Accurately characterising the design space of a process is critical for the development of robust and economic purification processes. However, this can require extensive experimentation which can be expensive and time consuming. Furthermore, this experimentation may require amounts of material that is not available during the early stages of process development, which can lead to delays in the development of the biopharmaceutical product. An approach that can ease the burden of characterisation is the use of scale-down methods and devices. Scale-down experiments require significantly smaller amounts of material and can be automated to improve throughput. Unfortunately various issues associated with scale-down experimentation limit their usefulness. One such issue is that various sources of variability can introduce noise into the results of these experiments. In this study Monte Carlo simulations were carried out in conjunction with a mechanistic general rate model of a typical ion exchange chromatographic purification to investigate the impact of variability on the results of these experiments. A comparison of alternative process modelling approaches was carried out to determine which was capable of producing the most accurate process models from noisy data. A Kriging algorithm was found to be the most capable technique. There are also various scaling effects that cause the scale down experiments to not be representative of the large scale process. Two approaches were developed that augmented extensive scale down experimentation with a small number of large scale experiments to produce representative models of the large scale design space. One of the approaches used derived transformation functions to transform the scale-down data into the large scale design space and the other used a Cokriging algorithm. Using the cation exchange chromatography purification of myoglobin from egg white proteins as a test platform it was shown that it was possible to produce accurate characterisations of the large scale process using only a fraction of the material that would be otherwise required for carrying out the experimentation at large scale. These approaches were further tested using a purification sequence consisting of a heat treatment step followed by a cation exchange chromatography step for the purification of an antibody fragment from E. coli lysate. In this study fractionation diagrams were adapted to describe the elution profiles of the product and its various impurities to enable the multi-objective optimisation of the process. It was demonstrated that this approach could be used to produce detailed characterisations that revealed the relationships between the process variables and multiple responses across the whole process sequence.

With the emerging interest in integrated biorefinery concepts, there is a need to identify and develop profitable product streams and ensure the utilisation of as many waste streams as possible. Early stage bioprocess development for these processes can be facilitated by the use of high throughput bioreactor platforms that enables rapid, quantitative and scalable data acquisition. This thesis aims to establish high throughput methodologies for the production and characterisation of industrial biocatalysts within an integrated biorefinery context. Specifically, the work focuses on the production of the CV2025 ω-Transaminase (CV2025 ω-TAm) in Escherichia coli BL21 (DE3) using sugar beet vinasse, a bioethanol waste stream, as a fermentation feedstock. The high throughput platform to be explored is a 24-well, controlled microbioreactor (MBR) that provides individual monitoring and control of process parameters at the well level. Initially, batch E. coli BL21 (DE3) fermentations expressing CV2025 ω-TAm were established in the controlled MBR using a synthetic medium to provide benchmark data on cell growth and enzyme expression. These cultures indicated a good degree of monitoring and control with respect to process parameters as well as culture reproducibility across the wells. Significant enhancements in relation to maximum biomass concentration (Xmax), yield of biomass on substrate (YX/S) and CV2025 ω-TAm specific activity of 3.7, 1.9 and 2.2-fold, respectively, were shown in the MBR compared to conventional shake flask system, also representing a 31-fold volumetric reduction. Optimisation of CV2025 ω-TAm production in the MBR showed that the best cell growth and enzyme titre was achieved with an early induction (6 h), 0.1 mM IPTG and 0.024 mmol IPTG gdcw-1, yielding enhancements in Xmax, YX/S and CV2025 ω-TAm specific activity of 1.04, 1.2 and 1.4-fold, respectively over the non-optimised cultures. Control of dissolved oxygen (DO) levels between 30 - 50% oxygen saturation had no significant impact on cell growth and CV2025 ω-TAm titre. Evaluation of vinasse as a fermentation feedstock for CV2025 ω-TAm production has led to several novel findings. Characterisation of vinasse showed that the feedstock comprised mainly of glycerol along with several reducing sugars, sugar alcohols, acetate, polyphenols and protein. Preliminary results showed E. coli BL21 (DE3) cell growth and CV2025 ω-TAm production were feasible in cultures using 17 to 25% (v/v) vinasse with higher concentrations demonstrating inhibitory effects. The D-galactose in vinasse was shown to facilitate auto-induction of the pQR801 plasmid leading to comparable CV2025 ω-TAm expression as obtained in IPTG-induced cultures. Assessment of different vinasse pre-processing options confirmed the relevance of