Heterogeneous SPP1-expressing esophageal cancer cells license immunotherapy resistance by organizing an immunosuppressive niche Authors: Jiakang Ma, Jianlin Zhu, Yiling Chen, Xiaodong Chu, Jie Zeng, Hongyu Chen, Yuhao Liu, Xiaosheng Lin, Lei Zhang, Jinyuan Yi, Guangping Zhang, Songwang Cai, Yusheng Lin, Lu Wang Published: 2026 ABSTRACT Highlights•A multi-omics approach identifies a novel SPP1-expressing malignant epithelial subpopulation in ESCC, functionally distinct from the well-characterized macrophage-derived SPP1, as a key driver of immunotherapy resistance.•Mechanistic analyses reveal that SPP1 drives immune evasion by upregulating PD-L1 via the mTORC1-STAT3 pathway, exhausting CD8⁺ T cells through SPP1-CD44-MAPK signaling, and polarizing macrophages via α5β1 integrin-FAK-AKT-mTOR activation.•Spatial transcriptomics validates that SPP1⁺ cells organize immunosuppressive niches, and therapeutic targeting of SPP1 synergizes with anti-PD-1 therapy to suppress tumor growth.•Drug sensitivity profiling nominates microtubule and kinase inhibitors as effective against SPP1⁺ cells, offering a strategy to address tumor heterogeneity and overcome resistance.Keywords: SPP1, Tumor heterogeneity, Immunotherapy resistance, Tumor microenvironment, Esophageal squamous cell carcinoma, Single-cell multi-omics analysis Introduction Esophageal Squamous Cell Carcinoma (ESCC) remains one of the most aggressive and lethal gastrointestinal cancers, with over 500,000 new cases globally and dismal 5-year survival rates of <20% in advanced stages [[1], [2], [3]]. Neoadjuvant chemoimmunotherapy (NACI), which combines PD-1/PD-L1 inhibitors with platinum-based chemotherapy, has shown promising efficacy in locally advanced ESCC [4]. However, primary resistance to NACI remains a significant obstacle, with >50% of patients failing to achieve a major pathological response, thereby limiting survival benefits [5]. The mechanisms underlying NACI resistance are multifactorial and involve intricate interactions within the tumor microenvironment (TME) [6,7]. Studies using single-cell RNA sequencing (scRNA-seq) have revealed the cellular heterogeneity and complex immune-stromal networks that contribute to the immunosuppressive landscape of ESCC, including the expansion of exhausted CD8+ T cells, regulatory T cells (Tregs), myeloid-derived suppressor cells (MDSCs), and tumor-associated macrophages (TAMs) [[8], [9], [10]]. These immune subsets, along with stromal components such as cancer-associated fibroblasts and endothelial cells, collectively drive resistance to immune checkpoint therapies [[11], [12], [13]]. However, while immune and stromal factors have been well characterized, the role of malignant epithelial cell heterogeneity in orchestrating immune evasion and therapy resistance remains poorly understood. Recent studies suggest that malignant epithelial cells exhibit functional and phenotypic heterogeneity, enabling them to dynamically integrate multiple signaling pathways to promote immune evasion and tumor progression. Although individual mechanisms such as PD-L1 upregulation, TGF-β secretion, and metabolic competition have been identified [[14], [15], [16]], how heterogeneous malignant cell subpopulations function as signaling hubs to synchronize immune responses within the TME remains a fundamental gap in our understanding. SPP1 (secreted phosphoprotein 1), also known as osteopontin, is a multifunctional extracellular matrix protein predominantly expressed in macrophages, with elevated levels correlating with worse survival outcomes across various cancers [17]. SPP1 interacts with receptors such as integrins and CD44, mediating immune cell functions and influencing the TME [18]. Despite its known role in immune regulation, the expression of SPP1 within tumor cells and its potential implications for driving tumor heterogeneity have yet to be fully explored. In this study, we identified a distinct SPP1⁺ malignant epithelial subpopulation enriched in NACI-resistant ESCC, representing a critical component of tumor cellular heterogeneity. Although SPP1 is well studied in macrophages, its role in malignant epithelial cells remains comparatively underexplored despite growing evidence of epithelial–immune crosstalk. We therefore investigated whether epithelial SPP1 contributes to therapy resistance through its heterogeneous expression and immune-modulatory activity. SPP1-expressing epithelial cells enhanced tumor-intrinsic malignancy and PD-L1 expression and were associated with broad immunosuppressive remodeling of the tumor microenvironment. Specifically, SPP1⁺ cells promoted CD8⁺ T-cell exhaustion through the SPP1–CD44 axis, with accompanying CD44-dependent activation of MAPK effectors, and promoted an immunosuppressive macrophage program through α5β1 integrin-associated FAK–AKT–mTOR signaling. Therapeutically, SPP1 blockade enhanced the antitumor effect of anti-PD-1 treatment and reduced markers of T-cell exhaustion and immunosuppressive macrophage infiltration in vivo. These findings establish SPP1⁺ malignant cells as central orchestrators of immunotherapy resistance and position SPP1 as an actionable therapeutic target to disrupt heterogeneity-driven resistance in ESCC. Methods and materials Data collection Single-cell RNA sequencing (scRNA-seq) dataset GSE203115, ESCC dataset GSE53625, ESCC cell line sequencing dataset GSE93641, and spatial transcriptomics data of head and neck squamous cell carcinoma (HNSCC) GSE181300 were obtained from the GEO database (https://www.ncbi.nlm.nih.gov/geo) [[19], [20], [21], [22]]. The GSE203115 dataset included tumor samples from ESCC patients who underwent two cycles of NACI, comprising both responders and non-responders. mRNA expression profiles of patients with ESCC were downloaded from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/) and defined as the TCGA-ESCC cohort. In addition, partial expression data visualizations and analysis results for TCGA, GSE53625, GSE17351, GSE20347, GSE23400, GSE38129, GSE45670, and GSE77861 were obtained from the ESCCEXPRESS database (http://www.bioinfo-zs.com/esccexpress/) [23]. Raw scRNA-seq data from 14 ESCC samples of patients receiving neoadjuvant immune checkpoint blockade therapy in the Shanghai Chest Hospital cohort were obtained from the Genome Sequence Archive (GSA) under accession number PRJCA012636 [24]. Raw reads were aligned to the human reference genome GRCh38 using Cell Ranger version 6.0.2 (10x Genomics) with default parameters. Single-cell RNA sequencing data analysis We used the Seurat package (version 4.3.0) [25] to process sparse expression matrix data and perform initial quality control. After quality control and filtering, cells with 200 to 5,000 genes and mitochondrial gene percentages below 20% were retained. Analysis of scRNA-seq data included normalization, dimensionality reduction, and clustering. Integrated and scaled data underwent principal component analysis (PCA), retaining the top 30 principal components, and batch effects were corrected using the Harmony R package [26]. The FindNeighbors and FindClusters functions (resolution = 0.4) were used to determine cell clusters. Cell types were identified through automatic and manual annotation, and the AddModuleScore algorithm was used to calculate gene signature scores for each cell subpopulation. Quantification of NACI treatment response enrichment in malignant cell subpopulations The ratio of observed to expected cell numbers (Ro/e) method [27] was employed to assess the enrichment of each malignant epithelial cell subpopulation across NACI treatment response groups. Ro/e values greater than 1 indicate enrichment, while values <1 indicate depletion in specific tissues or cancer types. Trajectory analysis The slingshot R package [28] was used to infer lineage and pseudotime for six malignant epithelial cell subpopulations. This method employs clustering-based minimum spanning trees to identify lineage structures and applies simultaneous principal curves to fit branching curves. Additionally, the scTour software [29] was used to dissect the dynamics of these six epithelial cell subpopulations. Using Monocle2 [30] to infer potential differentiation associations between T cell subclusters or macrophage subclusters. Cell-cell communication analysis The R package “CellChat” (version 1.6.1) [31] was used to systematically infer and visualize intercellular interactions. Gene expression data were processed