Incipient Melanoma Brain Metastases Instigate Astrogliosis and Neuroinflammation

Malignant melanoma is the deadliest of all skin cancers. The major cause of melanoma mortality is metastasis to distant organs, frequently to the brain. Autopsy reports show that 75% of melanoma patients who died of this disease developed brain metastases, and the incidence of brain metastasis is rising (1). Brain metastases are currently incurable, and are associated with a dismal median survival of less than one year (2). Therefore, it is essential to identify factors that play a role during the earliest stages of the metastatic process that may allow preventive therapeutics following removal of the primary tumor. Dissemination of cancer cells to distant organs is a multistage process, affected by various cells in the microenvironment (3). However, while the role of the microenvironment at the primary tumor site is well documented, the crosstalk between disseminated cancer cells and stromal cells at the metastatic site are poorly characterized. Recent studies have shown that changes in the metastatic microenvironment precede the formation of macrometastases (4). Nevertheless, despite the clear clinical implications, changes in the brain microenvironment that enable the growth of metastatic melanoma cells are poorly characterized.

One of the major obstacles for molecular characterization of early metastatic niches is the lack of tractable preclinical models of spontaneous brain metastasis. Existing models of melanoma brain metastasis mostly rely on injection of tumor cells via an intracardiac or intracarotid route, resulting in rapidly forming experimental macrometastases (2, 5). While these models were instrumental in contributing to our understanding of advanced stage metastatic disease, they do not allow comprehensive studies of the multi-stage process of metastasis. Additional models were developed by injection of patient-derived melanoma cell lines, giving rise to spontaneous brain metastases in immunodeficient mice (6, 7). Transgenic models of melanoma (8–12) are valuable for studying primary tumor initiation and progression. However, they encompass infrequent brain metastases with very long latency and thus preclude systematic analyses of the early changes in the brain microenvironment that enable metastatic growth (13).

To overcome this challenge, we established and characterized a transplantable mouse model system of spontaneous melanoma brain metastasis following orthotropic inoculation of melanoma-derived cell line in immunocompetent mice. We chose the Ret-melanoma model (12). Importantly, the RET oncogene is mutated in human melanoma (14, 15), particularly in desmoplastic melanoma, which has an increased risk for brain metastasis (16). While mutations in the RET oncogene are not very frequent in human melanoma, they result in activation of common oncogenic downstream signaling pathways, such as MEK kinases and p38 MAPK (12). In addition, Ret-melanoma cells express typical melanoma antigenic markers including TRP2, gp100 and TRP1 (17). The transplantable model we established recapitulates the pathologic multistep process of metastasis, including a relatively high penetrance of brain macrometastases.

We characterized the formation course of micro- and macrometastases and established molecular tools by which metastases can be quantitatively assessed. Moreover, we show that brain micrometastases can be detected intravitally by quantitative analysis of melanoma-derived transcripts in peripheral blood and in cerebrospinal fluid (CSF).

Astrocytes play a principal role in the repair and scarring process of the brain following injuries. Dysregulation of their function contributes to the pathogenesis of several diseases, including neurodegenerative disorders (18), brain cancer, and metastasis (19). Reactive astrogliosis is the primary response of astrocytes to brain insult, characterized by proliferation, migration to the injured site, and extensive upregulation of glial fibrillary acidic protein (GFAP) (20). We and others have previously shown that activated astrocytes surround and infiltrate experimental brain metastases (21–23). However, the role of astrocytes in facilitating the formation of spontaneous brain metastases is largely unknown. We therefore utilized our spontaneous model to characterize early changes in the brain microenvironment, and the role of astrocytes in promoting incipient growth of melanoma cells. Here we demonstrate early changes in the brain microenvironment that precede the formation of spontaneous brain metastases, including breakdown of the blood–brain barrier (BBB) and vascular hyperpermeability. Moreover, we show activation of astrogliosis and neuroinflammation in incipient brain metastases, and that astrocytes are activated by paracrine signaling from melanoma cells to express a gene signature associated with brain tissue damage. Finally, we demonstrate a functional role for astrocytes in facilitating the initial growth of melanoma cells in the brain. Thus, our approach for systematic detection of micrometastases in a clinically relevant mouse model provides a platform to study the early interactions between metastatic tumor cells and stromal cells in the brain microenvironment that regulate metastatic growth.

