A thorough analysis of cardiotoxicity of eight TKIs on cardiomyocytes and in patient data

At first, an in vitro human cardiomyocyte-based cell culture system was established to classify TKIs with different levels of cardiotoxicity. We selected TKIs with low cardiotoxicity (afatinib, gefitinib), medium-levels of cardiotoxicity (crizotinib, dasatinib, nilotinib), and high-levels of cardiotoxicity (sorafenib, sunitinib and ponatinib) based on literature reports (Table S1) [2, 4, 26,27,28,29,30,31,32,33,34,35,36]. Afatinib and gefitinib were reported without or with low cardiotoxicity. Crizotinib, dasatinib, and nilotinib induced cardiotoxicity at medium-level, including bradycardia, cardiac ischemia, and periphery vascular occlusion. Ponatinib, sorafenib, and sunitinib can target VEGFR2 and PDGFRs, so that cause high-levels of cardiotoxicity, such as hypertension, heart failure, myocardial infarction, and cardiac arrhythmias (Table S1). As it remains debatable, TKIs cause cardiotoxicity mainly through on-target or off-target effects and limited studies compared the cardiotoxic mechanisms of TKIs with similar targets, we chose 2–3 drugs that target EGFR (afatinib, gefitinib), Bcr-Abl (dasatinib, nilotinib), or VEGFR/PDGFR (ponatinib, sorafenib and sunitinib) so as to address these questions. The 8 TKIs selected are also widely used to treat different cancer types, from leukemia to solid tumors. Two previous studies have evaluated over 20 FDA-approved TKIs at a single dose and treatment duration on hiPSC-CMs or primary human cardiac cells [9, 37]; we think that we could gain different insights from the previous studies by studying more doses and treatment durations which mimic more variables associated with clinic usage of these drugs in order to find critical cardiotoxic mechanisms.

To evaluate the toxic effects of TKIs on cardiomyocytes, we assessed the ATP level, contraction, and respiration in response to different doses and treatment durations. ATP levels were measured over a dose range from 0.32 to 10 µM (spanning therapeutic relevant doses of the drugs) and over a time range from day 1 to day 5. Most drugs (afatinib, crizotinib, nilotinib, ponatinib, and sunitinib) caused dose-dependent and time-dependent reduction in ATP levels in hiPSC-CMs, whereas dasatinib and gefitinib did not inhibit ATP at any dose or time point tested (Fig. 1A–F, H). Sorafenib caused slight increase in the ATP level at day 1, but decrease at day 5 (Fig. 1G). We fitted a four-parameter log-logistic model to each dose response curve to calculate EC50. For drugs that shown inhibitory effects on ATP, EC50s were from 1 to 5 μM at day 5 (Fig. S1A-H) and decreased from day 1 to day 5, indicating an increase in toxicity over time (Fig. S1I). EC50s of sorafenib and nilotinib were lower than or similar to their maximal plasma concentrations (Cmax), while EC50s were higher than Cmax for the other drugs (Fig. S1A-H). We also used microelectrode array to measure base impedance and extracellular field potential (EFP) of hiPSC-CMs in response to these TKIs. Sorafenib (10 µM) and crizotinib (3.16 µM) stopped beating and contraction of hiPSC-CMs after 0.5 h of treatment (Fig. 1I, J). Afatinib (10 µM) and ponatinib (3.16 µM) reduced beat rate and corrected field potential duration (Fig. 1I, L), while nilotinib, gefitinib, and dasatinib had minimal effects on these parameters (Fig. 1I–L). Sunitinib (10 µM) increased beat rate initially and reduced the base impedance very acutely, indicating cell dissociation or death induced by the drug (Fig. 1I, K).