to identify highly expressed ligands and receptors in specific cell populations. Communication probabilities for all ligand-receptor pairs associated with each signaling pathway were calculated, constructing a comprehensive cell-cell communication network to understand interactions and communication mechanisms. High-dimensional weighted gene co-expression network analysis To identify gene co-expression regulatory networks associated with SPP1+ malignant epithelial cells, we performed high-dimensional weighted gene co-expression network analysis (hdWGCNA) using the “hdWGCNA” package (version v0.4.6) [32]. We began by constructing the expression correlation matrix for the malignant epithelial cell subpopulation, followed by selecting the optimal soft-thresholding power and building the co-expression network. Next, we computed harmonized module eigengenes, assessed module connectivity, generated the module assignment table, and carried out a series of visualization analyses. Finally, key gene modules were subjected to enrichment analysis using the R package enrichR to gain deeper biological insights. Gene set variation analysis (GSVA) 50 Hallmark gene sets were retrieved from the GSEA database (http://www.broadinstitute.org/gsea/msigdb). The enrichment scores were computed using the "gsva" algorithm from the GSVA R package [33] and significantly altered pathways were identified using the limma R package. Pathways with |t value of GSVA scores| > 1 were deemed to exhibit significant alterations. Clinical patients and specimens All study procedures were reviewed and approved by the Ethics Committee of Jinan University (Institutional Review Board approval No JNUKY-2024-0050). Written informed consent was obtained from all patients in accordance with the Declaration of Helsinki. We collected 25 clinical samples from ESCC patients treated with neoadjuvant immunotherapy using anti-PD-1 therapy (Camrelizumab) at the First Affiliated Hospital of Jinan University, including 10 samples from non-responders and 15 from responders. In addition, 32 newly diagnosed ESCC patients who had received no prior anticancer therapy were enrolled. ESCC diagnoses were confirmed by reviewing pathology records. IHC staining for SPP1 was performed on clinical specimens and scored using the IRS scoring system, which ranges from 0 to 12 points; a score greater than 4 was defined as high expression. Cell culture and plasmid transfection Human ESCC cell lines TE-1 and KYSE-150 and the murine ESCC cell line AKR were maintained under standard culture conditions at 37°C in a humidified incubator with 5% CO₂. The full-length human SPP1 cDNA cloned into pcDNA3.1, as well as the lentiviral plasmids pLVX-Spp1-IRES-Puro and the corresponding pLVX-IRES-Puro empty vector for murine cells, were synthesized by GenePharma (Shanghai, China). Transient transfection was performed using Lipofectamine 3000 (Invitrogen, USA) according to the manufacturer’s instructions. Stable overexpression was established by lentiviral transduction. Total RNA extraction and quantitative reverse transcription PCR (qRT-PCR) Total RNA was extracted from cells using the Total RNA Extraction Kit (M5105, NCM Biotech). cDNA synthesis was performed using PrimeScript RT Master Mix (#RR036A, Takara). qRT-PCR was conducted using the Bio-RAD real-time system (CFX connect, USA) and MagicSYBR Mixture (#CW3008M, CWBIO). GAPDH was used as the internal reference for normalization, and relative mRNA expression was calculated using the 2-ΔΔCT method. Primer sequences were as follows: SPP1 forward: 5′-ATGGAAAGCGAGGAGTTGAA-3′; reverse: 5′-ACTGCTTGTGGCTGTGGGT-3′. CD274 forward: 5′-CAGAAAAGCCTCATTCGTTGT-3′; reverse: 5′-CCAGGGCATCTGAATCTCG-3′. TIM-3 forward: 5′-AGGCAGCAACCCTCACAAC-3′; reverse: 5′-GGACAGAACCTCCAAAACCAG-3′. PD-1 forward: 5′-GCCTGTGTTCTCTGTGGACT-3′; reverse: 5′-ATGGTGGCATACTCCGTCTG-3′. LAG-3 forward: 5′-GGCTGCATCCTCACCTACAG-3′; reverse: 5′-TTGGCAGTGAGGAAAGACCG-3′. CD206 forward: 5′-CTGGTGGAAGAAGAAGCAGTCT-3′; reverse: 5′-GATTAGTCAAGGAAGGGTCGG-3′. CD163 forward: 5′-AGACCTGGACTATTGAATGGAGC-3′; reverse: 5′-TCAGCAGCAGTCTTAGGAATC-3′. iNOS forward: 5′-ATGGCCAAGGCCTGAACTAC-3′; reverse: 5′-CACCACTCGCTCCAGGATAC-3′. CD86 forward: 5′-CTCCAGCTCTGCTCCGTATG-3′; reverse: 5′-TGGCCCATAAGTGTGCTCTG-3′. GAPDH forward: 5′- GTCTCCTCTGACTTCAACAGCG-3′; reverse: 5′- ACCACCCTGTTGCTGTAGCCAA −3′. Western blotting Cells were lysed using RIPA buffer. Equal amounts of protein extracts were separated by 10% SDS/PAGE and transferred to polyvinylidene fluoride membranes. Membranes were blocked with 5% skim milk in TBS buffer containing 0.1% Tween-20 and incubated overnight at 4°C with primary antibodies against SPP1 (22952-1-AP, Proteintech), S6 (#2317, CST), phospho-S6 (Ser235/236) (#4857, CST), STAT3 (#9139, CST), phospho-STAT3 (Tyr705) (#9131, CST), PD-L1 (66248-1-Ig, Proteintech), FAK (#13009, CST), phospho-FAK (Tyr397) (#3283, CST), AKT (#9272, CST), phospho-AKT (Ser473) (#9271, CST), mTOR (#2983, CST), phospho-mTOR (Ser2448) (#5536, CST), p38 (#8690, CST ), phospho-p38 (Thr180/Tyr182) (#4511, CST ), ERK (#4695, CST), phospho-ERK (Thr202/Tyr204) (#4370, CST), JNK (ab208035, Abcam), and phospho-JNK (Thr183/Tyr185) (#9255, CST ) and β-actin (60008-1-Ig, Proteintech). Secondary antibodies were added and incubated for 2 hours at room temperature. Immunoreactive bands were visualized using enhanced chemiluminescence reagent (P10300, NCM Biotech) and X-ray film (Eastman Kodak). RNA sequencing RNA sequencing was performed in TE-1 cells transfected with empty vector (TE-1-vector) or an SPP1 overexpression construct (TE-1-SPP1) (n = 3 per group). Total RNA was extracted and subjected to quality control, with RNA concentration measured using a Qubit 4.0 fluorometer and RNA integrity assessed using a Qsep400 Bioanalyzer. Qualified samples were processed for mRNA isolation, followed by cDNA library construction via standard procedures including end repair, A-tailing, adapter ligation, purification, and PCR amplification. Final libraries were quantified (Qubit), assessed for insert size (fragment analyzer), converted into DNA nanoballs (DNBs), and sequenced on an MGI platform (Metware Biotechnology, Wuhan, China). For bioinformatic analysis, raw sequencing reads were quality-filtered using fastp (v0.23.2) to remove adapter sequences, discard reads containing >10% ambiguous bases (N), and exclude read pairs where >50% of bases had a quality score (Q) ≤20. Clean reads were then aligned to the human reference genome (Homo_sapiens. GRCh38.109) using HISAT2 (v2.2.1). Gene-level read counts were generated with featureCounts (v2.0.3). Differential expression analysis was conducted with DESeq2 (v1.38.3). Genes with an absolute log2 fold change > 0.5 and a false discovery rate (FDR) < 0.05 (Benjamini–Hochberg adjusted) were considered significantly differentially expressed. Spatial transcriptomics profiling and data analysis (Visium HD platform) Spatial transcriptomic profiling was conducted on a formalin-fixed, paraffin-embedded (FFPE) tumor block from ESCC Patient #1 using the Visium HD Spatial Gene Expression for FFPE assay (10x Genomics). RNA was extracted with the Qiagen RNeasy FFPE Kit, and quality was assessed by the DV200 metric. A 5 μm tissue section was mounted on a Poly Prep slide (Sigma-Aldrich, 10x Genomics; CG000684 Rev A), air-dried overnight, baked at 60°C for 2 hours, and then deparaffinized and decrosslinked per the manufacturer’s protocol. The section was H&E-stained and imaged at 20x magnification on a Leica Aperio Versa8 scanner. The Visium HD workflow, including probe hybridization/ligation, RNase treatment, tissue permeabilization, and probe capture by spatially barcoded oligonucleotides, was performed on the Visium HD instrument. The resulting libraries were sequenced on an Illumina platform. Raw sequencing data (FASTQ files) and histology images were processed with Space Ranger (v3.1.0; 10x Genomics) using the 'spaceranger count' pipeline with the human reference genome (GRCh38) to generate a filtered feature-barcode matrix. The processed data were imported into Seurat version 5.0.1 for downstream analysis. Spatial features were analyzed at a 16 × 16 μm bin resolution, which provided a balance between spatial granularity and transcript capture. Cell-type annotation was performed by reference-based deconvolution using RCTD, with tumor samples from the public ESCC scRNA-seq dataset PRJNA777911 serving as the reference. CCK-8 assay 