We first assessed the expression of STING in normal tissues using the Human Protein Atlas (HPA) database, and found that STING was expressed at high levels in tissues of the respiratory system, bone marrow and lymphoid tissues, as well as tissues of the breast and female reproductive system ( Figure 1A ). We also explored the onset of STING expression in the blood cells, and STING was intensely expressed in CD4+ T cells followed by hematopoietic stem cells (HSCs) and T/NK cells according to the Hemap database ( Figure 1B ). CCLE database analysis showed that STING was relatively highly expressed in acute myeloid leukemia (AML) and chronic myelogenous leukemia (CML) relative to most cancer cell lines ( Figure 1C ). Subsequently, we collected 16 leukemia cell lines in our laboratory, extracted RNA, and assessed the expression levels of STING using Real-time quantitative PCR (RT-qPCR). Our analysis revealed a significantly elevated expression of STING in the THP-1, NOMO-1, and K562 cell lines ( Figure 1D ). Next, Hemap database analysis confirmed that STING was clearly overexpressed in AML compared with that in other hematological malignancies ( Figure 1E ). Moreover, pan-cancer analysis based on the TCGA and GTEx databases showed that STING mRNA was significantly differentially expressed in both normal and tumor tissues in 25 out of 29 tumor types, including AML ( Figure 1F ). Furthermore, STING was differentially expressed across hematological malignancies, with diffuse large B-cell lymphoma (DLBCL) and AML showing the most prominent upregulation ( Figure 1G ). In AML, patients with high STING expression harbored more FLT3 and DNMT3A mutations (P = 0.038 and P = 0.047, respectively) but fewer RUNX1 and TP53 mutations (P = 0.048 and P = 0.018, respectively) than those with low STING expression ( Supplementary Figure S1 ).

This study investigated the relationship between STING expression and prognosis of patients with hematological malignancies. As shown in Figures 2A–C , a meta-analysis of Cox regression values (P value and hazard ratio (HR)) showed that STING is a low-risk gene for DLBCL ( Figure 2A ) with HR < 1 in GSE11318 (P < 0.0001), GSE117556 (P < 0.0001), GSE181063 (P < 0.0001), GSE23501 (P = 0.038), GSE32918 (P < 0.0001), GSE34171 GPL570 (P = 0.011), GSE69053 GPL14951 (P < 0.0001), GSE69053 GPL8432 (P = 0.001), GSE87371 (P = 0.034), GSE98588 (P = 0.002), NCICCR DLBCL (P = 0.006), and Reddy DLBCL (P < 0.0001). STING is a high-risk gene for AML ( Figure 2B ) with HR > 1 in Bullinger AML (P = 0.017), GSE10358 (P < 0.0001), GSE106291 (P = 0.025), GSE12417 GPL570 (P = 0.003), GSE1427 GPL97 (P = 0.05), GSE37642 GPL570 (P = 0.009), GSE6891 (P = 0.002), GSE71014 (P < 0.001), and TCGA AML (P = 0.026). And STING is a high-risk gene for multiple myeloma (MM) ( Figure 2C ) with HR > 1 in CoMMpass (P =0.012) and GSE57317 (P = 0.002). Afterward, we specifically chose the GSE10358 ( Figures 2D–G ) and GSE6891 ( Figures 2H–K ) datasets for Kaplan-Meier analysis due to their relatively large sample sizes, inclusion of both overall survival (OS) and event-free survival (EFS) data, and availability of cytogenetic information. These features allowed us to validate the prognostic value of STING for both OS and EFS, as well as in cytogenetically normal AML subgroups. Kaplan-Meier survival analysis revealed higher STING expression correlated with poorer EFS and OS in both the whole and NK subgroups ( Figures 2D–K ).

