SPP1 macrophage-induced T-cell stress promotes colon cancer liver metastasis through SPP1/CD44/PI3K/AKT signaling.

Background
Colon cancer (CC) is one of the most prevalent malignancies of the digestive system, with the majority of patients diagnosed at advanced stages, often accompanied by synchronous lymph node or distant organ metastasis.1 Colon cancer liver metastases (CCLM) represent the primary cause of mortality among patients with CC.2 Although immunotherapy has demonstrated promising results in patients with CC, its efficacy in patients with CCLM remains suboptimal.3 The intricate interactions among various cellular components within liver metastatic lesions play a critical role in establishing an immunosuppressive microenvironment and facilitating the formation of the pre-metastatic niche.4 Consequently, strategies aimed at targeting immune components within liver metastases (LM) to reverse the immunosuppressive microenvironment have emerged as a pivotal approach in the current treatment of CCLM.
Immune cells constitute the fundamental basis of tumor immunotherapy.5 Tumor-associated macrophages (TAMs) are among the most abundant immune cell populations within the tumor immune microenvironment (TIME) and play dual roles in both promoting tumor progression and mediating antitumor immunity.6 7 Secreted phosphoprotein 1 (SPP1), also known as osteopontin, is a multifunctional, secreted, and phosphorylated glycoprotein initially identified in the bone matrix. It plays a critical role in osteoblast differentiation, bone formation, and bone resorption.810 In a groundbreaking study published in Science, Pittet et al11 employed single-cell RNA sequencing (scRNA-seq) technology to uncover a novel pattern of macrophage polarization characterized by the CXCL9:SPP1 axis. The expression levels of CXCL9 and SPP1 in macrophages exhibit an antagonistic relationship, with elevated SPP1 expression in macrophages correlating with poor clinical outcomes in patients.12 Recent studies have further reported that SPP1 is specifically upregulated in macrophages within colorectal cancer (CRC) LM lesions.13
Tumor-infiltrating lymphocytes (TILs) are a pivotal component of the TIME and have demonstrated substantial antitumor efficacy in various therapeutic modalities, including chimeric antigen receptor T-cell therapy and immune checkpoint blockade (ICB) therapy.14 However, the phenotypic and functional heterogeneity of TILs significantly impacts the effectiveness of antitumor therapies and the likelihood of adverse side effects. In a landmark study, Wang et al15 employed scRNA-seq technology to construct a comprehensive pan-cancer atlas of T cells, identifying a distinct T-cell stress response state (TSTR). TSTR is characterized by the upregulation of heat shock protein genes and is frequently enriched in metastatic tumors.15 Notably, in patients with a low response rate to ICB therapy, the proportion of TSTR cells within CD4+ and CD8+ T-cell populations is significantly elevated, suggesting that TSTR cells may contribute to resistance to immunotherapy.16 Previous studies have shown that SPP1-expressing macrophages can induce T cells to adopt an exhausted phenotype.17 Furthermore, the crosstalk between SPP1-expressing macrophages and T cells plays a critical role in establishing the immunosuppressive TIME of gastric cancer LM.18 However, whether there is any interaction between SPP1+ macrophages and TSTR cells remains unexplored and warrants further investigation.
In this study, we conducted scRNA-seq analysis on primary CC tissues, matched LM, and adjacent normal tissues from four untreated patients with CCLM. This approach enabled us to comprehensively delineate the single-cell landscape across different lesion types. Additionally, we systematically identified subgroups of T cells and myeloid cells through differential gene expression analysis, and elucidated the distinct transcriptional profiles of these cell subgroups in LM compared with primary lesions. In our study, we identified a TSTR, which is distinct from exhausted T cells and is significantly enriched in LM. We further revealed that Macro_SPP1high macrophages drive T cells into a stress response state through the SPP1/CD44/PI3K/AKT signaling pathway, thereby contributing to the progression of CCLM. Targeting this pivotal signaling axis and alleviating the dysfunctional stress state of T cells could offer a promising therapeutic approach to suppress CCLM progression.

Methods

Human sample collection
At the Gastrointestinal Surgery Department of the Third Affiliated Hospital of Sun Yat-sen University between January 1, 2024, and December 31, 2024, we collected primary CC tissues, matched LM, and corresponding adjacent normal tissues from four newly diagnosed patients with CC presenting with synchronous CCLM who had not received immunotherapy as part of any neoadjuvant treatment regimen. Each tissue sample was divided into two aliquots: one aliquot was immediately placed in a pre-cooled phosphate-buffered saline (PBS) dish, where the tissue blocks were thoroughly washed with cold PBS solution to remove adipose tissue and visible blood vessels. The minced tissue was subsequently stored at 4°C and processed for single-cell sequencing analysis within 24 hours. The second aliquot was fixed and embedded in paraffin for subsequent immunofluorescent staining analysis. Notably, none of the enrolled patients had received preoperative radiochemotherapy.

