Bioprocess design requires substantial resources for the required experimental investigation of the options for each bioprocess step. With the aim of reducing the amount of experimentation needed for bioprocess development, a new computational framework called Bioprocess Data and Knowledge Framework (BDKF) has been developed to explore the data and knowledge systematically. In BDKF, the representation of four types of data and knowledge i.e. experimental data, ontologies, theoretical knowledge and empirical knowledge, have been established. The experimental data is the data that comes from previous experiments. The ontologies are the systematic description of the bioprocess terminologies used in the experimental data and knowledge. It can organize the terminologies of a domain as a hierarchy that allows the experimental data to be searched. The theoretical knowledge is the knowledge represented by formal definitions in the bioprocess, such as fundamental equations. The empirical knowledge is the knowledge obtained from practical studies, e.g. the relationships between different scales established through ultra scale-down experimentation. Three reasoning functionalities, search, prediction and suggestion, have been established to imitate human reasoning on using data and knowledge. The search functionality finds relevant experimental data to the bioprocess design problems. With this data, the prediction functionality analyses the data and estimates the possible performance of the bioprocess step. The suggestion functionality produces solutions for further experiments that either confirm the solutions or narrow down the design space. A prototype applying the BDKF approach to illustrate how to capture data and knowledge and how reasoning functionalities work for the operating conditions identification was developed for a case study on centrifugation. Design queries that represented relevant process material information and separation requirements were generated to initiate the BDKF approach. The prototype demonstrated that data from strain variants and data from different scales can be utilized through ontologies, theoretical knowledge and empirical knowledge. A more complicated prototype was developed for the chromatography case study. The prototype introduced a hierarchical heuristic approach to solve the chromatographic process design problems, such as column selection, buffer composition identification and operating conditions determination. This prototype demonstrated that BDKF can be used for both screening and optimisation to propose several potential bioprocess solutions. Evaluation results of each prototype showed that the BDKF approach can make good performance predictions and suggestions for further experiments. It is very promising as an early stage process development tool. Finally, a method for finding a design solution for a giving sequence by using mass balance calculations has been developed. A case study including centrifugation, filtration and chromatography has been examined. This demonstrated that BDKE method had the potential to allow all of the data and knowledge to be used for the whole bioprocess design. Therefore, the BDKF approach can provide a systematic way to harness bioprocess data and knowledge to enhance the efficiency of bioprocess development.

This thesis investigates bioprocess development approaches for the precipitation-based purification of IgG monoclonal antibodies (mAb) from clarified cell culture supernatant. A high throughput platform using microgram quantities of IgG was developed for the rapid identification and selection of suitable conditions for precipitation of IgG from a complex feedstock. The most effective conditions were then assessed in an ultra scale-down (USD) device run with 60 mL of material in order to characterise the precipitates in terms of their particle size distribution (PSD), morphology, strength and density with the aim of predicting likely precipitation and centrifugation separation performance upon scale-up. The influence of the physiochemical environment on precipitation was investigated and the effectiveness of the USD device in predicting laboratory scale and pilot scale performance was also evaluated. A high throughput micro-scale approach conducted in a 96 microwell plate format was developed for scouting favourable precipitation conditions. Optical measurements were used to locate suitable conditions. Turbidity was measured by absorbance at 600 nm as a rapid measure of the solids content and to monitor the progress of precipitation. Coupled with high throughput analytical techniques including Protein A HPLC and capillary SDS electrophoresis, the strategy for rapid process screening was defined. The capacity of this system to deliver rapid process understanding was then illustrated by applying it to the full factorial investigation of Polyethylene Glycol (PEG) precipitation for an IgG mAb as a function of antibody concentration, precipitant concentration and pH. Results showed that PEG concentrations required for maximum yield and purity were dependent on the IgG concentration. A window of operation was defined for all mAb concentrations tested where PEG concentrations of 12 to 20 %w/v, pH 8.0 gave desirable levels of yield and purity. The effect of reactor conditions on the IgG precipitate characteristics generated from batch precipitations at USD scale showed a marked dependence of PSD on mixing, IgG concentration, PEG concentration, and PEG addition rate. The results also showed that particle strength and