Robust mechanistic modeling of protein ion-exchange chromatography

An examination of process models and model risk frameworks for pharmaceutical manufacturing

Process models are a growing tool for pharmaceutical manufacturing process design and control. The Industry 4.0 paradigm promises to increase the amount of data available to understand manufacturing processes. Tools such as Artificial Intelligence (AI) might accelerate process development and allow better predictions of process trajectories. Several examples of process improvements realized through the application of process models have been shown in lyophilization, chromatography, fluid bed drying, bioreactor control, continuous direct compression, and wet granulation. An important consideration of implementing a process model is determining the impact of the model on the quality of the product and the risks associated with model maintenance over the product lifecycle. Several regulatory documents address risk-based considerations for process models. This work discusses existing risk-based frameworks for model validation and lifecycle maintenance that could aid the adoption of process models in pharmaceutical manufacturing. Hypothetical case studies illustrate the implications of applying a model risk framework to facilitate model validation and lifecycle maintenance in the manufacture of pharmaceuticals and biological products.

Coupling cation and anion exchange chromatography for fast separation of monoclonal antibody charge variants

A design procedure for the separation of charge variants of a monoclonal antibody (mAb) was developed, which was based on the coupling of cation-exchange chromatography (CEX) and anion-exchange chromatography (AEX) under high loading conditions. The design of the coupled process was supported by a dynamic model. The model was calibrated on the basis of band profiles of variants determined experimentally for the mAb materials of different variant compositions. The numerical simulations were used to select the coupling configuration and the loading conditions that allowed for efficient separation of the mAb materials into three products enriched with each individual variant: the acidic (av), main (mv) and basic (bv) one. In the CEX section, a two-step pH gradient was used to split the loaded mass of mAb into a weakly bound fraction enriched with av and mv, and a strongly bound fraction containing the bv-rich product. The weakly bound fraction was further processed in the AEX section, where the mv-rich product was eluted in flowthrough, while the av-rich product was collected by a step change in pH. The choice of flow distribution and the number of columns in the CEX and AEX sections depended on the variant composition of the mAb material. For the selected configurations, the optimized mAb loading density in the CEX columns ranged from 10 to 26 mg mL-1, while in the AEX columns it was as high as 300 or 600 mg mL-1, depending on the variant composition of the mAb material. By proper selection of the loading condition, a trade-off between yield and purity of the products could be reached.

High-throughput in silico workflow for optimization and characterization of multimodal chromatographic processes

While high-throughput (HT) experimentation and mechanistic modeling have long been employed in chromatographic process development, it remains unclear how these techniques should be used in concert within development workflows. In this work, a process development workflow based on HT experiments and mechanistic modeling was constructed. The integration of HT and modeling approaches offers improved workflow efficiency and speed. This high-throughput in silico (HT-IS) workflow was employed to develop a Capto MMC polishing step for mAb aggregate removal. High-throughput batch isotherm data was first generated over a range of mobile phase conditions and a suite of analytics were employed. Parameters for the extended steric mass action (SMA) isotherm were regressed for the multicomponent system. Model validation was performed using the extended SMA isotherm in concert with the general rate model of chromatography using the CADET modeling software. Here, step elution profiles were predicted for eight RoboColumn runs across a range of ionic strength, pH, and load density. Optimized processes were generated through minimization of a complex objective function based on key process metrics. Processes were evaluated at lab-scale using two feedstocks, differing in composition. The results confirmed that both processes obtained high monomer yield (>85%) and removed∼ 50 % $$ \sim 50\% $$ of aggregate species. Column simulations were then carried out to determine sensitivity to a wide range of process inputs. Elution buffer pH was found to be the most critical process parameter, followed by resin ionic capacity. Overall, this study demonstrated the utility of the HT-IS workflow for rapid process development and characterization.

