Bioprocessing offers a sustainable and green approach to the production of chemicals. However, a bottleneck in introducing bioprocesses is cell factory development, which is costly and time-consuming. A systems biology approach can expedite cell factory design by using genome-wide analyses alongside mathematical modeling to characterize and predict cellular physiology. This approach can drive cycles of design, build, test, and learn implemented by metabolic engineers to optimize the cell factory performance.<\/p>\n
Streamlining of the design phase requires a clearer understanding of metabolism and its regulation, which can be achieved using quantitative and integrated omic characterization, alongside more advanced analytical methods. We discuss here the current impact of systems biology and challenges of closing the gap between bioprocessing and more traditional methods of chemical production.<\/p>\n
Trends<\/p>\n
Bioprocessing offers a green and sustainable alternative to fossil fuel-based chemical production. However, cell factory development is currently too inefficient to allow bioprocessing to become cost-competitive.<\/p>\n
Systems biology can streamline factory design by combining genome-wide datasets with computational models to characterize metabolism.<\/p>\n
To increase the contribution of systems biology, large-scale omic data analysis and interpretation need to occur faster and more accurately, and the predictive strength of in silico metabolic models needs to be improved.<\/p>\n
This can be achieved by (i) developing new computational methods for data analysis, such as machine learning, and (ii) integrating into models regulatory structures that coordinate metabolism.<\/p>\n
By increasing the predictive strength of models, the cost and time for cell factory development can be reduced.<\/p>\n
The Growing Demand for Green and Sustainable Chemical Production<\/p>\n
As the world population grows and non-renewable energy sources decline, there is an increasing demand to produce chemicals in a sustainable and stable manner. Bioprocessing offers an attractive alternative to traditional chemical production by using microbial cells as a factory for chemical synthesis (Table 1). These typically originate from cell factories used for the production of classical products, but some have also served as model microorganisms used in basic research, whereby much of their metabolism has been characterized and understood. They include the baker\u2019s yeast Saccharomyces cerevisiae, the Gram-negative bacterium Escherichia coli, the Gram-positive bacteria Corynebacterium glutamicum and Bacillus subtilis, and the filamentous fungus Aspergillus niger. To transform these microbes into cell factories, metabolic engineers rewire microbial metabolism to reroute metabolic flux towards chemical production, a process which may include installation of heterologous genes and deliberate deregulation of native pathways. If successful, this engineering would enable chemical production to occur efficiently at large scale (Figure 1A) and ideally at low cost, with such a cell factory being referred to as having high titer, rate, and yield (TRY).<\/p>\n
As well as being sustainable, bioprocessing is also considered to be a green approach owing to its environmentally favorable properties. For example, renewable feedstock can be leveraged as an initial energy source (Figure 1B), as opposed to burning fossil fuels. Furthermore, chemicals produced using bio-based methods can possess \u2018cleaner’ attributes in comparison to their fossil-based equivalents, helping to reduce greenhouse gas emission and concomitant effects on global warming [1]. However, despite its environmental benefits, bioprocessing cannot currently compete with the fossil-based chemical industry because of the time and money required for cell factory development (Figure 2). This process typically takes more than 5 years, and costs in excess of 50 million $US. For example, to enable opioid synthesis in yeast, strain construction took 10 years, and to scale up production of the antimalarial drug artemisinin, it took 5 years of development and cost >50 million $US [2, 3].<\/p>\n
One reason that strain production was previously so cumbersome was because of the metabolic engineering (ME) (see Glossary) [4] approach used. This process was usually untargeted and involved multiple rounds of random mutagenesis and screening, resulting in only a subset of the genome of a microorganism being analyzed. As a result, any deleterious side effects that arose could not easily be interpreted and were challenging to circumvent. Since then metabolic engineers have implemented a more structured approach to strain production. This involves performing iterative cycles of design, build, test, and learn (DBTL) that integrate a systems biology [5, 6, 7, 8] approach (Figure 3). This review aims to summaries the most recent developments in systems biology within a bioprocessing context. Because this topic and those of a similar nature have been explored in the past, we direct the reader to such reviews for additional information [8, 9, 10, 11, 12, 13, 14].<\/p>\n
Systems Biology and Bioprocessing<\/p>\n
