The enzyme Escherichia coli (E. coli) transketolase (TK; E.C. 2.2.1.1) occupies a pivotal place in metabolic regulation. This enzyme catalyses the interconversion of sugars by transferring a two-carbon ketol unit from a ketose donor substrate to an aldolase acceptor substrate. It is also an important biocatalyst in stereo-specific carbon-carbon bond synthesis with potential industrial application for the synthesis of pharmaceuticals, agrochemicals and fine chemicals. Although many useful reactions have been reported for TK, many of the substrates and products are unstable or insoluble at the pH or temperature for which the enzyme has optimum activity. Understanding the structural stability of transketolase using mutagenesis under bioprocess conditions will improve our capacity to comprehend and ultimately to engineer it to make it work in a broader range of pH or temperature to potentially help in the reduction of process time and to increase the quality and solubility of products. In this thesis, Chapter 3 studied the denaturation of E. coli TK at extreme temperature and pH. DLS data for thermal study of holo-TK at pH 7 indicated that aggregation occurred at above 58 °C using a fixed temperature ramping rate. However, the thermal denaturation profile measured by CD for holo-TK showed a gradual non-cooperative loss of secondary structure as the temperature increased between 5 and 50 °C, and the mid-point of unfolding/aggregation under the same conditions was confirmed to occur at 58.3 °C. The enzyme was observed to be more tolerant at high pH, both structurally and catalytically, where no loss of tertiary structure was evident at pH 9 and 50% of enzyme activity was retained at pH 11. All the results presented in this chapter were published in Jahromi et al. 2011 which provide a good and solid base for protein engineering of the E. coli TK. Chapter 4 investigated the impact of dimer interface mutations on the thermal stability of the E. coli TK. The aim of this Chapter was to determine regions of the dimer interface that were more susceptible to aggregation. Mutations in this study 4 were limited to the dimer interface region and at least at a 4Å distance from the active-site. Computational tools such as Aggrescan, Tango and Pasta were used to predict the aggregation prone regions in the structure. None of the mutants in this study showed a higher specific activity compared to the wild-type as expected. The denaturation profiles from CD and DLS measurements showed mutants E647Q and D617N to have higher unfolding/aggregation temperatures compared to wild-type. The double mutants H94A-E647Q and D617N-K621M were shown to have additive effects compared to their corresponding single mutants. This study revealed that mutations of the PP-domain and Pyr-domain of E. coli TK are more prone to aggregation. This could be explained by the fact that the residues in these two domains are the ones that mainly contribute to the dimer interface stability. Chapter 5 studies mutagenesis of E. coli transketolase using consensus concept. Mutations were spread in all three domains and located either at the dimer interface or buried in the structure. This study revealed remarkable results with all of the seven mutants having stabilising effect on the enzyme compared to wild type. Previous studies have reported this technique to identify only 30%-50% of the mutants to be stabilising. The result of temperature ramping on the secondary structure of the consensus mutants, as measured by CD, showed that all of the mutants apart from F554L, displayed a stabilising effect, with an increase in the mid-point of unfolding/aggregation temperature in the range of 0.3°C to 4.3°C. Interestingly, F554L also had higher CD and DLS thermal unfolding/aggregation onset points compared to wild-type TK. The I365L mutant showed the highest increase in the mid-point of unfolding/aggregation of about 4.3 °C. Another remarkable result was that the I365L mutant also showed the highest specific activity retained both in purified and lysate form which were approximately 67% and 20% respectively. G506A was another interesting mutant that showed above 50% specific activity retained in the purified form. Overall, the consensus sequence approach has proven to be the most effective technique for generating enzyme variants with improved aggregation propensity.

