A High-Content Screen Identifies Drugs That Restrict Tumor Cell Extravasation across the Endothelial Barrier

Distant metastasis frequently transforms a localized and potentially curable tumor disease into an extensive multifocal pathology with limited therapeutic options (1). In fact, for some cancers, like pancreatic cancer, the high incidence of distant metastasis is in large part responsible for poor prognosis (2). Newly diagnosed patients with solid tumors are at constant risk of metastatic tumor cell dissemination either prior to or during the resection procedure (3, 4). The vascular system is the primary route for tumor cells to spread to distant organs (5). After entering the circulation, tumor cells must survive a hostile intravascular environment where they are exposed to mechanical shear stress and immune surveillance (6). Once physically restricted in small capillaries, tumor cells form attachments with the endothelium, destabilize the endothelial barrier, transmigrate, and implant into the adjacent extravascular tissue (5, 7, 8).

The vascular endothelium forms a dynamic and highly selective barrier that tightly regulates the trafficking of circulating cells between the intravascular and extravascular compartments (8, 9). While immune cells are equipped with a specialized machinery to easily transmigrate across the endothelium, circulating tumor cells are larger and apply less evolutionarily conserved steps to overcome the vascular barrier (7, 8). Among those, transcellular diapedesis and tumor cell–induced endothelial cell death have been reported (10, 11), in addition to paracellular transmigration through disruption of endothelial cell-cell junctions (7, 8, 12, 13). Thus, the dynamic regulation of biomechanical structures such as intercellular junctions, cell–matrix adhesions, and cytoskeletal organization in both tumor cells and the endothelium, as well as heterotypic cross-talk between both cell types, are of particular importance for the extravasation step (12, 14, 15).

The contribution of the endothelial barrier to tumor cell extravasation has been underscored by different gain- and loss-of-function studies (7, 8). These experiments highlighted critical roles for VE-cadherin and tight junctions (16, 17), as well as endothelial Tie1, Tie2, and Notch1 (18–20). Understanding these molecular mechanisms provide a platform to exploit blockade of tumor cell extravasation through the enhancement of vascular barrier stability (21–23). While many of these studies have offered proof of principle, they have not significantly expanded the repertoire of drugs to target specific steps in the metastatic cascade (24). Here, we performed a high-content screen using a transendothelial invasion coculture assay to identify compounds that hinder tumor cell extravasation. Hits were further characterized using a series of in vitro assays, a zebrafish model that enables three-dimensional (3D) visualization of direct tumor cell–endothelium interactions, and two distinct mouse models of lung metastasis. Finally, single-cell sequencing and proteomics were used to gain mechanistic insights of drug effects on endothelial cells and shed light on their impact to the process of metastasis.

Human umbilical vein endothelial cells (HUVEC; Lonza, #C2519A) were cultured in MCDB-131 endothelial growth medium (VEC Technologies). The human pancreatic cancer cell lines PaTu-8902 (ACC-179) and MIA PaCa-2 (CRL-1420) were a gift from Dr. Doug Hanahan (EPFL, Switzerland); 4T1 (CRL-2539) and LLC1 (CRL-1642) were purchased from ATCC. All tumor cell lines were cultured in DMEM (Gibco). Both types of culture media were supplemented with 10% FBS (Omega Scientific). Prior to injection into mice, cell lines were also tested for Mycoplasma and evaluated for pathogens (Charles River Laboratories).

BALB/cJ and C57BL/6 adult (8–10 weeks) mice were purchased from Jackson Laboratory (000651 and 000664). All experiments were previously approved and performed in accordance to NIH guidelines and Northwestern University Institutional Animal Care and Use Committee. Zebrafish experiments were performed according to approved Institutional Animal Care and Use Committee protocols at University of California, Los Angeles (Los Angeles, CA).

Lentiviral vectors (CMV-FP-IRES-puromycin) were used to express enhanced blue fluorescent protein (EBFP), green fluorescent protein (GFP), and red fluorescent protein (RFP) in endothelial and tumor cells. Generation of lentiviral particles and transduction of cells was performed as described previously (25).

For 3D surface renderings of transendothelial tumor cell invasion, EBFP-labeled HUVECs were plated onto gelatin-coated 8-wells (100,000 cells/well; Ibidi, #80826). After 48 hours, GFP-labeled PaTu-8902 (50,000 cells/well) were added onto HUVECs. Confocal microscopy was done following 1, 8, and 24 hours post coculture. For high-content screening, compounds or DMSO (vehicle control) were plated with MCDB-131 medium (30 μL/well) into 384 wells (Greiner bio-one, #781091) using a multi-mode dispenser (MultiFlo FX, BioTek), and a pin-tool (Biomek FX automated workstation, Beckman Coulter). Positive control CN03 was purchased from Cytoskeleton, Inc. (#CN03). EBFP-labeled HUVECs (5,000 cells in 40 μL/well) were added and cultured with compounds at 5 μmol/L for 40 hours, then GFP-labeled tumor cells were added onto the confluent endothelial monolayers (2,000 cells for PaTu-8902, or 1,000 cells for MIA PaCa-2, both delivered in 30 μL/well). Fluorescence imaging was performed every 24 hours using the ImageXpress MicroXL (Molecular Devices). Dose response of selected compounds was tested at 18 concentrations (50 μmol/L to 0.76 nmol/L). Total EBFP- and GFP-positive areas were automatically determined using the MetaXpress analysis software (Molecular Devices, version 6.2.2.707). A total of 3,520 compounds from LOPAC¹²⁸⁰, Prestwick Chemical Library, and NIH Clinical Collection, were screened. Compounds with a z-score of ≥4.0 or ≤−4.0 for the EBFP-positive area 72 hours after the addition of tumor cells were considered as initial hits and were further validated in triplicate. The data were then analyzed by hierarchical clustering, including EBFP- and GFP-positive areas and all 4 imaging timepoints.

