Umetrics Suite Blog

Which Data Analytics Methods Best Support a Quality by Design Approach for Biosimilars?

November 12, 2019

Biosimilars are an exciting route to increasing access to the highly effective therapy made possible by biologics, but ensuring a biosimilar meets the critical quality attributes (CQA) of the original biologic is a major challenge. Optimizing production at full scale is impractical, which makes a quality by design (QbD) approach using a reliable scale down model of the process an attractive alternative. A process development team at Zhejiang Hisun Pharmaceuticals, Taizhou, China, therefore developed a scaled down model of the cell culture process used to produce the biosimilar adalimumab. They qualified the model using multivariate data analysis (SIMCA), and explored the design space for key process attributes (KPA) and CQAs using MODDE.

quality by design for biosimilar development pharma

Using advanced data analytics methods such as MVDA and DOE provides an effective quality by design approach (QbD) for biosimilar development.

The promise and challenge of biosimilars

As patents for high cost biologics expire, many pharmaceutical companies are pursuing biosimilars as a low-cost alternative with equivalent efficacy, especially in markets where high-cost treatment is not feasible¹. However, there is great concern about the safety and efficacy of biosimilars. While meeting CQAs is vital in the manufacture of biopharmaceutical products, the need for CQAs of biosimilars to consistently meet the targets set for the original drug is particularly challenging for bioprocess development.

Scale-down models are powerful predictive tools

Optimizing processes used to manufacture biologics such as biosimilars at full scale is impractical due to high costs, limited resources, and the work needed to evaluate hundreds of independent variables. A viable alternative to support process design that is in line with ICH guidelines (2) is to develop a scaled-down model (SDM) that represents the proposed large-scale commercial process, backed up by a qualification process that confirms the SDM has a predictable relationship with the full-scale process. This approach was tested by the process development team at Zhejiang Hisun Pharmaceuticals for the cell culture process designed to produce the biosimilar adalimumab³.

Production-scale manufacturing of the biosimilar adalimumab involved a 750 L stainless-steel bioreactor with a working volume of 500 L that was used to produce nine batches with a temperature-shift fed-batch process for performance qualification and clinical studies. In-process data and process and product quality attributes were collected for scale-down model qualification. A 3 L bioreactor with an effective volume of 2 L was used to develop bench scale models and explore the design space. The culture duration was 11 to 12 days. Scale-independent variables comprising seed density, temperature, dissolved oxygen (DO), feeding viable cell density (VCD), and pH, were kept consistent across scales. Constant power per volume (P/V) was selected as scaling down criterion for dependent variables including agitation.

Daily samples were collected aseptically for analysis. The final antibody titer and CQAs including aggregates, acid peak, total afucosylation level and high mannose content were evaluated as dependent variables.

As patents for high cost biologics expire, many pharmaceutical companies are pursuing biosimilars as a low-cost alternative with equivalent efficacy, especially in markets where high-cost treatment is not feasible¹.


Testing comparability using multivariate data analysis

Several approaches have been used to qualify scale-down models, including risk analysis, Student’s t test, quality range approach and equivalence, but these are limited by individual CQA or CPP comparability and therefore cannot capture the linkage and relationships among variables. The team therefore chose another method, multivariate data analysis (MVDA) to compare the full-scale process and SDM by projecting the multidimensional datasets into a few principle components to enable exploration of linkage between variables. MVDA has been used to evaluate comparability in other cell culture processes (3,4,5).

In this case, SIMCA software (v14.1) was used to generate a batch level partial least squares (PLS) model based on nine batches of cell culture process datasets at the full scale of 500 L. Eight independent variables were projected into four principal components, which were linear combinations of the original variables and Principal Component 1 (PC1) explained as much as 92.8% of the total variance. The batch statistical process control charts for PC1 showed a well-controlled process (Fig. 1a), and the model was used to successfully predict the performance of the small-scale culture (Fig 1b).

batch level partial least squares model for multivariate data analysis

Figure 1. The scores of PC1 over time with 3 standard deviations from full-scale marked as red dotted lines: (a) Real data from full-scale production, (b) predicted performance of the scale-down model based on a PLS batch evolution model. From Figure 2, Nie et al, 2019.

Watch Webinar: Design of Experiments (DOE) and Quality by design (QBD) in  Bioassay Optimization

Testing for equivalence

An equivalence test was used to compare KPAs, including final product titer, iVCD, and viability, while the CQAs included aggregates, acid peak, total afucosylation level, and high mannose content. The analysis confirmed equivalence for all the attributes between the two scales.

