Abstract
The tumor microenvironment is composed of numerous cell types, including tumor, immune and stromal cells. Cancer cells interact with the tumor microenvironment to suppress anticancer immunity. In this study, we molecularly dissected the tumor microenvironment of breast cancer by single-cell RNA-seq. We profiled the breast cancer tumor microenvironment by analyzing the single-cell transcriptomes of 52,163 cells from the tumor tissues of 15 breast cancer patients. The tumor cells and immune cells from individual patients were analyzed simultaneously at the single-cell level. This study explores the diversity of the cell types in the tumor microenvironment and provides information on the mechanisms of escape from clearance by immune cells in breast cancer.
One Sentence Summary Landscape of tumor cells and immune cells in breast cancer by single cell RNA-seq
Breast cancer is the most common cancer and the leading cause of death from cancer in women worldwide(1). Four subtypes of breast cancer with distinct expression profiles have been classified based on gene expression signatures associated with highly variable clinical characteristics(2, 3). To design targeted treatment for such a diverse disease, understanding its molecular mechanism of initiation and progression is essential(4). Several studies have shown that the presence of tumor-infiltrating lymphocytes (TILs) is associated with breast cancer progression and neoadjuvant chemotherapy response(5-7). TIL levels within and between different subtypes of breast cancer vary(8). Based on the CD8+ T cell infiltration phenotype, tumors can be categorized as T-cell-inflamed and non-T-cell-inflamed tumors(9). In T-cell-inflamed tumors, the tumor cells act together with the tumor microenvironment (TME) to inhibit the antitumor functions of T cells. This process results in an exhausted T cell phenotype. Non-T-cell-inflamed tumors escape immune system clearance by preventing T cell infiltration into the TME(10). Cancer-associated fibroblasts (CAFs), tumor-associated macrophages (TAMs) and oncogenic pathway alterations in tumor cells are reportedly responsible for the non-T-cell-inflamed phenotype(10-14). However, the mechanism of immune evasion in breast cancer remains unclear. A recent study showed that a high number of TILs was beneficial for triple negative breast cancer (TNBC) and HER-2-positive breast cancer patient survival, but this parameter was an adverse prognostic factor for luminal HER2-negative breast cancer patient survival(15). High-resolution mapping of the TME composition and cell states of different breast cancer subtypes will help us understand the different effects of TILs and the mechanism of tumor immune evasion in breast cancer.
In the past five years, the development of high-throughput single-cell RNA sequencing has enabled high-resolution studies of biological processes(16). Single-cell transcriptome profiling of tumor cells has been used to characterize heterogeneous tumor cells and tumor-associated stromal and immune cells (17-19). To understand the mechanisms of escape from clearance by immune cells in breast cancer further, we molecularly dissected the breast cancer TME by using single-cell RNA-seq.
We performed single-cell RNA-seq on cells isolated from the tumor tissues of 15 human breast cancer patients (Table S1). Our samples included four subtypes of breast cancer: luminal A (P01, P02, P03, P04, P05, P06, P07A, P07B and P08), luminal B (P09 and P10), HER2+ (P11 and P12) and TNBC (P13, P14 and P15). We isolated single cells from tumor tissues without surface marker preselection. With proper quality control and filtration, we obtained single-cell transcriptome data for 52,163 individual cells (Fig. 1A).
To analyze the composition of the breast cancer TME, we clustered the cells according to their expression profiles. Preliminary clustering was applied to all cells using the Seurat package (Fig. S2, A, B and C). Each cluster was designated as epithelial or non-epithelial, and the epithelial cells were distinguished further as tumor cells or non-tumor epithelial cells based on the inferred copy number variation information (Fig. S3). The clusters of non-epithelial cells were annotated as T cells, macrophages, B cells, dendritic cells, NK cells and CAFs (Fig. 1, B and C). Based on specific cell markers, the T cell clusters were annotated further as CD4+ T cells, regulatory T cells (Tregs), CD8+ T cells, and CD8+ exhausted T cells (Fig. 1D). The precise identification of T cells will assist in analyzing the immune evasion mechanism of each sample. The tumor cells from different patients clustered separately, suggesting a high degree of inter-tumor heterogeneity (Fig. 1E). Finally, the cell type percentages of individual patients were calculated (Fig. 1F). The concentration and composition of the TILs were variable between different patients, which revealed the complexity of the tumor tissues. In the P13 and P14 samples, which were both TNBC and had a high number of TILs, most of the TILs were T cells and B cells. In many of the luminal A and luminal B samples (P02, P04, P06, P07A, P07B, P08 and P09), macrophages constituted the first or second largest TIL population.
