Abstract
Immunotherapies against brain metastases have shown clinical benefits when applied to asymptomatic patients, but they are largely ineffective in symptomatic cases for unknown reasons. Here, we dissect the heterogeneity in metastasis-associated astrocytes using single-cell RNA sequencing and report a population that blocks the antitumoral activity of infiltrating T cells. This protumoral activity is mediated by the secretion of tissue inhibitor of metalloproteinase-1 (TIMP1) from a cluster of pSTAT3+ astrocytes that acts on CD63+ CD8+ T cells to modulate their function. Using genetic and pharmacologic approaches in mouse and human brain metastasis models, we demonstrate that combining immune checkpoint blockade antibodies with the inhibition of astrocyte-mediated local immunosuppression may benefit patients with symptomatic brain metastases. We further reveal that the presence of tissue inhibitor of metalloproteinase-1 in liquid biopsies provides a biomarker to select patients for this combined immunotherapy. Overall, our findings demonstrate an unexpected immunomodulatory role for astrocytes in brain metastases with clinical implications.
Significance: This study presents a significant advancement in understanding immune modulation in brain tumors and offers new insights into the potential therapeutic interventions for brain metastases.
Introduction
The general dismal diagnosis of brain metastasis is starting to evolve into a more complex situation in which significant differences in prognosis exist depending on the state of the disease (i.e., local only vs. local and systemic; ref. 1), or the presence of vulnerabilities for which specific targeted drugs have demonstrated substantial benefits (1). Similarly, immunotherapies based on immune checkpoint blockade (ICB) have proved equally effective both on intracranial and on extracranial metastases in several clinical trials including patients with melanoma and lung cancer (2–8). Although variability of the responses is broad and not all patients benefit from it, the use of ICB to treat brain metastasis is widespread. However, many questions remain such as whether or not the therapeutic antibodies do get access to the brain or instead play their role extracranially and then activated T cells infiltrate the CNS (1, 9–11). Even more interesting is that almost all clinical trials have been developed on asymptomatic brain metastases. Thus, the symptomatic state, which is the clinically relevant one, remains poorly studied in the context of immunotherapy. Interestingly, in the limited reports in which ICB has been tested on symptomatic brain metastases, the therapeutic response rate dropped dramatically (2, 7). Although the reason behind the differential response of brain metastases to ICB is unknown, several reasons have been put forward. One of the main explanations is the use of corticoids in symptomatic brain metastasis as the cornerstone strategy to control edema. As potent immunosuppressor corticoids have been suggested to impair the effect of ICB, available preclinical data and metanalysis of clinical trials however cannot assign full responsibility to this (12–14).
The colonization of the brain by metastatic cells involves changes in the microenvironment. Initially, metastatic cells face a reactive glial response eliminating many of the cancer cells that completed extravasation (15). Subsequently, as the surviving cancer cells resume their growth, they start modifying the environment. As such, altered molecular patterns emerge de novo in specific cellular components of the brain. STAT3 is activated in a subpopulation of reactive astrocytes only in the advanced stages of the disease when metastasis has reached a certain size (16). This disease-associated altered molecular pattern contributes significantly to maintain the viability of metastasis by protecting cancer cells (16). Remarkably, this dependency on a component of the microenvironment was translated into a novel therapeutic opportunity validated in patients (16), which is now in clinical trials (NCT05689619).
Here, we report for the first time an unbiased approach to dissect the heterogeneity within metastasis-associated reactive astrocytes at the single cell level. We uncover various populations with distinct gene expression signatures suggesting previously unappreciated complexity at the functional level. Given the immediate clinical implications, we functionally dissected an immunomodulatory program present in a subpopulation within, previously identified as STAT3+ reactive astrocytes (16), acting on CD8+ T cells. We exploit this finding to favor the efficacy of ICB in patients with brain metastases and propose a novel combined immunotherapy compatible with advanced stages of the disease and agnostic to the primary source of metastasis. The core finding of the novel immunosuppressive mechanism demonstrated in relevant preclinical models and in patient-derived samples involves astrocyte-derived tissue inhibitor of metalloproteinase-1 (TIMP1) binding to the CD63 receptor on CD8+ T cells, which blocks their activated state. The validation of this phenotype using genetic and pharmacologic approaches allowed us to rationalize a novel combination immunotherapy to target local immunosuppression in the brain, thus favoring complementary efforts to activate T cells systemically. Such a strategy is complemented with the detection of TIMP1 in liquid biopsy to stratify such patients who could benefit the most from the combined immunotherapy.
In summary, our data not only uncover the unexplored role of reactive astrocytes as modulators of T-cell function in brain tumors by dissecting disease-associated glial heterogeneity but also exploit its functional implication on modulating brain-infiltrating T cells. We report the potential of developing organ-specific immunotherapies by dissecting the emerging cross-talk between two previously unconnected cell types in the tumor microenvironment.
Results
Clusters of Brain Metastasis–Associated Reactive Astrocytes Suggest Functional Diversity Including Immune Modulation
As previously reported by us and others (16–20), brain metastasis–associated astrocytes are heterogeneous. However, an unbiased approach to characterize this glial cell type in this pathological context has been lacking. We applied single-cell RNA sequencing (scRNA-seq) on melanoma brain metastasis generated by B16/F10-BrM (16) and enriched the resident glial population by ACSA2 (Fig. 1A), an established marker for astrocytes (21). Our approach efficiently enriched astrocytes in the single-cell population (Supplementary Fig. S1A and S1B) in a nonexclusive way because we detected other cell types after sequencing (Supplementary Fig. S1C). A total of 7,762 ACSA2+ astrocytes were profiled to identify nine clusters (Fig. 1B; Supplementary Table S1), in which three of them (clusters 3, 7, and 6) increased in the presence of brain metastasis (Supplementary Fig. S1D; Supplementary Tables S2 and S3). Interestingly, cluster 3 and cluster 7 represent a previously described subpopulation of brain metastasis–associated astrocytes characterized by enrichment in STAT3 expression and activation (Fig. 1B and C; Supplementary Fig. S1E and S1F; ref. 16). Given the enlarged complexity within the STAT3+ subpopulation, we dissected these two clusters, attending to their pathway enrichment. Interestingly, STAT3+ cluster 3 and STAT3+ cluster 7 astrocytes seem to represent functionally different subtypes with nonoverlapping top enriched pathways (Fig. 1D and E; Supplementary Fig. S1G; Supplementary Table S4). When analyzing cluster 6, corresponding to a STAT3− brain metastasis–enriched astrocytes cluster, the absence of STAT3+ identity pathways (i.e., interferon–antigen presentation, extracellular matrix, and cytokine/integrin signaling) was confirmed (Supplementary Fig. S1H; Supplementary Table S4). The functional diversity in STAT3+ clusters could be explained by the different patterns of receptors that activate STAT3 present in these subpopulations of reactive astrocytes. Although cluster 3 presents the highest expression of Il6r (Supplementary Fig. S1I–S1K), cluster 7 expresses growth factor receptors that are absent in cluster 3 (Supplementary Fig. S1I, S1J, and S1L). Additionally, the expression of interaction pairs between clusters 3 and 7 suggests a dynamic evolution of STAT3+ clusters that could potentially modulate each other (Supplementary Fig. S1M; Supplementary Table S5). STAT3+ astrocyte clusters (with high STAT3 expression and activation of STAT3 pathways) were further demonstrated in human brain metastases by scRNA-seq (Fig. 1F–I; Supplementary Fig. S1M–S1P; Supplementary Tables S6 and S7). Human STAT3+ brain metastasis–associated reactive astrocytes present an increased heterogeneity with a diverse set of functions that include those found in mice clusters (Supplementary Tables S8–S10). Thus, our findings suggest that STAT3+ clusters include a previously described protumoral component of astrocytes [cluster 3 in mice and cluster 4 in humans are enriched in interferon (Fig. 1D and I; Supplementary Tables S2, S4, and S8; ref. 22)] and also an unexplored compartment [cluster 7 in mice and cluster 5 in humans are enriched in extracellular matrix, cytokines, and interleukins (Fig. 1E and I; Supplementary Tables S3, S4, and S9)].
Clusters of brain metastasis–associated reactive astrocytes suggest functional diversity including immune modulation. A, Schema of the experimental design. Three different brains from C57BL/6J mice intracranially injected with B16/F10-BrM cells were enzymatically digested and pooled. ACSA2 labeling was used to enrich the sample in glial cells, obtaining 7,762 cells identified as astrocytes. A pool of three brains without tumor was used as the control for comparisons. B, Uniform manifold approximation and projection (UMAP) plot (0.2 resolution) of the different subpopulations of reactive astrocytes in brain metastasis. Dotted lines surround Stat3+ clusters. C, Stat3 expression in the different clusters of brain metastasis–associated astrocytes. Dot size represents the dimension of the subpopulation compared with total cells and a colored scale indicates the level of expression: blue, low expression and red, high expression. D and E, Representation of the top upregulated gene set enrichment analysis (GSEA) pathways in Stat3+ astrocytic clusters of brain metastasis according to the normalized enrichment score (NES) and a cutoff of P value < 0.05 and FDR <0.25. ECM, extracellular matrix. Colored pathways according to the biological category the gene sets belong to, correspond to more than half of the total pathways analyzed (total percentage of 100%). Detailed information of the pathways in Supplementary Table S4. F, Schema of the experimental design. Two human brain metastases from patients with lung cancer and breast cancer were fixed, digested and profiled for single-cell RNA-sequencing (scRNA-seq), 2,612 astrocytes and 1,338 astrocytes were identified, respectively. G, UMAP plot (k = 20) of the different subpopulations of reactive astrocytes in human brain metastasis. Dotted lines surround clusters with STAT3 high expression. H, STAT3 expression in the different clusters of brain metastasis-associated astrocytes. Dot size represents the dimension of the subpopulation compared with total cells and a colored scale indicates the level of expression: blue, low expression and red, high expression. I, Normalized enrichment score (NES) of GSEA pathways comparing clusters 3, 4, and 5 of human brain metastases–associated astrocytes. KEGG_Cytokine–cytokine receptor interaction, p.adjust = 1,05E−05; Reactome_Extracellular matrix (ECM) organization, p.adjust = 1,03E−03; Reactome_Signaling by Interleukins, p.adjust = 8,65E−03; Reactome_Antigen processing: Ubiquitination and Proteasome degradation, p.adjust = 1,67E−02; Reactome_Cell Cycle Checkpoints, p.adjust = 3,43E−03; Hallmark_Epithelial_mesenchymal_transition (EMT), p.adjust = 5E−09; KEGG_ECM–receptor interaction, p.adjust = 7.08E−05; HALLMARK_Interferon_alpha response, p.adjust = 1.11E−02; KEGG_Proteasome, p.adjust = 6.53E−04; HALLMARK_Myc Targets V1, p.adjust = 2.88E−07.
Clusters of brain metastasis–associated reactive astrocytes suggest functional diversity including immune modulation. A, Schema of the experimental design. Three different brains from C57BL/6J mice intracranially injected with B16/F10-BrM cells were enzymatically digested and pooled. ACSA2 labeling was used to enrich the sample in glial cells, obtaining 7,762 cells identified as astrocytes. A pool of three brains without tumor was used as the control for comparisons. B, Uniform manifold approximation and projection (UMAP) plot (0.2 resolution) of the different subpopulations of reactive astrocytes in brain metastasis. Dotted lines surround Stat3+ clusters. C, Stat3 expression in the different clusters of brain metastasis–associated astrocytes. Dot size represents the dimension of the subpopulation compared with total cells and a colored scale indicates the level of expression: blue, low expression and red, high expression. D and E, Representation of the top upregulated gene set enrichment analysis (GSEA) pathways in Stat3+ astrocytic clusters of brain metastasis according to the normalized enrichment score (NES) and a cutoff of P value < 0.05 and FDR <0.25. ECM, extracellular matrix. Colored pathways according to the biological category the gene sets belong to, correspond to more than half of the total pathways analyzed (total percentage of 100%). Detailed information of the pathways in Supplementary Table S4. F, Schema of the experimental design. Two human brain metastases from patients with lung cancer and breast cancer were fixed, digested and profiled for single-cell RNA-sequencing (scRNA-seq), 2,612 astrocytes and 1,338 astrocytes were identified, respectively. G, UMAP plot (k = 20) of the different subpopulations of reactive astrocytes in human brain metastasis. Dotted lines surround clusters with STAT3 high expression. H, STAT3 expression in the different clusters of brain metastasis-associated astrocytes. Dot size represents the dimension of the subpopulation compared with total cells and a colored scale indicates the level of expression: blue, low expression and red, high expression. I, Normalized enrichment score (NES) of GSEA pathways comparing clusters 3, 4, and 5 of human brain metastases–associated astrocytes. KEGG_Cytokine–cytokine receptor interaction, p.adjust = 1,05E−05; Reactome_Extracellular matrix (ECM) organization, p.adjust = 1,03E−03; Reactome_Signaling by Interleukins, p.adjust = 8,65E−03; Reactome_Antigen processing: Ubiquitination and Proteasome degradation, p.adjust = 1,67E−02; Reactome_Cell Cycle Checkpoints, p.adjust = 3,43E−03; Hallmark_Epithelial_mesenchymal_transition (EMT), p.adjust = 5E−09; KEGG_ECM–receptor interaction, p.adjust = 7.08E−05; HALLMARK_Interferon_alpha response, p.adjust = 1.11E−02; KEGG_Proteasome, p.adjust = 6.53E−04; HALLMARK_Myc Targets V1, p.adjust = 2.88E−07.
Given that the link between STAT3+ astrocytes and the immune system that we have previously suggested (16) was reinforced through the dissection of this astrocyte subpopulation at the single-cell level with the identification of various immunomodulatory molecules, we decided to functionally test this possibility. We confirmed the immunosuppressive nature of the secretome from pSTAT3+ astrocytes by interrogating CD8+ T cells in vitro (16) at the molecular level when incubated with the astrocyte conditioned media (CM; Fig. 2A and B; Supplementary Fig. S2A; Supplementary Table 11). To confirm this finding in vivo, we evaluated whether CD8+ T cells associated with brain metastasis were dependent on the presence and activity of STAT3+ astrocytes using the STAT3 inhibitor silibinin (16, 23). Although other cell types could be affected by silibinin, the levels of pSTAT3 observed in astrocytes are much higher than in CD8+ T cells (Supplementary Fig. S2B; ref. 16), which could suggest an increased functional dependency on this pathway. Additionally, we previously demonstrated that genetically engineering STAT3 loss of function in astrocytes phenocopied the pharmacological intervention (16). With this limitation in mind, we profiled the B16/F10-BrM brain metastasis–associated immune compartment, which includes CD8+ T cells (Supplementary Fig. S2C and S2D) among other cell types (Supplementary Fig. S2E), from mice treated with silibinin (Fig. 2C). Our findings demonstrate that pharmacological blockade of STAT3 alters specifically the proportion of T-cell subpopulations in the brain, increasing those clusters expressing known cytotoxic markers (Cxcr6, Gzmk, Gzma, Gzmb, Ccl5, Gimap7, Xcl1, Klrc1, Klrk1, and Cd160), which are not found upregulated in naïve T-cell clusters (Fig. 2D; Supplementary Fig. S2F and S2G). In order to evaluate the lack of cell type specificity of pharmacological intervention, we validated the STAT3-dependent modulation of tumor-infiltrating lymphocytes using STAT3 depleted mice in reactive astrocytes (GFAP-CreERT2; Stat3loxP/loxP, abbreviated as cKOGFAP-Stat3; Fig. 2E; ref. 16). We observed a general increase of brain metastasis–associated CD8+ T cells (Supplementary Fig. S2H and S2I) accompanied with the induction of Granzyme B (Fig 2F and G), which was in agreement with the strong increase of granzyme genes Gzmb and Gzmk (Supplementary Fig. S2J and S2K). However, no significant alteration in perforin and IFNγ-expressing CD8+ T cells was observed (Supplementary Fig. S2L and S2M). Thus, inhibition of STAT3, using either pharmacologic or genetic interventions, in brain metastasis–associated reactive astrocytes modulates the phenotype of CD8+ T cells in vivo. In order to demonstrate the functional relevance of this finding, we evaluated the ability of a CD8-blocking antibody to rescue the reduced brain metastases burden in cKOGFAP-Stat3 (Fig. 2H). Remarkably, blocking CD8+ T cells in cKOGFAP-Stat3 mice reverted the antimetastasis phenotype suggesting that the infiltrating immune population is actively suppressed by STAT3+ astrocytes in vivo (Fig. 2I and J; Supplementary Fig. S2N).
