The landscape of HCs in TME cells in BC

We examined the landscape of HCs using the BC scRNA-seq dataset described previously (Fig. 1A) [11]. We identified 40,036 TME cells in 6 samples from BC patients with brain and liver metastases. These cells were categorized into 7 different types, including malignant cells, endothelial cells, mural cells, CAFs, myeloid cells, B cells, and T cells (Fig. 1B). We also used tSNE to reduce the dimensions and explore the distribution between different metastasis location groups (Fig. 1C). The cell proportions in each patient were also assessed and shown in Fig. 1D and Supplementary Table S2. Besides, the annotated cell types were confirmed through the expression of canonical markers and the findings were presented using a heatmap (Fig. 1E), and a bubble plot provided a scaled expression level and proportion of specific markers expressed by each cell type (Fig. 1F). By cell-chat analysis, we also found that these cell types interacted in diverse and distinct manners (Fig. 1G). Moreover, it was clear that HCs were indeed expressed differently in BC metastases according to the scRNA-seq dataset. For example, ASF1A, ASF1B, CHAF1A, CHAF1B, HIRA, HJURP, IPO4, MCM2, NPM2, NPM3, and TONSL exhibited low expression levels in almost all cell types. In contrast, HSP90AA1, HSP90AB1, HSPA8, NAP1L1, NCL, and NPM1 are highly expressed in all cell types (Fig. 1H). Besides, we compared the expression level of HCs between BM and LM and observed that, in BM group, most HCs expression levels are higher than LM in malignant and B cells, whereas are lower in other cell types (Supplementary Figure S1 and S2).

Novel HCs‑mediated CAFs contributed to the TME of BC

We first extracted the CAFs subgroup from the scRNA-seq dataset. Based on the pseudotime analysis, we found that the HCs were crucial to CAFs trajectory process (Fig. 2A). By NMF algorithm, we identified 4 HCs‑mediated CAFs subgroups which named as HSP90AB1 + CAF-C1, DEK + CAF-C2, NASP + CAF-C3, and NoneHistone_CAF-C4 (Supplementary FigureS3). We then used cell-chat analysis and found that each HCs‑mediated CAFs subgroup had different numbers of ligand-receptor connections, that is, HSP90AB1 + CAF-C1 and DEK + CAF-C2 subgroups had more connections whereas NASP + CAF-C3 and NoneHistone_CAF-C4 possessed less connections (Fig. 2B). Among these subgroups, the HSP90AB1 + CAF-C1 and DEK + CAF-C2 proportions had higher percentages in BC LM samples than that in BM samples while the NASP + CAF-C3 and NoneHistone_CAF-C4 proportions had lower percentages in BC LM samples (Fig. 2C). Besides, the result of the KEGG enrichment analysis showed that the HSP90AB1 + CAF-C1 subgroup was related to numerous classic biological processes such as apoptosis, cellular senescence, TCA cycle, DNA replication, HIF-1 signaling pathway, etc., and the DEK + CAF-C2 subgroup exhibited activities in proteasome, ribosome, and TGF-beta signaling pathway. The NASP + CAF-C3 subgroup was found participated in cell adhesion molecules while the NoneHistone_CAF-C4 did not display a specific biological process (Fig. 2D).

Fig. 2

HCs modified the features of CAFs. (A) Trajectory analysis revealed the role of HCs in CAFs. (B) Cell–Cell communications from HCs-mediated CAFs to malignant cells. (C) Bar plot for 4 HCs-mediated CAFs clusters between brain metastasis and liver metastasis patients. (D) Heatmap showing the activated KEGG pathways in HCs‑mediated CAFs. (E) Different HCs-mediated CAFs clusters were correlated with the previous signatures. (F) Heatmap showing the significantly different TFs among HCs-mediated CAFs. (G) Heatmap showing the different average expression of common signaling pathway genes in the HCs-mediated CAFs, including collagens, ECM, MMPs, TGFβ, Neo-Angio, Contractile, RAS and Proinflammatory

