Access to this full-text is provided by Frontiers.
Content available from Frontiers in Medicine
This content is subject to copyright.
PERSPECTIVE
published: 06 June 2022
doi: 10.3389/fmed.2022.913287
Frontiers in Medicine | www.frontiersin.org 1June 2022 | Volume 9 | Article 913287
Edited by:
Hans-Dieter Volk,
Charité Medical University of
Berlin, Germany
Reviewed by:
Guido Moll,
Charité Universitätsmedizin
Berlin, Germany
*Correspondence:
Simon Hort
simon.hort@ipt.fraunhofer.de
Specialty section:
This article was submitted to
Translational Medicine,
a section of the journal
Frontiers in Medicine
Received: 05 April 2022
Accepted: 11 May 2022
Published: 06 June 2022
Citation:
Hort S, Herbst L, Bäckel N, Erkens F,
Niessing B, Frye M, König N,
Papantoniou I, Hudecek M,
Jacobs JJL and Schmitt RH (2022)
Toward Rapid, Widely Available
Autologous CAR-T Cell Therapy –
Artificial Intelligence and Automation
Enabling the Smart Manufacturing
Hospital. Front. Med. 9:913287.
doi: 10.3389/fmed.2022.913287
Toward Rapid, Widely Available
Autologous CAR-T Cell Therapy –
Artificial Intelligence and Automation
Enabling the Smart Manufacturing
Hospital
Simon Hort 1
*, Laura Herbst 1, Niklas Bäckel 1, Frederik Erkens 1, Bastian Niessing 1,
Maik Frye 1, Niels König 1, Ioannis Papantoniou 2,3, 4, Michael Hudecek 5, John J. L. Jacobs 6
and Robert H. Schmitt 1,7
1Fraunhofer Institute for Production Technology IPT, Aachen, Germany, 2Institute of Chemical Engineering Sciences,
Foundation for Research and Technology-Greece (FORTH), Patras, Greece, 3Skeletal Biology and Engineering Research
Centre, Department of Development and Regeneration, KU Leuven, Leuven, Belgium, 4Prometheus the Leuven R&D
Translational Division of Skeletal Tissue Engineering, KU Leuven, Leuven, Belgium, 5Lehrstuhl für Zelluläre Immuntherapie,
Medizinische Klinik und Poliklinik II, Universitätsklinikum Würzburg, Würzburg, Germany, 6ORTEC BV, Zoetermeer,
Netherlands, 7Laboratory for Machine Tools and Production Engineering (WZL) of RWTH Aachen University, Aachen,
Germany
CAR-T cell therapy is a promising treatment for acute leukemia and lymphoma.
CAR-T cell therapies take a pioneering role in autologous gene therapy with three
EMA-approved products. However, the chance of clinical success remains relatively low
as the applicability of CAR-T cell therapy suffers from long, labor-intensive manufacturing
and a lack of comprehensive insight into the bioprocess. This leads to high manufacturing
costs and limited clinical success, preventing the widespread use of CAR-T cell therapies.
New manufacturing approaches are needed to lower costs to improve manufacturing
capacity and shorten provision times. Semi-automated devices such as the Miltenyi
Prodigy®were developed to reduce hands-on production time. However, these devices
are not equipped with the process analytical technology necessary to fully characterize
and control the process. An automated AI-driven CAR-T cell manufacturing platform in
smart manufacturing hospitals (SMH) is being developed to address these challenges.
Automation will increase the cost-effectiveness and robustness of manufacturing. Using
Artificial Intelligence (AI) to interpret the data collected on the platform will provide
valuable process insights and drive decisions for process optimization. The smart
integration of automated CAR-T cell manufacturing platforms into hospitals enables the
independent manufacture of autologous CAR-T cell products. In this perspective, we
will be discussing current challenges and opportunities of the patient-specific but highly
automated, AI-enabled CAR-T cell manufacturing. A first automation concept will be
shown, including a system architecture based on current Industry 4.0 approaches for
AI integration.
