High-throughput screening of tyrosine kinase inhibitor cardiotoxicity with human induced pluripotent stem cells

See allHide authors and affiliations

Science Translational Medicine  15 Feb 2017:
Vol. 9, Issue 377, eaaf2584
DOI: 10.1126/scitranslmed.aaf2584

Failing fast for tyrosine kinase inhibitors

Discovery early in its life cycle that an anticancer drug causes heart damage (a common side effect) can halt development—saving money, time, and perhaps lives. To this end, Sharma and colleagues derived heart cells from human induced pluripotent stem cells and then examined how a battery of anticancer tyrosine kinase inhibitors altered their physiology. By measuring cell death, contraction, excitability, calcium dynamics, and signal transduction and integrating the results, they calculated a drug-specific “cardiac safety index.” This index proved highly informative, with low values corresponding to those drugs known to cause heart problems in patients. The analysis even revealed that VEGFR2-inhibiting drugs caused cells to try to compensate for the toxic effects by up-regulating protective insulin/IGF pathways, prompting the authors to devise a combination treatment that may limit the toxicity of this class of drug. This screening method is expected to reveal early on whether potential anticancer drugs are cardiotoxic.


Tyrosine kinase inhibitors (TKIs), despite their efficacy as anticancer therapeutics, are associated with cardiovascular side effects ranging from induced arrhythmias to heart failure. We used human induced pluripotent stem cell–derived cardiomyocytes (hiPSC-CMs), generated from 11 healthy individuals and 2 patients receiving cancer treatment, to screen U.S. Food and Drug Administration–approved TKIs for cardiotoxicities by measuring alterations in cardiomyocyte viability, contractility, electrophysiology, calcium handling, and signaling. With these data, we generated a “cardiac safety index” to reflect the cardiotoxicities of existing TKIs. TKIs with low cardiac safety indices exhibit cardiotoxicity in patients. We also derived endothelial cells (hiPSC-ECs) and cardiac fibroblasts (hiPSC-CFs) to examine cell type–specific cardiotoxicities. Using high-throughput screening, we determined that vascular endothelial growth factor receptor 2 (VEGFR2)/platelet-derived growth factor receptor (PDGFR)–inhibiting TKIs caused cardiotoxicity in hiPSC-CMs, hiPSC-ECs, and hiPSC-CFs. With phosphoprotein analysis, we determined that VEGFR2/PDGFR-inhibiting TKIs led to a compensatory increase in cardioprotective insulin and insulin-like growth factor (IGF) signaling in hiPSC-CMs. Up-regulating cardioprotective signaling with exogenous insulin or IGF1 improved hiPSC-CM viability during cotreatment with cardiotoxic VEGFR2/PDGFR-inhibiting TKIs. Thus, hiPSC-CMs can be used to screen for cardiovascular toxicities associated with anticancer TKIs, and the results correlate with clinical phenotypes. This approach provides unexpected insights, as illustrated by our finding that toxicity can be alleviated via cardioprotective insulin/IGF signaling.


Small-molecule tyrosine kinase inhibitors (TKIs) have markedly improved life expectancy for cancer patients (1). Since the U.S. Food and Drug Administration (FDA) approval of imatinib for treating chronic myeloid leukemia, dozens of TKIs have been developed. TKIs inhibit the phosphorylation activity of hyperactive receptor tyrosine kinases (RTKs) in cancer cells, stymieing enhanced cell survival, proliferation, and migration phenotypes associated with cancer progression. However, some TKIs are linked to severe cardiotoxicities including heart failure, reduced left ventricular ejection fraction, myocardial infarction, or arrhythmias (2, 3). Given these life-threatening complications, new approaches are needed to assess the cardiotoxicity of anticancer drugs.

Preclinical platforms for evaluating drug cardiotoxicity use animal models, which inaccurately predict human cardiac pathophysiology because of interspecies differences in cardiac structure, electrophysiology, and genetics (4). In vitro drug cardiotoxicity assessments also use nonhuman cells transfected with the human ether-à-go-go–related gene (hERG), which encodes a cardiac potassium channel, to evaluate drug-induced alterations in cardiac electrophysiology (5). Primary human cardiomyocytes, which are ideal for assessing drug cardiotoxicities, are difficult to procure and maintain (6). Because primary human cardiomyocytes are terminally differentiated, it is impossible to obtain sufficient quantities for cardiotoxicity screening. Human induced pluripotent stem cells (hiPSCs), however, provide an alternative (7). Human cardiomyocytes can be mass-produced from hiPSCs with chemically defined differentiation (8). Patient-specific hiPSC-derived cardiomyocytes (hiPSC-CMs) recapitulate cardiovascular disease phenotypes for dilated cardiomyopathy, hypertrophic cardiomyopathy, left ventricular noncompaction, long QT syndrome, viral cardiomyopathy, and others (914).

Here, we used patient-specific hiPSC-CMs, hiPSC-derived endothelial cells (hiPSC-ECs), and hiPSC-derived cardiac fibroblasts (hiPSC-CFs) from 11 healthy individuals and 2 cancer patients receiving TKIs to evaluate the cardiotoxicities of 21 FDA-approved TKIs. We also used cytotoxicity and high-throughput cell contractility assessments to establish a TKI “cardiac safety index.”


Expression of cardiomyocyte markers and RTKs in hiPSC-CMs

Eleven hiPSC lines were produced from the somatic tissues of 11 healthy individuals by cellular reprogramming with lentivirus or Sendai virus–based vectors expressing the transcription factors OCT4, SOX2, KLF4, and MYC. These individuals were a diverse group of males and females of various ages. Two additional hiPSC lines were created from two individuals receiving TKIs for cancer treatment (fig. S1A). All hiPSC lines expressed pluripotency markers (fig. S1B). hiPSC-CMs were produced with a chemically defined differentiation protocol (Fig. 1A). The hiPSC-CMs expressed standard cardiomyocyte markers (Fig. 1B) (8). Cardiomyocytes exhibited spontaneous beating and were purified for downstream assays (movie S1). Five healthy control hiPSC-CM lines were chosen for RTK expression analysis; all exhibited near-identical RTK expression (Fig. 1C). KDR encoding vascular endothelial growth factor receptor 2 (VEGFR2), PDGFRA encoding platelet-derived growth factor receptor α (PDGFRα), INSR encoding insulin receptor, and IGF1R encoding insulin-like growth factor 1 (IGF1) receptor were highly expressed.

