Abstract
Bone is the most common site of breast cancer metastasis. Bone metastasis is incurable and is associated with severe morbidity. Utilizing an immunocompetent mouse model of spontaneous breast cancer bone metastasis, we profiled the immune transcriptome of bone metastatic lesions and peripheral bone marrow at distinct metastatic stages, revealing dynamic changes during the metastatic process. We show that cross-talk between granulocytes and T cells is central to shaping an immunosuppressive microenvironment. Specifically, we identified the PD-1 and TIGIT signaling axes and the proinflammatory cytokine IL1β as central players in the interactions between granulocytes and T cells. Targeting these pathways in vivo resulted in attenuated bone metastasis and improved survival, by reactivating antitumor immunity. Analysis of patient samples revealed that TIGIT and IL1β are prominent in human bone metastasis. Our findings suggest that cotargeting immunosuppressive granulocytes and dysfunctional T cells may be a promising novel therapeutic strategy to inhibit bone metastasis.
Significance: Temporal transcriptome profiling of the immune microenvironment in breast cancer bone metastasis revealed key communication pathways between dysfunctional T cells and myeloid derived suppressor cells. Cotargeting of TIGIT and IL1β inhibited bone metastasis and improved survival. Validation in patient data implicated these targets as a novel promising approach to treat human bone metastasis.
Introduction
Despite major advances in early diagnostics and treatment strategies, breast cancer is still the most common malignancy in women and the second leading cause of cancer-related death (1). Mortality from breast cancer is almost exclusively due to metastatic relapse in distant organs. Bone is the most prevalent metastatic site of breast cancer, occurring in up to 70% of patients with advanced disease (2, 3). Bone metastasis from breast cancer is mostly osteolytic and encompasses severe morbidities including pathologic fractures, hypercalcemia, and spinal cord compression (4, 5). Specifically, spinal metastasis is frequent in patients with breast cancer, account- ing for 60% of skeletal metastases (6). Patients with breast cancer bone metastasis are treated with antiosteolytic drugs (e.g., bisphosphonates or anti-RANKL) to reduce bone destruction and alleviate symptoms. Because there are currently no efficient treatments to prevent or inhibit bone metastasis, a bet- ter understanding of the underlying mechanisms is essential to enable the development of novel therapeutic strategies.
The metastatic microenvironment is a crucial facilitator of the seeding and growth of disseminated cancer cells (7). To thrive at the metastatic site, disseminated cancer cells must adapt to organ-specific microenvironments. This requires distinct adaptation pathways and organ-specific cross-talk with resident and recruited stromal cells. Thus, prevention of metastatic relapse requires a better understanding of the unique landscape changes and molecular pathways in each metastatic site.
Bone metastatic relapse of breast cancer was found to be associated with systemic changes, immune modulation, and accumulation of protumorigenic myeloid cells even before the formation of evident metastases (8, 9). Specifically, one of the most abundant immunosuppressive cells are myeloid-derived suppressor cells (MDSC), a heterogeneous population of immature myeloid cells capable of inhibiting cytotoxic T-cell activity, promoting protumorigenic inflammation and recruiting additional inflammatory and immunosuppressive cell subsets (10). MDSCs were found to be increased in blood samples from patients with breast cancer bone metastases compared with patients with visceral metastases (11).
Breast cancer tumors and metastasis are often characterized by an immunosuppressive microenvironment and the presence of dysfunctional T cells. The conundrum of T-cell dysfunction propelled the development of multiple immunotherapy strategies, aiming to restore their normal functions (12). Specifically, immune-checkpoint blockade (ICB), designed to overcome the immune-inhibitory pathways and reactivate killing of tumor cells by cytotoxic T cells, has revolutionized the treatment protocols for several cancer types (13). However, ICB has thus far shown limited efficacy for breast cancer (14).
The composition and functional states of different cells in the metastatic microenvironment can profoundly affect the efficacy of ICB, leading to acquired immune resistance despite possible initial response (15, 16). Therefore, mapping of the transcriptional changes and functional interactions between immune cells in the bone metastatic microenvironment, and, specifically, understanding the distinct mechanisms by which MDSCs interfere with T-cell–mediated antitumor immunity are an urgent need to enable improved immune therapy strategies to inhibit breast cancer metastasis.
We previously established and characterized a model of spontaneous breast cancer bone metastasis following resection of the primary tumor (8). We found that the immune system is markedly disrupted during breast cancer bone metastasis, already at early metastatic stages (8). In the current study, we profiled the transcriptome of myeloid cells and T lymphocytes isolated from bone metastatic lesions and peripheral bone marrow (BM) at distinct metastatic stages, revealing the dynamic changes during the metastatic process.
We identified the key interaction pathways between granulocytes and T cells that are important in shaping an immuno- suppressive bone microenvironment. Functionally, we show that cotargeting immunosuppressive granulocytes and dysfunctional T cells in preclinical trials resulted in attenuated bone metastasis and improved survival.
Results
Breast Cancer Bone Metastases Are Highly Infiltrated by T Cells
To characterize the composition and changes in the immune microenvironment during breast cancer bone metastasis and assess the differences between metastatic lesions and peripheral changes, we used 4T-Bone, a transplantable model of spontaneous breast cancer bone metastasis in immunocompetent mice (Fig. 1A). 4T-Bone is a bone-tropic variant of 4T1 cells, generated by repeated in vivo passaging through the bone (8, 17). Following resection of the primary tumor, aggressive osteolytic bone lesions are formed, which can be assessed by CT imaging showing severe bone destruction (Fig. 1B). We isolated cells from early metastatic BM and late metastatic peripheral BM (i.e., BM from the femur and tibia of bone metastasis–bearing mice, from areas without evident metastases), as well as from bone metastatic lesions, and analyzed the immune milieu by flow cytometry, compared with normal BM (Fig. 1C; Supplementary Fig. S1A and S1B). As expected, we found elevated levels of myeloid cells compared with normal BM, specifically granulocytes (Fig. 1D; Supplementary Fig. S1B–S1D), likely representing enhanced myelopoiesis (8, 18). Surprisingly, although T lymphocytes were dramatically decreased in the BM at both early and late metastatic stages, the metastatic core was highly infiltrated by different T-cell populations (Fig. 1D and 0): T helper (CD3+CD4+), T cytotoxic (CD3+CD8+), and T regulatory (CD3+CD4+CD25+) cell levels were all significantly increased in the metastatic core compared with peripheral BM or with their corresponding normal levels (Fig. 1F–H). Immunostaining of granulocytes and T cells in bone metastatic lesions confirmed vast infiltration of both cell types (Fig. 1I). These findings imply that whereas T cells successfully infiltrate the metastatic lesion, they are nevertheless incapable of efficiently killing cancer cells. We hypothesized that interactions with immunosuppressive granulocytes may be responsible for the ineffectiveness of metastasis-infiltrating T cells. To test this hypothesis, we analyzed gene expression in total RNA isolated from normal or metastatic bones. Analysis of the results revealed that an immunosuppressive gene signature was highly expressed in the metastatic core compared with both normal region-matched control or with peripheral BM from metastasis-bearing mice (Fig. 1J). Upregulated genes included known MDSC markers such as PD-L1, IL1β, IL4Rα, and arginase1 (19). Specific analysis of granulocytes isolated from early and late bone metastatic stages revealed elevated levels of known MDSC markers such as DR5 and SiglecF (Supplementary Fig. S1E and S1F). Thus, immunosuppressive granulocytes may play a role in the dysfunction of metastasis-infiltrating T cells. One of the ways by which MDSCs inhibit T-cell function is by inhibiting their proliferation. To functionally assess the ability of granulocytes to inhibit T-cell activity, we cocultured bone metastasis–infiltrated granulocytes or granulocytes isolated from normal BM with CFSE (carboxyfluorescein succinimidyl ester)-stained CD8+ T cells and analyzed their proliferation rate by flow cytometry (Fig. 1K). Cytotoxic T cells cocultured with granulocytes isolated from bone metastatic lesions had reduced proliferation and increased percentage of nondividing T cells, compared with T cells cultured with control granulocytes (0). To further evaluate the immunosuppressive function of metastasis-associated granulocytes, we measured their arginase activity, characteristic of MDSCs (20), and found that granulocytes isolated from metastases had significantly enhanced arginase activity (Fig. 1O). Moreover, to directly determine the effect of granulocytes on cytotoxic T-cell killing, we immunized mice with 4T-Bone cell lysate, isolated 4T-Bone-antigen–specific cytotoxic T lymphocytes (CTL) and cocultured them with metastasis-associated granulocytes (Fig. 1P). Analysis of the T-cell activation marker CD44 revealed that it was downregulated following incubation of CTLs with metastasis-associated granulocytes (0 and 0). Importantly, we also analyzed CD107a, a functional marker of CTL degranulation during killing activity (21) and found that metastasis-associated granulocytes significantly reduced CD107a on the CTL cell surface, indicating inhibited killing activity (0 and 0). Moreover, CTL-derived GZMB and IFNγ secretion were significantly downregulated upon coculture with metastasis-associated granulocytes, further implicating granulocytes in restraining CTL killing activity (Supplementary Fig. S2A–S2C). Together, these results indicate that granulocytes play a functional role in T-cell inhibition in bone metastatic lesions.
Bone metastases are highly infiltrated by immunosuppressive granulocytes and T cells. A, Scheme of experimental design: orthotopic injection of 4T-Bone to mammary fat pad, followed by primary tumor resection and monitoring of spontaneous bone metastasis. TNBC, triple negative breast cancer. B, Representative CT scans of normal spine and spine with osteolytic metastases, indicated by red arrow. C, Scheme of experimental timeline and tissues analyzed. D–H, Immune landscape of normal BM and BM/lesions at early and late metastatic stages, analyzed by flow cytometry. n = 9, 8, 9, and 9 mice in normal, early BM, late peripheral BM and metastatic core, respectively. D, Proportions of immune cell populations in bones. Tregs; T regulatory cells. E, Representative FACS plots showing T-cell gates. Quantification of infiltrated T cells: (F) CD45+CD3+CD4+ T helper, (G) CD45+CD3+CD8+ T cytotoxic and (H) CD45+CD3+CD4+CD25+ T regulatory cells. Dots represent separate biological repeats. Data represent at least two independent experiments. Dashed lines represent normal BM levels. Data are presented as mean ± SEM, P values were calculated using two-tailed Welch t test. I, IHC of bone metastases stained for Ly6G (granulocytes) or CD3 (T cells), as indicated scale bar = 200 μm. J, Heat map representing analysis of immunosuppression-related gene expression from total RNA isolated from bones of normal or bone metastasis-bearing mice, normalized to MDSCs markers (S100a8 and S100a9). Data represent the average Z-score of n = 3 per normal sample and n = 7 for 4T-Bone groups. P < 0.0001 for Normal BM/ Spinal BM/ Periph BM versus Met-core, P values were calculated using two-way ANOVA, Tukey multiple comparisons test. K, Scheme of CFSE-based CD8+ T-cell proliferation assay following coculture with granulocytes isolated from either normal BM or from bone metastatic lesions. L–N, Representative histogram plots and quantification of proliferation (M) and nondividing T cells (N) in stimulated CD8+ T cells cocultured with granulocytes (n = 4 mice per group). Dots represent technical repeats of the 4 biological repeats. Data are presented as mean ± SEM. P values were calculated using two-tailed unpaired t test. O, Arginase activity in sorted granulocytes isolated from bone metastatic lesions or normal BM. P values were calculated using two-tailed unpaired t test. P, Scheme of killing assay of 4T-Bone–specific CD8+ T cells following coculture with granulocytes isolated from bone metastatic lesions. CD8+ T cells were isolated from 4T-Bone–immunized/vaccinated mice’s spleen. Q–T, Flow cytometry analysis of activation-CD44 and killing-CD107a of cytotoxic T cells following 24-hour w/wo metastasis-associated granulocytes: Q, Representative FACS plots and quantification of CD44+ CD8+ T cells; FMO; fluorescence minus one; R, Histogram of CD44 expression on CD8+ T cells; S, Representative FACS plots and quantification of membrane-bound CD107a+; T, Histogram of CD44 expression on CD8+ T cells. P values were calculated using a two-tailed unpaired t test. MFI, mean fluorescence intensity. Illustrations were created with BioRender.com.
