Quality by Design in the CMO Environment

December 1, 2007
Susan Cook

,
L. Rochelle Bazemore, PhD

,
Katherine A. Patton

BioPharm International, BioPharm International-12-01-2007, Volume 20, Issue 12
Page Number: 40–45

How the authors used design of experiments and quality by design principles to develop a hydrophobic interaction chromatography step.

ABSTRACT

Incorporating quality by design into process development in a contract manufacturing environment can be challenging. In this case study, design of experiments was used to identify the key and critical operational parameters and their targets for hydrophobic interaction chromatography (HIC) used in the purification of an antibody. Six parameters that typically influence HIC performance were assessed for their effect on elution profile, product yield, and resolution between product and product-related impurities. Statistical software was used to design experiments and analyze the data. The data established wide ranges for some parameters and tighter controls were required for others that had a significant effect on product yield. Some of the parameters also influenced the elution profile and the resolution of product from product-related impurities. These bench-scale studies established a design space that delivered product with reproducible quality and yields during manufacturing.

Process development in a contract manufacturing organization (CMO) is a delicate balance between various and often conflicting demands. The CMO strives to deliver defined, quality processes and product to a client, while meeting predetermined budget and timeline constraints. These limits are particularly narrow for early phase clinical programs before the candidate drug has proven its value. Although the specific deliverables for the CMO are dependent on the client requirements and clinical phase of the product, all campaigns seek to deliver robust processes that satisfy client and regulatory requirements.

The concept of quality by design is not new. It was best explained by Juran1 as a means of efficient quality control that has been applied to a wide range of goods and services. Quality by design was introduced to pharmaceuticals in the International Conference on Harmonization (ICH) Q8 guideline, approved in May 2006, which provides guidance for pharmaceutical process development.2

The aim of pharmaceutical development is to design a manufacturing process to consistently deliver a quality product. The knowledge gained from development studies and manufacturing experience provide scientific understanding to support the establishment of design space.2 The immediate benefit of a well-established design space is knowledge of process parameters that need to be monitored or controlled, which results in a robust manufacturing process.2 Product quality attributes can be accurately and reliably predicted over the design space.3 This allows continuous improvement, which, in turn, improves efficiency by optimizing a process and eliminating wasted efforts in production.4 Furthermore, the process can be optimized using less time, effort, and cost because changes in the design space do not require regulatory pre-approval.

Developing a design space requires extensive process characterization studies, and may seem an unwieldy task for a CMO, given the cost and time constraints. Design of experiments (DOE) is an efficient experimental approach that can reduce the number of runs required as compared to testing each parameter individually, and gives information on the interaction between parameters.4 In this case study, DOE was used to determine the key and critical operational parameters and their targets for a hydrophobic interaction chromatography (HIC) step in the purification process of a monoclonal antibody. Critical parameters are variables that are known to affect product quality and are difficult to control. Key parameters can also affect product quality, but are well-controlled and present less risk than critical parameters. A model for the process was created that generated targets for the key and critical parameters. Using this information, four 2,000-L scale runs were performed. Analysis of the runs demonstrated the predictive power of the bench-scale model.

Figure 1

EXPERIMENTAL APPROACH

The HIC resin used for this case study was Octyl Sepharose 4 Fast Flow (GE Healthcare, catalog number 17-0946). Octyl Sepharose demonstrated separation of different species with a reverse salt gradient. Protein A capture of the antibody produced a partially purified feed stream with three major impurities, referred to in this paper as species A, B, and C. Figure 1 shows size-exclusion high-performance liquid chromatography (SEC-HPLC) of a side fraction from the Octyl column enriched for the three product variants. On Octyl sepharose, species B co-elutes with the beginning of the product peak, and species A and C elute toward the tail of the product peak (Figure 2). Although there is overlap, initial screening experiments indicated that certain conditions increased the resolution between the species.

Figure 2

The DOE looked at six parameters: column bed height, protein loading density, pH, temperature, flow rate, and salt concentration of the load. Dependent variables analyzed were yield, purity, elution gate, and resolution between product and impurity peaks. JMP 6 software (SAS Institute) was used to design the experiments and analyze the data. Because of the number of variables (6) and limited material and time, a fractional factorial design with two center points was chosen (Table 1). It is sufficient to identify main effects and some two-factor interactions.

Table 1

Three 1.0-cm diameter jacketed columns were packed with Octyl resin to represent a tall (27.2 cm), short (14.6 cm), and center-point (22.5 cm) bed height. Packing efficiency tests were run on the columns to ensure consistency between the columns. The columns were sanitized with 0.5 M NaOH and stored in 20% ethanol before the first run and after each run to minimize the possibility of column fouling. Equilibration and elution buffers were prepared at pH 6.0, 6.5, and 7.0.

Water baths were set up to recirculate water at the specified temperature through the jacket of the column. In addition, a stainless steel coil was plumbed to the inlet of the column and submerged in a jacketed beaker of water, also connected to the recirculating bath to ensure temperature control. The chromatography systems used were GE Healthcare ÄKTA Explorers with inline flow restrictors.