Fig. 1. hiPSC-CMs exhibit sarcomeric proteins and express human RTK families.

(A) Diagram of study workflow. Somatic tissue samples were obtained from 13 individuals and reprogrammed into hiPSC colonies with either Sendai virus or lentivirus vectors expressing the transcription factors OCT4, SOX2, KLF4, and MYC (OKSM). hiPSCs were differentiated into hiPSC-CMs, hiPSC-ECs, and hiPSC-CFs. Purified cardiomyocytes were treated with TKIs and examined for alterations in cell viability, contractility, cellular signaling, and gene expression. (B) Confocal microscopic immunofluorescence images of differentiated hiPSC-CMs expressing the sarcomeric markers cardiac troponin T (TNNT2) and α-actinin (ACTN2). DAPI, 4′,6-diamidino-2-phenylindole. (C) Day 30 hiPSC-CMs from five healthy control lines expressing major RTKs including INSR, IGF1R, PDGFRA, and KDR. High TNNT2 and low PECAM1 expression indicates a pure hiPSC-CM population devoid of ECs. n = 3 biological replicates conducted for gene expression analysis in each hiPSC-CM line. Data are means ± SEM. RPKM, reads per kilobase per million.

High-throughput analysis of TKI-induced cytotoxicity and contractility in hiPSC-CMs

Twenty-one small-molecule TKIs were used for a high-throughput cardiotoxicity screen in hiPSC-CMs (table S1). Many TKIs inhibit multiple RTK families and induce cardiotoxicities including left ventricular dysfunction, myocardial infarction, and arrhythmias. However, the net benefit with respect to cancer treatment outweighs these risks, and these drugs are frequently prescribed at major cancer treatment centers (table S2). We included the highly cardiotoxic anthracycline doxorubicin as a positive control for toxicity. Using the PrestoBlue cell viability assay, we found that the VEGFR2/PDGFR-inhibiting TKIs sorafenib, regorafenib, and ponatinib induced the most cell death in hiPSC-CMs, with median lethal dose (LD50) values of 3.4, 7.1, and 4.3 μM, respectively (Fig. 2A). Doxorubicin was extremely cytotoxic to hiPSC-CMs at an LD50 of 0.78 μM. TKIs not strongly associated with cytotoxicity, such as imatinib or erlotinib, had LD50 values of 78.20 and 87.60 μM, respectively. Sorafenib, regorafenib, and ponatinib were highly cytotoxic in all 11 healthy control hiPSC-CM lines, as measured with quantitative and qualitative viability assays (figs. S2, A to C, and S3). We also performed cytotoxicity assays in hiPSC-CMs and hiPSC-ECs derived from two individuals with kidney cancer (fig. S4). These individuals received two TKIs each: sunitinib as first-line treatment and axitinib as second-line treatment. These patients experienced no significant clinical cardiotoxicity from either agent. As expected, we did not observe a significant difference in cytotoxicity between TKI-receiving patient hiPSC-CMs and healthy control hiPSC-CM lines after subjecting them to sunitinib or axitinib.

Fig. 2. High-throughput analysis of TKI toxicity in purified hiPSC-CMs allows for the development of a TKI cardiac safety index.

(A) Dose-response curves quantifying cytotoxicity after a 72-hour TKI treatment of five healthy control hiPSC-CM lines using a PrestoBlue viability assay. n = 5 biological replicates conducted per line. Data are means ± SEM. (B) Evaluation of hiPSC-CM contractility after a 72-hour TKI treatment with the IC200 Kinetic Imaging Cytometer. Average results from triplicate wells shown at each concentration. Red indicates decreased contraction rate, whereas green indicates increased contraction rate. (C) Values gathered from cytotoxicity and contractility analyses in hiPSC-CMs. Green shading indicates values associated with less cardiotoxicity. Red shading indicates values associated with higher cardiotoxicity. Cessation of beating is the concentration at which >50% of triplicate wells ceased beating. Effective concentration is the concentration at which a significant alteration in all listed contractility parameters was detected (see fig. S7 and Materials and Methods for details). Amplitude of effect is the degree to which all listed contractility parameters were altered at the effective concentration (see Materials and Methods for details). LD50 is the TKI concentration at which a 50% loss in viability is observed from viability assays, averaged across patient hiPSC-CM lines. Patient Cmax represents the maximum TKI blood plasma concentration experienced by patients reported in FDA literature. The cardiac safety index is a value from 0 to 1 that normalizes contractility and viability parameters to patient Cmax and combines these parameters to provide a relative metric for TKI cardiotoxicity. Highlighted drugs (surrounded by a red rectangle) have a safety index at or below 0.10, our threshold for highly cardiotoxic compounds. Clinically reported cardiotoxicities are alterations in patient cardiac function (see table S1). QT, QT interval prolongation; Hy, hypertension; LV, left ventricular ejection fraction decrease; HF, heart failure; MI, myocardial infarction; TdP, Torsades de pointes; SCD, sudden cardiac death; Brady, bradycardia; PE, pericardial effusion; Vas, vascular abnormalities; Afib, atrial fibrillation; **cardiovascular toxicity–associated boxed warning; #noncardiovascular toxicity–associated boxed warning.

To avoid lab-to-lab biases and variations in hiPSC-CM quality, we performed contractility assessment in CMs derived from commercially available, healthy control hiPSCs. We observed alterations in hiPSC-CM beating rate and other parameters at doses lower than the LD50 cytotoxicity values after treatment with multiple TKIs such as nilotinib and vandetanib, suggesting that irregular beating arises before cardiomyocyte death (Fig. 2B and fig. S5). We also determined hiPSC-CM contractility parameters in response to increasing TKI concentrations, effective drug concentrations at which contractility alterations initially appeared, and TKI concentrations at which hiPSC-CM contraction ceased (Fig. 2C and figs. S6 and S7). To accurately assess TKI toxicity, we investigated whether toxic TKI concentrations observed in cytotoxicity and contractility assays matched doses experienced by patients. We obtained patient Cmax values from FDA literature, providing an estimate of maximum TKI blood plasma concentrations in patients (Fig. 2C). By normalizing our in vitro data on cessation of beating, effective concentration, and LD50 cytotoxicity values to literature-reported Cmax values, we developed a cardiac safety index, a metric that identifies clinically cardiotoxic TKIs (see Materials and Methods, Fig. 2C, and fig. S7 for details).

