Invasive growth of brain metastases is linked to CHI3L1 release from pSTAT3-positive astrocytes

Abstract

Background

Compared to minimally invasive brain metastases (MI BrM), highly invasive (HI) lesions form abundant contacts with cells in the peritumoral brain parenchyma and are associated with poor prognosis. Reactive astrocytes (RAs) labeled by phosphorylated STAT3 (pSTAT3) have recently emerged as a promising therapeutic target for BrM. Here, we explore whether the BrM invasion pattern is influenced by pSTAT3⁺ RAs and may serve as a predictive biomarker for STAT3 inhibition.

Methods

We used immunohistochemistry to identify pSTAT3⁺ RAs in HI and MI human and patient-derived xenograft (PDX) BrM. Using PDX, syngeneic, and transgenic mouse models of HI and MI BrM, we assessed how pharmacological STAT3 inhibition or RA-specific STAT3 genetic ablation affected BrM growth in vivo. Cancer cell invasion was modeled in vitro using a brain slice-tumor co-culture assay. We performed single-cell RNA sequencing of human BrM and adjacent brain tissue.

Results

RAs expressing pSTAT3 are situated at the brain–tumor interface and drive BrM invasive growth. HI BrM invasion pattern was associated with delayed growth in the context of STAT3 inhibition or genetic ablation. We demonstrate that pSTAT3⁺ RAs secrete Chitinase 3-like-1 (CHI3L1), which is a known STAT3 transcriptional target. Furthermore, single-cell RNA sequencing identified CHI3L1-expressing RAs in human HI BrM. STAT3 activation, or recombinant CHI3L1 alone, induced cancer cell invasion into the brain parenchyma using a brain slice-tumor plug co-culture assay.

Conclusions

Together, these data reveal that pSTAT3⁺ RA-derived CHI3L1 is associated with BrM invasion, implicating STAT3 and CHI3L1 as clinically relevant therapeutic targets for the treatment of HI BrM.

Metastasis to the brain is a common complication of cancer, occurring in approximately 30% of patients with advanced solid tumors,¹ particularly of the lung, breast, and skin.² Treatment options for brain metastases (BrM) remain limited for patients without targetable genetic alterations,³ largely relying on a combination of surgery, radiotherapy, and systemic therapies, with generally poor clinical outcomes.^4,5 It is estimated that disease spread to the central nervous system is the cause of death for approximately half of patients with BrM.⁵ Our group has previously identified that peritumoral invasion into the adjacent brain is associated with poor outcomes in patients with surgically resected BrM.⁶ Patients with highly invasive (HI) BrM experience more frequent local recurrence and shortened overall survival compared to those with BrM exhibiting a minimally invasive (MI) pattern at the brain–tumor interface.⁶ Unlike MI disease that remains localized, HI BrM forms abundant contacts with parenchymal cells in the peritumoral brain.

Astrocytes are a highly abundant glial cell type that have been shown to localize predominantly to the periphery of BrM.^7,8 They can enter into reactive states in response to brain injury, which includes the colonization of the brain by metastatic cancer cells.^9–11 Secreted molecules from BrM-associated reactive astrocytes (RAs) can alter cancer cell phenotypes, including promoting chemoresistance,¹² tumor growth,^13–15 and invasiveness.^16,17 A subpopulation of RAs with activated signal transducer and activator of transcription 3 (STAT3) signaling have been shown to be required for metastatic colonization of the brain.⁸ This pro-metastatic program is initiated and maintained through the co-option of RAs by cancer cells.⁸ Pharmacological inhibition of this pathway with Legasil, a silibinin-based therapeutic targeting STAT3, is a promising avenue for BrM treatment. Legasil has shown efficacy in vitro, in preclinical animal models, and preliminary BrM patient cohorts.^8,18–20 While promising, recent studies have found that BrM patients do not universally respond to Legasil treatment,^8,20 suggesting that underlying mechanisms or features exist that can distinguish patients who are most likely to derive benefit from STAT3 inhibition.

Here we demonstrate that HI BrM are associated with abundant RAs, which stain positively for phosphorylated STAT3 (pSTAT3⁺ RAs) when compared to MI lesions, thus rendering HI BrM preferentially sensitive to Legasil. Integrating single-cell RNA sequencing of human HI BrM with the secretome of pSTAT3⁺ RAs identified CHI3L1 as a STAT3 target overexpressed in pSTAT3⁺ RAs. Activation of brain slice cultures with a STAT3 activating cocktail, or administration of recombinant CHI3L1, induces cancer cell invasion into the brain parenchyma. Our findings describe a novel mechanism through which pSTAT3⁺ RAs promote the growth of invasive BrM and identify STAT3 and CHI3L1 as clinically relevant, stromal-derived therapeutic targets for HI BrM.