Mice were maintained at the SPF facilities of the Tel Aviv University and the Center for Preclinical Research at the German Cancer Research Center, Heidelberg. All experiments involving animals were approved by the TAU Institutional Animal Care and Use Committee (approval # M-13-078) or by the local regulatory authorities in Karlsruhe, Germany (license No. G116/13). Five- to 8-week-old male C57BL/6 (Harlan) or female C57BL/6 (Charles River Laboratories) mice were used. Use of human samples was approved by the Institutional Review Board Committee at the University Medical Center, Göttingen, Germany.

A total of 5 × 10⁵ low passage (< p15) Ret-melanoma sorted (RMS) cells were resuspended in PBS and mixed 1:1 with growth factor–reduced Matrigel (356231, BD Biosciences) to a final volume of 50 μL. Mice were anesthetized by isoflurane, and injected subdermally at the right dorsal side, rostral to the flank, with a 29G insulin syringe (BD Biosciences).

Tumors were measured 4 times weekly by calipers. Tumor volumes were calculated using the formula X²×Y×0.5 (X-smaller diameter, Y-larger diameter).

Mice were anesthetized with ketamine (100 mg/kg) xylazine (10 mg/kg). An incision was made in the skin medial to the tumor. Tumors were detached from inner skin with clean margins to prevent recurrence. Next, tumor-associated connective tissue and blood vessels were detached. The incision was sutured using vicryl threads (J304H, ETHICON). Mice were weighed weekly and monitored until relapse.

Mice were anaesthetized with ketamine/xylazine, heart-perfused with PBS and brains, lungs, and livers were harvested. Brains were macroscopically examined for abnormal lesions and cut mid-sagittaly. Right hemispheres were taken to histology, and left hemispheres were flash-frozen in liquid nitrogen.

Normal brains were harvested and cut mid-sagittaly. Serial dilutions of RMS cells (10²–10⁶) were added to M-tubes (Miltenyi Biotec) and homogenized with the hemispheres.

Three to 5 μL cerebrospinal fluid (CSF) samples were obtained as described previously (24) with the following modifications: Hirschmann Micropipettes capillaries (Z611239, Sigma) were pulled at 63.5°C with a pipette puller (CP-10, NARISHIGE). The sharp edge of the pulled capillary was cut open. Mice were anesthetized with ketamine/xylazine, placed in Kopf Stereotaxic Alignment System, and CSF was collected from the cisterna magna. Capillaries were gently removed and placed on a 25G needle attached to a 1-mL syringe, in which the plunger was prepulled. The attachment point was sealed (to ensure outflow only), the CSF was slowly released and samples were stored at −80°C. Only blood-free samples were analyzed.

For gene expression analysis, samples were thawed on ice, and direct reverse transcription was next performed without RNA purification step (K1671, MAXIMA-RT kit, Life Technologies). CSF from normal mice was used as negative control, and purified exosomes from RMS cells as positive controls (ExoQuick-TC kit, EXOTC10A-1, SBI). cDNA was diluted 1:2 and qPCR was performed (StepOne, Applied Biosystems). CSF samples from spontaneous metastases were further amplified with additional 40 cycles of PCR (C1000, Bio-Rad). Two microliters of PCR products were run on 2% agarose gels.

Data were analyzed with Student's t test. Correlation analyses were performed with Fisher exact test (2 × 2 contingency table). Data were considered significant when P < 0.05. All statistical tests were two-tailed.