Keywords: ATMP, CAR-T cell, artificial intelligence, automation, autologous, manufacturing, Industry 4.0, smart
manufacturing hospital
Hort et al. Autologous CAR-T Cell Therapy
INTRODUCTION
With the transformation of hospitals toward smart treatment
centers, digitalization is entering the health sector and supporting
hospital employees through intuitive digital data management
and robotics. Successes have already been achieved in diagnostics,
treatments, and surgical intervention. In recent years, Advanced
Therapy Medicinal Products (ATMPs) have gained importance
for curing genetic and cellular diseases. One therapy already
being applied to treat acute leukemia and lymphoma is CAR-T
cell therapy. In contrast to traditional cancer treatments, CAR-T
cell therapy allows for the specific targeting of tumor cells. The
approved therapies Kymriah R
, Yescarta R
, and Tecartus R
(1–3)
target the CD19 antigen in hematological malignancies but differ
in cell composition, manufacturing process, and costimulatory
domain. These therapies use an autologous approach, where the
patient’s cells are engineered instead of allogeneic cell therapies
where cells are extracted from a healthy donor, engineered, and
expanded to treat multiple other patients. Allogeneic CAR-T cells
offer the opportunity for large-scale production. However, they
cause significant graft-vs.-host disease and are rapidly terminated
by the host’s immune system, currently limiting their applicability
(4). Autologous therapies have seen clinical approval but face
manufacturing and large-scale deployment challenges. Figure 1
visualizes these challenges along the six main steps of CAR-T
cell therapy.
FIGURE 1 | Challenges for wide-scale deployment of autologous CAR-T cell therapy.
The entire process from provision of starting material (e.g.,
apheresis, blood donation) to injection is currently dominated
by impractical manual processes. These processes are highly
complex, requiring much personnel and generating high costs
due to their labor-intensiveness, cost of materials, and use of large
cleanroom suites. Additionally, manual manufacturing leads to
frequent interaction of personnel and product, increasing the
risk for contamination and subsequent product loss. A transition
away from these manual and static manufacturing protocols is
needed to shorten production cycles to improve vein-to-vein
timelines. As autologous therapies are keyed to an individual
patient, the current centralized production increases overall
manufacturing times and generates avoidable logistics due to
laborious transportation of apheresis, viral vectors, and CAR-T
cell product. Compared to the established therapies the vein-to-
vein timelines [e.g., 17 days for Yescarta (5)] can be reduced and
consequently the patient’s chances of recovery increased.
Closed, semi-automated systems have been developed to
address these issues, such as the Miltenyi Prodigy R
and the
Lonza Cocoon R
(6,7). These devices follow a “one-device-per-
patient approach” to minimize the risk of cross-contamination.
Unfortunately, this manufacturing approach is unsuited for
large-scale deployment, limiting the reduction of manufacturing
costs and widespread application of CAR-T cell therapy. These
devices are time-consuming to adapt to technological advances
in the field due to their high level of integration and technological
Frontiers in Medicine | www.frontiersin.org 2June 2022 | Volume 9 | Article 913287
Hort et al. Autologous CAR-T Cell Therapy
complexity. Additionally, these devices do not provide the
necessary process insights to assess cell quality and provide
early information on the performance of the cells and potential
therapeutic outcome.
Installing more process analytical technology to generate a
broad data basis combined with data analytics and AI approaches
is needed to overcome a lack of process understanding. Even if
regulatory hurdles often still have to be overcome, AI is already
being used successfully in the hospital context (8). McKinsey
created an overview of the AI solutions currently used in Europe
and classified them according to a patient-centered healthcare
framework (9). Most of the identified use cases are in diagnostics
and clinical decision making, whereas a typical use case in
diagnostics is the automated counting of living and dead cells in
a blood sample. CAR-T cell therapy can highly benefit from the
solutions already in use and the overall potential of AI.
A novel automated manufacturing approach is needed to treat
high numbers of patients with autologous CAR-T cell therapies
at the state-of-the-art. This automated system should allow for
parallelized production of autologous CAR-T cell products to
decrease costs and increase the product’s availability. It needs
to be designed in an integrated but modular manner to allow
for rapid adaption to technological advancements in the field of
ATMP manufacturing but also decreases hands-on interventions
to a minimum.
The automation and AI integration require the second
transformation of hospitals into smart manufacturing hospitals
by enabling them to produce CAR-T cells directly at the point
of care. A smart manufacturing hospital is defined as a
hospital specialized on ATMPs that incorporates an end-to-end
automated manufacturing platform for personalized treatment
in an adjacent GMP facility. The facility connects to existing
logistic and IT infrastructure, offering extensive patient and
manufacturing data availability as well as AI-driven clinical
decision support while taking all regulations (e.g., G(A)MP,
MDR, cybersecurity) into account. For a flexible and modular
integration of the manufacturing platform into the GMP facility
and the hospital, an IT infrastructure based on reliable Industry
4.0 and IIoT (Industrial Internet of Things) is needed to
cope with the rapidly changing environment of automated cell
and gene therapy. Existing approaches (10–12) already show
applicability to the hospital context but do not meet the new
demands of the smart manufacturing hospital. In particular,
the reliable provision of patient and manufacturing data and
comprehensible decision support for the manufacturing process
requires a novel, holistic approach to exploit the full potential of
automated CAR-T cell therapy.
