Background
Tyrosine kinase inhibitors (TKIs) are widely prescribed in the clinic to treat cancer from blood malignancy to advanced solid tumors. As drugs that specifically target overexpressed or hyperactivated signaling downstream of receptor tyrosine kinases, TKIs are viewed as much safer than traditional chemotherapy, such as doxorubicin. But TKIs are often used for many months without a dosage cap, some of them still cause serious cardiac adverse events. For example, sunitinib is associated with 4.1% of congestive heart failure in a meta-analysis of 6935 patients [
1] and sorafenib causes 2.7–3% of myocardial ischemia in clinical trials [
2,
3]. Both TKIs cause hypertension in up to 47% of patients, which is probably due to inhibition of the VEGF signaling [
4]. In another study, 14% of patients with nilotinib treatment have one or more cardiovascular events, including peripheral artery disease (10%) and myocardial infarction (4%) [
5]. Ponatinib is associated with a significant incidence of arterial occlusion events (26%) [
6] and other adverse cardiac events, including arrhythmia, hypertension, and myocardial infarction [
7]. Those TKIs cause a wide array of cardiovascular toxicities in noticeable fractions of patients; yet clinical management is limited. Drug holiday, dose reduction, or standard anti-heart failure therapies are used depending on the type and severity of cardiotoxicity. To improve life quality and clinical treatment of patients with drug-induced cardiotoxicity, we need to understand molecular mechanisms of the TKI-induced cardiotoxicity.
TKI-induced cardiotoxicity can be classified into two categories: “on-target” and “off-target” effects. In these categories, cardiac cells develop adaptive and maladaptive responses to the pharmacological effects of TKIs. The physiologically inherited responses of the hearts, such as the hypertrophic response, the fetal gene program, the unfolded protein response, and the antioxidant response, are often induced by exogenous chemical insults and may regulate cardiotoxicity adaptively over time. TKI-induced cardiotoxicity is determined by both the pharmacological inhibition of putative targets or off-targets and the stress responses of cardiac cells. By profiling 4 TKIs at three different doses and four time points, we found that transcriptomic changes facilitate identification of cellular stress responses [
8]. Cardiac cells share a similar adaptive or drug-resistant pathway of aerobic glycolysis with cancer cells in response to sorafenib [
8]. A later study expanded the transcriptome profiling to 26 TKIs at a fixed dose and time and found that cardiac cell-based transcriptomic changes in combination with structure–activity relation of TKIs are predictive of drug-induced cardiotoxicity risks [
9]. Therefore, the high-throughput transcriptomics of human cardiomyocytes can improve our understanding on mechanisms of cardiotoxicity caused by TKIs.
One of the most conserved stress responses is the endoplasmic reticulum (ER) stress, which is active in nearly all cell types of human tissues. Changes in the oxidative, calcium, lipid modification, and unfolded protein levels in the ER can trigger this response, which activates three effectors, inositol-requiring enzyme 1α (IRE1α), protein kinase R-like endoplasmic reticulum kinase (PERK) and ATF6 [
10]. IRE1α activates the expression of ER chaperons and ER-associated degradation (ERAD) components to reduce ER stress [
11]. However, it may engage TRAF2 to cause apoptosis [
12] or NF-κB to induce inflammation [
13]. PERK reduces folding load in ER through phosphorylating translation initiation factor 2α (eIF2α), which inhibits mRNA translation [
14]; however, it may induce apoptosis through ATF4 and CHOP (official gene name:
DDIT3) upregulation [
15] or activate IL6 to promote inflammation [
16]. The primary role of ATF6 in the heart is to preserve proteostasis [
17]. Therefore, the three effectors downstream of ER stress have both positive and negative effects on cell fate, dependent on the selective pathways activated, or the duration of ER stress. Genetic modulation of the three effectors in mice hearts show a protective role of ER stress in response to pressure overload or ischemic diseases [
18‐
21]. However, if CHOP downstream of the PERK-ATF4 signaling is activated, this causes aggravated cardiac damage induced by ER stress [
22]. Since cardiac adverse effects induced by TKIs share similar phenotypes as pressure overload or ischemia, ER stress may be activated in TKI-induced cardiotoxicity. Indeed, imatinib-induced cardiotoxicity is associated with PERK and IRE1α activation and nilotinib activates ER stress and cell death in rat cardiac H9C2 cells [
23,
24]. However, which downstream selected pathway(s) or how the activation duration of ER stress regulates cardiotoxicity remains unknown.
