Smart process development: Application of machine-learning and integrated process modeling for inclusion body purification processes.

The development of a biopharmaceutical production process usually occurs sequentially, and tedious optimization of each individual unit operation is very time-consuming. Here, the conditions established as optimal for one-step serve as input for the following step. Yet, this strategy does not consider potential interactions between a priori distant process steps and therefore cannot guarantee for optimal overall process performance. To overcome these limitations, we established a smart approach to develop and utilize integrated process models using machine learning techniques and genetic algorithms. We evaluated the application of the data-driven models to explore potential efficiency increases and compared them to a conventional development approach for one of our development products. First, we developed a data-driven integrated process model using gradient boosting machines and Gaussian processes as machine learning techniques and a genetic algorithm as recommendation engine for two downstream unit operations, namely solubilization and refolding. Through projection of the results into our large-scale facility, we predicted a twofold increase in productivity. Second, we extended the model to a three-step model by including the capture chromatography. Here, depending on the selected baseline-process chosen for comparison, we obtained between 50% and 100% increase in productivity. These data show the successful application of machine learning techniques and optimization algorithms for downstream process development. Finally, our results highlight the importance of considering integrated process models for the whole process chain, including all unit operations.

Technology transfer of a monitoring system to predict product concentration and purity of biopharmaceuticals in real-time during chromatographic separation.

Technological developments require the transfer to their location of application to make use of them. We describe the transfer of a real-time monitoring system for lab-scale preparative chromatography to two new sites where it will be used and developed further. Equivalent equipment was used. The capture of a biopharmaceutical model protein, human fibroblast growth factor 2 (FGF-2) was used to evaluate the system transfer. Predictive models for five quality attributes based on partial least squares regression were transferred. Six out of seven online sensors (UV/VIS, pH, conductivity, IR, RI, and MALS) showed comparable signals between the sites while one sensor (fluorescence) showed different signal profiles. A direct transfer of the models for real-time monitoring was not possible, mainly due to differences in sensor signals. Adaptation of the models was necessary. Then, among five prediction models, the prediction errors of the test run at the new sites were on average twice as high as at the training site (model-wise 0.9-5.7 times). Additionally, new prediction models for different products were trained at each new site. These allowed monitoring the critical quality attributes of two new biopharmaceutical products during their purification processes with mean relative deviations between 1% and 33%.

Ethanol precipitation for purification of recombinant antibodies.

Currently, the golden standard for the purification of recombinant humanized antibodies (rhAbs) from CHO cell culture is protein A chromatography. However, due to increasing rhAbs titers alternative methods have come into focus. A new strategy for purification of recombinant human antibodies from CHO cell culture supernatant based on cold ethanol precipitation (CEP) and CaCl2 precipitation has been developed. This method is based on the cold ethanol precipitation, the process used for purification of antibodies and other components from blood plasma. We proof the applicability of the developed process for four different antibodies resulting in similar yield and purity as a protein A chromatography based process. This process can be further improved using an anion-exchange chromatography in flowthrough mode e.g. a monolith as last step so that residual host cell protein is reduced to a minimum. Beside the ethanol based process, our data also suggest that ethanol could be replaced with methanol or isopropanol. The process is suited for continuous operation.

Economics of recombinant antibody production processes at various scales: Industry-standard compared to continuous precipitation.

Standard industry processes for recombinant antibody production employ protein A affinity chromatography in combination with other chromatography steps and ultra-/diafiltration. This study compares a generic antibody production process with a recently developed purification process based on a series of selective precipitation steps. The new process makes two of the usual three chromatographic steps obsolete and can be performed in a continuous fashion. Cost of Goods (CoGs) analyses were done for: (i) a generic chromatography-based antibody standard purification; (ii) the continuous precipitation-based purification process coupled to a continuous perfusion production system; and (iii) a hybrid process, coupling the continuous purification process to an upstream batch process. The results of this economic analysis show that the precipitation-based process offers cost reductions at all stages of the life cycle of a therapeutic antibody, (i.e. clinical phase I, II and III, as well as full commercial production). The savings in clinical phase production are largely attributed to the fact that expensive chromatographic resins are omitted. These economic analyses will help to determine the strategies that are best suited for small-scale production in parallel fashion, which is of importance for antibody production in non-privileged countries and for personalized medicine.

Host cell protein analysis in therapeutic protein bioprocessing - methods and applications.

