Novel Screening Method Identifies PI3Kα, mTOR, and IGF1R as Key Kinases Regulating Cardiomyocyte Survival.

Background Small molecule kinase inhibitors (KIs) are a class of agents currently used for treatment of various cancers. Unfortunately, treatment of cancer patients with some of the KIs is associated with cardiotoxicity, and there is an unmet need for methods to predict their cardiotoxicity. Here, we utilized a novel computational method to identify protein kinases crucial for cardiomyocyte viability. Methods and Results One hundred forty KIs were screened for their toxicity in cultured neonatal cardiomyocytes. The kinase targets of KIs were determined based on integrated data from binding assays. The key kinases mediating the toxicity of KIs to cardiomyocytes were identified by using a novel machine learning method for target deconvolution that combines the information from the toxicity screen and from the kinase profiling assays. The top kinases identified by the model were phosphoinositide 3-kinase catalytic subunit alpha, mammalian target of rapamycin, and insulin-like growth factor 1 receptor. Knockdown of the individual kinases in cardiomyocytes confirmed their role in regulating cardiomyocyte viability. Conclusions Combining the data from analysis of KI toxicity on cardiomyocytes and KI target profiling provides a novel method to predict cardiomyocyte toxicity of KIs.

Clinical Perspective

What Is New?

We show that a novel computational method can be used to in vitro screen cardiotoxicity of new kinase inhibitors, a group of widely used anticancer drugs.

Our data identify novel kinase inhibitor targets, such as dual inhibition of phosphoinositide 3‐kinase catalytic subunit alpha and mammalian target of rapamycin, as potential mediators of cardiomyocyte viability and cardiotoxicity.

What Are the Clinical Implications?

The clinical relevance of this novel assay was supported by identification of well‐known cardiotoxic kinase inhibitors.

Toxicity of dual phosphoinositide 3‐kinase catalytic subunit alpha/mammalian target of rapamycin inhibition on cardiomyocytes may bear relevance for future drug design targeting the phosphoinositide 3‐kinase catalytic subunit alpha/mammalian target of rapamycin pathway.

Intoduction

Treatment of patients with kinase inhibitors (KIs) has dramatically improved the prognosis of many cancers. Unfortunately, treatment of patients with some of the KIs is associated with cardiotoxicity, and has raised concerns for adverse cardiac effects of kinase inhibition. There are 518 kinases encoded by the kinase genome (kinome).1 In many cancers, mutations in 1 or several genes encoding kinases can lead to failure in tumor suppression or overexpression of proto‐oncogenic proteins. Kinase inhibition in cancer cells can efficiently kill it or stop its growth but, unfortunately, many kinases that drive tumorigenesis are also important for cardiomyocyte survival and function.2 In addition, most KIs exhibit poor selectivity in their kinase targets because of high homology of ATP pockets across the kinases. Because of the diverse kinase targets found for each KI, many approved KIs are associated with adverse events including cardiomyopathy.3, 4, 5 Currently, among the clinically approved KIs, ponatinib and vandetanib are labeled with the black box warning for cardiovascular implications while for pazopanib, sunitinib, nilotinib, sorafenib, ceritinib, crizotinib, and dasatinib, the cardiovascular events are listed as possible adverse side effects.5

Cardiotoxicity has accounted for nearly half of the drug withdrawals, indicating that the predictive value of currently available toxicity screening methods is poor.6 Cardiotoxicity caused by a KI could result from both (1) on‐target toxicity where the kinase that enhances the tumor cell growth is also important for the cardiomyocyte survival; or (2) off‐target toxicity where a KI leads to toxicity via inhibition of a kinase not intended to be a target of the drug.2, 3 Off‐target toxicity is mainly observed in multitargeted KIs that interact with a wide array of kinase targets. A growing number of patients, many of them with cardiovascular comorbidities, are receiving treatments with KIs and currently, predicting cardiotoxicity of KIs in the preclinical phase is a challenge.

Herein, we carried out a systematic approach for identifying kinases, whose inhibition is detrimental for cardiomyocyte viability. We performed analysis of 140 KIs for their ability to induce cardiotoxicity in neonatal rat ventricular cardiomyocytes in vitro. Cardiomyocyte viability was analyzed by measuring the ATP levels in cardiomyocytes after exposure of cells to a 3‐log concentration range of each KI for 24 hours. KIs were then analyzed for their kinase targets based on integrated data from binding assays for wild‐type kinases (418) considering both on‐ and off‐targets of the KIs. Finally, we developed a novel machine learning method for target deconvolution that combines the information from our toxicity screen and from the kinase profiling assays to identify kinases crucial for cardiomyocyte viability.

