Fibroblasts in the tumor microenvironment: heterogeneity and dynamic interactions in tumor progression revealed by spatial transcriptomics

Abstract

Cancer-associated fibroblasts (CAFs) constitute the most abundant and functionally versatile stromal component of the tumor microenvironment (TME), with their phenotype and spatial location jointly governing tumor growth, immune evasion, metastasis, and therapeutic resistance. While traditional single-cell RNA sequencing has unveiled CAF transcriptional heterogeneity, it forfeits crucial tissue-contextual information; the advent of high-resolution spatial transcriptomics (ST)—encompassing platforms such as 10xVisium, Slide-seq, Stereo-seq, and MERFISH—now overcomes this limitation by preserving native tissue architecture while simultaneously capturing whole-transcriptome data and single-cell spatial coordinates, enabling comprehensive mapping of CAF “spatial atlases.” Recent literature has consolidated CAFs into five functional subtypes: myofibroblastic CAFs (myCAFs), inflammatory CAFs (iCAFs), antigen-presenting CAFs (apCAFs), matrix-remodeling CAFs (matCAFs), and proliferative CAFs (pCAFs), each exhibiting spatial preferences and dynamic plasticity within tumor cores, hypoxic niches, invasive fronts, and tertiary lymphoid structures. Distinct subpopulations form sub-micron-scale interaction networks with SPP1⁺ macrophages, CXCL13⁺ CD8⁺ T cells, natural killer cells, or endothelial cells to orchestrate either immune exclusion or activation. Multi-cancer investigations in colorectal, pancreatic, hepatic, and lung malignancies demonstrate that peri-tumoral enrichment of POSTN⁺ myCAFs predicts immune exclusion and shortened survival, whereas therapeutic targeting of CAF-immune or CAF-cancer signaling axes—such as IL-34/CSF1R, TGF-β/LOXL2, and JAG1/NOTCH1—can reverse immunotherapy resistance. Looking forward, integrative multi-omics, subcellular-resolution in-vivo tracking, and AI-driven spatial interaction modeling will further decode CAF spatial phenotypes and expedite their incorporation into precision oncology frameworks.

Introduction

Cancer is a leading global cause of death [1]. Research shows that tumor microenvironment (TME) is a decisive factor in tumor growth, invasion, and metastasis, and significantly impacts cancer patients’ treatment responses [2–5]. The dynamic changes in TME during cancer development and its complex interactions with cancer cells increase tumor heterogeneity, posing a central challenge in cancer therapy [6]. To explore new therapeutic strategies, in-depth research on TMEs cellular composition and spatial interactions is crucial [7]. TME is a complex network containing extracellular matrix (ECM), cancer-associated fibroblasts (CAFs), endothelial cells, adipocytes, nerves, smooth muscle, vascular cells, and immune cells [8]. Evidence shows that cells in different spatial locations within TME have diverse interactions, which functionally reshape TME, promoting tumor progression and evolution [9–11]. CAFs and immune cells are the most common stromal cells in TME. In addition to immune cell heterogeneity, CAFs’ heterogeneity is important in regulating tumor growth [12]. CAFs’ characteristic markers include α-smooth muscle actin (α-SMA), S100 calcium-binding protein A4 (S100A4), vimentin (VIM), fibroblast activation protein (FAP), platelet-derived growth factor receptor α (PDGFR α), and platelet-derived growth factor receptor β (PDGFRβ) [12]. TME also contains other CAF subtypes, which exhibit different biomarker expression profiles and functional properties, and participate in maintaining cancer stem cells (CSCs), promoting chemoresistance, and facilitating metastasis by secreting various matrix factors [13].

The emergence of single-cell RNA sequencing technology (scRNA-seq) has provided a powerful tool for analyzing TMEs biological properties and cellular heterogeneity [14–16]. However, scRNA-seq requires tissue digestion to release cells, leading to loss of spatial information. To overcome this, researchers use immunohistochemistry (IHC) and immunofluorescence (IF) staining on frozen or formalin-fixed paraffin-embedded (FFPE) tissue sections for precise cell and protein localization. Despite effectively labeling pre-selected proteins, these techniques’ low throughput limits their ability to discover new proteins or map cell types, prompting the development of high-throughput spatial proteomics technologies [1]. In this context, spatial transcriptomics (ST)—defined here as the broad family of technologies that preserve spatial context while profiling gene expression, encompassing both imaging- and sequencing-based approaches—has emerged. ST enables simultaneous analysis of a sample’s molecular and cellular composition and cellular interactions while preserving tissue structure [5]. It visualizes immune and cancer cell spatial distribution, detects molecular mechanisms driving or inhibiting effective antitumor responses, and aids in understanding therapy responses and resistance across various treatment modalities, including immunotherapy [17–19]. Compared to traditional low-throughput techniques, STs high-throughput advantage captures whole-transcriptome information, showing great potential in constructing spatial cell maps of complex tissues like embryos, brains, and hearts [20]. ST is also widely used in describing cancer’s spatial heterogeneity, helping discover new cell types and infer cellular interactions through co-localization patterns [21]. This supports understanding spatial-specific prognostic factors, revealing mechanisms affecting tumorigenesis, identifying malignant tumor characteristics, monitoring tumor responses to therapy, and opening up new diagnostic and therapeutic strategies for cancer research [22–25].