We further investigated the biological characteristics of STING in patients with hematological malignancies using the GSVA package ( Figure 3A , Supplementary Table S1 ). In general, STING expression was positively correlated with immune-related pathways such as the interferon-gamma (IFN-γ) response, IFN-alpha response, inflammatory response, JAK/STAT3 signaling, complement, and allograft-rejection pathways. In myelodysplastic syndrome (MDS), STING expression was positively correlated with TNF-α signaling via NFKB (r = 2.18, P < 0.001), allograft rejection (r = 2.13, P < 0.001), inflammatory response (r = 1.96, P < 0.001), interferon alpha response (r = 1.80, P < 0.001), and interferon-gamma response (r = 1.71, P < 0.001). In addition, STING expression was significantly negatively associated with E2F targets (r = −2.86, P < 0.001), MYC targets V1 (r = −2.57, P < 0.001), G2M checkpoint (r = −2.51, P < 0.001), and MYC targets V2 (r = −2.28, P < 0.001). In AML, STING expression was positively correlated with allograft rejection (r = 1.35, P < 0.05) and negatively associated with HEME metabolism (r = −2.72, P < 0.001), and TNF-α signaling via NFKB (r = −2.34, P < 0.001). Furthermore, in DLBCL, there was a significant positive correlation between STING expression and epithelial-mesenchymal transition (r = 3.02, P < 0.001), TNF-α signaling via NFKB (r = 2.82, P < 0.001), coagulation (r = 2.77, P < 0.001), inflammatory response (r = 2.70, P < 0.001), and interferon-gamma response (r = 2.66, P < 0.001). A significant negative correlation was observed between STING and E2F targets (r = −2.93, P < 0.001), MYC targets V1 (r = −2.62, P < 0.001), G2M checkpoint (r = −2.57, P < 0.001), and MYC targets V2 (r = -2.16, P < 0.001). To further validate the above findings in AML, we performed GSVA enrichment analysis using scRNA-seq data from AML (GSE116256). The bar plot showing the first 50 GSVA results indicated that STING expression was positively correlated with oxidative phosphorylation and MYC targets V1, and negatively correlated with KRAS signaling, apical surface, and myogenesis ( Figure 3B ).

A total of 29 TME-related signatures representing the stromal compartments (e.g., angiogenesis and cancer-associated fibroblasts), pro-tumor microenvironment via macrophages and MDSCs, anti-tumor microenvironment (e.g., B cells and T cells), and cancer cell properties (e.g., EMT signature) were selected by Bagaev et al. (22). The four microenvironment subtypes were conserved across cancer types and could predict response to immunotherapy (22). We explored the association between STING expression and the enrichment of 29 TME-related signatures across hematological malignancies ( Figure 4A , Supplementary Table S2 ). STING expression was significantly correlated with 29 TME signatures. In chronic lymphocytic leukemia (CLL), STING expression was negatively correlated with stromal-related signatures such as endothelium (r = −0.52, P < 0.001), cancer-associated fibroblast (CAF) (r = −0.58, P < 0.001), matrix (r = −0.67, P < 0.0001), matrix remodeling (r = −0.56, P < 0.0001), and protumor cytokines (r = −0.63, P < 0.0001). Also, STING expression was positively correlated with macrophages (r = 0.43, P < 0.0001), MDSC traffic (r = 0.27, P < 0.0001), and MDSC (r = 0.33, P < 0.0001). Besides, STING expression was significantly inversely correlated with anti-tumor signatures such as B cells, MHC I, EMT signature, and proliferation rate. Still, it positively correlated with macrophages, in the majority of hematologic malignancies. Thus, correlations between STING expression levels and TME in blood cancers play an important role in tumor growth and survival. Further, CIBERSORT and MCP-counter algorithms were used to analyze the correlation between STING expression levels and the composition of tumor-associated immune cells in diverse blood cancers. Consistent with the 29 TME-related signatures, STING expression was positively correlated with both M1 and M2 macrophages in most hematological malignancies ( Figure 4B ). However, in myeloid cancers such as MDS, CML, and CLL, STING expression was negatively correlated with M2 macrophages ( Figure 4B ), suggesting a potential context-dependent role of STING in shaping the immune microenvironment. These findings underscore the complex interactions between STING expression and macrophage subtypes across different hematological malignancies. Additionally, STING expression was negatively associated with B cell enrichment in the tumor microenvironment ( Figure 4B ). In addition to controlling tumor growth through their effects on the immune microenvironment, B cells can enhance the function of T cells by providing a more conducive environment for their activities (23). The relationship between STING expression and 72 immune genes was also analyzed ( Figure 4C ). STING expression was significantly positively related to that of inhibitory immune genes, including c10orf54 and TGFB1, and negatively related to stimulatory immune genes such as CD40 and HMGB1. In addition, STING expression was positively related to that of stimulatory immune genes such as ITGB2 and PRF1 in most hematological tumors. This finding needs further exploration and confirmation.