Three of seven compounds with cardiac safety indices at or below 0.10 (doxorubicin, nilotinib, and vandetanib) were previously labeled with FDA black box cardiotoxicity warnings. A safety index value of 0.10 was chosen as our threshold for highly cardiotoxic drugs because it marked a separation in the safety index between clinically cardiotoxic, black boxed drugs (doxorubicin, nilotinib, and vandetanib) and other compounds not commonly associated with cardiotoxicity. Nilotinib and vandetanib, which cause QT interval prolongation and arrhythmias, were selected for further analysis. Three of the TKIs with safety indices under 0.10 were VEGFR2/PDGFR-inhibiting TKIs (regorafenib, sorafenib, and vandetanib). Regorafenib and sorafenib had safety indices comparable to that of the cardiotoxic anthracycline doxorubicin. Thus, VEGFR2/PDGFR-inhibiting TKIs induced cardiotoxicities in hiPSC-CMs at clinically relevant concentrations comparable to the doses that patients experience. In patients, VEGFR2/PDGFR-inhibiting TKIs cause various toxicities including hypertension, heart failure, and QT interval prolongation (15).

Confirmation of toxicity for the known QT interval–prolonging TKIs nilotinib and vandetanib in hiPSC-CMs

QT interval prolongation remains a major concern during drug development (16). Because nilotinib and vandetanib cause dangerous QT interval prolongation and arrhythmias clinically, we conducted additional contractility, calcium imaging, and electrophysiological analyses in four healthy control hiPSC-CM lines treated with nilotinib or vandetanib (fig. S8). We selected dimethyl sulfoxide (DMSO) and axitinib, which are not associated with contractility abnormalities at clinically relevant doses prescribed to patients (per clinical literature and our previous data), as negative controls for toxicity. We observed a prolongation in cardiomyocyte contraction time after nilotinib or vandetanib treatment at clinically relevant concentrations as low as 3.7 μM (fig. S8). Neither DMSO nor axitinib elicited contraction irregularities at clinically relevant concentrations.

We also conducted calcium imaging of hiPSC-CMs after a 2-hour nilotinib or vandetanib treatment (Fig. 3A). At clinically relevant concentrations (Fig. 2C and table S1), nilotinib and vandetanib prolonged calcium transient duration and decreased beat rate in hiPSC-CMs (Fig. 3, B and C). The electrophysiologically “safe” drugs (DMSO control, imatinib, and axitinib) did not significantly alter calcium transient amplitude, transient duration, or beating rate (Fig. 3C); nilotinib and vandetanib altered cardiomyocyte electrophysiology (Fig. 4). We subjected hiPSC-CMs to an acute TKI treatment up to 10 min or to a longer treatment for 2 hours and recorded cellular electrophysiology with patch clamping (Fig. 4A). With acute treatment, TKIs at clinically relevant concentrations did not alter action potential (AP) duration (Fig. 4B). However, after 2 hours, the QT interval–prolonging TKIs nilotinib and vandetanib significantly prolonged AP duration and decreased cellular beating rate (Fig. 4, C and D). Drugs not known to alter cardiac electrophysiology, such as DMSO, imatinib, and axitinib, did not alter AP duration. These results demonstrate that detrimental arrhythmogenic effects of known QT interval–prolonging TKIs such as nilotinib and vandetanib can be recapitulated in hiPSC-CMs, as assessed by contractility assays, calcium imaging, and patch clamp electrophysiology.

Fig. 3. hiPSC-CMs exhibit alterations in intracellular calcium handling after a 2-hour treatment with known QT interval–prolonging TKIs.

(A) Schematic illustrating TKI treatment regimen for hiPSC-CMs before calcium imaging. (B) Raw line scans of individual hiPSC-CM calcium transients after TKI treatment at indicated clinically relevant concentrations and calcium dye treatment over multiple beats. (C) Quantification of hiPSC-CM calcium imaging parameters after a 2-hour TKI treatment. n = 10 cells recorded for each condition. Data are presented as box-and-whisker plots showing the minimum, first quartile, median, mean, third quartile, and maximum of the data set. Student’s t test indicates significance compared to control (*P < 0.05 and **P < 0.01).

Fig. 4. hiPSC-CMs exhibit alterations in cellular electrophysiology after a 2-hour treatment with known QT interval–prolonging TKIs.

(A) Schematic illustrating setup for acute and 2-hour TKI treatment before AP recording. (B) Representative AP tracings after acute TKI treatment for up to 10 min at clinically relevant concentrations in hiPSC-CMs. (C) Representative AP tracings after a 2-hour TKI treatment at clinically relevant concentrations in hiPSC-CMs. (D) Quantification of hiPSC-CM electrophysiological parameters after a 2-hour TKI treatment. Data are means ± SEM. *P < 0.05, compared to DMSO, Student’s t test. n = 10 cells recorded for each condition.