3,000 SPP1-transfected TE-1 and KYSE-150 cells were seeded in 96-well plates. Then, 10 µL of CCK-8 reagent (abs50003, Absin) was added to each well and incubated at 37°C with 5% CO2 for 2 hours. OD450 values were measured at 0, 24, 48, and 72 hours using a microplate reader. Colony formation assay For colony formation assays, transfected cells (1,000 cells) were cultured in 6-well plates for 8–10 days, fixed with 4% paraformaldehyde, and stained with 0.1% crystal violet. Wound healing assay Cells were seeded in 6-well plates and cultured overnight until 90% confluent. A sterile 200 µL pipette tip was used to scratch the cell monolayer, and detached cells were gently washed away with PBS three times. The monolayer was then cultured in serum-free RPMI 1640 medium, and wound images were captured at 0, 24, and 48 hours using an inverted microscope. Migration and invasion assay Migration and invasion assays were performed using 24-well Transwell chambers with 8 µm pores (Corning Costar, Kennebunk, ME, USA). For migration or invasion assays, cells were resuspended in serum-free medium and seeded into chambers with or without Matrigel Matrix (#356234, Corning). The lower chamber contained RPMI 1640 medium with 10% FBS. After 24 hours of incubation, cells were fixed with 4% paraformaldehyde and stained with 0.1% crystal violet. Migrated cells were counted under a microscope. Immunohistochemistry staining Tumor tissues from clinical specimens and tumor-bearing mice were fixed in 10% formalin and embedded in paraffin. The paraffin blocks were sectioned into 4-μm slices and mounted onto polylysine-coated slides. After deparaffinization and rehydration, tissue sections were incubated with primary antibodies against SPP1 (22952-1-AP, Proteintech) or PD-L1 (66248-1-Ig, Proteintech). Subsequently, the sections were treated with a biotin-labeled secondary antibody followed by an avidin–biotin–peroxidase complex to enhance signal detection. Visualization of the target proteins was achieved using diaminobenzidine (DAB) as the chromogen. CD8+T cell responses to tumor-derived conditioned media Peripheral blood samples were obtained from healthy volunteers after written informed consent. Peripheral blood mononuclear cells were isolated by standard density gradient centrifugation using Lymphoprep (Serumwerk Bernburg). Total CD8+T cells were then purified from PBMCs using the BD IMag™ Anti-Human CD8 Magnetic Particles kit(557766, BD) and activated with CD3/CD28 Dynabeads (11131D, Gibco) for 24 h. TE-1 and KYSE-150 cells transfected with either the SPP1-overexpression plasmid or the empty vector were cultured for 3 days, after which the conditioned medium (CM) was collected. CM was prepared by mixing 50% tumor cell supernatant with 50% fresh culture medium and used to culture CD8+ T cells, or with supplementation of anti-CD44 antibody (NBP2-22530, Novus Biologicals). After 48 h of culture, CD8⁺ T cells were collected for qRT-PCR to quantify the expression of T-cell exhaustion–related markers, and western blotting was performed to evaluate the indicated molecules. In parallel, cytokine levels in the culture medium were measured by ELISA using commercial kits for IFN-γ (PI511, Beyotime) and TNF-α (PT518, Beyotime). Differentiation of THP-1 cells and stimulation with conditioned medium THP-1 cells were treated with 100 ng/mL phorbol-12-myristate-13-acetate (PMA, S1819-1 mg, Beyotime) for 48 h to induce differentiation into adherent M0 macrophages. CM was collected from TE-1 and KYSE-150 cells transfected with either SPP1-overexpression plasmid or empty vector. The M0 macrophages were then cultured with CM derived from ESCC cells for 2 consecutive days, or with supplementation of with a monoclonal antibody against integrin α5β1 (Volociximab, #HY‐P99333, MCE). After treatment, macrophages were harvested for qPCR and western blotting analysis. All experiments were performed in triplicate. Currently, cytokine levels in the culture medium were measured by ELISA using commercial kits for IL-1β (PI305, Beyotime) and IL-10 (BMS215-2, Invitrogen). Multiplex immunohistochemistry staining We performed Multiplex immunohistochemistry (mIHC) based on tyramide signal amplification (TSA) to characterize changes in the tumor immune microenvironment between SPP1-high and SPP1-low tumor tissues. mIHC staining was carried out using the PANO Multiplex IHC kit (cat. no. 10001100020, Panovue) according to the manufacturer’s instructions. The immune markers evaluated were CD8 (ab101500, Abcam), PD-1 (66220-1-Ig, Proteintech), CD68 (ab213363, Abcam), and CD163 (ab182422, Abcam). The antibodies used for murine tissues were as follows: CD8α (#98941, CST), TIM-3 (ab241332, Abcam), F4/80 (ab240946, Abcam), CD206 (ab64693, Abcam). After staining, slides were counterstained with DAPI for 10 minutes and mounted with ProLong™ Diamond Antifade Mountant (Thermo Fisher Scientific). The results were derived by performing a t-test on the ratio of positively stained cells to tissue area (cells/mm²). Animal experiments C57BL/6 mice (5-6 weeks old) were purchased from Beijing Charles River. The mice were randomly assigned to groups, fed a normal diet, and given free access to water, with each group consisting of 5 mice. For the subcutaneous model, AKR cells (1 × 106) stably expressing the overexpression vector and Spp1 were injected subcutaneously into the right side of the back of each mouse. Tumor growth was monitored every three days, and tumor size was calculated using the formula: 1/6 × π × length × width². After 18 days, all mice were euthanized, and the tumors were collected and excised, weighed, photographed, and subjected to IHC. For in vivo treatment experiments, AKR cells (1 × 10⁶) were subcutaneously injected into the right flank of each mouse. When tumors reached approximately 100 mm³, tumor-bearing mice were randomly assigned to four groups (n = 5 per group). Mice received intraperitoneal injections every three days with 200 μg of antibody in 100 μL PBS: IgG isotype control, anti-SPP1, anti-PD-1, or anti-SPP1 plus anti-PD-1. All antibodies were purchased from Bio X Cell. On day 21, mice were euthanized, and tumors were excised, weighed, photographed, and subjected to mIHC analysis. Sensitive drug prediction scDrug constructed a workflow for comprehensive analysis of scRNA-seq data, providing functions from data analysis to drug response prediction and therapeutic recommendations, including estimating half-maximal inhibitory concentration (IC50) for cell clusters and reporting cell death percentages for drug efficacy [34]. CB-Dock2 (https://cadd.labshare.cn/cb-dock2/php/index.php) focuses on protein-ligand blind docking, integrating cavity identification, docking, and homologous template matching [35]. By providing 3D structures of proteins and ligands, it predicts binding sites and affinities to aid computer-aided drug discovery. The 3D protein structure of SPP1 (AF-P10451-F1) was obtained from the UniProt database (https://www.uniprot.org/). Small drug molecule structures were retrieved from PubChem (https://pubchem.ncbi.nlm.nih.gov/). These tools were used to identify small-molecule drugs targeting Epi-SPP1 cells and the SPP1 protein. Statistical analysis Data are presented as mean ± standard deviation. Unpaired two-tailed t-test or Wilcoxon-test were used for comparisons between two groups, and one-way ANOVA was used for comparisons involving more than two groups. Categorical variables were compared using the χ² test where appropriate. Statistical significance was calculated using R 4.2.2 and GraphPad Prism 8.0. All experiments were repeated at least three times to ensure reliability. A p-value < 0.05 was considered statistically significant. Results SPP1⁺ malignant epithelial cells are enriched in neoadjuvant immunotherapy-resistant ESCC To explore the potential mechanisms of immunotherapy resistance driven by tumor heterogeneity, we constructed two ESCC datasets by integrating data from independent neoadjuvant immunotherapy cohorts. The discovery cohort integrated single-cell RNA sequencing (scRNA-seq) data from GSE203115 (2 responders, 1 non-responder) and PRJCA012636 (paired pre- and post-treatment samples from 7 patients) (Fig. 1A). We complemented this with spatial transcriptomics, bulk RNA sequencing, and functional experiments for mechanistic validation (Fig. 1A). Analysis of the GSE203115 dataset (11,445 high-quality cells after processing) identified 13 major cell