CCL2-mediated endothelial injury drives cardiac dysfunction in long COVID

Abstract

Evidence linking the endothelium to cardiac injury in long coronavirus disease (COVID) is well documented, but the underlying mechanisms remain unknown. Here we show that cytokines released by endothelial cells (ECs) contribute to long-COVID-associated cardiac dysfunction. Using thrombotic vascular tissues from patients with long COVID and induced pluripotent stem cell-derived ECs (iPSC-ECs), we modeled endotheliitis and observed similar dysfunction and cytokine upregulation, notably CCL2. Cardiac organoids comprising iPSC-ECs and iPSC-derived cardiomyocytes showed cardiac dysfunction after severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) exposure, driven by CCL2. Profiling of chromatin accessibility and gene expression at a single-cell resolution linked CCL2 to 'phenotype switching' and cardiac dysfunction, validated by high-throughput proteomics. Disease modeling of cardiac organoids and exposure of human ACE2 transgenic mice to SARS-CoV-2 spike proteins revealed that CCL2-induced oxidative stress promoted post-translational modification of cardiac proteins, leading to cardiac dysfunction. These findings suggest that EC-released cytokines contribute to cardiac dysfunction in long COVID, highlighting the importance of early vascular health monitoring in patients with long COVID.

Extended Data Fig. 1 |. Characterization of long COVID patient’s arterial tissue.

a, Computed Tomography Angiography (CTA) of long COVID patients (Pt. 5 and Pt. 7). Pt. 5 exhibits healthy iliac vessels without any evidence of disease (arrow, left panel), however, has bilateral occlusion of popliteal arteries (middle panel). Pt. 7 showed severe narrowing/near complete occlusion of the right common femoral artery (right panel). b, CTA of non-COVID patients (Pt. 9 and Pt. 10). Pt. 10 exhibits high-grade stenosis of the left common femoral artery, without complete occlusion (arrow, left panel). c, Photograph of the excised anterior tibial artery (ATA) of long COVID Pt. 5 showing complete occlusion (arrow, left panel) and cut section after removal of plaque (right panel). d,e, Bar graphs showing isometric measurements of contraction in healthy controls (HC), non-COVID and long COVID patient’s arteries when exposed to (d) 60 mM KCl or 3 mM U46619 and (e) 1 μM nicardipine or 10 μM pinacidil. (n = 3; three separate segments of all patients’ arteries). f, PCA plot from bulk RNA-seq of harvested arteries from HC, non-COVID and long COVID patients show a closer correlation of total RNA expression in patients within the same group. g, Volcano plot of differentially expressed genes (DEGs) from bulk RNA-seq data of HC and non-COVID ATA show downregulation of inflammatory markers (FDR < 0.05). Linear mixed-effect regression model was used to determine the differential expression. h, Enrichment analysis of DEGs show GO terms that do not represent inflammatory and immune pathways in ATA from non-COVID patients. i, Bar graphs of RPKM values from bulk RNA-seq show increased expression of CXCL8, CXCL10, and TNFα only in long COVID patient’s arterial samples (n = 2-4 samples). P values for HC v/s long COVID (CXCL8 < 0.0001; CXCL10 = 0.0035; TNFα = 0.0091). All data are represented as mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analyses were performed using one-way ANOVA corrected with Bonferroni method.

Extended Data Fig. 2 |. scRNA-seq of arteries from long COVID, non-COVID and HC.

a, Dot plot shows mean expression of select genes in the major cell types. b–h, Representative UMAPs show relative expression of several marker genes in the major cell types of the arterial tissues. (b) EC markers, PECAM1 and EMP1; (c) contractile SMC markers, TAGLN and MYL9; (d) proliferative SMC markers, TPM2 and FTH1. (e) stress response markers, JUN and DUSP1; (f) lncRNAs, MALAT1 and NEAT1; (g) fibroblast markers, COL8A1 and FN1; and (h) mesenchymal cell markers, NR2F2 and CEBPD. i, Violin plots show increased immune score in clusters that are representative of mesenchymal cells (MSC). j, Representative UMAP show relative expression of inflammatory marker CXCL8 in the major cell types of the arterial tissues. k, Relative expression of inflammatory marker IL6 as represented by a UMAP (left panel) and violin plot (right panel) show increase expression of IL6 only in long COVID patient’s arteries. l, Violin plots show increased EndoMT score in clusters that are representative of mesenchymal cells (MSC).

