Driving towards digital biomanufacturing by CHO genome-scale models

Abstract

Genome-scale metabolic models (GEMs) of Chinese hamster ovary (CHO) cells are valuable for gaining mechanistic understanding of mammalian cell metabolism and cultures. We provide a comprehensive overview of the past and present developments of CHO-GEMs and in-silico methods within the flux balance analysis framework, focusing on their practical utility in rational cell line development and bioprocess improvements. There are many opportunities for further augmenting the model coverage and establishing integrative models that account for different cellular processes and data for future applications. With supportive collaborative efforts by the research community, we envision that CHO-GEMs will be critical to the increasingly digitized and dynamically-controlled bioprocessing pipelines, especially as they are successfully deployed in conjunction with artificial intelligence and systems engineering algorithms.

Unleashing the potential: Refining biomanufacturing with genome-scale metabolic models of CHO cells

Chinese hamster ovary (CHO) cells have been the dominant host for producing recombinant therapeutic proteins (RTPs) (see Glossary) in biomanufacturing [1]. With their proficiency in generating human-compatible glycosylation patterns and continuous progress in both cell line and bioprocess development, they have been considerably valuable for enhancing production of RTPs [2]. These improvements have relied on empirical methods, demanding extensive time and resources. However, the complex, non-linear nature of biological systems poses a significant challenge, impeding the development of precise control systems essential for consistent product quality and intensified productivity. In response, the biopharmaceutical industry is pivoting from these empirical methods to a new era of process digitalization and automation. This paradigm shift aims to forge a more streamlined, reliable, and cost-efficient bioprocessing platform to reduce deviations and accelerate the manufacturing of high-quality RTPs [3]. A comprehensive mechanistic understanding of cellular characteristics and their dynamic behaviors during cell culture will help the industry achieve a practical digital representation of bioprocesses.

A critical research milestone in the cell culture field involves the sequencing of the CHO cell genome [4], which provided a valuable resource of the myriad genetic elements associated with culture phenotypes, including transfection efficiency, genetic stability, growth rates, and productivity. These data also enabled the reconstruction of genome-scale metabolic models (GEMs), which mathematically linked genotype with phenotype. These CHO-GEMs can act as digital replica of CHO cells, providing lens to interpret cellular behaviors and metabolic states in biomanufacturing [3]. However, despite their great potential for the rational design of CHO cell lines and associated cell culture processes, we continue to face challenges and limitations in terms of model reliability, method development, and practical application. Hence, here we present the evolution and current status of CHO-GEMs, exploring their industrial or industrially relevant applications. Moreover, we propose future directions for model enhancement, focusing on overcoming existing hurdles and moving towards a fully virtual CHO cell system.

Mapping: Decoding mammalian cell metabolism with CHO genome-based models

The initial CHO-GEM, iCHO1766 [5], emerged from a synergistic collaboration within the scientific community, and was built from the CHO cell genome sequence unveiled in 2011 [6]. This model enabled subsequent advancements, where its applicability, quality, and reliability saw continuous improvement. Furthermore, researchers have been able to further incorporate high-throughput omics data (e.g., transcriptomics, proteomics, metabolomics) to model different culture conditions using diverse in-silico methods within the framework of flux balance analysis (FBA) (Box 1). Consequently, this comprehensive approach allowed for a deeper mechanistic understanding of CHO cell metabolism, paving the way for systematic optimization of cell cultures, and pinpointing potential engineering targets.

Box 1. Flux balance analysis.

Flux balance analysis (FBA) is the most basic method of constraint-based modeling, which enables the representation of intracellular biochemical and transport reactions, as well as the characterization of cell state, using flux values with the steady-state assumption [60]. The stoichiometric matrix (S) represents the relationships between the various metabolites (m) involved in the reactions (n) of the system. Thus, the stoichiometric matrix captures the conversion of reactants into products in each reaction. Each row corresponds to a metabolite, and each column corresponds to a reaction, with entries in the matrix indicating the stoichiometric coefficients. The central idea of FBA is to optimize a cellular objective, often the production of biomass or another cellular component (e.g., ATP), subject to the constraints imposed by the stoichiometric matrix and other environmental conditions. The optimization problem is formulated as a linear programming problem, aiming to find a set of fluxes that maximize or minimize the objective function. For example, the FBA framework idealizes cellular metabolism through the optimization of a specific metabolic objective, such as maximizing the growth rate as with the biomass objective function (BOF) [60].

