Breast cancer exploits neural signaling pathways for bone-to-meninges metastasis

Abstract

The molecular mechanisms that regulate breast cancer cell (BCC) metastasis and proliferation within the leptomeninges (LM) are poorly understood, limiting development of effective therapies. Here we show that BCCs in mice can invade the LM by abluminal migration along blood vessels that connect vertebral/calvarial bone marrow and meninges, bypassing the blood-brain barrier. This process is dependent on BCC engagement with vascular basement membrane laminin through expression of the neuronal pathfinding molecule integrin α6. Once in the LM, BCCs co-localize with perivascular meningeal macrophages and induce their expression of the pro-survival neurotrophin, GDNF. Intrathecal GDNF blockade, macrophage-specific GDNF ablation, or deletion of the GDNF receptor, NCAM, from BCCs, inhibit BC growth within LM. These data suggest integrin α6 and the GDNF signaling axis as new therapeutic targets against BC LM metastasis.

One-Sentence Summary:

BC hijacks CNS signaling pathways for metastasis to the LM niche.

Breast cancer (BC) patients diagnosed with metastases to the leptomeninges (LM), the cerebrospinal fluid (CSF)-containing membranes surrounding the brain and spinal cord (Fig. 1A), have a median survival of less than six months (1, 2). Despite recent advances in treating brain parenchymal metastases, there has been little improvement in leptomeningeal disease (LMD) outcomes (3–5). A significant barrier to the development of effective, targeted therapies for LMD has been our limited understanding of the molecular mechanisms that regulate metastatic invasion and survival within the unique LM microenvironment, which is both anatomically and immunologically sequestered and nutrient-poor (6, 7).

Fig. 1. BCCs invade the LM along the abluminal surface of emissary vessels.

(A) Schematic of leptomeninges (LM). (B) Incidence of bilateral hindlimb paralysis (BHLP) in 1833-P-engrafted mice. (n = 8 mice). (C) tdTomato (tdT) IHC image of spine section from a 1833-P-engrafted mouse. Dotted lines = LM. SA = subarachnoid space. (D) Schematic of EO-LM2 derivation. (E) BHLP scores of EO-Parental (n = 8), -LM1 (n = 6), or -LM2 (n = 8) mice at endpoint. Kruskal-Wallis with Dunn’s multiple comparisons test. (F) H&E of spine showing EO-LM2 cells entirely filling the meninges. Below dotted line = meninges. (G) Spine microCT of healthy vs. EO-LM2-tumored mice with BHLP. Dotted lines outline spinal canal. Brackets = bone channels encasing emissary vessels. Bone channel measurements from microCT images: healthy control, n = 3 mice/group, n = 48–109 channels/mouse. One-way ANOVA with Tukey’s post-hoc. (H) H&E and tdT IHC showing EO-LM2 and 1833-P cells surrounding emissary vessels in the spine. Brackets = bone channels. Arrowheads = BCCs. Scale bars = 50 μm. (I) Immunofluorescence images of optically cleared skulls showing tumor cells (red) migrating on α-SMA+ emissary vessels (green) that passage through the BM (white) to the LM. Brackets = bone channels. Arrowheads = BCCs. *P < 0.05, **P <0.01, ***P < 0.001. ns = not significant. ± s.e.m. (J) Quantification of tdT+ EO-LM2 cells detected in the brain parenchyma, choroid plexus, or LM by whole-mount confocal imaging, IHC, or IVM respectively (n = 3–5 mice/group). One-sample t-test against zero for LM Day 3 (D3). (K) Quantification of tdT+ 1833-P cells detected in the brain parenchyma, choroid plexus, or LM by whole-mount confocal imaging, IHC, or IVM respectively (n = 3–4 mice/group). Day = D. (L) Intravital confocal microscopy (IVM) Z-stack images of the LM of mice at multiple time-points post-EO-LM2 engraftment. BM = bone marrow. Arrowhead = perivascular tumor cluster. Arrows = BCCs in BM.

Elegant experiments have shown that breast cancer cells (BCCs) can colonize the LM by compromising the blood-CSF barrier of the choroid plexus (8) – a CSF-producing secretory epithelium within the ventricles of the brain. However, many BC patients with LMD do not have evidence of choroid plexus involvement (9). We thus investigated a potential alternative anatomic route for BC LM metastasis. We also probed the BCC crosstalk that enables tumor cells to subvert CNS-protective immunologic response mechanisms and thrive within the LM niche.