Tissue dissociation
The tissue samples were retrieved from the 4°C refrigerator and subjected to centrifugation to remove the preservation solution, followed by two washes with Dulbecco’s Modified Eagle Medium. Using sterile surgical scalpels, the tissues were rapidly minced into 2–3 mm³ fragments. Tissue digestion was then performed at 37°C for 1 hour in Roswell Park Memorial Institute (RPMI) 1640 medium supplemented with 10% fetal bovine serum (FBS), 100 U/mL penicillin, and 100 mg/mL streptomycin, containing 0.38 mg/mL collagenase VIII and 0.1 mg/mL deoxyribonuclease I. The resulting cell suspension was sequentially filtered through a 40 µm nylon cell strainer and centrifuged at 300×g for 5 min. The freshly isolated single-cell suspension was subsequently prepared for scRNA-seq analysis.

Single-cell RNA sequencing
Single-cell suspensions were processed immediately following the manufacturer’s protocol using the 10x Chromium Single Cell 3' v3 Reagent Kit (10x Genomics, Pleasanton, California, USA). Library preparation was conducted according to the standard workflow, and subsequent sequencing was performed on an Illumina NovaSeq 6000 platform (Illumina, San Diego, California, USA) at BGI Genomics (Shenzhen, China).

Processing of scRNA-seq data
The raw sequencing data generated from scRNA-seq were processed using the standard Cell Ranger pipeline for sequencing analysis. This process yielded a standardized output file for each sample, comprising three distinct file types: barcodes, features, and matrix. We employed the lapply function to batch-process paired samples and the merge function to integrate data across multiple samples. Using the encapsulated functions available in the Seurat R package (V.4.3.0.1), we assembled the combined data into a Seurat object and applied stringent quality control filters to remove low-quality cells. Specifically, we required that each gene be expressed in at least three cells, with the number of detectable genes per cell ranging between 200 and 6,000. Only cells with unique molecular identifier counts between 1,000 and 25,000 were retained, and the proportion of mitochondrial genes was restricted to less than 15%. Cells that met these quality control criteria were subsequently used for downstream analysis. The NormalizeData, FindVariableFeatures, and ScaleData functions were used to standardize the quality-controlled Seurat objects, identify highly variable genes, and perform normalization, respectively. Additionally, we retained key environmental variables associated with the quality-controlled cells to facilitate subsequent downstream analyses.

Dimension reduction and unsupervised clustering
We loaded the rigorously filtered single-cell data and conducted principal component analysis (PCA) based on the top 2,000 highly variable genes identified by the FindVariableFeatures function. To address batch effects across samples, we employed the Harmony R package (V.1.2.0). By analyzing the elbow point on the PCA SD plot, we selected the top 30 principal components for subsequent clustering and grouping analyses. The clustering results were visualized using the Uniform Manifold Approximation and Projection (UMAP) algorithm. Additionally, we used loop statements to systematically output cell clustering results at varying resolutions.

Identification of cellular clusters and extraction of subclusters
We employed the FindAllMarkers function to identify marker genes that differentiate distinct cell clusters, extracting the top 20 genes ranked by expression level for each cluster. The expression profiles of these marker genes were validated using the Cell Taxonomy tool and cross-referenced with established cell population markers reported in prior literature. Furthermore, we quantified the cell counts for each population and visualized the distribution using a bar chart. To investigate the composition of the TIME in patients with CCLM, myeloid and T-cell subgroups were isolated, reconstructed into separate Seurat objects, and subjected to a sequential analytical workflow. This workflow included normalization, dimensionality reduction, clustering, subclustering, and cell type identification.

Differential expression analysis
We extracted distinct cell clusters and reconstructed Seurat objects to perform normalization, identify highly variable genes, and carry out subsequent normalization steps. The data were categorized into LM and primary CC lesions, and the FindMarkers function was applied to identify differentially expressed genes (DEGs) across cell clusters in these groups. Finally, we used the EnhancedVolcano R package to generate volcano plots, effectively visualizing the DEGs.

Construction of single-cell trajectories
To investigate the differentiation origins of various cell subgroups, we selected specific clusters of interest for trajectory analysis. Using the Monocle2 R package (V.2.26.0), we constructed a cellData object. The differentialGeneTest function was employed to identify DEGs among cells, which were then ranked by q value. The top 1,000 DEGs were used for cell ordering. Finally, we applied the DDRTree method from the reduceDimension function in Monocle2 to map the cells and visualize the developmental trajectories of the different cell types.

