CX-5461 and doxorubicin activate a shared DNA damage-associated transcriptional response in human cardiomyocytes

Abstract

CX-5461 (CX) is under investigation for the treatment of late-stage cancers. While CX was first described as an RNA polymerase I inhibitor, it has recently been shown to primarily inhibit the beta isoform of topoisomerase II. This isoform is also inhibited by anthracycline drugs including doxorubicin (DOX) and mediates the toxic effects of these drugs on the heart. It is unclear whether CX will similarly cause cardiotoxicity. We therefore tested the effects of CX on iPSC-derived cardiomyocytes from 6 individuals. CX induces cell death in cardiomyocytes at micromolar concentrations. Transcriptome profiling following treatment over time reveals gene expression programs that correspond to the DNA damage response, which are pathways shared with DOX response genes. Micromolar CX concentrations affect heart-specific genes and 14 functionally validated genes in loci associated with DOX cardiotoxicity. Our data demonstrate the impact of CX on the transcriptome of cardiomyocytes, a potential off-target cell type of the drug.

Introduction

CX-5461 (CX) is a compound that was first described as a selective inhibitor of RNA polymerase I (Haddach et al. 2012). RNA polymerase I is involved in transcribing ribosomal DNA genes, which are highly transcribed in cancer cells. CX has therefore been established as a therapeutic agent to selectively target cancer cells including those from B-lymphoma (Bywater et al. 2012), acute lymphoblastic leukemia (Negi and Brown 2015), and colorectal cancer (Otto et al. 2022). Given the in vitro efficacy of CX, it is currently in clinical trials for use in advanced solid tumors with DNA repair deficiencies (Hilton et al. 2022), and advanced hematologic cancers (Khot et al. 2019). Phase I data from both trials suggests that CX is safe for further consideration.

Following the initial description of CX as an RNA polymerase I inhibitor, CX has been implicated in other cellular processes including activation of the DNA damage response in cancer and non-cancer cells (Sanij et al. 2020; Yan et al. 2021; Lehman et al. 2022; Liu et al. 2024). Mechanisms behind the toxicity induced by CX in DNA repair-deficient tumors include action as a G-quadruplex ligand resulting in stabilization of G-quadruplex DNA structures (Xu et al. 2017).

It has recently been demonstrated that the primary mechanism of CX toxicity is through topoisomerase II (TOP2) poisoning and not RNA polymerase I inhibition (Bruno et al. 2020; Olivieri et al. 2020; Pan et al. 2021). This is perhaps not surprising given that CX was originally derived from fluoroquinolones that interact with TOP2 and G-quadruplexes (Xu and Hurley 2022). There are 2 isoforms of TOP2—TOP2A and TOP2B. CX preferentially targets the TOP2B isoform (Pan et al. 2021 ). TOP2A has also been suggested to mediate effects of G-quadruplex ligands including CX (Bossaert et al. 2021), and the DNA-damaging effects of CX in ribosomal regions (Cameron et al. 2024).

TOP2 inhibitors, including doxorubicin (DOX), are effective anti-cancer drugs; however, they can cause off-target effects on the heart given the high levels of TOP2B expression in this tissue (Zhang et al. 2012). Because CX targets TOP2B, it has been suggested that CX might cause adverse effects on the heart (Pan et al. 2021). CX treatment also induces mutations across human cell lines further questioning the safety profile of the drug (Koh et al. 2024).

The risk of cardiotoxicity is an important consideration for any drug in development. DOX-induced cardiotoxicity is a well-established clinical phenomenon that can be effectively modeled in vitro using induced pluripotent stem cell-derived cardiomyocytes (iPSC-CMs) (Burridge et al. 2016). The role of TOP2B in mediating the cardiotoxicity associated with DOX has also been demonstrated in this system (Maillet et al. 2016). We have previously shown widespread global transcriptional effects of submicromolar treatment of TOP2 inhibitors including DOX and structurally related anthracyclines epirubicin and daunorubicin, and the anthracenedione, mitoxantrone, on iPSC-CMs (Matthews et al. 2024). It is unclear whether CX, which similarly inhibits TOP2B, will induce similar effects on the transcriptome of iPSC-CMs.

We therefore designed a study to determine the effects of CX treatment on iPSC-CMs from 6 healthy individuals compared to DOX. We found that CX is cytotoxic at micromolar concentrations and induces gene expression changes shared with DOX.

Materials and methods

iPSC lines

We used human iPSC lines derived from dermal fibroblasts from 6 healthy individuals with no known disease. All lines were obtained from the publicly available iPSCORE collection, developed by Dr. Kelly A. Frazer at the University of California, San Diego, as part of the NHLBI Next Generation Consortium (Panopoulos et al. 2017). The 6 iPSC lines were generated with approval from the Institutional Review Boards of the University of California, San Diego and The Salk Institute (Project no. 110776ZF) and informed written consent of participants. Cell lines are available through the biorepository at WiCell Research Institute (Madison, WI, United States), or through contacting Dr. Kelly A. Frazer at the University of California, San Diego.

The 6 iPSC lines were derived from 3 females and 3 males of Asian ancestry, with age at donation ranging from 18 to 30 years. Detailed metadata for each individual is listed below: Individual 1 (UCSD129i-75-1/iPSCORE_75_1): female, age 30, Asian (Irani); Individual 2 (UCSD132i-78-1/iPSCORE_78_1): female, age 21, Asian (Chinese); Individual 3 (UCSD143i-87-1/iPSCORE_87_1): female, age 21, Asian (Chinese); Individual 4 (UCSD178i-17-3/iPSCORE_17_3): male, age 18, Asian (Japanese); Individual 5 (UCSD138i-84-1/iPSCORE_84_1): male, age 21, Asian (Chinese); Individual 6 (UCSD154i-90-1/iPSCORE_90_1): male, age 23, Asian (Chinese).

iPSC maintenance and cardiomyocyte differentiation

iPSCs were cultured at 37 °C in a humidified atmosphere with 5% CO₂ and ambient oxygen. Cells were maintained on Matrigel hESC-qualified Matrix (354277, Corning, Bedford, MA, United States) at a 1:100 dilution in mTeSR1 medium (85850, StemCell Technologies, Vancouver, BC, Canada), supplemented with 1% penicillin-streptomycin (30-002-Cl, Corning). Cultures were passaged every 4 to 6 d using a dissociation reagent containing 0.5 mM EDTA and 300 mM NaCl in PBS when they reached 70% to 80% confluence.

Directed differentiation of iPSCs into cardiomyocytes was performed as previously described (Matthews et al. 2024). iPSCs were seeded in 10 cm Matrigel-coated dishes and cultured to 85% to 90% confluence. On day 0, differentiation was initiated by replacing the culture medium with cardiomyocyte differentiation medium (CDM) containing 12 μM CHIR99021 trihydrochloride (4953, Tocris Bioscience, Bristol, United Kingdom). The CDM formulation included 500 mL RPMI 1640 (15-040-CM, Corning), 10 mL B-27 minus insulin supplement (A1895601, Thermo Fisher Scientific, Waltham, MA), 5 mL GlutaMAX (35050-061, Thermo Fisher Scientific), and 5 mL penicillin-streptomycin. After 24 h (day 1), the media was replaced with fresh CDM. On day 3, cells were treated with 2 μM Wnt-C59 (5148, Tocris Bioscience) in CDM to inhibit Wnt signaling. CDM was replaced on days 5, 7, 10, and 12 to support continued cardiomyocyte differentiation.

