PEER Stage2 10.1002 Biot.200800333
PEER Stage2 10.1002 Biot.200800333
PEER Stage2 10.1002 Biot.200800333
pharmaceuticals
Carl-Fredrik Mandenius
Wiley-VCH
Page 1 of 28 Biotechnology Journal
25
26 Products Agency, Uppsala, Sweden
27
28
er
29
30
31 Keywords: Design-of-Experiments; Design space; On-line monitoring and control;
32 Biopharmaceuticals; Active pharmaceutical ingredients
Re
33
34
35
36
vi
40
41 E-mail: cfm@ifm.liu.se
42
43
44
45
46 Abbreviations: CPP, critical process parameters; CQA, critical quality attribute; DoE,
47 design of experiments; EMEA, European Medicines Agency; FDA, United States Food
48
49 and Drug Administration, ICH, International Conference on Harmonization; MAA,
50
51 marketing authorization application; MVA, multivariate analysis; NIR, near-infrared;
52
53 NME, new molecular entities; PAR, proven acceptable range; PAT, process analytical
54
55
technology; QbD, quality by design; SPC, statistical process control; TPP, target product
56 profile
57
58
59
60
1
Wiley-VCH
Biotechnology Journal Page 2 of 28
25
26
27
28
er
29
30
31
32
Re
33
34
35
36
vi
37
38
39
ew
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
2
Wiley-VCH
Page 3 of 28 Biotechnology Journal
25
26
27 competitiveness where time constraints and increased customer quality demands are
28 significant. These goals are explained in a variety of guideline documents from national
er
29
30 bodies and worldwide industrial organizations, e.g. the ICH quality guidelines [1-4]. The
31
32 United States’ Food and Drug Administration (FDA) and the European Medicines
Re
33
34 Agency (EMEA) have adopted these guidelines in their regulatory framework for drug
35
development and production [5].
36
vi
37
38 The basics of QbD are illustrated in Figure 1. The critical quality attributes (CQAs) are
39
the properties of the product that characterize its quality; they must be guaranteed in
ew
40
41 manufacture, otherwise the product must be discarded. Typical examples of such
42
43 attributes are purity, stability, solubility, and product integrity, but ease of analysis is not
44
45 uniform. The CQAs are a result of the product itself, but are also highly dependent on
46
47 how the product is manufactured. This is controlled by the critical process parameters
48 (CPPs). If the CPPs are properly selected and tuned, the right CQAs will be achieved.
49
50 These must be maintained over time, a non-trivial task for biological processes given the
51
52 natural variations in such systems and the time-dependent behavior of most batch
53
54 operations.
55
56
57
58
59
60
3
Wiley-VCH
Biotechnology Journal Page 4 of 28
25
26 Figure 1. The design space and control space as defined in QbD
27
28
er
29
30
31 The design space is the region where the parameter values can lie, while the control
32
space defines the limits for their control. The design space is the multidimensional
Re
33
34 combination and interaction of input variables (e.g. material attributes) and process
35
36 parameters that have been demonstrated to provide assurance of quality. The control
vi
37
38 space is the subsection of the design space where the manufacturer chooses to operate. To
39
do this successfully, a detailed and time-dependent understanding is essential. This is
ew
40
41 where science can contribute significantly. To accomplish this, we need reliable process
42
43 analytical tools.
44
45 PAT covers those methods that are useful for designing, analyzing, and controlling
46
47 manufacturing through timely measurements (i.e. during processing) of critical quality
48
49 and performance attributes of raw and in-process materials and processes, with the goal
50
51 of ensuring final product quality. The term analytical is viewed broadly, and includes
52
53
chemical, physical, microbiological, and mathematical aspects, as well as risk analysis.
54 Typical PAT tools are those that enable scientific, risk-managed pharmaceutical
55
56 development, manufacture, and quality assurance.
