GM-CSF signaling is critical for HER2+ breast leptomeningeal carcinomatosis


 Leptomeningeal carcinomatosis (LC), when tumor cells spread to leptomeninges surrounding the brain and spinal cord. HER2+ breast cancer is the most common origin of LC. HER2+ LC remains incurable, with few treatment options and response rates of <20%. One major limitation in development of HER2+ LC therapies has been lack of clinically relevant HER2+ LC primary cell-lines and animal models. To address this, we generated cell lines and patient-derived xenograft models using nodular HER2+ LC. This led to identification of granulocyte-macrophage colony-stimulating factor (GM-CSF) as an oncogenic autocrine driver of HER2+ LC. We observed that oligodendrocyte progenitor cells (OPCs) inhibit growth of HER2+ LC in vitro and in vivo. Furthermore, OPC-derived factor TPP1 degrades GM-CSF, decreasing GM-CSF signaling and suppressing HER2+ LC growth. Lastly, we determined that synergistic inactivation of GM-CSF signaling via the intrathecal delivery neutralizing anti-GM-CSF antibodies and a pan-Aurora kinase inhibitor (CCT137690) antagonizes development of HER2+ LC in vivo.

Introduction 1 HER2+ breast cancer leptomeningeal carcinomatosis (HER2+ LC) is an ominous complication of breast 2 cancer with a dire prognosis1, 2, 3. 10-30% of patients with metastatic breast tumors develop CNS metastases4, 3 5. HER2+ LC occurs when HER2+ breast tumor cells spread to the cerebrospinal fluid (CSF)-containing 4 leptomeninges surrounding the brain and spinal cord1, 2, 3, 6. Several factors positively correlate with a greater risk 5 of brain metastases, including young age; poorly differentiated tumors; HER2-enriched, luminal-HER2, basal-6 like, and TN breast cancer subtypes; and having four or more metastatic lymph-nodes7, 8. Once established, 7 HER2+ LC can invade the parenchyma to produce focal neurologic damage9. Any malignancy can spread to the 8 leptomeninges; however, given the high incidence of breast cancer (and particularly the HER2+ subtype) 9 worldwide, breast cancer is the most common origin of LC10. Although significant progress has been made in 10 developing breast cancer treatments that target systemic disease, efficacy in the central nervous system (CNS) 11 remains a challenge, with HER2+ LC typically developing while the systemic tumor burden is well-managed11, 12 12, 13, 14. Indeed, 30% of HER2+ LC cases are diagnosed as the first manifestation of cancer after a substantial 13 disease-free interval13, 15, 16, 17. 14 Unfortunately, although it was discovered nearly 150 years ago18, HER2+ LC remains incurable, with few 15 treatment options and response rates often <20%19, 20, 21, 22, 23, 24, 25. At present, HER2+ LC management is 16 multidisciplinary and includes radiotherapy (RT) and intrathecal chemotherapy (ITC)26, 27, 28, 29. Methotrexate 17 (MTX), a DNA alkylating drug, is frequently used as palliative ITC for HER2+ LC9,30,31,32,33. However, this 18 approach has very limited success and causes serious side effects26, 27,28,29. Furthermore,patients with HER2+ 19 LC are excluded from clinical trials because of their poor prognosis and to minimize results that are not 20 reproducible1, 34, 35, 36. Thus, our capacity to identify effective drugs to treat HER2+ LC is limited. 21 In addition, little is known about how HER2+ breast cancer cells proliferate in the compositionally simple 22 context of the leptomeninges, which are acellular and poor in protein, glucose, and cytokine content1, 2, 3, 6,. This 23 lack of understanding regarding the molecular mechanisms that govern HER2+ LC development is due in part to 24 a lack of clinically relevant primary HER2+ LC cell lines and patient-derived xenograft (PDX) models. Until 25 now, HER2+ LC analysis has been conducted primarily using floating HER2+ LC cells removed from the CSF 26 of patients in numbers insufficient to generate cell lines. Moreover, these floating cells differ from nodular tissue, 1 which represents the major clinical burden of HER2+ LC. 2 To address this limitation, we generated the first-ever expandable primary HER2+ LC cell lines and PDX 3 models. Using these models, we observed a non-uniform distribution of HER2+ LC inside the brain and spinal 4 cord, with a paucity of HER2+ LC metastases in white matter compared to the astrocyte-rich gray matter of the 5 cerebral hemispheres. This pattern recapitulates the distribution of metastases in the human CNS, where the 6 preponderance of metastases is in gray matter regions, with scarcely any in white matter. Further analysis revealed 7 that granulocyte-macrophage colony-stimulating factor (GM-CSF) is an oncogenic autocrine factor that promotes 8 HER2+ LC development in the CSF. In addition, we observed that oligodendrocyte precursor cells (OPCs), which 9 are abundant in white matter, disrupt GM-CSF HER2+ LC-derived oncogenic autocrine signaling, likely by 10 promoting GM-CSF degradation via TriPeptide Peptidase 1 (TPP1). Finally, we found that the synergistic 11 inactivation of GM-CSF signaling using anti-GM-CSF neutralizing antibodies and a pan-Aurora kinase inhibitor 12 (CCT137690) abrogates HER2+ LC growth both in vitro (using our novel HER2+ LC lines) and in vivo (using 13 our HER2+ LC PDX models). These findings bring us closer to discovering and developing more effective, novel 14 approaches for treating patients with HER2+ LC in the clinic.