Isolation of murine bone marrow-derived macrophage
Balb/c mice aged 5–6 weeks were euthanized by cervical dislocation and sterilized with 75% ethanol. The hind limbs were dissected, and bone marrow was flushed out using cold PBS. The resulting cell suspension was filtered through a 70 µm cell strainer to obtain single-cell suspensions. Erythrocytes were removed by incubation with erythrocyte lysis buffer. The cells were then cultured in RPMI 1640 medium supplemented with 10% FBS and 50 ng/mL macrophage colony-stimulating factor (PeproTech) for 7 days to generate bone marrow-derived macrophages (BMDMs).

Induction of SPP1high macrophage from BMDM
BMDMs were cultured in 6-well plates and maintained at 37°C under either normoxic (21% O2 and 5% CO2) or hypoxic (1% O2, 5% CO2, and balanced N2) conditions for 72 hours. Following incubation, SPP1high and SPP1low macrophages were isolated for further analysis.

Identification of SPP1 protein expression
The expression of SPP1 in macrophages was assessed by western blot analysis. Total proteins were extracted on ice using radioimmunoprecipitation assay lysis buffer, and protein concentrations were quantified using a BCA Assay Kit (Thermo Fisher Scientific). Subsequently, 20 µg of protein was resolved by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE). The separated proteins were transferred onto polyvinylidene difluoride membranes and incubated overnight at 4°C with a primary anti-SPP1 antibody (Abcam). Following this, the membranes were incubated with a horseradish peroxidase-conjugated secondary antibody, and protein bands were visualized using enhanced chemiluminescence detection.

Development of a BMDM-T cell co-culture system
Spleen-derived T cells from BALB/C mice were isolated using Lymphocyte Separation Medium (DAKEWE, Germany). BMDMs and T cells were co-cultured in a 12-well plate at a 1:5 ratio in RPMI-1640 medium (Gibco, USA) supplemented with 10% FBS (Gibco, USA), and incubated at 37°C with 5% CO₂ for either 24 or 48 hours prior to analysis.

Flow cytometry analysis of T cells in co-culture system
Following 24 or 48 hours of co-culture, T cells were harvested and stained with anti-CD4-BV605 and anti-CD8-PE/Cy7 antibodies. For intracellular staining, cells were fixed and permeabilized using the Transcription Factor Buffer Set (BD Pharmingen, USA) and subsequently stained with anti-heat shock protein 70 (HSP70)-Alexa Fluor 488, anti-pAKT, anti-pPI3K, anti-AKT, or anti-PI3K antibodies. The proportions of CD4+ and CD8+ T cells, as well as the expression levels of HSP70, pAKT, AKT, pPI3K, and PI3K, were quantified using flow cytometry.

Establishment of a CT26 cell-derived CCLM model
Male BALB/c mice aged 4–5 weeks were purchased from Yancheng Biotechnology (Guangzhou, China) for the establishment of CCLM models. All animal experimental procedures were approved by the Institutional Animal Care and Use Committee of Sun Yat-sen University. Briefly, approximately 1×106 CT26-Luc cells suspended in 50 µL of PBS were slowly injected into the spleen of each mouse. After 5 min, the splenic blood vessels were ligated, and the spleen was surgically removed to complete the procedure.

Synergistic immunotherapy combining anti-SPP1 and anti-PD-1 antibodies
1 week after establishing the CCLM models, bioluminescence imaging was performed to evaluate model consistency. Based on luminescence intensity, the mice were randomly divided into four groups, with six mice per group: PBS, anti-SPP1 (aSPP1), anti-programmed cell death protein-1 (PD-1) (aPD-1), and aSPP1+aPD-1 combination therapy. Therapeutic antibodies (aPD-1 and aSPP1, each administered at 5 mg/kg body weight) were injected intraperitoneally on days 1, 4, 7, 10, and 13. On completion of the treatment regimen, the mice were euthanized, and LM were collected for further analysis of the TIME using flow cytometry and pathological staining techniques.

Evaluation of tumor immune microenvironment
At the end of the treatment period, the TIME was analyzed using flow cytometry. Single-cell suspensions were prepared from LM tumor tissues. The suspensions were stained with a panel of fluorescently labeled antibodies, including anti-mouse CD3-PerCP/Cyanine5.5, anti-mouse CD4-FITC, anti-mouse CD8-APC, anti-mouse HSP70-FITC, anti-mouse CD11c-PE/Cy7, anti-mouse CD80-FITC, anti-mouse CD86-APC, anti-mouse CD11b-FITC, anti-mouse F4/80-APC, and anti-mouse CD206-BV421.