Bone metastases are highly infiltrated by immunosuppressive granulocytes and T cells. A, Scheme of experimental design: orthotopic injection of 4T-Bone to mammary fat pad, followed by primary tumor resection and monitoring of spontaneous bone metastasis. TNBC, triple negative breast cancer. B, Representative CT scans of normal spine and spine with osteolytic metastases, indicated by red arrow. C, Scheme of experimental timeline and tissues analyzed. D–H, Immune landscape of normal BM and BM/lesions at early and late metastatic stages, analyzed by flow cytometry. n = 9, 8, 9, and 9 mice in normal, early BM, late peripheral BM and metastatic core, respectively. D, Proportions of immune cell populations in bones. Tregs; T regulatory cells. E, Representative FACS plots showing T-cell gates. Quantification of infiltrated T cells: (F) CD45+CD3+CD4+ T helper, (G) CD45+CD3+CD8+ T cytotoxic and (H) CD45+CD3+CD4+CD25+ T regulatory cells. Dots represent separate biological repeats. Data represent at least two independent experiments. Dashed lines represent normal BM levels. Data are presented as mean ± SEM, P values were calculated using two-tailed Welch t test. I, IHC of bone metastases stained for Ly6G (granulocytes) or CD3 (T cells), as indicated scale bar = 200 μm. J, Heat map representing analysis of immunosuppression-related gene expression from total RNA isolated from bones of normal or bone metastasis-bearing mice, normalized to MDSCs markers (S100a8 and S100a9). Data represent the average Z-score of n = 3 per normal sample and n = 7 for 4T-Bone groups. P < 0.0001 for Normal BM/ Spinal BM/ Periph BM versus Met-core, P values were calculated using two-way ANOVA, Tukey multiple comparisons test. K, Scheme of CFSE-based CD8+ T-cell proliferation assay following coculture with granulocytes isolated from either normal BM or from bone metastatic lesions. L–N, Representative histogram plots and quantification of proliferation (M) and nondividing T cells (N) in stimulated CD8+ T cells cocultured with granulocytes (n = 4 mice per group). Dots represent technical repeats of the 4 biological repeats. Data are presented as mean ± SEM. P values were calculated using two-tailed unpaired t test. O, Arginase activity in sorted granulocytes isolated from bone metastatic lesions or normal BM. P values were calculated using two-tailed unpaired t test. P, Scheme of killing assay of 4T-Bone–specific CD8+ T cells following coculture with granulocytes isolated from bone metastatic lesions. CD8+ T cells were isolated from 4T-Bone–immunized/vaccinated mice’s spleen. Q–T, Flow cytometry analysis of activation-CD44 and killing-CD107a of cytotoxic T cells following 24-hour w/wo metastasis-associated granulocytes: Q, Representative FACS plots and quantification of CD44+ CD8+ T cells; FMO; fluorescence minus one; R, Histogram of CD44 expression on CD8+ T cells; S, Representative FACS plots and quantification of membrane-bound CD107a+; T, Histogram of CD44 expression on CD8+ T cells. P values were calculated using a two-tailed unpaired t test. MFI, mean fluorescence intensity. Illustrations were created with BioRender.com.
Transcriptome Analysis of Granulocytes during Bone Metastasis Progression Reveals MDSC Characteristics
We, therefore, focused our further analyses on the cross-talk between granulocytes and T cells in the bone metastatic microenvironment. We isolated by FACS sorting four cell populations—granulocytes (CD11b+Ly6CintLy6G+), CD3+CD4+, CD3+CD8+, and CD3+CD4+CD25+ T cells—and profiled their transcriptome by RNA sequencing at three time points during breast cancer bone metastasis. Of note, CD3+CD4+CD25+ T cells may not be exclusively T regulatory cells, as CD25 may be expressed by other populations of activated T cells. Analyses included (1) normal BM, representing the baseline, (2) early metastasis, analyzed at the time of primary tumor resection, and (3) two distinct bone regions from overt metastasis: the metastatic core and peripheral BM of the same mice (Fig. 2A; Supplementary Fig. S2D). Analysis of the transcriptome changes indicated that all cell populations were transcriptionally reprogrammed in bone metastasis (Fig. 2B; Supplementary Figs. S3A–S3C and S4A). To assess the changes that occurred in each specific cell type, we performed multiple comparisons that revealed distinct gene-expression changes across time and locations (Fig. 2C; Supplementary Fig. S4A). We therefore further compared the changes in immune cells in metastatic core or peripheral BM in late metastasis. Pathway analysis using multiple algorithms, including gene set enrichment analysis (GSEA), KEGG, and TRRUST, revealed that the most enriched pathways in bone metastasis–associated granulocytes were classic granulocyte activation signatures such as “neutrophil degranulation”; “inflammatory response”; “TNF signaling”; “complement” and “regulation of cytokine production.” In addition, these granulocytes also upregulated pathways associated with T-cell activity: “PD-1–PD-L1 pathway”; “T-cell receptor signaling”; “regulation of adaptive immune response” and “T-cell receptor signaling” (Fig. 2D and E). Moreover, we analyzed the transcription factors involved and found NFKB1, SP1, and JUN to be the most highly enriched in metastasis-associated granulocytes (Supplementary Fig. S4B). Notably, temporal analysis of changes in granulocyte gene expression revealed enhancement in their expression of an inflammation-related gene signature (Ccl5, Cxcl3, Il1b, Cxcl2, C1qa, and C1qb; Fig. 2F, left, and 0), accompanied by an increase in the expression of lipid metabolism–related genes (G0s2 and Hilpda), previously reported to mediate the protumorigenic effects of neutrophils (Fig. 2F, middle, and 0; refs. 22, 23). Importantly, we also found that metastasis-associated granulocytes exhibited a gradual decreased expression of maturation markers (Cpne3, Trem3, Camp, or Ltf; Fig. 2F, middle, and 0; ref. 24). These changes in gene expression are reminiscent of MDSCs, commonly defined as immature myeloid cells with immunosuppressive capacity. Indeed, staining of isolated metastasis-associated granulocytes showed marked immature cell morphology compared with cells isolated from normal BM: metastasis-associated granulocytes harbored banded nuclei typical of immature granulocytes already at early metastatic stages, whereas nor- mal granulocytes exhibited hypersegmented nuclei typical of polymorphonuclear cells (Fig. 2H; Supplementary Fig. S4C–S4E). Moreover, granulocytes started expressing immunosuppressive genes including Il4ra, Lfitm1, and Lfitm2 already at early metastatic stages followed by upregulation of immunosuppressive genes like Lgals1, Ll1r2, Arg1, Cd274, or Nos2 (10) in late-stage metastasis-associated granulocytes (Fig. 2F, right, and 0). Thus, during bone metastatic progression, granulocytes gradually exhibit morphology and gene signatures characteristic of MDSCs, which was most pronounced in the metastatic core.
Temporal and spatial profiling of bone metastasis–infiltrated leukocytes. A, Experimental design of RNA-seq of granulocytes, T helper, T cytotoxic, and Treg cells from distinct metastatic stages, metastatic core or peripheral BM. B and C, Volcano plots of significantly altered genes P < 1e-06; logFC>2; in normal versus all cancer groups from all 4 cell types (B) and in granulocytes from different comparisons (C). D–F, Analysis of granulocytes isolated from peripheral BM versus metastatic core: D, TRRUST analysis of enriched pathways; E, selection of enriched pathways in granulocytes: GSEA from KEGG pathways enrichment analysis. F, Heat maps of signature genes from inflammation, maturation, metabolism, and immunosuppression pathways in granulocytes. G, Temporal quantification of gene signature scores presented in F. P values were calculated using Tukey multiple comparisons test. H, Representative H&E images of granulocytes from normal BM or 4T-Bone–injected mice at early or late metastatic stages, isolated by FACS and Cytospin. Illustrations were created with BioRender.com.
Temporal and spatial profiling of bone metastasis–infiltrated leukocytes. A, Experimental design of RNA-seq of granulocytes, T helper, T cytotoxic, and Treg cells from distinct metastatic stages, metastatic core or peripheral BM. B and C, Volcano plots of significantly altered genes P < 1e-06; logFC>2; in normal versus all cancer groups from all 4 cell types (B) and in granulocytes from different comparisons (C). D–F, Analysis of granulocytes isolated from peripheral BM versus metastatic core: D, TRRUST analysis of enriched pathways; E, selection of enriched pathways in granulocytes: GSEA from KEGG pathways enrichment analysis. F, Heat maps of signature genes from inflammation, maturation, metabolism, and immunosuppression pathways in granulocytes. G, Temporal quantification of gene signature scores presented in F. P values were calculated using Tukey multiple comparisons test. H, Representative H&E images of granulocytes from normal BM or 4T-Bone–injected mice at early or late metastatic stages, isolated by FACS and Cytospin. Illustrations were created with BioRender.com.
Bone Metastases Are Infiltrated by Dysfunctional Cytotoxic T Cells and T Regulatory Cells
A central protumorigenic role of MDSCs is their ability to inhibit the function of CTLs (refs. 25, 26). We therefore next investigated the transcriptional changes in metastasis- associated CD3+CD8+ CTLs. Analysis of the transcriptomic data revealed that CD8+ cells isolated from bone metastasis exhibited dramatically different gene expression compared with peripheral BM from the same mouse, or with CTLs from early metastatic stage or normal BM (Fig. 3A; Supplementary Figs. S3B, S4A, and S4F). Metastasis-associated CTLs exhibited several upregulated pathways including “TGFβ signaling” and “cytokine-cytokine receptor interaction” (Fig. 3B), as well as enhanced SP1 and NFKB1 activity (Fig. 3C). More detailed analysis of specific gene expression indicated that CD8+ cells isolated from bone metastatic core expressed dramatically increased levels of multiple granzymes [Gzme, Gzmd, Gzmf, Gzmb (27); Fig. 3D, left], accompanied by decreased expression of genes associated with T-cell normal activity (Cd44, Cxcr5, Cxcr3, Sell; Fig. 3D, middle), indicating excessive antigen stim- ulation typical of “exhausted/dysfunctional” CTLs. Indeed, metastasis-associated CD8+ cells expressed high levels of several dysfunction markers including Pdcd1, Fasl, Tox, Tigit, Lag3, Havcr2, and Ctla-4 (Fig. 3D, right; refs. 28, 29). Thus, effector functions of CTLs in bone metastases may be hindered by their conversion into a dysfunctional state marked by increased expression of exhaustion/dysfunction markers.
