Umetrics Suite Blog

Can data analytics improve biopharma manufacturing processes for autologous treatments like CAR-T?

December 13, 2018

Advancements in cell and gene therapy hold promise for the future of personalized medicine, especially for cancer treatments. However, bioprocessing methods for autologous cellular therapies, and CAR-T in particular, often present unique challenges in manufacturing due to the variability of the starting material and unique nature of each batch. Is there a way to create more efficient processes in order to bring down costs and make personalized medicine a viable option for more patients?

MVDA could revolutionize cancer treatment with CAR-T cells

Advanced analytics has the potential to revolutionize the way autologous treatments such as CAR-T for cancer patients are produced.

At Sartorius Stedim Data Analytics, we propose that off-the-shelf data analytics tools and techniques can be applied to improve autologous therapies. Through better categorization of starting materials, process optimization, flexibility in manufacturing, enhanced real-time fault detection and prediction of critical quality attributes (CQAs), it can be made possible. Let’s see how.

Opportunities in CAR-T manufacturing

The cost of CAR-T therapies are incredibly high, coming in at about $300,000-$500,000 per treatment [1]. Manufacturing involves two production processes:

  • CAR-T cell production (CAR=chimeric antigen receptor)
  • Lentiviral vector production (lentiviral vectors are used to introduce genes into the T-Cells)

CAR-T manufacturing involves small batch processes that create individual treatments for each patient from their own biological material. Autologous cellular therapy means the patient’s own cells are used as starting material (in this case T cells, a type of white blood cells that play an important role in the immune system).

Individual treatments increase the likelihood of meeting patient-specific needs — while also introducing challenges. For example, in autologous individual therapies, the starting material may be quite different from one sample to another, making it challenging to apply a standard production process to each batch that’s being run. Small batch sizes also create challenges in applying the type of efficiencies that are typically used in large-scale production of more traditional bioprocesses.

The lentiviral vector production process also involves a smaller scale production than typical biopharma manufacturing processes. Although more of the traditional scalable technologies can be applied than with CAR-T cell manufacturing, there is still a strong demand to improve the processes and reduce bottlenecks in production to ultimately help these gene therapies to become more globally available.

Why multivariate data analytics?

Advanced data analytics can be applied to support and improve CAR-T manufacturing and lentiviral vector manufacturing. Multivariate data analytics (MVDA) helps uncover and categorize the variability in the results. Variability in the raw material for autologous treatments increases the risk of variability in the final product. With sufficient data, we can model the variability in quality and perhaps even predict patient response to the treatment.

By applying projection methods to associated data, clinical samples (or, starting material) can be characterized and any groups revealed and correlated to final results. Depending on the nature of the variability among the samples, they may benefit from different operating conditions through the manufacturing process in order to obtain consistent batch to batch results.

mvda data many sources

Multivariate data analysis allows data collected from a variety of sources to be analyzed in a way that provides more actionable insights compared to univariate data analytics.

How can design of experiments be used?

Design of experiments (DOE) is a systematic method of determining the relationship between factors affecting a process and the output of that process. It’s a way of finding cause-and-effect relationships. Carefully planned experiments allow you to gain the most information with the least effort or repetitions. The information resulting from the experiments can then be used to optimize the process and improve the output. DOE is an important component in Quality by Design (QbD) and Design Space as defined in ICH Q8.

A combined powerful approach

DOE and MVDA, along with real-time data analytics technology, provide a unique and powerful opportunity to support small batch processing.

Provided adequate data collection and automation are available, a combination of DOE and MVDA can be used to establish a design space that corresponds to high product quality in a process that allows flexible operating conditions while accounting for variability in the starting material. Using model predictive control, processes can become more automated and less dependent on operators making manual adjustments when process deviations occur. Incorporating combinations of data from various data sources is possible with MVDA.

CAR-T Manufacturing Process

Opportunities for analytics in CAR-T manufacturing exist throughout the different steps of the processes, wherever data is available.

Leukapheresis 

In autologous therapies, properties of collected cells may vary significantly from patient to patient and, as a result, different operating conditions may be necessary to achieve the same process outcome. It’s possible that patient response to the final treatment may vary, as well.