Immunohistochemical and immunofluorescence staining analysis
LM were fixed and subjected to staining for Ki67, caspase 3, CD86, CD206, CD4, CD8, SPP1, transforming growth factor (TGF)-β, and interleukin (IL)-10. For immunohistochemical analysis, tumor sections were processed by defatting, quenching endogenous peroxidase activity, and blocking with bovine serum albumin (BSA). The sections were then incubated with primary and secondary antibodies. Stained sections were examined under an optical microscope, and regions of interest were captured for further analysis. For immunofluorescence analysis, the sections were deparaffinized and blocked with BSA. Subsequently, they were stained with primary and secondary fluorescent antibodies. The slides were visualized under an optical microscope, and areas of interest were documented.

Statistical analysis
Data are shown as means±SD. Unpaired Student’s t-test was performed to assess the statistical significance between two groups. Correlations between two parameters were evaluated using Spearman’s rank correlation coefficient test. Prognostic differences between two groups were investigated using Kaplan-Meier survival curves. Statistical significance is denoted as *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001.

Results

Comprehensive single-cell transcriptomic atlas of human colon cancer and corresponding liver metastases
To delineate the single-cell landscape of the TIME in CCLM, we collected clinical samples comprising primary CC, matched LM, and corresponding adjacent normal tissues from four patients which were pathologically confirmed as microsatellite stable/mismatch repair proficient (online supplemental figure S1). These samples were subjected to scRNA-seq analysis (figure 1A). Following stringent quality control and filtering procedures, transcriptomic data from a total of 63,782 single cells across the four patients were retained for subsequent analysis. After normalizing gene expression levels, we identified the top 2,000 highly variable genes for PCA to reduce the dimensionality of the sequencing data. Subsequently, the top 30 principal components were used to implement Harmony, effectively mitigating batch effects across samples. We then employed graph-based clustering to partition the cells and UMAP for visualization (figure 1B). Cell clusters were annotated using the top 20 feature genes identified by the FindAllMarkers function and marker genes described in previous literature.19 20 The cells were classified into ten major cell types: mast cells (characterized by TPSAB1 and TPSB2 expression), B cells (identified by MS4A1 and CD79A), endothelial cells (positive for VWF and PECAM1), myeloid cells (defined by classical markers LYZ and CD14), plasma cells (marked by IGHA1 and MZB1), epithelial cells (expressing EPCAM and KRT18), astrocytes (defined by PLP1 and NRXN1 expression), plasmacytoid dendritic cells (pDCs; marked by LILRA4 and CLEC4C), fibroblasts (identified by DCN and COL1A2 expression), and T cells (expressing CD3E and CD3D) (figure 1C and D). The proportions of these major cell types varied across different samples, highlighting the heterogeneous nature of the TIME in patients with CCLM. Notably, CCLM exhibited a higher proportion of immune cells compared with primary tumors. Specifically, CCLM showed a relatively higher proportion of T cells and lower proportions of fibroblasts, plasma cells, and mast cells compared with primary CC (figure 1E). Furthermore, primary CC tissues contained more fibroblasts than adjacent colon tissues, while LM tissues exhibited a higher proportion of endothelial cells compared with adjacent liver tissues (online supplemental figure S2 and S3). These observations suggest distinct TIME profiles among primary CC, LM, and their corresponding adjacent tissues. Understanding the immune landscape of these tissues may hold significant implications for developing strategies to inhibit CCLM progression.

