Active remodeling of capillary endothelium via cancer cell-derived MMP9 promotes metastatic brain colonization

Abstract

Crossing the blood-brain barrier is a crucial, rate-limiting step of brain metastasis. Understanding of the mechanisms of cancer cell extravasation from brain microcapillaries is limited as the underlying cellular and molecular processes cannot be adequately investigated using in vitro models and end-point in vivo experiments. Using ultrastructural and functional imaging, we demonstrate that dynamic changes of activated brain microcapillaries promote the mandatory first steps of brain colonization. Successful extravasation of arrested cancer cells occurred when adjacent capillary endothelial cells (ECs) entered into a distinct remodeling process. After extravasation, capillary loops were formed, which was characteristic of aggressive metastatic growth. Upon cancer cell arrest in brain microcapillaries, matrix-metalloprotease 9 (MMP9) was expressed. Inhibition of MMP2/9 and genetic perturbation of MMP9 in cancer cells, but not the host, reduced EC projections, extravasation, and brain metastasis outgrowth. These findings establish an active role of ECs in the process of cancer cell extravasation, facilitated by crosstalk between the two cell types. This extends our understanding of how host cells can contribute to brain metastasis formation and how to prevent it.

Introduction

Brain metastasis (BM) is the most common tumor in the central nervous system and cause high morbidity and mortality in cancer patients (1), but the initial steps of BM development are still poorly understood. Early brain colonization is an inefficient process: only few arrested cells manage to extravasate out of the microcapillaries into the brain parenchyma and grow further to clinically relevant BM (2,3).

Cancer cell extravasation has been proposed to show similarities to leukocyte transmigration, involving disruption of endothelial cell (EC) junctions and subsequent transmigration across the endothelial layer (4). Current evidence supports a more complex concept, in which metastatic extravasation might take place by various mechanisms: interactions between cancer cells and ECs (5–7), immune cells (8,9), platelets (10), and also cancer cell-specific factors like induced programmed necrosis of lung ECs (11). Thus, the extravasation of cancer cells seems to depend on important interplays between malignant cells and host factors. However, many key mechanisms of this first crucial and rate-limiting step of the metastatic process remain obscure (4,12).

Generally, the blood-brain barrier (BBB) restricts the transfer of molecules, immune cells and therapeutics that may harm the brain, thus creating an especially difficult border for cancer cells to cross (2,4,7,13). Upon vascular arrest, brain metastatic cancer cells express specific proteins to increase their adhesion to the brain EC (2,12), and can actively contribute to the subsequent disruption of EC junctions (14–16). Compared to cancer cells in other organs, brain-metastatic cells remain intravascular for an extended time before extravasation (2,3), indicating that extravasation from the brain microcapillaries is a complex and time-consuming process. Recent work demonstrated structural changes to the brain ECs during early arrest and extravasation (7), without clarifying its functional role and molecular mechanism.

Studying the mechanisms of cancer cell extravasation from brain microcapillaries is challenging since the complex and dynamic cellular and molecular processes during cancer cell extravasation cannot be adequately understood using in vitro models and end-point in vivo experiments (4). Here, to optimally address these long-standing questions, a newly developed multimodal correlative microscopy technology was employed to track single extravasating cancer cells in various mouse models of BM. This allowed discovering a collaboration between cancer cells and ECs, resulting in extravasation of the cancer cell. MMP9 released by the cancer cell induces formation of EC filopodia-like projections, similar to a known vascular self-clearing mechanism from thrombotic material.

The novel extravasation strategy described here allows the cancer cell to overcome an anatomical structure as tight as the BBB.

Materials and methods

Cell lines

The murine tdTomato-expressing brain-passaged mammary adenocarcinoma E0771 cells, triple negative (17), a kind gift from Cyrus M. Ghajar and Patricia S. Steeg, originated from: RRID:CVCL_GR23) were cultured in DMEM with 10% fetal bovine serum (FBS), 1% penicillin/streptomycin (pen/strep). The murine Ret cells were isolated from spontaneous primary cutaneous melanoma from the ret-transgenic mouse model (18). The Ret cells (also referred to as CM #43) were transfected with px330 expressing a single guide RNA targeting the tyrosinase (Tyr) gene (targeting sequence tgcctcctctaagaacttgt). Single cell clones were established and successful targeting of the Tyr locus was verified by NGS (MiSeq platform). One isolated clone was color-coded by retroviral transduction using the pRp retroviral plasmid encoding for tdTomato. After retroviral infection polyclonal cell cultures were passaged for at least more than three times before use in experiments. To increase the brain metastatic potential, a requirement for meaningful in vivo microscopy studies with these cells, typical brain-passaging was performed. Here, 5*10⁵ parental Ret cells were intracardially injected in C57BL/6J mice. After a maximum of two weeks, the brains were removed and the cortex was dissociated and put into culture. Cancer cells were expanded in vitro and the same process was repeated to generate four rounds of brain-passage, generating the Ret-Br cell line. The brain-passaged human breast cancer cell lines Jimt-1 (ER-, PR-, HER2 amplification, trastuzumab resistant, p53-/-, a kind gift from Patricia Steeg, originated from: RRID:CVCL_2077) and A2058 (BRAF-V600E+/-, PTEN+/-, RB1+/-, p53-/-, RRID:CVCL_1059) were cultured in DMEM, 10% FBS and 1% pen/strep. For fluorescence imaging, the cells were either transduced with cytoplasmic green fluorescent protein (plKO.1-puro-CMV-TurboGFP, SHC003, Sigma-Aldrich, USA) or cytoplasmic tdTomato (LeGo-T2, plasmid #27342, Addgene, USA). All human cell lines were authenticated based on Single Nucleotide Polymorphysm (SNP) typing. Cells were used for experiments between passage two and six after thawing. During the whole duration of the study, all cell lines were tested via PCR every three months for mycoplasma contaminations.

Animals and surgical procedures

To establish constitutive labeling ECs in VE-Cad Cre^ERT2xRosa26-YFP^fl/fl mice (19,20), 4–5-week-old animals were administered four doses of tamoxifen (100 mg/kg) via intraperotoneal injection twice weekly, as described previously (20). VE-Cad Cre^ERT2xRosa26-YFP^fl/fl mice, Foxn1 Nu/Nu (Charles River, Germany) or NOD-scid IL2rγ^null (NSG, initially generated by Jackson Laboratory) mice (>8 weeks old), received a chronic cranial window supported by a titanium ring as described previously (3). Female mice were used for the breast cancer BM models, and for the melanoma BM models and microsphere injections, male mice were used. Minimally three weeks following window implantation, the mice were adequately anesthetized with ketamine/xylazine and injected into the left ventricle with 5 x 10⁵ cancer cells suspended in sterile PBS or with fluorescent microspheres in 0.9% sterile NaCl solution (1.25 x 10⁴ microspheres/mL, 100 μL, FluoSpheres red fluorescent polystyrene Microspheres, Ø 10 μm, F8834, Thermo Fisher Scientific USA). For histological analyses, 8-12-week-old female C57BL/6 mice (Janvier Labs) were used. All animal procedures were performed in accordance with the institutional laboratory animal research guidelines after approval of the local governmental Animal Care and Use Committee (Regional Council Karlsruhe, Germany, 35-9185.81/G-220/16 and 35-9185.81/G-273/19). Efforts were made to minimize animal suffering and to reduce the number of animals used according to the 3R’s principles. All mice were routinely checked for clinical endpoint criteria.