Afterward, the relationship between STING expression and the immunological microenvironment was investigated. The cytolytic score, which serves as a biomarker of anti-tumor immune activity, is an immune activity score estimated by the average gene expression of GZMA and PRF1 (24). In this study, cytolytic score and STING expression were positively correlated in some hematological tumors, especially CLL and Mantle Cell Lymphoma (MCL) ( Figure 4D ).

HLA class I antigens are crucial for the activation of cytotoxic CD8 +T cells. HLA class II molecules present antigenic peptides to CD4+ T cells, which are critical for antitumor immunity (25). We found that STING expression was positively correlated with HLA I in most blood tumor types, the most prominent being DLBCL, Classical Hodgkin lymphoma (CHL), and mucosa-associated lymphoid tissue (MALT) lymphoma ( Figure 4E ). Furthermore, STING expression was positively correlated with HLA II in MALT, MDS, peripheral T-cell lymphoma-not otherwise specified (PTCLNOS), T-cell acute lymphoblastic leukemia (T-ALL), and DLBCL ( Figure 4F ).

Then, ESTIMATE analysis was performed to determine the association of STING expression with immune scores, stromal scores, and tumor purity. MCL had the highest immune score and stromal score with the lowest tumor purity, while anaplastic large-cell lymphoma (ALCL) had the lowest immune score and stromal score with the highest tumor purity ( Figures 4G–I ).

The above results indicate that STING influences patient prognosis by potentially affecting the activity of immune cells and the immune microenvironment in blood tumors. This observation requires further exploration.

Several studies have reported that patients may benefit from CAR-T therapy, immune checkpoint inhibitors (ICI), and tumor antigen-related vaccines (26–29). Accordingly, we predicted the likelihood of a response to immunotherapy in different STING expression levels using the TIDE algorithm. In the BeatAML database, the results showed a strong negative correlation between STING expression and IFN-γ, Merck18 (T-cell-inflamed signature), CD274, CD8, and T-cell dysfunction-score signatures, whereas a strong positive correlation was observed between STING expression and T-cell exclusion signatures, including exclusion, and the M2 subtype of tumor-associated macrophages (TAMs) ( Figure 5A ). In addition, according to the TCGA AML database, STING expression was negatively correlated with biomarkers of TIDE score and CAF levels ( Figure 5B ). A low TIDE score indicates a lower likelihood of immune evasion, which indicates more benefit from ICI treatment for the patient (30). However, there was no difference in TIDE score, IFN-γ, MSI, Merck18, CD274, CD8, dysfunction, exclusion, MDSC, CAF, and M2 between STING low-expression and STING high-expression groups in the GSE58831 MDS dataset ( Figure 5C ). Furthermore, different results were found in GSE22762 CLL, and STING expression was significantly positively related to IFN-γ, MSI, Merck18, CD8, and T-cell dysfunction and significantly negatively related to T-cell exclusion, MDSC, and TAM M2 ( Figure 5D ).

