Introduction
Brain metastases, most often originating from lung cancer, breast cancer and melanoma, have a particularly dismal prognosis. Despite significant therapeutic advances in extracranial malignancies, management of brain metastases is still an unmet clinical challenge. Several mechanisms may be responsible for the ineffective treatment strategies, including exclusion of systemic agents by the blood-brain barrier (BBB) [
17], dense vascularization of the brain [
35] and protective effects of astrocytes on tumour cells [
42]. Nevertheless, cells of the neurovascular unit (NVU) have a decisive role in the fate of brain metastatic tumour cells [
40].
The concept of the NVU emerged to emphasize the unique intimate relationship between brain cells and the cerebral vasculature [
12]. Regulated by input from pericytes, glial cells and neurons, cerebral endothelial cells (CECs) take the centre stage of the NVU to form and maintain the BBB. In contrast to peripheral endothelial cells, CECs are connected to each other by continuous tight junctions (TJs), formed by transmembrane proteins like claudin-5 or occludin and a cytoplasmic plaque. Brain metastatic cells have to take up the challenge of opening or overcoming this tight endothelial barrier before diapedesis through brain microvessels. Accordingly, tumour cells spend several days arrested in the lumen of cerebral capillaries before migrating through the vessel wall [
19], despite the severe stress they are exposed to in the blood stream [
32]. It is largely unknown what changes are induced in the endothelium during this long time spent by the tumour cells intravascularly. Using in vitro and ex vivo methods, we have previously shown that metastatic cells may up-regulate expression of N-cadherin and mesenchymal markers in CECs during a process called endothelial-mesenchymal transition (EndMT), which can facilitate transmigration of the tumour cells [
16]. EndMT is indeed a long-lasting process, which was seen after 2 days in CECs exposed to factors secreted by melanoma or breast cancer cells. Ex vivo, we detected endothelial N-cadherin upregulation in the vicinity of extravasating breast cancer cells and close to the growing metastatic lesions at later time points; however, breast cancer cells migrated through the cerebral endothelium in an N-cadherin-independent manner [
10].
Diapedesis itself has been even less characterized in in vivo
/ex vivo models. It was suggested to take place exclusively from capillaries [
19] with the tumour cell being pinched at the transmigration hole on the vessel wall [
15]. CECs were shown to extend protrusions covering extravasating mammary carcinoma cells [
10], indicating active involvement of the endothelium. However, several questions remain to be answered. What are the changes induced in the cerebral endothelium by tumour cell extravasation in vivo? Is tumour cell-induced endothelial dysfunction reversible or irreversible? Is there any endothelial apoptosis, proliferation or any type of new vessel formation taking place during initial steps of brain metastasis formation?
After completing transvascular migration, metastatic cells can only survive in the brain environment if attached to the abluminal surface of the vessel wall [
3]. By this time, tumour cells come in contact with other cells of the NVU. Immediate and persistent peritumoural astrogliosis and microglial reactions are the most important in this respect [
40]. It is not known, however, how the glia limitans perivascularis is affected in initial and later phases of brain metastasis formation. Vascular changes during metastatic outgrowth – except for vessel co-option – also need better understanding.
By using real-time in vivo and ex vivo microscopy, here we aimed at unravelling morphological and functional changes in cerebral vessels and cells of the NVU before, during and after transmigration of breast cancer cells through the BBB.
Material and methods
Cell culture and in vitro models
4T1 mouse triple negative mammary carcinoma cells were kept in Roswell Park Memorial Institute (RPMI) 1640 medium (Pan Biotech, Aidenbach, Germany) supplemented with 5% foetal bovine serum (FBS, PAA Laboratories, Linz, Austria) and Glutamax (Thermo Fischer Scientific, Waltham, MA, USA). 4T1 cells were transfected with pcDNA3.1(+)/Luc2 = tdT plasmid using Lipofectamine 2000 (Thermo Fischer Scientific) and underwent single-cell cloning, after sorting of red fluorescent cells using a BD FACSAria Fusion flow cytometer (BD Biosciences, San Jose, CA, USA). The media of tdTomato-4T1 cells contained 500 μg/ml G418 (Thermo Fischer Scientific) for further selection and maintenance of red fluorescence. Emerald GFP-expressing EmGFP-4T1 cells were obtained by retroviral transfection and selected on blasticidin S (Sigma-Aldrich, St. Louis, MO, USA). All cell lines were regularly tested for mycoplasma contamination using the MycoAlert Mycoplasma Detection Kit (Lonza, Basel, Switzerland). Only mycoplasma-negative cultures were used for experiments.