Extended Data Fig. 3 |. Characterization of long COVID patient’s iPSC-ECs.

a-d, Flow cytometry of iPSC-ECs show high percentage of ACE2⁺/CD144⁺ double-positive cells. e, Confocal images of iPSC-ECs show expression and co-localization of ACE2 and CD144. Repeated three times with similar results. Scale bar: 40 μm. f, Bar graphs of nanoluciferase assay (that represents cellular entry and replication) show high luciferase activity in iPSC-ECs following the first and repeated SARS-CoV-2 infection (n = 2). g, Plaque assays that measures infectious virus particles show formation of plaques at 10⁻¹, and 10⁻² concentrations and complete clearance at neat concentration. h, Immunofluorescent images show expression of ACE2 in iPSC-ECs and binding of the virus in only those iPSC-ECs that were exposed to SARS-CoV-2. Scale bar: 100 μm. i, Bar graph shows quantification of green fluorescence produced by SARS-CoV-2 in iPSC-ECs. Pretreatment of iPSC-ECs with anti-spike antibody blocked the SARS-CoV-2 (n = 8). P values (vehicle v/s SARS-CoV-2 < 0.0001). j, Bar graph of qPCR data show high expression of SARS-CoV-2 ‘S’ gene that encodes the spike protein in iPSC-ECs treated with SARS-CoV-2 (n = 6). P values (vehicle v/s SARS-CoV-2 at 24, 48, and 72 hr <0.0001). All data are represented as mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analyses were performed using one-way ANOVA corrected with Bonferroni method.

Extended Data Fig. 4 |. Characterization of primary ECs isolated from long COVID patients.

a, Representative bright-field images of primary ECs from HC, long COVID Pt. 5 and Pt. 6 (VECs), long COVID Pt. 8 (BECs), and non-COVID Pt. 9 (VECs). b,c, Bar graphs of qPCR data show lower expression of (b) CD31 and (c) NOS3 in primary ECs from long COVID patients compared to HC (n = 6). P values (HC v/s long COVID < 0.0001; HC v/s non-COVID < 0.0001). d,e, Bar graphs of qPCR data show higher expression of (d) CCL2 and (e) IL6 in primary ECs from long COVID patients compared to HC (n = 6). P values (HC v/s long COVID < 0.0001). f, Representative images of capillary-like networks show impaired tube formation by primary ECs from long COVID patients compared to HC. Bar graph shows quantification of the number of tubes (n = 6). P values (HC v/s long COVID < 0.0001). g, Quantification of NO production shows impaired capacity of primary ECs from long COVID patients to produce NO in response to Ach (n = 6). h,i, qPCR data show (h) lower expression of PECAM1 and (i) higher expression of IL6 at D15 of the ‘washout experiment’ in long COVID patient’s iPSC-ECs (n = 6). P values for PECAM 1 (D0 v/s D3/D6 < 0.0001; D0 v/s D9 = 0.0002; D0 v/s D12 = 0.0035; D0 v/s D15 = 0.0213). P values for IL6 (D0 v/s D3/D6/D9 < 0.0001; D0 v/s D12 = 0.0012; D0 v/s D15 = 0.0420). j, Representative brightfield images of capillary-like networks show impaired ability to form tubular structures at D15 of the ‘washout experiment’ in long COVID patient’s iPSC-ECs. All data are represented as mean ± SEM; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analyses were performed using one-way ANOVA corrected with Bonferroni method. Scale bar: 50 μm.

Extended Data Fig. 5 |. Characterization of cardiac organoids following exposure to SARS-CoV-2.