On day 14, cardiomyocytes were metabolically selected for using glucose-free, lactate-supplemented purification media. Purification media consists of 500 mL RPMI without glucose (11879, Thermo Fisher Scientific), 106.5 mg L-ascorbic acid 2-phosphate sesquimagnesium salt (sc228390, Santa Cruz Biotechnology, Santa Cruz, CA), 3.33 mL of 75 mg/mL human recombinant albumin (A0237, Sigma-Aldrich, St. Louis, MO), 2.5 mL of 1 M sodium lactate in HEPES buffer (L7022, Sigma-Aldrich), and 5 mL penicillin-streptomycin. Purification media was exchanged on days 16 and 18. On day 20, iPSC-CMs were dissociated using 0.05% trypsin-EDTA (25-053 Cl, Thermo Fisher Scientific) for 10 to 15 min and neutralized with cardiomyocyte maintenance medium (CMM), composed of 500 mL glucose-free DMEM (A14430-01, Thermo Fisher Scientific), 50 mL FBS (S1200-500, Genemate), 990 mg galactose (G5388, Sigma-Aldrich), 5 mL 100 mM sodium pyruvate (11360-070, Thermo Fisher Scientific), 2.5 mL 1 M HEPES (SH3023701, Thermo Fisher Scientific), 5 mL GlutaMAX, and 5 mL penicillin-streptomycin. Cells were sequentially filtered through 100 and 40 μm nylon mesh strainers to generate a single-cell suspension.

iPSC-CMs were quantified and plated onto 0.1% gelatin-coated culture plates in CMM media. 50,000 iPSC-CMs were plated per well of a 96-well plate for viability assays, 150,000 cells per well of a 24-well plate for immunofluorescence staining, and 600,000 cells per well of a 12-well plate for RNA-seq. Cells were matured in culture for an additional 10 d, with media replaced on days 23, 25, 27, 28, and 30.

Quantification of iPSC-CM purity

Cardiomyocyte differentiation efficiency for each individual was assessed by measuring the expression of cardiac troponin T (TNNT2) by flow cytometry. Day 27 to 31 iPSC-CMs were dissociated with 0.05% trypsin-EDTA to generate a single-cell suspension. One million cells per sample were stained with Zombie Violet Fixable Viability Dye (423113, BioLegend, San Diego, CA, United States) for 15 min at 4 °C in the dark, followed by fixation and permeabilization using the FOXP3/Transcription Factor Staining Buffer Set (00-5523, Thermo Fisher Scientific) for 30 min at 4 °C. Fixed and permeabilized cells were stained with 5 µL of PE-conjugated anti-cardiac Troponin T antibody (Clone 13-11, 564767, BD Biosciences, San Jose, CA, United States) diluted in permeabilization buffer and incubated for 45 min at 4 °C in the dark. After 3 washes in permeabilization buffer, cells were resuspended in autoMACS Running Buffer (130-091-221, Miltenyi Biotec, Bergisch Gladbach, Germany) for acquisition. Control samples included iPSC-CMs that were unlabeled, iPSC-CMs labeled with TNNT2 antibody only, and iPSC-CMs labeled with viability stain only. Flow cytometry was performed on a FACSymphony Cell Analyzer (BD Biosciences). 10,000 events were collected per sample. A set of live cells, based on the viability stain, was obtained following removal of debris and selection of singlets. iPSC-CM purity was quantified as the proportion of live, TNNT2-positive cells relative to the total live population.

Drug preparation

CX-5461 (50-196-9852; MedChem Express, Monmouth, NJ, United States) was reconstituted in DMSO to generate a 1 mM stock solution and stored at −80 °C. Doxorubicin hydrochloride (D1515; Sigma-Aldrich) was dissolved in DMSO to a 10 mM stock and stored at −80 °C. Working concentrations for both compounds were prepared in CMM media immediately prior to treatment. DMSO (vehicle; VEH) controls were volume-matched to the highest corresponding drug concentration for each treatment.

Cell viability assay

Toxicity assays were performed using iPSC-CMs from 3 individuals (Individuals 1, 2, 3) on day 27 ± 1 of differentiation. iPSC-CMs from each individual were treated on 96-well plates in quadruplicate with VEH (DMSO, 0 µM) and a range of concentrations (0.1 to 10 µM) for CX and DOX, and a 50 µM concentration for CX. Four wells contained untreated iPSC-CMs and 4 wells contained CMM media only.

Toxicity assays were performed using iPSC lines from 3 individuals (Individuals 1, 2, 3). iPSCs were seeded in Matrigel-coated 96-well plates at a density of 30,000 cells per well and cultured in mTeSR1 medium at 37 °C with 5% CO₂ until they reached ∼80% confluence. Cells were treated in quadruplicate with CX (0.1 to 50 µM), DOX (0.1 to 10 µM), or VEH (DMSO, 0 µM) for 48 h. Untreated wells and media-only wells were included as negative and background controls, respectively.

Viability assays for iPSC-CMs were performed using 3 individuals (Individuals 1, 2, 3). Cells were treated with 0.1 or 0.5 µM CX or DOX and a corresponding VEH for 3, 24, 48, or 96 h in quadruplicate. Quadruplicate untreated iPSC-CM samples and media-only samples were also included.

Fluorescent intensity was measured at 0 h and the appropriate post-treatment time using the PrestoBlue Cell Viability Reagent (A13262; Invitrogen), according to the manufacturer's instructions. Fluorescence was measured using a Synergy H1 plate reader (BioTek) at 560 nm excitation and 590 nm emission.

Fluorescence values were averaged across quadruplicate wells at each time point. The 4 media-only wells were averaged and denoted as background fluorescence at each time point. Timepoint-matched background fluorescence values were subtracted from VEH and drug-treated sample fluorescence values. These background-corrected fluorescence values were divided by the untreated background-corrected fluorescence values. Normalized, background-corrected fluorescence values at each time point were divided by normalized, background-corrected fluorescence values at 0 h to obtain viability measurements for each treatment. For toxicity assays, dose response curves were fitted for each drug in each individual using viability values and a 4-parameter log-logistic regression model (LL.4) as implemented in the drc package (v3.0-1) in R (Ritz et al. 2015). The half-maximal lethal dose (LD50) was extracted from the fitted models. LD50 values were compared across drugs using a paired 2-tailed t-test. Viability effects were also determined at individual drug concentrations (0.1, 0.5, 2.5 µM) by comparing to the VEH containing volume-matched DMSO with paired t-tests. P < 0.05 is considered significant.

γ-H2AX immunofluorescence staining and quantification

Day 27 to 31 iPSC-CMs were treated with 0.1, 0.5, 1, 2.5, 5, or 10 µM CX, DOX, or VEH for 3, 24, or 48 h. Following treatment, iPSC-CMs were fixed in 4% paraformaldehyde for 15 min at room temperature, then permeabilized with 0.25% DPBS-T (0.25% Triton X-100 in DPBS) for 10 min. After washing, cells were blocked with 5% BSA in DPBS-T for 30 min at room temperature and incubated overnight at 4 °C with anti-phospho-Histone H2A.X (Ser139) rabbit monoclonal antibody (1:500 dilution; NC1602516; Fisher Scientific) prepared in 1% BSA in DPBS-T. The following day, cells were washed and incubated for 1 h at room temperature with a donkey anti-rabbit Alexa Fluor 594-conjugated secondary antibody (1:1000 dilution; A-21207; Invitrogen). Nuclei were counterstained using Hoechst 33342 (PI62249; Thermo Scientific) for 10 min in the dark.