The protumoral role of STAT3+ reactive astrocytes involves immune modulation. A, Schema of the experimental design. Green cells: pSTAT3− astrocytes; red cells: pSTAT3+ astrocytes. Preactivated CD8+ lymphocytes incubated with CM generated by pSTAT3− and pSTAT3+ astrospheres (as described in the “Methods” section) were processed for bulk RNA-seq. B, GSEA of biological process of T-cell activation downregulated in T cells incubated with pSTAT3+ astrospheres CM compared with pSTAT3− astrospheres CM. n = 3 independent T cells in in vitro cultures per condition. C, Schema of the experimental design. C57BL/6J mice were intracranially injected with B16/F10-BrM cells, control brains and brains from mice treated during 6 days with the STAT3 inhibitor, silibinin (Legasil 200 mg/kg, daily) were processed to obtain the immune infiltrate fraction, which was depleted from monocytes. Rhapsody system was used to single-cell sequence a total of 3,055 immune cells identifying different CD3+ T-cell clusters. D, Quantification showing the percentage of cytotoxic-like T cells (clusters 4, 7, and 13; Supplementary Fig. S2C–S2E) in the brains of control and silibinin treated mice. Values are shown in box-and-whisker plots, in which each dot is a mouse and the line in the box corresponds to the median. The boxes go from top to bottom quartiles, and the whiskers go from minimum to maximum values (n = 8 control mice; n = 9 mice treated with silibinin). P value was calculated using two-tailed t test between control and silibinin experimental groups. E, Schema of the experimental design. Tamoxifen (Tmx)-treated and untreated cKOGFAP-Stat3 mice intracranially injected with B16/F10-BrM cells were sacrificed at the experimental endpoint, their brains were processed to obtain the immune fraction for flow cytometry analysis or sorted for CD3+CD8+ lymphocytes for RNA isolation and qRT-PCR analysis of gene expression. F, Representative flow cytometry analysis of Granzyme B expression in CD3+CD8+ T cells from control and cKOGFAP-Stat3 brains intracranially injected with B16/F10-BrM cells. G, Quantification of the experiment in F. Error bars, SEM. Every dot is a different animal (n = 8). P value was calculated using the two-tailed t test. H, Schema of the experimental design. Brains from untreated or Tmx-treated cKOGFAP-Stat3 with IgG2 or anti-CD8 (10 mg/kg, every 2 days starting at day 3 after inoculation of cancer cells), 2 weeks after being inoculated with B16/F10-BrM cells intracardially, were analyzed. I, Representative images of ex vivo brains in H. Images show the BLI intensity. J, Quantification of ex vivo BLI. Values are shown in box-and-whisker plots in which every dot represents a different animal. Values were obtained from normalizing the ex vivo brain signal to the in vivo head signal 3 days after intracardiac injection when treatment was initiated (n = 39/28/28 mice per experimental condition, eight independent experiments). P value was calculated using the two-tailed t test.
The protumoral role of STAT3+ reactive astrocytes involves immune modulation. A, Schema of the experimental design. Green cells: pSTAT3− astrocytes; red cells: pSTAT3+ astrocytes. Preactivated CD8+ lymphocytes incubated with CM generated by pSTAT3− and pSTAT3+ astrospheres (as described in the “Methods” section) were processed for bulk RNA-seq. B, GSEA of biological process of T-cell activation downregulated in T cells incubated with pSTAT3+ astrospheres CM compared with pSTAT3− astrospheres CM. n = 3 independent T cells in in vitro cultures per condition. C, Schema of the experimental design. C57BL/6J mice were intracranially injected with B16/F10-BrM cells, control brains and brains from mice treated during 6 days with the STAT3 inhibitor, silibinin (Legasil 200 mg/kg, daily) were processed to obtain the immune infiltrate fraction, which was depleted from monocytes. Rhapsody system was used to single-cell sequence a total of 3,055 immune cells identifying different CD3+ T-cell clusters. D, Quantification showing the percentage of cytotoxic-like T cells (clusters 4, 7, and 13; Supplementary Fig. S2C–S2E) in the brains of control and silibinin treated mice. Values are shown in box-and-whisker plots, in which each dot is a mouse and the line in the box corresponds to the median. The boxes go from top to bottom quartiles, and the whiskers go from minimum to maximum values (n = 8 control mice; n = 9 mice treated with silibinin). P value was calculated using two-tailed t test between control and silibinin experimental groups. E, Schema of the experimental design. Tamoxifen (Tmx)-treated and untreated cKOGFAP-Stat3 mice intracranially injected with B16/F10-BrM cells were sacrificed at the experimental endpoint, their brains were processed to obtain the immune fraction for flow cytometry analysis or sorted for CD3+CD8+ lymphocytes for RNA isolation and qRT-PCR analysis of gene expression. F, Representative flow cytometry analysis of Granzyme B expression in CD3+CD8+ T cells from control and cKOGFAP-Stat3 brains intracranially injected with B16/F10-BrM cells. G, Quantification of the experiment in F. Error bars, SEM. Every dot is a different animal (n = 8). P value was calculated using the two-tailed t test. H, Schema of the experimental design. Brains from untreated or Tmx-treated cKOGFAP-Stat3 with IgG2 or anti-CD8 (10 mg/kg, every 2 days starting at day 3 after inoculation of cancer cells), 2 weeks after being inoculated with B16/F10-BrM cells intracardially, were analyzed. I, Representative images of ex vivo brains in H. Images show the BLI intensity. J, Quantification of ex vivo BLI. Values are shown in box-and-whisker plots in which every dot represents a different animal. Values were obtained from normalizing the ex vivo brain signal to the in vivo head signal 3 days after intracardiac injection when treatment was initiated (n = 39/28/28 mice per experimental condition, eight independent experiments). P value was calculated using the two-tailed t test.
TIMP1 and STAT3 in Reactive Astrocytes Correlate with a High Immune Cluster Classifier in Human Brain Metastases
Within the secretome of pSTAT3+ astrospheres (16), several candidates were previously suggested to play a role on the immunosuppressive properties of this glial cell subpopulation (16). Among them, we became particularly interested in TIMP1 because it was recently reported as one of the top deregulated proteins within the CD45− cell fraction of human brain metastases, which includes astrocytes (24). Our proteomics data (16) show high TIMP1-specific enrichment in pSTAT3+ astrospheres (Supplementary Fig. S3A). We further prove that TIMP1 derives from the microenvironment in human brain metastases (Supplementary Fig. S3B and S3C) and that its highest expression in available scRNA-seq data from experimental brain metastases (Supplementary Fig. S1C) corresponds to astrocytes as compared with other glial cells or macrophages (Supplementary Fig. S3D). Indeed, Timp1 expression colocalizes with pSTAT3+ astrocytes in astrospheres and in experimental brain metastasis (Fig. 3A), in particular with STAT3+ cluster 7 (Supplementary Fig. S3E; Supplementary Table S3). Furthermore, we confirmed that the major source of TIMP1 in human brain metastases are pSTAT3+ reactive astrocytes (Fig. 3A; Supplementary Fig. S3F–S3I), in which the TIMP1 highest expression is found in the cluster of astrocytes with greatest induction of STAT3 (cluster 5; Supplementary Fig. S3J). To demonstrate the contribution of astrocytes to microenvironment-derived TIMP1, we used the genetically modified mouse model (GEMM) GFAP-Cre; Timp1-loxP/loxP (for brevity, cKOGFAP-Timp1; Supplementary Fig. S3K–S3Q; ref. 25). We validated the absence of TIMP1 in the CM of pSTAT3+ astrospheres (16) derived from cKOGFAP-Timp1 GEMM (Supplementary Fig. S3O and S3P), in which we were unable to detect any influence of TIMP1 on the established phenotype of this in vitro surrogate for pSTAT3+ astrocytes (Supplementary Fig. S3N; ref. 16). Accordingly, GFAP+ pSTAT3+ brain metastasis–associated reactive astrocytes in cKOGFAP-Timp1 GEMM remain indistinguishable from the wild-type (WT) ones (Supplementary Fig. S3K–S3M). No additional analyses were performed to characterize astrocytes in the cKOGFAP-Timp1 GEMM. Importantly, depleting Timp1 from astrocytes decreases brain metastasis–induced TIMP1 to nontumor levels in the cerebrospinal fluid (CSF; Supplementary Fig. S3Q). Finally, we confirmed the STAT3 dependency of TIMP1 in vivo with both cKOGFAP-Stat3 mice (Supplementary Fig. S3O) and pharmacological inhibition of STAT3 (Supplementary Fig. S3R–S3T).
TIMP1 and STAT3 in reactive astrocytes correlate with a high immune cluster classifier in human brain metastases. A, Representative images showing pSTAT3+ TIMP1+ reactive astrocytes (arrowheads) in different samples: astrospheres enriched in STAT3, established brain metastasis induced by intracardiac inoculation of B16/F10-BrM cells and human breast cancer brain metastasis. Dotted line surrounds the cancer cells (CC). Scale bar, 20 μm. B, Schema of the experimental design. Sequencing data from patients’ samples with brain metastases were stratified into low, medium, and high immune categories or clusters. Immune clusters were calculated according to an initial algorithm and then complemented with a three-gene classifier representing key cell types of the microenvironment. C and D, STAT3 (C) and TIMP1 (D) expression in human samples from low, medium, and high immune clusters. Values are shown in box-and-whisker plots, in which each dot is a patient and the line in the box corresponds to the median. The boxes go from top to bottom quartiles, and the whiskers go from minimum to maximum values (n = 32 samples, low; n = 64 samples, medium; n = 12 samples, high). P value was calculated using the two-tailed t test. One-way ANOVA is shown to compare the three immune categories. E, Schema of the experimental design. A cohort of 12 human samples with extended resection including peritumoral microenvironment was used to validate sequencing data with IHC profile. In the IHC image, STAT3+ reactive astrocytes are shown. cc, cancer cells; RA, reactive astrocytes. Scale bar, 40 μm. F, Multiplex representative images of low/medium/high immune clusters in the cohort of human samples in E. STAT3 staining and TIMP1 RNAscope were performed in consecutive sections and allocated on the specific patient categories. n = 4 samples in low immune cluster, n = 4 samples in medium immune cluster, n = 4 samples in high immune cluster. Scale bar, 50 μm, magnification 15 μm. G, Graph showing the correlation between the percentage of immune cells as quantified by multiplex and the percentage of TIMP1+ events per cell in the microenvironment of 12 brain metastasis samples. Dots are colored according to the immune cluster calculated for the cohort of samples: low (green)/medium (grey)/high (red) immune clusters. P value was calculated using the two-tailed t test. H, Representative image of a patient with melanoma brain metastasis treated with ICB showing pSTAT3+ reactive astrocytes surrounding brain metastasis lesion next to CD8+ T cells. The patient showed extracranial response but failed to respond to ICB intracranially. The dotted line surrounds the cancer cells (cc). Scale bar, 15 μm. I, Representative image of multiplex in a sample of a patient in H. Magnification showing CD8+ Granzyme B+ T cells (yellow arrowheads) and CD8+ Granzyme B 3 T cells (pink arrowheads). Scale bar, 20 μm. J, Quantification of experiment in I. The graph represents the number of pSTAT3+ reactive astrocytes surrounding a CD8+ T cell with or without Granzyme B positivity in a ratio of 100 μm. A total of 40 CD8+ T cells from five different patients in which GRZ+CD8+ T cells could be identified belonging to the cohort in H were quantified. Error bars, SEM. Every dot is a different CD8+ T cell. P value was calculated using the two-tailed t test.
TIMP1 and STAT3 in reactive astrocytes correlate with a high immune cluster classifier in human brain metastases. A, Representative images showing pSTAT3+ TIMP1+ reactive astrocytes (arrowheads) in different samples: astrospheres enriched in STAT3, established brain metastasis induced by intracardiac inoculation of B16/F10-BrM cells and human breast cancer brain metastasis. Dotted line surrounds the cancer cells (CC). Scale bar, 20 μm. B, Schema of the experimental design. Sequencing data from patients’ samples with brain metastases were stratified into low, medium, and high immune categories or clusters. Immune clusters were calculated according to an initial algorithm and then complemented with a three-gene classifier representing key cell types of the microenvironment. C and D, STAT3 (C) and TIMP1 (D) expression in human samples from low, medium, and high immune clusters. Values are shown in box-and-whisker plots, in which each dot is a patient and the line in the box corresponds to the median. The boxes go from top to bottom quartiles, and the whiskers go from minimum to maximum values (n = 32 samples, low; n = 64 samples, medium; n = 12 samples, high). P value was calculated using the two-tailed t test. One-way ANOVA is shown to compare the three immune categories. E, Schema of the experimental design. A cohort of 12 human samples with extended resection including peritumoral microenvironment was used to validate sequencing data with IHC profile. In the IHC image, STAT3+ reactive astrocytes are shown. cc, cancer cells; RA, reactive astrocytes. Scale bar, 40 μm. F, Multiplex representative images of low/medium/high immune clusters in the cohort of human samples in E. STAT3 staining and TIMP1 RNAscope were performed in consecutive sections and allocated on the specific patient categories. n = 4 samples in low immune cluster, n = 4 samples in medium immune cluster, n = 4 samples in high immune cluster. Scale bar, 50 μm, magnification 15 μm. G, Graph showing the correlation between the percentage of immune cells as quantified by multiplex and the percentage of TIMP1+ events per cell in the microenvironment of 12 brain metastasis samples. Dots are colored according to the immune cluster calculated for the cohort of samples: low (green)/medium (grey)/high (red) immune clusters. P value was calculated using the two-tailed t test. H, Representative image of a patient with melanoma brain metastasis treated with ICB showing pSTAT3+ reactive astrocytes surrounding brain metastasis lesion next to CD8+ T cells. The patient showed extracranial response but failed to respond to ICB intracranially. The dotted line surrounds the cancer cells (cc). Scale bar, 15 μm. I, Representative image of multiplex in a sample of a patient in H. Magnification showing CD8+ Granzyme B+ T cells (yellow arrowheads) and CD8+ Granzyme B 3 T cells (pink arrowheads). Scale bar, 20 μm. J, Quantification of experiment in I. The graph represents the number of pSTAT3+ reactive astrocytes surrounding a CD8+ T cell with or without Granzyme B positivity in a ratio of 100 μm. A total of 40 CD8+ T cells from five different patients in which GRZ+CD8+ T cells could be identified belonging to the cohort in H were quantified. Error bars, SEM. Every dot is a different CD8+ T cell. P value was calculated using the two-tailed t test.
As we hypothesized that STAT3+ astrocytes are major contributors to local immunosuppression, we asked whether this astrocyte population correlated with the degree of immune infiltration in the microenvironment of human brain metastases. We interrogated the expression of STAT3 and TIMP1 in patient samples previously profiled with transcriptomics, annotated with respect to low, medium, and high immune categories (Fig. 3B; Supplementary Fig. S4A; ref. 26). Remarkably, both STAT3 and TIMP1 expression levels were enriched among human brain metastases classified as the high immune fraction (Fig. 3C and D). Of note, scored samples were compatible with reporting gene expression patterns from the microenvironment compartment (Supplementary Fig. S4B). The correlation between the genes of interest and the immune compartment was validated in a second cohort of human brain metastases (Supplementary Fig. S4C and S4D; Supplementary Table S12). This finding could suggest that the expression of STAT3 and TIMP1 genes is compatible with a dense immune landscape, broadly speaking, which could potentially involve the ability of these fractions of brain metastases to respond to immune checkpoint inhibitors if properly stimulated. Interestingly, we realized that the definition of human samples according to the different immune categories was reproduced by a reduced gene classifier composed of genes representative of key cell types from the microenvironment including CD8a (for CD8+ T cells and some subsets of dendritic cells), CD68 (for microglia/macrophages), and ITGAX (mainly for dendritic cells, and also for macrophages, NK cells, and granulocytes; Fig. 3B; Supplementary Fig. S4E and S4F). The use of these reduced number of markers to assess the immune infiltration of human brain metastasis could provide a clinically compatible assay that might be useful to stratify patients. Consequently, we develop a multiplex analysis applying the corresponding antibodies for these cell types to a cohort of 12 selected brain metastases in RENACER (list of supplementary figures, supplementary tables, and authors included in the RENACER signature; ref. 27). The selection criteria applied responded to the inclusion of samples obtained through extended neurosurgical resection (Fig. 3E; Supplementary Table S13) to make sure a substantial peritumoral microenvironment, in which astrocytes are exclusively located, was present (Fig. 3E; ref. 15). Samples were categorized into low/medium/high based on the combined score of the three antibodies (Fig. 3F; Supplementary Fig. S4G), which nicely correlated with the transcriptomic scoring (Fig. 3G). Analysis of the abundance of TIMP1 in the microenvironment of these samples confirmed correlation with the high immune cluster (Fig. 3G). Based on these findings, we hypothesized that patients with brain metastasis treated with immune checkpoint blocking antibodies, even in the presence of an immune-rich microenvironment, might not benefit from this immunotherapy given the concomitant presence of a local immunosuppressive compartment (i.e., pSTAT3+ astrocytes). Although an adequate comparison with the responders is a requisite, to preliminarily evaluate our hypothesis, we identified in RENACER eight patients affected with extracranial metastases that responded to ICB systemically but that later relapsed in the brain (Fig. 3H; Supplementary Table S14). Our ability to get access to these tissues from the RENACER cohort (27) allowed us to confirm the presence of pSTAT3+ reactive astrocytes enriched in TIMP1 (Fig. 3H). As CD8+ T cells are present in limited numbers infiltrating the tumor core, but mainly in the peritumoral area intermingled with reactive astrocytes (Supplementary Fig. S4H), we hypothesized that a correlation between the potential antitumor quality of CD8+ T cells and the distance to pSTAT3+ reactive astrocytes might exist. Interestingly, we found that this cohort of patients showed an inverse correlation between the density of pSTAT3+GFAP+ cells and granzyme-positive CD8+ T cells (Fig. 3I and J), by focusing on areas within the range of influence of cytokines (28). Thus, our findings provide the rationale to improve responses to ICB in brain metastases with high immune infiltration by targeting STAT3+ astrocyte–dependent local immunosuppression.
