Brain metastasis, the most lethal form of melanoma and carcinoma, is the consequence of favorable interactions between the invading cancer cells and the brain cells. Peroxisome proliferator–activated receptor γ (PPARγ) has ambiguous functions in cancer development, and its relevance in advanced brain metastasis remains unclear. Here, we demonstrate that astrocytes, the unique brain glial cells, activate PPARγ in brain metastatic cancer cells. PPARγ activation enhances cell proliferation and metastatic outgrowth in the brain. Mechanistically, astrocytes have a high content of polyunsaturated fatty acids that act as “donors” of PPARγ activators to the invading cancer cells. In clinical samples, PPARγ signaling is significantly higher in brain metastatic lesions. Notably, systemic administration of PPARγ antagonists significantly reduces brain metastatic burden in vivo. Our study clarifies a prometastatic role for PPARγ signaling in cancer metastasis in the lipid-rich brain microenvironment and argues for the use of PPARγ blockade to treat brain metastasis.
Brain-tropic cancer cells take advantage of the lipid-rich brain microenvironment to facilitate their proliferation by activating PPARγ signaling. This protumor effect of PPARγ in advanced brain metastases is in contrast to its antitumor function in carcinogenesis and early metastatic steps, indicating that PPARγ has diverse functions at different stages of cancer development.
This article is highlighted in the In This Issue feature, p. 1631
Metastasis, the spread of cancer from primary tumor sites to distal organs, is the cause of 80% of deaths from cancer. The brain is one of the common metastasis locations for carcinoma (e.g., breast and lung carcinomas) and melanoma (1). Compared with carcinomas, melanoma has a much higher propensity to metastasize to the brain: More than one third of patients with metastatic melanoma develop a clinically apparent brain metastasis (2). Brain metastasis typically occurs at a late stage of disease progression after patients have already survived primary tumors and metastatic disease in other organs. Therapeutic strategies, including novel chemotherapies and targeted inhibitors, have historically focused on controlling the disease at primary sites and visceral organs (3). However, these therapies have shown limited efficacy in brain metastatic lesions. As a consequence, brain metastasis is a significant problem in the clinic due to its rising incidence and its resistance to existing therapies (1, 3, 4). There is an urgent need to expand our knowledge on the mechanistic underpinnings of brain metastasis as a disease, from which new targeted therapies can be developed.
As envisioned in the “seed and soil” hypothesis, cancer metastasis depends on the complex interplay between cancer cells and the microenvironments in distal organs (5–8). The “soil,” representing the microenvironment, not only regulates the outgrowth of metastatic cancer cells but also contributes to therapy resistance. The brain has a unique microenvironment. At the cellular level, it is composed of brain-specific cell types: functional neurons and supporting glial cells. Astrocytes are the most abundant glial cells in the brain and contribute to the pathogenesis of many brain disorders (9, 10). A common hallmark of brain pathologies is reactive astrogliosis, where astrocytes increase glial fibrillary acidic protein (GFAP) expression and cellular processes (9, 10). Examination of the very early steps of brain colonization in experimental mice revealed that activated astrocytes associate with invading cancer cells, and this interaction persists throughout the formation of large metastatic lesions (11, 12). At the molecular level, one distinct feature is that the brain is the fattiest organ in the body. Lipids constitute ∼50% of the brain (13). Unlike adipose tissue, the fatty-acid component of the brain is enriched in polyunsaturated fatty acids (13). Of note, glial cells, but not neurons, regulate fatty-acid synthesis and metabolism in order to maintain the normal function of the brain (14–16). The “seed,” the invaded cancer cells, needs to survive, proliferate, and eventually form metastatic lesions. It is still unclear how the brain-tropic cancer cells adapt to the unique brain microenvironment.
In this study, we observed that brain metastatic cancer cells took advantage of the high-fat microenvironment in the brain for metastatic outgrowth. As a major cellular source of fatty-acid synthesis, astrocytes supplied arachidonic acid (AA; 20:4) and mead acid (20:3) to activate the proliferator-activated receptor γ (PPARγ) pathway in the surrounding cancer cells. PPARγ has diverse functions in different cancer types and stages, and when combined with different therapeutic strategies (17–19). Here, we identified a proproliferative function of the PPARγ pathway in brain metastatic cancer cells. Furthermore, systemic blockage of the PPARγ pathway specifically decreased brain metastases, but not extracranial tumor growth, in the preclinical mouse models.
Prometastatic Effect of Astrocytes on Brain Metastatic Cancer Cells
To develop clinically relevant brain metastasis models, we used patient-derived xenografts (PDX) established from surgically removed melanoma brain lesions (20) and performed in vivo selection to isolate brain-tropic melanoma cells, which we termed BrM cells (refs. 21, 22; Fig. 1A; Supplementary Fig. S1A). To track the growth of cancer cells in the experimental mice, we stably labeled the melanoma cells with far-red luciferase and fluorescent protein (Supplementary Fig. S1A). Of note, WM5265.2 cells from the brain metastasis PDX model remained brain-tropic in the experimental mice with very limited ability to form metastases in other organs (e.g., lung; Supplementary Fig. S1B). In contrast, WM1366 or WM793 cells, both from the primary melanoma PDX models (Supplementary Fig. S1A), either formed no metastatic outgrowth or formed massive metastases throughout the whole body (Supplementary Fig. S1B). In parallel, we developed a syngeneic melanoma brain metastasis model using the mouse Yumm1.7 melanoma cell line, established from a BrafV600E/Pten−/−/Cdkn2a−/− transgenic mouse (Fig. 1A; Supplementary Fig. S1A; ref. 23).