BCCs enter the CNS along emissary vessels bridging BM and LM

Our group recently discovered that leukemia cells in the vertebral/calvarial bone marrow (BM) can invade the LM by migrating along the laminin-rich abluminal/external surface of emissary vessels that directly connect the BM and LM, bypassing the need to cross the blood-brain barrier (Fig. 1A)(10, 11). Subsequent work has revealed that benign immune cells also utilize this transit corridor to efficiently enter the meninges (12–15). Previous findings from a large autopsy series indicate that the majority of BC patients with LM metastasis have coexisting vertebral or calvarial bone metastases located near prominent sites of LMD involvement (9). This observation suggests the possibility of direct BM-to-LM metastasis, though a mechanism has never been identified in solid tumors (9). We therefore utilized a well-established xenograft model and a new, immunocompetent mouse model of bone metastatic BC to investigate whether an emissary vessel corridor traversing the BM, pachymeninges (dura) and LM (arachnoid and pia mater) could be employed by malignant epithelial cells to colonize the LM.

Overt LMD involvement is an infrequent occurrence in systemically-engrafted xenograft mouse models of metastatic BC (16). In the bone tropic 1833-tdTomato-Parental (1833-P) subclone SCID mouse model of the human MDA-MB-231 cell line, however, we found that mice commonly develop progressive, bilateral hindlimb paralysis (BHLP), a potential sequela of LM metastases and cauda equina syndrome (Fig. 1B and table S1). As BHLP in BC patients can result from pathologic vertebral fracture with spinal cord compression, we used microCT to image the lumbosacral vertebrae of mice with BHLP. This demonstrated the integrity of the spinal canal (fig. S1A). By histologic analysis, we then confirmed the presence of brain and spinal LM metastases in mice with BHLP (Fig. 1C and fig. S1B), indicating that BHLP was a proxy for advanced LMD in these mice.

In syngeneic mouse BC models, spontaneous LM metastasis has not been reported. By serial intracardiac engraftment of murine EO771-tdTomato-Parental (EO-P) BC cells harvested from the meninges of a mouse with a rare occurrence of BHLP, however, we generated a subclone (EO-LM2) that displays a 100% rate of BHLP and LMD involving the spinal cord and brain (Fig. 1, D to F and fig. S1B). Similar to 1833-P BCCs, EO-LM2 cells are bone tropic and metastasize to the skull and vertebral column, but do not cause vertebral fractures (fig. S1, A to C).

To determine whether BCCs metastasized to the LM by migrating along the abluminal surface of BM-LM bridging emissary vessels, we performed detailed radiographic and histologic analyses of lumbar vertebrae and ex vivo immunofluorescence (IF) imaging of calvaria (17) in 1833-P and EO-LM2-engrafted mice. High resolution microCT imaging demonstrated that mice with LMD had a widening in the mean diameter of vertebral body emissary vessel bone channels, the apertures through which the BM emissary vessels enter into the spinal canal and LM, (Fig. 1G and fig. S2A and Movie S1), suggesting involvement by disease. We confirmed these radiographic findings by analysis of histologic cross-sections of the spines (Fig. 1H and fig. S2B). At vertebral levels corresponding to BM metastatic involvement in tumored mice, we identified bone channels containing perivascular BCC clusters in transit between BM and adjacent LM (Fig. 1H and fig. S2B). Consistent with these data, ex vivo imaging of optically cleared calvaria revealed BCCs engaging the abluminal surface of α-SMA+ vessels (10) within intact emissary bone channels (Fig. 1I, fig. S2C and Movie S2). These findings provide evidence of an efficient route for BCC migration into the LM that is independent of the bloodstream and does not rely on the breach of normal anatomic barriers, but instead exploits BCC BM tropism and hematopoietic trafficking pathways that link the marrow and CNS.

