Complement Component 3 Adapts the Cerebrospinal Fluid for Leptomeningeal Metastasis

Abstract

We molecularly dissected leptomeningeal metastasis, or spread of cancer to the cerebrospinal fluid (CSF), a frequent and fatal condition mediated by unknown mechanisms. We selected lung and breast cancer cell lines for the ability to infiltrate and grow in the CSF, a remarkably acellular, mitogen-poor metastasis microenvironment. Complement component 3 (C3) was upregulated in four leptomeningeal metastatic models and proved necessary for cancer growth within the leptomeningeal space. In human disease, cancer cells within the CSF produced C3 in correlation with clinical course. C3 expression in primary tumors was predictive of leptomeningeal relapse. Mechanistically, we found that cancer cell-derived C3 activates the C3a receptor in the choroid plexus epithelium to disrupt the blood-CSF barrier. This effect allows plasma components including amphiregulin and other mitogens to enter the CSF and promote cancer cell growth. Pharmacologic interference with C3 signaling proved therapeutically beneficial to suppress leptomeningeal metastasis in these preclinical models.

INTRODUCTION

The leptomeninges surround the brain and spinal cord and contain the cerebrospinal fluid (CSF). Cancer spread into the CSF compartment, or leptomeningeal metastasis, presents a formidable clinical challenge. Metastases to this fluid-filled space disseminate rapidly over the entirety of the central nervous system, settling on and invading into the brain, spinal cord, cranial and spinal nerves, resulting in rapid neurologic disability and death. Untreated, patients succumb to leptomeningeal tumor burden 6–8 weeks after diagnosis (Kesari and Batchelor, 2003; Scott and Kesari, 2013); current treatments offer little improvement on this grim prognosis (DeAngelis and Boutros, 2005; Waki et al., 2009). Although any systemic cancer may seed the leptomeninges, the majority of leptomeningeal metastases from solid malignancy arise from primary breast and lung cancers (Waki et al., 2009). Approximately 5–10% of patients with solid tumors harbor leptomeningeal metastasis, and this number is expected to rise (Brower et al., 2016; Clarke et al., 2010). The molecular basis of this morbid, increasingly prevalent complication of cancer remains unknown.

Cancer cells may access the leptomeningeal space through a variety of routes: Cells within the venous circulation may access the space through Bateson’s plexus (Glover et al., 2014), cells adjacent to spinal and cranial nerves may directly invade (Kokkoris, 1983), and cells within the brain parenchymal may broach the glia limitans (Boyle et al., 1980). Cancer cells within the arterial circulation pass through the choroid plexus to enter the CSF (Kokkoris, 1983). The choroid plexus is a polarized secretory epithelium that resides within the ventricles, secretes CSF and restricts entry of cells and plasma components into the leptomeningeal space. Within the leptomeningeal space, metastatic cancer cells may circulate freely. Outgrowth occurs in suspension, as well as in contact with the pia matter, a thin mesenchymal tissue layer that coats the neuro-axis, including the spinal cord and roots. Once established, leptomeningeal metastases may invade the parenchyma to produce focal neurologic damage (Figure 1A).

Figure 1. Leptomeninges and brain parenchyma select for distinct metastatic phenotypes.

(A) Leptomeningeal metastases (purple cells) enter the CSF via the choroid plexus (upper left) and reside within the CSF where they adhere to the pia mater covering the brain and spinal cord (upper right and lower right). Parenchymal metastases (blue cells) enter the parenchyma through the endothelium, surrounding the vasculature and contacting astrocytes (lower right).

(B) Top line: Iterative in vivo selection of leptomeningeal derivative cell lines. 20,000 parental cells are injected into the cisterna magna. After leptomeningeal metastases develop, the mouse is euthanized, cells are collected and grown in culture before injection into a second recipient mouse. This is carried out three times to generate intermediate (Int) cells. 50,000 Int cells are injected intracardially. Mice bearing leptomeningeal metastases are euthanized, cells are collected from the meninges and maintained in culture. These are denoted LeptoM cells. Bottom line: In vivo selection of parenchymal metastatic derivatives. 50,000 parental cells are injected intracardially. Mice with parenchymal metastases are euthanized, brain metastases dissected, dissociated and grown in culture. These cells are injected intracardially once more to generate BrM derivatives.

(C–D) 50,000 MDA231 Parental (green), LeptoM (purple) or BrM (blue) cells are injected intracardially into recipient mice. Tumor growth is monitored by BLI once weekly, neuro-anatomic localization (leptomeningeal, c; parenchymal, d) is determined by histology (inset). Scale bar = 100 μm. n = 7 mice Par, 15 mice LeptoM, 10 mice BrM cells. Refer also to Fig. S1.

(E) Principle component analysis (PCA) plots of transcriptome from BrM (blue), LeptoM (purple) and Int (orange) cell lines after RNASeq. Genes with base mean ≥ 50 fold change ≥ 2 or ≤ 0.5 and p < 0.01 were included for analysis.

(F) Intracranial localization of LeptoM cells after hematogenous dissemination at various time points. 50,000 MDA231-LeptoM cells were injected intracardially on day 0. Three mice were euthanized at each time point. The skull and meninges remained intact throughout tissue processing to maintain architecture. 10 coronal paraffin sections were stained by IHC for GFP; all GFP positive cells were counted on each section as follows choroid plexus (red), leptomeninges (green), parenchyma (blue).

The CSF is acellular and poor in protein, glucose and cytokine content (Spector et al., 2015b). The CSF-filled leptomeningeal space is markedly different as a metastasis microenvironment compared to the parenchyma of other major metastasis organ sites such as the brain, bone marrow, liver, or lungs. The stromal components of these other sites include mesenchymal, immune, epithelial and endothelial cells, extracellular matrix, local and systemic signals, which together provide support for metastatic outgrowth. Much has been learned about the cellular and molecular determinants of metastasis at these sites (Massague and Obenauf, 2016; Quail and Joyce, 2013). By contrast, very little is known about how cancer cells that infiltrate the leptomeningeal space can proliferate in the compositionally simple context of the CSF. We therefore focused our inquiry into mechanisms contributing to cancer cell survival within the CSF.

We postulated that cancer cells capable of thriving in the CSF must possess unique traits that modify the CSF in their favor. We demonstrate here that leptomeningeal metastatic cells grow in the CSF by secreting C3. Acting through its receptor C3aR on the choroid plexus, C3 disrupts the blood-CSF-barrier and allows for passage of select plasma components, including growth factors such as amphiregulin, and others into the CSF. Thus modified, the CSF supports cancer cell growth. Inhibition of this axis genetically or pharmacologically suppresses leptomeningeal metastasis, suggesting a therapeutic opportunity.

RESULTS

Leptomeninges and brain parenchyma select for distinct metastatic phenotypes

To generate mouse models of leptomeningeal metastasis from breast and lung cancers, we employed a two-stage selection strategy, first selecting for cancer cells that were capable of growing within the CSF and then selecting the resulting populations for the ability to reach the leptomeningeal space after hematogenous dissemination (Figure 1B). As sources, we chose cell lines derived from human (MDA231, HCC1954, PC9) or mouse (Lewis Lung Carcinoma, LLC) breast or lung carcinomas (Figure S1A). Cells were engineered to stably express GFP and luciferase and were inoculated directly into the CSF of immunodeficient (MDA231, HCC1954, PC9) or immunocompetent mice (LLC) (Figure S1A). Once tumor growth encompassed the leptomeningeal space, cells were collected from the meninges and recovered in vitro to derive intermediate cell lines. These “Inter” cells were then hematogenously disseminated by inoculation intro the arterial circulation, and allowed to form metastases. Cancer cells that grew in the leptomeningeal space were isolated and termed “LeptoM”. These cells posses the abilities to reach the leptomeninges from the blood circulation and to grow within the CSF environment (Figure 1B).

Metastasis to the anatomically closest site, the brain parenchyma, occurs in a densely cellular stroma dominated by endothelial cells, astrocytes, neurons, microglia, and the blood-brain barrier (Bos et al., 2009; Bowman et al., 2016; Chen et al., 2016; Sevenich et al., 2014; Valiente et al., 2014; Zhang et al., 2015). We compared the neuroanatomic localization of LeptoM derivatives with that of brain parenchymal metastatic derivatives (BrM cells) that were previously isolated from the same parental lines (Bos et al., 2009; Chen et al., 2016; Malladi et al., 2016; Nguyen et al., 2009) (Figure 1B). Parental, LeptoM or BrM cells were hematogenously disseminated into recipient mice; bioluminescent signal was monitored and neuro-anatomic localization of metastases in these animals was assayed by histopathology (Figures 1C–D, S1B–D). In each model system, the LeptoM derivative was aggressively metastatic to the leptomeninges but not to the parenchyma, whereas the BrM derivative displayed the opposite tropism. Analysis of the transcriptome of these cell lines by RNA-Seq demonstrated that the gene expression profiles of the BrM and LeptoM cell lines segregate independently by principal component analysis (Figure 1E). Thus, each LeptoM population was both phenotypically and transcriptomally distinct from its matched BrM population.

