Advanced physiological maturation of human iPSC-derived cardiomyocytes using an algorithm-directed optimization of defined media components

Abstract

Human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) hold tremendous promise for in vitro modeling to assess native myocardial function and disease mechanisms, as well as testing drug safety and efficacy. However, current hiPSC-CMs are functionally immature, resembling in vivo CMs of fetal or neonatal developmental states. The use of targeted culture media and organoid formats have been identified as potential high-yield contributors to improve CM maturation. This study presents an hiPSC-CM maturation medium formulation, designed using a differential evolutionary approach targeting metabolic functionality for iterative optimization. Relative to existing high-performing reference formulations, our medium significantly matured morphology, Ca²⁺ handling, electrophysiology, and metabolism, which was further validated by multi-omic screening, for cells in either pure or co-cultured microtissue formats. Together, these findings not only provide a reliable workflow for highly functional hiPSC-CMs for downstream use, but also demonstrate the power of high-dimensional optimization processes in evoking advanced biological function in vitro.

Using an evolutionary optimization approach, the authors report a culture medium that matures stem cell-derived heart cells, improving their structure, metabolism and electrical activity to advance highly functional models for in vitro testing.

Introduction

Cardiovascular diseases (CVDs), including heart failure, myocardial infarction, and arrhythmias, often have multifactorial etiologies, are progressive and prone to sequelae, and cannot usually be curatively treated; for this reason, they form the single most common cause of death in developed countries¹. For decades, understanding of CVDs has been advanced using animal-derived models of function and disease, but the need for robust and accurate in vitro models to inform human-specific pathophysiology has become apparent². Similarly, the incidence of cardiotoxicity, which has remained unpredictable by preclinical in vivo screening with murine models, is a leading reason for failure of new drug trials or post-marketing drug recalls^3,4. hiPSC-CMs provide a promising avenue for cardiac pharmaceutical efficacy screening, both for preclinical screening and personalized efficacy testing^5–7. Both disease phenotypes and drug responses are emergent composites of interactions across multiple aspects of physiology^8,9, necessitating higher-fidelity in vitro models of myocardium. Important advances have been made in evoking physiological maturation in both hPSC-CMs monolayers and organoids in translational and preclinical applications^10–12, including transcriptomic or proteomic profiles^13–30, morphometrics^{13,17–19,21–24,26–31}, metabolic profile^{13–15,18–24,28–31}, Ca²⁺ handling^{15–19,21,22,24,27–31}, electrophysiology^{15,16,18,19,21–24,28,31}, and disease modeling^{13,16,18,19,21,22,24,32,33}. However, despite the advancements of many physiological parameters of current hiPSC-CMs to neonatal levels or beyond, they remain phenotypically immature in most aspects of CM-specific physiology; they do not recapitulate the contractile force, morphology, electrophysiology, Ca²⁺ handling characteristics, or metabolic profile associated with adult myocardium³⁴.

Culture medium optimization has been explored as a high-yield avenue for further hiPSC-CMs maturation. Notable advances have been made using one- or two-factor experiments, including metabolic substrates such as fatty acids or galactose^15,16,30, and hormones such as triiodothyronine (T3)^17,35–39, insulin⁴⁰ or insulin-like growth factor^36,41, glucocorticoids^17,36,40, or the B27 supplement comprising a mixture of hormones and small molecules optimized for functional neuronal culture^{15,19,20,29,35,37,42}. However, there has been comparatively little focus to date on supplementation of endogenous cofactors or otherwise functional small molecules. The energetic demands of a fully mature CM to fuel electrophysiological and contractile function (e.g., via oxidative phosphorylation) may preclude secondary metabolic pathways that would produce such molecules from more basal media. Furthermore, pairwise and higher-order interactions have been found to be ubiquitous across biological systems, such as cell fate, comprising differentiation or maturation and their associated homeostatic changes^43,44. As a result, small stepwise optimization experiments are unlikely to converge, even in aggregate, on high-efficacy maturation medium formulations.

In this study, we sought to mature hiPSC-CMs using a suite of soluble factors comprising metabolic substrates, hormones, cofactors, and other small molecules based on the potential for wide-scale interactions between factors. Our general strategy in choosing factors was to mirror the exogenous availability of soluble signals to which the myocardium is exposed during late cardiac development, specifically during the perinatal window and early childhood⁴⁵, developmental phases that coincide with a switch to a predominantly oxidative metabolism⁴⁶. This metabolic specialization prioritizes oxidative catabolic flux and outsources peripheral metabolic processes (e.g., anaerobic catabolism, glycogen-related processes, cofactor synthesis, etc.) to other tissues. This transition to a highly efficient energetic phenotype may drive maturation in other functional metrics by increasing the accessible short-term energetic capacity within the cell⁴⁷. To maximize factor synergies and high-order interactions, a high-dimensional, differential evolution (HD-DE) algorithm was employed to functionally interrogate a large factor space, even in the presence of factor interactions or nonlinear responses⁴⁸. By combining a set of 17 independent soluble factors each at 5 dosage levels, a traditional full factorial screen would require c.a. 763 billion runs to cover the solution space; using the HD-DE methodology, we developed a high-performing custom formulation after 4 iterative generations comprising 169 unique formulations. Candidate formulations were scored based on their oxidative uncoupled:control ratio using cell respirometry as a self-normalizing correlate of overall functional maturation. This approach identified a medium formulation that best enhances hiPSC-CMs' maturity to maximize multiple functional performance metrics compared to existing high-performing commercial and published formulations. Recent high-performing hPSC-CM maturation efforts^13,25,49 have resulted in cell phenotypes that eclipse the specificity of single expressional markers of CM maturity⁵⁰, and the field has accordingly embraced more comprehensive physiological profiling both for the purpose of demonstrating application and for more precisely stratifying maturation efficacy. Accordingly, our formulation demonstrated an unmatched degree of advancement of many specific functional metrics of maturity, including morphology, contractility, electrophysiology, Ca²⁺ handling, metabolism, and gene and protein expression profiles.

Results

Differential evolution and selection of an hiPSC-CM maturation medium

To efficiently optimize hiPSC-CM maturation medium without full mechanistic understanding of the processes involved, we implemented an HD-DE algorithm-driven process (Fig. 1a), using oxidative uncoupled:control ratio (UCR; the ratio of pharmacologically electron transport chain-uncoupled to steady-state oxygen consumption rates) as an objective metric to score successive generations of candidate formulations relative to a commercially available control medium (STEMCELL Cardiomyocyte Maintenance Medium) (Fig. 1b). A total of 17 soluble factors (Supplementary Table 1, Fig. 1c) were chosen for iterative optimization; 6 of these factors as well as M199 basal medium had not previously been tested, nor appreciable analogues, for their use in CM maturation. Factors were surveyed at any of 5 discrete doses (including null; Supplementary Table 1) within a given candidate formulation. In all, 169 unique formulations were tested over 204 runs across 4 iterative generations, each consisting of 3 weeks of treatment with candidate formulations (culture timelines in Supplementary Fig. 1a, b). Furthermore, the first generation of formulations was defined using a low-discrepancy sequence so that the entire solution space was evenly queried at the widest possible coverage. This generation included at least one formulation that did not include each soluble factor, to ensure their relative contributions to the utility of the medium as a whole. By principal component analysis (PCA), the top 10 performing formulations formed 2 main clusters of composition, with 2 outliers (Fig. 1d, e); metabolic substrates and cofactors were correlated most strongly in PC1, while hormones and small molecules primarily associated with signaling were aligned within PC2 (Supplementary Table 2). A final formulation for further characterization was manually defined by including specific dose-factor combinations present in the top-right cluster of PCA but absent from the bottom-left quadrant, resulting in a composite of high-performing candidate formulations and containing 15 of the 17 queried factor additives in addition to the M199 base. This 16-component hiPSC-CM maturation formulation (“C16”) was confirmed to be a viable and high-performing candidate against high-performing control formulations from STEMCELL, iCell, and the homemade RPMI + B27 cocktail, using both morphological and contractile comparisons after 3 weeks of culture. Compared to treatment with both STEMCELL and a leading high-efficacy maturation formulation by Feyen et al. ²¹, hiPSC-CM monolayers treated with C16 medium exhibited increased spontaneous alignment (circular variance p < 0.0001), cell elongation (eccentricity p = 0.0019), and intracellular organization (α-actinin:connexin-43 (Cx43) colocalization coefficient p < 0.0001; Fig. 1f), while confocal microscopy demonstrated that C16 medium enhanced myofibrillar density, α-actinin striation, and Cx43 junctional expression (Fig. 1g). C16-treated cells demonstrated clear sarcomeric striations under phase-contrast in C16-treated monolayers (Supplementary Fig. 1d, e) with increased sarcomere length (p = 0.0003; Fig. 1h), as well as increased hypertrophy (projected area p < 0.0001; perimeter p < 0.0001) and anisotropy (form factor p < 0.0001), but not α-actinin integrated density (p = 0.91) (Supplementary Fig. 1f). Furthermore, C16 treatment induced striated expression of hallmark CM functional proteins DHPR, SERCA (Fig. 1i), and RyR2 (Fig. 1j). Transmission electron microscopy demonstrated increased C16-specific prevalence of t-tubules (Fig. 1k) and mitochondria (Fig. 1l). Optical contractile traces, obtained using particle image velocimetry, demonstrated accelerated full contractile cycles in C16-cultured monolayers derived from two hiPSC-CM lines from healthy donors⁵¹ (PGPC17 and PGPC14), as well as pathologically slowed and biphasic contraction in a MYBPC3-KO CRISPR-generated PGPC17 mutant hiPSC line (Supplementary Fig. 1g).

