Optimization of biopharmaceutical downstream processes supported by mechanistic models and artificial neural networks.

Downstream process development is a major area of importance within the field of bioengineering. During the design of such a downstream process, important decisions have to be made regarding the type of unit operations as well as their sequence and their operating conditions. Current computational approaches addressing these issues either show a high level of simplification or struggle with computational speed. Therefore, this article presents a new approach that combines detailed mechanistic models and speed-enhancing artificial neural networks. This approach was able to simultaneously optimize a process with three different chromatographic columns toward yield with a minimum purity of 99.9%. The addition of artificial neural networks greatly accelerated this optimization. Due to high computational speed, the approach is easily extendable to include more unit operations. Therefore, it can be of great help in the acceleration of downstream process development. © 2017 American Institute of Chemical Engineers Biotechnol. Prog., 33:696-707, 2017.

3D-liquid chromatography as a complex mixture characterization tool for knowledge-based downstream process development.

Knowledge-based development of chromatographic separation processes requires efficient techniques to determine the physicochemical properties of the product and the impurities to be removed. These characterization techniques are usually divided into approaches that determine molecular properties, such as charge, hydrophobicity and size, or molecular interactions with auxiliary materials, commonly in the form of adsorption isotherms. In this study we demonstrate the application of a three-dimensional liquid chromatography approach to a clarified cell homogenate containing a therapeutic enzyme. Each separation dimension determines a molecular property relevant to the chromatographic behavior of each component. Matching of the peaks across the different separation dimensions and against a high-resolution reference chromatogram allows to assign the determined parameters to pseudo-components, allowing to determine the most promising technique for the removal of each impurity. More detailed process design using mechanistic models requires isotherm parameters. For this purpose, the second dimension consists of multiple linear gradient separations on columns in a high-throughput screening compatible format, that allow regression of isotherm parameters with an average standard error of 8%. © 2016 American Institute of Chemical Engineers Biotechnol. Prog., 32:1283-1291, 2016.

Prediction of protein retention times in hydrophobic interaction chromatography by robust statistical characterization of their atomic-level surface properties.

The correlation between the dimensionless retention times (DRT) of proteins in hydrophobic interaction chromatography (HIC) and their surface properties were investigated. A ternary atomic-level hydrophobicity scale was used to calculate the distribution of local average hydrophobicity across the proteins surfaces. These distributions were characterized by robust descriptive statistics to reduce their sensitivity to small changes in the three-dimensional structure. The applicability of these statistics for the prediction of protein retention behaviour was looked into. A linear combination of robust statistics describing the central tendency, heterogeneity and frequency of highly hydrophobic clusters was found to have a good predictive capability (R2 = 0.78), when combined a factor to account for protein size differences. The achieved error of prediction was 35% lower than for a similar model based on a description of the protein surface on an amino acid level. This indicates that a robust and mathematically simple model based on an atomic description of the protein surface can be used for the prediction of the retention behaviour of conformationally stable globular proteins with a well determined 3D structure in HIC. © 2016 American Institute of Chemical Engineers Biotechnol. Prog., 32:372-381, 2016.

Fourier transform assisted deconvolution of skewed peaks in complex multi-dimensional chromatograms.

Lower order peak moments of individual peaks in heavily fused peak clusters can be determined by fitting peak models to the experimental data. The success of such an approach depends on two main aspects: the generation of meaningful initial estimates on the number and position of the peaks, and the choice of a suitable peak model. For the detection of meaningful peaks in multi-dimensional chromatograms, a fast data scanning algorithm was combined with prior resolution enhancement through the reduction of column and system broadening effects with the help of two-dimensional fast Fourier transforms. To capture the shape of skewed peaks in multi-dimensional chromatograms a formalism for the accurate calculation of exponentially modified Gaussian peaks, one of the most popular models for skewed peaks, was extended for direct fitting of two-dimensional data. The method is demonstrated to successfully identify and deconvolute peaks hidden in strongly fused peak clusters. Incorporation of automatic analysis and reporting of the statistics of the fitted peak parameters and calculated properties allows to easily identify in which regions of the chromatograms additional resolution is required for robust quantification.

Evaluation and characterization of axial distribution in expanded bed: II. Liquid mixing and local effective axial dispersion.

Expanded bed adsorption (EBA) is a promising technology to capture proteins directly from unclarified feedstock. In order to better understand liquid mixing along the bed height in expanded beds, an in-bed sampling method was used to measure residence time distribution at different bed heights. A 2cm diameter nozzle column was tested with agarose raw beads (3% crosslinked agarose containing tungsten carbide). Two settled bed heights (11.5 and 23.1cm) with different expansion factors (1.4-2.6) were investigated and the number of theoretical plates (N), the height equivalent of theoretical plate (HETP) and the local effective axial dispersion coefficient (Dax) were calculated for each bed height-defined zone. The effects of expansion factor, settled bed height and mobile phase were evaluated. The results showed that N increased with the increase of expansion factors, but Dax was unaffected under fixed bed heights. Dax and HETP were found similar as a function of relative bed height for two settled bed heights tested. Higher mobile phase viscosity resulted in stronger axial dispersion. In addition, the local effective Dax under the expansion factor near 2.0 had a different profile which showed minimum values at 0.6-0.8 relative bed height, and the potential mechanism was discussed. These results would be useful for the characterization of axial dispersion and modeling protein adsorption in expanded beds under varying operation conditions.