Choroid plexus and leptomeningeal infiltration

In order to discern the sequence of events leading to leptomeningeal metastases, we performed histopathological analysis at various time points after hematogenous dissemination of MDA231-LeptoM cells (Figure 1F). The cells immediately populated the choroid plexus, where they maintained a steady presence over the course of the disease. This structure, located within the ventricles, consists of a polarized, secretory epithelium enclosing an arcade of fenestrated capillaries (Spector et al., 2015a). Three days after intracardiac inoculation, cells began to accumulate in the CSF. Once the cancer cells were within the leptomeningeal space, they accumulated rapidly, filling the space and eventually accessing and invading the brain parenchyma through the Virchow-Robin spaces, the CSF-filled areas adjacent to cerebral vessels.

C3 upregulation in leptomeningeal metastasis

To molecularly characterize the mechanisms that underlie these cellular events, we compared the gene expression profiles of the parental cell lines with those of the Inter and LeptoM derivatives to identify differentially expressed genes (Figure S2A). Twenty differentially expressed genes were in common between all four models (Figure S2B). To discriminate between these 20 genes, we further focused on genes that were upregulated at the intermediate stage of selection and maintained in the LeptoM stage, reasoning that these genes would be specifically associated with cell growth within the CSF. This approach uncovered two such genes: lipocalin 2 (LCN2) in 4 of 4 models, and complement component 3 (C3) in 3 of 4 models and upregulated in LeptoM in all 4 models (Figure S2C). Additionally, we performed unbiased gene ontology (GSEA) analysis of upregulated genes within each model, which uncovered a single pathway in common: the Complement and Coagulation cascade (Figure 2A). C3 is a central protein in this cascade (Bajic et al., 2015). Examination of differential gene expression as applied to this pathway predicts C3 generation and suppression of thrombin generation (Figure S2D). Indeed, C3 expression was upregulated in the LeptoM derivative cells in comparison to the parental cells and BrM derivatives in all models at the mRNA and protein level (Figures 2B and S2E).

Figure 2. Upregulation of complement component 3 in leptomeningeal metastasis.

(A) GSEA analysis of upregulated genes in Par vs. Int and Par vs. LeptoM analyses for each model. KEGG Complement and Coagulation Cascade pathway result is shown.

(B) Cancer cell C3 expression by ELISA. Conditioned medium from 1 × 10⁶ cells grown in a 10 cm plate for 48 hours was collected. n = 3 experiments performed in duplicate.

C3 within the CSF may originate from either the carcinoma or the host. We therefore measured C3 in CSF from mice harboring MDA231-LeptoM metastasis using species-specific C3 ELISAs. With this approach, we detected human C3 exclusively in the CSF of mice with leptomeningeal metastasis and not in those with parenchymal brain metastases or extra-cranial metastases (Figure S2F).

C3 expression in human leptomeningeal metastasis

To address the clinical relevance of these findings we determined C3 levels by ELISA of CSF from 69 people with solid tumors and clinical symptoms of CNS metastasis (Table S1). Of these patients, 37 were confirmed to have leptomeningeal metastasis, 20 were diagnosed with parenchymal brain metastasis and 12 were found to have no CNS metastasis. The patients with leptomeningeal metastasis had the highest levels of C3 in the CSF (Figure 3A).

Figure 3. Complement component C3 in human leptomeningeal metastasis.

(A) ELISA of CSF obtained from patients with solid tumor primary and clinical symptoms suspicious for leptomeningeal metastasis. Final clinical diagnosis is indicated. n = 37 patients with leptomeningeal metastasis, 20 patients with parenchymal metastases, 12 patients with no CNS metastasis. Refer also to table S1. *** indicates p < 0.001.

(B) C3 in CSF by ELISA versus clinical disease burden. Refer also to Figure S3C for calculation of Disease Burden Score. n = 30 patients. ** indicates p < 0.05.

(C) C3 in CSF by ELISA versus site of CSF sample. CSF samples were obtained from patients harboring leptomeningeal metastasis, either from lateral ventricle (n = 18) or lumbar cistern (n = 22). *** indicates p < 0.001.

(D) C3 expression by qPCR in patient CSF. CD45-depleted cells from CSF were collected from breast and lung cancer patients (n = 5 patients) with leptomeningeal metastasis. Circulating peripheral leukocytes (n = 3), mammary epithelial cells (n = 2) or bronchial epithelial cells (n = 2) were also collected. mRNA is presented as C3:GAPDH on y-axis; B2MG:GAPDH on the × axis. All samples were analyzed in quadruplicate.

(E) Immunofluorescence of patient-derived CSF cells for CD45 and C3. n = 6 patients. 100 cells counted per patient. Representative images are presented in (F). **** indicates p < 0.0001.

(G) Representative images of primary tumors of patients stained by IHC for cytokeratin and C3.

(H) Proportion of total tumor area (as determined by cytokeratin staining) positive for C3 was correlated with clinical outcome. Refer also to figure S3F. n = 76 patients.

In this first cohort, we noted a wide range of C3 CSF concentrations in patients harboring leptomeningeal metastasis. To address this, we examined the clinical charts of 30 patients with leptomeningeal metastasis and complete CNS staging consisting of gadolinium-enhanced MRI of the brain, total spine and CSF cytologic analysis. The extent of leptomeningeal disease burden was quantified by assessment of the number of neuro-anatomic sites involved on MRI (Figure S3A–B). The lowest levels of C3 detected in the CSF were found in patients harboring LM that was not detected on MRI. We found that CSF C3 levels correlated well with such measures of disease burden (Figure 3B). Because the CSF circulates, traveling from the choroid plexi, around the CNS before resorption by the arachnoid granulations, we reasoned that levels of C3 would be lower in CSF sampled adjacent to the choroid plexus in the lateral ventricles than when sampled further downstream. Indeed, CSF from patients with leptomeningeal metastasis displayed such a pattern (Figure 3C). Importantly, this did not reflect the location of cancer cell growth (Figure S3C).

These indirect measures suggested that cancer cells within the CSF produce complement C3. To demonstrate this, we isolated cancer cells from lung and breast cancer patient CSF by centrifugation followed by CD45 immunodepletion to remove leukocytes. These cells were not expanded in culture prior to analysis. These cells expressed C3 mRNA as measured by quantitative PCR at higher levels than did bronchial and mammary epithelial cells or leukocytes (Figure 3D). C3 expression in CSF cancer cells was confirmed by immunofluorescence (Figure 3E–F). Together, these clinical data support the conclusion that cancer cells within the CSF express complement C3.

We obtained primary tumor samples from 76 patients with breast or lung primary cancers. Elevated C3 expression by IHC in the primary tumor correlated with development of clinically apparent leptomeningeal metastasis (Figure 3G–H). These two groups of patients were followed clinically for 5.2 +/− 1.1 years (low C3) and 4.3 +/− 0.4 years (high C3) after diagnosis of the primary tumor (Figure S3F). C3 expression in primary tumors did not correlate with C3 expression in parenchymal brain metastases in either unmatched (Figure S3G) or matched (Figure S3H) tumor specimens. Collectively, these results indicated a selective and specific association of tumor-derived C3 with metastatic growth in the leptomeningeal space.

C3 promotes cancer growth within the leptomeningeal space

Given the association of C3 expression with clinical and experimental leptomeningeal metastasis, we assessed the necessity of C3 for metastasis to this site. Two independent short hairpin RNAs (shRNAs) that suppressed C3 expression in LeptoM cell lines (Figures S4A–B) strongly reduced the leptomeningeal metastatic activity of MDA231-, PC9- and LLC-LeptoM models (Figure 4A–C, Figure S4C–D). In contrast, metastatic outgrowth of LLC-LeptoM cells in C3 null mice (Wessels et al., 1995) was indistinguishable from growth in C3 wild-type mice (Figure S4E), suggesting that leptomeningeal metastasis is dependent on cancer cell-derived C3, not C3 from the circulation.

Figure 4. C3 is necessary for cancer cell growth within the leptomeninges.

(A–B) C3 expression was stably knocked down using two independent short hairpins (shA and shB) or vector control (Ctl) in MDA231-LeptoM derivative cell lines. 2,000 cells were injected intracisternally, BLI was monitored every 5 days. n = 5 mice per group in each of two independent experiments; 10 mice total per group. Histograms represent in vivo BLI imaging at day 28 post inoculation. ***indicates p < 0.001

(C) Kaplan-Meier plot of overall survival of mice injected with MDA231-LeptoM cells with either vCtl or shA as described above.

(D) 1,000 MDA231-LeptoM cells were seeded in each well of a tissue-culture treated 96-well plate and allowed to grow in CSF from solid tumor patients with or without LM with 50% artificial CSF. Cell growth was monitored by CellTiter Glo assay at t = 1h and 72h. Data represent two independent experiments performed in quadruplicate. *** indicates p < 0.001

(E–F) 500 MDA231-LeptoM cells were seeded into a 384-well plate containing CSF collected from mice harboring no malignancy. Mice were treated with 1 mg/kg recombinant mouse C3a (rmC3a) or PBS I.P. 30 min prior to CSF collection. For mice treated with PBS, rmC3a was added ex vivo to a final concentration of 20 ng/mL to mouse CSF. Data represent two independent experiments performed in quadruplicate. *** indicates p < 0.001

(G) 2,000 Parental cells were introduced into the cisterna magna with either recombinant C3a or vehicle. Additional rmC3a or vehicle was delivered intracisternally every 7 days.

(H) BLI at day 14 of mice inoculated with MDA231 Parental cells, treated as described in (G). * indicates p < 0.01

(I) Kaplan-Meier plot of overall survival of mice in (H). n = 10 mice per group.

