CAFs originate from resident bone-marrow mesenchymal stem cells (MSCs) and adopt heterogeneous phenotypes. CAFs constitute a dominant stromal component in bone sarcomas and are increasingly recognized as co-architects of the immunosuppressive niche (35). scRNA-seq of human OS biopsies resolved two transcriptionally stable CAF subsets-myofibroblastic CAFs (myCAFs, ACTA2^high) and inflammatory CAFs (iCAFs, IL-6^highCXCL12^high), that together account for 25–35% of all non-malignant cells (35). Lineage-tracing studies in genetically engineered mouse models indicate that the majority of CAFs arise from local bone-marrow-derived mesenchymal stromal/stem cells (BM-MSCs) through TGF-β- and bone morphogenetic protein-2 (BMP-2)-driven reprogramming, a process that is molecularly indistinguishable from the osteoblastic differentiation cascade that gives rise to the malignant clone itself (35). Consequently, OS cells and CAFs share an overlapping mesenchymal developmental program, including expression of platelet-derived growth factor receptor (PDGFR)-β, CD90 and the osteoblastic transcription factor RUNX2, suggesting that the same oncogenic cues that initiate the tumor concurrently instruct CAF fate. This ontological convergence explains why CAFs are uniquely equipped to deposit a collagen- and versican-rich ECM that matches the stiffness (Young's modulus 25–35 kPa) of the surrounding mineralized bone, thereby shielding tumor cells from immune attack and impeding chimeric antigen receptor T-cell (CAR-T) cell penetration (36). Two transcriptionally distinct CAF subpopulations have been delineated: myCAFs (ACTA2^high) and iCAFs (IL-6^high CXCL12^high) (37). In OS, myCAFs aligned along collagen bundles associate with increased tissue stiffness (Young's modulus 25–35 kPa) and impaired CAR-T penetration, whereas iCAFs secrete CXCL12 that recruits CXCR4⁺ Tregs (38). Targeting CXCL12 with plerixafor sensitized OS xenografts to disialoganglioside 2 (GD2). CAR-T therapy, increasing intratumoral CAR-T density 4-fold (39). These data highlight CAFs subtyping as a prerequisite for effective stromal reprogramming.

Osteoblast-lineage cells, often dismissed as bystanders, actively shape immune privilege (40). Single-cell interactome analyses revealed 21 ligand-receptor pairs between OS cells and osteoblasts, including RANKL-RANK and Jagged1-Notch2 axes that polarize macrophages toward an M2 phenotype and upregulate PD-L1 (41). Pharmacologic RANKL inhibition with denosumab not only reduced osteoclastogenesis but also decreased M2-TAMs abundance by 40% in a syngeneic OS model, suggesting dual anti-resorptive and immunomodulatory benefits (42). Conversely, osteoclasts fuel a vicious cycle of bone destruction and TGF-β release; TGF-β concentrations >1 ng/ml were sufficient to convert 50% of CD8+ T cells into FOXP3+ Tregs in Transwell assays (43). These observations position osteoclast-directed therapy as a combinatorial partner for TGF-β inhibition.

Bone sarcomas develop a chaotic vasculature characterized by tortuous, CD31-high and NG2-low vessels with poor pericyte coverage (44). Microvessel density is highest in ES (mean 104 vessels mm⁻²) and lowest in CS (42 vessels mm⁻²), paralleling VEGF-A expression levels (45). Yet, bevacizumab achieved only modest disease stabilization [6-month progression-free survival (PFS) 18%] in a phase II OS trial, possibly because VEGF blockade paradoxically increased hypoxia and HIF-1α-driven PD-L1 expression on tumor cells (46). Alternative strategies that normalize rather than prune vessels, such as low-dose metronomic cyclophosphamide enhanced CART cells infiltration in murine ES models without exacerbating hypoxia, underscoring the need to balance anti-angiogenic and immunotherapeutic dosing schedules (47).