Immunofluorescence images were acquired under consistent exposure settings across conditions. One hundred to 600 nuclei were counted across 2 or 3 fields of view per treatment and timepoint, and scored for presence or absence of γ-H2AX using a custom macros in ImageJ version 1.54i (Schneider et al. 2012). The percentage of γ-H2AX-positive nuclei was calculated by dividing the number of positive nuclei by the total number of nuclei. The percentage of drug-treated iPSC-CMs positive for γ-H2AX was compared to the percentage in VEH-treated iPSC-CMs by a 2-tailed paired t-test. P < 0.05 is considered significant.

RNA extraction

Day 30 iPSC-CMs were treated with 0.1 or 0.5 µM CX, DOX or VEH for 3, 24, or 48 h. Cells were flash-frozen and stored at −80 °C. Total RNA was extracted from flash-frozen iPSC-CMs using the RNeasy Mini Kit 250 (74106; QIAGEN, Germantown, MD, United States), following the manufacturer's instructions. Extractions were performed in treatment- and timepoint-balanced batches of 12, with all conditions from a given individual processed in 1 batch. RNA concentration and integrity were assessed using the Agilent 2100 Bioanalyzer. All samples exhibited high-quality RNA with RNA Integrity Number (RIN) scores exceeding 8.0. Median RIN values across treatments for individuals 1 to 6 were 9.95, 9.75, 9.9, 9.05, 9.1, and 9.7, respectively.

Day 30 iPSC-CMs from Individuals 1 and 3 were treated with 2.5 µM CX or VEH in triplicate for 3, 24, or 48 h and processed as described above. Median RIN values across treatments for Individuals 1 and 3 were both 9.95.

RNA-seq library preparation

Polyadenylated RNA was isolated from 150 ng of total RNA using the NEBNext Poly(A) mRNA Magnetic Isolation Module (E7490L, New England Biolabs, Ipswich, MA, United States), and RNA-seq libraries were prepared using the NEBNext Ultra II Directional RNA Library Prep Kit with Sample Purification Beads (E7765S; NEB) and indexed using NEBNext Multiplex Oligos for Illumina (96 Unique Dual Index Primer Pairs, E6440S).

Sub-micromolar drug concentration libraries were prepared in 2 treatment- and timepoint-balanced batches consisting of all samples per individual for 3 individuals (Batch 1: Individuals 4 to 6; Batch 2: Individuals 1 to 3). Library quality was assessed using the Agilent 2100 Bioanalyzer. All 108 libraries were quantified, pooled, and sequenced across 3 AVITI paired-end sequencing runs (2 × 75 bp), yielding a median of 26,402,233 paired-end read pairs across samples.

Micromolar CX drug concentration libraries for Individuals 1 and 3 were prepared in 1 batch. All 36 samples were quantified, pooled, and sequenced on 1 AVITI paired-end sequencing run (2 × 75 bp), yielding a median of 33,065,154 paired-end read pairs across samples.

RNA-seq analysis

Raw sequencing reads were assessed for quality using MultiQC (v1.26) (Ewels et al. 2016). Reads were aligned to the human reference genome (GRCh38/hg38) using the Rsubread package (v2.16.1) (Liao et al. 2013), and gene-level counts were quantified using the featureCounts function with built-in exon-based annotations (Liao et al. 2014). Count matrices were imported into R for downstream processing. All custom analysis scripts used for this project are available at https://github.com/mward-lab/Paul_CX_2025 made possible by the workflowr package (Blischak et al. 2019).

Counts were transformed into log₂ counts per million (log₂cpm) using the cpm() function from the edgeR package (v4.0.16) (Chen et al. 2016). For the submicromolar CX and DOX treatments, we excluded genes with a mean log₂cpm ≤ 0 across all 108 samples, yielding 14,279 expressed genes for downstream analysis. The same approach for the micromolar CX treatments yielded 13,862 genes across 36 samples.

Gene expression correlation between iPSC-CMs and human tissues

Gene expression data across 24 human tissues were obtained from Jiang et al. (Jiang et al. 2020). We calculated the median log₂cpm expression across all VEH-treated iPSC-CM samples. Pearson correlations were computed between VEH iPSC-CMs and each tissue using the 14,279 genes expressed in iPSC-CMs. Correlation values were visualized as a heatmap using the ComplexHeatmap package (v2.18.0) (Gu 2022).

Principal component analysis

We performed principal component analysis on log₂cpm values using the prcomp() function in the R Stats package (v4.3.0) (Team 2023), with log₂cpm values centered prior to analysis. The top principal components (PCs) were visualized using ggfortify (v0.4.17) (Tang et al. 2016). Five covariates in the study: individual, sex, drug treatment, drug concentration, and treatment time were assessed for association with the top PCs using a linear model. P values were derived from the F-statistic, where P < 0.05 is considered significant.

Differential gene expression analysis

We identified differentially expressed genes (DEGs) across all treatment conditions using the edgeR-voom-limma pipeline (Law et al. 2016) and raw counts from the set of expressed genes. Raw counts were normalized using the trimmed mean of M-values (TMM) method to account for differences in library size and composition and transformed with voom to estimate and account for the mean–variance relationship. DEGs were determined by linear modeling and empirical Bayes moderation. CX and DOX treatments were contrasted against the time- and concentration-matched VEH samples. Individual was treated as a random effect using the duplicate correlation function. Genes with an adjusted P < 0.05 (Benjamini–Hochberg correction) are classified as differentially expressed.

Gene expression cluster analysis

We jointly modeled gene expression responses using the Cormotif package in R (v1.48.0) (Wei et al. 2015), which implements a Bayesian clustering approach to identify shared expression patterns, or correlation motifs, across multiple comparisons. TMM-normalized log₂cpm values were used as input. We paired each drug treatment with its corresponding VEH control at 3, 24, and 48 h. Separate models were constructed for the 0.1 and 0.5 µM datasets, each encompassing 6 pairwise comparisons. For both concentrations, we fit models across a range of motif numbers (K = 1 to 8) and selected the optimal model based on Bayesian information criterion and Akaike information criterion. Following model fitting, we extracted gene-wise posterior probabilities of differential expression from the best-fit Cormotif model for each concentration. Genes were classified into discrete clusters based on their joint probability patterns. Genes with posterior probabilities < 0.5 across all treatments were designated as the Non-response cluster for both drug concentrations. For the 0.1 µM concentration: the CX_DOX_1 cluster is defined as genes with posterior probability > 0.5 in DOX_24 hr, DOX_48 hr, CX_24 hr, CX_48 hr, and P < 0.5 in DOX_3 hr and CX_3 hr. The DOX-specific_1 cluster is defined by posterior probability > 0.5 in DOX_24 h, DOX_48 hr, and P < 0.5 in DOX_3 hr and all CX timepoints. For the 0.5 µM concentration: the CX_DOX_2 cluster is defined by posterior probability > 0.5 in CX_24 hr, CX_48 hr, and P ≥ 0.1 for CX_3 hr, P ≥ 0.02 for DOX_3 hr, P < 0.5 for DOX_24 hr and DOX_48 hr. The CX_DOX_3 cluster has P > 0.5 in DOX_24 hr, DOX_48 hr, CX_24 hr, CX_48 hr, 0.02 ≤ P < 0.5 for DOX_3 hr and P < 0.5 CX_3 hr. DOX-specific_2 has P > 0.5 in DOX_3 hr, DOX_24 hr, P ≥ 0.02 in DOX_48 hr and P < 0.5 in all CX timepoints. DOX-specific_3 has P > 0.5 for DOX_24 hr, DOX_48 hr, P < 0.5 for CX_3 hr, CX_24 hr, CX_48 hr, DOX_3 hr.