In vivo multiphoton laser scanning microscopy

A ZEISS LSM 7MP equipped with a Coherent Chameleon Ultra II or a Coherent Discovery NX was used for intravital microscopy as described previously (3). Images were acquired using a 20x/1.0 W-Plan-Apochromat obejective (ZEISS) and a BP500-550 / BP575-610 Filter (ZEISS), at gains set between 600-800 while keeping the laser power as low as possible to prevent photo-toxicity. Z-interval of the stacks was 1 or 3 μm, depending on the zoom of the image. Angiograms were obtained by tail vein injection of 100 μL of tetramethylrhodamine-isothiocyanate-Dextran (5 mg mL^-1, TRITC Dextran; average MW: 500,000, Sigma-Aldrich). The excitation wavelength was selected depending on the fluorophore imaged: 850 nm for GFP and TRITC-Dextran and 950 nm to visualize YFP, fluorescent microspheres, and tdTomato. The mice were narcotized with 0.5%-2% isofluran and painlessly stabilized using the titanium ring in a custom-built fixation system. By correlating the stereotactic coordinates of the microscope and the superficial blood vessel architecture it was possible to retrieve the same region at different time-points over weeks.

Multimodal correlative microscopy

Fluorescent Jimt-1 cells were retrieved for 3DEM by multimodal correlative microscopy as described before (21). Three to seven days following cancer cell injection, the fluorescent cells were visualized using in vivo imaging or confocal microscopy (LSM 780 NLO, ZEISS, Germany) of 100 μm thick vibratome sections of the perfused mouse brain. Following perfusion fixation, near-infrared branding was used to mark the region of interest on the tissue surface; either on the surface of the brain (following intravital microscopy) or on 100 μm thick brain vibratome sections. A <1 mm³ tissue block containing the cancer cell was dissected and the samples processed for electron microscopy in a PELCO Biowave Pro microwave (Ted Pella, USA). The samples were washed in cacodylate buffer (pH 7.4) and fixed with 1% OsO₄ (Electron Microscopy Sciences, USA) and 1.5% K₄Fe(CN)₆, rinses and secondary fixation with 1% OsO₄, staining with 1% uranyl acetate (Serva Electrophoresis GmbH, Germany) and stepwise dehydrating in ethanol and resin infiltration (21). The resin was polymerized at 60°C for 1 or 3-4 days, depending on the volume of the sample. A flat blockface and sides were trimmed around the EM-processed sample and a small resinblock containing the biopsy was selected and imaged with the x-ray microCT Bruker Skyscan 1272 (Bruker Biospin MRI GmbH, Germany) or Phoenix Nanotom m (GE Sensing & Inspection Technologies, USA), as described before (21). The imaged volumes obtained with X-ray microCT are correlated to the 3D fluorescence microscopy datasets in Amira (Thermo Fischer, USA) enabling to determine the position of the cancer cell within the resinblock with ~5 μm accuracy in x, y and z. The region of interest was approached with targeted microtomy and subsequently the sample block was prepared for focused ion beam-scanning electron microscopy (FIB-SEM, Auriga 60, ZEISS) (21) or serial sectioning for 3D transmission electron microscopy. Here, 60-70 nm thick serial sections were mounted on formvar-film coated single slot grids, stained with lead citrate and imaged with a Biotwin transmission electron microscope at 120 kV (Thermo Fisher Scientific).

EM image processing and data analyses

Electron microscopy was performed of 53 cancer cells from 36 unique positions from 9 mice on day 3, 4, 5 and 7 post intracardiac injection. For serial section TEM, generally every 10^th section was imaged (i.e. with an interval of approximately 600 nm). When a region of particular interest was recorded, every single section was acquired. Electron micrographs were viewed and semi-automatically or manually aligned in TrakEM2 (22), a plugin for Fiji (23). Segmentation was performed in TrakEM2 or 3dmod, part of IMOD (24) and resulting 3D reconstructions of the imaged volume were visualized in Amira (Thermo Fischer), Imaris (Oxford Instruments, United Kingdom) or 3dmod. Remodeling phenotypes (Neo-Lumen, Projections, Blocked lumen) were counted as such if occurring once or more in the imaged volume containing the cancer cell(s). The following criteria were attributed to the different phenotypes: Neo-lumen: an electron-lucent, partially or completely unperfused region within a single or created by multiple ECs lacking an open connection to the main lumen. Projections: <1 μm thick EC membrane protrusions towards the lumen or opposing ECs. Projections may enclose clots, debris, cytoplasts, platelets or erythrocytes. Blocked lumen: pathologically swollen ECs are blocking >70% of the lumen diameter or the vessel at ≥1 side of the cancer cell position. For quantification of pathological basement membrane, in total 35 tissue blocks from intravascular, intravascular + extravascular, extravascular and extravasating cells were analysed (see legend Fig. 2 for detailed description). EM micrographs were randomly selected from the imaged volume and scored for the presence of pathological basement membrane, based on one or more of the following criteria: 3 or more layers of basement membrane; and/or >150% of thickness; and/or large variations in thickness within the same image.

In vivo microscopy data analyses and quantification of the brain metastatic cascade

For initial image processing and quantifications Fiji (23) (RRID:SCR_002285) or ZEN Black (Zeiss, Germany, RRID:SCR_018163) was used. Adjustment to channel brightness or contrast was applied to the whole image. Segmentation and three-dimensional reconstruction and volume measurements were performed in Amira (Thermo Fischer, USA, RRID:SCR_007353) or Imaris (Oxford Instruments, United Kingdom, RRID:SCR_007370). For quantification of remodeling phenotypes, their occurrence was scored in direct proximity of the arrested or extravasated cancer cell (within 50 μm in the respective vessel). Formation of 1-2 μm thick endothelium protrusions projecting into the capillary lumen were scored as “Projections”, local increase in EC thickness (>300%, semi-quantitative measurement) were scored as “Blocked lumen” and “Swollen Lumen” indicated a >150% diameter increase with respect to the adjacent position. Post-extravasation vascular remodeling phenotype “capillary loops” was defined as a local increase in vessel curvature and density. Extravascular clusters of 3-50 cancer cells were scored as micrometastases, and metastases with >50 cancer cells were defined as macrometastases. Each cell included in the quantification was monitored until it disappeared from the field of view (by cell death or washout from the vessel) or until the end of the experiment. Blood flow velocities were measured by a line scan with a minimum length of 10 μm in vivo or from previously acquired timeseries.

Immunohistochemistry

Mice with brain metastases were anesthetized with a mixture of ketamin/xylazin and intracadrially perfused with PBS followed by 4.5% formaldehyde in phosphate-buffer (Roti- Histofix 4.5%, 2213, Carl Roth, Germany). Before cryosectioning, the brain was infiltrated overnight with 30% aqueous sucrose solution and snap frozen. Sections were labeled with the primary antibodies (anti-Aquaporin-4 (1:200, ab9512, Abcam, United Kingdom, RRID:AB_307299), anti-CD31/PECAM-1 (1:100, AF3628, R&D Systems, USA, RRID:AB_2161028), anti-CD41 (1:100, ab33661, Abcam, United Kingdom, RRID:AB_ 726487), anti-Collagen IV (1:100, AB756P, Sigma-Aldrich, USA, RRID:AB_2276457), anti-Delta-like ligand 4 (Dll4, 1:100, AF1389, R&D Systems, USA, RRID:AB_354770), anti-Endocan/ESM-1 (1:100, AF1999, R&D Systems, USA, RRID:AB_2101810), anti-GFAP (1:200, ab53554, Abcam, United Kingdom, RRID:AB_880202), anti-HIF1α (1:100, NB100-479, Novus Biologicals, USA, RRID:AB_10000633), anti-Ki67 (1:100, ab15580, Abcam, United Kingdom, RRID:AB_443209), anti-Laminin (1:100, ab30320, Abcam, United Kingdom, RRID:AB_775970), anti-Laminin 2 alpha (1:200, ab11576, Abcam, United Kingdom, RRID:AB_298180), anti-MMP9 (1:100, ab38898, Abcam, United Kingdom, RRID:AB_776512), and secondary antibodies donkey-anti-goat IgG Alexa 633 (1:300, A21082, Thermo Fisher Scientific, USA, RRID:AB_2535739), donkey-anti-rabbit IgG Alexa 488 (1:300, A21206, Thermo Fisher Scientific, USA, RRID:AB_2535792), donkey-anti-Rat IgG (1:300, A21208, Thermo Fisher Scientific, USA, RRID:AB_141709). Sections were embedded in Vectashield antifade mounting medium with DAPI (H-1200, Vector Laboratories, USA) or stained with DAPI (D8417, Sigma-Aldrich, USA) and subsequently embedded in Vectashield Hard set mounting medium (VEC.H.1400, Vector Laboratories, USA). Immunofluorescently stained sections were imaged on Zeiss microscopes (LSM700, LSM710 or LSM780 Spinning Disk, Carl Zeiss GmbH, Germany).