One of the main challenges in studying brain metastasis is the lack of murine models that fully recapitulate invasion, migration, and distant organ colonization by tumor cells (11). We therefore set out to establish a model of spontaneous melanoma brain metastasis that will incorporate all steps of metastatic disease in immunocompetent mice. To that end, we utilized a transplantable melanoma cell line derived from a spontaneously occurring skin tumor in Ret transgenic mice (25). To facilitate the detection of metastasizing melanoma cells in vivo, we engineered the Ret-melanoma cells to express the fluorescent reporter gene mCherry and selected for highly fluorescent cells by FACS. Transduced Ret cells will be referred to hereafter as Ret-mCherry Sorted cells (RMS). Importantly, expression of mCherry did not affect proliferation rates as compared with the parental cell line (data not shown). A total of 5 × 10⁵ RMS cells were inoculated orthotopically by subdermal injection (previously demonstrated to be preferential for spontaneous metastasis; ref. 26) into syngeneic mice, and primary tumor growth was analyzed (Fig. 1A and B). Notably, the formation of aggressive local tumors suggested that the expression of mCherry did not induce an immunologic host response. To adequately represent the clinical settings, local tumors were surgically excised and mice were monitored for brain metastases. When analyzing injected mice that did not succumb to pulmonary metastases, 50% of the remaining injected mice developed brain macrometastases approximately 3–6 months after primary tumor removal (Supplementary Table S1). The overall incidence of brain macrometastases was 23% (n = 60). Brain macrometastases were detected intravitally by MRI imaging (Fig. 1C; Supplementary Fig. S1), or visualized by ex vivo fluorescent imaging of mCherry-positive foci (Fig. 1D), and by macroscopic examination (Fig. 1E). The calculated average volume of macrometastases by MRI was 2 mm³ (Supplementary Fig. S1). The presence of mCherry-expressing cells in brains of injected mice was confirmed by FACS analysis. mCherry-positive cells represented a distinct population, which consisted of 5%–9% of total cells in the brain (Fig. 1F). Histology and immunostaining of tissue sections from various regions of brain lesions further validated the presence of parenchymal macrometastases (Fig. 1G–K and Supplementary Fig. S1). Moreover, the pattern of brain metastases in mice (as reflected by the analysis of MRI and histology) resembled meningeal spread, characteristic of human metastatic disease, which represents a devastating complication of brain metastases (27).

Macrometastases are the final stage of a long complex process. To gain insight on the initial steps of metastasis, we analyzed the formation of brain micrometastases in this model. We first evaluated whether the expression of mCherry can be quantitatively assessed as a reporter for the presence of micrometastases. Expression analysis of RNA from local tumors confirmed the expression of mCherry in vivo (Supplementary Fig. S2A). We next tested whether quantitative detection of mCherry could be utilized to determine brain metastatic load. To that end, we established an ex vivo modeling system of micrometastases by mixing known numbers of RMS cells with normal brains followed by combined homogenization and RNA extraction (Fig. 2A). qPCR analysis of mCherry revealed a linear correlation between melanoma cell numbers and mCherry expression (r² = 0.98; Fig. 2B). The same linearity was obtained for the known melanoma markers Trp-1, Trp-2 (r² = 0.99; Fig. 2C and D), Mart-1 and Mitf-v2 (not shown), confirming that this ex vivo calibration system can be used to quantify the number of melanoma cells. Strikingly, quantification of melanoma cells in brains of mice with no detectable macrometastases revealed mCherry-positive signal equivalent to as few as 100 cells (Fig. 2E and F). Notably, mice were heart perfused to ascertain that the mCherry signal did not originate in circulating melanoma cells, but rather from parenchymal metastases. Thus, this molecular quantification system provides a reliable tool to identify incipient metastatic lesions and to study the earliest stages of metastatic disease.

Utilizing this molecular detection system, we next quantified the percentage of micrometastases. Analysis of brains from 40 injected mice revealed that one month after primary tumor removal, 40%–50% of the mice had micrometastases composed of <10⁴ cells (Fig. 2E and F). Similarly, we quantified micrometastases in other organs (Supplementary Table S2; Supplementary Fig. S2B and S2C). In addition, micrometastases were detected and quantified by FACS analysis. Micrometastases consisting of less than 10⁴ cells corresponded to a population of approximately 1%–1.5% mCherry-positive cells of total brain cells (Fig. 2G and H and Supplementary Fig. S3).