In the brain lesions formed by WM4265.2-BrM1 cells and Yumm1.7-BrM cells, we detected GFAP+ astrocytes surrounding the cancer cells (Fig. 1B; Supplementary Fig. S1C). This is consistent with observations in the breast cancer brain metastasis model using MDA231-BrM cells, in which activated astrocytes associate with invading cancer cells, and this interaction persists throughout the formation of large metastatic lesions (12). We further confirmed the presence of activated astrocytes in the brain metastatic lesions from patients with melanoma (Fig. 1C). To detect the contribution of astrocytes to the growth of BrM cancer cells, we established cancer–astrocyte coculture assays under both two-dimensional (2-D) and three-dimensional (3-D) conditions (Fig. 1D–K). We tracked and quantified the growth of cancer cells by their fluorescence (Fig. 1E), luciferase labeling (Fig. 1F and J), and cell number counts (Supplementary Fig. S2A) in the coculture experiments. In nutrition-restricted culture medium (1% serum), astrocytes promoted the growth of both melanoma WM4265.2-BrM1 and Yumm1.7-BrM cells and breast cancer MDA231-BrM cells (Fig. 1D-K; Supplementary Fig. S2A). In complete media (10% serum), this astrocyte-promoted growth was abolished or much less significant, as the cancer cells grew relatively faster than in the nutrition-restricted condition (Supplementary Fig. S2B and C). Notably, astrocytes elicited more progrowth effects on brain metastatic cancer cells in physiologically relevant 3-D cocultures. We confirmed that the 3-D cocultured spheroids mimicked the cancer cell–astrocyte interactions in the brain metastatic lesions in vivo (Fig. 1I; Supplementary Fig. S2D). In addition, consistent with previously published work (24), astrocytes protected MDA231-BrM cells from apoptosis induced by tumor necrosis factor–related apoptosis-inducing ligand (TRAIL; Supplementary Fig. S2E). Overall, our data indicate that astrocytes have a progrowth and prosurvival effect on brain metastatic cancer cells.
Gene-Expression Profiling Predicts PPAR Signaling as a Brain Metastasis Mediator
We established two BrM derivatives from parental WM4265.2 cells, designated WM4265.2-BrM1 and WM4265.2-BrM2. WM4265.1-BrM2 cells showed significantly lower brain metastasis potential relative to WM4265.2-BrM1 despite the fact that they were selected from the highly brain metastatic parental WM4256.2 cells (Fig. 2A and B). To form brain metastases from circulating cancer cells, sequential steps are required: (i) cancer-cell migration across the blood–brain barrier (BBB), (ii) cancer-cell survival in the brain microenvironment, and (iii) cancer-cell growth. Thus, we compared the parental WM4256.2, the highly brain metastatic WM4265.2-BrM1, and the low–brain metastatic WM4265.2-BrM2 cell lines for these aspects. We first quantified the number of cancer cells extravasated across the BBB in the experimental mice. Seven days are required for the cancer cells to completely pass the BBB to get into the brain parenchyma (12). There was no difference in the migration of the three WM4256.2 cell lines across the BBB (Fig. 2C). We further compared the survival and growth of the cancer cells in vitro. None of the WM4256.2 cell lines were sensitive to natural apoptosis inducers (sFasL or TRAIL). Thus, we applied two drugs with different killing mechanisms, cisplatin (DNA- damage inducer) and staurosporine (broad protein kinase inhibitor), to induce cell death (Fig. 2D; Supplementary Fig. S3A). We did not detect any difference in either IC50 (Fig. 2E; Supplementary Fig. S3B) or drug-induced apoptosis (detected by cleaved caspase-3; Fig. 2F; Supplementary Fig. S3C) in the three WM4256.2 cell lines. Lastly, we tracked the growth rate of the three WM4256.2 cell lines in culture. Both highly brain metastatic parental WM4265.2 and WM4265.2-BrM1 cells showed a growth advantage over low–brain metastatic WM4265.2-BrM2 cells (Fig. 2G). In addition, in the low-serum condition (1% serum), we observed a significant progrowth effect of astrocytes only in parental WM4265.2 and WM4265.2-BrM1 cells, but not in WM4265.2-BrM2 cells (Fig. 2H). This was not due to the difference in astrocyte interactions during coculture in vitro (Supplementary Fig. S3D) or in brain metastatic lesions in vivo (Supplementary Fig. S3E). Thus, the growth of WM4265.2-BrM1 cells is faster than WM4265.2-BrM2 cells and can be further enhanced by astrocyte coculture.