To systematically study and compare cardiotoxicity mechanisms of TKIs, we profiled phenotype and transcriptome of human cardiomyocytes in a high throughput manner with over 100 treatments of eight TKIs. The transcriptome analysis not only helps us understand the specific biological processes regulated by the TKIs in cardiomyocytes but also enables comparison among TKIs with similar or different targets. ER stress is one of the highly enriched pathways regulated by three TKIs and how ER stress induced by TKIs promotes cardiotoxicity through inflammation and fetal gene re-expression, rather than cell death, is explored in this study.
Methods
Human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CM) culture and drug treatment
Cor.4U hiPSC-CMs derived from female human cells were ordered from Ncardia (
https://www.ncardia.com) and maintained in the manufacturer’s media. Cells were grown in growth media (Cor.4U Complete Culture Medium which contains 10% FBS and other essential nutrients, Catalog number Ax-M-HC250 from Ncardia) and treated in minimal media (BMCC Serum-Free Culture Medium, Catalog Number Ax-MBMCC250 from Ncardia, supplemented with 1% fetal bovine serum from Thermo Fisher Scientific) in a humidified incubator with 5% CO
2 and 37 °C. Cryo-preserved hiPSC-CMs were thawed into 75-cm
2 flasks, cultured in growth media for 3 days, and reseeded into multi-well plates. Then, hiPSC-CMs were cultured in minimal media for 1.5 days prior to drug treatment and maintained in minimal media during drug treatment with media exchanged every other day. Drug stocks were usually prepared in DMSO at 10 mM and stored at − 20 °C until use.
Neonatal rat cardiac myocytes (NRCM) isolation, culture, and treatment
Dissection of the hearts was performed on 1–2 days old Sprague–Dawley rats (Beijing Vital River Laboratory Animal Technology) and the procedure did not involve the use of anesthetics. Rats were euthanized by decapitation and thoracic cavity was opened with scissors and hearts were popped out of the opening gently with fingers. Hearts were clipped and put in HBSS immediately; 20–40 hearts were collected into one sterile vial containing a magnetic stir bar and 1 mL digestive media made of 0.1% trypsin (Macklin) and 0.05% collagenase (Gibco) in HBSS. Hearts were minced until they were uniform in size. Digestive media was added to 10 mL for 40 hearts and stirred gently for 6 min. Supernatant was discarded and the digestion was repeated once; 10 mL digestive media was added and stirred gently for 10 min. The supernatant with cells were collected and the digestion and collection were repeated 4 times. Cells were centrifuged at 1200 rpm for 5 min, resuspended in media (Dulbecco’s modified Eagleʼs medium (Gibco)) with 10% fetal bovine serum (FBS; VISTECH), 1% penicillin–streptomycin, and 1% 0.1 M 5-bromo-2-deoxyuridine solution (Shanghai Yuanye) and filtered through a 40-μm cell strainer. Cells were seeded on a 100-mm plastic dish and cultured in a 5% CO2 incubator at 37 °C for 2 h to allow non-myocytes to adhere to the plate. The dish was swirled and media gently pipetted up and down to detach lightly attached cardiomyocytes. The collected myocytes were seeded in the same media described before; 48 h after plating, medium was exchanged to fresh DMEM (Gibco) with or without 2% FBS. All drug treatments were performed after this media change.
3’digital gene expression with unique molecular identifiers (3’DGE-UMI) RNAseq and data analysis
Cor.4U hiPSC-CMs were cultured in 96-well plates, precultured as described in “human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CM) culture and drug treatment” and treated with drugs as shown in Fig.