We next asked whether metastases formation in this model is a continuous process. To that end, we averaged the relative expression signal from mCherry-positive brains at different micrometastases endpoints: 7, 14, and 56 days after subdermal inoculation. Analysis indicated that the mCherry signal increased exponentially with time in three independent cohorts analyzed (Fig. 2I). These results indicate that micrometastases are proliferative; or alternatively, that additional micrometastatic lesions form with disease progression. To obtain spatial insight of micrometastases, we utilized mCherry detection to analyze the histopathology of brain micrometastases. Analysis of staining revealed that disseminated cells were located in the choroid plexus and in the brain parenchyma (Fig. 2J and K and Supplementary Fig. S2D–S2K). To test whether micrometastases detected by qPCR correlate with the histologic findings, we examined the corresponding hemispheres by qPCR, as above. We found a significant correlation between the presence of melanoma cells in brain sections and their detection by qPCR in the contralateral hemisphere, thus enabling efficient and accurate screening for micrometastases-bearing brains (Fig. 2L).

Seeking to extend the clinical relevance of micrometastases detection, we next tested the feasibility of intravital analysis and monitoring of micrometastases. To that end, we initially analyzed peripheral blood from injected mice. Calibration curves for mCherry and Trp-2 detection by qPCR were analyzed by mixing (“spiking”) known amounts of RMS cells with 15 μL samples of normal peripheral blood (Supplementary Fig. S4A and S4B). We found that circulating melanoma cells could be readily detected and quantified. Moreover, there was a significant correlation between the presence of melanoma cells in blood and brain metastases (Supplementary Fig. S4D). Interestingly, the signal was much higher in blood than in brain and corresponded to an average of 60 cells/μL (Supplementary Fig. S4C), reflecting the abundance of circulating melanoma cells, only part of which will eventually colonize the brain.

As circulating tumor cells probably indicate the presence of metastases in other organs in addition to the brain and are therefore not an unequivocal predictor of brain micrometastases, we further analyzed samples of CSF. We initially audited the presence of melanoma-derived transcripts in 3-μL CSF isolated from mice bearing experimental brain macrometastases (intracardiac or intracranial injections). Analysis by qPCR revealed expression of mCherry and the melanoma transcripts Trp-2 and Mart-1 (Fig. 3A and Supplementary Fig. S5). Notably, expression of the housekeeping genes Hprt (Fig. 3B) and Gus (not shown) was undetermined, and no mCherry-containing cells could be detected (Supplementary Fig. S5), suggesting that melanoma-derived transcripts may be located in vesicles secreted by metastatic cells into the CSF (e.g., exosomes or melanosomes). This hypothesis is further supported by the observation that mCherry and Trp-2 were detected in exosomes isolated from RMS cells in vitro, while the expression of Hprt was undetermined (Fig. 3C and D). Detection of vesicles in the range of 30–100 nm by transmission electron microscopy (TEM) confirmed the presence of exosomes (Fig. 3E and Supplementary Fig. S5). Encouraged by these results, we tested whether CSF analysis could be utilized as a diagnostic tool for detection of spontaneous brain metastases. Analysis of CSF from subdermally injected mice confirmed the presence of mCherry transcripts. Furthermore, the signal intensity corresponded to metastatic load, with stronger signals in mice with macrometastases (Fig. 3F). Importantly, there was a significant correlation between detection of melanoma transcripts in CSF and metastases in the corresponding brains (Fig. 3G). Imaging by TEM confirmed the presence of exosomes in CSF of mice bearing spontaneous brain macrometastases (Fig. 3H). Thus, brain micrometastases can be detected intravitally, providing a system to study the earliest stages of metastases formation.