Onward and upward of CHO-GEMs: Model applicability and quality in focus

The iCHO1766 reconstruction provided a genetic and metabolic backbone for deriving cell line-specific models (Figure 1). For example, iCHO1766 has been tailored using relevant experimental and omics data to reflect the metabolism of parental CHO cell lines such as CHO-K1, -S, and -DG44 and thus predicting their growth phenotype, auxotrophy, and potential productivity upon various biochemical treatments [5]. Similarly, a GEM for CHO-K1 derived producer (i.e., CHO-SH87) was generated with the necessary amino acid composition for antibody synthesis and served as a scaffold for combining multi-omics profiles (e.g., transcriptomics, metabolomics, lipidomics, and glycomics). This study elucidated the genomic rearrangements and metabolic adjustments occur in CHO cells, following the transgene integration [7]. Likewise, time-series transcriptomic and exometabolomic data were used to refine parental and producer GEMs to understand their metabolic behaviors under various culture conditions [8-10]. Recently, the cell line specificity of CHO-GEMs was further assessed by exploring their compositional variations in biomass constituents, such as DNA, RNA, proteins, lipids, and other relevant components [11,12].

Figure 1. Development and improvement of CHO-GEMs for deciphering metabolic complexity.

(A) The iCHO1766 provided a genetic and metabolic backbone for further deriving cell line- and product-specific models by incorporating secretory pathways, enzyme capacity constraints, and a systematic gap-filling algorithm and combining omics datasets. (B) Various versions of reconstructed CHO-GEMs are presented in chronological order, providing information on the model components for generic models (iCHO1766 [5], iCHO2291 [14], iCHO2048s [13], iCHO2101[15], and iCHO2441[16]) s and contextualized model models (represented as CHO-K1 [7], CHO-DG44 [9], CHO-biomass [12] and CHO-parameters [8]). Abbreviations: CHO-GEM, Chinese hamster ovary cell-genome scale metabolic model; COPII, coat protein complex II; GPR, gene-protein-reaction association.

The quality of the CHO-GEMs has been persistently enhanced by refining genomic contents (i.e., annotated sets of genes, reactions, and metabolic gaps) and extending coverage by including functional processes and regulatory elements based on secretome and kinetome. For instance, iCHO2048s captured the protein post translational biosynthetic machinery and secretory pathways spanning across multiple cellular organelles such as endoplasmic reticulum (ER), the Golgi apparatus, and endomembrane system onto the genome-scale metabolic network, and thus allowing the prediction of metabolic costs and bottlenecks linked with various RTPs [13]. Comparably, iCHO2291 incorporated publicly available kinetic information such as turnover numbers (k_cat) and molecular weights of all the metabolic enzymes accounted in the CHO-GEM to estimate a more plausible internal flux distribution by properly allocating enzymatic costs and constraining the total enzyme capacity within the cells [14]. In parallel with the model expansions, additional efforts improved and the metabolic coverage of GEMs by augmenting the lumped and multi-step expression of biosynthetic pathways (e.g., lipid metabolism and glycan biosynthesis) [14], and resolving the gaps and dead-ends present within the CHO metabolic network using various algorithms [15,16].