Fig. 1

Toxicity of eight TKIs measured by cellular assays, FAERS analysis, and literature review. A–H Cor.4U hiPSC-CMs were treated with TKIs at doses from 0.32 to 10 µM and duration from 1 to 5 days. Fold changes in ATP were calculated relative to vehicle controls at the same time point and shown as a surface plot. Treatments selected for subsequent RNAseq profiling were enclosed in red circles. Toxicity that was significantly different from controls was labeled with red asterisks. I–L Base impedance and extracellular field potential were measured by CardioExcyte 96 microelectrode array in HELP hiPSC-CMs treated with fixed doses of TKIs (afatinib 10 µM, gefitinib 10 µM, crizotinib 3.16 µM, dasatinib 10 µM, nilotinib 10 µM, ponatinib 3.16 µM, sorafenib 10 µM, sunitinib 10 µM) for 24 h. Beat rate (I), amplitude of impedance (J), base impedance (K), and corrected field potential duration (FPDc, L) were calculated for different treatments. Data were presented as mean ± SD of three replicated wells. M–N Mitochondrial oxygen consumption and extracellular acidification were measured in rat cardiomyocytes treated with fixed doses of TKIs (same as in I) for 24 h. A representative experiment from three independent repeats was shown. O–R Basal, maximal, spare, and non-mitochondrial oxygen consumption were derived from the seahorse experiment from M. Data were presented as mean ± SEM of three independent experiments with 5–6 replicated wells each. *p < 0.05, **p < 0.01, ***p < 0.001 versus the DMSO vehicle control group. S Heatmap of reporting odds ratios (RORs) calculated based on the event numbers of cardiotoxicity-related medical terms mined from the FDA adverse events reporting system (FAERS). T Toxicity rankings of eight TKIs based on literature, FAERS, ATP level, mitochondrial respiration and beating properties. Drug with the highest toxicity is on the top

Since many TKIs affected the ATP level in hiPSC-CMs, we evaluated the effects of these TKIs on mitochondrion, the main organelle for ATP generation, through both the seahorse assay and measuring mitochondrial membrane potential. After 24 h of treatment, ponatinib, sorafenib, and sunitinib inhibited maximal, spare, and non-mitochondrial oxygen consumption significantly (from p < 0.05 to p < 0.001, Fig. 1M, P, Q, R). Many TKIs showed the trend of inhibiting basal oxygen consumption rate (OCR), such as afatinib, crizotinib, dasatinib, ponatinib, and sunitinib, but only sorafenib had a significant inhibition (p < 0.01, Fig. 1O). In the acute treatment, most TKIs did not inhibit basal OCR, indicating that these TKIs are not acutely toxic to mitochondria (Fig. S2A). Based on the quantification of mitochondrial membrane potentials, only sorafenib decreased the mitochondrial membrane potential significantly after 24 h of treatment; nilotinib, ponatinib, and sunitinib showed the trend of decreasing mitochondrial membrane potentials (Fig. S3). The inhibitory effects of TKIs on mitochondria cannot fully explain the changes in the ATP level. Extracellular acidification rate, which correlates with cellular glycolysis rate, was not significantly affected by these TKIs (Figs. 1N and S2B). To further quantify the cardiotoxicity events induced by these TKIs in patients, we analyzed the USA federal drug administration adverse event reporting systems (FAERS) and showed that nilotinib and ponatinib caused the most cardiotoxicity with significant reporting odds ratios (RORs), followed by crizotinib and dasatinib (Fig. 1S). The overall rankings of toxicity of these TKIs based on the literature review, the FAERS, the ATP level, mitochondrial respiration, and beating properties were similar, but the toxicity of afatinib was rated higher in cellular assays than in FAERS or literature, and the toxicity of dasatinib was rated lower in cellular assays than in FAERS or literature (Fig. 1T).

TKI-induced transcriptome changes are grouped into 7 clusters, with two enriched in ER stress

To further elucidate molecular mechanisms of cardiotoxicity induced by TKIs, we profiled transcriptome of hiPSC-CMs at different levels of toxicity in response to TKIs (the toxicity levels were determined based on the effects of TKIs on the ATP level, mitochondrial respiration, and beating properties). The transcriptome data followed a L- or T-shaped design (Fig. 2A). There were 129 samples in total with the vehicle controls at days 1, 3, and 5. To increase efficacy and reduce cost, we measured the transcriptome using the 3’digital gene expression with unique molecular identifiers (3’DGE-UMI) RNA-seq method [25, 38] where the UMI barcodes were used to label each drug treatment, rather than individual cells.