density were a factor of aging time, PEG concentration and mixing intensity. Data on scale-up from USD to lab (330 mL) and pilot (7 to 8 L) scales indicated that the trends and parameters identified at USD scale were important at large scale. Scale-up based on the local energy dissipation rate of the impeller region and the ratio of circulation time to feed addition time was shown to allow accurate prediction of large scale performance from USD scale. A USD rotating shear device providing feed material to a laboratory batch centrifuge allowed successful prediction of pilot plant scale clarification performance using the equivalent settling area theory. Expectatedly, preliminary analysis of the bulk re-solubilised precipitates following the precipitation step showed that the level of impurity clearance achieved did not match the performance of traditional Protein A affinity chromatography. However, integrating the precipitation step into a generic antibody purification platform and substituting the Protein A chromatography capture step with a precipitation step demonstrated the potential of precipitation albeit purity and yield of the final purified product was lower than the process incorporating a Protein A capture step. Nevertheless, this work indicated that further development activities would be required in order to use precipitation in the biopharmaceutical manufacture of mAbs. The use of precipitation for antibody recovery could potentially reduce costs associated with downstream operations, increase plant throughput and manage large intermediate process volumes associated with high titre high volume mAb production processes. Furthermore, such a development and scale-up approach could reduce the timelines for bioprocess design and development to facilitate faster drug to clinic initiatives. In conclusion, this study has demonstrated that micro-scale and ultra scale-down evaluation methods can be used successfully to predict precipitation conditions for use at laboratory and pilot scales. Rapid development of precipitation conditions can allow this unit operation to be used for a range of recombinant proteins (including mAbs) with associated reductions in cost of manufacturing.

The widespread use of biocatalysis in industry will require conversions to achieve high space-time yields. For many next generation bioconversions the low thermodynamic equilibrium constant of reactions currently limits their industrial implementation. This thesis aims to establish a series of generic microscale methods for the rapid evaluation of process options to enhance the yield of such equilibrium-controlled bioconversions. These are evaluated based on the asymmetric synthesis of chiral amino alcohols by the CV2025 ω-transaminase (ω-TAm) from C. violaceum DSM30191. Using 50 mM each of L-Erythrulose (Ery) as substrate and S-α-methylbenzylamine (MBA) as amino donor the standard yield of the product 2−amino−1,3,4−butanetriol (ABT) is just 26 % (mol.mol-1). This reaction produces acetophenone (AP) as a by-product which is also inhibitory to the CV2025 ω-TAm. Microscale methods to evaluate four process options for increasing the bioconversion yield were established each operating at 300 µL volume. The first option involves screening of alternative amino donors to the widely-used MBA. The second couples ω-TAm with an alcohol dehydrogenase (ADH) to convert the inhibitory AP by-product into the non-inhibitory (R)-1-phenylethanol (a glucose dehydrogenase (GDH) is also present to facilitate co-factor recycling). The third approach involves physical in-situ product removal (ISPR) of the volatile AP by operation at reduced pressure in a 96-well vacuum manifold. The final approach involves ISPR using polymeric resins, such as the AmberliteTM XAD series, for selective adsorption of AP from the bioconversion medium. For the particular reaction studied here, use of alternative amino donors such as isopropylamine (IPA) enabled a 2.8-fold increase in the reaction yield to 72 % (mol.mol-1) while the second enzyme system, though more expensive to implement, achieved a 3.8-fold increase in yield and almost quantitative conversion (98 % (mol.mol-1)). Each of the microscale methods was subsequently implemented as process options in an automated micro-scale process sequence linking biocatalyst production, bioconversion and product analysis. The high throughput potential of the methods was then illustrated using a focused library of ten novel ω-TAms from various strains. The bioconversion rates and yields of each enzyme were evaluated using a range of alternative amino donors. This highlight one novel ω-TAm, able to utilise L-α-serine as amino donor at improved yield than the CV2025 ω-TAm. This opens up the possibility of engineering whole cell TAm biocatalysts where the normally expensive amino donors are synthesised in vivo by normal amino acid metabolism starting from simple, cheap substrates. The microscale methods also enabled rapid optimisation of bioconversion conditions using the second enzyme system. This identified the possibility of reducing by 40-fold the standard concentrations of the co-factor recycling enzymes thus reducing the contribution of the enzymes to the overall Costs of Goods (CoG). Overall, this work has established an efficient and generic microscale approach to aid bioprocess design and specifically to help increase the yield of equilibrium-controlled bioconversions. The bioconversion rates and yields for the optimised reaction systems were subsequently verified for preparative scale bioconversions.