We hypothesized that C3 promotes leptomeningeal metastasis by improving cancer cell survival and growth within the CSF. Indeed, both MDA231-LeptoM and PC9-LeptoM cells grew in vitro within CSF from cancer patients with leptomeningeal metastasis but not in CSF from cancer patients without leptomeningeal metastasis (Figures 4D, S4F). Interestingly, CSF obtained from mice treated with parenteral C3 supported LeptoM cell growth in vitro, but direct addition of C3 to mouse CSF ex vivo did not (Figures 4E–F, S4G–H). These results indicated that C3 does not promote cancer cell growth directly, but rather by a non-cell autonomous process whereby C3 acts on surrounding tissue in the leptomeninges.

In the complement cascade, proteolysis of C3 by C3 convertase generates the active molecules C3a, an aphalotoxin, and C3b, an opsin (Merle et al., 2015). Parental cancer cell lines were introduced into the leptomeningeal space with or without exogenous recombinant mouse C3a (Figure 4G). This addition alone supported parental MDA231, PC9 and LLC cell growth within the leptomeninges (Figures 4H–I, S4I–L), identifying C3a as the pro-metastatic derivative of C3.

C3 perturbs choroid plexus barrier function

Briskly perfused at a rate of 4 ml/min per gram, the choroid plexus receives about 10-fold greater blood supply than the brain parenchyma, ensuring adequate CSF production (Spector 2015). The choroid plexus capillaries are separated from the leptomeningeal space by a simple epithelium with tight junctions. This epithelium controls the entry of fluid from blood into the CSF and excludes circulating cells, constituting the blood-CSF-barrier. Choroid plexus epithelial cells (CPEpi cells) share many characteristics with renal tubule cells, including barrier function and selective transport (Spector and Johanson, 2006). C3 has been implicated in loss of renal tubule barrier function in proteinuria (Bao et al., 2011; Floege and Amann, 2016; Ogrodowski et al., 1991). Notably, we found that CPEpi cells express the C3 receptor C3aR, a G-protein coupled receptor (Ames et al., 1996), on both the luminal and basolateral surfaces as demonstrated by immunofluorescence (Figure 5A). We therefore hypothesized that cancer cell-derived C3 plays a barrier disruptive role in this pathophysiological context.

Figure 5. C3 perturbs choroid plexus barrier function.

(A) Immunofluorescence of C57/Bl6 mouse choroid plexus for C3aR (red) or SPAK (green), a marker of the luminal CPEpi surface. Scale bar represents 100 μm.

(B) Human CPEpi cells are grown on poly-L-Lysine coated transwells until TEER is greater than 200 Ω/cm². Media was changed to conditioned media (CM) as indicated and TEER measurements were obtained every 12 hours. Two independent experiments performed in triplicate are averaged.

(C) Conditioned media from MDA231 parental cells supplemented with recombinant mouse C3a at 5 ng/mL (rmC3a), open circles or equal volume PBS, closed circles.

(D) Conditioned media from MDA231-LeptoM derivative cells immunodepleted with anti-C3 (open circles), vehicle (closed circles) or isotype control (grey circles).

(E–G) C57/Bl6 mice (C3aR^+/+ or C3aR^−/−) without malignancy were treated with 1 mg/mL rmC3a I.P. or vehicle alone prior to intracardiac introduction of mixed dextrans (Cascade Blue-10 kDa, FITC-40 kDa, Texas Red-500 kDa 1:1:1 in PBS). CSF and peripheral blood sampling occurred 30 min later. n = 5 mice, each fluorescence measurement performed in triplicate. *** p < 0.001; **** p < 0.0001; NS = not significant

(H–I) Choroid plexus from the lateral ventricles of C57/Bl6 mice were treated 37°C with conditioned media from LLC Parental cells supplemented with 10 ng/mL rmC3a or vehicle alone. After 0, 30 or 120 minutes at 37C, specimens were lysed for western blotting (H). After three hours at 37C, specimens were either fixed for IF (G) against ZO-1 (red) and claudin (green) nuclei are marked with DAPI (blue).

(J–K) Migration of 20,000 serum-starved Par cells, LeptoM control short hairpin (vCtl) or C3 knock down short hairpin (shA) across a confluent monolayer of human CPEpi cells toward artificial CSF + 1% FBS. Cells were allowed to migrate 12 hours before quantification of cells. Two independent experiments were performed in duplicate. Ten high-power fields (40×) were counted per condition per replicate. NS = not significant.

Tight junctions preclude efficient passage of ions and therefore create electrical resistance across the epithelial monolayer. To determine tight junction integrity in CPEpi cells, we assayed trans-epithelial electrical resistance (TEER) across monolayers of human CPEpi cells grown on a permeable membrane (Matter and Balda, 2003). Once electrical resistance measured with a voltmeter indicated good barrier function, medium in the upper compartment was exchanged for media conditioned by test cells (Figure 5B). Conditioned media from MDA231-LeptoM and LLC-LeptoM cells increased the electrical conductance across the CPEpi barrier, whereas conditioned media from matched parental cell lines did not (Figures 5C–D, S5A–B). Immunodepletion of C3a from MDA231-LeptoM or LLC-LeptoM conditioned media abolished this activity, whereas addition of recombinant C3a to conditioned media from parental cells increased electrical conductance (Figures 5C–D, S5A–B). These results indicate that C3a can reduce the barrier function of CPEpi cell monolayers.

We next assessed the effect of C3a on blood-CSF barrier function and the molecular size exclusion of this barrier in vivo. Fluorescently conjugated dextrans of various molecular sizes were introduced into the systemic circulation of mice pretreated with intraperitoneal administration of recombinant C3a (Figure 5E). CSF sampling 30 min later demonstrated that dextrans of up to 40 kDa were present in the CSF and their level was increased by the treatment with C3a (Figure 5C); this effect was dependent upon the presence of C3aR (Figure 5G).

C3 disrupts choroid plexus tight junctions and the blood-CSF barrier

Choroid plexus epithelial tight junctions are regulated by Gαi and protein kinase C signaling (Lippoldt et al., 2000). In vitro, passage of 40 kDa fluorescently-conjugated dextran across human CPEpi cell monolayers was inhibited by addition of pertussis toxin, an inhibitor of Gαi, and induced by phorbol ester, an activator of protein kinase C (Figure S5C–D). These effects were consistent with the involvement of C3aR G-protein coupled receptors (Ames et al., 1996; Lippoldt et al., 2000). C3a addition transiently increased myosin light chain kinase (MLCK) phosphorylation, a key regulator of tight junction permeability (Cunningham and Turner, 2012) (Figure 5H). The effect of C3a on tight junction disorganization was evident by immunofluorescence of whole-mount mouse choroid plexus after addition of C3a (Figure 5I). These results with choroid plexus epithelia are in line with previously observed effect of C3a in renal and pulmonary polarized epithelia (Bao et al., 2011; Conyers et al., 1990; Drouin et al., 2001).

Disruption of choroid plexus tight junctions led us to investigate the role of C3 in cancer cell passage across the choroid plexus epithelium. Although LeptoM derivative cells traversed the B-CSF-B more readily than their parental counterparts in vitro, LeptoM cancer cell migration across CPEpi cell monolayers was not altered by C3 knockdown in the cancer cells (Figures 5J–K and S5E), suggesting that C3 was not rate limiting for infiltration of circulating cancer cells into the CSF in these models. Importantly, in vivo, cancer cells entered the CSF space equivalently in C3aR^+/+ and C3aR^−/− host mice (Figure S5F), again supporting the conclusion that C3a-C3aR signaling does not mediate cancer cell entry into the CSF; alternative gene expression likely supports this activity. Indeed, LeptoM cells expressed various known mediators of cancer cell extravasation some of which were upregulated relative to the parental cells. These mediators include: cyclooxygenase 2/PTGS2 (Gupta et al., 2007), angiopoietin-like 4 (Padua et al., 2008), CCL2 (Qian et al., 2011), HB-EGF (Bos et al., 2009), MMP1 (Gupta et al., 2007) and VEGF (Lee et al., 2003) (Table S2).

Cancer cell C3 alters the CSF composition to promote tumor growth

To determine the relevance of host tissue C3 signaling in cancer cell outgrowth within the leptomeningeal space, we compared the growth of LLC-LeptoM cells that were inoculated into the CSF of wild-type mice or C3aR^−/− mice (Humbles et al., 2000). The C3aR deficient mice showed a very reduced ability to support the growth of the LeptoM cells in the leptomeningeal space (Figure 6A), arguing that growth of these cells depends not only on their ability to produce C3 but also on host C3aR. None of the LeptoM derivatives expressed C3aR (Figure S6A).

Figure 6. C3a mediates amphiregulin influx to adapt CSF for cancer cell growth.

(A) 2,000 LLC-LeptoM cells were introduced intracisternally into C3aR−/− or C3aR +/+ mice in C57/Bl6 background. n = 5 −/− mice, 10 +/+ mice. * p < 0.05

(B) ELISA for human amphiregulin of CSF obtained from patients with solid tumor primary and clinical symptoms suspicious for leptomeningeal metastasis. Final clinical diagnosis is indicated. n = 11 patients with leptomeningeal metastasis, 11 patients with no CNS metastasis. *** p < 0.001

(C) 1,000 MDA-LeptoM cells were seeded in each well of a tissue-culture treated 96-well plate and allowed to grow in artificial CSF supplemented with recombinant amphiregulin. Cell growth was monitored by CellTiter-Glo assay at 72 h. Data represent two independent experiments performed in quadruplicate. *** p < 0.001

(D) 1,000 PC9-LeptoM cells were seeded in each well of a tissue-culture treated 96-well plate and allowed to grow in artificial CSF supplemented with recombinant amphiregulin. Cell growth was monitored by CellTiter-Glo assay at the indicated time points. Data represent two independent experiments performed in quadruplicate.

