Glycolysis-driven cancer-associated fibroblasts shape T-Cell exclusion via the CXCL16–CXCR6 Axis: a metabolic-chemokine mechanism in tumor immunity

Abstract

Background

Immune-excluded tumors are characterized by abundant CD8⁺ T cells at the invasive margin but scarce infiltration within tumor nests, leading to limited responses to immunotherapy. Emerging evidence identifies cancer-associated fibroblasts (CAFs) as key mediators of immune exclusion.

Main body

This review highlights a glycolysis-driven CAFs (glyCAFs) pathway that orchestrates immune exclusion. Glycolysis CAFs upregulate Glucose Transporter 1 (GLUT1) to sustain CXCL16 secretion, which in turn engages CXCR6 on CD8⁺ T cells and effectively traps them at the tumor margin. We summarize the evidence supporting this glyCAF–GLUT1–CXCL16–CXCR6 signaling circuit and its functional impact on T-cell positioning. A distinct advantage of this axis lies in its ability to integrate a targetable metabolic phenotype with a spatially measurable immunologic outcome. We discuss potential spatial biomarkers—such as glyCAFs enrichment at the margin, elevated GLUT1 and CXCL16 expression, and CXCL16-high stromal niches closely associated with CD8⁺ T cells—and outline therapeutic strategies aimed at modulating this pathway. Pharmacologic inhibition of GLUT1 (e.g. BAY-876) can suppress glycolysis and CXCL16 production, while blockade of CXCL16 or CXCR6 may release T-cell retention and enhance responses to chemotherapy, radiotherapy, and immune-checkpoint blockade. Finally, we highlight open questions, regarding the cellular origins and regulatory networks of glyCAFs, the classification of the glyCAFs state as a transient adaptation or a stable entity and its biomarker potential, the spatial mechanisms of glyCAF–immune interplay through integrated metabolic and proteomic mapping, the preclinical validation of multi-target strategies informed by spatial biomarkers, and the preclinical validation of combinatorial therapeutic strategies using biomarker-guided approaches.

Conclusions

The glyCAF–GLUT1–CXCL16–CXCR6 axis establishes a clear mechanistic and translational framework linking stromal metabolism to immune spatial architecture, paving the way for precision immunometabolic strategies to overcome T-cell exclusion in solid tumors.

Introduction

Immune-excluded tumors harbor abundant cytotoxic T lymphocytes, particularly CD8⁺ T cells, at the invasive margin but few within tumor nests, thereby defining a spatially and prognostically distinct tumor microenvironment (TME) that differs fundamentally from the immune-inflamed phenotype [1–3]. Immune exclusion in tumors typically involves two sequential bottlenecks. The first is an access barrier, where physical constraints such as dense extracellular matrix and non‑permissive vasculature limit the number of effector T cells that reach the tumor–stroma interface [4–6]. Even if T cells arrive at the margin, a second retention barrier can trap them in stromal niches via chemokine and adhesion signals, preventing their entry into tumor nests [7]. Clinical benefit correlates with the density of CD8⁺ T cells in the tumor parenchyma rather than their numbers at the margin; when T cells remain confined to the margin, responses to immunotherapy are limited, making this stromal retention a central obstacle to immunotherapy efficacy [2]. Accumulating evidence indicates that immune exclusion, including resistance to immune checkpoint therapy, often results from the failure of effector T cells to infiltrate the tumor core, ultimately causing their retention in the stromal or tumor margin. Addressing this spatial barrier is essential for converting non-responders to responders in immunotherapy. Among various stromal components contributing to immune exclusion, cancer-associated fibroblasts (CAFs) have emerged as dominant architects shaping the spatial immune landscape [8–10]. Indeed, CAFs have become recognized as principal mediators of immune exclusion as well as central organizers of the TME, contributing significantly to tumor progression, metastasis, and therapy resistance [11].

Recent single-cell and spatial omics studies have further delineated that CAFs can be categorized into functionally distinct subtypes: myofibroblastic, matrix-remodeling, and immunomodulatory. This classification offers a conceptual framework for understanding CAFs heterogeneity and its role in shaping TME [12]. Within the tumor microenvironment, oncogenic stressors and inflammatory signals drive metabolic reprogramming in CAFs, leading to a glucose-dependent, highly glycolytic phenotype known as glycolysis-driven CAFs (glyCAFs), as a metabolic state within the CAF continuum. This metabolic shift is consistent with the “reverse Warburg effect”, in which tumor-derived factors such as ROS and hypoxia induce mitochondrial dysfunction and aerobic glycolysis in fibroblasts. These glyCAFs subsequently export metabolites such as lactate and pyruvate, supporting tumor growth and therapy resistance [13–15]. Notably, recent studies in soft-tissue sarcomas have linked the glycolytic program in CAFs to immune exclusion. GlyCAFs rely on Glucose Transporter 1 (GLUT1) to sustain high expression of CXCL16, a stromal chemokine previously implicated in monocyte recruitment and stromal activation in triple-negative breast cancer [16]. CXCL16 accumulates at the tumor margin, where it engages CXCR6 on T cells, promoting their peripheral retention and limiting infiltration into the tumor parenchyma. Genetic or pharmacological disruption of GLUT1 reduces CXCL16 production, and blockade of the CXCL16–CXCR6 axis restores CD8⁺ T cells migration into tumors and enhances chemotherapy response [17, 18]. These findings position the glyCAF–GLUT1–CXCL16–CXCR6 axis as a novel, targetable immune exclusion pathway to break the retention barrier at the tumor margin. This mechanism aligns with broader evidence that CXCR6 governs the spatial distribution and residency of effector T cells in solid tumors [19].

In this review, we discuss how glycolytic metabolic reprogramming in CAFs upregulates local CXCL16 chemokine signaling, thereby directing CD8⁺ T‑cell accumulation within the immune‑excluded tumor margin, and discuss how this spatial regulation can be translated into therapeutic strategies. We further detail this peripheral retention mechanism and detection approaches whereby glyCAFs-secreted CXCL16 engages CXCR6 on CD8⁺ T cells, and we distinguish glyCAFs from canonical CAF subtypes by highlighting their functional and transcriptional heterogeneity. We also systematically compare glyCAFs with established CAF subtypes, clarifying their unique position within the broader CAF taxonomy and refining our understanding of stromal cell heterogeneity. Additionally, we compare this axis with other chemokine pathways to underscore its distinct mechanistic and therapeutic features. Finally, we critically evaluate the translational potential of targeting this axis, including the promise and challenges of dual-targeting strategies in preclinical and clinical studies. By synthesizing this information, we aim to provide a new theoretical and experimental reference for developing novel combination immunotherapies. Overall, this review provides a conceptual and experimental framework linking CAF metabolism to immune exclusion and identifies translational opportunities for precision immunometabolic therapies.

Characteristics and Identification of glycolysis cancer-associated fibroblasts (glyCafs)

Cancer-associated fibroblasts are heterogeneous and encompass multiple phenotypic states. The term ‘glyCAFs’ refers to a functional CAF program characterized by persistent activation of glycolysis and heightened glucose utilization. This represents a metabolically plastic state that can be reprogrammed by targeted perturbation(2). GlyCAFs can be operationally defined across three complementary levels: molecular features and detection, spatial ecology, and functional validation. At the molecular level, they display a glycolysis-enriched transcriptomic profile; spatially, they localize to tumor margins; functionally, they modulate immune cell positioning and therapy resistance.

Induction and regulation of the glyCafs phenotype

GlyCAFs are a metabolically distinct fibroblast subtype, hallmarked by elevated expression of GLUT1 (SLC2A1) and coordinated activation of glycolysis gene programs [17]. A summary of the triggers and functional outcomes that shape the glyCAFs phenotype is provided (Table 1). In hypoxic invasive margins and fibrotic niches, HIF-1α activation transcriptionally upregulates GLUT1, broadly reprogramming glucose metabolism and predisposing fibroblasts to a glycolytic phenotype. Moreover, as chemokine signaling is highly sensitive to hypoxia, HIF-dependent CXCL16 induction directly links metabolic stress to immune cell positioning [16, 30–33].

Table 1.

From microenvironmental stress to glyCafs: triggers and outcomes

Trigger/signal	Central hubs	Metabolic executors	Readouts and outcomes	Key refs
Hypoxia; inflammatory stress	hypoxia-inducible factor-1α(HIF-1α)	GLUT1↑; HK2; PFKFB3; LDHA; MCT4; CAIX	Higher glucose uptake; higher lactate export; lower pH (acidic TME); impaired effector T-cell function.	[20–22]
Paracrine growth factors (TGF-β1, PDGF)	IDH3α↓, succinate↑, prolyl-hydroxylase inhibition, HIF-1α stabilization	GLUT1; HK2; PFKFB3; LDHA	Switch from OXPHOS to aerobic glycolysis; miR-424 drives IDH3α loss.	[23]
ROS; Caveolin-1(Cav-1)loss(Reverse Warburg effect)	HIF-1α; NF-κB (context dependent)	MCT4↑; glycolytic enzymes up	Stromal Cav-1 low/MCT4 high; increased lactate shuttling; association with poor prognosis.	[15, 24, 25]
Extracellular matrix stiffness/mechanics	YAP/TAZ (mechanotransduction)	GLUT transporters; PFKFB3 (and other glycolysis genes)	Nuclear YAP/TAZ; increased expression of glycolytic genes.	[26, 27]
Lactate/acidosis feedback	HIF-1α stabilization; immune-metabolic crosstalk	MCT4; reinforced glycolytic program	Suppressed effector T-cell function; enhanced myeloid immunosuppressive networks.	[28, 29]
Tumor cues in soft-tissue sarcoma (STS)	GLUT1-dependent glycolytic program	GLUT1; glycolytic enzymes; CXCL16 upregulation downstream	CXCL16 enriched at the margin; retention of CXCR6+ CD8+ /T_RM at the edge; glycolysis inhibition increases intratumoral entry and improves chemotherapy response.	[17]

Abbreviations: ↑ upregulated; ↓ downregulated; CAIX, carbonic anhydrase IX; Cav-1, caveolin-1; HIF-1α, hypoxia-inducible factor-1α; LDHA, lactate dehydrogenase A; MCT4, monocarboxylate transporter 4; NF-κB, nuclear factor kappa-B; OXPHOS, oxidative phosphorylation; PDGF, platelet-derived growth factor; PFKFB3, 6-phosphofructo-2-kinase/fructose-2,6-bisphosphatase-3; ROS, reactive oxygen species; STS, soft-tissue sarcoma; TME, tumor microenvironment; T_RM, tissue-resident memory T cell

The integration of single-cell RNA sequencing with spatial transcriptomics shows that glyCAFs preferentially occupy the invasive margins of immune-excluded tumors in close spatial proximity to CD8⁺ T cells, thereby pointing to a metabolically driven spatial barrier that limits T cells infiltration [12, 34]. Functional studies in soft-tissue sarcoma confirm that inhibition of GLUT1 or glycolytic flux reduces glyCAFs abundance and CXCL16 levels, restores CD8⁺ T-cell infiltration into the tumor core, and improves therapeutic responses [17]. Collectively, these findings establish a metabolic–spatial–immune triad wherein stromal metabolism maintains the spatial constraints on T cells infiltration, revealing actionable targets at both metabolic and spatial levels.