The analysis of host cell proteins (HCPs) is one of the most important analytical requirements during bioprocess development of therapeutic moieties. In this review, we focus on the comparison of different methods for the analysis of HCPs and how cell lines, fermentation conditions, and unit operations influence HCP distribution during the process chain. Current guidelines typically require reduction of HCPs to the ppm level, depending on the intended use, the route of administration of the product, and the production system. A range of immunospecific and non-specific methods are available that have been globally accepted by regulatory bodies. Immunospecific methods, such as ELISA, are simple to use in routine analysis and can quantify low levels of HCPs when specific antibodies are available. Non-specific methods are more complex; however, they provide a holistic view of the HCP profile and qualitative information of the composition of HCP in the sample. Different methods for the comparison of bioprocessing strategies during scale-up and purification development are compared herein. The methods include immunospecific methods, such as ELISA, western blot, and threshold, and non-specific methods, such as 2D-DIGE and 2D-HPLC combined with MS.

Protein precipitation by polyethylene glycol: a generalized model based on hydrodynamic radius.

PEGs for protein precipitation are usually classified by molecular weight. The higher molecular weight precipitants are more efficient but result in higher viscosity. Following empirical evidence that the precipitation efficiency is more comprehensively characterized by PEG hydrodynamic radius (r(h,PEG)) than molecular weight, this paper proposes a model to explicate the significance of r(h,PEG). A general expression was formulated to characterize the PEG effect exclusively by r(h,PEG). The coefficients of a linearized form were then fitted using empirical solubility data. The result is a simple numerical relation that models the efficiency of general-shaped PEG precipitants as a function of r(h,PEG) and protein hydrodynamic radius (r(h,prot)). This equation also explains the effects of environmental conditions and PEG branching. While predictions by the proposed correlation agree reasonably well with independent solubility data, its simplicity gives rise to potential quantitative deviations when involving small proteins, large proteins and protein mixtures. Nonetheless, the model offers a new insight into the precipitation mechanism by clarifying the significance of r(h,PEG). This in turn helps to refine the selection criterion for PEG precipitants.

Branched polyethylene glycol for protein precipitation.

The use of linear PEGs for protein precipitation raises the issues of high viscosity and limited selectivity. This paper explores PEG branching as a way to alleviate the first problem, by using 3-arm star as the model branched structure. 3-arm star PEGs of 4,000 to 9,000 Da were synthesized and characterized. The effects of PEG branching were then elucidated by comparing the branched PEG precipitants to linear versions of equivalent molecular weights, in terms of IgG recovery from CHO cell culture supernatant, precipitation selectivity, solubility of different purified proteins, and precipitation kinetics. Two distinct effects were observed: PEG branching reduced dynamic viscosity; secondly, the branched PEGs precipitated less proteins and did so more slowly. Precipitation selectivity was largely unaffected. When the branched PEGs were used at concentrations higher than their linear counterparts to give similar precipitation yields, the dynamic viscosity of the branched PEGs were noticeably lower. Interestingly, the precipitation outcome was found to be a strong function of PEG hydrodynamic radius, regardless of PEG shape and molecular weight. These observations are consistent with steric mechanisms such as volume exclusion and attractive depletion.

Two-dimensional difference fluorescence gel electrophoresis to verify the scale-up of a non-affinity-based downstream process for isolation of a therapeutic recombinant antibody.

For therapeutic antibody production Protein A chromatography is often replaced by non-affinity-based purification sequences, which are considered as more economical. 2-D DIGE was applied for evaluation of scale-up of non-affinity based process of a humanized monoclonal antibody, anti-Rh(D) IgG(1), in comparison with other conventional analytical methods, like SDS-PAGE, Western blot, or SEC. Due to a high sensitivity of this technique (125 pg protein/spot) and high dynamic range of five orders of magnitude, low molecular weight impurities were detected in purified samples. Cation exchange chromatography was efficient capture step for IgG(1) purification in laboratory and pilot scale. The differences between samples after first purification step in laboratory and pilot scale were compensated with second purification step where almost the same protein pattern was observed. 2-D DIGE is a helpful tool for monitoring of purification effects and for scale-up verification of downstream processes.

Stem cell separation: a bottleneck in stem cell therapy.