AUTOMATED AI-DRIVEN CAR-T CELL
MANUFACTURING
For decentralized onsite manufacturing of ATMPs, the hospital
infrastructure needs to be adapted. As such, a manufacturing
platform should operate mostly independent of highly trained
personnel producing autologous CAR-T cells autonomously.
The deployment of an automated AI-driven CAR-T cell
manufacturing platform requires extensive knowledge of
the underlying biological process, hardware (e.g., devices,
machines), and software components (e.g., control software,
data management, AI models). Since the three areas are
highly interconnected, a close interdisciplinary exchange of all
stakeholders is essential for success. Our automated AI-driven
CAR-T manufacturing concept, developed within the scope
of EU H2020 project AIDPATH (AI-driven, Decentralized
Production for Advanced Therapies in the Hospital) (13,14),
focuses on these three areas and approaches the technological
challenges and potential solutions. Figure 2 gives an overview of
the manufacturing platform executing the CAR-T cell process
and the software components enabling process control, AI
integration, and data management.
The CAR-T cell process poses several challenges toward
automation. Firstly, as an ATMP production process, it must
be conducted under aseptic conditions and according to
Good Manufacturing Practice (GMP) (15). Maintaining large
cleanroom suites in a hospital for ATMP manufacturing is
impractical, therefore aseptic production in cleanrooms with
lower grade and smaller footprints is essential. Secondly, for
sustained use of the automated manufacturing platform in
CAR-T cell manufacture, it should accommodate variations
in bioprocess set-up and design. Each step in CAR-T cell
manufacturing may differ dependent on the kind of CAR-
T cell therapy to be produced. As a hospital treats many
different patients with different needs, the platform has to be
suited to a multitude of ATMP products and allow for easy
implementation of new ones (16). For instance, while genetic
engineering has typically been achieved by viral transduction,
more and more processes utilize non-viral transfection methods
to transfer the CAR-DNA (17,18). Therefore, a universally
acceptable manufacturing platform should accommodate both
viral and non-viral methods. Studies have also shown the efficacy
of therapies with both CD4+and CD8+CAR-T cells (19).
Currently, this is achieved by culturing both cell types separately
and then combining them for the formulation of the therapy. To
improve treatment efficacy without increasing process burden,
a co-cultivation of CD4+and CD8+T cell populations is
preferable to eliminate the need for two separate cell cultures
running in parallel. Lastly, while some CAR-T cell therapies
are cryopreserved before injection, some are held until product
release and then directly transferred to the patient without prior
freezing (20). Additionally, cryopeservation can thereby have a
high impact on the outcome of the therapy (21). All of these
different modes of operation need to be represented by the
different hardware required for each individual process step, but
also by a highly flexible software architecture allowing for these
adaptions to different CAR-T products.
To automate autologous CAR-T cell therapy manufacturing,
all required hardware, such as machines and devices, need to
be combined in one integrated process pipeline. As depicted in
Figure 2 the manufacturing plant consists of two sections–one
for manufacturing and one for quality control. Both sections are
automated and centrally controlled. The manufacturing section
incorporates devices for cell washing, selection, electroporation,
expansion, harvest, and formulation. Our approach, developed
Frontiers in Medicine | www.frontiersin.org 3June 2022 | Volume 9 | Article 913287
Hort et al. Autologous CAR-T Cell Therapy
FIGURE 2 | Automated, AI-driven CAR-T cell manufacturing concept considering the CAR-T cell process, hardware, and software components.
in AIDPATH, realizes CAR-T cell manufacturing by automating
tubing-kit-based devices and interconnecting the tubing kits
with sterile connectors and tube welding. Cells are automatically
transferred to their next process step by connecting tubing
assemblies. This significantly reduces the need for a large number
of highly trained personnel to manufacture CAR-T cell products
while also greatly reducing the direct interaction of personnel and
product and thus risk of contamination.
A key element of the manufacturing section is the integration
of a sophisticated perfusion bioreactor, which not only enables
the much-needed co-cultivation of CD4+and CD8+CAR-T
cells but is also equipped with various sensors allowing the
deployment of AI-supported control strategies, which will be
developed during the AIDPATH project.
As manufacturing protocols integrate feedback loops based
on the outcome of analytical measurements and product release
is highly dependent on the time required for analytical assays,
the quality control section tightly integrates quality control
processes. Therefore, the platform is designed to pass cell
samples aseptically and automated from manufacturing to
quality control. The quality control section features devices
to conduct analytics for cell quantity, viability, identity, and
characterization of the subpopulations present. A liquid handler,
flow cytometer, and cell counter are integrated using a six-
axis industrial robot. The quality control section is completed
by integrating automation enabling solutions for common
laboratory tasks such as container capping and de-capping of
material restocking.