Fig. 2

Major biological processes regulated by different TKIs. A The L- or T-shaped designs that span three doses and three time points for each TKI, selected based on results from Fig. 1A–H; 129 samples with three biological replicates per condition were measured in 3’DGE-UMI RNAseq. B Transcriptome changes induced by eight TKIs over dose and time were grouped into 7 clusters based on tSNE analysis, and the corresponding drugs of each cluster were shown in C. D Expression of top 20 gene markers for each cluster. E In Cluster 0, mitochondrial tRNA genes were expressed at a higher level than the other clusters. F Biological processes enriched for gene markers of Cluster 2. Expression of representative genes in GO term of mitotic nuclear division was shown on the right. G Biological processes enriched for gene markers of Cluster 3. Expression of representative genes in this GO endoplasmic reticulum stress was shown on the right. H Biological processes enriched for gene markers of Cluster 4. Expression of representative genes in GO term of heart contraction was shown on the right. I Biological processes enriched for gene markers of Cluster 6. Expression of representative genes in GO term of response to topologically incorrect protein was shown on the right

Based on tSNE analysis at a resolution of 2, drug-induced transcriptome responses were classified into 7 clusters (Fig. 2B). Clusters 2, 5, and 6 were composed of single drugs, nilotinib, gefitinib, and afatinib, respectively (Fig. 2B, C). Cluster 3 contained sorafenib and ponatinib; Cluster 0 was composed of crizotinib and sunitinib; Clusters 1 and 4 contained over three drug treatments (Fig. 2C). When the resolution was increased to 2.5 or 3 in the tSNE analysis, it did not change the overall clustering significantly, but split Cluster 0 and Cluster 3 into two clusters which mainly contained a single drug treatment (Fig. S4). As the variables of drug dose and treatment duration in the experimental design also affected transcriptional responses, the clustering separated most of drugs rather than each individual. Among the clusters, 0, 2, 3, 4, and 6 contained over 10 significantly differentially expressed genes (DEGs) that were used for gene ontology enrichment analysis (Figs. S5 and 2D). The primary biological process enriched in Cluster 3 was GO:0,034,976 response to endoplasmic reticulum stress and in Cluster 6 was GO: 0,035,966 response to topologically incorrect proteins (Fig. 2G, I). ER stress-related genes, including HERPUD1, SESN2, CHAC1, DDIT3, NUPR1, and TRIB3, were upregulated in both Clusters 3 and 6, but cytoplasmic chaperons, such as DNAJB1 and HSPA1B, were only activated in Cluster 6. Cluster 3 was also enriched for tRNA aminoacylation of protein translation and the gene makers (such as SARS and GARS) were increased by both sorafenib and ponatinib (Figs. 2G and S6). Cluster 2 was enriched for mitotic nuclear division (Fig. 2F). Cluster 4 was enriched for genes associated with heart contraction, and these genes were upregulated by the treatments (gefitinib, dasatinib, and ponatinib at day 3 or day 5) (Fig. 2H). Even though not enriched, Cluster 0 was associated with higher expression of mitochondrial tRNA genes (such as MT-TV, MT-TT, and MV-TY, Fig. 2E). By comparing transcriptome changes induced by these drugs, we found that drugs with the same targets did not induce similar transcriptome changes, e.g., afatinib and gefitinib, dasatinib and nilotinib, sorafenib and sunitinib. Only sorafenib and ponatinib partially overlapped in Cluster 3. These results indicate that TKIs’ effect on transcriptome is drug-specific rather than target-specific.

Transcriptome data had good quality and consistency; 75 samples from day 1 treatments were barcoded twice to serve as technical replicates. In t-SNE plots, 75 technical replicates clustered closely (Fig. S7). The tSNE analysis was based on the top 10 principal components (PCs, p < 0.001, Fig. S8). About 10,000 unique genes and 105 total counts were detected as median values for different samples (Fig. S9A, B). Percentage of mitochondrial DNA was high (10–50%) in our data, which is probably caused by mitochondrial damage of some TKIs [39, 40] (Fig. S9C). Additionally, the percentage of mitochondrial DNA negatively correlated with total RNA counts of each sample (correlation coefficient =  − 0.66 and p < 0.001, Fig. S10A), indicating that samples with mitochondrial damage had lower number of total reads measured. As expected, the number of unique genes detected positively correlated with total RNA counts (correlation coefficient = 0.89 and p < 0.001, Fig. S10B). When comparing the same drug treatments (sorafenib or sunitinib at 3.16 µM and 3 days) of the current 3’DGE RNA-seq with the published bulk RNAseq data (GEO GSE114686) [8, 41], we found that ~ 33.5% DEGs of sorafenib and ~ 38.4% DEGs of sunitinib overlapped; Pearson correlation was 0.91 for sorafenib and 0.87 for sunitinib (Fig. S11). In summary, transcriptome data revealed the major biological processes regulated by eight TKIs and ER stress was a shared response by three of them.

