Astrocyte-induced Cdk5 expedites breast cancer brain metastasis by suppressing MHC-I expression to evade immune recognition

Abstract

Brain metastases (BrMs) evade the immune response to develop in the brain, yet the mechanisms of BrM immune evasion remains unclear. This study shows that brain astrocytes induce the overexpression of neuronal-specific cyclin-dependent kinase 5 (Cdk5) in breast cancer-derived BrMs, which facilitates BrM outgrowth in mice. Cdk5-overexpressing BrMs exhibit reduced expression and function of the class I major histocompatibility complex (MHC-I) and antigen-presentation pathway, which are restored by inhibiting Cdk5 genetically or pharmacologically, as evidenced by single-cell RNA sequencing and functional studies. Mechanistically, Cdk5 suppresses MHC-I expression on the cancer cell membrane through the Irf2bp1-Stat1-importin α-Nlrc5 pathway, enabling BrMs to avoid recognition by T cells. Treatment with roscovitine-a clinically applicable Cdk5 inhibitor-alone or combined with immune checkpoint inhibitors, significantly reduces BrM burden and increases tumour-infiltrating functional CD8⁺ lymphocytes in mice. Thus, astrocyte-induced Cdk5 overexpression endorses BrM immune evasion, whereas therapeutically targeting Cdk5 markedly improves the efficacy of immune checkpoint inhibitors and inhibits BrM growth.

Extended Data Fig. 1 |. Cdk5 is overexpressed in BrM and correlates with reduced overall survival from breast cancer but does not directly promote the proliferation of breast cancer BrM in vitro and in vivo.

A. CDKs expression in MDA-MB-231.Br3 BrM was normalized to their corresponding expression in MFP tumours, and ranked based on BrM-specific upregulation (N = 6 and 5, respectively). B. CDK5 expression in MFP tumours and BrMs from A. N = 5 and 6 biological replicates. Two-sided T-test. C. CDK5 expression in breast cancers and matched BrM (N = 13). Each dot indicates a patient. Paired two-sided T-test. D. Cdk immunoblotting in the indicated cell lines. E. Survival of mice bearing BrMs from 4T1.Br3.shCtl and 4T1.Br3.shCdk5#2 cells (N = 11 and 12 mice, respectively). Logrank test. F. Cdk5 immunoblotting in BrMs from EO771.Br3. shCtl and EO771.Br3.shCdk5 cells (N = 2 mice per group). G. p-Cdk5 staining in BrMs of mice ICA-injected with indicated cells. Scale bar = 50 μm. H, I Ki-67⁺ (H) and TUNEL⁺ (I) cells in BrMs of mice ICA-injected with indicated cells. Each dot indicates a lesion. Data pooled from N = 4 and 3 mice per group, respectively. 2-way ANOVA with Tukey’s multiple comparisons test. J. Correlation analysis between CDK5 expression and DNA replication signature across human BrM datasets. K, L Proliferation of EO771.Br3 (K) and 4T1.Br3 (L) cells transfected with control or Cdk5 shRNA (N = 2 biological replicates). Each dot indicates the mean. M. Hard agar assay of indicated cells. N = 4 biological replicates. Two-sided T-test. N. 4-day BrM seeding experiment of EO771.Br3.GFP.shCtl and EO771.Br3.GFP.shCdk5 cells. Brains stained as indicated. Scale bar = 20 μm. O. Quantification of GFP⁺ cells in brains from N. 15 fields of view per mouse. Each dot indicates a mouse. Two-sided Mann–Whitney U Test. P. UMAP plot. Q, R. Expression of Mki67 (Q) and Cdk5 (R) in cancer cells sorted from BrMs from EO771.Br3.shCtl or EO771.Br3.shCdk5 cells (N = 3 mice per group). Each dot indicates a cell, violin shape indicates data distribution, red dot indicates the mean, bounds of the box indicate 1^st and 3^rd quartiles, whiskers indicate the range. Two-sided Wilcoxon rank-sum test.Throughout, results are shown as means ± SEM. N.S. – not significant.

Extended Data Fig. 2 |. Cdk5 promotes CD8⁺ T cell infiltration in mouse and human BrM and boosts the antigen-specific CD8⁺ T cell killing in vitro.