Metastasis-infiltrated T cells are highly immunosuppressive and dysfunctional. A, Volcano plot of significantly altered genes P < 1e−06; logFC>2; in CD8+ T cells isolated from peripheral BM versus metastatic core. B, GSEA from KEGG pathways enrichment analysis of enriched pathways between CD8+ cells isolated from peripheral BM versus metastatic core. C, TRRUST analysis of transcription factor enrichment in metastatic core versus periphery, list of genes with count >100 and logFC>0.5 (D) Heat map of signature genes in cytotoxic T cells from pathways related to cytotoxic activity, T-cell activity state and dysfunction. E, Volcano plot of significantly altered genes P < 1e–06; logFC>2; of CD4+CD25+ T cells isolated from normal BM vs. metastatic core. F, GSEA from KEGG pathways enrichment analysis of enriched pathways between CD4+CD25+ cells isolated from normal BM versus metastatic core. G, TRRUST analysis of transcription factor enrichment in bone metastasis versus normal, list of genes with count >100 and logFC>0.5. H, Heat map of signature genes in Tregs from pathways related to T-cell activity, dysfunction, Treg/immune-suppressive activity, and IL17 pathway.
Metastasis-infiltrated T cells are highly immunosuppressive and dysfunctional. A, Volcano plot of significantly altered genes P < 1e−06; logFC>2; in CD8+ T cells isolated from peripheral BM versus metastatic core. B, GSEA from KEGG pathways enrichment analysis of enriched pathways between CD8+ cells isolated from peripheral BM versus metastatic core. C, TRRUST analysis of transcription factor enrichment in metastatic core versus periphery, list of genes with count >100 and logFC>0.5 (D) Heat map of signature genes in cytotoxic T cells from pathways related to cytotoxic activity, T-cell activity state and dysfunction. E, Volcano plot of significantly altered genes P < 1e–06; logFC>2; of CD4+CD25+ T cells isolated from normal BM vs. metastatic core. F, GSEA from KEGG pathways enrichment analysis of enriched pathways between CD4+CD25+ cells isolated from normal BM versus metastatic core. G, TRRUST analysis of transcription factor enrichment in bone metastasis versus normal, list of genes with count >100 and logFC>0.5. H, Heat map of signature genes in Tregs from pathways related to T-cell activity, dysfunction, Treg/immune-suppressive activity, and IL17 pathway.
We next analyzed the transcriptome of CD3+CD4+CD25+ isolated T cells and found they were transcriptionally reprogrammed upon bone metastatic progression (Fig. 3E; Supplementary Figs. S3C, S4A, and S4G), into immunosuppressive Tregs. Tregs classically function as central modulators of self-tolerance and have been shown to inhibit T-cell proliferation and function in breast cancer, leading to therapeutic resistance (30). Pathway analysis of differentially expressed genes revealed that metastasis-associated Tregs compared with Tregs from the peripheral BM of the same mouse were enriched for “cytokine signaling”; “T-cell receptor signaling”; and “Th1 Th2 cell differentiation” pathways (Fig. 3F). Moreover, transcription factors analysis revealed highly increased Nfkb1-induced gene regulation (Fig. 3G). Among the upregulated genes, we found genes encoding proteins important for T-cell activation (Cd44, Cd69, Ccr7; ref. 31) and the immunosuppressive features of Tregs [Icos, Tox, Havcr2, Tigit, Ctla4, Il10, Klrg1, and Tnfrsf9 (32); Fig. 3H, left and middle]. Additionally, there was elevated expression of genes linked to the IL17 pathway (Fig. 3H, right). Interestingly, in addition to their classic immunosuppressive function, there is growing evidence that Tregs also produce proinflammatory cytokines, such as IL17A, shown to be correlated with breast cancer bone metastasis (33). Together, these findings suggest that although T cells infiltrate the metastatic core, they are ineffective in killing of cancer cells, due to an immunosuppressive microenvironment conferred by the function of MDSCs and Tregs.
Immune-Checkpoint Interactions Are Highly Enriched between Metastasis-Associated Granulocytes and T Cells
To further test our hypothesis that granulocytes and Tregs mediate the dysfunctional phenotype of metastasis-associated CTLs, we performed receptor–ligand analysis using the ICELLNET transcriptome–based framework (34). ICELLNET uses a database containing over 380 receptor–ligand interactions, which are then classified into families. Communication score is computed for each family and for each receptor–ligand pair proportionally to the mean expression of both (Fig. 4A; ref. 35). Analysis of the results revealed that granulocytes had a higher number of interactions with CD8+ cells compared with either CD4+ or Tregs. Notably, the two cell populations that presented the highest number of interactions were CD8+ and Tregs (Fig. 4B). Moreover, metastasis-associated cells presented overall higher communication scores compared with normal controls (Fig. 4C), further highlighting the profound reshaping of the metastatic microenvironment. Analysis of immune-checkpoint signaling indicated that granulocytes had the highest communication score with CD8+ compared with other T-cell populations (Fig. 4D). Immune-checkpoint interactions were also increased between Treg and CD8+ T cells in bone metastases (Fig. 4E). Interestingly, analysis of the individual inter-actions between granulocytes and CD8+ T cells revealed that genes coding for immune-checkpoint ligands (Tnfsf9, Cd274, or Cd80) were expressed by granulocytes, presumably inter- acting with their corresponding receptors expressed on CD8+ cells (Tnfrsf9, Ctla4, Pdcd1, or Havcr2; Fig. 4F). Similarly, CD4+ T helper and Tregs expressed immune-checkpoint ligands (Pvr, Cd274 for CD4+ and Cd274 for CD25+; 0 and 0). These findings suggest that immune-checkpoint interactions between CTLs and metastasis-associated granulocytes, as well as T helper and Treg cells, shape the immunosuppressive and growth-permissive microenvironment of bone metastasis.
Immune-checkpoint molecules are highly expressed in bone metastasis. A, Illustration of the ICELLNET receptor–ligand analysis platform. Adapted from (34). B, Number of interactions per cell type, scaled 1 = higher score, 0 = lower score. C, Communication scores per cell type, in normal tissue (green) or metastasis-associated (pink). Scores are divided into gene families as indicated. D and E, Communication scores of “immune checkpoint” family in granulocytes (D) and Tregs (E). F–H, Balloon plots of individual interaction scores related to checkpoint receptor–ligand interactions between granulocytes and T-cell subsets (F), CD4+ and CD4+CD25+ cells with other cell subsets (G and H, respectively). I and J, Flow cytometry analysis: representative histograms and quantification of PD-1 (I) and TIGIT (J) expression on CD3+CD8+ from peripheral BM (pink) or metastatic core (purple), FMO (gray). Paired t test metastatic core versus peripheral BM. K and L, Representative dot plots and quantification of CD3+CD8+PD1+ (K) or CD3+CD8+TIGIT+ (L) cell percentage. Dots represent separate biological repeats (n = 7 mice per group). Data are presented as mean ± SEM. P values were calculated using a two-tailed Mann–Whitney test. M, Scheme of immune-checkpoint receptor–ligand interactions between T cells and granulocytes. N and O, Flow cytometry analysis and representative histograms and quantification of PD-L1 (N) and CD155 (O) expression on granulocytes from peripheral BM (pink) and metastatic core (purple), FMO (gray). n = 4 mice per group. Paired t test metastatic core versus peripheral BM. P and Q, Representative histogram and FACS plot of PD-1+ (P) and TIGIT+ (Q) in T-cell populations. R, Average gene expression of checkpoint receptors and ligands in T cells. Illustrations were created with BioRender.com.
Immune-checkpoint molecules are highly expressed in bone metastasis. A, Illustration of the ICELLNET receptor–ligand analysis platform. Adapted from (34). B, Number of interactions per cell type, scaled 1 = higher score, 0 = lower score. C, Communication scores per cell type, in normal tissue (green) or metastasis-associated (pink). Scores are divided into gene families as indicated. D and E, Communication scores of “immune checkpoint” family in granulocytes (D) and Tregs (E). F–H, Balloon plots of individual interaction scores related to checkpoint receptor–ligand interactions between granulocytes and T-cell subsets (F), CD4+ and CD4+CD25+ cells with other cell subsets (G and H, respectively). I and J, Flow cytometry analysis: representative histograms and quantification of PD-1 (I) and TIGIT (J) expression on CD3+CD8+ from peripheral BM (pink) or metastatic core (purple), FMO (gray). Paired t test metastatic core versus peripheral BM. K and L, Representative dot plots and quantification of CD3+CD8+PD1+ (K) or CD3+CD8+TIGIT+ (L) cell percentage. Dots represent separate biological repeats (n = 7 mice per group). Data are presented as mean ± SEM. P values were calculated using a two-tailed Mann–Whitney test. M, Scheme of immune-checkpoint receptor–ligand interactions between T cells and granulocytes. N and O, Flow cytometry analysis and representative histograms and quantification of PD-L1 (N) and CD155 (O) expression on granulocytes from peripheral BM (pink) and metastatic core (purple), FMO (gray). n = 4 mice per group. Paired t test metastatic core versus peripheral BM. P and Q, Representative histogram and FACS plot of PD-1+ (P) and TIGIT+ (Q) in T-cell populations. R, Average gene expression of checkpoint receptors and ligands in T cells. Illustrations were created with BioRender.com.
To validate these predictions in vivo, we initially analyzed the expression of the predicted checkpoint receptors TIM3, LAG3, PD-1, and TIGIT on CD8+ T cells from bone metastatic lesions or from peripheral BM. We found that CD8+ T cells in metastases expressed significantly higher levels of PD-1, TIGIT, and TIM-3 compared with controls (0 and 0; Supplementary Fig. S5A and S5B). Moreover, analysis of the percentage of metastasis-infiltrated CD8+ T cells which express these checkpoint receptors revealed that almost all CD8+ T cells expressed PD-1 and TIGIT (98.5% for PD-1 and 92.8% for TIGIT), whereas only a small fraction expressed TIM-3 (0 and 0; Supplementary Fig. S5C). We therefore decided to further focus on these two immune-checkpoint signaling axes (Fig. 4M). We next analyzed the expression of the corresponding ligands of PD-1 and TIGIT in granulocytes isolated from bone metastatic lesions. The analysis confirmed elevated expression of the immune-checkpoint ligands PD-L1 (PD-1 ligand) and CD155 (TIGIT ligand) compared with granulocytes isolated from peripheral BM of the same mice (0 and 0), confirming the transcriptome analysis results (Supplementary Fig. S5D). Notably, analysis of PD-L1 and CD155 expression on other myeloid cell populations indicated that granulocytes had the highest expression of both ligands (Supplementary Fig. S5E and S5F).