To obtain a broad picture of the putative underlying mechanisms, we performed RNA sequencing (RNA-seq) to unbiasedly compare (i) gene-expression profiles between WM4265.2-BrM1 and WM4265.2-BrM2 cells and (ii) astrocyte-induced changes between WM4265.2-BrM1 and WM4265.2-BrM2 cells. Using Ingenuity Pathway Analysis (IPA), we identified multiple pathways that were differentially activated under each comparison (Fig. 2I and J). Between these two comparisons, we identified two shared pathways: the PPAR and eukaryotic initiation factor 2 (EIF2) pathways (Fig. 2I and J; Supplementary Fig. S4A–S4C). EIF2, an enhancer of the translation of specific stress-related mRNA transcripts, has been shown to promote proliferation and survival of cancer cells (25). This is consistent with our functional assays showing increased cell growth of WM4265.2-BrM1 cells both cultured alone (Fig. 2G) and together with astrocytes (Fig. 2H) relative to WM4265.2-BrM2 cells. Another shared activated pathway was the PPAR signaling pathway. Therefore, we set out to further validate the activity and the functional relevance of PPAR pathways in brain metastasis.
PPARγ Signaling Is a Brain Metastasis Mediator
PPARs are ligand-activated transcription factors of the nuclear hormone receptor superfamily comprising the following three subtypes: PPARα, PPARβ/δ, and PPARγ. All of these PPAR members form heterodimers with nuclear retinoid X receptor (RXR) and bind to the common peroxisome proliferator–activated receptor response element (PPRE) to activate target genes. However, differential activation of PPARs elicits distinct biological activities (26, 27). To validate our IPA and to begin to identify the subtype(s) of PPARs involved, we assessed the binding of individual PPARs to PPRE in three WM4265.2 cell lines. We detected increased PPRE binding by PPARβ/δ and PPARγ, but not PPARα, in parental WM4265.2 and WM4265.2-BrM1 relative to WM4265.2-BrM2 cells (Fig. 3A). These results validate our IPA from RNA-seq and suggest a correlation between increased PPAR activity and enhanced brain metastasis potential. To further elucidate the nature of the PPAR signaling, we determined the level of protein expression and the location of PPARγ and PPARβ/δ. We observed elevated PPARγ expression, but not PPARβ/δ expression, in parental WM4265.2 and WM4265.2-BrM1 cells relative to WM4265.2-BrM2 cells (Fig. 3B). This elevated protein expression of PPARγ was not due to an increase in its steady-state mRNA level (Supplementary Fig. S5A). Importantly, in all of the melanoma and breast cancer BrM cells tested, the majority of PPARγ was localized to the nucleus, where PPARγ binds to RXR to activate gene expression (Fig. 3B; Supplementary Fig. S5B). In contrast, the majority of PPARβ/δ was localized to the cytosol (Fig. 3B; Supplementary Fig. S5B). These data suggest the PPARγ pathway is activated in cancer cells that possess a high ability to form brain metastases.
To further elucidate the PPAR pathway underlying the growth advantage of BrM cells, we applied specific antagonists of PPARγ (T0070907) or PPARβ/δ (GSK3787) to assess their effects on the growth of WM4265.2-BrM1 cells. A PPARγ antagonist, but not PPARβ/δ antagonist, inhibited cell growth in complete cell culture medium (10% serum; Fig. 3C). Moreover, the growth advantage provided by the astrocyte coculture under low-serum conditions (1% serum) was inhibited by the PPARγ antagonist, but not by the PPARβ/δ antagonist (Fig. 3D). In contrast, a PPARγ agonist (rosiglitazone), but not a PPARβ/δ agonist (GW501516), promoted the growth of WM4265.2-BrM1 cells (Supplementary Fig. S5C). We also compared the responses to rosiglitazone between high PPARγ–signaling WM4265.2-BrM1 and low PPARγ–signaling WM4265.2-BrM2 cells. A higher dose of rosiglitazone was required to elicit progrowth effect in WM4265.2-BrM2 cells (Fig. 3E). We replicated these experiments to confirm the PPARγ-activated cell growth in high brain metastatic breast cancer MDA231-BrM and mouse melanoma Yumm1.7-BrM cells (Supplementary Fig. S5C-5E).
To specifically ascertain the functional relevance of PPARγ signaling in brain metastasis, we used genetic approaches to deplete PPARγ expression in both WM4265.2-BrM cells using short hairpin RNA (shRNA). Because constitutive depletion of PPARγ may affect proliferation and subsequently the heterogeneity of the WM4265.2-BrM cells, we used doxycycline-inducible shRNA to knock down the expression of PPARγ (Fig. 3F; Supplementary Fig. S5F). PPARγ depletion in WM4265.2-BrM1 cells decreased cell growth (Fig. 3G) and astrocyte-induced enhancement in cell growth (Fig. 3H) in vitro. In contrast, decreasing the level of PPARγ did not affect the growth of WM4265.2-BrM2 cells (Fig. 3G). To test the effect of PPARγ depletion on brain metastasis in vivo, we injected these doxycycline-inducible PPARγ knockdown WM4265.2-BrM1 cells into experimental mice treated with doxycycline-infused food and water. As shown previously (24), doxycycline can cross the BBB into the brain metastatic lesions. Depleting PPARγ in WM4265.2-BrM1 cells significantly decreased brain metastatic burden (Fig. 3I). Thus, PPARγ signaling facilitates metastatic outgrowth in the brain.