Load material was produced by processing clarified harvest material from two 10-L cultures over a Protein A capture column. The Protein A eluate was adjusted to pH 6.0 and frozen in aliquots at –70 °C. For each chromatography run, thawed Protein A eluate was filtered and diluted to the appropriate ammonium sulfate concentration with 2 M (NH4)2SO4 at pH 6.0, 6.5, or 7.0. The protein solution was filtered again and UV absorbance at 280 nm was measured to calculate the total protein concentration of the the load volume required for each run. The load was equilibrated in the water bath to attain the specified temperature before starting the chromatography run.

Each run was performed using the method outlined in Table 2 while incorporating the specified pH, temperature, and flow rate parameters from the DOE scheme (Table 1). One column-volume (CV) fractions were collected throughout the elution with an automated fraction collector. Absorbance at 280 nm was used to determine concentration, and SEC-HPLC was used to determine purity for each fraction.

Table 2

The SEC-HPLC assay was used to determine the percent of product and species A, B, and C present in a sample. A Tosoh G3000SWXL column (catalog number 08541) with a guard column was used to analyze Octyl fractions. The mobile phase used was 100 mM sodium phosphate, 200 mM sodium chloride, pH 6.5. The target load was 30 μg of protein per injection, and total run time was 35 minutes. A gel filtration standard was used (Bio-Rad, 151–1901) to verify system suitability.

RESULTS

Analysis of DOE Runs

For each run, SEC-HPLC was performed on the load, unbound, gradient elution, and strip fractions to give a purity profile of each fraction. Additionally, the concentration of each of these fractions was determined by absorbance at 280 nm. Theoretical pools of the gradient elution fractions were then calculated using this data. The first criterion for pooling was that the pool must contain less than 2% of species A, because downstream processes did not demonstrate removal of this impurity. Furthermore, the pool must contain 10% species B and species C combined. In some runs, a pool could not be made to satisfy these requirements, so the yield for those runs was 0%. The yield calculated for the theoretical pools is product-specific, i.e., product in pool ÷ product in load x 100. Table 3 shows the raw data from DOE run 2.

Table 3

An absorbance threshold was used as an indicator of where in the salt gradient protein elution occurs, and was expressed as column volumes. The start of protein elution was defined as the point where absorbance at 280 nm increased above 50 mAU. A value of 1 would indicate that the 50 mAU threshold was surpassed at 1 CV into the gradient. A negative value indicates that elution began before the start of the gradient, that is, during post-load wash. The eluate was not collected until the gradient phase of the method, so any protein eluted during the wash was lost, reducing the yield.

Table 4

To determine resolution between two peaks, the number of CVs between the two peaks was divided by the average width in CV of the peaks being compared. A resolution value of 0 indicates no separation, and a value of 1 indicates complete separation between the species. JMP 6 statistical software was used to analyze the data from the DOE using seven input variables (the original six in the design plus percent species A in the load) and the output variables of yield, elution start, and resolution. Effect screening was used to identify the main effects with a 95% confidence level. Table 4 contains the output variables for each run in the DOE. Figure 3 is the prediction profiler generated by JMP 6 software, which shows trends as parameters were varied. Table 5 is a summary of the analysis with JMP 6.

Figure 3

The percent species A in the load was not included in the original design. However, it was identified as a variable after analysis of fractions by SEC-HPLC. Upstream development was concurrent with downstream development to save time so there was some variability in the feed stream. This turned out to be a boon: adding the parameter of percent A in the load to the analysis revealed additional correlations. Their validity was supported by the large increase in the adjusted R2 value for the models for percent yield (0.30 to 0.74), resolution of species C (–0.35 to 0.65), and start of elution (0.62 to 0.89).

The study results showed the Octyl chromatography to be quite robust. It was insensitive to pH from 6.0 to 7.0, linear flow rate from 100 to 180 cm/hr, and bed height from 15 to 25 cm. The bed height and flow rate combine for a range of residence time from 5 to 15 minutes with no impact.

Increasing loading density, as expected, reduced the resolution of species A from product. Less resolution, in turn, resulted in fewer fractions pooled and thus lower yield. There was a strong correlation between high loading density and early elution. Low ammonium sulfate in the load was correlated with early elution and improved resolution of product from species A, possibly because the product tended to elute early. In contrast, species A eluted at about the same point in the gradient regardless of loading conditions. The trends discussed above, which are highlighted in yellow in Table 5, were significant whether or not the percent species A in the load was included in the analysis.

Table 5

Additional trends were revealed with the inclusion of percent A in the load. Higher temperatures favored better yield, presumably because of the general phenomenon observed with HIC resins of improved binding of protein at higher temperatures. There was also a correlation between higher temperature and later elution, probably caused by the same mechanism. Larger amounts of species A in the load led to lower yield because fewer fractions could be pooled. The correlations of species A with early elution and less resolution of species C do not have readily apparent explanations. They may be collinear with other factors. No two-factor interactions were revealed in the analysis of the data. The targets for the key variables determined for this study were loading density 10 g/L of resin, 1.1 M ammonium sulfate in the load, and temperature ≥18 °C.