A. UMAP plots illustrate predicted immune populations in BrM from EO771. Br3.shCtl or EO771.Br3.shCdk5 cells (N = 3 mice per group). T cell clusters are circled in red. B. Major immune populations in BrM from EO771.Br3.shCtl or EO771.Br3.shCdk5 cells (N = 3 mice per group) based on scRNA-seq analysis. N = 4 technical replicates. C. Major T cell subsets in BrM from EO771.Br3.shCtl or EO771.Br3.shCdk5 cells (N = 3 mice per group) based on scRNA-seq analysis. N = 4 technical replicates. D. IHC profiling of major immune populations in BrM lesions of mice ICA-injected with EO771.Br3 or 4T1.Br3 cells transfected with control or Cdk5 shRNA (N = 3 mice per group). Cell number was normalized to BrM area. Each dot indicates a BrM lesion. Two-sided T-test. E. Representative IHC staining of CD8⁺ TILs in BrM lesions of mice ICA-injected with T11.Br1 cells transfected with control or Cdk5 shRNA (N = 3–5 mice per group). Scale bar = 100 μm. F. Two-sided Spearman’s correlation analysis of the expression of CDK5 and 3-gene TIL signature (CX3CR1, FGFBP2, FCGR3A) in 35 breast cancer BrM tissues from the GSE52604 human dataset. Each dot represents a patient. G. Spearman’s correlation analysis of the expression of CDK4/CDK6 and 12-gene TIL signature (CD8A, CCL2, CCL3, CCL4, CXCL9, CXCL10, ICOS, GZMK, IRF1, HLA.DMB, HLA.DOA, and HLA.DOB) in 35 breast cancer BrM tissues from the GSE52604 human dataset. Each dot represents a patient. H. Design of an experiment to test whether Cdk5 affects antigen-specific T cell killing of BrM cells in vitro. I. Quantification of remaining live cells 48 h after co-culture of indicated cancer cells with or without OT-I T cells (1:1 effector:target ratio). N = 3 biological replicates. Two-way ANOVA with Tukey’s post-hoc test. J. Representative images of culture wells from the experiment in J. Scale bar = 400 μm. Throughout, results are shown as means ± SEM. N.S. – not significant. TIL – tumour-infiltrating lymphocytes.

Extended Data Fig. 3 |. Cdk5 does not affect Pd-l1 in human or experimental BrM but negatively regulates APP and MHC-I gene expression in BrM both in vitro and in vivo.

A. Mean fluorescence intensity of Pd-l1 on the surface of indicated cells. One-way ANOVA with Dunnett’s post-hoc test. B. Immunoblotting for Pd-l1 in the indicated cell lines transfected with control or Cdk5 shRNA. C. mRNA expression of Cd274 in GFP⁺ single cells sorted from mice bearing BrM from GFP⁺ EO771.Br3.shCtl (N = 2,676 cells) or EO771.Br3.shCdk5 cells (N = 2,606 cells). Error bars indicate maximum and minimum values, box bounds indicate 1^rd and 3^rd quartiles, centre values indicate median, dots indicate outliers. Two-sided Wilcoxon rank-sum test. D. Two-sided Spearman’s correlation analysis of the expression of CD274 and CDK5 in BrM patients from GSE164150, GSE52604 and Varešlija et al. human datasets. N = 47, 35, and 21 patients, respectively. Each dot represents a patient. E. Expression of indicated APP genes in GFP⁺ single cells sorted from mice bearing BrM from GFP⁺ EO771.Br3.shCtl (N = 2,676 cells) or EO771.Br3.shCdk5 cells (N = 2,606 cells). Each dot indicates a cell, violin shape indicates data distribution, red dot indicates the mean, bounds of the box indicate 1st and 3rd quartiles, whiskers indicate the range. Two-sided Wilcoxon rank-sum test. F. Two-sided Spearman’s correlation analysis of the expression of human CDK5 and human HLA-A/B/C expression in mouse BrM lesions from MDA-MB-231.Br cells (GSE19184). Each dot represents a biological replicate. G. Two-sided Spearman’s correlation analysis of the expression of CDK5 and HLA-A/B/C in single cancer cells from human breast cancer BrM lesions (GSE186344). Tables below display a cross-correlation (Spearman’s) between HLA-A/B/C genes. Only samples with R²>0.5 for all cross-correlations were considered as passed the quality check. For A, results are shown as means ± SEM. N.S., not significant. QC – quality check, BrM – brain metastasis, Corr – correlation.

Extended Data Fig. 4 |. Cdk5 knockdown promotes MHC-I re-expression in BrM cells, and it is dependent on Cdk5’s kinase activity.

