Establishing Process Design Space for a Chromatography Purification Step: Application of Quality-by-Design Principles

Published on: 
BioPharm International, BioPharm International-11-01-2015, Volume 28, Issue 11
Pages: 28–35, 45

This case study reviews how quality-by-design principles can be implemented in an intermediate chromatography purification step that uses cation-exchange chromatography.Abstract


This case study reviews how quality-by-design principles can be implemented in an intermediate chromatography purification step that uses cation-exchange chromatography. The drug substance is a protein molecule expressed in microbial host cells. The purification process involves a capture step and an intermediate purification step, followed by a polishing step.  

Since FDA introduced the risk-based approach as a means to improve the regulation of pharmaceutical manufacturing and product quality (1), the concept of quality by design (QbD) has been gradually implemented into biologics development processes. In contrast to the traditional quality-by-testing (QbT) mentality, the QbD approach requires that quality “be built in by design” based on the knowledge of the product and process (2). It is a holistic approach that involves a thorough understanding of the relationship between process inputs and process outputs so that any potential risks affecting product quality can be mitigated. The aim is to ensure that the final drug product meets the pre-established quality requirements defined by a set of critical quality attributes (CQAs), which will ultimately determine its clinical performance.


This article presents a case study in which a QbD approach is applied to establish the design space for an intermediate chromatography purification step. The drug substance is a protein molecule expressed in microbial host cells, which is purified by a process consisting of capture, intermediate purification, and polishing steps. The intermediate purification step is achieved by cation-exchange chromatography (CEC) (see Figure 1). This step removes both process-related and product-related impurities. The product-related impurities are inactive variants that are closely related to the drug substance; a delicate design of conditions is, therefore, required to separate the product-related impurities from the drug substance. The objective of development is to achieve maximum purity and recovery at this intermediate purification step.

Experimental approach
Screening experiments were performed to identify the resins and chromatography conditions that separate the impurities from the drug substance molecule. The process was developed, scaled-up, and transferred to the manufacturing group to perform production batches. A scale-down process model was developed and qualified, and baseline studies with small-scale runs were performed to generate a history of step performance, which helped to establish the performance acceptance criteria for step yield and purity.

Process characterization was performed based on a series of design-of-experiment (DOE) studies. A screening design was employed to identify the ranges of parameters with potential risks. The parameters identified to have significant impact were further studied using a full factorial design. The results refined the parameter ranges based on the responses of purity and step recovery. Least square fit models of the results were used to perform Monte Carlo-based simulations to identify the process capability and the potential number of failures using the JMP software version 9.0 (3). Based on the regression model and Monte Carlo simulation results, targets and ranges for the parameters were proposed for this intermediate purification step.

Materials and methods
Resins used for screening were obtained from suppliers such as GE Healthcare, Toso Bioscience, Bio-Rad, and Applied Biosystems. Chemicals used in buffers were purchased from Mallinckrodt, J.T. Baker, or EMD.

Chromatography runs were performed on the AKTA explorer (or purifier) and custom-built chromatography skid. Protein concentration was measured with a DU 720 UV/vis spectrophotometer (Beckman Coulter). Buffer pH and conductivity were measured with a pH/conductivity meter (Mettler Toledo). Product yield was calculated based on enzyme-linked immunosorbent assay (ELISA) results, and purity was measured by reverse-phase high-performance liquid chromatography (RP–HPLC). Both ELISA and RP–HPLC were qualified for their intended purposes.



Results and discussionScreening and early development
The screening experiments identified a POROS resin to separate the product-related impurities from the drug substance molecule. With low pH and low salt concentration, the capture step eluate was suitable for loading directly onto the CEC column. The dynamic binding capacity at 10% breakthrough (DBC10) was determined to be approximately 25 g/Lbed by frontal analysis, which was well above the load under the processing conditions.