CAR-T cell manufacturing is a cost-intensive process, not
only because of its labor-intensiveness but also because of the
resources required. Current semi-automated devices rely on
a one-device-per-patient approach, allowing for parallelization
only by increasing device numbers. This leads to a linear increase
in investment costs for parallelized CAR-T cell manufacturing.
To make autologous cell therapy manufacturing economically
more attractive, parallelized production without a linear
increase of costs needs to be implemented. In our concept, this is
achieved by increasing the number of cartridges in the bioreactor
Frontiers in Medicine | www.frontiersin.org 4June 2022 | Volume 9 | Article 913287
Hort et al. Autologous CAR-T Cell Therapy
system. The longest step during manufacturing is expansion.
Thus, parallelization can be achieved by exchanging each patient’s
incubation cartridge. As all manufacturing devices use closed
tubing kits and these are replaced before the respective process
step, the material of several patients may be processed in parallel
without the risk of cross-contamination.
As CAR-T therapy is becoming a more established research
area, the technologies applied for the generation of the therapy
also evolve. Keeping up with technological advancements will
require the integration of new devices. This is enabled by a
modular approach to the manufacturing and quality control
section. Integration of new devices for the manufacturing section
is facilitated by the straightforward reconfiguration of the tubing
assemblies, as these are equipped with standardized interfaces for
sterile connection or welding. Retrofitting devices in the quality
control section is also uncomplicated as the six-axis robotic
handler features sufficient flexibility to provide different sample
materials and carriers. Overall, this allows for independence on
manufacturers, as various devices from different manufacturers
can be integrated. Furthermore, although the platform is built
for CAR-T cell manufacturing, the design is agnostic of cell
type as the implemented technologies are also applied in other
manufacturing processes. This makes the cell type manufactured
on the platform in future developments flexible to a encompass
variety of non-adherent, genetically engineered ATMPs.
For the software concept, two significant challenges arise.
To enable end-to-end automation and eliminate all avoidable
manual steps, a centralized execution, monitoring, and control of
the entire process chain is required while integrating the various
devices (22). Secondly, the AI models must be incorporated
into the system to guarantee continuous data supply for
model building and training and AI decision support for
manufacturing control. Also, the software concept must consider
the boundary conditions such as the modularity of the system,
the volatile environment of ATMP manufacturing, and the
regulatory landscape.
Currently, most devices in biotechnology are still being
developed for manual operation. This makes automation
difficult, as there are no interfaces for controlling the devices with
external software or reading out data. Although standardized
communication protocols such as OPC-UA and SiLA 2 are
becoming increasingly important, they are not yet offered by
most device manufacturers (23). Therefore, a middleware is
needed for central control that generates a driver for each
device. Each driver collects and sends data via a device’s physical
interface (e.g., USB, Ethernet) and then enables service-oriented
communication with the control software via a standardized
communication protocol. Here, the individual capabilities of
the device (e.g., set temperature) are semantically described
as a service. This semantically uniform description enables
the flexible creation of protocols, integrating decisions (e.g.,
if the temperature is higher than X, then Y), and the low-
effort integration of new devices (23,24). A scheduling module
ensures optimal machine utilization while scheduling all process
steps including the parallelized manufacturing for the bioreactor
cartridges. Digital batch records are generated automatically
to avoid laborious manual documentation. In AIDPATH the
software COPE is used and adapted to the requirements of
the CAR-T cell process. The software was developed in several
research projects for stem cell manufacturing (24,25).
The making of AI models enables an intensive insight into
biological processes and informed decision-making. However,
they also require various patient data (e.g., age, gender, previous
illnesses) and the manufacturing process (e.g., process and cell
parameters, device information). These heterogeneous data sets
are available in different qualities and formats and collected at
different frequencies. Therefore, a data management framework
must process all data using a standardized model, such as
the OMOP Common Data Model (26), to ensure a general
understanding and a straightforward analysis. Furthermore, data
of different velocities must be integrated. Continuous data
from sensors and devices in the manufacturing platform must
be collected by a stream data platform (e.g., Apache Kafka)
and made available to the AI models in aggregated form.
Furthermore, a data storage platform is required that processes
and stores batch data from patients and historical data sets. These
data processing procedures and components form the foundation
for the AI framework in which the various models are built,
trained, and then deployed (27).
Another essential part of the automated AI-driven CAR-T cell
manufacturing concept and the smart manufacturing hospital is
the involvement of clinicians and technicians. Therefore, data
can be integrated manually, automatically, and displayed in a
user-specific way. Without expertise in software development,
clinicians can create and customize process protocols using a
drag-and-drop process creator. The decision support system,
as part of the AI framework, transforms the results of the
AI models into decisions and comprehensibly prepares them
for human control and execution. This is brought together
in a unified user interface, enabling centralized patient-specific
process monitoring, data management, and manufacturing
platform control.
