samedan logo
 
 
 
spacer
home > ebr > autumn 2011 > drawn to scale
PUBLICATIONS
European Biopharmaceutical Review

Drawn to Scale

In recent years there have been initiatives from regulatory agencies and the industry that have emphasised the importance of process understanding in biopharmaceutical development. In response, new evidence shows how univariate and multivariate techniques can be used in the verification of small scale fermentation.

n 2002 the US Food and Drug Administration began its initiative to modernise the regulation of pharmaceutical manufacturing and product quality, and developed the concept of ‘quality by design’ (QbD) for pharmaceuticals. QbD principles and advice for implementation are outlined in regulatory guidances such as ICH Q8, Q9 and Q10. One of the foundations, underpinning QbD’s operation is that products and processes are sufficiently understood so that those elements most critical for product quality can be identified, codified and controlled.

While process understanding can come from a variety of sources – including early process development activities, as well as experience with similar products and manufacturing campaigns for clinical supply – scaled down process models of unit operations have particular utility. Advantages such as the ease of setup compared with pilot or full-scale activities, less material requirements, as well as possible automation through the use of robotics, allow for a depth and span of investigation that is impractical at larger scale. For example, in design space and control strategy development they allow a comprehensive understanding to be developed of the multivariate effect of operating conditions and material attributes on product quality. Once established they also allow for troubleshooting issues that can be encountered during commercial manufacturing, the testing of pathogen, and impurity removal and reduction strategies. Additionally, they can complement process performance qualification studies reducing the dependence on testing of commercial batches (1). Small scale process models (SSPM) also provide information that can be used in simulation studies to predict the performance of a commercial process, thereby avoiding problems before they occur.

Considerations for SSPM Development

There are a number of considerations that have to be taken into account in the development of a SSPM including:

  • Is the final commercial scale process defined prior to SSPM development? SSPMs can be used prospectively as a development and investigational tool before the final commercial process has been defined. In this instance, qualification, while ideally prospective, can be a retrospective activity. If data from the final scale is not yet available, the SSPM should be qualified against the largest scale for which data exists. The degree of variability in the large scale data set should be understood
  • Demonstration of equivalence across operating ranges. While most SSPM are qualified at set operating conditions, there is value in including additional runs at off-centre conditions to test the relevance of the model under conditions that may be expected in manufacture
  • The equipment used in the SSPM should have equivalent design characteristics and process control capabilities, with good engineering principles always in place
  • Identification of the scale independent parameters that will be used to establish the SSPM. For example, pH, dissolved oxygen, temperature, aeration, nutrient addition and the impact of scale dependant parameters such as working volumes
  • Limitations of the equipment utilised in the SSPM should be understood. While most unit operations can be scaled down effectively, even the best SSPM can have differences in terms of dead volumes, materials of construction, mixing patterns, process times and so on
  • Identification of the key process outcomes that will be used to judge the significance of the model, such as performance measures and product quality attributes
  • Raw materials used in the SSPM should be identical to those used in large scale operations. A good SSPM can be used to look for the impact of raw material variation on process performance
  • Assays utilised at both scales should be identical

The usefulness of a SSPM is dependant on its ability to mimic and predict performance of full scale in a meaningful way and to demonstrate equivalency of key process outcomes (1). If equivalency cannot be demonstrated between scales, an understanding for the unequivalency and its impact on the relevance of the information should be obtained. Inadequate models can mislead process understanding and thus a critical step in the application of the SSPM lies in how the verification and qualification of a model takes place.

Qualification of SSPM

One approach to SSPM qualification, which remains the most common, is to compare the key quality and performance attributes from both the small and commercial scale operations performed at set point conditions (2). For fermentation operations, a model can be qualified both by overlaying continuous data such as the growth profiles, gas evolution rates and by comparing the data with a set acceptance criteria. This would typically be the addition of the mean plus and minus three times sample standard deviation, which is derived using statistical analysis on historical batches. Additionally, univariate data analysis (UVDA) on discrete data can be used; for example the Students t-test and F-test are also applied to demonstrate equivalency by comparison of means and variance of key attributes with a certain confidence level (generally 95 per cent). In some instances where the normal distribution of the data cannot be established, then non-parametric tests like Mann-Whitney can be used to demonstrate the equivalency between different scales (3). Figure 1 demonstrates the application of the classical approach using the UVDA method.

The classical approach of UVDA typically only compares the final end determination of the key attributes, making it a straightforward and readilyapplied process. However, to fully understand the multidimensional link between the various critical parameters and their effect on the product critical quality attribute throughout the whole batch profile (for example with respect to batch evolution) requires the application of multivariate data analysis (MVDA) techniques. MVDA techniques, including principal component analysis (PCA) and partial least squares (PLS), not only take into account the multidimensionality of the data but also any missing values, variation due to different scale and experimental noise or error (5,6). The resulting model derived from the application of MVDA is more sensitive than the one developed using UVDA methods because it can detect observations that don’t fit the predicted response patterns while resulting in fewer false positive signals (7).

