Advances in manufacturing cardiomyocytes from human pluripotent stem cells

Abstract

The emergence of human pluripotent stem cell (hPSC) technology over the past two decades has provided a source of normal and diseased human cells for a wide variety of in vitro and in vivo applications. Notably, hPSC-derived cardiomyocytes (hPSC-CMs) are widely used to model human heart development and disease and are in clinical trials for treating heart disease. Success of hPSC-CMs in these applications requires robust, scalable approaches to manufacture large numbers of safe and potent cells. While significant advances have been made over the past decade in improving the purity and yield of hPSC-CMs and scaling the differentiation process from 2D to 3D, efforts to induce maturation phenotypes during manufacturing have been slow. Process monitoring and closed-loop manufacturing strategies are just being developed. In this review we discuss recent advances in hPSC-CM manufacturing including differentiation process development and scaling, and downstream processes including separation and stabilization.

1. Emerging Applications of Stem Cell-Derived Cardiomyocytes

Cardiovascular diseases (CVD) are the leading cause of death globally (1). Recovery from CVD has been hampered by the low regenerative capacity of the adult heart. Pluripotent stem cells (PSCs) can self-renew indefinitely and differentiate to any cell-type and hence PSCs offer a lucrative opportunity to generate cells for diverse applications including cell-based regenerative therapies to treat CVD, developmental biology, disease modelling, and drug discovery (Figure 1). PSC-derived cells are particularly powerful for applications in which primary human cells are difficult to obtain or culture, including cardiomyocytes (CMs).

Figure 1:

Applications of hPSC-CMs include cellular therapy, acellular therapy, engineered tissues patches, drug development and discovery, and disease modeling.

While animal CMs have been invaluable in understanding development and disease, they often differ from human CMs in characteristics such as ion channels, gene expression, metabolism, and electrophysiological properties (2). Human PSC-derived CMs (hPSC-CMs) have been shown to accurately mimic disease phenotypes in both inherited and environmental disorders, including increased arrhythmic calcium traces in CMs differentiated from Duchenne muscular dystrophy patient-derived iPSCs (3), calcium dysregulation, sarcomere disorganization, cellular hypertrophy and arrhythmia in hypertrophic cardiomyopathy (HCM) patient derived iPSC-CMs (4), and increased arrhythmogenicity in congenital long QT syndrome (LQTS) patient derived iPSC-CMs (5). Another approach to modeling cardiac diseases using hPSC-CMs is to introduce mutations that are linked to disease via genome editing. By creating variants of the HCM-causing mutation C9123T-MYH7 in hPSC lines using CRISPR/Cas9, Mosqueira et al. were able to recapitulate hallmarks of HCM including hypertrophy, multi-nucleation, hypertrophic marker expression, and sarcomeric disarray (6). Diseases caused primarily by environmental factors, such as diabetes, are comparatively complex to model using hPSCs. Under prolonged exposure to a high-glucose environment, hPSC-CMs exhibited reduced contractility and increased amplitude of calcium transients, mimicking cardiac pathology of diabetes patients (7).

The probability of success of drugs targeted to cardiovascular diseases to pass through clinical trials is approximately 25% (8). Hence, better preclinical models are needed to reduce costs associated with failed clinical trials. hPSC-CMs have been proven to accurately predict cardiotoxicity of numerous compounds and have paved the way to “clinical trial in a dish” applications to discover treatments for specific disease conditions. For example, Liang et al. generated CMs from iPSCs from patients with various hereditary disorders and were able to predict the susceptibilities to cardiotoxic drugs (9). McKeithan et al. developed an optical method to monitor action potential (AP) kinetics and applied a high-throughput screen in a 384 well-plate format to predict drug-induced proarrhythmias and identified drugs to reverse congenital and pharmacological arrhythmia (10). Kirby et al. screened 48,640 molecules to identify molecules with cardioprotective activity against oxidative stress using luminescence-based assay and impedance-based readings of monolayer integrity and contractility (11). Recently, Burnett et al. used hiPSC-CMs from five donors to test the cardiotoxicity of 1029 chemicals. The characterization of cardiotoxicity in this study recapitulated known drug effects and suggested possible risks of some environmental chemicals (12).

Preclinical animal models have established hPSC-CMs as a potential therapeutic to treat CVD. For example, human embryonic stem cell-derived CMs (hESC-CMs) transplanted into a mouse myocardial infarction (MI) model showed survival, organization, and maturation of the transplanted cells through time (13). Injection of hESC-CMs into the guinea pig heart demonstrated host-graft electromechanical coupling (14). hiPCS-derived cardiac patches on MI porcine hearts showed electromechanical coupling and an increase in genes associated with calcium-handling and contractility, suggesting maturation of CMs (15). The patches had cardioprotective and reparative effects on the hearts, such as a reduction in infarct size, wall stress, and apoptosis without causing arrhythmic complications. Transplantation of major histocompatibility complex (MHC)-matched allogeneic iPSC-CMs in a cynomolgus monkey MI model showed host-graft coupling and improved contractile function (16). Intramuscular transplantation of hESC-CMs in infarcted macaque monkeys showed electromechanical coupling, reduced ejection fraction, and improved contractile function, but a subset presented with graft-associated ventricular arrhythmias (17). Based on these and other promising preclinical studies, there are currently two ongoing clinical trials for hPSC-CMs: a phase I allogeneic iPSC-CM patch for ischemic heart disease patients (18) and a phase I/II allogeneic injection of iPSC-CMs to treat heart failure patients (19).