Methods

Animal handling and cardiomyocyte isolation were carried out in accordance with University of Oulu institutional guidelines, which conform to the National Research Council (US) Guide for the Care and Use of Laboratory Animals. An expanded methods section is provided in Data S1.

Data Availability

The authors declare that all supporting data are available within the article and its online supplementary files.

Predictive Target Deconvolution

We developed and applied a novel probabilistic machine learning–based predictive target deconvolution approach (PTD). The key assumption of PTD is that by modeling the KI response as an outcome of the target binding affinities, one can deconvolute and identify the key targets that are responsible for the prediction of the response. Specifically, PTD extends sparse linear regression7 to the case where predictor variables may contain missing values (here, binding affinities, see below). This is achieved by modeling missing predictor variables as parameters of the model that are learned during the inference procedure. As a result, PTD can be considered as a sparse linear regression for noncomplete predictor variables. PTD was used to model the dose–response measurements of a set of KIs as regression outcomes of the corresponding binding affinities. The model is formulated in the Bayesian paradigm as:

Here, forms the dose–response measurements while are the binding affinities. The outcome variables are regressed from with regression coefficients . X represent the unobserved and missing binding affinities. We assume to be centered around zero to match the background data distribution. We also assume normal distributions on and for simplicity, while varying these could further improve the results. We assume to have a Laplace prior inspired from sparse linear regression7 to induce sparsity for the kinase feature selection. Finally, the hyperparameters (α=1, γ=1) are also initialized to induce sparsity in both and X to match the assumption of sparse feature selection in and sparsely observed binding affinities in . The noise parameter σ is assumed to have a noninformative symmetric prior with shape and rate parameters a,b set to 1. The model was implemented using Gibbs sampling obtaining 200 posterior samples with a thinning factor of 10 and a burnin of 2000 iterations.

To quantify the KI responses, the dose–response curves were summarized using the quantitative drug sensitivity score (DSS).8 The area under the dose–response curve (AUC) was calculated by fitting a 4‐paremeter logistic function to the normalized %inhibition values over the concentration points. A predefined noise threshold (10% inhibition) was further used in the calculation of the drug sensitivity score (DSS), which classifies each compound with dose–response curved below the threshold as inactive (DSS=0). DSS data from 103 KIs used at maximum 10 μmol/L concentration were used in the PTD analysis. We used both on‐ and off‐target binding affinity profiles to model the polypharmacological effects of KIs. The kinase binding affinities of 132 of the 140 compounds were extracted from the FIMM in‐house Drug Target Commons database.9 For the remaining 8 compounds, the target affinity data were obtained from Davis et al.10 The integration of the bioactivity profiles was based on the KIBA algorithm.11 Kinases with affinity values indicating low potency (>1000 nmol/L) were removed as nonrelevant. Furthermore, kinases targeted by ≤3 inhibitors were excluded since they were considered statistically unreliable for PTD inference. This resulted in a total of 295 kinases that were used for modeling the response profiles of the 103 KIs (Data Set 1).

The PTD model was used to infer the top kinases associated with the DSS toxicity responses, where the PTD parameters were estimated using leave‐one‐out cross‐validation. The learned importance scores (regression weights) from each fold were then averaged over the 103‐folds to identify the kinases that are robustly linked to the toxicity response. We selected the top‐10 kinases based on the regression weights that were found to be consistent across the cross‐validation folds using t test with Bonferroni multiple‐hypothesis correction (differentiating the weights from zero null hypothesis).

Statistical Analysis

Statistical analysis was performed with GraphPad software. Normally distributed data were analyzed with t test for 2‐sample comparisons. For analysis of 3 or more groups, ANOVA followed by Dunnett's test was used. Differences were considered statistically significant at the level of P<0.05. Data are shown as mean±SD.

Results

Effect of KIs on Cardiomyocyte Viability

Measuring the ATP content in viable cells is a widely used approach in cytotoxicity screening applications in vitro. In cardiomyocytes, the number of cells plated showed a linear relationship and were highly correlated with both the ATP measured (r 2=0.98, Figure S1A) and the amount of total protein measured at the end of the experiment (r 2=0.98, Figure S1B).