Tumor vasculatures are known to be pathologically hyperpermeable in advanced tumor lesions (28, 29). However, very little is known about changes in the permeability of blood vessels in micrometastases. To test whether vessel hyperpermeability is a feature of brain micrometastases, we performed a modified Miles assay. Brains of injected mice analyzed were free of macrometastases, as confirmed by ex vivo fluorescent imaging and qPCR analysis of mCherry expression (Supplementary Fig. S6). Analysis of Evans blue extravasation into brain tissue revealed that spontaneous formation of brain micrometastases was associated with increased vascular hyperpermeability (Fig. 4A). To further quantify the leakiness of brain blood vessels and the integrity of the blood–brain barrier (BBB), mice were injected with FITC-Dextran (70 kDa). Quantification of staining indicated that mice with micrometastases had significantly more leaky vessels than control mice (Fig. 4B–E). Furthermore, staining of melanoma cells in mice injected with FITC-Dextran indicated that vessel permeability is correlated with the presence of disseminated melanoma cells around leaky blood vessels (Fig. 4C–H). These findings suggest that spontaneous brain micrometastases instigate breakdown of the blood–brain barrier.

We next wanted to test the applicability of the model to characterize the formation of a hospitable metastatic niche. We previously demonstrated that astrocytes are recruited to experimental melanoma macrometastases in brain (21). Therefore, we set out to characterize the role of astrocytes in spontaneously occurring brain metastases. To that end, we analyzed astrocyte recruitment and activation by immunostaining with GFAP, a marker of activated astrocytes, in brain sections from mice injected subdermally with RMS cells. Analysis of the results confirmed that activated astrocytes surrounded macrometastatic lesions (Fig. 5A). Moreover, we analyzed the expression of GFAP in brain sections from patients and validated that astrogliosis is evident also in human melanoma brain metastasis (Fig. 5B). Importantly, analysis of staining in brains bearing micrometastases indicated that astrocyte recruitment and activation is an early event (Fig. 5C). Therefore, we next sought to determine when astrogliosis is first instigated during brain metastases formation. Analysis of brains bearing micrometastases (as determined by mCherry expression) indicated that the expression of GFAP is upregulated as early as 2–4 weeks after primary tumor removal (Fig. 5D), suggesting that astrogliosis precedes metastases formation.

Neuroinflammation, manifested by upregulation of proinflammatory cytokines and chemokines is a feature of astrogliosis. To test whether this physiologic characteristic of astrogliosis is operative also in the metastatic niche, we analyzed the expression of multiple proinflammatory mediators in brains of mice bearing micrometastases, as above. Analysis of the expression results revealed upregulation in known proinflammatory cytokines and chemokines, including CCL17, CCL2, CXCL10, IL6, and IL-1β (Fig. 5E–I), suggesting that instigation of neuroinflammation, associated with astrogliosis, is an early event in the metastatic cascade.

We previously demonstrated that astrocytes facilitate the invasiveness of brain-tropic human melanoma cells in vitro (21). The functional role of astrocytes in facilitating the invasiveness of melanoma cells was further supported in an organotypic 3D intact brain slices cocultured with a tumor plug containing melanoma cells (Fig. 5J). Analysis of brain slices by immunostaining indicated that activated astrocytes and microglia are recruited to the tumor–brain interface (Fig. 5K–N). Seeking to obtain functional evidence that astrogliosis could be induced by paracrine signaling from disseminated melanoma cells, we tested whether conditioned medium (CM) from RMS cells could activate primary adult astrocytes that we had isolated from brains of normal mice. In addition to auditing the expression of Gfap, we analyzed the expression of an astrogliosis-related wound-healing gene signature, including Cxcl10, Lcn-2, Timp-1, SerpinE1, and Serpina3n (Fig. 5O–S). This gene signature was recently demonstrated to be operative during stroke and LPS-induced astrogliosis (30). Expression analysis of astrocytes incubated with melanoma cells CM indicated that secreted factors from melanoma cells could upregulate the astrogliosis-related wound-healing gene signature in normal astrocytes. These results suggest that the same molecular pathways that are induced during brain tissue damage and in neuroinflammation are operative also during brain metastases formation.

In addition to invasiveness, persistence of disseminated tumor cells at a hostile microenvironment is a critical limiting step of metastasis. We therefore asked whether astrocytes affect the initial growth of disseminated melanoma cells. To that end, we established a 3D coculture system to model incipient micrometastases. When dispersed melanoma cells (as few as 25 cells) were seeded in the presence of primary astrocytes they were more proliferative than control melanoma cells (Fig. 6A–C), suggesting that astrocytes are functionally necessary for the initial growth of melanoma cells.