Fig. 1. Iterative development and performance of hiPSC-CM maturation medium candidates relative to existing high-performing control formulations.

a Visual representation of High Dimensional, Differential Evolution (HD-DE) algorithm workflow; from a quasi-random distribution of vectors (formulations) in an extensive but discrete solution space, low-performing candidates are discontinued while high-performing candidates are kept and iterated using mutation and crossover operations. b Ranked performance of all tested formulation candidates (n = 169 unique vectors over 204 instances) against STEMCELL Technologies Cardiomyocyte Maintenance Medium (dashed line), using respirometric uncoupled-control ratio as an objective metric. c Soluble additive factor doses (y-axes) by their tested levels (x-axes) included in the HD-DE optimization, demonstrating exponential coverage of solution space over 17 dimensions. d Final composition of C16 hiPSC-CM maturation medium by factor (orange), superimposed over factor levels of the top 10 formulation candidates in the optimization phase (gray violin plots), and analogous reference levels, where applicable, from the literature (dark gray) or in the commonly used hiPSC-CM culture medium RPMI + B27 (blue). e Principal component analysis biplot visualization of the top ten global performing formulations during optimization (points; red indicates 1^st place formulation, blue indicates 2^nd place formulation, and purple indicates 3^rd place formulation) in addition to a composite high-performing formulation (“C16”; gold); red vector arrows represent correlation of medium additives (A-Q) corresponding to the order of the additives in Supplementary Table 1. Circled formulation groups indicate high-performing clusters derived from common lineages over generations of HD-DE optimization. f Morphometric analyses of STEMCELL-, Feyen maturation media-, and C16-treated monolayer cultures, including circular variance of whole-cell major axis alignment, whole-cell eccentricity, and α-actinin:connexin 43 (Cx43) colocalization under confocal microscopy (violin plots represent biological replicate distribution for alignment and technical replicate distribution for eccentricity and colocalization; points represent biological replicates (n = 15/treatment). Due to differences in cell projected area, different numbers of cells were analyzed between treatments and samples. g Confocal images of α-actinin- (magenta) and Cx43- (yellow) stained monolayers (DAPI-stained nuclei represented in cyan) cultured in iCell, RPMI + B27, STEMCELL, or C16 media. h Length of sarcomeres stained for α-actinin in STEMCELL-, Feyen- and C16-treated monolayer cultures (biological replicates in points (n = 15/treatment), distribution by violin plots). i Confocal detail (DAPI in cyan) of sarco/endoplasmic reticulum Ca²⁺ ATPase 2a- (SERCA2a; magenta) and dihydropyridine receptor- (DHPR; yellow) stained monolayers cultured in STEMCELL or C16 media. i Confocal detail (DAPI in cyan) of SERCA2a- (magenta) and ryanodine receptor- (RyR; yellow) staining in STEMCELL- or C16-treated monolayer cultures. j Peri-myofibrillar T-tubules (white arrowheads) in transmission electron micrographs of hiPSC-CMs from monolayers cultured in STEMCELL or C16 media. k Peri-myofibrillar mitochondria (white arrowheads) in transmission electron micrographs of hiPSC-CMs from monolayers cultured in STEMCELL or C16 media. All analyses were performed after 3 weeks of medium treatment post-differentiation. Capture and display parameters constant within subpanels of micrographs to allow for direct comparison of signal intensity. All scale bars are 50 µm in confocal images and 500 nm in TEM images. For analyses in panels 1 f and 1 h, normality was tested using a D’Agostino-Pearson test, and conditions compared using a one-way ANOVA with Tukey’s post-hoc test. Micrographs are demonstrative of at least 6 independent experiments each. Dotted lines in morphometric charts indicate median and quartiles; *, **, and **** denote p < 0.05, 0.01, and 0.0001, respectively.

Electrophysiological characterization and I_K1 identification in maturing hiPSC-CMs

The presence of stable resting (diastolic) membrane potentials and the absence of spontaneous beating are hallmarks of mature working myocardium. Importantly, in addition to the maturation metrics already listed, hiPSC-CM monolayers treated with C16 medium were not spontaneously contractile, unlike RPMI + B27-treated cells, suggesting marked electrophysiological differences between treatments. To assess the electrical properties of our cultures, we measured intracellular voltages in randomly-selected cells after 6 weeks of treatment with either C16 or RPMI + B27 (Fig. 2a). These electrical differences were associated with classical differences in the resting diastolic membrane potentials. Specifically, RPMI + B27 cells underwent spontaneous depolarization after reaching minimum diastolic membrane potentials (MDPs) of −56.1 ± 6.8 mV (Fig. 2b) a pattern that is characteristic of immature cardiomyocytes, while C16-treated cells displayed fixed resting diastolic membrane potentials (−81.1 ± 7.7 mV). RPMI + B27-treated cell APs were accompanied simultaneously by contraction and Ca²⁺ transients (Fig. 2c) at a rate of ~ 0.2 Hz. By contrast, C16-treated cells did not display spontaneous action potentials (APs), beating, or Ca²⁺ transients in the absence of electrical stimulation. Previous studies have established that the absence of spontaneous APs in mature tissues is linked to the presence of background inward rectifier K⁺ channels (I_k1), which are able to “clamp” the resting membrane potential at values close to the equilibrium potential for K⁺ ions (E_K)⁵². To assess whether I_K1 underlies the differences in spontaneous APs and beating between C16 and RPMI + B27, we applied extracellular Ba²⁺ at a dose (400 µmol L⁻¹) that potentially blocks I_K1 channels. While Ba²⁺ addition had minimal effect on either MDPs (49.2 ± 5.3 mV) or the beating rates of RPMI + B27 cells, it caused C16-treated cells to display spontaneous beating and APs accompanied by spontaneous depolarizations from MDPs of −53.7 ± 1.8 mV. Our findings suggest that C16-treated cells express a high number of I_K1 channels compared to RPMI + B27-treated cells. Indeed, in voltage-clamp studies, C16-treated cells demonstrated robust background K⁺ currents with characteristic kinetic behavior, classical inward rectification properties (Fig. 2d, e) and reversal properties (i.e., ~E_K) which are hallmark features of I_K1. Moreover, these background currents in the C16 treatment were eliminated by 400 µmol L⁻¹ Ba²⁺ (Fig. 2e). RPMI + B27 cells also displayed Ba²⁺-sensitive background K⁺ currents but these were approximately four-fold smaller than in C16-treated cells (i.e., Ba²⁺-sensitive densities measured at −115 mV were −21.9 ± 8.2 pA pF⁻¹ in C16 versus −5.5 ± 3.9 pA pF⁻¹ in the RPMI + B27 treatment, Fig. 2f) which consistent with the observed effects of Ba²⁺ on APs and beating of C16- vs. RPMI + B27-treated cells. Measurement of I_K1 density slope between −115 and −95 mV further demonstrated significant differences between C16-treated cells and those treated with RPMI + B27 or the high-performing medium of Feyen et al.²¹ (Fig. 2g).

Fig. 2. Electrophysiological enhancement of hiPSC-CMs matured using C16 medium.

a Sample traces of intracellular electrode recordings of hiPSC-CMs treated with RPMI + B27 (left) or C16 maturation medium (right) for 6 weeks, in neat Tyrode’s solution, followed by sequential supplementation with 100 and 400 mol L⁻¹ BaCl₂, before eventual washout. b Minimum diastolic potentials (resting membrane potentials) from intracellular recording in cell sheets, with or without 400 µM BaCl₂ perfusion (n = 31 and 7 biological replicates for RPMI + B27 without and with BaCl₂, and 16 and 3 biological replicates for C16 without and with BaCl₂). c Spontaneous action potential firing rates measured during action potential recordings, with or without 400 µmol L⁻¹ BaCl₂ perfusion (n = 8 and 3 biological replicates, respectively, for RPMI + B27 and C16 treatments). d Inward rectifier (I_K1) currents in neat Tyrode’s and supplemented with 400 µmol L⁻¹ BaCl₂; inset indicates voltage protocol and scale. Voltage sensitivity curves (e) of I_K1 (inset demonstrates positive component of I_K1 at high voltage) and I_K1 current density at −115 mV (f) (n = 8 and 4, respectively). g I_K1 current density slope between −115 and −95 mV for treatments of RPMI + B27, cardiac maturation medium as described by Feyen et al. (“Feyen”)²¹, or C16 (n = 6, 10, and 6 biological replicates, respectively). Data points denote biological replicates. For panel 2b, two-way ANOVA demonstrated statistical significance for medium treatment (p = 0.0012), Ba²⁺ treatment (p < 0.0001), and interaction (p = 0.042). For panel 2c, two-way ANOVA demonstrated statistical significance for medium treatment (p < 0.0001) and Ba²⁺ treatment (p < 0.0001). Tukey’s post-hoc test was used to compare individual conditions in both panels 2b, c (significance shown in the graph). For panel 2 f, a two-tailed t-test was used. For panel 2 g, conditions were compared using a one-way ANOVA with Tukey’s post-hoc test. Error bars denote mean ± SD; *, **, ***, and **** denote p < 0.05, 0.01, 0.001 and 0.0001, respectively.

Physiological characterization of maturing hiPSC-CMs

hiPSC-CM monolayers treated with C16 medium demonstrated differential Ca²⁺ handling and metabolic flux compared to high-performing control formulations after 6 weeks of culture. Measurement of steady-state Ca²⁺ handling at 1 Hz pacing using line-scanning microscopy demonstrated accelerated onset (Fig. 3a, b) and decay kinetics (Fig. 3c–e) in randomly-selected C16-treated cells vs. controls, which may reflect a greater dependence of Ca²⁺ cycling on the sarcoplasmic reticulum (SR) seen in mature versus immature myocardium³⁴. More direct evidence of enhanced SR function in the C16 culture conditions was uncovered by perfusing cells with high levels of caffeine to cause complete release of the SR’s Ca²⁺ into the cytosol. Specifically, when cells were pre-treated with verapamil to prevent Ca²⁺ entry into the cytosol via L-type channels, a bolus of caffeine generated a rise in intracellular Ca²⁺ within C16 cells that was approximately two-fold higher than for cells cultured in STEMCELL medium (p < 0.0001) (Fig. 3f–h). Also consistent with increased SR contributions, administration of the β-adrenergic receptor agonist, isoproterenol, further accelerated Ca²⁺ transient onset in C16-treated cells only (Fig. 3i).