To investigate the kinetics of LM colonization and alternative routes of LM metastasis in these models, we used a combination of ex vivo and in vivo approaches (fig. S3). First, we analyzed brain histologic sections for evidence of BC metastasis via the bloodstream through the choroid plexus vasculature, but did not detect EO-LM2 or 1833-P BCCs within the choroid plexus at any post-engraftment time-points (Fig. 1, J and K and fig. S4A). Next, we used single-cell resolution, video-rate intravital microscopy (IVM) through thinned skull windows of anesthetized mice (fig. S4, B to D) to examine interactions with LM vessels in real time. On the day of intracardiac engraftment (day 0), we detected occasional BCCs circulating through LM vessels, but none arrested within the vascular lumen or transmigrated from circulation (Fig. 1, J to L and fig. S4, E and F and Movie S3). In marked contrast, BCCs rapidly adhered to the lumen of vessels in the adjacent calvarial BM and began to extravasate into interstitial marrow spaces within hours (fig. S4, E, G to I). On day 3 and day 23, respectively, EO-LM2 and 1833-P BCCs first appeared in the LM on the abluminal surface of the LM vasculature, despite their continued absence from the circulation and from the adjacent surface of the brain (Fig. 1, J to L, fig. S4, E and F, J and K). In contrast, similar to observations in other CNS-tropic BC models (18, 19), EO-LM2 and 1833-P BCCs lodged within the lumen of brain parenchymal vessels immediately after intracardiac engraftment (day 0) and remained intraluminal before extravasating across the blood-brain barrier on days 3–14 (Fig. 1, J and K and fig. S5A). At timepoints corresponding to LM colonization, however, brain parenchymal metastases remained in close association with their vessels of origin, did not contact the LM, and were no closer than 190 μm from the pia mater, making it unlikely that brain-metastatic BCCs were the source of LM seeding (fig. S5B). Taken together, these findings suggest that while 1833-P and EO-LM2 BCCs can efficiently colonize the brain parenchyma and BM from the bloodstream, their entry to the LM originates via abluminal migration (fig. S6).

Integrin α6 expression by BCCs mediates abluminal vascular LM invasion

In mouse models of B-ALL, the laminin receptor integrin α6 plays a key role in abluminal LM metastasis along the laminin-rich basement membrane of emissary vessels (10). We therefore investigated whether laminin receptors also regulate BC LM metastasis. We first confirmed the cell surface expression of the high affinity laminin receptor, integrin α6 (20, 21), on the 1833-P and EO-LM2 cell lines (fig. S7, A to D). We next utilized CRISPR-Cas-9 to delete (KO) integrin α6 from these cells. While integrin α6 KO had no effect on in vitro proliferation (fig. S7, E and F), it resulted in a substantial decrease in BCC invasion and migration on laminin (fig. S7, G and H). To examine the contribution of integrin α6 to LM invasion in vivo, we engrafted mice intracardially with either the parental or integrin α6 KO 1833 and EO-LM2 cell lines. In both models, we found that integrin α6 KO resulted in prolonged survival due to a decreased incidence of BHLP as a clinical endpoint (Fig. 2, A, B, E and F, fig. S8A). Through analysis of histologic brain sections, we confirmed a marked decrease in LMD in 1833- and LM2-α6 KO engrafted mice (Fig. 2, C, D, G and H). In contrast, integrin α6 KO BCCs demonstrated equivalent growth in the brain parenchyma, lung, and BM compared to parental BCCs, suggesting that α6 expression was dispensable for metastasis at these other tissue sites (fig. S8, B to M).

Fig. 2. Integrin α6 mediates BCC metastasis to the LM.

(A) Kaplan–Meier survival curve for 1833-P vs. -α6KO mice. Two-sided log rank Mantel-Cox test, n = 7 mice per group. (B) Bilateral hindlimb paralysis (BHLP) incidence for paired 1833-P vs. -α6KO mice euthanized when either reached a clinical endpoint. Parental n = 8, α6KO n= 7. Fisher’s exact test. (C) Representative tdT IHC of brain sections from mice in (B). Dotted lines = leptomeninges (LM). (D) Quantification of the number of tdTomato+ (tdT) BCCs in the leptomeninges (LM) of mice engrafted with parental or α6 KO cells at matched time-points. n = 10 brain sections quantified per mouse. n = 6 mice/group. The difference in cell number between matched mouse pairs was taken and then log-transformed prior to running a One-sample t-test against zero. (E) Kaplan–Meier survival curve for EO-LM2 vs. -α6KO mice. Two-sided log rank Mantel-Cox test, EO-LM2 n = 7, EO-LM2-α6KO n = 8. (F) BHLP incidence for paired EO-LM2 vs. -α6KO mice euthanized when either reached a clinical endpoint. n = 8 per group. Fisher’s exact test. (G) Representative tdT IHC of brain sections from mice in (F). Dotted lines = LM. (H) Quantification of the number of tdTomato+ (tdT) BCCs in the leptomeninges (LM) of mice engrafted with parental or α6 KO cells at matched time-points. n = 10 brain sections quantified per mouse. n = 6 mice/group. The difference in cell number between matched mouse pairs was taken and then log-transformed prior to running a One-sample t-test against zero. (I) Incidence of LMD in the brains and/or spines of MCF7-P (n = 5) vs. -α6 OE (n = 6) engrafted mice. Fisher’s exact test. (J) Representative tdT IHC of brain section of mice in (I). Dotted lines = LM. (K) Number of spine or brain sections containing tdT+ tumor cells in the leptomeninges (LM) of mice from (I). 10 brain and 14 spine sections per mouse. One-sample t-test against zero. ± s.e.m. (L) Representative intravital microscopy images and quantification of EO-LM2 and EO-P vs. -α6KO cells (red) in the LM of mice on day 3 post-engraftment and quantification of the number of BCCs in the LM. n = 3 mice per group. One-way ANOVA. ± s.e.m. (M) Percentage of mice in which EO-LM2 or EO-LM2-α6 KO tumor cells were detected in α-SMA+ emissary vessel channels of the skull on day 3 post-engraftment as determined by IF imaging of the entire optically cleared skull. n = 4 EO-LM2; n = 3 EO-LM2-α6 KO. All scale bars = 50 μm. *P < 0.05, **P <0.01, ***P < 0.001. ns = not significant. ± s.e.m.