Drug-specific effects on transcriptome dominate dose-, time-, or toxicity-induced effects

To better visualize dose- or time-dependent effect of each drug, data were sliced and viewed based on either a concentration gradient or a time gradient (Fig. 3A, E). From the dose perspective, concentration was increased from the lower right to upper left direction for most drugs, except nilotinib, in the tSNE clustering (Fig. 3B, C). Afatinib, crizotinib, nilotinib, ponatinib, sorafenib, and sunitinib caused dose-dependent toxicity, whereas, dasatinib and gefitinib did not (Fig. 3B–D). When data were viewed longitudinally, afatinib, sorafenib, sunitinib, and dasatinib showed increasing toxicity with treatment duration, albeit to different degrees (Fig. 3E–H). Crizotinib, nilotinib, and ponatinib did not have time-dependent toxicity (Fig. 3F–H). When comparing changes from the dose and time gradients, four TKIs (afatinib, sorafenib, sunitinib, and dasatinib) showed similar directions of change in tSNE space; whereas three TKIs (ponatinib, crizotinib, and gefitinib) showed different, or even opposite, directions of change. In all the above cases, drug-specific effects on transcriptome dominated over dose-, time-, or toxicity-induced effects.

Fig. 3

Drug specific effects on transcriptome dominate dose-, time- or toxicity-induced effects. A Transcriptome data were sliced to retain all drug treatments at day 1 with different doses. The orange arrow represents the concentration gradient of data. Red squares denote the doses for six TKIs (labeled in red in B) and blue diamonds denote the doses for two TKIs (labeled in blue in B). B–D Data selected as in A were projected into the tSNE space and shown with the properties of drug, concentration, or toxicity (represented by the ATP fold changes). Darker gray corresponds to higher toxicity. E Transcriptome data were sliced to retain drug treatments at a fixed dose over 5 days. The blue arrow represents the time gradient of data. Red squares denote the time points for six TKIs (labeled in red in F) and blue diamonds denote the time points for two TKIs (labeled in blue in F). F–H Data selected as in E were projected into the tSNE space and shown with the properties of drug, time or toxicity (represented by the ATP fold changes). Toxicity that was significantly different from controls was labeled with red asterisks in H. I Principal component analysis (PCA) of transcriptome changes induced by eight TKIs. J Drug concentration of each condition projected into the PCA space. K Treatment duration of each condition projected into the PCA space. L Toxicity level of each condition defined by the percent ATP of controls projected into the PCA space. Darker gray corresponds to higher toxicity. M–N Expression of genes with top and bottom 30 highest loadings of PC1 and PC2 grouped by drugs

To analyze relation of TKI-induced transcriptome changes, principal component analysis (PCA) was run based on the top 2000 variable genes in the data. PC1 explained 24% of total variance, and PC2 explained 15% (Fig. 3I). Afatinib, sorafenib, and a few of ponatinib treatments were segregated from the other TKIs or vehicle controls from PC1 (Fig. 3I). This is consistent with the tSNE result where Cluster 3 (sorafenib and high-dose ponatinib-treated) and Cluster 6 (afatinib-treated) were segregated from the other TKIs (Fig. 2B, C). The PC2 axis distinguished crizotinib- and sunitinib-treated conditions, which were grouped into Cluster 0 in tSNE, from the rest of the drugs (Figs. 2B, C and 3I). Expression of genes with loadings ranked in the top 30 or the bottom 30 on PC1 or PC2 was plotted based on drugs (Fig. 3M, N). Consistent with Fig. 3I, the expression of genes with high PC1 loadings differentiated afatinib, sorafenib, and high-dose ponatinib from the rest of the drugs. Genes associated with ER stress (e.g., ATF4, NUPR1, DDIT3, TRIB3, CHAC1, SESN2, HERPUD1) were upregulated and genes related to cardiomyopathy and sarcomeric structure (e.g., MYH7, SYNPO2L, LDB3, ACTN2, MYLK3) were downregulated by afatinib, sorafenib, or high-dose ponatinib. Expression levels of high loading genes on PC2 distinguished the effect of crizotinib and sunitinib from the other TKIs (Fig. 3N). Among the high loading genes in PC2, genes associated with mitochondrial tRNAs (e.g., MT-TH, MT-TP, MT-TL1, MT-TV), as well as those function in mitochondrial electron transfer chain (e.g., MT-ND3, UQCR11, COX7B, MTND3P19, ATP5IF1, COX6C), were upregulated, similar to changes observed in gene markers of Cluster 0 (Figs. 3N and 2E). Consistent with the tSNE analysis, segregation of samples along PC1 or PC2 was dependent on drug type, rather than concentration, treatment duration, or toxicity (Fig. 3I–L). Therefore, transcriptome data revealed drug-specific, concentration-, and time-dependent responses in human cardiomyocytes to TKIs, with the drug-specific effect as the dominating factor.