Fig. 3. Functional enhancement of hiPSC-CMs matured in C16 medium.

a Representative line-scanning Fluo-4 traces during steady-state Ca²⁺ transients at 1 Hz pacing of single hiPSC-CMs treated 6 weeks with C16 maturation medium compared to high-performing control hiPSC-CM media (iCell Cardiomyocytes Maintenance Medium, RPMI + B27, and STEMCELL Cardiomyocyte Maintenance Medium); kinetic metrics isolated from trace sets include time to peak (b), and times to 10% (c), 25% (d), and 50% (e) decay from peak fluorescence, respectively. Comparisons between high-performing comparative treatments vs. the C16 formulation by Brown-Forsythe ANOVA to account for unequal variances between treatments and Dunnett’s multiple comparisons post-hoc test. n = 54 individual cells for C16, n = 47 for iCell, n = 38 for RPMI, and n = 37 for STEMCELL. Fluo-4-enabled Ca²⁺ traces of C16 (f) and STEMCELL (g)-treated cultures after 6 weeks with steady-state pacing at 1 Hz, followed by sequential additions of verapamil (20 µmol L⁻¹) and caffeine (10 mmol L⁻¹). h Fractional Fluo-4-enabled sarcoplasmic reticulum (SR)-derived calcium load upon addition of caffeine normalized to steady-state transient amplitudes for each cell; comparisons did not assume equal variance and were analyzed by Welch’s t-test; n = 12 biological replicates for C16, n = 8 for STEMCELL. i Relative change (%) of baseline time to peak fluorescence in single cells upon addition of isoproterenol (5 µmol L⁻¹); comparisons between high-performing comparative treatments vs. the C16 formulation by Brown-Forsythe ANOVA to account for unequal variances between treatments and Dunnett’s multiple comparisons post-hoc test. n = 6 biological replicates for each of C16, iCell, and RPMI, and n = 5 for STEMCELL. Representative mitochondrial stress test traces demonstrating (j) oxygen consumption (OCR) and (k) extracellular acidification (ECAR) rates of hiPSC-CMs cultured 6 weeks in C16 medium or existing high-performing control formulations, normalized to cell number. Normalized basal oxygen consumption rates (l) and uncoupled-control ratios (m) by treatment; comparisons of high-performing comparative treatments vs. the C16 formulation by repeated measures ANOVA and Dunnett’s multiple comparisons post-hoc test to account for batch effects of the respirometric assay. Enrichment of central carbon metabolic pathways (n–q) and peripherally-associated metabolite pools (r-u) with either uniformly ¹³C-labeled glucose or lactate; including citrate (n), succinate (o), malate (p), glutamate (q), lactate (r), fructose-1,6-bisphosphate (s), gluconate (t), and sedoheptulose-7-phosphate (u); conditions were compared using a one-way ANOVA with Tukey’s post-hoc test separately for the m = 0 and final isotopologue (i.e., m + 3–5) for each metabolite; n = 4 biological replicates except for ¹³C-glucose-labeled samples treated with the Feyen et al. medium (n = 3). Error bars denote mean ± SD; *, **, ***, and **** denote p < 0.05, 0.01, 0.001 and 0.0001, respectively.

C16-treated hiPSC-CMs also diverged from high-performing control formulations in metabolic profiling using Seahorse XFe96 respirometry (Fig. 3j, k). Specifically, the per-cell basal oxygen consumption rate outpaced control treatments (Fig. 3l), while the FCCP-enabled UCR of C16-cultured cells was higher than those of the iCell (p = 0.035), RPMI + B27 (p = 0.023), and STEMCELL (p = 0.013) control treatments (Fig. 3m). These findings were consistent with targeted metabolomic analyses of hiPSC-CM monolayers at the same timepoint, treated with either the C16 formulation, the formulation of Feyen et al.²¹, or RPMI + B27 as a control. C16-treated monolayers demonstrated greater incorporation (i.e., lowered m + 0 proportion) of labeled glucose and lactate into aerobic pathways (Fig. 3n–q), with reduced anaerobic metabolism (Fig. 3r–t) while maintaining pentose phosphate flux (Fig. 3u).

Transcriptomic profiling of maturing hiPSC-CMs

RNA sequencing was used to evaluate the transcriptomic profiles of hiPSC-CMs treated with C16 or high-performing control maintenance media to help identify targets for further physiological analysis. PCA was used to compare the overall transcriptional profiles of each medium relative to immature day 20 (d20) controls, demonstrating that C16-treated cells were highly divergent from both controls and other media (Fig. 4a). Many of the GO terms significantly upregulated in C16 treatments were associated with general cardiomyocyte functions, including the cardiac action potential, contraction, regulation of contractility, and oxidative metabolism, while GO terms heavily downregulated in C16 treatments included differentiation, motility, and anaerobic metabolism (Fig. 4b-j). Similarly, gene set enrichment analysis (GSEA) comparing C16 to high-performing control treatments revealed GO terms pertaining to heart function, lipid metabolism and oxidative phosphorylation, cell-cell adhesion, and transcriptional regulation (Supplementary Figs. 2, 3).

Fig. 4. Transcriptomic profiling of matured hiPSC-CMs and RNA sequencing-enabled meta-analysis to ex vivo tissues and historical maturation experiments.

a Principal component analysis of hiPSC-CMs cultured in C16 medium or a high-performing control formulation for 6 weeks, or at d20 of differentiation and prior to treatment (control). Colors denote different sample groups. b–j Heatmaps displaying expression of select genes from the GO terms; top 5 upregulated (red) and downregulated (blue) genes within C16-treated cells per GO term, listed with their respective Z scores. n = 3 biological replicates per treatment. Head-to-head treatment comparisons found in Supplementary Figs. 2, 3. k, l Meta-analysis of bulk RNA sequencing of hiPSC-CM in vitro maturation compared to native CMs at fetal, infant, and adult life stages, including tSNE (k) and principal component analysis (l). Sample descriptions found in Supplementary Table 3; validation found in Supplementary Fig. 4, principal component attribution in Supplementary Fig. 5, and cluster analysis in Supplementary Fig. 6.

Meta-analysis of (hiPSC-)CM maturation

RNA-seq datasets from several studies of CM maturation were pooled into a single meta-analysis to better understand transcriptional trajectories, or the gene expressional landscape around the physiological maturation process. To enable interpretation, external samples were divided into one of 7 broad categories. In vivo myocardial or CM samples were assigned to the ‘Adult’, ‘Infant’ or ‘Fetal’ category based on developmental timepoint at isolation. Matured in vitro hiPSC-CM samples were assigned to the “2D”, “3D” or “EB” categories, depending on the culture methodology and type of maturation approach employed (see Supplementary Table 3 for detailed descriptions). ‘Control’ samples were immature hiPSC-CMs, collected at ~d18–20 of differentiation, after the onset of spontaneous beating but prior to starting any type of maturation protocol.

To adjust for batch effects introduced by the inclusion of multiple datasets, the count matrix was adjusted to minimize variation among samples from different studies, while preserving variation among the designated sample categories (Supplementary Fig. 4). Following batch correction, tSNE and PCA analyses demonstrated that the external samples could agnostically be re-separated into distinct clusters according to their pre-assigned categories (Fig. 4k, l). Both GSEA of principal components (Supplementary Fig. 5) and differential expression analysis (Supplementary Fig. 6) revealed similar trends in functional gene expression between unique clusters; the PC1 positive direction (2D cluster) was enriched for electrophysiological function and electrical conduction, PC2 positive (3D cluster) enriched for contractile function and Ca²⁺ handling, and PC1 negative (EB cluster) enriched for cardiac developmental and morphogenic processes. DEA further identified specific functions enriched in each cluster that were not identified according to the PCA loadings, such as GPCR signaling and lipid handling in the “3D” cluster, and Ca²⁺ handling in the “2D” cluster.

Proteomic profiling of maturing hiPSC-CMs

Transcriptomic analysis suggested marked changes to transcriptional regulation, protein synthesis and turnover, and protein trafficking. To directly examine protein changes in the 2D system, global proteomic analysis of urea-soluble fractions was conducted on hiPSC-CMs after 6 weeks of treatment in C16, iCell, RPMI + B27, and STEMCELL media, as well as the formulation designed by Feyen et al.²¹ (Fig. 5). Principal component analysis (PCA), conducted on a set of 1869 proteins that had valid (greater than 0 LFQ intensity) values in all three samples from at least one experimental treatment, demonstrated that PC1 largely separated the STEMCELL and C16 conditions, while PC2 separated the high-performing Feyen medium from the other four conditions (Fig. 5a). Both PC loadings included metabolic and mitochondrial-specific proteins; the positive direction of PC1 also heavily featured sarcoplasmic synthesis and maintenance ontologies, ionoregulatory proteins, and membrane trafficking components, while the positive direction of PC2 heavily featured cytoskeletal proteins associated with contractility. Enriched GO terms associated with maturation conditions included membrane functionalization, oxidative metabolism, and mRNA processing, while downregulated terms centered on apoptotic induction and protein localization (Fig. 5b, c). Of particular note, MYH7 protein expression was not significantly downregulated in C16-treated samples, in contrast with its transcriptional downregulation in identical samples. This may suggest that regulation beyond the level of mRNA transcription may contribute to the functional expression of maturity-associated protein products.

