How To Use QbD Software To Improve An Existing Identification Method

By Ashraf Madian, Ph.D., Sr Group Leader, Pfizer Global Technology Services Biomanufacturing Sciences, and Shen Chen, Ph.D., Director, R&D, Lisa Cherry, Ph.D., Pharmaceutical Sciences Manager, Irish Gibson, Ph.D., Associate Research Scientist, all three from Pfizer CentreOne

A new approach to applying the USP monograph method for the identification of human serum albumin

What can you do when the identification method defined by the USP monograph does not provide clear guidance? In our case, we decided to improve the method with a Quality by Design (QbD) approach using QbD software designed to facilitate the development of the most desirable operating conditions. Our new test method meets the USP requirements, is robust, and reduces the run time by 84 percent.

How did we enhance the USP monograph approach?

Human serum albumin (HSA) is used as an excipient in several formulations. Prior to using it, we wanted to perform an identification test. The USP identification method for protein is accepted as the industry standard, but the method does not provide sufficient instruction for the many steps of the analysis, and modernization of the method by the USP is an ongoing effort.

According to the USP monograph for HSA, we needed to perform a peptide map identification test, which involves the molecule being digested with the enzyme trypsin, which cleaves proteins into peptide fragments. The peptides are then separated and identified using high-performance liquid chromatography (HPLC).

We attempted to follow the USP monograph testing approach but the chromatography results concerned us for two reasons:

• The method lacked robustness. Two different analysts ran the method and produced different results.
• The very long method run time of 120 minutes was not efficient and therefore impractical for a commercial manufacturing environment.

Upon investigation, we found even small changes in the organic amount of the mobile phase caused the chromatographic variability. We wanted a much more thorough understanding of the method and a quicker method as well. We decided to use a QbD approach to build quality into the method.

How do you build quality INTO a method?

With the goal of developing a proactive and risk-based approach, we started by asking questions about the method focusing on two areas: chromatography and peptide mapping.

1) Chromatography

• How can we reduce the run time?
• How can we achieve sufficient separation or resolution?
• How can we improve reproducibility?

2) Peptide mapping

• How should we choose peaks, which represent peptides, for evaluation? The USP chapter for peptide mapping provides only very general criteria and high-level instructions on methodology. For example, for a chromatogram, it states “acceptable resolution between peptide peaks”. Regarding the selection of peptide peaks for evaluation, it states to perform the “comparison between the sample and the USP reference standard.”
• Which regions of the molecule should we look at for identification? Because the molecule did not have a specific region of interest, we intentionally chose to evaluate peaks that spanned all three domains of the protein.

Next, we assessed and documented risks upfront.

• We identified risks in the following categories: material, method, instrument, measures, environment, and people.
• We described and made suggestions for risk mitigation upfront. Several of the method-related risks such as column selection, column temperature, and gradient time were optimized in the designed experiments.

How did we reduce the chromatography run time?

To reduce the run time, we used the Waters® Columns Calculator to convert the HPLC method to an ultra performance liquid chromatography (UPLC) method. The original HPLC method required 120 minutes of run time. Converting from HPLC method to UPLC reduced the run time by 84 percent to 19 minutes.

The USP monograph criteria state that the standard and sample chromatographs should match. Therefore, we compared the profile of the standard to the sample as shown in figure 1. The comparison of the peaks selected for evaluation met our acceptance criteria.

Figure 1

How did we decide which chromatography peaks to evaluate?

At this point, the shorter UPLC method looked viable. We realized we would need to thoroughly examine the method capabilities for identifying the peptides of interest. When changing parameters to use the UPLC method, the peptides may elute differently. We decided to use a second powerful separation method called mass spectrometry in combination with the UPLC detection for two reasons:

• We wanted to consistently examine the peptides of interest throughout development. While assessing method conditions during developement such as solvent, temperature, and column, the peptides might scramble, eluting at different retention times for different method conditions.
• We also wanted to know the peptide(s) sequence to ensure we evaluated peaks spanning the entire range of the molecule.

Figure 2 shows the peaks chosen for evaluation. The selected peaks met the initial criteria we established:

• All three domains of the HSA protein are included.
• Selected peaks contained a single peptide and are well resolved.
• The seven peaks selected also allow two resolutions per peak to be evaluated, a practical evaluation approach.

Figure 2

How to define and optimize the key method parameters with QbD software?

QbD is a systematic approach to method development rooted in statistical knowledge and complicated calculations. In the past, QbD methods required complex statistical expertise and a significant amount of time.