TIMP1 Mediates Brain Metastasis in a CD8+ T-Cell–Dependent Manner
To address the potential contribution of astrocyte-derived TIMP1 to the immunosuppressive phenotype on CD8+ T cells (Fig. 2A–J), we performed in vitro cytotoxicity assays. CD8+ T-cell cytotoxicity was analyzed using OT-I transgenic CD8+ T cells specific for the ovalbumin (OVA)-derived SIINFEKL peptide (29) and targeted B16/F10-BrM-OVAGFP cells (Fig. 4A; Supplementary Fig. S5A and S5B). As previously reported, activated CD8+ T cells cultured in the secretome of pSTAT3+ astrospheres reduced their cytotoxicity compared with pSTAT3− secretome addition, on a melanoma brain metastatic cell line (Fig. 4B; Supplementary Fig. S5C and S5D; ref. 16). We found that the addition of TIMP1 mimics the effect of the immunosuppressive pSTAT3+ secretome (Fig. 4B; Supplementary Fig. S5C and S5D), in the same line as described by Oelmann and colleagues (30). In addition, pSTAT3+ astrospheres generated from cKOGFAP-Timp1 were unable to influence the cytotoxicity of activated T cells (Fig. 4B; Supplementary Fig. S5C and S5D). These results were complemented within in vitro experiments with activated T cells, wherein anti-TIMP1 blocking antibody reverted the effect of the otherwise immunosuppressive pSTAT3+ secretome (Supplementary Fig. S5E–S5G). In order to expand this finding to more relevant models, we applied the blocking antibody against TIMP1 to organotypic cultures of both experimental (Fig. 4C) and patient-derived brain metastases ex vivo (Fig. 4D; Supplementary Table S15) that included the surrounding microenvironment in which astrocytes and T cells coexist (Supplementary Fig. S5H and S5I; ref. 16). Remarkably, blocking TIMP1 activity correlated with reduced metastasis-derived bioluminescence that was rescued by blocking CD8+ T cells (Fig. 4E). Targeting human TIMP1 in 11 patient-derived brain metastasis organotypic cultures (PDOC) from different primary tumors confirmed the decrease in viability of metastases (Fig. 4F). We further demonstrate that the phenotype was not direct on cancer cells because anti-TIMP1 blocking antibody did not significantly influence metastatic cells in isolation (Supplementary Fig. S5J–S5L). Consistent with the mouse model, the reduction in viability of human brain metastatic cells was rescued by targeting the CD8+ T cells infiltrating the PDOC in an additional cohort of seven patients (Fig. 4G). Remarkably, patients stratified as high immune cluster (Supplementary Fig. S5M; Supplementary Table S16), which we hypothesized responded better to anti-TIMP1 blockade in PDOCs, showed a greater decrease in cancer cell viability when compared with patients with limited CD8+ T-cell infiltration, low levels of STAT3 and TIMP1, and similar levels of dendritic cell and macrophage markers (Supplementary Fig. S5N; Supplementary Table S16). To expand the involvement of TIMP1 in vivo, we performed metastasis assays with two experimental models. A melanoma brain metastasis model (16) and a triple-negative breast cancer model (31) were inoculated in the cKOGFAP-Timp1 GEMM (Fig. 4H; ref. 25). Brains with conditional knockout of Timp1 in reactive astrocytes correlated with a decreased ability of metastatic cells to survive in this organ (Fig. 4I–L; Supplementary Fig. S5O and S5P). An analysis of the histology showed increased numbers of antitumoral brain metastasis–associated CD8+ T cells infiltrating the metastasis (Fig. 4M and N), which strongly suggest a potential negative influence of resident glial cells on the acquired immune system at the core of local immunosuppression.
TIMP1 mediates brain metastasis in a CD8+ T-cell–dependent manner. A, Schema of the experimental design. pSTAT3− and pSTAT3+wt and pSTAT3+ cKOGFAP-Timp1 CM (with or without rTIMP1 100 ng/mL or control IgG/anti-TIMP1 10 μg/mL) was added to CD8+ T cells and cultured with BrM cells in a 1:4 ratio (BrM-OVA cancer cells: OT-I T cells specific for the OVA-derived SIINFEKL peptide) or a 1:5 ratio (BrM cancer cells:CD8+ T cells previously activated). B, Quantification of the BLI signal from the experiment shown in A and representative images of B16/F10-BrM-OVA–derived BLI at the initial time point and 24 hours after adding CD8+ lymphocytes preincubated with CM. Light orange condition refers to coculture of OT-I T cells with B16/F10-BrM no OVA (control for antigen-specific killing). Values correspond to 24 hours BLI normalized to BLI before adding CD8+ T cells expressed in percentage with respect to the mean of control experimental condition (BrM cells). Error bars, SEM. n = 3 different cocultures per condition. P value was calculated using the two-tailed t test. C and D, Schema of the experimental design. Control IgG or anti-TIMP1 (10 μg/mL) was added to the medium in organotypic cultures of mouse brain with B16/F10-BrM established lesions (C) and PDOC that include the brain metastasis–associated microenvironment (D). E, Quantification of the BLI signal emitted by B16/F10-BrM cells in each brain slice normalized by the initial value obtained at day 0, before the addition of control IgG, anti-TIMP1 (10 μg/mL) or anti-CD8 (100 μg/mL). Values are shown in box-and-whisker plots in which every dot represents a different organotypic culture and the line in the box corresponds to the median. Whiskers go from minimum to maximum values (n = 42 IgG, 39 anti-TIMP1 and 27 anti-TIMP1 plus anti-CD8 independent organotypic cultures). Quantification is accompanied by representative images of wells containing brain organotypic cultures with established B16/F10-BrM metastases grown ex vivo for 3 days. The image shows the BLI intensity in each condition for each brain slice. P values were calculated using the two-tailed t test. F, Quantification of the number of Ki67+ cancer cells found in IgG2 and anti-TIMP1-treated PDOCs. Values are shown in box-and-whisker plots in which every dot represents a patient and each patient is an independent experiment (n = 11). The pie chart shows all BrM-PDOCs quantified in the graph and classified according to the specific primary tumor. P value was calculated using two-tailed t test. G, Quantification of the number of Ki67+ cancer cells found in IgG2, anti-TIMP1 (10 μg/mL), and anti-TIMP1 (10 μg/mL) plus anti-CD8 (10 μg/mL) PDOCs. Values are shown in box-and-whisker plots in which every dot represents a patient and each patient is an independent experiment (n = 7). P value was calculated using two-tailed t test. H, Schema of the experimental design. cKOGFAP-Timp1 mice were inoculated with BrM cells intracardially, and after 2 weeks, ex vivo brain BLI was analyzed. I and J, Representative images of brains from control and cKOGFAP-Timp1 mice intracardially injected with B16/F10-BrM (I) or E0771-BrM (J) cells. The image shows the BLI intensity in each condition. K and L, Quantification of ex vivo brain BLI. Values are shown in box-and-whisker plots in which every dot represents a different animal. Values were obtained from normalizing the ex vivo brain signal to the in vivo head signal 3 days after intracardiac injection with either B16/F10-BrM (K) or E0771-BrM (L) cells (n = 26/29 mice four independent experiments in K and n = 28/25 mice three independent experiments in L). P value was calculated using the two-tailed t test. M, Representative images of CD8+ T cells in metastatic lesions growing in the brains from control or cKOGFAP-Timp1 mice intracardially injected with E0771-BrM at experimental endpoint. White arrowhead indicates CD8+ T cells and red arrowheads indicate Ki67+CD8+ T cells. Scale bar, 25 μm, magnification 5 μm. N, Quantification of the total number of CD8+ T cells in control and cKOGFAP-Timp1 mice intracardially injected with E0771-BrM at human endpoint. Values are shown in box-and-whisker plots in which every dot represents a different animal. Ten brains were analyzed in each condition. P value was calculated using the two-tailed t test.
TIMP1 mediates brain metastasis in a CD8+ T-cell–dependent manner. A, Schema of the experimental design. pSTAT3− and pSTAT3+wt and pSTAT3+ cKOGFAP-Timp1 CM (with or without rTIMP1 100 ng/mL or control IgG/anti-TIMP1 10 μg/mL) was added to CD8+ T cells and cultured with BrM cells in a 1:4 ratio (BrM-OVA cancer cells: OT-I T cells specific for the OVA-derived SIINFEKL peptide) or a 1:5 ratio (BrM cancer cells:CD8+ T cells previously activated). B, Quantification of the BLI signal from the experiment shown in A and representative images of B16/F10-BrM-OVA–derived BLI at the initial time point and 24 hours after adding CD8+ lymphocytes preincubated with CM. Light orange condition refers to coculture of OT-I T cells with B16/F10-BrM no OVA (control for antigen-specific killing). Values correspond to 24 hours BLI normalized to BLI before adding CD8+ T cells expressed in percentage with respect to the mean of control experimental condition (BrM cells). Error bars, SEM. n = 3 different cocultures per condition. P value was calculated using the two-tailed t test. C and D, Schema of the experimental design. Control IgG or anti-TIMP1 (10 μg/mL) was added to the medium in organotypic cultures of mouse brain with B16/F10-BrM established lesions (C) and PDOC that include the brain metastasis–associated microenvironment (D). E, Quantification of the BLI signal emitted by B16/F10-BrM cells in each brain slice normalized by the initial value obtained at day 0, before the addition of control IgG, anti-TIMP1 (10 μg/mL) or anti-CD8 (100 μg/mL). Values are shown in box-and-whisker plots in which every dot represents a different organotypic culture and the line in the box corresponds to the median. Whiskers go from minimum to maximum values (n = 42 IgG, 39 anti-TIMP1 and 27 anti-TIMP1 plus anti-CD8 independent organotypic cultures). Quantification is accompanied by representative images of wells containing brain organotypic cultures with established B16/F10-BrM metastases grown ex vivo for 3 days. The image shows the BLI intensity in each condition for each brain slice. P values were calculated using the two-tailed t test. F, Quantification of the number of Ki67+ cancer cells found in IgG2 and anti-TIMP1-treated PDOCs. Values are shown in box-and-whisker plots in which every dot represents a patient and each patient is an independent experiment (n = 11). The pie chart shows all BrM-PDOCs quantified in the graph and classified according to the specific primary tumor. P value was calculated using two-tailed t test. G, Quantification of the number of Ki67+ cancer cells found in IgG2, anti-TIMP1 (10 μg/mL), and anti-TIMP1 (10 μg/mL) plus anti-CD8 (10 μg/mL) PDOCs. Values are shown in box-and-whisker plots in which every dot represents a patient and each patient is an independent experiment (n = 7). P value was calculated using two-tailed t test. H, Schema of the experimental design. cKOGFAP-Timp1 mice were inoculated with BrM cells intracardially, and after 2 weeks, ex vivo brain BLI was analyzed. I and J, Representative images of brains from control and cKOGFAP-Timp1 mice intracardially injected with B16/F10-BrM (I) or E0771-BrM (J) cells. The image shows the BLI intensity in each condition. K and L, Quantification of ex vivo brain BLI. Values are shown in box-and-whisker plots in which every dot represents a different animal. Values were obtained from normalizing the ex vivo brain signal to the in vivo head signal 3 days after intracardiac injection with either B16/F10-BrM (K) or E0771-BrM (L) cells (n = 26/29 mice four independent experiments in K and n = 28/25 mice three independent experiments in L). P value was calculated using the two-tailed t test. M, Representative images of CD8+ T cells in metastatic lesions growing in the brains from control or cKOGFAP-Timp1 mice intracardially injected with E0771-BrM at experimental endpoint. White arrowhead indicates CD8+ T cells and red arrowheads indicate Ki67+CD8+ T cells. Scale bar, 25 μm, magnification 5 μm. N, Quantification of the total number of CD8+ T cells in control and cKOGFAP-Timp1 mice intracardially injected with E0771-BrM at human endpoint. Values are shown in box-and-whisker plots in which every dot represents a different animal. Ten brains were analyzed in each condition. P value was calculated using the two-tailed t test.
Characterization of the Influence of TIMP1 in CD8+ T Cells
We characterized the influence of STAT3/TIMP1 on CD8+ T cells using immunophenotyping with different coactivatory, coinhibitory markers and cytokines. Flow cytometry analysis confirmed that according to the decrease in cytotoxicity that we observed previously (Fig. 4B; Supplementary Fig. S5C–S5G), pSTAT3+ CM decreased the expression of CD25 in effector CD8+ T cells (Fig. 5A–C). Furthermore, CD25 downregulation was rescued upon depletion of Timp1 in astrocytes (Fig. 5B and C). The absence of TIMP1 downstream of STAT3 leads to the increase of CD8+ T cells expressing inflammatory cytokines (Fig. 5A and D) and a decrease in exhausted CD8+ T cells (Fig. 5A and E). Furthermore, brain metastasis–associated CD8+ T cells increased CD44 and INFγ levels, whereas reduced exhaustion markers when TIMP1 was depleted from reactive astrocytes in vivo (Fig. 5F–L). TIMP1 has been mostly studied as a regulator of matrix metalloproteinase (MMP) (32), however its role as a ligand binding to CD63 receptor (33) has not been addressed until recently (32, 34, 35). We tested if the protumoral role of TIMP1 in brain metastasis depends on its interaction with MMPs or on its cytokine activity in organotypic cultures. Only blocking TIMP1 regions noninteracting with MMPs leads to a decrease in brain metastasis (Supplementary Fig. S6A–S6C). CD63 has been previously suggested as a marker of CD8+ T-cell activation (36), which we reproduced in vitro (Supplementary Fig. S6D). Although we detected a trend toward an increased percentage of circulating CD8+ T cells expressing CD63 when there is systemic disease in preclinical models (Supplementary Fig. S6E), a robust and significant increase in the surface levels of CD63 was only detected when the CD8+ T-cell fraction was evaluated in established brain metastases (Fig. 6A–D; Supplementary Fig. S6F and S6G; ref. 24). Furthermore, we confirmed the presence of CD63+CD8+ T cells in situ in both experimental and patient-derived brain metastases (Fig. 6C and D). We probed the binding of astrocyte-derived TIMP1 and CD63 on the surfaces of CD8+ T cells in cocultures of pSTAT3+ astrospheres and in vitro activated CD8+ T cells, even though the culture of these two cell types independent of each other did not reproduce the binding of the two molecules (Fig. 6E; Supplementary Fig. S6H). This finding was further validated in situ in human brain metastasis samples, detecting specific signals in proximity ligation assays (Fig. 6F; Supplementary Fig. S6I). The fact that the level of CD63 receptor increases along with the activation state of CD8+ T cells (Supplementary Fig. S6D) and that the binding of its ligand, TIMP1, triggers an immunosuppressive phenotype (Figs. 4B, 5A–E, and G–L) might be suggestive of a potential paracrine immune checkpoint. To consolidate this hypothesis, we first evaluated whether CD8+ T cells from CD63− null mice (37) exhibit improved antitumor ability (Fig. 6G). The addition of WT CD8+ T cells to organotypic cultures of established brain metastases generated with the B16/F10-BrM cell line was not sufficient to reduce significantly the viability of cancer cells (Fig. 6H), which reinforces the influence of the immunosuppressive microenvironment. By contrast, the absence of CD63− TIMP1 signaling when CD63 KO CD8+ T cells were added to organotypic cultures, allowed an effective T-cell–mediated killing of cancer cells (Fig. 6H). To further confirm the differential impact of TIMP1 among CD63low and CD63high CD8+ T cells, sorted CD8+ T cells with low or high CD63 levels were treated with CM from either WT or cKOGFAP-Timp1 pSTAT3+ astrospheres (Supplementary Fig. S6J–S6N). Although CD8+/CD63low T cells did not respond to the presence of TIMP1 from STAT3+ astrospheres CM, CD8+/CD63high T cells increased CD44/CD62L levels when TIMP1 was not present (Supplementary Fig. S6L–S6N). Additionally, cytotoxicity genes Gzmb and Gzmk were induced in sorted brain metastasis–associated CD8+/CD63high T cells when Timp1 was depleted from reactive astrocytes (Fig. 6I; Supplementary Fig. S6O). Our findings report a novel molecular cross-talk between STAT3+ reactive astrocytes and CD8+ T cells through TIMP1-CD63, leading to decrease in antitumor activity of this component of the acquired immune system infiltrating the brain. However, due to the lack of knowledge on the signaling pathways downstream of CD63 in lymphocytes upon TIMP1 binding, we performed phosphoproteomics analysis to deepen our findings on T-cell immunosuppression in brain metastasis. In vitro activated CD8+ T cells were analyzed by using LC/MS-MS–based proteomics after incubation with CM from pSTAT3+ astrospheres derived from WT or cKOGFAP-Timp1 astrocytes (Fig. 6J; Supplementary Fig. S7A). The lack of TIMP1 signaling on CD8+ T cells lead to a main enrichment of signatures related to T-cell activation as the top finding (Supplementary Fig. S7B; Supplementary Table S17). Dissecting the phosphosites significantly altered when the immunosuppressive signal activated by TIMP1 was not present, revealing several kinases with altered levels of their phosphorylated substrates (Fig. 6K; Supplementary Table S18). Among them, we validated changes in ERK1/2 phosphorylation (pERK1/2) in CD8+ T cells infiltrating metastases when targeting TIMP1 in astrocytes (cKOGFAP-Timp1; Fig. 6L; Supplementary Fig. S7C). Furthermore, analysis of human brain metastases scored with multiplex (Fig. 3E–G) showed a correlation between the quality of infiltrating CD8+ T cells on their pERK1/2 status and their immune cluster category (Fig. 6M; Supplementary Fig. S7D; Supplementary Table S19). Finally, we validated the modulation of ERK activity using rTIMP1 or anti-TIMP1 on CD8+ T cells while incubated in astrospheres CM (Supplementary Fig. S7E). Overall, we report that signaling downstream of CD63 receptor has major implications in antitumor activity of CD8+ T cells upon TIMP1 binding through the modulation of multiple kinases including ERK1/2 (Fig. 6N).