the dilution step in reducing polyphenol concentrations to below inhibitory levels. Moreover, the use of pasteurised vinasse was found to be promising for large scale applications. Further medium optimisation studies in the MBR showed the benefit of supplementing vinasse with specific media components. Supplementation of diluted vinasse medium with 10 g L-1 yeast extract enabled enhancements of 2.8, 2.5, 5.4 and 3-fold in specific growth rate, Xmax, CV2025 ω-TAm volumetric and specific activity, respectively, over those achieved in non-supplemented cultures. Additionally, the CV2025 ω-TAm titre attained here represented 81% of that obtained using an optimised synthetic medium. Investigation into the metabolic preferences of E. coli BL21 (DE3) when grown on a complex carbon source like vinasse showed the sequential metabolism of D-mannitol before glycerol utilisation, which was followed by the simultaneous metabolism of glycerol, D-xylitol, D-dulcitol and acetate thereafter. Finally, scale-up of the optimal conditions for CV2025 ω-TAm production using both synthetic and vinasse-based media, from the controlled MBR to a 7.5 L stirred tank reactor (STR) was shown based on matched kLa values and specific aeration rates. Results showed a good reproducibility with respect to cell growth, substrate consumption and CV2025 ω-TAm production between the scales, representing a 769-fold volumetric scale translation. The feasibility of further intensification of CV2025 ω-TAm production in STR at higher kLa values using both synthetic and vinasse-based media was also demonstrated leading to enhancements of 1.4 and 1.9-fold in enzyme titre, respectively. Overall, this work has established high throughput methodologies for the characterisation, optimisation and scale-up of industrial biocatalyst production. The approach was demonstrated within the context of an integrated sugar beet biorefinery. However, the utility of the high throughput approach is considered generally applicable across the industrial biotechnology sector.

The high level of innovation in drug discovery in recent years has presented a significant challenge for drug manufacturing process development, which must constantly evolve to meet this increasingly diverse demand. As a result, novel process monitoring technologies to rapidly optimise these processes, reduce development costs and improve time to market are in high demand in the biopharmaceutical industry. Within the context of bioprocess development and manufacturing, the main focus of this work is on fermentation and its impact on, and interaction with, primary recovery. Although E. coli is the most widely researched host organism for recombinant protein production and cell death during fermentation has been observed for decades, very little is understood about how to quantify and detect cell lysis in late stage fermentation, which leads to a number of problems in the downstream process such as product loss and poor operational performance. The complex nature of the cell broth means that it is difficult to observe lysis directly, and current analytical technologies are unable to rapidly and accurately monitor the shift between optimum intracellular product concentration and leakage to the cell broth. This thesis proposes that by monitoring the physical properties of the cell broth, i.e. by monitoring the viscosity, it may be possible to indirectly infer cell lysis, as the release of intracellular content, such as host cell protein and nucleic acids, to the cell broth at the point of lysis are known to cause an increase in the broth viscosity. In this thesis, cell lysis was first characterised in an industrially relevant E. coli fermentation producing antibody fragments (Fab'), using a range of common analytical techniques. Following this, a method has been developed to rapidly detect cell lysis and product loss using at-line viscosity monitoring, and a strong correlation was shown to exist between DNA release, product leakage, cytotoxicity and viscosity. Viscosity monitoring to detect cell lysis was shown to perform better than optical density measurements and online capacitance probes, and could detect lysis faster than HPLC, flow cytometry, cytotoxicity assays and DNA quantification. Subsequently, a model has been developed to quantify cell lysis using rapid viscosity monitoring. Viscoelasticity studies have also been performed to provide novel insight into changes in cell strength during fermentation. Finally, a case study has been carried out to demonstrate an application of viscosity monitoring in process development, and to enable insight into the impact of upstream processing conditions on the efficiency of downstream unit operations. A novel process design using crossflow filtration and flocculation achieved a 2.53-fold improvement in total product recovered, a 3-fold improvement in solids removal and a 3.6-fold improvement in product purity, in comparison to the existing Fab' primary recovery process. This work presents the novel use of viscosity monitoring in biopharmaceutical fermentation to rapidly detect cell lysis and product loss. In doing so, a deeper understanding of changes in the physical properties of cell broths during fermentation has been obtained, as well as insight into the impact of lysis on various primary recovery unit operations. The use of viscosity monitoring to rapidly detect lysis and product loss has been shown to be a promising analytical tool to enable optimisation in process development and facilitate harvest decision making for large scale operation.