Analysis of TKI cardiotoxicity in hiPSC-ECs and hiPSC-CFs

We next derived endothelial cells (hiPSC-ECs) and cardiac fibroblasts (hiPSC-CFs) from hiPSCs to determine cell type–specific differences in cardiotoxicity. hiPSC-ECs were produced with a chemically defined differentiation protocol using small-molecule Wnt signaling modulators, fibroblast growth factor (FGF) and VEGF stimulation, and magnetic-activated cell sorting of CD31+/CD144+ populations (fig. S9A). These hiPSC-ECs exhibited standard EC morphologies, markers, and functionality (fig. S9, B to D). As in hiPSC-CMs, sorafenib, regorafenib, and ponatinib were the most cytotoxic TKIs in hiPSC-ECs (fig. S9E). We also developed a custom hiPSC-CF differentiation protocol using small-molecule Wnt signaling modulation, FGF2 and VEGFA stimulation, and negative sorting for ECs (fig. S10A). These hiPSC-CFs were negative for cardiomyocyte markers, expressed mesenchymal and myofibroblast markers, and were morphologically similar to primary cardiac fibroblasts (fig. S10, B to E). These hiPSC-CFs exhibited TKI cytotoxicity profiles similar to those of hiPSC-CMs and hiPSC-ECs, with sorafenib, regorafenib, and ponatinib eliciting the highest cytotoxicities (fig. S10F). We next treated undifferentiated hiPSCs with our TKI panel to determine whether noncardiovascular cell types exhibit toxicities similar to hiPSC-CMs, hiPSC-ECs, and hiPSC-CFs (fig. S11). hiPSCs exhibited a unique TKI cytotoxicity profile, showing higher toxicity from VEGFR2/PDGFR dual inhibitors than did cardiovascular derivatives (hiPSC-CMs, hiPSC-ECs, and hiPSC-CFs). For example, axitinib, the least cytotoxic TKI in hiPSC-CMs, was extremely toxic to hiPSCs. Doxorubicin was also substantially more toxic in hiPSCs than in cardiovascular cell types. These results suggest that the VEGFR2/PDGFR-inhibiting TKIs sorafenib, regorafenib, and ponatinib exhibit cell type–specific cytotoxicities that differ between cardiovascular and noncardiovascular cell types.

Evaluation of RTK phosphorylation status in hiPSC-CMs after TKI treatment

To elucidate TKI-induced signaling alterations, we used an RTK proteome profiler to assess RTK phosphorylation after treatment with VEGFR2/PDGFR-inhibiting TKIs (Fig. 5 and fig. S12). Drugs were added to hiPSC-CMs at subcytotoxic concentrations. We observed a dose-dependent decline in VEGFR2 and PDGFRα phosphorylation after VEGFR2/PDGFR-inhibiting TKI treatment, with axitinib eliciting the strongest dual inhibition, suggesting that these TKIs can inhibit functionally relevant signaling pathways in hiPSC-CMs. ErbB2, ErbB4, and epidermal growth factor receptor 2 (EGFR2) phosphorylation remained constant over increasing TKI concentrations. Cabozantinib, a known Axl inhibitor, decreased Axl phosphorylation. Notably, we observed increased INSR and IGF1R phosphorylation after treatment with ponatinib and axitinib, suggesting a compensatory augmentation in insulin/IGF signaling during VEGFR2/PDGFR inhibition.

Fig. 5. Treatment with VEGFR2/PDGFR-inhibiting TKIs causes dose-dependent alterations in RTK signaling in hiPSC-CMs.

Normalized quantification of RTK phosphorylation in purified hiPSC-CMs treated with 0 to 1 μM of the VEGFR2/PDGFR-inhibiting TKIs sorafenib, cabozantinib, ponatinib, axitinib, regorafenib, or sunitinib for 72 hours. Phosphorylation array blots are shown in fig. S12. n = 3 biological replicates conducted. Data are means ± SEM.

Evaluation of insulin- and IGF1-mediated compensatory cardioprotection during TKI treatment

Treatment with insulin or IGF1 can enhance cardiac function during adverse events (17, 18). Given that insulin/IGF signaling was up-regulated after treatment with VEGFR2/PDGFR-inhibiting TKIs, we hypothesized that this compensatory up-regulation protects hiPSC-CMs from TKI toxicity. To determine whether exogenous insulin or IGF1 could enhance cardioprotective signaling in hiPSC-CMs, we used a high-throughput kinase phosphorylation array (Fig. 6A and fig. S13). Both IGF1 and insulin enhanced phosphorylation of the antiapoptotic Akt protein network. Cell survival was enhanced when hiPSC-CMs exposed to ponatinib were concurrently treated with IGF1 or insulin (Fig. 6B). This observation was confirmed quantitatively with CellTiter-Glo viability assays (Fig. 6C). Additionally, we observed that IGF1 and insulin treatment rescued hiPSC-CMs from doxorubicin cytotoxicity (fig. S14). To confirm that the effect of insulin/IGF1 was a result of enhanced cardiomyocyte survival rather than proliferation, we assessed the cell number at early time points after TKI treatment. We observed an increase in viability merely 12 hours after TKI treatment, confirming that insulin and IGF augment cardiomyocyte survival (fig. S15). We next evaluated the gene expression response in hiPSC-CMs during treatment with the VEGFR2/PDGFR-inhibiting TKIs sorafenib, regorafenib, and ponatinib and observed an increase in growth factor receptor gene expression (Fig. 7). NRP2, encoding for the noncanonical VEGFR neuropilin 2, was up-regulated in our microarray. We subsequently conducted RNA sequencing (RNA-seq) analysis of five healthy control hiPSC-CM lines treated with 1 μM VEGFR2/PDGFR-inhibiting TKI sorafenib for 72 hours and observed increased expression of KDR, encoding for the VEGFR2 receptor, and VEGFC, encoding for the VEGFC ligand (fig. S16). These gene expression analyses suggest that noncanonical VEGF-binding receptors and VEGF signaling pathway members are up-regulated to compensate for losing canonical VEGFR signaling after treatment with VEGFR2/PDGFR-inhibiting TKIs. A summary of the compensatory cardioprotective signaling model is shown in Fig. 8. Together, our data suggest that VEGFR2/PDGFR-inhibiting TKIs elicit a compensatory increase in cardioprotective insulin/IGF1 signaling (phosphorylation) in hiPSC-CMs. This cardioprotective signaling can be harnessed with exogenous insulin/IGF1 ligands to enhance cardiomyocyte survival. Simultaneously, VEGFR2/PDGFR-inhibiting TKIs increase the downstream gene expression of VEGF pathway members to compensate for an upstream loss in VEGF signal transduction.

Fig. 6. Insulin and IGF1 activate cardioprotective signaling pathways and alleviate cytotoxicity in hiPSC-CMs.