Forward with prediction fidelity: Improving reliability of CHO-GEMs

One of the well-known challenges in FBA while implementing it in large-scale GEMs such as CHO cells is the difficulty of reliably predicting cell culture behavior of CHO cells and interpreting the metabolic activity based on the resulting intracellular flux distributions. These challenges stem from the complexity and regulatory uncertainty in relating the multiple inputs (e.g., glucose, and amino acids) with multiple outputs (e.g., biomass, toxic metabolites, antibody proteins, etc.) whereby numerous alternative feasible solutions are possible for the same phenotype. This inherent limitation is due to the underdetermined nature of the models, which contain more flux variables than metabolite balance equations, as well as the unknown regulation of metabolic enzymes. Thus, it is desirable to narrow down the solution space and determine more plausible metabolic states [17], which can be achieved by imposing additional constraints and using different objective functions within the FBA framework [8].

Implementation of improved constraints, based on condition-dependent empirically derived assumptions, have yielded better predictions on cellular phenotypes (e.g., growth and nutrient uptake rates). For example, measurement variations for nutrients and byproducts in cell cultures were deduced by the principal component analysis of 21 datasets in reactor experiments. There were used to relax the constraints on their exchange rates (HybridFBA), thus effectively dealing with the potential technical errors in analytical measurements [18]. In addition, the essentiality of key nutritional components was predicted among three CHO cell lines (CHO-K1, -S, -DG44) by minimizing their uptake rates (Uptake rate objective function, UOF) instead of maximizing the cell growth which is routinely used as objective function in the FBA [19]. Experimentally-determined and computationally-estimated non-growth associated maintenance energy (i.e., maintenance ATP) was also parameterized in CHO-GEMs to reflect conditional energy requirements in producers and non-producers [20].

The reliability of intracellular flux predictions can be enhanced by FBA variants. Flux sampling can capture a broader spectrum of flux distributions, including suboptimal states without considering any cellular objectives [16,21], while parsimonious FBA (pFBA) predicts the most plausible flux distribution among multiple others based on the assumption that cells minimize the overall metabolic activity during resource utilization [22]. In carbon constraint FBA (ccFBA) [23] and nitrogen/carbon constraints FBA (nccFBA) [16], the total amounts of carbon/nitrogen in the culture media were added into the elemental balance constraints to remove unrealistic internal loops or futile cycles, thus constraining the bounds of model-predicted flux to a more realistic range. Therefore, the utilization of various carbon and nitrogen sources (glucose, lactate, amino acids) required for cell growth and RTP production was better estimated, possibly identifying desirable media conditions for the enhanced antibody productivity. Notably, enzyme capacity-constrained FBA (ecFBA) was applied to iCHO2291, by integrating kinetic parameters (kinetome) to improve its prediction fidelity, thus more realistically portraying resource allocation and regulatory processes in CHO cell cultures [14]. This approach successfully demonstrated the reliability of flux predictions from iCHO2291 with ¹³C isotope-labeled measurements and enabled a flux-based understanding of lactate overflow metabolism, which is rarely captured by conventional approaches.

Searching and navigating: Leveraging CHO-GEMs in bioprocess development

With growing interest in utilizing CHO-GEMs for bioprocess development, their practical applications have primarily focused on upstream processes; only a limited number of studies have explored their potential in cell line development. Here we review how CHO-GEMs can guide the systematic and rational development of both cell lines and culture processes (Figure 2). The insights from these models are advancing our mechanistic understanding and proficiency in bioprocessing.

Figure 2. Key applications of CHO-GEMs.

A phylogenetic tree is used to depict these publications with each box corresponding to a specific FBA method (upper smaller segment), GEM model (lower larger segment), and research target(s) such as growth, productivity, product quality, and toxic effects. The generic models are represented in yellow, while the cell line-specific models are depicted using different colors: CHO-K1 (orange), CHO-S (green), and CHO-DG44 (blue). In the box with branching arm(s), research targets such as growth, productivity, product quality, and toxic effects are indicated. Depending on the specific focus of each study, the research targets are represented by larger circles. Additionally, for cases involving improvements in media, instances of target enhancements are denoted by blue circles, with the action(s) at the top of the box. The arrows above represent an increase in the concentration of the mentioned component(s), while the arrows below indicate a decrease, ‘+’ denotes the addition of new additives to the media. Abbreviations: AA, amino acid; ARA, arachidonate; ASN, asparagine; ASP, aspartate; ENM, essential nutrient minimization; FBAwMC, flux balance analysis with molecular crowding; FVSEOF, flux variability scanning based on enforced objective flux; GLN, glutamine; GLU, glutamate; LAMOS, linear algorithm for finding multiple optimal solutions; LEU, leucine; ML, machine learning; PEA, pathway enrichment analysis; THR, threonine; VAL, valine; ecFBA, enzyme capacity-constrained flux balance analysis.