Evaluation and characterization of axial distribution in expanded bed. I. Bead size, bead density and local bed voidage.

Expanded bed adsorption (EBA) is an innovative chromatography technology that allows the adsorption of target proteins directly from unclarified feedstock, and the most important property of an expanded bed is the perfectly classified fluidization of resin beads in the column. Due to the variation of both size and density of bulk resin beads, the axial distributions of bead size, bead density and bed voidage are the inherent characteristics of an expanded bed. However, the understanding on these properties is quite limited. In this study, raw beads (3% crosslinked agarose containing tungsten carbide) and 2cm-diameter nozzle column were used as the model system and mean bead size, bead density and local bed voidage along the bed height were measured systematically with the in-bed sampling method for two settled bed heights (11.5 and 23.1cm) and different expansion factors (1.4-2.6). With the increase of bed height, mean bead size and wet density of the beads decreased from 140 to 90µm and from 4 to 2g/ml, respectively. The local bed voidage increased from 0.6 to 0.9 with the increasing bed height. The relative bed height and relative bed voidage were introduced to describe the general rule of axial distribution. Some empirical equations were used to correlate the mean bead size, bead density and local bed voidage along the bed height with the standard deviations of 10.6%, 6.1% and 5.5, respectively. In addition, a general equation was proposed to predict the axial distributions of bead size, bead density and local bed voidage in the column with standard deviations less than 10% for most of the experimental data, which would be useful for the characterization of resin beads distribution in an expanded bed under varying operation conditions.

Multi-dimensional fractionation and characterization of crude protein mixtures: toward establishment of a database of protein purification process development parameters.

A multi-dimensional fractionation and characterization scheme was developed for fast acquisition of the relevant molecular properties for protein separation from crude biological feedstocks by ion-exchange chromatography (IEX), hydrophobic interaction chromatography (HIC), and size-exclusion chromatography. In this approach, the linear IEX isotherm parameters were estimated from multiple linear salt-gradient IEX data, while the nonlinear IEX parameters as well as the HIC isotherm parameters were obtained by the inverse method under column overloading conditions. Collected chromatographic fractions were analyzed by gel electrophoresis for estimation of molecular mass, followed by mass spectrometry for protein identification. The usefulness of the generated molecular properties data for rational decision-making during downstream process development was equally demonstrated. Monoclonal antibody purification from crude hybridoma cell culture supernatant was used as case study. The obtained chromatographic parameters only apply to the employed stationary phases and operating conditions, hence prior high throughput screening of different chromatographic resins and mobile phase conditions is still a prerequisite. Nevertheless, it provides a quick, knowledge-based approach for rationally synthesizing purification cascades prior to more detailed process optimization and evaluation.

Selection of pH-related parameters in ion-exchange chromatography using pH-gradient operations.

This work demonstrates that the type of ion-exchanger (anion or cation), the mode of operation (bind-and-elute or flow-through), and the operational pH of ion-exchange chromatography (IEX) can be selected in a fast and rational way by analytical pH-gradient IEX operations, thereby eliminating the need for pH scouting or high-throughput screening. The developed approach was applied for the selection of an IEX process for the capture of a monoclonal antibody (MAb) from hybridoma cell culture supernatant (CCS). It was found within a day that MAb can optimally be captured by bind-and-elute mode cation-exchange chromatography (CEX) at pH 4.5 or anion-exchange chromatography (AEX) at pH 7.2 without lowering the salt concentration in the CCS. The performance of both CEX and AEX was predicted to be equal for this particular MAb capture.

pH-gradient ion-exchange chromatography: an analytical tool for design and optimization of protein separations.

This work demonstrates that a highly linear, controllable and wide-ranged pH-gradient can be generated through an ion-exchange chromatography (IEC) column. Such a pH-gradient anion-exchange chromatography was evaluated with 17 model proteins and found that acidic (pI<6) and basic (pI>8) proteins elute roughly at their pI, whereas neutral proteins (pI 6-8) elute at pH 8-9 regardless their pI values. Because of the flat nature of protein titration curves from pH approximately 6 to approximately 9, neutral proteins indeed exhibit nearly zero net charge at pH approximately 9. The elution-pH in pH-gradient IEC or the titration curve, but not the pI, was identified as the key parameter for pH optimization of preparative IEC in a fast and rational way. The pH-gradient IEC was also applied and found to be an excellent analytical tool for the fractionation of crude protein mixtures.