Mechanistic model-based characterization of size-exclusion-mixed-mode resins for removal of monoclonal antibody fragments

Although antibody fragments are a critical impurity to remove from process streams, few platformable purification techniques have been developed to this end. In this work, a novel size-exclusion-mixed-mode (SEMM) resin was characterized with respect to its efficacy in mAb fragment removal. Inverse size-exclusion chromatography showed that the silica-based resin had a narrow pore size distribution and a median pore radius of roughly 6.2 nm. Model-based characterization was carried out with Chromatography Analysis and Design Toolkit (CADET), using the general rate model and the multicomponent Langmuir isotherm. Model parameters were obtained from fitting breakthrough curves, performed at multiple residence times, for a mixture of mAb, aggregates, and an array of fragments (varying in size). Accurate fits were obtained to the frontal chromatographic data across a range of residence times. Model validation was then performed with a scaled-up column, altering residence time and feed composition from the calibration run. Accurate predictions were obtained, thereby illustrating the model's interpolative and extrapolative capabilities. Additionally, the SEMM resin achieved 90% mAb yield, 37% aggregate removal, 29% [Formula: see text] removal, 54% Fab/Fc removal, 100% Fc fragments removal, and a productivity of 72.3 g mAbL×h. Model predictions for these statistics were all within 5%. Simulated batch uptake experiments showed that resin penetration depth was directly related to protein size, with the exception of the aggregate species, and that separation was governed by differential pore diffusion rates. Additional simulations were performed to characterize the dependence of fragment removal on column dimension, load density, and feed composition. Fragment removal was found to be highly dependent on column load density, where optimal purification was achieved below 100 mg protein/mL column. Furthermore, fragment removal was dependent on column volume (constant load mass), but agnostic to whether column length or diameter was changed. Lastly, the dependence on feed composition was shown to be complex. While fragment removal was inversely related to fragment mass fraction in the feed, the extent depended on fragment size. Overall, the results from this study illustrated the efficacy of the SEMM resin in fragment and aggregate removal and elucidated relationships with key operational parameters through model-based characterization.

Application of mechanistic modelling in membrane and fiber chromatography for purification of biotherapeutics - A review

Mechanistic modelling is a simulation tool which has been effectively applied in downstream bioprocessing to model resin chromatography. Membrane and fiber chromatography are newer approaches that offer higher rates of mass transfer and consequently higher flow rates and reduced processing times. This review describes the key considerations in the development of mechanistic models for these unit operations. Mass transfer is less complex than in resin columns, but internal housing volumes can make modelling difficult, particularly for laboratory-scale devices. Flow paths are often non-linear and the dead volume is often a larger fraction of the overall volume, which may require more complex hydrodynamic models to capture residence time distributions accurately. In this respect, the combination of computational fluid dynamics with appropriate protein binding models is emerging as an ideal approach.

Parameter-by-parameter estimation method for adsorption isotherm in hydrophobic interaction chromatography

Hydrophobic interaction chromatography (HIC) is used as a critical polishing step in the downstream processing of biopharmaceuticals. Normally the process development of HIC is a cumbersome and time-consuming task, and the mechanical models can provide a powerful tool to characterize the process, assist process design and accelerate process development. However, the current estimation of model parameters relies on the inverse method, which lacks an efficient and logical parameter estimation strategy. In this study, a parameter-by-parameter (PbP) method based on the theoretical derivation and simplifying assumptions was proposed to estimate the Mollerup isotherm parameters for HIC. The method involves three key steps: (1) linear regression (LR) to estimate the salt-protein interaction parameter and the equilibrium constant; (2) linear approximation (LA) to estimate the stoichiometric parameter and the maximum binding capacity; and (3) inverse method to estimate the protein-protein interaction parameter and the kinetic coefficient. The results indicated that the LR step should be used for dilution condition (loading factor below 5%), while the LA step should be conducted when the isotherm is in the transition or nonlinear regions. Six numerical experiments were conducted to implement the PbP method. The results demonstrated that the PbP method developed allows for the systematic estimation of HIC parameters one-by-one, effectively reducing the number of parameters required for inverse method estimation from six to two. This helps prevent non-identifiability of structural parameters. The feasibility of the PbP-HIC method was further validated by real-world experiments. Moreover, the PbP method enhances the mechanistic understanding of adsorption behavior of HIC and shows a promising application to other stoichiometric displacement model-derived isotherms.

