Circadian pattern subtyping unveiling distinct immune landscapes in breast cancer patients for better immunotherapy

Abstract

Background

While epidemiological studies have established a firm link between circadian disruption and tumorigenesis, the role and mechanism are not fully understood, complicating the design of therapeutic targets related to circadian rhythms (CR). Here, we aimed to explore the intertumoral heterogeneity of CR and elucidate its impact on the tumor microenvironment (TME), drug sensitivity, and immunotherapy.

Methods

Based on unsupervised clustering of 28 CR genes, two distinct CR subtypes (cluster-A and cluster-B) were identified in the TCGA cohort. We further constructed a circadian rhythm signature (CRS) based on the CR genes primarily responsible for clustering to quantify CR activity and to distinguish CR subtypes of individual patients from external datasets. CR subtypes were evaluated by TME characteristics, functional annotation, clinical features, and therapeutic response.

Results

The cluster-B (low-CRS) group was characterized by highly enriched immune-related pathways, high immune cell infiltration, and high anti-tumor immunity, while the cluster-A (high-CRS) group was associated with immunosuppression, synaptic transmission pathways, EMT activation, poor prognosis, and drug resistance. Immunohistochemistry (IHC) results demonstrated that high CD8⁺ T cell infiltration was associated with low-CR-protein expression. Importantly, patients with low CRS were more likely to benefit from immune checkpoint blockade (ICB) treatment, possibly due to their higher tumor mutation burden (TMB), increased immune checkpoint expression, and higher proportion of “hot” immunophenotype.

Conclusion

In a nutshell, the cross talk in CR could reflect the TME immunoreactivity in breast cancer. Besides providing the first comprehensive pathway-level analysis of CR in breast cancer, this work highlights the potential clinical utility of CR for immunotherapy.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00262-023-03495-3.

Background

Worldwide, breast cancer (BC) has overtaken lung cancer as the most commonly diagnosed cancer [1]. Previous studies have suggested that the occurrence and development of BC are associated with genetic mutations [2]. However, compared to genetic causes, non-genetic causes are more prevalent in the onset of BC [1]. These include lifestyle factors that affect circadian rhythms, such as shift work [3], mistimed eating patterns [4], and the outdoor light at night [5]. In fact, the International Agency for Research on Cancer (IARC) classified circadian disruption as a “probable” Group 2A carcinogen back in 2007 [3].

Circadian rhythms (CR), which regulate periodic changes in physiological functions to maintain homeostasis, are the main manifestation of the biological clock. Not surprisingly, the immune system is under the control of CR [6]. For instance, by modulating cellular metabolism, CR can regulate the response of macrophages to immune stimulation [7]. The infiltration and function of immune cells in the tumor microenvironment (TME) are also influenced by CR. Changes in the immunological component of the TME can be detected in BC mice after chronic jet-lag treatment, including decreased levels of CD8⁺ T cells and anti-tumor chemokines (IFN-γ and CXCL13), and increased levels of tumor-promoting MHC-II^low tumor-associated macrophages (TAMs) and immunosuppressive chemokines (IL1β and IL10) [8]. Moreover, pan-cancer analysis has demonstrated dysregulated expression of CR genes in multiple tumor tissues including BC [9]. Among them, core negative regulators (PER1, PER2, PER3, CRY1, and CRY2) in the CR pathway are consistently downregulated. However, as a highly heterogeneous disease, the CR of BC tissues may not be consistent among different individuals. Currently, there is a lack of research on comparing CR between BC samples. Meanwhile, the influence of CR on infiltrating immune cells within the TME at the pathway level has rarely been reported.

In our study, we used bioinformatics tools to evaluate the patterns of CR genes expression and developed CR signature to quantify circadian activity in individual patients. We found that CR patterns were extensively associated not only with immune-related pathways, but also with tissue immunophenotypes and infiltration of multiple immune cell types. The association between CD8⁺ T cells and CR was confirmed by immunohistochemistry (IHC). Next, we proposed related potential mechanisms from the perspective of cell–cell communication. Finally, we demonstrated the prognostic relevance of CR patterns and evaluated their therapeutic value in immunotherapy.

Methods

Data collection and processing

mRNA expression, somatic mutation, and clinical data, including histology subtype, gender, and overall survival times of BC samples from The Cancer Genome Atlas (TCGA) were downloaded from the XENA database (http://xena.ucsc.edu/public/). After excluding male, metastatic, and duplicate samples, 1078 tumor samples and 112 normal samples were obtained. FPKM data were converted into TPM data for subsequent analysis. The “genefu” package [10] was used to classify the PAM50 intrinsic molecular subtypes based on the gene expression data. Single-cell RNA sequencing (scRNA-seq) dataset GSE176078 containing bulk RNA-seq data and corresponding scRNA-seq data from 24 patients was included for further analysis.