From a regulatory point of view, our automated CAR-
T cell manufacturing concept must comply with GMP and
consider the GAMP guidelines (28). Also, the MDR [Medical
Device Regulations (29)] will be taken into account. The
smart manufacturing hospital’s infrastructure will be designed to
provide layers of in-depth cyber security and resilience to the
manufacturing process. In case of a cyber incident, compromised
segments are easily isolated to allow the infrastructure’s
continuous functioning.
ARTIFICIAL INTELLIGENCE IN CAR-T
CELL MANUFACTURING
AI can gain crucial process insights into the cell’s characteristics
and behavior. This offers a great advantage for adaptive control
of the whole process and the creation of personalized process
protocols. Furthermore, AI can support economic platform
operation in the smart manufacturing hospital by optimizing
manufacturing schedules and resource management. Therefore,
AIDPATH will develop different AI applications along the CAR-
T cell manufacturing and therapy process.
Frontiers in Medicine | www.frontiersin.org 5June 2022 | Volume 9 | Article 913287
Hort et al. Autologous CAR-T Cell Therapy
To get deeper process insights on the CAR-T cell process
and understand how patient-specific characteristics influence
it, a digital twin will track the product through the entire
manufacturing process and perform simulations on the cell
behavior. Based on these insights the control software can
adaptively control the bioreactor in the time-consuming cell
expansion process. From the recorded process data of the
bioreactor, such as oxygen or lactose, the cells’ status can be
determined and possible expansion strategies simulated (30–32).
The process data is thereby supplemented by metabolomics data,
due to their promising characteristics for quality control in
personalized therapy (33,34).
The planning of the therapy requires solving a complex
resource allocation problem under substantial uncertainty and
with frequent replanning. The complexity comes from varying
production times and the number of resources needed, such
as medical equipment or intensive care beds. Additionally, the
time frame of therapy has to be adjusted on a patient-by-patient
basis during therapy, depending on the progression. While
conventional optimization algorithms reach here their limits,
reinforcement learning is a promising method with the ability
to cope with these challenges. Also, adaptive scheduling can
integrate the manufacturing process on the platform optimally
into the overall therapy process (35). In therapy planning,
decision support for the physician facilitates central decisions.
This enables a personalized therapy for each patient independent
of predetermined values.
Since all these AI applications are used in a sensitive
environment, one of the crucial aspects is trustworthiness (36).
In the AI application domain, trustworthiness can be made
tangible by asking two guiding questions: How well can one
specify the application’s behavior? What risks are introduced by
the application, and how can they be dealt with? The specificity
can be divided into the main pillars explainability, robustness,
and security. Here, trust is increased by explainable results,
robust predictions, and safe behavior of the application outside
the actual work domain. Risks that continue to exist can be
quantified and dealt with by risk management methods (37,
38).
DISCUSSION
This perspective has discussed an initial concept of how
automated AI-driven CAR-T cell manufacturing can be
implemented directly at the point of care in a smart
manufacturing hospital. It focuses on the engineering perspective
and how hardware and software components must be designed to
manufacture autologous CAR-T cells efficiently. The regulatory
framework is a significant obstacle that needs to be overcome
before a wide-scale deployment is possible. While the facility
design has been GMP-complaint and GAMP guidelines for
the software have been considered, there is a need for precise
regulatory guidance from EMA and the FDA on using AI-
driven manufacturing platforms. This refers to a validation of
a reliable functioning of the AI algorithms and the assurance
of trustworthiness (e.g., appropriate data quality and quantity
for training, possibility for continouos training) (39,40).
Another issue is economic considerations. Reduced manual,
cost-intensive handling steps are set against automation costs.
Comparing a similar system for automated stem cell production
shows the potential for overall cost reduction (41). However, a
health economic assessment for this concept will be the subject
of future development in AIDPATH. In addition to purely
economic considerations, the supply situation for patients
must, of course, also be considered. Due to the parallelization
of the bioreactor, high scalability and high throughput can
be aimed. The resulting shorter production and delivery times
positively affect the number of patient treatments. Nevertheless, a
discussion is needed to what extent centralized and decentralized
CAR-T cell production can coexist in the future. Another point
that is still up for discussion is the operator model. Although
the automated processes and an intuitive user interface allow
operation by non-highly qualified personnel, it is still unclear
to what extent such a system can be operated by hospitals or
external service providers, such as pharmaceutical companies. In
particular, it must be taken into account that the operation and
maintenance of the hardware, AI and IT infrastructure will result
in new tasks for the operator.