A. Flow cytometry histograms showing surface expression of MHC-I in indicated cell lines. B. Mean fluorescence intensity of samples in A. C. Quantification of surface MHC-I staining intensity in cultured 4T1.Br3.shCtl and 4T1.Br3.shCdk5 #2 cells. Each dot indicates a field of view. N = 11 fields of view for 4T1.Br3.shCtl and 10 fields of view for 4T1.Br3.shCdk5 #2, pooled from at least 3 culture wells. Two-sided T-test. D. Representative MHC-I staining of 4T1.Br3.shCtl and 4T1. Br3.shCdk5#2 cells. Scale bar = 20 μm. E. mRNA expression of APP genes in 4T1.Br3.shCtl and 4T1.Br3.shCdk5#1 cells. N = 3 technical replicates. F. mRNA expression of APP genes in T11.Br1.shCtl and T11.Br1.shCdk5#2 cells. Und. – signal undetected. N = 3 technical replicates. G. Histograms showing surface expression of HLA-A/B/C in the indicated cell lines. H. mRNA expression of HLA-A/B/C genes, ERAP1, and CDK5 in the indicated cell lines. N = 3 biological replicates. Two-sided T-test. I. EO771.Br3 cells were treated with 7.5 μM RSV, harvested at indicated timepoints, and probed for indicated proteins. J. DKAT cells were transfected with an empty vector, CDK5-overexpressing vector (CDK5^wt), and a vector containing dominant negative CDK5 mutation (CDK5^kd), followed by probing for indicated proteins. Kinase activity of CDK5^kd cells is impaired compared to CDK5^wt cells. K. Flow cytometry histograms showing surface expression of MHC-I in DKAT cells transfected with vectors overexpressing wild-type CDK5 (CDK5^wt), defective dominant negative CDK5 mutant (CDK5^kd), or control (Vec. Ctl). L. Expression of HLA-A and HLA-B genes in MDA-BM-231.Br3 primary MFP breast tumours (N = 5 mice) and MDA-BM-231.Br3 BrM (N = 6 mice). Two-sided T-test. M. Expression of HLA-A and HLA-B genes in breast tumours and matched BrM (N = 21 patients) from the dataset of Varešlija et al. Paired two-sided T-test. N. Expression of HLA-A and HLA-B genes in breast tumours and matched BrM (N = 16 patients) from the GSE125989 dataset. Paired two-sided T-test. For L-N, results are shown as means (red line). For other panels, results are shown as means ± SEM. N.S. – not significant. BrM – brain metastasis.

Extended Data Fig. 5 |. MHC-I overexpression diminishes experimental BrM in immunocompetent mice.

A. mRNA expression of H2-k1 in mouse primary MFP tumours and matched BrMs derived from EO771.Br3 cells (N = 2 biological replicates). B. Immunoblotting of indicated proteins in HEK923FT producer cell lines transfected with pLenti-C-Myc-DDK-P2A-Puro (Vec. Ctl) or H2-k1-C-Myc-DDK-P2A-Puro (H2-k1) vectors. C. Representative H&E staining scans of brains from mice ICA-injected using 4T1.Br3.Vec.Ctl or 4T1.Br3.H2-k1 cells. Metastatic lesions are outlined in yellow. Scale bar = 4 mm. D. Frequency of metastatic lesions (total) in immunocompetent mice bearing BrM from 4T1.Br3.Vec.Ctl or 4T1.Br3.H2-k1 cells. Each bar indicates an animal. E. Frequency of metastatic lesions (>100 μm²) in immunocompetent mice bearing BrM from 4T1.Br3.Vec. Ctl or 4T1.Br3.H2-k1 cells. Each bar indicates an animal. F. Representative IHC staining of CD8, CD4, GZMB and Ki-67 in immunocompetent mice bearing BrM from 4T1.Br3.Vec.Ctl or 4T1.Br3.H2-k1 cells (N = 4 mice per group). Scale bar = 100 μm. G. IHC quantification of CD8, CD4, GZMB and Ki-67 in immunocompetent mice bearing BrM from 4T1.Br3.Vec.Ctl or 4T1.Br3.H2-k1 cells (N = 4 mice per group). Cell number was normalized to the BrM area. Each dot indicates a BrM lesion. Two-sided Mann–Whitney U test. Throughout, results are shown as means ± SEM. BrM – brain metastasis, MFP – mammary fat pad.

Extended Data Fig. 6 |. Cdk5 regulates the expression and nuclear localization of Stat1 and Nlrc5 to control the surface MHC-I expression in BrM-seeking cells.

A. mRNA expression of MHC-I-driving transcription factors in 4T1.Br3 and T11.Br1 cells transfected with control or Cdk5 shRNA. Cdk5 expression was probed as a control. N = 3 technical replicates. B. mRNA expression of Stat1 and Nlrc5 in EO771.Br3 cells transfected with control or Cdk5 shRNA. Cdk5 expression was probed as a control. N = 4 technical replicates except for Stat1 (N = 3 technical replicates). C. Stat1 expression in GFP⁺ cancer cells sorted from mice bearing BrM from EO771.Br3.shCtl or EO771.Br3.shCdk5 cells (N = 3 mice per group). Error bars indicate maximum and minimum values, box bounds indicate 1^rd and 3^rd quartiles, centre values indicate median, dots indicate outliers. Wilcoxon rank-sum test. D. Immunoblotting of indicated proteins in cytoplasm and nuclear fractions from EO771.Br3 cells transfected with control or Cdk5 shRNA. E. mRNA expression of Nlrc5 in 4T1.Br3.shCdk5 cells transfected with control or Stat1 shRNA. N = 3 technical replicates. F. Immunoblotting of indicated proteins in 4T1.Br3.shCdk5 cells transfected with control or Stat1 shRNA. G. Flow cytometry histograms showing surface expression of MHC-I in 4T1.Br3 cells transfected with shRNA targeting Cdk5, Cdk5+Stat1, or control. H. mRNA expression of Kpna2 (3 primer pairs) in EO771.Br3 and 4T1.Br3 cells transfected with control or Cdk5 shRNA. N = 3 biological replicates. Two-sided T-test. I. Immunoblotting of Nlrc5 and Importin ɑ in cytoplasm and nuclear fractions from 4T1.Br3.shCdk5 cells treated with vehicle or ivermectin for 24 h. J. IHC analyses of mouse BrM (N = 3 per group) from T11.Br1 cells transfected with control or Cdk5 shRNA. Representative staining images are shown. Scale bar = 100 μm. K. IHC quantification of staining in J. Each dot indicates a BrM lesion. Two-sided T-test. Throughout, results are shown as means ± SEM. N.S. - not significant.