Compounds, or DMSO (vehicle control), were plated with MCDB-131 medium into 384-well microplates. Subsequently, PaTu-8902 (750 cells/well) or MIA PaCa-2 (200 cells/well) were added. Final treatment concentration of test compounds was 1.0 μmol/L and 5.0 μmol/L. After 72 hours, viable cells were quantified using CellTiter-Glo according to manufacturer's instructions (Promega, #G7572).

Tg(kdrl:GFP)^la116 transgenic zebrafish embryos (26) were used as described previously (27). Forty-eight hours postfertilization (hpf), zebrafish larvae were dechorionated, anesthetized with 160 μg/mL tricaine (Sigma-Aldrich), and positioned in wedge-shaped troughs of a 1.5% agarose pad. RFP-labeled PaTu-8902 were injected into the duct of Cuvier (∼100 cells in 5 nL PBS) using a microinjection system (pico-injector, PLI-100, Harvard Apparatus) under a stereo microscope (SZX12, Olympus). Intravascular tumor cell load of freshly injected fish embryos was graded into high/medium and low (fluorescence microscopy). Embryos with equivalent numbers of tumor cells were randomized into treatment groups and exposed to compounds or DMSO at a range of dosages (prior dose–response analysis) to assess possible toxic effects and a dose around the EC₅₀ was selected for evaluation of metastasis. After 30 hours of incubation at 33°C, zebrafish were transferred into a 384-well format, anesthetized, and placed into a confocal high-content imaging system (ImageXpress Micro Confocal, Molecular Devices). Z-stack confocal imaging was performed by imaging of 9 tiles (488 nm and 561 nm; 10× objective). Tiles were stitched and cropped using MetaXpress image analysis software, and color-merged z-stack files of the zebrafish embryos were generated. Number of intra- and extravascular tumor cells were obtained in the intersegmental vessels using Imaris.

Cultures of PaTu-8902 (375,000 cells) and HUVECs (200,000 cells), alone or as cocultures (PaTu-8902 cells onto HUVEC monolayers), previously exposed to either 1.0 μmol/L niclosamide or DMSO were isolated after 8 hours. For the generation of single-cell GEMs, 5,200 suspended cells were loaded on a Chromium Single Cell Instrument (10X Genomics) per sample. Single-cell RNA sequencing (RNA-seq) libraries were prepared using the Chromium Single Cell 3′ Library & Gel Bead Kit v2 (10X Genomics). Sequencing was performed on Illumina HiSeq2500, and the digital expression matrix was generated by demultiplexing, barcode processing, and gene UMI (unique molecular index) counting using the Cell Ranger v2.1.1 pipeline (10X Genomics). Transcriptional analysis was performed using R (Seurat). Cells that expressed less than 100 genes or 500 transcripts were eliminated, and 2,000 variable genes were selected by Seurat for further analysis. The data were regressed by sequencing depth to remove unwanted source of variation. Principal component analysis (PCA) and uniform manifold approximation and projection (UMAP) were used to reduce the dimensionality of the data. A graph-based clustering approach was later used, and cell types were defined for each cluster using signature genes. Endothelial or tumor cell identity was defined by expression of either CDH5 and PECAM1, or EPCAM and KRT18. For the ligand receptor analysis, the genes were first filtered by an average expression of 0.01 normalized counts. Ligand receptor pairs were collected from the iTALK database (28). The pair was considered differentially regulated if at least one gene (ligand or receptor) was differentially expressed. The scRNA-seq data were deposited in the GEO database (accession number: GSE160915).

Cells were exposed to six concentrations (0.1 μg/mL to 3.2 μg/mL) of niclosamide or vehicle (5% EtOH, 10% DMSO, 30% cremophor, 55% HBSS) in 48-well plates and the impact of drug on tumor cell proliferation was evaluated for 96 hours. Plates were transferred to the IncuCyte Live-Cell Analysis System (Sartorius), images were acquired every hour and evaluated (IncuCyte analysis software). Concentration–response curves were generated and the half maximal effective concentration (EC₅₀) values of niclosamide were determined at 96 hours (GraphPad Prism).

Two distinct sygeneic models were used based on the well-known metastatic abilities of 4T1 mammary cancer cells and Lewis lung carcinoma (LLC) cells. Both male and female mice between 10–14 weeks were used and total body weight was obtained prior initiating, and at time of termination. Tumor weight was also obtained (in the case of 4T1). Cells were expanded and always freshly plated 24 hours prior to injection. Four hours prior to trypsinization, cultures were washed and incubated in serum-free media to remove FBS, then trypsinized and counted. Trypsin was inactivated by soybean trypsin inhibitor (Thermo Fisher Scientific). For 4T1 cells, a total of 500,000 cells were injected subcutaneously on the right flank of BALBc mice. Upon implantation, 4T1 cells formed solid tumors, palpable by day 7 as a firm mass. Two weeks postimplantation, animals were randomized for treatment. Intraperitoneal administration of niclosamide (20 mg/kg/day) or vehicle control (10% DMSO, 55% HBSS, 30% Cremophor, 5% ethanol) was performed daily intraperitoneally. Tumor growth was monitored and mice were sacrificed 4 weeks after injection. For LLC metastasis assays, a suspension consisting of 500,000 cells in 100 μL of DMEM was injected intravenously (tail vein) in C57BL/6 mice. Niclosamide (20 mg/kg/day) or vehicle control were administered intraperitoneally according to the trial. Three different trials were conducted: (i) prevention trial (niclosamide or vehicle was injected 4 days prior tumor cell injection), (ii) intervention trial (treatment was initiated the day after tumor cell injection for 7 days) and (iii) prevention + intervention trial (continuous treatment starting 4 days prior tumor cell injection and continued for 7 days after tumor cell injection). All trials terminated 4 weeks after the initial tumor injection. Blood was collected for proteomics and metabolomics at the end of the experiment. Subsequently, mice were perfused in the right ventricle with 2% paraformaldehyde (PFA) followed by intratracheal injection of low melting agarose (1%) containing 2% PFA. Tissue was fixed overnight in 2% PFA and washed 4–5 times with PBS the next day. Specimens were either embedded in paraffin (left lobes and solid tumors) or in 7% agarose for vibratome sectioning (right lobe). For quantification of tumor metastasis, two blinded investigators counted the entire right lobe of each mouse (12–16 sections each at 400 μm). After decoding, the numbers were compared and found to be within 90%–95% overlap. Data were plotted and significance assessed by t test.