Approximately 10⁹ successfully engrafted CMs are thought to be needed for a hPSC-CM cell-based therapy to treat MI in non-human primates (20); manufacturing CMs at the scale required for clinical applications poses its challenges. Current protocols for hPSC-CM differentiation yield cells with immature phenotypes. This can be detrimental in clinical settings, as transplantation of immature cells may cause arrhythmias as observed in animal models (17). Immature hPSC-CMs also hamper the ability to accurately model specific diseases and may cause inaccurate effects in drug screening or toxicity studies. Importantly, current protocols fail to consistently produce homogeneous batches, and issues of batch-to-batch variability, low CM yield, and purity are multiplied when considering large-scale production of cells. The progress, considerations, and limitations associated with clinical and industrial-scale manufacturing of high quality hPSC-CMs will be discussed in this review.

2. Differentiation of Human Pluripotent Stem Cells to Cardiomyocytes

CMs provide mechanical functionality to the heart, but account for only about 25–35% of all heart cells (21). Within CMs, there are distinct subtypes, including ventricular, atrial, and nodal CMs, which exhibit unique electromechanical functionality. Because different regions of the heart and CM subtypes are impacted by different diseases, there is significant interest in generating specific CM subtypes.

Early efforts to generate CMs involved forming embryoid bodies (EBs) from hESCs (22). Since then, development of directed differentiation protocols that guide hPSCs through developmental stages via modulation of developmental signaling pathways using growth factors, small molecules, and extracellular matrix (ECM) has led to significantly more efficient CM differentiation. For example, Laflamme et al. used BMP4 and activin A to generate greater than 30% cardiomyocyte purity (23). Kattman et al. treated differentiating hPSCs with Activin A, BMP4, bFGF, VEGF and DKK1 in specific differentiation stages to robustly generate greater than 50% CMs (24). Small molecules are increasingly preferred due to their high activity and stability and low cost, although off-target effects must be considered (25). A small-molecule based CM directed differentiation protocol using Wnt pathway activation to drive mesoderm differentiation followed by Wnt inhibition generated greater than 80% pure CMs (26). A summary of the developmental stages and key modulation pathways in the differentiation of hPSC-CMs are shown in Figure 2.

Figure 2:

Schematic of the key developmental stages, signaling pathways, and markers involved in the differentiation of hPSC-CMs (22, 25, 26, 30, 31, 34, 42, 67).

Initial CM differentiation protocols generated mainly immature, ventricular-like CMs but also contained small populations of atrial-like and nodal-like cells (27). Recent studies have made advances in generating specific CM subtypes from hPSCs. For example, modulating the retinoic acid (RA) pathway can specify atrial versus ventricular subtypes (28, 29). Activation of RA signaling in the presence of Noggin generated cells expressing atrial markers and exhibiting atrial-like electrophysiology, while RA inhibition enriched the population for ventricular-like cells (28). Addition of RA at the Wnt inhibition step of the small-molecule based CM differentiation protocol achieved a CM population that was greater than 90% atrial-like cells based on MLC2A and MLC2V expression (29).

Stage-specific activation of the BMP, Wnt, and RA pathways (22, 30–32) generates pacemaker CMs. Culture of hPSCs as embryoid bodies with BMP, TGFβ inhibitor, and RA on day 3 and subsequent isolation of SIRPA+CD90− cells generated greater than 80% NKX2.5− pacemaker-like cells (30). Another protocol modulated Wnt signaling on day 5 (NKX2.5+ progenitor stage) which promoted expression of pacemaker-related genes (SHOX2, HCN4, ISL1, TBX3, and TBX18) as well as pacemaker-like action potentials (APs) (31).

As hPSC-CMs advance to the clinic, GMP-compliant, defined xenogeneic-free manufacturing processes are desirable for both safety and process robustness. The differences in sources in animal-derived components can also lead to batch-to-batch variability and unreliable results. Advances in developing xeno-free, chemically defined culture conditions have led to successful alternatives for hPSC and hPSC-CM protocols. For culture substrates, human recombinant ECM components like laminin (33), vitronectin (34, 35), and fibronectin (36) and synthetic substrates (37–41) have successfully replaced Matrigel. Various basal media formulations have been developed for hPSC-CM differentiation; Burridge et al. used RPMI 1640, L-ascorbic acid 2-phosphate, and recombinant human albumin and showed 80–95% differentiation efficiency (42). Lian et al. used RPMI supplemented with transferrin, sodium selenite, progesterone, putrescine, and BSA and showed 90% cTnT+ cells (43). Lin et al. showed 80–95% purity using E8 basal medium, lipid concentrate and heparin (44).

3. Maturation of hPSC-CMs

3.1. Defining hPSC-CM Maturation

A major limitation of current CM differentiation protocols is that they generate structurally and functionally immature cells compared to CMs in the adult heart. While maturation is often discussed as a unidimensional entity, there is emerging evidence of the independent regulation of different mature characteristics during CM differentiation. Here, we bin the concept of “maturation” into 4 categories: structural, electromechanical, bioenergetic maturation, and the molecular changes that elicit these phenotypic shifts (45–50). Structural phenotype includes the organization of the cells and intracellular components. Electromechanics include the functional phenotypic outputs of APs and contraction cascades. Bioenergetics encompass the dominant energy-converting processes including central energy metabolism and biomass production. Finally, molecular maturation incorporates the molecular abundance shifts which underpin the observed phenotypic outputs, including key gene/protein isoform switches. Table 1 details properties within each maturation category and provides representative phenotypes observed in adult CMs as compared to hPSC-CMs. Figure 3 depicts this representation of cardiomyocyte maturation, providing examples of methods shown to elicit various degrees of maturation in each of these maturation axes.