Predictive mechanistic modeling of loading and elution in protein A chromatography

Protein A chromatography is an enabling technology in current manufacturing processes of monoclonal antibodies (mAbs) and mAb derivatives, largely due to its ability to reduce the levels of process-related impurities by several orders of magnitude. Despite its widespread application, the use of mathematical modeling capable of accurately predicting the full protein A chromatographic process, including loading, post-loading wash and elution stages, has been limited. This work describes a mechanistic modeling approach utilizing the general rate model (GRM), the capabilities of which are explored and optimized using two isotherm models. Isotherm parameters were estimated by inverse-fitting simulated breakthrough curves to experimental data at various pH values. The parameter values so obtained were interpolated across the relevant pH range using a best-fit curve, thus enabling their use in predictive modeling, including of elution over a range of pH. The model provides accurate predictions (< 3% mean error in 10% dynamic binding capacity predictions and ∼ 5% mean error in elution mass and pool volume predictions, both on scale-up) for various residence times, buffer conditions and elution schemes and its effectiveness for use in scale-up and process development is shown by applying the same parameters to larger columns and a wider range of residence times.

Model based process optimization of an industrial chromatographic process for separation of lactoferrin from bovine milk

Model based process development using predictive mechanistic models is a powerful tool for in-silico downstream process development. It allows to obtain a thorough understanding of the process reducing experimental effort. While in pharma industry, mechanistic modeling becomes more common in the last years, it is rarely applied in food industry. This case study investigates risk ranking and possible optimization of the industrial process of purifying lactoferrin from bovine milk using SP Sepharose Big Beads with a resin particle diameter of 200 µm, based on a minimal number of lab-scale experiments combining traditional scale-down experiments with mechanistic modeling. Depending on the location and season, process water pH and the composition of raw milk can vary, posing a challenge for highly efficient process development. A predictive model based on the general rate model with steric mass action binding, extended for pH dependence, was calibrated to describe the elution behavior of lactoferrin and main impurities. The gained model was evaluated against changes in flow rate, step elution conditions, and higher loading and showed excellent agreement with the observed experimental data. The model was then used to investigate the critical process parameters, such as water pH, conductivity of elution steps, and flow rate, on process performance and purity. It was found that the elution behavior of lactoferrin is relatively consistent over the pH range of 5.5 to 7.6, while the elution behavior of the main impurities varies greatly with elution pH. As a result, a significant loss in lactoferrin is unavoidable to achieve desired purities at pH levels below pH 6.0. Optimal process parameters were identified to reduce water and salt consumption and increase purity, depending on water pH and raw milk composition. The optimal conductivity for impurity removal in a low conductivity elution step was found to be 43 mS/cm, while a conductivity of 95 mS/cm leads to the lowest overall salt usage during lactoferrin elution. Further increasing the conductivity during lactoferrin elution can only slightly lower the elution volume thus can also lead to higher total salt usage. Low flow rates during elution of 0.2 column volume per minute are beneficial compared to higher flow rates of 1 column volume per minute. The, on lab-scale, calibrated model allows predicting elution volume and impurity removal for large-scale experiments in a commercial plant processing over 106 liters of milk per day. The successful model extrapolation was possible without recalibration or detailed knowledge of the manufacturing plant. This study therefore provides a possible pathway for rapid process development of chromatographic purification in the food industries combining traditional scale-down experiments with mechanistic modeling.

Physics-informed neural networks to solve lumped kinetic model for chromatography process

Numerical method is widely used for solving the mechanistic models of chromatography process, but it is time-consuming and hard to response in real-time. Physics-informed neural network (PINN) as an emerging technology combines the structure of neural network with physics laws, and is getting noticed for solving physics problems with a balanced accuracy and calculation speed. In this research, a proof-of-concept study was carried out to apply PINN to chromatography process simulation. The PINN model structure was designed for the lumped kinetic model (LKM) with all LKM parameters. The PINN structure, training data and model complexity were optimized, and an optimal mode was obtained by adopting an in-series structure with a nonuniform training data set focusing on the breakthrough transition region. A PINN for LKM (LKM-PINN) consisting of four neural networks, 12 layers and 606 neurons was then used for the simulation of breakthrough curves of chromatography processes. The LKM parameters were estimated with two breakthrough curves and used to infer the breakthrough curves at different residence times, loading concentrations and column sizes. The results were comparable to that obtained with numerical methods. With the same raw data and constraints, the average fitting error for LKM-PINN model was 0.075, which was 0.081 for numerical method. With the same initial guess, the LKM-PINN model took 160 s to complete the fitting, while the numerical method took 7 to 72 min, depending on the fitting settings. The fitting speed of LKM-PINN model was further improved to 30 s with random initial guess. Thus, the LKM-PINN model developed in this study is capable to be applied to real-time simulation for digital twin.