Cell line development and engineering

The conventional procedure for CHO cell lines development requires considerable time and resource investments. Indeed, it involves multiple steps, ranging from screening and isolation of cells to clone selection and characterization. Thus, we need methods for more affordable and efficient development of high-yielding expression hosts; such methods can be based on GEM-guided analysis to help understand mechanisms underlying clonal variation, stability, productivity, and product quality, and to identify cell engineering targets. To this end, recent studies have combined enzyme assays [24] and time-series transcriptomics [10] with CHO-GEMs. In such studies, a comparative flux analysis between high and low producers identified metabolic bottlenecks within the pentose phosphate pathway (PPP) and tricarboxylic acid (TCA) cycle as well as suggested possible engineering strategies for enhanced growth and productivity. Similarly, metabolic connectivity information derived from CHO-GEMs has been combined with multi-omics profiles for gene set enrichment analysis for characterizing various parental CHO cells including CHO-K1, -DG44 and -DXB11. This analysis revealed variations in cell cycle progression and protein processing from the absence of the dihydrofolate reductase (DHFR) [25]. Another study demonstrated the utility of CHO-GEM for the engineering of CHO-S derived producers was introduced by computationally evaluating the lethality of gene candidates for integration sites [26] and CRISPR/Cas9 based gene editing [27]. For next-generation cell line development and engineering, omics-based model-guided strain design algorithms can be applied to systematically identify and prioritize context-specific targets for genome editing [28].

Process development

CHO GEM-based FBA was successfully demonstrated for process development in multiple studies. Mechanistic insights into phenotypic responses from metabolic changes during cell cultures were revealed under various environmental conditions, specifically changes in nutritional components of the media, process parameters (e.g., dissolved oxygen (DO), temperature, pH, aeration, etc.) and operational modes. For model-driven media development, phenotypic specificities and metabolic bottlenecks in CHO cells could be characterized to more systematically develop strategies to improve culture performance of CHO cells. For example, a flux scanning approach was applied to shortlist candidate nutrients that may increase mAb productivity, thus identifying 15 metabolites for a subsequent step of the design of experiment (DoE) [29]. In addition, their combinations were further explored by incorporating dietary reactions into the FBA framework, showing their optimal and synergistic sets, which await an experimental validation [30]. Another study aimed to predict optimal nutrient levels in the media. The UOF approach was employed to simulate the uptake rates of three essential amino acids (leucine, lysine and valine), and guide their supplementation strategies [31]. Similarly, the growth-inhibiting effects of vital nutrients (e.g., arginine, glucose, lysine, phenylalanine, and valine) were quantified through flux sensitivity analysis, suggesting subsequent basal media designs [32]. Recently, a data-driven and CHO GEM-guided systematic framework was developed to evaluate metabolic bottlenecks, disclosing insufficient nutrients and metabolic wastes in antitrypsin-producing CHO-DG44 cells [33]. As part of the industrial applications of the framework for rational media improvement, nutritional additives for basal media (i.e., coenzyme Q10) [34], feed media (i.e., glutamate, and asparagine) [35], and dipeptide feed media (i.e., glutamine and asparagine) [36] were targeted to debottleneck energy imbalance, byproduct accumulation, and excessive utilization of TCA cycle intermediates, respectively. These interventions were validated for their ability to increase cell growth and final titer. Concurrently, the physiological and metabolic effects of operating conditions were also explored using GEM-based approaches. For instance, considering the balance between cell density and dilution rate, the cell-specific perfusion rate was parameterized into the FBA framework to represent multiple metabolic phases (i.e., excessive nutrient, limited nutrient and maximized yield) during continuous CHO cell cultures [37]. Interestingly, this approach was further exploited to reduce the accumulation of unused nutrients or toxic byproducts, thus indicating the requirements for the potential perfusion media [38]. The impact of gas sparging, one of the most important operating parameters, was also investigated by simulating the metabolic response to sparging stress under fed-batch bioreactor cultures. It revealed that the elevated H₂O₂ turnover rates from the increased uptake of amino acids in higher sparging conditions leads to oxidative stress which could be restored via manipulating the redox homeostasis [39].