a, Representative brightfield images of cardiac organoids fabricated from iPSC-CMs alone or iPSC-CMs and iPSC-ECs from HC following vehicle or SARS-CoV-2 infection. b, Immunofluorescent images of COs show expression of ACE2 and binding of the virus in only those COs that were fabricated with iPSC-ECs when exposed to SARS-CoV-2. c, Bar graph shows quantification of green fluorescence produced by SARS-CoV-2 in COs. Pretreatment of COs with anti-spike antibody blocked the SARS-CoV-2 (n = 8). P values (vehicle v/s SARS-CoV-2 < 0.0001). d, Bar graph of qPCR data show high expression of SARS-CoV-2 ‘S’ gene that encodes the spike protein in COs treated with SARS-CoV-2 (n = 6). P values for CMs + ECs (vehicle v/s SARS-CoV-2 < 0.0001). e,f, Quantification of the contractile properties show spontaneous beating rate in COs fabricated from iPSC-CMs alone or iPSC-CMs and iPSC-ECs from (e) HC or (f) long COVID patients after exposure to SARS-CoV-2 (n = 40 for HC and n = 64 for long COVID). g,h, Quantification of the contractile properties show impaired (g) contraction and (h) relaxation velocities in COs fabricated with iPSC-CMs and iPSC-ECs from HC after SARS-CoV-2 infection compared to COs fabricated with iPSC-CMs alone and vehicle (n = 40). P values for CMs + ECs (vehicle v/s SARS-CoV-2 < 0.0001). i, Quantification of Ca²⁺ imaging parameters show a decrease in the Ca²⁺ transient amplitude in COs fabricated with iPSC-CMs and iPSC-ECs from long COVID patients after exposure to SARS-CoV-2 (n = 24). P values for vehicle v/s SARS-CoV-2 (CMs=0.0016 and CMs + ECs <0.0001). All data are represented as mean ± SEM; n = 5 for healthy controls and n = 8 for long COVID patients; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analyses were performed using one-way ANOVA corrected with Bonferroni method. Scale bar: 100 μm.

Extended Data Fig. 6 |. Characterization of cardiac organoids following treatment with ‘conditioned media’ from iPSC-ECs that were exposed to SARS-CoV-2.

a, Schematic workflow showing layout of the experimental design. b, Representative brightfield images of cardiac organoids (COs) fabricated from iPSC-CMs alone or iPSC-CMs and iPSC-ECs following treatment with ‘conditioned media’ from vehicle- or SARS-CoV-2 exposed iPSC-ECs. c,d, Quantification of the contractile properties shows impaired (c) contraction and (d) relaxation velocities in both groups of COs, including those fabricated from iPSC-CMs alone or iPSC-CMs and iPSC-ECs following treatment with ‘conditioned media’ from SARS-CoV-2 exposed iPSC-ECs (n = 20). P values for vehicle v/s SARS-CoV-2 (CMs=0.0046 and CMs + ECs <0.0001). e, qPCR data shows downregulation of CM markers, MYH6, MYH7 and RyR2 in both groups of cardiac organoids following treatment with ‘conditioned media’ from SARS-CoV-2 exposed iPSC-ECs (n = 3). P values for vehicle v/s SARS-CoV-2 in CMs (MYH6 = 0.0174; MYH7 = 0.0040; RyR2 = 0.0049) and in CMs + ECs (MYH6 = 0.0048; MYH7 = 0.0007; RyR2 = 0.0024). f,g, Gene expression profile from bulk RNA-seq show downregulation of (f) EC-specific genes, PECAM1, NOS3 and KLF2 and (g) CM-specific genes, MYH6/7 and RyR2 in COs fabricated from iPSC-CMs and iPSC-ECs after exposure to SARS-CoV-2 (n = 2). h, Volcano plot of differentially expressed genes (DEGs) from bulk RNA-seq data of COs fabricated with iPSC-CMs alone fail to show upregulation of inflammatory markers after SARS-CoV-2 infection compared to vehicle. Linear mixed-effect regression model was used to determine the differential expression. All data are represented as mean ± SEM; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analyses were performed using one-way ANOVA corrected with Bonferroni method. Scale bar: 100 μm.

Extended Data Fig. 7 |. Transcriptional profiling of cardiac organoids following treatment with ‘live’ virus strain of SARS-CoV-2.

a, Schematic workflow showing layout of the experimental design. b, Plaque assay that measures infectious virus particles show complete clearance for infected samples at neat, 10⁻¹, and 10⁻² concentrations. c, Bar graph of Nano-Glo assay show higher luminescence measurements in COs that were fabricated with iPSC-CMs and iPSC-ECs when compared to COs fabricated with iPSC-CMs alone (n = 2). P values for vehicle v/s SARS-CoV-2 (CMs + ECs=0.0171). d, Hierarchical clustering of bulk RNA-seq data show differentially regulated genes in COs fabricated with iPSC-CMs and iPSC-ECs after ‘live’ SARS-CoV-2 infection compared to COs fabricated with iPSC-CMs alone and vehicle. e,f, Volcano plots of differentially expressed genes show (e) upregulation of inflammatory markers in COs fabricated with iPSC-CMs and iPSC-ECs when exposed to ‘live’ SARS-CoV-2 compared to (f) COs fabricated with iPSC-CMs alone (FDR < 0.05). Linear mixed-effect regression model was used to determine the differential expression. g, Bar graphs of RPKM values from bulk RNA-seq show increased expression of IL1B (top panel) and CCL2 (bottom panel) in COs fabricated with iPSC-CMs and iPSC-ECs after ‘live’ SARS-CoV-2 infection compared to COs fabricated with iPSC-CMs alone and vehicle (n = 2). P values for vehicle v/s SARS-CoV-2 for IL1B (CMs + ECs <0.0001) and CCL2 (CMs + ECs <0.0001). h,i, Enrichment analysis of differentially expressed genes identified by bulk RNA-seq show (h) GO terms representative of inflammatory and immune pathways in COs fabricated with iPSC-CMs and iPSC-ECs after ‘live’ SARS-CoV-2 infection compared to (i) COs fabricated with iPSC-CMs alone. All data are represented as mean ± SEM; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analyses were performed using one-way ANOVA corrected with Bonferroni method.