Derivation and characterization of nodular HER2+ LC cells 2
To address the need for primary HER2+ LC lines that are functionally comparable to conventional nodular 3 HER2+ LC tumors, we dissociated fresh HER2+ nodular LC tissues obtained surgically for pathological 4 confirmation of LC or to decompress localized symptomatic lesions. The tissues were dissociated into single cells 5 by mechanical and enzymatic methods and then expanded on collagen-coated plates in media supplemented with 6 hCSF (Fig. 1A). We generated two primary Lepto lines (Lepto1 and Lepto2) using tissues from two different 7 patients and scaled them up to produce a cryobank of low-passage cells used for the experiments described herein. 8 Based on our observations that Lepto1 and Lepto2 cells behave similarly both in vitro and in vivo, we have 9 presented data primarily from the Lepto1 line. 10 To compare the metastatic organotropism of the primary Lepto lines, we first transduced them, as well as 11 the brain-tropic breast cancer cell line MDA-MB-231 (231-BR cells), with lentiviruses encoding mCherry and 12 firefly luciferase (mCherry:LUC). We then injected the cells (100K from each line) into the CSF space via the 13 cisterna magna puncture of NOD/SCID mice (n=12) and monitored tumor cell growth over time (~50 days) by 14 BLI. As expected, 231-BR cells generated tumors only within the brain parenchyma37, 38. In contrast, Lepto1 15 bioluminescence appeared on the surface of the brain and spinal cord (Fig. 1B). In addition, although we observed 16 fewer metastatic Leptol cells than 231-BR cells in the brain, there were overall more metastatic Lepto1 cells 17 across both the brain and spinal cord (Fig. 1B). Furthermore, Kaplan-Meier curve analysis showed that mice 18 harboring Lepto1 cells had significantly shorter survival periods than mice bearing 231-BR cells (Fig. 1C). 19 Importantly, the transcriptomic analysis demonstrated that Lepto1 lines retrieved post-implantation from the 20 leptomeninges of NOD/SCID mice exhibited gene signatures similar to those of pre-implanted primary Lepto1 21 cells (Fig. 1D). 22 Histological examination confirmed that Lepto1 cells colonized the leptomeningeal surface of the brain, 23 brain stem, and spinal cord (Fig. 1E). Moreover, we observed no Lepto1 cells were observed inside of the white 24 matter-rich brain stem and spinal cord (Fig. 1E). To confirm their regional preference (white vs. gray matter), we 25 implanted Lepto1 cells into the brain stem (white matter) or cerebral cortex (gray matter) of NOD/SCID mice. 26 BLI showed larger metastatic tumor deposits in mice with Lepto1 cells implanted in the cerebral cortex than in 1 mice with Lepto1 cells implanted in the brain stem (Fig. 1F). Furthermore, Lepto1 cell implantation into the 2 cerebral cortex significantly shortened animal survival compared to implantation into the brain stem (Fig. 1F). 3 Consistent with findings in patients that white matter is relatively tumor-resistant39 (Fig. 1G), we observed Lepto1 4 cell grafting efficiency of only ~5% (1/16) after implantation into white matter, compared to ~95% (15/16) after 5 implantation into gray matter (Fig. 1H). In addition, the immunofluorescence-based analysis of mCherry: LUC-6 labeled Lepto1 cells and their neighboring glial cell populations demonstrated that the Lepto1 cells were 7 juxtaposed to reactive (GFAP+) astrocytes in the brain (Suppl. Fig. 1A). A similar analysis revealed that Lepto1 8 cells were similarly juxtaposed to a layer of reactive (GFAP+) astrocytes on the surface of the brain stem and 9 spinal cord (Fig. 1I). In contrast, we observed that the inner core of the tissue, which is rich in OPCs, was devoid 10 of mCherry+ Lepto1 cells (Fig. 1H), suggesting that white matter resists Lepto1 cells colonization. 11

12
OPCs reduce HER2+ LC cell viability 13 To determine whether host glial cells modulate the heterogeneous distribution of metastases in the human 14 CNS, we immuno-panned CNS cell types from human iPSC-derived NPCs. Specifically, we used anti-CD45, 15 anti-GALC, anti-CD90, anti-NG2, and anti-HepaCAM antibodies to FACS sort microglia, oligodendrocytes, 16 neurons, OPCs, and astrocytes, respectively ( Fig. 2A)40. Cells were maintained in hCSF for various durations, 17 during which all cell types maintained their typical morphology and marker expression patterns ( Fig. 2A). We 18 then co-cultured the Lepto1 or Lepto2 cells along with each glial cell type in Boyden chambers to assess the 19 influence of the various cell populations on their viability (Fig. 2B). Co-culture of both Lepto1 and Lepto2 cells 20 with astrocytes increased their rate of proliferation, whereas co-culture with OPCs reduced their proliferation and 21 induced apoptosis (Fig. 2B, 2C, and Suppl. Fig. 1D). Because Lepto lines can initiate and drive tumorigenesis 22 (Fig. 1B, 1F), we performed tumorsphere formation assays to assess the effects of astrocytes or OPCs on the 23 tumor-initiating capacity of Lepto1 cells. We co-cultured Lepto1 cells