(E) 2,000 MDA Par cells were introduced into the cisterna magna with either recombinant amphiregulin or vehicle. Additional amphiregulin or vehicle was delivered intracisternally every 7 days. BLI on day 14 is illustrated. * p < 0.05; *** p < 0.001

(F) Naïve C57/Bl6 mice were treated with either 10 ng/mL rmC3a or vehicle prior to CSF and blood sampling. n = 3 mice in each treatment group. Amphiregulin was assayed by ELISA in serum and CSF from each mouse in triplicate. Data are presented as ratio of CSF Amphiregulin : Serum Amphiregulin; * p < 0.05

(G) Patients with leptomeningeal metastasis and intention to treat with intra-ventricular trastuzumab have an intra-ventricular (Ommaya) reservoir placed in the lateral ventricles. CSF adjacent to the choroid plexus is sampled just prior to treatment and analyzed for C2, C3, C4, Factor D and amphiregulin (AR) by ELISA. T1 post-contrast MRI from Patient 1 demonstrating clinical response is shown at right. Deposits of leptomeningeal metastasis appear as white plaques over the cervical spine and are indicated by red arrowheads.

(H) Analysis of CSF from Patient 1 for indicated analytes at three time points.

(I) Cancer cells within the cerebrospinal fluid produce C3. C3a binds to and activates C3aR on the choroid plexus leading to loosening of tight junctions and impaired barrier function. Select plasma components, including amphiregulin, gain entry to the cerebrospinal fluid where they support cancer cell growth.

We investigated whether cancer cell-derived C3 acts on the choroid plexus to disrupt the blood-CSF barrier to allow the influx of growth promoting components into the CSF. We compared the cytokine composition of the CSF before and after a diagnosis of leptomeningeal metastasis in six patients, using a commercially available dot blot array. This analysis uncovered multiple proteins whose levels increased as leptomeningeal metastasis progressed (Figure S6B). Each patient’s CSF displayed a distinct cytokine profile, reflecting patient-to-patient variability. All patients analyzed showed a three-fold or higher increase in the levels of at least 4 analytes (range 4–13) (Figure S6B). The most salient increases were observed in the growth factors amphiregulin, glial-derived neurotrophic factor (GDNF), persephin, artemin, platelet-derived growth factor (PDGF), and fibroblast growth factor-4.

To test the hypothesis that these mitogens supported growth of cancer cells in the CSF, we focused on amphiregulin. Amphiregulin is an EGF/TGF-α family member produced predominantly in the liver (Berasain and Avila, 2014) and present in blood at a normal level of approximately 20 pg/mL (Lemos-Gonzalez et al., 2007). Elevated levels of blood amphiregulin occur with various types of carcinoma (Addison et al., 2010). Using ELISA, we detected amphiregulin at levels ranging between 2.7–1950 pg/mL in the CSF of 9 of 11 patients with leptomeningeal metastasis but only in 1 case (11.2 pg/mL) out of 12 patients without leptomeningeal metastasis (Figure 6B). In vitro, supplementation of a mitogen-poor CSF-like media with amphiregulin stimulated the growth of PC9-LeptoM and MDA231-LeptoM cells (Figure 6C–D). In vivo, inoculation of amphiregulin into the CSF enhanced the growth of parental MDA231 cells within the leptomeningeal space (Figure 6E).

Amphiregulin was not produced by the LeptoM cancer cell lines themselves (Figure S6C) nor by mouse choroid plexus as compared to liver (Figure S6D). However, systemic treatment with C3a significantly increased amphiregulin concentration within the CSF (Figure 6F). Consistent with CSF flow dynamics, concentrations of amphiregulin were elevated in CSF sampled from the lumbar cistern when compared with CSF sampled from the lateral ventricles of patients with leptomeningeal metastasis (Figure S6E).

To synthesize these findings and apply them to a relevant clinical context, we analyzed both clinical data and cerebrospinal fluid from patients with leptomeningeal metastasis undergoing intra-ventricular treatment with trastuzumab (Herceptin). CSF collection from the Ommaya reservoir allows for sampling of the CSF adjacent to the choroid plexus (Figure 6G). Analysis of clinical cytology, components of the classical and alternative complement cascades, complement C3 and amphiregulin were undertaken at each of three time points (Figures 6H and S6F–G). As can be appreciated by the index case (Figure 6G–H), C3 and amphiregulin levels correlated with clinical course. In other patients, both responsive (Figure S6F) and non-responsive (Figure S6G) to the treatment, C3 consistently correlated with clinical course. CSF amphiregulin was elevated in some of these patients, but not in others, consistent with the variability observed in other case series (Figure S6B). In patients with elevated CSF amphiregulin, response to treatment was associated with a drop in CSF amphiregulin concentration. Taken together, these observations support a model whereby cancer cell-derived C3a acting through C3aR on the choroid plexus epithelium reduces the barrier function of these cells and allows for passage of growth factors such as amphiregulin and other components from the circulation into the CSF to support cancer cell growth (Figure 6I).

Targeting C3aR inhibits leptomeningeal metastasis

To complement the molecular and genetic evidence for this model with pharmacologic evidence, and to address the possible utility of targeting the C3aR as an anti-leptomeningeal metastasis therapy, we employed a commercially available non-peptide C3aR agonist (Benzeneacetamide,α-cyclohexyl-N-[1-[1-oxo-3-(3-pyridinyl)propyl]-4-piperidinyl]-, a-cyclohexyl-N-[1-[1-oxo-3-(3-pyridinyl)propyl]-4-piperidinyl]-benzeneacetamide) and a C3aR antagonist (SB 290157) (Ames et al., 2001). This C3aR antagonist has been employed in studies of reactive airway disease (Mizutani et al., 2009), autoimmune arthritis (Hutamekalin et al., 2010) and proteinuric nephropathy (Tang et al., 2009). In the MDA231 model, a once weekly parenteral treatment with the C3aR antagonist inhibited leptomeningeal metastasis, and provided a survival benefit to the mice (Figure 7A–B). In contrast, parenteral treatment with C3aR agonist promoted leptomeningeal metastasis (Figure 7A), and hastened disease progression (Figure 7B). The treatment with C3aR antagonist was effective at suppressing leptomeningeal metastasis from breast cancer and lung cancer in all four models (Figure 7C).

Figure 7. C3aR as a therapeutic target in leptomeningeal metastasis.

(A) 2,000 MDA231-LeptoM cells were introduced into the cisterna magna on day 0. Mice were treated with 10 mg/kg C3aR agonist (Ag), 10 mg/kg antagonist (Ant) or vehicle (Veh) I.P. twice weekly, tumor cell growth was monitored by BLI. n = 10 mice per group. **** p < 0.0001

(B) Survival analysis of mice treated in (A).

(C) 2,000 MDA231-LeptoM, PC9-LeptoM, HCC1954-LeptoM or LCC-LeptoM cells were introduced into the CSF on day 0 and treated with Veh or Ant as described in (A). n = 10 mice per group. BLI on day 14 is illustrated. * p < 0.05; ** p < 0.01; *** p < 0.001.

DISCUSSION

The relative immunological privilege and protection from circulating chemotherapy of the leptomeningeal space offers seemingly safe harbor for invading cancer cells. However, these same characteristics pose a microenvironmental challenge for the cancer cell: how to procure the necessary growth factors and nutrients for growth in the otherwise inhospitable CSF. We have uncovered a cancer cell-host interaction whereby the malignant cell overcomes an epithelial barrier to actively enrich the CSF with plasma-derived components.

Adaptating a barren metastasis microenvironment

As a metastatic microenvironment, the CSF is markedly different from the microenvironment of other major site of metastasis, such as the bone marrow, lung, liver or brain parenchyma. After infiltrating these other sites, metastatic cells meet a rich environment with many different cell types, extracellular matrix structures, perivascular nutrients, and intercellular signals. Immune surveillance in these host tissues eliminates a majority of the infiltrated cells and selects for cancer cells that evade immunity by entering quiescence or by other means (Malladi et al., 2016). Under the selective pressure of the microenvironment, metastasis initiating cells develop traits for the cooption of stromal components including osteoclasts in the bone marrow (Kakonen et al., 2002; Kang et al., 2003; Waning and Guise, 2014), macrophages, monocytes and neutrophils in the lungs (Acharyya et al., 2012; Chen et al., 2011; Qian et al., 2011; Wculek and Malanchi, 2015), and astrocytes in the brain parenchyma (Chen et al., 2016; Schwartz et al., 2016; Zhang et al., 2015). In sharp contrast to this paradigm, metastatic cells that enter the CSF confront an acellular environment that is relatively poor in nutrients and growth supporting signals. Here, we provide evidence that leptomeningeal metastatic cells produce C3 that disrupts the choroidal blood-CSF barrier to adapt the CSF for cancer cell growth.