MSCs are recruited to primary tumors via CCL5-CCR5 signaling and differentiate into either CAFs or osteoblast-like cells depending on TGF-β and BMP-2 gradients (48). In OS, MSCs-secreted PGE2 induces IDO1 in DCs, leading to a 60% reduction in T-cell proliferation in mixed lymphocyte reactions (49). Genetic ablation of cyclooxygenase-2 (COX-2) in MSCs restored DCs function and improved anti-PD-1 efficacy in vivo (50). These findings suggest that MSCs-educated immune suppression is reversible and that COX-2 inhibition could serve as an adjuvant to checkpoint blockade.

In summary, osteoblast-lineage cells, osteoclasts, MSCs and endothelial cells each contribute distinct immunosuppressive cues, including cytokine secretion, ECM remodeling, and metabolic reprogramming. These interactions not only reinforce tumor survival but also impede effective immune surveillance. A summary of key bone-related cells, their ontogeny, immunosuppressive mediators, and therapeutic implications is presented in Table I (14,34,42,47,50), highlighting potential targets for microenvironment-directed interventions.

Multiplex cytokine profiling of OS biopsies identified a high-TGF-β/low-interferon-γ (IFN-γ) signature that independently predicted poor histologic response to neoadjuvant MAP chemotherapy (53). Tumor-derived IL-6 and CXCL12 establish an autocrine/paracrine loop that recruits CXCR4⁺ MDSCs and polarizes macrophages to an M2 phenotype (54). These soluble factors collectively sculpt an immunosuppressive milieu and establish the rationale for co-targeting IL-6, CXCL12 and TGF-β in combination regimens.

Hypoxia (pO₂ <10 mmHg) affects 40% of OS and 60% of ES lesions, driving HIF-1α stabilization and downstream glycolytic reprogramming (55). A meta-analysis of 487 patients with OS associated elevated HIF-1α to a 2.1-fold increased risk of metastasis. While HIF-1α knockdown radiosensitizes OS cells, it simultaneously upregulates PD-L1, implying that radiation-HIF-1α-PD-L1 combinations warrant evaluation (56). Hypoxia-activated prodrugs such as TH-302 demonstrated modest single-agent activity (disease control rate 28%), but when combined with anti-cytotoxic T-lymphocyte associated protein 4 (CTLA-4), metastatic lung lesions shrank by 50% in a murine ES model (55).

Lactate concentrations >15 mM in OS interstitial fluid, suppressing NKs cytotoxicity by 50% via GPR81-dependent mechanisms (57). Pharmacologic buffering with oral sodium bicarbonate restored NK function and improved anti-GD2 antibody efficacy in vivo, supporting a readily translatable adjunct (58). Pharmacologic buffering with oral sodium bicarbonate restored NK cytotoxicity and enhanced anti-GD2 antibody efficacy in orthotopic OS models, confirming rapid translatability of this metabolic adjunct (58).

Physical barriers and aberrant chemokine gradients cooperatively limit intratumoral T-cell accumulation (59). The ECM of OS is enriched in collagen I/III, fibronectin and hyaluronan; high tissue stiffness reduces T-cell motility and cytotoxicity in vitro and associates with low CD8+ infiltration in vivo (60). Chemokine profiling of 22 treatment-naïve OSs revealed a marked paucity of CXCL9/10 and CCL5 concomitant with overexpression of CXCL12 and CCL18, a pattern that preferentially recruits CXCR4+ Tregs and CCR8+ M2 macrophages while excluding CXCR3+ CD8+ T cells (61). Consistently, spatial transcriptomics showed an inverse association between CXCL9/10 levels and distance of CD8+ T cells from tumor nests (Spearman ρ=−0.63, P<0.01) (62). The clinical implication is that forced intratumoral release of CXCL9 or CXCL10 via oncolytic viruses or STING agonists could enhance ICI efficacy, an approach currently being tested in early-phase trials (NCT05115319).