with astrocytes or OPCs for 72 h and then 24 allowed the Lepto1 cells to generate spheroids on adherent culture plates. After 7 days, control (mono-cultured) 25 Lepto1 cells developed round tumorspheres, and cells co-cultured with astrocytes generated tumorspheres in even 26 greater quantity and size. In contrast, Lepto1 cells co-cultured with OPCs either remained as single cells or formed 1 very small spheroids (Suppl. Fig. 1B). Furthermore, quantitative analysis of cell viability using CCK assays 2 showed that, compared to mono-culture, co-culture with OPCs reduced the proportion of sphere-initiating Lepto1 3 cells to about 10%, whereas co-culture with astrocytes increased cell viability (Suppl. Fig. 1C). Co-culture with 4 Lepto1 cells did not alter the proliferation of microglia, oligodendrocytes, neurons, OPCs, or astrocytes (Fig. 2D). 5 To characterize the effects of OPCs on Lepto1 cells in vivo, we injected mCherry:LUC-labeled Lepto1 6 cells (100K) into the cisternae magna of adult NOD/SCID mice (Day 0). We then injected OPCs (100K) into the 7 cisternae magna on Days 7 and 14 ( Fig. 2E) and monitored tumor growth via BLI from Days 14 to 50. Mice co-8 implanted with Lepto1 cells followed by OPCs showed reduced tumor growth, based on BLI, and prolonged 9 survival relative to mice that received Lepto1 cells only ( Fig. 2F and 2G). This finding confirms the inhibitory 10 properties of OPCs seen in vitro and strongly suggests that intrathecal OPC delivery suppresses HER2+ LC 11 progression in vivo. 12 13 GM-CSF acts as an oncogenic autocrine driver of HER2+ LC cell growth 14 To determine how Lepto cells initiate tumors and drive tumorigenesis in the leptomeningeal space 15 surrounding the brain and spinal cord, we co-cultured Lepto1 cells with OPCs for 72 h in a Boyden chamber and 16 compared them to Lepto1 control cells grown for the same duration as independent monolayers. When we 17 subjected growth medium from each sample to immunoblotting using a Cytokine XL array, we observed 18 significantly higher GM-CSF levels in the growth media of mono-cultured Lepto1 cells than in that of co-cultured 19 OPC/Lepto1 cells (Fig. 3A, 3B). Indeed, GM-CSF is expressed at higher levels in Lepto1 cells than in any of the 20 CNS cell types analyzed (Suppl. Fig. 2A) and was elevated in nodular HER2+ LC tissues compared to other 21 primary tumor, normal breast, and normal brain tissues (Suppl. Fig. 2A). Furthermore, surface GM-CSF receptors 22 (GM-CSFRα, or CD116) were homogenously expressed on ~99% of Lepto1 cells (Suppl. Fig. 2B), suggesting 23 autocrine activation of the GM-CSF signaling pathway in Lepto1 lines via binding of Lepto1-secreted  Furthermore, western blot analysis of cultured Lepto1 cells showed that the presence of OPCs suppressed the 25 phosphorylation-dependent activation of GM-CSFRα and its downstream targets, STAT5, AKT, and ERK1/2 26 (Fig. 3C). Furthermore, immunohistochemical analyses of pGM-CSFRα revealed receptor activation in tumor 1 tissue but not in surrounding brain tissue (Fig. 3D). Overall, these analyses suggest that OPCs inhibit the secretion 2 of GM-CSF from Lepto1 cells, which in turn inhibits the activation of GM-CSFRα and its effectors. 3 To test this hypothesis, we analyzed GM-CSF signaling in Lepto1 cells grown in either hCSF-4 supplemented media (controls), OPC-conditioned hCSF, or hCSF containing anti-GM-CSF neutralizing 5 antibodies (Fig. 3E, 3F). Based on immunoblot analysis, GM-CSF signaling was disrupted in Lepto1 cells 6 cultured in OPC-conditioned media or in hCSF with GM-CSF neutralizing antibodies (Fig. 3E), and cells in both 7 conditions were significantly more apoptotic than control Lepto1 cells (Fig. 3F). Conditional overexpression of 8 GM-CSF in Lepto1 cells using a Tet-On 3G inducible expression system with ZsGreenI (Suppl. Fig. 2C-E) not 9 only decreased the OPC-mediated induction of apoptosis (Suppl. Fig 2F) but also reduced the OPC-induced 10 suppression of Lepto1 cell viability (Suppl. Fig. 2G). To determine whether OPC-derived or secreted factors 11 inhibit GM-CSF expression and secretion by Lepto1 cells, we generated a luciferase reporter construct containing 12 the GM-CSF proximal promoter (+1 to -500 bp) upstream of the luciferase gene. We then transfected Lepto1 and 13 Lepto2 cells with this construct and co-cultured them with OPC cells, OPC-conditioned media, or control media 14 (Suppl. Fig. 2I). Interestingly, luciferase activity (Suppl. Fig. 2I) and GM-CSF expression (Suppl. Fig. 2H) 15 were comparable across all experimental and control groups, suggesting that OPCs or OPC-derived factors do 16 not regulate GM-CSF expression. These results suggest that OPCs secrete factors that inhibit GM-CSF signaling 17 via a different mechanism (Fig. 3G). 