Moreover, we calculated pan-CAF signatures activities among these subgroups, and we found that the HSP90AB1 + CAF-C1 and DEK + CAF-C2 subgroups were obviously correlated with desmoplastic CAF (pan-dCAF), inflammatory CAF (pan-iCAF), and proliferating CAF (pan-pCAF), whereas the NASP + CAF-C3 and NoneHistone_CAF-C4 subgroups were more closely to myofibroblast-like CAF (pan-myCAF) (Fig. 2E). Additionally, analysis of gene regulatory networks among HCs‑mediated CAFs revealed significant differences in TFs. Notably, the HSP90AB1 + CAF-C1 subgroup was characterized by enhanced TF activities of FOS, FOSB, JUN, JUNB, STAT1, etc., and the DEK + CAF-C2 subgroup exhibited upregulated TF activities of IRF1, CEBPD, TBX2, etc. As for the NASP + CAF-C3 and NoneHistone_CAF-C4 subgroups, TF activities like ETS1, ELF1, FOXP1, STAT2 were increasing (Fig. 2F, Supplementary FigureS4). Furthermore, we collected key CAF phenotype markers surface protein genes and compared their expression levels among the HCs-mediated CAFs subgroups. The result indicated that most of them were upregulated in the HSP90AB1 + CAF-C1 and DEK + CAF-C2 subgroups (Fig. 2G).

HCs‑mediated macrophages/B cells resembled classical characteristics

Myeloid cells were extracted from the scRNA-seq dataset and split into 4 minor cell types included dendritic cells (DCs), macrophages, mast cells, and monocytes (Fig. 3A). We screened out macrophages and pseudotime analysis also revealed that the HCs were vital to macrophages trajectory process (Supplementary FigureS5). We then performed NMF algorithm analysis based on HCs expression. We identified 10 clusters named as RSF1 + Macro-C1, NAP1L1 + Macro-C2, DEK + Macro-C3, ATRX + Macro-C4, NPM1 + Macro-C5, SET + Macro-C6, NCL + Macro-C7, HSPA8 + Macro-C8, HSP90AB1 + Macro-C9, and NoneHistone_ Macro-C10 (Supplementary FigureS6). We compared each cluster proportion between liver and BM samples, and we found that the NAP1L1 + Macro-C2 cluster possessed a significantly higher proportion in LM samples while the RSF1 + Macro-C1 cluster was more concentrated in BM samples (Fig. 3B). Similar to CAFs, we also noticed varying connections between HCs‑mediated macrophages and malignant cells, that is, the RSF1 + Macro-C1 cluster, NAP1L1 + Macro-C2 and DEK + Macro-C3 clusters had a large number of links whereas the HSP90AB1 + Macro-C9 and NoneHistone_ Macro-C10 clusters had the less (Fig. 3C). Afterwards, we calculated scores of the macrophage-related signatures in each cluster, and the result showed that the RSF1 + Macro-C1, NAP1L1 + Macro-C2 and DEK + Macro-C3 clusters were significantly associated with M1-like macrophage while the HSP90AB1 + Macro-C9 and NoneHistone_Macro-C10 clusters were strongly related to M2-like macrophage (Fig. 3D). Enrichment analysis also found obvious differences among these clusters (Supplementary FigureS7). Besides, we performed SCENIC analysis and found that multiple TFs, such as FOS, FOSB, JUN, JUNB, JUND, etc. were activated in the RSF1 + Macro-C1 and NAP1L1 + Macro-C2 clusters. However, we only observed YY1 activation in the HSP90AB1 + Macro-C9 and NoneHistone_ Macro-C10 clusters (Fig. 3H, Supplementary FigureS8). Previous studies have confirmed that macrophages play an essential role in metabolism. Therefore, we used ssGSEA algorithm to identify the relationship between metabolic pathway activities and each HCs‑mediated macrophage cluster. Interestingly, significant differences were detected among these clusters. The RSF1 + Macro-C1, NAP1L1 + Macro-C2, and DEK + Macro-C3 clusters showed higher metabolic activities in TCA cycle and glycolysis, etc. whereas other clusters fixed on metabolic pathways related to linoleic and taurine/hypotaurine acid metabolism (Fig. 3E).