57
58
59
60
4
Wiley-VCH
Page 5 of 28 Biotechnology Journal
25 In the lifecycle of a product, QbD starts with the definition of the desired product
26
27 properties. Then, during the process development phase, a multitude of input variables
28
er
29 are linked systematically with output variables. Identification of the critical variables and
30
31 parameters which have an impact on quality attributes (i.e. the CQAs) takes place at this
32 stage. In the interest of reproducibility and process economics, those variables and
Re
33
34 parameters that have an impact on process robustness should also be identified and
35
36 addressed thoroughly. At the same time, an appropriate process and product control
vi
37
38 strategy leading to a final drug product with pre-determined specifications should be
39
developed.
ew
40
41
42 As pointed out above, the final state of the product, the target product profile (TPP), must
43
44 be defined before process development for clinical supplies can begin. Exhaustive
45 physiochemical and bioanalytical characterization applying state-of-the-art methods in
46
47 the field of protein characterization are vital, in order to create a comprehensive picture
48
49 of the product and lower the risk of failures during the clinical development stage.
50
51 Information on the composition of amino acid sequence variants, the secondary and
52
tertiary structure, and post-translational modifications in detail; all this sets the stage for
53
54 the design of expression systems, cell line screening, and process development. For
55
56 example, the distribution of glycoforms of a given product may vary within mammalian
57
58 cell populations, and can influence the biological properties dramatically. Careful
59
60
5
Wiley-VCH
Biotechnology Journal Page 6 of 28
25
26 Source: Adapted from EFPIA, PAT Topic Group, 2005
27
28
er
29
30
31 Table 2 Important features of the QbD vision
32
Re
33
34
Aspect Starting point Final QbD approach (Vision)
35 (Minimal approach)
36 • systematic, relating input material attributes,
vi
25
26 production development, the most important question is that of which process parameters
27
28
impact CQAs, and to which degree and how controllably. A risk assessment may capture
er
29 the likelihood of occurrence and the potential effects of certain process parameters on
30
31 process robustness and, above all, on quality. A thorough risk assessment will therefore
32
Re
37 tightly. Furthermore, the ICH Q8 guidelines propose a control strategy that ensures the
38
39 TPP specification is reached in a reproducible manner [1]. The control strategy will be
ew
40
41
focused on (the critical) sources of variability, such as certain raw materials, but also the
42 inherent biological processes as a whole. The control strategy therefore encompasses all
43
44 input materials, unit operations, in-process testing (off-line, at-line, or on-line), and
45
46 finally the release tests. The level of process understanding strongly influences the design
47
48 of such a control strategy.
49
50 During development, the analytical results recurrently contribute to adaptation and
51
52 optimization of the production system, leading to continuous improvement and
53 refinement of both the process and the product specifications. PAT-based development
54
55 tools, such as Design-of-Experiments (DoE), mathematical modeling, and multivariate
56
57 data analysis (MVA), when carefully applied during the characterization, will create a
58
59
60
7
Wiley-VCH
Biotechnology Journal Page 8 of 28
29
30 robustness. As a result, a QbD design space will be created which could justify regulatory
31
32 flexibility. Important issues associated with this task are the inhomogeneities of signal
Re
33
34 traces due to spikes, and the generally much lower number of batches used for
35
36 biopharmaceuticals in comparison to pharmaceutical products.
vi
37
38 A number of spectroscopic tools are being developed in order to improve the control
39
ew
40 options for biological processes. One example is PTR mass spectrometry (Ionimed
41
42
Analytik GmbH, Innsbruck, Austria) for assessment of patterns of volatile fermentation
43 metabolites in real-time, thereby giving more insight into process performance and,
44
45 ultimately, better process robustness.
46
47 For a company like Sandoz, QbD means a substantial investment in technologies,
48
49 strategies, and skilled people. Implementation of QbD is, no doubt, a long-term
50
51 undertaking and requires resources not only in technical development but also in
52
53 production, quality, and regulatory support functions (see Table 2). Such investment
54
should ultimately pay off by reducing production costs, by a more consistent quality, and
55
56 by providing a more profound, science-based insight into biopharmaceutical processes.