Alexa Fluor cadaverine-555 (1 kDa) and dextran-FITC (10 kDa; Invitrogen) were injected intravenously at 10 μg/g and allowed to circulate for 2 hours. Mice were anesthetized and perfused with 10 mL of HBSS. Tissues were removed, homogenized in lysis buffer, supernatants were centrifuged at 13,000 rpm for 20 minutes, and fluorescence was measured on a fluorometer Synergy HT271167 plate reader. Results were normalized for total protein.

Mice were perfused with DMEM to remove circulating blood. Lung and tumors were dissected and washed in DMEM. Tissue was minced and enzymatically digested following manufacture protocol (Miltenyi Lung Tissue Dissociation Kit, 130–095–927). Single-cell suspensions were then lysed with RBC lysis buffer (Thermo Fisher 00–4333–57). Final single-cell suspension was stained with CD45-APCY7 (BioLegend, 103116), GR1-PB (BioLegend, 108430), CD11b-PE (BD BIosciences, 552850), CD-19-FITC (BioLegend, 152404), CD3-APC (BioLegend, 100236), and DAPI (live/dead) on ice and then analyzed on BD FORTESSA. Cells were gated on DAPI negative, CD45⁺, and further defined as monocytes (Gr^loCD11b⁺, CD3⁻CD19⁻), granulocytes (Gr^hiCD11b⁺, CD3⁻CD19⁻), T cells (CD3⁺), and B cells (CD19⁺).

Mouse plasma protein concentration was measured by a BCA assay (Pierce BCA Protein Assay Kit, Thermo Fisher Scientific). One milligram each was subjected to highly abundant protein depletion by Proteome Purify 2 Mouse Serum Protein Immunodepletion Resin (R&D Systems) following the manufacturer's protocol. Protein was concentrated and purified by acetone/trichloroacetic acid precipitation overnight at −20°C. After washing the pellet with ice-cold acetone, the resulting protein pellet was resuspended in 50 μL of 8 mol/L urea in 400 mmol/L ammonium bicarbonate, pH 7.8, reduced with 4 mmol/L dithiothreitol at 50°C for 30 minutes, and cysteines were alkylated with 18 mmol/L iodoacetamide for 30 minutes. The solution was then diluted to <2 mol/L urea and trypsin (Promega) was added at final trypsin/protein ratio of 1:50 prior to overnight incubation at 37°C. The resulting peptides were desalted using solid-phase extraction on a Pierce C18 Spin column and eluted in 80 μL of 80% acetonitrile in 0.1% formic acid. After lyophilization, peptides were reconstituted with 5% acetonitrile in 0.1% formic acid.

Peptides were analyzed by LC/MS-MS using a Dionex UltiMate 3000 Rapid Separation nanoLC and a Q Exactive HF Hybrid Quadrupole-Orbitrap Mass Spectrometer (Thermo Fisher Scientific Inc). Approximately 1 μg of peptide samples was loaded onto the trap column of 3-μm C18 beads. The analytic column was a 75 μm × 10.5 cm PicoChip column packed with 3-μm C18 beads (New Objective, Inc.). Flow rate was kept at 300 nL/minute. Solvent A was 0.1% FA in water and Solvent B was 0.1% FA in ACN. Peptides were separated on 120-minute analytic gradient from 5% ACN/0.1% FA to 40% ACN/0.1% FA. The top 20 most abundant precursor ions in each MS¹ scan were selected for fragmentation. Precursors were selected with an isolation width of 2 kDa and fragmented by Higher-energy collisional dissociation (HCD) at 30% normalized collision energy in the HCD cell. Previously selected ions were dynamically excluded from reselection for 20 seconds. The MS² AGC was set to 1 × 10⁵. All samples were run in duplicate.

Plasma was immediately mixed with cold 80% methanol at a 10:1 ratio and incubated overnight. Subsequently, samples were centrifuged at 20,000 × g for 30 minutes at 4°C. The metabolite-containing supernatant was then transferred to an Eppendorf. The protein pellet was used for total protein quantification. Extraction solution was dried using SpeedVac. Fifty-percent acetonitrile was added to the tube for reconstitution following by overtaxing for 30 seconds. Sample solution was then centrifuged for 30 minutes at 20,000 × g, 4°C. Supernatant was collected for LC-MS analysis. Samples were analyzed by high-performance liquid chromatography (HPLC) and high-resolution mass spectrometry and tandem mass spectrometry (HPLC/MS-MS). Specifically, system consisted of a Thermo Q-Exactive in line with an electrospray source and an Ultimate3000 (Thermo Fisher Scientific) series HPLC consisting of a binary pump, degasser, and autosampler outfitted with a Xbridge Amide column (Waters; dimensions of 2.3 mm × 100 mm and a 3.5-μm particle size). The automatic gain control (AGC) target was set at 1 × 10⁶ and the maximum injection time was 200 ms. The top 5 precursor ions were subsequently fragmented, in a data-dependent manner, using the higher energy collisional dissociation (HCD) cell set to 30% normalized collision energy in MS2 at a resolution power of 17,500. Besides matching m/z, metabolites are identified by matching either retention time with analytic standards and/or MS2 fragmentation pattern. Data acquisition and analysis were carried out by Xcalibur 4.1 software and Tracefinder 4.1 software, respectively (both from Thermo Fisher Scientific).