Many ocular diseases result in visual impairment often leading to blindness with disease progression, one such example is Retinitis pigmentosa (RP). The underlying cause for blindness is often due to the degeneration of photoreceptors and supporting cells in the retina. Photoreceptors are responsible for the conversion of light signals into electrical impulses which are processed by the brain to form a visual image. Unlike other cells, photoreceptors cannot be regenerated by the body. For this reason current treatment for many ocular diseases aims to preserve what little vision patients hold. The only way such diseases could be cured is by regenerating photoreceptors and transplanting them into patients to replace lost cells and function. The differentiation of retinal cells from human embryonic stem cells (hESCs) is a difficult process. Current strategies produce low efficiencies of retinal cells which could not supply cells in a clinical environment. It is evident that the key lies within the tight control of the microenvironment of these cells at various stages of the differentiation process. In this study various optimisation avenues were investigated in order to increase the yield of retinal cells. By investigating the impact of various dissociation reagents and enzymes it was found that, the dissociation of cells using TrypLE Express significantly increased retinal gene expression. Using this reagent resulted in over a 100 fold increase in expression of photoreceptor precursor marker Nrl in comparison to the mechanical dissection with Collagenase IV (the control). In this investigation it was also found that the dissociation of cells during the process, in order to reduce cell densities is highly detrimental. The investigations were successful in finding some changes to the published protocols that optimised the differentiation process of pluripotent Shef3 hESCs into retinal cells.

This thesis describes the use of ultra scale-down (USD) methods to address the primary recovery challenges posed by the processing of a novel, highly glycosylated Fc-fusion protein expressed in a Chinese Hamster Ovary (CHO) cell line. A feeding strategy coupled with an early cell culture harvesting time point delivered a broth whereby the Fc-fusion protein (~ 8.5 mg L-1)was minimally contaminated with "near-neighbour" variants of the glycosylated fusion protein. Such variants are not purified from the product during protein A recovery, making their reduced presence a critical feature of process success. However, USD shear studies demonstrated the need for low shear stress processing of the cell culture broth if the Fc-fusion protein was not to be contaminated by such variants due to intracellular release during cell removal. USD centrifugation studies showed that the processing of cell culture material using a continuous centrifuge equipped with low stress feed zones would avoid such contaminants. USD depth filtration, again a low stress operation,produced well-clarified process fluids without thecontamination of glycosylated variants. In addition, the charged filter material facilitated the reduction of host cell contaminants, such as DNA (reduced by ~ 50 %) and lipids (reduced by ~ 40 %), in the process fluidsto a greater extent than achieved by centrifugation. The predictionsfrom both centrifugation and filtrationstudies were successfully verifiedat pilot-scale. Depth filtration operated at constant flux rate was used as an alternative to constant pressure operation used in the USD studies. The preliminary analysis showed similarpredicted filter capacities provided the filtrate quality based on % solids remaining was the key design parameter. Again, the benefits of low shear stress operation were realised with no release of Fc-fusion protein contaminants. Finally, preliminary studies of packed beds of "big" beads for protein A affinity capture of the Fc-fusion protein directly from cell culture broth were carried out. Almost complete flow through of the cells was achieved when loading the cell culture into the packed beds and the majority of the Fc-fusion protein was recovered. Again, the characteristics of low shear stress operation were observed with no cell damage or release of glycosylated variants in the column flow-through fractions.

Tracheal abnormalities, congenital or acquired, represent a currently unmet clinical need. Tissue engineering has recently advanced and has been used to engineer hollow organs, including tracheae, for clinical use on a compassionate basis. However the current clinically used method for tracheal decellularisation has received mixed success and requires development to achieve GMP translation, quality manufacturing standards and ultimately routine clinical use. This thesis first examined the current clinically used detergent-enzymatic method (DEM) of decellularisation, a highly manual process that takes twenty-eight days to complete. Although the method achieved full decellularisation of non-cartilaginous regions of the tracheae, it failed to reduce the donor nuclear material sufficiently and resulted in the loss of key biochemical components, glycosaminoglycans (GAG) and collagen Type II, and the loss of biomechanical strength. A novel method for rapid, effective and non-detrimental tracheal decellularisation was required. Pressurised transmural flow was hypothesised to meet those requirements. A dual chamber bioreactor system was designed, fabricated and optimised to enable pressurised transmural decellularisation (PTD) to be investigated. Optimal PTD process parameters were ascertained and shown to produce tracheal scaffolds that achieved full decellularisation of the non-cartilaginous regions of the tracheae, a reduction of donor nuclear material (95%) which met the currently recommended levels of residual donor DNA for tissue engineered products, as well as maintaining GAG, collagen and biomechanical strength at comparable levels to the native tracheae. Additional to this, the new PTD process achieved this effective and non-detrimental decellularisation of tracheae in five working days with a ten-fold cost of goods reduction. With further development, the PTD method could become a fully automatable and closed process which could progress tissue engineered tracheae towards becoming a validated and regulated advanced therapy medicinal product (ATMP) and enable the advancement of tissue-engineered tracheae into regular clinical use.