We further examined the relevance of STING expression in relation to therapeutic response to potential ICB. Excitingly, significant differences in STING expression were observed between the responder and non-responder groups across AML, MDS, and CLL ( Figures 5E–H ). These findings indirectly indicate that STING expression plays a critical role in mediating the immunotherapy response, and patients with high STING expression respond better to immunotherapy.

The LSC17 score had strong prognostic value and contributed to the accurate prediction of initial therapy resistance (31). STING expression was positively related to the LSC17 score in CLL, DLBCL, follicular lymphoma (FL), MCL, MM, and AML, but negatively related to the LSC17 score in CML ( Figure 6A ). Patients with higher LSC17 scores had stronger myelodysplasia, and higher recurrence rates, and experienced lower chemotherapeutic effectiveness (31).

To depict the stemness of STING expression in hematological tumors, the mRNA expression-based stemness index (mRNAsi) was evaluated. Significant negative correlations between mRNAsi and STING gene expression levels were found in AML, CHL, CLL, CML, DLBCL, MCL, MDS, MM, pre-B cell acute lymphoblastic leukemias (pre-B-ALL), PTCLNOS, T-cell acute lymphocytic leukemia (T-ALL), and angioimmunoblastic T-cell lymphoma (AITL) ( Figure 6B ).

Using four independent ICB cohorts from four cancer types ( Figures 6C–F ), Kaplan-Meier survival analysis was performed. The MetastaticMelanoma-Liu_NatMed_2019 cohort included anti-PD1-treated metastatic melanoma patients; the AdvancedMelanoma-Snyder_NEJM_2014 cohort included advanced melanoma patients treated with anti-CTLA-4 therapy; the KidneyCancer-Miao_Science_2018 cohort consisted of kidney cancer patients treated with PD-1 or PD-L1 blockade therapy alone or in combination with anti-CTLA-4 therapy; and the MetastaticMelanoma-VanAllen_Science_2015 cohort included anti-CTLA-4-treated metastatic melanoma patients. Results showed that low STING expression was correlated with poor progression-free survival (PFS) and OS across these cohorts, suggesting that STING expression may serve as a biomarker of ICB response in solid tumors.

Data on STING expression in normal tissues were obtained from the HPA database (https://www.proteinatlas.org/). We also analyzed the expression of STING in blood cell types and hematologic cancers by using data from the Hemap dataset (http://hemap.uta.fi/). Levels of STING expression were analyzed across cancer cell lines from the Cancer Cell Line Encyclopedia (https://www.broadinstitute.org/ccle). The combined pan-cancer data for the STING gene were sourced from the TCGA, TARGET, and GTEx downloaded from the UCSC Xena Browser (https://xenabrowser.net). To confirm whether STING expression differs between blood tumors and normal samples, we utilized the Pan-Hem-Diff cohort containing 22 hematologic cancer types with matched tumor and normal samples. Survival information for patients with AML, DLBCL, and MM was downloaded from datasets containing survival information, including the GEO database (https://www.ncbi.nlm.nih.gov/geo/), cBioPortal for Cancer Genomics (http://www.cbioportal.org/), GDC data portal (https://portal.gdc.cancer.gov/), and PREdiction of Clinical Outcomes from Genomic Profiles (PRECOG, https://precog.stanford.edu/). The STING gene mutation data was also obtained from the cBioPortal for Cancer Genomics (http://www.cbioportal.org/). Our analyses covered major subtypes of myeloid leukemias, myelomas, B and T cell lymphomas, and lymphoid leukemias, including AML, AITL, ALCL, T-ALL, CML, CLL, CHL, DLBCL, FL, LCH, MM, MDS, MCL, MALT, PTCLNOS, and pre-B-ALL.

Total RNA was extracted from cell lines by using Trizol reagent (Invitrogen, USA). Reverse transcription of RNA was performed to generate cDNA with a Reverse Transcription Kit (Takara, Japan). RT-qPCR was conducted to determine the expression levels of STING in the 16 leukemia cell lines from our laboratory using SYBR Green Master Mix (Takara, Japan). The upstream and downstream primer sequences were 5’- CACTTGGATGCTTGCCCTC-3’ and 5’- GCCACGTTGAAATTCCCTTTTT-3’ respectively. GAPDH was used as the internal reference gene, and STING transcript levels were calculated using the 2^−ΔΔCT method.