For the toxicity analysis, cells were plated at a density of 60 000 cells/well, which is in the linear range of the ATP assay (Figure S1A). After exposure of cardiomyocytes to a 3‐log concentration range of each KI for 24 hours, an ATP assay was performed to determine the number of viable cells in culture. Eighty‐one of 140 KIs tested showed >10% decrease in cell viability and 27 showed a >30% decrease in cell viability (Figure 1A). The dose–response curves for 33 clinically approved KIs included in the screen are shown in Figure 1B. Twenty‐five of the 33 clinically approved KIs included in the study demonstrated a moderate‐to‐high toxicity to cardiomyocytes. Gefitinib, bosutinib, regorafenib, ponatinib, and midostaurin reduced cardiomyocyte viability by >50% at the maximum concentration tested, while imatinib, erlotinib, sorafenib, sunitinib, dasatinib, lapatinib, nilotinib, pazopanib, axitinib, trametinib, afatinib, ibrutinib, idelalisib, nintedainb, palbociclib, neratinib, binimetinib, dacomitinib, fostamitinib, and everolimus reduced cardiomyocyte viability by 10% to 50% at the maximum concentration tested. The dose–response curves for all KIs analyzed are shown in Table S1.

Figure 1

Toxicity of kinase inhibitors (KIs) on cardiomyocytes. Neonatal rat cardiomyocytes were treated with a 3‐log concentration range for each kinase inhibitor, and cardiomyocyte viability was assessed by ATP assay. A, Classification of cardiomyocyte toxicity of 140 KIs according to percent of viability decreased at maximum concentration. B, Dose–response curves for cardiomyocyte viability (% viable cells) of approved KIs analyzed. Each point on the curves represents a mean of 2 replicates.

Identification of Kinases Necessary for Cardiomyocyte Survival

The DSS calculation is based on the integration of the area under the dose–response curve to quantify the summary response intensity. The DSS score is defined within the range of 0 to 50, with close to 0 specifying no response, >5 indicating moderate response, and values close to 50 representing maximally toxic response.8 DSS score indicated moderate‐to‐high cardiomyocyte toxicity for 35 of the KIs tested (DSS >5, Figure 2A). Nineteen of the compounds induced low‐to‐moderate toxicity on cardiomyocytes (2<DSS<5, Figure 2B). DSS values for all the KIs tested are shown in Table S2.

Figure 2

Determination of drug sensitivity score (DSS) and kinases associated with cardiomyocyte survival. A, Kinase inhibitors (KIs) with moderate to high toxicity (DSS >5) and their primary kinase targets. B, KIs with low‐to‐moderate toxicity (5>DSS>2) and their primary kinase targets. C, DSS and the primary kinase targets for KIs tested at 10 μmol/L maximum concentration. Shown are KIs with DSS>5 and the corresponding DSS value at 3 and 1 μmol/L cutoffs. D, Corresponding regression weights of top 10 kinase targets identified by using 10 mmol/L maximum concentration. E, Corresponding regression weights of top 10 kinase targets identified by using 3 mmol/L maximum concentration. P values are calculated using 1‐sided t test.

To identify the kinase targets conferring cardiomyocyte toxicity, we analyzed a subset of 103 KIs that were used at maximum 10 μmol/L concentration. Thirty of the KIs had DSS >5 with 10 μmol/L concentration, with expected dose response with 3 and 1 μmol/L concentrations (Figure 2C). A linear regression model was then used to infer the kinases linked with the toxicity response in cardiomyocytes (Figure 2D, Table). The top kinases identified by the model were phosphoinositide 3‐kinase catalytic subunit alpha (PI3Kα, p110α), mammalian target of rapamycin (mTOR), and insulin‐like growth factor 1 receptor (IGF1R). The rank for Mek1 that has been previously shown to protect the cardiomyocytes from various stresses was 11 (Table).12 To exclude possible nonspecific toxicities of the kinase inhibitors, we next repeated the PTD model analysis by omitting the toxicity data with the 10 μmol/L KI concentration. Analysis for kinases mediating the toxicity with maximum 3 μmol/L concentration identified the same top 3 kinases (PI3Kα, IGF1R, and mTOR) having the highest regression weights (Figure 2E). Other kinases identified by both concentration thresholds were B‐Raf, PI3K catalytic subunit delta (PI3Kδ), RAF proto‐oncogene serine/threonine‐protein kinase (Raf‐1), PI3K catalytic subunit type 2 beta (PI3K2β), and mitogen‐activated protein kinase kinase kinase 7 (MAP3K7 or TAK1).