Single-cell and metabolic imaging approaches for localizing glyCafs

ScRNA-seq enables unbiased identification of the transcriptional program that defines glyCAFs state. This state is characterized by high expression of SLC2A1 (encoding GLUT1) along with canonical glycolytic enzymes such as HK2 and LDHA, which serve as practical markers for defining the phenotype [17, 35, 36]. By integrating spatial transcriptomics with scRNA-seq data, this glyCAFs subset was mapped onto spatial tissue coordinates, revealing significant enrichment at the invasive tumor margin [37–40].

At the histologic level, multiplex immunofluorescence (mIF) and spatial omics have been employed to visualize the co-localization of glyCAFs and CD8⁺ T cells in situ. Using markers such as CD90 and CD73 to identify glyCAFs, and CD8 for cytotoxic T cells, mIF reveals their spatial co-localization at the invasive margin. To quantitatively assess this interaction, the nearest-neighbor distance between CD73⁺ CD90⁺ glyCAFs and CD8⁺ T cells was measured directly on tissue images, supporting the concept that glyCAFs spatially retain CD8⁺ T cells at the margin [17, 41, 42].

Beyond spatial localization, functional and metabolic validation of glyCAFs has relied on imaging-based approaches to assess glucose uptake. The fluorescent glucose analog 2-NBDG, for instance, is widely used in flow cytometry and tissue section imaging as a proxy for glycolysis activity and is often correlated with GLUT1 expression to identify glycolytic programs in vivo [43–45]. In living mouse tumors and tissue models, 2-NBDG accumulation has been observed in fibroblasts, aligning with transcriptomic evidence of glycolysis and supporting their classification as glyCAFs in situ [45]. However, a growing body of genetic and pharmacologic evidence calls into question the specificity of 2-NBDG as a reporter of GLUT-mediated transport. In myeloma cells, CRISPR-Cas9 knockout of GLUT1 abolishes radiolabeled glucose uptake but does not affect 2-NBDG incorporation [44, 46]. Similarly, glucose transporter inhibitors such as cytochalasin B, WZB-117, and BAY-876 fail to suppress 2-NBDG signal in multiple models. This transporter-independent uptake has also been documented in L929 fibroblasts, human astrocytes, and T cells, where neither competitive glucose analogs nor GLUT blockers consistently reduce 2-NBDG labeling [47, 48]. In light of these findings, 2-NBDG should therefore be interpreted as a qualitative, transporter-independent indicator of glucose analog uptake and should be used with caution when inferring transporter-specific flux. Consequently, there is a pressing need to identify alternative quantitative metrics capable of specifically assessing glycolysis activity in glyCAFs. Together, these multi-scale approaches—from transcriptomic mapping to metabolic imaging—provide convergent evidence for the spatial and metabolic identity of glyCAFs.

Mechanism: the glyCafs–CXCL16–CXCR6 axis retains CD8⁺ T cells at the tumor margin

Building on the molecular and spatial characterization of glyCAFs outlined above, this section discusses how the glyCAFs actively shape immune exclusion. Across multiple solid tumors, the spatial organization of CAFs and their assembled extracellular matrix has been correlated with CD8⁺ T cells retention at the invasive margin, forming a histological basis for immune exclusion [35]. In soft-tissue sarcoma models, a metabolically distinct GLUT1⁺ glycolysis CAF subset is termed glyCAFs. These fibroblasts maintain high expression of the chemokine CXCL16, which binds to the receptor CXCR6 on CD8⁺ T cells and retains them at the tumor margin. Importantly, suppressing glycolysis in glyCAFs reduces CXCL16 levels, reverses T-cell exclusion, and enhances response to chemotherapy. This defines a causal link between stromal metabolic reprogramming and chemokine-mediated spatial control of immunity [17].

Mechanistically, CXCL16 exists in both transmembrane and soluble forms [1], while its cognate receptor CXCR6 shows preferential expression on effector and tissue resident memory like CD8⁺ T cells, where it governs their tissue localization and persistence [49]. To systematically dissect this axis, the following sections break down the process into three components: the upstream metabolic drivers inducing the glyCAF phenotype, the ligand-receptor interactions mediating peripheral T-cell retention, and the downstream implications for antitumor immunity, including reversibility and druggable targets (Fig. 1).

Fig. 1.

The glyCAF-GLUT1–CXCL16–CXCR6 axis: chemokine metabolism, edge retention, and therapeutic levers. (a) Hypoxia shifts fibroblasts to glycolysis. GLUT1 fuels this glyCAF state and drives CXCL16. CXCL16 appears as a soluble gradient from the margin toward the core. CXCL16 mediates T cells recruitment through two distinct mechanisms: the membrane-tethered form facilitates adhesion, whereas the proteolytically released soluble form guides chemotaxis. Both act on CXCR6⁺ CD8⁺ T cells to keep them at the edge and limit entry into the tumor core. (b) Targeting the glyCAF–CXCL16–CXCR6 axis to overcome immune exclusion. Disrupting this axis with a GLUT1 inhibitor, CXCR6 antagonist, or CXCL16-neutralizing antibody enhances intratumoral CD8⁺ T-cell infiltration. This remodeling of the tumor microenvironment primes tumors for a synergistic enhancement of response to radiotherapy, chemotherapy, or immune-checkpoint inhibitors

The glycolytic niche for immune exclusion

At the invasive margins of tumors, glyCAFs adopt a GLUT1-dependent metabolic program that sustains the expression of CXCL16, effectively trapping CD8⁺ T cells at the periphery and limiting their infiltration into the tumor core. This exclusionary function is sustained by a specific metabolic state defined by persistent glycolysis, enhanced glucose uptake, and elevated GLUT1 expression. Crucially, this state is not fixed but plastic, as demonstrated in soft‑tissue sarcoma where disruption of the GLUT1‑dependent program in glyCAFs remodels the stromal barrier and promotes T‑cell infiltration [17]. A parallel observation comes from pancreatic ductal adenocarcinoma, where high “metabolic CAFs” (meCAFs) abundance portends poor prognosis [50]. However, these meCAFs also promote CD8⁺ T cell infiltration and predict improved PD-1 blockade response. Thus meCAFs exemplify a functionally distinct CAF entity in which a glycolysis-high program is coupled to immune activation. This further illustrates the plasticity of CAF metabolic states. These divergent outcomes demonstrate that the relationship between CAF glycolysis and tumor immunity is context-dependent and mechanistically complex, requiring further investigation to resolve.

The upstream metabolic drivers inducing the glyCAF phenotype are functionally linked to the immune exclusionary mechanism. BAY-876 is a highly selective GLUT1 inhibitor with sub-nanomolar potency and minimal off-target effects [51]. Pharmacological inhibition of GLUT1 with agents such as BAY-876 directly suppresses glycolysis in glyCAFs, reducing their lactate production and reversing their activated phenotype [17]. Critically, by curtailing glyCAFs glycolysis, GLUT1 inhibition reduces the expression and secretion of CXCL16. The loss of this chemokine signal dismantles the spatial barrier at the tumor margin, liberating CXCR6⁺ CD8⁺ T cells from stromal retention and enabling their infiltration into the core. This metabolic reprogramming simultaneously alleviates the glucose-depleted, lactate-rich immunosuppressive milieu, further supporting T-cell function. In preclinical models, specifically suppressing glyCAFs glycolysis via GLUT1 inhibition thus impairs tumor growth and potently synergizes with chemotherapy and immune checkpoint blockade, thereby dismantling the glyCAFs-mediated barrier to directly enable and enhance a cytotoxic T-cell response [52]. Thus, GLUT1-driven glycolysis acts upstream to promote T-cell exclusion via the CXCL16–CXCR6 axis, thereby identifying both GLUT1 and this chemokine pathway as rational immunometabolic targets.

Collectively, these findings position the glycolytic niche not merely as a metabolic feature, but as an integrated immunometabolic axis that enforces spatial immune exclusion, revealing an actionable way for therapeutic intervention.

The CXCL16–CXCR6 axis in T cells positioning

Evidence supporting the CXCL16–CXCR6 axis as a spatial positioning module for CD8⁺ T cells falls into two complementary categories. Associative evidence stems from the spatial co‑localization of CXCL16‑enriched stromal compartments with CXCR6‑positive CD8⁺ T cells at tumor margin. Causal evidence, in turn, comes from functional perturbation studies: loss or blockade of CXCR6, neutralization or depletion of CXCL16, and stromal metabolic interventions that reduce CXCL16 levels. In preclinical models, these manipulations can successfully reposition CD8⁺ T cells, and in several contexts, this spatial shift correlates with improved tumor control.