In all, 30 CR genes were gathered from the Kyoto Encyclopedia of Genes and Genomes database (KEGG, hsa04710, https://www.kegg.jp/, PER1, PER2, PER3, CRY1, CRY2, NR1D1, BHLHE40, BHLHE41, CSNK1D, CSNK1E, ARNTL, CLOCK, RORA, RORB, RORC, RBX1, CUL1, SKP1, BTRC, FBXW11, FBXL3, NPAS2, PRKAA1, PRKAA2, PRKAB1, PRKAB2, PRKAG1, PRKAG2, PRKAG3, and CREB1). Among them, 28 genes with TPM > 0 in at least half of the samples were selected for further analysis (RORB, PRKAG3 were removed).

Clustering of the expression patterns of 28 CR genes

Unsupervised clustering algorithm was applied to cluster analysis of CR genes in 1078 BC samples to detect the CR diversity in BC. The R packages "factoextra" and "stats" were used for the above steps. The hclust method with the “ward.D” algorithm was used as the class agglomeration method. The clustering results were visualized using the ComplexHeatmap package.

Constructing the circadian rhythm signature (CRS) to evaluate individual CR status

1. Identification of CRS genes. We used univariate and multivariate logistic regression analysis to screen for CR genes that predominantly contribute to clustering. A total of 13 genes that were consistently statistically significant (p  <  0.05 in both univariate and multivariate logistic regression) were used to construct the CRS.

2. Verification of the efficacy of the CRS. The CRS genes in the signature were incorporated into the multivariate logistic regression model. AUC (area under the receiver operating characteristic curve) and C-index (concordance index) were calculated to verify the discrimination of the model, while HL (Hosmer–Lemeshow) test was used to verify the calibration. These three indices were combined to evaluate whether the signature matched the distribution of CR clusters.

3. Setup a scoring procedure. The Gene set variation analysis (GSVA) algorithm [11] was used to score the CRS of tumor patients to obtain a corresponding CRS score.

Transcriptome and gene set analysis

The edgeR [12] package was used to analyze the difference between two circadian subtypes. Genes with fold change (FC) > 2 and FDR < 0.05 were regarded as significantly differentially expressed genes (DEGs). Single-cell sequencing data analysis was implemented by R packages “Seurat” [13] and “Cellchat” [14]. Overrepresentation analysis (ORA) and gene set enrichment analysis (GSEA) were performed using WebGestalt (WEB-based GEne SeT AnaLysis Toolkit, https://www.webgestalt.org/) [15]. Pathways enriched in different circadian patterns were identified from Hallmark50, KEGG and Gene Ontology (GO) with a false discovery rate (FDR) < 0.05. The estrogen receptor (ER) and estrogen signaling (ES) pathways were selected from the “c2.cp.v7.4” gene set downloaded from MSigDB (http://www.gsea-msigdb.org). GSVA algorithm was used to quantify activity of both estrogen-related pathways.

Estimation of TME cell infiltration and immune-related signature

The ESTIMATE algorithm was used to assess immune cell infiltration (immune score), tumor purity (ESTIMATE score), and stromal content (stromal score) [16]. The relative abundance of immune cell infiltration of 22 different subtypes was quantified using the CIBERSORT algorithm [17] and performed 1000 times for permutation test to ensure the stability of the calculation. The sample with low quantitative accuracy (p > 0.05) was removed. Normalized scores from TIP [18] (Tracking Tumor Immunophenotype) were downloaded to track the status of anti-cancer immunity across seven-step cancer-immunity cycle in the TCGA-BRCA cohort. The gene sets of “TME-associated signatures” were downloaded and scored by Single-sample Gene Set Enrichment Analysis (ssGSEA) algorithm using the “IOBR” [19] R package.

Solid tumors can be immunologically classified into two groups according to their immune status: “Cold” tumors are characterized by a lack of T lymphocyte infiltration within the TME, whereas “hot” tumors are infiltrated with predominantly CD8⁺ T lymphocytes. Following the method presented by Ringel et al. [20], BC patients from the TCGA cohort were clustered based on a list of CD8⁺ signature genes by using the ConsensusClusterPlus [21] package. In total, 1000 repetitions were conducted to ensure the stability of the classification.