Table 1

Kinases Regulating Cardiomyocyte Survival Identified by Predictive Target Deconvolution Model

Rank	Targets	Regression Weights	P Values
1	PI3Kα	1.025	1.68 E‐27
2	mTOR	0.697	9.57 E‐30
3	IGF1R	0.578	1.86 E‐25
4	CHEK2	0.511	0.008
5	PI3KCδ	0.364	2.30 E‐19
6	PI3Kβ	0.356	1.60 E‐08
7	Raf‐1	0.274	1.55 E‐18
8	FLT3	0.242	6.53 E06
9	B‐RAF	0.219	6.97 E‐18
10	MAP3K7	0.217	1.54 E‐06
11	MAP2K1	0.198	2.33 E‐05
12	PDGFRA	0.183	1.99 E‐12
13	PI3Kγ	0.165	5.65 E‐21
14	AURKA	0.154	4.57 E‐14
15	MET	0.146	2.07 E‐06
16	NTRK3	0.142	1.92 E‐12
17	MAP4K5	0.126	1.22 E‐11
18	CSNK1D	0.125	2.23 E‐05
19	FGFR3	0.117	1.82 E‐08
20	AKT1	0.108	0.004
21	MST1R	0.100	6.37 E‐06

The kinases ranking is based on the regression weights using the data from 103 kinase inhibitors tested at maximum concentration 10 μmol/L. AKT1 indicates RAC‐alpha serine/threonine‐protein kinase; AURKA, aurora kinase A; B‐RAF, Serine/threonine‐protein kinase B‐raf; CHEK2, check point kinase 2; CSNK1D, casein kinase 1 delta; FGFR3, fibroblast growth factor receptor 3; FLT3, Fms‐related tyrosine kinase 3; IGF1R, insulin‐like growth factor‐1 receptor; MAPK, mitogen‐activated protein kinase; MET, Hepatocyte growth factor receptor; MST1R, macrophage stimulating‐1 receptor; mTOR, mammalian target of rapamycin; NTRK3, neurotrophic receptor tyrosine kinase 3; PDGFRA, platelet‐derived growth factor receptor alpha; PI3K, phosphoinositide 3‐kinase; Raf‐1, RAF proto‐oncogene serine/threonine‐protein kinase.

Dual Inhibition of PI3Kα and RICTOR Induces Necrotic Cardiomyocyte Death

mTOR inhibitors analyzed in the study target either mTOR complex 1 (mTORC1) (everolimus and sirolimus) or both mTORC1 and mTORC2 (AZD8055, vistusertib, PP242, sapanasertib, and OSI‐027), with markedly higher cardiomyocyte toxicity (DSS >4.3) observed with mTORC1/2 inhibitors. To further investigate the role of PI3Kα, IGF1R, and mTOR, we treated cardiomyocytes with KIs targeting either PI3K, IGF1R, mTOR, or both PI3K and mTOR (dual PI3K/mTOR inhibitors) and analyzed for both ATP levels and necrotic cardiomyocyte death. Cardiomyocytes were incubated with 2‐log concentration range of each inhibitor for 24 and 48 hours. Twenty‐four‐hour incubation of compounds induced a dose‐dependent decrease in cardiomyocyte ATP levels, while only PI3K and dual PI3K/mTOR inhibitors induced consistent increase in necrotic cell death (Figure S2 and Table S3). While treatment of cardiomyocytes with some inhibitors (pilaralisib, buparlisib, pictilisib, BMS754807, and omipalisib) induced necrotic cell death with only modest decrease in ATP levels, our data suggest that there is a threshold for the decrease in ATP levels (with ≈50–60% decrease) that results in necrotic cardiomyocyte death (adenylate kinase release). Forty‐eight‐hour treatment with KIs induced a further decrease in ATP levels, and cardiomyocyte toxicity was observed even with the lowest KI concentration (100 nmol/L) with most of the inhibitors (Figure 3). However, necrotic cell death was consistently observed with only dual PI3K/mTOR inhibitors.