Table 1. Maturation benchmarks of hPSC-CMs vs in vivo adult CMs.

Compiled markers of maturation and their previously-reported values in stem cell-derived and in vivo CMs. Several phenotypic markers within each of the 4 proposed maturation veins are presented.

Property		Adult CM	hPSC-CM	Refs
Bioenergetic	metabolism	oxidative / fatty acid oxidation	glycolysis/glucose oxidation	(108)
	mitochondrial cristae	densely packed	sparse, absent	(109)
	dominant growth mechanism	hypertrophy	hyperplasia	(110, 111)
	% mitochondrial volume	30%	5–10%	(108)
Electromechanical	conduction velocity	40 – 130 cm/s	2–43 cm/s	(112–115)
	contraction force	20–45 mN/mm2	4.4 mN/mm2	(116)
	resting membrane potential	−76 – −85 mV	−40 – −70 mV	(117, 118)
	twitch force	> 1 uN	< 300 nN	(119, 120)
	calcium influx origin	sarcoplasmic reticulum, t-tubules	trans-sarcolemmal	(121)
	upstroke velocity	180–400 V/s	15–200 V/s	(107, 113, 122–124)
	present cardiac currents	INa, ICa,L, Ito, IKr, Iks, IK1	INa, ICa,L, Ito, IKr, Iks, If	(117, 125, 126)
Structural	morphology	rod-like	round/oblong	(121, 127, 128)
	t-tubule	extensive network	absent	(121)
	morphology	elongated; larger	round/rod	(128, 129)
	nucleation	25–30% bi / multi	mono, sporadic bi	(127, 128, 130)
	sarcomere	2.2 μm; 40% of cell volume	1.6–1.8 μm; 10% of cell volume	(107, 131–133)
	gap junctions (Connexin 43)	polarized at ends	circumferential	(110, 132, 134, 135)
	size	4500 μm2	500–2000 μm2	(107)
Molecular	troponin I dominant isoform	cTnI (TNNI3)	ssTnI (TNNI1)	(136)
	titin dominant isoform	TTN-N2B	TTN-N2BA	(137, 138)
	myosin light chain dominant isoform	MLC2V (MYL2)	MLC2A (MYL7)	(139)
	α-actin	cardiac (ACTC1)	skeletal (ACTA1)	(140)
	myosin heavy chain dominant isoform	MHC-β (MYH7)	MHC-α (MYH6)	(107, 141)

Figure 3: Three-dimensional representation of hPSC-CM maturation.

Three principal axes of maturation: electromechanical, bioenergetic, and structural and the molecular underpinnings which give rise to these provide better context for the concept of hPSC-CM maturation. Here, immature CMs (early fetal or hPSC-CMs) represent the least mature state, with fully adult CMs representing the theoretical maximum in all 3 axes. Qualitative assignment of 3 examples of hPSC-CM maturation strategies reported in the literature is represented by the 3 points: intensity trained electro-stimulated hPSC-CMs (red)(105); GFAM-cultured hPSC-CMs (blue)(106); and ~100 day cultured hPSC-CMs (gray)(107). GFAM = glucose-depleted media supplemented with fatty acids and galactose (106).

3.2. Enhancing hPSC-CM Maturation

Efforts to induce maturation during hPSC-CM differentiation have typically involved application of microenvironmental cues found in the developing heart to differentiating CMs. Strategies to induce maturation have been reviewed extensively elsewhere and are summarized in Table 2 and Figure 4, marking the feasibility of incorporating these methods into a scalable CM manufacturing pipeline.

Table 2. Meta-review of recent review papers detailing various methods to mature hPSC-CMs.

Identification of various hPSC-CM maturation methods covered in eleven recent reviews. Column 2 provides an assessment of the unique aspects/benefits of each review paper. The bottom row provides a qualitative assessment of the feasibility of implementation of each of the existing maturation methods in a scaled manufacturing context based on existing processes and instrumentation.