Characteristics of T cells in CC and matched LM
Based on our single-cell data, T cells were significantly more infiltrated in LM compared with primary CC lesions. Consequently, we isolated 32,593 T cells for further analysis to explore their heterogeneity between CC and CCLM. The transcriptional data of these T cells underwent re-standardization, normalization, and re-clustering. Subclustering of T cells revealed 13 distinct subpopulations (figure 2A). Among these, CD8+ exhausted T cells (CD8Tex) were identified by the expression of TOX, a gene critical for inducing T-cell exhaustion. Another subpopulation, CD4_ANK3, was defined by the expression of ANK3, CAMK4, HIVEP2, ETV6, and ZBTB20, which exhibited a gene expression profile similar to that of CD8Tex (figure 2B, online supplemental figure S4-S6). Both CD4_ANK3 and CD8Tex showed increased infiltration in LM compared with CC (figure 2A, online supplemental figure S4-S6), leading us to hypothesize that CD4_ANK3 represents a subset of exhausted CD4+ T cells analogous to CD8Tex. Using GZMK and GZMB as markers, we differentiated two subsets of CD8+ effector T cells (CD8Teff), designated as CD8Teff_GZMK and CD8Teff_GZMB. Notably, CD8Teff_GZMK exhibited significantly higher expression levels of HSPA1A and HSPA1B compared with CD8Teff_GZMB (figure 2B), suggesting that GZMK-dominant effector T cells may experience more pronounced extrinsic stimulation. CD4+ regulatory T cells (CD4Treg) were characterized by high expression of FOXP3, CTLA4, and TIGIT. Naive T cells were identified based on their expression of CCR7 and IL7R. Additionally, we identified a subset of T cells in a TSTR, marked by the expression of HSPA1A and HSPA1B. Interestingly, TSTR cells also expressed hallmark genes associated with naive T cells, leading us to hypothesize that TSTR cells may differentiate from naive T cells. Furthermore, we identified a population of CD4+ T cells predominantly expressing CXCL13. Mucosal-associated invariant T (MAIT) cells were characterized by the expression of KLRB1 and SLC4A10. Progenitor T (Pro_T) cells were marked by MKI67 and STMN1, while natural killer precursor cells were identified by KIT and AFF3. We also classified two distinct subsets of natural killer (NK) cells: one expressing NCAM1 (CD56) and GNLY, designated as CD56-positive NK (NK_CD56) cells, and another expressing FCGR3A (CD16) and NKG7, designated as CD16-positive NK (NK_CD16) cells (figure 2B and C).
Single-cell trajectory analysis is a computational approach used to investigate the dynamic transitions of cells from one state to another. In this study, we applied this method to six T-cell subsets with potential shared origins in LM. Naive T cells were positioned at the initial segment of the developmental trajectory, while CD4_ANK3 and CD8Tex were located at the terminal end, underscoring their similarity to exhausted T cells. The developmental trajectory of TSTR cells closely paralleled that of naive T cells but diverged from that of exhausted T cells, suggesting distinct differentiation pathways for these populations (figure 2D). The developmental progression of CD8Teff_GZMK closely aligned with that of TSTR cells, whereas CD8Teff_GZMB was evenly distributed across all developmental stages. This finding provides a plausible explanation for the significant upregulation of HSPA1A and HSPA1B observed in CD8Teff_GZMK cells. In summary, our analysis revealed that naive T cells in CCLM follow four distinct differentiation pathways: (1) differentiation into CD8Teff cells, (2) differentiation into TSTR cells, (3) initial differentiation into CD8Teff followed by further differentiation into CD8Tex cells, or (4) initial differentiation into CD4_ANK3 cells (figure 2D).

T-cell subclusters, excluding CD4_ANK3 and CD8Tex, exhibit a stress response state in liver metastasis
In this study, T cells in LM were classified into 10 distinct subclusters, predominantly comprising exhausted T cells, including CD4_ANK3 and CD8Tex, alongside other subpopulations. We subsequently investigated the DEGs in T cells between primary CC and matched LM using the EnhancedVolcano R package. Notably, the HSP70 gene family, including HSPA1A, HSPA1B, and HSPA6, exhibited significantly upregulated expression in subclusters such as TSTR, Pro_T, CD4_CXCL13, CD4Treg, Naive_T, MAIT, CD8Teff_GZMK, and CD8Teff_GZMB in LM compared with their expression levels in matched CC tissues (figure 3A–H). To validate these findings, we performed immunofluorescence staining to detect HSP70 expression on CD4+ and CD8+ T cells in CC LM tissues. The results confirmed that HSP70 expression was markedly higher on CD4+ and CD8+ T cells in LM compared with those in primary CC (figure 3I, online supplemental figure S7). Unlike effector T cells, exhausted T cells and TSTR cells are functionally impaired. However, their gene expression profiles and developmental trajectories differ significantly. In our study, the expression of HSP70 in CD4_ANK3 and CD8Tex showed no significant difference between CC and LM. In contrast, genes such as IGLC2, AC060765.2, and AC079950.1 were highly expressed in exhausted T cells in LM compared with their expression levels in CC (figure 3J and K). Based on these observations, T cells in LM were categorized into two main groups: HSP70high T cells, representing stress response state T cells, and exhausted T cells. This classification highlights the distinct functional and molecular characteristics of T-cell subpopulations within the TIME of LM.