To further assess the functional importance of astrocytes to the initial growth of melanoma cells in vivo, we coinjected a limiting number of melanoma cells admixed with adult primary astrocytes to mice brains. Strikingly, melanoma cells coinjected with astrocytes gave rise to significantly (9-fold) larger brain lesions than RMS-only controls (Fig. 6D–F), implying that astrocytes play a central role in supporting the growth of melanoma cells in the brain. Moreover, quantification of reactive astrocytes in vivo by immunostaining demonstrated that larger brain lesions are characterized by increased GFAP expression in the brain, reflecting recruitment and activation of host astrocytes (Fig. 6G–I). This astrogliosis was specifically instigated by tumor cells rather than by the tissue damage caused by intracranial injection, as intracranial injection of serum-free medium (SFM) to control mice did not induce astrogliosis at the indicated timepoint (Fig. 6I). To get molecular insight on the effect of the brain microenvironment on brain-metastasizing melanoma cells, we established a variant of brain-tropic melanoma cells (BT-RMS) by isolating tumor cells from brain macrometastases and reinjecting them to mice. Analysis of the ensuing brain metastasis after two rounds of selection revealed that the brain-tropic variants exhibited an increased incidence of brain metastasis and a shortened timeline from subdermal injection to the formation of spontaneous brain metastasis (Supplementary Fig. S7). Moreover, analysis of known signaling pathways indicated that the MAPK pathway was activated in brain-tropic melanoma cells as compared with the parental tumor cells (Fig. 6J and K). Interestingly, the PI3K pathway was also mildly activated (Supplementary Fig. S7). These results indicate that the MAPK pathway is activated by the brain microenvironment. Importantly, in vitro experiments, in which melanoma cells were incubated with secreted factors of activated astrocytes, revealed that astrocytes induce MAPK activation in RMS melanoma cells (Fig. 6L and M and Supplementary Fig. S7). These results imply that the in vivo activation of this pathway in brain-tropic melanoma cells is mediated, at least partially, by astrocytes. Thus, astrocytes support the growth of melanoma cells and activate signaling pathways associated with enhanced proliferation.

Finally, we analyzed the expression of the gliosis wound-healing gene signature in brains of mice coinjected with astrocytes as compared with brains of mice injected with melanoma cells only (Fig. 6N–R). Expression results revealed upregulation of genes known to be associated with astrogliosis. Notably, the upregulation of most genes was significant even when analyzed in total brain, containing multiple other cells in addition to activated astrocytes. These results functionally implicate the induction of astrogliosis and a wound-healing program in facilitating metastatic colonization and the initial growth of melanoma cells.

Taken together, utilizing a model of spontaneous melanoma brain metastasis, combined with molecular detection of micrometastases, we show that metastatic melanoma cells instigate blood vessel hyperpermeability, recruitment, and activation of astrocytes accompanied by induction of neuroinflammation, and that activated astrocytes functionally facilitate early metastatic growth (Fig. 7).

Reciprocal interactions of metastasizing tumor cells with stromal cells in secondary sites are a key factor throughout the metastatic cascade (3). Here we show that incipient melanoma brain metastases instigate astrogliosis and neuroinflammation, which precede the formation of macrometastases.

Utilizing a model of spontaneous brain metastasis in immunocompetent mice, we demonstrate the initial changes in the brain metastatic niche. Early changes in the brain microenvironment included breakdown of the blood–brain barrier, recruitment of astrocytes, and instigation of proinflammatory mediators and a wound-healing gene signature that were associated with enhanced growth of melanoma cells in brain.

Reproducible models of brain metastasis that recapitulate the multistep process of metastases have been a major challenge for several decades (11, 31, 32). We have established a clinically relevant murine model of spontaneous melanoma brain metastasis in immunocompetent mice that provides a unique platform to study the pathophysiology of brain metastasis. The brain metastatic spread and histopathologic features of brain metastases in this model are comparable with the human disease (27), further supporting its clinical relevance. Notably, the RMS tumor cells used in this model were not selected for brain tropism by repeated injections, and thus represent authentic brain metastasis of cutaneous cells.