18 Anti-GM-CSF neutralizing antibodies have been used clinically in cancer treatment41, 42. Therefore, we 19 assessed the potential anti-tumor effects of anti-GM-CSF neutralizing antibodies in our HER2+ LC PDX model, 20 administering the antibodies intrathecally and analyzing HER2+ LC formation by BLI (Fig. 3H). Relative to 21 vehicle-treated controls, antibody-treated mice showed reduced tumor progression and increased overall survival 22 ( Fig. 3I). Moreover, immunohistochemical analysis of brain sections (Fig. 3J) and western blot analysis of 23 extracted tumor lysates (Fig. 3K) revealed that intrathecal administration of anti-GM-CSF neutralizing antibodies 24 suppressed not only canonical GM-CSFRα activation and EGFR phosphorylation but also the phosphorylation-25 mediated activation of STAT5, AKT, and mTOR (downstream GM-CSF pathway effectors43, 44) in HER2+ LC 26 PDX tumors. Taken together, these results confirm the critical role of GM-CSF signaling in HER2+ LC tumor 1 progression and suggest that treatment with anti-GM-CSF neutralizing antibodies could serve as a potential 2 strategy to target HER2+ LC. 3 4 OPC-derived TPP1 is a candidate regulator of GM-CSF in HER2+ LC cells 5 Given that OPCs suppress HER2+ LC cell viability and tumorsphere initiation in vitro (Fig. 2B, 2C) and 6 inhibit HER2+ LC tumor growth in vivo (Fig. 2F), we asked whether OPCs secrete factors that inhibit GM-CSF 7 signaling and induce Lepto cell apoptosis. To do so, we used liquid chromatography-tandem mass spectrometry 8 (LC-MS/MS) to analyze the secretomes of human astrocytes or OPCs, cultured either alone or with Lepto1 cells. 9 We identified 38 unique proteins in OPC-conditioned hCSF secreted exclusively by OPCs and whose levels 10 remained unchanged in OPC-Lepto1 co-cultures ( Fig. 4A Table 1). None of the identified proteins with extracellular localization can bind to or chelate 13 GM-CSF, and only TPP1 (Fig. 4B), a serine protease in the sedolisin family45, possesses the ability to degrade 14 GM-CSF. Therefore, we hypothesized that OPC-derived TPP1 could proteolytically degrade GM-CSF, leading 15 to suppression of the GM-CSF signaling pathway and consequent inhibition of Lepto1 cell growth and viability. 16 To determine whether TPP1 regulates GM-CSF signaling, we incubated Lepto1 cells with 50 or 100 ng/mL of 17 recombinant TPP1 protein in hCSF for 24 hours and monitored GM-CSF levels by ELISA. Both concentrations 18 of TPP1 significantly reduced GM-CSF levels in Lepto1 cells (Fig. 4C) and suppressed cell viability (Fig. 4D). 19 TPP1 treatment (100 ng/mL) also reduced the phosphorylation and hence the activation of the GM-CSF effectors 20 STAT5, AKT, and mTOR (Fig. 4E). To determine the effects of this suppression, we treated Lepto1 cells with 21 an AKT inhibitor (AZD5363) or an mTOR inhibitor (Rapamycin), both currently in use or under development to 22 target solid tumors46, 47, 48, 49, 50. Treatment with either suppressed Lepto1 cell viability with IC50 values in the 23 nanomolar range (Suppl. Fig. 3A). Moreover, co-culture with OPCs and treatment with TPP1 (50 ng/mL) reduced 24 GM-CSF levels in the culture media of Lepto1 cells conditionally overexpressing GM-CSF (following treatment 25 with 5 µg/mL doxycycline, DOX), as measured by ELISA (Fig. 4F). Subsequent western blot analyses showed 26 that co-culturing GM-CSF-overexpressing Lepto1 cells with OPCs and/or TPP1 (50 ng/mL), suppressed the 1 phosphorylation-dependent activation of GM-CSFRα and its downstream effectors, AKT and mTOR, suggesting 2 decreased oncogenic autocrine GM-CSF signaling (Fig. 4G). Futhermore, co-culturing Lepto1 cells with OPCs, 3 with and without TPP1 (50 ng/mL), induced apoptosis (Suppl. Fig. 3D), suggesting that OPC-derived TPP1 4 inhibits or degrades Lepto1-secreted GM-CSF and suppresses GM-CSF effectors. 5 To assess the effects of TPP1 loss-of-function on Lepto1 lines conditionally overexpressing GM-CSF, we 6 transiently transduced OPC cells with siTPP1 and the following day co-cultured them (or control OPCs 7 transduced with siGFP or siLUC) with the Lepto1 cells (Suppl. Fig. 3E). After one day of co-culture, siTPP1-8 transduced OPC cells had significantly lower intracellular TPP1 levels than the control OPCs (Suppl. Fig. 3F). 9 Interestingly, intracellular GM-CSF mRNA levels were not affected in Lepto1 cells or OPCs transduced with 10 siTPP1, siGFP, or siLUC (Suppl. Fig. 3G). However, GM-CSF levels were significantly higher in media obtained 11 from Lepto1 cells co-cultured with TPP1-depleted OPCs than in media from Lepto1 cells co-cultured with control 12 OPCs (Suppl. Fig. 3H). Similarly, the phosphorylation of signaling effectors downstream of GM-CSF was 13 significantly greater in Lepto1 cells co-cultured with TPP1-depleted OPCs (Suppl. Fig. 3I). 14 To assess the effects of OPC-derived TPP1 on Lepto1 cells in vivo, we co-implanted Lepto1 cells and 15 OPCs into the cisternae magna of NOD/SCID mice (Fig. 4H). As anticipated, we observed OPC density-16 dependent suppression of HER2+ LC tumor progression and augmented animal survival ( Fig. 4I, J). Analyses of 17 mouse CSF revealed that TPP1 protein levels also increased with increasing OPC density ( Fig. 4K), accompanied 18 by a corresponding decrease in GM-CSF levels (Fig. 4K). Interestingly, the OPCs co-implanted with Lepto1 cells 19 did not exhibit altered viability between Days 5 and 20 and expressed appropriate cell-specific markers (Suppl. 