Systems biology uses a data-driven approach to acquire a comprehensive and holistic understanding of cell metabolism. It allows cell physiology to be fully characterized by measuring different components of a cell including mRNA, protein, and metabolites (Box 1). Computational models can then be used as a framework for integrating these genome-wide datasets, with the aim of representing cellular metabolism entirely in silico. The potential of these models is considerable because they allow a large number of combinatorial gene changes to be tested without experimentalists testing each option in the laboratory; furthermore, they offer predictions for what the best metabolic route or pathway may be. However, for these models to produce accurate predictions, quantitative and comprehensive datasets from different omic layers must first be acquired.<\/p>\n
Omics Integration<\/p>\n
Revolutionary developments in sequencing instruments have helped to acquire genome and transcriptome data, and, increasingly, quantitative mass spectrometry methods have enabled analysis of the proteome and metabolome. Metabolic flux, the final output of metabolism and its regulation, can also be traced using fluxomics [15]. This method uses 13C-labeled glucose to trace where carbon propagates through the metabolic network, enabling a sensitive \u2018fingerprint’ of flux distribution [16]. Advances in technical and analytical methods have also enabled increasingly less time to be invested in omic measurements and analysis [17].<\/p>\n
Metabolic engineers can also reduce time by (i) recruiting tools from synthetic biology [18, 19] and (ii) coupling systems biology with evolutionary engineering [20, 21]. Since its emergence, CRISPR\/Cas technology has enabled candidate strains to be built in an unprecedentedly short timescale, leading to exponential advances in the initial stages of strain development [22, 23]. The emergence of in vivo biosensors has also shown promise for expediting the DBTL cycle by helping to rebalance the build and test phases of strain development [19]. Via high-throughput fast screening and selection, biosensors increase efficiency by enriching for only a subset of strains with high performance. Therefore, fewer strains containing a greater proportion of optimal candidates proceed to downstream in-depth physiological characterization [24]. If, after iterative attempts at gene editing, cells show no further improvement in TRY, evolutionary engineering can also be applied. This approach exploits the natural ability of cells to evolve genetic mutations, particularly when they are grown under high selection pressure. When chemical production is coupled to growth, adaptive laboratory evolution (ALE) can produce evolved strains whose causal mutations can be identified by sequencing and then reverse engineered into the parental strain. When eukaryotes are used, chemical production can also be improved via spatial engineering, which aims to redirect metabolic flux towards subcellular compartments where conditions for chemical synthesis may be more favorable [25].<\/p>\n
When impediments in cell metabolism still lead to less than optimal TRY, the open-ended nature of systems biology analyses enable the identification of alternative microorganisms, which have a natural ability to produce compounds of value. For example, a recent analysis of the genomic diversity of Penicillium species revealed that these fungi are an untapped source for new antibiotics and pharmaceuticals [26]. A systems biology approach was also applied to the oleaginous yeast, Yarrowia lipolytica, which accumulates most of its biomass as lipids, making it a promising cell factory for advanced biofuel production. By combining transcriptomics and metabolomics, lipid metabolism was found to have limited transcriptional regulation, suggesting that metabolic engineering efforts could be more effective if post-translational regulation was targeted instead [27].<\/p>\n
Systems biology may also optimize the bioprocessing approach by helping to increase the substrate range for a given cell factory. This can reduce the cost of microbial fermentation by utilizing low-cost and abundant sources of carbon for feedstock. For example, next-generation bioprocessing proposes the use of non-food feedstock to mitigate resource competition between food and fuel. Lignocellulose and its degradation constituents are one such second-generation feedstock that is abundant, inexpensive, and renewable. A systems biology approach can help utilize this biomass by designing yeast strains that can metabolize hexose and pentose sugars found in lignocellulose hydrolysates. Host strains can also be adapted to resist toxic compounds arising from lignocellulose pretreatment and recalcitrant lignin removal, which can inhibit growth and productivity [28]. For example, bacterial fermentation of switchgrass, one of the dominant grasses in North America, was analyzed using multi-omics and revealed that cells respond to lignocellulose pretreatment by restructuring their cell membrane and shifting metabolic flux towards the pentose phosphate pathway over time [29].<\/p>\n
Marine macroalgae such as seaweed are another abundant source of sugar, and have been proposed as a third-generation feedstock. This type of biomass, unlike lignocellulose, does not compete for arable land with edible crops, and has no or little lignin, reducing the costs and complexity of biomass pretreatment [30]. However, despite its promise, pretreatment and saccharification methods to access sugar monomers have yet to be established. Using transcriptomics and metabolomics, however, recent work on red macroalgae led to the discovery of two key genes and intermediate metabolites involved in carbohydrate metabolism, helping progress towards future bioconversion of algal biomass into biofuel and chemicals [31].<\/p>\n