(A) Phosphorylation arrays demonstrating alterations in hiPSC-CM kinase activity after a 12-hour IGF1 or insulin treatment. n = 3 biological replicate phosphorylation arrays conducted. Data are means ± SEM. *P < 0.05, Student’s t test. We observed a significant increase in phosphorylation of the following protein amino acid residues after IGF1 treatment (P values listed): Akt1/2/3-S473 (0.01), Akt1/2/3-T308 (0.04), GSK3α/β-S21/S9 (0.03), p53-S15 (0.002), p53-S392 (0.02), p53-S46 (0.003), PRAS40-T246 (0.001), TOR-S2448 (0.0001), and WNK1-T60 (0.002). We observed a significant increase in phosphorylation of the following protein amino acid residues after insulin treatment (P values listed): Akt1/2/3-S473 (0.01), Akt1/2/3-T308 (0.04), GSK3α/β-S21/S9 (0.005), p53-S46 (0.04), PRAS40-T246 (0.007), TOR-S2448 (0.01), and WNK1-T60 (0.006). (B) Immunofluorescence of hiPSC-CMs treated with sorafenib, regorafenib, or ponatinib at increasing concentrations for 72 hours in the presence of IGF1 or insulin. Calcein-AM stains viable cells. (C) CellTiter-Glo quantification of hiPSC-CM viability with or without insulin/IGF1 cotreatment during TKI treatment. n = 5 biological replicates conducted. IGF and insulin treatment significantly rescued ponatinib toxicity (P = 0.004). Data are means ± SEM.

Fig. 7. Treatment with sorafenib, regorafenib, or ponatinib leads to hyperactivation of compensatory signaling through noncanonical VEGFRs.

(A) Microarray heat map illustrating differentially expressed genes after a 72-hour sorafenib, regorafenib, and ponatinib treatment in hiPSC-CMs. Cells were treated with 1 μM TKI to avoid cytotoxicity at higher doses. Red indicates high gene expression, and blue indicates low gene expression. (B) Graph represents fold expression change (compared to control) of significantly altered genes after drug treatment. Significantly altered genes defined by P < 0.05 compared to untreated control. Multiple P-value comparisons made using one-way between-subject analysis of variance (ANOVA).

Fig. 8. Model for the activation of compensatory survival signaling in hiPSC-CMs in response to treatment with VEGFR2/PDGFR-inhibiting TKIs.

(A) In hiPSC-CMs, the RTKs VEGFR2, PDGFRα, INSR, and IGF1R are upstream of prosurvival signaling pathways. (B) Our results suggest that VEGFR2/PDGFR-inhibiting TKIs up-regulate INSR and IGF1R signaling (phosphorylation) to compensate for the loss of VEGFR/PDGFR signaling (phosphorylation). This compensatory effect can augment cardiomyocyte survival during TKI treatment via introduction of exogenous insulin and IGF1 ligands. We observed an increase in the downstream gene expression of noncanonical VEGF-binding receptors and VEGFR pathway members, presumably to compensate for VEGFR2/PDGFR-inhibiting TKI-induced loss in upstream VEGFR signaling (phosphorylation).


TKIs are a major class of cancer therapeutics, with revenues from these drugs annually reaching billions of dollars (19). However, many TKIs, like other chemotherapeutics, exhibit substantial cardiotoxicities (3). Our results demonstrate that hiPSC-CMs can assess TKI cardiotoxicity in a high-throughput fashion. We evaluated 21 FDA-approved TKIs using hiPSC-CMs derived from 11 healthy individuals and 2 patients receiving TKIs as cancer therapy. From the data obtained, we developed a cardiac safety index integrating TKI-induced cytotoxicity measurements, contractility assessments, and literature-reported TKI blood plasma concentrations in patients. We also validated the negative effects of known cardiotoxic TKIs.

Previous studies evaluated TKI cardiotoxicity using animals and other in vitro models (20). Sorafenib, one of the three most cytotoxic TKIs in our study, induces cardiomyocyte death and contractility defects in the zebrafish heart and causes ventricular dysfunction and heart failure clinically (table S2) (20). Ponatinib, withdrawn briefly for adverse vascular events, was reported to induce mitochondrial stress and elicit contraction abnormalities in hiPSC-CMs, corroborating our results (21, 22). Although regorafenib has a black box warning for liver toxicity, our study demonstrates that regorafenib can induce cardiotoxicity at clinically relevant doses similar to other VEGFR2/PDGFR-inhibiting TKIs (23). This is reasonable because sorafenib and regorafenib (also known as fluoro-sorafenib) have similar molecular structures (24).

The seven drugs with cardiac safety indices at or below 0.10, our threshold for highly cardiotoxic compounds, cause clinical cardiotoxicities including heart failure, ventricular dysfunction, myocardial infarction, and arrhythmias (table S2). Six of these seven drugs are widely prescribed at major cancer treatment centers at greater than 10,000 doses annually (table S2). Three of the seven compounds with safety indices at or below 0.10 (doxorubicin, vandetanib, and nilotinib) have cardiotoxicity-associated FDA black box warnings, lending validity to our cardiac safety index. We also confirmed the clinical arrhythmia-inducing potential of nilotinib and vandetanib by contractility assay, patch clamp electrophysiology, and calcium imaging in hiPSC-CMs.

Three of the four TKIs with the lowest safety indices (sorafenib, regorafenib, and vandetanib) were VEGFR2/PDGFR-inhibiting TKIs, piquing our interest in further studying this class of drugs. Clinically, these TKIs cause cardiovascular toxicities including hypertension, heart failure, and QT prolongation (15). The VEGFR2/PDGFR-inhibiting TKIs sorafenib, regorafenib, and ponatinib induced the highest levels of cytotoxicity in hiPSC-ECs, hiPSC-CFs, and hiPSC-CMs alike. The high hiPSC-EC toxicity we observed with VEGFR2/PDGFR-inhibiting TKIs such as ponatinib is consistent with the vascular toxicity observed clinically with these TKIs (1). The cytotoxicity profile for undifferentiated hiPSCs was substantially different from that of hiPSC-derived cardiovascular derivatives. For example, the VEGFR2/PDGFR inhibitor axitinib was the least cytotoxic TKI in hiPSC-CMs. However, it was extremely toxic to hiPSCs, perhaps because hiPSCs are highly sensitive to alterations to RTK signaling due to their pluripotent state. Doxorubicin was lethal to hiPSCs even at 0.1 μM, likely because it is a highly effective DNA intercalating agent for killing hyperproliferative cell types such as cancer cells and hiPSCs. The hiPSC-CMs also expressed the major RTK kinase families targeted by TKIs in our panel (INSR, IGF1R, PDGFRα, and VEGFR2), lending validity to our model. However, we did not observe a subject-specific difference in TKI-induced cardiotoxicity between the 11 hiPSC-CM lines made from healthy individuals and the 2 made from TKI-treated patients. This was expected because neither of the individuals receiving TKI treatment developed severe clinical cardiotoxicity from TKI treatment, as would have been seen if any of the subjects exhibited a genetic predisposition to TKI-induced cardiotoxicity that could be recapitulated in vitro.

