Introduction

Endothelial cells form the inner layer of blood vessels and have primary functions in controlling vascular tone and blood flow, mediating inflammation, promoting angiogenesis, and supplying tissues with oxygen and nutrients [1–3]. Depending on the tissue of origin and vascular bed, endothelial cells display a diverse range of morphological and biochemical signaling pathways, gene expression patterns, and membrane protein profiles. Specifically, the brain endothelial cells are different from the endothelial cells of other organs by exhibiting (i) higher levels of tight junction proteins, (ii) increased mitochondrial content, (iii) reduced pinocytic activity, and (iv) elevated levels of intake transporter proteins and efflux pumps [2,4]. Differences in cell-membrane post-translational modifications, e.g., glycosylation, can alter signaling, inflammation, vascular permeability, and mechanotransduction [1,2,5,6].

The endothelial cells are also one of the major stromal cells of the tumor microenvironment (TME). Signaling processes between the tumor and TME cells play a crucial role not only in stimulating tumor growth, migration and metastatic behavior, but also in remodeling the behavior of endothelial cells and angiogenic programs. Tumor and tumor-associated stromal cells secrete various soluble pro-angiogenic proteins, importantly VEGF, FGF2, and PDGF that activate the Tyr kinase receptors on endothelial cells to further upregulate downstream angiogenesis pathways [7]. Through connexin-43 dependent gap junctions, tumor cells exchange ions and small metabolites with endothelial cells, which can support then tumor progression and metastasis [8]. Furthermore, by releasing exosomes containing pro-angiogenic proteins, mRNAs and microRNAs, the tumor cells trigger angiogenic gene expression in endothelial cells to support their own growth and metastasis [9]. Indirectly, the tumor cells modulate the behavior of other cells in the TME by altering the physiological conditions (pH, O₂, temperature, TME composition) to induce endothelial cell responses [10]. For example, as a result of the Warburg effect, cancer cells produce lactate that is released in the TME via the MCT-4 transporter. When imported by endothelial cells through the MCT-1 transporter, lactate activates the NF-kB/IL-8 pathway. IL-8 produced through this mechanism acts in an autocrine manner to activate endothelial cells and stimulate angiogenesis [11].

To investigate the interactions between cancer and endothelial cells, tumor intravasation, angiogenic or immune signaling, or drug screening and delivery, several model systems have been developed [12–17]. In vitro, two-dimensional (2D) cancer/endothelial co-culture models have been extensively explored. Such 2D models are relatively cost-effective, easy to generate and use, and can provide valuable insights into cell morphology, the dynamics of fundamental biological processes, and pharmacokinetic behavior. Nonetheless, such models fail to recapitulate the intricate three-dimensional (3D) architecture of the in-vivo cellular environment. To address this shortcoming, alternative 3D-spheroid and organoid models that simulate tumor morphology, extracellular matrix (ECM) composition, the effect of chemical gradients (cytokines, chemokines, and other soluble factors), and cell-cell and cell-matrix interactions have been developed [18–20]. To provide structural support and imitate the ECM, the 3D co-culture models are typically constructed on a scaffold prepared from various natural or synthetic materials including decellularized tissues, polymers, hydrogels, or hybrid materials [18,20–22]. Among these, animal-derived matrices such as the ECM hydrogel derived from mouse sarcoma (Matrigel) and the rat tail collagen I are commonly used due to their widespread availability [23]. To further increase biological relevance, such cancer/endothelial co-culture models have been also integrated in microfluidic devices, which enable the investigation of physiological flow conditions and shear stress effects on cellular behavior [12,13,17,24]. These model systems raise, however, concerns about the potential impact of contaminants of non-human origin, immunogenicity, and poor reproducibility due to variations in the composition of the scaffolds. Additionally, due to increased complexity and heterogeneity of the culture environment, the 3D systems are more difficult to reproduce than the 2D models [18]. A scaffold-free strategy relying on the ability of cells to self-organize without needing synthetic or animal-derived matrices would address these issues, and would offer a physiologically more authentic tool for representing the in-vivo tumor biology and performing high-throughput drug screening [25]. Whether evaluating therapeutic drugs that target specific cell-membrane receptors (e.g., Lapatinib, T-DM1/Kadcyla), chemotherapeutic agents (Capecitabine), or immunotherapeutic cells/viruses or antibodies, the use of self-organizing 3D multicellular architectures that enable the direct visualization of cell behaviors and interactions in the presence of these agents would greatly benefit the modeling of cancer-supportive or destructive processes.

For simulating the in-vivo TME and studying angiogenic processes and vascular biology, most existing endothelial models utilize macrovasculature human umbilical vein endothelial cells (HUVEC). Due to ease in accessibility, isolation, and culture, HUVEC cells have been even utilized for developing in vitro microfluidic models of the BBB network, despite lacking many properties of the brain microvasculature endothelial cells [26]. A study by Eigenmann et al. has shown, however, that human brain microvasculature endothelial cells (HBMEC) that form a more restrictive barrier with a more favorable trans endothelial electrical resistance (TEER) and lower permeability, represents a more suitable cell line for modeling the BBB and performing high-throughput permeability screening [27]. Another study demonstrated that the immortalized human cerebral microvascular endothelial cell line (HCMEC/D3) isolated from the temporal lobe, and expressing platelet endothelial cell adhesion molecule 1 (PECAM-1), junctional adhesion molecule-A (JAM-A), VE-cadherin (CDH5), and tight junction CLDN3/CLDN5 and OCLN, is suitable for developing physiologically relevant in vitro 3D BBB models for studying barrier function and regulation, signaling mechanisms, and screening for drug uptake and neurotoxicity [28]. As an alternative to HCMEC/D3 cells, the cerebral cortex microvascular endothelial cell line (HBEC-5i) was used by Bolden et. al., due to its high expression of tight junction proteins, high TEER, and low permeability when compared to other immortalized BMECs [29]. In addition to the above-described immortalized brain endothelial cells, primary or low passage BMECs have been also utilized for building in vitro BBB models. The limited culture period can, however, compromise the barrier maturity and TEER stability. Human pluripotent stem cells (hPSC) represent another resource for producing endothelial cells, notwithstanding their limitations related to incomplete maturation, heterogeneous cellular identity, early senescence, and poor in vivo mimicry [30].