The potential application of olfactory ensheathing cells (OECs) to spinal cord injury has been the focus of a lot of research over the past ten years. Currently there are many challenges associated with the use of these cells as a therapy. These include the inherent plasticity of the cells and the fact they are challenging to sustain in culture for prolonged periods due to poor survival and proliferation. They possess unique properties as they are able to support neuronal survival and facilitate the regeneration of severed axons and therefore overcoming these challenges would be a step towards developing a cell therapy for spinal cord injury. Initially rat models were used to determine culture conditions that enhance the protein expression of key OEC markers p75NTR and S100β. An increase in p75NTR expression was achieved by co-culturing the primary rat OECs with a conditionally immortalised immobilised human mucosal fibroblast cell line feeder layer. OECs cultured on feeders were found to adopt a more spindle-like appearance compared with cells cultured on laminin which adopted an enlarged morphology. This morphological change is significant as spindle-shaped OECs are associated with neural regeneration function. Conditioned media collected from the human feeders resulted in an increase in Thy1.1 protein expression, an undesirable marker, with no increase in p75NTR expression and co-culture of primary OECs with mouse feeders (Ms3T3) gave similar results to co-culture with human feeders. It was determined that OECs benefit from the cell to cell contact and not necessarily trophic factors present in the media. The best culture conditions for primary rat OECs were found to be Ms3T3 feeders with DMEM/F12 Glutamax media. Testing with conditionally immortalised human OEC cell lines found contrasting results where an increase in S100β was observed when cells were cultured on laminin and lower levels of expression observed during feeder co-culture. Similarly to primary rat OECs, conditioned media from human mucosal fibroblasts was detrimental to S100β expression. The best culture conditions for human OECs were found to be laminin coated wells with DMEM/F12 Glutamax media. These data sets show that care has to be taken when translating animal models to studies with human cells as the data does not always correlate. Studies continued to characterise optimum culture conditions for the conditionally immortalised human OEC cell line. MACS purification technology was used to remove Thy1 positive cells from the polyclonal population. Although this removal was successful, it was found that the complete removal of Thy1 from the population does not ensure Thy1 is not present in the future population. After 5 days in culture Thy1 was being expressed in the Thy1 negative population. Time point staining determined that Thy1 turns on and off during time in culture and the removal of Triton X from the staining protocol is vital to visualising the presence of this protein. This time point study also revealed p75NTR is not a stable marker for OECs as turns off over time in culture. Further study towards understanding the role Thy1 and p75NTR take in OEC function would be beneficial to development of an OEC cell therapy for spinal cord injury. After the identification of optimum culture conditions for enhanced p75NTR and S100β expression, co-culture with neurons was carried out in order to determine if these conditions would link to an improvement in functional support. Neurite length was measured after 5 days of co-culture and was normalised against the number of neurites and neurons, which is an established method of relating the behaviour of the neurons to functional response after implantation. It was found that conditions that related to higher expression of p75NTR and S100β (laminin coated wells, standard media, shorter time in culture) led to longer and more numerous neurites. From this it can be established that levels of p75NTR and S100β expression are good predictive tools for the extent to which OECs can support neural regeneration in culture. The next step would be to relate the expression of these markers to the myelination of neurons. Thy1 expression was not found to be related to neurite extension and purified populations of negative and positive Thy1 OECs resulted in longer neurites than the original mixed population. This could be due to lateral inhibition but further work is required to confirm this theory. Results described in this thesis have demonstrated that caution needs to be applied when scaling rat studies to human cell work. It has also shown that the method and timing of detecting protein expression can be vital to the results observed. These are key aspects that need to be considered in order to fully characterise cell populations, especially one as variable as OECs. The methods used in this work showed an increase in p75NTR and S100β expression led to longer neurites extended from neurons. Further work should be carried out in order to fully understand the interaction between OECs and neurons and to explain the potential lateral inhibition pattern observed.