Afatinib, sorafenib, and ponatinib induce ER stress in rat cardiomyocytes

Each batch of hiPSC-CMs took about a month to differentiate and were very expensive to culture or purchase. Neonatal rat cardiac myocytes (NRCMs) took 1 day to isolate from neonatal rat hearts and were less expensive to culture; therefore, we chose NRCMs over hiPSC-CMs for the subsequent mechanistic studies. Admittedly, NRCMs were derived from rat and could have species differences from human, but these cells had been widely used to study molecular mechanisms of cardiac hypertrophy and provided translatable findings related to cardiac diseases [42]. Additionally, NRCMs shared similar sensitivity to afatinib, sorafenib, and ponatinib with hiPSC-CMs (Fig. S12). So, we used NRCMs to validate the upregulation of ER stress observed in previous RNA-seq data. NRCMs were treated with afatinib at 5.62 or 10 μM, sorafenib at 3.16 or 10 μM and ponatinib at 1.78 or 5.62 μM for 24 h and with the lower doses of the three TKIs for 72 h. Gene targets of three ER stress effectors, Atf4, Xbp1s, and Atf6, were upregulated by the three TKIs at high doses and 24 h (p < 0.001, Fig. 4A–C). Chac1, Ddit3, and Trib3 are gene targets of Atf4. Consistent with upregulation of Atf4, Chac1, Ddit3, and Trib3 were upregulated by the three TKIs at high doses and 24 h (p < 0.01, Fig. 4A–C). Dnajb9 is a downstream target of Xbp1s. Ponatinib treatment caused the most robust upregulation of Dnajb9 among the 3 TKIs (p < 0.001, Fig. 4A–C). Hspa5 and Herpud1 are downstream targets of Atf6 (Hspa5 is also up-regulated by Atf4). Herpud1 was most induced by ponatinib (p < 0.001, Fig. 4A–C). Hspa5 was increased similarly by the three TKIs (p < 0.01, Fig. 4A–C). Despite the drug- and dose-dependent increase in the expression of ER stress genes, activation of three UPR branches was biased. With high-dose treatment, afatinib and sorafenib activated the ATF4 axis more robustly than the XBP1s and the ATF6 axes, while ponatinib induced activation of three axes to similar degrees (Fig. 4D, left). The three TKIs induced different temporal activation of ER stress genes. Afatinib caused a higher expression of Atf4, but a lower expression of Xbp1s and Atf6 at 72- than 24-h treatment (Figs. 4D, right, and S13A). For sorafenib at 3.16 μM, most gene targets of ER stress were downregulated over time, indicating a transient ER stress response (Fig. 4D, right, and S13B). For ponatinib treated at 1.78 μM, the ATF4 and the ATF6 axes, but not XBP1s, showed time-dependent upregulation (Fig. 4D, right, and S13C). In summary, afatinib, sorafenib, and ponatinib not only induced ER stress at high doses acutely but also at low doses chronically.