CXCL16 exists in two functionally distinct forms. A membrane bound form supports adhesion. A soluble form is generated by proteolytic shedding and acts as a chemoattractant [53–55]. Its receptor, CXCR6, is enriched on effector and tissue resident like CD8⁺ T cells and supports their positioning and persistence in tissues [19, 56]. CXCL16 can also function through direct cell‑surface contact. In glioblastoma, CXCL16 expressed by tumor‑associated myeloid cells directs CXCR6‑mediated T‑cell engagement toward these myeloid populations, an interaction linked to impaired T‑cell activity [18]. In glyCAFs rich soft tissue sarcoma, CXCL16 is concentrated at the invasive margin and aligns with CXCR6 positive CD8⁺ T cells, a pattern clearly illustrates CD8⁺ T cells boundary retention and its role in immune exclusion [17].

Functional studies reveal that the CXCL16–CXCR6 axis guides CD8⁺ T cells positioning in two context‑dependent ways. Soluble CXCL16 can serve as a chemoattractant that promotes T‑cell entry. In transwell assays, it robustly recruits CXCR6‑expressing T cells, an effect abolished by CXCR6 disruption [57]. This directional role is also observed in vivo: following radiotherapy, tumor cells release more CXCL16, which promotes the infiltration of activated CXCR6 positive CD8⁺ T cells, a response that is diminished in CXCR6‑deficient settings [58, 59]. In contrast, within stromal niches rich in membrane‑associated CXCL16, the same axis supports T cells retention and maintenance. In a perivascular niche, CXCR6 positions effector‑like CD8⁺ T cells near CXCL16‑expressing dendritic cells that trans‑present IL‑15, thereby supporting local T cells survival and expansion [49, 60, 61]. Consistent with this retention function, CXCR6 promotes the persistence of memory CD8⁺ T cells in ovarian tumors and contributes to tumor control. Conversely, loss of CXCR6 in both mouse and human tumors impairs the maintenance of intratumoral effector‑like and tissue‑resident CD8⁺ T cells, weakening immune surveillance [62]. This biology also sets a therapeutic caveat. Prolonged or systemic CXCR6 blockade may compromise beneficial residency programs in settings where CXCR6 supports intratumoral persistence [49, 63].

In immune‑excluded tumors, where T cells are predominantly trapped at the margin, disrupting the CXCL16–CXCR6 axis can reverse this retention and enhance therapy. In a breast cancer model, impairing CXCR6‑dependent retention at the primary site enabled tumor‑experienced T cells to exit, seed the lung as tissue‑resident memory populations, and better control metastasis [64]. Likewise, in soft‑tissue sarcoma, a glycolysis CAF‑driven program elevates CXCL16 at the invasive margin. Inhibiting this metabolism reduces CXCL16, loosens the marginal barrier, restores CD8⁺ T‑cells entry into the core, and thereby sensitizes tumors to chemotherapy [17].

The CXCL16–CXCR6 axis directs CD8⁺ T cells through two opposing functions: recruitment and retention. This duality presents a clear therapeutic trade-off and, correspondingly, points to two distinct strategies for intervention. Whether the axis ultimately promotes immune exclusion or supports protective T cells residency depends on key contextual factors, including the spatial architecture of the tumor, the cellular source of CXCL16, and the treatment setting [19, 62, 63]. One approach targets the stromal metabolic source. In soft‑tissue sarcoma, inhibiting GLUT1 or glycolysis in glyCAFs lowers CXCL16 production, weakens the restrictive barrier at the invasive margin, and enables CD8⁺ T cells infiltration into the tumor core, thereby enhancing chemotherapy efficacy [17]. The other strategy intercepts the axis directly. In a mouse breast cancer model, disrupting CXCL16–CXCR6‑mediated retention at the primary tumor site promoted the redistribution of tumor‑experienced T cells. These cells then seeded the lung as protective tissue‑resident memory populations, improving control of metastatic spread [64]. The two strategies work in complementary ways: one lowers stromal CXCL16 production, and the other interferes with its spatial signal. Both approaches, however, should be applied in specific contexts. Prolonged or systemic blockade of CXCR6 could unintentionally harm beneficial T cells residency in tumors where this axis helps sustain antitumor immunity [19, 63].

Whether the CXCL16–CXCR6 axis promotes T cells recruitment or retention depends on three key factors: the soluble versus membrane bound form of CXCL16, the cellular source (e.g., tumor cells, myeloid cells, CAFs), and the spatial architecture of the tumor margin versus perivascular niches. Recognizing this duality is essential for designing context appropriate therapeutic strategies that avoid unintended disruption of protective T cell residency. In summary, the CXCL16–CXCR6 axis is a central driver of CD8⁺ T cells margin retention, and its net effect depends on tumor type, TME composition, and treatment context. Targeting this pathway at the level of metabolism, receptor, or ligand has clear translational promise (Table 2 summarizes the functions and key intervention nodes across the glyCAFs axis).

Table 2.

Mechanistic map of the glyCAF—GLUT1—CXCL16—CXCR6 pathway

Mechanism step	Representative molecules/markers	Detection and validation (recommended)	Functional/phenotypic outcome	Therapeutic/translational significance
1. Origin and plasticity	CAF core: FAP, PDPN, α-SMA; lineage/plasticity: ISLR (Meflin), LRRC15	scRNA-seq with pseudotime; lineage tracing in mouse models; spatial transcriptomics	Distinct origins and niches bias CAF function (metabolic, matrix, inflammatory)	Source mapping guides target choice (e.g., different strategies for Meflin+ rCAF vs glyCAF)
2. Metabolic reprogramming and markers	GLUT1 (SLC2A1), HK2, LDHA, MCT4; high 2-NBDG uptake; high glycolysis signature	Flow cytometry for 2-NBDG; GLUT1 protein by mIF/IHC; Seahorse assays; scRNA-seq pathway scoring	CAFs enter a high-glycolysis state (glyCAF) and enrich at the margin	Druggable upstream node (GLUT1 inhibitors) for metabolic intervention
3. Secretory effector upregulation	CXCL16 (membrane-bound and soluble); other chemokines	qPCR/ELISA/multiplex IF; spatial transcriptomics; protein imaging (IMC/CODEX)	CXCL16 forms a ligand-rich band at the tumor edge	Spatial biomarker; neutralizing-antibody target
4. Receptor expression and response	CXCR6 on CD8+ T cells, T_RM, some NK/NKT; controls positioning and survival	scRNA-seq; flow cytometry; tissue IF; migration/chemotaxis assays	CXCR6 draws CD8+ T cells to CXCL16-rich niches and promotes retention	Receptor-level target; engineering handle for guided T-cell therapies
5. Spatial pattern and exclusion	glyCAF (GLUT1+ /CXCL16+) concentrated at the invasive front; CD8+ T cells stall at the edge	Spatial transcriptomics (Visium/GeoMx); mIF/IMC; nearest-neighbor distance (NND) statistics	Failure of CD8+ T cells entry into the parenchyma; immune-cold/excluded zones	Spatial panel (glyCAFs abundance + GLUT1/CXCL16 edge enrichment + low intratumoral CD8) as a predictive biomarker
6. Functional consequences	—	In vivo transplant/knockout/antibody-block studies; tumor growth and immune-infiltration readouts	Weakened immune surveillance; reduced benefit from ICI, chemotherapy, and radiotherapy	Necessitates combined metabolic, chemotactic, and immune strategies to reverse exclusion
7. Intervention points and strategies	Upstream: GLUT1 inhibition (BAY-876; shRNA/CRISPR). Midstream: CXCL16 neutralization. Downstream: CXCR6 antagonism or receptor engineering	Pharmacology (dose and route); combinations (GLUT1 inhibitor with ICI/radiotherapy/chemotherapy); safety evaluation	Lower CXCL16 or block CXCR6 to restore CD8+ T cells infiltration and efficacy	Supports multi-node combinations; local delivery can reduce toxicity
8. Clinical/diagnostic translation	Spatial biomarker panels (mIF/spatial omics) for patient stratification	Define thresholds (edge glyCAFs density; edge enrichment index for GLUT1/CXCL16; parenchyma-to-edge CD8 ratio); prospective validation	Predict response to ICI and chemoradiotherapy; guide trial stratification	Select candidates for metabolism–immunity combinations

Functional consequences and therapeutic reversal of immune exclusion

CD8⁺ T cells retention at the tumor margin characterizes the immune-excluded state and limits immunotherapy efficacy [65–69]. Immune exclusion often reflects a series of sequential barriers, including stromal and extracellular matrix organization, endothelial selectivity, and competing chemokine signals, which together determine whether CD8⁺ T cells accumulate at the tumor margin or successfully enter the nests [70–72]. Here, we focus on the glyCAF–CXCL16–CXCR6 axis as a druggable stromal circuit that actively retains CD8+ T cells at the invasive margin, this retention mechanism is distinct from the physical barriers [73–75], such as dense ECM and restrictive endothelium, which can even restrict T cells access to the invasive tumor margin.

Current evidence, largely based on spatial co‑localization and functional perturbation studies, suggests that the immune exclusion may be actively orchestrated. GlyCAFs at the invasive margin express CXCL16 through a GLUT1‑dependent metabolic program [17]. CXCL16 can engage CXCR6 on CD8⁺ T cells and, in principle, could reinforce an adhesion‑chemotaxis circuit that increases stromal attachment. This would be expected to limit T cells movement into the tumor core, although direct dynamic evidence for such immobilization is still lacking. Mechanistically, CXCL16 biology offers a plausible explanation for margin retention that does not rely on a purely physical blockade. CXCL16 exists as a transmembrane, surface-anchored form that can support firm adhesion of CXCR6⁺ T cells, and it can be shed to generate a soluble chemokine that provides chemotactic cues. By engaging CXCR6 through both adhesive and chemotactic functions, CXCL16 can enhance the retention of T cells specifically along stromal interfaces at the invasive margin [76–80]. ECM-rich architecture often colocalizes with these interfaces and may provide a spatial scaffold for retention signals rather than being the signal itself. We note that these spatial co-localization data are correlative and do not directly demonstrate dynamic T-cell retention or migration. Future studies using intravital imaging or genetic lineage-tracing will be required to establish causality and to distinguish retention from other barriers such as ECM or vasculature.