Currently, elucidating mechanisms regulating CAFs’ function is a key focus in cancer research. Advances in flow cytometry and single-cell transcriptomics have provided unprecedented high-resolution analysis of CAFs’ state. Therefore, this review focuses on exploring CAFs’ impact on tumor progression in TME using ST, emphasizing how emerging ST and imaging technologies provide mechanistic insights, deepening our understanding of cancer pathogenesis, prognosis prediction, treatment, and prevention.

Cancer-associated fibroblasts: origins and roles in tumor progression

CAFs, the predominant stromal component in the TME, exhibit remarkable phenotypic heterogeneity. This heterogeneity arises from diverse cell types and transformation processes, including activated normal fibroblasts, epithelial-mesenchymal transition, and endothelial-mesenchymal transition. CAFs may originate from tissue-resident cells or bone marrow-derived precursors recruited from the circulation and differentiated upon tumor colonization [26]. Additionally, pericytes, endothelial cells, and epithelial cells can transdifferentiate into CAF-like phenotypes [27–29]. As activated fibroblasts, CAFs are a major stromal component in malignant solid tumors, contributing to TME remodeling, inflammatory cell activation [2, 30], and poor prognosis across various cancer types [31]. They significantly impact metastasis, immune evasion, and therapy resistance in cancers like pancreatic, biliary tract, lung, and breast cancer, making them promising therapeutic targets [32–36]. However, CAFs’ roles in cancer are complex and context-dependent, with different subpopulations potentially promoting or inhibiting tumor progression [37–39]. Recent single-cell and ST work has converged on a five-subtype taxonomy of CAFs that couples discrete molecular signatures to distinct functional programs. myCAFs express high α-SMA, PDGFRβ and extracellular matrix (ECM) components such as COL1A1 and POSTN; by generating contractile force and depositing ECM they stiffen the tumor microenvironment, thereby promoting metastasis and—via TGF-β signaling—immunotherapy resistance in pancreatic cancer and portending poor prognosis in lung cancer. iCAFs, marked by abundant IL-6, IL-11, LIF and CXCL12, fuel JAK/STAT3-mediated chemoresistance and immunosuppression; in pancreatic ductal adenocarcinoma (PDAC) they adopt a CXCR4- and FGG⁺/CRP⁺ signature and replace islets, amplifying local inflammation. Antigen-presenting CAFs (apCAFs) are rare cells that display major histocompatibility complex class II (MHC-II) molecules (HLA-DRA, CD74) and co-stimulatory ligands (CD80/86), enabling them to prime CD4⁺ T cells; paradoxically, in pancreatic tumors they appear to expand regulatory T cells and dampen anti-tumor immunity. Matrix CAFs (matCAFs) specialize in ECM remodeling, secreting Matrix metallopeptidase 11 (MMP11), Matrix metallopeptidase 14 (MMP14) and the collagen cross-linker Lysyl oxidase-like 2 (LOXL2); in breast cancer they co-localize with dense collagen fibers to construct metastatic niches. Finally, proliferative CAFs (pCAFs), distinguished by Ki-67 and cell-cycle genes (CCNB1, MKI67), undergo clonal expansion and drive aggressive disease; in cervical cancer they secrete SEMA3C and CXCL6 to enhance cancer-cell stemness. Collectively, this pentad provides a refined framework for understanding CAF heterogeneity and its therapeutic implications.

Notably, these subtypes display plasticity, interconverting in response to TME cues (e.g., TGF-βdrives myCAF differentiation, while interleukin-1α (IL-1α) promotes iCAF phenotypes) [40]. ST further reveals their spatial segregation: myCAFs dominate tumor cores [41], iCAFs accumulate in hypoxic niches [42], and apCAFs reside near tertiary lymphoid structures [43]. This refined classification underscores CAF heterogeneity as a critical determinant of tumor progression and therapeutic outcomes. (Fig. 1).

Fig. 1.

Subtypes and Functions of CAF

Advances in spatial transcriptomics technology

Building on the conceptual framework outlined above, we next summarize the technological principles that enable ST to capture CAF heterogeneity in situ. ST, an integration of imaging, biomarker analysis, sequencing, and bioinformatics, can precisely locate and quantify mRNA transcripts, enabling accurate spatial mapping of gene expression in tissue sections [21, 44]. It also reveals the spatial distribution of diverse cell types, explores cell population interactions, and constructs gene expression maps of different tissue regions. Recent technological advances have accelerated data acquisition and processing, enhanced quality, and achieved higher spatial resolution. This innovative technology shows great potential in uncovering complex mechanisms of various diseases, including cancer. Depending on the detection method, spatial transcriptomic techniques are mainly categorized into two types: imaging-based and sequencing-based [45]. (Fig. 2).