Driving: Propelling CHO-GEMs towards digital bioprocessing

To date, a variety of CHO-GEMs and algorithmic innovations have facilitated practical implementation for cell line and bioprocess improvements. Looking ahead, we anticipate a shift towards more digitized and dynamically managed culture processes. This evolution is expected to generate a wealth of high-throughput omics and real-time culture data, supported by state-of-the-art analytical technologies ranging from next-generation sequencing to non-invasive optical sensors. Despite these advancements, the industrial application of CHO-GEMs is still at its nascent stages, as the current models do not fully capture condition-specific cellular regulations that affect culture performance [40]. Thus, the community needs to (1) expand and refine the models by covering and integrating more regulatory elements spanning multiple layers and scales, and (2) relate them with key process variables by formulating the dependency of process parameters within the GEM-FBA framework.

CHO-GEMs also need to more accurately emulate and decipher phenotypic responses to genetic and environmental shifts during cell culture. For this, further expansions of the existing CHO-GEMs will increase their comprehensiveness by including more metabolic and regulatory mechanisms. In this regard, more contextualized models have been developed based on a variety of omics datasets (e.g., genomics, transcriptomics, proteomics, and metabolomics) as outlined in Table 1. In addition to the gene-protein-reaction (GPR) and their metabolic associations, the models could integrate the mechanistic details of transcriptional regulation, protein secretion, post-translational modifications (PTMs), and enzyme catalytic capacity (Figure 3). For example, a detailed representation of the transcription factors (TFs) and other pathway machinery (e.g., protein transport, unfolded proteins response, chaperones) can be accounted for by expanding the scope of CHO-GEMs; meanwhile, logical and probabilistic integration of transcriptional interactions and major features of protein synthesis, folding, trafficking, and eventual release were partially presented in the models [13,41]. Among the PTMs, N-linked glycosylation is an important process connected to metabolism, affecting critical quality attributes (CQAs) of secreted RTPs [42,43]. To modularize this regulatory component at the molecular scale, multistep glycosylation of N-linked glycans in ER and Golgi apparatus could be mechanistically described to relate the abundance of nucleotide sugar precursors and glycan profiles. Then, a hybrid CHO-GEM and glycosylation model can link nutritional environment and critical process parameters (CPPs) with product quality during the production culture via nucleotide sugars [44], which are differentially synthesized through cellular metabolism in a condition-dependent manner. Thus, adjustable and actionable targets can be identified to maintain desirable glycoforms of RTPs [45]. At the reaction scale, kinetic information (i.e., enzyme capacity constraints with k_cat values) was incorporated within the ecFBA to obtain more plausible metabolic states. However, the predictive accuracy of the model largely depends on the CHO-specific k_cat values which can be further updated by resorting to machine learning/deep learning algorithms on the basis of protein sequences and/or substrate structures for a diverse array of enzymes as similarly done for Escherichia coli [46] and, yeast and fungi species [47].

Table 1.

Example omics integration algorithms already used with CHO-GEMs for different cell lines.