All in all, this is a promising concept that needs to be adapted
and further developed to the rapidly changing market of cell and
gene therapies in the coming years (42). Here, the focus must be
on the transferability of the concept because CAR-T cell therapy
is only the beginning of ATMP development and deployment.
DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included
in the article/supplementary material, further inquiries can be
directed to the corresponding author.
AUTHOR CONTRIBUTIONS
All authors listed have made a substantial, direct, and intellectual
contribution to the work and approved it for publication.
FUNDING
The paper was written within the framework of the EU project
AIDPATH (grant agreement number 101016909). All mentioned
colleagues/companies are part of AIDPATH and therefore
received funding from the EU within the scope of AIDPATH.
ACKNOWLEDGMENTS
The authors would like to acknowledge the contributions of
their colleagues from Fraunhofer Institute for Cell Therapy and
Immunology IZI, University College London, Foundation for
Research and Technology (FORTH)-Hellas, SZTAKI, University
Clinics Würzburg, Aglaris Cell SL, Sartorius Cell Genix GmbH,
Fundació Clínic per a la Recerca Biomèdica, IRIS Technology
Solutions, Red Alert Labs, Panaxea b.v., ORTEC b.v. This
information reflects the consortium’s view, but the consortium is
not liable for any use that may be made of any of the information
contained therein.
Frontiers in Medicine | www.frontiersin.org 6June 2022 | Volume 9 | Article 913287
Hort et al. Autologous CAR-T Cell Therapy
REFERENCES
1. Kymriah: tisagenlecleucel. Available online at: https://www.ema.europa.eu/
en/medicines/human/EPAR/kymriah (accessed April 4, 2022).
2. Tecartus: Autologous peripheral blood T cells CD4 and CD8 selected and
CD3 and CD28 activated transduced with retroviral vector expressing anti-
CD19 CD28/CD3-zeta chimeric antigen receptor and cultured. Available
online at: https://www.ema.europa.eu/en/medicines/human/EPAR/tecartus
(accessed April 4, 2022).
3. Yescarta: axicabtagene ciloleucel. Available online at: https://www.ema.
europa.eu/en/medicines/human/EPAR/yescarta (accessed April 4, 2022).
4. Sanber K, Savani B, Jain T. Graft-versus-host disease risk after chimeric
antigen receptor T-cell therapy: the diametric opposition of T cells. Br J
Haematol. (2021) 195:660–8. doi: 10.1111/bjh.17544
5. Neelapu SS, Locke FL, Bartlett NL, Lekakis LJ, Miklos DB, Jacobson CA,
et al. Axicabtagene ciloleucel CAR T-cell therapy in refractory large B-
Cell lymphoma. N Engl J Med. (2017) 377:2531–44. doi: 10.1056/NEJMoa17
07447
6. Mock U, Nickolay L, Philip B, Cheung GW-K, Zhan H, Johnston ICD, et al.
Automated manufacturing of chimeric antigen receptor T cells for adoptive
immunotherapy using CliniMACS prodigy. Cytotherapy. (2016) 18:1002–
11. doi: 10.1016/j.jcyt.2016.05.009
7. Sei JJ, Lin K, Kelifa S, Daita K, O’ Connor J, Hewitt M, et al. The Cocoon (R)
platform: an automated, closed system for manufacturing patient-scale cell
therapies. Mol Ther. (2020) 28:385.
8. Klumpp M, Hintze M, Immonen M, Ródenas-Rigla F, Pilati F, Aparicio-
Martínez F, et al. Artificial intelligence for hospital health care: application
cases and answers to challenges in European hospitals. Healthcare. (2021)
9:961. doi: 10.3390/healthcare9080961
9. McKinsey & Company. Transforming healthcare with AI: the impact on the
workforce and organizations (2020). Available online: https://eithealth.eu/
wp-content/uploads/2020/03/EIT-Health- and-McKinsey_Transforming-
Healthcare-with- AI.pdf.
10. Dhariwal K, Mehta A. Architecture and plan of smart hospital based on
internet of things. Int Res J Eng Technol. (2017) 4:1976–80.
11. Rodrigues L, Gonçalves I, Fé I, Endo PT, Silva FA.Performance and availability
evaluation of an smart hospital architecture. Computing. (2021) 103:2401–
35. doi: 10.1007/s00607-021-00979-x
12. Uslu BÇ, Okay E, Dursun E. Analysis of factors affecting
IoT-based smart hospital design. J Cloud Comput. (2020)
9:67. doi: 10.1186/s13677-020-00215-5
13. AIDPATH: AI-driven, Decentralized Production for Advanced Therapies
in the Hospital (2021). Available online at: http://www.aidpath-project.eu
(accessed April 4, 2022).
