Bench-Scale Characterization of Cleaning Process Design Space for Biopharmaceuticals

ABSTRACT

The characterization and validation of equipment cleanliness are key requirements for biopharmaceutical facilities to assure that the cleaning process can meet predetermined cleanability criteria consistently and reproducibly. For a multi-product facility, the cost of performing such process characterization at large-scale could be substantial. This article describes how a robust bench-scale model can be used to characterize the key operating parameters of the cleaning process. A scale-down model that evaluates the cleanability of various protein drug products on stainless steel coupons was used to explore the process performance over a wide design space. The bench-scale cleaning model is also a useful tool for comparing the cleanability of various products and for developing new cleaning cycles.

The design space concept as introduced by the International Committee on Harmonization (ICH) for a unit operation can be defined as the linkage between the input variables or process parameters and critical quality attributes.¹ Characterizing the design space involves understanding these linkages and identifying variables and ranges within which consistent quality can be achieved. For the cleaning process, the design space can be considered the interactions among various operating parameters (e.g., temperature, soilant, cleaning agent concentration, dirty hold time) and their effect on performance parameters such as residual soil levels at the end of the cleaning process.

(Digital Vision/Getty Images)

Drug product processing involves various process parameters that can affect the safety, quality, efficacy, and purity of the final dosage form.² Cleaning processes also can have a direct effect on the critical quality attributes of a product by means of contaminant carryover through product contact equipment surfaces. Regulatory expectations call for biopharmaceutical companies to have a sound, documented cleaning validation approach that uses effective and consistent cleaning and keeps the carryover below the acceptable limits.^3–5 Such an approach warrants a cleanability assessment before the introduction of a new product into the manufacturing equipment. Bench-scale cleaning studies are a useful model to evaluate the relative cleanability of new products with very low material requirements, well in advance of technology transfer to the manufacturing facility.^6–9 Bench-scale models also provide the benefit of performing cleaning evaluations under controlled simulated conditions, thereby offering a useful tool to characterize the process design space.

There are several elements of cleaning process characterization, control, and validation including, but not limited to, cleaning cycle, equipment, sampling techniques, analytical assays, and acceptance criteria.^5,7,10 This article focuses on the characterization of the cleaning process with respect to process parameters. A bench-scale model is used to evaluate four different protein drug product formulations over a wide range of operating conditions. Both single parameter and cross interaction among different operating conditions are studied to fully characterize the design space associated with the cleaning process. Results show that temperature and cleaning agent concentration are strongly coupled, and different protein products behave differently with respect to their cleanability because the process conditions are varied. The work also provides useful insights into the development of an optimized cleaning cycle for biopharmaceuticals.

CLEANING PROCESS

In a typical cleaning process, cleaning agents such as alkaline and acidic reagents are used at elevated temperatures in combination with mechanical action to achieve removal of protein soilants from equipment surfaces. A combination of water and caustic acid rinses is used to achieve the desired level of cleanability needed to ensure minimal product carryover to the next product lot. A hot alkaline wash is considered the critical cleaning step in which conditions of high pH and high temperature are used to remove the protein by degrading it into smaller fragments, and to solubilize any hydrophobic residues.

A cleaning cycle relies on two pathways for soilant removal from equipment surfaces.^9,11 The first pathway is the physical removal commonly achieved through mechanical action resulting from the convective action of the fluid flow. The efficacy of this mechanism is governed by the soil–surface interactions and the extent of adhesion. The second mechanism uses chemical interaction between the cleaning agents and the protein soils such as protein degradation, solubilization, wetting, and emulsification to remove the soilant from the surface.¹¹ There are various inputs to the cleaning process that affect these cleaning mechanisms and thereby the overall performance of the cleaning cycle. These key operating parameters include the temperature of the cleaning solution, the concentration of the cleaning agent, the strength of the mechanical action, and contact time with the cleaning agent. In addition, other factors such as soil type, dirty hold time (i.e., how long the soiled equipment is held before cleaning), material of construction (stainless steel, glass, Teflon), and surface finish also affect the cleanability of the equipment.

