Despite advances in treatment, cardiovascular disease is the leading cause of death worldwide1. Globally, about 12% of adults are diagnosed with cardiovascular disease and over 30% of all deaths are caused by cardiovascular disease1. The excessive demand of heart transplantation has outpaced the limited number of healthy and functional heart donors2. Cell-based regenerative therapy provides a promising treatment for patients suffering from cardiac tissue injury3, 4. However, cardiomyocytes (CMs) are terminally differentiated cells with no regenerative capacity5. Hence, cost-effective and time-efficient platforms to generate functional CMs with high quality has emerged as an urgent need for cardiac medicine in drug screening, toxicity testing, disease modeling, and regenerative cell therapy.
Human pluripotent stem cells (hPSCs) can differentiate into cells from all three germ layers6-8. A variety of methods have been established to generate CMs from hPSCs9-11. These hPSC-derived CMs exhibit similar functional phenotypes to their in vivo counterparts11, including self-contractility and action potentials. hPSC-derived CMs have been used in disease modeling12, 13 and drug screening14, and hold great potential for regenerative medicine15, 16. However, batch-to-batch and line-to-line variability in the differentiation process of hPSCs into CMs has impeded the scale-up of CM manufacturing17. For safety, the quality of clinical-graded hPSC-derived CMs must be rigorously evaluated before they can be used for regenerative cell therapy in patients18. Current approaches to quantify CM differentiation rely on low-throughput, labor-intensive, and destructive immunofluorescence labelling and electrophysiological measurements11. New technologies that can non-invasively monitor CM differentiation in real time and evaluate the differentiation outcome at early stages are needed to effectively optimize biomanufacturing of CMs from stem cells.
Previous studies indicate that hPSC-derived CMs undergo dramatic metabolic changes throughout differentiation19. Reduced nicotinamide adenine dinucleotide (phosphate) (NAD(P)H) and oxidized flavin adenine dinucleotide (FAD) are autofluorescent cellular metabolic co-enzymes that can be imaged to collect metabolic information at a single-cell level20. The ratio of NAD(P)H to FAD intensity is the “optical redox ratio”, which reflects the relative oxidation-reduction state of the cell. The fluorescence lifetimes of NAD(P)H and FAD are distinct in the free and protein-bound conformations, so changes in these fluorescence lifetimes reflect changes in protein-binding activity21, 22. Optical metabolic imaging (OMI) quantifies both NAD(P)H and FAD intensity and lifetime variables. Several groups have demonstrated that autofluorescence imaging can non-invasively track stem cell metabolic activities in real time, including monitoring mesenchymal stem cell differentiation into adipocytes23, 24, osteocytes24, 25, and chondrocytes25, distinguishing differentiation of hPSCs into dermal and epidermal lineages26, metabolic difference between hPSCs and feeder cells27, and hematopoietic stem cells at different stages28. These prior studies indicate that OMI is suitable to detect the metabolic changes that occur during CM differentiation.
The goal of this study is to build a predictive model based on OMI to determine whether OMI can predict CM differentiation efficiency early in the differentiation process. Early prediction of CM differentiation outcome can benefit CM manufacturing. We demonstrate a facile method to non-invasively monitor metabolic changes during hPSC differentiation into CMs by combining OMI with quantitative image analysis. OMI is performed at multiple time points during a 12-day differentiation process under varying conditions (cell density, concentration of Wnt signaling activator) and different hPSC lines (human embryonic pluripotent stem cells and human induced pluripotent stem cells). Differentiation efficiency is quantified by flow cytometry with cTnT labelling on day 12. During the differentiation process all 13 OMI variables, including both NAD(P)H and FAD intensity and lifetime variables, change distinctively between low (< 50% cTnT+ on day 12) and high (≥ 50% cTnT+ on day 12) CM differentiation efficiency conditions. Multivariate analysis finds that day 1 cells (24 hours after Wnt activation) form a distinct cluster from cells at other time points. Logistic regression models based on OMI variables from cells at day 1 perform well for separating low and high differentiation efficiency conditions with a model performance at 0.91 (receiver operating characteristic (ROC) area under the curve (AUC)). Compared to previous studies23-28, we specifically contribute a predictive model based on OMI to determine CM differentiation outcome as early as day 1. This label-free and non-destructive method could be used for quality control for CM manufacturing from hPSCs.