14. Artificial Intelligence-driven, Decentralized Production for Advanced
Therapies in the Hospital. Available online at: https://cordis.europa.eu/
project/id/101016909/de (accessed April 4, 2022).
15. European Medicines Agency. New guidelines on good manufacturing
practices for advanced therapies (2017). Available online at: https://www.
ema.europa.eu/en/news/new-guidelines- good-manufacturing- practices-
advanced-therapies (accessed April 4, 2022).
16. Freitag F, Maucher M, Riester Z, Hudecek M. New targets
and technologies for CAR-T cells. Curr Opin Oncol. (2020)
32:510–7. doi: 10.1097/CCO.0000000000000653
17. Harris E, Elmer JJ. Optimization of electroporation and other non-
viral gene delivery strategies for T cells. Biotechnol Prog. (2021)
37:e3066. doi: 10.1002/btpr.3066
18. Poorebrahim M, Sadeghi S, Fakhr E, Abazari MF, Poortahmasebi V,
Kheirollahi A, et al. Production of CAR T-cells by GMP-grade lentiviral
vectors: latest advances and future prospects. Crit Rev Clin Lab Sci. (2019)
56:393–419. doi: 10.1080/10408363.2019.1633512
19. Turtle CJ, Hanafi L-A, Berger C, Gooley TA, Cherian S, Hudecek M,
et al. CD19 CAR-T cells of defined CD4+:CD8+composition in adult
B cell ALL patients. J Clin Invest. (2016) 126:2123–38. doi: 10.1172/JCI
85309
20. Su T, Ying Z, Lu X-A, He T, Song Y, Wang X, et al. The clinical outcomes
of fresh versus cryopreserved CD19-directed chimeric antigen receptor
T cells in non-Hodgkin lymphoma patients. Cryobiology. (2020) 96:106–
13. doi: 10.1016/j.cryobiol.2020.07.009
21. Cottle C, Porter AP, Lipat A, Turner-Lyles C, Nguyen J, Moll G,
et al. Impact of cryopreservation and freeze-thawing on therapeutic
properties of mesenchymal stromal/stem cells and other common cellular
therapeutics. Curr Stem Cell Rep. (2022) 8:1–21. doi: 10.1007/s40778-022-00
212-1
22. Jung S, Grunert D, Schmitt R. Service-oriented communication and control
system architecture for dynamically interconnected assembly systems. In:
Schüppstuhl T, Tracht K, Franke J, editors. Tagungsband des 3. Kongresses
Montage Handhabung Industrieroboter. Berlin, Heidelberg: Springer Berlin
Heidelberg (2018). p. 223–9.
23. Biermann F, Mathews J, Nießing B, König N, Schmitt RH. Automating
laboratory processes by connecting biotech and robotic devices—an overview
of the current challenges, existing solutions and ongoing developments.
Processes. (2021) 9:966. doi: 10.3390/pr9060966
24. Jung S, Ochs J, Kulik M, König N, Schmitt RH. Highly modular
and generic control software for adaptive cell processing on
automated production platforms. Procedia CIRP. (2018) 72:1245–
50. doi: 10.1016/j.procir.2018.03.189
25. Doulgkeroglou M-N, Di Nubila A, Niessing B, König N, Schmitt
RH, Damen J, et al. Automation, monitoring, and standardization
of cell product manufacturing. Front Bioeng Biotechnol. (2020)
8:811. doi: 10.3389/fbioe.2020.00811
26. OMOP Common Data Model. Available online at: https://ohdsi.github.io/
CommonDataModel/ (accessed April 4, 2022).
27. Al-Gumaei K, Muller A, Weskamp JN, Longo CS, Pethig F, Windmann S.
Scalable analytics platform for machine learning in smart production systems.
In: 2019 24th IEEE International Conference on Emerging Technologies and
Factory Automation (ETFA). Zaragoza: IEEE (2019). p. 1155–62.
28. ISPE. GAMP 5 Guide: Compliant GxP Computerized Systems. (2008).
Available online at: https://ispe.org/publications/guidance-documents/gamp-
5 (accessed May 4, 2022).
29. Regulation (EU) 2017/745 (2017). Available online at: https://eur-lex.europa.
eu/legal-content/EN/TXT/?uri=CELEX%3A32017R0745 (accessed May 4,
2022).