The role of complement in permeabilization of epithelial barriers is not unprecedented. It was previously described in renal and pulmonary epithelia (Conyers et al., 1990; Ricklin et al., 2016). C3a may also permeabilize these epithelia in the context of metastasis. The present work expands the range of C3 epithelial targets to include the choroid plexus, at least in the context of metastasis. As proof of principle, we show that this pathway allows entry of the EGFR ligand amphiregulin into the CSF. Given the diversity of CSF composition in leptomeningeal metastasis, clearly amphiregulin is one of many other cytokines and nutrients that enrich the CSF. Moreover, given the dynamic nature of CSF production and C3a generation, we anticipate that CSF composition will be similarly in flux, and would be unlikely to achieve a steady state. Previous attempts to establish CSF biomarkers of metastasis including myo-inositol (An et al., 2015), β-microglobolin (Svatonova et al., 2014), and others (Walbert and Groves, 2010) have failed to prove useful across a diverse tumor types. Remarkably, we find that C3 is present in the CSF of patients with leptomeningeal metastasis from different types of cancer, and at levels above those of CSF from patients with brain parenchymal metastases or with no CNS metastasis. Given the consistent elevation of C3 in CSF in leptomeningeal metastasis, this molecule may well play roles outside of blood-CSF-barrier disruption to support leptomeningeal metastasis.

Also of interest for future studies is lipocalin 2, whose expression is elevated in all our LeptoM models compared to the matched parental cell lines. Lipocalin 2 is a secreted component of the innate immune system that binds siderophores and is upregulated via the Toll-like receptors for IL-1 and IL-17 in the context of tissue injury and infection. Lipocalin 2 has been implicated in aspects of development, inflammation, tissue regeneration and cancer, but its exact role remains uncertain (Li and Chan, 2011).

Clinical implications

We show that leptomeningeal metastatic cells are phenotypically distinct from brain metastatic cells from the same source. Leptomeningeal metastases are also distinct from brain metastases in clinical presentation. Whereas the focal nature of parenchymal brain metastases allows for local treatment via surgery or radiation, the diffuse quality of leptomeningeal metastases renders focal treatment with radiation palliative at best, and current chemotherapeutic treatments are largely ineffective. The mechanism that we have uncovered brings forward C3 targeting as a novel therapeutic approach for treatment of leptomeningeal metastasis. Pharmacologic intervention in C3 signaling has been explored in other contexts, including reactive airway disease (Khan et al., 2014) and autoimmune arthritis (Hutamekalin et al., 2010). The preclinical results presented here suggest that pharmacologic interference with C3 signaling is therapeutically beneficial to suppress leptomeningeal metastasis.

This work raises a number of intriguing questions for further study. Pharmacologic manipulation of blood-CSF-barrier integrity might also be employed to allow for improved access of systemic chemotherapy into the CSF. Alternatively, maintaining normal barrier function in the setting of leptomeningeal metastasis may prove to be a superior therapeutic approach. These possibilities warrant additional preclinical study.

STAR METHODS

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit anti-human C3	Abcam	ab129945
Mouse anti-human Cytokeratin	Dako	M3515
Alexa Fluor 633 goat anti-mouse IgG	Life Technologies	A-21052
Alexa Fluor 488 goat anti-rabbit IgG	Life Technologies	A-32731
Chicken anti-C3a	Abcam	ab48581
Chicken anti-GFP	Aves	1020
Rabbit anti-β-catenin	Cell Signaling	D10A8
Rabbit anti-claudin-1	Cell Signaling	D5H1D
Mouse anti-MLCK	Abcam	ab55475
Mouse anti-pMLCK	Abcam	Ab200809
Bacterial and Virus Strains
Biological Samples
Cerebrospinal Fluid	MSKCC Brain Tumor Center	Mskcc.org/Brain/Cancer-Care
Primary Tumor Tissues	MSKCC Pathology Core	Mskcc.org/research-advantage/core-facilities/pathology#
Chemicals, Peptides, and Recombinant Proteins
Recombinant Mouse Complement Component C3a protein, carrier-free	R&D Systems	8085-C3-025
Recombinant Human Complement Component C3a protein, carrier-free	R&D Systems	3677-C3-025
Recombinant Human Amphiregulin protein	R&D Systems	262-AR-100
C3aR Antagonist SB 290157	Santa Cruz	Sc-222291
C3aR Agonist	Sigma-Aldrich	C4494
Critical Commercial Assays
Human Complement C3 ELISA Kit	abcam	ab108822
Mouse Complement C3 ELISA Kit	abcam	ab157711
Human Amphiregulin ELISA Kit	abcam	Ab99975
Mouse Amphiregulin ELISA Kit	abcam	Ab100668
Proteome Profiler Human XL Cytokine Array Kit	R&D Systems	ARY022B
Human Complement C2 ELISA Kit	Abcam	Ab154132
Human Complement Factor D ELISA Kit	R&D Systems	DFD00
Deposited Data
Raw and analyzed data	This paper	GEO: GSE83132
Experimental Models: Cell Lines
MDA231 Parental	Laboratory of Joan Massagué	Bos PD et al 2009
HCC1954 Parental	Laboratory of Joan Massagué	Malladi S et al 2016
PC9 Parental	Laboratory of Joan Massagué	Nguyen DX et al 2009
LLC Parental	Laboratory of Joan Massagué	Chen Q et al 2016
MDA231 LeptoM	This paper
HCC1954 LeptoM	This paper
PC9 LeptoM	This paper
LLC LeptoM	This paper
Experimental Models: Organisms/Strains
Athymic NCR nu/nu mice	Charles River	Strain 490
C3a−/−	Jackson Laboratory	129S4-C3tm1Crr/J
C57/Bl/6J	Jackson Laboratory	C57/Bl/6J
C3aR −/−	Jackson Laboratory	129S4-C3ar1tm1Cge/J
Oligonucleotides
Taqman probe mouse C3	Applied Biosystems	Mm01232779_m1
Taqman probe human C3	Applied Biosystems	Hs00163811_m1
Taqman probe mouse amphiregulin	Applied Biosystems	Mm01354339_m1
Taqman probe human amphiregulin	Applied Biosystems	Hs00950669_m1
Taqman probe human β2-microglobulin	Applied Biosystems	Hs99999907_m1
Taqman probe mouse β2-microglobulin	Applied Biosystems	Mm00437762_m1
shRNA against mouse C3 shA: CCAGAGTTTATTCCTTCATTT	Dharmacon	TRCN0000066878
shRNA against mouse C3 shB: CCATCAAGATTCCAGCCAGTA	Dharmacon	TRCN0000066882
shRNA against human C3 shA: ATCTTTAGCCTCCTGCAGC	Dharmacon	V2LHS_89157
shRNA against human C3 shB: TTCGAACAACAGAGTAGGG	Dharmacon	V3LHS_390612
Control sh CCGGGCGCGATAGCGCTAATAATTTCTC	Sigma	SHC016
Recombinant DNA
Software and Algorithms
Living Image software v.2.50	Caliper Life Sciences
R ver. 3.2.3	The R Project	https://cran.r-project.org/bin/windows/base/
GraphPad Prisim 6	Graph Pad
ImageJ	https://imagej.nih.gov/ij/	Github.com/imagej/imagej1
DESeq2	Love et al., 2014	www.bioconductor.org/packages/release/bioc/html/DESeq2.html
HTSeq v0.5.4	Anders and Huber, 2010	www-huber.embl.de/users/anders/HTSeq/
Metamorph software	Molecular Devices	www.moleculardevices.com
PanoramicViewer	3DHISTECH	www.3dhistech.com/pannoramic_viewer
FastQC v0.11.3.	Babraham Bioinformatics	www.bioinformatics.babraham.ac.uk/projects/download.html
STAR2.3.0e	Dobin et al., 2013	code.google.com/p/rna-star
Other

CONTACT FOR RESOURCES SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Joan Massagué (j-massague@ski.mskcc.org).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Animal Studies

All experiments using animals were performed in accordance with protocols approved by the MSKCC Institutional Animal Care and Use Committee. Athymic NCR nu/nu mice (NCI-Frederick) were housed in maximum barrier facilities, with individually ventilated cages, sterilized food and water. C3^−/− mice (129S4-C3tm1Crr/J), C57/Bl/6J and 129S4-C3ar1tm1Cge/J (Jackson Laboratory) were housed in pathogen-free rooms with individually ventilated cages. To generate C3aR^−/− mice on the C57/Bl/6J background, 129S4-C3ar1tm1Cge/J were backcrossed with C57/Bl/6J mice for 7 generations, and intercrossed to produce C3aR^−/− mice. C3aR genotype was confirmed by PCR of tail tip; please refer to Key Resources Table. For experiments, all mice were used at 5–6 weeks of age. For all experiments, breast cancer models were hosted in female mice, lung cancer models were hosted in male and female mice at 1:1 ratios. For allocation into experimental groups, littermates of the same sex were randomly assigned to experimental groups.

Human Studies

All tissues and CSF were obtained with Informed consent compliance with the MSKCC Institutional Review Board under MSKCC biospecimen protocols 13–039 and 12–123. CSF in excess of that needed for clinical studies was collected from patients harboring breast or lung cancer undergoing CSF sampling by lumbar puncture, cisternal or Ommaya tap. These patients had all previously received a variety of systemic treatments, including chemotherapeutics and targeted therapies. Similarly, archival formalin-fixed paraffin embedded primary tumors were obtained. All patients were assigned an identification code. Clinical details, including patient sex and age at time of CSF collection as well as date of primary tumor diagnosis, leptomeningeal metastasis, parenchymal brain metastasis and date of death were recorded in Table S1.