Methods

Ethics Statement

All patient samples were obtained after informed, written consent and were de-identified. Studies were approved by the institutional review boards of McGill University and the Montreal Neurological Institute-Hospital (IRB #2018-4150) and were conducted in accordance with the 1996 Declaration of Helsinki. All animal work was conducted in accordance with animal use protocols approved by the McGill University Animal Resources Centre (#2001-4830) in accordance with guidelines established by the Canadian Council on Animal Care and the Spanish National Cancer Research Centre (protocol approved by the CNIO (IACUC.030-2015), Instituto de Salud Carlos III (CBA35_2015-v2), and Comunidad de Madrid Institutional Animal Care and Use Committee (PROEX135/19).

Clinical Specimen Analysis and Growth Pattern Classification

H&E-stained slide specimens of surgically resected BrM were evaluated by a neuropathologist (M-C.G.). The criteria for BrM histopathological growth pattern classification have been previously described.⁶

Organotypic Brain Slice-Tumor Co-Culture Model

Assays using tdTomato transgenic mice.

Previously published protocols for organotypic slice cultures were adapted for our co-culture model.^25–27 Briefly, 3–12 week-old tdTomato transgenic mice were ethically anesthetized and decapitated. Whole brains were quickly removed and transferred to ice-cold dissection media containing minimum essential media (MEM) supplemented with 100 IU/mL penicillin/streptomycin, 0.5 µg/mL amphotericin B, and 50 µg/mL gentamycin sulfate. The cerebellum was removed from the brain using a sterile razor blade. A Leica VT1200/S vibratome was used to cut 250 µm thick coronal slices. Individual brain slices were transferred to 0.4 µm transparent polyethylene terephthalate (PET) membrane inserts in a 6-well plate. Media was changed to incubation media (50% MEM, 25% Hank’s balanced salt solution (HBSS), 25% normal horse serum (NHS), 100 IU/mL P/S, 0.5 µg/mL amphotericin b and 50 µg/mL gentamycin sulfate), with 1 mL added below the insert and 20 µL added directly onto the brain slice. To generate the 3D tumor plug, 10⁵ tumor cells suspended in 14 µL Matrigel® (Corning; Cat. #: 356231) were added to a sterile 5 mm diameter plastic spacer adjacent to the brain slice. The plugs were incubated at 37°C for 2 h. The spacers were subsequently removed, ensuring the 3D tumor plug was in direct contact with the cortex of the brain slice. Slices were incubated for 72 h at 37°C with 5% CO₂ to allow for cell invasion into the brain. Media was changed once, 48 h after tumor plug placement. Following 72 h of incubation, brain slices were gently washed twice with PBS and fixed for 24 h in 10% formalin at room temperature. Slices were then washed twice with PBS, and all liquid was aspirated prior to imaging (LSM710 confocal microscope; Zeiss; 5× FLUAR lens with a Numerical Aperture of 0.25; image size 512 × 512 pixels). Spectral imaging laser unmixing were used to obtain images of the brain-tumor plug interface. Imaris 9.2.1 software (Oxford Instruments) was used to quantify cell invasion; invaded cells were defined as green fluorescence objects at least 12 µm in diameter within the region of red fluorescence (brain). Cells that had invaded greater than 200 µm into the brain were quantified to ensure that true brain invasion, rather than cell migration to the brain-plug interface, was measured.

Assays using GFAP-Cre/ ERT2; Stat3^loxP/loxP mice.

Brain slice-tumor co-culture experiments were performed as previously described.⁸ Briefly, brains were dissected in HBSS supplemented with HEPES (pH 7.4, 2.5 mM), d-glucose (30 mM), CaCl₂ (1 mM), MgCl₂ (1 mM), and NaHCO₃ (4 mM) and embedded in 4% low-melting agarose (Lonza) preheated at 42°C. A vibratome (Leica) was used to cut brain slices 250 μm thick, which were sectioned at the hemisphere into 2 pieces. Brain slices were transferred to 0.8 μm pore membranes (Sigma-Aldrich) and incubated in slice culture media (DMEM supplemented with HBSS, 5% FBS, 30 mM d-glucose, 1mM 1-glutamine, and 100 IU/mL penicillin/streptomycin). After 1 h incubation, 3 × 10⁴ luciferase-expressing cancer cells were suspended in 2 μL of slice culture media and placed on the surface of the slice. Brain slices with cancer cells underwent bioluminescent imaging (BLI) using the IVIS-200 imaging system 12–16 h after the addition of the cancer cells (day 0) and on day 3. After 72 h of invasion, slices are fixed in 4% paraformaldehyde overnight. Slices were subsequently stained with antibodies against GFP (1:1000, Aves Labs, Cat. #: GFP-1020), GFAP (1:1000, Millipore, Cat. #: MAB360), and pSTAT3 (Y705, 1:100, Cell Signaling Technologies, Cat. #: 9145). Depth images of 27 μm were obtained using a Leica SP5 up-right confocal microscope (20× objective with 2.5 zoom); 19 z-slices were collected at each field of view with a distance of 1.51 μm between z-slices. ImageJ was used for image analysis; the percentage of GFP⁺ area present in the brain slice 19.63 μm deep (z-slice #13) was calculated relative to the GFP⁺ area at the brain surface (z-slice #1).