Metastasized cancer cells co-opt the brain microvessels to draw oxygen and nutrients, and to support their own survival and growth while they adapt their metabolism to the new TME [31,32]. In this manuscript we describe the development of a novel, proof-of-concept, scaffold-free 3D cell co-culture model aimed at mimicking endothelial/cancer cell interactions in the brain metastatic niche. The model leverages two novel findings in our laboratory, i.e.,: (i) Preliminary work has identified an affinity between brain endothelial and cancer cells that enables them to self-assemble in 3D constructs in a scaffold-free environment, and simulate the attachment of metastasized cancer cells to the brain microvasculature; and (ii) The proteomic profiling of brain and cancer cell secretomes revealed that the endothelial cells secrete considerably more collagen into their environment than other cell types. The model was built with (i) HBEC-5i cells which present multiple advantages over other immortalized brain endothelial cells due to high expression of proteins that delineate key interactions with cancer cells, and (ii) three distinct cancer cell lines of two different tissues of origin, i.e., MDA-MB-231 (triple negative/TN) and SK-BR-3 (HER2+) – representing frequently brain metastasized breast cancers, and SK-OV-3 – representing aggressive epithelial ovarian cancers with poor prognosis that seldom metastasize to the brain, limiting thereby the opportunities for their study. Our scaffold-free co-culture model exhibits a 3D networked architecture, direct cell-to-cell contact, cancer-dependent changes in endothelial cell morphology, and long-term stability, establishing a valuable tool for investigating the dynamic molecular crosstalk between endothelial and cancer cells. We envision that the platform will find broad utility for investigating survival, angiogenic, adhesion, and migratory programs in the TME, and for performing drug screening and developing immunotherapeutic approaches.

Methods

Reagents. Stable turboFP602 transfected MDA-MB-231 and SK-BR-3 breast and turboGFP transfected SK-OV-3 human ovarian cancer cell lines were purchased from Cells Online LLC (Milpitas, CA). HBEC-5i human brain endothelial cells were obtained from ATCC (Manassas, VA). Cell culture media [RPMI 1640 (1X), McCoy’s 5A (modified), DMEM high glucose (HG)], fetal bovine serum (FBS), GlutaMAX, Pen-Strep, puromycin, G418, Dulbecco’s phosphate buffered saline (DPBS), trypsin (0.25%)/EDTA (0.53 mM), and propidium iodide (PI) dead cell stain were purchased from ThermoFisher (Carlsbad, CA). Lapatinib ditosylate was acquired from Selleck (Houston, TX), and endothelial cell growth supplement (ECGS) and FITC-Dextran from Sigma Aldrich (St. Louis, MO). Primary ZO1 rabbit IgG mAb and secondary anti-rabbit IgG antibody conjugated to AlexaFluor 488 were purchased from Cell Signaling Technology (Danvers, MA), and primary Vimentin mouse IgG mAb and secondary anti-mouse IgG antibody conjugated to CruzFluor 488 from Santa Cruz Biotechnology (Dallas, TX). ActinRed 555 ReadyProbe Rhodamine phalloidin F-actin stain was from ThermoFisher, and DAPI nuclear counter stain from Cell Signaling Technology. EasyProbe green 488 dead cell stain was purchased from ABP Biosciences (Rockville, MD). PermaCell 24 well plates with cell culture inserts with 8 µm pore polycarbonate membranes and 8-well chamber slides were ordered from MatTek Life Sciences (Ashland, MA). Pipettes and Nunc EasYFlask polystyrene cell culture flasks with Nunclon Delta surface treatment were obtained from Life Technologies (Carlsbad, CA).

Cell culture. For stock preparation, the cells were initially cultured in the manufacturer’s recommended growth medium, i.e., turboGFP SK-OV-3 and turboFP602 SK-BR-3 in McCoy’s 5A supplemented with FBS (10–15%), glutamine (2 mM), and G418 (250 µg/mL); turboFP602 MDA-MB-231 in RPMI 1640 (1X) supplemented with FBS (10%) and puromycin (10 µg/mL); and, HBEC-5i in DMEM/HG supplemented with FBS (~15%), ECGS (~27 µg/mL), and Pen-Strep (0.5%). For downstream experiments, the cells were grown in DMEM/HG supplemented with FBS (10%) and Pen-Strep (0.5%). All concentrations represent the final concentrations in the cell culture medium, with liquid solution percentages representing v/v concentrations. The cell cultures were kept in a humidified incubator at 37 °C with 5% CO₂, and the culture medium was changed every 2–3 days.

Cell co-cultures. Two distinct methods were used for developing the scaffold-free co-culture model of cancer and endothelial cells. In the first approach, the endothelial cells were mixed with either green fluorescent SK-OV-3 or red fluorescent MDA-MB-231 cells in T25 cm² Nunc flasks in a ratio of 9:1 (i.e., 180,000 HBEC-5i: 20,000 cancer cells), or a ratio of 1:1 at double density seeding (i.e., 200,000 of each HBEC-5i and cancer cells). The co-culture mixtures were maintained in DMEM/HG medium supplemented with ECGS (~27 µg/mL), FBS (10%), and Pen-Strep (0.5%). For HBEC-5i/MDA-MB-231 co-cultures, ECGS was added to the culture medium only until day 16. In the second approach, the HBEC-5i endothelial cells were seeded at low confluence (200,000 cells) in a T25 cm² Nunc flask and maintained in DMEM/HG medium supplemented with ECGS (~27 µg/mL), FBS (10%), and Pen-Strep (0.5%) for ~10 days until 3D networked structures developed. At that stage, 200,000 cancer cells were added to the flask. Two biological replicates were performed for each experiment, i.e., by starting with fresh cell batches from liquid nitrogen, and with cells being propagated independently in 2–3 T25 cm² flasks.