Although originally immune-checkpoint receptors were thought to be exclusively expressed by dysfunctional cytotoxic T cells, recent evidence suggested that other cell types also express these receptors (36, 37). Indeed, we found elevated levels of both PD-1 (Fig. 4P) and TIGIT (Fig. 4Q) on metastasis-associated CD4+ T cells and Tregs (Supplementary Fig. S5G–S5J). Importantly, expression levels were higher in CD8+ T cells than in other T cells (0 and 0), in agreement with the transcriptomics data (Fig. 4R). Thus, bone metastases are highly infiltrated by immunosuppressive granulocytes and dysfunctional T cells, presumably interacting via the TIGIT– CD155 and PD-1–PD-L1 immune-checkpoint signaling axes.
IL1β Is a Central Player in the Immunosuppressive Phenotype of Granulocytes
Intrigued by these findings, we hypothesized that functional inhibition of the cross-talk between immunosuppressive cells (granulocytes and Tregs) and CTLs cells during breast cancer progression will be effective in preventing/attenuating bone metastatic relapse.
To identify potential drug targets for these cell subsets and signaling pathways, we used “SAveRUNNER” (38), a network medicine approach that takes advantage of reported drug–target interactions (Fig. 5A), to analyze our transcriptomics data of differentially expressed gene profiles in T cells and granulocytes. This analysis provided a list of potential FDA- approved drugs for each cell subset (we selected only drugs with a similarity index>0.7; Fig. 5B). The predicted therapeutics included numerous immune-checkpoint neutralizing antibodies as potential candidates to target cytotoxic T cells (Fig. 5C) and general T-cell inhibitors or IL1 inhibitors to target Tregs (Fig. 5D). Because we hypothesized that targeting the cross-talk of T cells with MDSCs would be beneficial, we further used SAveRUNNER to identify a drug that would target immunosuppressive granulocytes, to design a combination strategy that would potentiate ICB. The identified drug list for granulocytes proposed various inhibitors for IL1β or TNFα with the highest similarity index (Fig. 5E). Notably, both IL1β and TNFα were highly enriched in granulocytes isolated from bone metastatic lesions compared with all other groups (0 and 0).
IL1β is a potential target to inhibit immunosuppressive granulocytes. A, Illustration of SAveRUNNER analysis. B, Numbers of identified potential drugs by crossing transcriptome data with drug targets. C–E, Selection of drugs targeting dysfunctional CD8+ T cells (C), Tregs (D), and granulocytes (E). Gene expression of Il1b (F) and tnfa (G) in granulocytes during breast cancer bone metastatic progression. n = 4 for normal, early and periph and n = 3 in met-core; Statistics were generated using the adjusted P values from the DEseq2 analysis. H, Illustration of bone microenvironment supernatant preparation for ELISA. I, IL1β and (J) TNFα protein levels in peripheral BM and metastatic core. Dots represent separate biological repeats (n = 8 mice). Line links between peripheral BM and specific met-core sample. P values were calculated using a two-tailed Mann–Whitney test. K, Correlation between expression of il1b and the granulocytic markers s100a8 and s100a9 in bones (r = 0.8611, P < 0.0001 and r = 0.7294, P < 0.0002 for s100a8 and s100a9, respectively). L, Correlation between tnfa and s100a8 and s100a9 expression in bones (r = −0.3177, ns; r = −0.3974, ns for s100a8 and s100a9, respectively). K and L, n = 21 mice. M, IL1β protein level in plasma of normal or 4T-Bone–injected mice (n = 6 and 5 mice for normal and 4T-Bone, respectively). For K and L, two-tailed Pearson correlation was calculated. N, Average Z-score of gene expression of immunosuppressive gene signature in BM cells following incubation with 20 ng/mL recombinant IL1β. n = 4 biological repeats. O, Quantification of CD155, PD-L1, and SiglecF MFI from in vitro generated MDSC ± IL1β. P values were calculated using a two-tailed unpaired t test. P, Experimental design of BMT experiment presented in Q–S, Lethally eradiated WT recipient mice received WT or IL1β−/− whole BM, followed by orthotopic injection of 4T-Bone and resection of primary tumor. At late metastatic stages, BM and spleens were analyzed (n = 6 mice per group). Q, Spleen weight. Error bars represent SEM, P value was calculated by two-tailed Mann–Whitney test. R, Absolute numbers of PD-L1+ and CD155+ granulocytes in BM and spleen from mice transplanted with IL1β−/− BM or WT BM. Error bars represent SEM, P value was calculated by multiple t tests. S, Representative FMO histograms of PD-L1 and CD155 expression on granulocytes in BM from mice transplanted with WT BM t (light blue) or IL1β−/− BM (red). T, Arginase activity in granulocytes isolated from the BM of 4T-Bone–injected mice treated with 5 doses of αIL1β neutralizing antibody or isotype control following primary tumor resection. P values were calculated with a two-tailed unpaired t test. U, CTL killing assay: activated CD8+ T cells were cocultured for 24 hours w/wo granulocytes isolated from the BM of αIL1β/IC-treated mice. Representative histogram and quanti- fication of the % of CD107a+ CTLs. P values were calculated by performing a one-way ANOVA multiple comparisons test. Illustrations were created with BioRender.com.
IL1β is a potential target to inhibit immunosuppressive granulocytes. A, Illustration of SAveRUNNER analysis. B, Numbers of identified potential drugs by crossing transcriptome data with drug targets. C–E, Selection of drugs targeting dysfunctional CD8+ T cells (C), Tregs (D), and granulocytes (E). Gene expression of Il1b (F) and tnfa (G) in granulocytes during breast cancer bone metastatic progression. n = 4 for normal, early and periph and n = 3 in met-core; Statistics were generated using the adjusted P values from the DEseq2 analysis. H, Illustration of bone microenvironment supernatant preparation for ELISA. I, IL1β and (J) TNFα protein levels in peripheral BM and metastatic core. Dots represent separate biological repeats (n = 8 mice). Line links between peripheral BM and specific met-core sample. P values were calculated using a two-tailed Mann–Whitney test. K, Correlation between expression of il1b and the granulocytic markers s100a8 and s100a9 in bones (r = 0.8611, P < 0.0001 and r = 0.7294, P < 0.0002 for s100a8 and s100a9, respectively). L, Correlation between tnfa and s100a8 and s100a9 expression in bones (r = −0.3177, ns; r = −0.3974, ns for s100a8 and s100a9, respectively). K and L, n = 21 mice. M, IL1β protein level in plasma of normal or 4T-Bone–injected mice (n = 6 and 5 mice for normal and 4T-Bone, respectively). For K and L, two-tailed Pearson correlation was calculated. N, Average Z-score of gene expression of immunosuppressive gene signature in BM cells following incubation with 20 ng/mL recombinant IL1β. n = 4 biological repeats. O, Quantification of CD155, PD-L1, and SiglecF MFI from in vitro generated MDSC ± IL1β. P values were calculated using a two-tailed unpaired t test. P, Experimental design of BMT experiment presented in Q–S, Lethally eradiated WT recipient mice received WT or IL1β−/− whole BM, followed by orthotopic injection of 4T-Bone and resection of primary tumor. At late metastatic stages, BM and spleens were analyzed (n = 6 mice per group). Q, Spleen weight. Error bars represent SEM, P value was calculated by two-tailed Mann–Whitney test. R, Absolute numbers of PD-L1+ and CD155+ granulocytes in BM and spleen from mice transplanted with IL1β−/− BM or WT BM. Error bars represent SEM, P value was calculated by multiple t tests. S, Representative FMO histograms of PD-L1 and CD155 expression on granulocytes in BM from mice transplanted with WT BM t (light blue) or IL1β−/− BM (red). T, Arginase activity in granulocytes isolated from the BM of 4T-Bone–injected mice treated with 5 doses of αIL1β neutralizing antibody or isotype control following primary tumor resection. P values were calculated with a two-tailed unpaired t test. U, CTL killing assay: activated CD8+ T cells were cocultured for 24 hours w/wo granulocytes isolated from the BM of αIL1β/IC-treated mice. Representative histogram and quanti- fication of the % of CD107a+ CTLs. P values were calculated by performing a one-way ANOVA multiple comparisons test. Illustrations were created with BioRender.com.
To select between these two proinflammatory cytokines, we further analyzed their protein levels in the bone metastatic microenvironment compared with peripheral BM (Fig. 5H). We found that TNFα was elevated in the bone metastatic core compared with peripheral BM, and IL1β levels were also elevated in the bone metastatic core, but did not reach statistical significance (Fig. 5I and J). However, analysis of the correlation between the MDSC markers S100A8 and S100A9 with either cytokine revealed that whereas high levels of IL1β significantly correlated with both S100A8 and S100A9 (Fig. 5K), TNFα did not correlate with MDSC markers (Fig. 5L). Moreover, Il1b mRNA levels correlated with both Pdcd1 and Tigit in bone metastases (Supplementary Fig. S6A), in agreement with previous studies linking checkpoint molecule expression with elevated Il1b levels (39, 40). Importantly, GSEA confirmed increased expression of “IL-1 family signaling” genes in metastasis-associated granulocytes (Supplementary Fig. S6B). Moreover, 4T-Bone injected mice exhibited markedly higher systemic levels of IL1β compared with normal mice (Fig. 5M).
IL1β was shown to be a key player in bone metastasis (41, 42). We therefore next asked whether IL1β can drive an immunosuppressive phenotype in granulocytes. To test this, we cultured BM cells with recombinant IL1β and analyzed the expression of multiple immunosuppressive genes. Indeed, IL1β induced an upregulation in the expression of immunosuppressive genes in BM cells (Fig. 5N), suggesting reprogramming of granulocytes to MDSCs via IL1β. Moreover, incubation of in vitro generated MDSCs with IL1β elevated their expression of the immunosuppressive markers PD-L1, CD155, and SiglecF (Fig. 5O), and enhanced the ability of MDSCs to inhibit T-cell proliferation (Supplementary Fig. S6C). To further assess the importance of IL1β in affecting immune suppression in vivo, we performed adoptive BM transplantation (BMT) from transgenic IL1β−/− or from WT mice to recipient mice following lethal BM irradiation, followed by injection of 4T-Bone cells and primary tumor resection (Fig. 5P). Interestingly, initial analysis of mice transplanted with IL1β−/− BM revealed a striking decrease in splenomegaly, typical of systemic accumulation of granulocytes, observed in 4T-Bone injected mice (Fig. 5Q; ref. 8). Moreover, IL1β−/− BMT mice had reduced numbers of PD-L1+ and CD155+ granulocytes in both spleen and BM (Fig. 5R), and their metastasis- associated granulocytes downregulated their expression levels of PD-L1 and CD155 (Suppelemtary Fig. 5S). Importantly, in vivo neutralization of IL1β with antibodies following primary tumor resection further confirmed the functional role of IL1β in modulating granulocyte immunosuppressive functions: granulocytes isolated from the BM of αIL1β-treated mice had significantly reduced arginase activity (Fig. 5T), and lost their ability to inhibit CTL killing, evident by restored levels of membranal bound CD107a (Fig. 5U) and secreted IFNγ (Supplementary Fig. S6D). Of note, neutralization of IL1β also attenuated the accumulation of granulocytes, typically observed during bone metastatic progression (Supplementary Fig. S6E). Thus, IL1β plays a central role in functionally driving the immunosuppressive function of MDSCs.