Ref	Brief assessment of scope/benefits/unique focus of review	Extended Culture	Exogenous hormone/ligand/synthetic addition	Energy substrate modulation	ECM/substrate modulation	Electromechanical stimulation	Non-CM coculture	3D/tissue culture	In vivo implantation
(45)	More detailed discussion of maturation of hPSC-CMs in 3D systems	X	X	X	X	X	X	X	X
(142)	particularly in-depth analysis of calcium signaling and its relation to maturation	X	X	X
(47)	good discussion of in vivo CM developmental maturation and of post-engraftment maturation of hPSC-CMs in vivo	X	X	X	X		X	X	X
(48)	good appreciation and overview of CM subtype derivation/specification		X	X	X	X	X	X	X
(50)	Review of single-cell maturation studies/analyses; decent discussion of pathways required for, pathways that promote, and pathways that block hPSC-CM maturation	X					X	X
(143)	More focused on electromechanical stimulation and surface modification effects on maturation		X		X	X		X
(144)	particularly robust discussion of in vivo CM developmental biology/maturation/regulation	X	X			X		X
(46)	intricate/focused review on metabolic underpinnings of/methods to enhance maturation			X*
(145)	additionally, offers discussion of transcriptional and posttranscriptional regulatory pathways implicated in CM maturation in addition to model systems		X	X	X	X	X	X
(49)	additionally provides discussion of/reference to systems-level analyses of hPSC-CM maturation (transcriptome, epigenetic, non-coding RNA, metabolomic and proteomic)	X	X	X	X		X	X	X
(146)	provides a broader eye towards scalability; more detailed discussion of FB and EC-specific co-culture effects, and co-differentiation of cell types (particularly from CPCs)		X	X	X		X	X
Feasibility of implementation in scaled manufacturing context:		−	+++	+++	++	+	+	+++	−

^*This review also details several genetic alteration methods used to manipulate metabolic pathway utilization and their effects on CM maturation.

FB = fibroblast; EC = endothelial cell; CPC = cardiac progenitor cell; ECM = extracellular matrix.

Figure 4: Existing methods for maturing hPSC-CMs.

Summary of eight common methods utilized to mature hPSC-CMs corresponding to the categories listed in Table 2. The width and color of the outer ring represent a qualitative assessment of the feasibility of implementation of each of the existing maturation methods in a scaled manufacturing context based on existing processes and instrumentation (greener and thicker = more feasible). ECM = extra-cellular matrix; T3 = triiodothyronine hormone; HIF = hypoxia inducible factor; PPAR = Peroxisome proliferator-activated receptor.

4. CM Manufacturing

hPSC-CM manufacturing uses many of the same processes utilized for manufacturing other therapeutic cells. As illustrated in Figure 4, the steps for hPSC-CM manufacturing include 1) hPSC selection and expansion, 2) CM differentiation and maturation, 3) purification, 4) harvesting, 5) formulation and final fill, and 6) delivery and scheduling. Importantly for hPSC-CM manufacturing, differentiation adds complexity as the composition of the cell population must be monitored throughout the process to ensure that the product follows the desired differentiation pathway and to assess the heterogeneity in the intermediates and final product.

4.1. hPSC selection and expansion

Starting cell material can typically come from two sources: the patient (autologous) or a donor (allogeneic). If the starting cell source is isolated from the patient, the manufacturing process is lengthened and additional potential variability is induced. Typically, autologous cell therapies are thought to be more amenable to a decentralized manufacturing approach because autologous cell therapies generate smaller batch sizes that will be delivered to a single patient. Decentralization requires each manufacturing site to have its own inventory and quality control which typically increases the skilled labor required. Alternatively, a master cell bank of hPSCs could be generated from healthy donors for allogeneic batches. Each of these donors would need to provide informed consent and undergo a screening process including but not limited to their medical history and infectious diseases. Centralized manufacturing is generally more conducive to allogeneic cell therapies where single batches are delivered to multiple patients, which also leads to centralization of workforce and support such as quality control analysis. For allogenic processes, large individual unit operations run in a batch or semi-continuous modes are feasible. Duration of the manufacturing process is not as important since allogeneic cell therapies are an off-the-shelf product and a supply distribution system could provide end users with stabilized products.

Allogeneic treatments require strategies to minimize immune rejection. One strategy is to generate an iPSC haplobank which would allow human leukocyte antigen (HLA) matching of patients and donors. It is estimated that 100 lines would be needed to cover approximately 80% of European Americans, 60% of Asians, 50% of Hispanics, and 45% of African Americans (51). Another approach to prevent immune rejection involves using immune stealth iPSCs. Knockout approaches of human leukocyte antigen (HLA) I/II using gene editing strategies have been shown to not be detrimental to hPSC-CM differentiation (52, 53). Additional knock-in of tolerance signals such as CD47, PD-L1, HLA-C, HLA-E, and HLA-G may be necessary to evade NK-mediated cell death (54). The key advantage to an iPSC-based cell therapy is that iPSCs can be expanded over many passages but they will need to be monitored for acquisition of genetic abnormalities and changes in growth rate or differentiation efficiency.

Generation of iPSCs can be performed from virtually any donor cell type, however mononuclear blood cells are often preferred as blood is easy to obtain. Ideally a non-integrating and non-viral reprogramming method, such as Sendai virus or mRNA, would be used for cell therapy applications. Isolation of iPSC colonies currently requires sorting or manual colony picking which are both time and labor intensive. Following this step, iPSCs would need to be validated for expression of pluripotency markers, high viability, capability of single clonal expansion, genetic stability through karyotyping analysis, and sterility. Careful choice of donors may be necessary as epigenetic signature and memory of the donor cells has been shown to affect CM differentiation specification and efficiency (55). Screening of the master cell bank and working cell bank would be required to maintain viability, purity, sterility, and mycoplasma/endotoxin free iPSCs.

iPSC expansion can easily be scaled in cell aggregates or in carrier-based methods such as the process prototyped by Lonza in a closed, automated 3L reactor with a 70-fold expansion in 12 days producing 1×10¹⁰ undifferentiated hPSCs (56). These undifferentiated iPSC manufacturing processes are now becoming commercially available and have been reviewed elsewhere (57, 58).