SB-3CT treatment in vivo

Tamoxifen-treated, window-bearing VE-Cad Cre^ERT2xRosa26-YFP^fl/fl mice were treated with SB-3CT (25 mg/kg bodyweight, S7430, Selleckchem, USA) in 10% dimethylsulfoxide / polyethylene glycerol / PBS (DDP) injected intraperitoneal daily, starting from one day before intracardiac injection of cancer cells for 11 days in total. Intravital microscopy was performed of SB-3CT treated and control mice to monitor endothelial remodeling and metastastic progression on day 2-8, day 10 and day 14 p.i.. On day 14, the mice were perfusion fixed and cryosections of the brains were obtained as described above. The cryosections were embedded in Vectashield antifade mounting medium with DAPI (H-1200, VECTOR Laboratories, USA) and imaged with an Axio Scan.Z1 slide scanner (Zeiss, Germany) in fluorescence configuration using a 20x/0.8 Plan-Apochromat objective (Zeiss, Germany). Metastatic burden was quantified by calculating the ratio of metastasis-area to total imaged brain area for each section in both control and SB-3CT-treated mice. For histological analyses, to study the effect of SB-3CT, 8-12-week-old C57BL/6 mice (Janvier Labs, France) were treated with SB-3CT or carrier solution and intracardiacally injected with E0771 as described above, and sacrificed by perfusion fixation 3 days p.i..

Treatment with acetylsalicylic acid, clopidogrel and tinzaparin in vivo

Therapy with ASA/clopidogrel (acetylsalicylic acid (ASA) 5 mg/kg bw and clopidogrel 25 mg/kg bw, in up to 0.2 mL MCT, administered daily by gavage), and tinzaparin (tinzaparin, 0.6 IU/g bw, daily s.c. injection, volume depending on weight 0.1-0.2 mL in sterile 0.9% NaCl solution) starts 2 days before cardiac injection, as described before (25). In the control group, daily s.c. Injection and gavage with carrier solution is performed. To observe EC remodeling, the VE-Cad Cre^ERT2xRosa26-YFP^fl/fl mouse model is used. Measurements of volumes of cancer cells and clots from microscopy datasets are performed in Imaris.

MMP9 KO mice and MMP9 KD cell lines

E0771-tdTomato cells were intracardiacally injected in B6.FVB(Cg)-Mmp9<tm1Yvu> / J (26) (“MMP9 KO” mice, #007084, JAX, RRID:IMSR_JAX: 007084, 8-10 weeks) and C57BL/6 (control, 8-10 weeks) mice. E0771 cells were stably transduced with plKO.1-puro-CMV-TurboGFP_shnon-target (control; cytoplasmatic GFP, SHC016, Sigma-Aldrich, USA) or pLKO.1_hPGK-puro-CMV-tGFP_shmMMP9 (Mmp9 knockdown, to create “E0771-MMP9KD” cells, NM_013599, TRCN0000031233, targeting sequence CAGTACCAAGACAAAGCCTAT, Sigma-Aldrich, USA). GFP and tdTomato positive cells were selected through several rounds of FACS and transduction was validated using qPCR. E0771-ControlKD and E0771-MMP9KD cells were intracardiacally injected in C57BL/6 (control) mice and E0771-MMP9KD in B6.FVB(Cg)-Mmp9<tm1Yvu> / J mice. The mice (3-4 mice per condition and time point) were sacrificed by perfusion fixation 2 or 4 days p.i.. 100 μm vibratome sections or 10-20 μm cryosections were made from the fixed brains, fluorescently stained to visualize vasculature and/or MMP9. For each condition, the number of intravascular/extravascular cancer cells was scored, as well as EC projection formation and/or capillary loop formation as indicated in the figure legends.

Single cell RNA-sequencing analysis of human brain metastases

Single-cell gene expression data (27) (Gene Expression Omnibus, accession number GSE186344) was implemented in Seurat V4 (28) and pre-processed with the standard workflow of this R Package. The published annotations obtained from GSE186344 were used to perform cell type assessment. The following analyses were all performed using Seurat V4 (28).

Previously, we determined genes that were upregulated in slow-cycling cells of brain metastases that were associated with brain metastasis-initiating cells (29). Here, we determined the gene expression score from the upregulated genes in slow-cycling cells in all single-cell RNA-sequencing clusters of human brain metastases using the AddModuleScore function in Seurat (28).

Statistical analysis

Statistical analysis was performed using GraphPad Prism Software (Version 8.4.2, GraphPad Sofrware, USA). Statistical significance was stated for P values < 0.05. The used tests are noted in the figure legends. Statistical significance between groups was assessed by two-sided Student’s t-test for normally distributed data, and Mann–Whitney tests or Wilcoxon matched-pairs signed rank tests. Normal distribution of the datasets was determined by a Shapiro-Wilk test. Fischer’s exact tests were used to test the difference between EC remodeling phenotypes in 3DEM samples and to compare differences in reaching a next step in the metastatic cascade.

Results

Capturing cancer cell extravasation in the brain

To uncover the key events that are crucial for extravasation of cancer cells in the brain, we refined a multimodal correlative microscopy approach (21). Combining fluorescence microscopy with electron microscopy (correlative light and electron microscopy, CLEM) uniquely enables capturing these extremely rare and transient events in a large tissue volume and subsequently studying them at high resolution (21). In vivo and ex vivo fluorescence microscopy was used to identify cancer cells in the mouse brain, followed by x-ray microCT-guided retrieval of the imaged volume in the processed sample, and finally 3DEM (Fig. 1a). Remarkably, a total of 98% of cancer cells could be relocated and imaged with 3D-CLEM (n=53 Jimt-1 cancer cells). Cancer cells were found intravascular and extravascular (Fig. 1b, Supplementary Fig. 1a, b). By day 7 post injection (p.i.), the majority of cancer cells were found outside of blood vessels, whereas at 3 days p.i. most cells are still intravascular (Fig. 1c, Supplementary Fig. 1c). Arrested cancer cells were typically found elongated (3,25) in brain capillaries of 4.4-10 μm (mean: 6.5 μm) in diameter, notably at vascular branch points (Fig. 1b, Supplementary Figure 1b, Supplementary Movie 1), in line with previous findings (3,7). Vascular arrest caused a significant local dilation of the vessel lumen at the position of the arrested cancer cell (Fig. 1d, e). In 60% of metastatic sites, so-called “cytoplasts” were observed, non-apoptotic cellular fragments that lack a nucleus, previously observed in response to shear forces in lung metastasis (30)(Supplementary Fig. 1b, d). The high throughput of the multimodal imaging approach allowed capturing rare extravasation events (Fig. 1f, Supplementary Movie 2 and 3). Here, the cancer cell nucleus was detected partly inside and partly outside of a brain capillary, and was largely deformed into an hourglass shape with a diameter of <1 μm in its most constricted position. These findings demonstrate that multimodal 3D-CLEM enables capturing the earliest steps of the brain metastatic process – providing the unique ability to gain new insights into its cellular mechanism.