In the early stages of downstream process development there is typically only limited availability of process material. Novel methods to obtain information from fewest experiments are essential to make informed choices between processing alternatives at the earliest stage. Design of chromatographic separation initially involves scouting of appropriate matrix type, mobile phase compositions followed by test runs at lab scale and verified at pilot scale. Traditional small-scale methods for chromatography development focus on the screening of separation media and feedstock conditions. It is still necessary to predict chromatography performance at different scales and operating conditions. In this work a new method has been developed to predict performance of larger scale columns using an ultra scale-down approach. The strategy breaks traditional geometric scaling rules, using models to correct for the differences in performance and also for prediction of the effect of changes in operating conditions. Micro-scale columns were used to scale down lab scale runs further challenging the traditional scale down strategies. The characteristics of antibody fragments in E.coli lysate were identified in terms of pH, precipitation and ionic strength to determine good binding conditions. Chromatography studies were carried out at laboratory scale (1 mL) to investigate the flowrate effects on the adsorption of antibody fragments on a strong cation exchange resin. The effect was successfully predicted using a general rate model, which describes the physical and chemical forces of resin-protein interactions but with modifications to allow for deviations noted in experimental performance possibly due to fouling and long loading times changing the rate of protein transfer. Further studies were carried out using micro-scale tip chromatography, mimicking the results obtained at 1 mL scale. A similar effect of flowrate was observed and the scale up factor to predict the performance of laboratory 1 mL scale from 40 μL micro-scale was investigated.

The large number of novel biocatalyst candidates available due to advances in protein engineering and evolution has driven research on automated microwell techniques for rapid catalyst evaluation and quantification of enzyme kinetics and stability. Interest in the further reduction in volume to the microfluidic scale has complemented these microwell approaches for the development of bioprocess operations due to their potential as inexpensive analytical tools with minute volumes and high throughput as well as for their potential for mass replication. This project involves the design and characterisation of a prototype immobilised enzyme microreactor (IEMR) on the inner surface of a 200 μm ID fused silica capillary. Immobilisation is achieved through affinity-based interaction between His6- tags engineered on the transketolase (TK) and transaminase (TAm) enzyme variants and Ni-NTA groups on the derivatised capillary surface. The microreactor concept was validated with two reactions, namely the transketolase-catalysed conversion of hydroxypyruvate (HPA) and glycolaldehyde (GA) to produce L-erythrulose followed by the conversion of erythrulose to 2−amino−1,3,4−butanetriol (erythrulose−aminotriol) in the presence of methybenzylamine (MBA) by CV (Chromobacterium violaceum)-derived ω- transaminase. These keto- and aminodiol synthons are synthetically very useful in the production of a range of compounds with pharmaceutical application. The principles of stop-flow (batch) kinetics were initially investigated with respect to the catalytic performance of both enzymes, where the reaction was shown to depend on substrate concentration and residence time. TK kinetic parameters, evaluated based on a Michaelis−Menten model, in the IEMR (Vmax(app) = 0.1 ± 0.02 mmol.min-1, Km(app) = 26 ± 4 mM) were shown to be comparable to those measured in free solution. Furthermore, the kcat for the microreactor of 2.1 s−1 was similar to the value of 3.9 s−1 for the bioconversion in free solution. This was attributed to the controlled orientation and monolayer surface coverage of the His6−immobilised TK. Furthermore the quantitative elution of the immobilised TK and the regeneration and reuse of the derivatised capillary over 5 cycles were also demonstrated. Whilst slower than TK, the TAm reaction in the IEMR showed similar catalytic performance to a