A series of linear salt gradient elution studies were performed to screen the elution profiles of bound proteins from the column. The parameters scouted included a linear gradient slope (or column volumes [CVs] from 0–1 M sodium chloride [NaCl]) and a linear flow rate (100–500 cm/hour). The elution profiles are illustrated in Figure 2. Peak 1 contained the product-related impurities and Peak 2 was the drug substance peak. The linear gradient runs helped to elucidate the effect of gradient slope and flow rate on resolution (Rs) between the product-related impurities and drug substance peaks (see Figure 3). The Rs is calculated from the ratio of the difference between the peak retention volumes (VR) to the average of the peak base width at 10% of peak height (wb). The results suggested that resolution increases with shallower gradient slope, but decreases with the increase of flow rate. A resolution of 1.5 indicates baseline separation, which requires greater than 10 CVs in 0–1 M NaCl gradient and allows a linear flow rate of up to 500 cm/hour.

Based on the linear gradient studies, the washing step and elution conditions were developed. The developed chromatography scheme consisted of equilibration, loading, wash 1, wash 2, elution, post-elution cleaning, and decontamination steps, as illustrated in Figure 2. The product-related impurities were desorbed from the column as the wash 2 peak, and the product molecule was eluted as the CEC eluate. The strongly bound impurities, mostly process-related impurities, were eluted under high salt conditions, shown as post-elution clean peaks in Figure 2. The column was decontaminated with 0.1 N sodium hydroxide (NaOH) at the end of chromatography.

Scaling down the process model and baseline runs
To prepare for the process characterization studies, a small-scale process model with a scale-down factor of 1/25 was established and qualified to match the production-scale batch performance. A baseline study was performed to assess the performance variation and thereby, establish performance baselines under small-scale purification conditions. A total of 15 small-scale runs were performed under target process conditions. The CEC results of the 15 runs gave a mean step yield by ELISA of 70%, with a 6σ range of 53–87%, and a mean purity by RP–HPLC of 96%, with a 6σ range of 94–97%.

Process characterization strategies
The primary objective of this intermediate purification step was to remove closely related product-related impurities from the capture step eluate. Risk assessment was performed using the failure mode and effect analysis (FMEA) on the parameters illustrated in the fish-bone diagram (see Figure 4); the parameters with potentially high risks on the recovery and/or purity step were selected for a DOE-screening study. Based on the baseline study results and risk assessment, the acceptance criteria of step yield and purity were set at ≥ 65% and ≥ 95%, respectively, as the goal for process characterization studies.

Among the parameters studied, those that had a relatively large effect on recovery and/or purity and were relatively difficult to control were selected for a higher resolution follow-on DOE study, which resolved the main effects, interactions, and quadratics. Monte Carlo simulations were performed under different scenarios, using proposed parameter ranges. At the same time, full-scale purification was performed at the manufacturing site to accumulate historical data, which were compared to the Monte Carlo simulation results. The process capability and failure rate, with proposed parameter ranges and pre-defined recovery and purity desirability, were calculated, which helped to establish the design space of processing conditions. The process characterization workflow is illustrated in Figure 5. This article focuses on DOE screening and follow-on DOE studies; the integration DOE study will be reported in a future article.



DOE screening study
A two-level, eight-factor, fractional factorial design was performed to screen the main effects of protein load, loading pH and conductivity, wash 2 pH and conductivity, elution pH and conductivity, and flow rate. The parameters and test ranges are summarized in Table I.

A total of 14 small-scale runs were performed, which included 12 experimental runs and two control (parameters set at the center points) runs. The experiments were considered valid, as the control runs met the acceptance criteria on step recovery and purity, based on previous baseline studies using qualified scale-down process model. The results were analyzed using JMP software, which indicated that at the ranges tested, wash 2 pH, wash 2 conductivity, and elution conductivity had the largest effect size and the effects were statistically significant. Based on the results, the parameters were scored based on their effect size, response importance, and control factor, and those with high scores were selected for a follow-on DOE study, as summarized in Table II. The ranges of other parameters were set at the test ranges.

Follow-on DOE study
A two-level, four-factor, full factorial DOE study was performed to further characterize the effect of protein load, wash 2 pH, wash 2 conductivity, and elution conductivity. The test ranges for these parameters are summarized in Table III. The DOE responses are step yield (%) by ELISA and purity (%) by RP–HPLC. The linear regression model was refined by backward stepwise regression taking away the model terms with statistically insignificant effect (p-value greater than 0.05).