This thesis, completed in collaboration with Purification Process Development of Pfizer Biotherapeutics, is concerned with how mechanistic models of chromatographic bioseparations can be applied in industry to accelerate development and increase robustness of industrial protein purification processes, whilst also realising the benefits of a systematic development approach based on fundamental process and product understanding. The first results chapter considers the application of mechanistic models to provide a link between high throughput screening (HTS) and scouting runs conducted during early process development. The chapter focuses on an anion exchange (AEX) weak partitioning chromatography (WPC) polishing step in a platform monoclonal antibody purification process. Adsorption isotherms are formulated from experimental multicomponent batch adsorption studies of monomer – aggregate. A new approach is taken where the adsorption equilibria is characterised as a function of the product partition coefficient, enabling the model to be applied to new candidate monoclonal antibodies without additional experimental effort. Stochastic simulations conducted at an early stage of process development identify promising operating parameter ranges for challenging separations, directs optimal performance, and reduces development times. A detailed analysis of model predictions increases fundamental knowledge and understanding of the complex WPC multidimensional design space, which enables better informed process development at Pfizer. Resin fouling over a chromatography columns lifetime can cause significant (undesired) changes in process performance. A lack of fundamental knowledge and mechanistic understanding of fouling in industrial bioseparations limits the application of mechanistic models in industry. An experimental investigation was conducted into fouling of the AEX WPC considered in the first results chapter. Analysis of fouled resin samples by batch uptake experiments, scanning electron microscopy, confocal laser scanning microscopy and scale down column studies, indicated significant blockage of the pores at the resin surface occurred that after successive batch cycles. Mass transport into resin particles was severely hindered, but saturation capacity was not affected. The increased understanding of resin fouling can direct future efforts to mitigate this detrimental phenomenon and maintain process performance, whilst providing a basis for the development of new fouling models. The third results chapter considers an industrial hydrophobic interaction chromatography (HIC) separation at a late stage of process development. Resin lot variability, combined with a variable feed stream, had resulted in serious performance issues during the purification of a therapeutic protein from crude feed material. The traditional approach to tackling this type of problem involves defining a design space based on an extensive experimental effort directed by factorial design of experiments conducted at great cost. The result is a fixed, inflexible manufacturing process, with a control strategy based on reproducibility rather than robustness, and little fundamental understanding of the source of the issue. In the third results chapter, the application of mechanistic models to identify robust operating conditions for the HIC is considered. A model is developed, validated experimentally, and used to generate probabilistic design spaces accounting the historical variability in the resin lots and load material. The stochastic simulation approach is extended to explore the impact of reducing variability in the load material on the design space. With historical process variability, no operating condition was found where the probability of meeting product quality specifications remained > 0.95 for all resin lots. Model simulations indicated that adopting an adaptive design space, where operating conditions are changed according to which resin lot is in use, is favorable for ensuring process robustness, which is a step change concept for bioprocessing. The conclusions and outcomes resulting from the application of mechanistic models to the two industrial systems in this thesis, provides a basis for the next generation purification process development platform.