20 The Lepto lines derived from nodular HER2+ LC tissues mimic the metastatic organotropism of HER2+ 1 LC seen in patients (Fig. 1). Thus, we investigated whether reagents selectively targeting the Lepto lines would 2 antagonize HER2+ LC in patients. To do so, we performed a chemical genetics screen using Lepto1 cells to assess 3 LOPAC-1280 compounds (Fig. 5A). Specifically, Lepto1 cells were seeded in 384-well plates at a density of 4 1,000 cells per well, and one day later, LOPAC-1280 compounds were added at 4 concentrations (100 nM, 200 5 nM, 400 nM, and 800nM in 0.1% DMSO). After 72 h, we analyzed cell viability using a CellTiter-Glo cell assay, 6 which indicated that both a pan-Aurora kinase inhibitor (CCT137690) and an SRC kinase inhibitor (Bosutinib) 7 inhibited Lepto1 cell viability by ~95% (Fig. 5B). Dose-response assays showed that Lepto1 lines were sensitive 8 to CCT137690 and Bosutinib at all concentrations tested, but the IC50 value of CCT137690 was below 100 nM 9 ( Fig. 5C), significantly lower than of Bosutinib. Thus, additional studies focused primarily on the effects of 10 CCT137690. Treatment with CCT137690 (100 nM) significantly induced apoptosis in Lepto1 cells (Fig. 5D). 11 CCT137690 inhibits Aurora-A, B, and C kinases. Aurora-A participates in crosstalk between SRC kinase 12 and GM-CSF signaling to regulate effectors such as STAT5, AKT, and mTOR, all of which play critical roles in 13 Lepto1 cell proliferation and viability ( Fig. 5E)51, 52, 53. Furthermore, we observed that Aurora-A levels were 14 particularly elevated in nodular HER2+ LC tissues relative to other primary tumors or normal breast or brain 15 tissues ( Fig. 5F). In addition, western blot analysis of Lepto1 cells showed that combining anti-GM-CSF 16 antibodies with either CCT137690 or Bosutinib suppressed the phosphorylation-dependent activation of GM-17 CSF/GM-CSFRα effectors, including STAT5, AKT, and mTOR, thus reducing oncogenic autocrine GM-CSF 18 signaling ( Fig. 5G) and inducing Lepto1 cell apoptosis (Suppl. Fig. 4E). 19 Next, we investigated whether treatment with TPP1 or a combination of anti-GM-CSF neutralizing 20 antibodies plus CCT137690 would antagonize Lepto1 cell viability or cancer progression. To do so, we treated 21 Lepto1 cells for 24 h with DMSO (control), TPP1, anti-GM-CSF neutralizing antibodies, CCT137690, or 22 antibodies+CCT137690 and then cultured them in conditions favoring tumorsphere formation. After 7 days, 23 control cells developed numerous, round tumorspheres, whereas cells treated with TPP1, anti-GM-CSF 24 antibodies, CCT137690, or antibodies+CCT137690 formed fewer and smaller tumorspheres (Suppl. Fig. 4A). 25 CCK assays showed that every treatment also reduced the proportion of sphere-initiating cells (Suppl. Fig. 4B). 26 However, combination treatment with anti-GM-CSF antibodies+CCT137690 reduced the proportion of live, 1 tumorsphere-initiating cells by ~80% relative to DMSO treatment (Suppl. Fig. 4B), an effect that was 2 significantly greater than single drug or antibody treatment alone. To assess the effects of these treatments on 3 Lepto tumor growth, we treated Lepto1 tumorspheres with DMSO (control), TPP1, anti-GM-CSF antibodies, 4 CCT137690, or anti-GM-CSF+CCT137690 for 2 days. As expected, combination treatment significantly reduced 5 primary tumorsphere size and promoted tumorsphere dissociation (Suppl. Fig. 4A). CCK assays confirmed that 6 every treatment, and most significantly the combination treatment, reduced primary tumorsphere cell viability 7 (Suppl. Fig. 4C). 8 To examine the effects of each treatment on mechanisms associated with relapse, we developed secondary 9 tumorspheres by culturing surviving primary tumorsphere cells in standard stem cell medium without drug 10 treatment for 12 days. Surviving cells from primary tumorspheres treated with DMSO for 2 days were dissociated 11 and allowed to form tumorspheres, these surviving cells formed secondary tumorspheres similar to that of the 12 primary tumorspheres (Suppl. Fig. 4A). In contrast, surviving cells from primary tumorspheres treated for 2 days 13 with TPP1, anti-GM-CSF antibodies, CCT137690, or antibodies+CCT137690 developed fewer, more irregularly-14 shaped secondary tumorspheres (Suppl. Fig. 4A). The number of viable cells was also significantly lower in 15 secondary tumorspheres derived from primary tumorspheres treated with anti-GM-CSF antibodies+CCT137690 16 compared to those derived from control or single-treatment primary tumorspheres (Suppl. Fig. 4D), suggesting 17 that the combination treatment efficiently suppresses tumor regrowth. We next treated Lepto1 cells in vitro with 18 anti-GM-CSF antibodies plus CCT137690 or Bosutinib and, in both cases, observed strong suppression of the 19 phosphorylation-mediated activation of AKT, STAT5, and mTOR ( Fig. 5G). Combination treatment