Fig. 1. Multimodal correlative microscopy of cancer cells at the blood-brain barrier.

a, Cartoon representation of the intravital multimodal CLEM workflow, combining fluorescence microscopy (3D FM/IVM), near-infrared branding (NIRB), microCT x-ray imaging and 3DEM. b, Examples of intravascular (3 days post injection (p.i.)) and extravascular Jimt-1 cancer cells (5 days p.i.) captured with multimodal CLEM. IVM: 3D rendering from intravital microscopy, EM: representative pseudo-colored EM section, 3DEM: 3D rendering from 3DEM imaging and segmentation. c, Total numbers of intravascular and extravascular Jimt-1 cells imaged with multimodal CLEM. d, Diameter of vessels with arrested Jimt-1 cells (n=23 from 8 mice, day 3, 4, 5 and 7 p.i.), in direct proximity of the arrested cancer cell (“Adjacent”) and at the position of the cancer cell (“Arrest Site”). P value from a Wilcoxon matched-pairs signed rank test, whiskers show Min to Max values. e, IVM and EM images of the same intravascular Jimt-1 cell (5 days p.i.), the vessel lumen is expanded at the arrest site (blue bar vs. pink bar). f, 3D rendering of part of an extravasating Jimt-1 cell (4 days p.i.) imaged with 3DEM. Different z-positions of the acquired 3DEM stack are shown revealing the complex remodeling of the ECs and the basement membrane. The Jimt-1 cell nucleus is part inside and part outside of the blood vessel (arrowheads).

Extensive remodeling of ECs at the site of cancer cell arrest and extravasation

To ultrastructurally characterize the process of extravasation, we investigated local changes to the microenvironment during the earliest steps of brain metastasis using multimodal 3D-CLEM. Profoundly altered microvascular architecture with remodeling of ECs and disorganization of basement membranes, indicative of vessel wall activation (31), was frequently observed at the site of vessel-arrested and extravasated cancer cells (Fig 2a, b), which is in sharp contrast to the orderly architecture of normal brain microvessels (Fig. 2c). Quantification of the degree of basement membrane disorganization, including thickening and multilayered composition, at arrested, extravasating and extravascular cancer cells revealed an increase towards the point of extravasation (Fig. 2d). Indeed, in those rare occasions (n=3) where an extravasating cancer cell was captured, pathological remodeling of the basement membrane was always observed at the site of trans-endothelial migration (Fig. 2b, d, Supplementary Movie 3). Cancer cells were observed to disrupt the basement membrane during the extravasation process, but other than that, basement membrane signal was always observed on the vessel-adjacent side of the extravasated cancer cell (Supplementary Fig. 2a, b). The extravasating cancer cells did not surpass the astrocyte and microglia layers, two cell types that are also part of the neuro-vascular unit (Supplementary Fig. 2c, d).

Fig. 2. Ultrastructural remodelling of the blood-brain barrier at the site of arrested and extravasating cancer cells.

a, EM of pathological and normal basement membrane at arrested and extravascular Jimt-1 cancer cells (left-to-right: 4, 7, 3 and 5 days p.i.). b, EM reveals structural remodeling (arrowheads) of the basement membrane close to an extravasating Jimt-1 cell (4 days p.i.). c, Ultrastructure of a non-pathological brain capillary. d, Percentage of randomly selected EM sections showing pathological basement membrane at normal capillaries and at sites with intravascular, both intravascular and extravascular, extravascular or extravasating Jimt-1 cells (Intravasc: 12 tissue blocks from 6 mice; Intra- and extravasc: 8 tissue blocks from 5 mice; Extravasc.: 12 blocks from 6 mice, Extravasating: 3 blocks from 2 mice, 3-7 days p.i.). Whiskers show Min. to Max. values. P value only indicated for “Intravasc.” vs. “Extravasating”, and determined by two-tailed Mann-Whitney tests. e-g: Morphological classification and quantification of pathological remodelling of ECs at the site of intra- and/or extravascular Jimt-1 cells vs. normal capillaries (Intravasc: n=15 positions from 7 mice, Intra-/extravasc: 9 positions from 5 mice, Extravasc.: 13 positions from 7 mice). Exemplary EM images depict Jimt-1 cancer cells and local remodelling phenotypes at 7 (panel e) and 3 (f, g) days p.i.. Data are mean ± s.e.m. P values from Fischer’s exact tests.

Reproducibly, three subtypes of EC remodeling were discovered using 3DEM, specifically at the sites of cancer cell arrest in brain capillaries, but never in unaffected microcapillaries of the normal mouse brain: 1) generation of a neo-lumen that runs parallel to the existing one (Fig. 2e, Supplementary Fig. 2e); 2) inward EC projections, often creating EC pockets (Fig. 2f; potentially an early stage of neo-lumen formation); and 3) partial- or complete obstruction of the vessel lumen by swollen ECs (Fig. 2g). The latter two EC remodeling phenotypes were frequently observed at the sites of intravascular arrested cancer cells, and less so at positions where only extravasated cells existed (Fig. 2f, g). In summary, 3D-CLEM uncovered striking changes to the ultrastructural architecture of the brain microvasculature at the site of metastatic cancer cell arrest and extravasation, supporting a dynamic interaction of ECs with cancer cells before and during extravasation.

Formation of EC projections is a key event for extravasation and successful BM

To understand how these EC remodeling types relate to the success of cancer cell extravasation in the brain, we next aimed to interrogate the dynamics of EC remodeling, cancer cell position and growth using longitudinal intravital microscopy (IVM). Brain-metastatic red-fluorescent mouse melanoma (Ret-Br) or mammary carcinoma (E0771) cells were injected intracardiacally into cranial window-bearing mice expressing a yellow fluorophore in ECs (VE-Cad Cre^ERT2xRosa26-YFP^fl/fl)(19,20). Using this model, we could track the structural changes of the same brain microvessels at subcellular resolution and in real time over 14 days, and determine their dynamic alterations during initial cancer cell arrest, extravasation, and early perivascular growth (Fig. 3a, Supplementary Fig. 3a, Supplementary Movie 4). In both models used, micro- and macrometastases formation was always preceded by extravasation of the cancer cells, unlike as described before for lung metastases (32), underscoring the importance of this event in the development of BM. In brain regions where no cancer cell arrest occurred, microvessels showed no apparent morphological change over weeks (n>1000 capillaries over time analyzed). Importantly, the distinct EC remodeling phenotypes detected with 3D-CLEM (Fig. 2e-g) were also observed in vivo, specifically at those sites where cancer cells arrested in brain microvessels over time, and during the process of cancer cell extravasation (Fig. 3a-c, Supplementary Fig. 3a-c, Supplementary Movie 4). Thus, IVM corroborated the findings made with 3D-CLEM: most notably, the EC projections and the partial or complete blockage of the capillary lumen by pathological ECs could readily be identified (Fig. 3b, Supplementary Fig. 3b). Indeed, a subset of the neo-lumens was closed off to larger particles in the circulation, as demonstrated by the fact that following intravascular injection with 500kD- TRITC-dextran some neo-lumens remained devoid of the dye (Supplementary Fig. 3b). All types of EC remodeling were associated with cancer cell arrest and preceded both successful extravasation and clearance of the cancer cell from the capillary, indicating that the remodeling was not the sole driver of extravasation. However, long-term dynamic imaging of the same metastatic cells from two different brain metastasis models enabled to determine that a specific EC remodeling phenotype was key to successful extravasation and to final development of macrometastases. Indeed, thin luminal projections from ECs were more often found in regions adjacent to cancer cells that extravasated compared to those which did not, and extravasation and later brain macrometastases formation was restricted to those places were EC projections had occurred during the intravascular stage (Fig. 3c, d; Supplementary Fig. 3c-f).