Fig. 2.

Development of imaging- and sequencing-based ST technologies

Imaging-based spatial transcriptomic technologies

Imaging-based techniques in ST, including In Situ Hybridization (ISH) and In Situ Sequencing (ISS), are crucial for visualizing and quantifying mRNA transcripts within tissue sections. ISH, leveraging complementary single-stranded nucleic acid molecules, enables target RNA detection and imaging in cells or tissues via oligonucleotide probes, allowing quantification and spatial localization under a microscope [46]. Since its inception in 1982 [47], ISH has evolved from radioactive probe labeling [46] to Fluorescence In Situ Hybridization (FISH) [47]. FISH, a common ISH method, uses fluorescent probes for RNA localization. However, traditional FISH has limitations in signal intensity and stability. To address these, single-molecule FISH (smFISH) was developed by integrating FISH with digital imaging microscopy, offering enhanced sensitivity and specificity for single-molecule RNA analysis [48]. This technique marks each DNA or RNA molecule with multiple fluorophores, enabling probe-labeled transcript imaging with stronger and more stable signals than traditional FISH [49]. Yet, due to its traditional riboprobe-based approach, smFISH still has limitations in specificity, sensitivity, and multiplexing capability. The emergence of RNAscope introduced a novel “double Z” probe design for signal amplification and background suppression [50]. However, it still faces issues like probe hybridization noise, non-specific detection, and photobleaching, which can affect the accuracy of RNA quantification and localization. Subsequent technologies like cyclic single-molecule FISH (osmFISH) [51], Sequential FISH (seqFISH) [49], Sequential FISH plus (seqFISH⁺) [52], and Multiplexed Error-Robust Fluorescence In Situ Hybridization (MERFISH) [53] further improved detection efficiency. MERFISH, through sequential hybridization and imaging rounds combined with barcode decoding, has been successfully applied in ST studies from single-cell transcript localization to tissue-level analysis [54, 55]. Despite these advancements, ISH techniques still have limitations in detecting novel or polymorphic transcripts, with challenges in probe design, cross-reactivity, and fluorescence drift.

In contrast to ISH, ISS significantly boosts throughput and captures transcripts at subcellular resolution. It uses micrometer or nanometer-sized DNA beads for signal amplification during sequencing. Notable ISS techniques include ISS, Fluorescent In Situ RNA Sequencing (FISSEQ), Hybridization-based In Situ Sequencing (HybISS), Expansion Sequencing (ExSeq), and Spatially-resolved Transcript Amplicon Readout Mapping (STARmap). The first ISS method, developed by Ke et al. in 2013, was based on single-molecule RNA detection technology [56, 57]. This method utilized barcode padlock probes linked to target RNA-generated cDNA, with signal enhancement through rolling circle amplification (RCA) and barcode decoding via ligation sequencing chemistry. By associating coordinate information from images with decoded gene-specific barcodes, it enabled RNA expression detection in the native tissue environment. Specifically, RNA underwent reverse transcription, followed by signal amplification through RCA and subsequent sequencing. This method employed target probes for reverse transcription and decoded barcodes via ligation sequencing [56]. It has been widely used in cancer [58], tuberculosis [59], and brain development research [60]. Later, technologies like BaristaSeq [61] and STARmap [62] further improved detection sensitivity and the number of detectable genes. STARmap, combining advancements in hydrogel chemistry, optimized padlock probes and primers, and introduced a ligation sequencing strategy, enabling high-resolution spatial expression analysis of thousands of genes in the mouse cortex [62]. Other synthetic sequencing methods, such as Barseq [63] and HybISS [64], increased throughput and cell barcode detection efficiency by extending read lengths. However, most ISS-based spatial transcriptomic technologies, while achieving subcellular resolution, are often limited to a small number of genes or have lower detection efficiency [65], restricting their application in specific scenarios.