Three-dimensional porous wood monolithic columns for efficient purification of spike glycoprotein of SARS-CoV-2

Considerable research has been devoted to finding a cost-effective chromatographic matrix with efficient adsorption and high throughput. Wood exhibits complex micro-network structures that make it a powerful contender for a novel environment-friendly chromatographic matrix material. We demonstrate a novel strategy to manufacture a wood monolithic column, which chemically modified the wood and imported diethyl aminoethyl, diethylamine, and amino groups. This wood monolithic column can maintain fully monolithic column performances and highly selective to spike glycoprotein of SARS-CoV-2 by ion exchange force. The wood monolithic column was evaluated by static adsorption, dynamic adsorption, and frontal analysis. The results showed that the static adsorption capacity of the wood monolithic column with 2-diethylaminoethylchloride hydrochloride for bovine serum albumin was 14.72 mg/g, and the adsorption process was chemisorption. In addition, it retained 80 % adsorption capacity after 110 repeated adsorption-elution cycles.

Coupling of chromatography and precipitation for adjusting acidic variant content in a monoclonal antibody pool

The acidic charge variants (av) of monoclonal antibodies (mAb) are often reported to have reduced therapeutic potency compared with the main (mv) and basic variants (bv), therefore reduction in the av content in mAb pools is often prioritized over reduction in the bv content. In previous studies we described two different methods for reducing the av content, which were based on either ion exchange chromatography or selective precipitation in polyethylene glycol (PEG) solutions. In this study, we have developed a coupled process, in which advantages of simplicity and ease in realization of PEG-aided precipitation and high separation selectivity of anion exchange chromatography (AEX) were exploited. The design of AEX was supported by the kinetic-dispersive model, which was supplemented with the colloidal particle adsorption isotherm, whereas the precipitation process and its coupling with AEX was quantified by simple mass balance equations and underlying thermodynamic dependencies. The model was used to assess the performance of the coupling of AEX and precipitation under different operating conditions. The advantage of the coupled process over the stand-alone AEX depended on the demand for the av reduction as well as the initial variant composition of the mAb pool, e.g., the improvement in the throughput provided by the optimized sequence of AEX and PREC varied from 70 to 600% for the initial av content changed from 35 to 50% w/w, and the reduction demand changed from 30 to 60%.

Engineering protein A ligands to mitigate antibody loss during high-pH washes in protein A chromatography

Protein A chromatography is a workhorse in monoclonal antibody (mAb) manufacture since it provides effective separation of mAbs from impurities such as host-cell proteins (HCPs) in a single capture step. HCP clearance can be aided by the inclusion of a wash step prior to low-pH elution. Although high-pH washes can be effective in removing additional HCPs from the loaded column, they may also contribute to a reduced mAb yield. In this work we show that this yield loss is reflected in a pH-dependent variation of the equilibrium binding capacity of the protein A resin, which is also observed for the capacity of the Fc fragments alone and therefore not a result of steric interactions involving the Fab fragments in the intact mAbs. We therefore hypothesized that the high-pH wash loss was due to protonation or deprotonation of ionizable residues on the protein A ligand. To evaluate this, we applied a rational protein engineering approach to the Z domain (the Fc-binding component of most commercial protein A ligands) and expressed engineered mutants in E. coli. Biolayer interferometry and affinity chromatography experiments showed that some of the Z domain mutants were able to mitigate wash loss at high pH while maintaining similar binding characteristics at neutral pH. These experiments enabled elucidation of the roles of specific interactions in the Z domain - Fc complex, but more importantly offer a route to ameliorating the disadvantages of high-pH washes in protein A chromatography.