The VEGF and PDGFR signaling pathways are critical regulators of cardiovascular development. VEGF regulates EC function and promotes cardiomyocyte survival (25, 26). Although nanomolar concentrations of sorafenib, regorafenib, and ponatinib caused complete VEGFR2/PDGFR inhibition, cytotoxicity was observed at micromolar concentrations, suggesting that VEGFR2/PDGFR inhibition may not directly cause hiPSC-CM cytotoxicity. However, VEGFR2/PDGFR inhibition may provide a secondary benefit to hiPSC-CMs. VEGFR2/PDGFR-inhibiting TKIs such as ponatinib and axitinib led to compensatory hyperactivation of cardioprotective insulin/IGF1 signaling. Axitinib, the strongest inhibitor of VEGFR2 and PDGFRα phosphorylation per our kinase assays, induced the strongest compensatory effect along with ponatinib, enhancing INSR/IGF1R phosphorylation twofold. This TKI-induced compensatory effect is fortuitous because insulin and IGF1 are cardioprotective during adverse cardiac events (17, 18). Although VEGFR2/PDGFR-inhibiting TKIs may cause cardiomyocyte cytotoxicity, they may “prime” hiPSC-CMs for stimulation with prosurvival factors.

We harnessed this compensatory effect to enhance hiPSC-CM survival during VEGFR2/PDGFR-inhibiting TKI treatment with either insulin or IGF1, augmenting antiapoptotic Akt signaling. We observed a cardioprotective effect with insulin and IGF1 during ponatinib treatment but not with sorafenib or regorafenib. This may be because, per our RTK phosphorylation arrays, ponatinib elicited the strongest compensatory increase in INSR/IGF1R phosphorylation among these three TKIs. We also observed reduced doxorubicin cytotoxicity after IGF1 and insulin pretreatment, suggesting that these growth factors might alleviate anthracycline cardiotoxicity in cardiomyocytes, corroborating other studies (27). Phosphorylation of RTKs such as ErbB2, ErbB4, EGFR2, and Axl was not significantly altered by VEGFR2/PDGFR-inhibiting TKIs, except by the known Axl inhibitor cabozantinib. Because insulin and IGF1 also induce cell cycle activity, we investigated whether they enhance hiPSC-CM proliferation instead of survival (28). To test for a prosurvival effect, we substantially shortened the TKI treatment but still observed an increase in hiPSC-CM viability after insulin/IGF1 pretreatment. To further safeguard against confounding effects of insulin/IGF1-induced proliferation, we used day 30 postdifferentiation hiPSC-CMs, which exhibit lower mitotic activity than do younger cells (28). Finally, we conducted gene expression analysis to elucidate gene networks altered by TKI treatment. The VEGFR2/PDGFR-inhibiting TKIs sorafenib, regorafenib, and ponatinib induced a twofold increase in the expression of neuropilin 2, a noncanonical VEGFR (29). Thus, hiPSC-CMs may compensate for VEGFR2 signaling loss by up-regulating auxiliary and canonical VEGFR expression, thus facilitating VEGF signaling in the absence of VEGFR2 kinase activity. RNA-seq analysis revealed an up-regulation of KDR (VEGFR2 receptor) and an up-regulation in the expression of VEGF ligands. This suggests that in hiPSC-CMs, the upstream blockade of the phosphorylation cascade by VEGFR2/PDGFR-inhibiting TKIs causes a downstream increase in the gene expression of VEGF pathway members, potentially to compensate for the loss of upstream VEGFR signaling (Fig. 8).

In summary, we used hiPSC-CMs, hiPSC-ECs, and hiPSC-CFs to screen for the cardiotoxicity of 21 FDA-approved TKIs and established a cardiac safety index for TKI toxicity. We also validated the known cardiotoxicities of FDA black boxed TKIs in our hiPSC-CM platform. Although TKIs target various RTK families, we observed that VEGFR2/PDGFR-inhibiting TKIs exhibited high cardiotoxicity on multiple cardiovascular cell types. These cardiotoxicities could be rescued by co-opting compensatory prosurvival insulin/IGF1 signaling pathways. There is substantial interest in deriving hiPSCs from a broader cohort of individuals to establish a population-level, prospective, preclinical toxicity assessment to better inform drug development. As presented here, a combination of electrophysiological, functional, and genetic assays and continued advances in hiPSC biology will enable the development of efficient high-throughput platforms for preemptively screening potential chemotherapeutic compounds for cardiotoxicities.


Study design

To investigate the cardiotoxicity of FDA-approved TKIs in hiPSC-CMs, 11 healthy individuals and 2 individuals receiving TKIs for cancer therapy were recruited and consented. Either skin punch biopsies or blood draws were performed to obtain primary tissue samples for hiPSC production. Reprogramming of either skin fibroblasts or peripheral blood mononuclear cells to hiPSCs was conducted according to previously published protocols (8, 30). A minimum of n = 3 biological replicates were conducted for each experiment, with details in the figure legends.

Chemically defined differentiation of hiPSC-CMs

The hiPSCs were differentiated into hiPSC-CMs using a chemically defined protocol and maintained in medium supplemented with human albumin and ascorbic acid (8). This cardiomyocyte maintenance medium was devoid of growth factors such as insulin and IGF1. Differentiated cells were glucose-starved and supplemented with 5 mM sodium dl-lactate to metabolically select hiPSC-CMs (31). When replating, hiPSC-CMs were dissociated with TrypLE Express (Life Technologies) and reseeded on Matrigel-coated plates.

TKI stocks and cardioprotective growth factor treatments

TKI stocks (LC Laboratories) were resuspended in 10 mM DMSO and stored at −80°C. Insulin (Life Technologies) was stored at −20°C, and Long-R3 IGF1 (Sigma-Aldrich) was stored at 4°C. For cell viability rescue experiments, day 30 to 35 postdifferentiation hiPSC-CMs were pretreated with insulin or IGF1 for 12 hours before TKI treatment, when insulin or IGF1 was resupplemented.