All analyses were performed using Prism (v8.0.2; GraphPad Software) unless otherwise indicated. P values <0.05 were considered significant. Data are presented as means ± SEM or SD as indicated. Differences between groups were evaluated using ANOVA followed with Tukey post hoc multiple comparison test. Data pairs were compared by using a Student unpaired two-tailed t test and Mann–Whitney test. The Z'-factor (screening window coefficient) of the screening assay was calculated using the formula Z' = 1 − [(3σ_c+ + 3σ_c−)/|μ_c+ − μ_c−|] (where σ: standard deviation, μ: mean, c+: positive control, c-: negative control). Hierarchical clustering of screening hits was performed in R using the heatmap.2 function. For correlation analyses, Spearman correlation coefficient (r_s) and slope were determined. EC₅₀ values were calculated using the nonlinear regression dose–response equation in Prism. Functional annotation cluster enrichment analyses were performed in DAVID v6.8 (https://david.ncifcrf.gov).

Protein tandem MS data was queried for protein identification and label-free quantification against the SwissProt Mus musculus database using MaxQuant v1.6.0.16 (29). The following modifications were set as search parameters: peptide mass tolerance at 6 ppm, trypsin digestion cleavage after K or R (except when followed by P), 2 allowed missed cleavage site, carbamidomethylated cystein (static modification), and oxidized methionine, protein N-term acetylation (variable modification). Search results were validated with peptide and protein FDR both at 0.01. Proteins that were identified with >1 peptide were subjected to a further statistical analysis using Perseus software (version 1.6.0.7; refs. 29, 31). Catalog numbers for reagents and RRID were included in the Supplementary Materials.

In hematogenous metastasis of solid tumors, such as pancreatic cancer, extravasation of circulating tumor cells requires multiple interactions with the endothelium (Fig. 1A; refs. 7, 8). Drugs targeting one or multiple steps could be effective inhibitors of tumor cell extravasation, and thus restrict distant metastasis. Using human pancreatic cancer cells (PaTu-8902 and MIA PaCa-2; Supplementary Fig. S1A), we developed a high-content screening assay based on the ability of these cells to disrupt endothelial monolayers. In fact, when plated onto a confluent monolayer of HUVECs, tumor cells adhere to the endothelium and progressively disrupt the endothelial barrier through invasive protrusions and cell expansion within a 24-hour time frame (Fig. 1B). To automate the process in a 384-well format, we used EBFP-labeled endothelial cells (HUVECs) and GFP-labeled tumor cells (PaTu-8902 and MIA PaCa-2) treated either with test compounds or DMSO (vehicle control; Fig. 1C). When unable to invade, tumor cells remained rounded in shape. In contrast, once tumor cells successfully overcame the endothelial barrier, they attached and spread, dislodging endothelial cells. This resulted in a larger GFP-positive area, and defects in the total EBFP-positive endothelial cell monolayer (Fig. 1D). The two tumor cell lines used, PaTu-8902 and MIA PaCa-2, represent a wide range of pancreatic cancer phenotypes. While PaTu-8902 displayed an epithelial phenotype, both in regard to cell shape and expression of epithelial markers; MIA PaCa-2 exhibited a mesenchymal-like phenotype (Supplementary Fig. S1A and S1B). The recombinant protein CN03, which blocks GTPase activity by deamidation of Rho Gln-63 (32) and improves endothelial barrier stability (33), was used as a positive control. CN03 prevented tumor cell invasion and restored the integrity of the HUVEC monolayer (Fig. 1D; Supplementary Fig. S1C). For computational quantification, total areas of EBFP-positive endothelial cells and GFP-positive tumor cells were determined using fluorescence intensity thresholding. The area of EBFP-positive HUVECs was reduced by both PaTu-8902 and MIA PaCa-2 72 hours after the addition of tumor cells, and mostly unaffected when endothelial monolayers were pretreated with CN03 (Fig. 1E; Supplementary Fig. S1D). Evaluation of the GFP-positive area covered by either PaTu-8902 or MIA PaCa-2 (Fig. 1F; Supplementary Fig. S1E) provided a complementary read-out. The suitability of the established assay for high-content screening was validated by determination of a Z'-factor (screening window coefficient) of 0.653 for PaTu-8902, and 0.693 for MIA PaCa-2, respectively (31).

A high-content screening and validation pipeline was established to identify novel compounds that impaired transmigration of tumor cells (Fig. 2A). As part of the workflow, initial hits were validated by dose response, effects on tumor cell proliferation, endothelial barrier function, and tumor cell motility. Subsequently, promising compounds were evaluated in vivo using zebrafish and mouse models. A total of 3,520 compounds (LOPAC¹²⁸⁰ Library, Prestwick Chemical Library, and NIH Clinical Collection) were tested using both PaTu-8902 and MIA PaCa-2 cell lines (Fig. 2A,–C). Hits with inhibitory effects on transendothelial invasion were defined by a z-score of ≥ 4.0 of the total EBFP-positive area, as they only permitted limited disruption to HUVEC monolayers (Fig. 2D; Supplementary Fig. S2A). In contrast, hits promoting transendothelial invasion were identified by a z-score of ≤−4.0 (EBFP-positive area). A hierarchical clustering analysis combined alterations in endothelial and tumor cell area, in triplicate and across both tumor cell types and four different timepoints, validated 55 compounds (Fig. 2E). Of these, 38 inhibited and 17 promoted transendothelial tumor cell invasion. Validated hits showed different levels of effectiveness as per quantified area changes for HUVECs, and PaTu-8902 and MIA PaCa-2, respectively (Fig. 2F; Supplementary Fig. S2B–S2D). Correlation of findings for PaTu-8902 and MIA PaCa-2 uncovered significant agreement on transendothelial invasion for both tumor cell lines, despite their broadly distinct phenotypes (Fig. 2G).