30. Mehrian M, Lambrechts T, Marechal M, Luyten FP, Papantoniou I, Geris
L. Predicting in vitro human mesenchymal stromal cell expansion based on
individual donor characteristics using machine learning. Cytotherapy. (2020)
22:82–90. doi: 10.1016/j.jcyt.2019.12.006
31. Duran-Villalobos CA, Goldrick S, Lennox B. Multivariate statistical
process control of an industrial-scale fed-batch simulator. Comput
Chem Eng. (2020) 132:1–13. doi: 10.1016/j.compchemeng.2019.10
6620
32. Goldrick S, Sandner V, Cheeks M, Turner R, Farid SS, McCreath G,
et al. Multivariate data analysis methodology to solve data challenges
related to scale-up model validation and missing data on a micro-
bioreactor system. Biotechnol J. (2020) 15:e1800684. doi: 10.1002/biot.
201800684
33. Vernardis SI, Goudar CT, Klapa MI. Metabolomics and network
biology for sensitive monitoring of how growth environment changes
affect the physiology of industrial-scale perfusion cultures. IFAC
Proceedings Volumes. (2013) 46:227–32. doi: 10.3182/20131216-3-IN-2044.
00056
34. Li S, Todor A, Luo R. Blood transcriptomics and metabolomics
for personalized medicine. Comput Struct Biotechnol J. (2016)
14:1–7. doi: 10.1016/j.csbj.2015.10.005
35. Egri P, Csáji BC, Kis KB, Monostori L, Váncza J, Ochs J, et al. Bio-inspired
control of automated stem cell production. Procedia CIRP. (2020) 88:600–
5. doi: 10.1016/j.procir.2020.05.105
36. ISO/IEC. Information technology-Artificial intelligence-Overview of
trustworthiness in artificial intelligence. (2020). Available online at:
https://www.vde-verlag.de/iec- normen/248797/iso-iec-tr-24028-2020.
html (accessed May 4, 2022).
37. Ethics Guidelines for Trusworthy AI: High-Level Expert Group on Artificial
Intelligence. (2019). Available online at: https://www.aepd.es/sites/default/
files/2019-12/ai- ethics-guidelines.pdf (accessed April 4, 2022).
Frontiers in Medicine | www.frontiersin.org 7June 2022 | Volume 9 | Article 913287
Hort et al. Autologous CAR-T Cell Therapy
38. Rajpurkar P, Chen E, Banerjee O, Topol EJ. AI in health and medicine. Nat
Med. (2022) 28:31–8. doi: 10.1038/s41591-021-01614-0
39. Iglesias-López C, Agustí A, Obach M, Vallano A. Regulatory framework for
advanced therapy medicinal products in Europe and United States. Front
Pharmacol. (2019) 10:921. doi: 10.3389/fphar.2019.00921
40. Proposed Regulatory Framework for Modifications to Artificial
Intelligence/Machine Learning (AI/ML) Based Software as a Medical
Device (SaMD) (2019). Available online at: https://www.fda.gov/files/medical
%20devices/published/US-FDA-Artificial-Intelligence-and- Machine-
Learning-Discussion- Paper.pdf (accessed April 4, 2022).
41. Nießing B, Kiesel R, Herbst L, Schmitt RH. Techno-economic analysis
of automated iPSC production. Processes. (2021) 9:240. doi: 10.3390/pr90
20240
42. Roland Berger GmbH. Pharma’s big opportunity to ride the next wave: How to
enter the cell and gene therapy market. Munich (2021).
Conflict of Interest: JJ was employed by ORTEC BV.
The remaining authors declare that the research was conducted in the absence of
any commercial or financial relationships that could be construed as a potential
conflict of interest.
Companies Fraunhofer Institute for Cell Therapy and Immunology IZI, University
College London, Foundation for Research and Technology (FORTH)-Hellas,
SZTAKI, University Clinics Würzburg, Aglaris Cell SL, Sartorius Cell Genix
GmbH, Fundació Clínic per a la Recerca Biomèdica, IRIS Technology Solutions,
Red Alert Labs, Panaxea b.v., ORTEC b.v. were involved in the elaboration of the
idea of the paper, the reviewing and the decision to submit the paper.
Publisher’s Note: All claims expressed in this article are solely those of the authors
and do not necessarily represent those of their affiliated organizations, or those of
the publisher, the editors and the reviewers. Any product that may be evaluated in
this article, or claim that may be made by its manufacturer, is not guaranteed or
endorsed by the publisher.
Copyright © 2022 Hort, Herbst, Bäckel, Erkens, Niessing, Frye, König, Papantoniou,
Hudecek, Jacobs and Schmitt. This is an open-access article distributed under the
terms of the Creative Commons Attribution License (CC BY). The use, distribution
or reproduction in other forums is permitted, provided the original author(s) and
the copyright owner(s) are credited and that the original publication in this journal
is cited, in accordance with accepted academic practice. No use, distribution or
reproduction is permitted which does not comply with these terms.
Frontiers in Medicine | www.frontiersin.org 8June 2022 | Volume 9 | Article 913287
Available via license: CC BY
Content may be subject to copyright.