Immuno-/chemotherapeutic response prediction

The Immunophenoscore (IPS) [22] and Tumor Immune Dysfunction and Exclusion (TIDE) [23] algorithms were used to predict response to immune checkpoint blockade (ICB) therapy. We also collected datasets about immunotherapy. The IMvigor210 cohort [24] and the GSE78220 dataset were included in the study to analyze the relationship between CRS and immunotherapy response. The IMvigor210 cohort comprises patient data for intervention treatment (atezolizumab, anti-PD-L1 antibody) of advanced urinary tract transitional cell carcinoma. Expression and detailed clinical information were downloaded from http://research-pub.gene.com/IMvigor210CoreBiologies/. The GSE78220 dataset contains mRNA expression data and corresponding clinical information in pre-treatment melanoma undergoing anti-PD-1 checkpoint inhibition therapy. Moreover, we used the R package “oncoPredict” [25] to predict the chemotherapy response of each sample in the TCGA cohort based on Genomics of Drug Sensitivity in Cancer (GDSC, https://www.cancerrxgene.org/). Spearman analysis was used to calculate correlation between drug sensitivity score and CRS score. Significantly correlated drugs were identified with thresholds of |R|> 0.2 and p value < 0.05.

Immunohistochemistry assay

Sections were obtained from paraffin-embedded BC tissues of patients (n = 11) who underwent surgery at the First Affiliated Hospital of Nanchang University (Nanchang, China). After heating, deparaffinization, and rehydration, sections were placed in sodium citrate buffer (pH 6.0) for antigen retrieval. 3% hydrogen peroxide (H₂O₂) was used to block endogenous HRP activity. Slides were blocked with 3% BSA and incubated with primary antibodies (anti-CLOCK, ab3517; anti-CRY1, ab54649; anti-PER2, ab227727; anti-CD8, ab178089; all purchased from Abcam, UK) at 4 °C overnight. Horseradish peroxidase-conjugated secondary antibodies and DAB were used for signal visualization. The sections were then counterstained with hematoxylin and dehydrated. Typical images were captured with a microscope (eclipse Ci-e; Nikon, Japan). The mean intensity of protein expression was quantified using Fiji software (1.53). High and low levels of protein expression were segmented by median.

Statistical analysis

To examine the differences between groups, we used Wilcoxon rank-sum test, Student's t-test, or Kruskal–Wallis test for continuous variables, and Chi-squared test or Fisher's exact test for categorical variables. Spearman’s correlation coefficient method was used for correlation analysis. Kaplan–Meier method was used for survival analysis, and the log-rank test was used to calculate the significance of the difference. All statistical analyses were two-tailed, and a p value < 0.05 was considered statistically significant.

Results

Identification of distinct CR patterns in breast cancer mediated by CR genes

To explore the relationship among CR genes, 1078 BC samples from TCGA were selected to calculate pairwise correlations between the expressions of 28 circadian genes (Fig. 1a). We found that positive correlations were prevalent among CR genes. Thus, the cross talk among the CR genes may be important for the generation of different CR patterns between individual tumors. Next, we applied hierarchical clustering based on the expression profiles of 28 CR genes to classify samples with qualitatively different CR patterns. In total, 514 samples were identified in the cluster-A, whereas the other 564 samples were identified in the cluster-B (Fig. 1b). Regarding expression levels, 27 of the 28 CR genes were found to be significantly higher in the cluster-A group, with only RBX1 showing the opposite trend (Fig. S1a). We observed an imbalanced distribution of the two CR subtypes in the PAM50 subtypes and different ER status (Fig. 1c). Only small proportions of ER-negative (14.77%), basal-like (10.00%), and Her2-enriched (17.39%) tumors exhibited the cluster-A subtype compared to 70.37% of Lum A tumors. Furthermore, principal component analysis suggested that there was a remarkable transcriptome difference between the two CR subtypes (Fig. 1d).

Fig. 1.

Patterns of CR genes and construction of CRS in the TCGA cohort. a Heatmap showing positive and negative correlations among CR genes. b Unsupervised clustering of 28 CR genes. Red: high expression; blue: low expression. c Sankey plot of the CR cluster with PAM50 subtype and ER status in the TCGA cohort. d Principal component analysis showing that CR genes could successfully discriminate two CR patterns. e–f CRS score and the mRNA expression level of 13 CRS genes among normal, cluster-A, and cluster-B samples. g An overview of the mapping between CRS score and other patient annotations in the TCGA cohort. h Correlation between CRS and estrogen-related pathway in normal and cancer samples, respectively. ER: estrogen receptor; ES: estrogen signaling (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001)