To determine whether integrin α6 is sufficient to enable LM metastasis, we overexpressed (OE) integrin α6 in the α6-negative MCF7 cell line (fig. S9A). MCF7 parental (MCF7-P) cells establish BM micrometastases after intracardiac engraftment, but do not metastasize to the LM (Fig. 2, I and J). In contrast, 100% of mice engrafted with MCF7-α6 OE cells were found to have histologically detectable LMD in the spine and brain, and 33% developed BHLP (Fig. 2K and fig. S9B). Metastases within the choroid plexus or brain parenchyma were not detected in either MCF7-P nor MCF7-α6 OE mice, suggesting that the CNS metastasis phenotype was specific for LMD (fig. S9, C and D). Furthermore, integrin α6 OE did not affect in vitro cell proliferation nor in vivo colonization of the lung, liver or BM tissues (fig. S9, E and F).

These collective data from three distinct mouse models indicate that integrin α6 plays an important role in LM metastasis, either during early invasion and/or subsequent LMD proliferation. To test the hypothesis that integrin α6 mediates BCC abluminal emissary vessel trafficking and the earliest stage of LM invasion, we used a combination of single cell resolution IVM and ex vivo confocal microscopy of optically-cleared calvaria to examine differences in EO-P, EO-LM2, EO-P-α6 KO, and EO-LM2-α6 LM migration on day 3, the earliest timepoint at which cells were identified in the meninges (Fig. 1, J and L). Although EO-P cells do not produce overt LMD in mice, they express equivalent levels of integrin α6 (fig. S10, A and B). Consistent with a critical role for α6 in LM colonization, EO-P and EO-LM2 cells were detected in the LM at similar levels on day 3 post-intracardiac engraftment, while only rare EO-P-α6 KO or EO-LM2-α6 KO cells were identified (Fig. 2L). Imaging of optically-cleared intact skulls harvested on post-engraftment day 3 demonstrated BCCs on the abluminal surface of emissary vessels within calvarial bone channels of all EO-LM2 engrafted mice, while no BCCs were identified in EO-LM2-α6 KO engrafted mice (Fig. 2M and fig. S10C). Together, these data provide evidence of a critical dependence on integrin α6 for BCC migration into the LM along emissary vasculature.

GDNF and NCAM signaling in the LM supports BCC survival in a perivascular macrophage niche

EO-P cells express functional integrin α6 receptors and are capable of LM invasion (Fig. 2L and fig. S10A), yet in contrast to the LM-trophic EO-LM2 subclone, rarely demonstrate LMD progression in vivo (Fig. 1E). These data suggest that while integrin α6 functions to mediate LM colonization, another factor(s) differentially expressed by EO-LM2 cells mediates their successful growth in the LM niche. To screen for signaling pathways in EO-LM2 cells that could support their survival or proliferation in the meningeal niche, we performed RNA-seq analysis of EO-LM2 vs. EO-P cells. We found that NCAM1 was one of the most upregulated transcripts in EO-LM2 vs. EO-P cells (Fig. 3A and figs. S11 and S12A). This candidate was particularly intriguing, given that NCAM1 expression has been associated with an increased risk of LM metastasis in acute leukemia patients (22–24). We confirmed the cell surface expression of NCAM1 on EO-LM2 cells and absence in EO-P cells (Fig. 3B and fig. S12B). NCAM1 deletion from EO-LM2 cells did not alter in vitro proliferation, systemic disease growth or initial colonization of the LM (Fig. 3C and figs. S7F and S12, C and D). However, NCAM1 KO inhibited LMD progression (Fig. 3D and fig. S12E). Analogously, we found that 1833-P cells express NCAM2 (fig. S12F). NCAM2 KO in 1833-P did not alter systemic disease burden but increased LMD-free and overall survival (OS) of engrafted mice (Fig. 3E, fig. S12, G to I). These data indicate a role for NCAMs in BCC survival subsequent to invasion of the LM niche.