2A. Total RNA was isolated using MagMAX™-96 total RNA isolation kit. RNA was transferred into 384-well plates and the subsequent library preparation was done according to this publication [
25] and using an automated liquid dispensing system. RNA was reverse transcribed using the Maxima H Minus Reverse Transcriptase (Thermo Fisher) and barcoded primers containing poly(T) sequence about 24 bases long, a random 10-nucleotide sequence as unique molecular identifiers, a random 6-nucleotide sequence as well barcodes, and a sequence complimentary to pre-amplification primers. The other end of cDNA sequences was filled with a sequence complementary to pre-amplication primers through a template switching oligo and the same reverse transcriptase. The cDNAs of different drug treatments were pooled and amplified by PCR.
The cDNA library was sequenced twice, once in a Nextseq 500 (illumina) in the Bauer Core of Harvard University with 17 cycles on Read 1 and 60 cycles on Read 2 and another time in a Novaseq 6000 (illumina) using BerryGenomics with 150 cycles on both Reads. Even though the second sequencing yielded double the amounts of counts per sample, the t-distributed Stochastic Neighbor Embedding (tSNE) clustering of sequencing results was very similar to the previous data and we used the later data for analysis. Data were analyzed using R version 4.1.0 and Seurat_4.0.3.
Statistical analyses
All data are presented as mean ± SEM. Data were analyzed using a two-tailed Student’s t-test (for two-grouped comparisons) or ANOVA (for multiple-group comparisons). Significance was assigned at p < 0.05, *p < 0.05, **p < 0.01, ***p < 0.001, n.s. not significant. In all figures, n referred to the sample size which was selected based on previous studies. Unless otherwise indicated, the results were based on a minimum of three independent experiments to ensure reproducibility. Statistical analyses were performed using GraphPad Prism 9.0 software (GraphPad Software Inc., La Jolla, CA, USA). Adobe illustrator 27.1.1 was used to create artwork (Adobe Inc., USA).
Detailed methods were presented in the supporting materials.
Discussion
Many TKIs have transformed cancers into chronic diseases with extended drug therapy; however, TKI-induced cardiotoxicity becomes a significant problem to cause death or decrease life quality. Treatment that reduces cardiac damage caused by TKIs is lacking due to limited understanding of molecular pathology. Many TKIs have multiple targets besides receptor tyrosine kinases and could cause cardiotoxicity through non-signaling mediated effects, such as transcriptomic regulation. Here, we discover enriched biological processes regulated transcriptionally by eight TKIs in hiPSC-CMs as candidates for their cardiotoxic mechanisms and validate that ER stress-induced inflammation is a common mechanism of cardiotoxicity induced by three TKIs in cardiomyocytes. TKI-specific transcriptome changes dominate over dose- and time-dependent effects. Three TKIs, afatinib, sorafenib, and ponatinib, induce ER stress in cardiomyocyte and ponatinib also increases the IRE1α-XBP1s pathway and fetal gene expression in rat left ventricles. Afatinib, sorafenib, and ponatinib induce expression of fetal and pro-inflammatory factors in cardiac myocytes through cooperative activation of PERK and IRE1α. Through validating the critical role of ER stress in regulating cardiotoxicity of many TKIs, our study suggests that transcriptome data can direct us to find important molecular pathology of cardiotoxicity. We also pinpoint specific pathways downstream of ER stress that can serve as potential treatments for TKI-induced cardiotoxicity.