Fig. 4

Afatinib, sorafenib and ponatinib induce ER stress in NRCMs and rat hearts. A Fold changes of ER stress related genes, Atf4 and its targets, Chac1, Ddit3, Trib3;Xbp1s, and its target Dnajb9; Atf6 and its targets Hspa5 and Herpud1, in NRCMs treated with afatinib at 5.62 or 10 µM for 24 h. B Fold changes of ER stress related genes in NRVMs treated with sorafenib at 3.16 or 10 µM for 24 h. C Fold changes of ER stress related genes in NRCMs treated with ponatinib at 1.78 or 5.62 µM for 24 h. D Left: heatmap of gene expression changes from A–C. Right: heatmap of gene expression changes caused by afatinib 5.62 µM, sorafenib 3.16 µM, or ponatinib 1.78 µM at 24 or 72 h. E Phospho-eIF2α (or EIF2S1), XBP1s, ATF6 (cleaved), and GAPDH expression at different time points under afatinib, sorafenib, and ponatinib treatment measured by Western blot. F–H Fold changes of Anp, Dnajb9, and Ddit3 in rat left ventricles from sorafenib or ponatinib gavage for 3 or 7 days. A-C Data were presented as mean ± SEM (n = 3) and analyzed using ANOVA analysis. *p < 0.05, **p < 0.01, ***p < 0.001 versus the DMSO vehicle control group. #p < 0.05, ##p < 0.01, ###p < 0.001 versus the lower dose. F–H Data were presented as mean ± SEM (n = 8) and analyzed using ANOVA analysis. *p < 0.05, **p < 0.01, ***p < 0.001 versus the vehicle control

We further explored the temporal activation of ER stress at protein level in response to the 3 TKIs. Phosphorylation of eIF2α (gene name: Eif2s1), which is downstream of PERK and upstream of Atf4, responded quickly to the 3 TKIs and was activated within 30 min (Fig. 4E). Ponatinib activated phosphorylation of eIF2α for 24 h, but afatinib and sorafenib activated it for 12 or 6 h, respectively (Fig. 4E). XBP1s was activated much later than eIF2α, after 3 or 6 h of the TKI treatment and remained elevated for 24 h (Fig. 4E). The expression of activated-ATF6 was not changed by any drug treatment (Fig. 4E). To validate that these TKIs also induce the expression of ER stress genes in human cardiomyocytes, we analyzed the public dataset GSE114686. The data were based on treating Cor.4U hiPSC-CMs, namely the same cells used in the current study, with four TKIs (only sorafenib was overlapped with the current study). In GSE114686, gene markers of ER stress (including ATF4, CHAC1, DDIT3, TRIB3, XBP1, DNAJB9, ATF6, HSPA5, HERPUD1) were all upregulated by sorafenib at 10 µM for 24 h or at 3.16 µM for 168 h (see Table S2), supporting that ER stress induced by sorafenib also happens in human cardiomyocytes and is probably independent of the species of cell models.

Sorafenib and ponatinib modulate ER stress in adult rat hearts

To validate TKI-induced ER stress in vivo, we established cardiotoxicity animal models by gavaging Sprague–Dawley rats with either ponatinib (15 mg/kg) or sorafenib (50 mg/kg) once daily (see Table S3 The ARRIVE checklist). We did not do it for afatinib because this drug was not reported highly cardiotoxic. The dose of the drugs was selected based on literature [43, 44], as well as human-to-rat dose conversion based on body surface area principle. As ER stress is usually a transient response, we reasoned that we should collect the hearts at early time points before obvious cardiotoxicity. Rats were gavaged for 3 or 7 days and the hearts were collected to measure gene markers of ER stress. Ponatinib treatment was more toxic than sorafenib as one rat died in the ponatinib-treated group. Heart or body weight was not changed by sorafenib, but decreased by ponatinib at day 7 (p < 0.01 and p < 0.001, respectively, Fig. S14A–C). The ratio of heart weight to body weight was not changed in any treatment. Ponatinib induced a significant increase in gene expression of Anp (official gene name: Nppa), which is a fetal gene re-expressed during cardiac stress or dysfunction, indicating that potential cardiotoxicity was induced by this drug (p < 0.01, Fig. 4F). Sorafenib did not induce a significant change in Anp, but 2–3 animals had a high fold increase in Anp expression and this reflected variation in drug response among individual rats (Fig. 4F). Consistent with the cardiotoxicity induced by ponatinib, Dnajb9 expression was significantly increased at day 7, indicating the activation of Xbp1s axis of ER stress (p < 0.01, Fig. 4G). Sorafenib caused a significant reduction of Xbp1s (p < 0.05, Fig. 4H), the consequences of which need further exploration. Neither drugs caused a significant change in Ddit3, which is downstream of the PERK-peIF2α-Atf4 pathway (Fig. S14D).