Cell Lines

Human MDA-MB-231 (MDA231) (female), murine LLC (male) cell lines and their metastatic derivatives were cultured in DMEM with 10% fetal bovine serum (FBS), 2 mM L-glutamine and 100 units/mL penicillin-streptomycin. Human HCC1954 (female) and PC9 (male) cells and their metastatic derivatives were cultured in RPMI 1640 medium supplemented with 10% FBS 2 mM L-glutamine and 100 units/mL penicillin-streptomycin (all from Gibco). For lentivirus production, 293T cells were cultured in DMEM supplemented with 10% FBS and 2 mM L-glutamine. Human primary choroid plexus epithelial cells (CPEpi) were cultured in media specified by the supplier (ScienCell), and used between passages 2–6. All cells tested negative for mycoplasma.

METHOD DETAILS

Metastasis Assays

For parenchymal brain metastasis assays we followed previously described procedures (Bos et al., 2009), by left cardiac ventricle injection of 5 × 10⁴ MDA231-BrM2 cells, 5 × 10⁴ PC9-BrM cells, 5 × 10⁴ LLC-BrM, or 5 × 10⁴ HCC1954-BrM cells suspended in 100 μl of PBS.

To generate leptomeningeal metastatic derivative cells (LeptoM), 2 × 10⁴ parental cells suspended in 10 μL of PBS were injected into the cisterna magna of anesthetized recipient mice, similar to previous techniques (Reijneveld et al., 1999; Ushio et al., 1977). This additional volume introduced into the leptomeninges did not alter mouse behavior on waking from anaesthesia; the mice tolerated the procedure well. Tumor burden was monitored by bioluminescence imaging (BLI), and morbidity was monitored daily. Once tumor burden encompassed the entire CNS, or on development of significant morbidity, the mice were euthanized, the cranial vault removed, the spinal cord transected and the brain lifted out. The basilar meninges were then rinsed three times with 1 mL sterile PBS, collecting all cells lifted from the cavity. Cells were pelleted at 1,000 × g for 5 min., and resuspended in growth medium as per above, seeding into a 30 mm tissue culture dish (Corning). Growth media was changed daily. After three passages, only GFP-positive cells remained, which were tested for presence of mycoplasma before subsequent analysis or use. After three rounds of this selection, the cells were designated “Int”. 50,000 Int cells were injected intracardially into recipient mice, and tumor burden was monitored by BLI once weekly. At significant morbidity or when tumor burden encompassed the entire CNS, cells were harvested from the meninges as described above and grown in culture for three passages and designated “LeptoM”.

Quantification of tumor burden by BLI was performed using an IVIS Spectrum Xenogen instrument (Caliper Life Sciences) and analyzed using Living Image software, v. 2.50. Date of first BLI signal in CNS was recorded to establish time of metastasis. Localization of metastasis was established by histopathology.

A priori sample size determination for animal experiments was determined by Mead’s Resource Equation: 10 animals per treatment group in an experimental design of three groups, without further stratification, gave 28 degrees of freedom, and 8 animals per treatment group gave 24 degrees of freedom, which was considered acceptable. Therefore, for leptomeningeal metastasis assays, 8–10 mice were used in each group; exact numbers for each experimental series are included in the relevant figure legends. For drug treatment experiments, mice were inoculated with cancer cells before randomization into treatment groups. Mice dying less than 24 h after tumor inoculation were excluded from analysis.

Mouse CSF collection and intrathecal treatent administration

CSF was sampled as previously described (Reijneveld et al., 1999). Briefly, anaesthetized mice were positioned prone over a 15 mL conical tube to place c-spine in flexion. The occiput was palpated and a Hamilton syringe fitted with a 31G beveled (cutting) needle was introduced between the occiput and C1 at an angle. The needle was advanced 4 mm before either withdrawing 10 μL of CSF or introducing 10 μL of sterile treatment solution. For intrathecal treatments, mice were randomly assigned to treatment groups. Recombinant mouse C3a (rmC3a) treatment was 500 pg/mL in PBS, and amphiregulin treatment was 2 ng/mL or 4 ng/mL in PBS. The CSF space was accessed a maximum of once every seven days for either treatment of CSF sampling. For CSF analysis, CSF with visible blood or blood product was discarded; this occurred in fewer than 10% of the CSF samples collected. CSF was pooled from 5–10 mice to generate sufficient volume for mouse C3 ELISA or amphiregulin ELISA (R & D Systems), which were performed according to the manufacturer’s instructions.

Analysis of human CSF

All CSF obtained in the course of routine neuro-oncologic care from patients with solid tumors from Jan 24, 2015 until June 29, 2016 were collected for analysis. Specimens grossly contaminated with blood were eliminated, as were specimens obtained from patients out of concern for viral or bacterial meningitis. Freshly obtained CSF was pelleted at 1,000 × g for 5 min. Supernatant fluid was aliquoted and stored at −80°C prior to subsequent analysis before resuspension in sterile PBS. CD45 (+) cells were immune-depleted using CD45 Microbeads (Milteny Biotech) and magnetic separation. CD45-depeleted cells were re-suspended in DMSO and stored in liquid nitrogen until time of analysis. At time of analysis, cells were quickly thawed, washed in sterile PBS, pelleted at 1,000 × g for 5 minutes and then resuspended in 0.5 mL PBS with 1% FBS. 250 uL of cell suspension was lysed in preparation for qPCR using Qiagen RNAeasy Mini Kit (Cat no 74104) according to the manufacturer’s protocol. The remaining 250 uL of cell suspension was immobilized onto permafrost slides by cytospin prior to immunofluorescence as described below. Cell-free CSF was subjected to ELISA analysis for human C2, C3, C4, Factor D or amphiregulin (R & D Systems) according to the manufacturer’s instructions. Dot-blot analysis for select proteins was undertaken with undiluted CSF with the Angiogenesis Proteomic Profiler (R & D Systems) according to the manufacturer’s instructions.

qPCR analysis of human CSF

The analysis incorporated two housekeeping genes, GAPDH and β2M, and the complement C3 gene. A PCA analysis was performed using Pearson Correlation (n-1) among the three genes, and β2M and C3 were used in the factor scores due to several reasons. GAPDH was the lowest among of the 3 genes for both Eigenvalue and variability, which were 0.171 and 5.69% respectively. Factor loading and correlation of GAPDH compared to β2M and C3 also showed low significance with all values less than |0.4| (≤0.4, ≥−0.4). Factor scores were plotted on a scatterplot with the abscissa as C3 and the ordinate as β2M. Averages of two populations showed a significant difference with a p-value of p<0.0001.

Cytospin immunofluorescence

Air-dried cytospin preparations were fixed in 4% PFA 5 minutes at room temperature, rinsed in PBS 3 × 5 minutes prior to blocking for 30 minutes with 10% Normal goat serum, 2% BSA, 0.25% Triton X. A Pap pen (Abcam) en circled the cells and divided the sample in two sections. Primary antibodies were applied: Rabbit anti-human C3 ab129945 1:100; Mouse anti-human C3 Dako 1:50 or isotype control 1:50 incubated overnight at 4 C. This was followed by three PBS washes and secondary antibody Alexa Fluor 633 goat anti-mouse IgG and Alexa Fluor 488 goat anti-rabbit IgG (Life Technologies) for 2 h at room temperature followed by three washes in PBS and mounting with Prolong Gold with DAPI (Molecular Probes).

Clinical Data Abstraction

Date of primary tumor diagnosis, leptomeningeal metastasis, parenchymal brain metastasis and date of death were recorded for all patients. Patients with incomplete CNS staging (gadolinium-enhanced MRI brain, MRI spine and CSF cytologic analysis), were excluded from analysis. Those patients with complete staging information were coded for extent of leptomeningeal metastasis as follows: Hyperintense signal in the leptomeningeal space present on T1 post-contrast sequences and absent on T1 pre-contrast and susceptiblity-weighted sequences. Axial and coronal images were reviewed for all patients. Site of disease was coded as: ventricles, midbrain, cranial nerves, cerebellum, cervical cord, thoracic cord, conus or cauda equine, pons, cerebrum, or absent on MRI.

Patient-derived tumor analysis

Primary tumor specimens from breast and lung cancer patients who underwent primary tumor resection at MSKCC between 2013 and 2014 were collected for analysis. Missing or damaged samples were rejected, as were samples from patients with incomplete clinical records. Immunohistochemical staining for C3 and pancytokeratin was performed by the MSKCC Pathology Core Facility using standardized, automated protocols: A Leica Bond system was used with standard protocol “F” as follows: heat-mediated antigen retrieval with citrate buffer (pH6) ×20 minutes prior to 20 minutes of 5 mg/mL rabbit anti human C3 (abcam ab129945) at RT. C3 staining was quantified by (C3 positive staining area – CD45 positive staining area)/(cytokeratin positive staining area), using Metamorph software as follows: Images were scanned using Miramax, rotated to match the respective slides, and the regions of interest were drawn and exported with CaseViewer and PanoramicViewer (3DHISTECH). After exporting, the ROIs for the immunogens were overlaid on the matching Cytokeratin regions on Volocity (Perkin Elmer) and exported as a TIFF file. A macro was used to analyze the C3 to cytokeratin ratio by splitting the RGB channels and calculating the area of the ROI images in ImageJ. The area of the tissue was calculated by the number of pixels below the threshold, which was set to 253 out of 255. Thresholds for the C3 and cytokeratin were set to 80 and 135 out of 255, respectively. Images were then placed into the image calculator to determine the area of the regions that overlapped. Tissues with C3 positive proportion of 0.5 or higher were designated “High C3” expressing, those with proportion 0.5 or less were “Low C3” expressing.