Design of experiments (DoE) is used in process development to optimise the operating conditions of unit operations in a cost-effective and time-saving manner. Along with high throughput technologies, the modern high throughput process development lab can turnover a tremendous amount of data with minimal feedstock. These benefits are most useful when applied to the purification bottleneck, which accounts for up to 80% of the total process operating costs. However due to complexities of biochemical reactions and the large number interacting factors in unit operations (which usually cross interact with each other), even carefully planned DoE experiments on high throughput platforms can become difficult to manage and/or not provide useful information. This thesis examines the simplex search method and develops a set of protocols for use of the search method in combination with traditional DoE experimental design protocols. It is that is demonstrated in the developed in chapter 3 whilst also optimising a ammonium sulphate based precipitation step of an industrially relevant feedstock. Comparisons were drawn between a high resolution brute force study, a response surface DoE, the simplex method and then a combination of DoE and the simplex method. Various strategies were demonstrated that get the most out of the simplex method and mitigate against potential pitfalls. The precipitation step was optimised for yield and purity over the 3 factors, pH, ammonium sulphate concentration and initial MAb concentration and the results showed the simplex method was capable of rapidly identifying the optimum conditions in a very large 3 factor design space on an average of 18 experiments. The expansive study not only served as a testing ground for the methods comparison but demonstrated precipitation as a high throughput, low cost substitute for the expensive Protein A step. The DoE –simplex search protocols are then refined in two complex case studies in chapter 4, a PEG precipitation primary capture step and an ammonium sulphate precipitation and centrifugation sequence. The five factor precipitation and centrifugation sequence was especially complicated and utilised ultrascale down models provide accurate scale up data. This involved calibrating an acoustic device to provide shear treatment to the precipitate pre-centrifugation and using jet mixing equations to correlate precipitate conditioning between the TTecan robot’s tips and an impeller in a stirred tank. The techniques developed were all applicable to microscale and high throughput. In both instances, the combined DoE-simplex approach retuned superior results both in terms of experimental savings and generating information-rich data from the final local regions DoE around the simplex located optimums. A microscale chromatography protocol was developed on the Tecan liquid handling robot and demonstrated on screening work with different Protein A and cation exchange media. The caveats encountered when creating the running methods and the analytical methods supporting it for the Atoll robocolumns were highlighted and mitigation solutions implemented. The automated microscale Protein A method was successfully scaled up 50x from a 200 μL robocolumn to a conventional 10 mL labscale column. After selecting a cation exchange resin for developing an aggregate removal step, the DoE-simplex methodology was applied to an antibody product with an extremely high aggregate level and a comparison optimisation was made with a central composite design DoE. The difficult four factor design space overwhelmed the DoE and having used more experiment numbers than the DoE-simplex methodology, only went as far to show the high levels of curvature in the system and offer a poor prediction of the surface. The DoE-simplex methodology was able to provide a general model of the whole surface from the DoE, locate the optimum with the simplex in fewer experiment numbers. This subsequently allowed a local DoE to be applied to the optimum region to determine a robust operating range for the cation exchange step.