Characterization of the influence of TIMP1 in CD8+ T cells. A, Schema of the experimental design. pSTAT3− and pSTAT3+wt and pSTAT3+ cKOGFAP-Timp1 CM were added to CD8+ T cells, and flow cytometry analysis was performed. B, Representative flow cytometry analysis using preactivated CD8+ T cells incubated with CM generated by pSTAT3− and pSTAT3+wt or pSTAT3+ cKOGFAP-Timp1 astrospheres. C, Quantification of CD25 geometric mean fluorescence intensity (gMFI) in effector CD8+ T cells from A. Error bars, SEM. n = 3 different T cells cultures per condition. P value was calculated using the two-tailed t test. D and E, Flow cytometry analysis showing the percentage of IFNγ+TNFα+ (D) and exhausted PD1+LAG3+TIM3+CD39+ (E) CD8+ T cells incubated with CM generated by pSTAT3− and pSTAT3+wt or pSTAT3+ cKOGFAP-Timp1 astrospheres. Error bars, SEM. n = 3 different T cells cultures per condition. P value was calculated using the two-tailed t test. F, Schema of the experimental design. CD8+ lymphocytes from wt and cKOGFAP-Timp1 brains intracranially injected with B16/F10-BrM cells were analyzed by flow cytometry. G and H, Representative flow cytometry analysis of CD44 (G) and quantification of the experiment (H). Error bars, SEM. Every dot is a different animal (n = 5 wt brains and n = 5 cKOGFAP-Timp1 brains). P value was calculated using the two-tailed t test. I and J, Representative flow cytometry analysis of TNFα (I) and quantification of the experiment (J). Error bars, SEM. Every dot is a different animal (n = 8 wt brains and n = 9 cKOGFAP-Timp1 brains). P value was calculated using the two-tailed t test. K and L, Representative flow cytometry analysis of CD39 and PD1 (K) and quantification of the experiment (L). Error bars, SEM. Every dot is a different animal (n = 8 wt brains and n = 9 cKOGFAP-Timp1 brains). P value was calculated using the two-tailed t test.
Characterization of the influence of TIMP1 in CD8+ T cells. A, Schema of the experimental design. pSTAT3− and pSTAT3+wt and pSTAT3+ cKOGFAP-Timp1 CM were added to CD8+ T cells, and flow cytometry analysis was performed. B, Representative flow cytometry analysis using preactivated CD8+ T cells incubated with CM generated by pSTAT3− and pSTAT3+wt or pSTAT3+ cKOGFAP-Timp1 astrospheres. C, Quantification of CD25 geometric mean fluorescence intensity (gMFI) in effector CD8+ T cells from A. Error bars, SEM. n = 3 different T cells cultures per condition. P value was calculated using the two-tailed t test. D and E, Flow cytometry analysis showing the percentage of IFNγ+TNFα+ (D) and exhausted PD1+LAG3+TIM3+CD39+ (E) CD8+ T cells incubated with CM generated by pSTAT3− and pSTAT3+wt or pSTAT3+ cKOGFAP-Timp1 astrospheres. Error bars, SEM. n = 3 different T cells cultures per condition. P value was calculated using the two-tailed t test. F, Schema of the experimental design. CD8+ lymphocytes from wt and cKOGFAP-Timp1 brains intracranially injected with B16/F10-BrM cells were analyzed by flow cytometry. G and H, Representative flow cytometry analysis of CD44 (G) and quantification of the experiment (H). Error bars, SEM. Every dot is a different animal (n = 5 wt brains and n = 5 cKOGFAP-Timp1 brains). P value was calculated using the two-tailed t test. I and J, Representative flow cytometry analysis of TNFα (I) and quantification of the experiment (J). Error bars, SEM. Every dot is a different animal (n = 8 wt brains and n = 9 cKOGFAP-Timp1 brains). P value was calculated using the two-tailed t test. K and L, Representative flow cytometry analysis of CD39 and PD1 (K) and quantification of the experiment (L). Error bars, SEM. Every dot is a different animal (n = 8 wt brains and n = 9 cKOGFAP-Timp1 brains). P value was calculated using the two-tailed t test.
TIMP1 modulates CD8+ T cells through CD63. A, Schema of the experimental design. CD63 expression was analyzed by flow cytometry gating on CD8+ T cells from metastasis free condition and brains intracranially injected with B16/F10-BrM cells. B, Flow cytometry analysis of CD63 expression gated on CD8+ T cells from brains without tumor and brains intracranially injected with B16/F10-BrM cells. Error bars, SEM. Every dot is a different animal (n = 3 metastasis free brains and n = 6 B16/F10-BrM brain metastases). P value was calculated using the two-tailed t test. C, Immunofluorescence of established B16/F10-BrM metastasis. CD63 is expressed on CD8+ T cells surrounding the lesion. Red arrowhead indicates a CD63+CD8+ T cell. Scale bar, 10 μm. D, Representative image showing colocalization of metastasis-associated CD8 and CD63 staining in a patient with lung cancer brain metastasis. White arrowhead indicates a CD8+ T cell and red arrowhead indicates a double CD63+CD8+ T cell. Scale bar, 10 μm. E, Immunoblotting using anti-TIMP1, anti-CD63, and Vinculin antibodies showing secreted TIMP1 and CD63 binding on CD8+ T cells when cocultured with pSTAT3+ astrospheres. Cell lysates (first line) were immunoprecipitated with IgG isotype as the control (second line) and anti-CD63 (third line). F, Proximity ligation assay performed on a melanoma brain metastasis sample showing TIMP1 and CD63 in close molecular proximity on CD8+ T cells. Magnification showing red dots of TIMP1-CD63 interaction (white arrowheads) on a CD8+ T cell highlighted with a red arrowhead in the main picture. Scale bar, 10 μm. G, Schema of the experimental design. Wt or CD63−null CD8+ T cells were used in ex vivo organotypic cultures with established B16/F10-BrM metastasis. H, Quantification of the BLI signal emitted by B16/F10-BrM cells in each brain slice normalized by the initial value obtained at time 0, before the addition of wt or CD63−null CD8+ T cells. Values are shown in box-and-whisker plots in which every dot represents a different organotypic culture and the line in the box corresponds to the median. Whiskers go from minimum to maximum values (n = 8 no CD8+ T cells, 7 wt CD8+ T cells, and 10 CD63−null CD8+ T cells independent organotypic cultures). Quantification is accompanied by representative images of wells containing brain organotypic cultures with established B16/F10-BrM metastases grown ex vivo for 24 hours. The image shows the BLI intensity in each condition for each brain slice. P values were calculated using the two-tailed t test. I, Heatmap generated with the qRT-PCR analysis performed on CD63highCD8+ T cells sorted from wt and cKOGFAP-Timp1 mice 10 days after intracranial injection of B16/F10-BrM cells. n = 12 brains per condition and six brains for control condition (not injected with BrM cells). J, Schema of the experimental design. CD8+ lymphocytes were cultured with STAT3− astrospheres CM and wt or cKOGFAP-Timp1 STAT3+ astrospheres CM and processed for phosphoproteomics analysis. K, Heatmap showing the top 10 enriched sequence motifs found in CD8+ T cells in the absence of TIMP1 from the CM of STAT3+ astrospheres. Clustering enrichment using Fisher exact test was performed. P value < 0.01, FDR < 2%. L, Quantification of the number of pERK+CD8+ T cells in control and cKOGFAP-Timp1 mice intracardially injected with E0771-BrM at the endpoint. Error bars, SEM. Every dot is a different animal (n = 3 brains per condition). M, Quantification of the number of pERK+CD8+ T cells in human brain metastases samples scored with multiplex. Violin plots show the median of percentage pERK+CD8+ T cells among the total CD8+ T cells per field of view (n = 5–10/patient) from three patients analyzed in each condition. P value was calculated using the two-tailed t test. N, Model summarizing main findings on the immunomodulatory role of TIMP1 derived from STAT3+ reactive astrocytes in brain metastasis. Secreted TIMP1 acts on its receptor CD63 receptor on the surface of CD8+ lymphocytes, modulating ERK-mediated signaling and downregulating activation of T-cell markers and cytolytic enzymes and upregulating exhaustion markers, thus affecting effective T-cell–mediated killing of brain metastatic cells.
TIMP1 modulates CD8+ T cells through CD63. A, Schema of the experimental design. CD63 expression was analyzed by flow cytometry gating on CD8+ T cells from metastasis free condition and brains intracranially injected with B16/F10-BrM cells. B, Flow cytometry analysis of CD63 expression gated on CD8+ T cells from brains without tumor and brains intracranially injected with B16/F10-BrM cells. Error bars, SEM. Every dot is a different animal (n = 3 metastasis free brains and n = 6 B16/F10-BrM brain metastases). P value was calculated using the two-tailed t test. C, Immunofluorescence of established B16/F10-BrM metastasis. CD63 is expressed on CD8+ T cells surrounding the lesion. Red arrowhead indicates a CD63+CD8+ T cell. Scale bar, 10 μm. D, Representative image showing colocalization of metastasis-associated CD8 and CD63 staining in a patient with lung cancer brain metastasis. White arrowhead indicates a CD8+ T cell and red arrowhead indicates a double CD63+CD8+ T cell. Scale bar, 10 μm. E, Immunoblotting using anti-TIMP1, anti-CD63, and Vinculin antibodies showing secreted TIMP1 and CD63 binding on CD8+ T cells when cocultured with pSTAT3+ astrospheres. Cell lysates (first line) were immunoprecipitated with IgG isotype as the control (second line) and anti-CD63 (third line). F, Proximity ligation assay performed on a melanoma brain metastasis sample showing TIMP1 and CD63 in close molecular proximity on CD8+ T cells. Magnification showing red dots of TIMP1-CD63 interaction (white arrowheads) on a CD8+ T cell highlighted with a red arrowhead in the main picture. Scale bar, 10 μm. G, Schema of the experimental design. Wt or CD63−null CD8+ T cells were used in ex vivo organotypic cultures with established B16/F10-BrM metastasis. H, Quantification of the BLI signal emitted by B16/F10-BrM cells in each brain slice normalized by the initial value obtained at time 0, before the addition of wt or CD63−null CD8+ T cells. Values are shown in box-and-whisker plots in which every dot represents a different organotypic culture and the line in the box corresponds to the median. Whiskers go from minimum to maximum values (n = 8 no CD8+ T cells, 7 wt CD8+ T cells, and 10 CD63−null CD8+ T cells independent organotypic cultures). Quantification is accompanied by representative images of wells containing brain organotypic cultures with established B16/F10-BrM metastases grown ex vivo for 24 hours. The image shows the BLI intensity in each condition for each brain slice. P values were calculated using the two-tailed t test. I, Heatmap generated with the qRT-PCR analysis performed on CD63highCD8+ T cells sorted from wt and cKOGFAP-Timp1 mice 10 days after intracranial injection of B16/F10-BrM cells. n = 12 brains per condition and six brains for control condition (not injected with BrM cells). J, Schema of the experimental design. CD8+ lymphocytes were cultured with STAT3− astrospheres CM and wt or cKOGFAP-Timp1 STAT3+ astrospheres CM and processed for phosphoproteomics analysis. K, Heatmap showing the top 10 enriched sequence motifs found in CD8+ T cells in the absence of TIMP1 from the CM of STAT3+ astrospheres. Clustering enrichment using Fisher exact test was performed. P value < 0.01, FDR < 2%. L, Quantification of the number of pERK+CD8+ T cells in control and cKOGFAP-Timp1 mice intracardially injected with E0771-BrM at the endpoint. Error bars, SEM. Every dot is a different animal (n = 3 brains per condition). M, Quantification of the number of pERK+CD8+ T cells in human brain metastases samples scored with multiplex. Violin plots show the median of percentage pERK+CD8+ T cells among the total CD8+ T cells per field of view (n = 5–10/patient) from three patients analyzed in each condition. P value was calculated using the two-tailed t test. N, Model summarizing main findings on the immunomodulatory role of TIMP1 derived from STAT3+ reactive astrocytes in brain metastasis. Secreted TIMP1 acts on its receptor CD63 receptor on the surface of CD8+ lymphocytes, modulating ERK-mediated signaling and downregulating activation of T-cell markers and cytolytic enzymes and upregulating exhaustion markers, thus affecting effective T-cell–mediated killing of brain metastatic cells.
A Combined Immunotherapy Targeting Local Immunosuppression Provides Superior Control of Brain Metastasis
In order to demonstrate the therapeutic implications of our findings, we decided to test whether inhibition of STAT3 could be combined with ICB to obtain better antitumor responses in the brain (Fig. 7A). The B16/F10-BrM model responded to anti-PD1/ anti-CTLA4 extracranially (Fig. 7B; Supplementary Fig. S8A) but did not decrease tumor burden in the brain (Fig. 7B; Supplementary Fig. S8B). Complementary, as previously reported (16), the STAT3 inhibitor silibinin achieved a significant control of brain metastases (Fig. 7B; Supplementary Fig. S8B) but with limited extracranial benefit (Fig. 7B; Supplementary Fig. S8A). Although brain bioluminescence imaging (BLI) ex vivo did not show any additional benefit of the ICB and silibinin combination beyond the response to silibinin monotherapy (Supplementary Fig. S8B), the histological examination of these brains demonstrated that the response was clearly superior (Fig. 7C; Supplementary Table S20). Interestingly, although large metastases were mainly controlled by silibinin, metastases of medium size were more effectively targeted by ICB with combination therapy (Fig. 7C; Supplementary Table S20). The apparent dissociation between BLI and histology might suggest that the data obtained with bioluminescence are mainly contributed by large lesions, thus lacking the sensitivity to scoring changes affecting metastases from other size categories. In fact, we previously reported that silibinin was not effective against smaller metastasis both in experimental models and patients because STAT3+ reactive astrocytes are not present (16). We hypothesized that combined immunotherapy including silibinin could sensitize experimental metastases to the attack of CD8+ T cells activated systemically with ICB. Accordingly, we evaluated whether the antitumor response was increased when targeting local immunosuppression with silibinin in the brains of ICB-treated mice. Histological analysis of the brains from mice treated with combined immunotherapy showed increased markers of cytotoxic activity (Fig. 7D and E) and cleaved caspase-3–staining in cancer cells (Supplementary Fig. S8C and S8D). In order to reinforce our finding, to discard any influence of extracranial metastasis in the brain phenotype (11) and to explore the potential additional benefit of a combination with radiotherapy, we repeated the combination therapy using intracranial injection in this model to apply local therapy (Fig. 7F), as previously reported (31). Because the B16/F10-BrM model lacks a recently described radioresistance mechanism (31), we found it does respond to fractionated radiotherapy (Supplementary Fig. S8E). Accordingly, we added ICB and silibinin to irradiated mice and scored whether any additional benefit in overall survival (OS) was detected beyond what is provided by local therapy. In this experimental setting, ICB showed a superior ability to target brain metastases mimicking the effect of silibinin (Supplementary Fig. S8F). More importantly, the triple combination therapy did add additional brain tumor control, increasing OS (Fig. 7G; Supplementary Fig. S8F). Consistently, the combined immunotherapy led to a more efficient cancer cell killing (Fig. 7H and I) and more proliferative CD8+ T cells locally (Fig. 7J and K). To expand our finding to other relevant preclinical models and test whether the improved control of brain metastasis when combining ICB and STAT3 inhibition was triggered by impairing TIMP1-mediated immunosuppression, we used cKOGFAP-Timp1 mice intracardially injected with E0771-BrM cells and treated with ICB (Supplementary Fig. S8G). We found that abolished secretion of the STAT3 downstream target TIMP1 in reactive astrocytes improved ICB benefit in brain metastasis (Supplementary Fig. S8H–S8J), affecting metastases of both medium and big sizes (Supplementary Fig. S8K and S8L; Supplementary Table S20).