Extended Data Fig. 8 |. Characterization of the long-term effects of SARS-CoV-2 on cardiac organoids.

a, Representative brightfield images of COs fabricated from iPSC-CMs alone or iPSC-CMs and iPSC-ECs at different time points (D0, D3, D6, D9, D12 and D15) following SARS-CoV-2 infection. COs were exposed to SARS-CoV-2 only for 3 days (up to D3) and then removed and replaced with normal medium for another 12 days (up to D15). b, Quantification of the contractile properties show spontaneous beating rate in COs on different days following SARS-CoV-2 infection (n = 20-40). c-e, Quantification of the contractile properties show (c) long-term impairment of relaxation properties in COs fabricated with iPSC-CMs and iPSC-ECs, and no change in (d) contraction and (e) relaxation velocities in COs fabricated with iPSC-CMs alone after SARS-CoV-2 infection compared to vehicle (n = 20-40). P values (vehicle v/s SARS-CoV-2 for D3/D6/D9/D12 < 0.0001 and D15 = 0.0023). All data are represented as mean ± SEM; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analyses were performed using one-way ANOVA corrected with Bonferroni method. Scale bar: 100 μm.

Extended Data Fig. 9 |. Gene expression profiling of the long-term effects of SARS-CoV-2 on cardiac organoids.

a, Principal-component analysis (PCA) plot of bulk RNA-seq data show abundance of variation in gene expression at different time-points following exposure to SARS-CoV-2 when compared to vehicle controls. b, Gene expression profile from bulk RNA-seq show downregulation of CM-specific genes, MYH7 and TNNT2 at D3 and D6 of the washout experiment in COs fabricated from iPSC-CMs and iPSC-ECs after exposure to SARS-CoV-2 (n = 2). P values for vehicle v/s SARS-CoV-2 for MYH7 (D3 = 0.043; D6 = 0.0441) and TNNT2 (D3 < 0.0001). c, Gene expression profile from bulk RNA-seq show upregulation of fibrosis genes, COL22A1 and FN1 at D15 of the ‘washout’ experiment in COs fabricated from iPSC-CMs and iPSC-ECs after exposure to SARS-CoV-2 (n = 2). P values for vehicle v/s SARS-CoV-2 for COL22A1 (D15 = 0.0396) and FN1 (D15 = 0.0062). d, Enrichment analysis of differentially expressed genes identified by bulk RNA-seq at D6 show GO terms representative of cardiac contraction in COs fabricated with iPSC-CMs alone after SARS-CoV-2 infection compared to vehicle. e, Enrichment analysis of differentially expressed genes identified by bulk RNA-seq at D15 show enriched GO terms for extracellular matrix organization in COs fabricated with iPSC-CMs alone after SARS-CoV-2 infection compared to vehicle. All data are represented as mean ± SEM; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analyses were performed using unpaired two-sided Student’s t-test.