Predicting cardiotoxicity of drugs using in vitro assays remain challenging and researchers try to improve the accuracy of prediction by increasing the number of cell lines or the types of measurements. A traditional assay to evaluate cardiotoxicity is testing the inhibitory effect on the inward rectifying potassium channel (hERG), yet it is criticized for a high false positive rate (36%) [
47] and a non-physiological cell type. The model of hiPSC-CMs is a promising cell type that preserves not only electrophysiology but also metabolic and contractile phenotypes of cardiomyocytes. So, it is proposed to serve as a new platform for in vitro cardiotoxicity screening (see the CiPA initiative [
48]). We use this cell type to do both phenotypic and transcriptomic analysis in the hope to find the most relevant changes to human and help improve the correlation between in vitro data and patient-observed cardiotoxicity. When we rank the toxicity of TKIs based on different cellular assays and patient-derived data, we find that the toxicity of the eight TKIs measured in cellular assays mimics that observed in FAERS or literature. Three out of five drugs with high toxicity, sunitinib, sorafenib, and ponatinib, are shared between FAERS/literature and cellular assays. Additionally, the toxicity of the TKIs on cell viability is similar to that observed in cellular contraction and electrophysiology measurement, which is consistent with a previous study [
49]. Afatinib is ranked as highly cardiotoxic based on mitochondrial and beating assays, with little toxicity found in FAERS or literature; this discrepancy could be due to that the dose of afatinib used in vitro (10 µM) is much higher than its Cmax (0.05 µM) in vivo
. But our results suggest that patients who cannot metabolize afatinib properly or timely may experience unduly cardiotoxicity. Dasatinib is ranked as highly cardiotoxic based on FAERS and literature, but does not affect the functions of cardiomyocytes significantly in vitro, indicating that dasatinib may target other cell types or other organ systems to negatively affect cardiovascular function. Different phenotypic assays put weights on different aspects of cardiomyocyte function, for example, the seahorse assay measures energy metabolism and the microelectrode array measures contractility, so it is hard to combine the results into a single risk factor in a simple way. By evaluating a larger number of drugs with different levels of cardiotoxicity and focusing on specific phenotypes (cardiac arrhythmia as in CiPA or heart failure), we may be able to construct an efficient in vitro assay with appropriate algorithms to consider different phenotypic data in order to predict cardiotoxicity.
Besides phenotypic data, transcriptomic changes can also contribute to cardiotoxicity risk stratification of TKIs. However, current transcriptome data do not stratify TKIs according to toxicity, but based on drug identity. Most drugs elicit distinct transcriptomic changes; only sorafenib and ponatinib or sunitinib and crizotinib share similar transcriptional signatures. While the similarity between sorafenib and ponatinib can be explained by shared targets, that between sunitinib and crizotinib remains unknown. How to utilize the transcriptomic signatures for in vitro cardiotoxicity evaluation requires deeper understanding of the role of these signatures in cardiotoxicity. To explore this, we focus on transcriptional changes of a stress response pathway that is repeatedly identified in dimension reduction and enrichment analyses. Both tSNE and PCA analyses show that afatinib, sorafenib, and ponatinib induce the most distinct transcriptomic changes from the other TKIs and the changes unify on ER stress activation. Afatinib and sorafenib primarily activate PERK and IRE1α, whereas ponatinib activates all three effectors of ER stress in NRCMs. Ponatinib also upregulates the IRE1α-XBP1s pathway in rat hearts, but sorafenib shows opposite regulation of this pathway in vivo versus in vitro. This could be due to different temporal activation of the ER stress, different cross-regulatory mechanisms or many more cell types present in vivo versus in vitro, and needs further clarification. Inhibition of ER stress pathways blocks the induction of fetal or proinflammatory genes by TKIs, suggesting that ER stress promotes cardiotoxicity in cardiomyocytes. Even though all three TKIs activate ER stress, afatinib has a much lower cardiotoxicity reported in clinic than sorafenib or ponatinib. The reason may be that afatinib’s EC50 on cells is ~ 60 folds higher than its
Cmax in vivo. Another possible reason may be that afatinib induces many compensatory pathways to reverse cardiotoxicity, such as the activation of cytosolic chaperons and the Nrf2 pathway [
50,
51], which are not induced by sorafenib or ponatinib in cardiomyocytes.