MODEL

All the experiments in this study were conducted using a bench-scale cleaning model comprising the following components:

• Reciprocating shaking water bath: A Precision Scientific model 25 shaking water bath (Precision Scientific, Winchester, VA) was used to control temperature and agitation. Oscillation frequency and amplitude were adjusted to obtain the desired linear velocity. The temperature in the water bath was maintained with an accuracy of ±1°C.

• Stainless steel coupons: 6 x 3-inch stainless steel (304L) nonelectropolished coupons (Q-Lab Corp., Cleveland, OH) were used in this study. Although the equipment used in biopharmaceutical manufacturing is typically made from 316-grade stainless steel, for the purpose of these cleanability experiments, it was assumed that the grade of steel (304 versus 316) would not have a significant effect on the relative cleanability of drug products. Differences in stainless steel surface finishes were not included in the scope of this evaluation either.

• Cleaning agents: The cleaning solution used in this study was either de-ionized water or a CIP-100 solution (Steris, Mentor, OH). In addition, CIP-200 (Steris) and the alcohol solution Septihol (Steris) were used for precleaning the coupons.

Four protein drug products (labeled as products A, B, E, and H in this article) were evaluated in these studies. Products A and B are antibodies and products E and H are protein products. These aqueous products contained different excipients and stabilizers, and were formulated at various concentrations (≥100 mg/mL).

METHOD

Stainless steel coupons were precleaned in a cycle that comprised a CIP-100 wash, CIP-200 wash, and water for injection (WFI) rinses, and dried before use. This precleaning procedure helped to cleanse any residual coating from the coupon fabrication process and also ensured that all new coupons were in the same initial condition before starting the experiment. Of each protein product, 250 μL was spotted on the coupon (with a maximum of six spots per coupon). Sufficient resolution was offered by the 250 μL to differentiate among the cleaning times of all four products in reasonable experiment duration. Each protein soil was spread into a circular spot that was 1.35 cm in diameter. Variability in spot size and shape and in the measured cleaning time was minimized by using new coupons for each evaluation. The spotted coupons were air dried at room temperature for 24 h before cleaning. It should be noted that the selection of 250 μL over a circular spot 1.35 cm in diameter might not be completely representative of the typical product monolayer observed on the manufacturing equipment. As a result, the absolute cleaning times reported in this study are not reflective of a large-scale cleaning cycle.

Before starting the cleaning experiment, the water bath tank was filled with the appropriate cleaning solution. This volume was large enough (>1,000 times the protein load) to ensure that the performance of the cleaning process was not limited by the amount of cleaning solution. The temperature and agitation speed of the water bath were set based on the selected cleaning conditions. Based on the amplitude and frequency of the reciprocal shaking motion of the coupons, the cleaning fluid velocity relative to the product spot was computed. The product spots on the coupons were monitored visually and the total time taken by the spot to appear visually clean was taken as the cleaning time for that spot. At least three spots were studied for each of the four products to obtain an estimate of standard error in the reported cleaning time. Visual inspection is an important element of cleaning validation and offers a simple, noninvasive technique for surface cleanability assessment.^6,12–15 If needed, the technique can be easily coupled with more sensitive assays (gravimetric or total organic carbon).

Design of Experiments: Single Parameter Evaluation

The key parameters explored in this evaluation included: temperature, concentration of cleaning solution, dirty hold time, and agitation during cleaning. The first phase of this study was designed to identify the parameters that have a significant impact on the cleaning process while minimizing cross interactions among input variables. Each parameter was varied within a specified range that was preselected to be much larger than the normal operating range. One parameter was varied while others were kept constant at the baseline cleaning conditions. These baseline conditions were selected to mimic the manufacturing cleaning cycle. Table 1 shows the baseline cleaning conditions and the selected evaluation range for each parameter.