Brain Slice Treatments

Pre-treatment with STAT3-activating cytokines.

Following vibratome sectioning, brain slices were incubated for 24 h in incubation media supplemented with a cytokine cocktail consisting of human recombinant TGF-α (0.1 µg/mL; PeproTech; Cat. #: 100-16A), MIF (0.1 µg/mL; PeproTech; Cat. #: 300-69) and EGF (0.01 µg/mL; PeproTech; Cat. #: AF-100-15). Media was subsequently removed, and brain slices were washed twice with PBS prior to the addition of new, untreated incubation media and subsequent tumor plug placement. Untreated incubation media was used for the media change at 48 h.

Treatment with recombinant human CHI3L1.

Following vibratome sectioning, brain slices were incubated for 6 h in untreated incubation media. Tumor plugs were then placed, as described above, and media containing 100 µg/mL recombinant human CHI3L1 (R&D Systems; Cat. #: 2599-CH) ± 10 µM capivasertib (MedChem Express; Cat. #: HY-15431) was added to the inserts and cells were allowed to invade into the brain slice. Incubation media supplemented with recombinant human CHI3L1 ± capivasertib was used for the media change at 48 h.

Treatment with OH-Tmx.

Brain slices from GFAP-Cre/ ERT2; Stat3^loxP/loxP mice were treated with 1 µM 4-hydroxytamoxifen (OH-Tmx, Sigma-Aldrich, Cat. #: H7904) at day 0 to induce STAT3 knockout, or a vehicle control.

Immunofluorescent Staining of Brain Slice Co-Cultures

Brain slice co-cultures were fixed overnight in 4% PFA at 4°C. They were then washed with PBS-T (PBS with 0.5% Triton X-100) and blocked with normal goat serum in PBS-T (1:20) for 1 h at room temperature (RT). Co-cultures were stained with anti-Ki67 monoclonal antibody (1:5; Roche; Cat. #: 790-4286) for 48 h at 4°C, then with goat anti-rabbit Alexa Fluor 555 (1:400; Invitrogen, A21429) for 1 h at RT. After counterstaining with DAPI (1:1000), cocultures were stored in with PBS and imaged on 35 mm petri dishes by confocal microscopy.

Mouse Experiments

PDX models of BrM were established as previously described.⁶ Briefly, fresh BrM patient material was received from the operating room at the time of surgery. Fragments were expanded by engraftment into the subcutaneous flanks or mammary fat pad (in the case of breast cancer BrM) of NSG mice (The Jackson Laboratories, strain #005557). Subcutaneous PDX tumors were enzymatically dissociated (Miltenyi; Cat. #: 130-095-929) to generate a cell suspension. 10⁵ cells (from dissociated PDX tumors or cell lines) suspended in 5 µL of PBS were intracranially injected into the right frontal lobe of mice using a guide screw method.⁶⁵ The experiment was terminated and all mice were euthanized when the first mouse in any arm reached the clinical endpoint, as determined by blinded animal health technicians. Legasil treatment: Mice were treated with Legasil (Silibinin; Eurosil 85 milk thistle dry extract, refined and standardized 77.5% silymarin dosed at 200 mg/kg) diluted in water or a vehicle control (water) by daily oral gavage. Every 3 days, mice were anesthetized (isoflurane), injected intraperitoneally with d-luciferin (150 mg/kg) and imaged to quantify tumor growth using IVIS bioluminescence imaging. For PDX experiments, Legasil treatment began the day following the timepoint at which all animals included in the study had detectable bioluminescence signals by IVIS imaging. Animals were randomized into vehicle vs. Legasil treatment arms based on the mean bioluminescent signal intensity at the pre-treatment imaging time point. For syngeneic mouse experiments, treatment was initiated on the day following intracranial injection of the cancer cells. Tamoxifen treatment: cKO-STAT3: GFAP-Cre/ ERT2; Stat3^loxP/loxP mice on a C57BL/6J background were used as previously described.⁸ Briefly, 3 days after intracranial injection of cancer cells, mice were randomized and tamoxifen (50 µL of 20 mg/mL suspension in corn oil) or vehicle treatment was initiated with daily intraperitoneal injections until the day of sacrifice. Anesthetized mice (isoflurane) were injected retro-orbitally with d-luciferin (150 mg/kg) and imaged with IVIS bioluminescence imaging. Bioluminescence analysis was performed using Living Image Software (PerkinElmer; version 3).