PPARγ Signaling Promotes the Proliferation of Brain Metastatic Cancer Cells
To determine how the PPARγ pathway contributes to the growth of BrM cells, we tested the effect of a PPARγ antagonist on cell proliferation and apoptosis. We prelabeled BrM cells with CellTrace dye to track and quantify their proliferation. In complete cell culture media (10% serum), T0070907 inhibited the proliferation of the highly brain metastatic melanoma WM4265.2-BrM1 and breast cancer MDA231-BrM cells (Fig. 4A). Compared with WM4265.2-BrM1 cells, this inhibitory effect was less in WM4265.2-BrM2 cells (Fig. 4A). We used Annexin V and DAPI staining to track and quantify the death of BrM cells. Under the more stringent low-serum condition (1% serum), T0070907 did not affect cell death (Fig. 4B). Similarly, neither rosiglitazone nor T0070907 altered cancer-cell apoptosis (detected by cleaved caspase-3) induced by TRAIL or drugs inducing cell death (Fig. 4C; Supplementary Fig. S5G). Lastly, under the low-serum condition (1% serum), astrocyte coculture increased the proliferation of highly brain metastatic WM4265.2-BrM1 and MDA231-BrM cells, and this effect was diminished by T0070907 (Fig. 4D). In contrast, T0070907 did not change the protective effect of astrocytes on MDA231-BrM cells treated with TRAIL (Fig. 4E). Therefore, we conclude that PPARγ activation contributes to BrM cell proliferation, but not survival, and is further enhanced by the presence of astrocytes.
Polyunsaturated Fatty Acids Released from Astrocytes Activate the PPARγ Pathway in Brain Metastatic Cancer Cells
PPARγ can be activated by both naturally occurring ligands (e.g., polyunsaturated fatty acids) and pharmacologically synthesized agents (e.g., rosiglitazone). The brain is enriched with polyunsaturated fatty acids (13), which are critical precursors to generate phospholipids for the cell membrane. In the brain, fatty-acid synthesis in astrocytes is critical for the normal function of the brain (14–16). For example, fatty acids produced by astrocytes are taken up by neurons to support synapse formation and function (28, 29). We hypothesize that astrocytes serve as a “donor” of polyunsaturated fatty acids to activate PPARγ signaling in the surrounding BrM cells. We collected whole-cell lysates from WM4265.2-BrM1 and MDA231-BrM cancer cells and human astrocytes to quantify 70 different types of fatty acids by mass spectrometry. We detected reliable peaks for 45 fatty acids (Fig. 5A). Overall, astrocytes have a much higher content of detected fatty acids than our brain metastatic cells (Fig. 5A). Compared with both melanoma and breast cancer BrM cell lines, the top three enriched fatty acids in astrocytes were AA (20:4), mead acid (20:3), and docosahexaenoic acid (DHA; 22:6; Fig. 5A and B). When directly added into the culture medium, AA and mead acid, two structurally similar fatty acids, promoted the growth of WM4265.2-BrM1 and MDA231-BrM cells (Fig. 5C). Compared with WM4265.2-BrM1, the WM4265.2-BrM2 cell line, which shows modest PPARγ signaling, had a markedly decreased response to AA (Supplementary Fig. S6A). In contrast, DHA did not change the growth of any of the BrM cells (Fig. 5C; Supplementary Fig. S6B). Furthermore, the progrowth effect of AA and mead acid was abolished by the specific PPARγ antagonist T0070907 (Fig. 5D). The combined data support the premise that AA and mead acid enhance BrM cell growth by activating PPARγ signaling.
We next investigated whether the fatty acids can be released from astrocytes to activate PPARγ signaling in the surrounding BrM cells. We first quantified AA, mead acid, and DHA in the conditioned medium (CM) of cultured astrocytes and BrM cells. Serum-free culture medium was used to collect CM to avoid any exogenous contamination of fatty acids. In astrocyte CM, AA was secreted at the highest concentration, followed by DHA and mead acid (Fig. 6A), which was consistent with our findings of high AA content in astrocyte lysates (Fig. 5A). About 2-fold more DHA than mead acid was detected in astrocyte CM (Fig. 6A), even though astrocyte lysates had higher mead acid content (Fig. 5A and B). These data suggest that the secretion of DHA may be more efficient than that of mead acid. Notably, compared with BrM cells, astrocytes secreted higher amounts of these detected fatty acids (Fig. 6A). Secondly, we tested whether astrocyte CM could activate PPARγ signaling in BrM cells. Astrocyte CM increased PPARγ-dependent PPRE binding in the BrM cells (Fig. 6B). Similarly, exogenous AA increased PPARγ–PPRE binding when directly added to the BrM cells (Fig. 6C). Lastly, astrocyte CM promoted their growth, and this progrowth effect was diminished by the specific PPARγ antagonist T0070907 (Fig. 6D). PPARγ activation has been shown to be regulated by CDK5-dependent phosphorylation of PPARγ at serine273 in adipocytes (30). However, in our BrM cells, neither astrocyte CM nor exogenous AA treatment altered this specific phosphorylation (Supplementary Fig. S6C). Overall, our data indicate that polyunsaturated fatty acids, including AA and mead acid, are released from astrocytes and activate PPARγ signaling in BrM cells to enhance their growth.