Algorithm	Input omics dataset	Description	Target cell line		Method Ref.
GIMME a	Transcriptome, Proteome	Non-classifying method Determine active reactions above the transcriptome threshold Calculate the penalty score based on their expression deficit between minimum flux and threshold By minimizing the total sum of (flux)×(penalty score), constrain low-expressed reactions with the estimated flux value	CHO-K1	[5,7,16]	[55]
			CHO-S	[5,7,16]
			CHO-DG44	[9]
			CHO-SH-87	[7]
			CHO-GS	[10]
iMAT b	Transcriptome, Proteome	Classify reactions into high- and low-expressed with the transcriptome or proteome threshold Determine active/inactive reactions by comparing expression level and possible fluxes based on MILP optimization Remove inactive reactions, while retaining low-expressed reactions which are required for feasible solutions	CHO-S	[56]	[57]
mCADRE c	Transcriptome, exometabolome	Classify reactions in high- and low-expressed with transcriptome threshold Define high-expressed reactions as conserved core reactions Remove unnecessary reactions that are unassociated with core reactions based on reaction connectivity	CHO-S	[56]	[58]
			CHO-DG44	[13]
CORDA d	Proteome, exometabolome	Classify reactions in high- and low- expressed reactions based on proteome abundancy Assign a zero-reaction cost to highly expressed reactions and a higher cost to low-expressing reactions By minimizing the total sum of (flux)×(cost), define inactive reactions with flux-based threshold	CHO-K1	[16]	[59]
			CHO-S	[16]

^aGene Inactivity Moderated by Metabolism and Expression

^bIntegrative metabolic analysis tool

^cMetabolic Context-specificity Assessed by Deterministic Reaction Evaluation

^dCost Optimization Reaction Dependency Assessment

Figure 3. Multiple layers and scales of CHO-GEMs to capture condition-specific cellular regulations affecting culture performance.

This figure illustrates how to expand and enhance models by integrating regulatory components such as transcriptional regulation, enzyme activity regulation, glycosylation network, and secretory pathway within a CHO-GEM, beyond gene-protein-reaction (GPR) associations. The concept of each content (in clockwise order starting from the upper left box) are introduced: transcriptional regulation (blue), enzyme activity regulation (orange), secretory pathway (green), and glycosylation network (yellow). In addition to the concepts, dotted boxes illustrate how the relevant constraint and network can be incorporated into CHO-GEM. Transcriptional regulation (upper left): Respond to a stimulus, transcription factors (TFs) are either expressed or inhibited, leading to changed expression levels of subsequent genes associated with metabolic enzymes. Probability calculation of gene expressions under certain condition and resultant enzyme expressions, and their formulation as a reaction constraint facilitates regulation-represented metabolic model. Enzyme activity regulation (upper right): Enzyme activity, or kinetome, determines metabolic state of a cell, making itself a prerequisite for construction of a precise GEM. However, kinetome is scarcely measured for mammalian cells. AI-driven k_cat prediction models utilize databases containing enzyme sequences and substrate structures to estimate enzyme activity based on organism or cellular compartment. Then, enzyme capacity constraints can be integrated into model in conjugation with compartment-specific enzyme allocation to construct an advanced CHO-GEM. Secretory pathway (lower right): Translation and glycosylation process consumes energy cofactors and proteins, therefore inclusion of these components is necessary. Secretory pathway-involved subsystems and their metabolic enzymes can be formulated into stepwise reactions [13], along with their consumption of energy, metabolites and proteins added to reaction formula. Glycosylation network (lower left): Glycosylation pathway occurs from ER to the Golgi apparatus. Through advanced analyses of glycome, the glycosylation patterns can be investigated, which facilitates the optimization of production processes for desired mAb quality. The N-linked glycan, which is determined by a sequence of interconnected pathways, can vary depending on intracellular metabolism and the levels of nucleotide sugar precursors and carbon sources. The CHO-GEM, expanded with a glycosylation network through a hybridized approach, can identify rerouted fluxes and possible PTM targets for a desired N-glycosylated pattern. Abbreviations: CNN, convolutional neural network; COPII, coat protein complex II; ER, endoplasmic reticulum; GCN, graph convolutional network; GPR, gene-protein-reaction; GalNAc, N-acetylgalactosamine; GlcNAc, N-acetylglucosamine; IgG, Immunoglobulin G; NeuAc, N-acetylneuraminic acid; P, probability; P4HB, prolyl 4-hydroxylase beta polypeptide; PTM, post-translational modification; TF, transcription factor.