Cell Migration

Cell tracking experiments were performed and analyzed as previously described.⁶⁶ Briefly, 1.5 × 10³ MDA-MB-231 cells were seeded onto μ-slide 8-well dishes (Ibidi, Cat. #: 80821) coated with 5 μg/cm² human fibronectin (Sigma-Aldrich, Cat. #: F-0895). Cells were incubated for 24 h in media alone or media supplemented with 100 ng/mL of recombinant CHI3L1 in the presence or absence of 10 µM capivasertib. Imaging was performed on a Zeiss Axiovert 200M Automated Inverted microscope (Carl Zeiss) equipped with a Chamlide TC-L-Z003 stage top incubation chamber (Live Cell Instrument), set to 37°C with 5% CO₂. A 10× PLAN NEOFLUAR lens with a Numerical Aperture of 0.3 and a Ph 1 phase ring was used for imaging. A halogen lamp (Carl Zeiss, HAL 100) set to 2.3 V was used. 3 regions per well were collected on an Axiocam 506 mono camera (Carl Zeiss) exposed for 750 ms per scene, every 5 min for 10 h. Pixels were binned at 2 × 2. Cell migration experiments were manually tracked and blinded to analysts in ImageJ. Data was interpreted using a Matlab (Mathworks) script generated by A.K. (co-author), which calculates individual cell velocity as cell displacement over time.

Statistical Analysis

GraphPad Prism 9 was used for all statistical analyses, with the exception of bioluminescence BrM growth curves, for which MedCalc (v9) (MedCalc Software, Mariakerke, Belgium) was used. All quantitative results are shown with means ± SEM. Statistical tests used are indicated in the appropriate figure legend.

Results

pSTAT3⁺ Reactive Astrocytes Are More Abundant in Highly Invasive Compared to Minimally Invasive Brain Metastases

We performed multiplex immunohistofluorescence (IHF) on MI and HI patient BrM to visualize the location of reactive astrocytes (GFAP⁺) and cancer cells (PanCK⁺: epithelial BrM; MART1⁺: melanoma BrM). In MI BrM, we observed a clear separation between cancer cells within the metastatic lesion and RAs that remain peripheral to the main tumor mass (Figure 1A, Supplementary Figure 1A). This contrasts with HI BrM, where cancer cells invade into the brain parenchyma and form more abundant contacts with RAs (Figure 1A, Supplementary Figure 1A).

Given that STAT3 signaling in RAs has been shown to contribute to a pro-metastatic environment,^8,21 we sought to determine whether RAs in MI and HI BrM display differential STAT3 activation. Using patient BrM samples, we first confirmed that phosphorylated STAT3 (pSTAT3: Tyr 705) is present in RAs and that RAs are identified using independent markers (GFAP or n-cadherin), consistent with previous literature^10,22 (Supplementary Figure 1B and C). We then performed 2-color immunohistochemistry (IHC) for pSTAT3 (Tyr 705) and GFAP in 62 human BrM samples (22 MI and 40 HI, Figure 1B, Supplementary Figure 1D and E) and 30 intracranially injected patient-derived xenograft (PDX) models of BrM⁶ (16 MI and 14 HI, Figure 1C, Supplementary Figure 1E). Analysis algorithms detected GFAP⁺ RAs that exhibited low (+1), medium (+2), and high (+3) intensity staining for pSTAT3, which was used for subsequent quantification of a pSTAT3 + H-Score (Supplementary Figure 1D). Quantification of these tissues revealed significantly more pSTAT3⁺ RAs in HI vs. MI BrM in both patient and PDX cohorts, while no differences in pSTAT3 staining were detected in non-GFAP expressing cells (Figure 1D–G).

STAT3 Activity Within Reactive Astrocytes Is Required for Efficient Growth of Highly Invasive Brain Metastases and Contributes to Cancer Cell Invasion in the Brain

We next evaluated how STAT3 inhibition alters the intracranial growth of MI and HI BrM. Legasil, a silibinin-based therapeutic in development for the treatment of BrM, is a well-characterized STAT3 inhibitor^23,24 that has proven effective in the treatment of BrM by targeting pSTAT3⁺ RAs.⁸ We leveraged our established panel of MI and HI PDX BrM models⁶ (Supplementary Figure 2A) and monitored tumor growth in response to Legasil treatment. MI PDX BrM (GCRC2084, GCRC2015) did not experience significant changes in growth upon treatment with Legasil when analyzed as a total luminescence signal (Supplementary Figure 2B–D) or following normalization to baseline luminescence to account for differences in the cancer cell inoculum at the time of treatment initiation (Figure 2A and B). However, Legasil-treated HI PDX BrM (GCRC1945, GCRC1987) showed a trend of decreased growth when total luminescence signals were analyzed (Supplementary Figure 2B, E and F), which became statistically significant when normalized to baseline luminescence to account for the cancer cell inoculum at the time of treatment initiation (Figure 2C and D). Legasil-mediated loss of pSTAT3 in GFAP⁺ cells was confirmed by IHC in the GCRC1987 PDX (Supplementary Figure 2G–I), with a non-significant effect on the GFAP^- cells. To validate the relationship between HI growth pattern and response to Legasil, we expanded our investigation to include intracranially injected syngeneic cell lines (Supplementary Figure 2J). Consistent with the PDX models, we observed no difference in an MI syngeneic model of BrM (MC38 colorectal cancer cells) following Legasil treatment (Figure 2E, Supplementary Figure 2K), whereas 2 HI syngeneic models (4T1 breast cancer cells and YUMM1.7 melanoma cells) showed significant reduction in intracranial growth following Legasil treatment (Figure 2F and G, Supplementary Figure 2K).