Extended Data Fig. 7 |. Cdk5 expression is induced by educated astrocytes but not cancer-associated fibroblasts or microglial cells.

A. In vitro cultured parental and paired BrM-seeking cells or patient-derived xenografts were probed for CDK5. B. Representative bright-field photograph of primary mouse astrocytes cultured in vitro. C. mRNA expression of the indicated genes in primary mouse astrocytes. N = 4 technical replicates for all genes except for Trem2, s100b, and Ndrg2 (N = 3 technical replicates). D-F. Indicated cell lines were cultured in CM from CAFs (D), BV-2 microglial cells (E) or primary mouse microglia (F) and probed for Cdk5 at indicated timepoints. G,H. EO771.GFP cells were directly co-cultured with CAFs (G), BV-2 microglial cells (G) or primary mouse microglia (H), GFP-sorted at indicated timepoints, and probed for Cdk5. I,J. 4T1 cells were treated as indicated followed by immunoblotting for Cdk5. K. Cytokine array of CM from primary mouse astrocytes, naïve or pre-educated by 4T1.Br3 CM. L. Pre-proteomics verification of educated astrocytes CM’s ability to induce Cdk5 in 4T1 cells by immunoblotting. M. Indicated cell lines were cultured in the presence of recombinant Versican at the indicated doses for 4 h or 8 h, followed by Cdk5 immunoblotting. N. 4T1 cell lysates were immunoprecipitated by Emilin-1 antibody or IgG, and immunoprecipitation efficiency was evaluated by Emilin-1 immunoblotting. PDX – patient-derived xenografts, CAF – cancer-associated fibroblast, CM – conditioned media, EV – extracellular vesicles, IgG – Immunoglobulin G.

Extended Data Fig. 8 |. RSV penetrates the blood-brain barrier when administered in vivo and does not cause major toxicities, while inhibiting the growth of breast cancer BrM.

A. 4T1.Br3 cells were treated with 7.5 μM RSV, harvested at different timepoints, and probed for indicated proteins. B. 4T1.Br3 and T11.Br1 cells were treated with RSV or vehicle as indicated, harvested, and probed for indicated proteins. C. 4T1.Br3 cells were pre-treated with 7.5 μM RSV or vehicle for 10 days, followed by the addition of the activated CD8⁺ T cells (E:T = 10:1). Bars indicate a lactate dehydrogenase activity indicative of cell death, in supernatant samples collected 48 h after addition of T cells. Each dot indicates a biological replicate. Two-sided T-test. D. Mass spectrometry analysis of RSV standard and BrM sample isolated from mice bearing EO771.Br3 BrM. The table shows RSV concentration in BrM lesions. N = 3 biological replicates. E. Dynamics of mouse weight changes upon twice daily oral gavage of RSV (100 mg/kg) or vehicle. Two-way ANOVA. F. Top: Representative H&E images of livers of EO771.Br3 BrM-bearing mice after 2 weeks of twice daily oral gavage treatment with vehicle or RSV. Scale bar = 100 μm. Bottom: Serum concentration of AST, ALT and LDH in EO771.Br3 BrM-bearing mice after 2 weeks of twice daily oral gavage treatment with vehicle or RSV. Two-sided T-test. G. Representative H&E staining of brains from mice ICA-injected using 4T1.Br3 or T11 cells and treated with vehicle or RSV. BrMs are outlined in yellow. Scale bar = 3 mm. H. Model illustrating the role of Cdk5 in breast cancer BrMs. Communication between BrM cells and BrM-educated astrocytes leads to the secretion of Emilin-1 by astrocytes, which in turn induces overexpression of Cdk5 in BrMs. Through phosphorylation of Irf2bp1, Cdk5 restricts phosphorylation and expression of Stat1 and Nlrc5, leading to reduced MHC-I expression on the surface of BrM cells, which eventually leads to immune evasion followed by metastatic outgrowth in the brain. Targeting Cdk5 by small molecule inhibitor roscovitine may control the outgrowth of breast cancer BrM. For C, results are shown as means ± SEM. N.S – not significant, RSV – roscovitine, Veh – vehicle, ICA – internal carotid artery, BrM – brain metastasis, LDH – lactate dehydrogenase, AST – aspartate transaminase, ALT – alanine transaminase.