Table 1. List of process operating parameters and the operating ranges for large-scale and bench-scale cleaning process

Augmented Design of Experiments: Cross-Interaction Among Parameters

After the first phase of experiment was completed, an augmented design of experiments was constructed using the JMP statistical software (SAS, Cary, NC), and additional experiments were conducted to assess the effect of variable cross interactions on the cleaning process.¹⁶ A leverage plot analysis showed that the cross-interactions were limited to two parameters: concentration and temperature of the cleaning solution.

RESULTS AND DISCUSSION

Key Operating Parameters

Temperature. The conventional approach to cleaning processes may drive us to believe that it is always better to use a cleaning solution at a higher temperature. Although this may be the case for products where solubility increases with temperature (mainly small-molecule based pharmaceuticals), protein products exhibit a different trend. Figure 1a shows how the cleaning time changed for four products as the temperature of the 1% v/v CIP-100 solution was increased while keeping other operating parameters the same as the baseline listed in Table 1. The longest cleaning time occurred at 55°C. Interestingly, the shortest cleaning time for all products is observed at the lower temperatures. We attribute this trend to the unique behavior of proteins, for which cleaning time is a combination of two competing phenomena: the dissolution of protein soil in the cleaning solution (controlled by solubility, wetability, etc.), and protein degradation under high pH and temperature conditions though a series of chemical reactions including hydrolysis, oxidation, and de-amidation. Protein solubility is maximized in the lower temperature range (20–30°C) where protein molecules are dissolved in the aqueous solution while maintaining their structure. As the temperature increases, protein products lose their structure, start to denature, stick to the surface, and become increasingly more difficult to dissolve. The onset of denaturation is driven by the melting temperature of each product. Product H, with the lowest denaturation temperature, shows an increase in cleaning time even for a temperature of 40°C. As the temperature is increased, CIP-100 solution (pH >11) also starts to degrade the product molecules into smaller fragments.^6,17 At an intermediate temperature of approximately 55°C, although protein degradation has started, it is the protein denaturation phenomenon that dominates, resulting in an overall increase in the cleaning time, the effect being more noticeable for antibody products (A and B).

Figure 1. Relationship between cleaning time and various key operating parameters: (a) temperature, (b) CIP-100 concentration, (c) dirty hold time and (d) agitation. Only one parameter was changed at a time, while others were kept constant at the baseline conditions listed in Table 1.

As the temperature increases beyond 70 °C, the alkaline solution degrades the protein molecules more effectively into smaller fragments and also enhances protein removal from the surface. It is therefore concluded that although protein dissolution is high near ambient temperatures, a temperature of ≥70°C is needed to degrade the protein molecules into smaller fragments. Cleaning validation may seek product degradation as a requirement in addition to product removal from surfaces because smaller peptides can be considered less immunogenic than native product.¹⁸

Similar experiments conducted with cleaning in de-ionized water at various temperatures also showed ambient-temperature water resulted in shorter cleaning times for all four products. Higher temperatures (≥55°C) caused protein denaturation, making the cleaning solution (water) turbid. In the absence of degradation action of the CIP-100 solution, none of the four products could be cleaned within the experimental duration of 2 h when subjected to water at 70°C. It is therefore recommended that the prerinse step, often used before the equipment hold, should be performed using ambient-temperature water to remove the majority of the protein load from the equipment.

CIP-100 concentration. The concentration of the cleaning agent also plays a critical role in governing the ability of the cleaning fluid to remove protein products from equipment surface. Although high temperatures favor denaturation (loss of secondary and tertiary structure), a higher CIP-100 concentration increases the OH¯ concentration, which results in increased degradation rates. The primary effect of alkaline agents is peptide bond hydrolysis, which results in protein molecules being broken down into smaller peptide fragments.

Figure 2 compares the formation of such smaller peptide fragments over time for products A and H when subjected to two different CIP-100 concentrations at 70°C. As the figure shows, product degradation is much faster for 5% v/v CIP-100 than for 1% v/v solution. The hot alkaline wash achieves its cleaning action through a combination of protein degradation and enhanced dissolution. It is therefore expected that the cleaning time would decrease as the CIP-100 concentration is increased. Figure 1b shows that cleaning times, in general, are lower at higher CIP-100 concentrations.