populations, including both immune and non-immune populations (Fig. 1B, C; Fig. S1A). Comparative analysis revealed significant differences in the proportions of epithelial, stromal, and immune compartments between responders and non-responders (Fig. 1D). Copy number variation (CNV) analysis using inferCNVpy (version 0.4.5) confirmed malignant transformation exclusively within epithelial cells (Fig. 1E). Enrichment and characterization of SPP1⁺ malignant cells in NACI-resistant ESCC. (A) Schematics of the experimental design. Created with BioRender.com. (B) UMAP visualization of 11,445 cells analyzed by single-cell RNA sequencing (GSE203115). (C) Heatmap of marker genes defining cell clusters. (D) Cell type proportions in responders (R) vs. non-responders (NR). (E) UMAP visualization of copy number variation (CNV) scores across all cell types. (F) UMAP projection showing six malignant epithelial subtypes. (G) Expression levels of ESCC-associated genes across malignant subtypes (bubble plot: dot size = fraction of expressing cells; color = expression level). (H) Slingshot trajectory analysis showing both lineage trajectories (1 and 2). (I) Slingshot trajectory analysis showing lineage 1. (J) Slingshot trajectory analysis showing lineage 2. (K) Pseudotime scores (left) and vector field (right) visualized on UMAP space. (L) Ro/e ratio plot indicating subtype distribution in NR versus R. (M) Volcano plot of differentially expressed genes in Epi-SPP1 vs. other malignant cells. (N) Representative thoracic CT images of ESCC patients before and after anti-PD-1 (Camrelizumab) treatment (left panel), with yellow circles indicating the tumor locations; representative immunohistochemistry (IHC) images demonstrating SPP1 expression in matched two patients (left panel). Response rates to camrelizumab by SPP1 expression level (n = 25) (right panel). p = 0.009 by χ2 test. Subclustering identified six transcriptionally distinct malignant epithelial subtypes (Fig. 1F), demonstrating substantial heterogeneity within the malignant compartment. All subtypes expressed established ESCC-specific oncogenes (Fig. 1G), with marker genes defining each subtype (Fig. S1B). Pseudotime trajectory analysis (Slingshot/scTour) positioned Epi-GPC3 at the differentiation origin, with Epi-SPP1 and Epi-CAV1 occupying key transition states indicative of elevated plasticity (Fig. 1H-K). Critically, the SPP1⁺ malignant subtype (Epi-SPP1) was significantly enriched in non-responders (Fig. 1L, M). Analysis of the second dataset (PRJCA012636) corroborated this finding, identifying a homologous SPP1-expressing malignant subpopulation (Epi-6) that was likewise enriched in treatment-refractory cases (Fig. S2A-I). This consistent enrichment across cohorts highlights SPP1+ cells as a key heterogeneous component of resistance. Having established this association across our discovery cohorts, we next assessed its clinical relevance in an independent cohort of 25 ESCC patients treated with neoadjuvant camrelizumab (anti-PD-1) plus chemotherapy. Patients with high tumor SPP1 expression exhibited a significantly poorer radiographic response (Fig. 1N, Table S1). Collectively, these multi-cohort results identify an SPP1+ malignant epithelial subpopulation that is consistently enriched in immunotherapy-resistant ESCC, representing a critical element of tumor heterogeneity that mediates treatment failure and presents a potential therapeutic target. Aberrant SPP1 confers ESCC progression and immunotherapy resistance Since SPP1 is most commonly associated with its production and role in macrophages within the tumor microenvironment (TME), it is striking to identify SPP1+ epithelial cells amidst the heterogeneous cellular landscape, and its role in malignant epithelium has been underestimated. To define the functional role of the identified SPP1⁺ malignant subpopulation, we generated SPP1-overexpressing models in TE-1 and KYSE-150 ESCC cell lines (Fig. 2A). SPP1 overexpression significantly enhanced colony formation (Fig. 2B), cell viability (Fig. 2C), migration, and invasion (Fig. 2D-G). This pro-tumorigenic function was confirmed in vivo, where Spp1-overexpressing cells formed substantially larger subcutaneous tumors in mice (Fig. 2H). SPP1 promotes ESCC malignancy. (A) Validation of SPP1 overexpression in TE-1 (left) and KYSE-150 (right) cells by RT-qPCR (n = 3) and immunoblotting. (B) Colony formation assay showing the effect of SPP1 overexpression on TE-1 (top) and KYSE-150 (bottom) cell proliferation (n = 3). (C) CCK-8 assay results in SPP1-overexpressing TE-1 (top) and KYSE-150 cells (bottom, n = 3). (D-E) Representative images and quantification of migration and invasion assays in SPP1-overexpressing TE-1 cells (D) and KYSE-150 cells (E). Scale bar: 100 µm (n = 3). (F-G) Wound healing assay showing the effect of SPP1 overexpression on TE-1 cell migration (n = 3) (F) and KYSE-150 cell migration (n = 3) (G). (H) C57BL/6 mice inoculated with AKR cells transduced with vector control (n = 5) or Spp1 overexpression group (n = 5). Left: representative tumors; middle: tumor volume; right: tumor weight. Data are presented as mean ± SD. **p < 0.01, *** p < 0.001 by unpaired t-test. Next, we assessed the clinical relevance of tumor-intrinsic SPP1 expression. We found that SPP1 expression predicted immunotherapy resistance in both the ESCC (GSE203115) cohort and pan-cancer ICBatlas (Fig. S3A, B). Spatial transcriptomic profiling of ICB-treated LIHC tumors (Mendeley Data [36]) further demonstrated elevated SPP1 in non-responder tumor regions (Fig. S3C, D). Cross-platform GEO datasets validated significant SPP1 upregulation in ESCC tissues and cell lines relative to normal controls (Fig. S3E, F). Additionally, by analyzing ESCC proteomics data [37], we found that SPP1 protein level remained higher in tumor samples (Fig. S3G). In the TCGA-ESCC cohort, high SPP1 was correlated with shorter progression-free (HR = 2.82, 95% CI 1.26 - 6.28; p = 0.0086) and disease-free survival (HR = 10.22, 95% CI 1.35 - 77.58; p = 0.0058). Notwithstanding non-significant overall survival (OS) in TCGA (p = 0.1725), validation cohort (GSE53625) linked high SPP1 to shorter OS (HR = 1.55, 95% CI 1.06–2.29; p = 0.0244; Fig. S3H). Gene set variation analysis (GSVA) associated elevated SPP1 with enrichment of aggressive biological pathways, including epithelial-mesenchymal transition, NOTCH signaling, and mTORC1 activation (Fig. S3I). In summary, SPP1 is a functional driver of ESCC progression whose expression is consistently associated with aggressive disease biology and treatment resistance across human datasets. SPP1 promotes PD-L1 upregulation via the mTORC1–STAT3 axis The convergence of SPP1-mediated malignancy and immunotherapy resistance prompted examination of its mechanistic interplay with PD-L1, a key immune checkpoint in immune evasion. Analysis of the single-cell data revealed preferential enrichment of CD274 (PD-L1) in SPP1⁺ malignant cells (Fig. 3A), a correlation confirmed at the bulk transcript level in the TCGA-ESCC cohort (R = 0.26, p = 0.021, Fig. 3B). To test causality, we overexpressed SPP1 in ESCC cell lines and found it robustly increased both CD274 mRNA and PD-L1 protein levels (Fig. 3C, D). This effect was recapitulated in vivo, as tumors derived from Spp1-overexpressing cells exhibited elevated PD-L1 by immunohistochemistry (Fig. 3E). To delineate the underlying mechanism, we performed RNA sequencing on control and SPP1-overexpressing cells. SPP1 itself was the most significantly upregulated transcript (Fig. 3F), and pathway analysis of differentially expressed genes identified mTORC1 signaling as a top enriched pathway (Fig. 3G, H). Given the established connection between mTORC1 signaling and STAT3 pathway activation, coupled with the recognized role of STAT3 in regulating PD‑L1 expression in cancer [38], we posited this signaling axis mediates the effect of SPP1. In line with this hypothesis, SPP1 overexpression robustly increased phosphorylation of S6 (a canonical mTORC1 activity marker) and STAT3 (Fig. 3I), demonstrating activation of the mTORC1–STAT3 axis. Collectively, these results establish a direct mechanistic link whereby SPP1 activates the mTORC1–STAT3 signaling cascade to drive PD-L1 expression, providing a molecular basis for its role in immune evasion within heterogeneous tumors. SPP1 upregulates PD‑L1 expression via the mTORC1–STAT3 signaling axis. (A) Bubble plot showing the