Fig. 3. Dynamic EC remodelling during cancer cell arrest and extravasation.

a, IVM of a E0771 cancer cell from vascular arrest to macrometastasis. Structural changes to the cancer cell (arrows) and the ECs (arrowheads) are shown. b, EC remodeling phenotypes (arrowheads). c, Percentage of E0771 cells remaining intravascular (IV, purple) vs. total cells that extravasate (EV, green) accompanied by different EC remodeling events, before or during extravasation (n=127 cells from 4 mice, IVM). One or more EC remodeling events can be scored per cancer cell. Data are mean ± s.d., P value determined by a Mann-Whitney test. d, Quantification of the fate of distinct E0771 cells (n=94 cells from 3 mice), tracked over 14 days with IVM. When EC projections are associated with cancer cells, arrows and percentages are shown in blue. Black indicates absence of EC projections. EC projections are only observed during the intravascular stage and/or while extravasating. P values from a Fischer’s exact test. e-f, Dynamics of EC remodelling recovery following recanalization (“Vessel clearance”). Quantification of EC remodelling (all subtypes) at the site of an intravascular cancer cell (“Intravasc.”) vs. day 1-4 following recanalization, e.g. death, washout or extravasation of the cancer cell (n=65 E0771 from 3 mice, IVM). Arrowheads indicate EC projection formation (white) followed by recovery of these areas (grey). g, Timepoint (days p.i.) of first extravascular sighting for cancer cells that survive until day 14 as either micro- or macrometastasis (“Successful Mets”) or those that don’t (“Unsuccessful Mets”) (n=38 E0771, 3 mice, IVM). Whiskers show Min. to Max. values. P value from a two-tailed t-test with Welch’s correction. h, Vessel diameter at the site of arrested cancer cells vs. directly adjecent to it (n=38 E0771, 3 mice, IVM). As in g, both succesfull and unsuccesgul cancer cells are analysed. Whiskers show Min. to Max. values. P values determined by a Wilconxon test.

ECs forming projections in the vicinity of cancer cells did not express the proliferation marker Ki67 (Supplementary Fig. 3g). Tip cell markers ESM1 and Dll4, indicative for frank angiogenesis, were also not expressed at EC projections (Supplementary Fig. 3h). However, tip cell-like structures could occasionally be observed in the proximity of capillaries harboring metastatic cells, but not at these vessels themselves (Supplementary Fig. 3i). After clearance of the cancer cells from the microvessel, i.e. by re-entering the circulation, by extravasation, or cell death, EC remodeling normalized (Supplementary Fig. 3j). Depending on the fate of the cancer cell, vessel clearance could take between 12 hrs to several days; EC remodeling was generally reversed within 4 days after the cancer cell left the lumen (Fig. 3e and f, Supplementary Movie 5).

Next, we set out to determine if specific parameters of the extravasation process were related to unsuccessful brain metastatic growth. We found that extravasated metastatic cells that do not survive in the perivascular niche remained significantly longer intravascular before extravasation, compared to their successful counterparts (Fig. 3g). In contrast, mere dilation of the normal microvascular diameter, found at the site of cancer cell arrest, occurred both in successful and unsuccessful metastatic cells. However, at the position of unsuccessful metastatic cells, the associated vessel dilation was more pronounced, suggesting that excessive swelling of the microcapillary before extravasation predicts restricted BM growth (Fig. 3h). These results collectively show that EC remodeling is a reversible process triggered upon the intravascular arrest of cancer cells, and that rapid initiation of extravasation with a distinct EC remodeling phenotype determines successful brain colonization.

Post-extravasation capillary remodeling supports brain metastatic growth

We next asked whether other forms of structural changes to the vasculature could be observed after extravasation, i.e. during early perivascular growth and BM development. Indeed, capillary loops (3,33) were reproducibly detected where extravasated cancer cells grew along brain microvessels, leading to highly tortuous capillaries during early brain metastasis (Fig. 4a, Supplementary Fig. 4a). Consistently, capillary loops were not associated with EC proliferation (Supplementary Fig. 4b), supporting that capillary loops are dynamically formed at the site of perivascular metastatic cells. Whereas EC remodeling by formation of EC projections was restricted to intravascular and extravasating cells, capillary remodeling by capillary loop formation coincided with, or followed extravasation (Fig. 4b, Supplementary Fig. 4c) and was most prominent when the metastases grew perivascular. Strikingly, when metastatic cancer cells moved along the microcapillaries, the loops followed the position of the cancer cells (Fig. 4c, Supplementary Movie 6). In all three different tumor models investigated (mouse- and human breast cancer, human melanoma), macrometastases were only formed in presence of capillary loops (Fig. 4d and e, Supplementary Fig. 4c). Those cancer cells that did extravasate but then failed to induce capillary loops rarely developed micrometastasis, and never macrometastases (Fig. 4e). Whereas capillary loops were directly (<24 hrs) formed following extravasation of mouse breast cancer cells, in the human models it could take up to 14 days following extravasation before the first occurrence of capillary loops. In metastatic nodules that managed to grow to macrometastases over the time of the experiment, capillary loop formation occurred significantly earlier compared to those that did not outgrow the micrometastatic stage (Fig. 4f). This speaks for a metastasis-promoting role of this post-extravasation vascular remodeling of brain microvessels, more likely due to an increase in perivascular niche surface for the developing BM (33) rather than an enhanced blood supply: dynamic imaging revealed that irregular slow velocities (1.524 μm/s ± 1.36 μm/s) could occur in capillary loops, which was significantly lower when compared to the flow rate in normal brain micro capillaries (628 μm/s ± 46(34)). Taken together, capillary loop formation is another form of dynamic vascular remodeling that occurs later during the brain metastatic cascade, and typical for those metastatic nodules that grow most aggressively.

Fig. 4. Capillary loop formation at successful metastatic growth in the brain.

a, IVM of capillary loop formation (arrowheads) during perivascular growth of E0771 metastasis. b, First observation of EC projections and capillary loops during three distinct stages of early brain colonization for different BM models. EC projections are formed in the intravascular stage or when both intravascular and extravascular E0771 (n=57 cells from 3 mice) or Ret-Br (n=94 cells from 3 mice) cancer cells were present, but never following extravasation. Capillary loops were never observed during the intravascular stage, and were only formed where cancer cells are found extravascular (E0771: 17 metastases, 3 mice; A2058: 36 metastases, 3 mice; Jimt-1: 18 metastases, 3 mice). c, IVM of a developing E0771 BM. Top panels show z-projections, bottom panels show zoomed 3D renderings. The arrowheads indicate capillary loop formation. d, 3D representations from IVM of a growing Jimt-1 BM showing capillary loops (top panels, arrows) and a slow-growing metastasis without capillary loops (bottom panels). e, IVM reveals the fate of metastases that eventually form capillary loops (left, yellow box) and of those that extravasate but fail to induce capillary loops (right, blue box). Capillary loops are formed at micro- or macrometastases. f, Number of days following extravasation before the first observation of capillary loops, quantified for cancer cells that result in micro- or macromets (Jimt-1: 18 micro- vs. 11 macromets, 5 mice, A2058: 12 micro- vs 23 macromets, 3 mice). Whiskers show Min. to Max. values, P values determined by Mann-Whitney tests.