with 20 CCT137690 also significantly induced Lepto1 cell apoptosis (Suppl. Fig. 4E). Because combination treatment 21 with antibodies+CCT137690 most potently suppressed Lepto1 cell viability and tumorsphere initiation, growth, 22 and relapse among all of the treatments, we evaluated the effects of this combination on HER2+ LC growth in 23 vivo. To do so, we administered anti-GM-CSF antibodies+CCT137690 on Days 5, 10, and 15 after implantating 24 Lepto1 cells into NOD/SCID mice (Fig. 5H). Compared to vehicle control treatment, this combination treatment 25 significantly reduced tumor volume over time, as indicated by BLI analysis (Fig. 5I), and augmented survival 1 (Fig. 5I). CNS metastases from breast cancer may involve the parenchymal brain or leptomeninges39, 54, 55, 56. 2 HER2+ breast cancer is the most common solid tumor origin of leptomeningeal metastasis57, 58, 59, 60, 61, 62, 63, 64. 3 Once the tumor cells reach the leptomeninges, they may spread via the CSF65. HER2+ LC is not always 4 discernible disease based on magnetic resonance imaging (MRI), so diagnoses are made via positive cytology of 5 aspirated CSF samples. However, in some cases, adherent nodular deposits may develop on the surface of the 6 brain, spinal cord, and spinal roots, allowing diagnosis based on MRI alone14, 22, 65, 66. The presence of nodular 7 deposits is associated with the greatest suffering from headaches and intractable pain due to cranial and spinal 8 nerve invation22, 65, 66.Therefore, there is an urgent need to develop therapies that effectively target HER2+ nodules 9 in the leptomeninges. So, better understanding of the molecular mechanisms governing metastatic organotropism 10 is needed to improve precision medicine for metastatic disease. 11 Alternatively, various research groups have demonstrated that approximately 84% of breast cancers 12 contain at least one genomic alteration that can be potentially exploited as a treatment target67. Indeed, genetic 13 screens have identified promising therapeutic targets in breast cancer68. However, only a few, including 14 phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha (PIK3CA), AKT1, and ERBB2 8, 68, have 15 been validated in clinical studies, suggesting that the success rate of these strategies is poor. 16 Despite the significant progress made over the past decade based on a combination of clinical profiling 17 data and experimental models, our understanding of HER2+ LC remains limited, partly due to the absence of 18 bonafide HER2+ LC in vitro and/or in vivo models. Furthermore, HER2+ LC research has been mostly based on 19 floating HER2+ LC cells removed from the CSF in numbers insufficient to generate cell lines. Interestingly these 20 HER2+ LC floating cells differ phenotypically from nodular tissue, which represents the major clinical burden of 21 HER2+ LC. Therefore, one of the major limitations in the development of effective therapies for HER2+ LC is 22 the lack of bonafide in vitro and in vivo HER2+ LC models that recapitulate the human disease. 23 So, to address this challenge, we developed expandable, low-passage HER2+ LC lines (Lepto1 and 24 Lepto2) using nodular HER2+ LC tumor tissues isolated from patients. In vivo analysis of our HER2+ LC PDX 25 mouse models showed that Lepto cells colonized the leptomeningeal surface of the brain, brain stem, and spinal 26 cord, mimicking the spatial distribution of nodular HER2+ LC tumors in patients. Furthermore, we observed that 1 the white matter of NOD/SCID mice was relatively tumor-resistant when Lepto1 cells were implanted directly 2 into the brain stem, exhibiting <5% tumor cell graft efficiency compared to 95% in gray matter. These 3 observations indicate that our nodular HER2+ LC-derived Lepto lines and PDX models recapitulate the human 4 disease and suggest that white matter is less hospitable to HER2+ LC metastatic colonization. Interestingly, the 5 Lepto cells extracted from our PDX models remained transcriptomically similar to pre-implanted, low-passage 6 Lepto cells, indicating minimal or no genetic drift between pre-and post-implanted Lepto cells. 7 We also showed that OPCs, which are found primarily in the white matter, inhibit HER2+ LC cell viability 8 in vitro and in vivo. Immunoblotting mediated quantification of various cytokines released into the culture media 9 by Lepto1 cells in the presence or absence of OPCs, we identified GM-CSF to acts as an autocrine driver of Lepto 10 cell growth in vitro and in vivo. We also demonstrated that conditional overexpression of GM-CSF in Lepto1 11 cells partially rescued them from OPC-induced apoptosis and suppression of viability in primary Lepto1 cells. In 12 addition, LC-MS/MS analyses of media from Lepto1 cells cultured individually or co-cultured with either OPCs 13 or astrocytes identified TPP1 to be secreted from OPCs and that treatment with either OPC-derived or exogeneous 14 TPP1 abrogated GM-CSF-mediated signaling. 