Extended Data Fig. 10 |. Proteomic assessment of SARS-CoV-2 exposed cardiac organoids.

a, Bar graph of protein concentrations (pg/ml) as relative fold change in COs fabricated with iPSC-CMs alone after exposure to SARS-CoV-2 or vehicle (n = 2). b,c, Bar graph of ELISA assays show increased production of (b) COL22A1 and (c) FN1 protein by COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection compared to COs fabricated with iPSC-CMs alone and vehicle (n = 8). P values for vehicle v/s SARS-CoV-2 for COL22A1 (CMs + ECs <0.0001) and FN1 (CMs + ECs <0.0001). d,e, Immunostaining of mice hearts show increase levels of (d) CCL2 and (e) phosphorylated RyR2 in mice exposed to SARS-CoV-2 spike protein compared to vehicle controls. f, Bar graph of ELISA assays show increased serum levels of CCL2 in individual long COVID patients compared to healthy controls (n = 6). g,h, Bar graph of ELISA assays show increased serum levels of (g) COL22A1 and (h) FN1 in long COVID patients compared to healthy controls (n = 12-16). P values for healthy v/s mild long COVID patients (COL22A1 < 0.0001; FN1 = 0.0002) and healthy v/s severe long COVID patients (COL22A1 < 0.0001; FN1 < 0.0001). All data are represented as mean ± SEM; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analyses were performed using one-way ANOVA corrected with Bonferroni method. Scale bar: 50 μm.

Fig. 1 |. Patients with long COVID exhibit arterial thrombosis and EC dysfunction.

a, Schematic workflow for the recruitment of patients with long COVID, patients with non-COVID PAD and HCs. b, 3D reconstruction image of CT angiogram of a patient with long COVID with maximum intensity projection of both lower extremities before below-knee amputation shows occlusion of both left and right femoral arteries. c, Line graphs of percent relaxation show impaired vasoreactivity to Ach in TAs of patients with long COVID compared to HCs (n = 3; three separate segments of all patients’ arteries). P values (0.1 μM = 0.0029; 0.3 μM = 0.0404; 1 μM = 0.0131; 3 μM = 0.0046; 10 μM = 0.0047). d, Heatmap from bulk RNA-seq of TAs from HCs, patients with long COVID and patients with non-COVID PAD shows differential gene expression pattern. e, Volcano plot of DEGs from bulk RNA-seq data of HCs and long COVID TAs shows upregulation of inflammatory markers (FDR < 0.05). Linear mixed-effect regression model was used to determine the differential expression. f, Enrichment analysis of DEGs identified by bulk RNA-seq shows GO terms representative of inflammatory and immune pathways in TAs from patients with long COVID compared to HCs. g, Box plots of normalized expression from bulk RNA-seq data show significant upregulation of IL6 (upper panel) and CCL2 (lower panel) in patients with long COVID compared to non-COVID patients and HCs. Box plots represent data points from each individual sample with means and minimum and maximum values. h,i, UMAP of cells from TAs shows transcriptionally distinct clusters representing different cell types based on gene expression (h) and overlap of clusters from HCs, patients with long COVID and non-COVID patients (i). j, Heatmap of gene expression shows differential expression across different clusters from TAs of HCs, patients with long COVID and non-COVID patients. k, UMAP (left panel) and violin plot (right panel) show increased expression of CCL2 in the TAs of patients with long COVID compared to HCs and non-COVID patients. All data are represented as mean ± s.e.m. *P < 0.05, **P < 0.01. Significance of effects of Ach in long COVID-19, non-COVID-19 and HC arteries was determined by one-way ANOVA followed by a Bonferroni post-test. FDR, false discovery rate.

Fig. 2 |. iPSC-ECs from patients with long COVID-19 show impaired phenotype.

a, Schematic workflow of the experimental design. b, Representative bright-field images of iPSC-ECs from HCs and patients with long COVID showing typical ‘cobblestone’ monolayer after vehicle or SARS-CoV-2 infection. c,d, qPCR data show lower expression of NOS3 (c) and higher expression of CCL2 (d) in iPSC-ECs from HCs and from patients with long COVID after SARS-CoV-2 infection compared to vehicle. P values (NOS3: vehicle versus SARS-CoV-2 for HC < 0.0001 and long COVID < 0.0001) (CCL2: vehicle versus SARS-CoV-2 for HC < 0.0001 and long COVID < 0.0001). e, Representative images of capillary-like networks show impaired tube formation by iPSC-ECs from HCs and patients with long COVID after SARS-CoV-2 infection compared to vehicle. Right panel shows quantification of the number of tubes. P values (vehicle versus SARS-CoV-2 for HC < 0.0001 and long COVID < 0.0001). f, Quantification of NO shows impaired capacity of iPSC-ECs from HCs and from patients with long COVID to produce NO in response to Ach when exposed to SARS-CoV-2 compared to vehicle. P values (vehicle versus SARS-CoV-2 for HC < 0.0001 and long COVID < 0.0001). g, Schematic workflow showing layout of the ‘washout’ experiment. h,i, qPCR data show temporal expression of NOS3 (h) and CCL2 (i) in iPSC-ECs from patients with long COVID after SARS-CoV-2 infection. P values (NOS3: D0 versus D3/D6/D9/D12 < 0.0001 and D0 versus D15 = 0.0005) (CCL2: D0 versus D3/D6/D9 < 0.0001 and D0 versus D12/D15 = 0.0002). j, Bar graph depicting temporal quantification of capillary-like networks in iPSC-ECs from patients with long COVID after SARS-CoV-2 infection. P values (D0 versus D3/D6/D9/D12 < 0.0001 and D0 versus D15 = 0.0004). k, Bar graph depicting temporal quantification of NO production in iPSC-ECs from patients with long COVID after SARS-CoV-2 infection. All data are represented as mean ± s.e.m.; n = 5 for HCs and n = 6 for patients with long COVID; **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analyses were performed using one-way ANOVA corrected with the Bonferroni method. Scale bar, 50 μm. D, day; NS, not significant; NOx, nitrite/nitrate oxide.

