Perspectives on Process Analytical Technology

Looker_Studio - Stock.adobe.com

Quality is paramount for pharmaceutical and biopharmaceutical products. The process analytical technology (PAT) and quality-by-design (QbD) paradigms from FDA imbibe this by encouraging manufacturers to ensure consistency in product quality by continuously monitoring process streams and unit operations (1,2). PAT is recognized as a critical fundamental shift in inspecting and approving procedures for the pharmaceutical production process continuous process verification. FDA’s Center for Drug Evaluation and Research (CDER) has acknowledged the need for FDA guidelines to make PAT implementation easier (3).

Despite the regulatory push and the considerable interest shown by pharmaceutical manufacturers at the initial stages, the industry gradually lost interest as it became clear that the purported promise of real-time release testing (RTRT) is unlikely to be realized anytime soon. However, as more and more pharmaceutical manufacturers look to implement continuous processing, efforts toward creation and implementation of PAT have been rekindled. This is because, as in continuous processing, individual batches are not distinct, and critical quality attributes (CQAs) of the therapeutic product must be steadily maintained, monitored, and controlled over the time of operation, rather than merely at the end of each unit operations or batch (4).

Traditional end-of-batch quality testing methods, such as analytical high-performance liquid chromatography (HPLC) and ultraviolet-280 (UV-280) spectroscopy, are being increasingly adapted into in-line modalities with periodic sampling from the various unit operations due to the need for real-time monitoring and control of CQAs and critical process parameters (CPPs) during continuous manufacturing (5–7). This has also sparked the emergence of new analytical tools that can be integrated into process flow streams and deliver information in near real-time. In general, spectroscopic probes and flow cells are thus well-suited to PAT applications, primarily when used in conjunction with multivariate data analytics (MVDA) tools, which are imperative for reducing the dimensionality of large spectral datasets and extorting statistically significant quantitative information (8).

The American Association of Pharmaceutical Scientists (AAPS) PAT community aims to bring together scientists from all over the world to share knowledge and perspectives on the current development of PAT in the pharmaceutical industry. This article summarizes the views expressed by some of the PAT researchers that presented in webinars hosted by the AAPS PAT community (9,10). It also outlines various bioprocess-related applications and presents the commensurate future vision for the biopharmaceutical industry to achieve Industry 4.0 goals while meeting regulatory obligations. The article presents recent PAT approaches combining novel sensors, digital twin models and data science, which together offer unique solutions for pharmaceutical and biopharmaceutical process control (Figure 1).

Figure 1. Illustration of the role played by process analytical technology (PAT), digital twins, and data analytics in implementation of a robust control strategy. Figure courtesy of the authors.

Process models are key enablers of PAT

Advanced process models and digital twins. A digital twin is a virtual clone of a process and a crucial element for digitalization and pharma 4.0. The overarching goal for digital twins is to integrate critical material attributes (CMA), CQAs, and CPPs into process understanding, as illustrated in Figure 2. As per Prof. Christoph Herwig from the Vienna University of Technology, a digital twin model can have different levels of complexity, ranging from first principles to advanced data-driven models (9). In general, mechanistic models will not be sufficiently calibrated to the system at hand to be used for control decisions, and thus the need for data-driven models exists. Taking the example of mammalian cell culture, developing a standard mechanistic model for mammalian cell culture is impossible because metabolic networks and reaction routes are not yet fully known. Thus, hybrid models are created by combining modeling approaches (11–13).

Figure 2. Approach used to connect process analytical technology (PAT) and digital twins. Figure courtesy of the authors.