The subset of compounds with inhibitory properties on transendothelial invasion encompassed several drug categories (Fig. 3A). Three of these, namely antimetabolites, topoisomerase inhibitors, and tubulin modulators, are also part of a broader group of antineoplastic agents. Because our main interest was on the enhancement of endothelial barrier, and/or suppression of tumor cell–endothelium interactions, we sought to exclude compounds with substantial effects on tumor cell proliferation. From the 38 compounds evaluated, 19 displayed substantial inhibition of tumor cell proliferation with a largely similar impact on both PaTu-8902 and MIA PaCa-2 (Fig 3B). The combinatorial effects on transendothelial invasion and tumor cell proliferation by compound and associated drug category are shown in Fig. 3C. As expected, most compounds that were part of the classical antineoplastic drug categories had a substantial impact on tumor cell proliferation and were excluded from further validation analyses. Hence, we selected 19 compounds with no or minor impact on tumor cell proliferation. These were subjected to further dose–response evaluation by using 18 different concentrations (50 μmol/L to 0.76 nmol/L). Fourteen of these compounds showed favorable EC₅₀ values at or below 10 μmol/L (Fig. 3D) and were continued through the pipeline.

The impact of the 14 candidate compounds on endothelial barrier stability was further assessed on established HUVEC monolayers (Supplementary Fig. S3A) by using electric cell-substrate impedance sensing (ECIS). Five of the 14 compounds, namely forskolin, niclosamide, dipyridamole, DMH4, and edelfosine, elevated baseline resistance levels by 218.4 ± 22.6, 143.0 ± 22.9, 102.9 ± 22.4, 94.4 ± 38.3, and 66.0 ± 13.8 ohms above control, respectively (absolute resistance of control was 1,286.4 ohms; P < 0.05; Supplementary Fig. S3B). Importantly, we also measured transendothelial resistance after challenging endothelial monolayers with tumor cells (Supplementary Fig. S3A and S3C). Two compounds, forskolin and niclosamide, significantly diminished PaTu-8902–mediated loss of barrier resistance by 435.5 ± 39.3 and 311.2 ± 54.0 ohms compared with DMSO control, respectively (absolute resistance of control was 975.8 ohms; P < 0.05), indicating a strong barrier stabilizing effect.

As recognized previously (15, 25), cortical organization of the actin cytoskeleton is required for the stability of the endothelial barrier; while stress fibers are frequently associated with low barrier tightness. To assess how the 14 candidate compounds affected actin fiber organization as a surrogate for barrier stability, HUVEC monolayers were treated with the different compounds and effects on actin cytoskeleton organization, as well as on migration were also evaluated (Supplementary Fig. S3D–S3J). While DMSO-treated HUVEC monolayers showed randomly distributed stress fibers, compounds like forskolin, niclosamide, and atorvastatin induced reorganization of the cortical actin cytoskeleton, reflected by a significantly higher percentage of cells that were automatically detected by their actin fiber outlines (60.4 ± 2.8, 58.7 ± 1.4, and 57.1 ± 5.2 vs. 35.0 ± 2.5% of cells in DMSO control; P < 0.05; Supplementary Figs. S3E and S4A).

Next, the effect of candidate compounds on tumor cell motility was determined using scratch assays. Because of the different phenotype of each tumor line used for the study, we evaluated migration for the less invasive tumor cell line PaTu-8902 (no additional matrix), and invasion for the more aggressive tumor cell line MIA PaCa-2 (Supplementary Figs. S3F and S4B). The extent of tumor cell motility was determined by the percentage of the initial scratch area covered with migrating or invading tumor cells at 12 hours, or 48 hours, respectively. For PaTu-8902, the compounds methylbenzethonium chloride, forskolin, itraconazole, josamycin, niclosamide, and mupirocin significantly reduced tumor cell migration at 12 hours (51.2 ± 4.0, 72.7 ± 5.2, 76.7 ± 3.7, 79.0 ± 5.2, 86.7 ± 3.6, and 88.3 ± 4.0, vs. 97.2 ± 1.9% relative scratch density for DMSO control; P < 0.05; Supplementary Fig. S3G and S3H). For MIA PaCa-2, a significant reduction of tumor cell invasion was observed when cells were exposed to niclosamide and diphenyleneiodonium chloride (12.8 ± 2.2 and 21.6 ± 3.4, vs. 36.2 ± 4.3 percent relative scratch density for DMSO control; P < 0.05; Supplementary Fig. S3I and S3J). To eliminate interference of proliferation on the migration read-outs, experiments were also performed in the presence of cell-cycle inhibitors (Supplementary Fig. S4C and S4D). For all in vitro experiments shown, optimal compound dosages were applied (see Fig. 3D).