57
58
59
60
8
Wiley-VCH
Page 9 of 28 Biotechnology Journal
25
26
27 The main enabling PAT tools of QbD, when applied to biopharmaceutical processes,
28 consist of (1) chemometrics computation methods and (2) bioanalytical methods and
er
29
30 devices. Below the basics are described for applying chemometrics to QbD followed by a
31
32 few examples of novel applications of bioanalysis.
Re
33
34
35
36
Chemometrics for QbD
vi
37
38
39 Design-of-Experiments (DoE) [6,7] has lately experienced a revival in the
ew
40
41
pharmaceutical industry due to initiatives towards QbD. The key factor in QbD is
42 knowledge of the process, and in general mechanistic models are seen as the final aim
43
44 and proof of complete understanding. DoE is, properly applied, a powerful tool both in
45
46 the development of knowledge and in the determination of mechanistic models.
47
48 In short, one can say that to gain a higher level of knowledge, experiments are necessary.
49
50 By planning the experiments according to DoE, the information gained from each
51
52 experiment is maximized, thus increasing the prospects for solving the problems
53
addressed. Using DoE as a basis for collecting information enables the calculation of
54
55 predictive mathematical models that describe the relationships between changes in
56
57 experimental conditions and the outcome of the experiments. Depending on the type of
58
59
60
9
Wiley-VCH
Biotechnology Journal Page 10 of 28
25
26 Once the DoE has been finalized, the data analysis remains. The length of the
27
28
fermentation processes means that sophisticated tools for data analysis are required, such
er
29 as MVA, which allows analysis of process evolution. With DoE and MVA in
30
31 combination a process can be characterized, the main drivers identified, and a deeper
32
Re
37
38 detection of deviations, and identification of the cause of deviation. One value of these
39
ew
40 models is that they generally detect drifts earlier than classical statistical process control
41
42 (SPC) charts, since multivariate models also identify deviations due to changes in
43
44
correlation pattern in the process.
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
10
Wiley-VCH
Page 11 of 28 Biotechnology Journal
25
26
27
28
er
29
30 Combine for final quality modeling
31
32
Figure 2. By combining DoE and multivariate data analysis techniques, a multi step
Re
33
34 process can be summarized in one overview model. This model links the entire history of
35
36 a lot or batch to its final properties.
vi
37
38
39
ew
40
41
Continuous development of DoE and MVA tools is required as new data types emerge
42 and the conditions for experimentation change; however, the tool box available today is
43
44 usually sufficient for the current situation. The main limitations today are rather related to
45
46 IT structure; that is, availability of data and data organization. There is also a need for
47
48 improvement of in-line analytical instruments.
49
50 In order to turn DoE and multivariate techniques into commonly used tools, both in
51
52 development and production, there is a need to simplify and adapt the toolbox for each
53
54
specific application. The chemometrics supplier company Umetrics AB (Umeå, Sweden)
55 provides highly qualified solutions for these techniques by creating user-friendly software
56
57 without compromising the technical functionality and complexity. One successful
58
59
60
11
Wiley-VCH
Biotechnology Journal Page 12 of 28
25
26 invasive and non-destructive techniques, no sampling, high frequency of spectra
27
28 acquisition, and large number of molecules potentially quantified. Their limitations
er
29
30 mainly concern analysis calibration, chemometric processing of the spectra, and previous
31
32 identification of strategic operating parameters and set-points. Papers reporting data on
Re
33 this subject are focused on off-line [8-9], on-line [10], or in situ [11-12] NIR analyses,
34
35 mainly performed with suspension cell culture of CHO or Sf-9 cell lines. Ren and Arnold
36
vi
40
41
with adherent cells cultivated in stirred reactors.
42
43 Proper evaluation of NIR and Raman spectroscopy methods is highly desirable for cell
44
45
cultures. One example is adherent VERO cells cultivated in serum-free medium and on
46 microcarriers in spinner flasks and stirred bioreactors (CNRS, Université-Nancy, France).