Based on these findings, we postulated that combining inhibition of immune-checkpoint receptors (either PD-1 or TIGIT) with inhibition of the proinflammatory/MDSC cytokine IL1β may be an efficient treatment strategy to prevent breast cancer bone metastatic relapse.
Cotargeting Dysfunctional T Cells and MDSCs Reduces Bone Metastasis Progression by Sustaining Antitumor Immunity
To test the therapeutic benefit of targeting the immuno-suppressive TME, we treated mice with neutralizing anti-bodies against either the pro-MDSC cytokine IL1β or the T-cell checkpoint receptors PD-1/TIGIT or with a combina- tion of both treatments (αIL1β + αPD-1 or αIL1β + αTIGIT). Mice were treated in an adjuvant setting, following surgical resection of the primary breast tumor and were longitudinally monitored for bone metastatic progression and survival (Fig. 6A). When first signs of advanced metastatic disease were observed (paralysis of hind legs, indicating spinal cord injury), we assessed bone metastatic load intravitally by CT imaging, confirming metastatic spinal cord compression as the cause of spinal injury. Bone lesions were categorized according to the severity of bone destruction (+ for minor osteolytic lesion, ++ for multiple osteolytic lesions and +++ for extended bone destruction, consuming most of the vertebra; Fig. 6B). Quantification of the results revealed that whereas all treatment groups had lower metastatic load compared with mice treated with isotype control antibodies (IC), treatment with αPD-1 or with the combination of αIL1β+αTIGIT was most efficient in significantly reducing severe bone metastatic lesions (Fig. 6C).
Combining IL1β neutralization with ICB attenuates breast cancer bone metastasis. A, Experimental scheme: following orthotopic injection of 4T-Bone cells and primary tumor resection, mice were treated with neutralizing antibodies or appropriate isotype control every 72 hours. Bone metastasis was monitored by CT imaging and clinical symptoms. B, Representative CT scans of spinal columns showing bone destruction at different degrees of severity (+ for one small osteolytic lesion, ++ for multiple osteolytic lesions, and +++ for extensive bone destruction). C, Quantification of overall incidence of bone metastasis analyzed by CT imaging. P values were calculated using two-tailed X2 test. Treated 4T-Bone–injected mice were assessed for disease-free survival (DFS; D and E) referring to the absence of spinal cord compression (SCC) and for overall survival (OS; F and G). n = 11, 10, 12, 11, 12 and 12 mice in control, αIL1β, αPD-1, αTIGIT, αIL1β + αPD-1 and αIL1β + αTIGIT, respectively. H, Incidence of immune-related events (cytokine storm), surviving mice and SCC in all mice. I, Forest plot of log-rank hazard ratio (HR) and 95% confidence interval (CI) in all treatments. J, Cox proportional hazard regression analysis for both SCC occurrence and all-cause mortality for each treatment. K, ELISA of IFNγ protein levels in BM and bone metastatic lysate. IFNγ levels are normalized to total protein levels. n = 37 and 41 for survivors and dead mice, respectively. Mann–Whitney test was performed. Dots represent separate biological repeats. P values were calculated using two-tailed Mann–Whitney test. L, Representative dot plots of bone metastasis–infiltrated CD4+ T helper and CD8+ T cytotoxic cells, gated from CD3+. M, Quantification of L; n = 10, 9, 4, 7, 9, and 4 mice in control, αIL1β, αPD-1, αTIGIT, αIL1β + αPD-1, and αIL1β + αTIGIT, respectively. Kolmogorov–Smirnov test was performed for each treatment. N, Heat map of T-cell activation gene signature in bone metastases. Illustrations were created with BioRender.com.
Combining IL1β neutralization with ICB attenuates breast cancer bone metastasis. A, Experimental scheme: following orthotopic injection of 4T-Bone cells and primary tumor resection, mice were treated with neutralizing antibodies or appropriate isotype control every 72 hours. Bone metastasis was monitored by CT imaging and clinical symptoms. B, Representative CT scans of spinal columns showing bone destruction at different degrees of severity (+ for one small osteolytic lesion, ++ for multiple osteolytic lesions, and +++ for extensive bone destruction). C, Quantification of overall incidence of bone metastasis analyzed by CT imaging. P values were calculated using two-tailed X2 test. Treated 4T-Bone–injected mice were assessed for disease-free survival (DFS; D and E) referring to the absence of spinal cord compression (SCC) and for overall survival (OS; F and G). n = 11, 10, 12, 11, 12 and 12 mice in control, αIL1β, αPD-1, αTIGIT, αIL1β + αPD-1 and αIL1β + αTIGIT, respectively. H, Incidence of immune-related events (cytokine storm), surviving mice and SCC in all mice. I, Forest plot of log-rank hazard ratio (HR) and 95% confidence interval (CI) in all treatments. J, Cox proportional hazard regression analysis for both SCC occurrence and all-cause mortality for each treatment. K, ELISA of IFNγ protein levels in BM and bone metastatic lysate. IFNγ levels are normalized to total protein levels. n = 37 and 41 for survivors and dead mice, respectively. Mann–Whitney test was performed. Dots represent separate biological repeats. P values were calculated using two-tailed Mann–Whitney test. L, Representative dot plots of bone metastasis–infiltrated CD4+ T helper and CD8+ T cytotoxic cells, gated from CD3+. M, Quantification of L; n = 10, 9, 4, 7, 9, and 4 mice in control, αIL1β, αPD-1, αTIGIT, αIL1β + αPD-1, and αIL1β + αTIGIT, respectively. Kolmogorov–Smirnov test was performed for each treatment. N, Heat map of T-cell activation gene signature in bone metastases. Illustrations were created with BioRender.com.
Combining immune-checkpoint inhibitors (ICI) with αIL1β neutralizing antibody led to significant improvement in clinical signs as well as in overall survival (Fig. 6D–G). One of the most debilitating complications of aggressive bone metastasis is spinal cord compression (SCC): bone metastatic lesions press on the spinal cord, causing irreversible spinal injury. Strikingly, all treatment regimens resulted in delayed and decreased SCC compared with control mice: in the isotype control group up to 82% of mice developed SCC, whereas only 25% of αIL1β+αPD-1–treated mice and 33% of αIL1β + αTIGIT- treated mice eventually developed SCC (0 and 0; Supplementary Fig. S7A). Monotherapies also decreased the occurrence of SCC down to 50% (αIL1β−treated mice), 14% (αPD-1-treated mice), and 45% (αTIGIT-treated mice; Fig. 6D, E, and 0; Supplementary Fig. S7A).
Importantly, although PD-1 inhibition appeared to be the most promising approach when analyzing SCC, treatment with αPD-1 caused dramatic immune-related events (IRE): up to 33% of the mice treated with αPD-1 presented with a lethal cytokine storm (Fig. 6H). Thus, although αPD-1 treat- ment seemingly reduced SCC occurrence, the overall survival of mice was in fact not significantly improved, as the IRE dramatically affected mice survival (0 and 0). Of note, adverse effects of repeated administration of PD-1 antibody in the 4T1 model have been previously reported (43). Strikingly, no such immune related events (IREs) were observed in the groups treated with either αIL1β or αTIGIT (0 and 0). Moreover, survival analysis revealed that although most of the tested treatment regimens resulted in increased overall survival (0 and 0), combining αIL1β with αTIGIT was the most effective treatment strategy, with the lowest log-rank hazard ratio (HR) compared with all treatment groups: HR 0.36 (with 95% confidence interval of 0.13–1). Indeed, treatment with αIL1β+αTIGIT resulted in a 64% decrease in death (Fig. 6I).
We further analyzed the independent effect of each antibody, by performing Cox proportional hazards regression analysis for both SCC and all-cause mortality. Analysis revealed that both αTIGIT and αPD-1 significantly decreased the probability to develop SCC (reduction of 62%, P = 0.027 and 85%, P = 0.004; respectively), while neutralizing IL1β alone did not reach statistical significance (P = 0.074) in decreasing the probability to develop SCC independently of other drugs (52%; Fig. 6J). Thus, ICIs showed higher protective potential from bone destruction symptoms compared with IL1β neutralization alone. However, although none of the antibodies reached statistical significance in decreasing all-cause mortality, αTIGIT decreased the probability of death by 45% (P = 0.07) (Fig. 6J).
In summary, although Cox hazard regression analysis revealed that each antibody independently reduced the risk of developing SCC, log-rank analysis revealed that combination therapy of αIL1β and αTIGIT was most efficient in improving survival.
Interestingly, IFNγ was elevated in mice that survived until the experimental endpoint (Fig. 6K), which could indicate increased antitumor immunity. To get further mechanistic insights about the beneficial effect of combining ICB and IL1β neutralization, we analyzed the immune microenvironment of bone metastatic lesions in treated mice. Analysis of T cells revealed that the combination of αIL1β + αTIGIT treatment resulted in a significantly increased proportion of CD8+ CTLs versus CD4+ helper cells (0 and 0), whereas other treatments did not significantly affect the relative levels of T cells. Moreover, analysis of a gene signature associated with T-cell activation in the bones of mice from each treatment indicated that combination of IL1β and TIGIT neutralization was most efficient in upregulating the expression of this gene signature (Fig. 6N; Supplementary Fig. S7B).
To dissect whether targeting TIGIT, PD-1, and IL1β directly inhibit granulocyte–T-cell cross-talk, we isolated cancer-associated granulocytes from 4T-Bone injected mice and cocultured them with T lymphocytes in the presence of neutralizing antibodies (Supplementary Fig. S7C). Analysis of T-cell proliferation revealed that although granulocytes inhibited CD8+ T-cell proliferation (Supplementary Fig. S7D), αTIGIT antibodies significantly attenuated this inhibitory effect, resulting in a decrease in nondividing CD8+ T cells compared with isotype control (Supplementary Fig. S7E). Furthermore, inhibiting TIGIT in granulocyte–T-cell coculture (but not in T-cell monoculture) increased their secretion of IFNγ and TNFα (Supplementary Fig. S7F and S7G), further indicating that blocking the TIGIT–CD155 signaling axis between granulocytes and cytotoxic T cells alleviates T-cell dysfunction. Interestingly, αPD-1 and αIL1β did not show the same beneficial effects on T-cell proliferation or activation (Supplementary Fig. S7E–S7G).