A combined immunotherapy targeting local immunosuppression provides superior control of brain metastasis. A, Schema of the experimental design. C57BL/6J mice were intracardially injected with B16/F10-BrM cells, 3 days after the following treatments were administrated: IgG2 (10 mg/kg), silibinin daily (200 mg/kg), or ICB every 2 days (anti-PD1, 10 mg/kg, plus anti-CTLA4, 10 mg/kg) alone or in combination with silibinin. After 2 weeks, ex vivo analysis and histological analysis of different organs were performed. B, Representative images of control, ICB, silibinin, and ICB+silibinin–treated mice 2 weeks (endpoint) after intracardiac inoculation of B16/F10-BrM cells. In in vivo images, dotted lines surround the brain and lungs, showed in the ex vivo representative images below. Images show BLI intensity. C, Distribution of lesions according to size (small: <5e4 μm2, medium: 2.5e4 μm2 to 2e5 μm2, big: >2e5 μm2). Values are represented as percentage with respect to the total number of lesions per each experimental condition. n = 4 to 6 brains per condition. P values of the different comparison calculated using the two-tailed t test are shown in Supplementary Table S16. D, Representative images of perforin and Granzyme B staining at endpoint in brains from mice treated with ICB and ICB+silibinin. Arrowheads indicate positive staining. Scale bar, 50 μm. E, Quantification showing the number of cells expressing cytotoxic markers in D. Values are shown in box-and-whisker plots in which every dot is a different lesion (n = 6 lesions in three brains are quantified in ICB and n = 4 lesions in three brains are quantified in ICB+silibinin). P value was calculated using the two-tailed t test. F, Schema of the experimental design. Three days after intracranial inoculation of B16/F10-BrM cells, five doses of 3 Gy WBRT and IgG2 (10 mg/kg), silibinin daily (200 mg/kg), or ICB every 2 days (anti-PD1, 10 mg/kg, plus anti-CTLA4, 10 mg/kg) alone or in combination with silibinin were administrated. G, Kaplan–Meier curve showing survival proportions of mice without radiotherapy (dotted gray line, n = 12) and with radiotherapy (Rx; IgG2, red line, n = 8; ICB, blue line, n = 8; silibinin, gray line, n = 8; and ICB+silibinin, green line, n = 8). P value was calculated using log-rank (Mantel–Cox) test between Rx and Rx+ICB+silibinin groups. H, Representative images of cleaved caspase-3 staining of intracranially inoculated brains with B16/F10-BrM cells at endpoint from irradiated mice treated with ICB and ICB+silibinin. Scale bar, 75 μm; magnification, 25 μm. I, Quantification of experiment in H. Percentage of cleaved caspase-3 is normalized with tumor area. Values are shown in box-and-whisker plots in which every dot is a different field of view. Four brains per condition are quantified. P value was calculated using the two-tailed t test. J, Representative images of Ki67− (white arrowheads) and Ki67+ (red arrowheads) CD8+ T cells infiltrating brain metastases from mice intracranially inoculated with B16/F10-BrM cells and treated with radiotherapy with either ICB or ICB+silibinin. Scale bar, 25 μm. K, Quantification of experiment in J. Values are shown in box-and-whisker plots in which every dot is a different field of view. Three brains per condition are quantified. P value was calculated using the two-tailed t test. L, Quantification of TIMP1 levels measured in patients’ CSF. Noncancer control condition: n = 6 and brain metastasis condition: n = 12 (matched CSF samples from the same patients in Supplementary Fig. S9A) plus n = 2 unmatched CSF values. Each dot represents a different patient. Patients shown in N are colored in green. P value was calculated using the two-tailed t test. M and N, Schema of the strategy to perform an ex vivo proof-of-concept validation of TIMP1 as a biomarker of response to blockade of CD8+ T-cell local immunosuppression. Heatmap showing immune cluster category (according to total percentage of immune cells, and mean percentage of immune cells present in low immune cluster samples in Fig. 3G is used as reference), TIMP1 levels in the CSF (mean of TIMP1 levels in the CSF of noncancer patients is used as the reference) and response to anti-TIMP1 and anti-TIMP1+anti-CD8 (viability of cancer cells in percentage of Ki67+ cancer cells; IgG2 condition is used as the reference) in PDOCs of patients shown in L (green dots). Results from the PDOCs are in Fig. 4F and G and Supplementary Table S15. Represented values are shown in Supplementary Fig. S9C.
A combined immunotherapy targeting local immunosuppression provides superior control of brain metastasis. A, Schema of the experimental design. C57BL/6J mice were intracardially injected with B16/F10-BrM cells, 3 days after the following treatments were administrated: IgG2 (10 mg/kg), silibinin daily (200 mg/kg), or ICB every 2 days (anti-PD1, 10 mg/kg, plus anti-CTLA4, 10 mg/kg) alone or in combination with silibinin. After 2 weeks, ex vivo analysis and histological analysis of different organs were performed. B, Representative images of control, ICB, silibinin, and ICB+silibinin–treated mice 2 weeks (endpoint) after intracardiac inoculation of B16/F10-BrM cells. In in vivo images, dotted lines surround the brain and lungs, showed in the ex vivo representative images below. Images show BLI intensity. C, Distribution of lesions according to size (small: <5e4 μm2, medium: 2.5e4 μm2 to 2e5 μm2, big: >2e5 μm2). Values are represented as percentage with respect to the total number of lesions per each experimental condition. n = 4 to 6 brains per condition. P values of the different comparison calculated using the two-tailed t test are shown in Supplementary Table S16. D, Representative images of perforin and Granzyme B staining at endpoint in brains from mice treated with ICB and ICB+silibinin. Arrowheads indicate positive staining. Scale bar, 50 μm. E, Quantification showing the number of cells expressing cytotoxic markers in D. Values are shown in box-and-whisker plots in which every dot is a different lesion (n = 6 lesions in three brains are quantified in ICB and n = 4 lesions in three brains are quantified in ICB+silibinin). P value was calculated using the two-tailed t test. F, Schema of the experimental design. Three days after intracranial inoculation of B16/F10-BrM cells, five doses of 3 Gy WBRT and IgG2 (10 mg/kg), silibinin daily (200 mg/kg), or ICB every 2 days (anti-PD1, 10 mg/kg, plus anti-CTLA4, 10 mg/kg) alone or in combination with silibinin were administrated. G, Kaplan–Meier curve showing survival proportions of mice without radiotherapy (dotted gray line, n = 12) and with radiotherapy (Rx; IgG2, red line, n = 8; ICB, blue line, n = 8; silibinin, gray line, n = 8; and ICB+silibinin, green line, n = 8). P value was calculated using log-rank (Mantel–Cox) test between Rx and Rx+ICB+silibinin groups. H, Representative images of cleaved caspase-3 staining of intracranially inoculated brains with B16/F10-BrM cells at endpoint from irradiated mice treated with ICB and ICB+silibinin. Scale bar, 75 μm; magnification, 25 μm. I, Quantification of experiment in H. Percentage of cleaved caspase-3 is normalized with tumor area. Values are shown in box-and-whisker plots in which every dot is a different field of view. Four brains per condition are quantified. P value was calculated using the two-tailed t test. J, Representative images of Ki67− (white arrowheads) and Ki67+ (red arrowheads) CD8+ T cells infiltrating brain metastases from mice intracranially inoculated with B16/F10-BrM cells and treated with radiotherapy with either ICB or ICB+silibinin. Scale bar, 25 μm. K, Quantification of experiment in J. Values are shown in box-and-whisker plots in which every dot is a different field of view. Three brains per condition are quantified. P value was calculated using the two-tailed t test. L, Quantification of TIMP1 levels measured in patients’ CSF. Noncancer control condition: n = 6 and brain metastasis condition: n = 12 (matched CSF samples from the same patients in Supplementary Fig. S9A) plus n = 2 unmatched CSF values. Each dot represents a different patient. Patients shown in N are colored in green. P value was calculated using the two-tailed t test. M and N, Schema of the strategy to perform an ex vivo proof-of-concept validation of TIMP1 as a biomarker of response to blockade of CD8+ T-cell local immunosuppression. Heatmap showing immune cluster category (according to total percentage of immune cells, and mean percentage of immune cells present in low immune cluster samples in Fig. 3G is used as reference), TIMP1 levels in the CSF (mean of TIMP1 levels in the CSF of noncancer patients is used as the reference) and response to anti-TIMP1 and anti-TIMP1+anti-CD8 (viability of cancer cells in percentage of Ki67+ cancer cells; IgG2 condition is used as the reference) in PDOCs of patients shown in L (green dots). Results from the PDOCs are in Fig. 4F and G and Supplementary Table S15. Represented values are shown in Supplementary Fig. S9C.
Our initial findings suggest the feasibility of using TIMP1 to stratify patients who could benefit from the combined immunotherapy (Fig. 3D). However, systemic treatment should not rely on a biomarker requiring neurosurgery to score tissue samples. Consequently, given the secretory ability of STAT3+ RA (16), which includes TIMP1 (Supplementary Fig. S3A), together with existing reports using astrocyte-derived biomarkers in liquid biopsies (38, 39), we evaluated such a possibility in patients with brain metastasis. The CSF has been suggested as a better surrogate to the brain parenchyma than blood (40–43), so we decided to evaluate TIMP1 in these two types of liquid biopsies from the RENACER cohort. Although the TIMP1 levels in blood did not differ from that in healthy control individuals (Supplementary Fig. S9A; Supplementary Table S21), the CSF from patients with brain metastasis was significantly enriched in the potential biomarker (Fig. 7L; Supplementary Table S21). Furthermore, high levels of TIMP1 in the CSF of patients with brain metastasis predict worse OS (Supplementary Fig. S9B; Supplementary Table S22). In order to evaluate the correlation between the biomarker and the susceptibility to respond to strategies that block local immunosuppression, we checked whether any of these patients also had PDOC established from extended neurosurgeries as part of the RENACER pipeline (27). A selected group of samples with confirmed presence of immune cells compatible with medium–high immune clusters (Fig. 3G; Supplementary Fig. S9C and S9D; Supplementary Tables S21 and S23–S25) with PDOC and values of TIMP1 in the CSF above the mean of healthy controls could be allocated. According to the data reported above, PDOC proved their sensitivity to the blocking anti-TIMP1 antibody (Fig. 4F, 7M, and N; Supplementary Fig. S9C; Supplementary Tables S15 and S21) in a CD8+ T-cell–dependent manner (Fig. 4G).
Thus, our data provides the rationale to test a novel combined immunotherapy consisting on ICB antibodies and silibinin as a strategy to maximize the access to metastases and antitumor activity of CD8+ T cells by blocking local immunosuppression. In addition, the therapeutic strategy described could potentially be guided by a biomarker compatible with liquid biopsy to improve patient stratification and evaluation of the therapeutic benefit. Overall, our finding represents the first comprehensive approach to target symptomatic brain metastases with biomarker-guided immunotherapy.
Discussion
Recent clinical trials have tested ICB antibodies in patients with brain metastasis derived from melanoma and lung cancer (2–8). The results indicate variable rates of positive responses that could oscillate between 0% and 60% of the patients. However, positive response rates were mainly attributable to asymptomatic brain metastasis, which tend to be smaller in size. Indeed, in those trials in which symptomatic brain metastases were considered, ICB benefits for intracranial lesions dropped significantly (2, 7), which has created concerns about their translation to real-world clinical practice (44). Although corticoids have been suggested to underlie this differential response among patients, it remains controversial (7, 12–14, 45).
The data reported here could potentially explain these clinical findings to some extent because our previous observations concluded that pSTAT3+ reactive astrocytes are not present in early but in advanced stages of the disease (16) and patients treated with silibinin, a STAT3 inhibitor (16, 23), decreased the size of metastasis to a certain point, which then remains stable (16). Thus, we conclude that the lack of local benefit from ICB in patients with symptomatic brain metastasis reflects, rather or in addition to a potential consequence of the use of corticoids, a pSTAT3+ reactive astrocytes–driven mechanism that is responsible for local immunosuppression affecting CD8+ T cells arriving from the periphery. Thus, although ICB might facilitate the access of active T cells into the brain, these potential cellular antitumor entities suffer the local immunosuppressive environment that might underlie the requirement of a combination therapy.
Our data indicate that the presence of brain metastasis alters the immune landscape in the brain increasing immune cells numbers, however brain metastasis–associated T cells remain ineffective to target cancer cells. By dissecting astrocyte heterogeneity, we found subpopulations of astrocytes enriched in potential immunomodulatory signatures. When exploring the molecular basis of immune modulation mediated by metastasis-associated astrocytes, we found that the STAT3-dependent gene TIMP1, previously reported as a top differentially expressed protein in human brain metastasis samples (24), imposes a local immunosuppressive hub affecting the quality of CD8+ T cells. We demonstrate that the main source of TIMP1 is in the tumor microenvironment, and specifically a subpopulation of reactive astrocytes. TIMP1 has been mostly considered an MMP inhibitor (32). However, TIMP1 also plays an MMP-independent role by binding to CD63 (32–35, 46). We report a novel function for TIMP1/CD63 on the surface of CD8+ T cells infiltrating brain metastasis mediating immunosuppression in an antigen–dependent and -independent manner. Although the acquired immune system is necessary for STAT3/TIMP1-mediated immunosuppression, considering the nonrestricted expression of CD63 on CD8+ T cells, it could be presumed that extracellular vesicles expressing CD63 and other cell types such as macrophages may be also affected by TIMP1 increased in the brain metastasis microenvironment. The potential involvement of this and other immune cell types, including dendritic cells that are also directly affected by STAT3 inhibition in reactive astrocytes or as a consequence of the improved immune landscape, should be further addressed. Additionally, although our genetic strategy confirmed that STAT3 inhibition with silibinin is recapitulated with an astrocyte-specific targeting approach on STAT3; we cannot fully discard that the pharmacological strategy also affects other cell types. Whether the immunosuppressive role of the reactive astrocyte subpopulation could play a role in other brain disorders remains to be addressed. Indeed, it is tempting to speculate that given the role of astrocytes to limit potential threats to the brain, this could include their ability to block infiltrating T cells, which might otherwise increase the risk of causing deleterious consequences in this low regenerative organ.
Given that silibinin targets pSTAT3+ reactive astrocytes (16), we propose that the combination with ICB will increase local responses by facilitating CD8+ T-cell antitumor activity in patients with brain metastasis. It should be noted that silibinin could be affecting systemic T cells, and its effects may be potentiated by the action of radiotherapy-promoted T-cell priming (47). Even more, the fact that the levels of STAT3 and TIMP1 are enriched in those patients in whom the local environment is compatible with a potential response to ICB (i.e., high immune cluster) justifies the use of TIMP1 as a potential biomarker. CSF liquid biopsy to detect TIMP1 would allow to not only select patients who would benefit the most from the combined immunotherapy but also follow the therapeutic response over time.
Overall, our study demonstrates that dissecting the heterogeneity within the metastasis-associated microenvironment to cell type–specific subpopulations defined functionally (i.e., mouse cluster 7 and human cluster 5 within STAT3+ reactive astrocytes) offers the possibility to develop novel therapeutic vulnerabilities. By exploring a specific cross-talk within the altered brain metastasis microenvironment (TIMP1 ligand binding to CD63 receptor), we might have contributed to clarify an unsolved clinical limitation (i.e., the lack of response in symptomatic brain metastases). Given the preliminary data that we show in patients, the rationale of combining silibinin with ICB as a more effective immunotherapy for brain metastases supports a follow-up clinical trial after completing the one currently ongoing with silibinin as monotherapy (NCT05689619).
Methods
Animal Studies
All animal experiments are in accordance with a protocol approved by the CNIO, Instituto de Salud Carlos III and Comunidad de Madrid Institutional Animal Care and Use Committee (IACUC.030-2015, CBA35_2015-v2 and PROEX135/19). The cKOGFAP-Stat3 model was generated by breeding GFAP-CRE/ERT2 mice [B6.Cg Tg(GFAP-cre/ERT2)505Fmv/J, The Jackson Laboratory, cat.# 012849] with STAT3loxP/loxP, cKOGFAP-Timp1 was generated as described by Sutter and colleagues (25), and CD63− null mice were generated as described by Schröder and colleagues (37). Tg(TcraTcrb)1100Mjb/J (OT-I mice) were kindly provided by D. Sancho (CNIC) for the isolation of OT-I T cells.
Brain colonization assays were performed in 10- to 15-week-old mice, both males and females (except for the E0771-BrM cells that were injected in females), as previously described (16), by injecting 100 μL of PBS into the left ventricle containing 100,000 cancer cells or 1 μL of PBS intracranially (right frontal cortex, approximately 1.5 mm lateral and 1 mm caudal from the bregma, and to a depth of 2 mm) containing 40,000 cancer cells by using a gas-tight Hamilton syringe and a stereotactic apparatus.
Brain colonization was analyzed in vivo and ex vivo by BLI. Mice were anesthetized with isofluorane and injected retro-orbitally with D-luciferin (150 mg/kg) and imaged with IVIS Lumina III in vivo Spectrum Imaging System (Caliper Life Sciences). Bioluminescence analysis was performed using Living Image software v64. Ex vivo values at the endpoint were normalized to the BLI values of the head in vivo 3 days after injection of the cancer cells before starting treating with different drugs. Tamoxifen (i.p., 1 mg/day) was administered 3 days after cancer cells inoculation until the end of the experiment, silibinin in the formula of silymarin 77.5% (EuroMed, code no. 345316.00) was administered by oral gavage daily (200 mg/kg), 3 days after cancer cells inoculation, and treatment continued until mice reached the endpoint of the experiment. Starting 3 days after cancer cells inoculation, Control IgG (i.p.; 10 mg per kg, Bio X Cell, cat.# BE0090), anti-CD8α (i.p.; 10 mg per kg, Bio X Cell, cat.# BE0061), anti-PD1 (i.p.; 10 mg/ kg Bio X Cell, cat.# BE0146), and anti-CTLA4 (i.p.; 10 mg/ kg, Bio X Cell, cat.# BE0032) antibodies were administrated every 2 days during the first 2 weeks of treatment and on nonconsecutive days during the last week of treatment.
Radiotherapy
Three days after intracranial injection of B16/F10-BrM cells, the presence of established brain metastases was confirmed by BLI. Whole brain radiotherapy (WBRT) protocols mimicking the clinical procedure were applied as previously described (31): fractionated dose of 3 Gy per day for 5 consecutive days or completed regimen with 3 Gy per day for additional 5 days after 2 days without irradiation. Mice were followed up by BLI until the humane endpoint was reached.