Patients were divided into STING high and low expression groups based on the best cut-off value of STING expression. The ‘best’ cut-off value for Kaplan-Meier analysis was determined using the maxstat method, implemented in the survminer R package. This approach identifies the cut-off that maximizes the log-rank statistic, providing the most statistically significant division of the cohort into high and low-expression groups. Univariate Cox regression analysis was carried out to examine the prognostic value of STING in DLBCL, AML, and MM. We used the “survcomp” package to conduct meta-analyses and combine P values and HR. Genes with HR < 1 predicted better OS, while genes with HR > 1 indicated dismal OS. Kaplan–Meier analysis was utilized to exhibit the correlation of STING expression with OS, PFS, and EFS in patients. Kaplan–Meier survival curves were used to visualize survival outcomes between groups using the “survminer” package. Differential gene expression analyses were conducted and signature score data was obtained using the “limma” R package.

We applied more stringent criteria for dichotomizing the expression of STING across the different blood cancer types to provide more meaningful biological interpretations. Differentially expressed genes (DEGs) were extracted between high STING expression (top 30%) and low STING expression (bottom 30%) groups for each blood cancer type in the Hemap dataset. Then, DEGs were subjected to gene set enrichment analysis (GSEA) using the R package clusterProfiler. Further, we selected hallmark gene sets for functional analysis of GSEA from MSigDB (http://www.broad.mit.edu/gsea/msigdb/). The signature scores in the GSE116256 scRNA-seq dataset were computed using the GSVA package.

According to Bagaev et al., 29 functional gene expression signatures were used to describe pan‐cancer TME characteristics and categorize the patterns into stromal compartments, pro-tumor microenvironment, anti-tumor microenvironment, and cancer cell properties (22). Single-sample gene set enrichment analysis (ssGSEA) was used to calculate the signature scores for the tumor samples in Hemap via the “GSVA” R packages. The CIBERSORT and MCP algorithms were used to quantify the immune cell composition in different blood cancers (58). We used computational algorithms (e.g., CIBERSORT, MCP-counter) to infer immune cell proportions in hematological malignancies based on gene expression data, distinguishing this analysis from immune infiltration typically studied in solid tumors. Immune cell profiling in our study reflects cellular enrichment and relative proportions derived from transcriptomic data rather than physical infiltration into a dense tissue matrix. As part of this analysis, the cytolytic scores and HLA scores were analyzed for the Hemap dataset, as reported previously (59). Then, we calculated the immune scores, stromal scores, and tumor purity for each type of blood cancer using the ESTIMATE algorithm.

ICB response, immune dysfunction, and exclusion status were predicted using the TIDE algorithm (http://tide.dfci.harvard.edu). We also analyzed the correlations between STING expression levels and programmed cell death 1/programmed cell death ligand 1 (PD-1/PD‐L1) in AML (BeatAML and TCGA), MDS (GSE58831) and CLL (GSE22762). The LSC17 score was calculated for each blood cancer based on previous studies (60, 61). The stemness index mRNAsi was determined from a previous study (https://bioinformaticsfmrp.github.io/PanCanStem_Web/) (62). To predict patients’ response to ICB therapy in real-world immunotherapy settings, we enrolled four independent immunotherapy cohorts covering three cancer types (metastatic melanoma, advanced melanoma, and kidney cancer) to determine whether there were differences in ICB therapy between high and low STING expression groups.

R software was employed for statistical analyses. The correlation between two continuous variables was determined using a non-parametric Spearman test. Categorical data were compared using the chi-square test or Fisher’s exact test. Data visualization was performed using the R packages “ggplot2” and survcomp package (for forest plots). P < 0.05 was considered to indicate significance.