Given our observations that the pSTAT3 signal does not differ in the tumor compartment of HI versus MI BrM, but is significantly more abundant in the RAs of HI vs. MI lesions (Figure 1), we sought to evaluate if loss of STAT3 in RAs could be contributing to the Legasil-mediated effects on HI BrM growth. To test this hypothesis, we utilized a conditional knockout (KO) mouse model (cKO-STAT3: GFAP-Cre/ERT2; Stat3^loxP/loxP) where STAT3 expression is lost in RAs following tamoxifen (Tmx) administration.⁸ Three days following intracranial injection of MC38 cells (colorectal MI syngeneic model) or YUMM1.7 cells (melanoma HI syngeneic model), cohorts of mice were established that were treated with either vehicle or Tmx to induce deletion of STAT3 in RAs (Supplementary Figure 2L). No difference in growth was observed in the MC38 MI model between the vehicle or Tmx-treated cohorts when the data was analyzed as a total luminescence signal (Supplementary Figure 2M), or following normalization to baseline luminescence to account for differences in the cancer cell inoculum at the time of treatment initiation (Figure 2H). A similar proliferative index in the 2 groups was observed at the experimental endpoint (Figure 2I). In contrast, when compared to vehicle controls, mice bearing the YUMM1.7 HI model treated with Tmx exhibited a trend towards impaired growth when total luminescence signals were analyzed (Supplementary Figure 2N), which became statistically significant when normalized to baseline luminescence to account for the cancer cell inoculum at the time of treatment initiation (Figure 2J). The observed reduction in tumor growth in the Tmx-treated mice was correlated with reduced proliferation (Figure 2K) relative to the vehicle treated cohort.

To explore a potential role for pSTAT3⁺ RAs in modulating cancer cell invasion in the brain, we established an in vitro murine brain slice-tumor plug co-culture model of brain invasion.^25–27 Briefly, coronal sections of tdTomato-expressing mouse brains were cultured in vitro in the presence or absence of a STAT3-activating cytokine cocktail (EGF, TGF-α, MIF) (Figure 3A). This cocktail contains cytokines ubiquitously represented in a panel of BrM cancer cell lines and has been shown to induce STAT3 phosphorylation in RAs.⁸ We employed human (PC9) and mouse (LLC) lung cancer cell lines, both of which form MI lesions upon intracranial injection (Figure 3B and C). Following the removal of the cytokines from the brain slice, zsGreen-tagged PC9 or LLC lung cancer cells were cultured as a 3D tumor plug adjacent to the brain slice (Supplementary Figure 3A). For both models, there was minimal invasion into the brain slice in the absence of pre-treatment with the cytokine cocktail (Figure 3D and E). Cytokine cocktail pre-treatment of the brain slice resulted in a significant increase in both the number and the summed distance of PC9 and LLC cells invading the brain beyond 200 µm (Figure 3D–G), without influencing cancer cell proliferation (Supplementary Figure 3B and C).

We next leveraged the cKO-STAT3 mouse model to investigate how the loss of STAT3 in RAs impacts cancer cell invasion. To accomplish this, we seeded cancer cells directly on top of the brain slice and a z-series was captured up to the depth of 27 µm vertically into the slice⁸ (Supplementary Figure 3D). Administration of 4-hydroxytamoxifen (OH-Tmx) induced loss of STAT3 in RAs of the brain slice (Supplementary Figure 3E). Given that STAT3 knockout in RAs exerts a proliferation defect in cancer cells (Supplementary Figure 3E and F), which may confound the analysis of cancer cell invasion in this context,⁸ we used bioluminescent imaging to measure cancer cell signal between OH-Tmx and vehicle-treated brain slices, and only measured invasion in regions of the brain surface with equal cancer cell abundance (Supplementary Figure 3G and H). YUMM1.7 melanoma cells invaded the brain slices treated with OH-Tmx to a greater distance compared to brain slices treated with vehicle control (Supplementary Figure 3G, I, and J), consistent with pSTAT3⁺ RAs contributing to cancer cell invasion into the brain.

