Recombinant protein production by mammalian cells is a core component of today’s multi-billion dollar biopharmaceutical industry. Transcriptome and proteome technologies have been used to probe for cellular components that correlate with higher cell-specific productivity, but have yet to yield results that can be translated into practical metabolic engineering strategies. The recognition of cellular complexity has led to an increasing adoption of systems biology, a holistic investigation approach that aims to bring together different omics technologies and to analyze the resulting datasets under a unifying context. Fluxomics is chosen as the platform context to investigate cell metabolism because it captures the integrated effects of gene expression, enzyme activity, metabolite availability and regulation, thereby providing a global picture of the cell’s metabolic phenotype. At present, the routine quantification of cell metabolism revolves around very basic cellular parameters: growth, substrate utilization and product formation. For a systems approach, however, just measuring gross metabolic features is insufficient; we are compelled to perform high-resolution, large-scale fluxomics in order to match the scale of other omics datasets. The challenges of performing large-scale fluxomics come from two opposing fronts. Metabolic flux analysis (MFA) is the estimation of intracellular fluxes from experimental data using a stoichiometric model, a process very much susceptible to modelling biases. The in silico challenge is to construct the most comprehensive model to represent the metabolism of a specific cell, while the in vivo challenge is to resolve as many fluxes as possible using experimental measurements or constraints. A compromise needs to be established between maximizing the resolution of the MFA model and working within technical limitations of the flux experiment. Conventional MFA models assembled from textbook pathways have been available for animal cell culture for the past 15 years. A state-of-the-art model was developed and used to analyse continuous hybridoma culture and batch CHO cell culture data (Chapter 3). Reasonable metabolic assumptions combined with constraint based analysis exploiting irreversibility constraints enabled the resolution of most fluxes in central carbon metabolism. However, while the results appear consistent, there is insufficient information in conventional measurement of uptake, secretion and growth data to assess the completeness of the model and validity of all assumptions. 13C metabolic flux analysis (13C MFA) can potentially resolve fluxes in the central carbon metabolism using flux constraints generated from 13C enrichment patterns of metabolites, but the multitude of substrate uptakes (glucose and amino acids) seen in mammalian cells, in addition to the lack of 13C enrichment data from proteinogenic amino acids, makes it very difficult to anticipate how a labelling experiment should be carried out. The challenges above have led to the development of a systematic workflow to perform large-scale MFA for mammalian cells. A genome-scale model (GeMs), an accurate compilation of gene-protein-reaction-metabolite associations, is the starting basis to perform whole-cell fluxomics. A semi-automated method was developed in order to rapidly extract a prototype of GeM from KEGG and UniProtKB databases (Chapter 4). Core metabolic pathways in the mouse GeM are mostly complete, suggesting that these databases are comprehensive and sufficient. The rapid prototyping system takes advantage of this, making long term maintenance of an accurate and up-to-date GeM by an individual possible. A large number of under-determined pathways in the mouse GeM cannot be resolved by 13C MFA because they do not produce any distinctive 13C enrichment patterns among the carbon metabolites. This has led to the development of SLIPs (short linearly independent pathways) for visualizing these under-determined metabolic pathways contained in large-scale GeMs (Chapter 5). Certain SLIPs are subsequently removed based on careful consideration of their pathway functions and the implications of their removal. A majority of SLIPs have a cyclic configuration, sharing similar redox or energy co-metabolites; very few represent true conversion of substrates to products. Of the 266 under-determined SLIPs generated from the mouse GeM, only 27 SLIPs were incorporated into the final working model under the criterion that they are significant pathways and are potentially resolvable by tracer experiments. Most of these SLIPs are degradation pathways of essential amino acids and inter-conversion of non-essential amino acids (Chapter 8). In parallel, OpenFLUX was developed to perform large-scale isotopic 13C MFA (Chapter 6). This software was built to accept multiple labelled substrates, and no restriction has been placed on the model type or enrichment data. These are necessary features to support large-scale flux analysis for mammalian cells. This was followed by the development of a design strategy that uses analytical gradients of isotopomer measurements to predict resolvability of free fluxes, from which the effectiveness of various 13C experimental scenarios using different combinations of input substrates and isotopomer measurements can be evaluated (Chapter 7). Hypothetical and experimental results have confirmed the predictions that, when glucose and glutamate/glutamine are simultaneously consumed, two separate experiments using [U-13C]- and [1-13C]-glucose, respectively, should be performed. If there is a restriction to a single experiment, then the 80:20 mixture of [U-13C]- and [1-13C]-glucose can provide a better resolution than other labelled glucose mixtures (Chapter 7 and Chapter 8). The tools and framework developed in this thesis brings us within reach of performing large-scale, high-resolution fluxomics for animal cells and hence realising systems-level investigation of mammalian metabolism. Moreover, with the establishment of a more rigorous, systematic modelling approach and higher functioning computational tools, we are now at a position to validate mammalian cell culture flux experiments performed 15 years ago.