Isotherm model discrimination for multimodal chromatography using mechanistic models derived from high-throughput batch isotherm data

In this work, we have examined an array of isotherm formalisms and characterized them based on their relative complexities and predictive abilities with multimodal chromatography. The set of isotherm models studied were all based on the stoichiometric displacement framework, with considerations for electrostatic interactions, hydrophobic interactions, and thermodynamic activities. Isotherm parameters for each model were first determined through twenty repeated fits to a set of mAb - Capto MMC batch isotherm data spanning a range of loading, ionic strength, and pH as well as a set of mAb - Capto Adhere batch data at constant pH. The batch isotherm data were used in two ways-spanning the full range of loading or consisting of only the high concentration data points. Predictive ability was defined through the model's capacity to capture prominent changes in salt gradient elution behavior with respect to pH for Capto MMC or unique elution patterns and yield losses with respect to gradient slope for Capto Adhere. In both cases, model performance was quantified using a scoring metric based on agreement in peak characteristics for column predictions and accuracy of fit for the batch data. These scores were evaluated for all twenty isotherm fits and their corresponding column predictions, thereby producing a statistical distribution of model performances. Model complexity (number of isotherm parameters) was then considered through use of the Akaike information criterion (AIC) calculated from the score distributions. While model performance for Capto MMC benefitted substantially from removal of low protein concentration data, this was not the case for Capto Adhere; this difference was likely due to the qualitatively different shapes of the isotherms between the two resins. Surprisingly, the top-performing (high accuracy with minimal number of parameters) isotherm model was the same for both resins. The extended steric mass action (SMA) isotherm (containing both protein-salt and protein-protein activity terms) accurately captured both the pH-dependent elution behavior for Capto MMC as well as loss in protein recovery with increasing gradient slope for Capto Adhere. In addition, this isotherm model achieved the highest median score in both resin systems, despite it lacking any explicit hydrophobic stoichiometric terms. The more complex isotherm models, which explicitly accounted for both electrostatic and hydrophobic interaction stoichiometries, were ill-suited for Capto MMC and had lower AIC model likelihoods for Capto Adhere due to their increased complexity. Interestingly, the ability of the extended SMA isotherm to predict the Capto Adhere results was largely due to the protein-salt activity coefficient, as determined via isotherm parameter sensitivity analyses. Further, parametric studies on this parameter demonstrated that it had a major impact on both binding affinity and elution behavior, therein fully capturing the impact of hydrophobic interactions. In summary, we were able to determine the isotherm formalisms most capable of consistently predicting a wide range of column behavior for both a multimodal cation-exchange and multimodal anion-exchange resin with high accuracy, while containing a minimized set of model parameters.

Anti-Langmuir elution behavior of a bispecific monoclonal antibody in cation exchange chromatography: Mechanistic modeling using a pH-dependent Self-Association Steric Mass Action isotherm

The objective of this scientific work was to model and simulate the complex anti-Langmuir elution behavior of a bispecific monoclonal antibody (bsAb) under high loading conditions on the strong cation exchange resin POROS™ XS. The bsAb exhibited anti-Langmuirian elution behavior as a consequence of self-association expressed both in uncommon retentions and peak shapes highly atypical for antibodies. The widely applied Steric Mass Action (SMA) model was unsuitable here because it can only describe Langmuirian elution behavior and is not able to describe protein-protein interactions in the form of self-association. For this reason, a Self-Association SMA (SAS-SMA) model was applied, which was extended by two activity coefficients for the salt and protein in solution. This model is able to describe protein-protein interactions in the form of self-dimerization and thus can describe anti-Langmuir elution behavior. Linear gradient elution (LGE) experiments were carried out to obtain a broad dataset ranging from pH 4.5 to 7.3 and from 50 to 375 mmol/L Na⁺ for model parameter determination. High loading LGE experiments were conducted with an increasing load from 0.5 up to 75.0 mg_bsAb/mL_resin. Thereby, pH-dependent empirical correlations for the activity coefficient of the solute protein, for the equilibrium constant of the self-dimerization process and for the shielding factor could be set up and ultimately incorporated into the SAS-SMA model. This pH-dependent SAS-SMA model was thus able to simulate anti-Langmuir behavior over extended ranges of pH, counterion concentration, and column loading. The model was confirmed by experimental verification of simulated linear pH gradient elutions up to a load of 75.0 mg_bsAb/mL_resin.