Tregs constitute 10–25% of CD4+ tumor-infiltrating lymphocytes (TILs) in OS and exhibit an activated phenotype (FOXP3^high, CD25^high and CTLA-4^high) (67). Functionally, Tregs suppress CD8+ proliferation in vitro via IL-10, TGF-β and IL-35 and physically deplete IL-2 through high-affinity IL-2Rα (68,69). Notably, the Treg/CD8 ratio increases from 0.2 in primary tumors to 0.7 in metastases (P<0.001), paralleling reduced objective response rates to anti-PD-1 therapy in metastatic cohorts (8 vs. 26% in localized disease) (70). Depletion models in murine OS (K7M2) show that transient Treg reduction (anti-CD25) doubles CD8+ granzyme-B production and restores sensitivity to PD-1 blockade (71).

M2-polarized TAMs (CD68⁺CD163⁺) and G-MDSCs (CD11b⁺CD15⁺) dominate the bone sarcoma TME (72). TAMs secrete arginase-1, which depletes extracellular arginine and arrests T-cell proliferation at the G₀/G₁ phase (73). In vitro co-culture assays demonstrate that arginase inhibition (nor-NOHA) rescues CD8+ proliferation by 70% (74). MDSCs generate nitric oxide (NO) via iNOS and release ROS; both suppress T-cell receptor (TCR) ζ-chain expression levels and promote T-cell apoptosis (75). Pre-clinical models indicate that targeting the colony-stimulating factor 1 receptor (CSF-1R) pathway or pharmacologic iNOS inhibition markedly reduces the density of these suppressive myeloid subsets and enhances immune-checkpoint blockade efficacy, yet prospective clinical validation of their predictive value and safety remains ongoing (76,77).

Chronic antigen exposure in bone sarcomas drives a progressive exhaustion program (78). sc TCR sequencing revealed oligoclonal expansion (median 47 unique clonotypes per patient) with over-representation of exhausted TCF-1⁻TOX⁺ PD-1^high CD8⁺ subsets (79). Continuous PD-1 signaling, reinforced by TIM-3 and LAG-3, upregulates diacylglycerol kinase-α, dampening TCR signaling (80). Metabolic stress further exacerbates dysfunction: Lactate (15–20 mM) suppresses nuclear factor of activated T-cells (NFAT) nuclear translocation, while adenosine (via CD73/CD39 axis) elevates intracellular cyclic adenosine monophosphate (cAMP), both impairing cytokine production (81). Pre-clinical blockade of the adenosine A2A receptor (A2AR) pathway restores T-cell effector function in hypoxic OS models, providing a rationale for combining adenosine-targeted agents with immune-checkpoint inhibitors in future trials (82).

Conventional type 1 dendritic cells are scarce (<1% of CD45⁺ cells) and display an immature phenotype (CD83^lowCCR7^low) (83).DDMaturation defects are associated with IL-6 and VEGF-A produced by tumor cells, which downregulate co-stimulatory molecules (CD80/86) and impair cross-presentation of tumor antigens (84). Emerging pre-clinical data indicate that intratumoral delivery of a TLR3 agonist can expand cDC1 numbers and amplify CXCL9/10 secretion, thereby enhancing CD8⁺ T-cell infiltration; however, formal clinical validation is still pending (85).

Metabolic reprogramming within the bone sarcoma TME actively suppresses antitumor immunity (86). Tumor cells exhibit high glycolytic flux, generating an acidic, lactate-rich milieu that impairs CD8⁺ T-cell cytotoxicity and cytokine production, notably via inhibition of NFAT signaling and histone acetylation (87). Concurrently, tumor and stromal cell upregulation of IDO1 depletes tryptophan and produces immunosuppressive kynurenine metabolites, activating the aryl hydrocarbon receptor in T cells and driving exhaustion (88). Furthermore, the CD39/CD73-A2AR adenosine axis is a dominant immunosuppressive pathway; adenosine generated extracellularly suppresses T-cell effector functions via elevated intracellular cAMP. Elevated CD73 expression associates with poor response to PD-1 blockade (88,89). These pathways collectively establish a metabolically hostile environment for effector immune cells.