MMP2/9-inhibition reduces formation of EC projections and cancer cell extravasation

Having uncovered a critical role for EC projections during extravasation and early brain colonization, we next set out to identify the molecular players initiating this process. Both VEGFA (35) and Ang2 (36) have been shown previously to regulate BBB permeability and could be potent targets to inhibit extravasation in the brain (36). However, neither therapy with a VEGFA-, nor an Ang2-, nor a bi-specific Ang2/VEGFA inhibiting antibody could inhibit cancer cell extravasation or formation of EC projections in vivo over time (Supplementary Fig. 5a, b). Importantly, we observed that the potent protease MMP9, a modulator of the extracellular matrix, accumulated at the site of arrested cancer cells in the mouse brain in various xenograft-, patient-derived xenograft-, and syngenic models of BM (Fig. 5a, b, Supplementary Fig. 5c). Thus, we next investigated whether EC projections are driven by MMP activity, and if they are functionally involved in cancer cell extravasation. Hereto, the effect of MMP2/9 inhibition on EC remodeling in vivo was tested. Mice were treated with the MMP2/9 inhibitor SB-3CT or with a vehicle control daily for 11 days, starting 1 day before intracardiac injection (Supplementary Fig. 5d). IVM was performed following cancer cell injection to closely monitor all steps of the brain metastatic cascade. Indeed, the EC projections formed at the position of arrested cancer cells were significantly reduced by the MMP inhibitor (Fig. 5c, d). In line, MMP inhibition led to a significant reduction of cancer cell extravasation in the brain (Fig. 5e). Overall, 9% of intravascular arrested cancer cells could be longitudinally followed to grow to macrometastases in the control group, but only when EC projections were detectable during vascular arrest and/or extravasation (Fig. 5e, left diagram). In contrast, under MMP2/9 inhibition, only 1% managed to form macrometastasis (Fig. 5e, right diagram), which led to a total reduction of metastatic burden at day 14 (Fig. 5f). Importantly, EC projections were rarely formed under MMP inhibition, and without any EC projections, no successful macrometastasis formation could be observed in both groups (Fig. 5e). Moreover, a prolonged overall survival of mice treated with SB-3CT was observed (Supplementary Fig. 5e). Interestingly, no difference in post-extravasation BM growth dynamics could be observed under control- or SB-3CT treatment, supporting that MMP2/9 inhibition mostly affects extravasation rather than subsequent BM growth (Fig. 5g). Although MMP2 and 9 have the potential to digest the basement membrane (37), we did not observe a significant difference in the thickness of basement membrane proteins laminin and collagen IV in mice treated with SB-3CT or vehicle control; both remained increased to a similar extend beneath extravasated cancer cells (Fig. 5 h, i). Moreover, in presence of SB-3CT, the basement membrane was breached during extravasation (Fig. 5j). Taken together, we demonstrate that MMP2/9 inhibition resulted in significant reduction of EC projections and associated cancer cell extravasation, which prevented the development of large BM.

Fig. 5. Inhibiting MMPs suppresses endothelial activation, extravasation, vascular remodeling, and macrometastasis growth.

a, Immuno-fluorescence images of intravascular arrested E0771 cancer cells (gray arrows) which show MMP9 labeling (right, yellow arrowheads) or not (left). The right panels depict an extravasating E0771. b, MMP9 expression at arrested E0771 cells versus control vessels (n=25 from 2 mice, 3-4 days p.i.). Line indicates the median. P value from a Wilcoxon matched-pairs signed rank test. c, Metastatic growth of E0771, monitored on days 1 through 4 p.i. by IVM in control (top) or SB-3CT treated mice (bottom). d, EC projection formation scored from IVM datasets, during arrest or extravasation of E0771 cancer cells in control and SB-3CT treated mice (Control: 65.4% from n=3 mice, 23.4% from n=3 SB-3CT treated mice). Data are mean ± s.d. P value from an unpaired two-tailed t test. e, Metastatic fate of E0771 cancer cells, accompanied by EC projections (blue arrows) or not (black arrows) in control and SB-3CT treated mice (3 vs. 3 mice), from IVM. P values determined by a Fischer’s exact test. f, Surface ratio of metastases to total tissue, quantified for brain sections of control and SB-3CT treated mice (3 vs. 3), 14 days p.i. P value from a Mann Whitney test. g, Growth dynamics of E0771 BM in control mice or mice treated SB-3CT (from IVM, 3 vs. 3 mice, individual growth curves are shown (thin lines) as well as median BM volume (thick lines) and interquartile range). h, i Thickness of basement membrane components collagen IV (g) and laminin (h) at site of arrested E0771 cancer cells vs. control vessels in control vs. SB-3CT treated mice (11-104 positions, 3 vs. 3 mice). Data are mean ± s.d. P values determined from a Mann Whitney test. j, Immunofluorescence of an extravasating E0771 (3 days p.i.) from an SB-3CT treated mouse. Yellow arrowhead indicates lack of laminin at the extravasation site.

Extravasation of inert particles differs from that of cancer cells

Next, we set out to unravel the underlying mechanism of MMP9 expression at the site of the extravasating cancer cell. IVM revealed that the flow of blood cells was blocked in 85.2% at the site of intravascular arrested cancer cells (n=27, Fig. 6a, Supplementary Fig. 6a). Molecular dyes that stain ECs (wheat germ agglutinin (WGA) – A633 conjugate, ~38 kD, and Evans Blue) did not manage to surpass cancer cells that block the lumen (Supplementary Fig. 6b). The blockage of the microcapillary, however, did not seem to result in relevant local hypoxia, as HIF1-alpha expression was lacking at these sites (Supplementary Fig. 6c). Von Willebrand Factor (vWF)- positive clots and platelets accumulated in the vessel lumen adjacent to the arrested cancer cell, which were engulfed by EC projections (Fig. 6b, Supplementary Fig. 6d-g). In a mouse model of ischemic stroke, increased MMP9 activity was detected in similarly occluded brain microcapillaries (38). This motivated us to investigate whether disrupted blood flow by local microvascular obstruction itself induces vessel remodeling and MMP9 expression, either directly or indirectly, independent of the presence of cancer cells. Hereto, the occlusion of cerebral microcapillaries was mimicked by intracardiac injection of 10 μm inorganic fluorescent microspheres. MMP9 expression was increased and platelets (CD41-positive) accumulated adjacent to 35.4% of intravascular microspheres (Fig. 6c, d). However, in this tumor-free model, only few extravasation events could be observed (7.7% of microspheres tracked in vivo over 5 days), compared to the situation seen with cancer cells (Supplementary Fig. 7a, b). EC projections were also formed less frequently at arrested microspheres when compared to cancer cells (Supplementary Fig. 7c, d); instead, obstruction of the micro capillary lumen by EC nuclei at these sites was very frequently observed (Supplementary Fig. 7c-e). In capillaries that did not harbor microspheres, no angiogenesis or EC remodeling was observed (Supplementary Fig. 7f). In agreement with previous work (39), we observed pruning of the vessel (Supplementary Fig. 7g) as a means of extravasation - something that was never observed for cancer cells. Therefore, we conclude that pure mechanical blockage of microvascular blood flow an inorganic object does not suffice to induce relevant EC remodeling and, importantly, that these inert objects extravasate less efficient and via a different mechanism compared to cancer cells.

Fig. 6. Microvascular embolization induces platelet accumulation and MMP9 expression, but is not a sufficient driver of extravasation.

a, IVM demonstrates arrested flow of blood cells (grey arrows) at intravascular E0771 cancer cells (white arrows, day 6 p.i.). In peripheral positions, regular flow patterns are observed (white arrowheads). b, EM reveals uptake of clot-material by EC projections (arrowheads) adjacent to an intravascular arrested cancer cell. c-d, CD41 and MMP9 fluorescence labeling at arrested microspheres (242 microspheres, 3 mice) vs. normal microcapillaries (80 microcapillaries, 3 mice). P values from unpaired two-tailed t-tests. e, Volumes of platelets found directly adjacent to E0771 cancer cells injected in mice treated with ASA+C+T or control, 3 days p.i. (73 positions, 3 vs. 3 mice). f, Fluorescence microscopy shows platelet accumulations (CD41, arrowheads) at an arrested E0771 cell, under ASA+C+T treatment or control (3 days p.i.). g, Percentages of intravascular (IV), intravascular/extravascular (IV/EV) and extravascular E0771 cells observed on 3 days p.i. in mice treated with ASA+C+T or control (n=81 cells vs. n=91 cells from 3 vs. 3 mice). h, EC projections formed during arrest or extravasation of E0771 cancer cells (86.4%; n=81 cells, 77.8%; n=72 cells, 3 vs. 3 mice). Data are mean ± s.d. P values from Mann-Whitney tests (e, g and h).