Gene expression profiling and massively parallel sequencing have identified four main molecular subtypes of breast cancer; these subtypes have significant genetic and phenotypic diversity and inter-tumor heterogeneity among patients(2, 3). To further analyze the tumor cell heterogeneity within a patient, we examined the breast cancer subtype and oncogenic pathways based on our single-cell RNA-seq data. Unsupervised clustering was applied to each sample, and each subpopulation had a distinct expression pattern (Fig. 2A and 2B). Differentially expressed genes were analyzed for each subpopulation, and their functional characteristics were classified (Fig. 2C and 2D).
We also used gene set variation analysis (GSVA) to identify the diversity of cancer phenotypes and signaling pathways among patients(20). EMT programs in cancer have been widely considered as potential triggers of drug resistance, invasion, and metastasis; however, using EMT data for cancer diagnosis and treatment is difficult because their patterns and significance in human epithelial tumors in vivo are unclear(21). To understand EMT patterns in breast cancer in vivo further, we analyzed the EMT-associated gene signatures of the patients. Our results showed that the EMT programs displayed both inter-tumor and intra-tumor heterogeneity. Different subpopulations in the same patient exhibited distinct patterns of EMT programming, which suggest that intermediate states of the EMT program exist in vivo (Fig. 2E). An analysis of gene signatures associated with proliferation and cell cycle identified proliferating clusters in several samples. In sample P15, almost all clusters displayed a high proliferation gene signature score, suggesting that most of the cells in this sample were in active proliferation states. Only a few TILs were detected in sample P15, suggesting that this cancer progressed rapidly without an antitumor immune response. We also analyzed the activation of a series of oncogenic pathways for each individual cluster in each sample, many of which were diverse across the clusters; these data provided us with the resources to understand tumor cell heterogeneity (Fig. S4).
We used the PAM50 classifier to predict breast cancer subtype (luminal A, luminal B, Her2, basal and normal-like) for individual tumor cell clusters in each sample(22). Intra-sample heterogeneity was present in our results showing that the subpopulations of one sample corresponded to different breast cancer subtypes (Fig. 2F). Six luminal A samples (P03, P05, P06, P7A, P7B and P08) also contained clusters conforming to the luminal B subtype. Our results have revealed intra-tumor heterogeneity in breast cancer that could not be detected by bulk RNA-seq or IHC staining.
Our scRNA-seq data highlight the immune cell and CAFs diversity in the breast cancer TME. We identified 9 clusters of T cells, 5 clusters of TAMs, 5 clusters of B cells, 2 clusters of DCs and 2 clusters of CAFs in our samples, and each of these clusters has a distinct expression pattern (Fig. S5A and S5B). To examine the functional state of these cell clusters, we performed GSVA to identify functional phenotype diversity in immune cells and CAFs.
Among the 9 clusters of T cells, 5 of them were annotated as CD8+ T cells, and 3 of them were annotated as CD4+ T cells. T cell-mediated cytotoxicity is critical for tumor cell clearance. Based on the functional T cell associated gene signature analysis, the 5 clusters of CD8+ T cells gradually activated cytotoxic and exhausted T cell gene signatures (Fig. 3A)(23). The clusters annotated as CD8+ Texhausted and CD8+ T-4 expressed similar levels of cytotoxic genes, but multiple inhibitory receptors were overexpressed in the exhausted T cell cluster; this overexpression may attenuate the cytotoxicity of this T cell cluster (Fig. 4C). Thus, among these five CD8+ T cell clusters, T cells in the CD8+ T-3 and CD8+ T-4 clusters may be able to kill tumor cells. The three CD4+ T cell clusters also had varied T cell functional states. One cluster was annotated as regulatory T cells, naïve CD4+ T-2s and CD4+ T-1s with an exhausted signature (Fig. 4B and 4C). T cells in these three CD4+ T cell clusters might regulate CD8+ T cell function to different degrees.