Another way in which systems biology can enhance the function of cell factories is by generating platform strains, also referred to as metabolic chassis [32]. These strains increase metabolic flux towards a biosynthetic precursor which can then be used as a building block for a wide array of chemicals. For example, much effort has been made in targeting central carbon metabolism given the high flux that occurs towards the precursor acetyl-coenzyme A (acetyl-CoA). This metabolite is used in 34 compartmentalized reactions and can be used as a chemical building block for a variety of commercially profitable compounds such as lipids (biodiesel, pharmaceuticals), polyketides (antibiotics and anticancer drugs) and isoprenoids (perfumes and food ingredients) [4, 33]. However, central carbon metabolism has evolved to be extensively regulated both transcriptionally and post-transcriptionally [34, 35]. Therefore, to redirect flux towards these central pathways, metabolic regulation needs to be comprehensively understood [36].<\/p>\n
Modeling Cell Behavior<\/p>\n
To interpret how metabolism is regulated, large omic datasets need to be integrated and interpreted easily. The use of genome-scale models (GEMs) can provide the necessary scaffold for such data integration [37]. GEMs create a mathematical framework to assemble and sort omic data. Subsequently, these data can be subjected to mathematical analysis and computational predictions. In addition, the open-ended nature of GEMs enables continual refinement in their predictive power by recurrently adding constraints from new experimental data. GEMs can be used to perform flux balance analysis (FBA), which uses stoichiometric coefficients of each metabolic reaction and an objective function, such as biomass, to predict cellular growth or production rate for a compound of interest [38]. By integrating disparate data types across functional levels, GEMs therefore help to discover novel correlations and non-intuitive regulatory features in metabolism [39].<\/p>\n
The aim of in silico predictions is to find an optimal solution space for cell factory designs that produce optimal chemical TRY. Predictive power, however, is currently handicapped by an incomplete understanding of the metabolic network. To improve estimates of an optimal ME strategy, both the physical limitations of metabolic reactions, such as enzyme Kcat and Km values and the regulatory features that govern metabolic output, such as feedback and feedforward loops, need to acknowledged [40]. Defining these parameters and incorporating them into constraint-based modeling with GEMs can subsequently improve quantitative analyses for how cells respond to changes in either genetic or environmental conditions [41, 42]. By adding additional constraints such as enzyme dependence on cofactors, as well as pH, thermodynamics, energy coupling, and enzyme promiscuity, it would then be possible to generate more accurate simulations [43].<\/p>\n
Genome-wide data in particular for metabolites is lacking [44]. This immaturity in data analysis is predominantly due to (i) the chemical nature and complexity of metabolites, (ii) their ability to change in abundance over sub-second timescales, as a result of their interactivity with proteins and fast turnover, and (iii) their concentrations varying by several orders of magnitude. As a result, it is inherently challenging to quantify metabolite levels accurately and reproducibly [45]. Despite these setbacks, new insight into gene\u2013metabolome associations have been discovered in unprecedented detail via systematic metabolomic profiling of single-gene knockouts for S. cerevisiae and E. coli [46, 47]. This integration of genome-wide associations with metabolomics can then be used to map which genes are linked directly and indirectly with which metabolites.<\/p>\n
Alongside reconciling biological networks, efforts are being made to integrate spatial limitation into models to understand how a cell may trade between space availability and enzyme allocation. Because many ME strategies overexpress membrane enzymes, new predictions on membrane spatial limitation could benefit these strategies greatly [48]. For example, flux balance analysis with membrane economics (FBAME) has already been developed to examine membrane composition of bacterial cells [49]. This method is based on the theory that transmembrane proteins, such as substrate transporters and metabolic enzymes, \u200b compete for membrane space. For example, although glycolytic enzymes have lower efficiency than respiratory enzymes, their higher turnover and ability to occupy relatively less space may enable more room for substrate transporters and therefore could promote higher catabolic activity in nutrient-rich environments. Based on relative membrane costs and enzyme efficiency, FBAME can therefore predict more accurately gene expression and cell function [49]. Using protein-constrained FBA to simulate yeast metabolism, it was also shown that cells may trade off long energy efficient pathways, such as the electron transport chain, with less energy-efficient but smaller enzymes used in fermentation, to maximize growth using a finite protein pool [50].<\/p>\n
Predictions could also be improved by incorporating protein structure and thermodynamic properties, which may reveal structural effects on enzyme promiscuity, rate of catalysis, and allosteric regulation [42, 51]. For example, GEMs that include structural protein data (GEM-PROs) can use protein melting temperature to determine how temperature may affect the growth rate of a given organism [52]. Because many cell factories are not thermotolerant, insight into protein stability could improve the development of these biocatalysts by pinpointing the most limiting metabolic processes. For example, using the GEM-PRO platform in E. coli, it was possible to determine that the cofactor metabolic processes, in particular involving CoA and biotin, were most growth-limiting under heat stress, therefore their exogenous supplementation could potentially improve thermotolerance [53].<\/p>\n