15 TPP1 is a lysosomal serine protease that also serves as a non-specific lysosomal peptidase69, 70. Although 16 TPP1 deficiency is associated with various fatal neurodegenerative diseases71, 72, 73, 74, 75, such as neuronal ceroid 17 lipofuscinoses, its role in inhibiting HER2+ LC tumor development in the white matter remains unexplored. Our 18 studies demonstrate that OPC-secreted TPP1 plays inhibit GM-CSF-induced HER2+ LC tumor growth. We were 19 able to also show that extracellular levels of GM-CSF drop significantly when Lepto1 cells are incubated with 20 either OPCs or recombinant TPP1. ELISA based quantification of GM-CSF in the CSF (derived from variously 21 treated PDX mice models) or in the Lepto1 culture media indicate OPC secreted TPP1 degrades GM-CSF, 22 suggesting that intrathecal administration of recombinant TPP1 may be a viable therapeutic option to target 23 HER2+ LC growth. 24 In addition, found that the administration of anti-GM-CSF neutralizing antibodies also abrogates GM-25 CSF-mediated signaling and significantly impairs HER2+ LC development in vivo. To identify other potential 26 drugs to suppress GM-CSF signaling, we performed a chemical genetics screen using a LOPAC-1280 library and 1 identified two antagonists of Lepto1 cell viability: CCT137690, a highly selective pan-Aurora kinase inhibitor, 2 and Bosutinib, an SRC tyrosine kinase inhibitor. Either drug, when combined with anti-GM-CSF neutralizing 3 antibodies, significantly inhibited the activation of downstream GM-CSF pathway effectors by suppressing their 4 phosphorylation. Subsequent studies revealed CCT137690 to be more potent than Bosutinib, with IC50 values less 5 than 100 nM; hence, our further analyses focused primarily on CCT137690. Interestingly, Aurora kinases regulate 6 key mitotic activities, including centrosome maturation, spindle assembly, chromosome segregation, and 7 cytokinesis76, 77, 78, 79, 80, 81. Indeed, Aurora kinase inhibitors have been extensively studied as novel anti-mitotic 8 drug targets82,83 and the overexpression of Aurora-A kinase has been observed in HER2+ LC tissues. 9 We demonstrated that CCT137690 and anti-GM-CSF neutralizing antibodies synergize not only to 10 inactivate GM-CSF effectors but also strongly inhibit primary and secondary Lepto tumorsphere initiation, 11 growth, and relapse. Moreover, in PDX mouse models, combination treatment with CCT137690 and anti-GM-12 CSF neutralizing antibodies inhibited Lepto1 tumor growth and augmented overall animal survival more potently 13 than single drug, TPP1, or anti-GM-CSF antibody treatments. These findings suggest that comparable strategies 14 could be used to target HER2+ LC tumors in patients, and future research is warranted to establish the optimal 15 schedule of intrathecal chemotherapy with CCT137690 and anti-GM-CSF therapy. It is still unclear if the 16 administration of more than one drug is superior to monotherapy; however, some reports suggest that 17 polychemotherapy regimens are superior. Therefore, optimization of TPP1, CCT137690 and/or anti-GM-CSF 18 treatments in clinical trials might be needed for application in the clinic. 19 In summary, we developed novel primary HER2+ LC cell lines that can be expanded for multiple passages 20 in vitro and in vivo. Critically, these cell lines demonstrate metastatic colonization of leptomeninges in vivo and 21 are thus useful for establishing reliable HER2+ LC PDX models via injection into the cisterna magna. Using these 22 cell lines, we determined that GM-CSF signaling is critical for HER2+ LC growth and that OPC-secreted TPP1 23 abrogates GM-CSF signaling, thus underlying the spatially heterogeneous distribution of HER2+ LC tumors in 24 our PDX models and in patients. In addition, we found that abrogating signaling downstream of GM-CSF via the 25 administration of TPP1, anti-GM-CSF neutralizing antibodies, or the selective pan-Aurora kinase inhibitor 26 CCT137690 significantly decreased HER2+ LC progression in vivo and in vitro. Overall, our identified novel 1 neural niche specific crosstalk between HER2+ LC tumors and OPCs residing predominantly in the white matter. 2 We were also able to identify GM-CSF as an autocrine oncogenic driver of HER2+ LC growth in vivo via 3 induction of downstream GM-CSF signaling. Lastly, intrathecal administration of TPP1, CCT137690 and/or anti-4 GM-CSF antibodies may be an excellent strategy against HER2+ LC in the clinic. 5 6 Methods and Materials 1

Ethics statements 2