Fluorescence microscopy. An inverted Eclipse TE2000-U epi-fluorescence microscope (Nikon Instruments Inc., Melville, NY, USA), with Plan Fl l0X DL PhObj NA 0.30 WD 16.0 mm PH1 and Plan Fl ELWD DM Phase 20X NA 0.45 WD 7 mm-8.lmm CC objectives, was used to visualize the cancer-endothelial co-culture model systems. The excitation/emission maxima for green fluorescent SK-OV-3 cells and red fluorescent MDA-MB-231/SK-BR-3 cells were 482/502 nm and 574/602 nm, respectively. An AURA III solid state triggerable light source (Lumencor, Inc. Beaverton, OR) using a DAPI/FITC/TRITC/CY5 Quad filter (Chroma Technology Corp., Bellows Falls, VT) was used to illuminated the cells. The exposure time was 100 ms. An ORCA-Flash 4.0 LT + sCMOS digital camera (Hamamatsu Photonics, Bridgewater township, NJ) and the Nikon’s NIS-Elements Advanced Research imaging platform, version 5.11.01, were utilized for acquiring and processing the images. For immunofluorescence staining, the cells were cultured in chamber slides, fixed with chilled methanol (−20°C) for 15 min, and then exposed to blocking buffer [PBS + BSA (5%) + Triton X-100 (0.3%)] for 1 h at room temperature, primary antibody overnight at 4°C, and secondary antibody for 1.5 h at room temperature in the dark. DAPI was added to all chambers and the cells were incubated for 5 min. The antibodies were prepared in a PBS solution supplemented with BSA (1%) and Triton X-100 (0.3%) in the following concentrations: primary ZO1 rabbit IgG (0.07 µg/mL), secondary anti-rabbit IgG-AlexaFluor 488 (2 µg/mL), primary vimentin mouse IgG (2 µg/mL), and secondary anti-mouse IgG-CruzFluor 488 (1 µg/mL). ActinRed 555 (2 drops/mL) and DAPI (1 µg/mL) staining was performed after completing the secondary antibody staining procedure. To visualize the multilayer vertical organization of the 3D co-culture models, z-stack images were captured with a slice interval of 25 µm. The acquisition of images was initiated at a base level of 0 µm when the bottom slice of cells came into focus, and was continued until focusing could be achieved on the top layer of fluorescent cells.

Transwell migration assay. Transmigration assays of green fluorescent SK-OV-3 and red fluorescent SK-BR-3 cells were performed in transwell plates with polycarbonate membrane inserts of 8 µm pore size. The cells were seeded independently in multiple wells per condition (2 control wells and 4 experimental wells for each cell line). HBEC-5i cells (50,000/well) were seeded in the bottom chamber in 600 µL medium [DMEM/HG, FBS (10%), ECGS (~27 µg/mL), Pen-Strep (0.5%)] and incubated for 24 h in a water-jacketed incubator at 37 °C with 5% CO₂. For the control wells, the bottom chamber contained only the cell culture medium, but no HBEC-5i cells. After 24 h, when the HBEC-5i cells reached ~60% confluence, cancer cells suspended in 100 µL DMEM/HG supplemented with FBS (10%) and Pen-Strep (0.5%) were added to the upper inserts (~50,000 cells/insert) for both experimental and control wells, and the transwell plates were incubated for up to 5 days in an incubator. The migration of fluorescent cancer cells from the inserts toward the endothelial cells in the bottom chambers was monitored by fluorescence microscopy. After 5 days of incubation, the inserts were removed and images were taken from both control and experimental bottom wells.

Drug treatment. HBEC-5i and red fluorescent SK-BR-3 cells were mixed in a proportion of 1:1 and co-cultured in DMEM/HG medium supplemented with FBS (10%), ECGS (~27 µg/mL), and Pen-Strep (0.5%) for up to 5 days, when Lapatinib drug (1 µM or 10 µM) was added to the co-culture system. HBEC-5i and SK-BR-3 monocultures were also incubated separately with Lapatinib (10 µM). All drug treatments were performed for up to 72 h. The experiments were performed in duplicate for each condition and the dead cells were visualized with EasyProbe green dead cell stain.

Live/dead cell assay. The long-term viability of cells was monitored by visualizing the presence of dead cells, up to 28 days in co-culture, by fluorescence microscopy. Propidium iodide was used for visualizing dead green fluorescent SK-OV-3 cells (excitation/emission maxima 535/617 nm when bound to dsDNA; 5 µg/mL in the culture medium), and EasyProbe green dead cell stain for visualizing dead red fluorescent MDA-MB-231 and SK-BR-3 cells (excitation/emission maxima 500/530 nm; 2 drops/ml in the culture medium). The cells were incubated for 30 min with the dead cell stains in the culture medium in a 37 °C humidified incubator with 5% CO₂, and visualized by fluorescence microscopy directly in the culture medium without any further preparation. The experiments were performed in two independent biological replicates for each model system.

Proteomic data analysis. A detailed description of proteomic sample generation and data acquisition and processing methodology is provided in previous work [33,34]. Briefly, the cells were initially cultured in DMEM/HG supplemented with FBS (10%) and Pen-Strep (0.5%). Next, the cell-membrane proteins were isolated from cultures produced in serum-rich and serum-free media [34], while the cell secretomes were collected from the supernatant of serum-starved cells for 48 h and concentrated with Amicon Ultra-15 centrifugal filters (3000 MWCO). After proteolytic digestion with trypsin (substrate:enzyme ratio ~50:1) and sample cleanup, data-dependent mass spectrometry (MS) proteomic data were acquired with a nano-liquid chromatography EASY-nLC 1200 separation system interfaced to a hybrid quadrupole-Orbitrap^TM mass spectrometer (Q Exactive^TM, ThermoFisher Scientific). Full mass spectra (400–1600 m/z) were acquired with a resolution of 70,000, and data dependent acquisition of MS2 spectra was performed at a normalized collision energy of 30% and 17,500 resolution. Three biological replicates were processed for each cell type and subcellular fraction, and each fraction (1–1.5 µL, ~ 2 µg/µL) was injected 3–5 times for LC-MS analysis on a 250 mm long x 75 µm ID separation column (2 µm C18 particles). RAW MS files were processed with the Proteome Discoverer (v.2.5) software package by using the Sequest HT search engine and a minimally redundant/reviewed Homo sapiens database (UniProt DB, August 30, 2022 download, 20399 entries). The results of replicate injections of the same sample were combined in a multiconsensus report. Peptide target false discovery rates (FDRs) were set to 0.01/0.03 (stringent/relaxed). Only fully tryptic peptides were considered in the analysis, with precursor/fragment ion tolerances of 15 ppm and 0.02 Da respectively, min 6/max 144 amino acids, and up to 2 missed cleavages. Semiquantitative comparisons of cell-membrane and secretome data were performed based on peptide spectrum match (PSM) counts, after normalizing the data from each of the three biological replicates of a cell line based on the total PSMs per dataset. Cord diagrams from the proteomic data were created with RAWGraphs tools (https://rawgraphs.io).