Fig. 3 |. SARS-CoV-2-induced endotheliitis imparts CM dysfunction in human COs.

a, Schematic workflow of the experimental design. b, Representative bright-field images of COs fabricated from iPSC-CMs alone or iPSC-CMs and iPSC-ECs from patients with long COVID after vehicle or SARS-CoV-2 infection. c,d, Quantification of the contractile properties shows impaired contraction (c) and relaxation (d) velocities in COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection compared to COs fabricated with iPSC-CMs alone (n = 64). P values (CMs + ECs: vehicle versus SARS-CoV-2 < 0.0001) (SARS-CoV-2: CMs versus CMs + ECs < 0.0001). e,f, Quantification of Ca²⁺ handling properties shows an increase in the diastolic Ca²⁺ (e) and a decrease in the rate of Ca²⁺ uptake (f) in COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection (n = 24). P values for diastolic Ca²⁺ (CMs + ECs: vehicle versus SARS-CoV-2 = 0.0003) and rate of Ca²⁺ uptake (CMs + ECs: vehicle versus SARS-CoV-2 < 0.0001). g, Quantification of NO production shows impaired ability of COs fabricated with iPSC-CMs and iPSC-ECs to produce NO after SARS-CoV-2 infection (n = 24). P values (CMs + ECs: vehicle versus SARS-CoV-2 < 0.0001). h, Hierarchical clustering of bulk RNA-seq data shows differentially regulated genes in COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection. i, Volcano plot of DEGs from bulk RNA-seq data of COs fabricated with iPSC-CMs and iPSC-ECs shows upregulation of inflammatory markers after SARS-CoV-2 infection (FDR < 0.05). Linear mixed-effect regression model was used to determine the differential expression. j, Bar graphs of RPKM values from bulk RNA-seq show increased expression of IL1B (left panel) and CCL2 (right panel) in COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection. P values for vehicle versus SARS-CoV-2 (IL1B = 0.0060 and CCL2 = 0.0455). k, Enrichment analysis of DEGs identified by bulk RNA-seq shows GO terms representative of inflammatory and immune pathways in COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection. All data are represented as mean ± s.e.m.; n = 5 for HCs and n = 8 for patients with long COVID; **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analyses were performed using one-way ANOVA corrected with the Bonferroni method or paired Student’s t-test. Scale bar, 100 μm. FDR, false discovery rate; NS, not significant; NOx, nitrite/nitrate oxide.

Fig. 4 |. SARS-CoV-2-induced endotheliitis causes long-term CM dysfunction.