proportion (dot size) and average expression level (color) of CD274 (PD-L1) across malignant epithelial subtypes in the GSE203115 (left panel) and PRJCA012636 (right panel) scRNA-seq cohorts. (B) Correlation between SPP1and CD274 mRNA expression in the TCGA-ESCC cohort. Pearson’s correlation coefficient and p-value are shown. (C)CD274 mRNA levels measured by RT-qPCR in TE-1 and KYSE-150 cells following SPP1 overexpression (OE) (n = 3). (D) Immunoblot analysis of PD‑L1 protein expression in TE‑1 and KYSE‑150 cells with SPP1 OE. (E) Representative immunohistochemical (IHC) staining and quantification for PD-L1 in tumors from the subcutaneous mouse model (from Fig. 2H) (n = 5 per group). (F) Volcano plot of differentially expressed genes (DEGs) from RNA‑sequencing of TE‑1‑SPP1 OE versus vector‑control cells (n = 3). Significantly up‑ and down‑regulated genes are highlighted. (G) Heatmap depicting expression patterns of all identified DEGs from (F). (H) Hallmark pathway enrichment analysis of genes upregulated by SPP1 OE. The most significantly enriched signaling pathways are shown. (I) Immunoblots showing p-S6, S6, p-STAT3, STAT3, and PD-L1 expression in TE-1 and KYSE-150 cells after vector control or SPP1 overexpression. Data are presented as mean ± SD. **p < 0.01 by unpaired t-test. SPP1⁺ malignant cells facilitate CD8⁺ T cell exhaustion To investigate whether SPP1⁺ malignant cells more broadly orchestrate immunosuppression within the TME, we first analyzed their transcriptional networks. High-dimensional weighted gene co-expression network analysis (hdWGCNA) of malignant epithelial cells identified gene modules M8 and M10 selectively enriched in SPP1⁺ subpopulations (Fig. S4A-F). SPP1 functioned as the topological hub of the key module (M8) associated with antigen presentation and immune regulation (Fig. S4G-J). Initial ligand-receptor analysis predicted that SPP1⁺ malignant cells preferentially interact with CD8⁺ T cells and macrophages (Fig. S4K). Further, CIBERSORT [39] and the T cell state identifier (TCellSI) analyses [40] revealed increased T cell exhaustion and immunosuppressive macrophage-associated signatures in high-SPP1 samples (Fig. S5A-C). LR_cal analysis of TCGA confirmed significant upregulation of the SPP1-CD44 pair in these samples (Fig. S5D-E). We next sought to functionally dissect the consequences of this crosstalk, beginning with its impact on CD8⁺ T cell fate. Re-clustering of CD8⁺ T and NK cells (included due to transcriptional similarity to CD8⁺ T cells) identified five functionally distinct populations: CD8⁺ effector memory T cells (CD8⁺ Tem-GZMK), CD8⁺ exhausted T cells (CD8⁺ Tex-TIGIT), CD8⁺ proliferating T cells (CD8⁺ Tprf-STMN1), CD8⁺ effector T cells (CD8⁺ Teff-IFNG), and NK cells (Fig. 4A). Molecular profiling defined characteristic signatures, with the Tex-TIGIT subset distinguished by high expression of canonical exhaustion markers (PDCD1, HAVCR2, LAG3, TIGIT) and terminal exhaustion signatures (Fig. 4B-G). Pseudotime analysis showed that these exhausted cells were broadly distributed along the differentiation trajectory, supporting the notion that exhaustion represents a dynamic state that emerges during CD8⁺ T-cell differentiation (Fig. 4H). Building on our network predictions, CellChat analysis confirmed and specified this crosstalk, revealing that SPP1⁺ malignant cells preferentially interact with the Tex-TIGIT subset in non-responders, with the SPP1–CD44 axis as one of the key predicted communication pathways (Fig. 4I-J). SPP1⁺ malignant cells drive CD8⁺ T cell exhaustion through tumor-immune crosstalk. (A) UMAP visualization of CD8+ T cell (four subtypes) and NK cell clusters. (B) Bubble plot of top differentially expressed genes across CD8+ T/NK cell subtypes (dot size: fraction of expressing cells; color: expression level). (C) Heatmap of canonical marker gene expression in CD8⁺ T/NK clusters. (D) UMAP feature plots depicting expression of key marker genes in CD8+ T cells. (E) The heatmap illustrates the expression patterns of 19 curated gene signatures across CD8⁺ T/NK clusters. (F) Spatial distribution of four key signatures on UMAP. (G) Bubble plot comparing progenitor and terminal exhaustion scores across CD8⁺ T/NK clusters. (H) The developmental trajectories of CD8⁺ T/NK clusters were inferred by Monocle2. Each dot represents an individual cell, with colors corresponding to different subpopulations. (I) Quantification of significant ligand-receptor interactions between SPP1⁺ malignant cells and CD8+ T cell subtypes in non-responders. (J) Bubble plot of upregulated ligand-receptor pairs between SPP1+ malignant cells and exhausted CD8+ T cells (CD8+ Tex-TIGIT) in non-responders (dot size: p-value; color: communication probability). (K) Exhaustion marker expression (PD-1, TIM-3, and LAG-3) in CD8⁺ T cells exposed to conditioned media (CM) from vector control vs. SPP1-overexpressing ESCC cells (with or without anti-CD44 treatment) (RT-qPCR, n = 3). (L) Secreted IFN-γ and TNF-α in CD8⁺ T cell cultures treated with CM as in (K) (ELISA, n = 3). (M) Representative multiplex immunohistochemistry (mIHC) images for CD8 (green), and PD-1 (red), and DAPI (blue) in ESCC specimens with low and high SPP1 expression (left panel). Scale bars, 100 µm. Quantitative determination of CD8-, and PD-1-positive cells (right panel). (N) Immunoblots showing phosphorylation of P38, ERK and JNK in CD8⁺ T cells treated with CM as in (K). Data are presented as mean ± SD. *p < 0.05, **p < 0.01, ***p < 0.001 by unpaired t-test. This interaction was functionally validated in vitro. Conditioned media from SPP1-overexpressing tumor cells upregulated surface inhibitory receptors (PD-1, TIM-3, LAG-3), while suppressed IFN-γ and TNF-α secretion in CD8⁺ T cells (Fig. 4K-L). Crucially, these immunosuppressive effects were abrogated by anti-CD44 neutralizing antibody, indicating that SPP1-mediated T cell dysfunction is CD44-dependent (Fig. 4K-L). In an independent ESCC cohort, tumors with high SPP1 expression contained significantly more PD-1⁺CD8⁺ T cells (Fig. 4M). While SPP1-CD44 signaling can activate PI3K-Akt or MAPK pathways [[41], [42], [43]], whether tumor cell-derived SPP1 similarly activates these pathways in CD8⁺ T cells to directly mediate exhaustion was unknown. Western blot analysis demonstrated that SPP1-conditioned media activated p38, ERK, and JNK phosphorylation in CD8⁺ T cells in a CD44-dependent manner (Fig. 4N, Fig. S5F). These findings define a pathway in which SPP1⁺ malignant cells drive CD8⁺ T cell exhaustion via the SPP1-CD44 axis, with associated activation of MAPK signaling components, establishing an immunosuppressive microenvironment. SPP1⁺ malignant cells induce immunosuppressive macrophage polarization via α5β1 integrin-mediated pathway We next characterized the interplay between SPP1⁺ malignant cells and tumor-associated macrophages (TAMs). Subclustering identified four distinct macrophage populations: Macro-C1QC, Macro-CCL18, Macro-MKI67, and Macro-SPP1, defined by unique marker profiles (Fig. 5A, B). Polarization analysis revealed that Macro-SPP1 exhibited a clear immunosuppressive phenotype, while Macro-C1QC and Macro-CCL18 displayed mixed inflammatory and immunoregulatory features. Macro-MKI67 showed weak polarization-related signatures [44] but high proliferative activity, consistent with a transitional state (Fig. 5C, D). Functional profiling showed Macro-SPP1 specifically upregulated T-cell inhibitory molecules VSIG4 and LGALS9 (Fig. 5E). Critically, macrophages in non-responder (NR) tumors expressed elevated immune-checkpoint/evasion genes, [18] with the most pronounced increase in Macro-SPP1 (Fig. 5F, G). GSVA further revealed Macro-SPP1 enrichment in angiogenesis, fatty-acid metabolism, glycolysis, and hypoxia pathways (Fig. 5H, I), suggesting HIF-1α–linked metabolic reprogramming that promotes immunosuppression [45]. Monocle2 trajectory analysis positioned Macro-SPP1 at the terminus of macrophage differentiation, consistent with a differentiated immunosuppressive effector state (Fig. 5J). SPP1⁺ malignant cells orchestrate immunosuppressive crosstalk with SPP1⁺ macrophages to confer immunotherapy resistance. (A) UMAP visualization of macrophage subtypes: Macro-C1QC, Macro-CCL18, Macro-MKI67, and Macro-SPP1. (B) Bubble plot showing expression levels (color) and cellular fractions (size) of selected subtype-defining genes. (C) Heatmap of normalized expression of classically described