Ethics declaration: Not applicable.

Results

Development of the scaffold-free 3D endothelial/cancer cell co-culture model. Focus was placed on developing a system in which co-evolving cancer and endothelial cells form assemblies that mimic the networked structure of the brain microvasculature to which cancer cells adhere in the initial stages of brain metastasis to facilitate their own survival and proliferation. HBEC-5i endothelial cells were chosen due to their ability to form dense 2D monolayer networks with well-developed functional tight junctions that accurately replicate the properties of the in-vivo brain microvasculature. The HBEC-5i cells also exhibit high-abundance expression of key proteins that are critical to the establishment of cancer cells in the TME, such as adhesion (CDH5, VCAM1/CD106, ICAM1/CD54), transport (ABC transporters), and tight junction (CLDN5, OCLN) proteins, as well as receptors critical to modulating immune and inflammatory responses (TNF receptor CD40) [35]. Co-culture with astrocytes, which we intend to incorporate in future work, has been shown to increase, or alter, the expression of adhesion/tight junction and ABC transporter proteins [36,37]. The three cancer cell lines that were investigated represent useful preclinical models that are often used to study the fundamental mechanisms of cancer biology, the efficacy of therapeutic drugs, and the mechanisms of drug resistance. Breast cancer MDA-MB-231 (TN) and ovarian cancer SK-OV-3 cells were used to develop the model, and breast cancer SK-BR-3 (HER2+) to test the response to exposure to a small-molecule receptor Tyr kinase (RTK) inhibitor, Lapatinib.

At low confluence, in monoculture, the HBEC-5i cells presented a spindle-shaped morphology. With increased confluence, typically after 10 days of culture, 3D networked structures formed spontaneously in regular culture medium, without any need for scaffolding support (Fig 1A-C). Fluorescence microscopy imaging revealed a healthy endothelial culture capable of forming tight junctions, a key indicator that the cells maintained their original functionality. This was evidenced by visualizing cell-membrane localized ZO-1 proteins, cortical ring actin, and cytoplasmic vimentin with a characteristic denser accumulation in the perinuclear region (Fig 1D-F). The ability to develop networked tubular structures is a known feature of endothelial cells grown in isolation [23]. When grown in co-culture, however, in this work, the endothelial and cancer cells displayed a unique self-assembly behavior that has been overlooked in previous studies. The use of transfected red-fluorescent breast (tFP602 red) and green-fluorescent ovarian cancer (tGFP green) cells enabled the observation of a remarkable behavior of cells, i.e., the cancer and endothelial cells self-assembled in 3D constructs that mimicked the topology of the brain microvasculature where the cancer cells attach and thrive until bursting in a new tumor (i.e., overall vessel radius <100 µm; capillary radius <3.5 µm; capillary branch lengths extending up to 1 mm but most commonly <200 µm, with median branch length ~45 µm [38]). Proteomic profiling studies in our laboratory also revealed that the HBEC-5i cells secrete a significant amount of collagen compared to the cancer cells, which we believe served as a natural scaffolding material that supported the necessary cell-cell interactions that led to the formation of the 3D constructs.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Microscopy images of proliferating HBEC-5i cells.

(A) Endothelial cells at low density seeding; (B) Multi-cellular assemblies of endothelial cells formed at higher confluence; (C) Self-organization of endothelial cells in 3D aggregates that mimic the microvasculature; (D) Immunofluorescent visualization of ZO-1 accumulation at nascent endothelial cell contact sites (green dots and lines); (E) Immunofluorescent staining of ZO-1 (green), cytoplasmic actin (red), and cell nuclei (blue); (F) Immunofluorescent staining of vimentin (green) and cell nuclei (blue).

https://doi.org/10.1371/journal.pone.0349061.g001

Based on these findings, we developed two approaches for building 3D endothelial/cancer co-culture models in which the endothelial and cancer cells were either pre-mixed and allowed to co-evolve, or the cancer cells were added to the pre-formed endothelial constructs. For the 1^st approach, the HBEC-5i and MDA-MB-231 or SK-OV-3 cancer cells were mixed in a proportion of 9:1 or 1:1 to mimic either early or later stages of metastatic growth, respectively (Figs 2 and 3). For the 2^nd approach, HBEC-5i cells were seeded in a flask and allowed to evolve for ~10 days until the cellular networks formed, time at which the cancer cells were added (Fig 4). The co-culture was then maintained using the same medium and cell culture conditions.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Development of the co-culture model system prepared from premixed HBEC-5i and stable transfected fluorescent cancer cells (1:1 mix).

Top panel: HBEC-5i/SK-OV-3 (turbo GFP/green) model; Bottom panel: HBEC-5i/MDA-MB-231 (turbo FP602/red) model. Cells were seeded in equal proportion (200,000 each, ~ 30% confluence) in a 25 cm² culture flask and allowed to evolve for ~30 days. Self-organization of endothelial cells in 3D aggregates and accumulation of cancer cells around these structures was observable after ~7 days (80-90% confluence). After ~21 days, clearly delineated 3D networked constructs incorporating the vast majority of cancer cells were observable.

https://doi.org/10.1371/journal.pone.0349061.g002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Co-culture model system prepared from premixed HBEC-5i and stable transfected fluorescent cancer cells (9:1 mix).

Top panel: HBEC-5i/SK-OV-3 (turbo GFP/green) model; Bottom panel: HBEC-5i/MDA-MB-231 (turbo FP602/red) model. Cells were seeded in a proportion of HBEC-5i (180,000 cells)/cancer cells (~20,000) in a 25 cm² culture flask and allowed to evolve for ~30 days. Self-organization of endothelial cells in 3D aggregates and accumulation of cancer cells in these structures was observable after ~7 days. After ~21 days, clearly delineated 3D networked constructs were observable.

https://doi.org/10.1371/journal.pone.0349061.g003

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Co-culture model system prepared from pre-formed HBEC-5i networked structures to which stable transfected fluorescent cancer cells were added.