CHI3L1 Is Secreted by pSTAT3⁺ Reactive Astrocytes and Is Overexpressed in Astrocytes Surrounding Highly Invasive Brain Metastases

Given the established pro-tumorigenic paracrine signaling loops between RAs and BrM,^13–17 we investigated secreted factors from pSTAT3⁺ RAs that could potentially stimulate the invasive growth of BrM. Primary murine astrocytes were cultured as astrospheres in the presence or absence of the STAT3 activating cytokine cocktail (EGF, TGF-α, MIF), as previously described.⁸ Conditioned media (CM) collected from these astrospheres underwent proteomic analysis (Figure 4A). From this analysis, 266 proteins were found to be significantly upregulated in CM from pSTAT3⁺ vs. pSTAT3^- astrospheres,⁸ which included the secreted glycoprotein Chitinase 3-like-1 (CHI3L1) (Figure 4B). Although not the most highly upregulated protein in the resulting proteomics dataset, CHI3L1 was further explored because it is a known STAT3 transcriptional target^28,29 with established pro-tumorigenic roles in several cancer types, including promoting cancer cell invasion, metastasis, proliferation, cell survival, immunosuppression, and therapeutic resistance.^30–39 Indeed, in the context of glioma, CHI3L1 has been shown to promote invasive growth into the brain.^40,41 We therefore sought to evaluate the differential expression of CHI3L1 in HI versus MI BrM. We conducted IHF staining and RNAscope analysis on patient samples and intracranially injected PDX models, respectively. Quantification of IHF staining revealed elevated CHI3L1 levels in the stromal cells of HI vs. MI patient samples (n = 10 HI, n = 10 MI, Figure 4C). In agreement with these results, we observed increased Chi3l1 mRNA levels in brain parenchymal cells surrounding HI, but not MI, PDX models using RNAscope (n = 8 HI, n = 12 MI, Figure 4D, Supplementary Figure 4A). We additionally confirmed decreased Chi3l1 mRNA levels in the brains of cKO STAT3 mice (P = 0.0491 for HI YUMM1.7 cells, P = 0.0512 for MI MC38 cells, Supplementary Figure 4B and C), consistent with CHI3L1 being a STAT3 transcriptional target.^28,29

We next sought to identify the cells that express CHI3L1 in HI BrM samples. We performed single-cell RNA sequencing (scRNAseq) on surgically resected tissue from 4 patients with HI BrM, with 2 samples sequenced from each patient: 1 from the center of the lesion (metastatic center, MC) and 1 from the peripheral surrounding brain (SB) (Figure 5A and B). A total of 27,306 cells were sequenced across the 4 patients, representing BrM from different primary sites (n = 2 melanoma, n = 1 breast, n = 1 lung, Figure 5B). Cell types were identified by generating a cell atlas and using the expression of canonical cell-type markers on the uniform manifold approximation and projection (UMAP) (Supplementary Figure 5A). Ten clusters were identified, including 3 cancer cell clusters (breast cancer, lung cancer, and melanoma), and 7 non-cancer clusters (vascular smooth muscle cells (vSMCs), oligodendrocytes, macrophages/microglia, endothelial cells, astrocytes, B cells, and T cells/NK cells) (Figure 5C and D, Supplementary Figure 5A). Copy number variation analyses were performed to confirm the identity of cancer cells (Supplementary Figure 5B and C). All non-cancer cell clusters were comprised of cells obtained from all 4 patients, except the B cell cluster (Supplementary Figure 5D). Consistent with the HI invasion pattern, as determined by post-operative histopathological assessment, cancer cells were identified in the SB samples of 3 of the 4 patients, accounting for 23.4% of all cancer cells sequenced (Supplementary Figure 5D and E).

We assessed CHI3L1 expression across the 10 clusters and found that CHI3L1 was absent from all cancer cell clusters (Figure 5E), consistent with a stromal source of CHI3L1. Astrocytes, macrophages/microglia, and vSMCs were the only cell types in our samples found to express CHI3L1, consistent with previous reports of low CHI3L1 levels in other resident cell types within the brain, including lymphocytes.^42–44 The astrocyte cluster had the highest percentage of CHI3L1-expressing cells (14/84,16.66%), followed by the macrophages/microglia cluster, which displayed both lower frequency (339/5690, 5.96%) and lower expression levels compared to astrocytes (Figure 5E). Interestingly, a small number of vSMCs (9/188, 4.78%) were found to have high CHI3L1 expression (Figure 5E), consistent with previous work implicating CHI3L1 in vascular remodeling^45,46 and atherosclerotic plaque stability.⁴⁷ Together, these data support astrocytes as the predominant source of CHI3L1 in HI BrM.

Given the limited sample size available for scRNAseq (n = 4), we sought to validate these findings with a larger cohort and at the protein level. We performed multiplex IHF staining of patient BrM samples (n = 9 MI, n = 8 HI) for CHI3L1, pSTAT3, GFAP (astrocyte marker) and CD68 (macrophage marker) (Figure 5F). CHI3L1⁺/GFAP⁺/CD68^- cells were significantly more abundant than CHI3L1⁺/GFAP^-/CD68⁺ cells (Figure 5G), consistent with our scRNAseq data implicating astrocytes rather than macrophages as the predominant source of CHI3L1. These data are consistent with previous reports.^{29,34,42,43,48} Interestingly, IHF staining also revealed a subset of stromal CHI3L1-expressing cells that were GFAP^-/CD68^-, implicating cells other than astrocytes and macrophages as sources of CHI3L1 (Figure 5G). These CHI3L1⁺/GFAP^-/CD68^- stromal cells may represent several possible cell types previously identified as sources of CHI3L1, including vascular smooth muscle cells,⁴⁷ as identified by our scRNAseq data.