Figure 2. SDS-PAGE gel to study degradation of protein products A and H during cleaning at 70 °C in (a) 1% v/v CIP-100 and (b) 5% CIP-100. The first lane represents molecular weight markers. The numbers on the tops represents the reaction time divided by contact time (in min) with CIP-100 solution corresponding to that sample.

However, the trend for the change in cleaning time is also product specific. Product H, for instance, exhibited a minimal cleaning time at an intermediate concentration of 1% CIP-100 (Figure 1b). Other products, such as B and E, also show an improved performance at 1% CIP-100 compared to 2% CIP-100. Such an increase in cleaning time at higher caustic concentration can potentially be attributed to factors such as a change in fluid properties and a change in surface morphology of the deposited product affecting the fluid diffusion into the deposit. This variability highlights the product-specific nature of the cleaning performance and the importance of characterization work in determining optimum cleaning conditions. However, this trend, established in Figure 1, is specific to a cleaning temperature of 70°C. As we report later, temperature and cleaning agent concentration are strongly coupled parameters and it is necessary to understand their cross interaction to understand the complete behavior of the performance of the cleaning process.

Dirty Hold Time. Another factor governing the cleanability of equipment under manufacturing conditions is dirty hold time—the duration of time the soiled equipment and equipment parts are held before subjecting them to the cleaning cycle. Cleaning validation is often performed using the worst-case scenario of maximum expected dirty hold time. We studied the effect of air-drying at ambient temperature for different hold times on the observed cleaning time for the four products. Figure 1c shows that there is a significant effect on cleanability during the early hours of hold time (<16 h) because the protein soil is not completely dried. We observed a small increment in cleaning time once the coupons have been dried for 24 h at room temperature. This also justifies the choice of 24 h as a baseline drying time for the remaining evaluations in this study.

Agitation. One of the key operating parameters that contributes to the removal of the product soils from the equipment surface is the mechanical action generated by the cleaning fluid. The impingement action of the cleaning fluid spray ball is often used to generate this effect in automated clean-out-of-place (COP) baths. Because of the specificity of the cleaning action generated by a spray ball and its dependence on the location in the COP bath, this remains one of the most challenging operating parameters to scale down. In fact, in the scaled-down model we only aspire to mimic the worst-case scenario of lowest shear expected to be observed in the equipment holes, cavities, or corners in the COP bath with minimal fluid flow. Figure 1d shows how the cleaning times changed over four different agitation speeds of the shaking water bath.

Protein Concentration. Protein products tend to get more difficult to clean as their concentration increases. Generally, a rise in concentration is accompanied by an increase in product viscosity and a stronger propensity to form a thicker soilant layer on the surface. To assess the challenges posed by new high concentration formulations on the cleaning process, it is important to evaluate how the cleanability is affected by the increase in protein concentration. We studied three different products: A, B, and H. A range of product concentrations was prepared by diluting the products in their respective buffers and concentrating them through 10-K centrifugal filters. Figure 3 shows that the cleaning time can be significantly higher for these products as the concentration is increased. It is also observed that the increase in cleaning time was more significant for the difficult-to-clean antibody products (A and B) compared to protein product (H).

Figure 3. Dependence of cleaning time on product concentration, using baseline cleaning conditions for temperature, CIP-100 concentration, dirty hold time, and agitation. Cleaning time increases with product concentration.

Cleaning Design Space Characterization

After completion of the single parameter study, an augmented design of experiments was constructed to evaluate the cross interactions between the two critical parameters: temperature and CIP-100 concentration. Additional experiments were conducted to sample the two-parameter space. JMP software was used to perform the statistical analysis using various operating parameters as effects and the cleaning time as a response. Linear regression analysis was used to fit the model to the experimental data. The model included both first order terms (temperature and concentration), and second order terms (self- and cross-interaction).