standard reaction in glass vials. Stopped−flow bioconversion results were complemented by continuous flow kinetics of the TK reaction with on-line UV detection (ActiPix, Paraytec), where the dependence of reaction kinetics on flow conditions was investigated. The Km(app), evaluated based on a continuous flow kinetic model, was shown to increase with flow rate, with the optimal being at the lowest flow rates used (0.2 μL.min-1). Furthermore, the value of Km(app) was shown to approach the value of the Michaelis constant of the free enzyme under zero flow (∼25 mM). The prototype microfluidic system was then implemented for the quantitative evaluation of multi-step TK-TAm bioconversion kinetics and the formation of chiral amino diol 2-amino-1,3,4- butanetriol (ABT) product from achiral substrates was demonstrated. The rate of accumulation of ABT (also referred to as EAT) by TAm was 0.02 mM.min-1.μgTAm -1, which was 4× slower than the rate of the TK−catalysed step. Demonstration of the synthesis of the product via the dual reaction and the monitoring of each component provided a full profile of the little known bioconversion and demonstrated the potential for creating novel multi−enzyme pathways in lab−on−a−chip systems, which further enables multi-substrate screening and screening of libraries of evolved enzymes of interest to be achieved rapidly and economically. This in vitro study of multi-step enzyme kinetics provides insight into the behaviour of these de novo engineered pathways to aid incorporation into suitable host cells.

Embryonic Stem Cells (ESC) have the potential to differentiate into any adult cell type even after extended in-vitro culture. These cells have the scope to be a source of specialized cells for pharmaceutical screening and potential cell based transplantation therapies. The successful commercialization of ESCderived products will be highly dependent on the development of cost-effective production bioprocesses. Currently, the most inefficient phase of these bioprocesses is cell differentiation, resulting in low numbers and purity of the target phenotype. Research described in this thesis investigated whether physiological oxygen tensions would influence the yield and purity of neuronal cells during mouse ESC (mESC) differentiation. A chemically defined media, monolayer protocol for the neuronal differentiation of mouse ESC in conjunction with a hypoxia chamber to control the oxygen tension on the growth surface was used. In the first part of the study, it was found that 2% O2 enhanced the yield of cells expressing neuron specific markers (βIII-tubulin and MAP2) as compared with traditional culture conditions (20% O2). Further experiments demonstrated that higher neuronal production was achieved when mES cells were differentiated in the range 4 – 10% O2; an increase in neural rosette diameter was detected within the same oxygen conditions. Embryo development is carried out under hypoxic conditions before week 11 (in humans) of gestation and before 9 days of gestation in the case of mouse embryo. After these indicated stages, blood starts to flow from mother to embryo and oxygen conditions rises from hypoxic to different physiological oxygen values (generally rises from 0% to 2%, 5% or 8% depending of the organ or tissue analyzed). Step changes in oxygen tension during mESC neuronal in vitro differentiation were performed in order to mimic the described in vivo conditions. These results highlight that mimicking in-vivo oxygen tensions during early embryo development can be used to achieve significant enhancements in the yield of neuronal cells from ESC differentiation protocols. Automated mES neuronal differentiation under static oxygen conditions was implemented in order to avoid transient changes from in vitro controlled physiological values to laboratory environmental 20% O2 conditions while manual processing. This lead to an increase of neuronal maturation as observed in the higher MAP2 (mature neuronal marker) expression. Interestingly cell metabolism was also influenced by static oxygen conditions. Results described in this thesis establish the importance of controlling oxygen tension either in a manual or automated way during the in vitro mESC neuronal differentiation. The strategies proposed in this thesis will allow developing higher efficiency and higher purity of neurons from a stem cell bioprocessing perspective.