Fig. 3. GDNF:NCAM signaling increases BCC survival in nutrient-poor conditions.

(A) Volcano plot of differentially expressed genes in EO-LM2 vs. -P cells. (B) Mean Ncam1 fluorescence intensity (MFI) by flow cytometry. n = 3 biological replicates. Two-sided, unpaired Student’s t-test. (C) Representative intravital microscopy (IVM) and quantification of EO-LM2 vs. -NCAM1 KO cells in the LM of mice on day 3 post-engraftment. n = 6 mice. Two-sided, unpaired Student’s t-test. (D) Representative H&E of brain section and bilateral hindlimb paralysis (BHLP) incidence of paired EO-LM2 vs. -NCAM1 KO mice euthanized when reached a clinical endpoint (EP). n = 5 mice. Fisher’s exact test. (E) Kaplan–Meier curves showing time to BHLP score of 0.5 in 1833-P vs 1833-NCAM2 KO mice. Two-sided log rank Mantel-Cox test, n = 8 per group. (F) Percentage of cells surviving after in vitro culture under nutrient deprivation conditions +/− GDNF, NCAM or GFRα1 neutralizing antibodies. n = 3 biological replicates. One-way ANOVA with Dunnett’s test. (G) GDNF (red) immunofluorescence (IF) on meninges cytospins from healthy or EO-LM2-engrafted CSF1R-GFP mice. Dotted line = isotype. Arrows = CSF1R+ cells (green). n = 3 mice; n = 18–43 cells quantified/mouse. Two-way ANOVA with Tukey’s multiple comparisons. (H) Representative GDNF IHC on brain sections from healthy vs. 1833-P mice at EP. Arrowheads = GDNF. (I) Representative CD206, GDNF, and tdT IF on brain section of 1833-P mouse at EP. Arrows = CD206+ macrophages. Arrowheads = tdT+ 1833 cells. (J) Flow cytometry quantification of GDNF expression in cells isolated from meninges of EO-LM2 mice. n = 3 biological replicates. One-way ANOVA with Dunnett’s test. Scale bars for (G) = 10 μm, all others = 50 μm. *P < 0.05, **P <0.01, ***P < 0.001, ****P < 0.0001. ns = not significant. ± s.e.m.

NCAM1 and its paralog NCAM2 (collectively referred to as NCAM) have been shown to function as alternative receptors for the neuronal survival factor, GDNF (25, 26). GDNF is minimally expressed in the healthy adult brain and LM, but is secreted by reactive CNS microglia and macrophages in response to stroke and other brain injury, where it is deposited in the extracellular matrix and serves to block apoptotic neuronal stress responses (27–32). The LM are a relatively nutrient-poor environment wherein tumor cells must overcome substantial cellular stress to survive (33). We therefore tested whether GDNF could protect NCAM+ 1833-P and EO-LM2 cells from cell death induced by nutrient deprivation. Recombinant GDNF treatment provided marked protection from cell death (“rescue”) in nutrient-poor conditions, and this effect was reversed by NCAM receptor and GFRα1 co-receptor blockade (Fig. 3F and fig. S13A). NCAM-negative EO-P cells, on the other hand, did not demonstrate a survival response to GDNF (fig. S13B). Consistent with GDNF activation of pro-survival signaling pathways downstream of NCAM (34, 35), FAK phosphorylation in nutrient-deprived EO-LM2 cells was increased by GDNF and reversed by NCAM blockade (fig. S13 C and D). Furthermore, RNAseq analysis revealed that numerous genes involved in cell survival and proliferation were significantly upregulated by GDNF:NCAM signaling (fig. S13 E-H).