Whether the glyCAF–CXCL16–CXCR6 axis is the dominant barrier or a secondary feature depends on the underlying context within a given tumor. In one setting, T cells consistently reach the invasive margin, are clearly extravascular, and navigate the stroma with little hindrance, yet they stop at the edge of the core [17]. That is retention‑dominant immune exclusion. In the other setting, T cells are scarce at the margin or remain largely trapped inside vessels, often accompanied by extensive desmoplasia or vascular abnormalities. Here the bottleneck lies upstream, where endothelial barriers or a dense ECM physically block T cells extravasation [6, 68, 73, 81]. This distinction matters clinically. Retention‑dominant tumors call for strategies that release trapped cells. Entry‑limited tumors may first require interventions that normalize vessels or remodel the matrix.

This spatial distinction shapes therapeutic strategy. For tumors where margin retention of CD8⁺ T cells drives immune exclusion, releasing trapped cells is the logical approach. In soft-tissue sarcoma, targeting GLUT1 disrupts the axis, reduces CXCL16, relieves T-cell retention, and promotes their infiltration into the tumor parenchyma, ultimately improving chemotherapy response [17]. This establishes that glyCAF-driven exclusion is a dynamic and druggable process, and its reversibility is precisely what makes it therapeutically valuable. But further studies are needed to confirm that the effect is directly due to altered T‑cell motility rather than secondary changes in ECM or vasculature.

Building on these insights, two principal therapeutic strategies have emerged to overcome immune exclusion. The first targets stromal metabolism by inhibiting GLUT1‑dependent glycolysis in CAFs, which reduces CXCL16 expression, depletes margin‑restrictive glyCAFs, and promotes CD8⁺ T cells infiltration. The second directly interrupts the CXCL16–CXCR6 adhesion–chemotaxis circuit using neutralizing antibodies, redistributing T cells and reversing marginal retention in preclinical model. These approaches are complementary, one reprogramming CAF metabolism, the other blocking the adhesion‑chemotaxis circuit, and can be rationally combined. Notably, some interventions primarily improve access by reducing ECM crowding [82, 83] and increasing endothelial permissiveness [73, 84], so that more CD8⁺ T cells can reach the invasive margin. In contrast, CXCL16–CXCR6 axis blockade targets a separate step, releasing T cells from margin retention and supporting subsequent entry into tumor core [85]. Viewed this way, CXCL16–CXCR6 represents one stromal T-cell retention program. It can operate alongside other chemokine axes, most notably CAF-derived CXCL12–CXCR4 [86]. Thus T-cell exclusion is not driven by one axis alone [85, 87–89]. Importantly, both interventions require contextual precision, as prolonged or systemic CXCR6 pathway blockade may inadvertently disrupt beneficial T‑cell residency in settings where the axis supports antitumor persistence.

Future validation needs for the GlyCAF–GLUT1–CXCL16–CXCR6 positioning axis

While the role of the CXCL16–CXCR6 axis in CD8⁺ T‑cell positioning is becoming clearer, key mechanistic questions and translational gaps persist.

First, we need direct validation in patient samples. Current evidence comes largely from preclinical models. To establish clinical relevance, multiplex spatial protein assays must be used to map CXCL16 and CXCR6 positive CD8⁺ T cells within the same tumor sections. Only then can we determine whether their spatial relationship predicts treatment outcomes in prospectively annotated patient cohorts.

Second, the functional contributions of the two CXCL16 isoforms are not well defined. Membrane‑bound CXCL16 mediates adhesion, whereas ADAM10/17‑cleaved soluble CXCL16 acts as a chemoattractant [1, 90, 91]. Disentangling these roles experimentally, especially within the same microenvironment, is essential to understand whether retention or recruitment dominates in a given tumor context.

Third, we require definitive causal evidence from glyCAF‑rich models. This calls for perturbation experiments that combine CAF‑specific suppression of CXCL16 with CXCR6 disruption in CD8⁺ T cells, followed by CXCL16 rescue. Such an approach can test whether stromal CXCL16 is strictly necessary and sufficient for marginal T cells retention, or merely a bystander within a permissive niche.

Addressing these points will solidify the mechanistic basis of the glyCAF–GLUT1–CXCL16–CXCR6 axis and accelerate its development into reliable spatial biomarkers and targeted therapies.

Clinical and translational implications

Translating the biology of the glyCAF–GLUT1–CXCL16–CXCR6 axis into clinical strategies represents a pivotal frontier in overcoming immune exclusion. A primary objective is to develop this pathway into a quantitative, spatially informed biomarker framework for patient stratification and therapeutic guidance [92].

Having established the mechanistic role of the glyCAF–GLUT1–CXCL16–CXCR6 axis in immune exclusion, we see it as an emerging, context‑dependent model rather than a universally established pathway. With that in mind, we now turn to its clinical translation. To ground this discussion in actionable evidence, we have stratified the key intervention points along this axis based on three practical criteria: the maturity of the supporting evidence (preclinical or clinical), translational feasibility, and foreseeable immune‑related liabilities. This tiered evaluation, detailed in Table 3, helps distinguish targets ready for clinical exploration from those requiring further validation, while outlining the opportunities and safety considerations that will guide their development.

Table 3.

Evidence stratification of actionable intervention nodes along the glyCAF–GLUT1–CXCL16–CXCR6 axis

Evidence tier	Target & modality	Key efficacy evidence	Practical feasibility	Spatial pharmacodynamic endpoints for translation	Key liabilities/immune-related toxicity
Tier 1	CAF GLUT1 (metabolic node)—small-molecule GLUT1 inhibitor (e.g., BAY-876)	In murine STS models and scRNA-seq data, glyCAFs restrict CD8+ T cell entry via GLUT1-dependent CXCL16. Inhibiting glycolysis or GLUT1 reduces this marginal barrier and improves chemotherapy response [17].	Tool compounds exist (BAY-876 widely used preclinically); true CAF-selective clinical agent not established.	Margin-localized glyCAFs abundance, GLUT1 and CXCL16 expression in glyCAFs, invasive-margin–to–tumor-core CD8+ T cells density ratio, nearest-neighbor distance between CD8+ T cells and CXCL16+ CAFs [17].	GLUT1 inhibition can raise CNS metabolic liability due to systemic GLUT1 biology (BBB glucose transport) [93–95]; GLUT1 inhibition can dampen CD4+ T cells proliferation and IFN-γ in vitro (dose and context dependent) [96].
Tier 1	CAF-targeted GLUT1 delivery—BAY-876-loaded CAF-derived extracellular vesicles	In mouse lung tumor models, BAY-876-loaded CAF-derived EVs reduce ECM stiffness, increase intratumoral CD8+ T cell infiltration, and synergize with anti-PD-L1. Public scRNA-seq data confirm GLUT1-high CAF subsets [52].	Translation is limited by incomplete standardization of EV isolation and characterization and the current difficulty of producing clinical-grade EV formulations at scale.	Same as above plus stromal stiffness/collagen readouts (imaging/biomechanics) and CAF activation markers in spatial maps [52].	Adds EV-platform risks (batch variability, biodistribution) while not removing core GLUT1 on-target systemic liability (CNS/immune metabolism).
Tier 1	CXCL16 (ligand) blockade—neutralizing antibody or knockdown	In mouse TNBC models, intratumoral CXCL16 neutralization (NAb) combined with anti-PD-1; human TNBC tissue analysis supports relevance [97].	Antibody modality is realistic; but tumor-targeted delivery and exposure control matter (systemic vs intratumoral).	Best-fit when CD8+ T cells accumulate at invasive margin with high CXCL16 signal [17]. “On-target in tumor” = ↓margin CXCL16 gradient and ↓stromal retention signature; “off-tumor” monitoring = TRM-linked compartments (e.g., airway TRM metrics in translational studies) [98].	Targeting CXCR6 may disrupt beneficial CTL residency. In tumors, CXCR6 organizes effector-like CTLs in a perivascular niche via CXCL16+ DC interactions supporting survival [49]. In lung, it supports tissue-resident memory T cells (TRM); blockade may impair airway TRM maintenance, increasing infection risk and reducing local immunosurveillance [98]
Tier 2	CXCR6 (receptor) antagonism—small-molecule antagonist (e.g., SBI-457)	SBI-457 tested in human hepatoma cell lines and SK-Hep-1 xenografts; disrupts CXCR6 pathway and shows combination potential with sorafenib [99]. CXCR6 knockdown reduces hepatoma cell invasion in vitro and inhibits tumor growth [100].	While a small-molecule antagonist is feasible, current evidence is largely context-limited, and both immunocompetent efficacy and T cells trafficking consequences require dedicated validation.	Would need CXCR6 expression and soluble CXCL16 shedding stratification (as proposed in hepatocellular carcinoma cell work) plus spatial immune context before considering translation [99]. Target engagement (↓CXCR6 signaling outputs; e.g., β-catenin readouts in HCC work) + tumor burden change; spatial immune PD must confirm no loss of beneficial CXCR6+ effector CTLs if combined with ICI.	Risk of harming anti-tumor CTL positioning and survival niches [49]; plus uncertain immune effects since key data come from xenografts.
Tier 1 (application-dependent)	CXCR6 (engineering)—forced CXCR6 expression in ACT/CAR-T to improve homing	In murine pancreatic tumor ACT/CAR-T settings, CXCR6-armed T cells show improved migration toward CXCL16 and better tumor control (preclinical efficacy) [57].	gene transfer is Feasible in cell-therapy manufacturing, but depends on ACT availability and tumor CXCL16 landscape.	↑increased engineered T-cell accumulation in CXCL16-rich regions; improved parenchymal penetration; persistence markers [57].	CXCL16–CXCR6 axis has been linked to CAR-T neurotoxicity (ICANS) via CNS recruitment of cytotoxic CAR CD4+ T cells (human samples) [101].