Fig. 5. Global proteomic profiling of C16-treated hiPSC-CMs relative to high-performing control formulations.

a Principal component analysis of differential protein expression in hiPSC-CMs after 6 weeks of treatment in C16, iCell, RPMI + B27, or STEMCELL media, as well as a top-performing formulation from the literature²¹. b Heat map demonstrating stratification of treatments by protein regulation. c Over-representation analysis of the top 5 GO terms differentially expressed in C16-treated hiPSC-CMs compared to other treatments for each cluster in panel (b); GO term size (in number of gene products) represented both by marker size and size on X-axis.

Maturation of hiPSC-CM-containing engineered cardiac microtissues

Marked differences in the level of attainable hiPSC-CM function relative to monolayers have previously been found in the context of engineered and co-cultured microtissues, leveraging physiological, mechanical and electrical cues to drive maturation. We used the Biowire II platform³² to examine further the impact of our maturation medium on parameters of CM function (Fig. 6). These studies involved the creation of compacted multicellular muscle strips³², followed by 3 weeks of culture in either modified C16 (see “Methods”) or RPMI + B27, and with or without continual electrical field pacing (1 Hz). As in monolayer culture, Biowire tissues in C16 media exhibited lower rates of spontaneous contractility than in RPMI + B27 (Fig. 6a; Supplemental Movies 1–4) after 3 weeks in of treatment. Contractile forces normalized to cross-sectional area (i.e., twitch/systolic stress) were approximately five-fold higher with C16 versus RPMI + B27 treatment after 1, 2, or 3 weeks in culture (Fig. 6b), with or without electrical stimulation during the culture period. After 3 weeks in culture with electrical stimulation, the twitch stress was 1.69 ± 1.42 mN mm⁻² with C16 treatment vs. 0.35 ± 0.37 mN mm⁻² with RPMI + B27 treatment. By 3-way ANOVA, C16 treatment (p = 0.0002) and culture time (p < 0.0001) both independently enhanced twitch stress, and these parameters positively interacted (p < 0.0001), establishing greater benefit of culture time in C16 medium versus the high-performing control. Neither tissue morphology nor diastolic stresses were significantly affected by treatments (Supplementary Fig. 7). Furthermore, positive force-frequency responses after 3 weeks in culture were enhanced when measured at either 2 Hz or 3 Hz stimulation (relative to tissue-matched baseline stresses at 1 Hz) by both culture time (p < 0.0001) and C16 treatment (p < 0.0001) (Fig. 6c); again, there was a positive interaction (p < 0.0001) between these factors.

Fig. 6. Contractile enhancement of engineered myocardial tissues cultured in C16 maturation medium.

a Decrease in tissue contractile spontaneity after 3 weeks of culture in C16 media (n = 4 independent experiments, each comprising 5–8 individual tissues representing technical replicates). b Progressive absolute twitch force increased over 3 weeks of Biowire II culture in C16 or RPMI + B27 medium formulations, with or without continuous electrical field stimulation (1 Hz). c Positive force-frequency relationship of contractile displacement of polymeric beams summatively increases with both medium- and electrical stimulation-mediated maturation of tissues after 3 weeks of tissue treatment in C16 or RPMI + B27 media; measurements are normalized to 1 Hz stimulation of the relevant treatment condition (dotted line). For panels 6b and 6c, n = 9 biological replicates, except n = 11 for the C16 + stimulation condition. d Confocal microscopy reveals differential expression of SERCA2a (magenta) and DHPR (yellow) in electrically-stimulated Biowires cultured for 3 weeks in C16 or RPMI + B27 media. e TEM of longitudinal and transverse sections of C16-treated Biowires induces sarcomeric development and higher ribosome-associated sarcoplasmic reticulum (scale bars 1 µm); detailed examination of C16-treated tissues (e–h; scale bars 500 nm) includes sarcomeres with defined M- and H- bands and Z-disks and myofibril-associated high mitochondrial density (f), electron-dense intercalated disks (g), and nuclear SR invaginations (h). i–l Transcriptomic analysis of Biowire II cultures treated with C16 or RPMI + B27 medium and with or without electrical field stimulation. Principal component analysis (i) and associated GO terms stratifying treatments, top 5 upregulated (red) and downregulated (blue) genes within all C16-treated tissues per GO term, listed with their respective Z scores; n = 4 biological replicates per C16 treatment and n = 3 biological replicates per RPMI + B27 treatment. Micrographs are demonstrative of at least 5 independent experiments each. Error bars denote mean ± SD. Significance indicated by * p < 0.05, ** p < 0.01, and *** p < 0.005. Stress value calibration and tissue characterization are found in Supplementary Fig. 7; breakdown of principal components in (i) is found in Supplementary Fig. 8.

Tissues treated for 3 weeks with the C16 formulation also exhibited higher-density SERCA2a and DHPR staining than those treated with RPMI + B27, similarly to in monolayers (Fig. 6d). However, DHPR expression was variable between cells, suggesting stratification of phenotypes within the tissue. Compared to RPMI + B27 treatment, C16-treated tissues under transmission electron microscopy demonstrated increased myofibrillar bundling, sarcomeric striation and zonation, mitochondrial density and myofibrillar association, electron-dense intercalated disks, and nuclear invaginations by the SR⁵³ (Fig. 6e–h). Consistent with both the functional and morphological differences observed in C16-treated tissues, the two formulations were highly stratified by PCA of bulk RNA sequencing primarily by medium treatment and not electrical field stimulation (Fig. 6i). Highly enriched GO terms in C16-treated cultures pertained most strongly to oxidative metabolic and contractile ontologies, while RPMI + B27-treated culture segregation was driven by developmental ontologies (Fig. 6j–l; Supplementary Fig. 8).

Discussion

In vitro myocardial models remain morphologically, molecularly, and functionally distinct from their in vivo counterparts, notwithstanding recent advancements in knowledge of CM maturation^34,45,54. Furthermore, the approaches used in culture medium design to date have been largely prescriptive, and factor selection has been limited by the feasibility of assaying quantitative and high-throughput markers of maturation.

Traditional regression-based optimization methodologies must be limited in scope to be logistically feasible and of sufficient statistical resolution to be actionable. We leveraged a differential evolutionary algorithm to survey a much wider set of additives over successive iterations, using a metric of metabolic organization and anabolism (UCR) that, by self-normalization, allowed for maximizing separate runs per generation. A range of UCR performance was observed across the multiple generations of iteration, but the final C16 formulation was validated to outperform existing high-performing commercial and homemade media in all metrics tested, both quantitative and qualitative, making it a key candidate for further benchmarking. C16-treated hiPSC-CMs displayed marked morphological changes, including hypertrophy, elongation, and sarcomeric striation evident under brightfield examination, and corresponding development of T-tubule-like (DHPR positive) and SR-like (SERCA2a and RyR2 positive) structures (Fig. 1). Differential contractility was evoked by C16 treatment across two healthy hiPSC-CM lines, as well as a CRISPR-induced mutant line.

The appearance of inward rectifier K⁺ currents (I_K1) is a key milestone in our CM maturation approach, as it is needed to prevent spontaneous AP firing in working atrial and ventricular CMs^55,56. This quiescence is essential in vivo to allow nodal control of ventricular contraction combined with proper coordination of ventricular contraction by Purkinje fibers, thereby ensuring efficient pumping action. C16-treated hiPSC-CMs exhibited several markers of advanced electrophysiological maturation (Fig. 2), including characteristic I_K1 current^21,55,57. The upregulation of I_K1 in genetically-engineered hiPSC-CMs was previously demonstrated to eliminate electrophysiological spontaneity⁵⁵, but these functional levels have not been replicated in wild-type hiPSC-CMs. It is notable that C16-treated cells achieved I_K1 densities roughly equivalent to those measured in isolated adult cardiomyocytes ( ~ 20–25 pA/pF at −115 mV), although I_K1 densities are known to display regional heterogeneity⁵⁸. Accordingly, C16-treated hiPSC-CMs were generally quiescent after approximately 3 weeks, and did not display APs or contractile spontaneity at 6 weeks unless I_K1 was blocked by targeted Ba²⁺ application. Furthermore, our measurement of the resting membrane potential (−81.1 ± 7.7 mV) well-approximates adult measurements around −80 to −85 mV, which are near the analytically-derived K⁺ equilibration potential and serve to prevent arrhythmias in vivo⁵⁶.

In addition to their electrophysiology, C16-matured hiPSC-CMs demonstrated advanced Ca²⁺ handling and metabolism (Fig. 3). Kinetics of steady-state paced Ca²⁺ transients were generally accelerated, and cells exhibited a marked kinetic response to isoproterenol stimulation, which is consistent with the observation of increased β-adrenergic response in maturing CMs^6,15,18. Importantly, this response relies not only on increased expression of the β-adrenergic receptor but also on the formation of functional dyads to potentiate SR²⁺ Ca store release. The presence of functional SR Ca²⁺ reuptake, as opposed to only sarcolemmal Ca²⁺ efflux, would also be consistent with the accelerated early decay kinetic parameters observed in baseline Ca²⁺ transients of C16-treated cells. Furthermore, upon challenge, matured cells demonstrated increased verapamil (an L-type current blocker) sensitivity and had significantly increased caffeine-sensitive SR Ca²⁺ stores. This recapitulation of advanced CM functionality was reflected in metabolic analysis, where cells demonstrated higher per-cell oxidative rates, as well as higher central carbon flux and a lower reliance on anaerobic glycolysis, all hallmarks of mature CM metabolism⁵⁹.