Fig. 3

NMF clusters of HCs for macrophages and B cells. (A) t-SNE plot of myeloid cells. (B) Bar plot for 10 HCs-mediated macrophages clusters between brain metastasis and liver metastasis patients. (C) Cell–Cell communications between main HCs-mediated macrophage cells to malignant cells by Cellchat analysis. (D) Violin plots of M1 and M2 macrophage-related signatures scores among HCs-mediated macrophages clusters. (E) Heatmap showing significantly different metabolic signaling pathways among HCs-mediated macrophages clusters. (F) Bar plot for 5 HCs-mediated B cells clusters between brain metastasis and liver metastasis patients. (G) Cell–Cell communications between main HCs-mediated B cells to T cells by Cellchat analysis. (H) Heatmap showing the significantly different TFs among HCs-mediated macrophages and B cell clusters

We also explored B cells heterogeneity based on the result of NMF algorithm analysis. 5 clusters were identified and named as RSF1 + B-C1, HSPA8 + B-C2, DEK + B-C3, ATRX + B-C4, and NoneHistone_ B-C5 (Supplementary FigureS9). We found that the proportions of the RSF1 + B-C1 and DEK + Macro-C3 clusters were consistently higher in BM samples, and the NoneHistone_ B-C5 cluster possessed a significantly higher proportion in LM samples (Fig. 3F). Cell-chat analysis showed that HCs‑mediated B cells clusters had similar links to T cells (Fig. 3G). However, the result of enrichment analysis still indicated that the ATRX + B-C4 and NoneHistone_ B-C5 clusters were weakly related to classic biological pathways (Supplementary FigureS10). Besides, we still found significant differences among these clusters during SCENIC analysis (Fig. 3H, Supplementary FigureS11).

HCs‑mediated T cell phenotypes underscored the antitumor immune response in BC

We renamed 6 main cell types among the detected T cells, including CD4+, CD8+, NK, NKT, Tregs, and other T cells (Fig. 4A). Monocle analyses confirmed that HCs were correlated with T cells trajectory process (Supplementary FigureS5). By NMF algorithm analysis, we identified 5 HCs‑mediated CD4 + T clusters (ATRX + CD4 + T-C1, HSP90AA1 + CD4 + T-C2, BAZ1A + CD4 + T-C3, SET + CD4 + T-C4, and NoneHistone_CD4 + T-C5), 9 HCs‑mediated CD8 + T clusters (ATRX + CD8 + T-C1, RSF1 + CD8 + T-C2, NCL + CD8 + T-C3, HSPA8 + CD8 + T-C4, SET + CD8 + T-C5, DEK + CD8 + T-C6, NAP1L1 + CD8 + T-C7, HSP90AB1 + CD8 + T-C8, and NoneHistone_CD8 + T-C9), 4 HCs‑mediated NK clusters (DEK + NK-C1, NPM1 + NK-C2, ATRX + NK-C3, and NoneHistone_NK-C4), and 4 HCs‑mediated Treg clusters (HSPA8 + Treg-C1, DEK + Treg-C2, RSF1 + Treg-C3, and NoneHistone_Treg-C4) (Supplementary FigureS12). We performed Cell-chat analyses and found that the NoneHistone_Treg-C4 had more ligand-receptor links compared with other HCs‑mediated Treg clusters, and the NoneHistone clusters in HCs‑mediated CD4 + T, CD8 + T, and NK cells possessed less links (Fig. 4B). We then assessed the proportion of each cluster. Although the proportion of the NoneHistone clusters were consistently higher in BM samples, we only found significant differences in Tregs group (Fig. 4C). Besides, these HCs-mediated T cell phenotypes expressed obvious differences among TFs based on network regulatory analysis (Fig. 4D, Supplementary Figure S13-16). Moreover, HCs- mediated T clusters were associated with numerous differences in the expression of immune co-inhibitors, co-stimulators, and functional T cell markers (Fig. 4E F).