From the 14 compounds characterized in vitro, four outperformed the rest on endothelial barrier and tumor cell motility (Fig. 4A). Forskolin, niclosamide, dipyridamole, and atorvastatin exhibited robust ability to block tumor cell invasion of endothelial monolayers at low dosages. To test these compounds in vivo, and monitor their impact on tumor cell extravasation in real time, we took advantage of the unique features of zebrafish (Fig. 4B). PaTu-8902 cells were directly injected into the duct of Cuvier of Tg(kdrl:GFP)^la116 zebrafish embryos two days post-fertilization. After injection, fish were immediately exposed to the four candidate compounds or DMSO at optimal dosages above the EC₅₀ concentration as determined by dose–response analysis. Three-dimensional fluorescence imaging was performed at 30 hours, and the percentage of tumor cells present in the extravascular or intravascular compartments of intersegmental vessels was determined using Imaris imaging software (Supplementary Fig. S5A). Evaluation of multiple compounds in a larger number of treated zebrafish embryos was achieved by incorporating a confocal high-content imaging system as part of the validation workflow (Supplementary Fig. S5B). In comparison with the DMSO-treated control group, treatment with niclosamide or forskolin significantly decreased the percentage of extravasated tumor cells per zebrafish (mean fold change 0.757 ± 0.05, and 0.770 ± 0.07 of DMSO control, respectively; Fig. 4C and D). No significant effect was observed with atorvastatin or dipyridamole. In the group of DMSO-treated zebrafish, an average of 40.75 ± 2.1% PaTu-8902 were extravasated 30 hours post tumor cell injection. As forskolin is a well-known modulator of barrier function (34), we continued mechanistic studies using niclosamide as our top molecule.

Single-cell RNA-seq was used to explore mechanistic effects of niclosamide on tumor cell extravasation and endothelial barrier properties. Niclosamide-treated cocultures were compared with cocultures exposed to DMSO and to monocultures of PaTu-8902 and HUVECs (Fig. 5A; Supplementary Fig. S6A and S6B). Using these three different conditions, we were able to identify transcriptional changes reflective of tumor–endothelial cell contact (coculture effect), and niclosamide treatment, on both tumor and endothelial cells. UMAP analysis of the obtained single-cell sequencing dataset identified distinct cell clusters of HUVEC and PaTu-8902, in each experimental condition (Fig. 5B). Endothelial cell and tumor cell identity were confirmed by the expression of the markers CDH5 and PECAM1, or EPCAM and KRT18, respectively (Fig. 5C). Heatmaps of genes with most significant changes of RNA expression levels per condition and cell type are shown in Fig. 5D and E. Of interest, the chemokines CCL2 and CXCL2, as well as the inhibitor of DNA-binding proteins ID1 and ID3, were significantly upregulated upon tumor–endothelial cell contact, and downregulated by niclosamide treatment. In addition, niclosamide treatment significantly reduced transcripts of MMP2 and VWF in tumor cells. Enrichment analysis of genes upregulated by tumor–endothelial cell contact highlighted inflammatory gene annotations such as NFκB, TNF signaling, and CXC chemokines, as well as cell–cell junctions and cytoskeletal regulation (Fig. 5F). Importantly, the top annotations modified by niclosamide were cell–cell adhesion and cell–cell junctions for both endothelial and tumor cells. The specific genes of these top clusters are shown in Supplementary Fig. S6C and D, and include cytoskeleton and transmembrane proteins involved in stabilizing barrier function for endothelial cells, and promoting cell–cell attachment for tumor cells.

Further analyses using the iTALK database to highlight ligand–receptor interactions affecting either heterotypic (tumor and endothelial cells) or homotypic (endothelial–endothelial cells; tumor–tumor cells) changes were also performed (Fig. 5G and H). Interaction pairs were considered differentially up- or downregulated when at least one gene (ligand or receptor) was differentially expressed upon niclosamide treatment. Functional annotation of up- or downregulated ligand–receptor genes identified a number of enriched clusters related to attachment, extracellular matrix, and cytoskeletal regulation, such as the top enriched clusters “focal adhesion, ECM organization, actin cytoskeletal regulation”, and “ECM organization, focal adhesion”, respectively (Fig. 5G). In addition, the circos plot illustration of all downregulated ligand–receptor interactions part of the top annotation cluster revealed the central role of integrin beta-1 (ITGB1) based on its numerous connections to other extravasation-related molecules (Fig. 5H; circos plot of top upregulated annotation genes, see Supplementary Fig. S6E).

Given the impact of niclosamide on the endothelial barrier, we further tested the compound on metastasis using immunocompetent models. Mouse mammary carcinoma cells (4T1) were injected subcutaneously in BALB/c mice and allowed to grow for two weeks, at which time, mice were treated with either vehicle or niclosamide (intervention trial; Fig. 6A). Because of the short stability of niclosamide at 37°C (Supplementary Fig. S7A–S7C) mice were injected daily. The number of metastatic nodules was determined on serial vibratome sections of the right lung lobe (Supplementary Fig. S7D and S7E). Evaluation of the data showed a statistically significant difference between vehicle and niclosamide-treated mice, despite little differences on the size of primary tumors or body weight (Fig. 6B–D). There was no difference in the response of male or female mice to the treatment.

We also tested the efficacy of niclosamide in Lewis lung carcinoma (LLC) colonization. Because LLC cells were directly injected in the circulation, the model enabled prevention trials to address the question as to whether niclosamide could tighten vascular barrier prior to injection of tumor cells and prevent tumor seeding. Thus, we conducted both prevention and intervention trials (Fig. 6E). In addition to counting the right lobe using vibratome sections, the left lung lobes were evaluated by histology. Differences in the presence and size of cancer nodules were evident by naked eye (Fig. 6F,–H). Detailed quantification was performed by counting metastatic nodules in vibratome sections of the entire left lobe in all animals, as done for the 4T1 experiments. The evaluation was carried out in blinded counts by two individuals with a 90% overlap in accuracy. The findings indicated that all three niclosamide treatment regimes resulted in a substantial reduction of metastatic disease (Fig. 6I–K).

