Minimization of Freeze/Thaw-Induced Protein Aggregation and Optimization of a Drug Substance Formulation Matrix

The authors explore the use of statistical experimental design and multivariate analysis to develop a drug substance formulation matrix.
Aug 01, 2015
Volume 28, Issue 8, pg 30–37

Science Photo Library/GettyimagesAbstract
The authors explore the use of statistical experimental design and multivariate analysis to develop a drug substance formulation matrix.

Protein aggregation is a major issue that can occur at any stage of biologics production and storage and often results in delays in the commercialization of protein drug candidates (1–3). To date, the mechanism of protein aggregation is still not well understood. The control and analysis of protein aggregation is a major challenge for many biopharmaceutical companies. While it is possible to reduce the risk of protein aggregation by optimizing buffer matrix, protein concentration, operating/storage conditions (i.e., temperature, agitation, contact material, etc.), and adding stabilizers, the measures to effectively preventing protein aggregates differ experimentally on a case-by-case basis. The presence of aggregates in drug products is undesirable, as it can stimulate immunogenic reactions and cause drug rejection upon administration to patients (4). In recent years, protein aggregation in drugs and drug candidates has gained increased scrutiny from worldwide health authorities (5–8). It is important to develop strategies to minimize protein aggregation early in process development and have proper analytical tools for monitoring, control, and characterization.

Frozen storage is an important preservation method for soluble proteins; however, freezing (and thawing) of a protein solution introduces complex changes to the buffer environment that may lead to protein aggregation. Such changes include formation of ice-water interfaces, adsorption to container surfaces, cryoconcentration of protein and solutes, pH changes, and phase separation (1, 2, 9). Cryoprotectants have been used to reduce freeze/thaw (F/T)-induced protein aggregation. Many cryoprotectants are rich in hydroxyl groups that replace water molecules to form hydrogen bonds with protein molecules needed to maintain the high-order structure during freezing. Cryoprotectants also function by delaying ice crystallization and eutectic transition of a protein solution (10).

In this study, protein aggregation was observed during F/T of a protein drug substance. To minimize the formation of F/T-induced protein aggregates, the effects of buffer, protein concentration, storage temperature, and cryoprotectants on protein aggregation, potency, and activity were investigated using dynamic light scattering (DLS), size-exclusion chromatography (SEC), potency, and activity assays, respectively. The operating ranges were optimized using statistical experimental design to establish the parameters and control limits for the drug substance formulation matrix. Finally, the drug substance formulated in the proposed matrix was subjected to multiple F/T cycles to confirm the acceptability of the optimized conditions.

Experimental approach
Protein aggregation was monitored using DLS and spectrophotometric methods. A number of conditions were screened, including buffering species, pH, salt concentration, protein concentration, cryoprotectants, cryoprotectant concentration, and storage temperature. Based on the screening results, the drug substance was optimized using a custom design-of-experiments (DOE) study, which refined the parameter ranges based on the responses of critical quality attributes of the drug substance (i.e., size-exclusion chromatography [SEC]-based purity, potency, and activity). The DOE results were used to create multivariate regression models for each of the responses. A prediction profile was constructed based on the multivariate regression models, which, along with the historical manufacturing data, were used to perform Monte Carlo simulations to identify process capability and potential number of failures (11). All the statistical experimental design and data analysis were performed using JMP software version 9.0 (SAS Institute, Cary, NC). Based on the DOE and Monte Carlo simulation results, targets and control limits for protein concentration and drug substance matrix components were determined. The operating ranges were confirmed by subjecting the drug substance to multiple F/T cycles.

Materials and methods
Polyethylene glycols (PEGs) of different molecular mass were used in this study. PEG-3350 was purchased from Dow, while PEG-4000 and PEG-6000 were from Fluka. Polyoxyethylene glycol dodecyl ether (BRIJ-35), PEG-300, and Tergitol NP-40 (nonylphenol ethoxylate) were purchased from Sigma-Aldrich. Poloxamer 188 (P-188) was from BASF. Human serum albumin was purchased from Talecris. Tween-20 and Tween-80 were from Croda. Sucrose and ammonium sulfate were from EMD Millipore, and other chemicals used in buffers were purchased from Mallinckrodt or J.T. Baker.

DLS measurement was performed using a Zen 3600 Zetasizer (Malvern Instruments), and protein concentration was measured with a DU720 UV/vis spectrophotometer (Beckman Coulter). Potency was determined using a cell-based assay, while the proteolytic activity was measured based on a enzyme-based assay. Both potency and activity were reported as relative to a reference standard. Purity was assessed by analytical SEC, which detects aggregates and low-molecular-weight proteins.



Results and discussion
Prevention of aggregation formation
The presence of protein aggregates was measured before and after the F/T cycles were monitored by DLS, which determines the size distribution profile of small particles in Brownian motion in a solution. The drug substance solutions, in either citrate or phosphate buffer, were used to screen solution conditions that prevented F/T-induced aggregation. To determine the effect of protein concentration, buffer species, pH, salt, cryoprotectant, and storage condition on the prevention of F/T-induced aggregation, a series of screening experiments were completed; see Table I for a summary. Based on the initial screening studies, it was determined that of the factors investigated, only addition of cyroprotectants prevented F/T-induced aggregation.