Clot formation is not necessary for cancer cell extravasation

Since we found MMP9 co-localizing with the clots formed next to the arrested microspheres (Fig. 6c, d), we next studied the contribution of platelets, cancer cells and ECs to MMP gelatinolytic activity in an in vitro perfusion assay (Supplementary Fig. 8a). Platelet coverage of brain ECs was increased significantly upon addition of cancer cells (1.9±0.8-fold) (Supplementary Fig. 8b). In situ zymography demonstrated that cancer cells in combination with platelets induced a 1.7-fold local increase in gelatinolytic activity on the EC surface (Supplementary Fig. 8c). Furthermore, in these conditions, pro-MMP9 and activated MMP9 was increased in the supernatant of the microfluidic slides (Supplementary Fig. 8d), but no relevant activity of pro-MMP2 and activated MMP2 was detected.

In order to determine if platelet accumulation and clot formation influences cancer cell extravasation in vivo, we combined platelet inhibition by acetylsalicylic acid (ASA) and clopidrogel (C) with thrombin inhibition by low molecular weight heparin (Tinzaperin, T), as described before(25). Although less clots were found at the site of intravascular arrested mouse breast cancer cells in ASACT-treated mice (Fig. 6e, f), no effect on extravasation was observed following clot inhibition vs. control treatment (Fig. 6g). Importantly, we also did not measure a difference in EC projections in this model (Fig. 6h), indicating that clot-formation is not mechanistically linked to the occurrence of EC projections.

Combined, these findings demonstrate that neither occlusion of the blood vessel nor clot formation is in itself sufficient to trigger EC projection formation and subsequent extravasation. This implies that extravasation is driven not only by the host, but by cancer cell factors, too.

MMP9 from the cancer cell, rather than from the microenvironment, is required for extravasation

To determine the cellular source of MMP9 during the extravasation process, we investigated whether ECs express MMP9 in vitro, which was indeed the case (Supplementary Fig 9a). We also demonstrated increased gelatinolytic activity of ECs following perfusion with cancer cells and platelets in vitro (Supplementary Fig. 8c). To determine if the MMP9 produced by ECs or other cells from the microenvironment influences cancer cell extravasation, we studied this process in vivo in MMP9 KO vs. control mice (Fig. 7a, b). Here, a slight retention of cancer cells inside the vasculature on day four p.i. in MMP9 KO mice was observed, but no difference in the amount of extravasated cells compared to control mice (Fig. 7b). Strikingly, MMP9 was still found in proximity of cancer cells in MMP9 KO mice (Fig. 7c), primarily in the cancer cell (Fig. 7d, e). We next set out to determine the potential role of the cancer cell in MMP9 secretion as part of the extravasation process. Brain-metastatic cancer cells indeed expressed MMP9 in vitro (Supplementary Fig 9b). We injected MMP9 KD cancer cells in control or MMP9 KO mice, and compared their extravasation dynamics to those from control KD cells in control mice, two or four days following intracardiac injection (Fig. 7f, Supplementary Fig 9c). Knockdown of MMP9 in cancer cells resulted in strongly reduced extravasation at the later time point, and this effect was only moderately enhanced when MMP9 KD cells were injected in MMP9 KO mice (Fig. 7g). Formation of EC projections was reduced in mice injected with MMP9 KD cells vs. control KD cells, which was not relevantly aggravated in MMP9 KO mice (Fig. 7h). In conclusion, the main source of MMP9 affecting EC projection formation and the extravasation process is the cancer cell itself.

Fig. 7. Cancer cell-produced MMP9 is the main driver of extravasation in vivo.

a, Experimental schedule describing the heart injection (HI) of E0771 cells in C75/Bl6 (control) and MMP9 KO mice, followed by perfusion on day 2 or 4 p.i.. b, Percentages of cancer cells (E0771) found intravascular (IV), intravascular/extravascular (IV/EV) and extravascular on day 2 and 4 p.i. (Control: n=102 and 94 cells on day 2 and 4. MMP9 KO: n=92 and 64 cells from each day 2 and 4. 3 mice per timepoint). P value from a Welch’s t-test. c, MMP9 expression at the site of cancer cells (BrM) or neighboring positions (Control)(Control: n=86 and n=94 cells. MMP9 KO: n=92 and n=87 cancer cells. 4 mice per timepoint). P values from a Wilcoxon matched-pairs signed rank test to compare C75/Bl6 and BrM, and with Mann-Whitney tests to compare MMP9 expression at BrM from C75/Bl6 and MMP9 KO mice. d, Distribution of MMP9 expression measured at the position of E0771 cells. e, MMP9 staining (yellow arrowheads) inside an arrested E0771 cell in a MMP9 KO mouse. f, As in a, but with MMP9 KD or control KD E0771 cells. g, Percentages of control KD and MMP9 KD E0771 cells found intravascular arrested (IV), intravascular/extravascular (IV/EV) and extravascular (Day 2: n= 169 control KD cells and n=144 MMP9 KD cells from 4 C75/Bl6 mice each and n=160 from 4 MMP9 KO mice. Day 4: n=117 control KD cells and 136 MMP9 KD cells from 4 C75/Bl6 mice each and n=82 MMP9 KD cells from 4 MMP9 KO mice). Data are mean ± s.d., P values from unpaired T tests, for with exception of EV MMP9 KD cells in C75/Bl6 vs. MMP9 KD mice; here a Mann-Whitney test was used. h, EC projection formation during intravascular arrest or extravasation of Control KD and MMP9 KD E0771 cells in C75/Bl6 and MMP9 KO mice, on day 2 p.i. (Control KD; n=67 cells, MMP9 KD cells in control mice; n=80 cells, MMP9 KD cells in MMP9 KO mice; n=110 cells. 4 mice per condition). Data are mean ± s.d., P value from an unpaired T test. i, j, Dot plots of expression of MMP9 and genes associated with slow-cycling, BMIC cells(29) (bottom panels) in different clusters of breast cancer (i) and melanoma brain metastatic cancer cells. Data was obtained in silico from single cell RNA sequencing data from 3 BM of both entities (27).

Next, we investigated if post-extravasation vascular remodeling, i.e. capillary loop formation, was also dependent on MMP9 expression of cancer cells. Using IVM, we found that EC projections at the same site preceded capillary loop formation in the majority (76.2%) of cases (Supplementary Fig. 9d). In only 9.5% of events, no remodeling of any kind was observed before capillary loop formation was first detected. Upon MMP2/9 inhibition, less capillary loops were formed at the position of extravasated cancer cells, compared to controls (Supplementary Fig. 9e). Since we have shown above that capillary loop formation is strongly associated with successful BM, we studied if the difference in capillary loop formation in mice treated with SB-3CT was due to a general reduction in BM growth (Fig. 5e, f), or if it was a direct effect of MMP2/9 inhibition. We identified by IVM at which point during BM growth capillary loops were first observed in control or SB-3CT treated mice. Under MMP9 inhibition and control treatment, capillary loops were formed at BMs of similar volumes, at similar time points during the experiment and within a similar timeframe of BM development (Supplementary Fig. 9f, g). Moreover, capillary loops were formed in BM of similar size from both control and MMP9 KD cancer cells (Supplementary Fig. 9h). This indicates that under MMP9 deficiency, capillary loop formation occurred at a similar rate during BM growth compared to the control situation. Together this makes it unlikely that MMP9 is directly involved, and more likely that biomechanical forces exerted from cancer cells, or other factors, are involved in the generation of capillary loops.

