As the market for novel biopharmaceutical products grow, so does the demand for productive, reliable manufacturing processes which can be rapidly developed at a large scale. Chromatography, a foundational process in pharmaceutical manufacturing has seen increased process efficiency and productivity with the extension from traditional batch to continuous counter-current operation, writes Niamh O'Connell.

This has led to increased use of reversed-phase liquid chromatography (RPLC), the gold standard in analytical chromatography, for large-scale preparative separations. Despite its intricacy and ambiguity in retention behaviour, RPLC offers high-selectivity separations which prove useful in the purification of complex biopharmaceutical products.

As the scale, range and market for these products grow there is increased industry focus on accelerated process development and rapid process optimisation.

This can be achieved by integrated process modelling from early process development, allowing for rapid design and process characterization, which then leads to ease of operation, control and optimisation procedures.

However with increased efficiency comes increased complexity, the design, control and optimisation of these chromatography systems is not straightforward. This has led to increased focus on accurate, comprehensive process models to support the process development.

In this project three dynamic chromatography models, the ideal model, the equilibrium dispersive model (EDM) and the general rate model (GRM) were developed to simulate the batch gradient elution of a protein-mixture using a RPLC system. The investigation was then extended to the simulation of continuous chromatography systems, and system performance was predicted for SMB and MCSGP systems.

Model development

RPLC retention mechanisms are well-characterised for small molecule products, however the application to biomolecules is limited due to specific retention characteristics such as slow diffusion, stationary phase accessibility and conformational changes (Jandera et al., 2011). Furthermore biomolecular separations can be easily manipulated by organic modifier concentration, pH or temperature and are almost exclusively run under gradient conditions.

The ideal model and EDM offer a simplified characterisation of a chromatographic process, in reality chromatographic separation involves a myriad of complex hydrodynamic, thermodynamic and kinetic phenomena, as shown schematically in fig. 1 below.

The ideal model assumes infinite efficiency and the EDM assumes that all non-idealities are captured by a lumped parameter. The GRM includes all contributions to column mass transport effects, axial dispersion (molecular & eddy diffusion), external liquid-film mass transfer resistance and intra-particle diffusion, which consists of both pore and surface diffusion (Guiochon et al., 2006).

As both the external and intraparticle mass transfer effects are considered, two mass balance equations are employed; one describing transfer through the bulk column and one describing transfer through the stationary phase pores. The adsorption-desorption kinetics are modelled using the modified multi-component Langmuir isotherm as defined by Melander et al. (1989).

The complex system of equations is solved using gPROMS ModelBuilder and the in-built numerical solvers which are based on a method-of-lines technique. The model consists of several PDEs, and two independent variables time and axial direction, thus spatial discretisation is required. The finite difference and  finite element discretisation methods are employed.

Figure 1:  Schematic of all mass transfer resistances present in chromatographic system adapted from Dizaji (2016)

To investigate model prediction accuracy the operating parameters and column characteristics simulated are identical to those investigated by Gu & Zheng (1999) for the binary separation of human growth hormone (hGH) and recombinant human growth hormone (hGHG120R) using a 250 x 10 mm preparative RP column with C4 packing material (Vydac 214TP510) (Gu & Zheng, 1999).

Batch chromatography

The simulated elution profiles at column outlet for batch gradient separation of hGHG120R & hGH are shown below (fig. 2) for all three models. hGHG120R elutes from the column first, and is the loosely-bound component, thus hGH is the strongly-bound component.

The ideal model and EDM predict highly similar elution profiles, with near-identical peak shape and identical retention time, this is due to the negligible axial dispersion term (Dax = 1.06 x 10-8 m/s).

The elution profile predicted by the GRM is significantly different, having a longer retention time and broader peak shape. The consideration of additional mass transfer resistances results in a 15% slower movement of solute through the column and increased peak broadening effects (Schmidt-Traub et al., 2020).

The models were validated against experimental date published by Gu & Zheng (1999). It was determined that the most accurate elution behaviour is predicted by the GRM model with a <6.5% relative error in the retention time behaviour predicted, compared to >10% for the other models.

Figure 2. Simulated elution profile at column outlet for batch gradient chromatographic separation of hGH & hGHG120R using the ideal model, equilibrium-dispersive model (edm) and general rate model (grm). Dimensionless protein concentration, cb,i is shown on the primary (RHS) axis and organic modifier concentration shown on the secondary (LHS) axis

The model was developed to include automatic cut-point identification (fig. 3), which allows for complete resolution of the components and may then be used to determine the product purity and yield at the column outlet. From initial observation it appears that the results obtained using the ideal model give the highest pure product recovery (100%) however this is not indicative of the real-life scenario, as  accuracy of the model employed may impact the predicted values.

Where the separation is simulated using the most accurate model (GRM) the cut-points are specified by the EDM model, there is zero yield of both solutes (table 2). For preparative separations where high value product must be recovered in high purities (>99%) it is imperative that the fraction collection points are correctly identified during process development.

Figure 3. Elution profile of hGH and hGHG120R simulated using the GRM, including the identified cut-points and associated fraction

Further parameters employed for process productivity analysis include specific productivity, PV,i (mol solute recovered per unit of stationary phase volume) and specific eluent consumption, Eci (unit of eluent volume required per mole of recovered solute).