The biotech community is under increasing pressure from the regulators to demonstrate effective implementation of the Quality by Design (QbD) principles and hence exhibit greater process understanding. Essential to achieving this goal is the implementation of a systematic framework that integrates experimental and predictive modeling techniques whilst accounting for the inherent variability of biologics. The aim of this thesis was to create a comprehensive framework for applying the principles of QbD throughout the lifecycle of a commercially distributed therapeutic protein produced in Chinese hamster ovary (CHO) cell culture. A systematic approach incorporating QbD principles was applied, involving risk assessment of potential process failure points, small-scale model development and qualification, design and execution of experiments, definition of operating parameter ranges and process performance acceptance criteria for validation and classification of operating parameters in the second generation process for this therapeutic protein. This methodology was illustrated through detailed process characterisation of a fed-batch production culture step and a virus inactivation step. One of the most prominent interactions found in the fedbatch production culture model was the three factor interaction of pH × temperature × seeding density. In the virus inactivation model, an increase in temperature (30 °C), incubation time (180 min), a reduction in pH (pH 3.5) at a relatively high protein concentration (5.5 g/L) resulted in a significant increase in the proportion of aggregated protein. Continuous process improvement and robustness are fundamental components in the QbD paradigm and are an integral component in the comprehensive framework described in this thesis. Following commercial implementation, Multivariate analysis (MVA) was employed to build upon the knowledge gained from process characterisation studies by identifying not only the process parameters but also the cell culture factors such as daily nutrient and metabolite levels which influenced protein titre and product quality attributes across scales. This analysis provided valuable knowledge for future troubleshooting and process refinement activities. Application of a statistical tool such as MVA adds scientific rigor to the data analysis thus improving the prediction of scale-up effects on the final product. Process robustness opportunities are often only realised when the process is run over multiple batches at manufacturing-scale. In this process, a perfusion culture which is employed to provide high cell densities to inoculate a production culture was identified as a potential risk to process and product supply and therefore identified as an area of focus for process optimisation activities. Application of a cell conditioning strategy enabled the generation of a seed fed-batch culture which achieved relatively high cell densities to inoculate the production culture. Shorter culture duration in the fed-batch culture also yielded comparable cell culture performance and productivity in production cultures relative to historical process performance. Successful commercial implementation and validation of the second generation process and the subsequent manufacture of hundreds of batches of this therapeutic protein verified the approaches described in this thesis as a suitable model for the development, scale-up and operation of any biopharmaceutical manufacturing process. This methodology is aligned with the QbD principles and can be applied to any biopharmaceutical manufacturing process, thus delivering sustainable, compliant and in-control processes throughout the product life-cycle.

Chiral amino alcohols are building blocks of several pharmaceuticals such as antibiotics, and retrovirals. De novo biocatalytic strategies have been developed for their production as chemical synthesis presents several challenges in establishing chiral centres. One of the successful approaches previously implemented in the Biochemical Engineering Department is the coupling of a transketolase with a transaminase step. This scheme depends on the use of hydroxypyruvate (HPA) and methylbenzylamine (MBA), which are expensive non-natural substrates limiting both the application of the pathway on an industrial scale, and its integration with the host metabolism. Therefore, this thesis examines the ability of two novel biocatalytic cascades to synthesize chiral amino alcohols from natural cheaper compounds, and the issues faced when attempting to integrate the pathways with Escherichia coli metabolism. Two different biocatalytic strategies were studied: a recycling cascade and a sequential pathway. The former, directly aminates erythrulose using serine as amino donor for the simultaneous synthesis of 2-amino-1,3,4-butanetriol (ABT) and HPA. The latter, involves three enzymatic steps carried out by two transaminases and one transketolase using serine, pyruvate and glycolaldehyde (GA) as substrates. Both coupled reactions depend on a transaminase able to use serine as amino donor. As a result, this work initiates with screening studies to identify the ideal candidates from a library of 100 transaminases. The selected enzymes were then characterized based on their main kinetic constants, KM and kcat, and their amino acceptors profiles when using serine as amino donor. From the results, two transaminases were selected for the proof of concept of the cascade reactions in a purified system. Due to the low bioconversions achieved, the work progressed towards their optimization implementing Response Surface Methodology (RSM) using a Central Composite Design (CCD). Finally, the optimized cascades were then evaluated in a whole cell environment using E. coli resting cells, exposing the challenges to be faced during their future metabolic integration. This study finalizes with a first attempt to design and build a more suitable expression system required for future work. Within this work, three new transaminases were described and characterized: two ω-transaminases isolated from Rhodobacter sphaeroides with broad amino acceptors profile using serine as amino donor; and a serine: pyruvate transaminase isolated from Deinococcus geothermalis with high substrates specificity. Both cascades were successfully tested, where the recycling system gave 30-times higher yields (30% conversion), and was more advantageous for the integration in comparison to the three-step sequential pathway. After the optimization, the yields of the recycling cascade and the sequential pathway were 2-fold and 3-fold improved, respectively. Finally based on the preliminary whole cell study of the recycling cascade and the first two steps of the sequential pathway, it became clear that the main challenges for the future metabolic integration are HPA consumption by the host, and the differential effect of the expression of different recombinant enzymes on the E. coli host cells.