Abbreviations: STS, soft tissue sarcoma; ACT, adoptive cell therapy; CNS, central nervous system; BBB, blood–brain barrier; EV, extracellular vesicle; ECM, extracellular matrix; ICANS, immune effector cell-associated neurotoxicity syndrome; PD, pharmacodynamics; NAb, neutralizing antibody; TNBC, triple-negative breast cancer; CTL, cytotoxic t lymphocyte; DC, dendritic cell; TRM, tissue-resident memory T cells; ICI, immune checkpoint inhibitor

Tier 1 (robust preclinical + human relevance): Demonstrates therapeutic benefit in immunocompetent and/or orthotopic animal models with a closed mechanistic loop, and is further supported by human relevance (e.g., patient cohorts, clinical specimens, or tissue-based correlative analyses)

Tier 2 (preclinical, context-limited): Evidence is primarily derived from xenografts, immunodeficient systems, or a single tumor lineage and context; generalizability is uncertain and clinical extrapolation should be cautious

It should be noted, however, that immune exclusion in solid tumors is not governed by this chemokine axis alone. Increased ECM stiffness has been shown to restrict the migration and penetration of T cells and other effector immune cells into the tumor core, contributing to immune exclusion and limiting responses to immunotherapy [102, 103]. Dysfunctional tumor vasculature establishes a barrier that reduces immune cell recruitment and sustains an immunosuppressive microenvironment, thereby reinforcing immune exclusion [72, 104]. Moreover, a broader chemokine landscape involving CXCL12–CXCR4 [105], CCL2–CCR2 [106, 107], and CCL5–CCR5 [108] also contributes to immune cell recruitment, retention, and exclusion across tumor types. Thus, these structural, vascular, and chemokine mechanisms can operate alongside the glyCAF–CXCL16–CXCR6 axis to collectively determine immune exclusion and influence therapeutic response.

Given the complex, multi‑layered nature of immune exclusion, treatments that focus on a central, druggable target within this network hold particular promise. The glyCAF–GLUT1–CXCL16–CXCR6 axis offers two such synergistic points for intervention. From a therapeutic perspective, this axis thus points to two synergistic therapeutic strategies: inhibiting glycolysis in CAFs to overcome T-cell exclusion and improve chemo-sensitivity, and directly targeting the CXCL16–CXCR6 chemokine retention axis to remodel the tumor microenvironment and enhance the efficacy of radio, chemo, or immunotherapy [17, 18].

To advance this paradigm, early-phase clinical studies should prospectively assess how GLUT1 inhibition modulates immune cell metabolism and intratumoral pharmacokinetics [96]. Such efforts will be essential to validate this axis and inform the rational design of combination therapies.

Predictive biomarkers (spatial TME perspective)

Spatially, the TME in immune‑excluded tumors is shaped by both physical barriers and active stromal signaling. Viewing the TME through the lens of stromal signaling reveals two spatial readouts: the abundance of glyCAFs at the tumor margin and the activity of the GLUT1–CXCL16 axis in this region. Together, these correlated features form a practical, spatially defined biomarker for identifying immune-excluded tumors along the tumor margin. In soft-tissue sarcoma models, by inhibiting glycolysis in glyCAFs to deplete CXCL16 and break down the marginal barrier, CD8⁺ T cells gain access to the tumor core, thus directly sensitizing tumors to chemotherapy [17]. This suggests that glycolysis blockade could be linked to both improved immune access and enhanced treatment efficacy. Therefore, the spatial glyCAF–CXCR6–CXCL16 axis constitutes a compelling mechanistic framework. When applying this biomarker, the CXCL16‑based margin‑retention signal should be interpreted in the context of local ECM and vascular features. Doing so helps distinguish true chemokine‑driven retention from cases where T cells access is physically restricted. Ultimately, the predictive value of this spatial signature should be validated in spatially resolved cohorts with detailed clinical annotation. The distinct spatial configuration of marginal glyCAFs and GLUT1‑driven CXCL16 positions this axis as a practical, mechanistically rational biomarker set for identifying immune-excluded tumors.

From a clinical prediction perspective, spatially resolved multimodal imaging and spatial transcriptomics can capture three key data layers at once: cell phenotype, metabolic markers, and spatial location. Using these layers, an actionable readout can be built, including the abundance of glyCAFs at the tumor margin, the spatial enrichment of GLUT1 and CXCL16 on CAFs, and the density of CD8⁺ T cells within the parenchyma. Moreover, the nearest-neighbor distance between CD8⁺ T cells and CXCL16⁺ glyCAFs can be measured. Spatial biomarkers, which quantify features such as cell-cell proximity and distribution, have been shown to more reliably predict response to immune checkpoint inhibitors than traditional biomarkers, including PD-L1, TMB, and MSI, or metrics based solely on total TIL density [92]. Therefore, a spatial glyCAFs–GLUT1–CXCL16 signature has clear predictive value and practical clinical utility (Fig. 2).

Fig. 2.

Spatial biomarkers and validation workflow for the glyCafs–GLUT1–CXCL16 signature. (a) Spatial predictive biomarkers. Cartoon tissue map showing edge-focused readouts: glyCafs abundance at the margin (FAP, PDPN, α-SMA), enrichment of GLUT1 and CXCL16 on CAFs at the edge, CD8⁺ T cells density in the parenchyma, nearest-neighbor distance from CD8⁺ T cells to CXCL16⁺ glyCafs, and the margin-to-parenchyma ratio. (b) Validation workflow. Step 1: use scRNA-seq to confirm the glycolytic/GLUT1 program in CAFs (for example, HK2 and LDHA) and quantify glycolysis activity. Step 2: use spatial transcriptomics (Visium, GeoMx, or in-situ methods) to map CXCL16 and verify an edge-to-core gradient with parenchymal drop-off. Step 3: validate at the protein level with multiplex imaging (mIF, IMC/MIBI-TOF, CODEX); measure CAF markers (FAP, PDPN, α-SMA), GLUT1, CXCL16, and CD8; compute spatial metrics such as nearest-neighbor distance and margin-to-parenchyma ratios. Step 4: test the signature in an independent cohort and compare predictive performance for response to immune-checkpoint blockade, chemotherapy, and radiotherapy

Methodologically, high-quality studies and reviews provided solid technical and statistical frameworks. We propose integrating these candidate biomarkers into a standard validation pipeline. Firstly, the GLUT1-high, glycolytic signature in CAFs is confirmed using single-cell RNA sequencing. This involves isolating CAFs from tumor stroma and measuring their glycolysis activity and GLUT1 expression, thereby establishing a molecular baseline for subsequent spatial analyzes [37, 109, 110]. Secondly, spatial transcriptomics (Visium, GeoMx, or high-resolution in situ methods) are employed to map the CXCL16 gradient and align it with CD8 signals. This typically shows a gradual decrease in CXCL16 from the tumor margin toward the tumor core [111–113]. Thirdly, validation at the protein level is performed using multiplex immunofluorescence (mIF), imaging mass cytometry (IMC/MIBI-TOF), or CODEX. Detection of 30–60 markers is performed on whole-slide or high-magnification sections. CAFs markers (FAP, PDPN, α-SMA), GLUT1, CXCL16, and CD8⁺ T cells are quantified, and spatial metrics such as nearest-neighbor distance and margin-to-parenchyma ratios are computed to set quantitative thresholds [114–116]. Finally, the signature is validated in independent patient cohorts, either prospectively or retrospectively, and its correlation and predictive performance are compared with responses to ICI, chemotherapy, and radiotherapy.

Collectively, these integrated technical approaches provide a framework for evaluating the spatial glyCAF–GLUT1–CXCL16 signature as a mechanistically informed biomarker candidate that can help identify immune-excluded tumors.

Therapeutic advantages of targeting GLUT1 in glyCafs

Conceptually, chemokines direct where T cells are retained within the tumor microenvironment, while physical and vascular barriers—such as dense ECM and abnormal vessels—determine whether T cells can enter and navigate through the tumor tissue [102, 117]. Targeting GLUT1 in CAFs represents a strategic approach to disrupt glyCAF-driven immune exclusion at its metabolic source. Suppressing GLUT1 dampens glycolysis in CAFs, which reduces CXCL16 expression at the tumor margin and promotes the release of CXCR6⁺ CD8⁺ T cells from stromal retention, facilitating their infiltration into the tumor core. In addition, this metabolic inhibition also attenuates CAF‑linked matrix stiffening [52]. This dual action alleviates both the retention cue mediated by CXCL16 and the physical barrier created by a stiffened ECM, enabling deeper penetration of CD8⁺ T cells toward the tumor core.

The therapeutic potential of targeting GLUT1 is well supported in experimental models. Inhibiting GLUT1 genetically or pharmacologically with agents like BAY‑876 can suppress glycolysis in glyCAFs and reduce their production of CXCL16. This drop in CXCL16 weakens the chemokine‑mediated retention at the tumor margin, permitting CD8⁺ T cells entry into the core. In vivo, combining GLUT1 inhibition with doxorubicin improves tumor control, extends survival, and enhances CD8⁺ T cells cytotoxicity, demonstrating clear chemosensitization [17]. Notably, a precision delivery approach using CAF‑derived extracellular vesicles to transport BAY‑876 specifically to GLUT1‑high stromal and tumor cells reverses the activated CAF phenotype, promotes profound intratumoral T cells infiltration, reduces ECM stiffness, and synergizes with anti‑PD‑L1 therapy to improve outcomes in preclinical lung cancer models [52]. These findings collectively validate GLUT1 as a tractable metabolic target within the stroma.