Comparison of DEGs with CX DEGs in colorectal cancer cells

We obtained RNA-seq data from colorectal cancer cells treated with 0.5 µM CX or VEH for 6, 24, or 48 h (Otto et al. 2022). First, we compared the log₂ fold change values for the colorectal cancer cells with our data in iPSC-CMs. We calculated the Pearson correlation coefficient across all drug-VEH comparisons. Second, we obtained the set of DEGs (FDR < 0.05) in colorectal cancer cells at each time point. We combined the DEGs across timepoints to obtain a set of 1,228 unique DEGs. We calculated the proportion of colorectal cancer CX DEGs amongst our set of DEGs across treatments. χ² tests were used to compare overlap proportions between CX and DOX treatments at matched concentrations and timepoints. P < 0.05 is denoted as significant.

Gene ontology enrichment analysis

We performed functional enrichment analysis for both DEGs and response clusters. For each gene set, enrichment of GO terms in the Biological Process (BP) category was performed using the clusterProfiler package (v4.10.1) (Wu et al. 2021) in R. G-quadruplex terms were obtained from the Molecular Function category. The background gene set consisted of all expressed genes in the dataset (n = 14,279 or n = 13,862). GO terms with FDR-adjusted P < 0.05 are considered significant.

DNA damage response gene expression analysis

We obtained the set of 69 DNA damage-associated genes from the Molecular Signatures Database (MSigDB v7.5.1) (Liberzon et al. 2015). 65 genes are expressed in our data. We calculated the enrichment of DNA damage genes in the set of DEGs for each treatment, compared to the set of genes that are not DEGs for that treatment, by using the odds ratio from a Fisher's exact test. DEG sets with P < 0.05 are considered enriched. The effect sizes (log₂ fold change) from the differential expression analysis across treatments were visualized for each gene.

P53 target gene expression analysis

We obtained a set of 346 p53 target genes (Fischer 2017). 300 genes are expressed in our data. We calculated the enrichment of p53 target genes in the set of DEGs for each treatment, compared to the set of genes that are not DEGs for that treatment, by using the odds ratio from a Fisher's exact test. DEG sets with P < 0.05 are considered enriched.

Heart-specific gene enrichment analysis

We obtained a set of 419 genes that are specifically expressed in the heart based on transcriptomic data (Uhlén et al. 2015). Genes were defined as heart-specific if they exhibited a ≥ 4-fold higher mRNA expression in heart tissue compared to all other human tissues. We calculated the enrichment of heart-specific genes in the set of DEGs for each treatment, compared to the set of genes that are not DEGs for that treatment, by using the odds ratio from a Fisher's exact test. DEG sets with P < 0.05 are considered enriched.

Tissue-specificity analysis

We obtained heart left ventricle tissue-specificity scores from GTEx RNA-seq data (Jiang et al. 2020). We collated tissue-specificity scores for all CX DEGs across concentrations. We compared scores between DEG sets using Wilcoxon rank-sum tests, where significant differences are denoted when P < 0.05.

Gene expression response in DOX-induced cardiotoxicity loci

We obtained a set of DOX-induced cardiotoxicity genes that have been functionally validated in iPSC-CMs (Fonoudi et al. 2024). 25 genes are expressed in our data. We calculated the enrichment of cardiotoxicity genes in the set of DEGs for each treatment, compared to the set of genes that are not DEGs for that treatment, by using the odds ratio from a Fisher's exact test. DEG sets with P < 0.05 are considered enriched. The effect sizes (log₂ fold change) from the differential expression analysis across treatments were also visualized for this gene set.

Gene expression response in heart failure-associated loci

We obtained a set of 273 loci associated with heart failure through the GWAS catalog (Cerezo et al. 2025), and extracted the mapped genes. 118 unique mapped genes are expressed in our data. We calculated the enrichment of heart failure genes in the set of DEGs for each treatment, compared to the set of genes that are not DEGs for that treatment, by using the odds ratio from a Fisher's exact test. DEG sets with P < 0.05 are considered enriched. The effect sizes (log₂ fold change) from the differential expression analysis across treatments were visualized for the set of top 30-associated genes.

Results

CX is cytotoxic to iPSC-CMs at micromolar concentrations

To assess the cardiotoxic potential of CX relative to DOX, we used an in vitro model based on iPSCs derived from 6 healthy individuals (3 females and 3 males) with no known cardiac disease or cancer diagnosis (Fig. 1a). We differentiated the 6 iPSC lines into beating cardiomyocytes. The majority of iPSC-CMs from each individual express the cardiac-specific marker, cardiac troponin T, as measured by flow cytometry (median purity = 98.4%; Supplementary Fig. 1 in File 1).

Fig. 1.

CX is cytotoxic at micromolar concentrations. a) Experimental design of the study. iPSCs derived from 3 males and 3 females were differentiated into cardiomyocytes (iPSC-CMs) and treated with TOP2 inhibitor drugs CX-5461 (CX) or doxorubicin (DOX) or a vehicle control (VEH) for 3, 24 or 48 h. Effects on cell viability, DNA damage, and gene expression were measured. b) Proportion of viable cardiomyocytes following exposure to increasing concentrations of CX (blue) or DOX (pink) for 48 h in 3 individuals. Viability data were collected in quadruplicate for each drug concentration, normalized, and expressed relative to pretreatment values. Dose response curves derive from mean viability measurements fit with a 4-parameter logistic regression. c) LD50 values following 48 h of CX treatment in iPSC-CMs and iPSCs from 3 individuals together with previously reported LD50 values for a panel of CX sensitive and resistant ovarian cancer cell lines and a normal immortalized ovarian epithelial cell line (Cornelison et al. 2017; Sanij et al. 2020).

To evaluate drug-induced cytotoxicity, we treated iPSC-CMs from 3 individuals with a range of concentrations of CX (0 to 50 µM) and DOX (0 to 10 µM) and a VEH control for 48 h and assessed cell viability using a resazurin-based fluorometric assay. Both drugs induced dose-dependent decreases in cell viability across individuals. However, DOX is approximately 20 times more cytotoxic than CX (median LD50 DOX = 0.53 µM, median LD50 CX = 10.03 µM; 2-tailed paired t-test; P = 0.02; Fig. 1b). To gain insight into the cell-type specificity of CX effects on viability, we treated iPSCs from the same 3 individuals with a range of concentrations of CX for the same time period. No viable cells were detected at concentrations greater than 0.75 µM CX treatment and the LD50 was in the submicromolar range for all individuals (median LD50 CX = 0.2 µM, Fig. 1c). The LD50 values we obtained for iPSCs are in line with values previously reported for CX-sensitive ovarian cancer cell lines (Fig. 1c) (Cornelison et al. 2017; Sanij et al. 2020). iPSCs treated with CX or DOX yield similar submicromolar LD50 values (Supplementary Fig. 2 in File 1).

The concentration of CX in plasma from patients treated with the drug ranges from 0.5 to 3.5 µM (Khot et al. 2019). We therefore selected 2 submicromolar, clinically relevant concentrations of CX, 0.1 and 0.5 µM, that have been used in prior in vitro studies, for further characterization (Pan et al. 2021; Hilton et al. 2022). DOX concentrations in plasma have been reported in the range of 0.002 to 1.73 µM and the effects of submicromolar concentrations of DOX on cardiomyocytes have been extensively studied (Burridge et al. 2016; Magdy et al. 2021; Matthews et al. 2024). For example, there are thousands of gene expression changes within 24 h of 0.5 µM DOX treatment in cardiomyocytes (Matthews et al. 2024). To investigate drug effects over time, we treated cells with either 0.1 or 0.5 µM CX or DOX or VEH for 3, 24, or 48 h.