Figure 3

Toxicity of selective PI3K, IGF1R, mTOR, and dual PI3K/mTOR on cardiomyocytes. Neonatal rat cardiomyocytes were treated with 4 different concentrations of each inhibitor for 48 hours. At the end of the experiment, medium samples were collected for adenylate kinase (AK) assay and cells were lysed for measurement of ATP levels. Shown are analyses for cardiomyocyte toxicity by ATP assay (left column) and AK assay (right column). Data for ATP assay are shown as percent of viable cells and for adenylate kinase as fold change vs control. Concentrations from left to right are 0.1, 1, 3, and 10 μmol/L. Data are represented as mean±SD **P<0.01; ***P<0.001 (Dunnett's test).

To confirm the role of the top kinase targets identified on regulating cardiomyocyte viability, we used RNAi to downregulate PI3Kα, IGF1R, and RICTOR (specific protein subunit in the mTORC2 complex) in cardiomyocytes. Quantitative polymerase chain reaction analysis showed that 70 nmol/L siRNA was sufficient to decrease the expression levels of the kinases by ≈50% (Figure S3). It is of note that the decrease in kinase levels achieved by the chosen siRNA concentrations was not complete, but rather mimicked the inhibition achieved with clinically relevant concentration of KIs. Analysis of cardiomyocyte toxicity by ATP assay indicated that PI3Kα and IGF1R silencing modestly reduced cardiomyocyte viability, while RICTOR silencing had no effect (Figure 4A). Interestingly, combined knockdown of the kinases (PI3Kα+IGF1R, PI3Kα+RICTOR, or IGF1R+RICTOR) resulted in toxicity on cardiomyocytes (Figure 4A). Analysis for adenylate kinase release showed that knockdown of individual kinases did not induce necrotic cell death, and only combined silencing of PI3Kα and RICTOR resulted in an increase in AK release (Figure 4B). Taken together, inhibition of either PI3Kα or IGF1R induces a decrease in cellular ATP levels indicative of reduced cardiomyocyte viability. Dual inhibition of PI3K and mTOR decreases cellular ATP levels and induces necrotic cardiomyocyte death.

Figure 4

Co‐knockdown of PI3Kα and RICTOR induces cardiomyocyte death. Neonatal rat cardiomyocytes were transfected with 70 nmol/L siRNAs targeting PI3Kα, IGF1R, and RICTOR alone, or in combinations. A, Analysis of cardiomyocyte viability by ATP assay presented as percent of viable cells. B, Analysis of cardiomyocyte necrosis by adenylate kinase (AK) assay presented as fold change compared with control. Scatters represent mean±SD (n=6–8) **P<0.01; ***P<0.001 (t test). IGF1R indicates insulin‐like growth factor‐1 receptor; PI3Kα, phosphoinositide 3‐kinase catalytic subunit alpha; RICTOR, Rapamycin‐insensitive companion of mTOR.

Discussion

Most of the current data concerning the role of protein kinases in maintenance of normal cardiomyocyte homeostasis are based on studies in mouse knockout models. However, predicting the KI toxicity based entirely on findings in genetically engineered mice may be misleading. Data from studies with germline knockout mice could be confounded by unique developmental roles of the gene. Conditional kinase knockout in cardiomyocytes, on the other hand, results in complete and permanent inhibition of a kinase, which is not achieved with a KI. Knockout models also disrupt protein–protein interactions that will generally not be disrupted with the drug treatment. This could cause phenotypes in the knockouts that would not be seen with KI treatment.

Currently, there are 50 KIs approved by the US Food and Drug Administration and ≈250 KIs are currently investigated in clinical trials. However, very few of the KIs target only 1 protein within physiologically relevant concentrations.13, 14 The nonclinical development of KIs as well as other anticancer products are mostly performed in line with the ICHS9 regulatory guideline, which only recommends electrocardiographic measurements and appropriate clinical observation to monitor cardiovascular safety.15 Thus, there is a high unmet need for screening methods that would identify cardiovascular toxicity, and specifically cardiomyocyte toxicity, in earlier phases of anticancer drug development. The neonatal cardiomyocytes provide a unique screening tool to study KI‐mediated cardiotoxicity because of their characteristic persistent rhythmic contraction and thus constant requirement for high level of ATP production. It has been shown that, using neonatal cardiomyocytes, many KIs in clinical use including lapatinib, erlotinib, gefitinib, imatinib, sorafenib, sunitinib, and dasatinib induced cardiomyocyte toxicity in a way that correlated with the clinical cardiotoxicity data.16, 17