From a translational perspective, integrating GLUT1 inhibition with CXCL16–CXCR6 axis blockade offers a multi-node strategy that spans metabolic reprogramming to chemokine signaling. This integrated approach relieves immune exclusion, enhances response to chemo and radiotherapy, and can sensitize tumors to immune checkpoint blockade [17, 52, 97, 118]. Beyond this core effect, in tumors where immune cell positioning is dominated by other chemokine modules (e.g., CXCL12–CXCR4 or CCL2–CCR2), this strategy may be rationally combined with these chemokine modules for a broader synergistic effect.

Together, targeting GLUT1 in glyCAFs represents a promising, stroma-focused strategy that may help overcome immune exclusion and improve the efficacy of conventional and immune-based therapies.

Antibodies targeting CXCL16–CXCR6 and combination strategies

Immune exclusion in tumors typically involves multiple barriers. Impaired vascular extravasation limits initial T cells entry [73]; dense and collagen‑rich ECM hinders interstitial migration [117]. Both are frequently coupled with chemokine‑mediated T cells retention in stromal niches. Against this backdrop, interventions against the CXCL16–CXCR6 axis can be classified into two groups: ligand blockade and receptor modulation. Blocking CXCL16 with neutralizing antibodies, or silencing its expression can reduce intratumoral CXCL16 level, eases CD8⁺ T cells stalling at the margin, and improves antitumor activity when combined with anti-PD-1 in mouse models [18, 97]. In contrast, when CD8⁺ T cells are sparse even at the tumor border, the dominant limitation often lies upstream: impaired vascular extravasation or a dense, stiff ECM that restricts interstitial migration [119, 120]. In such cases, blocking CXCL16 should be viewed as a complementary strategy, most effective when combined with interventions that first improve vascular access or normalize the ECM.

Genetic or antibody interference with CXCR6 can modulate the distribution of T cells residing in tissues and lessen CXCL16-driven peripheral retention in select settings [57]. The CXCL16–CXCR6 axis can also be used to guide engineered T cells. In pancreatic tumor models, engineering T cells or CAR‑T cells to express CXCR6 enhanced their migration toward CXCL16, improved their accumulation and motility within tumors, and ultimately led to better tumor control and survival. These effects were observed across multiple solid tumor model systems, including subcutaneous and orthotopic tumors as well as patient‑derived xenografts, and align with the success of similar T‑cell engineering strategies in CAR‑T therapies for hematologic malignancies [57, 121]. Summarily, engineering T cells to express CXCR6 improves their tumor homing to CXCL16-rich tumors and boosts the efficacy of CAR-T or TCR-T therapies, potentially converting a chemokine retention into a homing advantage [19, 122].

Collectively, these findings support the development of CXCL16-neutralizing antibodies and CXCR6-directed strategies, either used alone or combined with ICIs or cell therapies, to overcome glyCAFs-mediated immune exclusion and improve control of solid tumors. Notably, in tumors where immune positioning is dominated by other chemokine axes [107, 123–125], these pathways can be rationally layered with CXCL16–CXCR6 blockade to build broader combination regimens.

In sum, targeting the CXCL16–CXCR6 axis therefore offers a viable strategy to reduce T cells retention in immune‑excluded tumors and could be rationally combined with therapies that address earlier physical or vascular barriers.

Feasibility, safety, and trial-design considerations

Advancing these strategies into the clinic requires a clear path for each target. Small-molecule GLUT1 inhibitors like BAY-876 provide a direct, mechanistically grounded point [52]. The primary translational challenge will be achieving sufficient stromal exposure within the tumor while minimizing systemic effects. Extracellular vesicles show considerable promise as a targeted delivery system in preclinical research. However, several practical barriers must be overcome before they can be widely adopted in the clinic. These include establishing consistent, large-scale manufacturing under good manufacturing practice standards, ensuring batch-to-batch reproducibility, developing validated assays to confirm product potency, and navigating a regulatory landscape that is still taking shape [126–129].

Safety profiling is crucial. Because GLUT1 is essential for glucose transport across the blood–brain barrier, systemic inhibition carries a risk of central nervous system toxicity [93]. Study also suggest that inhibiting GLUT1 can suppress CD4⁺ T‑cell function in certain contexts, highlighting the need for careful immune monitoring if combining such agents with immunotherapy [96]. For the CXCL16–CXCR6 axis, neutralizing the ligand CXCL16 with antibodies represents the most straightforward approach in biomarker-selected, immune-excluded tumors. Blocking the receptor CXCR6 requires more caution, as it may disrupt beneficial T cells residency programs that support antitumor immunity [54]. Blockade makes sense when tumor margin retention is clearly the dominant problem. In that setting, CXCL16 is concentrated in CAF‑rich stroma at the invasive front, and CXCR6⁺ CD8⁺ T cells are trapped along the margin. Disrupting the axis under these conditions can relieve retention, let T cells penetrate, and improve immune response [17]. However, the situation [118] changes when intratumoral CXCR6⁺ CD8⁺ resident memory T (TRM) cells are already abundant and show resident markers like CD69 or CD103 [19, 49, 62]. In several models, CXCR6 supports intratumoral CD8⁺ TRM cells persistence and antitumor efficacy [49, 63]. Losing it can weaken tumor control or blunt PD‑1 response [63, 118, 130]. In that context, blockade may do more harm than good.

Given these considerations, initial clinical trials should adopt a biomarker-guided design. This involves enrolling patients whose tumors exhibit a defined “retention-dominant” phenotype—characterized by glyCAF/GLUT1/CXCL16 enrichment at the invasive margin coupled with a margin-high, core-low CD8⁺ T‑cell distribution [17]. The pharmacodynamic effects of treatment should be assessed using spatial readouts, such as changes in the margin-to-core T cells ratio, the proximity of T cells to CXCL16 positive CAFs, and the gradient of CXCL16 itself. Concurrent safety monitoring must be tailored to the specific mechanism, including neurologic assessments for GLUT1 inhibitors and immune function tests for combinations involving CXCR6 pathway blockade [131].

Taken together, moving these targeted strategies into the clinic will require rigorous safety monitoring, biomarker‑selected trials, and spatial pharmacodynamic measures to assess therapeutic potential.

Distinctions from conventional CAFs, subtype heterogeneity, and plasticity within the CAF continuum

A full view of glyCAFs and their role in immune exclusion requires placing them within the wider context of CAF heterogeneity, plasticity, and continuum, across which they can emerge as a distinct metabolic state. Single-cell RNA sequencing and spatial omics have substantially advanced our view of the tumor microenvironment by revealing a broad spectrum of distinct CAF states. Canonical frameworks distinguish myofibroblastic CAFs (myCAFs) from inflammatory CAFs (iCAFs). MyCAFs express contractile modules such as ACTA2/α-SMA, TAGLN, MYL9 and TPM1/2 and are enriched for TGF-β/SMAD and RhoA/ROCK signaling with matrix-remodeling functions. iCAFs secrete cytokines and chemokines including IL-6, LIF and CXCL12 and exhibit immunoregulatory activity [132]. Antigen-presenting CAFs (apCAFs) constitute a third state marked by MHC-II machinery such as HLA-DRA and CD74 with potential to modulate CD4⁺ T-cell response [133]. Spatial and functional classifications have identified additional subtypes, including vascular-associated CAFs (vCAFs), matrix-CAFs (mCAFs) and developmental-CAFs (dCAFs) described in breast cancer, as well as cycling or proliferative CAFs and Meflin-positive cancer-restraining CAFs (rCAFs) [134, 135].

The pronounced heterogeneity of CAFs, now mapped in Table 4, provides the essential context for recognizing the distinct nature of the glyCAFs state. Against the backdrop of CAF heterogeneity, the glyCAFs state is defined by a core set of metabolic and spatially distinct features with a unique immunomodulatory footprint. Unlike classical myCAFs or iCAFs, which are primarily defined by contractile or secretory phenotypes, glyCAFs are fundamentally orchestrated by a GLUT1-high, glycolysis-dependent metabolic program that fuels sustained CXCL16 production. This metabolic specialization directly dictates a defined spatial function: it establishes a barrier mediated by CXCL16 and its receptor CXCR6, which restrains CD8 positive T cells at the tumor-stroma interface. What distinguishes glyCAFs is thus not merely their metabolic state, but its functional convergence with an exclusionary immune checkpoint. This paradigm challenges the adequacy lineage-based CAF taxonomies and underscores the necessity of incorporating metabolic circuitry and spatial context into any cohesive framework of stromal cell function [17, 148].

Table 4.