PPARγ Is a Therapeutic Target for Brain Metastasis
To validate the activation of the PPARγ pathway in clinical samples of brain metastasis, we performed IHC staining of PPARγ to detect its expression as well as its nuclear localization in the cancer cells. For melanoma, we compared normal skin, benign nevi, primary tumors, extracranial metastases [including lymph node and gastrointestinal (GI) tract], and brain metastases. Our results showed a significantly higher proportion of PPARγ-positive samples in brain metastasis lesions (Fig. 7A). Notably, in all PPARγ-positive brain metastasis lesions, we detected distinct nuclear distribution of PPARγ in melanoma cells (Fig. 7A). For breast cancer, we obtained 13 paired samples of primary and brain metastatic tumors from the same patients. These paired samples were processed in the same pathology department and stained with PPARγ at the same time. Thus, we scored the expression of PPARγ by the staining intensity (Supplementary Fig. S7A) and confirmed significantly increased PPARγ staining in the brain metastases (Fig. 7B). However, the nuclear localization of PPARγ in the breast cancer brain metastatic samples was less distinct relative to the melanoma brain metastasis samples (Fig. 7B). Our data confirm the increased expression and nuclear localization of PPARγ in brain metastases compared with primary tumors, lymph nodes, and GI tract metastases in clinical samples.
Lastly, we systemically administered the PPARγ antagonist T0070907 in our melanoma and breast cancer brain metastasis animal models to assess its therapeutic potential on brain metastatic outgrowth. Daily administration of T0070907 significantly decreased brain metastatic burden in both melanoma and breast cancer metastases, using both female and male experimental mice (Fig. 7C; Supplementary Fig. S7B). Our in vitro data suggested that T0070907 inhibits BrM cell proliferation, the step after extravasation, to establish brain metastatic lesions. Thus, we initiated treatment of the experimental mice 7 days after initial cancer cell inoculation, which is required for the extravasation of cancer cells cross the BBB (12). We observed a similar inhibitory effect of T0070907 on brain metastases (Fig. 7D). In contrast, systemic application of T0070907 did not change the growth of subcutaneously implanted tumors (Fig. 7E) or lung metastases (Fig. 7F). Thus, the therapeutic effect of the PPARγ antagonist was specific to the polyunsaturated fat–rich brain microenvironment. PPARγ signaling is one of the key pathways that regulate metabolism, particularly glucose homeostasis and fat metabolism. Thus, we assessed for any changes in body weight of the experimental mice as a potential side effect. Our data showed that daily administration of T0070907 for 28 days did not significantly decrease the body weight of either male or female experimental mice (Fig. 7G; Supplementary Fig. S7C), indicating that this drug is well tolerated in mice. Our combined data strongly support the premise that inhibiting the PPARγ pathway may be a viable therapeutic strategy to control brain metastases.
Astrocytes have diverse functions in brain metastasis (12, 24, 31–36). On the one hand, activated astrocytes release the killing factor soluble FasL in the microenvironment to induce cancer cell death (12). On the other hand, most studies suggest prometastatic function for astrocytes. Astrocytes have been known to facilitate brain metastasis by increasing the survival, trans-BBB migration, and stemness of the invading cancer cells (12, 24, 33–37), as well as by modulating the immune cells in the brain metastatic lesions (31, 32). Here, we identified a proproliferative function of astrocytes by supplying unsaturated fatty acids to activate PPARγ signaling in the invading brain metastatic cancer cells (Fig. 7H). This PPARγ activation in cancer cells is not through CDK5-dependent phosphorylation of PPARγ at serine273. As the most abundant glial cells, astrocytes are a major source of fatty-acid synthesis in the brain (14–16). Fatty acids produced by astrocytes are taken up by neurons to support synapse formation and function (28, 29). Our data suggest that, once they have migrated across the BBB to the brain parenchyma, cancer cells take advantage of this high fatty-acid microenvironment to proliferate, which is the ultimate step required for metastatic outgrowth to form macrometastases. Moreover, inflammation-activated astrocytes have been shown to increase the production and secretion of polyunsaturated fatty acids (38). This raises an intriguing possibility, in which the metastatic cancer cells themselves instigate the manipulation of the surrounding astrocytes to obtain PPARγ activators in the microenvironment.
The role of PPARγ in cancer remains controversial, and may depend on cancer types and stages (17, 18). Most studies show that activating the PPARγ pathway suppresses cancer development. Perhaps this is why all relevant cancer clinical trials (from https://clinicaltrials.gov) are using PPARγ agonists to prevent cancer development or treat primary tumors. One of the identified suppressive mechanisms shows that the PPARγ pathway modulates the transition between epithelial and mesenchymal phenotypes of cancer cells. PPARγ activation induces reverse epithelial–mesenchymal transition, also termed MET, by directly increasing the expression of E-cadherin (39, 40) and indirectly inhibiting the canonical WNT/β-catenin pathway (41–43). Consequently, cancer cells exhibit decreased invasion and migration, blocking the spreading of metastatic cells from the primary tumor to secondary metastatic organs. Little is known about the effect of PPARγ-activated MET on the proliferation of cancer cells after they achieve migration into distal metastatic organs. Moreover, increasing evidence indicates that PPARγ can act as a cancer promoter, particularly in a lipid-rich environment (44, 45). A recent study on primary brain tumors shows that PPARγ induces the production of reactive oxygen species in glioblastoma to promote tumor growth (45). Our current work focuses on the most aggressive form of cancer, the brain metastasis, and identifies PPARγ as a promoter of cancer-cell proliferation in the brain. The results not only expand our knowledge on the mechanistic underpinnings of cancer brain metastasis, but also highlight that therapeutic strategies using targeted agonists or inhibitors will have to be more context-dependent and personalized.