inflammatory/immunosuppressive polarization-associated genes among cell subtypes. (D) Bubble plot showing expression levels (color) and cellular fractions (size) of macrophage polarization-associated signature scores across macrophage subtypes. (E) Bubble plot shows the proportion of cells expressing immune checkpoint and evasion genes in each macrophage subtype. (F) Bubble plot showing the proportion of cells expressing immune checkpoint and evasion genes in the R and NR groups. (G) Bubble plot shows the proportion of cells expressing immune checkpoint and evasion genes in Macro-SPP1 within the R and NR groups. (H) Heatmap showing GSVA scores for 50 hallmark pathways across various macrophage clusters. (I) Density heatmaps illustrating the distribution of hypoxia, glycolysis, fatty acid metabolism, and angiogenesis scores across different macrophage subtypes. (J) The developmental trajectories of macrophages were inferred by Monocle2. (K) Quantification of significant ligand-receptor interactions between Epi-SPP1 malignant cells and macrophage subtypes in non-responders. (L) Bubble plot of upregulated ligand-receptor pairs from Epi-SPP1 to Macro-SPP1 cells in non-responders (dot size: p-value; color: communication probability). (M) Quantification of significant ligand-receptor interactions from Macro-SPP1 to Epi-SPP1 cells in non-responders. (N) Bubble plot of upregulated ligand-receptor pairs from Macro-SPP1 to Epi-SPP1 cells in non-responders (dot size: P-value; color: communication probability). (O) Immunoregulatory state of THP-1-derived macrophages cultured with conditioned media (CM) from control or SPP1-overexpressing ESCC cells, with or without volociximab (anti-α5β1) treatment. Expression of immunosuppressive-associated markers (CD206, CD163) and inflammatory-associated markers (CD86, iNOS) was assessed by RT-qPCR (n = 3). (P) IL-1β and IL-10 secretion by THP-1-derived macrophages treated as in (O) (ELISA, n = 3). (Q) Representative mIHC images for CD68 (green), CD163 (red), and DAPI (blue) in ESCC specimens with low and high SPP1 expression (top panel). Scale bars, 100 µm. Quantitative determination of CD68- and CD163-positive cells (bottom panel). (R) Immunoblot analysis of FAK, AKT, and mTOR phosphorylation in macrophages treated with CM as in (O). Data are presented as mean ± SD. *p < 0.05, **p < 0.01, ***p < 0.001 by unpaired t-test. CellChat analysis suggested that Epi-SPP1 and Macro-SPP1 cells engaged in extensive ligand-receptor interactions in treatment-resistant tumors. Beyond the established SPP1-CD44 axis, we identified multiple predicted SPP1-associated interactions mediated by integrins. Further screening of candidate receptors revealed several potential integrin receptor complexes between Epi-SPP1 and Macro-SPP1, including αVβ5 (ITGAV+ITGB5), αVβ1 (ITGAV+ITGB1), α5β1 (ITGA5+ITGB1), and α4β1 (ITGA4+ITGB1). Expression profiling showed that ITGA5 was relatively enriched in SPP1⁺ macrophages, suggesting that SPP1-associated integrin signaling may be active in this population. In addition, CellChat analysis showed that the SPP1-(ITGA5+ITGB1) receptor complex had a relatively high predicted interaction probability. Based on the combined evidence from receptor enrichment and intercellular communication strength, we prioritized the α5β1 integrin axis as a candidate pathway mediating SPP1-associated signal transduction between malignant cells and macrophages (Fig. 5K, L, Fig. S6A). Bidirectional signaling analysis confirmed active crosstalk via SPP1-CD44 and key integrin dimers, including SPP1-(ITGAV+ITGB1) and SPP1-(ITGAV+ITGB6) (Fig. 5M, N). Functional validation showed conditioned media from SPP1-overexpressing tumor cells promoted immunosuppressive macrophage polarization (Fig. 5O), reduced secretion of pro-inflammatory IL-1β, and increased immunosuppressive IL-10 (Fig. 5P). These effects were abolished by treatment with Volociximab, a monoclonal antibody against integrin α5β1 (Fig. 5O, P). Consistent with these findings, mIHC revealed significantly reduced CD163⁺CD68⁺ macrophage infiltration in SPP1-low versus SPP1-high tumors (Fig. 5Q). Mechanistically, SPP1-conditioned media activated phosphorylation of FAK, AKT, and mTOR in macrophages, signaling events that were attenuated by Volociximab (Fig. 5R, Fig. S6B). These data demonstrate that SPP1⁺ malignant cells educate TAMs toward an immunosuppressive phenotype primarily through activation of the α5β1 integrin–FAK–AKT–mTOR signaling axis. Spatial transcriptomics validates immunosuppressive niches orchestrated by SPP1⁺ malignant cells To validate whether the cellular interactions predicted by our single-cell analyses spatially organize into immunosuppressive niches in situ, we performed high-resolution spatial transcriptomic profiling of an ESCC tumor (Patient #1) using the Visium HD platform (Fig. 6A). We leveraged a public ESCC single-cell RNA-seq dataset (PRJNA777911) as a reference for deconvolution, which contained major cell types including malignant epithelial, T cell, and myeloid populations (Fig. 6B, C). RCTD-based deconvolution generated a detailed cellular map of the tumor microenvironment (Fig. 6D), with cell-type annotations confirmed by the expression of canonical marker genes (Fig. 6E). Spatial mapping of SPP1 expression confirmed its classical association with myeloid cells while also revealing its prominent enrichment within malignant epithelial regions (Fig. 6F, Fig. S7A). Co-localization analysis of the three central cellular players, SPP1⁺ malignant cells, CD8⁺ exhausted T cells, and SPP1⁺ macrophages, revealed their pronounced spatial overlap, defining organized immunosuppressive niches (Fig. 6G). Spatial analysis of two head and neck squamous cell carcinoma (HNSCC) further confirmed the specific co-localization of SPP1⁺ malignant cells with both exhausted CD8⁺ T cells and SPP1⁺ macrophages (Fig. S7B-G). This spatial expression pattern, together with our ligand-receptor analysis, strongly indicates that SPP1⁺ malignant cells, exhausted CD8⁺ T cells, and SPP1⁺ macrophages form organized immunosuppressive niches. Spatial transcriptomics reveals immunosuppressive niches enriched for SPP1⁺ malignant cells. (A) H&E-stained section of an ESCC tumor (Patient #1) subjected to Visium HD spatial transcriptomics (upper panel); spatial distribution of total transcript counts (nCount_Spatial, 16 × 16 µm bins) overlaid on the tissue image (bottom panel). (B) UMAP visualization of the major cell types identified in the public ESCC scRNA-seq dataset (PRJNA777911), used as a reference for deconvolution. (C) Dot plot showing expression of canonical marker genes for each cell type in the scRNA-seq reference (PRJNA777911). Color indicates average expression; dot size indicates the fraction of expressing cells. (D) Spatial map of the Visium HD transcriptome from patient #1, with each 16 × 16 µm bin colored by its predominant cell type as annotated by RCTD deconvolution. (E) Validation of cell-type annotations in the Visium HD data via the expression of canonical marker genes, presented as in (C). (F) Distribution of SPP1 expression across cell types in the scRNA-seq reference (top) and in the spatially resolved Visium HD data (bottom). (G) Spatial co-localization analysis defines immunosuppressive niches. Density plots show the spatial distribution of SPP1⁺ malignant cells (top left), CD8⁺ exhausted T cells (top right), and SPP1⁺ macrophages (bottom left) in patient #1. Two distinct, co-localized regions (Niche 1 and Niche 2) are indicated (dashed circles). High-resolution views of the cellular composition within each niche are shown (bottom right). Blocking SPP1 improves the anti-tumor activity of PD-1 blockade in ESCC To evaluate the therapeutic potential of SPP1 inhibition, we tested whether blocking SPP1 could alleviate T-cell exhaustion and reduce immunosuppressive macrophages, thereby sensitizing tumors to anti-PD-1 therapy. We utilized a syngeneic mouse model in which AKR cells were implanted subcutaneously. Six days post-inoculation, mice were treated every three days via intraperitoneal injection with: IgG isotype control, anti-SPP1, anti-PD-1, or combination therapy (Fig. 7A). Tumor volume was measured on days 6, 9, 12, 15, 18, and 21, with endpoint analysis on day 21. Anti-SPP1 monotherapy significantly suppressed tumor growth compared to the control group, while combination treatment resulted in the most pronounced inhibition (Fig. 7B–D). Survival was