4.2. CM Differentiation

CM differentiation is typically optimized at the lab scale in a 2D adherent process using xeno-free chemically-defined reagents, as discussed in section 2. However, scale out of 2D differentiation to commercial or clinical requirements is inefficient due to flask-to-flask variability in differentiation efficiency, a high surface area to volume ratio, and requirement for a large GMP clean room facility and high labor costs due to large numbers of vessels (59). Alternatively, differentiation could be scaled up in a 3D bioreactor which would increase cell density compared to 2D differentiation and therefore be more cost effective, require less space, and less labor. Stirred-tank reactors are highly promising as cells are exposed to similar forces at a variety of scales. Several groups have shown that CM differentiation can be successfully performed in aggregates of 150–300 μm with the optimum size being cell line specific (60–65). Typically, stirred bioreactor systems can achieve up to 0.5–3×10⁶ cells/mL in bioreactors ranging from 40 mL to 3L, with similar yields for microcarrier and scaffold-free systems (60–65). To improve CM purity, one group identified that addition of ascorbic acid during the differentiation followed by a lactate selection generated >80% cTnT⁺ cells in both a stirred tank and rocker platform system (65). Alternatively, new reactor designs, such as the single-use vertical wheel, may improve differentiation efficiency by uniform particle suspension and reduced shear stress compared to stirred tanks. One of the biggest challenges in scaling 3D hPSC-CM differentiation is optimization of oxygen levels and waste removal with sufficient mixing while minimizing mechanical cell damage (66). Following differentiation, additional steps may be needed to enhance hPSC-CM maturation.

4.3. CM Purification

While directed differentiation advances have enabled generation of relatively pure populations of hPSC-CMs, residual nonmyocytes are present in even the most efficient differentiations. Single cell RNA sequencing has determined that these nonmyocyte populations contain a variety of cell types, including endothelial cells, epicardial cells, stromal and smooth muscle-like cells, and cells from endoderm and ectoderm germ lineages, and that the composition of these populations changes through the differentiation process (67, 68). In some applications these nonmyocytes do not need to be removed prior to use of the CMs, and in fact these nonmyocytes may even be beneficial by supporting cardiac tissue formation or cardiac function. However, one may wish to purify CMs following differentiation to better define the cell population and improve manufacturing reproducibility.

The most widely-used method to purify hPSC-CMs leverages the ability of CMs to metabolize lactate as a carbon source. Culture of a population of CMs and nonmyocytes in a medium lacking glucose and containing lactate enables purification of cardiomyocytes up to 99% (69). The advantage of metabolic selection is that it is easy to implement and highly scalable. Lactate purification has been successfully integrated into 2D multi-layered culture plate and 3D microcarrier-based CM differentiation platforms (65, 70). One must consider that replacing glucose with lactate changes the metabolic phenotype of the purified CMs. Metabolic selection led to greater utilization of oxidative phosphorylation in the surviving cells and an increase in expression of molecular markers associated with structural and functional maturation, but did not result in more mature contractility in 3D engineered cardiac tissues constructed with lactate-purified CMs (71, 72).

Molecular markers are often used to purify desired cell types from mixed populations using fluorescence activated cell sorting (FACS) or magnetic activated cell sorting (MACS). However, CMs lack unique, specific cell surface markers that enable live cell sorting. SIRPA1 (CD172a) was proposed as a CM cell surface marker based on a screen of CD antibodies and VCAM1 has also been suggested as a CM surface marker (73, 74). While these markers enrich CM populations, not all CMs express SIRPA1 and VCAM1, and SIRPA1 and VCAM1 expression is not restricted to CMs. Additionally, FACS and MACS are more rapid but less scalable than metabolic selection, making them better suited to lineage studies than scalable manufacturing.

While specific CM surface markers remain elusive, numerous specific intracellular markers of CMs exist, including transcription factors and structural proteins. Genetic-based separations combine the specificity of these markers with the scalability of selection-based approaches, such as antibiotic resistance, or sortability using fluorescent protein expression coupled with FACS. For example, MYL2 promoter-driven expression of GFP enabled purification of hPSC-CMs by FACS while MYH6 promoter-driven expression of the puromycin resistance gene enriched CMs via antibiotic selection (75, 76). MicroRNA-based switches have also been designed to separate CMs based on specific miR expression, using fluorescent protein or selectable marker expression (77). Recent advances in genome editing have improved the efficiency and safety of line construction. Nevertheless, genetic strategies are likely more suited to allogeneic rather than autologous applications.

Percoll gradient centrifugation can enrich CMs from nonmyocytes based on differences in cell density (78). CMs can also be enriched by generating cardiospheres from a heterogeneous differentiation mixture and maintaining the cardiospheres in 3D (79). CMs have also been enriched using microfluidic devices to fractionate cells based on adhesion to a substrate or cell size and shape (80, 81). These microfluidic approaches provide rapid, label-free enrichment, but scaling to commercial and clinical demands remains a challenge.