Venus-YFP-expressing primary mouse brain endothelial cells (MBECs) were isolated from 6- to 8-week-old FVB/Ant:TgCAG-yfp_sb #27 female mice (obtained from Institute of Experimental Medicine, Budapest, Hungary). After collection of the brains, the meninges were removed and cerebral cortices were cut into small pieces and digested in two steps with collagenase and collagenase/dispase. Microvessel fragments were collected after 10 min 1000 · g centrifugation on Percoll (Sigma-Aldrich) gradient, and plated onto fibronectin/collagen-coated dishes. Endothelial cells growing out of the microvessels were cultured in DMEM/F12 (Thermo Fisher Scientific), 10% plasma-derived serum (PDS, First Link, Birmingham, UK) and growth factors. In the first two days, 4 μg/ml puromycin (Sigma-Aldrich) was added to remove contaminating cells.
YFP-MBECs and tdTomato-4T1 cells were used for endothelial-tumour cell co-cultures. First, we cultured endothelial cells on the abluminal side of the filter inserts (Corning-Costar Transwell Clear, Corning, NY, USA, #3450) coated with collagen. Tumour cells were seeded on the luminal side in a number of 4.5 · 104/cm2 and co-cultured for 48 h.
Experimental animals and surgeries
All surgeries were carried out on 8-week old female BALB/c (The Jackson Laboratory) or FVB/Ant:TgCAG-yfp_sb #27 mice. Before every procedure, mice were anaesthetized via inhaled isoflurane 4% (v/v) in oxygen for induction and 1–2% (v/v) for maintenance, from a precision vaporizer (Open Circuit Isoflurane Tabletop System, Stoelting, Dublin, Ireland). Depth of anaesthesia was monitored by toe pinch tests.
For all intravital experiments, cranial windows were used to obtain optical access to the cortex. Briefly, anaesthetized animals were mounted on a stereotaxic frame incorporating a heating pad. Craniotomy (d = 3.5 mm) was performed over the right parietal cortex with a micro drill (H.MH-170, High Speed Rotary Handpiece, Foredom, Blackstone Industries, Bethel, CT, USA) fitted with a 0.5 mm burr, followed by the removal of the dura. In some experiments, astrocytes were labelled by topical application of 10 μM SR101 (Sigma-Aldrich) in Ringer-HEPES solution for 2–3 min before the window installation. A coverslip of 5 mm diameter was then placed over the exposed brain and the edge of the glass was sealed with cyanoacrylate glue. An aluminium plate was glued onto the skull for head fixation. The exposed bone and the aluminium bar were covered with cyanoacrylate glue and dental cement (Unifast III, GC Europe, Leuven, Belgium) to increase stability. A recovery period of at least one month was allowed between implantation of the cranial window and intravital microscopy observation of endothelial-tumour cell interactions. Astrocyte-tumour cell interactions were investigated in a time frame of 30 min to 2 h after cranial window installation due to the temporary astrocyte staining. After recovery, either 106 tdTomato-4T1 cells were inoculated into the right common carotid artery or 3 · 106 tdTomato-4T1 cells were injected intracardially into FVB/Ant:TgCAG-yfp_sb #27 female mice with chronic cranial window for two-photon microscopy, or without craniotomy for ex vivo investigations. BALB/c female mice were intracardially injected with 3 · 106 EmGFP-4T1, then 1 day, 5 days or 10 days later cranial window was installed right before two-photon microscopy imaging.