Top panel: HBEC-5i/SK-OV-3 (turbo GFP/green) model; Bottom panel: HBEC-5i/MDA-MB-231 (turbo FP602/red) model. HBEC-5i cells (~200,000) were seeded in a 25 cm² culture flask and allowed to evolve until they formed 3D networked structures (~10 days), time when ~200,000 cancer cells were added to the flask. The migration of SK-OV-3 and MDA-MB-231 cells toward HBEC-5i structures was observable already after 2 days of co-culture. The evolution of endothelial/cancer constructs was monitored for an additional 3 weeks.

https://doi.org/10.1371/journal.pone.0349061.g004

Characterization of the endothelial/cancer cell constructs. Whether added from start or after the endothelial structures were formed, the cancer cells migrated toward, attached themselves, influenced the morphology, and continued to grow with the endothelial constructs. The endothelial/cancer assemblies formed on un-coated plastic surfaces, without the addition of contaminating animal or synthetic ECM/lamina or scaffold components, and were stable for >4 weeks. The final constructs had consistent morphology in replicate preparations and demonstrated uniform growth, long-term stability, and reproducible behavior, leading to progressively better delineated fluorescent cancer cell accumulations as the co-cultures evolved from day 2 to 23 (Figs 2–4).

The morphology of constructs that evolved in co-culture varied, however, from one cancer cell line to another. The HBEC-5i/SK-OV-3 co-cultures generated network constructs similar to the ones produced from HBEC-5i alone, while the HBEC-5i/breast cancer co-cultures generated rather discontinuous, spotty networks that formed slower than the HBEC-5i/SK-OV-3 networks. The cancer cells accumulated, however, always in the areas of higher-density endothelial structures, suggesting the presence of paracrine chemokine gradients and inter-cellular interactions that support cancer cell migration and attachment in order to gain survival benefits. The mixing ratio of endothelial-to-cancer cells did not appear to impact the formation of constructs, but the HBEC-5i cells proliferated at a faster rate in the pre-mixed co-cultures, compared to the HBEC-5i monocultures, with prominent structures being already observable after 6-7 days of co-culture vs ~10 days in monoculture, respectively, when the cultures were initiated at ~25-30% confluence. Follow-up experiments confirmed the time-line for the development of the constructs and the migratory behavior of cancer cells. As evidence for 3D cellular organization, z-stack images of sequential optical sections, acquired with 25 µm resolution, revealed multiple cellular layers along the z-axis and the formation of 200-400 µm vertical assemblies in the co-culture models (S1 Fig). The cancer cells stimulated the formation of 3D endothelial cellular arrangements and network branching points which entirely covered the small flask area within two weeks of culture. The topological dimensions of well-established structures that formed within ~20 days were larger than previously described for brain capillary networks [38], but within the reported ranges. Areas with denser networks were characterized by branch lengths with medians of 442/471 µm, means of 449/528 µm, and mean standard deviations of 123/164 µm; and branch width dimensions with medians of 59/70 µm, means of 57/70 µm, and mean standard deviations of 14/17 µm. Areas with sparser networks had branch lengths with medians of 792/689 µm, means of 823/767 µm, and mean standard deviations of 322/401 µm; and branch width dimensions with medians of 96/64 µm, means of 111/79 µm, and mean standard deviations of 38/40 µm. The measurements were visualized and summarized in supporting S2 Fig.

Both SK-OV-3 and MDA-MB-231 cells secrete pro-angiogenic factors known to facilitate angiogenic sprouting and vascular tube formation by upregulating the downstream fatty acid-binding protein 4 (FABP4) expression of VEGF signaling in endothelial cells [39]. While the formation of endothelial networks and ability to stimulate their formation with cancer cells aligned with the findings of previous studies [e.g., HUVEC stimulated by lung cancer cells via upregulated PI3K/AKT and COX2 signaling pathways [40]], the networked architectures that formed by using our experimental conditions did not have a hollow lumen. Images acquired from HBEC-5i cultures that exhibited 3D constructs incubated with different concentrations of FITC-labeled dextran (50–500 µg/mL in the cell culture medium) for 15 min, 30 min and 24 h, did not reveal hollow structures in which the FITC-dextran could have accumulated to highlight the networks (S3 Fig). We note that a high molecular weight dextran (70,000) was used to prevent cell-membrane penetration and artifactual labeling of the networks. This outcome was not surprising, however, as we did not add any specific pro-angiogenic supplements to the culture that could have promoted the formation and long-term stability of tubular vascular networks.

In the 2^nd approach, within two days of co-culture, the cancer cells were observed to start migrating toward the pre-formed, high-density HBEC-5i structures and generate shortly thereafter clearly delineated endothelial/cancer constructs (Fig 4). Both cancer cell types exhibited chemotactic migratory behavior, with SK-OV-3 cells establishing themselves faster in the endothelial networks than the MDA-MB-231 cells. Unlike in the 1^st approach when the cells were pre-mixed from start, the continuous network architecture of HBEC-5i/MDA-MB-231 co-cultures was preserved in this scenario.

The migratory behavior of cancer cells (SK-OV-3 and SK-BR-3) was corroborated by a transwell assay, where HBEC-5i cells were seeded in the bottom wells and the cancer cells in the upper inserts comprising a membrane with 8 µm pores (Fig 5). The control wells contained cancer cells in the upper inserts and only culture medium in the bottom wells. After 5 days of incubation, the inserts were removed, and images from the bottom experimental and control wells were acquired. A much larger number of cancer cells were observable in each of the experimental wells that were pre-seeded with HBEC-5i cells, clustering, again, in the areas of higher density endothelial cells, and supporting the observation that the migration of cancer cells in the model system is controlled by chemical cues secreted by the endothelial constructs (Fig 5).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Transwell assay images depicting the migration of cancer cells toward endothelial cells.