Human WM4265.2, WM793, WM1366, and MDA-MB-231 (MDA231) cells, murine Yumm1.7 cells, and their brain metastatic derivatives were cultured in DMEM with 10% fetal bovine serum (FBS) and 2 mM L-glutamine. For lentivirus production, 293T cells were cultured in DMEM supplemented with 10% FBS and 2 mM L-glutamine. Human and mouse primary astrocytes were cultured in media specified by the supplier (ScienCell), and used between passages 2 and 6. Patient-derived WM4265.2, WM793, and WM1366 cells were obtained from Dr. Meenhard Herlyn's lab. Yumm1.7 cells were obtained from Dr. Ashani Weeraratna's lab. MDA231-BrM cells were generated in Joan Massagué's lab. All these cells have been well characterized by fingerprint or exon sequencing (25, 46, 47). Cell authentication on newly generated BrM cells was not performed. All cells tested negative for Mycoplasma. Mycoplasma testing with MycoAlert Mycoplasma Detection Kit (Lonza) was performed at Cell Center Services, University of Pennsylvania. Each cell line was tested after isolation from the experimental mice and routinely retested every 3 to 6 months.
All experiments using animals were done in accordance with protocols approved by the Wistar Institutional Animal Care and Use Committee. Athymic NCr nu/nu mice (Charles River Laboratories), C57BL/6J mice (Jackson Laboratory), and NSG mice (The Wistar Institute) were use at 5 to 6 weeks of age. Sex of the experimental mice are indicated in the individual experiment. To establish the BrM cells by in vivo selection, we followed previously described procedures (21, 22). In brief, 5 × 104 cancer suspended in 100 μL of PBS were injected into the left cardiac ventricle. Metastasis growth was monitored by bioluminescence imaging (BLI) after retro-orbitally injecting the experimental mice with D-luciferin (150 mg/kg). At the experimental endpoint, we anesthetized mice (ketamine 100 mg/kg, xylazine 10 mg/kg), retro-orbitally injected D-luciferin, identified the brain colonization by ex vivo BLI, cultured the single-cell suspension from the brain metastatic lesions, and sorted out fluorescent-labeled cancer cells 2 weeks after in vitro culture. For both WM4265.2 and Yumm1.7 cells, we performed two rounds of in vivo selection to obtain WM4265.2-BrM1, WM4265.2-BrM2, and Yumm1.7-BrM cells. For brain metastasis assays, we followed previously described procedures (24). In brief, 104 MDA231-BrM2 cells, 5 × 104 WM4265.2-BrM cells, or 5 × 104 Yumm1.7-BrM cells suspended in 100 μL of PBS were injected into the left cardiac ventricle. At the experimental endpoint, brain colonization was quantified by ex vivo BLI. For inducible knockdown experiments, mice were given doxycycline hyclate (Sigma-Aldrich) in the drinking water (2 mg/mL) and the diet (Harlan). For lung metastasis assays, 2 × 105 cancer cells in 100 μL PBS were injected into the lateral tail vein. For subcutaneously implanted tumor growth, 4 × 104 cancer cells in 50 μL PBS were injected. Female athymic NCr nu/nu mice were used for MDA231-BrM cells. Male NSG mice were used for WM4265.2-BrM1 cells because these cells did not grow subcutaneously in athymic NCr nu/nu mice. For drug treatment experiments, mice were intraperitoneally injected with T0070907 (Selleck Chemicals or synthesized by Wistar Molecular Screening and Protein Expression Facility; 5 mg/kg/day). Vehicle (5% DMSO, 45% polyethylene glycol 300 in water) was used in control mice. Body weight of every experimental mouse was measured on day 1 and every 7 days. Normalized body weight was calculated as the measured weight divided the initial body weight from day 1. BLI was performed using an IVIS SpectrumCT In Vivo Imaging System (PerkinElmer) and analyzed using Living Image software. In the selected experiments to count the number of lesions in the brain, a diffuse light imaging tomography sequence was set up within the Living Image software in conjunction with the IVIS SpectrumCT. A combination of five 2-D luminescence images was obtained at varying filters between 560 and 640 nm on the specimen. Then the specimen was lowered to the CT portion of the IVIS SpectrumCT, where a microCT image was obtained of the specimen to showcase the regions and surface of the brain. The count values were used to determine correctly adjusted camera settings and exposure times, with a minimum of 600 counts to reach a quantifiable luminescence signal. For brain metastasis assays, 8 to 10 mice were used in each group. For drug treatment experiments, mice were inoculated with cancer cells and randomly assigned to treatment groups. Following the established approach to quantify the number of BrM cells that migrated across the BBB to enter the brain parenchyma (12), a short-term brain metastasis assay was performed by injecting 5 × 105 cells into the left cardiac ventricle. Seven days after cancer-cell inoculation, whole-mount staining of GFP was applied to 1 of 10 of the whole-brain tissues, and the number of GFP-positive cancer cells was quantified under a fluorescence microscope.