To determine if BCCs could stimulate GDNF expression in meningeal macrophages, we isolated meningeal cells from healthy vs. EO-LM2-engrafted CSF1R-GFP C57BL/6 mice, in which myeloid cells report as green-fluorescent, and performed GDNF IF. We found that CSF1R+ (myeloid) cells exhibited a greater than 3-fold increase in GDNF expression, while CSF1R- (non-myeloid) meningeal cells displayed only a modest increase in GDNF expression (Fig. 3G). GDNF IHC confirmed a marked increase in GDNF deposition within the LM matrix of EO-LM2 and 1833 diseased mice (Fig. 3H and fig. S14A). Co-immunofluorescence staining of brain histologic sections from 1833-engrafted mice revealed that CD206+ macrophages, but not tdT+ tumor cells, were a source of GDNF (Fig. 3I and fig. S14B). To further define the GDNF+ cell populations within the meninges, we conducted flow cytometry on meningeal isolates from EO-LM2 endpoint mice (fig. S15). This revealed that border associated macrophages (BAMs), including subdural, dural MHCII^low and dural MHCII^high BAMs (36), and monocyte-derived macrophages (MDMs) were the predominant source of GDNF, while other meningeal cell populations, including monocytes, neutrophils, lymphocytes, natural killer (NK) cells, stromal cells, and tumor cells, showed no or minimal GDNF expression (Fig. 3J). Macrophage numbers were significantly increased in isolates from meninges of LM2 mice (fig. S16, A and B). There were, however, no significant increases in regulatory or exhausted T cells to suggest a link between GDNF+ macrophages and suppression of T cell-mediated tumor immunity (fig. S17).

We next examined the evidence for in vivo interactions of LM macrophages and metastatic BCCs by performing IVM of the LM of tumored CSF1R-GFP mice, in which monocytes and macrophages express green fluorescence (37). We observed a near 100% co-localization of EO-LM2 and CSF1R+ cells (fig. S18A). 3D reconstruction of Z-stack images revealed BCCs tightly insinuated between blood vessels and perivascular macrophages (Fig. 4A and Movie S4), suggesting their protective encasement by CSF1R+ cells.

Fig. 4. Macrophage-derived GDNF promotes BCC survival in the LM niche.

(A) Representative 3D IVM image of EO-LM2 cells (red) co-localizing with CSF1R+ cells (green) surrounding LM vessels (white). (B) GDNF IF (magenta) on meninges cytospin from veh. or PLX5622-treated EO-LM2-engrafted CSF1R-GFP (green) mice at EP. n = 4 mice; n = 25–110 cells/mouse. Two-sided, unpaired Student’s t-test. (C) Kaplan–Meier curves showing time to BHLP score of 0.5 in veh. vs. PLX5622-treated EO-LM2 mice. Two-sided log rank Mantel-Cox test, n = 9 control, n = 10 PLX5622. (D) Representative 3D IVM images of EO-LM2 cells (red, arrowheads) co-localizing with CSF1R+ cells (green) surrounding LM vessels (white) in veh. vs. PLX5622-treated mice on day 3 or 10 post-engraftment. (E) Quantification of EO-LM2 and CSF1R co-localizing (co-loc.) or non-co-localizing (no-coloc.) cells in the LM of mice in (D). n = 3 mice. Two-way ANOVA with Tukey’s multiple comparisons. (F) Dosing strategy for intracerebroventricular (ICV) GDNF rescue experiment. (G) Kaplan–Meier survival curve of time to BHLP score of 0.5 of mice in (F). Two-sided log rank Mantel-Cox test, n = 9 veh. ICV, n = 10 GDNF ICV. (H) Breeding strategy for Csf1r-specific Gdnf knockout mice. (I) Kaplan-Meier survival curve of time to BHLP score of 0.5 for Cre-;Gdnf f/f vs. Cre+;Gdnf f/f mice engrafted with EO-LM2. Two-sided log rank Mantel-Cox test, n = 13 for Cre-;Gdnf f/f, n = 8 for Cre+;Gdnf/f. (J) Intrathecal GDNF antibody treatment strategy for mice in (K). (K) Kaplan–Meier curves showing time to BHLP score of 0.5 in anti-GDNF vs. isotype control antibody treated mice. Two-sided log rank Mantel-Cox test, n =6 isotype Ab group, n = 8 anti-GDNF group. Scale bars for (B) = 10 μm, all others = 50 μm. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.001. ns = not significant. ± s.e.m.

In order to investigate whether CSF1R+ cells in the LM niche were critical for BC LMD growth, we treated mice with the CSF1R inhibitor, PLX5622, which has been shown to dramatically deplete CSF1R+ macrophages/microglia from the CNS with modest effects on systemic CSF1R+ populations (fig. S18, B to E) (38, 39). Residual macrophages isolated from meninges of PLX5622-treated mice also showed a marked decrease in GDNF expression in subdural and Dural^lo BAMs (Fig. 4B and fig. S18, F and G). PLX5622 treatment did not affect BCC colonization of the LM as measured by IVM; however, growth of BCC within the LM was substantially decreased and, correspondingly, the time to clinical LMD progression (BHLP) was prolonged in treated mice (Fig. 4, C to E). These data indicate that resident myeloid cells promote LMD by supporting BCC survival and proliferation within the LM niche.