Our results on endothelial barrier using electrical impedance analyses (Supplementary Fig. S3A) suggested niclosamide might restrict permeability in vivo. To address this question directly, we performed permeability assays using tracers and immunocytochemical evaluation of fibrinogen, a plasma protein in tumor-bearing mice. We noted significant accumulation of fibrinogen in the extracellular space of the tumors in vehicle-treated, in contrast to low levels in niclosamide-treated animals (Fig. 7A). Intravenous injection of cadaverin (1K) and dextran (10K) tracers followed by assessment of their leakage into the extravascular space showed that niclosamide significant prevented permeability in the lungs of tumor-bearing mice (Fig. 7B and C). Importantly, niclosamide did not affect trafficking of immune cells, as determined by FACS and immunocytochemistry in the tumors or lungs of 4T1-injected mice (Fig. 7D–F).

We also interrogated possible effects (and biomarkers) of niclosamide treatment by evaluating plasma proteins and metabolites in tumor-free and tumor-bearing mice. Proteomics analyses revealed multiple proteins affected by exposure to niclosamide in tumor-naïve mice (Fig. 7G). Several proteins altered in tumor-free mice were also affected in tumor-bearing mice. One of these, kininogen, was further validated in tumor-bearing mice by immunoblots because of its relevance in the generation of bradykinin, a potent permeability mediator. Importantly, we noticed a significant reduction of bradykinin in the plasma of tumor-bearing mice that were treated with niclosamide (Fig. 7H). Metabolomics analyses of plasma were also performed in tumor-free (Fig. 7I and J) and tumor-bearing mice (Fig. 7K and L). The findings showed overlap in three metabolites: d-gluconic acid, pyruvic acid, and alpha-ketoglutarate, which could serve as potential indicators of niclosamide response in clinical trials, as they correlate with increased barrier post niclosamide treatment.

Tumor cell extravasation is a complex process regulated by the metastatic phenotype of circulating tumor cells and their molecular interactions with the vascular endothelium. The importance of endothelial barrier integrity in this context has been long established and concur that tightening the vascular barrier prevents metastatic events (7, 8). Unfortunately, previous studies are yet to translate this knowledge into clinical practice. Moreover, while in recent years specific targets on either tumor cells or the endothelium have been tested (21, 22), the complexity of direct and indirect tumor–endothelial interactions and the efficacy of drugs with dual cell-specific effects have been neglected. Our work takes advantage of an unbiased screening approach of tumor–endothelial cell cocultures, which initially advanced 38 compounds as potential hits for a more comprehensive characterization in vitro. From these, 4 compounds were selected for further evaluation in a zebrafish model that enables expeditious 3D visualization of tumor cell extravasation through a high-content imaging system. Two compounds, niclosamide and forskolin, significantly reduced tumor cell extravasation in vivo. Finally, prevention and intervention trials in mice underscored the effect of niclosamide in preventing hematogenous metastasis. Mechanistically, single-cell sequencing was used to reveal drug effects on tumor and endothelial cell signaling, and the interaction network between both cell types in systems treated with niclosamide. Our findings identify niclosamide as an effective drug to restrict tumor cell extravasation, and mechanistically link its effect to modulation of cell signaling and heterotypic cross-talk critical for intercellular junctions, cell–matrix interactions, permeability-related cytokines, and cytoskeletal regulation.

Approved in 1982 by the FDA as an anti-helminthic drug, niclosamide has been recently explored for other therapeutic purposes (35). In relation to cancer, niclosamide exerts antitumor activity in a wide range of cancer types through inhibition of a number of different pathways including Wnt, mTORC1, STAT3, NFκB, and Notch (36–40). Functionally, niclosamide impairs tumor cell migration, invasion, and survival in vitro (41–42), and it reduces tumor growth in vivo (43, 44). These results align with our findings supporting the notion that niclosamide is a multifunctional drug affecting a broad range of cellular pathways and functions. Importantly, in mouse models of breast, colon, and skin cancer, niclosamide treatment has led to reduced numbers of distant metastatic lesions (42–44), although the role of tumor cell extravasation across the endothelial barrier as a relevant mechanism was not investigated.

Our screen focused exclusively on endothelial barrier properties. We determined that exposure of endothelial monolayers to niclosamide prevented tumor cell extravasation by significantly enhancing transendothelial resistance levels, restoring barrier stability upon direct tumor cell contact, and improving the cortical organization of the actin cytoskeleton. Using a zebrafish model, we were able to directly visualize tumor cell extravasation in vivo, and to associate previously observed antitumor effects of niclosamide in mouse models to this particular step of the metastatic cascade. We also verified that niclosamide affected hematogenous metastasis. Importantly, exposure of the vasculature to niclosamide four days prior to tumor cell injection was sufficient to significantly impair lung colonization by tumor cells to levels similar to intervention trials. These findings, in particular, underscore the concept that effects on vascular barrier stability are a significant component of the mechanism associated with niclosamide antimetastatic effects.

Currently, based on its reportedly broad antitumor activity, niclosamide is being evaluated in four clinical trials for colon (phase I, NCT02687009; phase II, NCT02519582), and prostate cancer (phase I, NCT03123978; phase II, NCT02807805; ClinicalTrials.gov registry). In light of these ongoing trials, a timely understanding of the functional and molecular mechanisms of niclosamide in the context of the metastatic cascade is critical. In particular, both the proteomics and metabolomics evaluations performed here might prove valuable as biomarkers for molecular activity.