For ex vivo observations, surgically untouched FVB/Ant:TgCAG-yfp_sb #27 mice received 3 · 106 tdTomato-4T1-cells intracardially. Certain animals were subjected to in vivo proliferation assay and treated intraperitoneally with 5-ethynyl-2′-deoxyuridine (EdU, Thermo Fisher Scientific, 100 mg/kg), a thymidine analogue, 24 h before tissue collection. After 1, 3, 5, 8 or 10 days, animals were anaesthetized and transcardially perfused with phosphate buffered saline (PBS, 10 mM, pH = 7.4), then with Karnovsky’s fixative (for electron microscopy) or 3% paraformaldehyde (for immunofluorescence) in PBS. Brains were removed and post-fixed by immersion in the same fixative at 4 °C overnight. The following day, the fixative was replaced with PBS (for vibratome section) or 30% sucrose solution in 0.1 M phosphate buffer (PB, for frozen sections), and the brains were stored at 4 °C until further processing.
Intravital two-photon imaging
Mice were anaesthetized with isoflurane and kept on a heating system-incorporated stereotaxic stage. The head was immobilized and positioned via the attached aluminium bar. This stable positioning and the unique pattern of the superficial pial vasculature allowed us to image the same cortex volume over days. Intravital microscopy was carried out with a FEMTO2DAlba microscope (Femtonics, Budapest, Hungary) using a 20x or 60x large working distance water immersion objective using MES software (v4.6.2336, Femtonics). Two-photon excitation was performed using a Mai Tai HP Ti-sapphire laser (Spectra-Physics, San Jose, CA, USA) at 810 nm, which was found optimal for EmGFP and CellTracker Red CMTPX excitation and also adequate for SR101 and at 900 nm, which optimally excited both tdTomato and Venus-YFP. Laser power was set to 10–40% depending on the depth of imaging (0–400 μm from the brain surface). Emission wavelengths were collected by GaAsP photomultipliers. Larger volumes (x: 500 μm; y: 500 μm; z: 250 μm) were recorded with 3 μm vertical steps to evaluate the cell number changes and dynamics of tumour cell intravascular location in the first 48 h following inoculation. To acquire sufficient red fluorescence signal, tdTomato-4T1 cells were also stained with CellTracker Red CMTPX (Thermo Fisher Scientific) for this experimental setup. ImageJ’s “3D Object Counter” plugin was used to assess the changes [
1]. High magnification z-stack images (x: 120 μm; y: 120 μm; z: 120 μm, with 1 μm steps) were recorded for studying cell morphology changes and transmigration and no additional labelling of 4T1-tdTomato was applied. Image stacks were auto-levelled, merged and converted to RGB colour in Fiji [
28].
Immunofluorescence and fluorescence microscopy
After fixation, the whole brain was mounted for freezing microtome (Reichert-Jung, Leica Biosystems, Wetzlar, Germany) or vibratome (Leica Biosystems) sectioning and sliced coronally. 50 μm brain sections were collected and stored in PBS with 0.05% sodium azide. Antigen retrieval was either omitted or performed by incubating slices at 85 °C for 60 min in PBS. Permeabilization was performed with 0.5% TritonX-100 in PBS for 30 min at room temperature, followed by blocking with 3% BSA (bovine serum albumin) in PBS. Primary antibody solutions were prepared in 3% BSA and 0.5% TritonX-100-containing PBS. Sections were incubated overnight under slow nutation. The following antibodies were used on vibratome sections: anti-AQP4 (1:100, Santa Cruz Biotechnology, Santa Cruz, CA, USA, #sc-20,812), anti-claudin-5 (1:100, Thermo Fisher Scientific, #35–2500), anti-cleaved caspase-3 (1:50, Cell Signaling, Boston, MA, USA, #9661,), anti-collagen IV (1:100, Abcam, Cambridge, UK, #ab6586) and anti-PECAM-1 (1:120, Novus Biologicals, Centennial, CO, USA, #NB100–2284); on frozen sections: anti-fibronectin (1:100, Abcam, #ab2413), anti-GFAP (1:100, Abcam, #ab7260) and anti-Iba-1 (1:100, Abcam, #ab5076). Sections were extensively washed in PBS, and the secondary antibody solution was afterward applied on them for 60 min at room temperature in the dark. Alexa Fluor 488, 594, 647 anti-rabbit, anti-mouse and anti-goat IgG (Jackson ImmunoResearch, Cambridgeshire, UK and Thermo Fisher Scientific) and STAR RED anti-mouse IgG (Abberior, Göttingen, Germany) were used as secondary antibodies in a dilution of 1:300–1:600 in 3% BSA-containing PBS. Sections were then washed with PBS, counterstained with a colour compatible nuclear staining (Hoechst 33342, Sigma-Aldrich) for 5 min, washed again with PBS, rinsed in water and mounted with an aqueous fluorescent mounting solution, FluoroMount-G media (SouthernBiotech, Birmingham, AL, USA). Visualization of nuclei with DNA synthesis was performed on sections from EdU-treated animals using Click-iT Plus EdU Alexa Fluor 647 Imaging Kit (Thermo Fischer Scientific) following the manufacturer’s instructions. ImageJ’s “3D Object Counter” plugin was used for manual assessment of EdU-positive tumour cells.