High-throughput imaging and quantitative viability assays

Day 30 to 35 postdifferentiation hiPSC-CMs were plated on Matrigel at 25,000 cells per well of a 384-well plate (Greiner Bio-One). Cells were treated with TKIs at 0 to 100 μM for 72 hours unless otherwise specified. Immunostaining qualitatively assessed cell viability per previous protocols (8). For quantitative viability measurements, cells were treated with CellTiter-Glo Viability Assay (Promega), CCK8 (Dojindo), or PrestoBlue reagent (Life Technologies) per manufacturer-recommended procedures. High-throughput imaging and viability assays were conducted using a Cytation 5 plate reader/imager (BioTek Instruments). Prism (GraphPad) was used for curve fitting, LD50 calculations, and statistical analysis.

High-throughput hiPSC-CM contractility assessment

After a 72-hour TKI treatment, hiPSC-CMs were washed with Tyrode’s solution. Imaging dye was prepared by diluting Hoechst 33258 (H3569, Life Technologies) to 4 μg/ml and wheat germ agglutinin–Alexa Fluor 488 conjugate (W11261, Life Technologies) to 5 μg/ml in Tyrode’s solution. Solution was added to hiPSC-CMs and incubated at 37°C and 5% CO2 for 15 min. After rewashing with Tyrode’s solution, hiPSC-CMs were incubated for 15 min before imaging. The IC200 Kinetic Imaging Cytometer (Vala Sciences) recorded a 6.5-s time series of contracting hiPSC-CMs at 100 Hz at 20× magnification per well of a 384-well plate.

Safety coefficients based on contractility parameters

Safety coefficients presented in this paper are computed as follows. Cessation of beating is the concentration at which >50% of triplicate wells ceased beating. Effective concentration for the ith metric (ECi) is a concentration that presents a statistically significant contractility difference (P < 0.05) from baseline. Amplitude of the effect (AEi) quantifies the magnitude of such departures from baseline as the log2 of the ratio of the metric value EVi at concentration ECi, hereinafter referred to as effective value (EVi), to the value of the metric BVi at baseline concentration. Additionally, i = 1…M, where M is the number of metrics obtained from the contractility analysis (M = 9 in this paper: Tpeak, Trise, Tfall, total contraction time, Dpeak, Dvalley, Dp2v, contraction rate, and relaxation rate).Embedded Image

Average amplitude of effect (AE) is obtained by averaging the different AEi’s.Embedded Image

Average effective concentration (EC) is a weighted average of the different effective concentrations using each respective amplitude of effect as weight.Embedded Image

The rationale for this EC expression is that it considers all different ECs and introduces a bias toward the ECs corresponding to metrics that most prominently alter normal cell behavior.

Calcium imaging

Day 30 to 40 postdifferentiation hiPSC-CMs were reseeded in Matrigel-coated eight-well Lab-Tek II chambers (Nalge Nunc International) and were treated with TKIs for 2 hours. Cells were treated with 5 μM Fluo-4 AM and 0.02% Pluronic F-127 (Molecular Probes) in Tyrode’s solution for 15 min at 37°C and washed with Tyrode’s solution afterward. Ca2+ imaging was conducted using a Zeiss LSM 510 Meta confocal microscope (Carl Zeiss AG) and analyzed using Zen software. Spontaneous Ca2+ transients were obtained at 37°C using a single-cell line scan mode.


Whole-cell APs were recorded with patch clamp technique, as previously described (8). Cultured hiPSC-CMs were dissociated using TrypLE and plated as single cells on glass coverslips coated with Matrigel. Cells were placed in an RC-26C recording chamber (Warner) and mounted onto an inverted microscope (Nikon). The chamber was continuously perfused with warm (35° to 37°C) extracellular solution of the following composition: 150 mM NaC1, 5.4 mM KCl, 1.8 mM CaCl2, 1.0 mM MgCl2, 1.0 mM Na pyruvate, 15 mM Hepes, and 15 mM glucose; pH was adjusted to 7.4 with NaOH. Glass micropipettes (2- to 3-megohm tip resistance) were fabricated from standard wall borosilicate glass capillary tubes (Sutter BF 100-50-10) and filled with the following intracellular solution: 120 mM KCl, 1.0 mM MgCl2, 10 mM Hepes, 10 mM EGTA, and 3 mM Mg-ATP; pH was adjusted to 7.2 with KOH. Single-beating cardiomyocytes were selected, and APs were recorded in whole-cell current clamp mode using an EPC 10 patch clamp amplifier (HEKA). External solution containing 0.1% DMSO (vehicle) was applied to establish the baseline. Then, cells were treated with TKI solution containing imatinib, axitinib, nilotinib, or vandetanib (LC Laboratories). Data were acquired using PATCHMASTER software (HEKA), digitized at 1.0 kHz, and analyzed using FITMASTER (HEKA), Igor Pro (WaveMetrics), and Prism 5 (GraphPad). For recordings on differentiated ventricular-like hiPSC-CMs, the maximum diastolic potential of single cardiomyocytes varied from −70 to −50 mV, action potential amplitude (APA) was greater than 90 mV, and action potential duration (APD)90/APD50 was less than 1.20. Only cardiomyocytes satisfying the aforementioned criteria were ventricular-like cardiomyocytes and selected for assessing the effects of TKIs. Baseline APs were recorded for 3 min before application of the drug and at 3, 5, and 10 min while keeping a continuous perfusion with the drug. In a separate series of experiments, the drugs were added for 2 hours at 37°C before patching. Average responses of n = 10 APs were analyzed per treatment. Significant APD90 prolongation is defined as >10% change in APD90.

Kinase phosphorylation profiling

Phosphorylation of human RTKs and other phosphoproteins was determined using Human Phospho-RTK Array or Human Phospho-Kinase Antibody Array (R&D Systems). Day 30 to 35 postdifferentiation hiPSC-CMs were treated with TKIs for 72 hours and lysed. Lysate was incubated overnight on an RTK or phosphokinase panel and subsequently with an anti–phosphotyrosine–horseradish peroxidase antibody to assess phosphorylation. Blots were developed using Gel Doc XR (Bio-Rad). Phosphorylation intensity was determined using ImageJ software.