Sequencing-based methods within spatial transcriptomics

Sequencing-based ST platforms, including the original array-based ST protocol, Slide-seq, and next-generation sequencing (NGS)-driven approaches, Laser Capture Microdissection (LCM), In Situ Barcoding (ISB), and 10x Genomics Visium, have revolutionized the field of gene expression analysis. The array-based ST protocol, for instance, captures, sequences, and maps RNA molecules within tissue samples at high spatial resolution, providing a powerful tool for gene expression profiling. In this method, reverse transcription primers immobilized on a solid phase release and capture RNA molecules at their original locations, generating cDNA tagged with positional barcodes. Tissue samples undergo fixation, staining, imaging, and permeabilization, allowing mRNA molecules to diffuse vertically to the chip surface, where they hybridize with poly(dT) reverse transcription (RT) primers for in situ reverse transcription. The synthesized barcoded cDNA is then sequenced and decoded, enabling the mapping of expressed genes back to their original tissue context [66]. In the initial ST protocol, the capture area consisted of 1,040 spots, each 100 μm in diameter. To achieve higher spatial resolution, Chen et al. developed Slide-seq, which utilizes DNA-barcoded microbeads to achieve a resolution of 10 μm. However, due to the random deposition of microbeads on the slide surface, sequencing is required to decode the barcodes and associate spatial information with captured RNA, a process that may reduce efficiency and increase costs [67]. In contrast, NGS-based ST methods have significantly enhanced the throughput and efficiency of gene detection. These methods build upon single-cell RNA sequencing technologies and incorporate spatial barcoding during library preparation [68]. In 2016, Stahl et al. first reported an NGS-based ST approach capable of capturing full transcriptome information from tissue sections [69]. The core innovation of this method lies in the use of microarray slides with spatial barcodes to capture polyadenylated RNA, ensuring that each transcript is mapped back to its original location in the tissue via a unique positional barcode. Each slide contains over 1,000 spots (100 μm in diameter, 200 μm center-to-center spacing), enabling unbiased analysis of large tissue areas without the need for pre-selecting specific regions or gene targets [70, 71]. This approach, subsequently commercialized as 10x Genomics Visium, was first validated in mouse olfactory bulb tissue [69] and has since been widely adopted and optimized by numerous research teams [72–76]. LCM-based methods, such as Geographical position-sequencing (Geo-seq) [77], Transcriptome In Vivo Analysis (TIVA) [78], and Neighbouring cell-Interaction Histology and Enriched-cell sequencing (NICHE-seq) [79], employ laser beams to excise specific tissue regions identified under a microscope [80, 81]. Compared to the original LCM approach, Geo-seq offers improved sensitivity but lower resolution; TIVA is applicable to live cells but has limited throughput; and NICHE-seq provides higher throughput but is not suitable for human samples. Overall, LCM-based ST techniques are labor-intensive, low-throughput, and unsuitable for batch processing.

ISB-based ST technologies capture RNA molecules in situ, followed by off-site cDNA sequencing. ISB methods can be divided into two groups based on barcoding strategies. The first group employs solid-phase barcoding capture (SPBC), where tissue is transferred to a substrate with pre-arranged DNA barcodes. This group includes 10x Genomics Visium [69], Slide sequencing (Slide-seq) [67], Slide sequencing 2.0 (Slide-seq2) [82], High-Definition ST (HDST) [83], Spatio-temporal enhanced resolution omics sequencing (Stereo-seq) [84], Sequencing-scope (Seq-Scope) [85], Ascorbate peroxidase-mediated proximity labeling and sequencing (APEX-seq) [86], and Pixelated ST sequencing (PIXEL-seq) [87]. The second group, including NanoString Digital Spatial Profiling (DSP) [88] and ZipSeq [89], utilizes selective barcoding, where DNA barcodes are collected from or delivered to specific tissue locations. ISB-based ST has been widely applied in studies of the mouse olfactory bulb, gingival tissue, adult heart tissue, and various cancers [20]. These ST technologies not only enable the analysis of cellular composition and distribution within the TME but also facilitate the construction of spatial trajectories and interaction networks, deepening our understanding of cancer pathogenesis, prognosis, treatment, and prevention. With the rapid advancement of ST technologies and their accompanying bioinformatics tools, challenges related to cost, sensitivity, and automation are expected to be addressed in the coming years. Furthermore, the emergence of spatial multi-omics technologies integrates transcriptomics with other omics data, such as proteomics, providing a more comprehensive landscape of cancer biology. This integration has the potential to transform our understanding of cancer and open new avenues for precision medicine.

This rapidly evolving suite of ST technologies provides powerful tools to dissect the complex roles of fibroblasts, particularly CAFs, revealing their remarkable spatial organization and functional heterogeneity within the tumor microenvironment.

CAFs in spatial transcriptomics

Identification and functional analysis of fibroblast subpopulations

Recent advancements in ST have significantly enhanced our understanding of CAFs, uncovering their heterogeneity in cellular composition, gene expression profiles, and spatial localization. Based on genotype, phenotype, or spatial positioning, CAFs can be classified into distinct subpopulations. In CRC, ST has identified three CAF subtypes: COL11A1⁺ INHBA⁺ CAFs [42], MFAP5⁺ CAFs [90], and FAP⁺ CAFs [91]. In CRC, Zheng et al. used 10x Genomics Visium to capture transcriptomes across whole tissue sections, thereby identifying COL11A1⁺ INHBA⁺ CAFs within hypoxic microenvironments; these cells correlate with maintaining cancer stemness traits. However, Visium’s 100 µm spot spacing was unable to resolve subcellular interactions between CAFs and adjacent cancer cells [42]. Peng et al. further employed higher-resolution Slide-seqV2 to map near-single-cell-level gene expression back to tissue, revealing tight spatial adjacency between MFAP5⁺ CAFs and C1QC⁺ macrophages at 10 µm resolution. Nonetheless, random bead deposition could introduce positioning errors [90]. Qi et al., also using Visium, confirmed co-localization of FAP⁺ CAFs with SPP1⁺ macrophages but could not demonstrate actual physical contact within the 55 µm spot diameter [91]. In CRC, Enfield et al. applied 10×Visium to show α-SMA⁺ myCAFs build ECM barriers that block CD8⁺ T-cell infiltration; however, the 55-µm spot spacing limited subcellular resolution [92].