That has changed in recent years as user-friendly software has made QbD implementation practical for non-statisticians. The FDA (U.S. Food and Drug Association) and EMA (European Medicines Agency) encourage its use in the industry and are currently developing review criteria for evaluation of QbD-based analytical methods.

In the following sections, we demonstrate how to use QbD software for method development for peptide mapping of HSA.

Phase 1: Initial Chemistry Screening

During the first phase, we evaluated different column types and gradient times. We chose three different C18 columns and two pore sizes, 130 and 300 angstrom, for evaluation. All three columns had a length of 100 mm. A 50 mm column would have been too short, compressing the chromatography with the complex mixture. A 150 mm column was not used, because the run time would have been significantly extended, and therefore would have contradicted the method goal to reduce run time.

We evaluated two different column chemistries, BEH (ethylene bridged hybrid) and CSH (charged surface hybrid). BEH particle columns are designed for pH stability. CSH particle columns have an applied charge on the surface of the resin and can help with improved peak shape for basic compounds under acidic, low ionic conditions.

With the QbD software (Fusion QbD® software product suite) guiding us through design of experiments or DOE, we examined the column and gradient time. The inputs are in figure 3 where the gradient time is listed as “variable” and ranges from 9.49 minutes to 13.99 minutes. The QbD software performs a statistical sampling on the variables of interest. This means the conditions to evaluate are chosen over a range and do not necessarily cover every possible combination. The experimental design output resulted in our examining three columns with five gradient times for a total of 16 runs, detailed in figure 4. Resolution was chosen as the main criteria for examining the experiments, and two resolutions per peak were calculated.

Figure 3

Figure 4

Figure 5 below shows the results from the experiments. The column type is listed at the top. The peptide being evaluated is listed on the left. The x-axis on each graph is the gradient time, and the y-axis on each graph is the resolution. Thus, a flat line (i.e. zero slope) indicates gradient time does not adversely affect resolution. An increasing or decreasing straight line or curved line indicates time does affect resolution.

Figure 5

Column Type = UPLC peptide BEH C18, 300A

Column Type = UPLC peptide BEH C18, 130A

Column Type = UPLC peptide CSH C18, 130A

The software produces a significant amount of data and we have minimized the amount shown for this article. Figure 6 below gives you an idea of the type and amount of data obtained for each peptide. From this data, the software provides the best answer for the variables examined. The column identified by the software for the best resolution with least adverse effect was the BEH C18 column with 300A pore size.

Figure 6 below demonstrates what an effective resolution looks like on a chromatogram. It Illustrates an enlarged section of the chromatograms for separation of two peptides with the three columns. T33 and T80-T81 were two of the peptides that were troublesome during this part of the development. Notice how they are completely resolved, or separated, with the chosen column, whereas they are not separated with the other two columns. Separation is necessary for proper identification of the peptide fragments.

Figure 6

Phase 2: Method optimization

In the second phase, we used the best column identified by the screening study. The goal of this phase was to further optimize the method by studying gradient times, oven column temperatures, and concentrations of the additive trifluoroacetic acid (TFA). We entered those factors into the DoE software. Eighteen runs were recommended.

Figure 7 includes examples of the resolution response surface graph for one of the peptides. On the right, there is a linear effect for gradient time and oven temperature at the 0.100% TFA. The effect is more complex with the 0.750% and 0.125% TFA on the left.

Figure 7

Figure 8 summarizes the best overall answer for the final optimized method: a gradient time of 9.49 minutes, oven temperature of 27.7 °C, and TFA concentration of 0.1107%. The predicted resolution for each peptide is shown in figure 9. The most challenging separation is the resolution between T71_T72 and the peak before it (see T71_T72_R1 for details).  It only achieved close to 1.5 resolutions.

Figure 8

Figure 9

The acceptable performance region in figure 10 is the unshaded, or white, region, which also coincides with where the “best overall answer” would fall. In the lower left corner you can see the edges of failure for T7_T8_R1 and T7_T8_R2 bordering the design space.

Figure 10

Finally, the QbD software produces the proven acceptable ranges in figure 11 for oven temperature and gradient time.

Figure 11

Key takeaways

1. Incorporating QbD into analytical method development is a powerful approach for proactively reducing the risk of failure. Developing a robust identification test for HSA using the QbD approach will reduce the chance of method failure for testing incoming HSA material.
2. The FDA and EMA are currently working together to define review criteria for evaluation of QbD-based analytical methods. This study demonstrates one of the areas where QbD approach provided valuable improvements to an existing methodology.

Source: Pfizer CentreOne Contract Manufacturing