a, Schematic workflow of the experimental design. b, Quantification of the contractile properties of COs shows long-term impairment of contraction velocity in COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection compared to vehicle (n = 20–40). P values (vehicle versus SARS-CoV-2 for D3/D6/D9/D12 < 0.0001 and D15 = 0.0002). c, Hierarchical clustering of temporal bulk RNA-seq data from COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection shows an orchestrated interplay of differentially regulated genes at D6 and D15. d, Volcano plot of bulk RNA-seq data from COs fabricated with iPSC-CMs and iPSC-ECs shows upregulation of inflammatory markers at D6 after SARS-CoV-2 infection (FDR < 0.05). Linear mixed-effect regression model was used to determine the differential expression. e, Bar graph of RPKM values from bulk RNA-seq shows increased expression of CCL2 at D3 and D6 in COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection. CCL2 levels return to normal at D15 of the washout experiment. P values (vehicle versus SARS-CoV-2 for D3 = 0.0031 and D6 = 0.0495). f, Enrichment analysis of DEGs identified by bulk RNA-seq at D6 shows GO terms representative of inflammatory and immune pathways in COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection. g, Volcano plot of bulk RNA-seq data from COs fabricated with iPSC-CMs and iPSC-ECs shows upregulation of cardiac remodeling markers at D15 after SARS-CoV-2 infection. h, Bar graph of RPKM values from bulk RNA-seq shows increased expression of COL6A2 at D15 in COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection. P values (vehicle versus SARS-CoV-2 for D15 = 0.0093). i, Enrichment analysis of DEGs identified by bulk RNA-seq at D15 shows enriched GO terms for collagen biosynthesis and post-translational protein modification in COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection. All data are represented as mean ± s.e.m.; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analyses were performed using one-way ANOVA corrected with the Bonferroni method or unpaired Student’s t-test. D, day; FDR, false discovery rate; NS, not significant.

Fig. 5 |. Integrative single-cell analysis of COs reveals CCL2, an EC-specific cytokine, as the prime source for SARS-CoV-2-induced cardiac dysfunction.

a, Schematic workflow showing layout of the same-sample single-cell multiome experiment of COs. b–e, UMAPs of cells from COs show six transcriptionally distinct clusters based on gene expression (snRNA-seq) (b), accessible chromatin regions (snATAC-seq) (c), combined integration of snRNA-seq and snATAC-seq (d) and overlap of clusters from COs exposed to SARS-CoV-2 infection and vehicle (e). f, Heatmap of gene expression from snRNA-seq data from COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection shows the majority of the differentially regulated genes only in those clusters (C4 and C5) that represent ECs compared to clusters represented by CMs (C1–C3). g, Heatmap of gene expression from snRNA-seq data shows upregulation of inflammatory genes, including CCL2 and COL22A1, in SARS-CoV-2-exposed COs compared to vehicle. h, Heatmap of snATAC-seq identified with bias-matched differential testing displays 35,520 marker peaks across all clusters derived from c. i,j, Genome track images show cell-type-specific sites for snATAC-seq around CCL2 (i) and COL22A1 (j) gene loci in clusters 4 and 5 of COs that were exposed to SARS-CoV-2. Track scale: units of fold enrichment relative to the total number of reads in transcription start site. k, Combined UMAPs of cells from COs show expression of CCL2 (left panel) and COL22A1 (right panel). t-SNE, t-distributed stochastic neighbor embedding.

Fig. 6 |. Knockdown of CCL2 in ECs rescues SARS-CoV-2-induced CM dysfunction.

a, Schematic workflow of the experimental design. b, Representative bright-field images of COs fabricated from scramble or CCL2-KD iPSC-ECs and iPSC-CMs. c,d, Quantification of the contractile properties shows a rescue in their contraction (c) and relaxation (d) velocities in COs fabricated with iPSC-CMs and CCL2-KD iPSC-ECs after SARS-CoV-2 infection when compared to COs fabricated with iPSC-CMs and scramble iPSC-ECs (n = 20). P values for contraction velocity (vehicle versus SARS-CoV-2 for scramble < 0.0001 and CCL2-KD = 0.0068). P values for relaxation velocity (vehicle versus SARS-CoV-2 for scramble < 0.0001 and CCL2-KD = 0.0048). e,f, Quantification of Ca²⁺ handling properties shows a rescue in their kinetics as seen by a decrease in the diastolic Ca²⁺ (e) and an increase in the Ca²⁺ transient amplitude (f) in COs fabricated with iPSC-CMs and CCL2-KD iPSC-ECs after SARS-CoV-2 infection (n = 10–20). P values for diastolic Ca²⁺ (vehicle versus SARS-CoV-2 for scramble < 0.0001). P values for amplitude (vehicle versus SARS-CoV-2 for scramble < 0.0001). g, Quantification of NO production shows a rescue in the ability of COs fabricated with iPSC-CMs and CCL2-KD iPSC-ECs to produce NO after SARS-CoV-2 infection (n = 20). P values (vehicle versus SARS-CoV-2 for scramble < 0.0001). h,i, Analysis of bulk RNA-seq as hierarchical clustering (h) and volcano plot (i) from COs fabricated with iPSC-CMs and CCL2-KD iPSC-ECs shows reversal in the upregulation of inflammatory and cytokine markers, including CCL2 (FDR < 0.05). Linear mixed-effect regression model was used to determine the differential expression. j, Bar graphs from bulk RNA-seq data show reversal in the RPKM values for CCL2 in COs fabricated with iPSC-CMs and CCL2-KD iPSC-ECs after SARS-CoV-2 infection. P values (vehicle versus SARS-CoV-2 for scramble = 0.0125). k, Enrichment analysis of DEGs shows GO terms that do not represent inflammatory or immune pathways in COs fabricated with iPSC-CMs and CCL2-KD iPSC-ECs after SARS-CoV-2 infection. l, Bar graphs from bulk RNA-seq data show reversal in the RPKM values for COL22A1 in COs fabricated with iPSC-CMs and CCL2-KD iPSC-ECs after SARS-CoV-2 infection. All data are represented as mean ± s.e.m.; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analyses were performed using one-way ANOVA corrected with the Bonferroni method. Scale bar, 100 μm. FDR, false discovery rate; NOx, nitrite/nitrate oxide.