Cell growth assays

For cell growth assays, CSF from either mouse or human subjects was diluted 1:1 with defined CSF-like medium (119 mM NaCl, 26.2 mM NaHCO₃, 2.5 mM KCl, 1mM NaH₂PO₄, 1.3 mM MgCl₂, 10 mM glucose, 0.3 g/L of serum albumin) (CSHL Protocols 2011) supplemented with 15 mg/L of phenol red. For mouse CSF growth experiments, 500 cells were deposited per well, with 50 μL of growth medium per well of a 384-well plate. For human CSF growth experimented, 1000 cells were deposited per well with 100 μL of growth medium per well of a 96-well plate. Cell viability was assessed at 0, 24, 48 and 72 h using Cell Titer Glo (BioRad) according to the manufacturer’s instructions.

Epithelial barrier function assays

Poly-L-lysine coated 6.5 mm transwell supports (Corning Costar) with 3.0 μm pore size were seeded with human CPEpi cells and maintained in growth medium. Trans-epithelial electrical resistance (TEER) was measured with EVOM resistance meter (World Precision Instruments) fitted with EVOM STX-2 electrode tips. TEER was measured daily. Once TEER was ≥ 200 Ω/cm², the barrier was deemed ready for assays.

Conditioned media (2 mL) from LLC or MDA231 LeptoM cells was immunodepleted with 100 μL of protein G Dynabeads (Life Technologies) coated with 10 μg of either anti-human C3a or anti-mouse C3a, or isotype control mouse IgG_1κ according to manufacturer’s protocol. C3 depletion was confirmed by ELISA for either mouse or human C3 (R & D Systems) as per manufacturer’s instructions. For TEER assays, lower compartment media was replaced with immunodepleted or control conditioned media. TEER was measured 30 min later and every 6 h thereafter.

Cell migration across CPEpi monolayers was assayed in 6.5 mm transwell units with 8 μm pore size, coated with poly-L-lysine, with human CPEpi cells growing on the underside of the membrane. 1 × 10⁴ serum-starved cancer cells in 100 μL were deposited in the upper compartment of the well, 300 μL of defined CSF-like media was deposited in the bottom well. Cells were allowed to migrate 40 h. At the endpoint, cells were removed from the upper compartment and upper membrane surface. GFP-positive cells in each high-power field were counted, 10 high-power fields per well.

Brain histopathology and immunohistochemical staining

For whole-head preparations, extracranial tissue was dissected and the skull was fixed in 4% paraformaldehyde prior to decalcification and mounting in paraffin blocks. For brain only preparations, mouse brains were fixed with 4% paraformaldehyde prior to mounting in paraffin blocks. Images were acquired with Zeiss Axio Imager Z1 microscope or Leica SP5 upright confocal microscope, and analyzed with ImageJ, Imaris, and Metamorph softwares. Antibodies used for immunostaining were chicken anti-C3a ab48581 (Abcam) and chicken anti-GFP 1020 (Aves).

Choroid plexus whole mount preparation

Anesthesized mice were euthanized, the brain sterilely dissected and placed in ice-cold sterile PBS. With the aid of a stereoscope (Zeiss Stemi 2000C), choroid plexus (CP) was removed from the lateral ventricles bilaterally and placed in ice-cold PBS prior to treatment. CPs were treated for 2 h at 37°C with conditioned medium from parental cell lines supplemented with 5 ng/mL recombinant mouse C3a (R & D Systems) or equivalent volume of PBS. After treatment, CPs were immediately fixed in 4% PFA 1 h, washed twice in PBS 5 min each, permeabilized twice with PBS-Triton 0.25% for 10 min before incubation in 10% Normal goat serum, 2% BSA, 0.25% Triton X for 24 h at 4 C with gentle rocking. Then, primary antibodies are added: goat anti-ZO-1 (Pierce) 1:1000 and or rabbit anti-claudin (Invitrogen) 1:50 for 18 h at 4°C. This is followed by six 10 min washes in PBS-Triton X 0.25% before addition of Alexa Fluor 633 goat anti-mouse IgG and Alexa Fluor 488 goat anti-rabbit IgG (Life Technologies) for 2 h. This was followed by 3 washes with 0.25% Triton-X in PBS and 3 washes with PBS before mounting with Prolong Gold with DAPI (Molecular Probes).

Transcriptomic analysis

Cells cultured in 10 cm plates at 75 % confluency were collected, and total RNA was extracted using the PrepEase RNA spin kit (USB). mRNA purified from cancer cells was used for library construction with TruSeq RNA Sample Prep Kit v2 (Illumina) following the manufacturer’s instructions. Samples were barcoded and run on a Hiseq 2000 platform in a 50bp/50bp paired-end run, using the TruSeq SBS Kit v3 (Illumina). An average of 40 million paired reads were generated per sample.

FASTQ files from RNA-Seq results were quality assessed by FastQC v0.11.3. Raw reads were mapped to human genome hg19 (GRCh37, Feb 2009) or mouse genome mm10 (GRCm38, Dec 2011) using STAR2.3.0e(Dobin et al., 2013) with standard settings for paired-end sequencing reads. In average 84% of raw reads were uniquely mapped. Mapped reads were counted to each gene by HTSeq v0.5.4(Anders and Huber, 2010) with default settings. Differential gene expression analysis were performed following the instructions of “DESeq2” package (Love et al., 2014) deposited in Bioconductor.

mRNA and protein detection

Total RNA was extracted using the PrepEase RNA spin kit (USB). To prepare cDNA, 1 μg of total RNA was treated using the Transcriptor First Strand cDNA synthesis kit (Roche). C3 and amphiregulin expression was quantified by Taqman gene expression assay primers: mouse C3 Mm01232779_m1, human C3 Hs00163811_m1, mouse amphiregulin Mm01354339_m1, human amphiregulin Hs00950669_m1 (Applied Biosystems). Relative gene expression was normalized relative to β2-microglobulin (Hs99999907_m1, Mm00437762_m1). Reactions were performed using SYBR Green I Master Mix (Applied Biosystems). Quantitative expression data were analyzed using an ABI Prism 7900HT Sequence Detection System (Applied Biosystems). For western immunoblotting, cells or choroid plexus samples were lysed with RIPA buffer and protein concentrations determined by BCA Protein Assay Kit (Pierce). Proteins were separated by SDS-PAGE and transferred to nitrocellulose membranes (BioRad). Antibodies used for western blotting were rabbit anti-β-catenin D10A8 (Cell Signaling); rabbit anti-claudin-1 D5H1D (Cell Signaling); mouse anti-MLCK ab55475 (Abcam); and, mouse anti-pMLCK ab200809 (Abcam)

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis and figure plotting were performed in R (ver. 3.2.3) or GraphPad software (Prism 6). Exact value of N and what N represents for each experiment are annotated in each figure legend. Values reported are averages ± standard error of the mean (S.E.M.). For Transcriptomic analysis, multiple hypothesis testing was adjusted using the Benjamini & Hochberg false-discovery-rate method.

DATA AND SOFTWARE AVAILABILITY

The accession number for RNA sequencing data deposited in NCBI Gene Expression Omnibus is GSE83132.

Supplementary Material

1

Figure S1. In vivo selection for leptomeningeal metastasis, Related to Figure 1

(A) Cell lines used in this study and their characteristics.

(B–D) 50,000 MDA231 Parental (Par, green), LeptoM (purple) or BrM (blue) cells are injected intracardially into recipient mice. Tumor growth is monitored by BLI twice weekly, neuro-anatomic localization (leptomeningeal, left; parenchymal, right) is determined by histology (inset). PC9 series n = 11 mice Par, 12 mice LeptoM, 7 mice BrM cells; HCC1954 series n = 5 mice Par, 9 mice LeptoM, 8 mice BrM; LLC series n = 5 mice Par, 7 mice LeptoM, 6 mice BrM.

Figure S2. RNASeq analysis of leptomeningeal metastasis derivative cell lines, Related to Figure 2

(A). Venn diagrams of genes differentially expressed between Par and Int and Par and LeptoM cell lines. Genes with base mean ≥ 50, fold change ≥ 2 or ≤ 0.5 and p < 0.01 were collected for analysis.

(B) Differentially expressed genes in common between all models. Genes differentially between Par and Int or Par and LeptoM with base mean ≥ 50 fold change ≥ 2 or ≤ 0.5 and p < 0.01 were collected for each model.

(C) Genes differentially expressed in common between all models are displayed with fold change noted in the chart. p < 0.05 are shown in grey, p < 0.01 are shown in black. *The mouse equivalent of the human gene C15orf48 is NMES1.

(D) Schematic of genes included in the KEGG complement and coagulation cascades. Genes differentially expressed between parental and LeptoM cells are colored according to expression pattern at left.