Taken together, these results suggest that combining IL1β and TIGIT neutralization is an effective treatment strategy to attenuate breast cancer bone metastasis by sustaining and promoting antitumor immunity.
Inhibition of TIGIT, PD-1, and IL1β May Be Efficient in Human Bone Metastasis
Human bone metastases are infiltrated by IL1β−expressing granulocytes and TIGIT-expressing T cells. A, Analyses of RNA-seq from human bone metastasis originating from different primary cancers (colorectal, prostate, lung, kidney, and unknown). B, Correlation between IL1B and granulocytic markers score (combining S100A8, S100A9, and CD33 expression). C, Correlation between IL1B and immune-checkpoint receptor score (combining PDCD1, TIGIT, TIM3, CTLA4, and LAG3). D, Correlation between IL1B and TIGIT. E, Integration and analysis of scRNA-seq data of human bone metastases originating from breast or lung cancer. F, All cells from scRNA-seq of 2 breast cancer bone metastases and 1 lung cancer bone metastases were projected onto a tSNE plot distinguished by different colors. G, Autoclusterization of all cells forming 9 clusters. H, Autoclusterization of immune cells forming 4 clusters: 2 myeloid and 2 lymphoid. I, Genes of interest in the myeloid/granulocytic cluster (cluster 0: orange); IL1B, S100A8, S100A9, and CD14 are expressed in the same cluster. Genes of interest in the T cytotoxic cluster (cluster 3: purple); ICR are expressed in CD8+ T cells. J, Feature plots of expression distribution for IL1B and TIGIT. Expression levels for each cell are color-coded and overlaid onto the UMAP plot. IL1B is almost exclusively expressed in the granulocytic-like cluster and TIGIT in the T-cell cluster. K, Representative images and (L) quantification of pathologic score of TIGIT or PD-1 in peripheral BM or metastatic core from human breast cancer bone metastases. n = 34 and 35 for TIGIT and PD-1, respectively. P value was calculated using X2 test. Pathologic scoring was given according to either intensity of staining or abundance of positive cells. M, Representative images of TIGIT staining in human bone metastasis from different primary origins. Renal cell carcinoma n = 10, lung cancer n = 6, prostate cancer n = 2, squamous cell carcinoma n = 1, and kidney cancer n = 3. Illustrations were created with BioRender.com.
Human bone metastases are infiltrated by IL1β−expressing granulocytes and TIGIT-expressing T cells. A, Analyses of RNA-seq from human bone metastasis originating from different primary cancers (colorectal, prostate, lung, kidney, and unknown). B, Correlation between IL1B and granulocytic markers score (combining S100A8, S100A9, and CD33 expression). C, Correlation between IL1B and immune-checkpoint receptor score (combining PDCD1, TIGIT, TIM3, CTLA4, and LAG3). D, Correlation between IL1B and TIGIT. E, Integration and analysis of scRNA-seq data of human bone metastases originating from breast or lung cancer. F, All cells from scRNA-seq of 2 breast cancer bone metastases and 1 lung cancer bone metastases were projected onto a tSNE plot distinguished by different colors. G, Autoclusterization of all cells forming 9 clusters. H, Autoclusterization of immune cells forming 4 clusters: 2 myeloid and 2 lymphoid. I, Genes of interest in the myeloid/granulocytic cluster (cluster 0: orange); IL1B, S100A8, S100A9, and CD14 are expressed in the same cluster. Genes of interest in the T cytotoxic cluster (cluster 3: purple); ICR are expressed in CD8+ T cells. J, Feature plots of expression distribution for IL1B and TIGIT. Expression levels for each cell are color-coded and overlaid onto the UMAP plot. IL1B is almost exclusively expressed in the granulocytic-like cluster and TIGIT in the T-cell cluster. K, Representative images and (L) quantification of pathologic score of TIGIT or PD-1 in peripheral BM or metastatic core from human breast cancer bone metastases. n = 34 and 35 for TIGIT and PD-1, respectively. P value was calculated using X2 test. Pathologic scoring was given according to either intensity of staining or abundance of positive cells. M, Representative images of TIGIT staining in human bone metastasis from different primary origins. Renal cell carcinoma n = 10, lung cancer n = 6, prostate cancer n = 2, squamous cell carcinoma n = 1, and kidney cancer n = 3. Illustrations were created with BioRender.com.
These analyses confirm that in human bone metastatic samples, from both lung and breast origin, IL1B is highly expressed by granulocytes whereas TIGIT is expressed by cytotoxic T cells (Fig. 7I and 0), implying that the IL1β and TIGIT signaling axes are relevant players in human bone metastasis.
To analyze these candidate targets at the protein level, we analyzed a cohort of 35 women with breast cancer bone metas- tasis (see Supplementary Table S1 for cohort details) by IHC of PD-1 and TIGIT (Fig. 7K). Strikingly, TIGIT expression was detected in the majority of the bone samples analyzed (more than 80%), whereas PD-1 was detected in only 20% of patients (Supplementary Fig. S8E). Pathologic spatial analysis indicated that TIGIT was enriched within the metastatic core of bone metastatic lesions, compared with peripheral bone tissue in the same patient samples (Fig. 7l), further implying that TIGIT is of central importance in breast cancer bone metastasis. Excited by these results, we analyzed TIGIT in a cohort of 27 bone metastasis samples from 9 different cancer types and found that TIGIT-expressing cells were detected in bone metastases from kidney, lung, prostate, renal cell carcinoma, and squamous cell carcinoma, suggesting that the TIGIT signaling axis may be an attractive therapeutic target in bone metastasis not only from breast cancer (Fig. 7M; Supplementary Fig. S8F).
Together, these findings confirm that in human bone metastasis, IL1β is highly expressed by granulocytes whereas TIGIT is expressed by cytotoxic T cells (Fig. 7I). Our findings position IL1β and TIGIT as central players in human bone metastasis, encouraging further clinical investigations.
Discussion
Our study maps the dynamic changes in the bone microenvironment during breast cancer bone metastasis and uncovers a key cross-talk between myeloid and lymphoid cells that facilitates the formation of an immunosuppressive and growth-permissive metastatic niche. Transcriptome profiling of metastasis-associated granulocytes and T cells isolated from distinct stages and locations of breast cancer bone metastases revealed differences between the metastatic core and the peripheral BM of metastasis-bearing mice. Specifically, we found that bone metastatic lesions are highly infiltrated by dysfunctional cytotoxic T cells, regulatory T cells, and immunosuppressive granulocytes and identified the PD-1–PD-L1 and TIGIT–CD155 axes as central in the interactions between MDSCs and CTLs. Furthermore, we showed that IL1β is a key factor in driving the immunosuppressive phenotype of metastasis-associated granulocytes. These findings were functionally important, as combination therapy using ICB with neutralization of IL1β resulted in attenuated bone metastasis and improved survival. Analysis of human bone metastasis from different primary origins revealed that TIGIT and IL1β are also prominent in human bone metastasis. Our findings imply that combining ICB with IL1β inhibition may be an attractive novel therapeutic approach to treat patients with bone metastasis.
It has become clear in recent years that the TME composition and the cross-talk between immune cells in the metastatic microenvironment are of central importance for treatment efficacy (44). We have previously demonstrated that immuno- suppressive granulocytes that function as MDSCs accumulate in the metastatic niche at early stages of breast cancer bone metastasis, presumably leading to a reduction in T-cell presence in the BM of bone metastasis-bearing mice (8). In this study, we profiled the transcriptome of granulocytes and T cells isolated from distinct time points during metastatic progression and compared the peripheral BM with bone metastatic lesions. We found that in addition to MDSCs, metastatic lesions are highly infiltrated by T cells, which are dysfunctional and therefore ineffective in restricting metastatic growth.
MDSCs were shown to interfere with T-cell activity in multiple ways (45). Our data analysis of gene expression and receptor–ligand interactions revealed that expression of the IC ligands PD-L1 and CD155 on granulocytes, and high levels of their respective receptors PD-1 and TIGIT on bone metastasis–infiltrated T cells may underlie the disruption of T-cell– mediated antitumor immunity. Interestingly, PD-1 and TIGIT were also expressed by T helper cells and Tregs, suggesting that their interactions with MDSCs may also contribute to the formation of an immunosuppressive niche.
The network medicine algorithm analysis we performed predicted that targeting immune-checkpoint signaling with ICB antibodies will be beneficial in treating breast cancer bone metastasis. Despite accumulating clinical data supporting the emergence of immunotherapy as a potential care for metastatic TNBC patients, its efficiency and safety in breast cancer bone metastasis remain largely unknown (46, 47). In general, the efficacy of ICB treatment for bone metastasis is multifaceted: ICB provided good results for renal cell carcinoma patients with bone metastasis, but was ineffective in prostate patients with bone metastases (48), and showed inconsistent efficacy in advanced non–small cell carcinoma patients with bone metastasis.
Based on the composition and functional interactions of the bone metastatic TME that we had identified, we designed combinatory approaches to target both immune checkpoints and MDSC signaling. Specifically, the algorithm we used predicted that inhibition of IL1β may be a therapeutically beneficial approach to inhibit metastasis-associated granulocytes. IL1β was previously shown to be upregulated in breast cancer–associated MDSCs (24) and in breast cancer bone metastasis (42). Moreover, targeting IL1β was shown to reverse the immunosuppressive microenvironment in several cancer types (42, 49, 50) and to reduce breast cancer bone metastasis in humanized mouse models (51). We found that IL1β is associated with an immunosuppressed microenvironment and with the expression of immune-checkpoint receptors in mouse and in human bone metastasis. Moreover, we showed that IL1β was sufficient to induce the expression of checkpoint ligands in granulocytes. Importantly, targeting IL1β in vivo by neutralizing antibodies, or by adoptive BM transplantation from IL1β−/− mice restored T-cell killing activity. Thus, we showed that IL1β is a central mediator of granulocyte immunosuppressive function.
We therefore proceeded to target IL1β in combination with ICB. Notably, the combination of αIL1β with αPD-1 was shown to be efficient in several cancer types (50, 52).
We performed preclinical trials to evaluate the efficacy of two different ICBs: αPD-1 and αTIGIT, as monotherapies or in combination with IL1β inhibition in treating breast cancer bone metastasis in mice. Although PD-1 is increasingly used in the clinic, including in breast cancer (53–55), the efficacy of targeting TIGIT, as a single agent or in combination with other anticancer therapies, in patients with locally advanced or metastatic tumors is still under investigation (NCT02794571; ref. 56). Importantly, our analysis of independent scRNA-seq data sets from human bone metastatic patients demonstrated that IL1β and TIGIT are relevant players and highlighted the potential benefit of targeting these candidates. Moreover, we demonstrated that TIGIT is highly expressed in human bone metastasis from six different cancer types.