CX elicits an attenuated transcriptomic response compared to DOX at equimolar concentrations

We first measured the effects of 0.1 and 0.5 µM CX and DOX treatments on iPSC-CM viability over 3, 24, and 48 h. At these concentrations we did not observe an effect on viability at any treatment time except following 48 h of 0.5 µM DOX treatment (Fig. 2a). Both DOX and CX have been reported to induce cytotoxicity by initiating DNA damage in cancer cells. While DOX has also been shown to induce DNA damage in iPSC-CMs, the effects of CX have not been tested. To investigate whether CX has genotoxic effects at sublethal concentrations, we quantified the expression of γ-H2AX, a marker of DNA double-strand breaks, in drug-treated iPSC-CMs by immunofluorescence staining. There is no difference in the proportion of γH2AX-positive cells in VEH- and CX-treated cells across 3-, 24- and 48-h time points and 0.1 and 0.5 µM concentrations (Fig. 2b, c). However, DOX induces an increase in DNA damage at all timepoints following 0.5 µM treatments and following 0.1 µM treatments at 24 and 48 h. When increasing the concentration of CX 10 to 20 times (5 and 10 µM), there is still no difference in the level of DNA damage between CX- and VEH-treated cells (Supplementary Fig. 3 in File 1).

Fig. 2.

At equimolar concentrations CX induces minimal effects on the iPSC-CM transcriptome compared to DOX. a) Cell viability following treatment with 0.1 and 0.5 µM CX and DOX and corresponding VEH for 3, 24, and 48 h. Data representative of 3 individuals. Asterisk represents a statistically significant difference in viability between drug and VEH treatment (*P < 0.05). b) Immunostaining of the DNA damage marker, γ-H2AX (red), and Hoechst nuclear stain (blue) in iPSC-CMs from a representative individual (Individual 1) treated with 0.5 µM CX, DOX or VEH over time. Scale bar: 100 μm. c) Percentage of VEH-, CX-, and DOX-treated iPSC-CMs from Individuals 1, 2, and 3 that stain positive for γ-H2AX following 3, 24, and 48 h of treatment with 0.1 or 0.5 µM CX, DOX, and VEH. Data representative of 100 to 600 cells per treatment per individual. Asterisk represents a statistically significant difference in γ-H2AX expression between drug and VEH treatment (*P < 0.05). d) Gene expression profiles (log₂cpm) across 14,279 expressed genes from 108 samples representing 6 individuals (1,2,3,4,5,6), 2 sexes (male, female), 3 treatments (DOX [red], CX [blue], and VEH [yellow]), 2 drug concentrations (0.1 µM [pale shade], 0.5 µM [dark shade]), and 3 treatment times (3 h: circles; 24 h: triangles; 48 h: squares). e) Proportion of differentially expressed genes (DEGs) between drug (CX and DOX) and VEH at either 0.1 or 0.5 µM drug concentrations following 3, 24, or 48 h of treatment amongst all expressed genes. DEGs are categorized by whether they are upregulated in response to drug (up) or downregulated in response to drug (down). f) Correlation between drug responses across drugs. The log₂ fold change of expression between each drug treatment and VEH was calculated for all 14,279 expressed genes at each concentration and timepoint. Values were compared across all treatment groups by hierarchical clustering of the pairwise Pearson correlation values. Drugs are colored by type (CX: blue; DOX; pink), concentration (0.1 µM: teal; 0.5 µM: magenta), and treatment time (3 h: light green; 24 h: orange; 48 h: purple). g) Top biological processes enriched across DEGs (adj. P < 0.05). The top 5 biological processes enriched in each drug treatment were collated, and the unique set interrogated across DEGs. Enrichment in each treatment is represented as −log₁₀  P values. Asterisk represents processes significantly enriched in a set of DEGs (adj. P < 0.05).

To characterize the impact of CX treatment on the transcriptome of human cardiomyocytes, we performed bulk RNA sequencing on iPSC-CMs from 6 individuals treated with 0.1 and 0.5 µM CX, DOX, or VEH for 3, 24, and 48 h, yielding data from 108 samples (Supplementary Table 1). Sequencing metrics indicated high data quality across all treatments, with a median yield of 26,402,233 read pairs across samples, and uniform read depth across treatment, concentration, and timepoint groups (Supplementary Fig. 4a to c in File 1; Supplementary Table 2). Following genome alignment, counting of reads mapping to genes, and removal of lowly expressed genes, we obtained a final set of 14,279 expressed genes.

To determine the physiological relevance of our in vitro human cardiomyocyte model to human heart tissue, we correlated our iPSC-CM gene expression data (median VEH log₂cpm) with gene expression data from human tissues (Jiang et al. 2020). Amongst the 24 tissues tested, iPSC-CMs are most similar to heart tissue (r = 0.81 for right atrium, r = 0.79 for left ventricle; Supplementary Fig. 5a in File 1). Similarly, iPSC-CMs express a range of cardiac genes including TTN, MYH6, ACTN2, and RYR2 (Supplementary Fig. 5b in File 1). When considering the gene expression data from all 108 samples, the data clusters primarily based on whether the iPSC-CMs were treated with DOX for 24 or 48 h or not (Supplementary Fig. 6 in File 1). Principal component analysis reveals that PC1, accounting for 31.6% of the variance, associates with drug type, drug concentration, and treatment time (Fig. 2d and Supplementary Fig. 7 in File 1). The second principal component, representing 15.6% of the variation is associated with sex and individual indicating that the primary sources of variation in the data associate with known biological factors (Supplementary Fig. 7 in File 1).

To quantify the impact of drug treatment on the iPSC-CM transcriptome, we identified DEGs between each drug treatment and the time-matched VEH (adj. P < 0.05; Fig. 2e; Supplementary Fig. 8 in File 1; Supplementary Tables 3 to 14). CX treatment induced hundreds of DEGs across drug concentrations and treatment time (3 h: 0.1 µM CX vs VEH n = 1; 0.5 µM CX vs VEH n = 2; 24 h: 0.1 µM CX vs VEH n = 205; 0.5 µM CX vs VEH n = 278; 48 h: 0.1 µM CX vs VEH n = 318; 0.5 µM CX vs VEH n = 370). DOX treatment induced thousands of DEGs (3 h: 0.1 µM DOX vs VEH n = 2; 0.5 µM DOX vs VEH n = 533; 24 h: 0.1 µM DOX vs VEH n = 5,064; 0.5 µM DOX vs VEH n = 9,828; 48 h: 0.1 µM DOX vs VEH n = 3,461; 0.5 µM DOX vs VEH n = 10,128). While DOX treatment results in a roughly even distribution of DEGs that are up- and downregulated at 24 and 48 h, the majority of DEGs induced by CX at these timepoints are downregulated (93% of all CX DEGs and 51% of all DOX DEGs; Fig. 2e and Supplementary Fig. 8 in File 1). Comparison of expression profiles across DEG sets reveals clustering primarily by treatment type and some variation in response between individuals (Supplementary Fig. 9 in File 1). Inter-individual variability is not unexpected given that the individual from which iPSCs are derived is a major source of transcriptional variability (Panopoulos et al. 2017), and that there are genetic variants associated with the transcriptional response to drugs including DOX (Knowles et al. 2018).

To directly compare drug responses across drug types and time we calculated all pairwise correlations of effect sizes (log₂ fold change). Samples separate primarily by whether they were treated for 3 h or not, regardless of drug type or concentration (Fig. 2f). Three-hour responses are generally weakly correlated consistent with the low number of DEGs at this timepoint (rho = 0.36 to 0.40). iPSC-CMs treated with 0.1 or 0.5 µM DOX or 0.5 µM CX are most similar in this group. Amongst the 24- and 48-h treatment time samples, 0.5 µM DOX-treated samples cluster separately from the other treatments and are the most strongly correlated responses (rho = 0.91), followed by 0.1 µM DOX-treated samples (rho = 0.89). CX responses are closest to low dose DOX responses and show strong correlations amongst each other (rho = 0.75 to 0.82).