Given the intertumoral heterogeneity of CR in BC, we constructed a CRS (circadian rhythm signature, see Methods) to quantify CR activity in individual patients (Table S1, Fig. S1b–c). We found that about half of the CRS genes expression was highest in the cluster-A group, more in the normal group (normal adjacent tissue from TCGA), and lowest in the cluster-B group (Fig. 1e). The distribution of the CRS score shows the same trend and can correctly stratify patients into their cluster-A/B subtypes (Fig. 1f). To visualize the distribution of CR clusters across the CRS score, we draw an annotated bar chart, PAM50 subtype and various pathological parameters were also included (Fig. 1g). In addition to cluster-B, basal-like, Her2-enriched, and ER-negative tumors crowded in the low-CRS-score side. Considering that the basal-like, Her2-enriched subtypes often correspond to ER-negative specimens, the regulation of CR in BC may be related to estrogen signaling. Correlation analysis results showed that both estrogen-related pathways were strongly positively correlated with CRS score (Fig. 1h). However, in normal samples, the CRS score is only positively correlated with the estrogen signaling pathway and not with the ER pathway. This implies a special link between ER signaling and CR in tumors.

Functional and pathway annotation of CR clusters

To explore the potential biological behavior of two CR patterns, we determined 659 DEGs between two CR clusters using the R package “edgeR” (Fig. 2a). In total, 228 upregulated genes and 431 downregulated genes were obtained in the cluster-A group. Notably, the expression of immune checkpoints including PD1 (PDCD1; logFC = − 1.16; FDR < 0.0001) and CTLA4 (logFC = − 1.06; FDR < 0.0001) were upregulated in the cluster-B group. Although the FDR value of PD-L1 (CD274) was significant, its log2(fold change) value was below the threshold (logFC = − 0.44; FDR < 0.0001). The full result of differential gene expression analysis was extracted to perform GSEA analysis of KEGG and Hallmark50 signature using WebGestalt (Fig. 2b–c). A normalized enrichment score (NES) > 0 indicates that the gene set is enriched in cluster-A, while a score < 0 indicates that it is enriched in cluster-B. For pathways enriched in different CR subtypes (NES > 0/ < 0), if there are more than 20, the top 20 gene set items will be used for the visualization. Circadian rhythm, early/late estrogen response, glutamatergic synapse, and neuroactive ligand–receptor interaction were enriched in cluster-A group, while immune response-related pathways including inflammation, interferon alpha/gamma response, antigen processing and presentation, natural killer cell-mediated cytotoxicity, and cytokine–cytokine receptor signaling were observed in cluster-B group.

Fig. 2.

Differential gene expression analysis and functional enrichment between CR clusters in the TCGA cohort. a Volcano plot of differential expressed genes between cluster-A and cluster-B groups. b–c Bar plot display of GSEA analysis results of Hallmark50 and KEGG pathway. GSEA: gene set enrichment analysis; FDR: false discovery rate. d The bar plot shows the top 20 enriched GOBP terms from the ORA analysis in two CR clusters (ORA: overrepresentation analysis; GOBP: gene ontology biological process; Ratio: enrichment ratio)

For further investigation of the biological functional differences between the different CR subtypes, we entered the up- and downregulated genes separately and performed the ORA analysis on GO biological process (BP) using WebGestalt (Fig. 2d). The annotations for cluster-A focused on neurotransmitter transmission, synaptic transmission, and cell–cell signaling. In addition, there are some interesting terms which do not rank in the top 20, including axon development, rhythmic process, mammary gland development, drug transport, and regulation of response to drug. GO annotations for cluster-B include immune response, cytokine-mediated signaling pathway, response to cytokine, and leukocyte activation. These analyses revealed that cluster-B subtype was intimately linked to signatures of the immune system, while cluster-A was linked with synaptic signaling.

Distinct circadian subtypes are associated with the immune status within TME

After discovering the correlation between circadian subtypes and immunity, we began to consider whether circadian subtypes are related to TME immune status. ESTIMATE algorithm [16] showed that immune score and ESTIMATE score were relatively higher in cluster-B subtype, while stromal score was lower (Fig. 3a). These results indicate that the cluster-B subtype had higher levels of immune cell infiltration, lower tumor purity, and less stromal content. Furthermore, events in the TME of different CR clusters could be distinguished by “TME-associated signatures” (Fig. 3b). Cluster-B subtype was associated with higher DNA repair, antigen processing machinery (APM), CD8 T cells, immune signaling, and TME score, while cluster-A was associated with higher epithelial–mesenchymal transition (EMT). These findings suggest a relatively “hot” tumor immune microenvironment in the cluster-B subtype. Then we clustered samples from TCGA into "hot", "intermediate", and "cold" immunophenotypes and found a significantly higher proportion of “hot” immunophenotypes in cluster-B subtype (Fig. 3c–d).

Fig. 3.