To shed light into the mechanistic underpinnings, gene expression profiles of tumor–endothelial cell cocultures exposed to vehicle and niclosamide were determined at a single-cell level. Tumor–endothelial cell contact itself led to enrichment of gene annotations related to either inflammatory signaling, or cell–cell adhesion, intercellular junctions, and cytoskeletal regulation. Interestingly, the entire set of genes enhanced by niclosamide was enriched in annotations relevant to tumor cell extravasation. These included: cell–cell adhesion, adherens junctions, ECM organization, and actin cytoskeleton regulation. In addition, niclosamide affected multiple transcripts previously known to drive tumor cell extravasation. For example, niclosamide promoted substantial reductions in CCL2 and CXCL2 transcripts, both significantly upregulated upon initial tumor–endothelial cell contact. In vivo binding of tumor cell–derived CCL2 to endothelial CCR2 facilitates tumor cell extravasation through endothelial retraction, induction of vascular permeability, and recruitment of monocytes (13, 45). Likewise, CXCL2 secreted by endothelial cells induced endothelial barrier disruption by enhancing force generation and cytoskeletal dynamics of CXCR2-expressing tumor cells (46). Furthermore, niclosamide reduced the expression of tumor-derived matrix metallopeptidase 2 (MMP2) and von Willebrand factor (VWF), both molecules associated with tumor cell extravasation (47, 48). A close evaluation of all ligand–receptor interactions impacted by niclosamide highlighted a key role for integrin beta-1 (ITGB1). In fact, ITGB1 is critical for tumor cell extravasation, such as invadopodia formation, direct interaction with the subendothelial ECM, recruitment of F-actin at the protruding edge, and transmigration past the intercellular gaps of the endothelium (49). Accordingly, ITGB1 has been reported to be crucial for tumor cell extravasation in a zebrafish and a chick embryo model, and plays a prometastatic role in different types of cancer (7, 27). Using proteomics, we also noted that niclosamide treatment reduced kininogen levels, the precursor for bradykinin, which is a potent endothelium-dependent vasodilator and vascular permeability factor (50). The reduction in bradykinin is likely an important systemic contributor to the antimetastatic effects mediated by niclosamide.

In light of novel immunotherapies for the treatment of solid tumors, the question arises how a drug impeding tumor cell extravasation would affect the trafficking of immune cells. The immature and leaky tumor vasculature substantially differs from the fairly healthy blood vessels in the pre-metastatic niche (51). In the context of immune cells that are highly specialized and well-equipped to transmigrate across the endothelial barrier (52), the prediction would be that drugs blocking tumor cell extravasation might not necessarily affect the well-orchestrated pathways used by immune cells. However, we have not investigated these aspects in depth at a molecular level, and future work on the effects of niclosamide on immune cell extravasation is needed.

In summary, our findings identify niclosamide as an effective drug for inhibition of tumor cell extravasation and hematogenous metastasis, highlighting its cell-specific properties and mechanistic effects. These results support the notion that drug-induced stabilization of cell–cell junctions and cell–matrix interactions attenuates the migratory and invasive capabilities of tumor cells, and simultaneously promotes endothelial barrier stabilization. Such drugs should be analyzed directly for their efficacy in human clinical settings, especially in association with surgical approaches that can enhance metastatic activity for short periods of time.

A.D. Langenbacher reports grants from NIH NHLBI during the conduct of the study. J.N. Chen reports grants from NIH/NHLBI during the conduct of the study. R. Damoiseaux reports personal fees and other from Forcyte Biotechnologies, personal fees from Amgen, personal fees from Holoclara, personal fees from Kirkland & Ellis, personal fees from NGM Biopharmaceuticals Inc, personal fees from Ophidion, and personal fees from Librede Inc. outside the submitted work. No disclosures were reported by the other authors.

G. Hilfenhaus: Conceptualization, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. A. Mompeón: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing-original draft. J. Freshman: Investigation, writing-review and editing. D.P. Prajapati: Investigation, writing-review and editing. G. Hernandez: Validation, investigation, visualization, methodology, writing-review and editing. V.M. Freitas: Writing-review and editing. F. Ma: Data curation, software, investigation, methodology, writing-review and editing. A.D. Langenbacher: Investigation, visualization, methodology. S. Mirkov: Investigation, writing-review and editing. D. Song: Investigation, writing-review and editing. B.-K. Cho: Software, formal analysis, writing-original draft, writing-review and editing. Y.A. Goo: Resources, supervision, writing-review and editing. M. Pellegrini: Resources, writing-review and editing. J.-N. Chen: Conceptualization, resources, writing-review and editing. R. Damoiseaux: Conceptualization, resources, data curation, software, supervision, writing-review and editing. M.L. Iruela-Arispe: Conceptualization, supervision, funding acquisition, writing-original draft, project administration, writing-review and editing.

The authors thank Douglas Hanahan for providing the pancreatic cancer cell lines; Austin Quach for support with R; the UCLA Zebrafish Core Facility; Sequencing Core Facility; Broad Stem Cell Institute Flow Cytometry Core; Integrated Molecular Technologies Core. All high-throughput screening experiments were performed at the UCLA Molecular Screening Shared Resource (supported by Comprehensive Cancer Center Grant NIH/NCI P30CA016042). Proteomics services were performed by the Northwestern Proteomics Core Facility, generously supported by NCI CCSG P30 CA060553 awarded to the Robert H. Lurie Comprehensive Cancer Center, instrumentation award (S10OD025194) from NIH Office of Director, and the National Resource for Translational and Developmental Proteomics supported by P41 GM108569. Metabolomics experiments were performed by the Metabolomics Core Facility at Feinberg School of Medicine at Northwestern University. This work was supported by NIH R01CA197943 and R35HL140014 (to M.L. Iruela-Arispe), R01 HL140472 (to J.-N. Chen), and a fellowship of the German Research Foundation (DFG; #HI1727/1–1 to G. Hilfenhaus) and of the Sao Paulo Research Foundation (FAPESP; #2016/19968–3 to V. Freitas).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.