Table I: Screening of drug substance formulation matrix conditions to minimize protein aggregation. A designation of “No” was given to those conditions and parameters that were not effective in preventing F/T-induced aggregation; a designation of “Yes” was assigned to conditions and parameters that were effective in preventing F/T-induced aggregation. F/T=freeze/thaw.

CLICK TO ENLARGE Figure 1: Prevention of freeze/thaw (F/T)-induced aggregation by adding PEG-3350. Dynamic light scattering results of drug substance solution at 0.5 g/L without (a) or with (b) 1.0% (w/w) of PEG-3350.  The samples were either stored in a refrigerator (2–8 ˚C) or after one cycle of freeze/thaw (post-F/T).Figure 1 shows typical results from the screening studies. Under refrigerated conditions (2–8 °C), the drug substance retained the monomer. However, following a single F/T cycle, protein aggregates were detected (see Figure 1a). As described in Table I, using DLS to detect aggregation, varying buffer species (acetate, citrate, or phosphate), protein concentration, pH, or salt level did not prevent F/T-induced protein aggregation. In contrast, many cryoprotectants were found to be effective in preventing F/T-induced aggregation, including Tween 20, Tween 80, P-188, NP-40, BRIJ-35, and all of the PEGs (see Figure 1b for an example using PEG). As little as 0.1% PEG in either citrate or phosphate buffer inhibited the formation of aggregates.

PEG is widely used in the pharmaceutical industry, readily available, easily dissolved, and does not interfere with UV absorbance. In addition, as a compound on FDA’s compounds generally regarded as safe (GRAS) list, PEG has low immunogenicity and itself is widely used as a drug (12). For this study, a medium-sized PEG was selected as the cryoprotectant, and only PEG results are presented hereafter.

Figure 2: The effect of drug substance buffer matrix on potency. Panel a) illustrates the experiment workflow and drug substance samples for testing; panel b) shows one-way analysis of variance (ANOVA) relative potency and Tukey-Kramer HSD [honestly significant difference] test results comparing the potency means of different samples at a significance level of 0.05. UF/DF=ultrafiltration/diafiltration.
The effect of PEG on potency
To assess the functional impact of the buffer or presence of PEG on drug substance potency--a critical quality attribute--the drug substance was prepared in replicates, in either citrate or phosphate buffer, supplemented with 1% of either PEG or blank buffer, and then tested for potency (Figure 2a). The potency data were analyzed by one-way analysis of variance (ANOVA) and a Tukey-Kramer HSD [honestly significant difference] test (Figure 2b). The results indicated that the buffer and PEG addition did not make a statistically significant difference on drug substance potency, at the significance level (α) of 0.05.


Custom-designed DOE study
To establish the ranges of drug substance matrix components, a multilevel, custom-designed DOE study with five factors and three responses was performed. The DOE evaluated the effects of ammonium sulfate concentration (0, 100, 140, 180, and 300 mM), pH (5.5, 6.0, 6.3, 6.5, 6.7, and 7.0), PEG concentration (0.1%, 0.5%, 1.0%, and 1.5%), protein concentration (multiple concentrations from 0.04 to 0.49 mg/mL, based on UV measurement results), and hold time (0, 2.5, and 5 days) on SEC purity, potency, and activity. A total of 32 runs were completed, and the results were analyzed using JMP Version 9.0. One run was identified as an outlier by Mahalanobis distance outlier analysis and was excluded from the calculations.

Figure 3: The constructed multivariate regression models for a) purity, b) potency, and c) activity. The models were made based on design-of-experiment (DOE) results of each response.

Multivariate analysis was performed to identify the high level correlations among the factors. Linear regression models for each response were constructed using terms of individual factors, two-level interactions, and quadratics. The models were further refined by stepwise regression, taking away terms with p-values greater than 0.05. The refined regression models are illustrated in Figure 3.

CLICK TO ENLARGE Table II: The effect size and design margin of parameter terms on responses. Pr. conc.=protein concentration; conc.=concentration; AS=ammonium sulfate, A278=ultraviolet light absorbance at 278 nm. Bold indicates statistical significance at significance level (∝) of 0.05.


The results indicated significant regression models for SEC purity (Panel A, adjusted R2: .81, p-value: <0.0001), potency (Panel B, adjusted R2: .67, p-value: 0.0024), and activity (Panel C, adjusted R2: .71, p-value: 0.0011), at α=0.05. The effect sizes and margins of parameter terms on responses are summarized in Table II. Protein concentration, PEG concentration, and their associated interaction and quadratics have the biggest effect sizes among the different model terms.

CLICK TO ENLARGE Figure 4: Screenshot of example prediction profiler and Monte Carlo simulations based on the multivariate regression models. The Monte Carlo simulation was performed under different scenarios. The profiles illustrated in this figure used the parameter standard deviations calculated from actual historical batches from scenario 2.