Provided sufficient data, the properties of the clinical material can be mapped and linked to cell concentrations and other quality attributes in the finished treatment. The possibility to categorize and predict patients as “responders”, “partial responders,” and “non-responders”, and a roadmap for how to optimize downstream operating conditions for different starting materials, could be of critical value, particularly in time-sensitive situations.

using historical data to map outcomes for CAR-T treatment responses

OPLS-DA is a common technique to categorize samples in a dataset and associate group separations with variable patterns. Left: Score plot from OPLS-DA model showing group separation. Right: Variables associated with separation of the groups.

USE OPLS-DA to classify patients

GC/MS and OPLS-DA applied to separate glioblastomas from oligodendrogliomas of different grades, followed by identification of metabolic patterns associated with survival rates. [2]

Cell expansion 

The cell expansion step of a CAR-T manufacturing process is a bioreactor process in which data can be continuously collected and stored using an automation system and a process historian.

With the right historical data, MVDA can be used to establish a map of representative process evolution that can be linked to cell concentrations in the finished product, enabling real-time process monitoring and early fault detection. Process forecasting and model predictive control are extensions to this application that can be used for open- or closed-loop control. [3]

 

predictive monitoring

Real-time process monitoring, forecasting and control allow early detection of process faults and immediate action.

Lentiviral production optimization

Lentiviral vectors are used to introduce genes into the T-cells to deliver chimeric antigen receptors (CARs) [5]. The lentiviral vectors are manufactured in a separate process for subsequent introduction in the gene transfer step of the CAR-T manufacturing process. Hence this process is subject to its own events that indirectly become a factor in manufacturing CAR-T cell therapies.

For example, lentivirusis sensitive to pH, temperature, shear and salt [4], and the virus produced can be cytotoxic to the producer cells, which may lead to variable amounts and quality of virus. Furthermore, quantitative measurement of infectious titer can take time and downstream processing may occur without knowing the infectious titer.

DOE and MVDA may be used to optimize operating conditions, increase process understanding, build soft sensors to predict yield and quality attributes, etc. Where automation and continuous data collection are available, real-time applications can further support a more optimal and consistent process. Example applications include:

  • Classification of cell and plasmid characteristics/correlation to physical/infectious titer
  • Definition of critical process parameters
  • Optimization of reactor conditions
  • Comparison of quality of first and second harvest
  • Process monitoring, early fault detection, forecasting and control

Summary and conclusions

There are many situations where data analytics can support and strengthen CAR-T cell therapy manufacturing. With a diversity of raw material, having a means of classifying unique individual samples supports production of safe and effective therapies.

MVDA used with DOE can help define critical process parameters and their relationship to the critical quality attributes thereby managing yield and quality, improving outcomes and securing risk management.

Key to making it work is availability of necessary instrumentation and sensory technology for characterization and quantification of samples and process, as well as automation and data infrastructure for real-time analytics.

What to know more?

Register to watch this previously recorded webinar on Batch Process Data analysis. 

Watch webinar

 

 You might also like: 

Using data analytics to optimize design space and setpoint conditions for bioreactors

 

 

References

1. Hernandez, Immaculad. (April 26, 2018). "Analysis determines true cost for CAR T-cell therapy," Helio - In the Journals Plus. Accessed Dec. 3, 2018.

2. Mörén, L., Bergenheim, A T., Ghasimi, S., Brännström, T., Johansson, M. et al. (2015) "Metabolomic screening of tumor tissue and serum in glioma patients reveals diagnostic and prognostic information," Metabolites, 5(3): 502-520

3. McCready, C., “Model Predictive Control for Bioprocess Forecasting and Optimization,” BioProcess International, 17 Nov, 2018.

4. Zimmermann, Katrin et al. "Highly efficient concentration of lenti- and retroviral vector preparations by membrane adsorbers and ultrafiltration," BMC biotechnology vol. 11 55. 20 May. 2011, doi:10.1186/1472-6750-11-55.

5. Milone, M., O’Doherty, U., “Clinical use of lentiviral vectors,” Leukemia 32, 1529-1541 (2018)

 

Topics: Real Time Data Analytics, Manufacturing Quality Control, Pharmaceutical manufacturing, Medicine/Health

Anna Persson

Written by Anna Persson

Anna Persson is the Senior Principal Data Scientist for Sartorius Stedim Biotech based in the Greater New York City Area. Anna is responsible for the online and off-line implementation of the Umetrics Suite of Solutions (SIMCA, SIMCA-online, Control Advisor, SIMCA-Q, MODDE, Easy Analytics, Active Dashboard) in customer environments. Anna is a graduate of Umeå University.