The substantial progress in embryonic stem cell (ESC) research could lead to new possibilities in the treatment of various diseases. Currently, applications of ESC for cell therapy are impeded by the presence of potentially teratoma-forming undifferentiated ESC. Thus, a selective and quantitative removal of undifferentiated ESC from a pool of differentiated and undifferentiated cells is essential before cell therapy. We evaluated the highly selective magnetic activated cell sorting (MACS) method for the quantitative removal of undifferentiated ESC. We found that the clearance rates for undifferentiated ESC decreased with decreasing amount of undifferentiated ESC in the cell pool. Using a simplified model calculation we could predict that, assuming an initial purity of 60%, an estimated 31 steps are required to achieve less than 10(-1) cell per 10(9) cells. Thus, a log clearance rate of 10, which would be necessary for a therapeutically application, is hard to achieve. Our work clearly indicates that the current MACS technology is insufficient to meet the purification needs for cell therapy.

Engineering of a two-step purification strategy for a panel of monoclonal immunoglobulin M directed against undifferentiated human embryonic stem cells.

A two-step purification strategy comprising of polyethylene glycol (PEG) precipitation and anion-exchange chromatography was developed for a panel of monoclonal immunoglobulin M (IgM) (pI 5.5-7.7) produced from hybridoma cultures. PEG precipitation was optimized with regards to concentration, pH and mixing. For anion-exchange chromatography, different resins were screened of which Fractogel EMD, a polymer grafted porous resin had the highest capacity. Despite its significantly slower mass transfer, the binding capacity was still higher compared to a convection driven resin (monolith). This purification strategy was successfully demonstrated for all 9 IgMs in the panel. In small scale most antibodies could be purified to >95% purity with the exception of two which gave a lower final purity (46% and 85%). The yield was dependent on the different antibodies ranging from 28% to 84%. Further improvement of recovery and purity was obtained by the digestion of DNA present in the hybridoma supernatant using an endonuclease, benzonase. So far this strategy has been applied for the purification of up to 2l hybridoma supernatants.

Two-dimensional fluorescence difference gel electrophoresis for comparison of affinity and non-affinity based downstream processing of recombinant monoclonal antibody.

Although Staphylococcus Protein A (SpA) affinity chromatography is the state of the art capture step for antibody purification, non-affinity methods are more economical. We used two-dimensional fluorescence difference gel electrophoresis (2-D DIGE) to evaluate the purification of a recombinant IgG(1) antibody from cultured cells, with two different processes: (1) SpA capture followed by cation-exchange chromatography (CEX); and (2) CEX capture, followed by anion exchanger, then hydrophobic interaction chromatography. Efficiencies were similar in sodium dodecylsulphate polyacrylamide gel electrophoresis and size-exclusion chromatography; however, 2-D DIGE revealed higher efficiency with SpA than with CEX capture. Thus, 2-D DIGE is a valuable tool for downstream process development.

2-D DIGE to expedite downstream process development for human monoclonal antibody purification.

Two-dimensional fluorescence difference gel electrophoresis (2-D DIGE) is an established method for assessing protein expression strategies, understanding pathogenesis mechanisms, characterizing biomarkers, and controlling therapeutic processes. We applied 2-D DIGE to facilitate the development of a purification process for a recombinant IgG1 antibody against Rhesus D antigen expressed by Chinese hamster ovary cells. The variability of two expression clones as well as the influence of cell viability on the host-cell protein pattern was assessed quantitatively. Up to 800 different spots were identified. 2-D DIGE showed that differences in cell viability had more influence on the protein expression pattern than did the expression clone itself. After purification of the IgG from different culture supernatants, the protein patterns on 2-D DIGE were identical, indicating the validity of purification scheme.

High-performance monolith affinity chromatography for fast quantitation of immunoglobulin G.

High-performance monolith affinity chromatography employing protein A resins has been introduced previously for the fast purification of IgG from different sources. Here we describe the design and evaluation of a fast and specific method for quantitation of IgG from purified samples as well as crude supernatant from Chinese hamster ovary (CHO) cells. We used a commercially available affinity monolith with protein A as affinity ligand (CIM protein A HLD disk). Interferences of CHO host cell proteins with the quantitation of IgG from CHO supernatant were eliminated by a careful choice of the equilibration buffer. With this method developed, it is possible to quantify IgG within 5 min in a concentration range of 23-250 microg/ml. The calibration range of the method could be extended from 4 to 1000 microg/ml by adjusting the injection volume. The method was successfully validated by measuring the low limit of detection and quantification, inter- and intra-day precision and selectivity.