Stem cells are unique cells that have both the capacity for self-renewal and, depending on their origin, the ability to form at least one, and sometimes many, specialised cell types of all three embryonic germ lineages - germ cells (endoderm, mesoderm and ectoderm), extra-embryonic tissue and trophoblast. Since the derivation of the first human embryonic stem cell (hESC) line in 1998, there has been substantial interest in the potential of these cells both for regenerative medicine and cell therapy, and as disease models for monogenic disorders. Aside from the need to improve derivation efficiency and further the understanding of the basic biology of these cells, the ability to work with hESC opens up three broad research areas. The first is the development of clinical grade culture systems with the aim of producing cell lines suitable for subsequent manipulation for therapy. The second is the opportunity to use these cells as a tool to study the earliest determinative events in mammalian development, such as the origins of patterning in the mammalian embryo. The third is the use of hESCs carrying clinically relevant genetic mutations as models for disease research and therapeutic target identification. The development of several methods of embryo manipulation tailored to the morphology of the blastocyst is described here, which resulted in the derivation of seven lines from four different procedures and provided the tools for subsequent research. Acknowledging that each laboratory in isolation is unlikely to derive sufficient lines to draw significant conclusions regarding manipulation methodology and culture parameters, an international collaboration was initiated with the aim of standardising the reporting of derivation and thus obtaining the maximum information from the generation of each new hESC line. To address the need for the development of clinical grade culture systems, alternative feeder cells were assessed for their suitability in hESC culture and derivation. Modified human foreskin fibroblasts and human amniotic epithelial cells (hAECs) were investigated, as both cell types can be fully qualified and validated. Whilst both were able to support the culture of existing lines, only the hAECs showed promise in supporting derivation. In addition, analysis of in-house and commercially available media showed that neither were physiologically optimal for the growth of inner cell mass (ICM) cells or putative hESC, as metabolite concentrations were in excess and subsequent catabolite levels exceeded known toxic levels. The timing and mechanisms establishing patterning and future polarity in the mammalian embryo remains a subject of intense debate. Here, the potential of single blastomeres to generate hESC was used as an assessment of pluripotency. The determination of the most appropriate day for attempting derivation was performed by assessing blastomere development and pluripotent marker expression, and the predicted success of derivation was considered in the light of division patterns. Putative stem-like cells were visible in several cultures. Furthermore, isolated blastomeres from two-, four-, and eight-cell embryos were analysed for the quantitative expression of multiple target genes known to be associated with lineage formation and the stem cell state. Analysis suggested that broad changes in gene expression were occurring with development stage. However, no consistent grouping structure for cells within embryos was observed, and no convincing pattern was seen when considering the individual embryo variance scores. Several approaches are discussed to differentiate between the biological and methodological variability in this experimental design. The suitability of hESC as models for genetic disease was studied following the derivation of two lines carrying Huntington disease (HD). Subsequent differentiation using a stromal co-culture neural induction protocol resulted in the establishment of a stable, highly proliferative cell population which was simple to culture and bank. The cells were of an astroglial phenotype, and therefore highly suited for subsequent studies regarding HD pathophysiology, as glial cells are severely affected in HD. During differentiation the CAG repeat size increased from 46 to 70, showing the salient feature of somatic instability of the huntingtin gene. Therefore this cell population provides a valuable tool in the study of disease pathogenesis and transmission.