Finally, transcriptomic analysis of matured hiPSC-CM monolayers (Fig. 4) suggested strong evidence of functional maturation, including downregulation of products associated with differentiation, proliferation, and motility, and upregulation of products associated with excitation-contraction coupling, electrophysiology, lipid transport, and oxidative metabolism. Proteomic analysis (Fig. 5) reflected many of the expressional changes expected to accompany increased contractility, improved Ca²⁺ handling, and metabolic function, including cytoskeletal, sarcoplasmic and trafficking, and substrate trafficking and OXPHOS ontologies, respectively. Targeted proteomic approaches may produce additional insight with respect to the electrophysiological profile of maturing hiPSC-CMs. Furthermore, the transcriptional downregulation of key traditional markers of hiPSC-CM maturity, such as MYH7 and TNNI3, in the C16 treatment was not reflected in the expression of their protein products, suggesting that transcriptional assessment without corresponding protein or functional verification may not be sufficient in determining the maturation status of advanced hiPSC-CM cultures. Additionally, given the electrophysiological basis for non-spontaneous contraction in the C16-treated hiPSC-CMs, it is possible that decreased cyclic mechanical strain contributed to lower turnover and, therefore, gene expression, despite the stability of corresponding protein expression.

The increase in hallmark CM function was also observed in Biowire II engineered myocardial tissues, where treatment with C16 medium was additive to continuous electrical field stimulation in evoking increased functional performance (Fig. 6). Tissues treated with C16 medium and subjected to non-progressive electrical stimulation demonstrated steady-state twitch stresses near the peak of measurements recorded of engineered myocardium at physiological [Ca²⁺] and at 1 Hz ( ~ 1.7 mN mm⁻²). These twitch stresses remain below the strongest recorded in engineered myocardial tissues (2–5 mN mm⁻² at equivalent conditions)¹⁹, but did not require the extensive fabrication or culture conditions employed in the study, suggesting a benefit to workflow and adaptation. Ex vivo measurements of healthy adult myocardial twitch stress from ventricular strip, trabecular, or papillary muscle preparations generally range from 5–20 mN mm^−260–66, although some estimates reach 44 mN mm⁻²⁶⁷; regardless, isometric preparations are not limited by the shortening ability of tissue and therefore can be expected to exceed the approximately isotonic measurements as measured in this study. Maturing function was further reflected in the highly positive force-frequency (Bowditch) response of C16-treated tissues, and in the tissue architecture, functional protein expression, and TEM-visible ultrastructure associated with myocardial maturation. Interestingly, the well-validated importance of electrical stimulation of PSC-containing microtissues in advancing physiological function^18,19,68 was replicated in our study, but overshadowed by the statistical effect of culture medium treatment in both functional and transcriptomic analysis. However, despite the transcriptomic similarity between unstimulated and stimulated C16-treated tissues, stimulated tissues consistently outperformed the unstimulated condition on physiological analyses, suggesting non-transcriptomic contributions to tissue function. Efforts to further improve tissue functionality could include extended culture periods, the use of progressive stimulation protocols^18,19, or increasing cell diversity in co-cultures.

Several previous studies have examined the transcriptomes of maturing PSC-CMs^{15,18,21,22,69–71} or native myocardium^15,23,72,73, but most maturation protocols have been characterized only relative to freshly-differentiated PSC-CMs or time-matched existing high-performing controls. Using these internal controls, we were able to construct a meta-analysis of CM maturation both in vitro and in vivo. The positioning of the maturation protocol clusters relative to the in vivo clusters appeared to indicate that 2D approaches promote maturation of the transcriptome most effectively, while EB approaches reduce maturity relative to controls. Analysis of the PC loadings challenges this interpretation, as functional categories associated with CM maturation could be identified in all PC loadings, diverging from freshly-differentiated PSC-CM controls. Furthermore, each maturation protocol included in the meta-analysis has been experimentally shown to promote functional maturation in some capacity, even in EB methods, which most oppose in vivo samples in the PCA. However, in this study and others, 3D co-culture of CMs and fibroblasts more closely replicates the mechanical and biochemical niche of myocardium and evokes emergent functional physiological characteristics and metrics that have not to date been achieved in simple monolayers; their transcriptomic trends are further consistent with marked downregulation of proliferative ontologies, but incomplete downregulation of developmental gene programs. Collectively with our insights from paired proteomic experiments, these observations therefore suggest that functional maturation may not be required to follow the same linear trajectory observed for the transition from fetal to adult CMs in vivo, and that achieving an ‘adult-like’ transcriptional profile may not be absolutely required for functional maturation. Additionally, the direct interpretation of in vitro transcriptomic datasets in the context of in vivo samples is complicated by the wealth of cell types in myocardial tissue, including endothelial cells, fibroblasts, pericytes, and immune cells⁷⁴, which, even if removed from digested samples bound for sequencing, may exert regulatory control on CM phenotype in situ.

The development of a highly-mature PSC-CM source will enable critical advancements in several downstream applications, particularly ones that require PSC-CMs at an advanced stage of electrophysiological development. Cell transplantation applications of PSC-CM technology have been complicated by the presence of graft-induced focal arrhythmias, ostensibly due to continued AP spontaneity⁷⁵. Furthermore, Torsades de Pointes ventricular tachyarrhythmias result from both genetic and pharmacological risk and are notoriously difficult to model in vitro; TdP stems from hERG (human ether-a-go-go; K_v11.1) current blockade, but this current is highly dependent on other co-existing CM currents, complicating pharmaceutical design and predictive preclinical evaluation^8,9. Further development and in-depth examination of high-fidelity and electrophysiologically mature hiPSC-CMs may provide insight into their suitability for cardiotoxicity screening applications.

In conclusion, the iteratively optimized C16 medium significantly advanced maturation of PSC-CMs toward that of mature adult tissues in both monolayer and co-culture organoids; this formulation may be beneficial for applications in basic physiology, drug screening, personalized medicine, or the design and execution of cell or tissue therapies. This study also directly compares a large collection of CM maturation datasets within a single analysis, providing insight into potential lacking physiological ontologies in existing maturation protocols (including the one presented in this study). Furthermore, within our own study, we demonstrate inconstancies between transcriptomic, proteomic, and functional analyses, demonstrating the importance of the use of different levels of assessment of physiological maturity. There is likely significant benefit to further testing of novel hiPSC-CM maturation medium formulations, with additional additives that may either push maturation signaling or supplement crucial metabolic processes. By providing a framework to achieve advanced levels of hallmark functional hiPSC-CM metrics, our findings suggest that the process and components underlying functional CM maturation may be more highly regulated and multifaceted than previously understood, and provide a basis for the generation of novel hypotheses regarding the process and regulation of CM maturation. More generally, this study provides a proof of concept for high-dimensional optimization of stem cell-derived cultures and engineered tissues.

Methods

No statistical methods were used to predetermine sample size. Assignment of experimental wells was randomized, but blinding of conditions could not be used during experiments or outcome assessment due to physical differences (e.g., medium color) between the treatments used and the morphological differences in the resulting cells.

Design of an iterative algorithmic search for optimized maturation medium

The C16 maturation medium was non-prescriptively optimized in a large solution space ( ~ 763 billion discrete formulations for an analogous 5 ¹⁷ full factorial experimental design), with generations of formulations iteratively scored and evaluated according to the respirometric uncoupled:control ratio (UCR) as measured by Seahorse XFe96 analysis as an objective metric. This design philosophy allowed for agnostic testing and the evolution of emergent physiological interactions between multiple defined soluble factors within a formulation; several of these factors had not been surveyed previously for efficacy toward PSC-CM maturation. Additionally, we used exclusively M199-based formulations for two reasons: firstly, the ionic profile of M199 is closer to that of human plasma than DMEM- or RPMI 1640-based media, especially in [Ca²⁺], for which homeostasis is vital to developing CMs; and secondly, that M199 contains a much more diverse source of cofactors and vitamins than either basal DMEM or RPMI 1640 media. For the latter consideration, the use of a defined formulation negates the possibility of serum providing these essential components, especially in a highly-specialized cell such as a maturing CM, which may outsource many non-core biosynthetic and secondary metabolic pathways to other tissues of the body. Soluble factors were chosen to be included in the screen based on inclusion in other published PSC-CM maturation media (e.g., lactate, galactose, lipid supplements, albumin), inclusion in adult primary cell culture media (e.g., creatine, carnitine, taurine, insulin, transferrin, selenium), hormones implicated in myocardial development (e.g., LIF, IGF-1, triiodothyronine, hydrocortisone, neuregulin), secondary metabolites implicated in membrane stability (e.g., putrescine, ethanolamine), and exogenous compounds with impacts on central carbon metabolism and mitochondrial activity (e.g., 2-deoxyglucose, metformin).

The formulation search was performed using an existing iterative High-Dimensional, Differential Evolution (HD-DE) decision tree algorithm programmed in MATLAB⁴⁸; the published script was modified for R2018a version compatibility, for the use of 17 variable factors, and for duplicate runs. Briefly, the algorithm generates formulation candidate vectors within a defined number of runs, factors, and levels of said factors. Maximal coverage during the initial generation is generated using a Sobol quasi-random distribution (e.g., minimizing discrepancy). After a generation is produced, objective metric scores are reported, theoretically forming a rough Pareto distribution. Formulations within 10% of the top scorer are chosen to continue in the next generation, while poor-scoring formulations are either replaced with randomly-generated vectors or with offspring vectors from the intergenerational memory of top scorers. The process is formally terminated when the median candidate of the Pareto-ranked generation does not improve by at least 10% after 3 consecutive generations. Via formulation screening by the UCR objective metric, the top 10 scoring formulations were subjected to PCA to establish trends and differences in high-performing formulations, as well as possible mechanistic explanations thereof. From analysis of the top 11 HD-DE-generated formulations, a manual adjustment was made from the clusters formed to generate the C16 medium. The C16 formulation maintained most of the characteristics from its parent cluster with 3 key changes: 1) lactate, which was deemed to be redundant in the presence of multiple other carbon sources, was removed to reduce manufacturing burden, 2) ethanolamine, which was otherwise ubiquitously present in other HD-DE lineages, was added at a low dose, and 3) growth hormone, which was variably present or excluded from lineages, was removed to reduce cost and manufacturing burden.