Application of MVDA in Scale Down Model Qualification

In this case study two data sets, one derived from commercial scale fermentation at 3,000L and one derived from laboratory scale fermentation at 15L, were used to demonstrate the application of MVDA in SSPM qualification. These two data sets were screened for any univariate outliers or faulty measurements and then imported into SIMCA-P+ software (version 12.0.1.0 from Umetrics AB, Sweden). For qualification studies using MVDA, only the scale-independent variables are included in the analysis (7). Hence offline measurements from scale-independent input and output variables such as temperature (oC), pH, pO2 (per cent), methanol flow rate (vvm), glycerol flow rate (vvm), exit CO2 (per cent), dry cell weight (g/L), wet cell weight (g/L) and OD600 (AU) were included as a predictor variables (X) in the analysis. In addition to this, the run time of fermenter representing the local batch time was also included as a response variable (Y) in the analysis.

The three dimensional data (batch x variables x time) was unfolded by preserving variable direction (6). In order to give all the variables an equal chance of contributing to the multivariate model after unfolding, each variable was first centred with respect to their means and then scaled to unit variance. To compare the evolution of batches at different scales, an observation level modelling approach was adopted for both data sets (8). In this approach, two models – one for commercial scale and the other for laboratory scale – were built by projecting observations on the hyper-planes and translating them into latent variables. PLS modelling approach was used to relate the process data (X-variables) to the run time of production fermenter (Y-variable) (5,6,9). This provided an appropriate maturity index model that was used to explore the batch trajectory and progress at both scales with respect to run time. Next, an iterative model diagnostics step was performed and in this step, if required, any outlying multivariate observations were eliminated, and a new model was fitted to the remaining data. Once a suitable model for both commercial and laboratory scale was determined, the final step involved generating various plots from each model for comparison to aid in the qualification of the SSPM. To highlight the advantage of MVDA in SSPM qualification two resulting plots from the analysis are explained here in detail.

Figure 2 shows the variable importance for projection (VIP) plot a key plot – obtained from PLS analysis. VIP plots indicate which variables included in the analysis have the greatest contribution to a batch evolution and thus this plot is helpful in scale-up comparison and in identifying influential variables (10). Both the VIP plots in Figure 2 show that the fermentation process has a strong impact on OD600, dry cell weight (DCW) and wet cell weight (WCW). Among the input parameters considered in the model, methanol flow rate was the most influential parameter (VIP index more than one). Comparatively, the temperature, pH and pO2 have less influence on the fermentation process as these process parameters are tightly controlled during the entire batch evolution. Comparison of VIP plot across different scales demonstrates that the partial pressure of oxygen (pO2) increased in terms of its relative importance for projection in laboratory scale. As a consequence of better mass transfer, efficient gas mixing and distribution, the small scale fermenter rarely needs additional oxygen supply; therefore, oxygen demand in this fermenter can be sufficiently sustained by the airflow and agitation. In contrast, the manufacturing scale fermentation process requires additional supplementation with pure oxygen in addition to the oxygen supplied from air. This explains the rational behind the change in the relative importance of pO2 across different scales, as well as provides additional information, which are important when considering scale up. Overall the VIP plot across the two scales agree well with each other, thus providing an additional quantitative assessment of the successful scaling of the production fermenter.

Further information towards the qualification of the SSPM was obtained from a three-dimensional scatter plot obtained from PLS analysis and is shown in Figure 3 (page 68). This plot was created from the commercial scale data using the first three predictive score with a 99 per cent confidence ellipsoid (7). Data from the laboratory runs were then used to predict the spatial coordinates of these batches in the multivariate space formed with the commercial scale batches. As shown in Figure 3, due to good prediction ability of the combined first three predictive score, the laboratory scale batches (blue triangles) reside well within the multivariate confidence ellipsoid of the commercial scale batches (black triangles) indicating that both data sets have comparable trends and possess similar correlation structures. This plot provides additional confidence that the 15L laboratory scale is an acceptable SSPM for the 3,000L commercial scale fermenter.

Conclusion

Successful manufacturing of biopharmaceuticals relies on good process design, scale up and control strategies. SSPMs have an important part to play in process design allowing for the generation of process understanding and defi nition of control strategies for commercial manufacture. However, to be useful, SSPMs have to be shown to represent the commercial scale process as closely as possible.

MVDA have advantages over traditional UVDA techniques in that they provide a fi ngerprint of the process and hence the outcome can be used to compare future batch profi le, fault detection and real-time process monitoring. Variation introduced by uncontrollable factors and any missing values are accounted in the model and the analysis does not rely on the assumption of normal distribution of data. The resulting MVDA plots, including the score plot, loading plot, VIP plot and batch control chart, illustrate batch evolution trends, determine clusters and outliers, and help in identifying infl uential variables dictating the batch trajectory and improving process understanding.

These statistics demonstrate that MVDA is a useful technique for extracting process understanding from multidimensional data sets and using that information, together with traditional one dimensional and UVDA, to gain confi dence in SSPM development.