DNA hypermethylation silences perforin 1 (PRF1), granzyme B (GZMB) and IFNγ promoters in CD8⁺ TILs, whereas H3K27me3 deposition represses T-box transcription factor 21 and eomesodermin loci, locking cells in an exhausted state (90). OS-derived exosomal miR-221-3p downregulates suppressor of cytokine signaling 3 in macrophages, enhancing JAK2-STAT3 signaling and sustaining M2 polarization (91). Conversely, pharmacologic inhibition of enhancer of zeste homolog 2 (EZH2; tazemetostat) reactivates PRF1 and GZMB transcription in murine models, improving response to PD-1 blockade (92). These data raise the prospect of combining epigenetic modifiers with immune checkpoint inhibitors (ICIs) to overcome adaptive resistance.

The advent of ICIs, particularly targeting the PD-1/PD-L1 and CTLA-4 axes, revolutionized oncology. However, their application in bone sarcomas, primarily OS, has yielded predominantly modest clinical benefits, underscoring the profound influence of the TME in shaping resistance (93,94).

Across the published pembrolizumab, nivolumab and nivolumab-plus-ipilimumab series, objective responses in bone sarcoma rarely exceed single digits. SARC028, the sentinel multicenter phase II trial, enrolled 22 patients with OS or CS and documented only one partial response lasting 32 weeks, yielding an objective response rate of 5% (93). Two confirmatory studies have since reproduced this figure: The PEMBROSARC cohort of 30 heavily pre-treated (≥2 prior lines of systemic cytotoxic chemotherapy or targeted therapy) patients with OS had no objective responses at all (94), while the pediatric ADVL1412 expansion treated 14 patients with OS with nivolumab plus ipilimumab and observed just one durable partial response (objective response rate, 7%) (95,96). Single-arm studies using pembrolizumab monotherapy (Boye et al (97), 28 patients) or the nivolumab-sunitinib combination [Palmerini et al (98); 15 patients] produced identical null response rates.

The mechanistic explanation for this uniformly low activity lies in three interconnected features of the bone-sarcoma TME detailed in ‘Mechanisms of immunosuppression within the bone sarcoma TME’, immunologically ‘cold’ tumors: Spatial mapping shows PD-L1-expressing cells physically separated from CD8+ CTLs, creating an excluded infiltrate refractory to checkpoint reactivation (99). Genomic quietness: Low mutational burden and minimal neoantigen presentation deprive T cells of the antigenic signal needed for ICI-driven expansion (93). Pronounced myeloid dominance: The SDF-1/CXCR4 axis recruits monocytic MDSCs that outcompete CTLs for glucose and secrete arginase-1, directly neutralizing anti-PD-1 efficacy in pre-clinical models (100). Notably, the lone responder in SARC028 had chondroblastic OS with high PD-L1 expression (≥50% tumor cells) and brisk peritumoural CD8 infiltration, illustrating that exceptional responders may arise when the tumor deviates from the archetypal ‘cold’ bone-sarcoma microenvironment (93).

PD-L1 expression remains the most interrogated biomarker, yet results are inconsistent. Discordant PD-L1 status between primary and metastatic lesions occurs in 35% of patients with OS, undermining archival tissue reliability (101). In PEMBROSARC, PD-L1 positivity (≥1%) was present in 30% of cases but was unassociated with benefit (94). Conversely, Le Cesne et al (94) reported that high PD-L1 combined with IDO1 expression enriched for a 10% long-term disease-control subgroup, hinting at combinatorial biomarker utility. Spatial transcriptomics now suggests that the spatial proximity of PD-L1⁺ macrophages to CTLs, rather than absolute PD-L1 levels, predicts non-progression (99). Beyond PD-L1, an ‘immune-hot’ signature (CD8⁺ CTL/Treg ratio >2 and low M-MDSC density) emerged as an independent predictor of PFS in the nivolumab-plus-ipilimumab ADVL1412 cohort (HR, 0.31; 95% CI, 0.12–0.78) (96). However, this signature was present in only 6/38 (16%) sarcoma biopsies, underscoring the rarity of immunologically favorable bone sarcomas.