Fig. 1 |. CDK5 is overexpressed in human BrM and restricts antigen presentation and the T cell response.

a, BrM-specific upregulation of CDKs. Median fold change in CDK gene expression levels in BrMs normalized to their corresponding expression in matched primary breast tumours and ranked based on their BrM-specific upregulation. b, CDK5 expression in primary breast cancers and matched BrMs. Each dot indicates a patient. Paired two-sided Student’s t-test. a,b, n = 21 patients. c, H-scores (left) and representative staining images (right) of CDK5 in primary breast cancers and BrMs. Box-and-whisker plots: the box bounds indicate the first and third quartiles, the centre values indicate the median, the whiskers indicate the range and each dot indicates a patient. Two-sided Student’s t-test. Magnified views of the regions outlined in red boxes are provided. d, Overall survival of patients with breast cancer BrMs expressing high (n = 7) or low (n = 14) CDK5 levels. HR, hazard ratio; z-score cutoff = 0.081; log-rank test. a,b,d, Data from Varešlija and colleagues. e, Survival of immunocompetent mice bearing BrMs of the indicated cell lines transfected with shCtl or shCdk5. Log-rank test; n = 7–12 biological replicates. f, Top pathways enriched in mouse EO771.Br3.shCdk5 compared with EO771. Br3.shCtl GFP⁺ BrMs based on scRNA-seq. MHC-I-associated pathways are highlighted in red. The gene enrichment test was hypergeometric based. False-detection rate was used to control for multiple testing error rate. Pathways associated with antigen presentation and MHC are highlighted in red. g, Immune populations, determined by IHC analysis, in the BrMs derived from the indicated cells (n = 3 mice per group). Each dot indicates a lesion; two-way analysis of variance (ANOVA) with Tukey’s multiple comparisons test. h, Representative IHC staining of CD8⁺ T cells in BrM lesions of mice ICA-injected with T11.Br1 cells transfected with shCtl or shCdk5 (n = 4 mice per group). i, Two-sided Spearman’s correlation analysis of CDK5 and a TIL signature expression in human breast cancer BrM tissues (GSE52604). Each dot represents a patient; n = 35. j, Survival of immunocompromised nude mice bearing BrMs of EO771.Br3.shCtl or EO771. Br3.shCdk5 cells. k, Survival of CD8-depleted immunocompetent C57BL/6J mice bearing BrMs derived from EO771.Br3.shCtl or EO771.Br3.shCdk5 cells. e,j,k, n, biological replicates); log-rank test. Scale bars, 100 μm. Results are shown as the mean ± s.e.m. NS, not significant.

Fig. 2 |. Cdk5 confers resistance to antigen-specific T cell killing in vitro.

a, Detection of Cr release from Cr⁵¹-loaded 4T1.Br3 cells incubated with activated CD8⁺ T cells (effector:target, 1:1) for 48 h. Cr⁵¹-unloaded cells were used as a negative control and 4T1.Br3.shCtl cells treated with 1% SDS were used as a positive control with 100% Cr release. One-way ANOVA with Dunnett’s post test; n = 4 or 3 biological replicates. b, Survival fraction of cells remaining at 48 h following co-culture of the indicated cancer cell lines with activated CD8⁺ T cells (effector:target, 1:1); n = 3 or 12 biological replicates. c, Levels of LDH activity in cell supernatants collected 48 h following co-culture of the indicated cell lines with activated CD8⁺ T cells (effector:target, 10:1). LDH activity in wells without CD8⁺ T cells was used as control; a.u., arbitrary units. d, EO771.Br3.shCtl and EO771.Br3. shCdk5 cells were pulsed with the OVA peptide, labelled using a two-colour live-tracking system, incubated with OT-I T cells and the fraction of surviving cells was quantified using flow cytometry. e, Representative flow cytometry plots of EO771. Br3.shCtl-OVA-DiO and EO771.Br3.shCdk5-OVA-DiD cells that were cultured separately or as a mix with and without the addition of OT-I T cells for 48 h. The percentages indicate the proportion of cells in the respective quadrant. f, Survival fraction of DiO⁺ (EO771.Br3.shCtl-OVA) and DiD⁺ (EO771.Br3.shCdk5-OVA) cells after co-culture with activated OT-I T cells for 48 h at the indicated effector:target ratios. g, Levels of LDH activity in cell supernatants collected 24 h after the start of the experiment in d and e. f,g, Two-way ANOVA with Tukey’s post test. h, Survival fraction of EO771.Br3.OVA.shCtl and EO771.Br3.OVA.shCdk5 cells after co-culture with activated OT-I T cells for 48 h at the indicated effector:target ratios. c,f–h, n = 3 biological replicates. a,b,f,h, Data were normalized to cell numbers in wells without added T cells. b,c,h, Two-sided Student’s t-test. Results are shown as the mean ± s.e.m. NS, not significant.