(A,D) Control wells with red fluorescent SK-BR-3 cells seeded in the upper inserts; (B,C,E,F) Experimental wells with red fluorescent SK-BR-3 cells seeded in the upper inserts and HBEC-5i in the bottom wells; (G,J) Control wells green fluorescent SK-OV-3 cells seeded in the upper inserts; (H,I,K,L) Experimental wells with green fluorescent SK-OV-3 cells seeded in the upper inserts and HBEC-5i in the bottom wells. The images were taken from the bottom wells after 5 days incubation.

https://doi.org/10.1371/journal.pone.0349061.g005

Proteomic analysis. Preliminary proteomic profiling of HBEC-5i, SK-OV-3, and MDA-MB-231 cell secretomes and cell-membrane proteins revealed the presence of key cytokines/chemokines, cell adhesion molecules, growth and angiogenic factors, and ECM components that could support the behavior of cells in the model system (Fig 6). While a full analysis of the proteome data is outside the scope of this study, several key findings warrant discussion. Mass spectrometric analysis detected multiple types of collagen in the secretome of all cell lines, by multiple unique peptides, and by thousands of high-confidence peptide spectrum matches-PSMs (FDRs < 0.01; see S1 Dataset for a semiquantitative assessment of relative protein abundances). As pinpointed above, the endothelial cells secreted collagens in much higher abundance than the cancer cells, in particular COL1A1, COL1A2, COL6A1, and COL6A3. These collagens are important for building the ECM, maintaining the integrity of the BBB, and promoting endothelial cell migration and angiogenesis. Moreover, type I collagen has been shown to initiate the formation of 3D capillary networks from proliferating endothelial cells [41]. The collagen composition of the brain ECM is however highly dynamic, the various members of this superfamily having diverse physiological roles. For example, in the basal lamina of the healthy CNS tissue, the endothelial cells deposit primarily collagen IV, collagens I and VI being rather associated with various diseased states and cerebrovascular pathologies [42,43]. In vitro, in cell cultures, the deposition of collagens I and VI could therefore be attributed to an inherent need to establish an ECM foundation that supports inter-cellular signaling and protects against stress. Interestingly, the SK-OV-3 cells also secreted collagens in high abundance, of a different type (COL12A1 and COL18A1), however, which were found in previous studies to be overexpressed in the tumor TME [44]. Altogether, collagen deposition can remodel the ECM and TME composition to promote tumor growth and progression, and in our model served as a scaffolding material that supported the establishment of cell-cell interactions and the formation of 3D constructs.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Cord diagrams showing the relative abundance of protein components identified in the cellular secretome or cell-membrane fractions of HBEC-5i, SK-OV-3, and MDA-MB-231 cells, relevant to the formation of 3D endothelial/cancer constructs.

(A) Chemokines in the cell secretomes (PSM range: 0-200), (B) Angiogenic factors in the cell secretomes (PSM range: (0-62), (C) Growth factors in the cell secretomes (PSM range: 0-796), (D) Adhesion and junction proteins in the cell-membrane proteomes (PSM range: 0-8905).

https://doi.org/10.1371/journal.pone.0349061.g006

Interestingly, the SK-OV-3 cells in particular, also secreted ECM remodeling matrix metalloproteinases in high abundance (MMP1 and MMP2). These enzymes mediate ECM degradation, enabling cell invasion, and could have been important contributors to the faster formation of HBEC-5i/SK-OV-3 constructs.

The cells secreted essential cytokines/chemokines involved in eliciting immune responses and angiogenic processes, modulating the expression of adhesion molecules, and supporting cancer proliferation, EMT, and migration [45–47] (Fig 6A). As also shown by our previous microarray profiling studies [33], the proteomic findings confirmed the presence of CCL2 and CXCL1/8/10/12 in the HBEC-5i secretome. Among these, CCL2 (or MCP-1), an inflammatory cytokine with multiple roles in promoting immune cell infiltration in the brain and tumor cell migration across the endothelium, was present in high abundance. CCL2 was shown to promote breast cancer brain metastasis and the invasion and adhesion of SK-OV-3 cells [48,49]. CXCL8 and CXCL1 – secreted by all cells, and CXCL12 – identified only in the HBEC-5i secretome, are known drivers of metastasis by promoting tumor cell motility, adhesion and invasion [47,50]. The overexpression of CXCL1 has been shown to increase the proliferation of ovarian cancer cells [51].

All cells secreted growth and angiogenic factors with key role in stimulating new blood vessel formation and cell growth, including various members of the VEGF, PDGF, FGF, PGH, and ANG family of proteins (Fig 6B). VEGF proteins were particularly abundant in all cancer and endothelial cell secretomes, whereas ANG and ANGPT2/ANGPTL4 in SK-OV-3, suggesting a potentially significant role in the formation of the HBEC-5i/SK-OV-3 networked constructs. While ANG (angiogenin) directly stimulates blood vessel formation, ANGPT2 (angiopoietin 2) contributes to vessel destabilization [52]. Nonetheless, it has been shown that ANGPT2 has a context-dependent role, facilitating endothelial cell proliferation and migration in the presence of VEGF [53,54]. Notable in the secretome of cancer cells was also the high level of CSF1 – a cytokine with role in inflammatory regulatory processes, adhesion and migration [52], and of GDF15 – a multifunctional protein involved in many processes of relevance to cancer progression and metastasis such as EMT, remodeling of the TME, immune escape, and development of therapeutic resistance (e.g., anti PD-1/PD-L1 therapies) [52,55].

In the cell-membrane fraction of all cells, cell adhesion molecules were detected in high abundance, in particular ITGB1, ITGA6, and CD44 (Fig 6C). ITGB1 and ITGA6 form a heterodimer α6β1 that binds laminin and collagen, mediating thus the attachment of cells to the ECM [56]. High expression in cancer cells was associated with stemness and increased metastatic potential [56]. CD44 – a receptor for various ECM components (e.g., hyaluronic acid, MMPs, osteopontin, collagen) that is overexpressed on cancer cells, is a cancer stem cell marker and promoter of metastasis by facilitating adhesion and migration processes [57,58]. CD44 and ICAM1 (or CD54, detected in high abundance in HBEC-5i and lower abundance in MDA-MB-231) have been also shown to promote cancer cell clustering ability along with adhesion to endothelial capillary structures [57,59].