1-step DSD method for process characterization

Traditionally, process characterization has involved a 2-step experimental design involving low-resolution screening followed by response surface design and exploration. The complexity of cell culture processes has stimulated the development of a 1-step method. In this Definitive Screening Design (DSD), the main effects are uncorrelated with two-factor interactions and quadratic effects, and two-factor interactions are not confounded with each other, a key advantage over standard screening design. As a result, DSD is efficient in identifying main linear effects, main quadratic effects, and two-factor Interactions, and is being increasingly used in process development and characterization of biologics.

The DSD method was used for experimental design and model development. Five independent variables recognized as CPPs that might not be well controlled were evaluated: pH, shifted temperature, inoculation seeding density, VCD at first feeding, and VCD at temperature shift. Dependent variables included final product titer and four CQAs including aggregates, acid peaks, total afucosylation and high mannose content, but acid peak and total afucosylation were excluded due to lack of relationships to independent variables. Only 13 runs were required to identify the main effects and interactions.

Three quadratic polynomial models with good fits for final product titer, aggregates, and high mannose content were established and could be used to investigate the effect of independent variables on the CQAs in preparation for design space development.

Read more: Creating a design of experiments study to predict formula robustness

Design space development

Design of Experiments (DOE) has been increasingly used in the biopharmaceutical industry for product and process development as a means to satisfy regulations around QbD and to derive a combination of parameters that ensure the product meets the defined quality attributes, also known as a Design Space.

With the SDM, the design space for both key process attributes (KPAs) and CQAs could be established with MODDE® Pro using a probability-based Monte Carlo simulation method (Fig. 2). The design space could be verified, indicating that the target specification can be achieved by conditions with the design space even in a worst-case scenario involving the lowest final product titer and highest content of aggregates and high mannose. The design space developed for the process can be extended to full-scale process control and will be further validated.


Figure 2. The design space. The green areas indicate a probability of failure of less than 1%. The largest hypercube design space is shown as a gray dashed frame and the optimal robust setpoint is shown as meeting arrows. A, pH; B, shifted temperature; C, inoculation seeding density; D, VCD at first feeding; and E, temperature shift. From Figure 7, Nie et al, 2019.

Watch webinar: Design of Experiments for the beginner

Combining MVDA and DOE – a powerful tool for quality by design

This case study proposes a roadmap for the characterization of a cell culture process that is inline with ICH guidelines and uses a QbD approach based on a reliable scale down model. The resulting design space confirmed how product CQAs could meet the specifications for the biosimilar. This combination of MVDA and DOE can be applied for other complex dataset comparisons, such as process changes, technology transfer, and process scale-up.

Find out more

Find out more about D-optimal design or the basics of using MODDE for Design of Experiments in one of our recorded webinars.

Watch webinar: D-Optimal Design- What it is and when to use it

Watch Webinar: Design of Experiments (DOE) and Quality by design (QBD) in  Bioassay Optimization


Get MODDE Free Trial


1. Gupta, SK et al, Opportunities and Challenges in Biosimilar Development. Bioprocess International, Thursday, May 18, 2017.

2. ICH Q11. Pharmaceutical quality system. 2012. quality/article/qualityguidelines.html. Accessed 23 Feb 2019.

3. Nie, L et al. Development and Qualification of a Scale-Down Mammalian Cell Culture Model and Application in Design Space Development by Definitive Screening Design AAPS PharmSciTech (2019) 20:246. DOI: 10.1208/s12249-019-1451-7

4. Tescione L, et al. Application of bioreactor design principles and multivariate analysis for development of cell culture scale down models. Biotechnol Bioeng. 2015;112(1):84–97. 10.1002/bit.25330.

5. Tsang VL, et al. Development of a scale down cell culture model using multivariate analysis as a qualification tool. Biotechnol Prog. 2014;30(1):152–60.

6. Rathore AS, et al. Monitoring quality of biotherapeutic products using multivariate data analysis. AAPS J. 2016;18(4):793–800.


Topics: Design of Experiments (DOE), Quality by Design (QbD), Pharmaceutical manufacturing

Kai Touw

Written by Kai Touw

Kai Touw is a (Bio)Pharma Market Manager at Sartorius Stedim Data Analytics. He is a driven and enthusiastic technology evangelist bridging the world of data science into pharmaceutical and biopharmaceutical processing.

Search the Blog

    Subscribe to the Blog

    View the:

    Data Analytics Glossary of Terms

    List of Webinars

    Get a Free Trial