Materials

For iterative formulation candidate screening and characterization of the C16 formulation, soluble components used were 2-deoxyglucose (DXG498; BioShop), sodium L-lactate (L7022; Sigma), Chemically Defined Lipid (11905031; Gibco), D-galactose (GAL500; BioShop), creatine monohydrate (CREE200; BioShop), L-carnitine hydrochloride (C0283; Sigma), taurine (TAU303; BioShop), insulin-transferrin-selenium (41400045; Gibco), bovine serum albumin fraction V (10735078001; Roche), β-mercaptoethanol (M3148; Sigma), putrescine dihydrochloride (PUT001; BioShop), ethanolamine hydrochloride (E6133; Sigma), triiodothyronine (T6397; Sigma), recombinant insulin-like growth factor 1 (PHG0071; Gibco), recombinant leukemia inhibitory factor (SRP3316; Sigma), growth hormone (869008; Millipore Sigma), hydrocortisone (H0888, Sigma), recombinant neuregulin 1β2 (ab73753; Abcam), and metformin hydrochloride (ICN15169101; MP Biomedicals), in a base of Medium 199 (M4350; Sigma). High-performing controls used were iCell cardiomyocytes maintenance medium (CMM-100-120-001; Cellular Dynamics), cardiomyocyte maintenance medium (05020; STEMCELL Technologies), and RPMI 1640 (R8758; Sigma) with 1X B27 supplement (17504044; Gibco). An additional top-performing formulation from recent literature (“Feyen”) was prepared as previously described²¹ and used in proteomic and metabolomic analyses as an exemplar of maturation progress.

For immunofluorescent staining, anti-α-actinin (ab9465; Abcam), anti-connexin 43 (ab217676; Abcam), anti-ryanodine receptor (ab2868; Abcam), anti-SERCA2 (ab137020; Abcam), anti-NCX1 (ab2869; Abcam), anti-DHPR (ab65266; Abcam), goat anti-mouse, Alexa Fluor™ Plus 488 (A32723, Invitrogen), goat anti-rabbit, Alexa Fluor™ 488 (A11008, Invitrogen), goat anti-mouse Alexa Fluor 568 (A11004; Invitrogen), and donkey anti-rabbit, Alexa Fluor™ Plus 594 (A32754, Invitrogen) were used, as well as phalloidin-tetramethylrhodamine B isothiocyanate, (P1951; Sigma), and DAPI (62248; Thermo Scientific). For respirometry, a DMEM-based XF medium (103334; Agilent) was used, with oligomycin (O4876; Sigma), FCCP (15218; Cayman Chemical), antimycin A (A8674; Sigma), and rotenone (NC0779735; Cayman Chemical) used as inhibitors. Mitochondrial staining was performed using MitoTacker Deep Red FM (M22426; Invitrogen). Unless otherwise noted, all other materials were obtained from Sigma.

Cells

An iterative formulation search and subsequent characterization of hiPSC-CM cultured in the chosen final maturation medium candidate (“C16”) was performed using the Personal Genome Project of Canada-17 hiPSC line (PGPC17), which has been extensively characterized across a wide range of differentiation pathways⁵¹. Further testing for cross-cell line compatibility was performed on PGPC14 hiPSC, and on MYBPC3-knockout PGPC17 cells generated via CRISPR; both lines were also previously characterized as hiPSC-CMs⁵¹.

Cells were frozen in clumps in mTeSR (STEMCELL Technologies) containing 10% DMSO, and thawed and cultured when ready for differentiation. Cells were cultured in mTeSR on Geltrex (Thermo-Fisher; LDEV-free, hESC-qualified; incubated at 1:100 in DMEM for 1 h on TCPS plates) and passaged in clumps using ReLeSR (STEMCELL Technologies), all according to manufacturer guidelines. Cells were passaged at least two times after thawing before being expanded for differentiation. PSC-CM differentiation was performed using the Cardiomyocyte Differentiation Kit (STEMCELL Technologies) according to manufacturer directions, with cell seeding densities optimized for each cell line (between 4–8 × 10⁵ cells well⁻¹ in a 12-well dish, previously coated with 1:100 Matrigel (Corning) in DMEM). Cells were cultured post-differentiation until d18 post induction before reseeding for maturation experiments. Obtained monolayers in 12-well format were dissociated within 1 week of differentiation by incubation for 1 h at 37 °C in 1 mL Hanks buffer, containing 200 U mL⁻¹ Collagenase Type II (Worthington), with 0.5 mL TryPLE Select (Gibco) added for 15 additional minutes. Cells were centrifuged 5 min at 300 x g, resuspended, and plated in RPMI + B27 for 1 d culture before applying treatments. For monolayer imaging, cells were plated in 96-well polymer coverslip-bottom imaging plates (89626; ibidi) and treated as above. Non-screening treatments were applied to monolayers for 6 weeks; all monolayer cultures were performed in the absence of antibiotics.

Formulation screening

Soluble factors used in the iterative screening and final C16 formulation (Supplementary Table 1) were selected with the goal of providing a range of substrates, cofactors, and hormones that would not necessarily be created by a maturing or mature CM but which would be conducive to increased biosynthesis and hypertrophy, oxidative metabolism to fuel highly-energetically demanding physiological processes, or activation of pathways implicated in maturation of myocardium or other functional tissues in vivo. In all, 169 unique formulations were queried over 4 generations of varying size, using the Seahorse XFe96 mitochondrial stress test to interrogate metabolic function.

Cells were dissociated from differentiation wells and seeded at 8 × 10⁴ cells well⁻¹ in CM support medium on Matrigel-coated XFe96 monolayer plates (Agilent Technologies). Culture medium was changed the next day to the well’s respective treatment. Cells were then cultured for 21 days in 60 µL medium well⁻¹, with full medium changes every second day for 3 weeks in either duplicates of a formulation candidate, or one of four wells of control Cardiomyocyte Maintenance Medium (05020; STEMCELL Technologies). The four corner wells of the plate were additionally left unseeded as internal measurement controls according to standard XFe96 manufacturer guidelines.

Respirometry

Cells were equilibrated 45 min before metabolic characterization in an initial volume 150 µL Seahorse XF base medium (103334-100, Agilent Technologies) containing additional (in mmol L⁻¹) glucose (5), pyruvate (1), glutamine (1), sodium lactate (5), and 1X Chemically Defined Lipid, to best supply the oxidative substrate flexibility of mature CMs⁷⁶. A mitochondrial stress test was performed with sequential injections of 25 µL each (final concentrations): oligomycin (2.5 µmol L⁻¹), carbonyl cyanide-4-phenylhydrazone (FCCP; 1 µmol L⁻¹ for iterative testing, or either 0.20, 0.40, 0.50, 0.65, 0.80, or 1.00 µmol L⁻¹ for characterization of hiPSC-CMs cultured in C16 medium), antimycin A and rotenone (2.5 µmol L⁻¹ each). Due to the high specific oxidative flux of CMs, a modified measurement protocol to avoid hypoxia was employed⁷⁷; cells were measured for 2 rather than 3 cycles at each step, with the minimum measurement time of 2 min.

Respirometric measurements were normalized to cell numbers per well. Each well was fixed in 2% formalin for 5 min, washed 3 times with Ca²⁺ and Mg²⁺-free PBS, stained with Hoechst (1 µg mL⁻¹) in PBS for 5 min, washed 3 additional times, and imaged using an IX71 inverted widefield fluorescent microscope (Olympus Corporation) with a FITC filter. Cell quantifications were performed by performing an automated count of nuclei using ImageJ 1.52p (NIH), by sequential use of the standard Otsu threshold, watershed segmentation, and particle count functions. Normalized per-cell oxidation rates (OCR) and uncoupled:control ratios (UCR) were compared between treatments using a repeated measures ANOVA and Dunnett’s multiple comparisons post-hoc test for high-performing comparative treatments vs. the C16 formulation, where indicated, to account for batch effects of the respirometric assay.

Immunofluorescence, confocal microscopy, and morphometric analysis

Cell immunofluorescence for manual analysis

Cultured cells were fixed with 2% paraformaldehyde for 10 min at room temperature, followed by 90% ice-cold methanol for 10 min, followed by permeabilization buffer (0.5% Triton X-100, 0.2% Tween-20 in PBS) for 30 min at 4 °C. Blocking buffer (5% FBS in permeabilization buffer) was then added and incubated for 1 h at room temperature. Cells were incubated with primary antibodies (α-actinin 1:500, Cx43 1:500) in blocking buffer overnight at 4 °C; incubation in rhodamine-red-conjugated phalloidin (Sigma) was 1 h at RT. Fluorophore-conjugated secondary antibody staining (anti-rabbit AlexaFluor® 488, anti-mouse AlexaFluor® 594, anti-mouse AlexaFluor® 647; Molecular Probes; 1:800) was performed at room temperature for 1 h in the dark. Nuclear counterstaining was performed using 1 μg mL⁻¹ DAPI at room temperature for 15 min in the dark. Images were acquired using an FV3000 confocal microscope with 405, 488, 561, and 640 nm lasers.