Tumor immune microenvironment analysis in the TCGA cohort. a Differences in ESTIMATE score, immune score, and stromal score between the two CR clusters. b The CR clusters were distinguished by a variety of TME-associated signatures quantified by the ssGSEA algorithm (APM: antigen processing machinery; EMT: epithelial–mesenchymal transition). c BC samples clustered into distinct immunophenotypes by CD8⁺ T cell signature genes. d Stacked bar graph illustrating the percentages of immunophenotypes in the two CR clusters. e Differences in TMB between two CR clusters (TMB: tumor mutation burden). The y-axis shows the log10 tumor mutation burden at 50mb capture size (MB: Megabase). f–g Difference in distribution and mutation frequency of 10 genes with highest mutation rate in TCGA cohort between two CR clusters (ns, p > 0.05; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001)

Since tumor mutation burden (TMB) is a crucial factor in the process of tumor antigen presentation and is correlated with immunotherapy response, we analyzed the corresponding somatic mutation data of TCGA samples using the R package "maftools". We found that the cluster-B subtype had a higher mutation load (Fig. 3e) despite its higher DNA repair activity. There was also a significant difference in the frequency of mutations between the two subtypes (Fig. 3f-g). TP53, TTN, and HMCN1 are mainly mutated in the cluster-B group, while PIK3CA, CDH1, GATA3, MAP3K1, and KMT2C are mainly mutated in the cluster-A group. Notably, more than half of the cluster-B patients had mutations in TP53.

Distinct circadian subtypes are associated with immune cell infiltration

To obtain an immune infiltration landscape, we scored the degree of 22 infiltrating immune cells for each sample in the TCGA cohort using CIBERSORT algorithm [17]. The difference in median value of abundance (cluster-A minus cluster-B) was used to represent the disparity in immune cell infiltration between the two circadian subtypes (Fig. 4a). We observed that the infiltration by Tregs, CD8⁺ T cells, activated NK cells, and M1 macrophages was higher in the cluster-B group, while the infiltration of M2 macrophages was higher in the cluster-A group. Depending on their type, infiltrating immune cells can exert both pro- and anti-tumor effects. Among them, M2 macrophages and Tregs are recognized as tumor-promoting cells. Although Tregs were also higher in the cluster-B group, this may be due to the infiltration balance between Tregs and CD8⁺ T cells [26], which was supported by our observations (Fig. 4b). And there was no difference in the infiltration balance between the two circadian subtypes (Fig. 4c). In addition, it is worth noting that there was a significant difference in the polarization of two types of macrophages (M1 and M2) between the two circadian subtypes (Fig. 4d). The macrophage balance fractions (MBF: M2 / (M1 + M2)) [27] of cluster-B subtype were lower, i.e., polarization in favor of M1 (anti-tumor subtype). Next, we analyzed the differences in anti-tumor immune processes between the two groups based on the TIP seven-step cancer-immunity cycle (Fig. 4e). The CR cluster-B group exhibited stronger systematic anti-tumor immunity, including release and presentation of cancer cell antigens, priming and recruitment of immune cells, recognition of cancer cells by T cells, and killing of cancer cells.

Fig. 4.

Differences in immune cell infiltration between two circadian subtypes in the TCGA cohort. a Difference in the abundance of infiltrating immune cells between two CR clusters. The x-axis indicates the median of cluster-A minus the median of cluster-B. b Spearman correlation between CD8⁺ T cells and Tregs in BC tissues. c Boxplot illustrating the difference in the ratio of CD8⁺ T cells and Tregs between two CR clusters. d Boxplot illustrating the difference in MBF between two CR clusters. MBF: macrophage balance fraction. e Difference in the 7 steps of the cancer immunity cycle between two CR clusters (ns, p > 0.05; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001)

Cytokines and ligand–receptor pairs differ in different circadian subtypes

To further elucidate the cellular-level mechanism of tumor immune microenvironment differences between circadian patterns, we introduce the scRNA-seq dataset GSE176078 [28] for single-cell analysis. Patients were divided into two CR groups (high CRS and low CRS) by median partitioning based on their CRS scores calculated from the bulk RNA-seq data. Consistent with the previous analysis, the low-CRS group, which largely corresponded to cluster-B subtypes, had higher immune scores and ESTIMATE scores (Fig. 5a).

Fig. 5.