Mechanistic modeling, simulation, and optimization of mixed-mode chromatography for an antibody polishing step

Mixed-mode chromatography combines features of ion-exchange chromatography and hydrophobic interaction chromatography and is increasingly used in antibody purification. As a replacement for flow-through operations on traditional unmixed resins or as a pH-controlled bind-and-elute step, the use of both interaction modes promises a better removal of product-specific impurities. However, the combination of the functionalities makes industrial process development significantly more complex, in particular the identification of the often small elution window that delivers the desired selectivity. Mechanistic modeling has proven that even difficult separation problems can be solved in a computer-optimized manner once the process dynamics have been modeled. The adsorption models described in the literature are also very complex, which makes model calibration difficult. In this work, we approach this problem with a newly constructed model that describes the adsorber saturation with the help of the surface coverage function of the colloidal particle adsorption model for ion-exchange chromatography. In a case study, a model for a pH-controlled antibody polishing step was created from six experiments. The behavior of fragments, aggregates, and host cell proteins was described with the help of offline analysis. After in silico optimization, a validation experiment confirmed an improved process performance in comparison to the historical process set point. In addition to these good results, the work also shows that the high dynamics of mixed-mode chromatography can produce unexpected results if process parameters deviate too far from tried and tested conditions.

The use of predictive models to develop chromatography-based purification processes

Chromatography is the workhorse of biopharmaceutical downstream processing because it can selectively enrich a target product while removing impurities from complex feed streams. This is achieved by exploiting differences in molecular properties, such as size, charge and hydrophobicity (alone or in different combinations). Accordingly, many parameters must be tested during process development in order to maximize product purity and recovery, including resin and ligand types, conductivity, pH, gradient profiles, and the sequence of separation operations. The number of possible experimental conditions quickly becomes unmanageable. Although the range of suitable conditions can be narrowed based on experience, the time and cost of the work remain high even when using high-throughput laboratory automation. In contrast, chromatography modeling using inexpensive, parallelized computer hardware can provide expert knowledge, predicting conditions that achieve high purity and efficient recovery. The prediction of suitable conditions in silico reduces the number of empirical tests required and provides in-depth process understanding, which is recommended by regulatory authorities. In this article, we discuss the benefits and specific challenges of chromatography modeling. We describe the experimental characterization of chromatography devices and settings prior to modeling, such as the determination of column porosity. We also consider the challenges that must be overcome when models are set up and calibrated, including the cross-validation and verification of data-driven and hybrid (combined data-driven and mechanistic) models. This review will therefore support researchers intending to establish a chromatography modeling workflow in their laboratory.

Design of Experiment (DoE) for Optimization of HPLC Conditions for the Simultaneous Fractionation of Seven α-Amylase/Trypsin Inhibitors from Wheat (Triticum aestivum L.)

Wheat alpha-amylase/trypsin inhibitors remain a subject of interest considering the latest findings showing their implication in wheat-related non-celiac sensitivity (NCWS). Understanding their functions in such a disorder is still unclear and for further study, the need for pure ATI molecules is one of the limiting problems. In this work, a simplified approach based on the successive fractionation of ATI extracts by reverse phase and ion exchange chromatography was developed. ATIs were first extracted from wheat flour using a combination of Tris buffer and chloroform/methanol methods. The separation of the extracts on a C18 column generated two main fractions of interest F1 and F2. The response surface methodology with the Doehlert design allowed optimizing the operating parameters of the strong anion exchange chromatography. Finally, the seven major wheat ATIs namely P01083, P17314, P16850, P01085, P16851, P16159, and P83207 were recovered with purity levels (according to the targeted LC-MS/MS analysis) of 98.2 ± 0.7; 98.1 ± 0.8; 97.9 ± 0.5; 95.1 ± 0.8; 98.3 ± 0.4; 96.9 ± 0.5, and 96.2 ± 0.4%, respectively. MALDI-TOF-MS analysis revealed single peaks in each of the pure fractions and the mass analysis yielded deviations of 0.4, 1.9, 0.1, 0.2, 0.2, 0.9, and 0.1% between the theoretical and the determined masses of P01083, P17314, P16850, P01085, P16851, P16159, and P83207, respectively. Overall, the study allowed establishing an efficient purification process of the most important wheat ATIs. This paves the way for further in-depth investigation of the ATIs to gain more knowledge related to their involvement in NCWS disease and to allow the absolute quantification in wheat samples.