extended in all treatment groups, with the most substantial improvement observed in the combination cohort (Fig. 7E). Immune profiling revealed that combination therapy significantly reduced the infiltration of TIM-3⁺CD8⁺ T cells and CD206⁺F4/80⁺ macrophages in the TME (Fig. 7F, G).These results indicate that SPP1 blockade remodels the TME by attenuating T-cell exhaustion and shifting macrophage polarization away from an immunosuppressive phenotype, thereby synergizing with PD-1 inhibition to enhance anti-tumor immunity. The combination of SPP1 inhibition and PD-1 blockade effectively suppresses tumor growth in vivo. (A) Treatment schedule for tumor inoculation and drug administration in C57BL/6 mice (n = 5/group). (B) Representative tumor images from AKR-bearing mice treated with IgG control, anti-SPP1, anti-PD-1, or combination therapy. (C) Tumor growth curves for AKR tumors across treatment groups. (D) Tumors harvested on day 21 post-treatment, and average tumor weights in each treatment group. (E) Kaplan-Meier survival curves for tumor-bearing mice under indicated treatments. (F) Representative mIHC images showing the fluorescence signals of TIM-3 (green), CD8 (red), and DAPI (blue) in tumors from each treatment group. Scale bars, 100 μm. (G) Representative mIHC images showing the fluorescence signals of CD206 (green), F4/80 (red), and DAPI (blue) in tumors from each treatment group. Scale bars, 100 μm. Data are presented as mean ± SD. *p < 0.05, **p < 0.01, ***p < 0.001 by one‑way ANOVA with Tukey's post‑hoc test. Drug sensitivity profiling identifies therapeutic vulnerabilities targeting SPP1⁺ malignant cells To identify therapeutic agents targeting heterogeneous SPP1⁺ malignant cells, drug sensitivity screening identified two potent inhibitor classes. Microtubule-targeting agents Epothilone B and Docetaxel exhibited low IC₅₀ values and high specificity (Fig. S8A), with molecular docking confirming strong binding to SPP1 (docking score = −7.3 for both; Fig. S8B, C). Parallel screening revealed kinase inhibitors Dasatinib and WH-4-023 as highly cytotoxic agents (Fig. S8D), with stable SPP1 binding demonstrated by docking scores of −6.6 and −7.1, respectively (Fig. S8E, F). These therapeutic vulnerabilities designate SPP1 as both a biomarker and direct molecular target in immunotherapy-resistant ESCC. Discussion Our study shifts the paradigm from macrophage-centric to epithelial-centric SPP1 biology in immune evasion. We characterize SPP1⁺ malignant epithelial cells in ESCC, elucidating their critical role as a key mediator of heterogeneity-driven resistance to immunotherapy. Our findings highlight how SPP1 intrinsically enhances malignant properties and PD-L1 expression amidst a heterogeneous cellular landscape. Additionally, SPP1+ cells contribute to immune suppression by establishing immunosuppressive niches, driving CD8⁺ T cell exhaustion through SPP1-CD44 signaling with associated MAPK activation and promoting immunosuppressive macrophage polarization via α5β1 integrin-dependent FAK-AKT-mTOR signaling. The therapeutic relevance of these findings is underscored by the enhanced antitumor effect of combined SPP1 blockade and anti-PD-1 therapy, which reversed both T cell exhaustion and immunosuppressive macrophage features and led to significant tumor suppression in vivo. Our study provides novel insights into a distinct SPP1⁺ malignant cell subset with immune resistance properties, complementing previous reports on SPP1⁺ macrophages and highlighting the underappreciated heterogeneity of epithelial contributions to resistance. This finding aligns with recent work in pancreatic ductal adenocarcinoma (PDAC), which established that SPP1 secreted by epithelial tumor cells acts as a master regulator of cellular communication within the TME [46]. In PDAC, epithelial-derived SPP1 engages the CD61 receptor on mesenchymal cells to initiate a critical signaling loop involving BMP2 and GREM1, which ultimately governs mesenchymal cell fate decisions [46]. Our work in ESCC reveals that epithelium-derived SPP1 also possesses a potent immunomodulatory capacity, driving immune evasion through parallel mechanisms. This convergence of evidence across cancer types solidifies the paradigm of tumor epithelial cells as a primary source of pathogenic SPP1, which can orchestrate diverse microenvironmental processes, from cellular differentiation to immune suppression, depending on the tissue context and the cellular heterogeneity present. Beyond confirming the epithelial source, our study delves deeply into the specific mechanisms of SPP1-mediated immune resistance. Through pseudotime trajectory analysis, we positioned these SPP1⁺ cells as pivotal nodes in cellular differentiation, suggesting they acquire their pro-tumorigenic and immunosuppressive functions through a process of transcriptional reprogramming. This subset was significantly enriched in pathways directly fueling cancer progression and immune evasion, including dysregulated interferon-gamma and TNFα/NF-κB signaling, p53 pathway alterations, and hypoxia-responsive metabolism. However, the contextual factors that determine whether epithelial SPP1 drives primarily immune-modulatory programs (as in ESCC) or cell-fate decisions (as in PDAC) are not yet fully understood. It is plausible that these functions are not mutually exclusive and may operate concurrently, with the dominant phenotype being influenced by the unique pressures of the specific TME. Future research employing single-cell proteomics and more complex preclinical models will be essential to dissect this complexity. In examining large independent datasets, we found that SPP1 is positively correlated with primary non-response to ICB therapies, enrichment of immunosuppressive TAM-associated signatures, and T-cell exhaustion programs. In ESCC, transcriptomic and proteomic analyses further confirm the overexpression of SPP1, establishing it as a key player in a tumor-intrinsic immunosuppressive network that drives malignant progression. Functionally, SPP1 overexpression augments colony formation, proliferation, and migration/invasion while promoting CD8⁺ T cell exhaustion and inducing immunosuppressive polarization in macrophages, accelerating in vivo tumor growth, and markedly increasing tumor-cell PD-L1 at both mRNA and protein levels, supporting a regulatory role of endogenous tumor-cell SPP1 in these processes. These findings are consistent with prior reports showing that OPN/SPP1 upregulates PD-L1 via NF-κB/STAT3 signaling, contributing to immune resistance mechanisms [47]. Whether SPP1 regulates additional checkpoint molecules or co-inhibitory pathways remains to be investigated in future studies. We investigated the mechanisms underlying SPP1-mediated immunotherapy resistance, identifying the SPP1-CD44 signaling axis between epithelial SPP1⁺ cells and exhausted CD8⁺ T cells (Tex-TIGIT) as central to immune suppression within the TME. Previous studies in renal cell carcinoma and hepatocellular carcinoma have suggested that this axis plays a significant role in promoting T-cell exhaustion [41,48]. Our work reveals that SPP1⁺ malignant cells directly promote CD8⁺ T cell dysfunction through the SPP1-CD44 signaling axis, which is accompanied by activation of MAPK effectors (p38, ERK, JNK) in our experimental system. This mechanism is spatially organized, with co-localization of SPP1⁺ epithelial cells and Tex-TIGIT cells within the TME. The coordinated expression of multiple inhibitory receptors (TIGIT, ENTPD1) on exhausted CD8⁺ T cells suggests potential for combined targeting strategies. The functional significance of this pathway is confirmed by the abrogation of exhaustion phenotypes upon CD44 blockade, establishing a direct mechanistic link between tumor-intrinsic SPP1 expression and T cell dysfunction. Parallel to T cell modulation, SPP1⁺ cells drive immunosuppressive macrophage polarization through α5β1 integrin-dependent signaling. This pathway activates the FAK-AKT-mTOR cascade, promoting metabolic reprogramming and immunosuppressive function in macrophages. The predominant expression of ITGA5 in SPP1⁺ macrophages and the specific inhibition by Volociximab highlight the therapeutic relevance of this interaction. Spatial analysis confirms the co-localization of malignant SPP1⁺ cells with SPP1⁺ macrophages, revealing an organized immunosuppressive niche where bidirectional signaling reinforces resistance mechanisms. These