Leverage Plots

Leverage plots were constructed to graphically view the significance of each term's effect on cleaning time. A P-value analysis shows a strong correlation between the cleaning time and the cross interaction between temperature and concentration (Figure 4). The cleaning time was modeled as a response to both first- and second-order interactions between temperature and concentration. Table 2 lists the scaled estimates for this model and provides a more relevant scale-invariant effect size. As shown in Table 2, the cross interaction between temperature and concentration has the maximum effect on cleaning time. This suggests that the effect of a temperature change on cleaning time is strongly dependent on the value of CIP-100 concentration, and vice-versa.

Table 2. Scaled estimates for various first- and second-order terms of the fitted model

A similar analysis was performed for the remaining three products (B, E, and H) and the cross interaction term was consistently found to be the most significant parameter with respect to its effect on cleaning time.

Figure 4. Leverage plot analysis for dependence of cleaning time on the most significant term: the cross interaction between temperature (T) and concentration (C)

Interaction Plots and Surface Response Profiles

The interaction plots offer a graphical view of any prospective interaction among the various factors. Any evidence of interaction between effects is depicted as nonparallel lines in the interaction plots. Figure 5 shows that the impact of CIP-100 concentration varies depending on the temperature of the cleaning fluid. At high temperatures, a high CIP-100 concentration shortens the cleaning time, whereas at low temperatures, improved cleaning is achieved at lower CIP-100 concentrations. Such interaction profiles as well as the process response surface may be product specific and should be established for each product during the development of the cleaning cycle.

Figure 5. Interaction profiles for cross interaction between temperature and concentration of CIP-100 cleaning solution. The intersecting profiles show that these two operating parameters are strongly coupled.

Two-Parameter Design Space

Figure 6 shows contour plots for cleaning time dependence on temperature and concentration for each of the four products. The plots depict the two-dimensional design space constructed for the cleaning process for all four products. The plots exhibit the effects of cross-interactions between temperature and cleaning agent concentration on the cleaning time as measured through the bench-scale cleaning model.

Figure 6. Representation of the design space constructed using two critical operating parameters (temperature and concentration) and the third performance parameter (observed cleaning time)

Based on these contour plots, it can be concluded that there are certain product-specific and general behaviors in the response profiles of cleaning process. All of four products exhibited longer cleaning times at high temperature water conditions, and at low temperature–high CIP-100 concentration conditions. Improved soilant removal is achieved at low temperature–low CIP-100 and high temperature–high CIP-100 values. At ambient temperatures, the cleaning time progressively increases with increase in CIP-100 concentration. The trend is reversed for temperatures ≥70°C. Although these general trends are common to the four products, the magnitude of the change, and the specific temperature and concentration values at which the change occurs, are unique to each product. Among the four products, product A is the most difficult to clean (H is least) under the conditions of high temperature and high concentration. Product H, however, becomes the most difficult to clean product in ambient temperature water (0% CIP-100).

These small-scale studies form the foundation for cleaning process characterization work and offer great value in resource savings with respect to both material and time. However, as described earlier in the article, certain elements of the cleaning process could be scale- and equipment-specific. These may include, but are not limited to, equipment shape, location in the bath, spray ball coverage, level of agitation, and fluid flow dynamics. Bridging studies can be conducted to derive the appropriate scaling factors needed to convert small-scale cleaning times to large-scale. The scaling factors would be specific to the equipment, the cleaning bath, and large-scale conditions. Historical monitoring data from the cleaning validation and operation runs also can offer useful information regarding worst-case conditions and suggestions for appropriate parameters for design space characterization.

CONCLUSIONS

A bench-scale model has been developed and used to perform cleanability assessments of protein drug products. The method evaluates the cleaning time of protein soilants deposited on stainless steel coupons and cleaned under simulated large-scale conditions. The model can successfully estimate the relative cleanability of drug products and support a worst-case–based cleaning validation approach.