Cell immunofluorescence for quantitative analysis

Cultured cells were cultured in 96-well polymer coverslip-bottom imaging plates. At 3 weeks of culture, cells were washed in PBS with calcium and magnesium (PBS+/+) and incubated at 37 °C in 500 nmol L⁻¹ MitoTacker Deep Red FM dye for 40 min. The dye was prepared according to manufacturer instructions by solubilizing in anhydrous DMSO to a 1 mM stock, then diluting to 500 nM in either Cardiomyocyte Maintenance Media or C16 media. The cells were then washed in PBS+/+ and fixed in 4% paraformaldehyde for 10 min at RT prior to washing 3 times in PBS+/+ for 5 min, then permeabilized and blocked with 10% normal goat serum (NGS) in 0.5% Triton X-100/0.2% Tween-20 for 90 min at RT. They were then incubated with primary mouse anti-α-actinin (1:500) and primary rabbit anti-connexin 43 (1:800) in 5% NGS in 0.5% Triton X-100/0.2% Tween-20, overnight at 4 °C on an orbital platform. The next day, they were washed 3x in PBS+/+, then incubated with secondary Alexa Fluor 568–conjugated goat anti-mouse IgG (1:200) and secondary Alexa Fluor 488–conjugated goat anti-rabbit IgG (1:200), for 90 min at RT on an orbital platform. Then, after washing again in PBS+/+ 3 times for 5 min each time, cells were incubated with DAPI counterstain (1:1000) for 5 min at RT. Finally, cells were washed in PBS+/+ twice, then in distilled water once, and mounted in a drop per well of Aqua-Mount (Epredia 13800). At this point, cells were sealed and stored at 4 °C away from light for up to 3 months. Confocal images were obtained using an Olympus FluoView 3000 laser scanning confocal microscope (Olympus Corporation). Acquisition followed in 3 sites in each of 10 wells per media treatment condition of each of the 4 channels (DAPI, AF488, AF568, and AF647) using a 60X/1.35 NA oil immersion objective. Images were saved in files appended with plate, well, and site metadata.

Automated image morphometric analysis

Images for automated analysis, as acquired above, were analyzed with CellProfiler⁷⁸. A custom CellProfiler pipeline was built based on the CellPainting assay⁷⁹. The pipeline imported all images and applied an illumination correction filter. Each image was labeled with plate, well, and site information by extracting its metadata from the original files. The DAPI channel was first used to identify nuclei through a global Otsu thresholding method. The AF568 (α-actinin) channel was used next to identify cell borders through a global thresholding method, minimizing cross-entropy. A threshold correction factor from 0.25–0.4 and a regularization factor of 0.005 were also applied through trial and error to give the best cell segmentation results. Cells touching the edges of the image were then excluded from subsequent analysis. 1047 parameters were calculated and recorded for each cell. Notable parameters included cell area, perimeter, eccentricity, form factor, orientation, co-localization of α-actinin and connexin-43, and α-actinin intensity. These parameters quantified for each cell in each image (10 images per media treatment condition) were exported to a comma-separated values (CSV) file for statistical analysis. All the above-mentioned parameters were generated per cell as technical replicates to generate 1 biological replicate per well, except cell orientation, for which the average circular variance per biological replicate was taken instead. Normality was tested using a D’Agostino and Pearson test, and conditions compared using a one-way ANOVA with Tukey’s post-hoc test.

Sarcomere spacing measurement and analysis

In each of the image stacks acquired (5 wells per treatment condition × 3 sites per well), the AF568 α-actinin channel was used to manually measure sarcomere spacing. Representative myofibrils (5 per image) were selected, and the average distance between peaks were calculated automatically detected using the ImageJ integrated ridge detection plugin, and sarcomere length was determined by measuring the distance between intensity peaks between adjacent filaments. One cell was selected per image, and the lengths of 5 sarcomeres for that cell were acquired. Equality of variance was tested using the D’Agostino and Pearson test, and conditions were compared using a one-way ANOVA with Tukey’s post-hoc test.

Monolayer transmission electron microscopy

hiPSC-CMs were fixed with 4% paraformaldehyde in PBS, and 1% glutaraldehyde for 1 h at room temperature, fresh fixative was added, and samples were incubated overnight at 4 °C. Samples were washed with PBS, and subsequently incubated in 1% osmium tetroxide in 0.1 mol L⁻¹ sodium cacodylate buffer for 1 h at room temperature and washed. Samples were then dehydrated with a graded ethanol series (30%, 50%, 70%, 90%, 100%). Samples were resin-embedded by increasing concentrations of Epon resin mixed with propylene oxide in the following sequence (propylene oxide: Epon resin): 1:0 (30 min), 2:1 (2 h, agitated), 1:2 (3 h, agitated), 0:1 (overnight), 0:1 (2 h, agitated), and polymerized for 48 h at 40 °C. Samples were sectioned at 80 nm using a Leica Reichert Ultracut E microtome and stained using 5% uranyl acetate and 5% Reynold’s lead citrate, prior to imaging with a Talos L120C TEM (Thermo) and a BM-Ceta camera.

Optical contractile tracking of hiPSC-CM monolayers

Contractile kinetics were assessed from phase-contrast videos recorded at ~20 fps with a 40X objective, using a particle-image velocimetry package for ImageJ⁸⁰ to quantify time-resolved displacement over the contractile cycle relative to the relaxed state. Individual traces from biological replicates were fitted with a quartic regression using Prism 9 (GraphPad Software Inc., San Diego, USA).

Electrophysiological characterization

Intracellular recordings

Action potentials (APs) were recorded in single hiPSC-CMs or small clusters of hiPSC-CMs (n = 14 for C16 and n = 30 for RPMI + B27) using sharp microelectrodes of resistances between 40 and 90 MΩ (filled with 3 mol L⁻¹ KCl) with an Axopatch 200B amplifier. Intracellular voltages were digitized at a sample rate of 10 kHz (Digidata 1440 A plus pCLAMP 10.3, Molecular Devices). Voltage recording was performed at room temperature and perfused with modified Tyrode’s solution (pH 7.4) containing (in mmol L⁻¹): NaCl (120), KCl (5) CaCl₂ (2), MgCl₂ (1), Na₂HPO₄ (0.84), MgSO₄ (0.28), KH₂PO₄ (0.22), NaHCO₃ (27), and glucose (5.6)⁵⁵.

Patch-clamp recordings

Voltage-clamp measurements were used to record whole-cell currents in single isolated hiPSC-CMs using the patch-clamp technique at room temperature. The hardware was the same as for the AP recordings. The pipette solution (pH 7.2 adjusted with KOH) contained (in mmol L⁻¹): K-gluconate (150), EGTA (5), HEPES (10), and MgATP (5). The series resistance of the pipettes was 3-4 MΩ. The external bath solution (pH 7.35–7.4 with NaOH) contained (in mmol L⁻¹): NaCl (148), KCl (5.4), MgCl₂ (1.0), CaCl₂ (1.8), NaH₂PO₄ (0.4), glucose (5.5), and HEPES (15). The external perfusion solutions also contained nifedipine (5 μmol L⁻¹) to block possible overlapping Ca²⁺ currents. I_K1 was assessed by using voltage-steps from −115 mV to −20 mV (in 5 mV intervals) were applied for 500 ms from a membrane voltage of −60 mV. Currents were recorded at sampling rates of 10 kHz and analyzed using Clampfit 10.7 (Molecular Devices). Current-voltage (I–V) relationships were generated using peak currents as a function of membrane voltage. Currents were also recorded following the perfusion of BaCl₂ (100–400 μmol L⁻¹), which is a potent and selective blocker of I_K1 at such low concentrations⁸¹. Normality was not tested due to sample size limitations, and treatments were compared using a two-tailed t-test or one-way ANOVA with Tukey’s post hoc test, as appropriate.

Ca²⁺ imaging

Cells cultured in 12-well plate format were dissociated as described above and replated in single-cell format on 96-well plates with a #1.5 coverslip bottom treated with 1:100 Matrigel, and cultured in their treatment medium to recover for 2 days before imaging. Frozen dry vials of 50 µg Fluo-4 AM (Thermo-Fisher) were reconstituted with 50 µL DMSO containing 20% (w/v) Pluronic F-127. Cells were treated with 50 µL well⁻¹ Fluo-4 AM solution, diluted 1:100 in Tyrode’s solution, for 30 min at RT. Cells were rinsed twice with Tyrode’s solution and maintained up to 2 h thereafter in 100 µL Tyrode’s solution, containing 5 µmol L⁻¹ S-blebbistatin (Toronto Research Chemicals) to prevent movement artifact.

Ca²⁺ imaging was performed on a FluoView FV3000 (FV3000; Olympus Corporation) confocal microscope heated to 37 °C. Linescan measurements with 488 nm excitation and collection from 505-550 nm at 1% laser power and c.a. 500 V at the HSD were taken at the 20X objective and 2X zoom with spatial and temporal resolutions of 0.62 µm and c.a. 1.2–1.8 ms, respectively. Cells were subjected to steady-state stimulation at 1 Hz monophasic square-wave impulses of 2 ms and 40 V at approximately 0.6 cm using custom-made graphite electrodes connected to an S48 physiological stimulator (Grass Technologies, Warwick, RI). For monolayer SR capacity experiments, cells were paced at 1 Hz at steady-state before consecutive additions of verapamil (20 µmol L⁻¹) and caffeine (10 mmol L⁻¹). Fractional SR load upon addition of caffeine was normalized to steady-state transient amplitudes for each cell; comparisons did not assume equal variance and were analyzed by Welch’s t-test. Fluo-4 fluorescence measurements were taken using a CCD camera at 30 Hz framerate on an IX71 inverted epifluorescence microscope (Olympus Corporation, Tokyo, Japan) equipped with a FITC filter cube. Transients were manually extracted using ImageJ. Epifluorescent SR capacity transients were directly reported and interpreted. Transient confocal linescans were smoothed with a 7-frame rolling-average filter and analyzed for kinetics, using a custom MATLAB script. Kinetics parameters assessed included time from baseline to peak fluorescence, and times from peak fluorescence to a decay of 10%, 25%, 50%, and 75% peak magnitudes. Kinetic metrics were compared between treatments using a Brown-Forsythe ANOVA and Dunnett’s multiple comparisons post-hoc test for high-performing comparative treatments vs. the C16 formulation, where indicated, to account for unequal variances between treatments.