Fig. 3 |. Cdk5 suppresses the expression of MHC-I and APP genes in breast cancer BrM.

a,b, Recurrence-free survival (RFS) of patients with breast cancer with low or high HLA-A, HLA-B and HLA-C expression levels assessed individually (a) and combined (b). Data derived from KMPlotter; log-rank test; n = 3,951 (a) and 846 (b) patients, respectively. c, Levels of mRNA expression of the indicated genes in tumour cells sorted from mice bearing BrMs derived from GFP⁺ EO771. Br3.shCtl (n = 2,676 cells) or EO771.Br3.shCdk5 cells (n = 2,606 cells). Each dot indicates a cell, the violin shape indicates the data distribution and the red dot indicates the mean (exact values are provided on the graphs). Box-and-whisker plots: the bounds of the box indicate the first and third quartiles and the whiskers indicate the range. Two-sided Wilcoxon rank-sum test. d, Membrane and whole-cell lysate (WCL) fractions of the indicated cell lines were probed for MHC-I and Cdk5, respectively. e, Levels of mRNA expression of MHC-I genes in the indicated cell lines. f, Levels of mRNA expression of APP genes in EO771.Br3.shCtl and EO771.Br3.shCdk5 cells. e,f, n = 3 technical replicates. g, Membrane and WCL fractions from the indicated cell lines were probed for MHC-I and CDK5, respectively. h, Membrane and WCL fractions from vector control and CDK5-overexpressing MDA-MB-231 cells treated with vehicle or ascending concentrations of RSV were probed for MHC-I (HLA-A/B/C) and CDK5, respectively. i, MHC-I mean fluorescence intensity (MFI) of DKAT cells transfected with vectors overexpressing wild-type CDK5 (CDK5^wt), defective dominant negative CDK5 mutant (CDK5^kd) or control. One-way ANOVA with Dunnett’s post-hoc test; n = 3 biological replicates. j, Histograms showing surface expression of MHC-I (H-2kd) in 4T1.Br3 cells transfected with the indicated vectors. k, Magnetic resonance images of mouse brains ICA-injected with 4T1. Br3.Vec.Ctl (top) or 4T1.Br3.H2-k1 (bottom) cells. The arrowhead indicates a lesion. Scale bar, 5 mm. l, Metastasis areas in mice that were ICA-injected with either 4T1.Br3.Vec.Ctl or 4T1.Br3.H2-k1 cells. The dashed red line indicates the mean. Two-sided Mann–Whitney U-test; n = 6 and 7 mice (biological replicates), respectively. e,f,i,l, Results are shown as the mean ± s.e.m.; NS, not significant. Vec.Ctl, vector control.

Fig. 4 |. Cdk5 inhibits MHC-I expression through the Stat1–importin α–Nlrc5 pathway.

a, Levels of mRNA expression, determined from RNA-seq, of MHC-I-driving transcription factors in cultured EO771.Br3.shCtl (n = 2 biological replicates) and EO771.Br3.shCdk5 cells (n = 2 shRNA clones). Rep., replicate. b, Immunoblots for the indicated proteins in cancer cell lines transfected with shCtl or shCdk5. c, Consensus MHC-I promoter region. Adapted with permission from ref. , Elsevier. d,e, Immunoblots for the indicated proteins in the cytoplasmic and nuclear fractions of 4T1.Br3 (d,e) and EO771.Br3 (e) cells transfected with shCtl or shCdk5. f, Immunoblots for the indicated proteins in the cytoplasmic and nuclear fractions of 4T1.Br3.shCdk5 cells transfected with shCtl or shRNA targeting Stat1 (shStat1). g, Membrane and WCL fractions of 4T1.Br3 cells transfected with shCdk5 and shStat1 or shCtl were probed for MHC-I (H2-kd) and Cdk5, respectively. h, Immunoblots for importin ɑ and β in EO771.Br3 and 4T1.Br3 cells transfected with shCtl or shCdk5. i, Immunoblots for importin ɑ, Stat1 and Cdk5 in 4T1.Br3.shCdk5 cells transfected with shCtl or shStat1. j, Immunoblots for NLRC5 and importin ɑ in the cytoplasmic and nuclear fractions of 4T1.Br3.shCdk5 cells treated with vehicle or ivermectin for 24 h. k, IHC analyses of BrMs from 4T1.Br3 cells transfected with shCtl or shCdk5 (n = 3 per group). Representative staining images (left) and quantification (right; each dot indicates a BrM lesion) are shown. Magnified views of the regions in the red boxes are provided next to the main images. Scale bar, 100 μm (main images) and 20 μm (magnified views). Two-sided Student’s t-test; results are shown as the mean ± s.e.m. l, Two-sided Pearson’s correlation analysis between HLA-A/B/C, NLRC5 and STAT1 expression in human BrM datasets. Gapdh was used as a control.