To test whether loss of meningeal GDNF contributed to the failure of BC LMD to progress in macrophage-depleted mice, we delivered recombinant GDNF intraventricularly in mice to rescue LM tumor growth (Fig. 4F). As shown in Fig. 4G, intraventricular administration of GDNF accelerated LMD progression in PLX5622-treated mice, suggesting that exogenous GDNF was sufficient to replace the pro-tumoral effects of ablated meningeal myeloid cells. As a complementary approach, we next bred Csf1r-Cre and Gdnf floxed mice to generate mice with specific ablation of GDNF expression in monocyte/macrophage cells (Fig. 4H and fig. S19, A and B). Systemic disease burden was equivalent in EO-LM2 Cre+;Gdnf^f/f and EO-LM2 Cre-;Gdnf^f/f control mice (fig. S19, C and D). However, EO-LM2 Cre+;Gdnf^f/f mice demonstrated prolonged LMD-free survival and decreased BHLP score at clinical endpoint compared to EO-LM2 Cre-;Gdnf^f/f control mice, supporting the importance of macrophage-derived GDNF in LMD progression, (Fig. 4I and fig. S19E).

Lastly, we tested whether GDNF blockade within the meninges could inhibit BC LMD. Using intraventricularly-implanted micro-osmotic pumps, we delivered GDNF neutralizing antibodies intrathecally by 7 days continuous infusion (Fig. 4J). Mice treated with anti-GDNF antibodies demonstrated a significant decrease in time to LMD symptom development and a significant survival benefit (Fig. 4K). Taken together, our findings from these genetic and pharmacologic approaches suggest that macrophage-derived GDNF is a key factor in BC LMD progression.

Integrin α6 and GDNF expression are associated with meningeal metastasis in BC patients

We next examined whether integrin α6 receptor expression correlated with meningeal metastasis in patients with BC diagnoses. We used IHC to compare tumor cell membrane integrin α6 expression in breast primary tumors, BM metastases, and brain biopsy specimens from patients with meningeal-based vs. brain parenchymal BC metastases (tables S2 and S3). CT and MRI imaging were reviewed to exclude any patients with breakdown of the calvarial or vertebral bone that would allow unconstrained spread to the LM. Although the rarity of biopsies performed on meningeal-based lesions limited our sample size, we found a statistically significant correlation between integrin α6 positivity and meningeal lesions (Fig. 5, A and B). Review of the medical record confirmed that all patients with meningeal lesions also had bone metastasis detected by bone scan or by BM biopsy. We next performed IHC staining to determine whether these cases expressed NCAM1 or NCAM2 in addition to integrin α6. While NCAM1 was detected in a minority of samples (3/7), all meningeal-based biopsies demonstrated extensive BC expression of NCAM2 (Fig. 5, C to E).

Fig. 5. Meninges-based BC metastases express integrin α6, NCAM, and GDNF.

(A) Tumor integrin α6+ expression in biopsies of breast cancer (BC) primary tumors (n = 35) and metastatic lesions to BM (n = 18), brain parenchyma (n = 23), or meninges (n = 7). Chi-squared test. (B) Representative integrin α6 (brown) IHC images from (A). (C) Tumor NCAM1 and NCAM2 expression in biopsies of BC metastatic lesions to meninges. (D) Representative NCAM1 IHC images from (C). (E) Representative NCAM2 IHC images from (C). (F, G) Representative IHC on serial sections of a meningeal-based breast cancer metastasis from the patient series in (A). Single stain (C): GDNF (brown), CD68 (magenta); Dual stain (D): GDNF (yellow), CD68 (magenta). Dotted lines surround tumor nests. (H) Machine learning quantification of GDNF+ area, CD68+ area, and GDNF+ vs. GDNF- CD68 cell area in patient samples from (C,D). n = 6 cases. n = 3 control cerebral cortex samples. (I) Manual quantification of GDNF/CD68 dual stained slides. n = 5 cases; n = 10 random GDNF+ 40x fields quantified per case. (J) Proposed model of integrin α6+ BCCs invading the LM along the laminin-rich abluminal surface of BM-LM bridging emissary vessels and then co-localizing with tumor-associated meningeal macrophages that secrete GDNF and activate NCAM-dependent tumor cell survival pathways. **P < 0.01;***P < 0.001. ± s.e.m.