47
48 The quantification of lactate and glucose was performed on culture media samples by off-
49
50 line measurements, and compared to data from reference enzymatic methods. NIR
51
52 spectroscopy data from chemometric analyses of spectra could be correlated to reference
53 results, contrary to the Raman results. Furthermore, no influence of microcarriers on
54
55 profile and quality of acquired NIR data was noticed, while Raman spectroscopy
56
57 presented an interesting sensitivity to cell physiological state. In a second experiment, in
58
59
60
12
Wiley-VCH
Page 13 of 28 Biotechnology Journal
29
30
31 is required to produce energy for both cell growth and energy maintenance. For example,
32 the aerobic degradation of carbohydrates (e.g. glucose) by glycolysis and the respiratory
Re
33
34 chain reaction in the mitochondrion both depend on oxygen.
35
36 Ideally, the conversion of 1 mol glucose and 6 mol O2 yields 6 mol CO2, biomass and
vi
37
38 energy. Thus, a strong relationship between glucose, oxygen consumption, and carbon
39
ew
25
26 molecules and other small molecular metabolites [16].
27
28 There is no doubt that analytical applications of genomic, proteomic, and transcriptomic
er
29
30 methods provide powerful tools to PAT. The omics tools have a unique possibility to
31
32 relate quality aspects to the CPP through a better understanding of systems biology
Re
33 aspects in biopharmaceutical production [17]. So far, relatively few reports are available
34
35 on how identification and reduction of metabolic/physiological bottlenecks of the
36
vi
37 biological systems can influence the quality of the production, and how this
38
39 understanding can subsequently be exploited with QbD methodology.
ew
40
41
42
43
44 4 Regulatory aspects on the introduction of QbD for biopharmaceuticals
45
46 Regulatory considerations for biological molecules as pharmaceutical products
47
48
The viewpoints reported above have to be in perfect compliance with the long term
49
50 regulators’ requirements on the biotechnology-related pharmaceuticals. Especially
51
52 biological processes and starting materials are more prone to variation, and the molecules
53
54 are often large and may be conjugated, for example as glycosylated and pegylated forms of
55
56
the product. These may be sensitive to oxidization, deamidation, and aggregation/
57 fragmentation. Changes may influence activity, but changes in for example glycosylation
58
59
60
14
Wiley-VCH
Page 15 of 28 Biotechnology Journal
25
26 flexibility within a given process is allowed if validated. The introduction of a properly
27
28
validated and documented QbD system and design space in the manufacture of biological
er
29 medicinal products therefore meets no formal obstacles. It should be noted that introduction
30
31 of QbD is optional, and the classical way with more emphasis on end product testing will
32
Re
37 the raw and starting material (including cell banks) is particularly important for a
38
39 biological product, where the biological origin of the material can be expected to
ew
40
41
introduce variability. In-process controls are performed to monitor whether the process is
42 behaving as expected. The process should be validated and shown to be robust and
43
44 reproducible. The production should be performed under Good Manufacturing Practices
45
46 (GMP), in order to ensure reliability, to assure adherence to approved production
47
48 methods, and to avoid the risk of contamination.
49
50 As a final part of the quality control comes the testing of the active substance and the
51
52 finished medicinal product. These tests do not cover all aspects of the product, but a
53 number of relevant tests are picked to mirror the product characteristics under normal
54
55 production conditions. From these, and by taking risk assessment aspects into account,
56
57 the control strategy is formed. All parts need to be included, but the focus put on each of
58
59
60
15
Wiley-VCH
Biotechnology Journal Page 16 of 28
25
26 simplify the establishment of models and allow the ruling out of non-significant
27
28
differences.
er
29
30 For a deeper understanding of the processes, it is also important to evaluate potential
31
32 interactions between process parameters; by multivariate analysis or other methods [4].
Re
33
34
35
36 Aspects to consider in the introduction of QbD-based approaches
vi
37
38
In the cases when similar molecules are produced, for example, monoclonal antibodies,
39
ew
40 platform technologies are often used. However, prediction models have been shown to be
41
42 invalidated by small differences in, for example, buffers. To be fully able to extrapolate
43
44 from one product to another, it is important that the model is shown to also be fully
45
46 relevant to products for which the platform was not initially established.