4.4. Harvesting

Cell harvesting will depend on several aspects, including whether the iPSC-CMs are manufactured in 2D or 3D and if the final product will be formulated as single cells, cell sheets, or aggregates. If cells are singularized, harvesting will require a delicate balance between mechanical shear forces and/or enzymatic activity to disrupt the matrix while maintaining cell viability and function. For final formulations that are aggregates or cell sheets, aggregates may need to be purified by size to focus on removing excess debris from the manufacturing process. An option for the harvesting and sorting includes continuous centrifugation such as Ksep(R) or Combisart(R) with single-use manifolds by Sartorius (82). New disruptive technologies such as acoustic based purification by MilliporeSigma, where aggregates are held at low energy states in a standing-wave based on size, could improve efficiency of sorting viable aggregates from debris and single cells.

4.5. Formulation, Final Fill and Finish

Final formulation will depend on how CMs will be stored and may be different for singularized cells and cell aggregates. Delivery of stable products from the manufacturing site to patients may be more difficult for centralized than decentralized manufacturing. To store the final cell product cryopreserved at −80°C or in liquid nitrogen, cryoprotectants need to be chosen carefully to reduce cell toxicity and maintain functionality upon thawing. These cryoprotectants will likely need to be removed prior to patient treatment. Current protocols suggest approximately 80% viability of hPSC-CMs after cryopreservation and full functional recovery after 5 days of culture in CryoStor(R) CS10 and STEMdiff cardiomyocyte freezing medium, two commercially available xeno-free and serum-free cryoprotectant formulations (83). Fill and finish materials are another challenge as large temperature changes typically require bags or vials which may not be conducive to the high cell number required for cellular therapies. As an alternative to freezing, the cell product could quickly be delivered from the manufacturing site to the patient, foregoing challenges associated with freezing and thawing but reducing shelf life of the cellular product. A study demonstrated the feasibility of this strategy by showing that ~250 μm diameter hPSC-CM aggregates can be stored at 4°C for 7 days with minimal cell death, loss of metabolic activity, and disruption to sarcomere structure (84).

4.6. Delivery and Scheduling

Cold chain transportation of cell products and reagents will require a well established distribution chain of qualified shippers that can trace packages and certify that the product remains at the required temperature and is intact upon delivery. The delivery site will need a location for storage of the shipment prior to use. Additionally, decisions on which patients should be treated will require complex decision trees based on disease state, urgency, and possible outcome of treatment. Inventory management will also be crucial as there are few suppliers for many key reagents which can be detrimental to steady-state or on-demand production as demonstrated through process and distribution simulations (85). One possible solution is to use 3^rd party logistics suppliers to aid in optimizing scheduling and delivery.

5. Quality, monitoring, and analytics

5.1. Process-specific GMP production considerations:

In obtaining regulatory approval for human cell-based therapies in the US, one must comply with 21 CFR 2171 and PHSA351 (86, 87). Current good manufacturing process (cGMP) standards underlying the approval of cell-based therapy products exist as one pillar of an overall quality management system (QMS). Here, we focus solely on cGMP-specific process/product considerations without delving further into the details of the overall QMS. A more detailed overview of an overall QMS for cell product manufacturing is discussed in (88).

5.2. Starting material/reagents qualification:

5.2.1. hPSCs

Qualified release assays and for-information-only (FIO) characterization assays of the starting hPSC banks should be established prior to hPSC-CM manufacture (89). Qualified release assays are required by government oversight agencies in accordance with their respective regulatory frameworks (90) and FIO assays are potentially useful to inform downstream processing parameters (e.g. a cell growth rate assay to determine the optimal inoculation density for a production run) and later process development and optimization studies or process modifications. Furthermore, cell bank stability characterization will also likely be required to determine storage lifetime and drifts in cell phenotype over time (89).

5.2.2. Cell culture reagents

Critical reagents utilized in the manufacturing process will be required to be obtained in cGMP grade except in specific circumstances wherein exceptions can be granted at the discretion of the regulatory agency. Wherever possible, these reagents should be obtained in xeno-free formulations. Furthermore, obtaining reagents with drug master files available for reference will likely facilitate regulatory approval (91).

Processes utilizing xeno-free, defined composition, cGMP-compliant reagents exist; however, many of these are proprietary formulations and/or sole-source materials (56). It is important to consider process robustness with an eye toward limiting or eliminating process reliance on such materials as they may confer undesired and uncontrollable constraints on the process, including supply chain issues from a sole-source supplier. Thus, it is important to consider at the onset the development of a process free of the use of proprietary, sole-source, and xeno-derived materials, and the validation of multiple sources of critical reagents with qualified materials testing assays.

5.3. Process, Instrumentation, and Assay Qualification

For cGMP-compliant manufacturing, rigorous qualification/validation of the process and analytical equipment and assays performed is required to comply with regulatory guidelines. The following FDA documents provide specific information on process and assay validation: (91, 92).

5.4. Product-Specific GMP Production Considerations

hPSC differentiation to CMs can be challenging, with even minor variations potentially derailing a successful outcome. As CM manufacturing processes migrate to 3D and clinical scale, the complexity of the process increases and the cost of failed batches becomes high. As such, detailed understanding and control of the process becomes increasingly crucial to ensure repeatable and robust product output. This process knowledge is also important to obtaining regulatory approval.