Fig. 4

NMF clusters of HCs for T/NK cells. (A) t-SNE plot of T/NK cells by six cell types, including CD4 + T cells, CD8 + T cells, Tregs, NK cells, NKT cells, and other T cells. (B) Cell–Cell communications from main HCs-mediated T/NK cells to other cells by Cellchat analysis. (C) Bar plot of main HCs-mediated T/NK cells clusters between brain metastasis and liver metastasis patients. (D) Heatmap showing significantly different TFs among HCs-mediated clusters in CD4 + T cells, CD8 + T cells, NK cells, and Tregs. (E, F) Heatmap showing significantly different features among T clusters in CD4 + T cells, CD8 + T cells, NK cells, and Tregs, including immune stimulators, inhibitors and T cell function marker genes, as well as four T function signatures (T exhaustion score, T cytotoxic score, T effector score, and T evasion score)

HCs‑mediated TME patterns guided Tumor prognosis and immunotherapy

Through the utilization of tumor samples and corresponding BM samples sourced from GSE173661, alongside normal and tumor tissues acquired from the TCGA database, our investigation has revealed a notable alteration in the HCs activity score, indicating the significance of HCs in the BC process (Fig. 5A and B). We also calculated the HCs activity scores between BM and LM in each cell type, and we found that the HCs activity scores in BM in malignant and B cells are higher than LM group, whereas are lower in other cell types (Supplementary FigureS17). In order to determine the predictive significance of HCs-mediated TME signature, we computed the enrichment score of each HCs-mediated TME cell subtype. Subsequently, the HR for overall survival (OS) was calculated by performing univariate Cox regression analysis for each HCs-related cell subtype in 5 BC cohorts. Notably, we observed significant differences in OS rates among these sub-clusters. For example, HSPA8 + Treg-C1 were identified as unfavorable for BC survival, whereas RSF1 + Macro-C1, NAP1L1 + Macro-C2, DEK + Macro-C3, ATRX + Macro-C4, NoneHistone_Macro-C10, RSF1 + B-C1, HSPA8 + B-C2, DEK + B-C3, ATRX + B-C4, ATRX + CD4 + T-C1, BAZ1A + CD4 + T-C3 and NCL + CD8 + T-C3 were associated with a favorable prognosis in BC (Fig. 5C). Furthermore, in order to forecast the immune response in individuals who received immunotherapy, we employed the logistic regression method to calculate the OR for immune response of each HCs-related cell subtype in 8 ICB cohorts. We observed similar significant phenomena that HCs were relevant to patients’ immunotherapy responses, especially for B cells and macrophages. For example, RSF1 + B-C1, HSPA8 + B-C2, DEK + B-C3, ATRX + B-C4, RSF1 + Macro-C1, NAP1L1 + Macro-C2, DEK + Macro-C3, and ATRX + Macro-C4 were associated with a favorable immunotherapy response in BC (Fig. 5D).

Fig. 5

Overall of the prognosis, immunotherapy response and immunity pathways correlations of HCs-mediated cells types in the public bulk RNA-seq cohorts. (A) Box plot of HCs activity between primary and paired brain metastatic tissues in GSE173661 cohort (* p < 0.05). (B) Violin plot of HCs activity between normal and tumor tissues in the TCGA-BRCA cohort (**** p < 0.0001). (C) Bubble plot of OS analyses (data from 5 BC cohorts). (D) Bubble plot of immunotherapy response analyses (data from 8 immunotherapy cohorts with response rate) (E) Heatmap showing significant correlations between cancer immunity cycles and immunoregulation-related pathways with all HCs-mediated cluster scores (* p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001)

The TME plays a vital role in influencing the immunotherapy effectiveness. Hence, we computed the cancer–immunity cycle scores of BC samples from the TCGA-BRCA. Subsequently, we conducted an analysis to examine the associations between the enrichment score of each HCs-mediated TME cell subtype and the cancer–immunity cycle scores. Notably, the levels of different anti-cancer immune responses, including the release of cancer cell antigens, T cell recruiting, CD8 T cell recruiting, Th1 cell recruiting, NK cell recruiting, and killing of cancer cells, were observed to be significantly elevated in RSF1 + Macro-C1, NAP1L1 + Macro-C2, DEK + Macro-C3, RSF1 + B-C1, HSPA8 + B-C2 and DEK + B-C3, etc. (Fig. 5E). We also computed the immunoregulation-related pathways scores of BC samples from the TCGA-BRCA dataset as well as examined the associations between the enrichment score of each HCs-mediated TME cell subtype with it, and we found similar significant phenomena that immunoregulation-related pathways scores were significantly elevated in RSF1 + Macro-C1, NAP1L1 + Macro-C2, DEK + Macro-C3, whereas were decreased in NCL + Macro-C7, HSPA8 + Macro-C8, and NoneHistone_Macro-C10, indicating that HCs might play an important role in TME (Fig. 5E).