Hanley et al. performed region-level transcriptome analysis on non-small cell lung cancer (NSCLC) tissue using 10x Visium, first linking CD34⁺ adventitial fibroblasts to an iCAF phenotype; however, the 55 µm spot spacing remained insufficient for capturing submicron-scale communication between fibroblasts and immune cells [93]. Yang et al. mapped the spatial landscape of CAFs in early lung adenocarcinoma using 10x Visium, finding THBS2⁺ fibroblasts enriched at the invasive front. Yet, the platform’s region-level resolution precluded precise boundary definition between CAFs and individual tumor cells [94]. Xun et al. leveraged subcellular-resolution Stereo-seq to precisely localize APSN⁺ CAFs at tumor boundaries while positioning SFRP2⁺ CAFs in non-malignant regions. Nanoscale precision distinguished “pro-invasive” and “pro-remodeling” CAF axes in HCC for the first time, though high costs and probe design complexity limited broad application [95]. Wang et al. integrated 10x Visium with single-cell transcriptomics, revealing primary lesions enriched with F3⁺ CAFs whereas liver metastases were dominated by MCAM⁺ CAFs. Visium provided tissue-wide context but required single-cell data to compensate its 55 µm resolution for inferring direct CAF—CXCL13⁺ CD8⁺ T cell interactions [96]. Sun et al. tracked OSCC progression using 10x Visium, observing T-stage-dependent increases in Mesen⁺ CAFs. The 55 µm spots, however, couldn’t resolve fine boundaries between CAFs and epithelial cells; subsequent single-cell data validated CAF-cancer cell ligand-receptor pairs [97]. Ou et al. combined snRNA-seq with 10x Visium, finding POSTN⁺ myCAFs clustered around cancer nests and associated with immune exclusion. Visium established a region-level immune microenvironment framework but lacked resolution to resolve submicron-scale CAF–T cell contacts [98].

Ferri-Borgogno et al. used GeoMx DSP to perform region-of-interest sequencing at the HGSC tumor-stroma interface, identifying short-term survivors enriched with α-SMA⁺ VIM⁺ PDGFRβ⁺ CAFs forming a physical immune barrier. Despite precise 50–100 µm UV-microdissection, low throughput limited capture of rare CAF-immune cell pairings [99]. In PDAC, diverse ST platforms collectively mapped the CAF “spatial atlas”: Moncada et al. used 10x Visium on whole sections to distinguish myCAFs, iCAFs, and apCAFs, finding iCAFs enriched in tumor peripheries and myCAFs in cores. The 55 µm resolution, however, couldn’t capture subcellular CAF-cancer cell contacts [74]; Ren et al. further applied nanoscale Stereo-seq to reveal replacement of islets by FGG⁺ CRP⁺ iCAFs, suggesting inflammatory CAFs influence endocrine cells through nanoscale adjacency [41]; Hwang et al. employed GeoMx DSP to compare CAF programs pre-/post-neoadjuvant chemotherapy, showing significant enrichment of immunomodulatory programs post-treatment. The low-throughput regional dissection limited detection of rare CAF-immune cell interactions [100]. These studies highlight the spatial interplay between fibroblasts and cancer cells. For instance, POSTN expression in M-type fibroblasts serves not only as a biomarker for Barrett’s esophagus (BE) progression but also correlates with immune suppression in the TME, further underscoring the critical role of fibroblasts in tumor invasion and metastasis [101]. (Fig. 3).

Fig. 3.

Identification of CAF subtypes in diverse cancers using imaging- and sequencing-based ST