Fig. 7 |. Cell-type-specific secretome profiling validates EC-specific CCL2 as the prime source for SARS-CoV-2-induced cardiac dysfunction.

a, Bar graph of protein concentrations (pg ml⁻¹) as relative fold change shows increased production of EC-specific cytokines from COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection compared to vehicle (n = 2) b,c, Protein analysis using ELISA (n = 8) (b) and immunoblotting (n = 2) (c) shows increased production of CCL2 by COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection. Right panel of c shows quantification of CCL2 protein expression normalized to β-actin. d, Bar graphs show increased production of ROS levels by COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection (n = 4). e, Immunoblot shows increased levels of RyR2 phosphorylation in protein extracts from COs fabricated with iPSC-CMs and iPSC-ECs after SARS-CoV-2 infection (n = 3). Right panel of e shows quantification of RyR2 phosphorylation as relative protein density normalized to β-actin. f, Bar graphs show reversal in the production of ROS levels by COs fabricated with iPSC-CMs and CCL2-KD iPSC-ECs after SARS-CoV-2 infection when compared to scramble COs (n = 4). g, Immunoblot shows reversal in the phosphorylation of RyR2 in protein extracts from COs fabricated with iPSC-CMs and CCL2-KD iPSC-ECs after SARS-CoV-2 infection (n = 2). Right panel of g shows quantification of RyR2 phosphorylation as relative protein density normalized to β-actin. h, Bar graph of ELISA assays shows increased serum levels of CCL2 in mice that were exposed to SARS-CoV-2 spike protein compared to vehicle controls. Data are represented from eight mice. i, Immunoblots show increased levels of CCL2 (upper panel) and RyR2 phosphorylation (lower panel) in protein extracts from mouse hearts that were exposed to SARS-CoV-2 spike protein compared to vehicle controls. j, Bar graphs show increased serum levels of CCL2 in patients with long COVID compared to HCs (n = 12–24). All data are represented as mean ± s.e.m.; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analyses were performed using one-way ANOVA corrected with the Bonferroni method. NS, not significant.

Fig. 8 |. Schematic representation summarizing the role of the endothelium in long-COVID-associated cardiac dysfunction using iPSC-derived 3D COs.

a, Schematic of the hypothesis depicting the different stages by which SARS-CoV-2 infection can lead to EC activation and cytokine production, thereby contributing to cardiovascular dysfunction. b, Schematic of the study design to unravel the role of ECs in long-COVID-19-associated cardiac dysfunction. Human vascular tissues from patients with long COVID with thrombosis and iPSC-derived COs were leveraged to model COVID-19-induced EC injury and hyperinflammation. Deep transcriptomic analysis at a single-cell resolution and proteomics of these human cells revealed EC-specific immune reaction and release of cardiotoxic cytokines that imparted cardiac dysfunction, suggesting impaired EC–CM crosstalk after SARS-CoV-2 infection. Notably, EC-induced oxidative stress in response to SARS-CoV-2 infection led to post-translational modification of cardiac proteins, further implicating the role of the ECs in COVID-19-associated cardiac dysfunction. Together, our data provide mechanistic insights into the pathological crosstalk of the endothelium in the development of cardiovascular dysfunction in patients with long COVID, where a better understanding can have substantial implications on the search for therapeutics. PBMC, peripheral blood mononuclear cell; t-SNE, t-distributed stochastic neighbor embedding.