We characterized the temporal and pathologic progression of micro- and macrometastases formation and established a system for quantitative detection of brain metastatic load. We have established molecular and intravital methodologies to diagnose micrometastases and quantitatively determine brain metastatic burden (qPCR of melanoma transcripts in brains, CSF analysis).

The presence of tumor cells in CSF was previously reported to be associated with advanced disease: cytologic and molecular analysis detected melanoma cells in CSF from stage IV melanoma patients with brain metastases (33). Here we demonstrate that CSF can be utilized to detect melanoma-derived transcripts in exosomes, rather than in tumor cells seeded in the CSF, thus implicating this analysis in early diagnosis and monitoring of occult micrometastases.

Activated astrocytes were shown to be recruited to brain metastases (2). Moreover, we previously showed that paracrine signaling from brain-metastasizing melanoma cells contributes to reprogramming and activation of astrocytes in vitro (21). Here we show in vivo that activation of astrocytes and upregulation of a gene signature associated with inflammation and immune cell recruitment (e.g., CCL2, CXCL10, CCL17), are instigated in the brain metastatic niche before the formation of brain macrometastases. Neuroinflammation is a prominent characteristic of astrogliosis, which includes release of proinflammatory cytokines and chemokines, increased blood–brain barrier permeability, and leukocyte infiltration (34, 35). Sustained inflammation is present in both acute CNS injury and in chronic neurodegenerative disorders, and astrocytes are major players in neuroinflammation (36), but its role in facilitating brain metastases formation is largely unknown.

In addition to upregulation of a proinflammatory gene signature in astrocytes, we show that a gene signature associated with stroke and LPS-induced brain tissue damage (30) is instigated in astrocytes by brain-metastasizing melanoma cells in vitro as well as in vivo. Functionally, upregulation of gliosis genes in vivo was associated with enhanced metastatic colonization when melanoma cells were coinjected with astrocytes. This gliosis-related wound-healing gene signature includes the chemokine CXCL10, a known chemoattractant for T cells expressed by astrocytes during neuroinflammation associated with chronic inflammatory diseases in the CNS (37). Thus, our findings suggest that astrogliosis and neuroinflammation, physiologically instigated as a response of astrocytes to overcome brain tissue damage, are hijacked by brain-metastasizing tumor cells to support their growth.

Survival and proliferation of disseminated cells in distant organs is a crucial junction in metastatic colonization. Astrocytes were previously shown to protect melanoma cells from chemotherapy via the endothelin–endothelin receptor signaling axis (38–40), and by direct contact through establishment of functional gap junctions via connexin 43 (41, 42). However, the role of astrocytes in facilitating the primary growth of brain-metastasizing melanoma cells is unknown. By seeding limiting numbers of melanoma cells and astrocytes in 3D cocultures, as well as in brains, we show, for the first time to the best of our knowledge, that astrocytes are functionally important in supporting the initial growth of metastatic melanoma cells. Molecular analysis of signaling pathways activated in brain-tropic melanoma cells (BT-RMS) suggested that this astrocyte-facilitated growth enhancement is mediated by activation of the MAPK signaling pathway in brain-metastasizing tumor cells.

Interestingly, the gliosis-related gene signature activated in astrocytes included SERPINs, recently shown to be involved in protection of brain-infiltrating cancer cells from FasL-mediated apoptosis (22). The study by Valiente and colleagues (22) demonstrated that secretion of SERPINs from tumor cells rescues them from astrocyte-mediated apoptosis. Our findings expand these observations and suggest that melanoma cells directly induce the expression of SERPINs in activated astrocytes, thus reprogramming astrocytes from growth inhibitory to growth promoting.