Gene expression

RTK expression in hiPSC-CMs was determined using Ion AmpliSeq (Life Technologies). RNA was extracted using the RNeasy Micro Kit (Qiagen). Complementary DNA libraries were synthesized using the Ion AmpliSeq Transcriptome Human Gene Expression Kit. Libraries were added to Ion PI chips and loaded onto an Ion Chef instrument for template preparation. Transcriptome sequencing was conducted on an Ion Proton sequencing system (Life Technologies). For expression analysis of hiPSC-CMs after TKI treatment, a GeneChip Human Gene 1.0 ST DNA microarray was used (Affymetrix).

Statistical analysis

Data are presented as means ± SEM unless otherwise specified. Comparisons are conducted via Student’s t test, unless otherwise specified, with significant differences defined by *P < 0.05 or **P < 0.01. For microarray experiments, multiple P-value comparisons were made using a one-way between-subject ANOVA (P < 0.05) and Affymetrix Transcriptome Analysis Console 2.0 software.


Materials and Methods

Fig. S1. hiPSCs exhibit characteristic morphologies and markers of pluripotent stem cells.

Fig. S2. Quantitative and qualitative cell viability assays illustrate sorafenib, regorafenib, and ponatinib cytotoxicity in hiPSC-CMs.

Fig. S3. Quantitative cell viability assays on additional hiPSC-CM lines demonstrate VEGFR2/PDGFR-inhibiting TKI toxicity.

Fig. S4. Quantitative cell viability assays in hiPSC-CMs and hiPSC-ECs derived from patients receiving TKI treatment.

Fig. S5. Commercially available, healthy control hiPSC-CMs exhibit alterations in cellular contractility after a 72-hour TKI treatment.

Fig. S6. Heat maps of high-throughput contractility analysis on commercially available, healthy control hiPSC-CMs treated with TKIs.

Fig. S7. Extended calculations for TKI safety index after a 72-hour TKI treatment on commercially available, healthy control hiPSC-CMs.

Fig. S8. hiPSC-CMs exhibit alterations in cellular contractility after a 72-hour treatment with known QT interval–prolonging TKIs.

Fig. S9. hiPSC-ECs exhibit EC characteristics and demonstrate cytotoxicity in response to TKI treatment.

Fig. S10. hiPSC-CFs exhibit properties of adult cardiac fibroblasts and demonstrate cytotoxicity in response to TKI treatment.

Fig. S11. hiPSCs demonstrate a TKI cytotoxicity profile that is unique from those of hiPSC-CMs, hiPSC-ECs, and hiPSC-CFs.

Fig. S12. VEGFR2/PDGFR-inhibiting TKI treatment in hiPSC-CMs results in activation of compensatory insulin/IGF1 signaling.

Fig. S13. IGF1 and insulin treatment activates cardioprotective Akt signaling in hiPSC-CMs.

Fig. S14. IGF1 and insulin treatment rescues doxorubicin toxicity in hiPSC-CMs.

Fig. S15. IGF1 and insulin treatment rescues ponatinib toxicity at early time points in hiPSC-CMs.

Fig. S16. RNA-seq of hiPSC-CMs treated with the VEGFR2/PDGFR-inhibiting TKI sorafenib illustrates compensatory hyperactivation of VEGF signaling.

Table S1. Small-molecule TKIs selected for high-throughput cardiotoxicity screen.

Table S2. Adverse cardiac events associated with small-molecule TKIs selected for high-throughput cardiotoxicity screen.

Movie S1. hiPSC-CMs before purification via glucose deprivation.

References (3250)


Acknowledgments: We thank K. Red-Horse for assistance with confocal imaging. We also thank E. Yeh from the MD Anderson Cancer Center for information regarding TKI usage and cardiotoxicity prevalence. We acknowledge the Stanford High-Throughput Bioscience Center for assistance with high-throughput imaging and plate reader assays. We thank A. Olson from the Stanford Neuroscience Microscopy Service for help with calcium imaging. Funding: We acknowledge support from the American Heart Association Predoctoral Fellowship (13PRE15770000) and NSF Graduate Research Fellowship (DGE-114747) (A.S.); NIH (K99/R00 HL121177) and American Heart Association Beginning Grant-in-Aid (14BGIA20480329) (P.W.B.); NIH Director’s Pioneer Award, American Heart Association Grant-in-Aid, and Endowed Faculty Scholar Award of the Lucile Packard Foundation for Children and Child Health Research Institute at Stanford (S.M.W.); and Burroughs Wellcome Foundation Innovation in Regulatory Science, American Heart Association Established Investigator Award, and NIH (R01 HL132875, R01 HL130020, R01 HL128170, R01 HL123968, and R24 HL117756) (J.C.W.). Author contributions: A.S., P.W.B., and J.C.W. designed the study and participated in data analysis and manuscript writing. A.S. and P.W.B. participated in all experimental work. W.L.M. and R.S. conducted high-throughput contractility assessments. P.S. performed patch clamp electrophysiology. N.S. assisted with deriving hiPSC-CMs and hiPSC-ECs from patients receiving TKI treatment. J.M.C. performed microarray and RNA-seq analysis. T.K. assisted with high-throughput toxicity analysis. H.W. conducted calcium imaging. A.H. conducted hiPSC-EC differentiation and characterization. E.M. assisted with gene expression analysis. Y.Z. performed immunocytochemistry and edited the manuscript. A.K. performed immunocytochemistry and edited the manuscript. A.C.F. assisted with patient recruitment. J.C.d.A., S.M.W., J.J.M., M.M., and J.C.W. assisted with study design and manuscript editing. Competing interests: M.M. holds equity in and is on the scientific advisory board for Vala Sciences, a company offering high-content screening services, particularly instrumentation used for measuring the electrical and contractile physiology of cardiomyocytes, and is on the scientific advisory board of Stem Cell Theranostics, a company that uses patient-specific hiPSC-CMs for drug discovery. J.C.W. is a cofounder and is on the scientific advisory board of Stem Cell Theranostics. Other authors declare that they have no competing interests. Data and materials availability: The gene expression data are found at Gene Expression Omnibus with accession no. GSE8894 and GSE89411.

Stay Connected to Science Translational Medicine

Navigate This Article