Brain Slice Assays
Organotypic slice cultures from adult mouse brain were prepared as previously described (16). Organotypic cultures included brains obtained at the endpoint of metastatic disease when brain lesions are established. Brains were dissected in Hank’s balanced salt solution (HBSS) supplemented with HEPES (pH 7.4, 2.5 mmol/L), D-glucose (30 mmol/L), CaCl2 (1 mmol/L), MgCl2 (1 mmol/L), and NaHCO3 (4 mmol/L), and embedded in low-melting agarose (Lonza) preheated at 42°C. The embedded brains were cut into 250-μm slices using a vibratome (Leica). Slices were divided at the hemisphere into two pieces. Brain slices were placed with flat spatulas on top of 0.8-μm pore membranes (Sigma-Aldrich) floating on slice culture media [DMEM, supplemented with HBSS, FBS 5%, D-glucose (30 mmol/L), L-glutamine (1 mmol/L), 100 IU/mL penicillin, and 100 mg/mL streptomycin]. BLI was acquired after generating brain slices (day 0) to confirm the presence of brain metastasis and 3 days after the addition of the inhibitor (day 3), considering for analysis of floating brain slices. Growth rate was obtained by comparing fold increases between day 3 and day 0. In the case of T-cell addition, 20,000 CD8+ T cells were seeded on the top of established brain metastasis in brain slices after 1 day in culture. Control IgG (10 μg/mL, Bio X Cell, cat.# BE0090) or preservative (0.05% sodium azide) was added in the control condition if necessary, and anti-TIMP1 antibody (102D1; 10 μg/mL, Thermo Fisher Scientific, cat.# MS608PABX), anti-TIMP1 (N-terminal; 10 μg/mL, Sigma-Aldrich, cat.# SAB2109118), anti-TIMP1 carboxy terminal end (10 μg/mL, Abcam, cat.# ab38978), and anti-mouse CD8α (100 μg/mL, Bio X Cell, cat.# BE0061) were added to the media on day 0. Brain slices were fixed in paraformaldehyde (PFA) (4%) overnight and followed up by free-floating immunofluorescence.
Cell Culture
Mouse brain metastatic cell lines have been generated as previously described (16, 31). All cell lines were tested negative for Mycoplasma (by qRT-PCR). We did not do cell authentication beyond visual, morphological, and growth rate analyses. The maximum number of passages between thawing and use are 15 for all the cell lines.
B16/F10-BrM were cultured in DMEM supplemented with 10% FBS, 2 mmol/L L-glutamine, 100 IU/mL penicillin/streptomycin, and 1 mg/mL amphotericin B, E0771-BrM were cultured in RPMI1640 medium supplemented with 10% FBS, 1% HEPES, 2 mmol/L L-glutamine, 100 IU/mL penicillin/streptomycin, and 1 mg/mL amphotericin B. B16/F10-BrM-OVAGFP cells were generated by lentiviral-mediated transduction of a truncated nonsecreted OVA-GFP fusion protein (bm1T OVA) generously supplied by D. Sancho (CNIC). HEK 293T cells (cultured in DMEM supplemented with 10% FBS, 2 mmol/L L-glutamine, 100 IU/mL penicillin/streptomycin, and 1 mg/mL amphotericin B) at 70% confluence were transfected in Opti-MEM with Lipofectamine 2000 (Invitrogen) and incubated at 37°C overnight with the corresponding plasmids. Mouse astrocytes were obtained from 1- to 3-day-old pups (16). Brains were mechanically dissociated and filtered through 70-μm filters, and cell suspension was cultured in a petri dish for the next 7 days. After gentle shaking at 37°C overnight, the media were changed.
Astrosphere Assays
Astrospheres were generated as previously described (16). Briefly, mouse astrocytes were obtained from mechanical dissociation of brains from 1- to 3-day-old pups. After 7 days in culture and gentle shaking overnight at 37°C, the media were changed, and astrocyte enrichment was confirmed. Astrocytes were treated with a cytokine cocktail including EGF (0.01 μg/mL, R&D Systems, cat.# 2028-EG-200), macrophage migration inhibitory factor (MIF) (0.1 μg/mL, R&D Systems, cat.# 1978-MF-0257CF), and TGFα (0.1 μg/mL, R&D Systems, cat.# 239-A-100) in DMEM with B27 (1×) for 96 hours. After treatment, 5 × 104 astrocytes were seeded in low attachment plates and incubated for 7 days in the presence of the same media to evaluate the ability to form astrospheres. CM was collected, filtered, and added to activated CD8+ T cells.
Immunoblotting
Lysis buffer (Cell Signaling Technology, cat.# 9803S) with the following protease inhibitors: 200 mmol/L Na3VO4, 500 mmol/L NaF, 100 mmol/L phenylmethylsulfonylfluoride, was used to extract total protein. Protein lysate from the microenvironment was obtained by dissecting luciferase tissue immediately adjacent to luciferase+ cancer cells. Microdissection was initially validated by confirming the absence of GFP+ cells using flow cytometry. Tissue was mechanically desegregated with the FastPrep-24 5G lysis system (MP Biomedicals) by using zirconium beads at 6.0 m/s for 15 seconds followed by 10 minutes of incubation on ice before lysis. For protein quantification, BCA protein color kit was used (Fisher Scientific, cat.# 23227). After denaturalization, 10 to 50 μg of protein lysates were resolved by SDS-PAGE. Transfer to polyvinylidene difluoride membranes (VWR, cat.# 10600021) was carried out in transfer buffer 1× (Alaos, cat.# TT5C-10) 20% methanol for 2 hours 100 V. Blocking was performed with 5% milk, and membranes were washed with TBS-Tween 0.1%. The following primary antibodies: p44/42 MAPK (Erk1/2; 1:1,000, Cell Signaling Technology, cat.# 9107), Phospho-p44/42 MAPK (Erk1/2; 1:1,000, Cell Signaling Technology, cat.# 4370), anti-TIMP1 (1:1,000, Thermo Fisher Scientific, cat.# MS608PABX), anti-CD63 (MX-49.129.5; 1:500, Santa Cruz Biotechnology, cat.# sc-5275), anti-tubulin (1:5,000, Santa Cruz Biotechnology, cat.# sc-17787), anti-vinculin (1:10,000, Sigma, cat.# V9131), and secondary antibodies from Invitrogen (AF680) and Li-Cor Odyssey CLx system (LICORbio) were used for visualization.
Immunoprecipitation
For immunoprecipitation, cocultures of STAT3+ astrospheres (as described in the section Astrospheres Assays) and CD8+ T cells (cultured in vitro) were performed. CD8+ T cells were added over STAT3+ astrospheres (washed with PBS 1× after gentle centrifugation) in a concentration of 6 × 105 CD8+ T cells/1.5 mL of coculture. After 72 hours in culture, 1,000 μg of total protein extract was incubated at 4°C overnight with anti-CD63 (MX-49.129.5; Santa Cruz Biotechnology, cat.# sc-5275) or isotype control (IgG1, Cell Signaling Technology, cat.# #5415) in a concentration of 10 μg/mg of protein. Dynabeads protein-G (Thermo Fisher Scientific, cat.# 10003D) was vortexed and washed two times. Then, 50 μL was incubated with the different fractions for 2 hours at 4°C. Finally, samples were washed and eluted for detection of CD63 and TIMP1 by immunoblotting.
RNA Isolation and cDNA Synthesis
QIAshredder columns (QIAGEN) were used to homogenize the preparation when needed and whole RNA was isolated using the RNeasy Mini Kit (QIAGEN; human and mouse tissue) or PicoPure RNA Isolation Kit (Thermo Fisher Scientific; CD8+ T cells). Then, 150 to 1,000 ng RNA was used to generate cDNA using iScript cDNA Synthesis Kit (Bio-Rad, cat.# 1708890). cDNA from sorted cells was amplified with SsoAdvanced PreAmp Supermix (Bio-Rad, cat.# 1725160).
qRT-PCR
Gene expression was analyzed using SYBR Green gene expression assays (GoTaq qPCR Master Mix, Promega, cat.# A6002). The following mouse genes were used (5′→3′, forward; reverse):
Actin (GGCACCACACCTTCTACAATG; GTGGTGGTGAAGCTGTAGCC),
Timp1 (GAGACACACCAGAGCAGATACC; TGGTCTCGTTGATTTCTGGGG),
Gzmk (GCCATTTATGGCGTCCATCC; CCGGACTGAAGTCGTGAGAA),
Gzmb (CAGGAGAAGACCCAGCAAGTCA; CTCACAGCTCTAGTCCTCTTGG),
S100b (CTGGAGAAGGCCATGGTTGC; CTCCAGGAAGTGAGAGAGCT),
Itgam (AAGCAGCTGAATGGGAGGAC; TAGATGCGATGGTGTCGAGC).
Quantitative PCR reaction was performed on QuantStudio 6 Flex Real-Time PCR System (Applied Biosystems) and analyzed using the software QuantStudio 6 and 7 Flex software.
Bulk RNA-Seq
Total RNA samples (500 ng), with RNA quality score of 9.4 on average (range, 9.0–9.8 on a PerkinElmer LabChip analyzer), were converted into sequencing libraries with the “NEBNext Ultra II Directional RNA Library Prep Kit for Illumina” (NEB #E7760). Briefly, polyA+ fraction was purified and randomly fragmented, converted to double-stranded cDNA, and processed through subsequent enzymatic treatments of end-repair, dA-tailing, and ligation to adapters. Adapter-ligated library was completed by PCR with Illumina PE primers. The resulting purified cDNA libraries were applied to an Illumina flow cell for cluster generation and sequenced on an Illumina NextSeq 550 (with v2.5 reagent kits) by following the manufacturer’s protocols. Raw images generated by the sequencer were submitted to analysis—per-cycle basecalling and quality score assignment with Illumina Real-Time Analysis (RTA) integrated primary analysis software. Conversion of BCL (base calls) binary files to FASTQ format was subsequently performed with Local Run Manager Generate FASTQ Analysis Module (Illumina). Eighty-six base-pair single-end sequenced reads followed adapter and polyA tail removal as indicated by Lexogen. Mouse reads were analyzed with the Nextpresso (https://doi.org/10.2174/1574893612666170810153850) pipeline as follows: sequencing quality was checked with FastQC v0.11.0 (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Reads were aligned to the mouse genome (GRCm39) with TopHat 2.0.10 (48) using Bowtie 1.0.0 (49) and samtools 0.1.19 (50), allowing three mismatches and 20 multihits. The gencode vM26 gene annotation for GRCm39 was used. Read counts were obtained with HTSeq (51). Differential expression and normalization were performed with DESeq2 (52), filtering out those genes in which the normalized count value was lower than two in more than 50% of the samples. From the remaining genes, those that had an adjusted P value below 0.05 FDR were selected.
Immunofluorescence and IHC
For immunofluorescence, fixation with PFA 4% at 4°C was applied to tissue prior to slicing of the brain by using a vibratome (250-μm slices; Leica) or sliding microtome (80-μm slices; Thermo Fisher Scientific). Both types of brain slices were blocked in normal goat serum (NGS) 10%, BSA 2%, Triton 0.25% in PBS for 2 hours at room temperature (RT). Primary antibodies: anti-KI67 (1:500, Abcam, cat.# ab15580, 1.500), anti-HMB45 (1:500, Abcam, cat.# ab732), anti-CD63 (1:100, Santa Cruz Biotechnology, cat.# sc-5275), anti-GFAP (1:1,000, Millipore, cat.# MAB360), anti-GFP (1:1,000, Aves Labs, cat.# e-1020), anti-cleaved caspase-3 (1:500, Cell Signaling Technology, cat.# 9661), anti-CD8 (1:100, Novus Biologicals, cat.# NB200-578), osteopontin (1:100, Santa Cruz Biotechnology, cat.# 21742), and anti-MHC class 1 H2 Db/H2-D1 (1:100, Abcam, cat.# ab25244) were incubated overnight at 4°C in blocking solution and the following day for 30 minutes at RT. After washing in PBS-Triton 0.25%, secondary antibodies: Alexa-Fluor anti-chicken488, anti-chicken647, anti-rabbit555, anti-mouse555, anti-mouse488, anti-mouse647, anti-rat555, and anti-rat488 (Invitrogen, dilution 1:300) were added in blocking solution and incubated for 2 hours. After washing in PBS-Triton 0.25%, the nuclei were stained with bisBenzimide (1 mg/mL; Sigma-Aldrich) for 7 minutes at RT.
IHC of paraffin embedded tissues was performed at the CNIO Histopathology Core Facility. For the different staining methods, the slides were deparaffinized in xylene and rehydrated by a graded ethanol series to water. Several immunohistochemical reactions were performed on an automated immunostaining platform (Autostainer Link 48, Agilent; Discovery XT-ULTRA, Roche/Ventana).
First, antigen retrieval was performed with the appropriate pH buffer and endogenous peroxidase was blocked (3% hydrogen peroxide). The slides were then incubated with the appropriate primary antibody, as detailed in Supplementary Table S23, for single, double, or triple staining. Following the primary antibody, the slides were incubated with appropriate secondary antibodies and with horseradish peroxidase (HRP)–conjugated visualization systems when needed.
The immunohistochemical reaction was revealed using ChromoMap DAB, DISCOVERY Purple, or Teal Kit (Roche/Ventana). Nuclei were counterstained with hematoxylin. Finally, slides were dehydrated, rinsed, and mounted for microscopic evaluation. Positive controls for primary antibodies were included in each staining series.
Lysozyme IHC and RNAscope staining method were performed in an automated immunostaining platform (Ventana DISCOVERY ULTRA, Roche), including deparaffinization and rehydration as a part of the platform protocol with the appropriate probe: TIMP1 mRNA (ACDBio, cat.# 567849 for human and ACD, cat.# 316849 for mouse). After the probe, slides were incubated with the corresponding probe amplification kit (RNAscope VS Universal HRP Reagent Kit, ACD, cat.# 323210), conjugated with HRP, and the reaction was developed using 3,3-diaminobenzidine tetrahydrochloride (DAB Detection Kit, Roche/Ventana, cat.# 760-224).
Proximity Ligation Assay
Interaction between CD63 and TIMP1 was investigated using in situ Duolink (Duolink In Situ Orange Starter Kit Mouse/Rabbit, cat.# DUO92102) according to the manufacturer’s instructions. Paraffin sections were deparaffinized, and antigen retrieval was done by heat-induced epitope retrieval (HIER) in citrate buffer, high pH. Next, the sections were blocked for 1 hour at 37°C and incubated with anti-TIMP1 antibody (1:1,000, Dako, cat.# M7293) and anti-CD63 antibody (1:500, Sigma-Aldrich, cat.# HPA010088) for 30 minutes at 37°C. Proximity ligation assay probes were added, and the sections were incubated for 1 hour at 37°C followed by ligase oligonucleotides added for 30 minutes at 37°C. Finally, amplification solution was added for 100 minutes at 37°C. Then, the slides were incubated with anti-GFAP (1:500, Abcam, cat.# ab4674) and anti-CD8 (1:100, Novus Biologicals, cat.# NB200-578) antibodies for 1 hour at RT followed by several washes and incubation for 1 hour at RT with secondary antibodies (Invitrogen, dilution 1:300). Coverslips were mounted using DAPI to visualize cell nuclei. Only primary antibodies or omission of primary antibodies were used as negative controls.
Image Acquisition and Analysis
Sample selection for analysis was done based on expert histopathological evaluation.
Images were acquired with a Leica SP5 upright confocal microscope 10×, 20×, 40×, and 63× objectives and analyzed with ImageJ software. Whole slides were acquired with a slide scanner (Axio Scan Z1, Zeiss), and images were captured with Zen Blue software (V3.1 Zeiss). Human samples were analyzed with QuPath (53).
Multiplex IHC
To investigate the immune architecture of human and murine brain metastases, we used Opal technology (Akoya Biosciences), which allows simultaneous imaging of several markers within one tissue section. The staining was performed on a Ventana DISCOVERY ULTRA instrument (Ventana Medical Systems) and imaged using the Vectra 3 automated quantitative pathology imaging system (Akoya Biosciences) as described previously (54). In brief, formalin-fixed, paraffin-embedded samples were deparaffinized, rehydrated, and subjected to heat-mediated antigen retrieval for 32 minutes at 95°C in Cell Conditioning Solution (CC1; Ventana Medical Systems, pH 9). Upon incubation of the primary antibody according to Supplementary Table S24, the matching HRP-coupled secondary OmniMap antibody (Ventana Medical Systems) was added for 12 minutes at 36°C. Subsequently, the signal was detected by incubation of the matching Opal fluorophore (Akoya Biosciences) for 8 minutes at RT. Afterward, the antibody complex was removed by heat-mediated stripping with CC2 buffer (Ventana Medical Systems, pH 6) for 24 minutes at 100°C. The incubation of the primary antibody and secondary antibody, fluorophore, and subsequent heat treatment were repeated until all markers were detected. Finally, the nuclei were counterstained with DAPI (Merck), and slides were mounted with a coverslip using Fluoromount-G medium (SouthernBiotech). After whole scanning (×100) of sections using Vectra 3.0 Automated Imaging System (Akoya Biosciences), regions of interest were defined in Phenochart software (Akoya Biosciences) and multispectral images were acquired (×200 magnification). The imaging data were then quantified using inForm (Akoya Biosciences) and R software. Briefly, multispectral images were unmixed using a previously built library consisting of single stained tissue slide for all used fluorophores and DAPI. Subsequently, tissue segmentation and cell segmentation were performed. For quantification of stained cells, a self-learning approach was applied to phenotype all cell types. The downstream analyses were performed in R software using the add-ins phenoptr and phenoptrReports (Akoya Biosciences).