interactions promote macrophage polarization toward an immunosuppressive phenotype through the conserved integrin-activated FAK-PI3K/Akt1/mTOR cascade [[49], [50], [51]]. Similar mechanisms have also been documented in nasopharyngeal carcinoma, where SPP1 activates the CD44/JAK2/STAT3 signaling pathway to recruit and polarize macrophages toward immunosuppressive states [52]. SPP1 activation of CD44 and RGD-binding integrins induces FAK/Src-ERK and PI3K-AKT/NF-κB signaling, promoting EMT, invasion, migration, stem-like properties, and therapy resistance [[53], [54], [55]]. Spatial transcriptomics predicted colocalization and ligand-receptor interactions between Epi-SPP1 cells, Tex-TIGIT cells, and MARCO+/SPP1+ macrophages, indicating that the SPP1-CD44/integrin axis links immunosuppressive macrophages to the exhausted T cell niche, collectively driving resistance to immunotherapy. The HNSCC spatial transcriptomic dataset was used as supportive cross-squamous evidence. This comparison was motivated by prior evidence that ESCC shares molecular and immunological features with other squamous carcinomas, particularly those arising from the upper aerodigestive tract[56,57]. Thus, the HNSCC analysis suggests that the spatial niche organization observed in ESCC may reflect a broader squamous tumor architecture. The role of SPP1 in recruiting and polarizing TAMs in ESCC via the reported CD44/integrin–PI3K–AKT axis [58] reinforces its function as a dual hub that links aggressive phenotypes and immune evasion, positioning it as a key target for combined therapies that involve PD-1/PD-L1 blockade. Therapeutically, our data reveal that SPP1 blockade enhances the antitumor efficacy of anti-PD-1 treatment to reverse both T cell exhaustion and macrophage polarization, suppressing tumor growth in vivo [59]. Therapeutic blockade of the SPP1 pathway, either directly via anti-SPP1 antibodies or indirectly via anti-ITGA5 antibodies, demonstrates a stronger antitumor effect when combined with anti-PD-1 treatment than either monotherapy alone. In glioblastoma, administration of an anti-ITGA5 antibody or a small-molecule inhibitor (Cilengitide) significantly reversed immunosuppressive macrophage polarization and T-cell exhaustion [51]. The findings advocate for clinical trials combining SPP1 or ITGA5 blockade with anti-PD-1 therapy in cancers with high SPP1+ cell infiltration. In exploring more effective therapeutic strategies targeting SPP1⁺ malignant cells, we identified microtubule inhibitors such as epothilone B and docetaxel as potent agents with specific cytotoxicity against SPP1⁺ cells. These agents inhibit tumor cell proliferation and migration, disrupting the aggressive behavior of SPP1⁺ cells while modulating their immune-evasive properties, thereby addressing cellular heterogeneity. Additionally, kinase inhibitors such as dasatinib and WH-4-023, which inhibit Src-family kinases and integrin signaling, showed promising cytotoxicity against SPP1+ cells, fitting the biology wherein SPP1 engages CD44/RGD-binding integrins to activate FAK/Src and downstream ERK and PI3K-AKT signaling [60,61]. These findings support a combination approach using microtubule inhibitors alongside Src/FAK-axis inhibitors to effectively target SPP1+ cells and overcome immunotherapy resistance, while additional mechanistic and translational studies are needed before clinical application can be proposed. Despite these promising findings, our study has limitations. The small cohort size in some datasets may limit a comprehensive analysis of the Epi-SPP1 subset and its clinical associations. Future multi-center studies with larger patient cohorts are necessary to validate SPP1⁺ cells as prognostic biomarkers [62]. Furthermore, the interactions between SPP1⁺ cells and other components of the TME, particularly cancer-associated fibroblasts and endothelial cells, remain underexplored. In addition, CellChat analysis predicts potential cell–cell communication based on ligand–receptor expression patterns rather than directly demonstrating functional interactions in vivo, highlighting the need for further in vivo experimental validation of these inferred interactions. Our in vivo experiments provide initial functional validation, but more clinically relevant models, such as patient-derived xenografts and humanized mouse models, will be valuable for better recapitulating the human tumor microenvironment and evaluating SPP1-mediated immunotherapy resistance. Future research incorporating single-cell proteomics, spatial transcriptomics, and additional functional assays will be critical to fully elucidate the spectrum of SPP1-mediated signaling pathways in ESCC. In conclusion, this study identifies SPP1⁺ malignant epithelial cells as key regulators of immune resistance in ESCC, driven by tumor heterogeneity. These cells drive aggressive tumor phenotypes and immune evasion through coordinated regulation of T cell exhaustion and macrophage polarization. The enhanced antitumor effect of combined SPP1 blockade and anti-PD-1 therapy provides a strong rationale for further investigation of combination strategies targeting this pathway. Overcoming SPP1-mediated resistance represents a promising approach to improve outcomes for ESCC patients receiving immunotherapy by addressing the challenges of tumor heterogeneity. Abbreviations ESCC: Esophageal squamous cell carcinoma NACI: Neoadjuvant chemoimmunotherapy TME: Tumor microenvironment scRNA-seq: single-cell RNA sequencing TAMs: Tumor-associated macrophages hdWGCNA: High-dimensional weighted gene co-expression network analysis GSVA: Gene set variation analysis OS: Overall survival PFS: Progression-free survival DFS: Disease-free survival GSA: Genome Sequence Archive TCGA: The Cancer Genome Atlas GEO: Gene Expression Omnibus Ethics approval and consent to participate All study procedures were reviewed and approved by the Ethics Committee of Jinan University (Institutional Review Board Approval No JNUKY-2024-0050). Written informed consent was obtained from all patients in accordance with the Declaration of Helsinki. All animal experiments were approved by the Ethics Review Committee of Jinan University (Approval No IACUC-20230322-05). Consent for publication All authors give consent for publication of the manuscript. Availability of data and material The datasets used and analysed during the current study are available from the corresponding author on reasonable request. Funding The work was supported by a grant in part by the National Natural Science Foundation of China (82403413 to Y.Lin.; 82472858 to L.W.; 82573209 to S.C.); the Guangdong Basic and Applied Basic Research Fund Project (2025A1515012301 and 2026A1515012806 to Y.Lin., 2024A1515012947 to L.W.); the Fundamental Research Funds for the Central Universities (21624303 to H.C.); China Postdoctoral Science Foundation (2025M773863 to X.C.); Natural Science Major Programs for Universities in Anhui Province (2024AH040192); General Project of Anhui Provincial Health Commission (2024Aa20053). CRediT authorship contribution statement Jiakang Ma: Writing – original draft, Validation. Jianlin Zhu: Validation, Methodology. Yiling Chen: Validation, Methodology. Xiaodong Chu: Validation, Methodology. Jie Zeng: Methodology. Hongyu Chen: Methodology. Yuhao Liu: Methodology. Xiaosheng Lin: Methodology. Lei Zhang: Methodology. Jinyuan Yi: Methodology. Guangping Zhang: Writing – original draft, Project administration. Songwang Cai: Writing – original draft, Project administration, Funding acquisition. Yusheng Lin: Writing – original draft, Validation, Project administration. Lu Wang: Writing – original draft, Project administration, Funding acquisition. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements We are very grateful to the team of Zhigang Li for providing us with the raw scRNA-seq data of 14 ESCC neoadjuvant immune checkpoint blockade therapy samples from the Shanghai Chest Hospital cohort (PRJCA012636). Footnotes Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.neo.2026.101331. Contributor Information Guangping Zhang, Email: yfyzgp@163.com. Songwang Cai, Email: SongWCJN@163.com. Yusheng Lin, Email: easonlin0423@126.com. Lu Wang, Email: luellawang@163.com. Supplementary materials References Associated Data Supplementary Materials Data Availability Statement The datasets used and analysed during the current study are available from the corresponding author on reasonable request.