As expected, the HBEC-5i cells expressed much higher levels of VCAM1 (vascular cell adhesion molecule, or CD106) – an adhesion molecule involved in the reorganization of endothelial tight junctions and mediation of immune cell infiltration from the bloodstream [60], and CDH2 (N-Cadherin or CD325) – a molecule that mediates cell-cell adhesion contributing to the stabilization of adherens junctions and integrity of the vascular barrier, as well as to angiogenesis and tumor progression [61,62]. The cancer cells expressed L1CAM (CD171) and EPCAM (CD326), which are prognostic/diagnostic biomarkers often overexpressed on cancer cells, with roles in supporting cell motility, invasion, proliferation, and metastatic progression [63,64]. LICAM also facilitates the adhesion of cancer cells to the endothelial layer to wrap around the brain capillaries for better access to oxygen and nutrients. CDH1 (E-cadherin or CD324), a classical cadherin expressed by epithelial cells that prevents cancer cell invasion by maintaining cell-cell adhesions [65], was identified only in the cancer cells, in very low abundance in MDA-MB-231 and somewhat higher abundance in SK-OV-3 cells. The SK-OV-3 cells expressed, however, a higher level of CDH2, suggesting the presence of a mesenchymal, more aggressive phenotype. The tight junction proteins TJP1/TJP2 (ZO-1/ZO-2) and OCLN were identifiable, but in much lower abundance than the other cell-membrane proteins, and CLDN5 was not detectable by MS.

Cell viability and long-term stability of the model system. The endothelial/cancer cell models were maintained for up to 30 days, and cell viability was assessed with red propidium iodide stain in HBEC-5i/SK-OV-3 and a green dead cell stain in HBEC-5i/MDA-MB-231 co-cultures (Fig 7). Fluorescence imaging revealed the presence of rather few dead cancer cells on the top of mostly live HBEC-5i cells, indicating that the culture environment and the cancer-endothelial interactions contributed to the survival of cells, the stability of the cellular structures, and the creation of a robust and biologically relevant 3D model of natural tissue.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Long-term viability of cells in the pre-mixed endothelial/cancer (1:1) co-culture models assessed by fluorescence microscopy and dead cell staining.

Top panel: HBEC-5i/SK-OV-3 (turbo GFP/green) model; (A) Green fluorescent SK-OV-3 cells (FITC filter); (B) Dead cells (red) visualized by PI staining (TRITC filter); (C) Merged image of transmitted, FITC- and TRITC-filtered images of co-cultured cells. Bottom panel: HBEC-5i/MDA-MB-231 (turbo FP602/red) model; (D) Red fluorescent MDA-MB-231 cells (TRITC filter); (E) Dead cells (green) visualized by staining with EasyProbe dye (FITC filter); (F) Merged image of transmitted, TRITC- and FITC-filtered images of co-cultured cells. Images were acquired after 28 days of co-culture.

https://doi.org/10.1371/journal.pone.0349061.g007

Application of the 3D co-culture model. As 3D co-culture systems replicate more accurately the in vivo environment and enable more reliable predictions of drug efficacy and resistance, the above-described model was evaluated for its suitability in anti-cancer drug screening applications. SK-BR-3 breast cancer cells, which overexpress HER2 receptors and can therefore be effectively targeted with drugs, were used for this purpose. HBEC-5i and red-fluorescent SK-BR-3 cells were mixed in a 1:1 ratio and co-cultured in regular DMEM/HG culture medium until the accumulation of SK-BR-3 cells in the endothelial structures was observed. The constructs presented similar morphological characteristics to the ones observed in the HBEC-5i/MDA-MB-231 model, with cancer cells clustered on the structures created by the endothelial cells. We investigated the effect of Lapatinib, a highly selective, reversible Tyr kinase inhibitor that targets the EGFR (HER1) and HER2 receptors [66], and which possesses the important property of being able to cross the BBB for therapeutic purposes. We have previously shown that the drug can exert broad effects on the SK-BR-3 cells, inhibiting cell-cycle proliferation, adhesion, and migration processes, and leading to cell death and detachment upon prolonged exposure to the drug [34]. Here, we compared the drug response of SK-BR-3 cells in 2D monoculture and 3D co-culture systems by visualizing the dead cells after exposure to the drug (Fig 8). Confluent monocultures of HBEC-5i and SK-BR-3 were used as controls. There was no impact of drug addition to the HBEC-5i cells, which displayed no increase in the numbers of dead cell counts even after 72 h exposure to Lapatinib (10 µM) (note the few cells labeled with green dead cell stain in Fig 8A,B). The SK-BR-3 controls, on the other hand, responded as expected, displaying progressively more dead cells with increased exposure time to the drug, ultimately detaching and leaving large surface areas of the culture plate exposed (note the red/live and yellow/dead SK-BR-3 cells in Fig 8C,D, as well as the exposed plate areas). Interestingly, in the 3D co-culture model, the viability of cancer cells was increased, indicating a protective role exerted by the endothelial cells. Major differences between the non-treated co-cultures and the ones treated with 1 µM or 10 µM Lapatinib solution, for 48 h or 72 h, were not observed. Moreover, the cancer cells did not detach from the endothelial cells (Fig 8E-H). While more work will be needed to uncover the mechanistic details of this behavior, the results support the hypothesis that the interaction of cancer cells with other cells in the tumor environment increases the resistance to treatment with therapeutic drugs [67]. To validate the potential of the model system for drug screening applications and enhance its predictive power, future quantitative investigations incorporating a broader range of drugs will also have to be conducted.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 8. Lapatinib treatment of HBEC-5i and red fluorescent SK-BR-3 cells in mono- and co-culture systems.