As computational models are developed further, the in silico role of systems biology will increasingly contribute to the cell factory design process. However, despite expanded models now containing thousands of reactions and metabolites, metabolic flux analysis can only be accurately determined for well-known pathways such as central carbon metabolism where intracellular fluxes are best understood. Similarly, precise kinetic modeling, which usually requires absolute metabolite concentrations, can only be used on well-defined subcomponents of the entire system [54]. Because these models do not yet integrate the thousands of metabolites present within a cell, it remains a challenge to predict metabolite and reaction interactivity on a global scale. Before cell metabolism can be dynamically modeled and accurately predicted, it is therefore necessary to bridge the gap between model scale and model precision [44].<\/p>\n
Systems Biology and Recombinant Protein Production<\/p>\n
Biopharmaceutical proteins such as antibodies, vaccines, blood factors, and hormones (e.g., insulin) currently make up 25% of total pharmaceuticals and \u223c40% of sales, and, after mammalian cells, E. coli and S. cerevisiae are the predominant host organisms for their production [55]. One way in which systems biology could guide production efforts is through GEMs for protein secretory machinery. The cell wall of S. cerevisiae can be a significant deterrent when trying to access intracellular chemical end-products, therefore information on its secretome could guide efforts towards end-product secretion [56]. Microfluidic screening can also be used to screen mutant libraries, and this enabled the isolation of S. cerevisiae strains with more than fivefold improved production of recombinant proteins [57]. Genome sequencing could identify causal mutations in these strains, furthermore, RNA-seq analysis could identify the underlying physiological changes in these mutated strains that govern the improved production of recombinant protein [58].<\/p>\n
Proteomics can also be applied to determine how genetic editing affects the host protein pool. Understanding proteome allocation would reveal whether a cell is expressing non-essential proteins during the bioprocess, such as proteins modulating nutrient adaptation or stress resistance, that might hinder recombinant protein production [59]. By removing the expression burden of such proteins, and \u2018streamlining\u2019 the proteome, ribosomes could have a greater capacity to translate target polypeptides [60]. Decreasing the abundance of non-essential proteins with high expression costs could lead to the largest improvements in bioprocess performance [61]. This could, for example, improve growth rate, which may then enhance chemical production if productivity is growth-coupled. For example, in a recent study that combined E. coli global absolute proteomic data with genome-scale metabolic modeling, it was revealed that under many environmental conditions a large fraction of the proteome is unused [59]. Subsequent proteomic data from hosts re-engineered with a reduced and optimized proteome could then be used to constrain metabolic models further, allowing more efficient engineering strategies in cell factory design.<\/p>\n
Systems Biology and Scaling<\/p>\n
Even when systems biology is exploited fully to retrofit cell metabolism, bridging the gap between a promising laboratory-scale cell factory (titers up to 5\u2005g\/l) and its industrial scale counterpart (\u226550\u2005g\/l) can be a formidable task. Nonetheless, to commercialize a given bioprocess, transitioning to the industrial scale is an essential stage. Industrial scale fermentations typically involve medium volumes greater than 1000\u2005l (and exceeding 100 000\u2005l for commodity chemicals and biofuels), necessitating constant mixing to obtain a homogenous environment, a challenge that increases alongside scale. Imperfect mixing can result in spatial fluctuations of medium composition, translating to temporal variation in the environment to which cells are exposed. Such delays in mass and heat transfer lead to gradient formations within the bioreactor for a variety of components including substrates, products, potentially toxic end- and byproducts, dissolved oxygen, temperature, pH, and carbon dioxide. Therefore, unlike their laboratory-scale equivalents, microbes fermenting at the industrial scale undergo constant changes in their microenvironment and regular feast\/famine cycles, resulting in dynamic metabolic behavior and heterogeneity in biomass, productivity, and yield [62].<\/p>\n