47
48 According to the control strategy concept, the sole fulfillment of a specification cannot be
49
50 considered a design space, since end testing is only one part of the strategy and the
51 specification will be process specific. Deviations from the accepted process may result in
52
53 impurities that will go unnoticed, as the specification is not designed to cover them.
54
55
56
57
58
59
60
16
Wiley-VCH
Page 17 of 28 Biotechnology Journal
25
26 The new concept will have an impact on the work of regulators. The basis for approval
27
28
will not only be assessment of documentation submitted in the MAA, but also
er
29 documentation considered during inspections and assessment of the quality systems [3]
30
31 and risk assessment strategies [2] in place. This requires a close collaboration between
32
Re
33 assessors and inspectors. To help with this and to keep up with the evolution of the field,
34
35 a specialized group, the EMEA PAT team, has been established. This group consists of
36 members from the Quality Working Party, the Biologicals Working Party, and the ad hoc
vi
37
38 Inspectors’ Working Party. It has regular contact with industry organizations and other
39
ew
40 regulatory agencies to discuss common issues and provide expert knowledge for
41
42 applications dealing with QbD/PAT issues. Reports and question-and-answer sessions are
43
44
published on the EMEA website [20-22].
45
46 As pointed out above, it is vital to make the link to the processes used to show safety and
47
48 efficacy in clinical trials. All flexibility introduced and all changes made to production
49 must result in a comparable product in all important aspects; relaxation of the level of
50
51 assurance in this respect compared to the current situation will not be accepted. A change
52
53 in the control strategy (e.g. less end product testing, increased flexibility within an
54
55 approved design space) is however allowed for, as long as equal or better quality can be
56
assured. Good transparency is of major importance for the successful introduction of
57
58
59
60
17
Wiley-VCH
Biotechnology Journal Page 18 of 28
25 The workshop resulted in a number of recommendations for how QbD in the area of
26
27 biotechnology-related pharmaceuticals can be improved and facilitated. The
28
er
29 recommendations are far from comprehensive, but are intended to highlight a few
30
31 pertinent issues. The following issues appear to be of particular urgency and relevance:
32
Re
37
38 overview, and so efficient analysis becomes unrealistic. Further development of
39
ew
40 knowledge management systems [3], including more powerful data management software
41
42 tools, would be a valuable asset in mining for useful information.
43
44 ii) Retrieval of historical data. When it comes to the use of historical data, one hurdle is
45
46 the availability of data. It is commonly stored for storing instead for stored for use,
47
48 meaning that it is difficult to retrieve and organize for multivariate modeling. This issue
49 has been recognized and addressed by many providers of data storage systems. It should
50
51 also be remembered that in most cases historical data does not contain enough variability
52
53 to yield all valuable information about a process. For existing processes, historical data is
54
55 usually the first step in a QbD investigation, but to fully explore the process a DoE is
56
57
required.
58
59
60
18
Wiley-VCH
Page 19 of 28 Biotechnology Journal
25 different origin and have a different impact on risk. Thus, a critical analysis of
26
27 parameters, occurrence, and measurability would facilitate prioritization of specific
28
needs.
er
29
30
31 v) New real-time or just-in-time biospecific analyses and assays. There is a need for
32
Re
37
38 and side-product analysis in biotech process steps – which are not easily covered with
39
ew
40 typical systems biology analysis but call for more tailor-made assays. Biochemical
41
42 engineering could contribute to further development of such methods on the basis of
43
44
existing bio-recognition principles (e.g. biosensors, advanced multidimensional
45 analyzers).