5.4.1. Final product release criteria, characterization assays, and CQAs

One must design the manufacturing process to achieve a target product profile (TPP), a well-defined molecular or phenotypic profile of the final formulated product. The TPP is typically linked to a specific clinical effect of the administered dose and can evolve over the product development lifecycle. Based on the TPP and final process and product specifications, a product must comply with a set of critical quality attributes (CQAs) which define the final deliverable. A CQA is defined by the FDA as “a physical, chemical, biological, or microbiological property or characteristic that should be within an appropriate limit, range, or distribution to ensure the desired product quality” (93). In relation to cell therapies, CQAs are defined as potency, sterility, purity, and identity (93). However, many prefer to identify “safety” as a CQA, with sterility being one component thereof. Although the specific assays employed to characterize a product can vary greatly, the field has been working toward harmonizing CQAs and associated analytical assays, thereby simplifying and facilitating regulatory review processes and analytical tools development (94, 95).

A minimum set of required and FIO CQAs to be assayed prior to lot release are detailed in Table 3. Purity assessment will be required to ensure an acceptable range of the target cell type (CMs) in the final formulation, with defined acceptable ranges of nonmyocytes. These analyses, assessing the heterogeneity of the cells within the product, will require single-cell resolution, likely to be performed via flow cytometry. Additional purity release assays will be required to assess acellular contaminants, with specific analytical techniques determined based on the contaminant in question. Identity assessment focuses on ensuring the product in question is that intended for use; at the least proof that the final product was derived from the appropriate donor source will be required, likely via STR profiling. Further identity assays, although not necessarily required for regulatory approval, will undoubtedly be performed to provide better understanding of the product and correlate clinical outcomes with these attributes. Potency metrics in the case of cardiomyocytes are potentially quite varied. At minimum, assessment of pre- and/or post-thaw viability will be required to ensure understanding of the live cell dose administered. Numerous additional characterization assays related to analyzing the expected method of action of the final cell product in affecting a clinical outcome. For example, the beating phenotype and dynamics thereof in the case of ventricular cardiomyocytes for infarct repair may be employed. At least one definitive assay will be required to be set as a mandatory release criterion. Safety considerations for hPSC-derived cell types are particularly focused on mitigating tumorigenicity of the cell product. Thus, proof of a normal karyotype and sufficiently low numbers of residual hPSCs will be required. Finally, standard sterility assays for any administered drug product or cell-based therapy will be mandated to include adventitious agent, bacterial/fungal, endotoxin, and mycoplasma testing.

Table 3. Putative release assays for cGMP manufacture of hPSC-CMs.

Required and likely characterization assays to be performed prior to lot release of hPSC-CMs. Although most assays will be performed on bulk samples, some will require single-cell resolution to assess cellular heterogeneity of the final product. Example analytical methods and specifications based on prior experience and compiled from references (94, 99, 147–150).

	Attribute	Sample resolution	Assay	Required/FIO	Example/proposed specification
Purity	%CMs	single-cell	FC	Required	>70% cTnT+
	% of specific non-CM populations	single-cell	FC	Required	absence or low abundance of user-defined panel of validated off-target cell markers
	Presence of acellular contaminants	bulk	determined by contaminant identity (DLS, ELISA, etc)	Required	absence or low abundance of user-defined contaminant species
Identity	STR profile	bulk	STR profiling	Required	match source hPSC STR profile
	CM subtype specification(s)	single-cell	FC, ICC	FIO
	beating phenotype	bulk	automated video analysis	FIO
	other specific molecular profiles	bulk / single-cell	RT-qPCR, FC, ICC, etc.	FIO
Potency	viability (pre-/post-thaw)	bulk	imaging, FC	Required	> 70% pre- and/or post-thaw viability
	electrophysiology	bulk / single-cell	MEA, patch-clamp	FIO
	calcium handling	bulk / single-cell	imaging	FIO
	contraction force	bulk	various (ex: imaging, MEA)	FIO
	chemotropic response	bulk	various (ex: imaging, MEA)	FIO
	chronotropic response	bulk	various (ex: imaging, MEA)	FIO
	metabolic activity	bulk	various (ex: Seahorse)	FIO
Safety	residual PSCs	bulk	teratoma formation (or PCR-based)	Required	absence or low abundance of residual PSCs (below teratoma formation threshold)
	karyotype	bulk	G-banded or other	Required	normal karyotype
Sterility	adventitious agent	bulk	in vitro and in vivo tests	Required	negative
	bacterial/fungal	bulk	colony growth	Required	negative
	endotoxin	bulk	LAL test	Required	< 0.5 EU/mL
	mycoplasma	bulk	PCR	Required	negative

FIO = for information only (characterization) assay; CM = cardiomyocyte; FC = flow cytometry; ICC = immunocytochemistry; DLS = dynamic light scattering; MEA = multi-electrode array.

Further information on the development and qualification of release assays can be obtained in the following references: (92, 93, 96).

7.4.2. In process characterization

In addition to defining the CQAs of the final formulated product, intermediate or in-process quality attributes should be established to guide successful acquisition of the TPP. This is particularly important in the case of lengthy and highly dynamic processes, such as the differentiation of hPSCs to CMs. In identifying in-process quality attributes, understanding of key intermediate differentiating cell populations is essential. Thus, assessing these intermediate states in real time can provide crucial information on process progression, and inform process development and control. Early characterization strategies based on the fundamental understanding of the underlying developmental biology of the differentiation process often rely on targeted gene and protein expression analyses, and as such often require sacrificial sampling of the process (60, 97). Additional characterization methodologies such as (single cell) RNA sequencing, proteomics, and metabolomics, among others, may facilitate more definitive cell characterization (98, 99) in addition to methods amenable to live cell analysis and potential sorting applications due to reliance on identified cell surface markers of hPSC-CMs and key intermediates (100). Non-invasive and label-free methods for real-time, in-process characterization will undoubtedly accelerate and improve process development timelines and robustness, such as the monitoring of key metabolites in cell culture media through time (56).