None of the variables tested had any impact on resolution between product and species B. Octyl Sepharose Fast Flow may be incapable of better separation, or the controlling parameter has yet to be discovered. Resolution of species C only seemed correlated with the percent A in the load. These observations reaffirm that the primary role of the Octyl column is to remove species A; other species may be reduced, but not removed, except at a high cost to yield. However, as long as the key variables are controlled, the prediction profiler generated by JMP 6 calculates that the resolution of species A should vary only between 0.53 and 0.59 for 6.9–12.0% A in the load for the Octyl column.

Figure 4

ANALYSIS OF MANUFACTURING RUNS

Following development of the purification process, four manufacturing runs were conducted at the 2,000-L culture scale. The bed volume of the Octyl column was 40.5 L. Figure 4 is an overlay of the UV absorbance chromatograms of the Octyl step at manufacturing scale, and Table 6 shows the major input and output variables for the runs. Column dimensions and temperature were constant, and flow rate was <150 cm/hr to maintain pressure less that 2 bar.

Table 6

Figure 4 and Table 6 show the consistency of the runs. The yield and purity were 66% ± 6% and 92% ± 2%, respectively. Those are narrow ranges, particularly for a product in its first manufacturing campaign. The only differences visible in the chromatograms in Figure 4 are the start of the elution and the peak height. The one output variable that showed high variance was resolution of species B from the main product, which confirmed one of the main predictions of the bench-scale model. Figure 5 shows the distribution of species during the elution gradient for one of the manufacturing runs, demonstrating the comparability of the bench-scale model and manufacturing scale runs.

Figure 5

The key and critical parameters for the manufacturing runs were analyzed using JMP 6. The caveat is that the number of data points and the breadth of the ranges of the data are insufficient to yield any statistically significant correlations. However, the trends shown in Figure 6 seem to confirm some of the main points of the model derived from the DOE. Loading density and conductivity, used here in place of ammonium sulfate concentration, appeared to result in earlier elution as predicted. Increasing conductivity resulted in less resolution of species A from the main product, and more species A in the load had a negative impact on yield.

Figure 6

CONCLUSION

A contract manufacturing organization must maximize efficiency during process development to gain the knowledge necessary to design quality manufacturing processes and still meet timelines and budget. In this case study, the key and critical parameters for Octyl chromatography and their ranges had to be determined for the first manufacturing campaign at the 2,000-L scale.

Several steps were taken to work within the limitations of the contract while extracting the most information possible to start building the design space for the process. A fractional factorial design was used for the DOE, which required only 18 chromatography runs to identify the main effects of seven process variables. Upstream and downstream development overlapped to condense the timeline. The resulting variability in impurities in the harvests revealed a critical parameter. Small-diameter columns were used to minimize load material and raw material needs for the experiments. Efficiency was further increased by automated methods for purification and analysis; running the instruments overnight reduced labor charges to the client. Use of JMP software provided objective analysis according to accepted statistical concepts, and thus provided a scientific basis for setting parameters.

The non-critical parameters that were identified were pH and flow rate or residence time. Key parameters were temperature, loading density, and the concentration of ammonium sulfate in the load. The critical parameter for this case study was the percent of species A present in the load. It affected yield and separation of product from impurities and was not well controlled at that point in development. The Octyl step was designed to consistently remove amounts of species A between 7 and 12% by controlling the key parameters. Meanwhile, upstream conditions were refined so that the percent of species A in the load during the manufacturing runs was 9.5% ± 0.6%. Thus percent A was down-graded from a critical to a key parameter.

The results of the manufacturing runs showed that the process was robust when the key parameters identified in process development were controlled. The small variation among the runs coincided with the results of the bench-scale experiments, demonstrating the predictive power of the model.

This work shows how quality by design principles can be applied during process development in the CMO environment. Design of experiments is a cost-effective method for developing robust processes and providing information for expanding the design space. Both the CMO and the client reap the benefits of consistent yields and quality as well as the high success rate that results from a well-characterized process for the production of biopharmaceuticals.

Susan Cook is a scientist I, Katherine A. Patton is a scientist I, and L. Rochelle Bazemore, PhD, is a staff scientist, all at Diosynth Biotechnology, NC, 919.388.5700, susan.cook@diosynth-RTP.com

REFERENCES

1. Juran J. Quality by Design. Mankato, MN: The Free Press; 1992.

2. US Food and Drug Administration. Guidance for industry: Q8 pharmaceutical development. Rockville, MD; 2006 May.

3. US FDA. Guidance for Industry: PAT—a framework for innovative pharmaceutical development, manufacturing, and quality assurance. Rockville, MD; 2004 Sept.

4. The PAT Team and Manufacturing Science Working Group. Innovation and continuous improvement in pharmaceutical manufacturing—Pharmaceutical cGMPs for the 21st century. Rockville, MD; Aug 2002.