Apoptosis, necrosis, and autophagy can all contribute to cardiomyocyte death induced by toxic stresses, and the type of cell death following exposure of cardiomyocytes to a given compound cannot be predicted. Thus, analysis for the amount of viable cardiomyocytes following exposure of cardiomyocytes to novel compounds is the most reliable method to detect toxicity. The technique of measuring ATP levels in viable cells is highly sensitive, reproducible, and less vulnerable to false positive results because of the fluorescence of compounds themselves that is commonly observed with fluorescence‐based cell viability assays.18 We found that the ATP assay provides a linear analysis method for measuring cardiomyocyte viability and has a very low threshold for detection of toxicity. Reduction in cardiomyocyte ATP levels may also be indicative of mitochondrial damage/dysfunction that is associated with cardiotoxicity of some KIs.19, 20 In the clinical setting, the decrease in cardiomyocyte ATP levels does not necessarily culminate in cell death, but may manifest as left ventricular dysfunction.

Some aspects need to be taken into account when extrapolating our data from neonatal cardiomyocytes to adult cardiomyocytes. Glycolysis is a predominant source of myocardial ATP production in neonatal cardiomyocytes, whereas the mature heart is mainly aerobic with free fatty acids as the predominant energy substrate.21 Moreover, cardiomyocytes have proliferative capacity in the neonatal heart, whereas the adult cardiomyocyte proliferation is very limited.22 Neonatal cardiomyocytes are more resistant to hypoxia23 and there are differences in calcium signaling between neonatal and adult cardiomyocytes.24 In the context of cell death, tumor necrosis factor α triggers apoptosis in adult cardiomyocytes but not in neonatal cardiomyocytes.25 Furthermore, there are some differences in adult and neonatal cardiomyocyte response to toxic stresses. For example, doxorubicin induces calpain‐mediated necrotic cell death in adult cardiomyocytes, whereas neonatal cardiomyocytes undergo caspase‐dependent apoptosis when exposed to doxorubicin.26, 27

In the current study, the dose–response curve for the toxicity of each inhibitor was summarized using the quantitative DSS, a metric used to determine area under the curve. Compared with other metrics such as IC50 and EC50, DSS has the advantage of being robust against sources of technical variability, because it is based on continuous interpolation and integration of multiple dose–response relationships in a high‐throughput formulation.8 By combining the data from DSS determination and KI target profiling, we identified key kinases necessary for cardiomyocyte viability. Specifically, a sparse linear regression was used to infer the kinases linked to the toxicity responses. The concentrations of KIs tested reached maximum free concentration of the compounds in patients.28 To verify the findings and to exclude possible nonspecific toxicity of the KIs, we also calculated the DSS values for each inhibitor by omitting the highest concentration for each inhibitor. Both of the analyses indicated that PI3Kα, IGF1R, and mTOR are the key top KI targets mediating the toxicity to cardiomyocytes. There are previous data indicating that cardiomyocyte damage correlates with lack of KI specificity.29 Among the studied KIs in clinical use, bosutinib and ponatinib are least selective,30 and those are also the KIs with the highest DSS value. However, for KIs in clinical use, lack of target specificity did not correlate with their toxicity to cardiomyocytes (R=0.31; P=0.14). Because cellular ATP levels may decrease without actual cell death, we assessed for rupture of the cell membrane, which is considered a hallmark of necrotic cell death. Our data indicate that only dual inhibition of PI3Kα and mTOR by a KI or kinase knockdown is consistently associated with rupture of the cardiomyocyte cell membrane. The regression analysis also identified Raf‐1 and B‐Raf among the key mediators of cardiomyocyte viability. Some Raf inhibitors, however, have been shown to paradoxically activate the MAPK pathway in cells bearing oncogenic Ras or elevated upstream receptor signaling.31, 32, 33 The data concerning the role of Raf‐1 and B‐Raf in regulating cardiomyocyte viability should therefore be interpreted with caution.