TAMs, a major type of leukocytes that infiltrate tumors, play an important role in tumor growth and progression(24). Macrophages in different TMEs are functionally distinct. Our single-cell RNA-seq data identified 5 clusters of macrophages, revealing a high complexity of TAMs in vivo. Clusters 3, 4 and 5 manifested mainly in the M2 state, which may promote tumor growth by suppressing antitumor immune responses (Fig. 3D). We also examined the expression of a series of genes associated with the immune regulatory function of macrophages in these clusters (Fig. 3E). Macrophages in clusters 2 and 4 expressed higher levels of HLA-II genes than the other 3 clusters. In cluster 4, we also detected the expression of CD274, PDCD1LG2 and IOO1, which have been reported to be associated with suppressing CD8+ T cell functions.
We also analyzed the expression pattern of several B cell markers, which were previously reported to be associated with B cell maturation and regulatory functions in five B cell clusters (Fig. 3F). Cluster 2 was HLA-II-positive and may be able to present antigens. Cluster 3 was TCL1A-positive; TCL1A-positive cells are reportedly involved in controlling cervical cancer development(25). B cells in clusters 4 and 5 expressed high levels of immunoglobulin-associated genes, suggesting that they were antibody secretion B cells, which might be involved in tumor immune regulation(26).
A recent study had identified four distinct CAF subsets in breast cancer; based on this classification, we examined the properties of CAFs in our data(27)(Fig. 3F and 3G). CAF-1 and CAF-2 showed distinct expression patterns of CAF markers and pathways. CAF-1 cells expressed higher levels of immune suppression associated genes. This finding suggests that CAF-1 cells may be associated with tumor immune evasion in breast cancer.
The cross-talk between tumor cells and stromal cells in tumor microenvironment is critical for cancer initiation and progression. Based on our single cell RNA-seq data, we analyzed intercellular communication in breast cancer TME using a dataset of human ligand receptor pairs(28). A network of potential cell-cell interactions was constructed showing extensive communications between tumor cells and stromal cells (Fig. 4A). Macrophages had the most potential connections with tumor cells (Fig. 4B). The interactions between macrophages and tumor cells mainly mediated by multiple ligand-receptor pairs, including CSF1-CSF1R, IGF1-IGF1R and IL1B-IL1R1, which may be associated with macrophages stimulation and response (Fig. 4C). CAFs densely communicated with tumor cells and macrophages through CXCL12-CXCR4, IGF1-IGF1R and FGF7-FGFR1(Fig. 4C). Macrophages recruited CXCR3 positive T cells by producing CXCL10 and CXCL9, and CCL4 secreted by T cells and DC may regulate recruitment of CCR1 positive macrophages to TME. Besides interactions with known functions, we also identified several potential intercommunications in breast cancer TME, which need further confirmation in future.
Class I human leukocyte antigen (HLA) is responsible for cancer-specific neoantigen presentation(29). Genetic alterations in HLA-I molecules are reportedly associated with the escape of tumor cells from immune cell clearance(30, 31). To analyze the mechanism of immune evasion in breast cancer comprehensively, we first examined HLA-I gene expression in the tumor cells at the single-cell level (Fig.5A). The expression of HLA-associated genes was diverse across patients. In most samples (P01, P03, P04, P05, P06, P07A, P07B, P09, P10, P11, P12, P14 and P15), HLA-I gene expression was positively correlated with the percentage of infiltrating T cells. In one TNBC sample (P15) that lacked lymphocyte infiltration, all of the HLA-I genes were downregulated. In contrast, HLA-I genes were overexpressed in sample P14, and the tumor tissue of this patient was highly T cell-inflamed with 50% infiltrating T cells. These results suggest that in samples P01, P05, and P15, HLA-I gene downregulation played a critical role in preventing immune cell infiltration into the tumor tissues.