Immunohistochemistry and immunofluorescence were visualized with Leica SP5 and Leica SP8 confocal laser scanning microscopes with 63x and 100x oil immersion objectives or a STED (stimulated emission depletion) super-resolution-capable STEDYCON (Abberior Instruments, Göttingen, Germany) built on an Axio Observer Z1 inverted epifluorescence microscope (Zeiss, Oberkochen, Germany) equipped with an alpha Plan-Apochromat 100x/1.46 oil immersion objective.
Preparation of ultrathin sections and transmission electron microscopy (TEM)
TdTomato-4T1-bearing FVB/Ant:TgCAG-yfp_sb #27 mouse brains were sectioned using a VT1000S (Leica Biosystems) vibratome. 100 μm sections were collected and post-fixed for 5 h in Karnovsky’s fixative at room temperature. After post-fixation, sections were cut into 1–2 mm2 pieces. Specimens containing tdTomato-4T1 cells were selected for further processing. Samples were rinsed and post-fixed in 2% OsO4. After dehydration with a graded series of ethanol, the samples were embedded in epoxy resin (Durcupan ACM, Sigma-Aldrich) and polymerized at 55 °C for 48 h. Ultrathin sections (50 nm) were prepared with an Ultracut UCT (Leica Biosystems) and contrasted with 2% uranyl acetate (Electron Microscopy Sciences, Hatfield, PA, USA) and 2% lead citrate (Electron Microscopy Sciences), then analysed with a JEM-1400Flash transmission electron microscope (JEOL, Tokyo, Japan) fitted with an 8 MP Matataki Flash CCD camera (JEOL).
Western-blot
4T1-tdTomato cells and YFP-Venus MBECs were co-cultured for 48 h on the two sides of the filter inserts and lysed separately in radioimmunoprecipitation assay buffer. After 30 min incubation on ice, cell lysates were centrifuged at 13,000 · g for 15 min at 4 °C. Protein concentration was determined with bicinchoninic acid (BCA) (Santa Cruz Biotechnology). Laemmli buffer was added to the samples, followed by heating at 95 °C for 3 min. Prepared samples were electrophoresed using standard denaturing SDS-PAGE procedures and blotted on polyvinylidene difluoride (0.2 μm pore size from Bio-Rad, Hercules, CA, USA and the 0.45 μm pore size from BioTrace, Pall Corporations, Port Washington, NY, USA) membranes (β-actin and fibronectin, respectively). Afterwards, the non-specific binding capacity of the membranes was blocked with 3% BSA or 5% non-fat milk in TBS-T (Tris-buffered saline with 0.1% Tween-20). Membranes were incubated with primary antibodies diluted in TBS-T: anti-β-actin (1:1000, Santa Cruz Biotechnology, #sc-47,778) or fibronectin (1:1000, Abcam, #ab2413). Blots were washed in TBS-T three times for 10 min. Horseradish peroxidase (HRP)-conjugated secondary antibodies were diluted in TBS-T as follows: 1:3000 anti-rabbit IgG and anti-mouse IgG (Jackson ImmunoResearch) and added for 1 h and then washed again in TBS-T. Immunoreaction was visualized with Clarity Chemiluminescent Substrate (Bio-Rad, Hercules, CA, USA) in a ChemiDoc MP System (Bio-Rad). Densitometry analysis was performed with the Image Lab Software, version 5.2 (Bio-Rad).
