A kinetic description of the Warburg effect in CHO cells

The Warburg effect has been known about for nearly 100 years. It is characterised by a high rate of glycolytic flux and production of lactate in highly proliferating cells, relative to their slower growing counterparts. This phenomenon is still observed in environments where there is sufficient oxygen to catabolise glucose into CO₂ via oxidative phosphorylation. High growth is a trait common among both cancer cells and cells engineered to produce biopharmaceuticals. The Warburg effect has often been described as an enigma due to it being less energy and carbon effecient than oxidative phosphorylation. This leads to the overall question of this thesis: How do glycolytic enzymes regulate the Warburg effect?.

The organism we are investigating the Warburg effect in are known as Chinese Hasmster Ovary (CHO) cells. CHO cells are a common cell line used in the production of biopharmaceuticals. The laboratory we are working in has developed a non–Warburgcell line, CHO-ZeLa, that has a similar growth rate to its parent cell line, CHO-S, which is Warburg positive. In this thesis we investigate this high growth, non-Warburg phenotype and compare it to its parent stain using a kinetic model.

We developed a Bayesian kinetic modelling framework called Maud that uses Hamiltonian Monte Carlo for posterior sampling. To evaluate our framework we carried out a case study of a representitive metabolic model, which compared full posterior sampling to an approximation of the posterior using point estimation. We showed that the posterior distribution is not well approximated by point estimation methods. Additionally, full Bayesian sampling translated to tighter confidence intervals and improved predictions. We made Maud available as a Python package that automates ODE construction, diagnostics, and posterior predictions.

We collected a dataset consisiting of 6 cell lines that included their proteome, metabolomeand fluxome (measured using ¹³C metabolic flux analysis). This dataset was crucial indeveloping a kinetic model of glycolysis in CHO cells that spaned the Embden-Meyerhof-Parnas (EMP) and pentose phosphate (PP) pathways.

Our comparative model of the CHO-S and CHO-ZeLa cell lines showed that hexokinase 2 is a strong regulator of glycolysis and may be mediated via mitochondrial binding,a mechanism not previously modelled. Furthermore, our results aligned with previous models of the Warburg effect in CHO cells that suggest that allosteric regulation of phosphofructokinase M is required for the Warburg effect. Kinetic modelling has the side effect of helping identify when mechanisms may be incorrectly understood. We were able to identify that the current understanding of triosphosphate isomerase and glyceraldehyde-3-phosphate dehydrogenase mechanisms are likely missing additional regulation, which was indicated by the surrounding metabolomics data not fitting the model.

The results also suggest that we require a method to account for phosphorylation in metabolic models. We therefore developed a Monod-Wynman-Changeaux (MWC) analogous model of phosphorylation, and a combinatorial model of phosphorylation. These models will futher be used to model the hexokinase 2 and 6-phosphofructo-2-kinase/fructose2,6-bisphosphatase mechansism, which were identified as essential to regulate the Warburg effect.