These parameters are typically used for economic assessments of process feasibility, as stationary phase and mobile phase are the two key contributors to operating costs. Comparing the results obtained through simulation with the ideal model, EDM & GRM different pictures of process productivity are given (table 3).

The GRM indicates relatively low eluent consumption but inefficient utilisation of the stationary phase. However for the Ideal Model/EDM the converse is true, with much higher eluent consumption in particular.

Continuous chromatography (SMB & MCSGP)

The model was then extended to continuous operation, namely simulated moving bed (SMB) and multi-column counter-current solvent gradient purification (MCSGP) which follows from work developed by Brennan (2018) and Maher (2020) for ion-exchange chromatography systems.

Figure 4. Schematic of SMB system, including zone locations, where A is the weakly bound (less retained) component and B is the strongly bound (more retained) component. Figure adapted from Carta & Jungbauer (2010)     

Figure 5. Schematic of 3-column MCSGP system, showing all inlet and outlet flows where A is the loosely-bound component (hGHG120R) and B is the strongly-bound component (hGH)

For both systems the operating conditions were determined using the EDM, and similar to batch operation the performance predicted using the ideal model & EDM showed high product purity and yield, but the GRM predicted low yield and purity. The GRM predicts contamination of the extract stream for both systems, and contamination of the raffinate stream for the SMB system. The product yield predicted is 3.5-3.7 times lower in the SMB system and 1.5 times lower in the MCSGP system.

Figure 6. Average dimensionless concentration in the raffinate stream of the SMB system for all three simulation models

Figure 7.  Average dimensionless concentration in the extract stream of the SMB system for all three simulation models

Thus, where the ideal model and EDM predict high recovery of pure product based on the design and operating parameters specified, the GRM model predicts contamination of both eluent streams and inefficient solute recovery. For the simulated process this would result in significant reductions in process profitability as the extract and raffinate streams would need further processing to recover the solute.

Figure 8.  Average dimensionless concentration in the raffinate (zone 1) stream of the MCSGP system for all three simulation models

Figure 9. Average dimensionless concentration in the extract stream (zone 3) of the MCSGP system for all three simulation models


The work completed in this project provides an argument against using “simplified” models or assumptions for large-scale RPLC development. This is shown to result in incorrect operating parameter identification, deceptive process performance evaluations and ultimately decreased process efficiency across a range of systems.

The significance of model accuracy in process design, development, optimisation and control is shown and as process modelling becomes an integral element of early process design, deliberate consideration must be given to model accuracy.

The use of incorrect model from the outset may negatively impact process feasibility assessments, operation of live processes and ultimately impact overall process performance and product quality.

Future work

It is recommended that further RPLC systems be investigated to determine whether the GRM is the most accurate model for a range of systems. The next system proposed is a small organic molecule solute, as the retention behaviour is expected to be significantly different to that of  a biomolecule.

Furthermore, similar assessments should be performed for other common chromatography processes. The models developed in this project are first-principle mechanistic models which describe the mass transfer and adsorption phenomena by a series of mathematical equations.

As the general principles of adsorption and desorption are identical across all elution chromatography the model is easily applicable to other “true” chromatography methods with small adjustments. This would offer a more wide-ranging judgement on the impact of model selection for simulation of batch and continuous chromatography processes. 

Author: Niamh O’Connell, University College Dublin


1.) Brennan, A (2018) Simulation of Continuous Chromatography. Unpublished BE Research Project. University College Dublin.

2.) Carta, G, Jungbauer, A (2010) Protein Chromatography – Process Development and Scale-Up. Weinheim: WILEY-VCH Verlag GmbH & Co.

3.) Close, EJ et al. (2014) ‘Modelling of industrial biopharmaceutical multicomponent chromatography’, Chemical Engineering Research and Design, 92(7), pp. 1304-1314.

4.) Degerman, M, Jakobssen, N, Nilsson, B (2007) ‘Modelling and optimisation of preparative reversed-phase liquid chromatography for insulin purification’, Journal of Chromatography A, 1162, pp. 41-49.

5.) Dizaji, NM (2016) Minor Whey Protein Purification Using Ion-Exchange Column Chromatography, Unpublished PhD Thesis. The University of Western Ontario.

6.) Gu, T et al. (1995) ‘Purification of a pyrogen-free human growth hormone antagonist’, Biotechnol. Bioeng., 48(5), pp. 520-528.

7.) Guiochon, G. et al. (2006) Fundamentals of Preparative and Nonlinear Chromatography 2nd ed. Amsterdam: Elsevier.

8.) Jandera, P, Kučerová, Z, Urban, J (2011) ‘Retention times and bandwidths in reversed-phase gradient liquid chromatography of peptides and proteins’, Journal of Chromatography A, 1218, pp. 8874-8889.

9.) Maher, R (2020) Productivity Analysis of Various Batch and Continuous Multi-Column Counter-Current Chromatographic Processes Simulated in gPROMS. Unpublished BE Research Project. University College Dublin.

10.) Melander, WR, Rassi, ZR, Horváth, C (1989) ‘Interplay of hydrophobic and electrostatic interactions in biopolymer chromatography : Effect of salts on the retention of proteins’, Journal of Chromatography , 469, pp. 3-27.