Discussion
The microenvironment is increasingly recognized as a profound determinant of tumour progression and therapeutic outcome. Composed of neurons, glial and vascular cells and protected by the BBB, the brain environment is very special, generally able to host only a few cancer types. Among these, triple negative breast cancer is one of the most relevant, having the worst outcome among mammary carcinoma subtypes [
13].
Due to the complex anatomy and intercellular communication network, in vitro models are less suitable to study mechanisms related to the brain environment. In addition, early steps of brain metastasis development cannot be observed in human, because of the detection limit of current diagnostic tools. Xenograft models are devoid of a full immune response, while injection of tumour cells into the brain parenchyma is not recapitulating the extravasation step [
34]. Therefore, we applied an in vivo animal model and followed the interactions of brain resident cells with metastatic breast carcinoma cells. Since only a limited number of metastatic lesions were observed in the peripheral organs of the mice, our in vivo setup is a relevant model of brain metastasis formation with the condition that tumour heterogeneity and interspecies differences (translation from mouse to human) are not addressed.
Among cells of the NVU, CECs and astrocytes are the most active in immediately responding to and continuously associating with invading tumour cells [
15,
19]. Both CECs and astrocytes have a Janus-faced attitude, bearing both tumour-destructive and -protective mechanisms [
40]. Much less is known about the role of pericytes, which regulate the permeability of the blood-tumour barrier [
20] and contribute to connective tissue accumulation in the metastases [
33].
During the long-lasting arrest in the lumens of cerebral capillaries, several changes may take place both in cancer and in endothelial cells. EndMT is one mechanism, the development of which is a slow process, lasting for a few days [
16]. EndMT is characterized by loss of TJ proteins, switch from VE- to N-cadherin and up-regulation of mesenchymal markers, like collagen, fibronectin and α-smooth muscle actin. We have previously described up-regulation of N-cadherin in CECs of tumour cell-hosting capillaries; however, N-cadherin proved to be dispensable for the extravasation of breast cancer cells [
10]. Here we observed no reduction in the expression of claudin-5 – the protein forming the backbone of endothelial TJs – throughout the metastatic process. Although we saw up-regulation of extracellular matrix proteins, which are also markers of EndMT; however, mainly in tumour and not in endothelial cells. Therefore, we conclude that EndMT is probably not a mandatory mechanism involved in extravasation of triple negative mammary carcinoma cells to the brain, at least in mouse.
Although the extravasation occurred late (after 4–5 days or even later), surviving and finally diapedesing cells reached their final position in the capillary lumen early (in the first 24 h). The majority of arrested tumour cells disappeared in the first 2 days. Time-dependent reduction of the number of arrested tumour cells may be a consequence of mechanical stress, anoikis or immune attacks [
32]. Mechanical stress may be primarily caused by deformation of the tumour cells in the narrow vessels, because fluid shear is practically completely excluded by the obstruction of the lumen by the tumour cell itself and the endothelial plugs, described here. Flow may not only induce shear stress, but may also influence different steps of tumour cell extravasation. In a zebrafish model, reduced flow was found to promote early arrest of tumour cells, while increased flow enhanced extravasation because integrin-dependent adhesion forces rapidly exceeded shear forces [
9]. Therefore, isolation of tumour cells from the blood flow may have complex consequences on the metastatic process.
According to our results, anoikis seems to be a rare event, since we could seldom find cleaved caspase-3-positive tumour cells. Cell death was also rarely observed in endothelial cells coming in contact with cancer cells, leading to the formation of large outlets for extravasation of the tumour cells through the vessel wall. However, the majority of metastatic transmigrations occurred through small (< 5 μm) pores. In more than one third of the cases, the lowest resistance point through which the diapedesis took place was found by the tumour cell after the development of multichannelled capillaries. Using in vitro models, we have previously shown that brain endothelial cells extend filopodia-like membrane protrusions towards breast cancer cells, incorporating them and facilitating the transcellular route of transendothelial migration [
10]. Endothelial protrusions covering extravasating cells were seen in vivo as well [
10], and this seems to be the mechanism of the formation of multiple channels within a vessel. This type of transmigration has been previously described as “endothelial covering-type extravasation” in a zebrafish model of HeLa tumour formation [
14] and “endothelization” in a mouse pulmonary melanoma metastasis model [
18,
26].