There is growing interest in the use of lentiviral vectors, particularly for cancer immunotherapy and the treatment of monogenic diseases. Manufacturing of these vectors is challenging primarily due to cytotoxic effects of vector components resulting in low cell culture titres and vector instability leading to low purification yields. In addition, currently used processes are typically not scalable as they rely on adherently cultured cells and unit operations such as batch centrifugation and gel filtration. To improve process scalability, suspension adaptation of a lentiviral vector packaging cell line was attempted, however, cell aggregation could not be prevented. For vector clarification it was found that membranes with pore sizes of 0.22 µm resulted in recoveries less than 50%, whereas the use of 0.45 µm membranes resulted in recoveries close to 100%. Successful vector concentration utilising centrifugal filters was possible with a membrane molecular weight cut-off (MWCO) of 100 kDa, whereas a 300 kDa MWCO led to low recoveries. Chromatography stationary phases that allow convective mass transfer, such as membranes and monoliths, are becoming increasingly popular for purification of large molecules. Lentiviral vector was found to bind monoliths with weak and strong anion exchangers over a wide range of conditions. Vector elution conditions determined for membrane- and monolith-based resins in high-throughput 96-well plate format were found to not be indicative of gradient elution conditions for 1 mL versions of these resins operated by a chromatography system. The thermolability of lentiviral vectors leads to a requirement for storage at less than -65°C. Inexpensive mixtures of sugars in combination with a glycine-derivative were studied for their ability to stabilise a lentiviral vector during freeze-drying and subsequent thermochallenge. The amorphous solid formed upon freeze-drying was able to stabilise the vector for up to 12 weeks at 4°C and eight weeks at 25°C. This is in comparison to formulation in phosphate buffered saline, where more than 90% of infectious titre was lost immediately upon freeze-drying.

Manufacturing of large biomolecules such as viral vectors used in emerging gene therapies suffers from low product yields due to limitations of traditional resin-based chromatography in downstream processing. Ultrafiltration based purification techniques are being considered for purification of such viral vectors by separating these viral vectors from impurities such as host cell proteins. Improvements in yields of viral vectors have so far concentrated on the design of new chromatography stationary phases such as monoliths, membrane adsorbers, nanofibers and gigaporous resins. Improvement in the architecture of ultrafiltration membranes has not been studied for application in viral vector purifications. Porous anodic alumina (PAA) membranes, exhibit narrow pore size distribution and straight through pore channels compared to traditional polymeric membranes which have broad pore size distributions and tortuous channels. The present work evaluated porous anodic alumina membranes for potential applications in virus ultrafiltration using model protein solutes. Protein nanoparticles of 80-90nm diameter and thyroglobulin of 20 nm diameter were used as the physical mimics for two of the most commonly used viral vectors, Adenovirus and Adeno-associated viruses respectively and a small protein of 8 nm, bovine serum albumin was used as the model impurity. A reproducible and high yielding protocol was developed for the synthesis of protein nanoparticles from bovine serum albumin using a de-solvation process. Based on comparable hydraulic permeability, dextran sieving curve and mean pore size 20 nm rated PAA membrane and 300 kDa rated polymeric ultrafiltration membranes were compared for filterability of the model solutes. PAA membranes were found to have superior fouling resistance (1.5-2.5 times higher values of recoverable membrane permeability) and up to 4 times higher transmission than the polymeric membranes for large model solutes. These findings were attributed to the differences in the membrane architecture resulting in different sieving behaviour. PAA membranes were found to be susceptible to leaky transmissions of large solutes due to the presence of surface defects. Separation performance of binary mixtures of model solutes was studied using a diafiltration process. Electrostatic interactions and transmembrane pressure were identified as crucial process parameters to improve separation performance of the alumina membranes. Lot-to-lot variations in the alumina membranes were also characterised using electron microscopy and were found to influence the separation performance. PAA membranes were found to be compatible for virus processing as similar infectivity recovery of 60-70% was observed for both PAA and the polymeric membranes along with 60 % removal of the impurities.