Fig. 5 |. Cdk5 inhibits Stat1 expression and Stat1^pS727 phosphorylation by phosphorylating Irf2bp1 at S66.

a, Schematic of the experimental design of stable-isotope labelling by amino acids in cell culture. b, Venn diagram (left) and list (right) of proteins significantly dephosphorylated (log₁₀(fold change) < −0.3) in both 4T1.Br3.shCdk5–1 and T11.Br1.shCdk5–1 cells compared with the corresponding shCtl controls. Red boxes indicate Irf2bp1. c, Irf2bp1 was immunoprecipitated from 4T1.Br3 (left) and T11.Br1 (right) cells transfected with shCtl or shCdk5, followed by immunoblotting for pSerine and Irf2bp1. d, Cell lysates of 4T1.Br3.shCdk5 and T11.Br1.shCdk5 cells as well as the corresponding shCtl controls were separated in parallel on 12% SuperSep Phos-tag (top) and 12% bis-Tris (bottom) gels and then probed for Irf2bp1. Lagging bands at the top indicate Irf2bp1 with varying degrees of phosphorylation (heavy or low). e, Cdk5 was immunoprecipitated from the indicated cells, followed by Irf2bp1 immunoblotting. f, In vitro kinase assay testing the ability of recombinant human CDK5 plus p25 complex (rhCDK5/p25) to directly phosphorylate mouse Irf2bp1 or human IRF2BP1. RSV was used as a control. Gels for [ɣ−³²P] Irf2bp1 assays and CDK5 immunoblotting were run in parallel. g, T11.Br1 and 4T1.Br3 cells transfected with single guide RNA (sgRNA) targeting Irf2bp1 (sgIrfbp1) or control sgRNA (sgCrl) were probed for the indicated proteins. For 4T1.Br3, seven single colonies from two different sgRNAs were used. For T11.Br3, two single colonies from two different sgRNAs were used. h, Levels of mRNA expression of the indicated genes in 4T1.Br3 cells transfected with sgCrl or sgIrf2bp1. Results are shown as the mean ± s.e.m.; two-sided Student’s t-test; NS, not significant; n = 3 biological replicates. i, 4T1.Br3.sgIrf2bp1 cells transfected with vectors expressing wild-type Irf2bp1 (WT), a non-phosphorylatable mutant (S66A) or a phosphomimetic mutant (S66D) were probed for the indicated proteins. IP, immunoprecipitate; WB, western blot.

Fig. 6 |. BrM-educated astrocytes promote Cdk5 production in BrM cells through emilin-1 secretion.

a, CDK5 (GSE149283) expression across mouse metastatic sites. One-way ANOVA with Dunnett’s post-hoc test; n = 20, 18, 20, 11 and 18 biological replicates. b, H-scores of CDK5 staining in BrMs and matched extracranial metastases (ECM). Two-sided Student’s t-test; n = 12 and 16 patients. c, Overview of the experimental design. WB, western blot. Credit: Alina Yuzhalina. d, Cdk5 staining in MFP tumours and paired BrMs. e, MFP tumours and paired BrMs of three mice were probed for Cdk5. f, MFP tumours and paired BrMs (day 0) as well as those explanted and cultured in vitro (day 6) were probed for Cdk5. g, The indicated cells were transwell co-cultured with astrocytes and probed for Cdk5 at the indicated time points. h,i, The indicated cells were cultured in CM from naive (h) or educated (i) astrocytes and probed for Cdk5 at the indicated time points. j, Overview of the experimental design. LC–MS/MS, liquid chromatography with tandem mass spectrometry. k, Proteins overrepresented in CM from educated astrocytes compared with control astrocytes (determined from LC–MS/MS). FC, fold change; n = 2 biological replicates. l, Astrocytes were cultured in DMEM medium or CM from 4T1.Br3 cells, followed by emilin-1 immunoblotting. m, Astrocytes were cultured in DMEM medium or CM from EO771.Br3 cells, followed by Emilin-1 profiling. Two-sided Student’s t-test. n, Mouse BrMs (top) and unaffected healthy brains (bottom) were stained for GFAP and emilin-1. o, Representative images of human BrMs co-stained for CDK5 and GFAP (left). Comparison of CDK5 positivity in astrocyte-rich and astrocyte-poor areas (right); each dot indicates a field of view; paired two-sided Student’s t-test. p, The indicated cell lines were cultured with recombinant mouse emilin-1, followed by Cdk5 immunoblotting at the indicated time points. q, Cdk5 immunoblot of 4T1 cells incubated in DMEM medium (Ctl) or CM from educated astrocytes previously treated overnight with IgG control or antibody to emilin-1 (Ab). r, Cdk5 positivity (left) and counts of CD8⁺ T cells (right) in BrMs of control and astrocyte-depleted mice (n = 5 mice per group). Each dot indicates a field of view. Two-sided Student’s t-test. Results are shown as the mean ± s.e.m.; NS, not significant; FOV, field of view. Scale bars, 50 μm.