The use of human specimens was approved by the City of Hope (COH) Institutional Review Board (IRB; 3 protocols #07047 and #16015) 84, 85, 86. Written informed consent was obtained from all the patients under protocol 4 #07047, and the study was conducted in accordance with the Declaration of Helsinki, institutional guidelines, and 5 all local, state, and federal regulations. All mouse studies were approved by the COH Institutional Animal Care 6 and Use Committee (protocol #10044). 7 shunts (under IRB#16015). These were mainly patients with obstructive or communicating hydrocephalus, and 3 none had infections, evidence of meningitis, multiple sclerosis, or any other known diseases. After collection, the 4 samples were centrifuged at 1,000 g (4°C, 15 min) to remove any cells and debris, filter-sterilized (pore size: 0.45 5 µm), and stored at -80°C until use. To condition hCSF with primary cell lines and astrocytes, cells were grown 6 in hCSF for 2 days, and then the hCSF was collected and purified by centrifugation at 4,000 g (15 min at 4°C). 7 8 FACS sorting, differentiation, and culture of various CNS cell types 9 Human induced pluripotent stem cell (iPSC)-derived multipotent neural progenitor cells (NPCs) were obtained 10 from EMD Millipore (Cat.# SCC035) and propagated using ENStem-A neural expansion medium (Cat.# 11 SCM004). The cells were differentiated to terminal neurons and oligodendrocytes using ENStem-A Neuronal 12 Differentiation Medium (Cat.# SCM017) and human OPC Expansion Media (Cat.# SCM107; basal medium with 13 PDGF-AA, NT3, FGF2, T3, and retinoic acid), respectively, following the supplier's recommendations. NIH-14 approved H9 human embryonic stem cell (hESC)-derived OPCs were also obtained and propagated using the 15 Human Oligocyte Differentiation Kit (Cat.# SCR600). Neurons and oligodendrocytes were differentiated from 16 these hESC-derived OPCs using EMD Millipore's Human OPC Spontaneous Differentiation Complete Media 17 (Cat. #SCM106), according to the supplier's instructions. GFAP+ reactive astrocytes were derived from NPCs87 18 and microglia were differentiated from human iPSCs88, as described previously. The differentiated microglia, 19 oligodendrocytes, neurons, OPCs, and reactive astrocytes were purified by immunostaining with anti-CD45, anti-20 GALC, anti-CD90, anti-NG2, and anti-HepaCAM antibodies, respectively, followed by FACS sorting as 21 described40, and propagated in supplier-recommended media. To recapitulate the in vivo microenvironment, the 22 cells were grown in hCSF for various durations during the in vitro experiments. Cell morphology and 23 differentiation status were monitored using immunofluorescence staining with anti-CD45, anti-GALC, anti-24 CD90, anti-NG2, and anti-HepaCAM antibodies. 25 26 siRNA transfections 1 Gene knockdown was accomplished using pools of four siRNAs targeting the same gene at different sites (Thermo 2 Fisher Scientific; listed in Resource Table 1). siRNA targeting GFP and Luciferase were used as the negative or 3 scramble controls, which should not target any mRNA in the Lepto1 cells. The transfection procedure was 4 performed according to the X-tremeGENE protocol (Sigma-Aldrich).

GM-CSF and TPP1 ELISA 22
The Quantikine Human GM-CSF Immunoassay, a 3.5-or 4.5-h solid-phase ELISA, was used to measure GM-23 CSF levels from Lepto1 cell culture media supernates (Human GM-CSF Quantikine ELISA Kit; R&D Systems). 24 A TPP1 ELISA kit (Ray Biotech) was used to measure TPP1 levels in the cell culture supernates and in mouse 25 CSF. Both ELISA-based quantification kits were used according to the manufacturers' instructions and data were 1 derived from three independent experiments. 2 3

Immunohistochemical and immunofluorescence analysis 4
Mouse brains were fixed in 4% paraformaldehyde, dehydrated in 30% sucrose, and then sectioned (10μm) and 5 mounted on microscopic slides. H&E staining to track tumor growth and immunofluorescence staining of 6 proteins, such as GFAP, Olig2, and MBP, were conducted as previously described84, 85, 86. 7 8 Cell viability and apoptosis assay 9 Cell viability was assessed using a CellTiter-Glo Luminescent Cell Viability Assay kit, according to the 10 manufacturer's protocol. Apoptosis was measured using Annexin V-FITC or staining of phycoerythrin PE-11 conjugated CD326 (EpCAM-PE). Annexin V-FITC binding was analzed by flow cytometry using an FITC signal 12 detector and EpCAM-PE staining was analyzed using a PE emission signal detector. Adherent Lepto1 cells were 13 trypsinized and washed once with FBS-containing media before incubation with Annexin V-FITC or EpCAM-14

Tumorsphere formation studies 17
To investigate the tumor-initiating ability of Lepto cell lines, Lepto1 and Lepto2 cells were treated with DMSO 18 or CCT137690 (100 nM) for 24 h, and 2×104 of the treated cells were seeded in Lepto medium on ultra-low 19 attachment six-well plates (Corning). After 7 days, sphere-initiating cells were detected using the Cell Counting 20 Kit-8 (CCK) assay, according to the manufacturer's protocol. To examine the effects of CCT137690 (100 nM), 21 TPP1 (100 ng/mL), and anti-GM-CSF antibodies (0.1-1μg/mL) on tumor growth, primary tumorspheres were 22 cultured for 5 days and then treated for 2 days before their viability was measured using the CCK assay. DMSO 23 was used as a negative control. To examine the effects of each treatment on tumor recurrence, primary 24 tumorspheres were trypzinized and re-seeded without drug treatment to produce secondary tumorspheres. After 25 12 days of treatment, secondary tumorsphere viability was measured using the CCK assay. 26 1