CR group partitioning and scRNA-seq analysis in the GSE176078 cohort. a Differences in ESTIMATE score, immune score, and stromal score between high- and low-CRS groups. b The t-SNE diagram with stacked bar graph depicts the landscape of various immune cell clusters and cancer cells from the two CR groups. c Upregulated cell–cell communication between macrophages, CD8⁺ T cells, and cancer epithelial under low-CRS group. L–R pairs: ligand–receptor pairs. d–e Bar graph of CTLA4, PDCD1, IFNG, and TGFB expression in different immune cell clusters (bars represent the mean + standard deviation) (ns, p > 0.05; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001)

Next, their corresponding scRNA-seq data were analyzed. T-distributed statistical neighbor embedding (tSNE) was applied to visualize immune cell clusters and cancer cells in different CR groups (Fig. 5b). We observed that the low-CRS group had more CD8⁺ T cells and macrophages, while the cancer epithelia were fewer. These three cell types were then subjected to cell–cell communication analysis using the CellChat [14] package (Fig. 5c). Ligand–receptor pairs (L–R pairs) with ligand.logFC > 0.3, receptor.logFC > 0.3, and p < 0.05 were regarded as significantly differentially expressed. A total of 26 differential L–R pairs were found, all of which were upregulated in the low-CRS group. We further visualized them at the molecular and pathway level, respectively. CD8⁺ T cells are the primary signal recipients, receiving MHC-I class antigens from tumor cells, as well as various chemokines and cytokines from other cells.

Consistent with the results of differential expression analysis in the TCGA cohort, we found that CTLA4 and PD1 (PDCD1) expressed by various T cell subtypes were generally upregulated in the low-CRS group (Fig. 5d). Interestingly, CTLA4 was less expressed by cycling T cells in the low-CRS group, suggesting that the proliferated and activated T cells were less restricted by CTLA4.

In order to explore the mechanisms by which macrophage polarization varies between the different circadian subtypes, we analyzed the L–R pairs that bind to macrophages (Figure S2). In addition to IFNG-related L–R binding, TGFB was upregulated in the low-CRS group, although less probably to communicate than IFNG. We observed similar results in the gene expression analysis, where the expression of TGFB and IFNG was generally upregulated in the immune cells of the low-CRS group (Fig. 5e).

Clinical significance of circadian subtypes and ability to predict therapeutic response

Survival analysis for CR clusters was performed using Kaplan–Meier curves in the TCGA cohort. Although there was no significant difference in the prognosis of different CR subtypes in all samples, the cluster-B group showed better overall survival in ER-positive and Lum A subtypes (Fig. 6a–c).

Fig. 6.

Immunotherapeutic and clinical significance of circadian subtypes. a–c Kaplan–Meier curves showing overall survival in cluster-A and cluster-B circadian subtypes in the TCGA cohort (log-rank test). d IPS score between two CR groups. e Relative distribution of CRS score between TIDE prediction in the TCGA cohort. f–g Stacked bar graph illustrating the percentages of immunophenotypes and immunotherapy responses in the high- and low-CRS groups for the IMvigor210 cohort (PD = progressive disease; SD = stable disease; PR = partial response; CR = complete response). h Kaplan–Meier curves showing overall survival after anti-PD1 therapy in high- and low-CRS groups in the GSE78220 cohort (log-rank test). i Correlation between CRS score and drug sensitivity in the TCGA cohort evaluated by the spearman analysis (**p < 0.01; ****p < 0.0001)

To date, the identification of predictive indicators has been critical to the development of immunotherapy strategies. In this study, we want to investigate whether circadian subtypes could predict the immunotherapy benefits of BC patients. The IPS (Fig. 6d) and TIDE (Fig. 6e) algorithms are widely used to assess response to immunotherapy. Our analysis results indicated that patients in the cluster-B group were more likely to benefit from immunotherapy, while patients with higher CRS scores were more likely to be non-responders to ICB therapy. Since there is a lack of published datasets for BC patients receiving immunotherapy, we used the immunotherapy cohort (IMvigor210: anti-PD-L1; GSE78220: anti-PD1) from other cancer types for instead. After the median partitioning, the samples were divided into high- and low-CRS groups based on CRS scores. We observed that the distribution of immunophenotypes in different CR groups was consistent in IMvigor210 samples and TCGA samples. The low-CRS subtype had a higher proportion of “inflamed” (equivalent to “hot”) and a lower proportion of “desert” (equivalent to "cold") immunophenotypes (Fig. 6f). Moreover, in both datasets, we observed that the low-CRS group was better able to benefit from immunotherapy (Fig. 6g–h).

To further understand the impact of CR patterns on chemotherapy response, we evaluated the correlation between the CRS score and estimated drug sensitivity score. The higher the drug sensitivity score, the higher the IC50 of the drug, which means it is less sensitive. Of the 287 drugs treated on BC cell lines (GDSC1000 resources), 196 significantly correlated pairs were identified (Fig. 6i). Among them, 180 drugs showed drug resistance (Rs > 0) correlated with the CRS score, while the other 16 drugs showed drug sensitivity (Rs < 0). This implies that patients with lower CRS scores, corresponding to the cluster-B subtype, are more sensitive to most chemotherapy drugs.