Insulin‐like growth factor I (IGF1), acting through IGF1R, is one of the key activators of the PI3K pathway, and activation of the PI3K/Akt pathway has been linked to protection from ischemic and toxic cardiac injuries.34, 35 Previously, pharmacological inhibition of PI3Kα has also been shown to abrogate the protective effects of ischemic preconditioning.36 Mammalian target of rapamycin is found in 2 distinct complexes, mTORC1 and mTORC2. Cardiomyocyte‐specific deletion of raptor (specific subunit of mTORC1 complex) disrupted cardiac function under normal physiological conditions and eventually resulted in heart failure in mice.37 Cardiomyocyte‐specific RICTOR deletion (specific subunit of mTORC2 complex), on the other hand, reduced cardiomyocyte survival and accelerated left ventricular dysfunction in mice in a pressure overload model.38 Moreover, inhibition of mTORC1 by rapamycin has been shown to protect against cardiac ischemia/reperfusion injury, while dual mTOR inhibitors abolished cardioprotection after ischemic preconditioning.39, 40 While the downstream signaling mechanisms regulating response are not fully elucidated, inhibition of mTORC2 (by RICTOR knockdown) has been shown to reduce Akt phosphorylation, while rapamycin enhances Akt phosphorylation.38, 41

The PI3K/AKT/mTOR pathway is activated at high frequency in human cancers.42, 43, 44 Given the fact that inhibiting mTORC1 by rapamycin and other rapalogs leads to activation of the PI3K/AKT pathway as a part of its negative feedback mechanism, much attention has been paid to combining mTOR and PI3K inhibitors for additional antitumor activity.45 Dual inhibition of PI3K and mTOR in human cancers inhibits Akt phosphorylation at 2 distinct phosphorylation sites, resulting in greater antitumor activity.43 Our current data, however, show that dual inhibition of PI3K and mTOR also results in reduced cardiomyocyte survival (Figure 5). Interestingly, the potency of the 4 dual PI3K/mTOR inhibitors against PI3Kα (Data set 1) highly correlated with the observed DSS value (R=0.82, P=0.14). Development of dual PI3K/mTOR inhibitors has been hampered by adverse effects of compounds, including stomatitis, noninfectious pneumonitis, rash, hyperglycemia, and immunosuppression.46 Currently, most of the clinical observations of adverse cardiac effects are limited to trials with everolimus or mTOR inhibitors,47 and cardiovascular effects of the dual PI3K/mTOR inhibition are not well known. Several of the clinically used KIs analyzed in the current study are associated with adverse cardiac effects. We found that 81% of clinically approved KIs included in the study induced toxicity to cardiomyocytes. Use of ponatinib, dasatinib, trametinib, and midostaurin, all highly toxic to cardiomyocytes in the current study, have been associated with development of left ventricular dysfunction and congestive heart failure in patients.48, 49, 50 Our data from isolated cardiomyocytes, however, are not able to detect adverse cardiovascular events stemming from KI toxicity to other resident cardiac cells (eg, endothelial cells).

Figure 5

The schematic presentation for the signaling of the top 3 kinases crucial for cardiomyocyte survival. IGF1R is upstream of the prosurvival PI3K/Akt signaling pathway. Dual PI3K/mTOR inhibitors simultaneously inhibit the phosphorylation of Akt and attenuate the prosurvival function of Akt. Akt indicates PKB, Akt serine/threonine kinase; IGF1, insulin‐like growth factor I; IGF1R, insulin‐like growth factor‐1 receptor; mTOR, mammalian target of rapamycin; mTORC2, mammalian target of rapamycin complex 2; PI3K, phosphoinositide 3‐kinase.

In summary, we developed here a novel in vitro cardiotoxicity screening method that identifies known cardiotoxic compounds and protective signaling pathways. The method is also readily applicable to other experimental models to identify key kinases mediating cell viability or other biological response. By using this approach, we identify highly adverse effects of PI3Kα/mTOR dual inhibition on cardiomyocytes that may have relevance for future drug design targeting the PI3Kα/mTOR pathway.

Sources of Funding

This work was supported by Academy of Finland grants 131020 and 297094 to Kerkelä and 268505 to Magga, and 296516 to Khan, and by Finnish Foundation for Cardiovascular Research (to Elmadani, Magga, and Kerkelä).

Disclosures

None.

Data S1. Supplemental methods.

Table S1. Dose–Response Curves of All KIs Included in the Screen

Table S2. DSS Values, Primary Kinase Targets, and Maximum Concentration Tested of All KIs Included in the Screen

Table S3. Effect of Selective PI3K

Figure S1. Validation of ATP assay.

Figure S2. Cardiomyocyte toxicity of selective inhibition of PI3K, IGF1R, mTOR, and PI3K/mTOR.

Figure S3. qPCR analysis of PI3Kα, IGF1R, and RICTOR mRNA levels.

Click here for additional data file.

Data Set 1. Values are shown as −log nmol/L.

Click here for additional data file.