The refined regression model, illustrated in Figure 6, explained 99% of the variation of the step yield (R2 = 0.99, F (12,5) = 75.59, p < 0.0001) and 96% of the variation of the step purity (R2 = 0.96, F (12,5) = 11.28, p = 0.0074). The R2 value depicted the goodness of fit of the model, and the adjusted R2 was a modification of R2 that adjusted for the number of explanatory terms in the model. In the models for step yield and step purity, both the R2 (0.99 and 0.96, respectively) and the adjusted R2 (0.98 and 0.88, respectively) are high, indicating good model fitting with statistical significance (p-value of <0.0001 and 0.0074, respectively, at α = 0.05)(see Figure 6).

The multi-collinearity was assessed by the variable inflation factor (VIF). In the constructed regression models, all the terms have VIF values around 1 (see Tables IV and V), indicating lack of over-fitting. The characteristics of the regression model are summarized in Tables IV and V, which include unstandardized coefficient (B), standardized coefficient (β), t-ratio, p-value, and VIF of the independent variables.

The regression model and available data from actual manufacturing batches were used in the Monte Carlo simulation. Due to limited data from production batches, Monte Carlo simulations were performed under different scenarios to assess how well the simulated model outputs meet the pre-established acceptable response ranges. A simulation example is illustrated in Figure 7.


The simulation was done under the following scenario:

  • The current parameter setting was set as target

  • The proposed parameter range was treated as the 6σ range, thus, the parameter standard deviation (SD) was calculated by dividing the parameter range by 6.


Based on 10,000 simulated runs, the predicted step yield was 90.50 ± 5.06%, and the predicted step purity was 96.52 ± 0.33%, with failure rates of 0.24 ppm and 1.57 ppm and process capability index (Cpk) of 1.68 and 1.55, respectively (see Table VI).




The effect size and margin of the parameters, as well as their interactions and quadratics, on step recovery and purity are summarized in Table VII, which shows that Wash 2 pH and its interaction with Wash 2 conductivity have the biggest effect on step recovery, while protein load and the three-level interaction of protein load, Wash 2 pH, and Wash 2 conductivity have the biggest effect on step purity.



Figure 8 illustrates a contour plot to show the effect of Wash 2 pH and conductivity on the step yield and purity under different protein loads. The proposed parameter target and ranges were labeled, and the design space was represented as a rectangle area in green, located within a white area surrounded by pink and blue. The white area represented the desired outputs of the CEC step that meet both the step yield and step purity requirements. It is clear that the Wash 2 pH and conductivity need to be tightly controlled to achieve the desired step performance under different protein load conditions.

The process conditions were developed for a CEC intermediate purification step to separate product-related impurities from the drug substance. The process was characterized by applying a QbD approach using risk assessment, DOE screening, and follow-on DOE studies. The proposed process conditions would be able to achieve the desired performance requirements for this purification step, based on the Monte Carlo simulation results. It is necessary to demonstrate the proposed process conditions at production scale and across the entire purification process. This work will be reported in a future article.


1. FDA, Pharmaceutical cGMPs for the 21st Century-A Risk-Based Approach: Final Report (Rockville, MD, September 2004).
2. A.S. Rathore and R. Mhatre, Quality by Design for Biopharmaceuticals: Principles and Case Studies (John Wiley & Sons, Hoboken, New Jersey, 2009).
3. SAS Institute Inc., “JMP Statistics and Graphics Guide,” accessed Oct 5, 2015.

The author would like to thank the Biopharmaceutical Process Sciences Group for performing the experiments and the Biopharmaceutical Analytical Sciences Group for testing the samples.

About the author: Hui Xiang;


Article submitted: May 19, 2015.
Article accepted: June 22, 2015.

Article Details

BioPharm International
Vol. 28, No. 11
Pages: 28–35, 45

Citation: When referring to this article, please cite it as H. Xiang, " Establishing Process Design Space for a Chromatography Purification Step: Application of Quality-by-Design Principles," BioPharm International 28 (11) 2015.