Emerging evidence suggests mast-cell stabilization as a novel resistance mechanism. Mast cells are tissue-resident, c-Kit⁺FcεRI⁺ innate immune cells that differentiate in the bone marrow, circulate as progenitors and mature locally under the influence of stem-cell factor, IL-3, IL-33 and CXCL12. In bone sarcomas, mast cells preferentially localize to the hypoxic tumor margin and along neo-vessels, where they act as central hubs integrating stromal and immune signals (102). Upon activation, mast cells rapidly degranulate histamine, tryptase and chymase, and additionally secrete IL-4, IL-10, VEGF-A, PDGF-B and TGF-β1. These mediators collectively enhance vascular permeability, stimulate CAF proliferation and promote M2-like macrophage polarization, thereby reinforcing an immunosuppressive niche that blunts cytotoxic T-cell influx. Pharmacological stabilization of mast cells with cromolyn or targeted depletion via c-Kit inhibition restores CD8⁺ T-cell infiltration and sensitizes OS xenografts to anti-PD-1 therapy, underscoring their therapeutic relevance (44). Mast-cell-derived IL-10 and TGF-β convert macrophages toward M2 and exclude CTLs; pharmacologic stabilization restored PD-L1 blockade sensitivity pre-clinically (44), validation in human studies is required. Finally, host-related factors (human leukocyte antigen loss of heterozygosity, low peripheral blood T-cell clonality) associated with primary resistance in soft-tissue sarcomas warrant investigation in bone sarcomas via ongoing multi-omics studies (such as NCT 04339738) (34,35,103).

In summary, current ICI trials confirm modest single-agent activity in bone sarcomas, driven by an immunologically ‘cold’ and myeloid-dominated TME. Objective responses cluster in rare cases with inflamed, PD-L1-high, CTL-rich tumors, but robust predictive biomarkers are still lacking. Future efforts must integrate spatially resolved TME profiling, systemic immune monitoring and rational combination strategies to overcome primary resistance.

T cell receptor-engineered T (TCR-T) cell therapy and TIL therapy represent alternative ACT strategies, but their application in bone sarcomas is less advanced than CAR T therapy and faces distinct challenges. TCR-T cells can target intracellular antigens presented by MHC molecules, potentially broadening the targetable antigen repertoire. Lu et al reported a clinical trial using an MHC class II-restricted TCR targeting the cancer germline antigen MAGE-A3. While the trial included patients with various types of metastatic cancer, one patient with metastatic sarcoma (unspecified subtype) achieved a complete response lasting >24 months, demonstrating the potential therapeutic activity of TCR-T cells in sarcoma (127). Recently, Hamada et al (128) developed TCR-T cells targeting the sarcoma-associated antigen papillomavirus binding factor (PBF). These TCR-T cells demonstrated specific cytotoxicity against PBF-positive sarcoma cell lines in vitro and inhibited tumor growth in xenograft models. Watanabe et al (114) targeted the cancer stem cell antigen DNAJB8 with TCR-T cells, showing efficacy against various solid tumors in vitro and in vivo, though specific bone sarcoma data were limited.

TCR-T therapy faces considerable hurdles in bone sarcomas. Key challenges include identifying truly tumor-specific intracellular antigens shared across subtypes, dependence on tumor MHC expression, intratumoral heterogeneity enabling escape and the critical risk of off-target toxicity due to TCR cross-reactivity (114,128). Furthermore, as with CAR T cells, TCR-T cells must overcome the profoundly immunosuppressive TME to traffic, persist and function effectively, with specific bone sarcoma TME interaction data currently limited.