Importantly, quantification of pSTAT3 and CHI3L1 staining intensity and prevalence (H-Score) across all stromal cells revealed a positive correlation between the 2 proteins (R² = 0.4143, P = 0.0053 Figure 5H), consistent with CHI3L1 being a STAT3 target.^28,29 This correlation was similarly observed when the analysis was limited to GFAP⁺/CD68^- cells, (R² = 0.4427, P = 0.0036, Figure 5I).

CHI3L1 Promotes Cancer Cell Invasion in a Brain Slice Culture Model

PC9 lung cancer cells treated with recombinant CHI3L1 exhibited increased phosphorylation of AKT by immunoblot (Figure 6A), in agreement with previous findings demonstrating PI3K/AKT signaling pathway activation in BrM^49,50 downstream of CHI3L1.^35,51,52 This effect was observed using recombinant CHI3L1 derived from either human or mouse sources (Figure 6A, Supplementary Figure 6A), consistent with interspecies CHI3L1-mediated signaling effects. CHI3L1-mediated activation of the AKT pathway has previously been implicated in cell migration and invasion in the context of breast cancer.³⁵ We therefore sought to evaluate the contribution of CHI3L1 and AKT signaling on cancer cell migration and invasion in our BrM model systems. We assessed the in vitro efficacy of capivasertib (AKT inhibitor) in several cancer cell lines (Supplementary Figure 6B–D), demonstrating the concentrations of inhibitor required to inhibit phosphorylation of S6 kinase, a molecule downstream of AKT in its signaling cascade.⁵³ Next, we demonstrate that recombinant human CHI3L1 increased MDA-MB-231 breast cancer cell motility, a phenotype that could be neutralized with capivasertib (Supplementary Figure 6E). Invasion into the brain was evaluated using our in vitro brain slice-tumor plug co-culture assay (Figure 3A), in which tumor plugs placed adjacent to the coronal brain slice were treated with or without recombinant human CHI3L1. The addition of recombinant CHI3L1 increased cancer cell invasion into the brain using 2 models of MI BrM (PC9, and MC38 cells), both of which show minimal invasion in the absence of recombinant CHI3L1 (Figure 6B and C). Moreover, the addition of the AKT inhibitor, capivasertib, neutralized the increased invasion observed when recombinant CHI3L1 was added to the culture media (Supplementary Figure 6F and G). Together, these data demonstrate that CHI3L1 can promote invasive growth of BrM, in part, via activation of AKT signaling (Figure 6D).

Discussion

Invasion patterns in BrM have been described in multiple studies,^54–56 but only recently have HI BrM been associated with inferior clinical outcomes.^6,55 We therefore sought to investigate potential mechanisms driving BrM invasion that can be exploited for therapeutic benefit. Our results indicate that HI BrM are associated with more abundant pSTAT3⁺ RAs, when compared to MI BrM, which may provide a potential therapeutic avenue for the treatment of a subset of patients with BrM. The involvement of pSTAT3⁺ RAs in the invasive growth of BrM is consistent with previous evidence that this subpopulation of RAs promotes and maintains a pro-metastatic environment in the brain.^8,21

STAT3 inhibition for the treatment of BrM has shown promise in multiple preclinical/clinical studies and is the subject of an ongoing randomized controlled trial investigating its efficacy in preventing BrM local recurrence following surgical resection (NCT05689619). However, it is already known that not all BrM patients respond to STAT3 inhibition, and no clinical or histopathological features have yet been identified to predict responsiveness to treatment.^8,20 Using pre-clinical models, our data demonstrate that HI BrM are preferentially sensitive to STAT3 inhibition via Legasil or genetic ablation. This suggests that with further development, invasion pattern may serve as a predictive biomarker for STAT3 targeted therapy in the treatment of active BrM or in the adjuvant setting after BrM resection. Although we observed decreased tumor burden in HI BrM in response to STAT3 inhibition, future work leveraging an expanded cohort of orthotopic PDX BrM models will be required to assess whether targeting STAT3 confers prolonged overall survival in HI BrM.

Several therapeutic strategies exist to target STAT3. These include direct STAT3 inhibition with molecules such as silibinin,²³ via antisense oligonucleotides targeting STAT3 mRNA,⁵⁷ or through STAT3 protein degradation with the novel proteolysis targeting chimera (PROTAC) molecules.^58,59 These emerging developments in STAT3 targeting and intracranial drug delivery approaches may serve as future avenues of interest for the management of HI BrM.

An alternative approach to directly targeting STAT3 involves the identification of STAT3 transcriptional targets that promote the invasive growth of BrM. Such molecules may serve as relevant and readily inhibited therapeutic targets. We identified CHI3L1 as a target fulfilling these criteria by integrating proteomics from the secretome of mouse pSTAT3⁺ RAs with scRNAseq data from human HI BrM-associated RAs. CHI3L1 is a pSTAT3⁺ RA-derived secreted factor that phenocopies the pro-tumor functions of pSTAT3⁺ RAs in vitro. This suggests that CHI3L1 may serve as an alternative or adjunct therapeutic target for the treatment of HI BrM.