In fact, the process of endothelial protrusion formation and consequent multilumination of the vessel resembles the first step of intussusceptive angiogenesis [
5], when endothelial bridges are formed [
25]. Collateral capillaries spanning the tumour cell-affected vessel may form through this mechanism. However, according to our observations, the main tumour vascularization mechanism [
41] in mouse brain metastases is vessel co-option. During this process, cancer cells proliferate attached to the abluminal surface of already existing capillaries [
15] interacting with the basement membrane [
3,
36].
It is still a question of debate whether tumour cell migration through the vessel wall induces a reversible or irreversible damage of the endothelium. Using an in vitro model, we have previously suggested that melanoma cells induce apoptosis of CECs [
8]. However, melanoma and breast cancer cells seem to have a distinct behaviour in the brain, showing different transmigration efficiencies and routes (paracellular of melanoma and mainly transcellular of breast cancer) [
10,
23]. Here we observed that triple negative breast cancer cells might induce endothelial death in the mouse brain during their extravasation; however, more often, the endothelium could recover even from severe structural changes. These changes included blebbing of the endothelial cell membrane, which was observed with a frequency of almost 80% in the neighbourhood of tumour cells. Blebbing is usually associated with apoptosis [
4]; however, non-apoptotic membrane blebbing has also been described in a wide variety of cell types in response to multiple stimuli [
7]. Non-apoptotic endothelial blebbing has been observed in oxidative stress [
37] and adhesion [
24]. The role of endothelial bleb formation in the metastatic process remains to be elucidated. Moreover, it is also not clear whether the extent of blebbing (from a few basolateral vacuoles to complete restructuration of the vessel wall) has any impact on the tumour cells.
Nevertheless, blebbing involves rearrangement of the membrane and the cytoskeleton, a process dependent on the Rho family of small GTPases, especially activation of Rho-kinase (ROCK) [
7]. In apoptotic cells, activation of ROCK by caspase-3 seems to be responsible for bleb formation [
29]. In addition, ROCK activation has also been linked to tumour cell motility. In line with this, we frequently observed tumour cell blebbing during extravasation in the brain. Generally, the blebbing membrane of the tumour cell was the first to squeeze through the vessel wall, followed by the cytoplasm and the nucleus. The bleb-associated mode of tumour cell motility that does not require proteolysis and is associated with a rounded cell morphology, the so-called amoeboid migration, is dependent on ROCK activation [
27]. However, our previous in vitro results indicated that fostering the amoeboid migration by Rac inhibition hampers not only melanoma, but also breast cancer cell transmigration through brain endothelial monolayers [
23,
39]. It remains to be established how small GTPase signalling and blebbing in endothelial and tumour cells affect breast cancer brain metastasis formation.
Another interesting observation of our study is that tumour cells breach the glia limitans perivascularis during their extravasation. This is important because immune cells – diapedesis of which is used as a reference in deciphering mechanisms of extravasation of tumour cells [
26,
31] – also have to migrate through the glia limitans perivascularis to induce neuro-inflammation [
6,
11]. However, as a metastatic lesion is growing, AQP4-positive astrocyte end-feet are disappearing from the vessel to cover the surface of the tumour, from which reactive astrocytes are completely expelled. Since in parallel with vanishing from the capillaries, AQP4 accumulates at the border of the tumour, we conclude that this is rather an extraction of astrocyte end-feet from the vessels than loss of polarization, as previously suggested [
20]. Interestingly, claudin-5 staining of the vessels remained continuous even after loss of direct contact with astrocyte end-feet, probably due to the presence of pericytes [
2]. Indeed, disappearance of astrocyte foot processes from metastatic vessels did not show any correlation with the permeability [
20]. Although we cannot exclude the possibility that some vessels become leaky due to partial opening of the TJs, our results suggest that this is not mandatory in brain metastatic lesions.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.