Spatial heterogeneity of fibroblasts in the tumor microenvironment

Stromal cells are a critical component of the TME, playing pivotal roles in tumor growth, progression, immune suppression, and metastasis [102–104]. ST has unveiled the spatial distribution preferences of stromal cells within the TME. For instance, in a mouse model of lung cancer, the loss of Tgfbr2 induces stromal remodeling and drives tumor development [105]. Among various cancer types, CAFs exhibit the most pronounced spatial co-localization patterns. In CRC, FAP⁺ CAFs are spatially co-localized with SPP1⁺ macrophages [91], while MFAP5⁺ CAF-derived microfibrils are closely associated with C1QC⁺ macrophages [90]. These interactions are critical in promoting invasive tumor phenotypes. Additionally, Zheng et al. identified a tight spatial interaction between COL11A1⁺ INHBA⁺ CAFs and CD44⁺ CRC cells, highlighting their role in maintaining cancer stem cells [42]. In the liver cancer microenvironment, SPP1⁺ macrophages and CAFs form a spatially defined tumor immune barrier (TIB) niche, which is essential for tumor progression and immune surveillance [106]. In the TME of skin cancers, particularly basal cell carcinoma (BCC), CAFs and macrophage subpopulations exhibit distinct spatial distribution patterns, with their complex interactions contributing to the establishment of an immunosuppressive environment [107]. Moreover, macrophages adjacent to CAFs participate in ECM remodeling, a process closely linked to the collective migration of tumor cells and ECM-remodeling fibroblasts at the invasive front, providing new insights into the mechanisms of BCC invasion [108]. CAFs also maintain strong co-localization patterns with endothelial cells and perivascular cells. In homologous recombination deficiency (HRD) samples, angiogenic interactions between iCAFs and endothelial cells are particularly prominent [109]. In glioblastoma (GBM), CAFs are enriched in tumor stem cell (GSC) and perivascular regions, where they support and nurture GSCs, potentially contributing to therapy resistance [110]. In HRD subsets of liver cancer, fibroblast-endothelial interactions related to angiogenesis are significant, and the enrichment of myCAF subpopulations is associated with poor prognosis, underscoring the role of fibroblasts in tumor angiogenesis and patient outcomes [109]. Davidson et al. found through 10×Visium that interstitial-like tumor cells and POSTN⁺ myCAFs are closely adjacent in the interstitial regions in ccRCC, although the platform provides macroscopic spatial relationships, the 55 µm resolution still fails to capture the potential subcellular-level signaling between them [111]. It is worth noting that although several recent studies ultimately assigned CAF identity using only a single marker such as APSN or SFRP2 [95], this simplification belies the pivotal role that high-throughput spatial transcriptomics played from the outset. An unbiased survey of the whole transcriptome first highlighted these genes as selectively enriched at the tumour–boundary interface, and nanometre-scale Stereo-seq mapping subsequently resolved APSN⁺ CAFs within a 10-µm rim precisely at the invasive margin of hepatocellular carcinoma. There, they establish direct membrane contact with tumour cells and neighbouring SPP1⁺macrophages to assemble an ECM-dense barrier—an anatomical configuration that neither conventional IHC nor dissociated scRNA-seq could have revealed.

In studies of high-grade serous ovarian cancer (HGSC), co-localization analysis of tumor cells and adjacent fibroblasts revealed that the absence of specific fibroblast subtypes is associated with favorable patient prognosis [99]. In pancreatic cancer, stromal tissue is predominantly composed of fibroblasts and endothelial cells, with their co-localization observed in the PDAC microenvironment [74]. Research has identified a strong association between inflammatory fibroblasts and cancer cells expressing the KRT19 gene module, as well as a significant correlation between stress response gene modules and inflammatory fibroblast signatures. These findings deepen our understanding of PDAC biology and treatment responses, providing a scientific basis for developing novel therapeutic strategies [74]. Furthermore, interactions between CAFs and natural killer (NK) cells highlight the critical role of fibroblasts in esophageal squamous cell carcinoma (ESCC) progression, not only in directly supporting tumor cells but also in modulating the immune microenvironment [101]. In the ESCC TME, the most prominent interactions are observed between epithelial cells and fibroblasts, while epithelial cells also exhibit strong interactions with NK cells and T cells, underscoring the importance of these cellular interactions in tumor development and immune surveillance [112].

The presence of CAFs in tumor tissue is closely linked to treatment responses. However, ST-based studies reveal significant spatial and functional heterogeneity among CAFs. In primary CRC tumors, F3⁺CAFs are enriched, whereas MCAM⁺ CAFs dominate in liver metastases, indicating substantial differences in stromal microenvironments across organs [96]. In NSCLC, fibroblasts occupying alveolar and adventitial regions are associated with overall survival in lung adenocarcinoma (LUAD) patients [93]. In liver cancer, APSN⁺ CAFs are enriched at tumor boundaries, SFRP2⁺ CAFs in non-malignant regions, and myCAFs in non-malignant spot areas, with their distribution differences significantly correlating with patient prognosis [95]. In metastatic ovarian cancer, the density of CAFs in stromal clusters is markedly higher in short-term survival samples compared to long-term survival samples, with high densities of α-SMA⁺ VIM⁺ PDGFRβ⁺ CAFs linked to reduced immune infiltration and shorter survival [113]. In ESCC, iCAFs are predominantly localized in stromal regions, while myCAFs show no significant distribution differences between tumor and stromal areas, further revealing the heterogeneous distribution of CAFs in the TME [112]. In PDAC, fibroblasts in the tumor core exhibit immunosuppressive activity, whereas those at the tumor periphery demonstrate immune-enhancing functions. This regional functional duality highlights the dual role of fibroblasts in anti-tumor immunity. Additionally, fibroblasts adjacent to tumor tissue exhibit tumor-suppressive effects, such as through B cell activation and leukocyte regulation [41]. (Fig. 4).