Bioprocessing groups must keep pace with the many biologics and vaccines entering development, while ensuring process robustness, controlling costs, and accelerating project timelines. Microscale techniques provide a means to cope with these challenges by enabling high-throughput investigations to identify problems early, reduce requirements for costly large-scale experiments, and promote quality-bydesign approaches for process optimisation. Micro-tip columns (packed sorbent in a pipette tip) for chromatography and Adaptive Focused Acoustics (AFA) for cell disruption are two such techniques with potential to deliver high-throughput process development. This thesis characterises these platforms and integrates them as elements of the development workflow. Firstly, the key parameters are defined for robust, automated micro-tip chromatography. Finite-bath methods for isotherms and kinetic measurements are demonstrated, with sorbent contact time found to be critical for uptake of proteins on porous adsorbents, consistent with pore diffusion being rate-determining. Based upon these micro-tip data, two data-driven models are applied to predict dynamic binding capacity, one employing a shrinking-core model, and the other, a stagedreaction model. Both show satisfactory agreement with experimental laboratory column results. Micro-tip chromatography is then illustrated as an accelerated process development strategy for a mixed-mode chromatography step, with the results found to be predictive of laboratory column-scale yield, purity and capacity. In a second application, micro-tip chromatography is used to evaluate the interaction of upstream fermentation changes upon the downstream chromatography. The microscale chromatography is predictive of laboratory-scale yield and purity, despite being 1000-times smaller, while increasing productivity by over ten-fold. The miniaturisation of the chromatography, however, necessitates the development of a microscale cell disruption method to fully realise the gains in throughput and volume reduction. The AFA technique meets this goal, providing representative feed material for chromatographic study. Together, micro-tip chromatography and AFA form the basis for a next-generation bioprocess development platform.

Pleated membrane cartridge filters are used extensively throughout a typical bioprocess. They are exposed to a range of operating conditions and feedstocks. Discrepancies between the performance of the flat sheet membrane and pleated membrane have previously been identified, although little has been done to fully characterise the effects of pleating. As current scale-up techniques use the flat sheet membrane to predict the performance of the large-scale pleated cartridge, the discrepancy in performance between flat sheet and pleated cartridge leads to inaccuracy in scale-up. This inaccuracy is accounted for by over-sizing of the equipment. In turn this reduces the efficiency of the bioprocess and increases capital costs. At the present time no accurate and reliable scale-up methodology exists that accounts for the effects of pleating. A systematic investigation into the effect of pleating has been conducted. By varying the key pleat characteristics: pleat height, type and packing density, the impact upon cartridge performance of these characteristics has been determined. Using this knowledge, new scale-down cartridge filters have been developed, fabricated and tested. When faced with both clean water and a pepsin protein solution, performance was within 10% of the large-scale 10” counterpart, whilst operating with a 1000 fold reduction in feed volume. This compares well to flat sheet membrane which showed up to 53% variation in performance to the pleated cartridge filter. The scale-down cartridge is limited to the degree in which reduction of feedstock can be achieved. So as to reduce feed volume requirements further, a ultra scaledown methodology has been developed that uses experimental models to account for the effect that pleating has upon cartridge performance. When coupled with experimental data derived from flat sheet discs, the scale-up performance improves predictions with flat sheet membrane however discrepancies still exist between the two scales, suggesting that the method is not yet robust. Based upon the work of this thesis the close performance between the scaledown cartridges and the large-scale cartridges, coupled with the low feed requirement, make the device an excellent method by which rapid scale-up can be achieved during the process development of biopharmaceutical products. However, it is recommended that the ultra scale-down approach is developed further, so as to build a robust method to predict the performance of industrial scale pleated filter cartridges using significantly reduced areas of flat sheet membrane.

This thesis summarises work done on the Escherichia coli strain MG1655 expressing a Green Fluorescent Protein (GFP) and the flavo-protein N-methyl-L-tryptophan oxidase (MTOX) product and examining the effect foreign protein production has on cell growth parameters. It also uses molecular modelling tools to generate data relating to FAD flux and MTOX production, comparable to that seen in E.coli fermentations. The MG1655 strain was chosen as it was the focus of the first K-12 complete sequencing project and closely related to the strain W3110, a second K strain that had been used to develop a number of deletion mutants which were central to the study. It presents data from shake flask and stirred tank reactor fermentations on minimal, carbon limited and complex media. Samples from these growth experiments were then analysed concentrating on biomass concentration, protein assays (both chemical and fluorimetric), high performance liquid chromatography and calculation of yield parameters. From this a baseline of growth was established with which to compare changes in growth after a shift in protein product from GFP to MTOX. An assay was also developed to measure the amount of active and inactive MTOX enzyme produced and this data compared with the level of FAD available to the cell at specific time points throughout growth. Finally modelling work is presented and in silico values compared with those generated in vitro. A discussion of the entire study concludes the work.