With advanced multiple-layered and -scaled CHO-GEMs, condition-dependent cellular behaviors can be better predicted for more reliable industrial applications by empirically or mechanistically relating nutritional environment and CPPs (e.g., pH, temperature, DO levels, and agitation rates) with model components (e.g., reaction, enzyme activity) through mathematical or logical representations. When it is unclear what the specific mechanisms are underlying the dynamics of nuanced and fluctuating conditions in bioreactors, the FBA framework could be expanded by effectively adopting machine learning approaches, thus better predicting cellular phenotypes or metabolite concentrations [48,49] and their future trajectories [50]. Recently, major operational variables, e.g., pH and temperature, were parameterized into GEMs for microbial systems. Specifically, the flux variables related to redox cofactor (e.g., NADH), proton transport, and the enzymatic turnover rate (k_cat) and maintenance energy were expressed as canonical functions of pH and temperature, respectively, reflecting their dependency in E. coli [51] and Saccharomyces cerevisiae [52]. Accordingly, the relevant formulations can be derived from multi-omics and culture profiles of CHO cells under various pH and temperature conditions (Box 2), thereby guiding experimental design for cell line engineering and improving process efficiency, adaptability, and operability under dynamic environments for the digital bioprocessing.

Box 2. Case study: integrating operational parameters and CHO-GEMs.

Genome-scale metabolic models (GEMs) poses can integrate operational parameters, such as pH and temperature, in order to mechanistically predict CHO cell growth, productivity, and product quality (Figure I). In CHO cell culture, pH has been reported as a key parameter that affects cell growth, viability, productivity, and product quality [61]. Mechanisms associated with pH can be illustrated as multiple layers of advanced GEM, specifically, transcriptional regulation, post-translational modifications, as well as enzyme kinetics, and can be represented as pH-dependent components. Especially regarding enzyme kinetics, pH acts as a critical parameter driving structural deformation. Thus, enzymatic activity can be represented as a function of environmental pH and the enzymatically optimal pH.

Figure I. A detailed structure of pH and temperature integration into GEM.

A model of pH dynamics and transport, which provides compartment-specific pH differences in a cell, applicable to changes in enzymatic activity. Temperature changes are utilized to calculate alterations in the enzyme activity. Experimental measurements taken under pH or temperature shifts can be utilized for comparative analysis to validate the integration process.

Combined with ecFBA, pH-dependent k_cat estimation will enable the simulation of cellular metabolic states under varying pH conditions. Moreover, adopting kinetic equations for acidic or basic metabolite transportation rates between compartments, such as the cytoplasm, extracellular space, and organelles, will facilitate the calculation of compartment-specific pH and resultant enzymatic activity. Predicting k_cat for mitochondria, ER, and the Golgi apparatus will provide accurate predictions of fluxes related to cell proliferation and product quality, which are prone to perturbations by culture phase or pH [62,63]. Similarly, temperature affects enzyme kinetics, thereby perturbing cellular systems with respect to VCD and productivity [64]. However, the lack of mechanistic relationships between temperature and cell response hinders the integration of temperature into GEMs. Similar to pH, enzyme kinetics can be formulated as a function of temperature. Furthermore, the undiscovered gap of temperature or pH-dependent effects on cell metabolism can be explored by incorporating comparative analysis of multi-omics data obtained from high-throughput condition-varying cultures.

Concluding remarks and future perspectives

Advances in CHO-GEMs and relevant analytical methods are already showing utility in several practical applications including rational cell line and process development. Further capabilities of these models will emerge following a sustained focus on augmenting model coverage and creating an integrative model structure with multiple layers, including additional cellular processes other than metabolism (e.g., enzyme and transcriptional regulation).