Bioprocess development is very labour intensive, requiring many experiments to characterize each unit operation in a process sequence to achieve product safety and process efficiency. Recent advances in microscale biochemical engineering have led to automated high throughput experimentations. The activities for bioprocess development are implemented sequentially in which 1) liquid handling system performs the wet lab experiments; 2) standalone analytical devices detect the data; and 3) specific software is used for data analysis and experiment design. The experiment design, data analysis and process understanding require substantial engineer’s time. It becomes one of the time consuming bottlenecks in bioprocess development particularly when a high number of experiments are needed to explore a large design space in order to discover high process performance. This thesis addresses this challenge by reducing the development time yet achieving high process efficiency. A closed-loop learning approach has been adopted in bioprocess design in which the design objectives are driven iteratively by intelligent data acquisition. A framework that brings all of the elements performed manually into an automated fashion has been established to deliver the closed-loop learning approach. Based on the framework, a novel prototype of Intelligent Automation Platform for Bioprocess Development (IAPBD) has been built using multi-agent architecture. A rational database and four agents including Coordinate Agent, Experiment Design Agent, Execution Agent and Assay Agent have been designed to perform individual tasks and worked as a team to deliver the bioprocess design objectives. The multi-agent architecture used a blackboard mechanism to connect the four agents so that they are able to communicate with each other during operations. Starting with a set of initial experiments from the user or database, IAPBD is able to drive the robotic arm to perform defined initial experiments, and drive the analytical devices to detect the data after the experiments are completed. Then it will pick up the data from the analytical devices, and carry out data processing and process evaluation based on the optimization objective to achieve the design solution. After evaluation of the design solution, it will decide to stop or continue to design the next round experiments for optimization. The prototype has been evaluated first by a lysozyme precipitation process design, which involves typical microscale experimental procedures such as mixing, shaking, sample preparation and high throughput data detection instruments. All of the devices used have programmable interface so agents can control them directly. An optimal design solution that maximizes the yield of lysozyme and maximizes the ammonium sulphate concentration was found within a few of iterations using simplex search algorithm or Artificial Neural Network (ANN) and the whole tasks were completed automatically without human intervention. The success of this case study proved the concept of IAPBD. The second case study was designed to further evaluate the prototype by the precipitation for monoclonal antibody purification. A new “watcher” algorithm in the Assay Agent has been further developed to communicate with instruments that do not have programmable interface e.g. HPLC. IAPBD carried out two sets of precipitation experiments automatically. The “watcher” algorithm has been proved robust and efficient to retrieve the data from HPLC. The third case study is designed to use the sequential Design of Experiment method to optimize the production of a whole cell biocatalyst in fermentation. An optimal solution for medium composition and operating conditions was found successfully. It was then confirmed by large scale fermentation experiment that the optimal solution has increased the biomass/product more than 5 folds. The benefit of this novel IAPBD is its automation that reduces the bioprocess design time significantly and frees engineer’s time for other intellectual tasks. Prove-of- concept of IAPBD has been achieved with limited real experimental evaluation. With further development and evaluation, its potential to significantly reduce the time and cost of the whole bioprocess development may be realized.