Fig. 4.

Spatial heterogeneity of CAF subtypes revealed by imaging- and sequencing-based ST

Fibroblast spatial interactions with diverse cell types

Clinically, predicting cancer patients’ responses to specific therapies is of paramount importance, yet the inherent complexity and heterogeneity of tumors make this task exceptionally challenging. scRNA-seq has revealed that specific cellular components can significantly influence therapeutic responses across various cancers [114]. ST further demonstrates that the spatial aggregation of CAF subpopulations and their interactions with other cell types can induce distinct treatment outcomes. Currently, beyond tumor cells, the spatial interactions between CAFs and macrophages have been extensively studied. For instance, in the context of CRC, MFAP5⁺ CAFs activate the M2 phenotype of macrophages via the IL34/CSF1R axis and promote an immunosuppressive environment through the MIF/CD74 signaling pathway. These activated myeloid cells, in turn, regulate MFAP5⁺ CAF activation by releasing pro-tumor ligands such as EGF superfamily proteins and NAMPT, thereby driving tumor growth. Additionally, the synergistic signaling between MFAP5⁺ CAFs and pro-tumorigenic FAP⁺ CAFs further consolidates the aggressive CRC microenvironment [90]. Moreover, FAP⁺ CAFs facilitate the transition of THBS1⁺ macrophages to SPP1⁺ macrophages through RARRES2/CMKLR1 signaling. SPP1⁺ macrophages enhance the expression of ECM-related genes in FAP⁺ CAFs by secreting cytokines such as IL1A, IL1B, and TGFB1, thereby modulating the connective tissue microenvironment. This microenvironment may impede lymphocyte infiltration into the core regions of CRC, reducing the efficacy of PD-L1 immunotherapy [91]. In the tumor boundary niche of hepatocellular carcinoma (HCC), the robust interactions between the macrophage subpopulation SPP1⁺Macro and the fibroblast subpopulation APSN⁺CAFs, as well as their interplay with tumor cells, are closely associated with biological processes such as ECM organization, collagen fibril organization, cell-matrix adhesion, positive regulation of chemotaxis, and responses to TGF-β [95]. Jain et al. localized CAFs in the perivascular niche of GBM based on 10× Visium and found that they are enriched in GSC regions; although the platform revealed the macroscopic colocalization of CAFs and blood vessels, the 55 µm sampling interval still requires subsequent single-cell sequencing to verify the direct dialogue between CAFs and endothelial cells [110]. In ESCC, the interactions between CAFs and macrophages (Mφs) are particularly critical at the tumor invasive front. CAFs not only correlate with the expansion density of Mφs but also drive the conversion of M1 to M2 macrophages, a process associated with increased numbers of CD68⁺ or CD163⁺ Mφs, collectively impairing immune cell function and anti-tumor immune responses [115]. (Fig. 5).

Fig. 5.

CAF-macrophage cell interactions revealed by imaging- and sequencing-based ST

Beyond CAF-macrophage interactions, the roles of CAFs in engaging with other immune cells have also garnered attention. For example, in liver metastatic cancer, the CAF subpopulation MCAM⁺CAF interacts with tumor-specific CD8_CXCL13 cells via the JAG1–NOTCH1 signaling pathway, potentially influencing tumor metastasis and immune responses. Additionally, the F3⁺ CAF subpopulation enhances tumor invasiveness through the NRG1–ERBB3 signaling pathway [96]. Kang et al. further observed that myCAFs and iCAFs exhibit spatially mutually exclusive relationships. HRD activates the NF-κB signaling pathway via DNA damage, subsequently triggering the JAK/STAT pathway, which has been shown to promote iCAF differentiation. Conversely, in non-HRD contexts, enhanced TGF-β signaling drives myCAF differentiation [109]. Enfield et al. combined Visium and scRNA-seq to show that, in CRC, α-SMA⁺ myCAFs erect a TGF-β-driven ECM barrier that physically excludes CD8⁺ T cells from tumors, curtailing infiltration and enabling immune evasion; depleting myCAFs or blocking LOXL2 restores T-cell influx and sensitizes tumors to PD-1 blockade [92]. (Fig. 6).

Fig. 6.