Single-cell profiling of myeloid cell subsets in colon cancer and liver metastasis tissues
After isolating mast cells, myeloid cells, and pDCs, we reconstructed the Seurat object, standardized the gene expression data, and performed dimensionality reduction and clustering analyses. A total of 6,455 myeloid cells were further reclassified into 13 distinct subclusters (figure 4A). Among these, we identified four macrophage subtypes: macro_SPP1, macro_DPYD, macro_CCL5, and macro_TFF3, characterized by the expression of SPP1, DPYD, CCL5, and TFF3, respectively. Monocytes were divided into two subsets, mono_CD14 and mono_CD16, based on the high expression of FCN1, S100A8, and S100A9 for mono_CD14, and FCGR3B (CD16) and FTL for mono_CD16. Additionally, we defined a subset of myeloid cells with high expression levels of MKI67 and STMN1, markers associated with cell proliferation, as the myeloid_MKI67 subset. Furthermore, we identified two pDC subtypes (pDC_JCHAIN and pDC_LILRA4) and three conventional dendritic cell (cDC) subtypes (cDC1_CLEC9A, cDC2_CD1C, and cDC3_LAMP3) based on their respective marker gene expression profiles (figure 4B and C).
To explore the developmental dynamics of myeloid cells in LM, we selected five myeloid cell subsets for trajectory analysis. Using mono_CD14 as the initial cell type, we analyzed the differentiation trajectories of the four macrophage subtypes: macro_SPP1, macro_DPYD, macro_CCL5, and macro_TFF3. The analysis revealed that macro_DPYD was positioned at the terminal stage of differentiation, macro_SPP1 and macro_TFF3 were at intermediate stages, and the majority of macro_CCL5 were at the initial stage of differentiation (figure 4D). These findings provide insights into the hierarchical differentiation and functional diversity of myeloid cell populations within the tumor microenvironment of LM.

SPP1, alongside APOC1 and APOE, was highly expressed in myeloid cell subclusters in LM compared with CC
In this study, myeloid cells in LM were classified into 13 distinct subclusters. Notably, the expression of SPP1 in macrophage subclusters was significantly higher in LM compared with primary CC (figure 5A). We further investigated the DEGs in monocyte-macrophages between CC and matched LM. Intriguingly, SPP1, along with apolipoprotein E (APOE) and apolipoprotein C1 (APOC1), exhibited remarkably upregulated expression in macrophage subsets, including Macro_SPP1, Macro_CCL5, Macro_TFF3, Macro_DPYD, Mono_CD14, and Mono_CD16, in LM compared with their expression levels in CC (figure 5B–G). To validate these findings, we performed immunofluorescence staining to assess SPP1 expression in macrophages within CCLM. The results revealed a significant increase in the infiltration of CD68+SPP1+ cells in LM compared with CC (figure 5H). APOE and APOC1 are traditionally known for their roles in lipoprotein metabolism. However, recent studies have implicated these proteins in processes such as cell proliferation, angiogenesis, and tumor metastasis.21 22 To explore the clinical relevance of SPP1, APOC1, and APOE in CC, we analyzed data from The Cancer Genome Atlas. The expression of SPP1 showed a significant correlation with APOC1 and APOE in patients with CC patients, and high expression levels of these genes were associated with poor prognosis (figure 5I–M). These findings suggest that the elevated expression of SPP1, accompanied by APOC1 and APOE in macrophages, may play a critical role in promoting CCLM, highlighting their potential as therapeutic targets.

Macro_SPP1 upregulated HSP70 in T cells via SPP1/CD44/PI3K/AKT signaling pathway
Here, BMDMs were subjected to hypoxic culture conditions to induce elevated SPP1 expression, designated as macro_SPP1high (figure 6A). Comparative analysis revealed that BMDMs cultured under hypoxic conditions demonstrated significantly higher SPP1 expression relative to those maintained under normoxic conditions (macro_SPP1low) (figure 6B). To identify the ligand-receptor pairs involved in the interaction between SPP1+ macrophages and T cells, we used the ligand-receptor database provided by the CellChat R package to infer the cell communication network between macrophages and T cells and discovered that SPP1+ macrophages primarily interact with T cells through four ligands: CD44, ITGAV+ITGB1, ITGA4+ITGB1, and ITGA5+ITGB1, facilitating intercellular communication (online supplemental Figure S8). To assess the influence of SPP1+ macrophages on TSTR, we established co-culture systems of BMDMs with T cells (figure 6C). Following 24 hours of co-culture, the macro_SPP1high group exhibited a marked reduction in CD4+ and CD8+ T-cell populations, concomitant with increased HSP70 expression in both T-cell subsets (figure 6D). These phenotypic changes were further accentuated after 48 hours of co-culture (figure 6E). To elucidate the mechanistic role of CD44 in these observed effects, we employed a CD44-blocking antibody (aCD44) (figure 6F). The administration of aCD44 in the macro_SPP1high-T-cell co-culture systems led to a partial restoration of CD4+ and CD8+ T-cell proportions, indicating that CD44 plays a role in the SPP1-mediated modulation of T-cell populations (figure 6G and H and online supplemental figure S9). T cells co-cultured with SPP1high macrophages exhibited significantly higher expression levels of PD-1, T-cell immunoglobulin and mucin-domain containing-3 (TIM-3), and lymphocyte activation gene 3 (LAG-3) compared with those co-cultured with SPP1low macrophages, indicating enhanced T-cell functional impairment. Importantly, CD44 blockade effectively reversed this T-cell dysfunction, suggesting that SPP1 mediates functional impairment of T cells predominantly through the CD44 signaling axis (online supplemental Figure S10). Notably, we examined the contribution of the PI3K/AKT signaling pathway to these cellular alterations using flow cytometry and western blotting. Our results demonstrated that Macro_SPP1high cells trigger a TSTR via activation of the SPP1/CD44/PI3K/AKT signaling axis (figure 6I and J, online supplemental figure S11 and S12).