Fig. 7 |. Pharmacological inhibition of Cdk5 delays BrM growth in vivo through re-expression of MHC-I and an augmented T cell response.

a, Membrane and WCL fractions from the indicated cell lines treated with vehicle control or RSV were probed for MHC-I (H2-k), phospho-Cdk5 (pCdk5) and Cdk5. b, Representative flow cytometry histograms of MHC-I (H2-k) expression in the indicated cell lines treated with vehicle or RSV. IgG2a was used as an isotype control. c, EO771.Br3 cells were pre-treated with 7.5 μM RSV or vehicle for ten days and pulsed with the OVA peptide SIINFEKL before co-culture with OT-I T cells (effector:target, 1:1). Survival fractions of live cells remaining after 48 h of co-culture are provided. Two-sided Student’s t-test; n = 3 biological replicates. **P = 0.0079. d, Number of metastatic lesions in mice bearing BrM from 4T1. Br3 cells following a course of vehicle or RSV (100 mg kg⁻¹) administration. e, Total metastasis area in mice bearing BrM from 4T1.Br3 cells following a course of vehicle or RSV administration. f, Number of metastatic lesions in mice bearing BrM from T11.Br1 cells following a course of vehicle or RSV (100 mg kg⁻¹) administration. d,f, Each bar indicates an animal. The dotted lines indicate the average (avg.) of the respective group. g, Total metastasis area in mice bearing BrM from T11.Br1 cells following a course of vehicle or RSV administration. e,g, Each dot indicates a mouse. Two-sided Mann–Whitney U-test. h,i, Representative staining (h) and numbers (i) of CD8⁺ TILs infiltrating 4T1.Br3 BrMs in mice treated with vehicle or RSV. h, Magnified views of the regions outlined in dashed red lines are provided (right). Scale bars, 100 μm (main images) and 50 μm (magnified views). i, The T cell numbers were normalized to the BrM area of each lesion. Each dot indicates a BrM lesion. Two-sided Student’s t-test; data were pooled from n = 3 mice per group. j, Two-sided Fisher’s exact test to examine the significance of the association between Gzmb (left) and perforin (right) positivity, and vehicle and RSV treatment of 4T1.Br3 BrMs. Results are shown as the mean ± s.e.m.; NS, not significant; Veh, vehicle.

Fig. 8 |. Pre-clinical trial of RSV + anti-PD-1 combination treatment to inhibit breast cancer BrM.

a, Progression-free survival of nivolumab (anti-PD-1)-treated patients with melanoma (GSE91061 dataset; n = 51 total patients) expressing high and low levels of CDK5. A z-score of −0.87 was used as the cutoff; log-rank test. b, EO771.Br3 cells were pre-treated with RSV or vehicle for five days, followed by the addition of the activated CD8⁺ T cells (effector:target, 10:1) in the presence of control IgG or anti-PD-1 and the surviving fraction of cells at 48 h of co-culture was determined. One-way ANOVA with Tukey’s post-hoc test. c, Schedule of the RSV + anti-PD-1 combination treatment in mice ICA-injected with T11 cells. d, Typical histopathological appearance of BrMs derived from T11 cells—that is, clusters of different sizes—stained with haematoxylin and eosin (H&E). The yellow box (right) is a magnification of part of a smaller cluster. Scale bar, 500 μm. e, Number of metastatic clusters of different sizes at the study end point. Two-way ANOVA with Dunnet’s post-hoc test (main row effect). f, Total BrM area at the study end point. One-way ANOVA with Tukey’s multiple comparisons test. e,f, n = 7, 7, 9 and 8 biological replicates. g, IHC analyses of mouse BrMs (n = 3 per group) from the experiment in c. Representative staining images are shown. Magnified views of the regions in boxes outlined in red are shown beneath the main images. Scale bars, 100 μm (main images) and 20 μm (magnified views). h, IHC quantification of BrM-infiltrating CD8⁺ TILs (top left), Gzmb⁺ cells (top middle) and Ki-67⁺ cells (bottom right) as well as pSTAT1 (S727) (bottom left), pCDK5 (bottom middle) and Nlrc5 (top right) H-scores of BrMs from the experiment in g (n = 3 per group). Cell numbers were normalized to the BrM area. Each dot indicates a BrM lesion. One-way ANOVA with Tukey’s post-hoc test. Results are shown as the mean ± s.e.m.; NS, not significant; Veh, vehicle.