Microtissue maturation

Culture and functional benchmarking

Microtissues were created according to established methods³². Briefly, hiPSC-CMs mixed with primary human fibroblasts were cast in fibrin gels within embossed polystyrene molds and suspended between flexible poly(octamethylene maleate (anhydride) citrate) (POMaC) posts to allow for contraction against resistance. All strips were cultured in 10 cm cell culture dishes, containing 10 mL of medium, changed twice weekly, with 0.1% penicillin-streptomycin. The C16 medium used for monolayer culture was found to be incompatible with tissue function due to toxicity to primary cardiac fibroblasts (as found in monolayer fibroblast culture); a modified version without 2-deoxyglucose allowed continued tissue function and was used as the working version of C16 for all Biowire II experiments. Strips of up to 8 functional tissues were cultured for 1 week in RPMI + B27 medium, to allow compaction, before assignments to treatments in either modified C16 or continued RPMI + B27 media, and with or without electrical field stimulation (10 V cm⁻¹ at 1 cm, 1 Hz monophasic square wave pulses for 3 ms) via two graphite electrodes connected to an S48 physiological stimulator (Grass Technologies, USA). Tissues were assessed for function at 0, 7, 14, and 21 d of treatment for baseline paced diastolic stress, peak systolic twitch stress, and stimulation force-frequency relationship up to 6 Hz using an IX71 microscope at 37 °C. Polymeric wires were separately assessed for stiffness using a MicroSquisher (CellScale, Waterloo, Canada) to calculate absolute tissue diastolic and active forces. Twitch forces were calculated as the difference between diastolic and active forces. Stresses were derived from forces normalized to the calculated cross-sectional tissue area (an ellipse of 5:3 aspect ratio at the tissue midpoint between polymeric wires based on existing cross-sectional imaging³²). Stress values were first analyzed using a three-way repeated measures ANOVA (Medium x Stimulation x Time) to assess interactions between the effects of medium versus stimulation on systolic stress. If the three-way interaction analyses failed to find significance, they were repeated by two-way repeated measures ANOVA combined with a within-timepoints Tukey’s HSD, as a post-hoc test. Sphericity was not assumed, and the Geisser-Greenhouse correction was applied when indicated.

Transmission electron microscopy

Microtissues were rinsed 3x with PBS before fixation in 4% paraformaldehyde and 1% glutaraldehyde in PBS. Secondary fixation was in 1% OsO₄ in 0.1 mol L⁻¹ sodium cacodylate buffer. Specimens were dried in a series of 50%, 70%, 90%, and 3 × 100% EtOH, then infiltrated in 50% and 70% (2 h steps) and 100% (overnight) Spurr resin. Samples were sectioned at 70 nm, stained with uranyl acetate and lead citrate, and imaged with a T20 transmission electron microscope (Thermo) and a 4k CCD (Gatan).

¹³C metabolic flux analysis

Cells were incubated for 24 h in one of three substrate-labeled culture media. The base medium consisted of glucose-free DMEM, containing 0.1% w/v Fraction V bovine serum albumin, 50 µmol L⁻¹ carnitine, and 1X Insulin-Transferrin-Selenium supplement (Gibco). Medium also contained 10 mmol L⁻¹ glucose, 3 mmol L⁻¹ lactate, and 120 µmol L⁻¹ palmitate; in any well, one of these three carbon sources was U-¹³C labeled while the remaining two were untagged. After incubation, cells were quickly washed twice in PBS and extracted with a 2:2:1 (v:v:v) mixture of acetonitrile, methanol, and water, all mass spectrometry grade and pre-cooled to −80 °C. Cells were scraped from plates and the slurry kept at −80 °C until analysis. Analysis was performed by LC-MS by The Metabolomics Innovation Center at McGill University according to previously-established methods⁸², and isotopomer relative balances were measured for metabolites of interest with each label.

Protein extraction and global proteomic profiling

Cells were washed 3x in PBS, scraped from 12-well plates, and centrifuged 5 min at 200 x g. Pellets were flash-frozen and stored at −80 °C until extraction. Protein extraction was in 8 M urea with 5 × 3 s sonication bursts at 30% power on ice. Protein concentrations were assessed by BCA assay (Pierce). Samples were reduced in 2.5 mmol L⁻¹ dithiothreitol (DTT) for 1 h at 37 °C, before alkylation in 5 mmol L⁻¹ iodoacetamide for 30 min at RT in the dark. Finally, samples were diluted 1:10 in 50 mM ammonium bicarbonate before the addition of 1:50 protein:protein by mass of sequencing-grade trypsin (Promega) for overnight digestion at 37 °C, followed by the terminating addition of formic acid at a final concentration of 1%. Mass spectrometry and analysis of proteomic samples were performed on three independent samples for each media treatment as previously described⁸³, detailed below.

Chromatography, DDA mass spectrometry, search and data processing methods

For each sample, 1 μg of total protein equivalent was loaded on C18 tips (Evosep, EV-2001) using manufacturer instructions and subjected to in-line chromatography using the Evosep One (EV-1000) instrument 15SPD LC protocol. This involved an 88-min gradient using the Endurance 15 cm analytical column (EV-1106) and a stainless-steel emitter (EV-1086), with a spray voltage of 1.9 kV. For DDA MS data acquisition, the Q Exactive Plus instrument (Thermo) was used, with each cycle involving one full MS scan (at 70,000 resolution) followed by the top 15 selected MS/MS scans at 17,500 resolution. The full MS scan range was set to 400–1500 m/z. The AGC target was set to 1e6 and 40 ms maximum injection time for full MS scans, and 1e5 AGC target and 50 ms maximum injection time for MS/MS scans. MS/MS scans were performed with a normalized collision energy of 27 and an isolation window of 1.6 m/z in centroid mode with an acquisition mass range between 200 and 2000 m/z. Generated RAW files were searched using the MaxQuant (v2.0.3.0) search engine. The data were searched using the canonical and isoform protein sequence FASTA files from UNIPROT (UP000005640_9606, February 2021 release). Three missed cleavage events and variable modifications for methionine oxidation and N-terminal acetylation were allowed, with carbamidomethylation of cysteines set as a fixed modification. The output was filtered based on a 0.01 precursor FDR, with the ‘match between runs (MBR)’ feature enabled. The remaining search parameters were left at default settings and are in the search data and RAW mass spectrometry files deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD036639. Search results were processed using Perseus version 1.6.15.0. LFQ values were used for quantitative analyses. Principal component analysis (PCA) and head-to-head comparisons were performed on the set of 1869 proteins that had valid (greater than 0 LFQ intensity) values in all three samples from at least one experimental treatment with data imputed from a normal distribution (width = 0.3, downshift = 1.8).

RNA sequencing

Sample preparation

Total RNA was extracted from hiPSC-CM monolayers using a Qiagen RNeasy Micro kit and prepared for sequencing at 100 ng with a Stranded Total RNA Prep kit (Illumina Inc., USA), and bulk-sequenced at the Princess Margaret Genomics Center (Toronto, Ontario) using 100 bp paired-end reads at a depth of 40 million reads. Biowire II tissues were extracted using a PicoPure™ kit (Thermo-Fisher), prepared with a SMARTer Stranded Total RNA-Seq Kit v3—Pico Input Mammalian kit (Takara Bio Inc., Japan) with 0.75 ng and bulk-sequenced using 100 bp paired-end reads at a depth of 50 million reads. Reads were aligned to the human reference transcriptome assembly (GR.Ch38) using Salmon⁸⁴.

Data analysis

All RNAseq data analyses were performed in R. Gene-level read counts were obtained using the Tximeta⁸⁵ package. For the meta-analysis, ComBat-seq⁸⁶ was used to correct for batch effects introduced by the inclusion of samples from different publications. For principal component and T-distributed stochastic neighbor embedding (tSNE) analyses, the raw count matrix was adjusted using the variance stabilizing transformation implemented in the DESeq2 package. PCA and PC-associated GO term enrichment were performed using the pcaExplorer⁸⁷ package. t-SNE analysis was performed using the Rtsne package, and clusters were determined from tSNE analysis using the Louvain method⁸⁸, implemented in the bluster package. Differential expression was determined using the DESeq2⁸⁹ package, and Benjamin-Hochberg adjusted p-values < 0.05 were considered significant. Gene set enrichment analysis (GSEA) was performed using the GAGE package⁹⁰, and terms with Benjamini-Hochberg adjusted p-values < 0.05 were considered significant. GO term similarity was determined using the kappa similarity index, and clusters were identified by hierarchical clustering. Pathway enrichment analysis was performed using the GAGE package⁹⁰, and pathway diagrams were generated using Pathview⁹¹. All plots were produced using the ggplot2 and ComplexHeatmap packages.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

N.I.C. and C.A.S. conceived the study. N.I.C., M.M.K., and J.A. contributed to the iterative optimization workflow. EYW and KW assisted with microtissue method development. N.I.C., L.J.D., W.C., U.K., M.Z.M., Y.D., Z.M., C.R., R.A.G., and R.G.I. collected and analyzed data. J.P.S., A.O.G., F.B., M.R., S.M., J.E., P.H.B., and C.A.S supervised the study. N.I.C., L.J.D., W.C., and R.A.G. prepared display items and drafted the manuscript. All authors edited the manuscript and approved the final version.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

The data generated in this study are provided in the Supplementary Information/Source Data file. Raw and processed RNAseq data have been deposited in the NCBI Gene Expression Omnibus under the GEO Series accession number GSE214617. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE⁹² partner repository with the dataset identifier PXD036639. Source data are provided with this paper.

Code availability

The HD-DE code is available at 10.5281/zenodo.18664141.

Competing interests

NIC and CAS have assigned their interest in the intellectual property associated with the C16 formulation and its applications to the University of Toronto, which has filed a patent application (US 63/280,388; Maturation Medium for Pluripotent Stem Cell-derived Cardiomyocytes). This IP has been licensed to Censo Biotechnologies Ltd. T/A Axol Bioscience (Cambridge, UK). All other authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-70550-9.