Combination of aSPP1 and aPD-1 therapy markedly suppressed the growth of liver metastases and reprogrammed the immunosuppressive microenvironment
To further validate the antitumor efficacy of aSPP1 in combination with aPD-1 in vivo, we established CCLM models using CT26-Luc cells (online supplemental figure S13). The detailed experimental design and treatment protocols are illustrated in figure 7A. Following successful model establishment, the uniformity of LM was assessed using bioluminescence imaging (online supplemental figure S14). Based on the imaging results, mice were randomly allocated into four groups and treated with PBS, aPD-1, aSPP1, or aSPP1+aPD-1 combination therapy. At the end of the treatment period, mice were euthanized, and LM were harvested for subsequent analysis. The aSPP1+aPD-1 combination therapy group exhibited the smallest tumor volume and the lowest tumor weight compared with the other treatment groups (figure 7B and C). Furthermore, CCLM tissues from the combination therapy group showed the lowest levels of cell proliferation and the highest rates of apoptosis (online supplemental figure S15). These results demonstrate that aSPP1 synergistically enhances the antitumor effects of aPD-1 immunotherapy, resulting in significantly greater inhibition of tumor growth compared with monotherapy.
In this study, we have elucidated that SPP1 serves as a regulatory factor in modulating the proportions of CD4+ and CD8+ T cells, while also inducing T cells to transition into a stress response state in vitro. To further explore the potential of aSPP1 in alleviating the immunosuppressive microenvironment within CCLM, we conducted a comprehensive analysis of the immune microenvironment in LM following various treatment regimens. Consistent with in vitro findings, aSPP1 treatment increased the infiltration levels of CD4+ and CD8+ T cells. Combination of aSPP1 and aPD-1 treatments markedly enhanced the infiltration of these T cells in the tumor microenvironment (figure 7D and E, online supplemental figure S16). Furthermore, flow cytometric analysis demonstrated that aSPP1 treatment markedly decreased HSP70 expression levels in both CD4+ and CD8+ T cells, indicating that targeting SPP1 can effectively alleviate TSTRs (figure 7F and G). A significant increase in granzyme B expression following aSPP1 blockade, which was further enhanced when combined with aPD-1 treatment, providing additional evidence that SPP1 blockade restores T-cell function (online supplemental figure S17). Then, we assessed the polarization ratios of TAMs in the TIME following different treatments. The proportion of M2-like immunosuppressive TAMs (CD11b+F4/80+CD206+) was significantly decreased in the aSPP1 treatment group, with an even more pronounced reduction observed in the aSPP1+aPD-1 combination therapy group (figure 7H and I, online supplemental figure S18). The levels of inhibitory cytokines IL-10 and TGF-β, which are closely associated with TAMs, were significantly reduced in the aSPP1+aPD-1 combination therapy group (online supplemental figure S19). Mature DC ratio was significantly increased in the aSPP1 treatment group compared with PBS group and aSPP1 combined with aPD-1 showed synergistic effect in promoting dendritic cell recruitment (online supplemental figure S20). Combination therapy of aSPP1 and aPD-1 significantly inhibited the growth of LM, promoted DC maturation, reduced the proportion of M2 macrophages, enhanced T-cell infiltration, and alleviated T-cell stress.