Correlation between CD8⁺ T cell infiltration and CR proteins expression in BC tissue

Since CRY1, PER2, and CLOCK are generally recognized as core CR genes and are included in the CRS, we selected them for verification in BC pathology samples. CD8⁺ T cell infiltration was represented by CD8 antigen expression. The expression of the different proteins was determined by IHC and shown as brown staining (Fig. 7a). The relative expression level of proteins was represented by the mean density. When analyzing individual CR proteins, only CRY1 showed a significant correlation with CD8 expression, and samples with low levels of CRY1 had higher levels of CD8 antigen (Fig. 7b–c). Related research by Lesicka et al. [29] showed that of 10 circadian genes, only CRY1 was upregulated in advanced tumors. This suggests that immune mechanisms may mediate this distinctive association between CRY1 and tumor progression. Given that the CR genes were considered as an integrated cluster in this study, we then divide the patients into two CR groups based on the expression of the three CR proteins (Fig. 7d). Among three CR protein types, patients with high expression of at least two of them are defined as the high-CR-protein group (n = 5) and others as the low-CR-protein group (n = 6). Surprisingly, a significantly higher level of CD8 expression was observed in patients in the low-CR-protein group.

Fig. 7.

CR protein (CRY1, PER2, CLOCK) and CD8 antigen expression in BC tissue. a Representative IHC staining of various proteins in human BC (scale bars: 100 μm). b Spearman correlation analysis between the expression of three CR proteins and CD8 antigen. c CD8 antigen expression was compared under different expression levels of 3 CR proteins (bars represent the mean + standard deviation). d Boxplot shows the distribution of different levels of CD8 expression under different CR groups (ns, p > 0.05; **p < 0.01)

Discussion

Growing evidence suggests that CR play an indispensable role in inflammation, anti-tumor activity, and TME remodeling [6, 8, 30]. The interrelationships and collaborative functions of multiple types of CR genes in cancer are not fully understood as most studies have focused on a single type of CR gene. Here, based on 28 CR genes, we provide a new CR-related typology in BC with relevance to the tumor immune microenvironment and therapeutic response. To quantify CR activity and evaluate CR subtypes in individual patients, we constructed the CRS, as the expression of the CRS gene can largely discriminate the CR cluster of TCGA patients. The cluster-B subtype (low CRS) was found to be associated with a more immunoreactive TME characterized by highly enriched immune-related pathways, high immune cell infiltration, high immune cell communication, and high anti-tumor immunity. On the contrary, the cluster-A subtype (high CRS) exhibits immunosuppressive TME. In addition, cluster-A may also be associated with a higher risk of metastasis due to high EMT activity, highly enriched synaptic transmission, and neurotransmitter release pathways, including glutamate signaling, which has been reported to be associated with breast-to-brain metastasis [31].

Although the CR pathway includes positive and negative arms, the changes in these two feedback directions are not opposite in BC. The consistent up- or downregulation of the CR genes, i.e., there is a general positive correlation, can also be observed in gliomas [32]. CR often play the role of tumor suppressors in cancer [30] and tend to be downregulated [9]; thus, it is increasingly hypothesized that enhancing CR in cancer cells may effectively control tumor progression [33, 34]. However, two CR subtypes identified in this study showed significant differences in CR gene expression, suggesting the diversity of CR among individual BC patients. Beyond that, the cluster-A subtype (CR enhanced) showed higher circadian activity than the normal samples, while the cluster-B subtype (CR reduced) was lower than that of the normal samples. These findings provide a clearer interpretation of two CR patterns and suggest that circadian enhancement may not be an appropriate choice for some patients.

One potential mechanism to explain CR alterations in BC may be the regulation of estrogen. Studies have shown that CRY2, PER1, and PER2 genes are downregulated in ER/PR-negative breast tumors compared to ER/PR-positive tumors [29]. Among them, PER2 has been reported to be able to link the circadian system and the ER [35]. In this study, besides the early/late estrogen response hallmark enriched in the cluster-A subtype, we also found a strong positive correlation between estrogen-related pathways and CRS. Intriguingly, compared to the estrogen signaling pathway, the correlation of the ER pathway was significant only in tumor tissue, not in normal tissue. These findings suggest that estrogen may mediate CR intertumoral heterogeneity in BC and mainly through ER, rather than GPR30 receptor or other pathways. Meanwhile, this ER-mediated CR heterogeneity could be the reason why CR subtyping has prognostic significance only in ER-positive patients. For the ER-positive subset of BC, it seems possible to target the CR to aid hormone therapy. Melatonin is one such potential candidate that can inhibit ER signaling and improve sensitivity to hormone therapy [36–38].