Single-Cell RNA Sequencing
Mouse brains were extracted in pre-cooled D-PBS 1× and processed with the Adult Brain Dissociation Kit (Miltenyi, cat.# 130-107-677) using gentleMACS C Tubes (Miltenyi, cat.# 130-093-237) and the gentleMACS Octo Dissociator (Miltenyi, cat.# 130-096-427). Cell suspension was filtered with a 70-μm strainer and was centrifuged at 300g for 10 minutes at 4°C. For myelin removal, the protocol described by Korin and colleagues (55) was followed. The pellet was resuspended with 7 mL of RPMI1640, at RT, and 3 mL of SIP solution (Stock isotonic Percoll, Sigma Aldrich, cat.# GE17-0891-02) was added to each tube, mixing gently. Gradually, 30% (vol/vol) percoll/cell mixture was layered on top of 2 mL of 70% (vol/vol) SIP in PBS 1×. Samples were centrifuged at 500g, for 30 minutes, at 18°C, with minimal deceleration. The top layer of myelin was removed using a 10 mL pipette, and the solution containing all cellular fractions was centrifuged at 500g, for 7 minutes, at 18°C. The supernatant was discarded, and the cells ready for staining were diluted in cold D-PBS/BSA buffer 0.5%. Cell suspension was magnetic labelled with anti-ACSA2 (Miltenyi, cat.# 130-097-678) microbeads, and the enrichment in glial populations was checked by flow cytometry (BD FACSCanto II) with anti-ACSA2-PE (1:100, Miltenyi, cat.# 130-123-284). For dead cell removal and washing prior to single-cell sequencing, Debris Removal Solution (Miltenyi, cat.# 130-109-398) was used. The effluent containing the live cell fraction was centrifuged at 300g for 10 minutes, washed, and finally resuspended in 1× PBS containing 0.04% BSA in a concentration of 7 × 105 cells/mL, placing the cells on ice. Cells suspended in PBS-BSA were tested for optimal viability and if free of debris and aggregates. Cell sample was loaded onto a 10× Chromium Single Cell Controller chip B (10× Genomics) as described in the manufacturer’s protocol (Chromium Single Cell 3′ GEM, Library & Gel Bead Kit v3, cat.# PN-1000075). Intended targeted cell recovery of ∼10,000 cells, generation of gel beads in emulsion (GEM), barcoding, GEM-RT clean-up, cDNA amplification, and library construction were all performed as recommended by the manufacturer. scRNA-seq libraries were sequenced with an Illumina NextSeq 550 (using v2.5 reagent kits) in paired-end fashion (28 bp + 56 bp bases). The bollito (56) pipeline was used to perform read analysis, as follows: sequencing quality was checked with FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Reads were aligned to the mouse reference genome [GRCm38, vM25 gene annotation from GENCODE (57) with STARsolo (STAR 2.7.3a; ref. 58)]. Seurat 3.2.2 (59) was used to check the quality of sequenced cells, explore and quantify single-cell data, and obtain cell clusters and specific gene markers. For the annotation of the different cell subtypes, signatures from Zeisel and colleagues (60); Habib and colleagues (61) and Batiuk, Martirosyan and colleagues (62) were used. For astrocytes annotation, signatures in Supplementary Table S13 were applied.
For analyzing potential interaction among clusters, we calculated the differentially expressed genes for each cluster and ran a protein–protein interaction network analysis with STRING database (63) information. Ligand–receptor interactions between cluster 3 and cluster 7 were selected and filtered on the basis of the experimental and combined scores.
In the case of human samples, for which written informed consent was obtained from all patients included in this study, Chromium Fixed RNA Profiling was used (10× Genomics). About 25 to 50 mg of fresh biopsies were fixed in 1 mL of fixation buffer containing 4% formaldehyde. After 22 hours at 4°C, fixed tissue was digested with a liberase-based solution using gentleMACS Octo Dissociator (Miltenyi, cat.# 130-096-427) and the following protocol was followed: 20 minutes at 37°C, 50 rpm, spin for 30 seconds at 37°C, 2,000 rpm (clockwise), and spin for 30 seconds at 37°C, 2,000 rpm (counter clockwise). Then, the sample was gently centrifuged, and the pellet was resuspended in 1 mL of chilled quenching buffer. After cell counting and for long-term storage, glycerol (10%) and enhancer were added for cryopreservation. Fixed cell suspensions obtained with the Chromium Next GEM Single Cell Fixed RNA Sample Preparation Kit (10× Genomics, PN-1000414) were processed with the Chromium Fixed RNA Profiling Reagent Kit according to the manufacturer’s instructions (10× Genomics, user guide CG000527). Briefly, samples were hybridized to a human transcriptome probe set (Chromium Fixed RNA Kit, 10× Genomics, PN-1000474) and encapsulated in GEM in a Chromium iX instrument (10× Genomics). GEM recovery and gene expression library construction were all performed as recommended by the manufacturer. Libraries were sequenced with an Illumina NextSeq 550 (using v2.5 reagent kits) in a paired-end fashion (28 bp + 56 bp bases). Raw images generated by the sequencer were submitted to analysis, per-cycle basecalling, and quality score assignment with Illumina RTA integrated primary analysis software (RTA v2). Conversion of BCL (base calls) binary files to FASTQ format was subsequently performed with bcl2fastq2 (Illumina). For data analysis, Cell Ranger 7.0.0 was used to generate the count matrices that were then subjected to QC procedures in R to discard cells with low counts across all genes. Filtered matrices were normalized by scaling and normalization (64) using the batchelor R package (https://bioconductor.org/packages/release/bioc/html/batchelor.html). Dimensionality reduction, graph-based cell clustering, and cluster visualization using the scran R package (https://bioconductor.org/packages/release/bioc/html/scran.html) were performed. Clusters were automatically annotated as described in Wang and colleagues (65). SingleR was used to predict the cell type using the Human Primary Cell Atlas (66) as the reference. Doublet detection on clustering results was based on two approaches from the scDblFinder R package (https://bioconductor.org/packages/release/bioc/html/scDblFinder.html). The first approach detects doublets as clusters with expression profiles lying between two other clusters, and the second involves artificially stimulating doublets from the expression data and then training a classifier to identify putative doublet calls among real cells. For integration, datasets were corrected for differences in gene detection and sequencing depth. Batch effects were addressed using the mutual nearest neighbors with the fastMNN function of the batchelor package (67).
Flow Cytometry
Cell suspensions were obtained from brains processed according to Korin and colleagues (55) or from the spleen of 10- to 15-week-old C57BL/6 mice. For T cells in in vitro experiments, the spleens were pressed through a 70-μm cell strainer, and red blood cells were lysed with ACK Lysing Buffer (Lonza, cat.# 10-548E). For intracellular staining of CD8+ T cells in in vitro culture, eBioscience Cell Stimulation Cocktail (plus protein transport inhibitors; 500×) were used (2 µL per ml, Invitrogen, cat.# 00-4975-9). The resulting cells suspensions were incubated for 10 minutes with Fc Block (1:100, BD Biosciences, cat.# 553141) in staining buffer (eBioscience, cat.# 00-4222-26) and incubated for 30 minutes with the corresponding primary antibodies (Supplementary Table S25) in staining buffer. In the case of intracellular staining, BD Cytofix/Cytoperm Fixation/Permeabilization Kit (BD Biosciences, cat.# 554714) was used. After washing, cells were resuspended in staining buffer and acquired on FACSymphony, LSRFortessa X20, or FACSCanto II flow cytometers (BD Biosciences) with optimized settings through voltration experiments. Cell sorting experiments were carried out on a FACSAria IIu cell sorter (BD Biosciences).
Rhapsody
For tissue dissociation, mouse brains were transferred to RPMI1640 medium and dissociated gently using a 15-mL Dounce homogenizer, and then, the protocol described by Korin and colleagues (55) was followed. The top layer of myelin was removed, cells from the interphase were collected with a Pasteur pipette and washed with staining buffer (PBS−/−, containing 5% FBS and 2 mmol/L EDTA). Cells were centrifuged (10,000g, 1 minute, 4°C) and stained for flow cytometry. Target population (DAPI-CD45+CX3CR1−) was sorted in an FACSAria Fusion sorter (BD Biosciences) into 1.5 mL LowBind Eppendorf tubes (Eppendorf, cat.# 0030122348). In some cases, cells were separated by magnetic beads using Mouse CX3CR1+ Selection Kit (MojoSort, cat.# 480056) to remove unwanted cells, and the negative fraction was collected in LowBind Eppendorf tubes. For scRNA-seq cell capture, library preparation, and sequencing and analysis, each sample was barcoded with the Single Cell Labelling of BD Single-Cell Multiplexing Kit following the manufacturer’s instructions. Single-cell capture and cDNA synthesis preparation were performed following the manufacturer’s instructions with BD Rhapsody. mRNA targeted and sample tag library preparations were done according to BD Rhapsody Targeted mRNA and AbSeq Amplification Kit protocol using BD Rhapsody Immune Response Panel Mm kit (cat.# 633753). The concentration of PCR products and amplified libraries were determined with a Qubit fluorometer using the Qubit dsDNA HS Assay Kit (Invitrogen, cat.# Q32854). Their size distribution was assessed by running an aliquot on an Agilent Technologies 2100 Bioanalyzer, using an Agilent High Sensitivity DNA chip (Agilent Technologies, cat.# 5067-4626). Sequencing was performed in a NovaSeq 6000 system. Library demultiplexing and targeted gene-expression library were aligned using Seven Bridges Genomics platform following the BD Rhapsody pipeline (BD Biosciences). Cell clustering and gene expression analysis was performed using Seurat v4.1.1 (68).
T Cells in In Vitro Culture
CD8+ T cells were obtained from the spleen of 10- to 15-week-old C57BL/6 female mice. The whole organ was pressed through a 70-μm cell strainer and red blood cells were lysed with ACK Lysing Buffer (Lonza, cat.# 10-548E). Cells were resuspended in HBSS 1× supplemented with 2% FBS and 1 mmol/L EDTA at a concentration of 108 cells/mL. EasySep Mouse CD8+ T cell Isolation Kit (StemCell Technologies Inc., cat.# 19853A) protocol was followed as indicated by the manufacturer to select total CD8+ T cells. Dynabeads Mouse T-Activator CD3/CD28 (Thermo Fisher Scientific, cat.# 11456D) was used to activate CD8+ T cells in culture. After 24 hours, the Dynabeads were removed from the culture with the help of a magnetic particle concentrator. CD8+ T cells were cultured in RPMI1640 medium supplemented with 10% FBS, 2 mmol/L L-glutamine, 1 mmol/L sodium pyruvate, 100 IU/mL penicillin/streptomycin, 50 µmol/L β-Mercaptoethanol, and 1 mmol/L HEPES and human IL2 (Miltenyi, cat.# 130097743). When using CD8+ T cells sorted from the spleen, T cells were activated with anti-mouse CD3e clone 145-2C11 (1 µg/mL, BD Biosciences, cat.# 553066) coated plates, soluble anti-mouse CD28 (37.51; 1 µg/mL, Tonbo Biosciences, ref. 70-0281-U500), and mouse IL2 (0.1 µg/mL, Miltenyi Biotec, cat.# 130-094-054), in RPMI medium supplemented with 10% FBS and penicillin–streptomycin. CD8+ T cells were maintained in culture for 1 day before CM from STAT3+ and STAT3− astrospheres was added. Two to three days after addition of CM, flow cytometry was performed using the appropriated conjugated antibodies. Activated CD8+ T cells incubated with CM from astrospheres were added to B16/F10-BrM cells in the ratio 1:5 (cancer cell:CD8+ T cell) for viability assays that were analyzed by bioluminescence.
OT-I T cells extracted from the spleen of Tg(TcraTcrb)1100Mjb/J and maintained in in vitro culture after stimulation with 40 pmol/L ovalbumin-derived SIINFEKL peptide (29) were used in cytolysis assays in the ratio 1:4 (cancer cell:CD8+ T cell).
Anti-TIMP1 antibody (102D1; 10 µg/mL, Thermo Fisher Scientific, cat.# MS608PABX) or rTIMP1 (100 ng/mL, R&D Systems, cat.# 980-MT) was added at day 0 when indicated.
Phosphoproteomics
CD8+ T cells were obtained from the spleen of 10- to 15-week-old C57BL/6 female mice and selected, activated, and expanded as described above by using EasySep Mouse CD8+ T cell Isolation Kit (StemCell Technologies Inc., cat.# 19853A). After CM from astrospheres was added, cell density was maintained at 500,000 cells/mL. Two days after the addition of CM, T-cell pellets were washed with PBS 1× three times, and the sample was prepared for proteomic analysis.
Lymphocytes were lysed for 15 minutes at 95°C in 5% SDS, 100 mmol/L Tris/HCl, pH 8.0. After cooling, the lysate was incubated at 25°C with 10 units of DNAse (Benzonase, Merk) and sonicated for 10 minutes in a Bioruptor for DNA shearing. Protein concentration was determined using BSA as the standard. Then, samples were digested on bead protein aggregation capture with MagReSyn Hydoxyl microparticles (ratio protein/beads 1:5) in an automated KingFisher instrument (Thermo Fisher Scientific). Proteins were digested for 16 hours at 37°C, with 300 µL of a mixture of trypsin/LysC in 50 mmol/L TEAB pH 8.0 (TrypZean trypsin, Sigma-Aldrich, LysC endoprotease, Wako, protein:enzyme ratio 1:100 each). Resulting peptides were speed-vac dried and redissolved in 100 μL of 200 mmol/L HEPES, pH 8.5.
Samples (approximately 100 μg) were labeled, 1 hour at 25°C, using Thermo Fisher Scientific TMTpro 18plex Isobaric Label Reagent. Reaction was quenched/stopped by adding 5% hydroxylamine. Samples were mixed in a 1:1 ratio based on the total peptide amount, which was determined from an aliquot by comparing overall signal intensities on a regular LC/MS-MS run. The final mixture was finally desalted using a Sep-Pak C18 cartridge (Waters) and dried prior to high pH reverse-phase high performance liquid chromatography (HPLC) prefractionation.
Labeled peptides were prefractionated offline by means of high pH reverse-phase chromatography using an Ultimate 3000 HPLC system equipped with a sample collector. Briefly, peptides were dissolved in 100 μL of phase A (10 mmol/L NH4OH) and loaded onto an XBridge BEH130 C18 column (3.5 μm, 150 mm length, and 1 mm ID; Waters). Phase B was 10 mmol/L NH4OH in 90% CH3CN. The following gradient (flow rate of 100 μL/minutes) was used: 0 to 50 minutes 0% to 25% B, 50 to 56 minutes 25% to 60% B, and 56 to 57 minutes 60% to 90% B. One-minute fractions from minute 15 to 65 were collected, neutralized with 10 μL of 10% formic acid and immediately vacuum dried. Based on UV absorbance at 280 nm, 40 fractions were pooled in eight fractions for phosphopeptide enrichment.
Phosphopeptides were enriched with MagReSyn Zr-IMAC HP beads in an automated KingFisher instrument, using the manufacturer's protocol. Eluted fractions, enriched in phosphopeptides, were immediately acidified with 10% formic acid and dried in a vacuum dryer. Flow-through for each pool was further fractionated by micro reversed phase (RP) high pH in four fractions and kept for total proteome analysis.
LC/MS-MS was done by coupling an UltiMate 3000 RSLCnano LC system to an Orbitrap Exploris 480 mass spectrometer (Thermo Fisher Scientific). Samples were loaded into a trap column (Acclaim PepMap 100 C18 LC columns 5 μm, 20 mm in length) for 3 minutes at a flow rate of 10 μL/minutes in 0.1% formic acid (FA). Then, peptides were transferred to an EASY-Spray PepMap RSLC C18 column (Thermo Fisher Scientific; 2 μm, 75 μm × 50 cm) operated at 45°C and separated using a 90-minute effective gradient (buffer A: 0.1% FA; buffer B: 100% ACN, 0.1% FA) at a flow rate of 250 nL/minutes. The gradient used was from 4% to 6% of buffer B in 5 minutes, from 6% to 25% B in 70 minutes, from 25% to 45% B in 14 minutes, and 10 additional minutes at 98% B.
The mass spectrometer was operated in a data-dependent mode, with an automatic switch between mass spectrometry (MS) and MS/MS scans using a top 15 method (intensity threshold ≥ 5e4, dynamic exclusion of 20 seconds, and excluding charges unassigned, +1 and ≥ +6). MS spectra were acquired from 350 to 1,500 m/z with a resolution of 60,000 full width and high maximum (FWHM) (200 m/z). Ion peptides were isolated using a 0.7 Th window and fragmented using higher energy collisional dissociation with a normalized collision energy of 36. MS/MS spectra were acquired with a fixed first mass of 120 m/z and a resolution of 45,000 FMHW (200 m/z). The ion target values were 3e6 for MS [maximum injection time (IT) 25 ms] and 1e5 for MS/MS (maximum IT, auto). For data analysis, raw files were processed with MaxQuant (v 2.1.4.0) using the standard settings against a mouse protein database (UniProtKB/TrEMBL, 21,990 sequences) supplemented with contaminants. Carbamidomethylation of cysteines was set as a fixed modification, whereas oxidation of methionines, protein N-term acetylation, and phosphorylation of S, T, Y, and N/Q deamidation were variable modifications. Minimal peptide length was set to seven amino acids, and a maximum of two tryptic missed-cleavages were allowed. Results were filtered at 0.01 FDR (peptide and protein level).