To gain insight into whether the gene expression changes induced by CX in iPSC-CMs are similar to those in cancer cells, we obtained DEGs and treatment effect sizes from colorectal cancer cells treated with 0.5 µM CX for 6, 24, or 72 h (Otto et al. 2022). Comparing our CX responses to responses in cancer cells revealed correlated effect sizes at the 24- and 48-h timepoints (median Pearson correlation = 0.22 to 0.34; P < 0.05; Supplementary Fig. 10a in File 1). Our DOX treatment effect sizes are also correlated with CX treatment in cancer cells albeit lower than that for CX (correlation = 0.13 to 0.22; P < 0.05; Supplementary Fig. 10a in File 1). These results suggest a degree of sharing of CX responses across cancer cells and cardiomyocytes, and across CX and DOX treatment. Indeed, we find that 33 to 40% of our 24- and 48-h iPSC-CM CX DEGs are DEGs in at least 1 time point following CX treatment in colorectal cancer cells (n = 1,228 DEGs), and that the proportion is significantly higher than those overlapping DOX DEGs at these timepoints (χ² test; P < 0.05; Supplementary Fig. 10b in File 1).

We next asked which biological processes are enriched amongst DEGs compared to all expressed genes (Fisher's exact test; adj. P < 0.05). The top processes enriched amongst CX DEGs relate to chromosome segregation and DNA replication (Fig. 2g). These processes are enriched across drug concentrations at 24- and 48-h timepoints, and are also amongst the most enriched in the 0.1 µM DOX-treated samples over time. Many biological processes are shared across CX and 0.1 µM DOX DEGs (Supplementary Fig. 11 in File 1). Samples treated with 0.5 µM DOX for 3 h are enriched for processes related to heart development, while there are no enriched processes at the later time points likely due to the large proportion of DEGs.

CX induces a subset of DNA damage response genes but not RNA polymerase I-associated genes

We identified hundreds of biological processes enriched amongst CX DEGs. We next asked whether 3 broad functional categories previously associated with CX treatment in cancer cells are contained in this set. First, we considered that CX selectively inhibits RNA polymerase I-mediated transcription in cancer cells. We did not find any of the 8 “transcription by RNA polymerase I”-associated terms enriched amongst CX or DOX DEGs suggesting cell-type specificity in this effect (Supplementary Fig. 12a in File 1). Second, we considered the ability of CX to stabilize G-quadruplex DNA structures and asked whether 6 G-quadruplex-associated terms are enriched in our data. The “G-quadruplex binding” function is enriched in the set of DEGs identified post-treatment with 0.5 µM CX at the 24- and 48-h timepoints, suggesting that this activity is involved in iPSC-CMs as well as in previously identified cell types (Supplementary Fig. 12b in File 1). The genes associated with G-quadruplex binding that are CX DEGs include DDX11, BLM, and PIF1. This function is not associated with any DOX DEG sets. Third, we considered 8 “DNA damage response”-associated terms. We found DNA repair and DNA damage signal transduction processes enriched amongst CX DEGs at 24- and 48-h timepoints across concentrations (Supplementary Fig. 12c in File 1). These terms are also enriched amongst 0.1 µM DOX DEGs. Considering the overlap in functions of a subset of DOX and CX DEGs, we next jointly modeled the data across drugs.

CX treatment induces gene expression changes shared with DOX

To directly compare CX and DOX-induced transcriptional changes, we jointly modeled the data to assess gene expression trajectories over time in a concentration-dependent manner (see Materials and methods). We determined that 3 clusters represent the predominant patterns in the data from the samples treated with 0.1 µM CX, DOX, and VEH (Supplementary Fig. 13a in File 1), and 5 clusters represent the samples treated with 0.5 µM CX, DOX, and VEH (Supplementary Fig. 13b in File 1). At 0.1 µM, the majority of genes have a low probability of being differentially expressed across timepoints for both CX and DOX, and are therefore classified as nonresponse genes (n = 12,308; Fig. 3a; Supplementary Table 15). There are 1,551 genes that have a high probability of differential expression following 24 and 48 h of DOX treatment only, which are denoted as the DOX-specific_1 cluster. There are no gene clusters that show CX-specific responses. Instead, there are 415 genes that respond following 24 and 48 h of CX and DOX treatment, denoted as the CX-DOX_1 cluster.

Fig. 3.

CX treatment induces gene expression changes shared with DOX. a) Gene expression clusters following joint modeling of pairs of tests from the 0.1 µM drug-treated samples. Posterior probabilities of genes being differentially expressed following a treatment are represented by white to black shading. Genes are categorized by their posterior probability in each treatment. The gray cluster represents Non-response genes that have a low probability of differential expression across drugs. The lilac cluster represents genes that respond to CX and DOX. The magenta cluster represents genes that respond only to DOX. b) Gene expression clusters following joint modeling of pairs of tests from the 0.5 µM drug-treated samples. Clusters are defined as described in (a). c) The proportion of DOX response genes that also respond to CX following 0.1 µM drug treatment. d) The proportion of DOX response genes that also respond to CX following 0.5 µM drug treatment. e) The log₂ fold change across CX and DOX treatments compared to VEH for genes in each cluster identified in the 0.1 µM drug-treated samples. f) The log₂ fold change across CX and DOX treatments compared to VEH for genes in each cluster identified in the 0.5 µM drug-treated samples. g) Top biological processes enriched across gene clusters (adj. P < 0.05). The top 5 biological processes enriched in each cluster were collated, and the unique set interrogated across clusters. Enrichment in each treatment is represented as −log₁₀  P values. Asterisk represents processes significantly enriched in a cluster (adj. P < 0.05).

Amongst the samples treated with 0.5 µM CX, DOX, and VEH, there are 7,134 nonresponse genes (Fig. 3b; Supplementary Table 16). Similar to the gene expression profiles from the 0.1 µM data, there are 6,450 genes that respond to DOX only at 24 and 48 h (DOX-specific_3 cluster). The higher concentration of DOX results in an additional gene cluster including genes that respond strongly at 3 and 24 h and weakly at 48 h (DOX-specific_2; n = 179). At this higher concentration of drug there is still no CX-specific response. There is a cluster of genes that responds to CX at 24 and 48 h and weakly to DOX at 3 h and strongly at 24 and 48 h (CX-DOX_3; n = 221), and another that responds weakly to CX at 3 h and strongly at 24 and 48 h and weakly to DOX at 3 h (CX-DOX_2; n = 142). The proportion of DOX response genes that also respond to CX is greater in the 0.1 µM drug group than the 0.5 µM drug group (χ² test; P < 0.05; Fig. 3c, d).

The genes in each cluster show the expected response to each drug based on the log₂ fold change distributions between pairwise drug and VEH comparisons (Fig. 3e, f). The Non-response groups show small drug effect sizes across drug concentrations, while the DOX-specific response clusters show increased absolute effect sizes in the DOX-treated groups. Clusters of genes with expression effects induced following both CX and DOX treatment tend to be downregulated in line with the majority of CX DE genes being downregulated. Individual genes in each of the response categories show the expected gene expression patterns (Supplementary Fig. 14 in File 1). For example, TRIP13 is a CX-DOX_1 response gene that is downregulated in response to both CX and DOX at 24 and 48 h following 0.1 µM treatment (Supplementary Fig. 14a in File 1), while BRCA1 is a CX-DOX_3 response gene that is downregulated at 24 and 48 h following 0.5 µM treatment (Supplementary Fig. 14b in File 1). These results show that CX-induced effects on gene expression are shared with DOX at each timepoint. We next asked whether the genes that respond to 0.1 µM CX treatment also respond to 0.5 µM CX treatment. Of the 457 genes that respond to CX at either concentration, 321 (70%), are shared across concentrations suggesting robust effects on gene expression (Supplementary Fig. 15 in File 1).