In healthy tissues, the immune system is thought to be under the control of CR [6], whereas this is largely unknown in TME. For example, in normal breast epithelium, ectopic expression of PER2 can inhibit the function of pro-inflammatory factors [39], but the existence of such inhibition after tumorigenesis is unclear. An encouraging finding in this report was that communication between immune cells within the TME is still associated with CR status. Many L–R interactions related to function of CD8⁺ T cells were upregulated in the low-CRS group, such as the CXCL9/10—CXCR3 axis and LCK activation. CD8⁺ T cells in the low-CRS group could also receive more MHC-I antigens from tumor cells, which is associated with the identification and elimination of transformed cells [40]. In addition, the modulation of macrophage polarization by other immune cells is also influenced by CR status. Most of the differential immune signals targeting macrophages were upregulated in the low-CRS group, with IFN-γ (pro-inflammatory) upregulated relatively higher than TGF-β (anti-inflammatory), which could lead to M1-favoring polarization. Consistently, we observed a shift in the polarization balance in the TCGA cohort, with a lower proportion of M2 macrophages in the cluster-B group (low CRS).

Immune checkpoints, a set of self-tolerance mechanisms normally used to protect the body from harmful immune responses, have been reported to be engaged in tumor immune escape [41]. The results of treatment with ICB in solid tumors have ushered in the era of modern immunotherapy, but effectively assessing whether patients can benefit from immunotherapy has become a barrier to clinical application. Since neo-antigens resulting from mutations could be presented by MHC proteins and recognized by CD8⁺ T cells, a patient with a higher TMB and an immunologically hot tumor is more likely to benefit from ICB [42, 43], while transcriptional inhibition of MHC I antigens contributes to immunotherapy resistance [44]. Our observations are entirely consistent with this hypothesis. BC patients with lower CRS scores tended to have a "hot" immunophenotype, higher MHC I expression, higher TMB, and were evaluated (IPS and TIDE algorithms) to have a better response to ICB treatment. In other cancer cohorts treated with ICB (imvigor210 and GSE78220), the low-CRS group also showed a better response. In addition to immunotherapy, the low-CRS samples was more sensitive to chemo drugs, the underlying mechanism could be a low level of BMAL1/CLOCK, which reduces the activation of the Nrf2 pathway-mediated drug resistance [45].

There are several limitations to this study, the most prominent being the unavailability of time points for collecting biospecimens from cancer patients. Given that CR may not be disrupted in all cancer patients, transcriptome data may still fluctuate rhythmically over the 24-h period in a large number of patients. Therefore, it is necessary to use a sizable sample to average out the biases caused by sampling at different time points and to understand the general role of CR genes in cancer. In addition, the validation experiment is limited by the sample size. Although there exists some statistical significance, the relationship between CR patterns and tumor immune microenvironment requires further experiments for validation.

Conclusion

In summary, our systematic, comprehensive analysis of CR patterns revealed an extensive regulatory mechanism around TME remodeling and the crucial clinical implications of circadian cross talk. This study also provides new insights into the assessment of individual CR status and helps to develop personalized chronotherapy strategies to promote the use of ICB in BC patients. Currently, the immunotherapy targets in BC are mainly ER-negative patients, including HER2 and TNBC subtypes. Given the strong positive correlation between the CR signature and the ER pathway, the combination of drugs targeting estrogen and CR may help ER-positive patients create a more suitable TME for immunotherapy.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

BC

Breast cancer

CR

Circadian rhythm

CRS

Circadian rhythm signature

ER

Estrogen receptor

ICB

Immune checkpoint blockade

IPS

Immunophenoscore

TIDE

Tumor Immune Dysfunction and Evolution

TME

Tumor microenvironment

TMB

Tumor mutation burden

Authors’ contributions

SX performed the bioinformatic analysis, statistical analysis, and drafting of the manuscript. WZ performed the experiment. LW, TZ, WW, OZ, and XX contributed to the investigation, methodology, and data curation. ZL and DL conceived of the study, participated in its design, and revised it substantially. All authors read and approved the final manuscript.

Funding

This research was funded by the National Natural Science Foundation of China, grant number 81560464 and 31960152 (D.L.).

Availability of data and materials

All online data and tools described in this article are available from their web servers and are free for any scientist to use for non-commercial purposes. Further information and source code are available from the corresponding author upon reasonable request.

Declarations

Conflict of interest

The authors declare that there are no conflicts of interest.

Ethics approval and consent to participate

This study was approved by the institutional review board of the First Affiliated Hospital of Nanchang University, and written informed consent was obtained from all patients.