The biopharmaceutical industry is at a pivotal moment of its short history, with ‘blockbuster’ monoclonal antibody (mAb) biosimilars emerging and the ‘next generation’ of molecules beginning to make their mark. Concomitantly, regulatory constraints on manufacturers have amplified, with processes now requiring in-depth characterisation. Scale-down mimics of process steps and supportive analytical methods are therefore becoming indispensable. The work presented in this thesis has focussed on the downstream processing of mammalian cell derived mAbs. With industrial relevance considered throughout, a scale-down primary recovery and chromatography process was implemented. This was supported by ELISA, mass spectrometry and gel-based methods to investigate host cell proteins (HCPs), which are rigorously controlled process-related impurities. The interface of upstream and downstream processing was identified as a critical aspect of bioprocessing which can dictate impurity profiles. Preliminary work demonstrated that small particle debris and cell properties, such as viability and size, all contribute to primary recovery performance. Subsequent investigations of mild hypothermic cell culture showed that cell membrane rigidity also plays an important role, as centrifugation clarification was up to 4% greater for temperature adapted cells. However, HCPs were adversely affected, increasing by ~50%. To further understand harvest material properties, a polyethylenimine flocculation process was developed. This improved centrifugation performance, by reducing shear susceptibility, and increased subsequent depth filtration capacity up to 2.1 fold. Benefits to purification were also demonstrated, with DNA and negatively charged HCPs being removed. Protein A chromatography was the final focus of work, owing to its prominence in purification. Four different resins were evaluated; non-specific adsorption of HCPs was only pronounced for the resin with a controlled pore glass back-bone, but ‘product association’ was significant for all. To help analyse this sub-set of HCPs, an immuno association disruption methodology was developed and may prove invaluable in future characterisation efforts to improve product purity.

This thesis illustrates the need to consider overall process performance through the use of biosensors, scale-down and low-volume analytical technologies to achieve processes that account for cell engineering, upstream processing and its effects on downstream processing performance; the upstream/downstream interface. An Escherichia coli fed-batch fermentation expressing antibody fragment (Fab’) was used as a system representative of current manufacturing challenges. Sensor surface characterisation using dual polarisation interferometry (DPI) was explored and a step-wise fabrication and characterisation of a multi-layer DPI based biosensor was carried out. A chemical characterisation of the sensor surface was performed ex-situ using x-ray photoelectron spectroscopy (XPS) and secondary ion mass spectrometry (SIMS) to investigate the chemical identity and homogeneity of the chemical layers. The studies uncovered UV/ozone cleaning as the method of choice in comparison with surface acid washing and the use of hydrophilic surfaces as optimal for antibody recognition in comparison to hydrophobic surfaces. The studies highlighted the use of DPI as a sensor characterisation tool. Biological characterisation assessed the potential of the sensor as a monitoring tool. Following characterisation, method development ensued, to quantify antibody fragments via interaction with protein G. The surfaces were assessed through non-specific binding studies and challenged with complex fermentation materials. The characterised biosensors were utilised to assess Fab’ product loss and Fab’ quality during bioprocessing. The biosensor was shown to be sensitive down to 1.7 µg/mL of Fab’ with low interference due to non-specific binding (NSB) and good percentage recovery in 4-fold diluted extracellular broth. The sensor surfaces were then used in conjunction with scale-down tools for primary recovery and low-volume analytical technologies to obtain insights into the interaction between cells from the upstream process and downstream processing operations. For the process studied, the net yield at the end of the primary recovery, which takes into account both product losses during fermentation and centrifugation, was the critical parameter due to the intercellular