Afatinib, sorafenib, and ponatinib induce different levels of lipid peroxidation, ROS, calcium defects, and TNNT2 loss in rat cardiomyocytes

Oxidative stress is an inducer of ER stress [45], so we also evaluated whether these TKIs induce reactive oxygen species (ROS) in NRCMs by flow cytometry. All three TKIs induced an increase in free ROS at 3 h of treatment. Specifically, the percent of ROS-high cells was increased by 14.1%, 16.2%, and 6.8% by afatinib, sorafenib, and ponatinib, respectively (Fig. 5A). For sorafenib treatment, the median fluorescence intensity of the ROS-high cells was about twofolds of that in the control group. These TKIs also induced an increase in ROS at 24 h, but the percent of ROS-high cells was lower than that at 3 h (Fig. 5B), indicating that ROS induction may be transient. Lipid peroxidation measured by the ratio of fluorescence intensity from the FITC channel to the PE channel was also assessed at 3 and 24 h. At 3 h of treatment, the percent of cells with high lipid peroxidation was increased slightly by afatinib and sorafenib (Fig. 5C). However, at 24 h of treatment, the three TKIs increased the percent of cells with high lipid peroxidation by 4 to 32% (Fig. 5D); the level induced by sorafenib was the closest to that of the positive control, cumene hydroperoxide (42%, Fig. S15). Ca2+ also plays a vital role in maintaining ER homeostasis and cardiomyocyte contraction [46], and calcium overload is associated with ER stress and contraction/relaxation defects. So we measured how these TKIs affect calcium homeostasis of NRCMs. Sorafenib induced a significant cytosolic overload of Ca2+ in NRCMs at 0.5 and 1 h (p < 0.01 and p < 0.001, respectively); whereas afatinib and ponatinib did not cause any significant changes in calcium concentration (Fig. 5E, F). We then tested whether antioxidants, such as trolox, can rescue the ER stress induced by these TKIs. While trolox reduced the level of lipid peroxidation induced by afatinib and sorafenib at 24 h (92.5% vs 83% and 94.6% vs 88%), it did not block the induction of ER stress by these TKIs (Fig. S16). Paradoxically, the combination of afatinib or sorafenib with trolox increased the expression of some ER stress genes compared with TKI alone (Fig. S16). As ER stress is dynamic, it could be that the anti-oxidant extended the activation duration of ER stress or that some other factors regulated ER stress cooperatively with oxidative stress. TNNT2 protein (or cardiac Troponin T2), a sarcomeric component and marker for cardiac damage, was inhibited by the 3 TKIs, among which sorafenib had the most effect (p < 0.001, Fig. 5G, H). In summary, ROS level and lipid peroxidation were increased by the three TKIs, and calcium overload induced by sorafenib prior to ER stress. The toxicity of these TKIs was also associated with a decrease in cardiac Troponin T2 protein expression, which may negatively affect myocardial cell contraction.

Fig. 5

The 3 TKIs induce different levels of lipid peroxidation, ROS, calcium defects, and TNNT2 loss in NRCMs. To measure oxidative stress and cardiotoxic effects induced by TKIs, NRCMs were treated with 10 µM afatinib, 10 µM sorafenib, or 5.62 µM ponatinib and stained with various dyes before quantification by flow cytometry or imaging. A-B Percent of ROS-high cells quantified by flow cytometry and staining with H2DCFDA after different TKI treatments for 3 or 24 h. C–D Histogram of the ratios between green (oxidized) and red (non-oxidized) fluorescence intensity of C11-bodipy581/591 in NRCMs under different treatments. Percentages by the condition names were the fraction of cells in the specified gate for each treatment. E Representative fluorescence images of live NRCMs pre-loaded with Calbryte™ 520 AM and treated with the indicated drug from 0 to 3 h. F Integrated density of green fluorescence in E. G Representative immunofluorescence images of TNNT2 (green) and nuclei (blue) in NRCMs treated with TKIs for 24 h. H Integrated density of green fluorescence in G. Data were presented as mean ± SEM (n = 3) and analyzed using ANOVA analysis. *p < 0.05, **p < 0.01, ***p < 0.001 versus the DMSO vehicle control group. Scale bar: 200 µm

Afatinib, sorafeni