A Transcriptomic Atlas of Drug-Induced Endothelial Dysfunction in Human Endothelial Cells

Dear Editor:

Drug-induced vascular burden is a critical challenge in both pharmaceutical development and clinical setting. In the past two decades, multiple drugs have been withdrawn from the market due to unanticipated adverse vascular complications, such as the increased risk of myocardial infarction and stroke (e.g., sibutramine and valdecoxib), heart valvular disease (e.g., pergolide and dexfenfluramine), and haemorrhagic stroke (e.g., phenylpropanolamine). Furthermore, many cancer drugs, such as anthracyclines, tyrosine kinase inhibitors (TKIs), and proteasome inhibitors, are also well-known to elicit a broad spectrum of vascular dysfunctions [1]. However, there has yet to be a study that systematically compares the molecular changes induced by different clinically-relevant chemical insults in vascular endothelial cells (ECs). Here, we established a transcriptomic atlas of endothelial injuries using both human primary aorta ECs and human induced pluripotent stem cell-derived endothelial cells (iPSC-ECs). Specifically, we performed RNA sequencing of 72 samples from 4 EC cell lines and 2 time points in response to 8 clinically-relevant compounds, most of which are chemotherapies known to raise vascular liability, including doxorubicin, bortezomib, carfilzomib, sunitinib, lapatinib, chloroquine, and cyclophosphamide, and aspirin as the control. This dataset provides a rich resource for the discovery of diagnostic markers, identification of therapeutic targets, and mechanistic investigation of drug-induced endothelial dysfunction.

For each compound, we treated ECs with increasing doses and measured cell viability after 24 hours (Figure 1A, B). Immunostaining confirmed that both the iPSC-ECs and primary ECs exhibit the typical cobblestone morphology and express the EC marker VE-cadherin (Figure 1C). Overall, we observed a consistent trend in dose response between iPSC-ECs and primary ECs (Figure 1D). The viability data appears to correlate with the clinical observations. For instance, aspirin is known to be a safe drug with a minimal vascular burden [2]. Our data shows aspirin has no effect on cell viability in both iPSC-ECs and primary ECs. In contrast, proteasome inhibitors carry high vascular risks such as hypertension, acute myocardial infarction, and angina [3]. Indeed, the viability assay indicates that proteasome inhibitors are highly toxic to the ECs. In particular, 0.01 μM (dose 1) of bortezomib reduced cell viability by 50% and 0.25 μM (dose 1) of carfilzomib reduced cell viability by over 80%.

Figure 1.

(A) Illustrative summary of the overall workflow and (B) dosing information of this study. Human iPSC-ECs were derived from iPSCs provided by Stanford Cardiovascular Institute Biobank and human aorta ECs were purchased from ATCC and Lonza. (C) Immunostaining of EC marker VE-cadherin in iPSC-ECs and primary aorta ECs. (D) Dose response of the 8 compounds in iPSC-ECs and primary ECs was determined using CellTiter-Glo viability assay. Cells were treated for 24 hours with increasing doses as indicated in (B). DMSO was used as the control. n=5 wells of cells for iPSC-ECs and n=6 wells of cells for primary ECs. Experiments were independently repeated twice. *p<0.05; **p<0.01; ***p<0.001. Unpaired Student’s t-test. Data are displayed as mean ± s.e.m. (E) PCA plots of 4 independent EC lines (2 iPSC-ECs and 2 human primary aorta ECs) treated with 8 compounds or DMSO for 24 hours (left) or 72 hours (right). PCA plots were generated using DESeq 2 normalized expression data of the DEGs. (F) Ranking of common pathways enriched by the upregulated DEGs and the downregulated DEGs, respectively. (G) Bubble plot of the top pathways from upregulated DEGs, including the PPAR signaling pathway, NOD-like receptor signaling pathway, fluid shear stress, and atherosclerosis in different drug treatments as well as from downregulated DEGs, including the PI3-Akt signaling, HIF-1 signaling, and AGE-RAGE signaling pathway by different drug treatments. (H) Heatmap of proteasome inhibitors-specific gene signature (24-hour time point). (I) Pathway enrichment analysis of drug-specific gene signature from the upregulated DEGs.

Next, we performed transcriptomic profiling on 2 lines of human iPSC-ECs and 2 lines of human primary aorta ECs treated with each compound for 24 or 72 hours at physiologically-relevant doses (Figure 1B). We first visualized the data using principal component analysis (PCA). At the 24-hour time point, we observed considerable variations between iPSC-ECs and primary ECs while the variations induced by the compounds seem less substantial (Figure 1E). However, after 72 hours of treatment, changes induced by drugs increased significantly (Figure 1E). Enrichment analysis of the differentially expressed genes (DEGs) revealed common pathways underlying drug-induced endothelial injuries. Specifically, the upregulated DEGs were mostly enriched in PPAR signaling pathway, the NOD-like receptor signaling pathway, and fluid shear stress and atherosclerosis (Figure 1F). PPAR activation in the endothelium may promote inflammation by upregulating IL-6, IL-8 and ET-1 while suppressing angiogenesis [4]. Similarly, NOD-like receptor signaling is a master regulator of inflammation that has been associated with EC senescence [5]. These data suggest that modulation of the PPAR or NOD-like receptor pathway may provide a generic solution to alleviate drug-induced endothelium dysfunction. Conversely, the downregulated DEGs were enriched in the AGE-RAGE signaling pathway and the PI3K-Akt signaling pathway in 7 drug treatments (Figure 1F). Dysregulation of the PI3K-Akt in ECs impairs critical cellular function (e.g., nitric oxide production) and cell survival [6]. Next, we reasoned that the degree of endothelial dysfunction provoked by each compound can be partially gauged by the disruption of these stress-related pathways using their enrichment metrics (gene number & false discovery rate (FDR)). By doing so, we found that cyclophosphamide, chloroquine, lapatinib, and carfilzomib appear to be inducing the strongest transcriptomic response (Figure 1G). Aspirin, on the other hand, is associated with a much milder transcriptomic change (Figure 1G). These data suggest the possibility of using the transcriptome to predict drug-induced vascular toxicity. Lastly, we analyzed drug-specific transcriptomic gene signatures. These genes were exclusively up- or down-regulated by a particular drug (or a type of drug) but not by any other drugs tested in this study. For instance, we found that proteasome inhibitors had 22 unique upregulated DEGs such as PSMC4, PSMB2, and HSPB8 (Figure 1H). Enriched pathways of these genes include proteasome, glutathione metabolism, and chemical carcinogenesis, matching the mechanism of action of proteasome inhibitors (Figure 1I) [7]. Similarly, cyclophosphamide, an alkylating agent known to induce DNA damages [8], had a unique signature enriched in DNA replication (Figure 1I). Interestingly, chloroquine had the most distinct transcriptomic signature, implicating pathways such as base excision repair, oxidative phosphorylation, and lysosome (Figure 1I).

In summary, data generated from this study may be utilized as a valuable resource for understanding drug-induced endothelial injury and the identification of novel markers for diagnostic applications. While overall our results showed consistency with the clinical observations, we acknowledge that an in vitro 2D model is intrinsically limited by its physiological relevance. In future studies, it is important to validate molecular insights in more sophisticated systems such as vascular organoids and animal models [9].