We next asked which biological processes are enriched amongst the 8 drug response clusters. Similar to the pairwise DEG analysis, we find terms related to chromosome segregation and DNA repair amongst the response clusters shared between DOX and CX at both drug concentrations (adj. P < 0.05; Fig. 3g; Supplementary Fig. 16 in File 1). In contrast, the 0.5 µM DOX-specific_3 response cluster is enriched for terms including GPCR signaling and cardiac development.

Micromolar CX concentrations induce similar gene expression changes to DOX

Given that we observed at most 380 DEGs following submicromolar concentrations of CX treatment across timepoints, we increased the concentration of CX used to treat iPSC-CMs 5-fold to 2.5 µM (Fig. 4a). This concentration is below the maximum tolerated dose of 4 µM reported by an early clinical trial, and within the 4 ± 2 µM plasma concentration measurement range in a later trial (Khot et al. 2019; Hilton et al. 2022). We therefore tested the effects of 2.5 µM CX and VEH treatment over 3, 24, and 48 h. There is no difference in cell viability between CX- and VEH-treated iPSC-CMs across 3 individuals at any of these timepoints (Fig. 4b). Similarly, there is no effect on viability when increasing the treatment time to 96 h (Supplementary Fig. 17 in File 1). There is also no difference in the level of DNA damage between CX- and VEH-treated iPSC-CMs across time (Fig. 4c). We profiled the transcriptome following 3, 24, and 48 h of treatment in triplicate for 2 individuals (Individuals 1 and 3; Supplementary Tables 1, 2). At this concentration we observed that 24 and 48 h of CX treatment is associated with the primary source of variation in the global gene expression data (Fig. 4d). We observed thousands of gene expression changes for Individual 1 (3 h: 2.5 µM CX vs VEH n = 2,205; 24 h: 2.5 µM CX vs VEH n = 3,439; 48 h: 2.5 µM CX vs VEH n = 4,804; Fig. 4e and Supplementary Fig. 18 in File 1 and Supplementary Tables 17 to 19). The 24- and 48-h 2.5 µM DEGs are enriched in terms similar to those identified to be enriched in the lower CX concentration DEGs including chromosome segregation and cell cycle (Fig. 4f). These results are recapitulated when considering data collected from Individual 3 (Supplementary Fig. 19 in File 1; Supplementary Tables 20 to 22).

Fig. 4.

Micromolar CX treatments induce DEGs that resemble DOX. a) Experimental design to determine the effects of micromolar concentrations of CX on iPSC-CMs. b) Cell viability following treatment with 2.5 µM CX and corresponding VEH for 3, 24, and 48 h. Data representative of 3 individuals (1, 2, 3). c) Percentage of iPSC-CMs from Individuals 1, 2, and 3 that stain positive for γ-H2AX following 3, 24, and 48 h of treatment with 2.5 µM CX and VEH. Data representative of 100 to 600 cells per treatment per individual. d) Gene expression profiles (log₂cpm) across 13,862 expressed genes from 18 samples representing Individual 1 including 2 treatments (CX [blue], VEH [yellow]), and 3 treatment times (3 h: circles; 24 h: triangles; 48 h: squares) where each treatment was performed in triplicate (1, 2, 3). e) Proportion of DEGs between 2.5 µM CX and VEH at 3, 24, or 48 h of treatment amongst all expressed genes. DEGs are categorized by whether they are upregulated in response to drug (up) or downregulated in response to drug (down). f) Top biological processes enriched across DEGs (adj. P < 0.05). The top 5 biological processes enriched in each drug treatment were collated, and the unique set interrogated across DEGs. Enrichment in each treatment is represented as −log₁₀  P values. Asterisk represents processes significantly enriched in a set of DEGs (adj. P < 0.05). g) Overlap of DEGs for each treatment at 3 h. h) Overlap of DEGs for each treatment at 24 h. i) Overlap of DEGs for each treatment at 48 h.

We next considered how the micromolar CX response genes compare to the response genes from the treatments with submicromolar concentrations of CX and DOX across time. While only 7% of 2.5 µM CX DEGs are shared with 0.1 or 0.5 µM DOX DEGs following 3 h of treatment, 81% of CX DEGs are shared with DOX following 24 h of treatment and 79% following 48 h of treatment (Fig. 4g to i). These results suggest convergence of response pathways induced by these drugs over time.

DNA damage response genes are enriched in CX and DOX DEGs

Given the known role of DOX in inducing DNA damage, and the enrichment of DNA damage response terms across CX DEG sets, we determined the transcriptional response of a curated set of DNA damage response genes from the Molecular Signatures Database (Liberzon et al. 2015). As expected, across both concentrations, DOX DEGs are enriched for DNA damage response genes compared to genes not classified as DEGs following 24 or 48 h of treatment (P < 0.05; Fig. 5a). CX DEGs are enriched for DNA damage response genes following 24 and 48 h of treatment at all concentrations (P < 0.05; Fig. 5a). The majority of DNA damage response genes that are affected by CX and DOX across concentrations are downregulated and include the cell cycle checkpoint-related genes (CCNB1, CDC25C, and CDK1) and DNA repair genes (BRCA1, RAD51, and CHEK2) for example (Fig. 5b). Following treatment with 2.5 µM CX there is an increase in the expression of apoptotic markers including FAS and TRAILR2 similar to the response across DOX concentrations. We therefore asked whether these genes are enriched for p53 target genes (Fischer 2017). P53 target genes are enriched in 24- and 48-h DOX-responsive gene sets and 2.5 µM CX-responsive genes, but not submicromolar CX-responsive genes (Fig. 5c).

Fig. 5.

CX treatment affects DNA damage response gene expression. a) Enrichment of DNA damage response genes amongst CX and DOX response genes (Liberzon et al. 2015). Asterisk represents those DEG sets that are enriched for DNA damage response genes compared to non-DEGs (*P < 0.05; ***P < 0.01). b) Drug responses (log₂ fold change) of DNA damage response genes. Asterisk represents genes that are classified as DEGs (adj. P < 0.05). c) Enrichment of p53 target genes amongst CX and DOX response genes (Fischer 2017). Asterisk represents those DEG sets that are enriched for p53 target genes compared to non-DEGs (*P < 0.05; ***P < 0.01).

Heart-specific genes are enriched amongst micromolar CX DEGs

Given the tissue-specific terms associated with DOX-specific response clusters, we asked whether our DEG sets are enriched for a set of 419 curated heart-specific genes (Uhlen et al. 2015) compared to genes that are not DEGs. Heart-specific genes are enriched amongst DOX DEGs identified following 24 and 48 h of treatment (Fig. 6a). DEGs identified following 24 and 48 h treatments with submicromolar concentrations of CX are depleted for heart-specific genes, while DEGs identified following treatments with 2.5 µM CX are enriched (Fig. 6a).

Fig. 6.