Combining hybrid models with noise-smoothening filters of real-time process data can provide real-time monitoring, control, and automated decision-making. In a work presented by Harini Narayanan, a PhD student at ETH Zurich (10), a hybrid model was used as a digital twin for a mammalian cell culture process, showing predictive monitoring accuracy improvements of 35% compared to industrial benchmark tools based on standard partial least squares regression models (14). Overall, fundamental mathematical models with hybrid parameters obtained from experiments, combined with rapid sensors and analytics, can result in effective digital twins that can decrease development time and costs and facilitate real-time control (15). The ultimate aim is to develop PAT tools and models for all unit operations so that one set of process control decisions propagates further corrective and maintenance actions in the subsequent process, resulting in a unified process control strategy in line with the objectives of Industry 4.0 as well as in compliance with regulatory requirements.

Machine learning. Due to the growing interest in continuous biotherapeutics manufacturing, machine learning (ML) techniques for real-time prediction of product quality and process control are becoming increasingly popular. Various different machine-learning algorithms for predicting crucial process parameters can be deployed in different unit operations. These approaches are particularly well-suited to downstream unit operations, including chromatography and filtration, as the process run times are often just a few hours long, which is very short compared to the upstream process, which runs for days or weeks. Due to these short process times, there is a need for information to be provided rapidly so that control decisions can be made within an effective time window. However, typical measurements for CQAs, such as glycosylation, aggregates, and charge variants, are done using HPLC and require method times of approximately one hour, as well as frequent sampling.

Thus, there is an advantage in training machine-learning based models that can correlate real-time process information commonly available from sensors such as pH, UV, conductivity, and pressure, with the CQAs of the biomolecule, bypassing the need for sampling and tedious analytics. In a case study presented by Nikita Saxena, PhD, a post-doctoral fellow at the Indian Institute of Technology Delhi (10), pH, UV, and conductivity sensors data were used to train tree-based regression models, including decision tree and random forest models. The results were promising, with less than 5% prediction errors for all predicted attributes in 40 chromatography cycles in a continuous monoclonal antibody (mAb) production train. The random forest model outperforms other methods for small process-scale datasets because they have minimal possibilities of overfitting, are computationally efficient, and do not require a graphics processing unit. Overall, these machine-learning techniques are powerful enablers of PAT.

Need for novel sensors to facilitate rapid analytics

Real-time monitoring of N-glycosylation. Real-time monitoring is critical to get information within the required time frame to execute control actions. Thus, a key focus area of PAT is to develop tools for monitoring CQAs that impact the efficacy and safety of the drug product. As per Prof. Shishir Chundawat from Rutgers University (10), a key challenge is measuring post-translational modifications of proteins such as N-linked glycosylation, which are important CQAs but difficult to measure in-line due to the minute differences between the different variants. A recent innovation is the sequential-injection-based PAT system “N-GLYcanyzer” to monitor mAb glycosylation in the upstream process (16). The key breakthrough is the development of an integrated mAb sample and derivation system that can perform antibody titer and glycoform analysis in fewer than two hours. The technique consisted of several steps: mAb capture, deglycosylation, fluorescent glycan tagging, and glycan enrichment. These are performed on an integrated HPLC system for direct injection and analysis. To enhance glycan labeling efficiency under aqueous conditions, a variety of fluorescent tags and reductants can be used, such as porous graphitized carbon (PGC), to maximize glycan recovery and enrichment. The N-GLYcanyzer platform was demonstrated for automated, near real-time glycosylation monitoring in a trastuzumab biosimilar manufacturing process.