A distinct MMP9 expression pattern in brain metastases of patients

Having demonstrated in a preclinical setting that cancer cell MMP9 is relevant for EC remodeling and extravasation, the question arises if this finding can be translated to the human disease. We interrogated a recently published single cell RNA sequencing dataset from patient BM from different entities, and their microenvironment (27). As part of an in silico analysis of these datasets, unbiased clustering of brain-metastatic cancer cells was performed. MMP9 was expressed in specific clusters of cancer cells (Fig. 7i, top panel). Since these samples are derived from established, symptomatic BM, this finding does not inform if MMP played a role during early brain colonization. In a pre-clinical model of breast cancer brain metastases, we previously identified brain metastases initiating cells (BMICs), which are slow-cycling at the moment of intravascular arrest and are characterized by a specific gene signature, including stemness markers (29). Strikingly, the cluster of cancer cells with highest MMP9 expression exhibited also the highest values for the gene expression signature characteristic for BMIC/slow-cycling cells (Fig. 7i, bottom panel). Similarly, we found a particular high expression of MMP9 in BMIC signature clusters of cancer cells from melanoma and lung cancer brain metastases (Fig. 7j, Supplementary Fig. 9i). These findings suggest that in human brain metastases a subpopulation of BMIC-like cells exist, potentially a direct progeny of the original “seeds” that first colonized the brain, which show a specifically high MMP9 expression. This provides a first hint that expression of MMP9 could also play a role in metastatic brain colonization in humans.

Discussion

Extravasation of metastatic cells is a critical step for the formation of brain metastases. Here, we interrogated the mechanisms of cancer cell extravasation from brain microcapillaries in vivo and discovered an MMP9-dependent, distinct EC remodeling process at the site of cancer cell arrest, which supports cancer cell extravasation. We furthermore demonstrate that post-extravasation remodeling of the microvasculature, i.e. capillary loop formation, supports growth of both melanoma and breast cancer BM, but does not rely on MMP9. MMP9 from the cancer cell, but not from the host, is required for EC remodeling and subsequent extravasation. All in all, this study describes a new, dynamic interplay of host blood vessels with cancer cells as a novel and crucial factor of BM formation.

In principle, different models of cancer cell extravasation from blood microcapillaries can be envisioned: an active process, initiated and driven by the cancer cell (3, 14-16); “passive extravasation” where EC protrusions engulf cancer cells and push them outside, as convincingly demonstrated for larger tail vein vasculature in the zebrafish embryo (34); or a more anatomical complex, reciprocal process that combines active participation of both cancer cells and ECs. We find that EC projections are formed before and during extravasation, and that inhibition of this type of EC remodeling in turn affects cancer cell extravasation. The EC projections are reminiscent of structures formed during endothelium-driven expulsion of microclots from cerebral microvessels in an alternative mechanism to fibrinolysis (38–40), which was also demonstrated in other organs (40,41). During this process, described as “angiophagy”, ECs extend membrane projections that cover, phagocytose and drive out blood clots (38,40) resulting in recanalization of the capillary. The membrane projections extend towards the opposing endothelium (38,40,41), leading to the formation of endothelial pockets or micro-lumen, also found in our current work. The precise underlying mechanism driving angiophagy is undefined (40), although MMPs have been implicated (38). During this process, extensive cytoskeletal remodeling and pruning of the ECs takes place, enabling extravasation of the blood clot (40). Our findings suggest that the ECs respond to the arrested cancer cell in a similar fashion as to the microclots: EC projections are formed aiming to drive out the cause of the blockage. Most likely, this remodeling of the endothelial layer affects cell junctions, resulting in a temporary weakness of the otherwise unsurpassable BBB, allowing for the cancer cell to enter and colonize the brain (Supplementary Fig. 9j). We have shown here and previously (25) that microclots co-localize with arrested cancer cells, but the cancer cell does not require the presence of clots to take advantage of the physiological mechanism that microvessels employ to clear the clot. Instead, microclots rather support vascular arrest (25) but not extravasation.

Interestingly, blocking of the blood flow per se does not prevent projection formation or extravasation. Another type of EC remodeling, a thickening of the EC, occurs proximal to both arrested microspheres and cancer cells. Clustering of migrated EC nuclei (7,42) (Supplementary Fig. 7e) or endothelial-to-mesenchymal transition (EndMT)(43) at the cancer cell arrest sites can lead to loosening of EC adherens- and tight junctions. However, this type of remodeling was not sufficient to support microsphere extravasation. This implies that even when the ECs initiate steps promoting extravasation, the cancer cell needs to actively participate in the process. Indeed, even though EC projections precede successful extravasation of cancer cells, these structures are also often associated with cells that do not extravasate (Fig. 3c, d, Supplementary Fig. 3c-f).

MMP9 from the cancer cell correlates with EC projection formation and cancer cell extravasation success. Interestingly, inhibition of MMP9 does not appear to influence digestion of the basement membrane during the extravasation process. This could be explained by the fact that other factors can also degrade components of the basement membrane. Macrophage and neutropil elastase, as well as cathepsins were found to be able to digest laminin (44,45). Moreover, MT1-MMP, MT2-MMP and MT3-MMP were found to potently degrade BM in an ex vivo model of tumor cell invasion (46).

Both the microenvironment as the cancer cell can contribute to MMP9 expression at arrest sites. Multiple cell types in the neurovascular unit excrete MMP2 and 9 upon hypoxia or ischemia, including ECs (8), astrocytes (2), or pericytes (47). Indeed, we also find MMP9 expression in proximity of arrested microspheres in our cancer-free model. However, cancer-cell-derived MMP9 appears to be more effective in inducing EC projection formation and extravasation, either due to the spatial positioning or the exact timing of activity.

Here, extravasation was studied using different syngenic, xenograft and PDX models where cancer cells were intracardiacally injected into mice. However, this methodology comes with some limitations: intracardiac injection does not fully recapitulate the metastatic cascade since it does not include the presence of a primary tumor, nor cancer cell invasion and subsequent intravasation into a bloodvessel. An interesting future direction would be to study if the pre-metastatic setting (48) influences the extravasation mechanism. Since extravasation of cancer cells in the brain is such a rare event and its timing is so unpredictable, this study does not include evidence from patient biopsies or pre-clinical models of spontaneous metastases, such as genetically engineered mouse models (GEMMs). We show that cancer cell-derived MMP9 is required for EC projection formation and extravasation, however future work will need to demonstrate what biomechanical, environmental or cancer cell-intrinsic factors trigger MMP9 expression and if MMP9 impacts the cancer cell directly during this process. EC projections were reduced pharmaceutically by MMP2/9 inhibition or by genetic perturbation of MMP9 in cancer cells. Another approach would be to inhibit EC projections directly, e.g. by actin depolarization, and to study the effect on extravasation. Lastly, in order to translate these basic research findings to the clinic, more studies are required to identify therapies to inhibit EC projection formation, extravasation and resulting BM disease.

The work presented here demonstrates how cancer cells extravasate into the brain parenchyma: by exploiting a moment of weakness in the otherwise unsurpassable BBB, specifically during the self-clearing attempt of the very microcapillary it has occluded. We show that the remodeling of the brain vasculature, taking place during and directly following extravasation and dependent on MMP9 from the cancer cell, is compulsory for successful brain colonization and therefore a promising therapeutic target. Indeed, the prevention of brain metastasis in high-risk patients suffering from solid cancers is an attractive clinical prospective (1,49). Therefore, the identification of another early and crucial step of metastatic colonization can help to better select novel targets for anti-metastatic treatments.

Statement of Significance.

Tracking single extravasating cancer cells using multimodal correlative microscopy uncovers a brain seeding mechanism involving endothelial remodeling driven by cancer cell-derived MMP9, which might enable development of approaches to prevent brain metastasis.

Data availability

The single cell RNA-sequencing data analyzed in this study were obtained from Gene Expression Omnibus, accession number GSE186344. The data generated in this study are available upon request from the corresponding authors.