As the field continues to grow and evolve, we foresee enhanced predictive prowess from GEMs, catalyzed by sustained collaboration and dedication within the biomanufacturing community, such as the International Biomanufacturing Network (IBioNe) and the European Society for Animal Cell Technology (ESACT). The community is envisaged to actively foster education through organized courses and workshops, providing valuable insights into the modeling of animal cell culture. This educational initiative not only contributes to academia but meets the growing demands in both CHO-GEMs and FBA applications within the industry (see Outstanding Questions).

Outstanding questions Box.

Which algorithmic methods can be integrated with CHO-GEMs to enhance their predictive accuracy of cellular behaviors in diverse conditions, especially considering complex biological phenomena?

Is it feasible to develop a universal framework or set of guidelines for the creation and application of GEMs across various CHO cell lines in industrial contexts?

How can we evolve CHO-GEMs to more precisely simulate dynamic cellular responses under fluctuating bioprocessing conditions?

Which strategies can be employed to fuse CHO-GEMs with regulatory components and process modifications, aiming to achieve a more holistic representation of biological interactions with the cell culture environment?

Which innovative models or methodologies hold potential to be synergized with CHO-GEMs for fully digitalizing bioprocesses?

What specific roles do academic and industrial partnerships play in propelling the advancement of biomanufacturing via CHO-GEMs, and how can these collaborations be further optimized?

CHO-GEMs, renowned for their adaptability across various mammalian cell lines, are poised to revolutionize fields beyond their current scope. From recombinant Adeno-associated virus (rAAV) production for gene therapy using HEK293 cells [53] to cultured meat production with animal tissues [54], they are poised to propel applications in biopharmaceutical and food production, since such models mechanistically encode the processes involved in biomass and macromolecule production. In addition, GEMs are pivotal in shaping the digital twins of bioprocesses, as an instrumental tool for accurately tracking, forecasting, and navigating cellular behaviors and metabolic states with timely monitored and digitalized culture data. Hence, we envision a future where GEMs, seamlessly integrated with data-driven AI models and sophisticated control algorithms, autonomously orchestrate bioreactor runs, meticulously steering into the optimal culture trajectory. In this context, AI and machine learning strategies can be applied to expand the adoption of GEMs across the biopharmaceutical industry for more innovative and efficient RTPs production [40]. This represents a profound culmination of scientific progress and innovation towards fully digital biomanufacturing.

Glossary

Biomass objective function (BOF)

A mathematical equation used in constraint-based models to represent the biomass components necessary to meet the growth requirements of a cell

Constraint-based methods

Computational techniques utilizing genetic, biochemical, and physiological constraints to model, simulate, and analyze biological systems, especially applied to GEMs for studying the behavior of metabolic networks under various conditions

Enzyme capacity

The kinetic parameters in terms of molecular weight and turnover number of each enzyme and the available intracellular protein content, obtained from publicly available databases (e.g., BRENDA, UNIPROT, and KEGG) and relevant studies

Gene set enrichment analysis

A computational method used to determine whether predefined sets of genes are differentially expressed under different biological states or conditions

Genome-scale metabolic models (GEMs)

Comprehensive mechanistic and mathematical representations that integrate information from an organism's genome to predict and analyze the complex network of biochemical reactions occurring within its metabolism

Kinetome

Ome that explains kinetic information such as kinetic parameters (turnover number) and molecular weights with gene-protein-reaction (GPR) information

Post-translational modifications (PTMs)

Protein modifications after protein synthesis. These modifications, such as phosphorylation, glycosylation, acetylation, methylation, and ubiquitination, play vital roles in protein maturation and functionality

Recombinant therapeutic proteins (RTPs)

Non-native proteins used to treat various diseases and disorders such as cancer, inflammation, and genetic disorders. CHO cells are the primary producers of RTPs, particularly for monoclonal and bispecific antibodies (mAbs and bsAb)

Uptake rate objective function (UOF)

Mathematical equation that is used to predict optimal nutrient uptake rates for cellular growth and function