Recombinant protein production by mammalian cells is a core component of today’s multi-billion dollar biopharmaceutical industry. Transcriptome and proteome technologies have been used to probe for cellular components that correlate with higher cell-specific productivity, but have yet to yield results that can be translated into practical metabolic engineering strategies. The recognition of cellular complexity has led to an increasing adoption of systems biology, a holistic investigation approach that aims to bring together different omics technologies and to analyze the resulting datasets under a unifying context. Fluxomics is chosen as the platform context to investigate cell metabolism because it captures the integrated effects of gene expression, enzyme activity, metabolite availability and regulation, thereby providing a global picture of the cell’s metabolic phenotype. At present, the routine quantification of cell metabolism revolves around very basic cellular parameters: growth, substrate utilization and product formation. For a systems approach, however, just measuring gross metabolic features is insufficient; we are compelled to perform high-resolution, large-scale fluxomics in order to match the scale of other omics datasets. The challenges of performing large-scale fluxomics come from two opposing fronts. Metabolic flux analysis (MFA) is the estimation of intracellular fluxes from experimental data using a stoichiometric model, a process very much susceptible to modelling biases. The in silico challenge is to construct the most comprehensive model to represent the metabolism of a specific cell, while the in vivo challenge is to resolve as many fluxes as possible using experimental measurements or constraints. A compromise needs to be established between maximizing the resolution of the MFA model and working within technical limitations of the flux experiment. Conventional MFA models assembled from textbook pathways have been available for animal cell culture for the past 15 years. A state-of-the-art model was developed and used to analyse continuous hybridoma culture and batch CHO cell culture data (Chapter 3). Reasonable metabolic assumptions combined with constraint based analysis exploiting irreversibility constraints enabled the resolution of most fluxes in central carbon metabolism. However, while the results appear consistent, there is insufficient information in conventional measurement of uptake, secretion and growth data to assess the completeness of the model and validity of all assumptions. 13C metabolic flux analysis (13C MFA) can potentially resolve fluxes in the central carbon metabolism using flux constraints generated from 13C enrichment patterns of metabolites, but the multitude of substrate uptakes (glucose and amino acids) seen in mammalian cells, in addition to the lack of 13C enrichment data from proteinogenic amino acids, makes it very difficult to anticipate how a labelling experiment should be carried out. The challenges above have led to the development of a systematic workflow to perform large-scale MFA for mammalian cells. A genome-scale model (GeMs), an accurate compilation of gene-protein-reaction-metabolite associations, is the starting basis to perform whole-cell fluxomics. A semi-automated method was developed in order to rapidly extract a prototype of GeM from KEGG and UniProtKB databases (Chapter 4). Core metabolic pathways in the mouse GeM are mostly complete, suggesting that these databases are comprehensive and sufficient. The rapid prototyping system takes advantage of this, making long term maintenance of an accurate and up-to-date GeM by an individual possible. A large number of under-determined pathways in the mouse GeM cannot be resolved by 13C MFA because they do not produce any distinctive 13C enrichment patterns among the carbon metabolites. This has led to the development of SLIPs (short linearly independent pathways) for visualizing these under-determined metabolic pathways contained in large-scale GeMs (Chapter 5). Certain SLIPs are subsequently removed based on careful consideration of their pathway functions and the implications of their removal. A majority of SLIPs have a cyclic configuration, sharing similar redox or energy co-metabolites; very few represent true conversion of substrates to products. Of the 266 under-determined SLIPs generated from the mouse GeM, only 27 SLIPs were incorporated into the final working model under the criterion that they are significant pathways and are potentially resolvable by tracer experiments. Most of these SLIPs are degradation pathways of essential amino acids and inter-conversion of non-essential amino acids (Chapter 8). In parallel, OpenFLUX was developed to perform large-scale isotopic 13C MFA (Chapter 6). This software was built to accept multiple labelled substrates, and no restriction has been placed on the model type or enrichment data. These are necessary features to support large-scale flux analysis for mammalian cells. This was followed by the development of a design strategy that uses analytical gradients of isotopomer measurements to predict resolvability of free fluxes, from which the effectiveness of various 13C experimental scenarios using different combinations of input substrates and isotopomer measurements can be evaluated (Chapter 7). Hypothetical and experimental results have confirmed the predictions that, when glucose and glutamate/glutamine are simultaneously consumed, two separate experiments using [U-13C]- and [1-13C]-glucose, respectively, should be performed. If there is a restriction to a single experiment, then the 80:20 mixture of [U-13C]- and [1-13C]-glucose can provide a better resolution than other labelled glucose mixtures (Chapter 7 and Chapter 8). The tools and framework developed in this thesis brings us within reach of performing large-scale, high-resolution fluxomics for animal cells and hence realising systems-level investigation of mammalian metabolism. Moreover, with the establishment of a more rigorous, systematic modelling approach and higher functioning computational tools, we are now at a position to validate mammalian cell culture flux experiments performed 15 years ago.