Knockdown and Cancer-Cell Labeling Constructs
For inducible knockdown, control and PPARG shRNAs in TRIPZ lentivial vector (Dharmacon) were used. Doxycycline hyclate (1 μg/mL; Sigma-Aldrich) was added to induce the expression of shRNA. Targeted sequences of PPARG shRNAs and sequence of nonsilencing control shRNA are listed in Supplementary Fig. 8. For stably labeling the cancer cells, we used pLenti-UBC vector to express far-red luciferase-GFP or far-red luciferase–RFP fusion protein. After stably labeling the cancer cells, GFP-positive cells were sorted using Astrios EQ (MoFlo).
mRNA and Protein Detection
Total RNA was extracted using the Direct-ZolRM RNA MiniPrep Plus (Zymo Research). To prepare cDNA, 1 μg of total RNA was treated using the RevertAid RT Kit (Thermo Fisher). Sequences of primers are listed in Supplementary Fig. 8. Relative gene expression was normalized relative to ACTB (encoding human β-actin). Reactions were performed using Powerup SYBR Green Master Mix (Applied Biosystems). Quantitative expression data were analyzed using QuantStudio 6 Flex and QuantStudio Real-Time PCR Software v.1.2 (Applied Biosystems). For total protein lysates, cell pellets were lysed with RIPA buffer and protein concentrations determined by BCA Protein Assay Kit (Pierce). Cytosol and nucleus fractions were isolated from cell pellets using NE-PER Nuclear and Cytoplasmic Extraction Kit (Pierce). For Western blotting, proteins were separated by SDS-PAGE and transferred to nitrocellulose membranes (Bio-Rad). Antibodies used for immunostaining are listed in Supplementary Fig. 8.
Cancer Cell–Astrocyte Coculture Experiments
Astrocytes and cancer cells were mixed at a ratio of 1:1 or 2:1. For experiments to detect the growth effect of astrocytes on cancer cells, cancer cells (2.5 × 103–5 × 103 cells/well) were seeded with or without astrocytes in tissue culture–treated 96-well plates. In 3-D culture condition, 96-well plates were coated with 50 μL 1.5% Difco Noble Agar (Becton Dickinson). For astrocyte CM experiments, astrocytes were cultured until 90% confluent and continued to be cultured in 1% FBS-containing media. CM was collected after 48 hours and went through 0.45-μm filter before being added to cancer cells. For fatty-acid experiments, AA, mead acid, or DHA (Cayman Chemical Company) was added in the cancer cells. The same amount of ethanol was used as vehicle control. In all these coculture, CM, and fatty-acid treatment experiments, the growth of cancer cells was quantified after 48 hours by BLI (by IVIS SpectrumCT) or AlamarBlue staining (Thermo Fisher; by Synergy HT from BioTek). Data were analyzed by Living Image software or Gen5 3.05. Final results were normalized by cancer cell culture–alone samples. To validate this approach to measure the growth of cancer cells, the number of GFP- or RFP-labeled cancer cells was quantified under fluorescence microscope or by flow cytometry in the indicated experiments. For cell-proliferation assays, cancer cells were prelabeled with CellTrace Violet dye before culture. The initial label and the final label intensity in GFP+ cancer cells were measured by flow cytometry (BD FACSAria II from BD Biosciences). Proliferation index was calculated by dividing the measured intensity after 48 hours by the initial intensity. For cell-death measurement, cancer cells were treated with TRAIL (PeproTech), cisplatin (Acros Organics), and staurosporine (Selleckchem) for 24 hours. Cell death was detected by caspase-3 cleavage (using Western blotting) or Annexin V/DAPI staining (Thermo Fisher; the apoptotic and necrotic cells were detected by flow cytometry). Specific antagonists and agonists, T0070907 (final concentration: 10 μmol/L), GSK3787 (5 μmol/L), rosiglitazone (10 μmol/L), and GW501516 (1 μmol/L), were all purchased from Tocris Bioscience and added to the cultured cells. The same amount of DMSO was used as vehicle control. The doses of agonists and antagonists were selected based on the related references recommended by Tocris Bioscience. We tested different doses of fatty acids, agonists, and antagonists, based on previously published work or recommended dose from Tocris Bioscience, in our initial experiments in the lab to optimize the doses for our BrM cells. For RNA-seq experiments, BrM cells were cocultured with astrocytes for 24 hours, and the GFP+ cancer cells were sorted using Astrios EQ (MoFlo).
RNA-seq and Bioinformatics Analysis
mRNA purified from cancer cells (n = 4 biologically independent experiments) was used. Sequencing libraries were prepared from RNA samples using QuantSeq (Lexogen). Samples were aligned using hg38 and 2-pass STAR alignment. Gene and transcript level counts were calculated using RSEM using Ensembl v75 annotation. All reads within any transcript's coding region were counted to get expression for each gene. Raw counts were tested for differential expression using the DESeq2 method after filtering out lowly expressed genes (genes with at least 10 raw counts in at least one sample were considered). DESeq2-normalized count values were used for determining expression differences. Differentially expressed genes (FDR = 0.05, as the cutoff) under each indicated comparison were used for IPA (QIAGEn Bioinformatics). Gene set variation analysis was performed on normalized raw counts using gene sets (PPAR and EIF2) curated from the IPA results. All these were performed by Wistar Genomics and Bioinformatics Facilities.