In searching for effective treatment to reduce cardiotoxicity without affecting the therapeutic effect of drugs, one needs to validate that the mechanism of cardiotoxicity is cardiac specific or have differential effects in the tumor versus in the heart. For example, if ER stress promotes cardiotoxicity but is dispensable for or even inhibits cancer killing by TKIs, this process may be a promising target for developing oncocardiology treatment. To compare the effect of ER stress in cardiomyocytes and cancer cells, we searched literature about TKI-induced ER stress in cancer cells. Afatinib induces the activation of the PERK-eIF2α pathway downstream of ER stress in the head and neck squamous cell carcinoma and sorafenib induces both PERK and IRE1α activation in leukemia and hepatocellular carcinoma cells [
52‐
54]. So ER stress is activated in both cardiomyocytes and cancer cells. Then, what is the role of ER stress in the therapeutic effect of these TKIs? In the case of afatinib, ER stress activation is required for drug-induced apoptosis [
52]. So ER stress inhibition may negatively affect afatinib’s therapeutic effect, thus not desirable for treating the associated cardiotoxicity. In the case of sorafenib, the role of ER stress in drug-induced apoptosis in cancer is still debatable. One study shows that blockade of IRE1α or PERK can enhance apoptosis induced by sorafenib [
53], whereas another study finds that enhancing ER stress increases apoptosis and inhibits tumor growth in vivo in response to sorafenib [
54]. ER stress also promotes drug resistance to sorafenib through various mechanisms in HCCs [
55‐
57]. Therefore, whether ER stress plays a positive or negative role in the anti-cancer therapeutic effect of sorafenib needs further clarification. Besides existing reports, to validate ER stress as a worthwhile target to treat cardiotoxicity without negatively affecting tumor treatment, one needs to carry out more experiments using both cells and tumor-bearing animal models.
ER stress is usually adaptive initially and causes detrimental effects if persisted. Upon afatinib, sorafenib, and ponatinib treatment, the initial induction of ER stress is probably beneficial for cardiac cells to restore ER homeostasis. However, if the repair attempts fail and the ER stress persists, this may lead to cellular damages, such as apoptosis or inflammation. ER stress induced by afatinib or ponatinib does not cause apoptosis, while that by sorafenib causes mild degree of apoptosis in cardiomyocytes. Yet the TKI-induced ER stress upregulates inflammation, including the inflammasome-IL1β pathway and the NF-κB pathway, in cardiomyocytes. Like the known pathways in immune cells [
58,
59], ER stress promotes expression of pro-inflammatory genes in cardiomyocytes, albeit dependent on coordinated activation of both PERK and IRE1α. Since inflammatory factors, such as IL1β and IL6, increase risk of cardiovascular diseases [
60,
61], the signaling from ER stress to inflammation induced by the TKIs may be a potential target for treating cardiotoxicity. Our finding is consistent with previous ones that sorafenib induces inflammation in skin [
62] and ponatinib causes inflammation in endothelial cells [
63] and in the ischemic brain of zebrafish [
64]. Since production of pro-inflammatory factors can exacerbate ER stress, whether inflammation is just a consequence of ER stress or it modulates ER stress and cardiotoxicity to TKIs still needs further investigation. As the three effectors of ER stress have adaptive and maladaptive functions depending on the duration and the selected pathways activated, maximizing the adaptive and minimizing the maladaptive functions of the ER stress may be an effective way to mitigate the cardiotoxicity that is caused by this pathway.
Our study is limited in many aspects. First, TKI-induced cardiotoxicity can be regulated by non-ER stress mechanisms that are not evaluated in the current study. For example, sorafenib inhibits mitochondria, induces ROS accumulation, and regulates metabolism of cardiomyocytes [
8,
44,
65]. How ER stress is associated with other pathways that regulate cardiotoxicity remain to be determined for many of the TKIs. Secondly, we established animal models based on short-term treatments of two cardiotoxic TKIs (ponatinib and sorafenib) and did not measure cardiac function, as the short-term intervention should not cause a noticeable change in cardiac function. It is worthwhile to treat the animals for a comparable duration with that of human therapy, or at least a month as generally done in literature, and measure ER stress at different time points over the course, as well as cardiac function at the end time point. Thirdly, we have not identified the upstream trigger of ER stress by TKIs. We tried to evaluate the role of ROS in ER stress activation, but our data suggest a paradoxical role of ROS. The antioxidant trolox, used in this paper, has a mild effect in inhibiting ROS or lipid peroxidation induced by TKIs, and the time points measured are limited. So, further study using some other antioxidants and at more time points is warranted. Last but not least, due to the high cost and long duration of using hiPSC-CMs, we switched to use NRCMs for mechanic studies. Our data and literature suggest that the two cell models agree in most of the phenotypic and molecular assays. To avoid species-related difference, one should further validate the ER stress-associated signaling pathways and gene expression induced by TKIs in more hiPSC-CMs or primary human cardiomyocytes.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.