Adaptation of hybridomas to protein-free media results in a simplified two-step immunoglobulin M purification process.

An IgM antibody was purified from hybridoma supernatant containing serum using a three-step purification process comprising of tangential flow filtration, anion-exchange chromatography and size-exclusion chromatography. Recovery and purity were significantly improved upon adaptation of the hybridoma to serum-free media. The process could even be simplified by omitting the initial tangential flow filtration step. Even with a two-step purification process a purity of >98% and a recovery of >60% was obtained.

Dispersion effects in preparative polymethacrylate monoliths operated in radial-flow columns.

Monolithic media have found widespread use as excellent tools for fast analytical separations of small molecules, proteins, pDNA and viruses. Polymethacrylate monoliths with large channels are attractive for capturing large molecules, like immunoglobulins, DNA, and viruses. For preparative purposes, these monoliths are operated in radial flow mode. Band spreading in monoliths is extremely low and mostly dominated by the contribution of extra column effects. The model used here had a single axial dispersion coefficient which lumps together extra column effects and the intrinsic band spreading of the monolithic material to characterize the adsorption of proteins and pDNA on polymethacrylate ion-exchange monoliths. Due to the fact that the performance of the monolith was unaffected by the velocity within the applied range, and due to highly favourable adsorption isotherms, a constant pattern model could be applied to predict preparative runs on radial flow units assuming axial flow for modelling.

Hydrophobic interaction chromatography of proteins IV. Kinetics of protein spreading.

Adsorption of proteins on surfaces of hydrophobic interaction chromatography media is at least a two-stage process. Application of pure protein pulses (bovine serum albumin and beta-lactoglobulin) to hydrophobic interaction chromatography media yielded two chromatographic peaks at low salt concentrations. At these salt concentrations, the adsorption process is affected by a second reaction, which can be interpreted as protein spreading or partial unfolding of the protein. The kinetic constants of the spreading reaction were derived from pulse response experiments at different residence times and varying concentrations by applying a modified adsorption model considering conformational changes. The obtained parameters were used to calculate uptake and breakthrough curves for spreading proteins. Although these parameters were determined at low saturation of the column, predictions of overloaded situations could match the experimental runs satisfactorily. Our findings suggest that proteins which are sensitive to conformational changes should be loaded at high salt concentrations in order to accelerate the adsorption reaction and to obtain steeper breakthrough curves.

Comparison of protein A affinity sorbents II. Mass transfer properties.

Protein A affinity chromatography is the standard purification method for isolation of therapeutic antibodies. Due to improvements in expression technology and optimization of fermentation, culture supernatants with high antibody content must be processed. Recently protein A affinity media with improved adsorption characteristics have been developed. The agarose media MabSelect Xtra and MabSelect SuRe are recent developments of the existing protein A affinity medium MabSelect. MabSelect Xtra is designed to exhibit a higher binding capacity for IgG, and MabSelect SuRe is functionalized with an alkaline stabilized protein A. ProSep-vA Ultra is a porous glass medium with a pore size of 70 nm, also developed to improve the binding capacity. Adsorption was measured in a finite and infinite bath. Mass transfer in these systems could be well described by a model including film and pore diffusion. Mass transfer parameters were used to accurately predict IgG breakthrough in packed bed mode. The dynamic binding capacity of all three media did not change when residence time was at least 4 min. All three media are suited for capture of feed stocks with high antibody content.

In situ determination of adsorption kinetics of proteins in a finite bath.

A method for fast in situ measurement of adsorption kinetics based on a finite bath was developed. We modified the conventional finite bath by replacing the external loop by a dip probe which enables in situ measurement of the concentration change in the contactor. Deposition of adsorbent particles on the reflection surface of the dip probe compromised measurements. Different membranes, a polyamide, a polypropylene and a nylon membrane were tested to protect the internal reflection surface of the dip probe from fouling with adsorbent particles. The nylon membrane provided efficient protection and high mass transfer evaluated by response time experiments. Unspecific adsorption of the model protein on the membrane could also be excluded. To corroborate the measurements of the dip probe the results were compared to a conventional finite bath and to a shallow-bed. The uptake curves for human polyclonal IgG at different concentrationes (0.1-3 g/l) on rProtein A Sepharose FF and MabSelect were used as model system. The effective diffusion coefficients were determined using a pore diffusion model. These values were in good agreement for all methods.