Indeed, CHI3L1 has emerged as a target of interest for anti-cancer therapy due to its pro-invasive and pro-metastatic roles in several cancer types, including hepatocellular carcinoma, lung cancer, gastric cancer, breast cancer, and glioma.^{30,31,33,40,41,60} The data described herein suggests that these pro-tumorigenic functions, at least in part, extend to metastatic brain tumors, offering further incentive for the development of CHI3L1-targeting therapeutics. Monoclonal antibodies targeting CHI3L1 have been described and are currently in pre-clinical development,^60,61 suggesting that such therapeutic agents may find utility in the treatment of central nervous system (CNS) tumors that originate in, or metastasize to, the brain.

Given the diverse pro-tumorigenic roles of CHI3L1,^30–38 the impact of CHI3L1 on the invasive growth of BrM is likely multifaceted. CHI3L1 has been implicated in cancer cell invasion in several cancer types, including primary brain tumors.^40,41 The pro-invasive effects of CHI3L1 are often ascribed to activation of the pro-invasive PI3K/AKT pathway.⁵¹ Indeed, herein we observe CHI3L1-mediated phosphorylation of AKT and induction of cancer cell invasion into the brain, a phenotype that can be neutralized following AKT inhibition. These data suggest that CHI3L1 may regulate BrM growth through additional mechanisms beyond simply regulating invasion. One possibility includes CHI3L1-mediated cancer cell proliferation or inhibition of apoptosis. Notably, astrocyte-derived CHI3L1 has been shown to promote glioblastoma cell proliferation and survival,³⁹ suggesting the possibility of similar functions in BrM. CHI3L1-mediated angiogenesis, an effect previously reported in breast cancer, cervical cancer, and glioblastoma, may also contribute to the CHI3L1-mediated pro-growth phenotypes observed in this study.^37,62,63 Additionally, there is growing evidence that CHI3L1 can create an immunosuppressive microenvironment in primary brain tumors.^36,41,64 CHI3L1 promotes activation and differentiation of pro-tumorigenic immune cells, including M2 macrophages, and is associated with decreased anti-tumorigenic cells in the tumor immune microenvironment.^41,60 Moreover, it was recently identified that CHI3L1 regulates immune checkpoint molecules, with ongoing work investigating exploiting this function for anti-cancer therapies.⁶¹

Together, this work identifies a tumor-stromal crosstalk mechanism that contributes to the invasive growth of BrM and which may provide opportunities for therapeutic intervention. Moreover, this work highlights that histopathological growth patterns may serve as a predictive biomarker for STAT3 or CHI3L1 inhibition in the context of BrM.

Conflict of interest statement

The authors have declared that no conflicts of interest exist.

Funding

M.D. and S.M.M acknowledge funding from the McGill MD/PhD Program and Vanier Canada Graduate Scholarships, and M.D. recognizes support from the Brain Tumor Foundation of Canada. N.P. is funded by the Spanish Association Against Cancer (AECC #POSTD19016PRIE). P.M.S. is a McGill University William Dawson Scholar. This research has been supported by grants from the Terry Fox Foundation and the Quebec Breast Cancer Foundation (Grant #: 251427-251690) and the Canadian Institutes of Health Research (CIHR PJT-175066) to P.M.S.

Acknowledgments

We would like to thank the patients and their families who donated the tissues studied in this work. We are grateful to the Goodman Cancer Institute and McGill University Life Sciences Complex core facilities for their support, including the Histology Core, the Advanced BioImaging Facility, and the McGill Comparative Medicine and Animal Resource Centre. We thank Talia James for technical assistance, Margarita Souleimanova, and Valentina Muñoz-Ramos for help with tissue biobanking and the operating room team at the MNIH. We thank Dr. Yojiro Yamanaka and Nobuko Yamanaka for TdTomato transgenic mice and Drs. Ian Watson, Sidong Huang, and Logan Walsh for cancer cell lines used in this study. We acknowledge Sherif Attalla and Tarek Taifour for helpful suggestions and members of the Siegel laboratory for thoughtful discussions and critical comments on the manuscript.

Author Contributions

Designing research studies: M.D., S.M.M., N.P., M.V., K.P. and P.M.S. Conducting Experiments and acquiring data: M.D., S.M.M., N.P,. G.K., R.Z., M.G.A., D.Z., A.N., M.B., C.M., P.L., F.C. and Y.I. Analyzing data: M.D., S.M.M., N.P., A.N-M., J.N., A.K., C.M., and L.T. Providing reagents: N.P., P.S., R.A.M., M-C.G. and K.P. Writing manuscript: M.D., S.M.M. and P.M.S. Reviewing and editing of manuscript: All authors. Supervision of study: M-C.G., R.A.M., W.J.M., M.P., M.V., K.P. and P.M.S.

References

Author notes