Discussion
Currently, we used single-cell transcriptomic sequencing to analyze key immune components in both primary CC and their corresponding LM, identifying potential targets to enhance immunotherapy. Macro_SPP1high cells significantly reduced T-cell populations and induced a stress response state in both CD4+ and CD8+ T cells, thereby mediating CCLM via the SPP1/CD44/PI3K/AKT signaling pathway. Therapeutic intervention with aSPP1 combined with aPD-1 not only markedly suppressed LM growth and reversed the immunosuppressive microenvironment but also offers a promising synergistic strategy to enhance the efficacy of immune checkpoint immunotherapy in patients with CCLM.
SPP1 has been recently studied for its immunomodulatory functions, and our findings further substantiate its critical role in promoting immunosuppression within CCLM. Accumulating evidence indicates that SPP1+ macrophages are prominently associated with the establishment of an immunosuppressive tumor microenvironment and actively promote tumor angiogenesis and progression.23 24 Furthermore, emerging studies have demonstrated that SPP1 contributes to the functional exhaustion of both CD8+ and CD4+ T-cell populations, with its expression levels showing significant positive correlations with T-cell exhaustion-related gene signatures across multiple malignancies, including CRC and thyroid carcinoma.17 25 Our investigation revealed a marked upregulation of SPP1 expression in TAMs within liver metastatic lesions compared with primary CC sites, correlating with adverse clinical outcomes. This finding is consistent with established literature demonstrating that SPP1-mediated immune dysregulation within the tumor microenvironment significantly contributes to unfavorable prognosis in patients with CRC.26 Furthermore, SPP1 orchestrates the establishment of an immunosuppressive niche through multiple mechanisms, including the induction of T-cell exhaustion and functional impairment, which are mechanistically associated with enhanced apoptotic signaling, metabolic dysregulation, and cellular senescence.27 28 These coordinated processes collectively facilitate immune evasion in metastatic progression.
Our investigation identified a unique T-cell subpopulation in LM displaying a stress response phenotype (designated as TSTR cells), characterized by significant upregulation of HSPA1A and HSPA1B expression. These molecular chaperones, belonging to the heat shock protein family, are well-established mediators of cellular stress responses and have been implicated in the phenotypic transition from functional effector T cells to a dysfunctional or exhausted state.29 30 Importantly, TSTR cells appear to represent a transitional state between effector and terminally exhausted T-cell populations, potentially providing crucial mechanistic insights into immune suppression pathways operative in metastatic CRC. Unlike terminally exhausted T cells, which exhibit an irreversible and persistent state of immune dysfunction, TSTR cells may retain partial capacity for functional restoration following therapeutic modulation, thereby representing a promising therapeutic target for immunotherapeutic interventions.
While both TSTR cells and exhausted T cells demonstrate functional impairment, including diminished proliferative capacity and altered cytokine production profiles, they represent distinct developmental stages and exhibit unique functional characteristics within the immune response hierarchy.15 31 Terminally exhausted T cells, which are predominantly observed in the context of chronic infections and malignant tumors, manifest profound and irreversible dysfunction resulting from persistent antigen exposure.32 This sustained activation leads to the coordinated upregulation of multiple immune checkpoint molecules, including PD-1, TIM-3, and LAG-3, which collectively impair cellular proliferation and cytokine secretion capabilities.33 In contrast, TSTR cells emerge as an intermediate cellular state under conditions of chronic stress, particularly within the tumor microenvironment. These cells are molecularly characterized by the prominent expression of stress-responsive proteins, particularly HSPA1A and HSPA1B.15 34 Importantly, unlike their terminally exhausted counterparts, TSTR cells maintain partial functional plasticity and retain the capacity for activation and potential differentiation into effector T-cell populations when provided with appropriate microenvironmental cues.
In this study, we identified TSTR cells by their distinct expression of HSPA1A and HSPA1B. In contrast, Tex exhibited elevated expression of genes such as ANK3, ZBTB20, and CAMK4, which were significantly downregulated in TSTR cells. Furthermore, HSPA1A and HSPA1B expression levels were substantially lower in Tex cells compared with TSTR cells. Although both TSTR and Tex cells exhibit immune dysfunction, TSTR cells retain the capacity to respond to therapeutic interventions, including ICB, which has the potential to reverse their dysfunctional state and restore immune responsiveness. These findings highlight TSTR cells as a promising therapeutic target for strategies aimed at rejuvenating T-cell function.

Conclusion
Overall, we have identified a novel population of HSPA1Ahigh/HSPA1Bhigh T cells (TSTRs) that exhibit high infiltration in LM. These T cells exist in a stress state and contribute to cancer immune escape. Notably, TSTR is characterized as an intermediate developmental state bridging effector and exhausted T cells, representing a distinct cellular state separate from exhausted T cells. Intriguingly, Macro_SPP1high cells upregulate HSPA1A and HSPA1B expression in T cells, inducing a stress response state in these cells and mediating CCLM through the SPP1/CD44/PI3K/AKT signaling pathway (figure 8). The blockade of Macro_SPP1high-TSTR crosstalk can reverse the TSTR state and inhibit the growth of LM, suggesting a potential novel therapeutic strategy for suppressing CCLM.

Supplementary material
10.1136/jitc-2025-012330online supplemental file 1