Landscape of CAF subtypes

Subtype	Representative markers/pathways	Key functions and phenotypes	Typical tumors/source	Key refs
glyCAF (glycolytic)	SLC2A1/GLUT1↑, HK2, LDHA, glycolysis gene set; CXCL16 secretion	Marked metabolic reprogramming; GLUT1-dependent CXCL16 upregulation acting on T cells CXCR6; linked to immune exclusion and margin retention; tractable pathway	Soft-tissue sarcoma; mouse models and multi-cohort validation	[17]
myCAF (myofibroblastic)	ACTA2/α-SMA, TAGLN, COL1A1/3A1; high TGF-β activity; LRRC15+ subset	ECM deposition and contraction; dense stroma and immune exclusion; LRRC15+ myCAF associated with poor ICB response	Multiple cancers (e.g., skin, pancreas)	[12, 136, 137]
iCAF (inflammatory, immunoregulatory)	IL-6, LIF, IL-11, CXCL12, DPP4	Cytokine and chemokine secretion; reshapes immune infiltration; plasticity with shifts toward ECM-rich myCAF state	Multiple cancers; supported by spatial transcriptomics	[37]
apCAF (antigen-presenting)	HLA-DRA/DP/DQ, CD74, CIITA	MHC-II–dependent antigen presentation; associated with ICI response in some settings	Gastric and pancreatic cancers	[138, 139]
mCAF (matrix, remodeling)	COL1A1, FAP, MMPs, POSTN	ECM production and remodeling; mechanical barrier	Skin and breast cancers	[12]
vCAF/pvCAF (vascular, perivascular-like)	RGS5, MCAM/CD146, PDGFRB, CAV1, TAGLN	Perivascular or pericyte-like traits; vascular niche	Breast cancer; defined by scRNA-seq and spatial data	[140]
cCAF (cycling/proliferative)	MKI67, TOP2A and other cell-cycle genes	High proliferative activity; often coexists with other CAF states	Breast and others	[136]
dCAF (developmental-like)	Developmental/stemness programs (cohort-dependent)	Developmental-like transcriptional state; linked to tissue niche and origin	Breast and others	[141]
rCAF (restraining, Meflin+)	ISLR/Meflin; BMP pathway regulation	Associated with better prognosis and ICB sensitivity; can be pharmacologically induced	Pancreas, NSCLC, genitourinary tumors	[135, 142]
LRRC15+ CAF (TGF-β–driven myCAF subset)	LRRC15; TGF-βR2 axis	Establishes protumor stromal homeostasis; suppresses CD8+ T cells function; limits ICB response	Multiple cancers; pancreatic models	[143]
IFN-licensed CAF (ilCAFs)	High IRF1; IFN-γ/STING pathway; CCL4, CCL5; SLC14A1+ (bladder)	IFN-driven, immune-promoting; increase treatment sensitivity with radiation, chemotherapy, or inflammation; linked to response	Colorectal, bladder; pan-cancer	[144, 145]
CD10+ GPR77+ CAF	CD10, GPR77	Maintains cancer stem cells and chemoresistance	Multiple cancers	[146]
EndoMT-CAF (endothelial–mesenchymal transition)	Dual endothelial and mesenchymal lineage features	Reflects vascular–stroma plasticity; associated with survival stratification	Pan-cancer single-cell integrations	[37]
SFRP4+ CAF	SFRP4	Inhibits or modulates tumor-cell migration (cohort-dependent)	Breast cancer	[147]

CAF identities are increasingly interpreted as transcriptional states that exist along a dynamic continuum, rather than as strictly fixed, discrete subtypes [149]. This perspective is supported by pan‑cancer single‑cell analyses, which reveal shared CAF programs and state plasticity across tumor types [150]. Metabolic stress drives CAF plasticity by inducing reprogramming that activates fibroblasts into tumor-promoting CAFs [151]. Hypoxia can induce epigenetic reprogramming in fibroblasts, promoting a transcriptome that is both pro-glycolytic and characteristic of activated CAFs. In breast cancer models, epigenetic reprogramming drives a shift in CAFs glucose metabolism, increasing lactate and pyruvate production. This metabolic remodeling is consistent with an inducible, high-glycolysis program [32]. These findings indicate that a high-glycolysis CAF state can be induced and reprogrammed within the dynamic CAF continuum, reflecting the inherent plasticity of glyCAF-like phenotypes.

Functionally, this metabolic plasticity allows glyCAFs to function within TME. In soft-tissue sarcoma models, glyCAFs utilize GLUT1-driven glycolysis to sustain CXCL16 production, thereby excluding CD8⁺ T cells from the tumor core. Inhibiting GLUT1 or glycolysis reduces CXCL16 and restores T‑cell infiltration, thus supporting the functional reprogrammability of this glycolysis‑high stromal state [17]. A related phenomenon is observed in pancreatic ductal adenocarcinoma, where so‑called “metabolic CAFs” exhibit a similar glycolysis‑linked program and have been associated with differential responses to immunotherapy in reported cohorts [50]. Together, current evidence supports glyCAFs as an inducible glycolysis high CAF state that can be remodeled by microenvironmental and metabolic constraints, while cross cancer lineage stability remains unproven and requires spatially resolved validation in patient cohorts.

Comparison with other chemotactic axes

The glyCAF–GLUT1–CXCL16–CXCR6 pathway differs fundamentally from the canonical CXCL12–CXCR4 axis in its underlying mechanism, spatial detectability, and translational therapeutic potential (Table 5). The CXCL12–CXCR4 axis establishes an immunosuppressive niche by recruiting regulatory T cells and upregulating PD-L1 via JAK-STAT and PI3K-AKT signaling [86, 152], and is predominantly regulated by transcriptional programs.

Table 5.

A comparative overview of the glyCAF–CXCL16–CXCR6 and CXCL12–CXCR4 chemotactic axes

Feature	glyCAF–CXCL16–CXCR6 Axis	CXCL12–CXCR4 Axis
Core Function	Spatial retention of CD8+ T cells at the tumor–stroma interface	Broad recruitment of immunosuppressive cells (e.g., Tregs, MDSCs) and induction of immune tolerance
Upstream Driver	GLUT1-dependent glycolysis in a specific CAF subset (glyCAFs)	Hypoxia, inflammatory cytokines (e.g., NF-κB signaling)
Key Downstream Effect	Enforces CD8+ T cells exclusion from the tumor core	Promotes an immunosuppressive niche and upregulates PD-L1 via PI3K-AKT and JAK-STAT pathways
Spatial Specificity	High; restricted to glyCAF-rich invasive margins	Low to moderate; often diffuse within the tumor bulk
Biomarker Potential	High; defined by spatial assays (e.g., spatial transcriptomics)	Moderate; typically assessed via bulk mRNA or circulating protein level
Therapeutic Targeting	Multi-node: Upstream (GLUT1, e.g., BAY-876); Downstream (CXCL16/CXCR6 blockade)	Primarily downstream: Receptor blockade (e.g., AMD3100)
Therapeutic Rationale	Reverses metabolic-immune chemokine retention axis to alleviate T-cell exclusion	Disrupts immunosuppressive cell recruitment and overcomes stromal barrier
Monotherapy Potential	Promising due to spatial and metabolic specificity	Limited due to pathway redundancy and broad receptor expression
Combination Synergy	Potentiates chemotherapy, radiotherapy, and immune checkpoint blockade	Largely confined to combination regimens to improve T-cell infiltration

In contrast, the glyCAF axis functions as a cohesive, metabolically regulated circuit. GLUT1-dependent glycolysis in CAFs sustains CXCL16 production, which directly mediates CD8⁺ T cells retention at the tumor periphery via CXCR6. This direct coupling of metabolism to spatial immune positioning offers druggable nodes at both levels.

Clinically, measurement of the CXCL12–CXCR4 axis is often confounded by its diffuse expression, whereas the glyCAF signature is spatially enriched at the invasive margin, enabling more accurate quantification via spatial transcriptomics or multiplex immunofluorescence [153, 154]. Therapeutically, CXCR4 monotherapy has yielded limited efficacy [86]. Conversely, upstream GLUT1 inhibition (e.g., with BAY-876) simultaneously disrupts the metabolic driver, chemokine output, and spatial architecture of immune exclusion, promoting T-cell infiltration [17] and sensitizing tumors to conventional therapy [155]. Alternatively, direct downstream blockade of the pathway—through neutralizing CXCL16 or inhibiting its receptor CXCR6—offers a complementary strategy to release trapped T cells and reverse immune exclusion.

Thus, the glyCAF–CXCL16–CXCR6 axis presents a more tractable and spatially defined target for overcoming stromal-driven resistance.

Future directions

Future research on the glyCAF–GLUT1–CXCL16–CXCR6 axis should advance along several complementary fronts to bridge mechanistic understanding and clinical translation.

First, defining the ontogeny and transcriptional regulation of glyCAFs is crucial. Key priorities include elucidating hypoxia and HIF-1α-driven transcription, identifying developmental lineages, and investigating potential transdifferentiation from endothelial or myeloid precursors. Such work will clarify why metabolically dominant CAF subsets emerge in specific tumor regions. Recent insights into CAF plasticity [156], hypoxia responses [31, 157], and HIF-1α–mediated metabolic control [158] strongly implicate the hypoxia–HIF–GLUT1 pathway as a key upstream driver.

Second, clarifying whether the glyCAFs state represents a recurrent metabolic phenotype within the broader CAF continuum will demand a detailed study of CAF transitions across activation trajectories. This will help distinguish whether the observed glycolysis‑high, GLUT1‑high profile reflects a transient adaptation or a durable, metastable CAF entity, a distinction that is essential for assessing its biomarker potential. Therefore, future work should employ definitive lineage‑tracing models, spatially resolved multi‑omics, and validation in annotated therapeutic cohorts to rigorously establish the stability and predictive utility of the glyCAFs state.

Third, a multimodal spatial-omics strategy is essential to dissect the niche-level interactions between glyCAFs and immune cells at the invasive margin. The primary goal is to map their dynamic co-evolution, specifically charting the co-localization of glyCAFs with CXCR6⁺ tissue-resident memory T cells, Tregs, and NKT cells. Understanding how this spatial proximity shapes immune cell metabolism, survival, and effector function requires a unified technical approach. This should integrate single-cell spatial transcriptomics, multiplexed protein imaging (e.g., CODEX), and spatial metabolomics using mass spectrometry imaging to map in situ nutrient gradients like glucose and lactate [159–161]. Crucially, these metabolic maps must be co-registered with spatial proteomic data from platforms such as imaging mass cytometry, which localize key effector proteins like CXCL16 within the same microenvironment [162–164]. This integration is non-negotiable because CXCL16 function is form-specific: membrane-bound CXCL16 mediates adhesion and retention, while its soluble form, cleaved by ADAM10, acts as a chemoattractant. Discerning these distinct functional pools spatially is key to separating mechanisms of immune cell retention from those of recruitment. Through such correlated multimodal analysis can we mechanistically link local glycolysis activity in glyCAFs to chemokine-directed immune positioning and exclusion.

Fourth, the translational path requires preclinical models that can evaluate multi‑target combination therapies while also elucidating their underlying mechanisms. Orthotopic or patient-derived xenograft models offer a relevant setting for testing regimens that concurrently inhibit GLUT1 using agents such as BAY‑876, neutralize CXCL16, and block CXCR6. Beyond assessing therapeutic efficacy, these models should also enable precise cell-type-specific perturbations to establish causality. A definitive experimental sequence would include: selectively deleting CXCL16 in CAFs to examine its effect on CD8⁺ T‑cell positioning at the tumor margin; disrupting CXCR6 in T cells to test ligand‑receptor dependency; and finally rescuing CXCL16 in CAFs to determine whether stromal-derived CXCL16 is both necessary and sufficient for marginal retention rather than a passive niche correlate. The therapeutic synergy of such strategies with chemotherapy, radiotherapy, or immune checkpoint blockade must then be rigorously evaluated using quantitative metrics, including depth of T‑cell infiltration, tumor microenvironment remodeling, and long‑term tumor control.