Our single-cell analysis of tumor cells also revealed diverse HLA-I expression in different subpopulations of the same sample. In sample P02, only cluster 3 showed HLA-I gene expression, which may be adequate for priming the antitumor immune response. In samples P04, P06 and P12, HLA-I gene expression could be detected, but only a few infiltrating T cells were found, suggesting that another T cell exclusion mechanism was involved in these samples.
In samples P02, P03, P08, P09, P10, P13 and P14, over 10% of infiltrating T cells were detected. In these samples, the antitumor immune response was primed, and T cells were recruited to the tumor tissues but failed to kill all of the tumor cells. The T cell composition of each patient, which was diverse among patients, revealed the functional T cell states of individual patients (Fig. 5B). These results showed that in infiltrating T cell populations, the percentage of cytotoxic CD8+ T cells (CD8+ T-3 and CD8+ T-4) in individual samples was less than 35%, which may be not adequate for killing all of the tumor cells. In addition, the existence of regulatory T cells in the tumor tissues of these patients may also inhibit the antitumor function of cytotoxic CD8+ T cells. We also used the GSVA scores of T cells and macrophages to further evaluate the functional T cell and macrophage states in individual samples (Fig. S6B and S7; Fig. 5C and Fig. 5D). Cytotoxic scores revealed the ability of T cells to destroy tumor cells, the exhaustion score showed the dysfunctional state of the T cells, and the presence of Tregs and M2s revealed inhibitory T cells and macrophages, respectively. In sample P09, more than 50% of the TILs were macrophages, and most of the macrophages displayed the M2 phenotype, which could protect the tumor cells from T cell attack. For this sample, macrophages may be targeted for therapy. In sample P14, tumor cells expressed high levels of IDO1, LGALS9, CEACAM1, TNFSF10 and PDCD1LG2, which were all reported to inhibit cytotoxic T cell function (Fig. S6A). In sample P14, 8% of the infiltrating T cells were CD8+ exhausted T cells, and 12% were CD4+ exhausted T cells, which may directly contact the tumor cells and become exhausted. Immune checkpoint therapy may be effective for this patient. In samples P10 and P13, most of the infiltrating T cells had neither cytotoxic nor exhausted signatures, which suggests that these T cells were not activated in the tumor tissues. Effectively activating the T cells in these samples may be critical for treatment. Taken together, analyzing tumor cells and infiltrating immune cells simultaneously by single-cell RNA-seq enables us to understand the precise status of tumor tissues in individual patients comprehensively. This information may be very helpful for clinical diagnosis and therapy.
Using single-cell transcriptomic data from 53,000 cells, we dissected precisely the TME of 15 breast cancer patients by determining the proportion of each cell type and their gene signatures. Several subpopulations of immune cells and CAFs were identified. This finding highlights the diversity of macrophages, CAFs and B cells in the breast cancer TME. The mechanism of T cell exclusion and T cell exhaustion in the breast cancer samples was analyzed comprehensively based on the landscape of the TME.
Our results revealed that breast cancer tumor cells appear to interact with TAMs and CAFs to escape immune system clearance. Several signaling pathways and molecular factors that are likely involved in this process were identified, as well as complex and diverse active mechanisms in each sample. Thus, single-cell RNA-seq provides comprehensive insights into the immune suppression network in the TME, and this information could be a critical guide for personally precise therapy design for cancer patients.
Acknowledgment
This work was supported by the Shenzhen Municipal Government of China (grant GJHZ20170314152701465), National Natural Science Foundation of China (No.31500694, 81672593 and 81272899), Natural Science Foundation of Guangdong Province, China (2017B020227012), “sanming” project of medicine in Shenzhen(SZSM201512015). We also acknowledge in part by the Intramural Research Program of the National Institutes of Health, National Cancer Institute, Center for Cancer Research, and Division of Cancer Epidemiology and Genetics from Leidos-Frederick under contract # HHSN261200800001E.