HSPA8 deficiency inhibits Tumor cell migration and invasion

In order to investigate the impact of HCs on tumor cells, we specifically chose the HSPA8 to examine its potential tumorigenic effect. We firstly explore its expression difference between tumor and normal tissues from the combination of the TCGA and GTEx databases. We found that HSPA8 was significantly upregulated in most cancer type while significantly downregulated in kidney renal clear cell carcinoma (KIRC), kidney renal papillary cell carcinoma (KIRP), and acute myeloid leukemia (LAML) (Fig. 6A). Then, we assessed the expression level of HSPA8 at the single-cell level, and Fig. 6B showed that HSPA8 widely distributed in various cell type. Next, we performed log-rank and Cox regression analyses to investigate the prognostic role of HSPA8, and we found the its heterogeneity in the prognostic value, that is, HSPA8 was a risk factor in BLCA, BRCA, CESC, HNSC, LIHC, etc. while was a protective factor in KIRC, LGG, OV, READ, etc. (Fig. 6C). Besides, we validated its prognostic value in BC RNA-seq and microarray datasets from bc-GenExMiner database, and the results were consistent with our findings (Fig. 6D). To evaluate the involvement of HSPA8 in the in vitro metastatic behavior of MDA-MB-231 and BT-549 cells, we employed siRNA (HSAP8) transfection to knock down HSPA8 expression. The expression levels of HSPA8 proteins were effectively decreased in MDA-MB-231 and BT-549 cells when compared to untransfected cells (Fig. 6E). As Fig. 6F H shown, knockdown of HSAP8 significantly decreased the migratory capacity of MDA-MB231 and BT-549 cells during transwell and wound healing assays.

Fig. 6

HSPA8 deficiency inhibits tumor cell migration and invasion. (A) Boxplots of the HSPA8 expression between tumor and normal tissues in the TCGA pan-cancer cohorts (* p < 0.05, ** p < 0.01, *** p < 0.001). (B) Heatmap of the HSPA8 expression among different cell types in TISCH database. (C) Heatmap of the prognostic value of HSPA8 in the TCGA pan-cancer cohorts. (D) Survival analyses of HSPA8 using K-M analyses in BC RNA-seq and microarray datasets from bc-GenExMiner database. (E) Western blot assays showing the efficacy of siRNAs targeting HSPA8 in BC cell lines. (F) Transwell migration assays were performed to measure the migration abilities of HSPA8 in BC cell lines. (G) Wound healing assays were performed to measure the migration abilities of HSPA8 in BC cell lines. (H) Boxplots of the number of cells migrated per field and relative healing area (% control) in BC cell lines (* p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001). ACC, adrenocortical carcinoma; BLCA, bladder urothelial carcinoma; BRCA, breast invasive carcinoma; CESC, cervical squamous cell carcinoma and endocervical adenocarcinoma; CHOL, cholangiocarcinoma; COAD, colon adenocarcinoma; DLBC, lymphoid neoplasm diffuse large B-cell lymphoma; ESCA, esophageal carcinoma; GBM, glioblastoma multiforme; HNSC, head and neck squamous cell carcinoma; KICH, kidney chromophobe; KIRC, kidney renal clear cell carcinoma; KIRP, kidney renal papillary cell carcinoma; LAML, acute myeloid leukemia; LGG, brain lower grade glioma; LIHC, liver hepatocellular carcinoma; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma; MESO, mesothelioma; OV, ovarian serous cystadenocarcinoma; PAAD, pancreatic adenocarcinoma; PCPG, pheochromocytoma and paraganglioma; PRAD, prostate adenocarcinoma; READ, rectum adenocarcinoma; SARC, sarcoma; SKCM, skin cutaneous melanoma; STAD, stomach adenocarcinoma; TGCT, testicular germ cell tumors; THCA, thyroid carcinoma; THYM, thymoma; UCEC, uterine corpus endometrial carcinoma; UCS, uterine carcinosarcoma; UVM, ocular melanomas. ALL, acute lymphoblastic leukemia; AML, acute myeloid leukemia; KIPAN, pan-kidney cancer; LSCC, laryngeal squamous cell carcinoma; OS, osteosarcoma