Pre-clinical studies show that oncolytic viruses can control OS growth, yet the immunocompetent microenvironment rapidly blunts efficacy (129–131). Intratumoral Semliki Forest virus encoding HSV-TK shrank orthotopic 143B-luc tumors by 74% in nude mice, but complete regressions were absent and viral genomes disappeared within 7 days, leaving the contribution of antitumor immunity unaddressed (129). Similarly, intravenous reovirus cleared Ras-activated xenografts yet failed against A673 or SJSA-1 lesions, and its benefit was lost in syngeneic K7M2 mice owing to neutralizing antibodies and T-cell-mediated viral clearance (130).

Systemic delivery is further complicated by metastatic burden. Christie et al (131) delivered mLIGHT-armed myxoma virus intravenously to syngeneic 66.1 lung-metastatic OS; median survival doubled when nodule counts were <100, but mice with >200 lesions showed rapid rebound growth driven by hypoxia-associated PD-L1 upregulation. These data indicate that high antigen load and immune suppression within the bone sarcoma TME impose a clear therapeutic ceiling.

Large-animal studies corroborate these limitations. Laborda et al (132) treated seven spontaneous canine OSs with intratumoral ICOVIR-16K; two dogs achieved partial metabolic responses and one complete pathological response, yet neutralizing antibody titres >1:1,000 were detected after the second dose, coinciding with viral clearance in serial biopsies. The findings underscore the challenge of maintaining viral replication in immunocompetent, outbred hosts and reinforce the need for transient B-cell depletion or alternative dosing schedules.

Live-attenuated Listeria monocytogenes vaccines offer a complementary strategy. Musser et al (133) administered Lm-LLO-E7 to 15 OS-bearing dogs; although only mild pyrexia and lymphopenia occurred, median survival appeared prolonged vs. historical controls (234 vs. 180 days). However, lack of randomization and heterogeneous adjuvant chemotherapy preclude definitive efficacy claims and anti-Listeria immunity may limit booster efficacy.

Finally, antigen heterogeneity threatens peptide-based vaccination. Chen et al (134) screened 87 treatment-naïve OSs; melanoma-associated antigen A4, New York esophageal squamous cell carcinoma-1 and preferentially expressed antigen in melanoma were expressed in 28, 20 and 15% of tumors respectively, with only 6% co-expressing ≥2 antigens. Taken together, these studies highlight the dual hurdles of TME-mediated immune suppression and tumor-intrinsic antigen diversity; rational combinations of oncolytic viruses or vaccines with checkpoint blockade, radiotherapy and personalized neoantigen libraries are required to unlock their clinical potential.

High-dimensional analyses concur OS is dominated by arginase-1⁺ M2-TAMs and Ly6G⁺ monocytic MDSCs (61,145). VEGF-A and P2RX7 signaling fuel tumor growth and instruct M2 polarization via PI3K/Akt/mTOR (135,136). Silencing VEGF-A (135) or inhibiting P2RX7 (136) reduced tumor growth and CD163⁺ macrophages by 50% in OS models. However, in ES, targeting extracellular vesicles restored DC maturation only without TGF-β, indicating context-dependent myeloid reprogramming signals (146). These convergent yet conditional data argue that multi-pathway blockade is required for durable reprogramming.

Translation strategies: Systemic CSF-1R inhibition (pexidartinib) eradicated >70% M2-TAMs in mice, tripled CD8⁺ TILs, but caused grade-3 hepatotoxicity in 25% of pediatric patients with only transient responses (145). While promising, repeated intralesional injection is impractical for multifocal disease, and long-term immunogenicity is unknown.

Collagen cross-linking enzyme LOXL2 is a central regulator of ECM stiffness in OS. LOXL2 knockdown reduced stiffness (35 to 15 kPa), doubled doxorubicin penetration and improved CAR-T infiltration (145). Paradoxically, partial ECM degradation increased metastasis by releasing VEGF-sequestered matrices (138), suggesting indiscriminate collagenolysis is counterproductive. This underscores the need for spatiotemporally controlled strategies, such as photo-activatable MMP-9 inhibitors.