The model was also used to characterize the effect of varying the operating parameters over a broad process design space on process performance. Statistical analysis using leverage plots show that temperature and concentration of the cleaning solution are critical process parameters, the cross-interaction term being most significant. Self-interaction plot analysis also demonstrates the strong coupling between these two parameters and the product-specific nature of the coupling. Cleanability trends are significantly different for the four products when the temperature and chemical conditions are altered. The findings offer key insights in the significance of various process parameters and the interplay among them that can be useful for both cycle development and optimization.

ACKNOWLEDGMENTS

The authors are thankful to Abe Germansderfer (Corporate Validation, Amgen), Erwin Freund, Anurag S. Rathore, and Ed Walls (Process Development, Amgen) for their critical review and valuable suggestions toward this work.

Nitin Rathore is a senior scientist, Cylia Chen is an associate scientist, and Wenchang Ji is a principal scientist, all in drug product and device development at Amgen, Inc., Thousand Oaks, CA, 805.313.6393, nrathore@amgen.comWei Qi is a PhD candidate at the department of chemical engineering at the University of Virginia, Charlottesville, VA.

REFERENCES

1. International Conference on Harmonization (ICH). Annex to Q8. Pharmaceutical Development. Version 2. 2007 Nov 1.

2. Rathore N, Rajan R. Current perspectives on stability of protein drug products during formulation, fill and finish operations. Biotechnol Prog. 2008;24(3):504–514.

3. US Food and Administration. Current good manufacturing practice for finished pharmaceuticals. Code for Federal Regulations. 2004, Title 21, Volume 4, Section 211.67. Washington DC: US Government Printing office.

4. Mollah AH. Cleaning Validation for Biopharmaceutical Manufacturing at Genentech. BioPharm Int. 2008 Mar;21(3):68–75.

5. Sanchez JAM. Equipment cleaning validation within a multi-product manufacturing facility. BioPharm Int. 2006 Feb;19(2):38–49.

6. Rathore N, Qi W, Ji W. Cleaning characterization of protein drug products using UV-vis spectroscopy. Biotechnol Prog. 2008;24(3):684–690.

7. Mollah AH, White EK. Risk-based cleaning validation in biopharmaceutical API manufacturing. BioPharm Intl. 2005 Nov;18(11):54–67.

8. Sharnez R, Lathia, J, Kahlenberg D, Prabhu S. In situ monitoring of soil dissolution dynamics: A rapid and simple method for determining worst-case soils for cleaning validation. PDA J Pharm Sci Technol. 2004;58:203–214.

9. Le Blanc DA. Validated cleaning technologies for pharmaceutical manufacturing. Boca Raton, FL:, CRC Press LLC; 2000.

10. Cardot J-M, Beyssac E. The basics of cleaning and cleaning validation. Bioprocess Int. 2007;5(6):66–76.

11. Verghese G. Selection of Cleaning Agents and Parameters for cGMP Processes. Proceedings of the Interphex Conference; 1998 Mar 17–19, Philadelphia, PA.

12. LeBlanc DA. Visually clean as the sole acceptance criteria for cleaning validation protocols. PDA J Pharm Sci Technol. 2002;56:31–36.

13. Forsyth RJ, Nostrand VV. Using visible residual limits for introducing new compounds into a pharmaceutical research facility. Pharm Technol. 2005;29(4):134–140.

14. Forsyth RJ, Roberts J, Lukievics T, Nostrand VV. Correlation of visible-residue limits with swab results for cleaning validation. Pharm Technol. 2006.

15. Forsyth RJ, Nostrand VV, Martin GP. Visible-residue limits for cleaning validation and its potential application in a pharmaceutical research facility. Pharm Technol. 2004;28(10):58–72.

16. JMP Statistical Discovery from SAS, Release 6, SAS Institute Inc, 2006.

17. Sheen SJ, Sheen VL. Characteristics of Fraction-1-protein Degradation by Chemical and Enzymatic Treatments. J Agric Food Chem. 1987;35:948–952.

18. Martinez JE. Immunogenic Potential of Therapeutic Protein Residues after Cleaning. Bioprocess Int. 2004;2(9):38–44.