Afterward, the phosphosite or protein intensities files were loaded in ProStaR [v1.30.0; Wieczorek and colleagues (69), Bioinformatics, 2017] using the intensity values for further statistical analysis. Briefly, proteins/sites with less than 18 valid values were filtered out. Then, a global normalization of log2-transformed intensities across samples was performed using the cyclic loess normalization (LOESS) function. Differential analysis was done using the empirical Bayes statistics limma. Proteins with a P value < 0.05 and a log2 ratio >0.3 or <−0.3 were defined as regulated. The FDR was estimated to be below 5% by Benjamini–Hochberg.
Sampling of Human Tissues
Human brain metastasis tissue, peripheral blood, and CSF were collected by CNIO Biobank as the backbone of a collaborative nationwide multicenter cohort, RENACER, integrated by 19 different hospitals and coordinated from the CNIO Biobank. Written informed consent for each donor was collected from each patient included in this study and surplus diagnostic samples were shipped to CNIO in less than 24 hours from surgery, under controlled temperature and other preanalytical variables, to warranty homogeneity and quality of the cohort. All the studies were conducted in accordance with recognized ethical guidelines (Declaration of Helsinki) and were approved by our institutional review board (IRB; CEI PI 25_2020-3). Comprehensive clinical information was also collected by CNIO Biobank associated to the samples.
Patient-Derived Organotypic Brain Cultures
Surgically resected human brain metastases which have the advantage of including the immune tumor microenvironment from patients with lung cancer (seven cases), breast cancer (two cases), melanoma (four cases), or other primary sources (two cases) were obtained from the CNIO Biobank that previously received them from Hospital Universitario 12 de Octubre, Complejo Hospitalario Universitario de Albacete, Hospital Álvaro Cunqueiro Vigo, Complejo Universitario de Navarra, Hospital Universitario de Burgos, and Hospital Universitario de Bellvitge. All samples were in compliance with protocols approved by our IRB (CEI PI 25_2020-3). Written informed consent was signed by all patients included in this study. PDOCs were generated as described previously (70). Briefly, after neurosurgical resection, brain metastasis samples were directly collected in Neurobasal-A media (Thermo Fisher Scientific, cat.# 21103049) supplemented with 1 μg/mL amphotericin B, 100 IU/mL penicillin/streptomycin, 25 ng/mL basic human fibroblast growth factor, 100 ng/mL IGF1, 25 ng/ mL EGF, 10 ng/mL neuroregulin1-β1 (R&D Systems, cat.# 396-HB), 1× N-2 supplement (Gibco, cat.# 17502048), and 1× B27 supplement. Organotypic brain cultures were prepared as described above. Slices from brain metastases were cultured in the presence of human IgG (Bio X Cell, cat.# BE0092), anti-TIMP1 (Thermo Fisher Scientific, cat.# MS608PABX), and anti-CD8 (Bio X Cell, cat.# BE0004-2) at 10 μg/μL for 3 days. Brain slices were fixed in 4% PFA overnight at 4°C, and then free-floating immunofluorescence was performed. Proliferation was evaluated by manually counting Ki67+ nuclei from cancer cells.
Spheroids Assays
Human samples were disaggregated mechanically, ACK Lysing Buffer (Lonza, cat.# 10-548E) was used to lyse red cells, and the samples were digested with DMEM supplemented with 0.125% collagenase III and 0.1% hyaluronidase at 37°C for 45 minutes. After PBS 1× washing, cells were resuspended in Neurobasal-A media supplemented as described for PDOCs, and astrospheres CM and drugs were added (anti-TIMP1, 10 μg/mL, Thermo Fisher Scientific, cat.# MS608PABX). Spheroids were maintained in culture in low attachment plates for a maximum of 3 days. For immunofluorescence staining, spheroids were fixed using Cytospin (Thermo Fisher Scientific) and PFA (4%).
Clinical Samples
Brain metastases from lung cancer (seven cases), breast cancer (three cases), melanoma (11 cases), or from other primary origins (four cases) were obtained from the CNIO Biobank that previously received them from Hospital Universitario 12 de Octubre, Complejo Hospitalario Universitario de Albacete, Hospital Álvaro Cunqueiro Vigo, Complejo Universitario de Navarra, Hospital Universitario de Burgos, and Hospital Universitario de Bellvitge. All samples were in compliance with protocols approved by our IRB (CEI PI 25_2020-3) and the Institutional Review Board of the Department of Neuroscience, University of Turin. Written informed consent was signed by each patient included in this study. Cases were selected to include only samples with peritumoral tissue in order to evaluate the microenvironment surrounding brain metastasis. IHC was performed at the CNIO Histopathology Core Facility using standardized automated protocols and multiplex was performed at the Institute of Immunology (Faculty of Medicine Carl Gustav Carus).
TIMP1 Detection in Liquid Biopsies
To determine the concentration of TIMP1 in mice plasma, around 500 μL of blood were centrifuged (500g for 10 minutes at 10°C, and the resulting supernatant fraction, again at 3,000g for 20 minutes at 10°C) immediately after the extraction. For detection of TIMP1 secreted in mice CSF, CSF was extracted from the cisterna magna of anesthetized animals with a capillary tube, then it was centrifuged 600g for 5 minutes at 4°C. TIMP1 levels were measured using ELISA as indicated by the manufacturer (Sigma-Aldrich, cat.# RAB0468).
For liquid biopsies, a patient cohort of six plasma samples from noncancer patients were obtained from the CIMA, University of Navarra, and from patients with lung cancer brain metastasis (six cases), breast cancer brain metastasis (two cases), melanoma brain metastasis (one case), and brain metastasis with other primary tumors (two cases) were obtained from the CNIO Biobank that previously received them from Hospital Universitario 12 de Octubre and Hospital Álvaro Cunqueiro Vigo. CSF samples from five noncancer patients were obtained from the Biobank of Hospital Universitario Virgen de la Macarena, and from patients with lung cancer brain metastasis (six cases), breast cancer brain metastasis (two cases), melanoma brain metastasis (one case), and brain metastasis with other primary tumors (two cases) were obtained from the CNIO Biobank that previously received them from Hospital Universitario 12 de Octubre and Hospital Álvaro Cunqueiro Vigo. All samples were in compliance with protocols approved by their respective IRB (B.0001601, CEI PI 25_2020-v2, and CEI PI 25_2020-3). Written informed consent was signed by each patient included in this study. TIMP1 levels in patients’ plasma and CSF were measured by ELISA following the manufacturer’s instructions (Sigma-Aldrich, cat.# RAB0466).
Survival Analysis
Survival data of 10 patients with brain metastases from different solid tumors were available. The mean (range) TIMP1 levels of the cohort (5–317 μg/mL) was used to determine high TIMP1 (>167 ng/mL) and low TIMP1 (<167 ng/mL). Kaplan–Meier product limit method was generated for survival estimations. Log-rank test was performed to analyze survival differences between TIMP1 levels in liquor (high vs. low). A two-sided P value of <0.05 was considered to indicate statistical significance.
Immune Cluster Analysis
Transcriptomic data were used to cluster a total of 108 brain metastatic samples into high, medium, and low immune following the methodology in García-Mulero and colleagues (26). Gene expression of selected genes was compared between the three groups by nonparametric methods. To select biomarkers of high immune metastases, the best combination of marker genes was selected from a list of candidate genes by a binary decision tree with cross-validation (k = 10) that identified the optimal classification model for high/low differentiation. R package caret was used to perform the selection. High and Low samples (n = 44) were randomly divided into Training (75%, n = 33) and Test (25%, n = 11) datasets. The Training datasets were used for classification and the Testing dataset for evaluation of prediction accuracy. Prediction accuracy was evaluated by calculating the sensitivity, specificity, and AUC.
For the validation with samples from the RENACER cohort (n = 135) or subcohorts with specific samples, raw reads preprocessing was performed as detailed: QuantSeq 3′ mRNA-Seq reads from brain metastatic samples were processed closely following Lexogen’s QuantSeq 3′ mRNA-Seq Kit and integrated data analysis pipeline on Bluebee platform (015UG108V0140). FastQC (v0.11.9) was used to generate QC reports of the sequencing reads. Raw reads were then trimmed with BBDuk (BBMap v38.93) to remove both the polyA tail and adapter sequences. Trimmed reads were aligned with STAR v2.7.8a (58) to the GRCh38 reference with custom ENCODE settings as suggested by the aforementioned protocol and indexed with samtools v1.14 (50), Finally, mapped reads were counted and aggregated to gene level counts with htseq-count v0.13.5 (51) and the GENCODE v38 comprehensive gene annotation. For count normalization and batch correction, normalization and variance stabilization of the raw counts were performed by DESeq2 v1.34.0, VST function (52). Then, we used limma v3.50.1 (71) to fit a linear model of the normalized counts including both the batch and primary site of each metastatic sample. Afterward, the batch component was removed using removeBatchEffect, although preserving the differences associated with the primary site of the sample. For the immune cluster classification, the normalized and regressed gene expression matrix was used to assess the immune cluster profile of each sample and to cluster them according to the methods of García-Mulero and colleagues (26). For the analysis of RENACER cohort or the specified subsets of samples, single-sample enrichment scores were calculated for a set of immune signatures defined by the authors using the GSVA package (72). Then, samples were grouped by agglomerative hierarchical clustering with ward.D2 as the linkage method over the Euclidean distance of the enrichment scores. Finally, the resulting dendrogram was split with the R package dendextend v1.16.0 to generate three categories, each representing different immune and inflammatory profiles. All the bioinformatics analyses were carried out in R v4.1.1.
Gene Set Enrichment Analysis
GSEAPreranked (73) was used to perform gene set enrichment analysis for the selected signature collections on a preranked gene list according to the t-statistic, setting 1,000 gene set permutations. Gene sets with significant enrichment levels (FDR q value <0.25) were considered.
Quantification and Statistics
Data were analyzed using GraphPad Prism 8 software (GraphPad Software). For comparisons between two experimental groups in datasets that followed a normal distribution, an unpaired, two-tailed Student t test was used. For multiple comparisons, ANOVA test was performed. For survival curves, P values were obtained with log-rank (Mantel–Cox) two-sided tests. χ2 test was performed for the comparison of group proportions. For CD8+CD63+ T cell qPCRs, a relative scale is used for the representation that takes the minimum and maximum values for each gene.
Data Availability
The MS proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD040436. Bulk RNAseq data from CD8+ T cells have been deposited to Gene Expression Omnibus (GEO) with the dataset identifier GSE228364. scRNA-seq data from experimental brain metastasis have been deposited to GEO with the dataset identifier GSE228368, and scRNA-seq data from human samples have been deposited to GEO with the dataset identifier GSE254379. Rhapsody scRNA-seq data have been deposited to GEO with the dataset identifier GSE228379.
Authors’ Disclosures
N. Priego reports grants from Asociación Española Contra el Cáncer during the conduct of the study. A. de Pablos-Aragoneses reports grants from Fundación La Caixa during the conduct of the study. L. Álvaro-Espinosa reports grants from Spanish Ministry of Economy and Competitiveness during the conduct of the study. R. Rudà reports grants from Bayer, as well as personal fees from Novocure, Servier, Genenta, and CureVac outside the submitted work. M. Schmitz reports grants from Federal Ministry of Education and Research, cofunded by the European Commission, and the Federal Ministry of Education and Research during the conduct of the study. M. Valiente reports grants from AstraZeneca outside the submitted work. No disclosures were reported by the other authors.
Authors’ Contributions
N. Priego: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. A. de Pablos-Aragoneses: Formal analysis, investigation, visualization, methodology. M. Perea-García: Formal analysis, investigation, visualization, methodology. V. Pieri: Formal analysis, investigation, visualization, methodology. C. Hernández-Oliver: Data curation, formal analysis, methodology. L. Alvaro-Espinosa: Formal analysis, investigation, visualization, methodology. A. Rojas: Formal analysis, investigation, visualization, methodology. O. Sánchez: Formal analysis, investigation, visualization, methodology. A. Steindl: Data curation, formal analysis, investigation, visualization, methodology. E. Caleiras: Formal analysis, supervision, visualization, methodology. F. García: Data curation, formal analysis, investigation, methodology. S. García-Martín: Data curation, formal analysis, methodology. O. Graña-Castro: Data curation, formal analysis, methodology. S. García-Mulero: Data curation, formal analysis, methodology. D. Serrano: Formal analysis, investigation, visualization, methodology. P. Velasco-Beltrán: Data curation, formal analysis, investigation, methodology. B. Jiménez-Lasheras: Formal analysis, investigation, visualization, methodology. L. Egia-Mendikute: Formal analysis, investigation, visualization, methodology. L. Rupp: Data curation, formal analysis, investigation, visualization, methodology. A. Stammberger: Data curation, formal analysis, investigation, visualization, methodology. M. Meinhardt: Supervision; provided and processed the human samples and collected clinical data. A. Chaachou-Charradi: Resources, formal analysis, supervision. E. Martínez-Saez: Human samples and clinical evaluation. L. Bertero: Supervision; provided and processed the human samples and collected clinical data. P. Cassoni: Supervision; provided and processed the human samples and collected clinical data. L. Mangherini: Supervision; provided and processed the human samples and collected clinical data. A. Pellerino: Resources, formal analysis, supervision. R. Rudà: Supervision; provided and processed the human samples and collected clinical data. R. Soffietti: Supervision; provided and processed the human samples and collected clinical data. F. Al-Shahrour: Formal analysis, supervision, visualization, methodology. P. Saftig: provided the CD63-null mice. R. Sanz-Pamplona: Formal analysis, supervision, validation. M. Schmitz: Resources, formal analysis, supervision. S.J. Crocker: provided the Timp1loxP/loxP mice. A. Calvo: Formal analysis, supervision, visualization, methodology. A. Palazón: Resources, formal analysis, supervision. RENACER: Resources, investigation, methodology; provided and processed the human samples and collected clinical data. M. Valiente: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.
Acknowledgments
We thank all members of the Brain Metastasis Group, A. Schietinger, A. Gros, and A.A. Boire for critical discussion of the article and the CNIO Core Facilities for their excellent assistance, especially the Monoclonal Antibodies Core Unit that provided us with the following plasmids: PCMV6-hTIMP1-myc-DDK, PCDNA3.1-mTIMP1, PCDNA3.1-mTIMP2, PCDNA3-hTIMP3 and PCMV6-mTIMP3-myc-DDK. We also thank EuroMed as a supplier of silymarin for in vivo experiments; D. Sancho (CNIC) and the members of the Melanoma Group from CNIO for Tg(TcraTcrb)1100Mjb/J mice, the OVA-OT-I system, and their help with cytolysis assays. We want to particularly acknowledge the patients and the Biobank Nodo Hospital Virgen Macarena (Biobanco del Sistema Sanitario Público de Andalucía) integrated in the Spanish National Biobanks Network (PT20/00069) supported by ISCIII and FEDER funds, for their collaboration in this work. This study was funded by MINECO SAF2017-89643-R, SAF2014-57243-R, SAF2015-62547-ERC (to M. Valiente), Fundació La Marató de TV3 (141; to M. Valiente and A. Calvo), Fundación Ramón Areces CIVP19S8163 (to M. Valiente) and CIVP20A6613 (to E. Ortega-Paino.), H2020-FETOPEN (828972; to M. Valiente), Cancer Research Institute—Clinic and Laboratory Integration Program CRI Award 2018 (54545; to M. Valiente), LAB AECC 2019 (LABAE19002VALI; to M. Valiente), ERC CoG (864759; to M. Valiente), ERANET-TRANSCAN-3 (TRANSCAN2021-2023; to M. Valiente), with funds from Instituto de Salud Carlos III/NextGenerationEU/PRTR (AC20/00114) and FC AECC (TRNSC213878VALI), Federal Ministry of Education and Research (03ZU1111LB) and co-funded by the European Commission (01KT2304B; to M. Schmitz), MICINN (PID2019-107956RA-I00; to A. Palazón), LAB AECC 2021 (LABAE211744PALA; to A. Palazón), ERC-StG (804236; to A. Palazón), NIH–NS078392 (to S.J. Crocker), La Caixa INPhINIT Fellowship (grant LCF/BQ/DI19/11730044; to A. de Pablos-Aragoneses), MINECO–Severo Ochoa PhD Fellowship (BES-2017-081995; to L. Alvaro-Espinosa); and an AECC postdoctoral fellowship (POSTD19016PRIE; to N. Priego), Gobierno Vasco predoctoral fellowship (PRE_2019_1_0320; to B. Jiménez-Lasheras), FPI predoctoral fellowship (PRE2020-092342; to P. Velasco-Beltran), Ramón y Cajal Programme fellowships: RYC2018-024183-I (to A. Palazón) and RYC2022-038084-I (to D. Serrano); M. Valiente is an EMBO YIP member (4053). CNIO is supported by the ISCIII, the Ministerio de Ciencia e Innovación, and is a Severo Ochoa Center of Excellence (SEV-2015-0510).
Note: Supplementary data for this article are available at Cancer Discovery Online (http://cancerdiscovery.aacrjournals.org/).