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
19
Wiley-VCH
Biotechnology Journal Page 20 of 28
29 [5] Federal Food and Drug Administration (USA) Centre for drug administration and
30
31 research. Guidance for industry, process analytical technology, a framework for
32
Re
37 [6] Eriksson, L., Johansson, E., Kettaneh-Wold, N., et al. Design of Experiments:
38
39 Principles and Applications. Umetrics AB, Umea, Sweden, 2008.
ew
40
41 [7] Mandenius, C.-F., Brundin, A., Review: Bioprocess optimization using DoE
42
43 methodology, Biotechnol. Prog. 2008, in press
44
45 [8] Arnold, S. A., Crowley, J., Woods, N., Harvey, L. M., McNeil, B., In-situ near
46
47 infrared spectroscopy to monitor key analytes in mammalian cell cultivation,
48
49 Biotechnol. Bioeng. 2003, 84, 13-19.
50
51 [9] Ren, M., Arnold, M., Comparison of multivariate calibration models for glucose,
52
53 urea, and lactate from near-infrared and Raman spectra. Anal. Bioanal. Chem. 2007,
54
55
387, 879-888.
56
57
58
59
60
20
Wiley-VCH
Page 21 of 28 Biotechnology Journal
25
26
27 [14] Oezemre, A., Heinzle, E., Measurement of oxygen uptake and carbon dioxide
28
production rates of mammalian cells using membrane mass spectrometry.
er
29
30 Cytotechnology 2001, 37, 153-162.
31
32
[15] Eliasson Lantz, A., Jorgensen, P., Poulsen E., Lindemann C., Olsson, L.,
Re
33
34 Determination of cell mass and polymyxin using multi-wavelength fluorescence.
35
36 J. Biotechnol. 2006, 121, 544-554.
vi
37
38
[16] Clementschitsch, F., Bayer, K., Improvement of bioprocess monitoring,
39
ew
25
26
27
28
er
29
30
31
32
Re
33
34
35
36
vi
37
38
39
ew
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
22
Wiley-VCH
Page 23 of 28 Biotechnology Journal
25
Source: Adapted from EFPIA, PAT Topic Group, 2005
26
27
28
er
29
30 Table 2 Important features of the QbD vision
31
32 Starting point
Re
37
• fixed • adjustable within design space
38 Manufacturing • focus on reproducibility • focus on control strategy and robustness
39 process • validation based on full-scale lifecycle approach to validation, continuous
ew
25
26
27
28
er
29
30
31
32
Re
33
34
35
36
vi
37
38
39
ew
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
24
Wiley-VCH
Page 25 of 28 Biotechnology Journal
25
26
27
28
er
29
30
31
32
Re
33
34
35
36
vi
37
38
39
ew
40
41
42
43
44
45
46
47
Figure 1. The design space and control space as defined in QbD
48
49
50
51
52
53
54
55
56
57
58
59
60
25
Wiley-VCH
Biotechnology Journal Page 26 of 28
25
26
27
28
er
29
30
31
32
Re
33
34
35
36
Combine for final quality modeling
vi
37
38
39
ew
40
41
42
43
44 Figure 2. By combining DoE and multivariate data analysis techniques, a multi step
45
46 process can be summarized in one overview model. This model links the entire history of
47
a lot or batch to its final properties.
48
49
50
51
52
53
54
55
56
57
58
59
60
26
Wiley-VCH
Page 27 of 28 Biotechnology Journal
1
2
3
4 Design space
5 (the area which has been demonstrated
6
7 to give adequate quality)
8
9
10
Fo
11
12
13
rP
14
15
Control space
16 (where the production
17
18
19 ee
will operate)
CQA 2
rR
20
21
22
23
24
25
ev
iew
26
27
28
29
30
3
31
A
32
CQ
33
34
35
36
37
38
39
CQA 1
40
41
42
43
44
45
46 Wiley-VCH
47
48
Biotechnology Journal Page 28 of 28
1
2
3
4
5
6
7
8
9
10
Fo
11
12
13
rP
14
15
16
17
18
19
Lab
scale
ee Seed
Fermentation
Primary
Fermentation
rR
20 s
21 Identify operating window of individual unit operations:
22
23
24
25
ev
iew
26
27
28
29
30
31
32
33
34
35 Combine for final quality modeling
36
37
38
39
40
41
42
43
44
45
46 Wiley-VCH
47
48