The direct ω-oxyfunctionalisation of aliphatic alkanes in a regio- and chemoselective manner remains difficult to perform by industrial organic chemistry. Monooxygenases such as the AlkBGT enzyme complex from Pseudomonas putida efficiently catalyse these readily available substrates to fatty alcohols, aldehydes and acids under mild conditions. However, numerous challenges remain to achieve industrially competitive space-time-yields for bio-oxidations. The ability to rapidly screen bioconversion reactions for characterisation and optimisation is of major importance in bioprocess development and biocatalyst selection; studies at lab scale are time consuming and labour intensive with low experimental throughput. This study developed and validated a robust high-throughput microwell platform for whole-cell two-liquid phase bio-oxidations of highly volatile alkanes. Using microwell plates machined from polytetrafluoroethylene and a sealing clamp, highly reproducible results were achieved with no significant variability such as edge effects determined. Further, the developed platform was extended by a fed-batch implementation revealing the large impact of feeding conditions on the resting cell bio-oxidation of volatile alkanes. The unpredictable nature and large differences between varying alkane substrates show the importance of being able to test fed-batch conditions early in development. Lastly, six co-solvents were screened to relieve organic phase toxicity and improve control over the product spectrum in the bio-oxidation of hydrocarbons. In combination with computational and statistical tools, it was shown that polar surfactants allow the extraction of the alcohol product, increasing the alcohol yield and reducing phase toxicity. Specifically, the solubility of co-solvents reaction substrate and product was revealed to be the determining factor for product selectivity. Overall, the developed microwell plate greatly improves experimental throughput, accelerating the screening procedures specifically for biocatalytic processes in non-conventional media. Its simplicity, robustness and standardisation ensure high reliability of results. The accelerated data collection on biological as well as process options allows obtaining key process design data early on, de-risking and speeding up the translation of new processes from laboratory to pilot-plant scale.

Vaccinia virus Lister strain, a large, structurally complex double stranded DNA pox virus, is being developed by a number of organisations around the world as an oncolytic and immunotherapeutic agent for the treatment of a broad range of cancers. Should these therapies prove to be efficacious in the clinic, large quantities of vaccinia virus will need to be produced at very high levels of purity as dosing requirements are expected to be as high as approximately 1x109 pfu/dose. In this thesis, the development of two convective interaction media (CIM) monolith capture steps for vaccinia virus with considerable purification factors is described; one uses cation exchange while the other uses hydrophobic interaction chromatography. The purification process development involved an extensive material characterisation study resulting in enhanced product understanding, a rapid resin screening study aimed at quickly identifying suitable resin chemistries, followed by process optimisation studies on the best performing monoliths. After being challenged with crude infectious vaccinia harvest, CIM OH monoliths are shown to be able to recover up to 90% of the infectious virus loaded whilst removing up to 99% of the contaminating DNA (without nuclease treatment) and 100% of quantifiable protein. Binding capacities were shown to be in the order of 1x109 pfu/mL. The high levels of both batch to batch and assay variability as well as the tendency of vaccinia virus to aggregate in the feed material, typical of viral processes especially when developed alongside un-optimised upstream conditions, are clearly demonstrated and the implications are explored. The results show that it can be challenging to draw robust conclusions on process performance. To minimise the effects of analytical variability, a number of orthogonal analytical methods have been used to quantify and characterise viral particles. These include TCID50, nanoparticle tracking analysis (NTA), tunable resistive pulse sensing (TRPS), transmission electron microscopy (TEM), scanning electron microscopy (SEM), and real-time PCR (qPCR). In order to put this into an industrial context, a comparative cost of goods analysis of monoliths and ultracentrifugation technologies for the purification of large viral particles is provided. This shows that both chromatography using monoliths and continuous flow ultracentrifugation can be economically viable, although both have limitations. The potential economic benefits of using a monolith-based process over an ultracentrifugation-based process are increased productivity, the ability to generate purer material and ease of scale-up. CIM monoliths are unique stationary phases that offer efficient separation and high productivity owing to fast cycle times and high binding capacities. Both cation exchange (CIM SO3) and hydrophobic interaction (CIM OH) monoliths are effectual at removing the majority of contaminants in a single purification step that can easily be scaled up to 8 L bed volumes. CIM monoliths have the potential to be an attractive option for future manufacturing processes for oncolytic viral therapies. They are shown in this thesis to achieve a higher percentage recovery and better removal of DNA, protein and aggregate than any other technology described in the literature to date.