CAF-immune cell interactions revealed by imaging- and sequencing-based ST

However, ongoing research continues to delve into the interactions between CAFs and tumor cells. For instance, in metastatic ovarian cancer, highly activated CAFs engage in bidirectional signaling with tumor cells, particularly in short-term survival (STS) samples. The prevalent presence of the APOE-LRP5 ligand-receptor pair suggests that these CAFs may promote tumor growth, rendering tumors more aggressive, poorly differentiated, and stem-like, thereby conferring chemotherapy resistance. Furthermore, CAFs recruit myeloid-derived suppressor cells, impairing T cell function and facilitating immune escape [113]. In NSCLC [93] and LUAD [94], myofibroblasts, particularly those originating from adventitial or alveolar fibroblasts, contribute to tumor progression through inflammatory and stress response signaling [93]. Research has delineated three sequential stages of fibroblast activation: upregulation of inflammatory cytokines, transmission of stress response signals, and eventual enrichment of fibrotic collagen expression, leading to myofibroblast accumulation in tumor samples [94]. In the LUAD TME, CAFs play a pivotal role in tumor progression via the TGF-β1-THBS2 feedback loop. Within this process, THBS2⁺ CAFs and O-glycosylation regulation are particularly significant, as they modulate the CDK4-pRB axis in CAFs at the tumor invasive front, potentially reducing the mesenchymal characteristics of cancer cells and offering a promising therapeutic strategy [116]. In cervical squamous cell carcinoma, myCAFs significantly enhance cancer cell stemness and proliferation by secreting specific factors such as SEMA3C, POSTN, and CXCL6. Simultaneously, myCAFs support tumor growth and metastasis by inhibiting lymphocyte infiltration and remodeling the tumor ECM, thereby inducing resistance to immune checkpoint blockade therapy [98]. In basal cell carcinoma, transcriptional reprogramming has been identified as a key driver of tumor cell invasiveness, underscoring the mutual influence between tumor cells and surrounding fibroblasts and their role in promoting tumor progression through bio-crosstalk, primarily involving epithelial cell characteristics and collective migration [108]. Similarly, in oral squamous cell carcinoma [97] and renal cell carcinoma [111], CAFs interact with cancer cells via signaling pathways such as TGF-β, playing critical roles in cancer initiation and progression and ultimately impacting patient survival. In esophageal cancer [117] and PDAC [118], TGF-β secreted by malignant epithelial cells plays a central role in shaping the local microenvironment phenotype, exacerbating ANXA1 protein loss, which disrupts normal fibroblast homeostasis and drives uncontrolled conversion to myCAFs [117], or inducing higher retinoic acid-related gene expression in tumor-proximal fibroblasts compared to tumor-distal fibroblasts, promoting tumor-proximal CAF activation and supporting tumor growth and dissemination [118]. (Fig. 7).

Fig. 7.

CAF-tumor cell interactions revealed by imaging- and sequencing-based ST

Future perspectives

Unlike dissociation-based approaches such as flow cytometry and scRNA-seq, spatial transcriptomics preserves tissue architecture and, when cross-referenced with single-cell profiles, exposes CAF subtypes that are confined to specific tumour regions. Trajectory analyses of integrated datasets have further traced a path of FAP-high CAFs from the tumour core toward the invasive margin—a spatial pattern that remains hidden in cell suspensions. These observations illustrate how ST complements single-cell technologies by anchoring CAF heterogeneity to physical tissue structure rather than to isolated expression signatures. Currently, ST, as an umbrella term for imaging- and sequencing-based approaches, has demonstrated remarkable capabilities in characterizing spatial heterogeneity and clinical applications, making it a powerful tool in the study of tumor heterogeneity. This technology has deepened our understanding of fibroblasts in the TME, revealing their pivotal roles not only in tumor progression and stromal composition but also in their intricate interactions with tumor cells and immune cells. A comprehensive understanding of fibroblast functions within the TME is not only critical for deciphering tumor biology but also provides a foundation for developing or refining precision therapeutic strategies. Future research should focus on elucidating how fibroblasts regulate stromal composition during tumor progression, as well as their interactions with tumor and immune cells, and explore how these processes can be leveraged to optimize treatment approaches. The rapid advancements in emerging fields such as single-cell genomics, ST, and computational biology have provided powerful tools for achieving the goals of precision oncology. These technologies not only uncover cellular heterogeneity and spatial dynamics within the TME but also offer novel insights for drug target screening, precise clinical diagnostics, and the development of innovative therapeutic strategies. By integrating multi-omics data and computational models, researchers can gain deeper insights into the molecular mechanisms of tumors, thereby advancing the design and optimization of personalized treatment regimens.

Author contributions

XJ and XW designed this study, performed the statistical analysis, and drafted the manuscript; WS and HZ supervised the study. All authors read and gave final approval of the manuscript.

Funding

Liaoning Province Applied Basic Research Program, (2022020225-JH2/1013; The Natural Science Foundation of Liaoning Province (2024-MS-054)).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethical approval

This article does not contain any studies with human participants or animals performed by any of the authors. The Creation of Mechanism Diagrams by Figdraw (https://www.figdraw.com/).

Consent for publication

Before the completed text could be considered for publication, it was first reviewed by each of the authors.

Competing interests

The authors declare no competing interests.