nature of the product. Studies showed that 30 h was the optimal harvest time with regard to product loss. To benefit from increased product titres at 40 h and above requires that cell strength and integrity be improved. The effect of cell engineering on the growth and productivity of the upstream process as well as its effect on cell integrity at harvest was conducted. This was conducted through the assessment of the growth and productivity impacts of periplasmic nuclease expression in an Escherichia coli Fab' fragment production strain. This modification allows the removal of DNA upon cell disruption. The Co-expression of staphylococcal nuclease and antibody product in the same sub-cellular compartment showed no impact on the growth characteristics and Fab’ yield of the host strain. However, introduction of the periplasmic nuclease did perturb kinetics of Fab’ production; increasing the rate of intracellular Fab’ accumulation and concomitantly increasing in the rate of appearance of Fab’ in the external medium. This thesis highlights the need to consider the effects of upstream processing on primary recovery and shows how harvest point determination is a critical process parameter to assess during the design of a bioprocess.

Despite their huge potential as therapeutic agents, protein molecules are not naturally optimised for stability and are susceptible to aggregation during bioprocessing. A good understanding the mechanisms that drive protein aggregation is therefore important to enable the development of strategies for eliminating protein aggregation bioprocessing. To better understand the factors affecting aggregation, a detailed kinetic and biophysical analysis of the Af-X molecule when subjected to a range of pH and NaCl concentrations was carried out. At pH 4.0, a salt-induced dimer was found to accumulate at low NaCl concentrations, but less so at higher salt where the rate of protein precipitation increased. Kinetic studies showed that this salt-induced dimer is not an intermediate on the Af-X precipitation pathway. Surface plasmon resonance, circular dichroism, and fluorescence spectroscopy highlighted the biological inactivity of the salt-induced dimer and confirmed its structural and functional similarity to a previously observed cell culture dimer. Analysis of the crystal structures of both dimers confirmed the biophysical characterisation. The post induction time and cellular location of Af-X dimer formation, as well as the effect of IPTG concentration, temperature, and post-induction feed rate on the propensity for dimerisation of the Af-X molecule, were investigated using fed-batch fermentation. While a decrease in both the IPTG concentration and the post-induction feed rate resulted in a 2% increase and a 5% decrease (respectively) in dimer content of the total Af-X yield, a reduction in the post-induction temperature resulted in a 12% increase in the dimer content of the total Af-X yield. This unexpected increase in dimer concentration was found to agree with the proposed effect of temperature on the Af-X mechanism of aggregation. Site-directed mutagenesis was used to generate L-XX-KVRST single and H-XX-Y/L-XX-KVRST double mutants that were evaluated for their propensity for aggregation both in solution and during E. coli cell culture. While the L-XX-K single mutant was found to be destabilising to the Af-X molecule, the L-XX-R single mutant was found to be stabilising. These contrasting observations can be explained by taking into account possible physical distance limitations or angular orientation of the lysine side chain that is less than optimal for salt bridge formation with the aspartic acid residue at position XX. Finally, samples of Af-X were incubated for 8 weeks at 25 °C without shaking, and compared to samples incubated for 6 days at 25 °C with shaking in order to establish a correlation between agitation and the long term stability studies commonly used in formulation development. The effect of ultrasonication on the propensity for aggregation of the Af-X molecule was also assessed using a time-course experiment. Both agitation and sonication had negligible effects on the propensity for Af-X dimerisation. In addition to providing useful insights into aggregation in the Af-X molecule, this work will serve as a starting point for future investigation of aggregation in antibody fragments in general. A number of potential future directions that build on these insights are also discussed.