PPAR Binding Assay
Nucleus fractions, isolated from cell pellets using the NE-PER Nuclear and Cytoplasmic Extraction Kit (Pierce), were used with the PPARα, δ, γ Complete Transcription Factor Assay (Cayman Chemical). In the astrocyte CM experiments, CM was collected 48 hours after treating human astrocytes with 1% serum culture medium. BrM cells were harvested after being treated with CM for 8 hours.
Mouse brains were fixed with 4% paraformaldehyde, 70-μm sections were cut by cryostat (Thermo Fisher) and whole-mount staining was applied following established protocols (12). For immunostaining of 3-D cultured cell spheroids, cells were fixed with 4% paraformaldehyde and stained. For clinical samples, paraffin sections were stained by the Wistar Histotechnology Facility. Antibodies used for immunochemical staining are listed in Supplementary Fig. 8. Images were acquired with Nikon 80i microscope (Nikon Instruments) or TCS SP5II upright confocal microscope (Leica), and analyzed with LAS AF and NIS-Elements software.
Human astrocytes and BrM cancer cells were cultured in 1% serum culture media for 48 hours. The cells were washed in PBS and then scraped into ice-cold methanol and transferred to glass vial for lipid extraction. The same volume of SPLASH! LIPIDOMIX (Avanti Polar Lipids) dissolved in methanol was introduced to each sample as an internal standard. To measure secreted fatty acids, cells were cultured in serum-free media for 48 hours. Chloroform and deoxygenated ice-cold PBS were added and samples were centrifuged. The lower phase was collected and an equivalent volume from each sample was dried under a nitrogen stream. Samples were then redissolved in 0.3 M potassium hydroxide in 90% methanol and incubated at 80°C for 1 hour. Formic acid and hexanes were added. The upper phase was dried under a nitrogen stream before resuspension in methanol. LC/MS analysis was performed using HILIC chromatography and Thermo Q-Exactive HF-X mass spectrometer (Thermo Fisher Scientific). Raw data analysis was performed using TraceFinder software (Thermo Fisher Scientific). Peak areas were normalized to internal standard. We used deuterated oleic acid as the internal standard. Fatty-acid content in cell lysates was normalized by the total protein amounts. Synthetic AA, mead acid, and DHA (Cayman Chemical Company) were used to generate standard curves to quantify and calculate the amount of each fatty acid secreted in the CM.
Clinical Sample Analysis
Normal skin, benign nevi, primary tumors, and lymph node and GI tract metastatic melanoma samples were purchased as tissue arrays from US Biomax. Paraffin-embedded tissues of melanoma brain metastases (14 cases) and paired primary tumors/brain metastases (13 pairs) were obtained from the University of Pennsylvania Departments of Pathology and Laboratory Medicine and ChristianaCare Health System, in compliance with the University of Pennsylvania or Wistar Institutional Review Board. The studies were conducted in accordance with recognized ethical guidelines. Written informed consent was obtained from all subjects. IHC staining for PPARγ was performed by the Wistar Histotechnology Facility. For the paired breast cancer samples, 3 to 8 images, depending on the tissue size, were obtained from each sample for PPARγ staining score.
Statistical analysis was performed using GraphPad software (Prism) and Student t test (two-tailed). P values <0.05 were considered statistically significant.
Disclosure of Potential Conflicts of Interest
X. Xu is a consultant for CureBiotech, Inc., and GlaxoSmithKline, has received a commercial research grant from CureBiotech, Inc., and has ownership interest (including patents) in Exio Biosciences, Inc., and CureBiotech, Inc. A.T. Weeraratna is an unpaid consultant/advisory board member for the Melanoma Research Foundation and Phoremost Technologies. Per the corresponding author, none of these activities are related to this manuscript. No potential conflicts of interest were disclosed by the other authors.
Conception and design: Y. Zou, Q. Chen
Development of methodology: Y. Zou, A. Watters, N. Cheng, G.M. Alicea, X. Xu, Q. Chen
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Zou, A. Watters, N. Cheng, C.E. Perry, G.M. Alicea, J.L.D. Parris, E. Baraban, P. Ray, A. Nayak, X. Xu, M. Herlyn, M.E. Murphy, A.T. Weeraratna, Z.T. Schug, Q. Chen
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Zou, N. Cheng, C.E. Perry, K. Xu, X. Xu, Q. Chen
Writing, review, and/or revision of the manuscript: Y. Zou, K. Xu, X. Xu, M.E. Murphy, A.T. Weeraratna, Z.T. Schug, Q. Chen
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Zou, C.E. Perry, M. Herlyn
Study supervision: Q. Chen
We thank Maria Cecilia Nunes, Brian Keith, and Dario Altieri for insightful discussions. This work was supported by P50 CA174523; Susan G. Komen CCR (CCR17487999); Jayne Koskinas Ted Giovanis Foundation for Health and Policy, a Maryland private foundation dedicated to effecting change in the health care industry for the greater public good; V Foundation for Cancer Research (Q. Chen); The Ching Jer Chern Memorial Award (Y. Zou); and T32 CA009171 (J.L.D. Parris). Core facilities used in this study are supported by P30CA010815.