Our findings in vivo indicated that although αPD-1 treatment considerably reduced bone metastasis, it also caused severe, and sometimes lethal, adverse immunologic effects. Conversely, targeting TIGIT did not cause any severe adverse effect in mice, as monotherapy or in combination therapy. We show that cotargeting of IL1β and TIGIT was the most efficient approach in increasing both disease-free survival (DFS) and overall survival (OS), whereas the combination of αPD-1 and IL1β enhanced DFS but not significantly affected OS. Our findings in mouse models, combined with the clinical data from patients, position TIGIT as a key signaling checkpoint receptor in bone metastatic progression and suggest that combining ICB with IL1β inhibition may be an attractive novel therapeutic approach to treat patients with bone metastasis.
There is currently no efficient treatment for patients with breast cancer with bone metastasis. Our study highlights the complexity of the cross-talk between immune cells in the bone metastatic microenvironment and their effects on clinical outcome and suggests that combinatorial therapy to cotarget immunosuppressive granulocytes and dysfunctional T cells may be a promising therapeutic strategy to inhibit breast cancer bone metastasis that should be considered for further clinical trials.
Methods
Mouse Strains
Animals were maintained in specific pathogen-free conditions with controlled temperature/humidity (22°C/55%) environment on a 12-hour light–dark cycle and with food and water ad libitum. All animals were maintained within the Tel Aviv University Specific Pathogen- Free facility. All Animal procedures included in the study were granted ethical approval by the Tel Aviv University Institutional Animal Care and Use Committee. BALB/c mice were purchased from Envigo, Israel. Mice were used for experiments at 6 to 9 weeks of age. BALB/c Il1b−/− transgenic mice were a kind gift from Prof. Elena Voronov.
Cell Lines
The 4T-Bone is a bone-tropic variant of 4T1 cells (8). Cells were grown in RPMI-1640 supplemented with 10% FCS, 1% penicillin–strep- tomycin, 1% sodium pyruvate, 1% HEPES 1 mol/L, and 0.5% glucose. All cell lines were routinely tested for Mycoplasma using the EZ-PCR- Mycoplasma test kit (Biological Industries; 20-700-20). The cell lines were not authenticated.
Orthotopic Tumors
A total of 0.5 × 106 4T-Bone cells were resuspended in PBS and mixed 1:1 with Matrigel (356,231, BD Biosciences) to a final volume of 100 μL. Cells were inoculated into the fourth mammary fat pad of female Balb/c mice. Tumors were excised approximately 16 days later, under anesthesia.
Analysis of Bone Samples
Isolation of BM Samples
BM samples were isolated from the femur and tibia from either normal mice or 4T-Bone–injected mice at different metastatic stages. At late stage, BM was flushed out from the noninvolved femur and tibia, referred to as “peripheral BM.” “Early” metastatic stage samples were isolated as previously described (8). Briefly, BM cells were isolated from the tibia and femur at the time of tumor removal, when there was no evidence of metastatic colonization.
Detection of Bone Macrometastases
Mice were routinely checked for morbidity (e.g., paraplegia or difficulty of movement). Macrometastatic bone lesions were detected by intravital imaging (μCT). Mice were euthanized at day 40, and bones from legs, sternum, ribs, and spinal column were analyzed.
Single-Cell Suspension and Sample Preparation
Bone metastases were isolated and dissociated. Briefly, bone tissues were harvested, washed in PBS, minced thoroughly with scissors, and incubated for 40 minutes with RPMI supplemented with 0.1% collagenase IV (Worthington, LS004177) and 0.1% diaspase II (Roche, 04942078001) on stir plate in 37°C water bath. BM cells were flushed out using RPMI supplemented with 10% FCS and 2 mmol/L EDTA. Spleens were harvested from 8- to 12-week-old mice, minced, and dissociated. Single-cell suspensions were filtered by 70-μm cell strainers (Corning) and red blood cells were lysed. Remaining single-cell suspensions were taken for further analysis.
Bone Histology and Tissue Preparation
Bones were harvested from mice and fixated in 4% paraformaldehyde for 6 hours followed by 48 hours sucrose or fresh-frozen tissues embedded in optimal cutting temperature compound (OCT; Tissue-Tek) on dry ice. Serial sections were obtained to ensure equal sampling of the examined specimens (10 μm trimming).
H&E Staining
Bone tissue sections were stained with H&E using Multistainer Leica ST5020.
Immunohistochemistry
Tissue sections were stained using Leica Bond III, with the following anti-mouse antibodies: CD3 (Invitrogen; 14-0032-82; clone 17A2), Ly6G (BioLegend; 127601; clone 1A8). Donkey anti-rat peroxidase-conjugated secondary antibody (Jackson ImmunoResearch Laboratories; 712-035-153). Slides were visualized and analyzed using confocal microscopy.
FACS Analysis and Sorting
FACS Analysis
BM or bone metastasis cells were isolated from mice. Single-cell suspensions were prepared according to the organ. Cells were counted and resuspended in FACS buffer (PBS with 2% FCS and 2 mmol/L EDTA). Cells were then incubated for 30 minutes with appropriate antibodies on top of Fc-block (anti-mouse CD16/CD32). anti-CD45-BV650 (BioLegend, BLG-103151, dilution 1:100); anti-CD11b-PeCy7 (BioLegend, BLG-101215, dilution 1:100); anti-CD11c-PerCP-Cy5.5 (eBioscience, 45-0114, dilution 1:100); anti-SiglecF-APC-R700 (BD Biosciences, BD565183, dilution 1:100); anti-Ly6G-APC (BioLegend, 127614, dilution 1:200); anti-Ly6C-FITC (BioLegend, 128006, dilution 1:200); anti-CD4-APC-Cy7 (BioLegend, BLG-100413, dilution 1:100); anti-CD4-PE (BioLegend, BLG-100512, dilution 1:100); anti-CD8a-APC (BioLegend, BLG-100712, dilution 1:100); anti-CD8a-PE (eBioscience, 12-0083, dilution 1:100); anti-CD3-FITC (BioLegend, BLG-100306, dilution 1.5:100); anti–PD-1-BV785 (BioLegend, BLG-135225, dilution 1:100); anti–TIGIT-APC (BioLegend, BLG-142106, dilution 4:100); anti–PD-L1-PE (BioLegend, BLG-124308, dilution 1:100); anti-CD155-BV785 (BioLegend, BLG-131525, dilution 1:100); anti–PD-L1-PerCP Cy5,5 (BioLegend, BLG-124334, dilution 1:100); anti-CD107a (LAMP-1) PE (BioLegend, BLG-121611, dilution 1:100); anti-DR5 (CD262) PE (BioLegend, BLG-119905, dilution 1:100); anti-CD44 AF488 (BioLegend, BLG-103016, dilution 1:100) and DAPI (Molecular Probes; D3571). The specificity of staining was validated by the appropriate FMO method. Analysis was performed with CytoFLEX Flow Cytometer (Beckman Coulter, Inc.), and data analysis was done with FlowJo Software (version X.0.7).
FACS Sorting of Granulocytes and T Cells for RNA Sequencing
Sorting was performed using BD FACSAria III. Granulocytes were isolated as CD45+CD11b+Ly6CintLy6G+. T cytotoxic cells were isolated as CD45+CD3+CD8+ cells. T helper cells were isolated as CD45+CD3+ CD4+CD25− cells. T regulatory cells were isolated as CD45+CD3+CD4+ CD25+ cells. Gating as described in Supplementary Fig. S2D. Sorts were performed using FACSDiva software v8.
Arginase Activity
Arginase activity was assessed using QuantiChrom Arginase Assay Kit (DARG-100, BioAssay Systems) according to the manufacturer’s instructions. Briefly, 3 × 105 CD45+CD11b+Ly6CintLy6G+ granulo- cytes were isolated by FACS from bones/BM/spleens of mice. Cells were washed in PBS and centrifuged at 1,000 × g for 10 minutes at 4°C. Cell pellets were lysed in 100 μL of 10 mmol/L Tris–HCl (pH 7.4) containing 1 μmol/L pepstatin A, 1 μmol/L leupeptin, and 0.4% (w/v) Triton X-100, followed by 10-minute centrifuge at 14,000 × g. The supernatant was used for arginase assay.
Immune Cells’ Transcriptome Analysis
RNA-seq
Granulocytes, T helper, T cytotoxic, and T regulatory cells were isolated by cell sorting from normal BM, early-stage metastatic BM, and late metastatic stage of mice as described above. At the late metastatic stage, two distinct sites were analyzed: the peripheral BM and the bone metastatic core of the same mouse. MARS-seq protocol was used to generate libraries. MARS-seq libraries were sequenced using Illumina NovaSeq 6000. Sample barcodes were extracted from read 2 and concatenated to the fastq header of read 1. Sequencing data were then trimmed using fastp 0.20.0 (57) and aligned to the GRCm38 assembly using STAR 2.6.0c (58), DESeq2 1.30.1 (59), and R 3.6 were used for normalization of count data and for statistical analysis of differential gene expression. Reads were mapped to the Mus musculus genome (mm10) using STAR. Read counts per gene were calculated using HTSeq-count and a Refseq gtf file. Heat maps were generated based on gene expression Z-scored per gene derived from DESeq2 analysis results.
Metascape Pathway Enrichment Analysis
Lists of genes upregulated (defined as logFC>0.5) in bone metastases as compared with peripheral BM or normal BM were selected as inputs for pathway enrichment analysis, genes were included if the count per sample was higher than 100. The list of genes was then submitted to the online bioinformatics tool Metascape for identification of enriched pathways (60).
Receptor–Ligand Analysis
Receptor–ligand analysis was performed using the ICELLNET algorithm (35). Briefly, significantly differentially expressed gene lists from either met-core vs. normal or met-core vs. peripheral BM were used as inputs.
Network Medicine Analysis
To identify candidate drugs potentially targeting metastatic cell populations, we utilized the SAveRUNNER tool (38). SAveRUNNER predicts drug-disease associations by quantifying the interplay between the drug targets and the disease-associated proteins in the human interactome including 217,160 unique protein–protein interactions connecting 15,970 proteins. Gene sets for network analysis were generated by selecting differentially expressed genes between cell populations isolated from metastatic core versus populations isolated from peripheral BM; cutoff was set at logFC>0.5. Secondary elimination was based on read count >100 for each gene to be included. Drug–target interactions were acquired from DrugBank (http://www.drugbank.ca/) including 1,875 FDA-approved drugs. A drug was considered a potential candidate if DEGs and the corresponding drug were nearby in the interactome more than expected by chance (P < 0.05) and the weight of their interaction manifested in similarity index is >0.7.
Cytospin Staining
Granulocytes were seeded on coated slides for Cytospin. Cells were then fixated for 10 minutes with PFA 4%, washed stained with H&E using Multistainer Leica ST5020.
Bone Homogenate Supernatant
Bones were crushed using mortar and baster. RPMI-1640 media (2 mL) were added and incubated for 5 minutes at room temperature. The solution was then centrifuged for 7 minutes at 500 × g. Supernatants were collected and filtered through 0.45-μm filters.
