been used to reliably predict metabolic engineering strategies for a variety of systems (Choi et al., 2010; Boghigian et al., 2012; Badri et al., 2019; Srinivasan et al., 2019). It has a simple and efficient framework and can identify both deletion and amplification targets. Flux Coupling Analysis (FCA) is yet another interesting algorithm that can identify which metabolites can be coupled together. But very few metabolites are innately coupled without interventions. Moreover, it excludes all the reactions that do not carry flux under a given set of conditions from the analysis. This affects the applicability of FCA because these reactions, though not coupled, can have an effect on target production in the presence of other interventions. co-FSEOF is able to identify more combinations of products that can be co-produced and also provides a wider range of intervention strategies for a given set of metabolites.

Using co-FSEOF, we examined the co-production of multiple pairs of metabolites, and both deletion and amplification targets were obtained in E. coli and S. cerevisiae under both aerobic and anaerobic conditions. Anaerobic fermentation enabled the co-production of a higher number of metabolites when compared to aerobic fermentation in both organisms. This could be due to the incomplete respiration in the absence of oxygen that leads to the formation of multiple by-products. Also, S. cerevisiae produces more industrially significant metabolites when compared to E. coli. Some of these proposed intervention strategies have been verified experimentally by other studies in literature, as mentioned in Section 3. This shows the efficacy of the algorithm in furnishing reliable targets. In addition to readily apparent intervention strategies, co-FSEOF also provides non-intuitive intervention strategies that are present in auxiliary biochemical pathways (as discussed in Section 3.2.2, 3.2.3).

The co-optimization analysis for all possible pairs of metabolites in the network is intended to be exploratory in order to give a larger picture of the metabolic capabilities of the organism. This analysis showed that around 200 pairs could be co-optimized in E. coli under aerobic conditions, and around 1,000 pairs of metabolites could be co-optimized in the other cases. An important class of metabolites observed in the analysis are alcohols such as ethanol, isobutanol, and 2,3-butanediol, which can be co-produced using various interventions. Co-optimization can thus enable the efficient production of biofuels (Collas et al., 2012). We can explore the metabolic capabilities of the organism to identify all possible pairs of metabolites that can be co-produced. Following this, important and commercially valuable pairs of metabolites can be further studied to obtain higher-order intervention strategies, as shown in Section 3.2.3. Exploring the higher-order strategies can expand the efficiency of the intervention targets obtained. Since the evaluation of higher-order intervention strategies is laborious and computationally expensive, we have limited the size to a maximum of three manipulations at a time. It not only enhances yield, but also provides alternate routes to achieve a similar yield. The advantageous strategies can be chosen based on the ease of manipulation in an experimental setup in such cases.

The evaluation of the results is carried out using FVA, which ensures the robustness of the targets obtained. While FBA provides one optimal solution from the solution space, FVA gives the entire range of values the flux can take up. This is a significant difference that sets co-FSEOF apart from other existing algorithms like OptKnock (Burgard et al., 2003) and OptReg (Pharkya and Maranas, 2006). Also, the algorithm validates and returns all the intervention strategies in a single run, contrary to the existing algorithms, most of which are sequential and require a separate run for each strategy obtained. The set of intervention strategies validated through FVA can be short-listed for experimental verification using the scores. The o v e r a l l   s c o r e   can be used to compare the effectiveness of different intervention strategies. The score is designed to incorporate both increase in product flux and decrease in biomass flux, so that both biomass and product production are favored in the mutant. Moreover, co-FSEOF uses biomass as the objective throughout the analysis, and the effect of the intervention strategies on product synthesis is studied when the organism optimizes growth. Thereby, the intervention strategies result in the co-optimization of product production and biomass formation. An intervention strategy with better o v e r a l l   s c o r e ensures better product synthesis along with good biomass formation. If one product is more favored than the others economically or otherwise, we can use the individual scores S c o r e i to choose the appropriate strategy for the process that is formulated. The products can be chosen based on their economic value, or ease of co-production. One drawback of co-production is the cost associated with downstream processing. But this can be overcome by choosing easily separable products or choosing metabolites such that one is accumulated in the cell and one is secreted out, as in the case of polyhydroxy butyrate and succinate, respectively (Kang et al., 2010). However, this problem does not occur in the case of biofuels where the alcohol mixture is optimized for and therefore does not require extensive separation of the products. Co-FSEOF not only identifies intervention strategies for co-production of a given set of metabolites, but also allows us to explore the different combinations of products that can be co-produced in an easy and efficient manner.

Co-production can open new avenues for the sustainable production of chemicals. Designing bioprocesses for co-production using laboratory experiments alone is cumbersome and can result in sub-optimal strategies. co-FSEOF empowers us to explore and exploit microbial systems in a better fashion. It can be used to computationally study and optimize co-production by identifying intervention strategies for multiple metabolites and thereby improve the efficiency of bioprocesses. It should be noted that the co-optimization analysis was limited to pairs of metabolites to reduce the computational time. Nevertheless, co-FSEOF can be easily extended to co-optimize a more extensive set of metabolites. To conclude, this study can be used to identify various genetic manipulations that can co-optimize a set of products, which might be challenging to achieve through pure experimentation. It provides a novel and critical approach to study co-production computationally. We hope this study will aid the design and development of more sustainable bioprocesses.