5.5. Process optimization

Knowledge and characterization of in-process attributes can dramatically improve the speed, robustness, and real-time decision making of independent process steps/unit operations. For example, knowing the desired intermediate profile can provide a shorter turnaround time for the optimization of the initial induction of hPSCs to mesodermal intermediates than running the complete process through to a final hPSC-CM state.

In attempting to optimize a process, one must identify the critical process parameters (CPPs). The FDA defines these as “process parameter[s] whose variability has an impact on a critical quality attribute and therefore should be monitored or controlled to ensure the process produces the desired quality” (86). However, one might not know which process parameters are indeed critical a priori, or the effect each might confer on the resultant CM product and/or cell intermediates. Thus, rational or exploratory analyses of the available parameter space for each independent operation can be performed to identify highly effective and robust process parameters and establish empirical bounds thereof (60). Often, groups employ sequential optimization of these and other parameters from initial 2D hPSC-CM exploratory platforms through smaller-scale 3D and into larger scale 3D systems (60). Overall, numerous studies have investigated a wide range of parameters in optimizing large-scale hPSC-CM production protocols, with highly varied resulting purities (97).

Many groups realize the importance of inductive molecule concentrations on hPSC-CM differentiation efficiency, and have presented efforts to optimize scaled production CM purity based on modulation of these critical reagents (101). Others have investigated the effects of media composition, culture platform, oxygenation level, and culture modality (batch, semi-continuous, perfusion) on resultant hPSC-CM yield, purity, and phenotype (60). Further, in 3D aggregate cultures, the additional requirement of agitation to mitigate the presence of nutrient and waste gradients and producing more homogenous conditions throughout the bioreactor system while limiting the potential negative effects of shear on cell signaling and aggregate dissolution has led some groups to establish computational models of such systems as a means to enhance process understanding and further optimize bioreactor processing parameters to achieve high-purity hPSC-CM batches (101). For example, in silico models of hPSC expansion in a stirred tank bioreactor were coupled to wet lab outputs of proliferation rates, aggregate size, pH, glucose, and lactose to generate an improvement in yield by over ten-fold and reduction in medium consumption by 75% (102).

5.6. Advanced process control and modeling

A more detailed process understanding of CM manufacturing will likely guide the establishment of process control schemes employing control decisions which can mitigate the effects of materials variability, enhance process robustness, and overall reduce manufacturing losses and consequently manufacturing costs. Current strategies under development use advanced computational techniques to model cell manufacturing and identify critical process parameters. For example, Picken et al. estimated probabilities associated with different process parameters (aspiration volume, total nucleated count and colony forming units) and used a Monte Carlo model to predict the yield of human bone marrow derived mesenchymal stromal cells (hBM-MSCs). This modelling framework was used to identify parameters which would allow a desired yield specification within a specified duration (103). A combination of macroscale and microscale modeling has been used to connect an autologous cell therapy factor to a macroscale supply distribution and determine the number of reactors required to recover from a supply disruption (85). Complex modeling and control methods are not well established in the therapeutic cell manufacturing setting, wherein many still operate under the “product is the process” adage, largely upheld to-date based on incomplete knowledge and/or characterization of starting materials properties and process parameters and their effects on product quality. However, collecting big data control and quality metrics and using that data to inform modeling approaches will be crucial for manufacturers to maintain consistent high-quality products and demonstrate that changes to the integrated process does not affect safety, identity, purity, or potency of the cellular process as the product moves through early clinical trials to scaled commercial production (66). Applying modeling strategies such as these to CM manufacturing will be crucial for development of robust manufacturing processes going forward.

6. Conclusions and Future Directions

hPSC-CM manufacturing and applications are a classic “chicken and egg problem.” A manufacturing process cannot be precisely defined without knowing the application but the application cannot succeed without an effective manufacturing process. Thus, current manufacturing strategies have co-evolved with in vitro and therapeutic use of hPSC-CMs. The field has made significant progress in developing efficient processes to differentiate hPSCs in 2D, transition those processes to scalable 3D platforms, and understand the molecular and function characteristics of the cells.

Clear and quantitative CQAs are needed to guide hPSC-CM manufacturing. We lack sufficient understanding of how cell attributes influence performance. CQAs will be application-specific. For example, mechanistic studies are needed to identify how maturation state affects the ability of hPSCs to predict drug safety and efficacy, or the ability of hPSC-CMs to engraft and functionally integrate in the myocardium in heart failure patients.

Once intermediate and final product attributes are identified, analytic tools to monitor these attributes will be needed. Ideally these measurements would be incorporated into a closed-loop process model that provides higher robustness than current manufacturing operations. Significant advances in process modeling and understanding are needed.

Current manufacturing processes are labor-intensive but process automation can reduce labor costs and process risk (104). We anticipate the growth of automated manufacturing platforms as the field evolves.

Figure 5: Process diagram of hPSC-CM manufacturing.

Steps include 1) iPSC selection and expansion, 2) CM differentiation and maturation, 3) purification, 4) harvesting, 5) formulation and final fill, and 6) delivery and scheduling.