(E) Quantitative PCR for C3 mRNA in all models, beta-2 microglobulin served as internal standard. Each sample assayed in quadruplicate in two independent experiments. * indicates p < 0.05; ** p < 0.01

(F) ELISA for human C3 in mouse CSF. CSF was sampled from mice harboring extracranial metastases “None”, parenchymal metastases “BrM” or leptomeningeal metastases “LeptoM”. n = 6 mice per group. **** p < 0.0001

Figure S3. C3 expression of leptomeningeal metastasis derivative cell lines and human disease, Related to Figure 3

(A–B) Rubric for assignment of leptomeningeal disease burden score. Sites of leptomeningeal metastasis are assigned: Site A: ventricles, midbrain or cranial nerves; Site B: cerebellum; Site C: cervical cord; Site D: thoracic cord; Site E conus medullaris or cauda equina; Site F: pons; Site G: cerebrum. Refer also to Figure 3B.

(C) Site of disease and relationship to concentration of C3 in CSF obtained from lumbar cistern. N = 76 patients.

(D) Period of time of active clinical follow up after initial primary tumor resection. Refer to Figure 2J. NS = not significant.

(E) and (F) IHC of primary tumors and parenchymal metastases for C3. n = 9 parenchymal metastases and 17 primary tumor samples, unmatched (F), n = 7 matched primary and parenchymal brain metastasis tissue samples (G). NS = not significant.

Figure S4. C3 knockdown inhibits leptomeningeal metastasis; C3 add-back promotes leptomeningeal metastasis, Related to Figure 4

(A) Short hairpin knockdown of C3 mRNA as measured by qPCR. Data are presented as fold change from vector control n= 6 samples per group.

(B) Short hairpin knockdown of C3 expression as measured by ELISA of conditioned media. n = 6 samples per group.

(C) 2,000 LLC LeptoM cells stably expressing vector control, C3 shA or shB were injected intracisternally into C57/Bl6 mice. n = 5 mice per group in two independent experiments. Left panel: bioluminescence quantification of metastatic burden. * p < 0.05; ** p < 0.01; Right panel: Kaplan-Meier plot of overall survival of mice injected with LLC-LeptoM cells with either vCtl, shA or shB.

(D) 2,000 PC9 LeptoM cells stably expressing vector control, C3 shA or shB were injected intracisternally into nude mice. n = 5 mice per group in two independent experiments. Left panel: bioluminescence quantification of metastatic burden. * p < 0.05; ** p < 0.01; Right panel: Kaplan-Meier plot of overall survival of mice injected with LLC-LeptoM cells with either vCtl, shA or shB.

(E) 2,000 LLC LeptoM cells were injected intracisternally into wild-type or C3 knockout mice in C57/Bl6 background. Left panel: bioluminescence quantification of metastatic burden. n = 10 mice per group. NS = not significant. Right panel: Kaplan-Meier plot of overall survival of mice in each group. NS = not significant.

(F) 1,000 MDA231-LeptoM (A) or PC9-LeptoM cells were seeded in each well of a tissue-culture treated 96-well plate and allowed to grow in CSF from solid tumor patients with or without LM with 50% artificial CSF. Cell growth was monitored by CellTiter Glo assay at t = 1h and 72h. Data represent two independent experiments performed in quadruplicate. *** p < 0.001

(G–H) 500 PC9-LeptoM cells were seeded into a 384-well plate containing CSF collected from mice harboring no malignancy. Mice were treated with 1 mg/kg recombinant mouse C3a (rmC3a) or PBS I.P. 30 min prior to CSF collection. For mice treated with PBS, rmC3a was added ex vivo to a final concentration of 20 ng/mL to mouse CSF. Data represent two independent experiments performed in quadruplicate. *** p < 0.001

(I) BLI at day 18 of mice inoculated with 2,000 PC9 Parental cells. Parental cells were introduced into the cisterna magna with either recombinant C3a or vehicle. Additional rmC3a or vehicle was delivered intracisternally every 7 days. **** p < 0.0001

(J) Kaplan-Meier plot of overall survival of mice in (H). n = 10 mice per group.

(K) BLI at day 7 of mice inoculated with LLC Parental cells. 2,000 parental cells were introduced into the cisterna magna with either recombinant C3a or vehicle. Additional rmC3a or vehicle was delivered intracisternally every 7 days.

(L) Kaplan-Meier plot of overall survival of mice in (K). n = 10 mice per group.

Figure S5. C3 alters Blood-CSF-Barrier permeability, Related to Figure 5

(A–B) Human CPEpi cells are grown on poly-L-Lysine coated transwells until TEER is greater than 200 Ω/cm². Media was changed to conditioned media (CM) as indicated and TEER measurements were obtained every 12 hours. Two independent experiments performed in triplicate are averaged.

(A) Conditioned media from LLC parental cells supplemented with recombinant mouse C3a at 5 ng/mL (rmC3a), open circles or equal volume PBS, closed circles.

(B) Conditioned media from LLC-LeptoM derivative cells immunodepleted with anti-C3 (open circles), vehicle (closed circles) or isotype control (grey circles).

(C–D) Confluent human CPEpi monolayer cultured on Poly-L-Lysine coated transwells were pretreated with 200 ng/mL pertussis toxin (PTX), 50 ng/mL phorbol ester (PMA) or vehicle alone for 30 min prior to addition of 50 ng/mL FITC-dextran 40 kDa and 5 ng/mL recombinant human C3a. After 3 hours, media was collected and fluorescence quantiated from upper and lower chambers. * indicates p < 0.05.

(E) Migration of 20,000 serum-starved Par cells, LeptoM control short hairpin (vCtl) or C3 knock down short hairpin (shA) across a confluent monolayer of human CPEpi cells toward artificial CSF + 1% FBS. Cells were allowed to migrate 12 hours before quantification of cells. Two independent experiments were performed in duplicate. Ten high-power fields (40×) were counted per condition per replicate. NS = not significant. Refer also to Figure 5J.

(F) Neuro-anatomic localization of LLC-LeptoM cells after intracardiac inoculation into C3aR^+/+ or C3aR^−/− mice. 50,000 cells were injected intracardially into recipient mice. Three days later, the mice were euthanized, brains harvested and fixed. FFPE sections were stained for GFP by IHC. GFP (+) cells were counted in each of 8 coronal sections per mouse. N = 3 mice per group. NS = not significant.

Figure S6: CSF composition of patients with leptomeningeal metastasis, Related to Figure 6.

(A) Expression of C3aR in mouse choroid plexus as determined by qPCR compared with leptomeningeal-derivative cell lines. **** indicates p < 0.0001.

(B) CSF from six solid tumor patients prior to diagnosis and just after diagnosis was compared by commercial cytokine dot blot array. Analytes expressed at higher levels after diagnosis with p < 0.01 by paired t-test in at least three patients are presented. Each dot represents the mean fold change for a particular patient.

(C) Amphiregulin in conditioned media from indicated model cell lines as measured by ELISA. Data represent two independent experiments performed in triplicate.

(D) Quantitative PCR for amphiregulin mRNA in indicated mouse tissues, beta 2 microglobulin served as internal standard. Each sample assayed in quadruplicate in two independent experiments. ND = amphiregulin not detected.

(E) Amphiregulin in CSF by ELISA versus site of CSF sample. CSF samples were obtained from patients harboring leptomeningeal metastasis, either from lateral ventricle (n = 13) or lumbar cistern (n = 11).

(F–G) Patients with clinical improvement (F) or lack of improvement (G) in response to treatment over three time points. ELISA for indicated analytes in CSF. Refer also to Figure 6G.

Table S1: Clinical Characteristics of Patients with Symptoms of Leptomeningeal Metastasis, Related to Figure 3A.

Patients with solid tumor primary malignancy and clinical symptoms suggestive of leptomeningeal metastasis underwent CSF sampling by lumbar puncture and MRI imaging of brain and spine. Site of CNS metastasis (leptomeninges vs. parenchyma vs. no metastasis) and date of final clinical diagnosis are indicated. Refer also to Figure 2C.

Table S2: Extravasation genes in LeptoM and Parental cell lines, Related to Figure 1 E and Supplementary Figure 2.

Transcriptome of Parental and LeptoM cell lines was obtained by RNASeq analysis. Mean read count of six genes associated with cancer cell extravasation is shown. P-value was determined with student’s t-test. NS = not significant. *Mouse ortholog of human CCL2 is CCL12.

Key Resource Table.

Your MS Word document “KRT Boire 2017.docx” cannot be opened and processed. Please see the common list of problems, and suggested resolutions below.

Common Problems When Creating a PDF from Microsoft Word Documents

When you open your document in MS Word, an alert box may appear with a message. This message may relate to margins or document size. You will need to find the piece of your Word document that is causing the problem. Selectively remove various pieces of the file, saving the modified file with a temporary file name. Then try to open modified file. Repeat this process until the alert box no longer appears when you open the document.

Embedded Macros

Your submission should not contain macros. If they do, an alert box may appear when you open your document (this alert box prevents EM from automatically converting your Word document into the PDF that Editors and Reviewers will use). You must adjust your Word document to remove these macros.

Corrupted Tables

Your document may contain a table that cannot be rendered correctly. This will be indicated by a warning alert box. Correct the content of the table that causes the problem, so that the alert box no longer appears.

Word 2002/Word XP files

At the present time, EM supports Word files in Word 2000 and earlier formats. If you are using a more recent version of MS Word, try saving your Word document in a format compatible with Word 2000, and resubmit to EM.

Other Problems

If you are able to get your Word document to open with no alert boxes appearing, and you have submitted it in Word 2000 (or earlier) format, and you still see an error indication in your PDF file (where your Word document should be appearing). please contact the journal via the ‘Contact Us’ button on the Navigation Bar.’

You will need to reformat your Word document, and then re-submit it.