Micromolar CX response genes are enriched for heart-specific genes. a) Enrichment of heart-specific genes amongst CX and DOX response genes (Uhlen et al. 2015). Asterisk represents those DEG sets that are enriched for heart-specific genes compared to non-DEGs (*P < 0.05, **P < 0.01, ***P < 0.001). b) Tissue-specificity scores of CX DEGs across CX concentrations and time. Tissue-specificity scores were obtained for each gene in left ventricle heart tissue (Jiang et al. 2020). Higher tissue-specificity scores indicate genes that have higher specificity for heart tissue. Asterisk represents significant differences between DEG sets (***P < 0.001).

We next considered tissue-specificity more broadly by comparing tissue-specificity scores for all genes expressed in the heart left ventricle across submicromolar and micromolar CX DEG sets (Jiang et al. 2020). Tissue-specificity scores are higher in the micromolar 24 and 48 h CX DEG sets compared to the submicromolar sets, indicating that CX induces changes in genes that are less broadly expressed at micromolar concentrations (P < 0.05; Fig. 6b). Together, these results indicate that micromolar CX treatment affects the expression of heart-specific genes.

CX affects the expression of genes in heart disease risk loci

DOX, the canonical anthracycline, is an effective anti-cancer drug. However, it can cause cardiotoxicity in some patients (Belger et al. 2024). DOX-induced cardiotoxicity is mediated through TOP2B (Zhang et al. 2012). Given that both DOX and CX inhibit TOP2, we evaluated whether CX and DOX influence the expression of genes implicated in DOX-induced cardiotoxicity. We focused on 25 DOX-induced cardiotoxicity genes that have been functionally validated (Fonoudi et al. 2024). This gene set is neither enriched nor depleted in DEG sets identified following any treatment (Fig. 7a). However, eighty percent of these genes are classified as DOX DEGs (20 of 25; Fig. 7b). While none of these genes are CX DEGs at any timepoint across submicromolar drug concentrations, 14 of these genes are DEGs following 3, 24, or 48 h of 2.5 µM CX treatment (Fig. 7b). Four of these genes respond in the same direction as DOX namely HAS3, PRDM2, MYH7, and CYP2J2.

Fig. 7.

Micromolar CX response genes include genes associated with risk for heart disease. a) Enrichment of functionally validated genes in DOX-induced cardiotoxicity loci amongst CX and DOX response genes (Fonoudi et al. 2024). b) Drug responses (log₂ fold change) of functionally validated genes in DOX-induced cardiotoxicity loci. Asterisk represents genes that are classified as DEGs (adj. P < 0.05). c) Enrichment of mapped genes in heart failure-associated loci (GWAS) amongst CX and DOX response genes (Cerezo et al. 2025). d) Drug responses (log₂ fold change) of mapped genes in heart failure-associated loci (GWAS) (Cerezo et al. 2025). Mapped genes were obtained from the 30 SNPs most significantly associated with heart failure. Asterisk represents genes that are classified as DEGs (adj. P < 0.05).

Anthracyclines such as DOX, are increasingly recognized as having the potential to lead to heart failure (Giordano et al. 2012; van der Zanden et al. 2021). We therefore examined the transcriptional responses of 118 genes associated with heart failure risk through GWAS across CX and DOX drug treatments (Cerezo et al. 2025). This gene set is enriched in 3-h 2.5 µM CX DEGs (Fig. 7c). Three-hour 0.5 µM DOX DEGs show the same direction of effect. When considering the 30 SNPs most significantly associated with heart failure risk, 28 respond to DOX in at least 1 time point or drug concentration (adj. P < 0.05; Fig. 7d). CX affects the expression of 18 of these genes, 13 of which respond in the same direction as genes affected by DOX. Three of these genes are identified at both submicromolar and micromolar CX concentrations: PSRC1, CDKN1A, and LSM3, while 15 respond only following micromolar CX treatments. Together, these results suggest that micromolar concentrations of CX impact genes in risk loci for DOX-induced cardiotoxicity and heart failure.

Discussion

CX is an anti-cancer drug currently in clinical trials for the treatment of advanced solid and hematological malignancies. CX has been described as an RNA polymerase I inhibitor, which takes advantage of the high levels of ribosomal RNA transcription in cancer cells. Recent studies have identified TOP2B, a topoisomerase isoform that is highly expressed in the heart, as the primary target of CX (Bruno et al. 2020; Pan et al. 2021). TOP2B is affected by several anti-cancer drugs including etoposide, mitoxantrone, DOX, and related anthracyclines. DOX and mitoxantrone are known to cause adverse effects on the heart (Coleman et al. 1984; Smith et al. 2010). It has therefore been hypothesized that CX might similarly adversely affect the heart (Pan et al. 2021). We therefore sought to investigate the cellular and molecular effects of CX treatment on iPSC-CMs and compare these to DOX. Our results indicate that CX-induced gene expression changes mirror a subset of changes induced by DOX.

We found that CX is cytotoxic to iPSC-CMs at micromolar concentrations albeit approximately 20-fold less toxic than DOX (median CX LD50 = 10.03 µM). CX and DOX were similarly cytotoxic to iPSCs at submicromolar concentrations. CX is an effective anti-cancer agent that decreases cancer cell line viability in ranges from 0.63 to 4.7 µM suggesting that CX is less toxic to postmitotic cardiomyocytes than rapidly dividing cancer cells (Seashore-Ludlow et al. 2015; Corsello et al. 2020). CX does not initiate DNA damage, measured as the presence of the DNA double-strand break marker γ-H2AX within 48 h of treatment at submicromolar or micromolar concentrations.

Transcriptome profiling indicated minimal effects of submicromolar CX treatments. However, the genes that are affected are similarly affected by DOX at the same concentrations. These gene sets include terms related to chromosome segregation and DNA replication. Transcriptome profiling of a micromolar concentration of CX-induced thousands of gene expression changes, the majority of which are shared with DOX. CX response genes identified at submicromolar and micromolar concentrations are enriched for genes in the DNA damage response pathway similar to DOX, while p53 target genes are enriched only at micromolar CX concentrations. Heart-specific genes and genes associated with DOX-induced cardiotoxicity and heart failure are only affected by micromolar CX concentrations.

Our results should be considered in light of the fact that iPSC-CMs tend to resemble fetal, rather than adult cardiomyocytes. Our day 30 iPSC-CMs express both the TOP2A and TOP2B isoforms, while human heart tissue predominantly expresses the TOP2B isoform (Cui et al. 2019; Consortium 2020). It has been shown that TOP2A expression decreases following culturing of iPSC-CMs for 60 d and that this correlates with reduced DOX susceptibility (Cui et al. 2019). To better phenocopy the human phenotype, future experiments could test the effects of CX following long-term culture of iPSC-CMs.

In summary, we characterized the human cardiomyocyte response to CX treatment and contrasted it to the well-known effects of DOX. We found CX to be cardiotoxic at micromolar concentrations and to induce a transcriptional profile resembling that of DOX treatment. The potential off target effects of CX on the heart should therefore be considered.

Data availability

All RNA-seq data have been deposited in the Gene Expression Omnibus (www.ncbi.nlm.nih.gov/geo/) under accession number GSE302122. All custom analysis scripts used for this project are available at https://github.com/mward-lab/Paul_CX_2025. Supplementary material includes Supplementary Tables 1 to 22 and Supplementary File 1 incorporating Supplementary Figs. 1 to 19.

Supplemental material available at G3 online.

Funding

This work was funded by a Cancer Prevention Research Institute of Texas (CPRIT) Recruitment of First-Time Faculty Award (RR190110) to M.C.W.

Conflicts of interest

None declared.

Author contributions

M.C.W conceived and designed the study. S.P., J.A.G., A.R.B., and E.R.M. performed experiments. S.P., E.R.M. and M.C.W analyzed the data. S.P. and M.C.W wrote the manuscript with input from co-authors. M.C.W supervised the work.