CAFs fortify barriers via paracrine TGF-β and CXCL12. CLTC-TFG signaling in CAFs upregulates TGF-β, excluding CD8⁺ T cells (139). The bifunctional anti-PD-L1/TGF-β antibody TQB2858 achieved a 33% objective response rate in relapsed OS (140), yet CAFs heterogeneity predicted variable CXCL12 levels and inconsistent T-cell recruitment. To address plasticity, membrane-anchored IL-12 T cells converted FAP⁺ CAFs to an inflammatory phenotype, reduced collagen by 60% and improved CAR-T persistence (141). Unfortunately, sustained IL-12 induced cachexia in 20% of mice, necessitating inducible expression systems.

VEGF-A overexpression drives chaotic vasculature (135,142). Anti-VEGF therapy normalizes vessels preclinically, but clinical bevacizumab monotherapy yielded only 18% 6-month PFS. Mechanistically, it exacerbated hypoxia, upregulated HIF-1α, PD-L1 and expanded MDSCs (145). Conversely, low-dose metronomic cyclophosphamide increased pericyte coverage and doubled CAR T cells infiltration without worsening hypoxia, illustrating the importance of dose scheduling.

Hypoxia-activated prodrugs such as TH-302 kill hypoxic cells. Combined with anti-PD-1, it improved survival in metastatic OS mice (55). However, variable hypoxic thresholds and a lack of consensus imaging biomarkers hinder application. Integrating ¹⁸F-MISO PET with circulating hypoxia gene signatures may stratify responders, but needs validation.

IDO1 is upregulated in 60% of OS and associated with kynurenine-mediated T-cell suppression (143). Epacadostat plus pembrolizumab produced a 33% ORR, yet IDO1 expression did not predict benefit, possibly due to compensatory tryptophan 2,3-dioxygenase activation (145). Dual IDO/TDO inhibitors are under investigation, with off-target hepatic toxicity a concern.

Elevated lactate suppresses NFAT signaling and granzyme-B transcription (147,148). Monocarboxylate transporter 1/4 blockade restored CD8⁺ function ex vivo, but caused rebound glycolysis and hyperlactatemia in mice (147). Adenosine pathway blockade enhanced GD2-CAR-T expansion and yielded partial responses in 2/6 canine patients with OS (145); however, variable CD73 expression (>45% across biopsies) emphasizes the need for companion diagnostics.

EZH2 overexpression silences MHC-I and T-cell chemokine genes (144). EZH2 inhibitor tazemetostat restored HLA-A/B/C expression and improved CAR-T recognition; combined with anti-PD-1, eradicated tumors in 60% of mice (144). Continuous dosing raised stem-cell toxicity concerns, necessitating intermittent delivery. Histone deacetylase (HDAC) inhibition (entinostat) increased CXCL9/10 and reduced Tregs, but caused dose-limiting thrombocytopenia. DNA methyltransferase (DNMT) inhibitor decitabine demethylated IFNG/GZMB promoters in exhausted TILs, yet myelosuppression halted dose escalation (145). Collectively, epigenetic therapy is a promising immune-sensitizing adjunct, but optimal scheduling and predictive biomarkers remain undefined.

Conventional preclinical models, including cell monolayers and immunodeficient xenografts, inadequately recapitulate the spatial, mechanical and immunosuppressive features of bone sarcoma TME. In vitro 2D cultures fail to model mineralized bone matrices, which physically impede CAR-T cell trafficking and alter metabolic crosstalk between tumor cells and stromal components such as osteoblasts or marrow adipocytes (149,150). Murine xenograft models (such as NOD-SCID) lack functional human immune systems, rendering them incapable of evaluating immunotherapies dependent on human-specific antigen presentation or checkpoint interactions. For instance, such models could not predict the clinical failure of HER2-targeted CAR T cells in OS, which later was attributed to TME-driven exhaustion markers (PD-1, TIM-3) and TGF-β-mediated suppression (149,151). Even syngeneic mouse models exhibit interspecies discrepancies in cytokine networks; the TGF-β/osteopontin (OPN) axis is key for osteoclast-mediated immunosuppression in human bone metastases, poorly mirrored in murine systems, leading to overstated efficacy of immune checkpoint blockade in bone tumors (152). These limitations underscore a key gap: Conventional models cannot simulate the dynamic immunosuppressive reprogramming induced by bone-specific niches, resulting in unreliable therapeutic predictions.