Enzyme-Linked Immunosorbent Assay
IL1β, IFNγ, and TNFα enzyme-linked immunosorbent assay (ELISA) were performed for bone homogenate supernatant using the R&D ELISA Kit, according to the manufacturer’s instructions (R&D Systems; DY401-05; DY485; DY410). For granulocytes–T-cell coculture experiments, cells were pelleted by 10-minute centrifuge at 600 × g; supernatants were collected and used for ELISA for TNFα, GZMB, or IFNγ (R&D Systems; DY1865).
BM Differentiation and MDSC Generation under IL1β Stimulation
For BM differentiation, BM cells were flushed out from naïve mice. Cells were cultured in RPMI medium supplemented with 10% FCS, 50 μmol/L β-mercaptoethanol; 1%NEAC; 1% HEPES 1M with/without 20 ng/mL recombinant IL1β for 24 hours (PeproTech,B-10). For MDSC generation, cells were cultured as stated above and supplemented with recombinant cytokines IL6 (PeproTech, 216-16-10) and GM-CSF (PeproTech, 250-05-10) for 4 days with/without IL1β. Media were renewed every other day.
BM Transplantations
Seven-week-old female Balb/c WT mice were lethally irradiated using an X-ray machine (160HF; Philips) at a total dose of 9 Gy. Twenty-four hours after irradiation, mice were injected intravenously (i.v.) with 2 × 106 unfractionated BM cells collected aseptically from flushed femurs and tibias of age-matched Balb/c WT or IL1β−/− female mice. Following transplantation, mice received antibiotics for 4 weeks in drinking water (enrofloxacin; 0.2 mg/mL). To ensure radiation lethality, one mouse from each group was irradiated with- out transplantation. Three weeks after transplantation, mice were injected orthotopically into the mammary fat pad as described above.
In Vivo Functional Experiments
One day following tumor resection, mice were injected intraperitoneally with 150 μg/mouse of αIL1β (Armenian hamster IgG; B122; Bio X Cell; BE0246); 150 μg/mouse of αPD-1 (Armenian hamster IgG; J43; Bio X Cell; BE0033-2); 150 μg/mouse of Armenian hamster IC (Armenian hamster polyclonal; Bio X Cell; BE0091); 100 μg/mouse of αTIGIT (mouse IgG1; 1B4; ABSOLUTE; Ab01258-1.1-25), or 100 μg/mouse of mouse IC (mouse IgG1; B1-8; ABSOLUTE; Ab00104-1.1). Neutralizing antibodies were administered alone or in combination twice weekly for a total of five doses. For the SCC analysis, mice that developed IREs or primary tumor recurrence were excluded from the analysis.
T-Cell Isolation
Spleens were harvested from 10- to 12-week-old mice, minced, and dissociated. T cells were isolated either using MojoSort Mouse CD3 (BioLegend, 480031) or CD8 (BioLegend, 480035) T Cell Isolation Kit.
Cancer-Associated Granulocyte Isolation
Spleens of 4T-Bone–injected mice were harvested, minced, and dissociated. Granulocytes were isolated using MDSC-cell isolation kit (Miltenyi Biotec, 130-094-538). The first fraction containing Ly6G+ cells was collected.
T-Cell Suppression Assay
Purified CD3+ or CD8+ were isolated from naïve BALB/c mice and labeled with the proliferation dye CellTrace CFSE (5 M, BioLegend, 423801). A total of 1 × 105 cells in 100 μL per well were plated in a 96-well plate precoated with anti-mouse CD3ε (1 μg/mL, Southern- Biotech, 1530-01). CD11b+Ly6CintLy6G+ granulocytes were isolated from BM or spleens of 4T-Bone–injected mice using FACS-sorting or magnetic-associated cell sorting. Suppressor cells were then incubated with stimulated T cells (1 × 105 cells granulocytes/well). Dilution of CFSE was evaluated 2 to 3 days later by flow cytometry. For neutralization studies, T cells were preincubated with one of the following antibodies for 1 hour: αIL1β(5 μg/mL); αPD-1(10 μg/mL); αTIGIT (25 μg/mL) or appropriate isotype controls mouse IC (25 μg/ mL); Armenian hamster IC (7.5 μg/mL).
Ex Vivo Killing Assay
CD8+ T cells were isolated from the spleen of 4T-Bone-vaccinated mice, granulocytes were isolated from bone metastatic lesions. Briefly, splenocytes were cultured for 48 hours in the presence of 4T-Bone cell lysate, enhancing survival of reactive T cells. CD8+ T cells were isolated using magnetic beads as described above. 4T-Bone–specific CD8+ T cells were cocultured with or without CD45+CD11b+Ly6CintLy6G+ metastasis-associated granulocytes at a ratio of 1:1 in a round bottom 96-well plate. After 24 hours, cells were stained and analyzed for CD44 and membrane-bound CD107a by flow cytometry. For killing activity, 8,000 4T-Bone cells were seeded and incubated with CD8+ T cells w/wo metastasis-associated granulocytes, and GZMB and IFNγ secretion was measured by ELISA.
4T-Bone Vaccination
Seven-week-old Balb/c female received at least 4 doses of 4T-Bone lysate. 4T-Bone lysate solution was prepared by heat-inactivating 30 × 106 cancer cells for 2 minutes at 95°C and passed the solution through a syringe.
RNA Isolation and Quantitative Real-Time PCR
RNA was isolated from total bone samples, and qRT-PCR was performed. Briefly, bones were enzymatically digested to prepare single-cell suspension. Single cells were pelleted and resuspended into RNA lysis buffer using the PureLink RNA Mini Kit (Invitrogen; 12183018A). RNA concentration and purity were analyzed using NanoDrop 2000c Spectrophotometer. cDNA synthesis was conducted using qScript cDNA Synthesis Kit (Quanta Biosciences, 95047-025). Quantitative real-time PCRs (qRT-PCR) for mouse genes were conducted using SYBR Green FastMIX (Quanta Biosciences, P/ N84071) in a StepOne Real-Time PCR System. All experiments were performed in triplicate. RQ (2−ΔCt) was calculated. Relative expression was normalized to ACTB, UBC, and RSP23. All primers and oligonucleotide sequences used are shown in Supplementary Table S2.
Human Data from Public Databases
Human Data from Bulk RNA-seq
Global gene expression of il1b, s100a8, s100a9, cd33, pdcd1, tigit, ctla4, tim3, and lag3 was analyzed in human bone metastasis samples from several primary origins (n = 3 kidney; n = 2 colorectal; n = 1 lung; n = 8 prostate and n = 3 unknown origin) based on a publicly available data set GSE101607.
Human Data from scRNA-seq Analysis
scRNA-seq data of human bone metastasis from lung and breast primary origin were retrieved from Gene-Expression Omnibus (GSE123902 and GSE190772, respectively). Data integration, processing, and analysis were performed using the Seurat package v4.3.0 under R v4.0.3. Briefly, Seurat objects were created and cells were removed from analysis if they had unique feature counts >6,000 or <300, or if mitochondrial counts were >10%. Objects were integrated using FindIntegrationAnchors() and IntegrateData() functions, correcting for technical differences including batch effects. Counts were normalized and scaled. Clustering and 2D projection by t-distributed stochastic neighbor embedding (t-SNE) and UMAP were performed after dimensional reduction using the first 20 principal components (PC). Plots were generated using the ggplot2 function.
Human Staining for TIGIT and PD-1
Sheba Medical Center Cohort
Human patient samples (n = 35) were collected with written informed consent and processed at the Sheba Medical Center, Israel, in accordance with recognized ethical guidelines, under an approved Institutional Review Board (8153-10). Tissue sections were stained for TIGIT (clone 3B5, LS-B16022, LSBio) or PD-1 (clone NAT105, 315M-96, Cell Marque) and analyzed by an expert pathologist.
Baylor College of Medicine Cohort
The protocols for the collection and use of human bone metastasis samples were performed in accordance with the Declaration of Helsinki and approved by the Institutional Review Boards at Baylor College of Medicine (H-49396) and University of Texas Medical Branch (H-46675). All the patients have provided written informed consent on the use of their samples for research purposes when undergoing orthopedic surgery. Tissue sections were stained for TIGIT (clone 3B5, LS-B16022, LSBio).
Graphical Illustrations
Graphical elements used to create experimental design schemes were created in Bio Render licensed by Tel Aviv University Medical Faculty.
Statistical Analysis
Statistical analyses were performed using GraphPad Prism software. For two groups, statistical significance was calculated using t test with Welch correction unless otherwise stated. For more than two comparisons, one-way ANOVA with Tukey correction for mul- tiple comparisons was applied unless otherwise stated. P ≤ 0.05 was considered statistically significant unless otherwise stated. All experiments represent at least three biological repeats. Correlation analysis was performed using Pearson correlation, P ≤ 0.05 was considered statistically significant. Outliers were identified and removed using the Grubbs’ or ROUT method.
Data Availability
The data generated in this study will be publicly available in Gene- Expression Omnibus (GSE253444).
Authors’ Disclosures
A. Sonnenblick reports grants and personal fees from Novartis and Roche, personal fees from MSD, Gilead, Astra-Zeneca, Eli Lilly, and Pfizer outside the submitted work. R. Satchi-Fainaro reports personal fees from Teva Pharmaceutical Industries, Ltd. outside the submitted work. No disclosures were reported by the other authors.
Authors’ Contributions
L. Monteran: Conceptualization, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. N. Ershaid: Conceptualization, software, formal analysis, validation, investigation, methodology. Y. Scharff: Software, formal analysis, validation, methodology. Y. Zoabi: Software, formal analysis. T. Sanalla: Formal analysis. Y. Ding: Formal analysis, investigation, methodology. A. Pavlovsky: Formal analysis. Y. Zait: Investigation. M. Langer: Investigation. T. Caller: Methodology. A. Eldar-Boock: Resources, formal analysis. C. Avivi: Formal analysis. A. Sonnenblick: Resources, methodology. R. Satchi-Fainaro: Resources. I. Barshack: Resources, methodology. N. Shomron: Software, resources. X.H. Zhang: Resources, methodology. N. Erez: Conceptualization, resources, supervision, funding acquisition, investigation, methodology, writing–original draft, writing–review and editing.
Acknowledgments
The authors thank Dr. Irena Shur and Dr. Daria Makarovsky at the Faculty of Medicine Interdepartmental Core Facility (SICF) for their help with imaging and FACS analyses. We thank Dr. Yaron Carmi for his advice and helpful discussions. This study was supported by grants to N. Erez from the United States DoD (A CDMRP Break-through Award BCRP Award ID: W81XWH2110394). The Israel Cancer Research Fund (ICRF Project Grant), Worldwide Cancer Research, The Israel Science Foundation (ISF IPMP #3495/19 and ISF #693/23). N. Erez and A. Sonnenblick received support from The Richard Eimert Research Fund on Solid Tumors.
Note Supplementary data for this article are available at Cancer Discovery Online (http://cancerdiscovery.aacrjournals.org/).