Spectroscopic sensors for rapid and simultaneous quantification of multiple analytes. Spectroscopic tools are already well-established in the pharmaceutical space, as presented by Yi Tao, PhD student at Rutgers University (9), in a series of case studies showcasing the deployment of spectroscopic tools for monitoring and control of API blending and tableting processes. In the biopharmaceutical space, in-line spectroscopic sensors using infrared, near-infrared, or Raman spectroscopy have many applications (17). Spectroscopy allows information to be gained about multiple analytes simultaneously in a non-degradative manner. Spectroscopic probes in tandem with chemometric algorithms are well-suited to PAT applications in bioprocessing. One example presented by Prof. Christoph Herwig of the Vienna University of Technology (9) showcased effective use of spectroscopy in the upstream bioreactor by combining near and mid-IR spectroscopy in a Penicillium chrysogenum fed-batch process (11). The spectroscopic data were fed into a calibrated chemometric model and successfully predicted nitrogen, product, and precursor concentrations. In the downstream process, spectroscopy has also gained many applications in monitoring and control of different unit operations, including chromatography and filtration. An example presented by Garima Thakur, a PhD student at the Indian Institute of Technology (9), demonstrated a control strategy for single pass tangential flow ultrafiltration using near infrared flow cells in the feed and retentate streams to make flux control decisions (18). The spectroscopic sensors were able to accurately measure the concentration of the target protein in a large range of 0.5–200 g/L on a time scale of seconds, enabling the membrane operating conditions to be adjusted in real time to maintain consistent drug product concentration.

Automation is a key requirement for deploying PAT

Real-time data analytics, modeling, PAT integration, and control techniques are indispensable in smart processes development. To ensure consistent high yields, prevent deviations, and provide an assurance of quality, decision-making needs to be highly automated without the need for human monitoring or intervention. One key challenge is real-time integration with third-party equipment and centralized control consoles for monitoring, visibility, performance tracking, and prediction. As presented by Amos Dor, CTO & Head of Pharma at Applied Materials Automation Product Group of Applied Materials (10), an example of a tool for enabling process automation is Applied SmartFactory RX, an end-to-end solution provided by the company (19). The solution includes smart process development, smart inspection, smart process, smart scheduling, and smart maintenance. The solution has been shown to lead to increased yield by 9%, quality by 75%, productivity by 20%, and equipment uptime by 150% in various pharmaceutical manufacturing processes. Other automation platforms can also be developed on a case-by-case basis by establishing read and write to the manufacturing equipment from a central controller, such as a programmable logic controller (PLC) or distributed control system (DCS) (20). Overall, automation systems are essential for effectively deploying PAT, and should facilitate improved process understanding, reduced costs, increased outputs, and, ultimately, increased profits, as illustrated in Figure 3 (21).

Figure 3. Advantages of automation and control systems in manufacturing. Figure courtsey of the authors.

Conclusion and future perspective

The biopharmaceutical industry is likely to undergo significant digital transformation. With considerable improvements in computing power and data management solutions, novel and effective analytics approaches are being developed. Future manufacturing facilities will include standalone elements that provide greater flexibility and more robust control. The path to full implementation of Industry 4.0 will include improvements and technologies that eliminate many of the risks and challenges associated with automation, monitoring, and control. Innovation in modeling and simulation, sensor systems, data management, data analysis, computation, and engineering approaches are needed to support the deployment of PAT at manufacturing scale. Implementation of artificial intelligence tools, such as machine learning, is also a vast untapped opportunity. Overall, the industry must move from the development of individual solutions toward realizing integrated monitoring, optimization, and control tools on a plant-wide scale. It is essential to gain the support of regulators and establish clear frameworks and guidelines on topics such as automation validation, model validation, model maintenance, and model-based decision-making. The success of implementing advanced data management, process modeling, real-time analytics, and automated controls will be critical in determining the future impact of PAT.

Acknowledgements

Authors acknowledge the American Association of Pharmaceutical Scientists (AAPS) for providing the PAT Community platform. Authors gratefully acknowledge the expert speakers at the webinars conducted by the AAPS PAT Community, whose ideas have been summarized in this article (in order of presentation): Christoph Herwig, PhD, professor, Institute of Chemical Engineering, Vienna University of Technology; Garima Thakur, doctoral researcher, Department of Chemical Engineering, Indian Institute of Technology Delhi; Yi Tao, doctoral researcher, Department of Chemical and Biochemical Engineering, Rutgers University; Amos Dor, CTO and Pharma general manager, Applied Materials AGS Automation Products Group; Shishir Chundawat, PhD, professor, Department of Chemical and Biochemical Engineering, Rutgers University; Nikita Saxena, PhD, post-doctoral researcher, Department of Chemical Engineering, Indian Institute of Technology Delhi; Harini Narayanan, doctoral researcher, Institute for Chemical and Bioengineering, ETH Zurich.