To address these shortcomings, genetically engineered and humanized murine platforms incorporating functional human immunity have emerged (153,154). huHSC-NOG mice reconstituted with human hematopoietic stem cells enable long-term (≥6 months) study of human T-cell differentiation and memory responses, making them ideal for evaluating sustained immunotherapy efficacy (153). For example, F. Hoffmann-La Roche Ltd's CEA-TCB bispecific antibody combined with atezolizumab (anti-PD-L1) showed synergistic tumor reduction in huHSC-NOG models of colorectal cancer, a finding later validated clinically (155,156). Conversely, huPBMC-NOG models, which rapidly reconstitute mature T cells within 2–4 weeks, are suited for short-term assessments of T-cell-engaging therapies such as blinatumomab (157). However, their utility is constrained by graft-vs.-host disease) and deficient B-cell reconstitution, limiting studies on antibody-dependent cellular cytotoxicity (158).

Advanced 3D engineered systems now integrate biomechanical cues to mimic bone TME pathophysiology. Stiffness-tunable hydrogels (such as poly-aspartate scaffolds) revealed that matrix rigidity >40 kPa, mimicking mineralized bone, induces YAP overexpression in OS cells, synergizing with TNF-α from TAMs to drive STAT3-mediated chemoresistance (159). These platforms also demonstrate context-dependent therapeutic vulnerabilities; STAT3 inhibition reversed doxorubicin resistance in stiff 3D models but showed minimal efficacy in 2D cultures (151). For giant cell tumor of bone (GCTB), mass cytometry of patient-derived samples identified denosumab-induced depletion of γδTCR+ osteoclast-like giant cells and expansion of cytotoxic CD8+ T cells, a shift not reproducible in non-humanized models (160). Despite progress, discrepancies persist: while 3D models predict enhanced CAR-T cytotoxicity under dynamic flow, clinical translation remains hampered by poor infiltration into osteoid-rich niches 5.

sc and spatial omics technologies are dissecting the cellular heterogeneity and topographical drivers of immunotherapy resistance in bone sarcomas. Mass cytometry (CyTOF) of GCTB identified a spatially restricted PD-1hiTIM-3+CD69+ CD8+ T-cell population juxtaposed with SIRPα+ TAMs, suggesting localized exhaustion within the TME (161). In OS, multi-omics integration of transcriptomic and sc data defined a palmitoylation-driven metabolic-immune subtype characterized by ZDHHC3/21/23 upregulation, which induces ‘cold tumor’ phenotypes via dual suppression of MAPK signaling and CD8+ T-cell infiltration. High-risk patients, as defined by a palmitoylation-driven prognostic score (PPS) above the 75th percentile, showed resistance to PD-L1 inhibitors in the IMvigor210 cohort, highlighting this subtype's clinical relevance (162).

Spatial transcriptomics further maps cytokine gradients underpinning immunosuppressive niches. In bone metastases, osteoclast-derived OPN creates CXCL12-rich ecological niches that sequester Tregs and exclude cytotoxic T cells, a mechanism validated through in situ hybridization in clinical biopsies (152). Radiation modeling in microphysiological systems revealed that proton therapy amplifies localized TGF-β secretion from damaged osteoblasts, spatially associating with PD-L1 upregulation in surviving tumor cells (159). However, technological constraints remain: Current spatial methods lack resolution for bone's mineralized matrix and sc datasets are biased toward non-mineralized zones due to tissue dissociation artifacts (163). Discrepancies also arise in immune subset classification; CyTOF-defined TAM subsets in GCTB contradict scRNA-seq annotations due to antibody specificity issues vs. dropout effects in sequencing (164,165).