Two ways in which systems biology can support strains through this bottleneck is (i) by multi-omic characterization, and (ii) integrating fluid dynamics and cell metabolism kinetics. Multi-scale comparative omic characterization can address phenotypic changes arising from scaling by pinpointing differences in metabolism for the same cell factory in small-scale batch culture compared to chemostats which begin to simulate industrial conditions. Computational platforms can also be used to combine transcriptomic, proteomic, and metabolomic data from the same fermentation run to understand how extracellular conditions affect cell phenotype. Metabolomic data in particular can be used to establish kinetic models that inform on pathway activity and the abundance of precursors, byproducts, and target chemicals [63]. 13C-based metabolic flux analysis can additionally be used to determine how cell physiology changes with cultivation scale. For example, in 13C fed Penicillium chrysogenum storage metabolites were found to fluctuate as cells were subjected to feast and famine cycles [64]. Recent advances in genome-scale metabolic modeling have also allowed the integration of such metabolic information with bioprocessing-specific parameters such as oxygenation conditions, substrate usage, and cofactor generation, enabling cell activity to be more precisely characterized [65, 66].<\/p>\n
Computational fluid dynamics (CFD) is another promising tool to predict volume-related performance. Based on numerical analysis (e.g., using Navier\u2013Stokes equations) and fluid mechanics, CFD helps to predict fluid flow, heat transfer, and mass transfer, and therefore possible rate-limiting steps at different cultivation scales [67]. By coupling CFD simulations of fluid flow with multi-scale analyses, the bioprocess scale-up can be rationally approached [62]. For example, CFD was recently employed to help design scale-down simulations in which both the dynamic extracellular conditions within a large-scale bioreactor as well as microorganism response, in terms of substrate uptake, are considered [68]. By solving challenges of scaling and outlining the dynamic metabolic response, it will eventually be possible to design and simulate cell behavior in bioprocessing as systematically as one would in the automotive industry.<\/p>\n
Concluding Remarks<\/p>\n
Since the first proof-of-principle industrial strains were proposed more than a century ago the panel of chemicals produced by microorganisms has continued to expand. Improved platform strains, alternative host microorganisms, increased substrate range, and advances in engineering and computational capabilities have all contributed to improved output. Successful production strains at industrial scale, however, seem to be dominated by microbes which naturally overproduce the compound of interest, indicating there is still room to develop the use of heterologous enzymes [69]. However, as technological advances are made in synthetic biology, ME, and systems biology, cells that have been significantly retrofitted for non-native chemical production are beginning to demonstrate commercially competitive titers [2].<\/p>\n
One major bottleneck in bioprocessing is that investment is not linear with scale. For example, costs can be difficult to estimate because of uncertainties in oil prices, government policies, and unclear costs in scaling up production [70]. In recent years the emergence of systems biology has improved cell factory development, allowing an increasing number of chemicals to be produced in response to commercial demand (Figure 4, Key Figure). To further improve the efficiency of cell factory construction, metabolic engineers need to progress through fewer and faster cycles of the design, build, test, and learn cycle (Figure 3). This can be achieved by beginning with an optimal design strategy using in silico predictive models. Because current genome-editing tools have allowed exponential advances in strain construction, the rate-limiting steps of cell factory development predominantly occur in the test and learn phases of DBTL (see Outstanding Questions). For example, although large datasets can lead to meaningful biological insight, the time taken to interpret them remains considerable and can be biased by sample preparation as well as by the analytical methods used for measurement and data processing.<\/p>\n
To improve this, more systematic and precise analytical approaches could be integrated into the omic analysis workflow to expedite learning. Future work should also be invested in the analysis of more host organisms, ideally to acquire in vivo absolute values for RNA, proteins, and metabolites under dynamic conditions. In addition, the various regulatory structures orchestrating metabolic output should be more clearly characterized. During the test phase, new synthetic biology and evolutionary engineering methods, such as biosensors and ALE respectively, can also be implemented more routinely to identify top-performing candidates in high throughput, thereby reducing the number of strains for multi-omic analysis [19, 21].<\/p>\n
It is clear that systems biology approaches have become an integrated part of cell factory development for any new bioprocess. However, we foresee that systems biology in particular will benefit cell factory development when there are major demands on TRY because here it is necessary to approach close to maximum theoretical yields, and this will require major rerouting of metabolic fluxes in the cell. For other processes, different systems biology approaches discussed here may also be applicable, a good case in point being the improvement of recombinant protein production by gaining better insight into the protein secretory pathway. In conclusion, systems biology has evolved to become an indispensable approach for next-generation bioprocessing. By continuing to advance this field, the efficiency and economic competitiveness of bioprocessing efforts will likely also increase as well, making it possible to address the growing demand for fuels and chemicals via greener and more sustainable production processes.<\/p>\n
<\/p>\n
Supplemental Information<\/p>\n