References

  1. ICH guideline Q11 (Draft) on development and manufacture of drug substances (chemical entities and biotechnological/biological entities), EMA/CHMPICH/425213/2011
  2. Rathore A, Krishnan R, Tozer S, Smiley D, Rausch S and Seely J, Scaling down biopharmaceutical operations, Part 1: Fermentation, Pharmaceutical Technology, April 2006
  3. NIST/SEMATECH e-Handbook of Statistical Methods www.itl.nist.gov/div898/handbook/ 
  4. Montgomery DC and Runger GC, Applied Statistics and Probability for Engineers, 4th edition, John Wiley & Sons, Inc, New York, 2006
  5. Kourti T and MacGregor JF, Process Analysis, Monitoring and diagnosis, using multivariate projection methods, Chemometrics Intelligent Laboratory System 28: pp3-21, 1995
  6. Martin EB and Morris AJ, Enhanced bio-manufacturing through advanced multivariate statistical technologies, Journal of Biotechnology Progress 99: pp223-235, 2002
  7. CMC Biotech Working Group, A-Mab: a case study in bioprocess development, Version 2.1, 30 October 2009
  8. Eriksson L, Johansson E, Kettaneh-Wold N, Trygg J, Wikstorm C and Wold S, Multi- and megavariate data analysis, Part-I, Basic principles and applications, 2nd edition, Umetrics Academy, 2006
  9. Wold S, Sjostrom M and Eriksson L, PLS-regression: A basic tool of chemometrics, Chemometrics Intelligent Laboratory System 58, pp109-130, 2001
  10. Kirdar AO, Conner JS, Baclaski J and Rathore AS, Application of multivariate analysis towards biotech processes: Case study of a cell-culture unit operation, Journal of Biotechnology Progress 23: pp61-67, 2007

Read full article from PDF >>

Rate this article You must be a member of the site to make a vote.  
Average rating:
0
     

There are no comments in regards to this article.

spacer

Graham McCreath is Head of Process Design at Fujifi lm Diosynth Biotechnologies where he is responsible for late stage process characterisation, process validation and QbD. He has 16 years of experience in industrial bioprocess development gained at PPL Therapeutics, Avecia, MSD and Fujifi lm, in addition to four years’ post-doctoral experience in purifi cation process development. He has developed and supported a number of products derived from transgenic animals, mammalian and microbial sources throughout all stages of pharmaceutical development from Phase 1 to launched commercial product. He holds a PhD in Chemical Engineering from the University of Cambridge. Email: graham.mccreath@fujifi lmdb.com

Mahesh Shivhare is a Senior Process Statistician in the process design group at Fujifi lm Diosynth Biotechnologies. He oversees and provides statistical support to all phases of biopharmaceutical drug development, from early phase process development to process and product characterisation, validation as well as commercial manufacturing. He has experience in providing statistical support for both the small molecules and biologics pharmaceutical industries and has held similar positions previously with Avecia, MSD and Pfi zer. He holds a PhD in Chemical Engineering from Newcastle University. He is also a certifi ed Six Sigma Green Belt professional. Email: mahesh.shivhare@fujifi lmdb.com

spacer
Graham McCreath
spacer
spacer
spacer
Mahesh Shivhare
spacer
spacer
Print this page
Send to a friend
Privacy statement
News and Press Releases

New laboratory and regulatory services for biotech customers

USA/Germany, September 21, 2020. Gerresheimer adds essential laboratory and regulatory services to its broad portfolio of products, solutions and services for biotech companies. Specialized laboratory technicians and an experienced team of further specialists offer these services in the two innovation and technology centers in Bünde (Germany) and Vineland (New Jersey/USA). Gx Biological Solutions offers full service for biotech companies regarding product development of primary packaging and drug delivery solutions, material and functional testing as well as comprehensive regulatory support. The partnership with Gerresheimer shortens the time to market for pharmaceutical companies, minimizes risks and saves resources.
More info >>

White Papers

Advantages of Quantitative NMR for the Determination of Relative Response Factors

Novatia, LLC

Quantitative NMR (qNMR) is a technique that is being applied broadly and at an increasing rate in the field of pharmaceutical analysis (1). This white paper highlights the advantages of using qNMR to determine Relative Response Factors (RRFs) for pharmaceutical impurities detectable by HPLC. A single determination of RRFs using qNMR allows for simple and accurate quantitation of impurities which eliminates the need for preparation, qualification, and storage of reference standards. An example is presented here, which demonstrates quantitation of known impurities that have variable responses to UV-VIS detection, thereby providing a more accurate assessment of impurity levels than UV-VIS response alone.
More info >>

 
Industry Events

12th World Meeting on Pharmaceutics, Biopharmaceutics and Pharmaceutical Technology

8-11 February 2021, Vienna, Austria

In continuation of the very successful scientific meetings in Budapest, Paris, Berlin, Florence, Geneva, Barcelona, Malta, Istanbul, Lisbon, Glasgow and Granada, the 12th PBP World will be held in Vienna from 8-11 Feb 2021. This every two year held conference has gained an ever increasing impact among the pharmaceutical scientists: With up to 1000 submitted abstracts and about 1300 participants it has become a well-established major meeting, attracting scientists from all over the world.
More info >>

 

 

©2000-2011 Samedan Ltd.
Add to favourites

Print this page

Send to a friend
Privacy statement