Finally, we performed IHC staining for GDNF and CD68 expression in available cases from our panel of meningeal-based metastasis biopsy specimens. We found that 5/6 of these samples showed GDNF positivity characterized by intense staining of the peri-tumoral stroma. Compared to healthy control samples, disease samples demonstrated significantly higher GDNF positivity and CD68 staining (Fig. 5, F to I), suggesting an increase in both GDNF secretion and monocyte/macrophage abundance during malignancy. Machine learning (Fig. 5H) and pathologist’s quantification (Fig. 5I) of GDNF and CD68 staining revealed that CD68+ cells in tumor samples were frequently GDNF+, and that the majority of cells within GDNF+ regions were CD68+. These data suggest macrophages as an important source of GDNF deposition in patients.

Discussion

The LM are considered a “sanctuary site” for tumor growth, due to the limited access of both cytotoxic immune cells and therapeutic agents in this anatomic compartment (40). At the same time, it is a relatively harsh environment for tumor cell survival, requiring adaptation to its hypoxic and nutrient-poor conditions (33). The molecular processes that allow BC and other malignant cells to navigate and flourish in this microenvironment are only beginning to be elucidated (41–43). Here we show that BCCs exploit a hematopoietic cell trafficking corridor (10) to enter the LM. There, they interact with resident myeloid cells to create a favorable microenvironment for growth. This process combines neuronal mimicry, immune cell hijacking, and a restructuring of the LM to induce a more protective stromal niche. Central to this process are integrin α6 and NCAM, each of which has well-described roles in neurogenesis and neuronal pathfinding (44–47). These molecules appear to play tandem roles in BC LMD, with BC integrin α6-vascular basement membrane laminin interactions stimulating migration along the vasculature (48) of BM-LM bridging vessels, and NCAM transducing growth and survival signals, including meningeal GDNF (Fig. 5J). This is highlighted by the observation that mice engrafted with the EO-P (integrin α6+, NCAM-) and MCF7-P (integrin α6-, NCAM+) cells do not develop LMD; however, upon acquisition of both molecules, the EO-LM2 (integrin α6+, NCAM+) and the MCF7-α6 OE (integrin α6+, NCAM+) cells are able to successfully invade and colonize the LM. Interestingly, NCAM and other integrins have been reported to co-localize at the cell surface, and functional interactions between integrins and NCAM can reciprocally enhance downstream signaling events (49, 50). Integrin α6 expression may therefore not only propel LM invasion but also strengthen the metastatic “fitness” of cells in the laminin-rich meningeal microenvironment.

Macrophages are increasingly recognized to play a myriad of roles in primary brain tumors and metastases (51–55), although their function in the unique microenvironment of the LM is only beginning to be understood (41, 56, 57). Residing in close apposition to the vasculature, meningeal macrophages are ideally situated to perform a gatekeeper function. Despite this, our findings suggest a model in which BCCs in the LM are able to subvert these macrophages, paradoxically by stimulating neuroprotective macrophage inflammatory responses. Arrival of BCCs into the LM by means of a perivascular migration route may also ensure that BCCs enjoy the protection of perivascular macrophages throughout their journey. Indeed, we observed invading LM BCCs to be tightly intercalated between the vasculature and perivascular myeloid cells. This close interaction, perhaps enhanced by homotypic NCAM cell adhesion between BCCs and NCAM+ macrophages, may provide a combined physical and molecular shield against cytotoxic stressors.

Our data provide evidence that BCCs co-opt intrinsic neuronal-stroma cross-talk to thrive in the LM. Additional, and likely related, pro-tumoral microenvironment interactions undoubtedly exist. Going forward, it will be valuable to identify other neutrophins and growth factors that may also support LMD proliferation and their cellular sources; to interrogate the origin and further characterize the phenotype of pro-tumoral LM myeloid cells; and to probe alternative molecular mechanisms that may contribute to abluminal vascular migration between the BM and LM. As other LM metastatic malignancies, including lung cancer and melanoma, can also express integrin α6 and/or NCAM, it is possible that these pathways are tumor agnostic. Future efforts to targets these signaling mechanisms could therefore benefit BC and other LM metastasis patients who have a high unmet clinical need for effective therapies.

Data and materials availability:

Please contact Dorothy A. Sipkins (dorothy.sipkins@duke.edu) for primary data and/or material requests. RNA-sequencing data will be deposited into GSA upon publication.