Scenario 1 used the proposed parameter targets, and the parameter standard deviation (SD) was calculated by dividing the proposed parameter range by six (i.e., the proposed parameter range was treated as the six sigma [6 x SD, or 6σ] range). Scenario 2 used the actual parameter targets and SDs from historical manufacturing batches. Scenario 3 used the actual parameter targets and SDs from historical manufacturing batches, with random noise adjusted so that the variation of predicted results matches that of manufacturing batch results. For simplicity, hold time was fixed at 0 day for the simulation as no hold is proposed at this step. The distribution of 10,000 simulated runs are compared with the distribution of the historical manufacturing batches in terms of mean, SD, failure rate (in % and ppm), and process capability (Cp or Cpk).


Figure 4 illustrates an example of simulation profiles under scenario 2 and the simulation results are summarized in Table III. The process capability indices, Cp and Cpk, under the different simulation scenarios, are all greater than or equal to 1.1.

Table III: Summary of actual and simulated results, failure rate in ppm, and process capability Cp/Cpk. SD=standard deviation; ppm= parts per million; Cp/Cpk=process capability indices; N/A=not applicable; SEC=size-exclusion chromatography.

CLICK TO ENLARGE Figure 5: The contour profiles of the PEG concentration and protein concentration with predicted a) purity (pink contour lines); b) potency (green contour lines); and c) activity (blue contour lines). The proposed design space is shown in a rectangle filled with light blue color. The area with purity < 95% is filled with pink color, and the area with potency outside the range of 0.6–1.5 is filled with green color. LSL=lower specification limit; USL=upper specification limit. Figure 5 illustrates the predicted response surface under the different protein concentrations and PEG concentrations. The rectangles filled with blue color represent the proposed operating ranges. Based on the results, parameter targets and ranges were established for the drug substance formulation matrix.


Confirmation study
To confirm the selected conditions, drug substance solutions in the proposed formulation matrix were frozen at -70 °C or -150 °C. Samples were subjected to multiple F/T cycles prior to analysis by DLS. As shown in Figure 6, repeated F/T cycles did not result in the development of aggregates, as detected by DLS.

Figure 6: Dynamic light scattering results of optimal drug substance formulation matrix conditions after multiple freeze/thaw cycles.

The objective of this study was to develop an appropriate drug substance matrix to prevent the formation of F/T-induced protein aggregates. In this study, freeze/thaw-induced protein aggregation was observed during process development. To minimize protein aggregation, PEG-3350 was added to the drug substance matrix, and the matrix formulation was optimized by a custom DOE study. Based on the DOE results, multivariate regression models were established, Monte Carlo simulations were performed under different scenarios, and drug substance matrix conditions were proposed. The proposed conditions were confirmed by experiments with multiple F/T cycles. Several batches of drug substance have been produced under the proposed conditions and have demonstrated good stability after storage.

1. W. Wang, Intl. J. Pharmaceutics 289, pp. 1–30 (2005).
2. H-C Mahler et al., J. Pharm. Sci. 98 (9), pp. 2909–2934 (2009).
3. M.E.M. Cromwell, E. Hilario, and F. Jacobson, AAPS J. 8 (3), pp. E572–579 (2006).
4. A.S. Rosenberg, AAPS J. 8 (3), pp. E501–507 (2006).
5. FDA, Guidance for Industry, Immunogenicity Assessment for Therapeutic Protein Products (Rockville, MD, Aug. 2014).
6. R.V. Cordoba-Rodriguez, Biopharm. Intl. 21 (11), (2008), accessed May 9, 2012.
7. T.M. Scherer et al., AAPS J. 14 (2), pp. 236–243 (2012).
8. USP, USP General Information <1787>, “Measurement of Subvisible Particulate Matter in Therapeutic Protein Injections,” Pharmacopeia Forum 39 (2), March–April (2013).
9. W. Wang, S. Nema, and D. Teagarden, Intl. J. Pharmaceutics 390 (2), pp. 89–99 (2010).
10. Y. Mi, G. Wood, and L. Thoma, AAPS J. 6 (3), pp. 1–10 (2004).
11. SAS Institute Inc., “JMP Statistics and Graphics Guide,” Cary, NC, accessed May 9, 2012.
12. J. Chen et al., Green Chem. 7, pp. 64–82 (2005).


Peer Reviewed
Article submitted: Apr. 10, 2015.
Article accepted: May 13, 2015. 


About the Authors
Hui Xiang is principal scientist; Derek Chan is principal scientist; both at Allergan. Ronald Bates was director, at Allergan, Inc., and is currently director at Bristol-Myers Squibb.


Article Details
BioPharm International
Vol. 28, No. 8
Pages: 30–37
Citation: When referring to this article, please cite it as H. Xiang, D. Chan, and R. Bates, "Minimization of Freeze/Thaw-Induced Protein Aggregation and Optimization of a Drug Substance Formulation Matrix," BioPharm International 28 (8) 2015.


lorem ipsum