Top panel: monoculture controls of HBEC-5i and SK-BR-3 cells treated with Lapatinib (10 µM); (A,B) Dead cells (green) in HBEC-5i monocultures; (C,D) Dead cells (yellow) in red fluorescent SK-BR-3 monocultures. Bottom panel: Co-culture HBEC-5i/SK-BR-3 (turbo FP602/red) cell model treated with Lapatinib (1 and 10 µM). (E-H) Dead cells (yellow) in the endothelial/cancer co-culture system. EasyProbe green dye (FITC filter) was used to visualize the dead cells in all cultures.

https://doi.org/10.1371/journal.pone.0349061.g008

Discussion

Intercellular interactions established in the complex TME are essential to cancer cell survival, growth, and metastasis. Due to a limited understanding of these interactions in the unique brain TME, cancer brain metastases are often associated with poor prognosis. While the development of novel in vivo mouse models (patient derived xenografts, syngeneic, or genetically engineered mouse models) has significantly advanced our knowledge of tumor biology, the establishment of such models is time consuming and expensive, and the visualization of cancer cell behavior within the TME is challenging. Importantly, interspecies genetic and physiological disparities lead to skewed or inaccurate representations of tumor-host TME interactions, further limiting the use of animal models. Alternative 3D models (e.g., spheroids or organoids) do not fully replicate the complexity of the TME but can simulate key aspects of tumor morphology, composition, and chemical signaling. Nonetheless, difficulties in reproducible culturing and the need for animal-origin or synthetic scaffolding materials that contaminate the environment and induce additional variable effects, impose further restrictions on their use.

In this work we established an in vitro scaffold-free 3D co-culture model for studying the interactions between cancer and brain endothelial cells, and their role in tumor development. Abundant collagen production by the HBEC-5i cells, as supported by proteomic data, created the requisite endogenous substrate for simulating 3D cellular interactions and enabling self-assembly in 3D networked constructs that mimic the natural tissue architecture, in a reproducible and cost-effective manner. The 3D constructs were established without the use of exogenous scaffolding materials or supportive techniques and technologies that help cells self-assemble into aggregates such as physical agitation, hanging drop, liquid overlay, magnetic levitation, bioprinting, or microfluidic guiding [20,68]. While the monoculture of endothelial cells showed the initiation of microvascular-mimicking structures after 10 days of culture, in endothelial-cancer co-cultures, these structures were already visible after 6–7 days. The cancer cells affected the phenotype and behavior of the cells in the 3D constructs, as evidenced by the different morphology of the generated constructs. The integrated brain endothelial and highly aggressive cancer cell constructs enabled the examination of cell migration, 3D structure formation, and long-term cell viability, offering valuable insights into the mechanisms that drive tumor cell behavior at the brain metastatic site. The co-culture model was observed up to 30 days, with live-dead cell staining showing only a small proportion of dead cells in the system.

The observed interactions between cancer and endothelial cells were driven by cell-membrane and secreted factors from both cell types. While collagen deposition by HBEC-5i cells provided for a fertile matrix that enabled the development of the 3D model, proteomic data revealed that the HBEC-5i, SK-OV-3, and MDA-MB-231 cells expressed a wide array of chemokines and adhesion, growth, and pro-angiogenic factors that supported migration, angiogenic, and proliferation pathways. Chemokine-driven processes in the TME, including signaling sustained by CCL2, CXCL1, CXCL8, CXCL10 and CXCL12, guide not only the migration of both cancer and endothelial cells [47–51], but also the remodeling of endothelial cells to ultimately support cancer progression [45]. Angiogenic factors secreted by cancer cells (e.g., VEGFs, PDGFs, TGFB1) have been shown to interact with VEGFRs, PDGFRs, Endoglin (ENG), Neuropilin1 (NRP1), and integrin receptors on endothelial cells to activate signaling pathways that promote sprouting angiogenesis and endothelial vasculature in the TME [52]. Moreover, studies that investigated endothelial-tumor interactions have emphasized the important role played by endothelial cell-surface integrins in modulating adhesion, migration, invasion, and vascular remodeling processes in cancer models [69]. Such intercellular interactions that steer the development of microvasculature structures and the migration of cancer cells towards the endothelial cells were observed in the newly developed 3D in vitro model system, supporting its physiological relevance. Validation with a transwell assay further confirmed the role of paracrine signaling in shaping the complex TME. Furthermore, the presence of key secretome cytokines with roles in innate immunity and inflammatory processes (e.g., CSF1 [52]), and of various growth factors implicated in the many steps of cancer progression, uncovered additional areas of research that could benefit from the model and prospects for the development of novel prognostic/diagnostic assays and targeted therapies. The co-culture system displayed prolonged stability, holding thus significant potential for investigating anti-cancer drug responses. Subjecting the HBEC-5i/SK-BR-3 3D model to the receptor Tyr kinase inhibitor Lapatinib revealed a differential drug response characterized by reduced cell morbidity relative to 2D SK-BR-3 monocultures, exposing thus the protective role conferred by the presence of endothelial cells. Although qualitative, these observations underscore the value of the model system as a preclinical drug-testing platform.

Conclusions

We developed a novel, scaffold-free, in vitro 3D brain endothelial/cancer cell model system that mimics the behavior of cancer cells in the early establishment of the metastatic TME. By using stable transfected fluorescent cancer cells, the model provided a platform for observing the dynamic interplay between cancer and endothelial cells, i.e., the migration, positioning, and attachment of cancer cells to the high-density endothelial networked microstructures, the cancer-induced remodeling of the co-culture environment, and the morphological changes in endothelial cells. The model presents several unique features when compared to other reported in vitro systems. First, the choice of HBEC-5i cells enabled this model to better recapitulate the brain microvasculature due to the expression of well-known tight junction, adhesion, and transport proteins. Second, the model was built based on the inherent high collagen secretion property of HBEC-5i cells without using any commercially available ECM components, thus eliminating contamination risks and making it cost effective. Third, the secretion of key chemokines, pro-angiogenic molecules, growth factors, and ECM enzymes by both cancer and endothelial cells, as well as the expression of specific tight junction and cell adhesion molecules, enabled the spontaneous formation of 3D constructs that mimicked the brain microvasculature with attached cancer cells without any need to control the cell positioning process. Finally, long-term stability and applicability to exploring drug responses underscored the physiological relevance of the model and its potential to accelerate the development of therapeutic interventions. Future work will explore the incorporation of other brain resident stromal and immune cells to better replicate the in vivo microenvironment and enable more accurate investigations of the dynamic cellular interactions that delineate the behavior of cancer cells at the brain metastatic site.

Supporting information

Acknowledgments

We thank Yunqian Zhang for establishing the fluorescent MDA-MB-231 cell culture stock. The authors acknowledge the use of Grammarly, Google Gemini, and Open AI ChatGPT for minor editorial improvements to the text, including grammar and sentence syntax. The research conception and design, data acquisition and analysis, and interpretation of results are the sole intellectual contribution of investigators.