Importantly, we show that astrocyte activation is an early event, which occurs even before the formation of overt macrometastases. This implies that systemic or paracrine factors derived from circulating and/or disseminated tumor cells induce the formation of a hospitable metastatic niche in the brain. This observation is supported by our findings that melanoma cells can reprogram normal astrocytes to express GFAP and astrogliosis-related genes. Furthermore, our findings of melanoma-derived transcripts in exosomes also suggest that exosomes may be operative in the paracrine crosstalk between melanoma cells and astrocytes (4). Tumor cell-derived transcripts in exosomes were recently reported to support the formation of a premetastatic niche (43–45). Thus, astrocyte activation and reprogramming may be at least partially mediated by systemic signaling via exosomes.

The development of targeted therapeutics (e.g., BRAF inhibitors and MEK inhibitors; ref. 46) and immune checkpoint blockade therapies has dramatically changed the landscape of melanoma treatment in recent years, making the immune system a central therapeutic target. The way forward in harnessing these new therapeutic options to provide survival benefit to patients is by understanding immune responses in the tumor microenvironment and investigating novel drug combinations (47). Importantly, Ret-melanoma cells are immunogenic, and induce the formation of TRP-2–specific T cells, which accumulate in the bone marrow and in tumors (48). Such T cells, specific for intrinsic melanoma antigens are also a feature of human melanoma, and are found in patients' blood (49). However, similar to human disease, immune modulation in the Ret-melanoma model results in immunosuppression; tumors are infiltrated with MDSCs that block antitumor T-cell activity and Tregs (50). Thus, our model may provide a clinically relevant platform to investigate combination therapeutics and to assess their effect on brain metastasis in an immunocompetent host. Moreover, we show that brain micrometastases are associated with vascular hyperpermeability, and with the presence of disseminated melanoma cells in proximity to leaky blood vessels. These findings will facilitate testing of therapeutics at early stages, as vascular leakiness was previously demonstrated to enable drug delivery (28). Application of these findings in human disease will enable the combination of intravital early diagnostics of brain metastasis with preventive therapeutics, which may be the key to limiting metastasis.

In summary, we have established and characterized a preclinical model of brain metastasis and utilized it to characterize the earliest changes in the brain metastatic niche. Molecular understanding of metastasis at its onset will enable the design of novel therapeutic approaches that may prevent brain metastatic relapse.

No potential conflicts of interest were disclosed.

Conception and design: H. Schwartz, K. Müller-Decker, R. Satchi-Fainaro, T. Pukrop, N. Erez

Development of methodology: H. Schwartz, E. Blacher, K. Müller-Decker, R. Satchi-Fainaro, N. Erez

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Schwartz, E. Blacher, N. Livneh, L. Abramovitz, D. Ben-Shushan, R. Blazquez, A. Barrantes-Freer, M. Müller, K. Müller-Decker, R. Stein, R. Satchi-Fainaro, V. Umansky, T. Pukrop

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Schwartz, E. Blacher, M. Amer, N. Livneh, R. Blazquez, A. Barrantes-Freer, G. Tsarfaty, R. Satchi-Fainaro, T. Pukrop, N. Erez

Writing, review, and/or revision of the manuscript: H. Schwartz, E. Blacher, D. Ben-Shushan, A. Barrantes-Freer, R. Satchi-Fainaro, V. Umansky, T. Pukrop, N. Erez

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E. Blacher, A. Klein, S. Soffer

Study supervision: R. Stein, R. Satchi-Fainaro, N. Erez

The authors thank Dr. Pablo Blinder and Dr. Limor Broday for granting access to their facilities, Irena Shur and the Sackler School of Medicine Interdepartmental Core Facility (SICF) for help with imaging, FACS and PCR analyses, Prof. Ilan Hammel for his help with staining interpretations, Nasma Aqaqe for her help in FACS analyses, and Muhammad Yassin for his help with viral infections.

This research was supported by grants from The German–Israeli Cooperation in Cancer Research (MOST–DKFZ, CA-152 to N. Erez and K. Müller-Decker), from Worldwide Cancer Research (formerly known as the AICR; to N. Erez), and The Melanoma Research Alliance (the Saban Family Foundation–MRA Team Science Award to N. Erez and R. Satchi-Fainaro).

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