Finally, to validate the clinical relevance of these spatially derived biomarkers (e.g., glyCAF/GLUT1/CXCL16 signatures), studies on multi-institutional cohorts are essential. This requires validating, in multi-institutional cohorts, whether the spatial abundance of glyCAFs and enrichment of GLUT1/CXCL16 at the tumor margin correlate with patient outcomes following chemo-radiotherapy or immunotherapy. These findings should then inform the design of biomarker-driven early-phase trials. For instance, patients exhibiting a high CXCL16/low CD8⁺ T cells infiltration signature could be prioritized for trials combining GLUT1 inhibition with checkpoint blockade or radiotherapy, creating an efficient pipeline from mechanistic insight to actionable combination therapy.

Conclusions

This review establishes a cohesive framework tracing the glyCAF–GLUT1–CXCL16–CXCR6 axis from stromal metabolic reprogramming to spatial immune exclusion and ultimately to therapeutic opportunity. Glycolytic reprogramming driven by GLUT1 in glyCAFs, upregulates the expression and secretion of CXCL16. At the tumor margin, CXCL16 binds CXCR6 on CD8⁺ T cells, this chemotactic and spatial coupling creates retention niches by preventing T‑cell entry into the tumor core, that lead to an immune-excluded phenotype. This spatial mechanism links stromal metabolism directly to immune positioning.

From a translational and clinical perspective, the glyCAFs—GLUT1—CXCL16—CXCR6 axis has two types of value: predictive value and actionable value. On the predictive side, its spatial readouts, including the abundance of glyCAFs at the margin, the enrichment of GLUT1 and CXCL16 on CAFs, and the nearest neighbor distance to CD8⁺ T cells, serve as practical markers of immune-excluded tumors. On the actionable side, the pathway can be targeted at both ends. Upstream GLUT1 can be inhibited; CXCL16 can be neutralized or CXCR6 can be blocked. Used alone or in combination, these methods alleviate margin retention, promote CD8⁺ T cells infiltration, and enhance the effects of chemotherapy, radiotherapy, and immune-checkpoint inhibitors. This creates a clear experimental and clinical route for combination strategies that relieve immune exclusion, strengthen antitumor immunity, and sensitize tumor killing.

However, the clinical translation of this mechanism requires systematic validation. Future efforts should prioritize five key objectives. First, elucidating the ontogeny and upstream drivers of glyCAFs—such as hypoxia-responsive transcriptional programs and cellular lineages—is essential to understanding their emergence in specific tumor niches. Second, determining whether the glyCAFs state is a stable, recurrent metabolic phenotype within the CAF continuum. This is crucial for its biomarker reliability and requires lineage-tracing, spatial multi-omics, and longitudinal validation. Third, multimodal spatial profiling must integrate metabolic and proteomic maps to resolve how glycolytic gradients and membrane-bound versus soluble CXCL16 colocalization collectively direct immune cell retention or recruitment at the tumor margin. Fourth, preclinical models must enable both multi‑target combination therapy testing and cell‑type‑specific mechanistic validation to establish direct causality and therapeutic synergy. Fifth, orthogonal preclinical models should be employed to evaluate combination therapies targeting GLUT1, CXCL16, and CXCR6, with clinical cohorts prospectively analyzed to link spatial biomarkers such as glyCAFs abundance and CXCL16 enrichment to patient outcomes, thereby enabling biomarker-stratified early-phase trials.

In summary, while the complete axis currently rests on limited studies and requires further validation across tumor types, the glyCAF-GLUT1–CXCL16–CXCR6 axis offers both a mechanistic framework linking metabolic reprogramming with immune exclusion and a translational roadmap from molecular discovery to biomarker-guided therapies, positioning it as a pivotal target for overcoming immunotherapy resistance in solid tumors. Precise intervention along this axis may contribute to overcome immune exclusion in solid tumors and improve the efficacy of immunotherapy, radiotherapy, and chemotherapy.

Abbreviations

CAFs

Cancer-associated fibroblasts

glyCAFs

glycolysis-driven cancer-associated fibroblasts

TME

Tumor Microenvironment

GLUT1

Glucose transporter 1

CAIX

Carbonic anhydrase IX

Cav-1

Caveolin-1

HIF-1α

Hypoxia-inducible factor-1α

LDHA

Lactate dehydrogenase A

MCT4

Monocarboxylate transporter 4

NF-κB

Nuclear factor kappa-B

OXPHOS

Oxidative phosphorylation

PDGF

Platelet-derived growth factor

PFKFB3

6-phosphofructo-2-kinase/fructose-2,6-bisphosphatase-3

ROS

Reactive oxygen species

STS

Soft-tissue sarcoma

ACT

Adoptive cell therapy

CNS

Central nervous system

BBB

Blood–brain barrier

EV

Extracellular vesicle

ICANS

Immune effector cell-associated neurotoxicity syndrome

PD

Pharmacodynamics

Nab

Neutralizing antibody

TNBC

Triple-negative breast cancer

CTL

Cytotoxic t lymphocyte

DC

Dendritic cell

T_RM

Tissue-resident memory T cell

mIF

Multiplex immunofluorescence

α-SMA

α-Smooth muscle actin

ACTA2

Actin alpha 2, smooth muscle

apCAF

Antigen-presenting cancer-associated fibroblasts

BMP

Bone morphogenetic protein

CD10

Cluster of differentiation 10

CCL4

C-C motif chemokine ligand 4

CIITA

Class II major histocompatibility complex transactivator

COL1A1

Collagen type I alpha 1 chain

CODEX

CO-detection by indEXing

CRISPR

Clustered regularly interspaced short palindromic repeats

cCAF

Cycling/proliferative cancer-associated fibroblasts

CXCL12

C-X-C motif chemokine ligand 12

CXCR6

C-X-C motif chemokine receptor 6

dCAF

Developmental-like cancer-associated fibroblasts

DPP4

Dipeptidyl peptidase 4

ECM

Extracellular matrix

EndoMT

Endothelial-mesenchymal transition

FAP

Fibroblast activation protein

GPR77

G Protein-coupled receptor 77

GeoMx

GeoMx digital spatial profiler

HLA-DRA

Human leukocyte antigen-D related alpha

HK2

Hexokinase 2

ICB

Immune checkpoint blockade

ICI

Immune checkpoint inhibitors

IHC

Immunohistochemistry

IL-6

Interleukin-6

IMC

Imaging mass cytometry

IFN

Interferon

IFN-γ

Interferon-gamma

ilCAFs

IFN-licensed cancer-associated fibroblasts

IRF1

Interferon regulatory factor 1

ISLR

Immunoglobulin superfamily containing leucine-rich repeat

iCAF

Inflammatory/Immunoregulatory cancer-associated fibroblasts

LIF

Leukemia inhibitory factor

LRRC15

Leucine-rich repeat containing 15

LRRC15⁺ CAF

LRRC15-expressing cancer-associated fibroblasts

mCAF

Matrix/Remodeling cancer-associated fibroblasts

MCAM

Melanoma cell adhesion molecule

Meflin

Mesenchymal stromal cell- and fibroblast-expressing linx paralogue

MHC-II

Major histocompatibility complex class II

MMPs

Matrix metalloproteinases

MKI67

Marker of proliferation Ki-67

NND

Nearest-neighbor distance

NK

Natural killer cells

NKT

Natural killer T cells

NSCLC

Non-small cell lung cancer

PDPN

Podoplanin

PDGFRB

Platelet-derived growth factor receptor beta

POSTN

Periostin

qPCR

Quantitative polymerase chain reaction

rCAF

Restraining/Meflin⁺ cancer-associated fibroblasts

RGS5

Regulator of G-protein signaling 5

scRNA-seq

Single-cell RNA sequencing

SFRP4

Secreted frizzled-related protein 4

SLC14A1

Solute carrier family 14 member 1

SLC2A1

Solute carrier family 2 member 1

shRNA

Short hairpin RNA

STING

Stimulator of interferon genes

TAGLN

Transgelin

TGF-β

Transforming growth factor-beta

TGF-βR2

Transforming growth factor-beta receptor type 2

TOP2A DNA

Topoisomerase II alpha

vCAF/pvCAF

Vascular/Perivascular-like cancer-associated fibroblasts

Visium

Visium spatial gene expression

2-NBDG

2-(N-(7-Nitrobenz-2-oxa-1,3-diazol-4-yl)Amino)-2-Deoxyglucose

PD-1

Programmed death-1

CAR-T

Chimeric antigen receptor T-cell therapy

TCR-T

T cell receptor-T cell therapy

Author contributions

Yunya Liu: Writing – review & editing, Writing – original draft, Formal analysis, Data curation. Chutong Xiong: Writing – original draft, Data curation. Guichen Huang: Writing – original draft, Data curation. Menghao Xu: Writing – original draft. Jinxiao Li: Writing – original draft. Rui Chen: Writing – review & editing, Conceptualization. Fengxia Liang: Writing – review & editing, Conceptualization. Minfeng Zhou: Writing – review & editing, Conceptualization.

Funding

We gratefully acknowledge the support provided by the following research grants: the National Natural Science Foundation of China (82374585), the Youth Science Foundation Project of National Natural Science Foundation of China (82505327), the China Postdoctoral Science Foundation - Hubei Joint Support Program (2025T068HB) and Natural Science Foundation of Hubei Province (2024AFD272).

Data availability

Not applicable

Declarations

Ethics approval and consent to participate

Not applicable

Consent for publication

Not applicable

Competing interests

The authors declare that there is no conflict of interest.