References

1. M.K. Maruthamuthu, et al., Trends Biotechnol. 38, 1169–1186 (2020).

2. H. Helgers, et al.,Processes 9, 172 (2021).

3. E.J. Kim, Pharmaceutics 13, 919 (2021).

4. A. Guerra, M. von Stosch, and J. Glassey, Crit. Rev. Biotechnol. 39, 289–305 (2019).

5. S. Hoffman, K. Soni, and J. Ferraiolo, “UV-Vis Based Determination of Protein Concentration: Validating and Implementing Slope Measurements Using Variable Pathlength Technology,” BioProcess International, Sept. 11, 2014.

6. M. Jiang, et al., Biotechnol. Bioeng. 114, 2445–2456 (2017).

7. A. Tiwari, et al., Anal. Chem. 90, 7824–7829 (2018).

8. G. Thakur and A. Rathore, Pharm Tech 45 (3) pp: 32–40 (2021

9. AAPS, PAT applications in automation and control of pharmaceutical and biopharmaceutical processes, Webinar, Feb. 9, 2022.

10. AAPS, Process Analytical Technology in Biopharmaceuticals: Recent Trends, Webinar, Dec. 6, 2021.

11. A. Golabgir, et al., Biotechnol. Prog. 31, 1703–1715 (2015).

12. P. Sinner, et al., IFAC-Pap. 52, 181–186 (2019).

13. J. Kager, et al., “Extension of a Particle Filter for Bioprocess State Estimation using Invasive and Non-Invasive IR Measurements,” in Computer Aided Chemical Engineering, A.A. Kiss, E. Zondervan, R. Lakerveld, and L. Özkan, eds., pp.1417–1422 (Elsevier, Vol. 46, 2019).

14. H. Narayanan, et al., Biotechnol. Bioeng. 117, 2703–2714 (2020).

15. V. Abt, et al., Curr. Opin. Chem. Eng. 22, 244–252 (2018).

16. A. Gyorgypal, and S.P.S Chundawat, “Integrated Process Analytical Platform for Automated Monitoring of Monoclonal Antibody N-Linked Glycosylation,” In Analytical Chemistry (American Chemical Society, 2022).

17. G. Thakur and A.S. Rathore, Pharm Tech 45 (3) 2021.

18. G. Thakur, S. Thori, and A. Rathore, J. Membr. Sci. 613, 118492 (2020).

19. Applied Materials, “Automation Software: APPLIED SMARTFACTORY RX ANALYTICS & CONTROL,” www.appliedmaterials.com, accessed March 28, 2022.

20. G. Thakur, et al., Biotechnology and Bioengineering 118(5), 1913–1931 (2021).

21. A.S. Rathore, et al., Current Opinion in Chemical Engineering 31, 100671 (2021).

About the authors

Anurag S. Rathore*, asrathore@biotechcmz.com, is professor and coordinator of the DBT Center of Excellence for Biopharmaceutical Technology, and Garima Thakur is a doctoral researcher; both are at the Department of Chemical Engineering, Indian Institute of Technology Delhi, India. Naveen G. Jesubalan is a doctoral researcher at the School of Interdisciplinary Research, Indian Institute of Technology Delhi, India.

* To whom all correspondence should be addressed.

Article details

BioPharm International
Volume 35, Number 5
May 2022
Pages: 31–35

Citation

When referring to this article, please cite it as A. S. Rathore, N. G. Jesubalan, and G. Thakur, “Perspectives on Process Analytical Technology,” BioPharm International 35 (5) (2022).