Transgelin defines pro-tumorigenic cancer-associated fibroblasts in pancreatic cancer

Abstract

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is characterised by a pronounced desmoplastic reaction, predominantly composed of cancer-associated fibroblasts (CAFs), including myofibroblastic CAFs (myCAFs).

Methods

We performed a single-cell assay for transposase-accessible chromatin with high-throughput sequencing (scATAC-seq) and single-cell RNA sequencing (scRNA-seq) on pancreas tissues from KPC mice (LSL-Kras^G12D/+; LSL-Trp53^R172H/+; Pdx-1-Cre) to characterise myCAF heterogeneity. A transgelin (Tagln) knockout orthotopic mouse model was used to determine the functional role of Tagln.

Results

Epigenetic profiling uncovered heterogeneity within the myCAF population, revealing distinct subclusters characterised by specific transcription factor (TF) motifs, such as Srf, Cebpb, Prrx1, and Smad4. We identified three transcriptionally distinct myCAF subtypes, each enriched for unique TF–associated signalling pathways. Among the identified myCAF subtypes, Tagln emerged as a potential functional driver. Tagln knockout mice exhibited significantly reduced PDAC tumour burden compared to wild-type. Analysis of TCGA revealed that high TAGLN expression in PDAC samples was associated with poor survival.

Conclusions

Our findings highlight the functional heterogeneity of myCAFs and identify TAGLN-expressing myCAFs as critical mediators of tumour progression, providing evidence that targeting stromal TAGLN may represent a promising therapeutic strategy for PDAC.

Introduction

Pancreatic ductal adenocarcinoma (PDAC), which accounts for approximately 85% of pancreatic malignancies, is associated with a dismal 5-year survival rate of only 9%, and its incidence continues to rise globally[1]. A hallmark of PDAC is the extensive desmoplastic stroma that constitutes the majority of the tumour mass. Among the various stromal components, cancer-associated fibroblasts (CAFs) represent a predominant population, comprising up to 90% of the tumour stroma [2].

Recent single-cell transcriptomic analyses of both mouse and human PDAC have identified three major subtypes of CAFs with distinct profiles. Myofibroblastic CAF (myCAF), located adjacent to the tumour cells, is driven by TGFβ signalling and expresses α-smooth muscle actin (αSMA; ACTA2) and Transgelin (TAGLN). Inflammatory CAF (iCAF), typically situated distally from the tumour cells, is activated via the JAK/STAT3 pathway and expresses interleukin 6 (IL6) and CXCL12 [3, 4]. A third subset, antigen-presenting CAFs (apCAFs), characterised by the expression of CD74 and MHC class II molecules, has been identified primarily in mouse models [4, 5]; their presence in human PDAC appears sporadic [4, 6].

During tumour progression, the myCAF/iCAF ratio increases, reflecting the growing contribution of myCAFs to tumour development [4, 7]. Importantly, high stromal activity, defined by elevated expression of αSMA, is associated with a poor prognosis in PDAC patients[8]. However, efforts to therapeutically target αSMA-positive CAFs in PDAC have produced conflicting outcomes[7, 9], suggesting that αSMA-positive myCAFs are functionally heterogeneous and may include both tumour-promoting and tumour-restraining subpopulations [10].

TAGLN, also known as SM22, is a calmodulin-related, actin-binding protein expressed in activated fibroblasts, where it contributes to cytoskeletal remodelling and cell migration [11, 12]. While TAGLN-deficient mice are viable and fertile, they exhibit impaired smooth muscle contractility [13, 14]. In the tumour context, TAGLN in lung CAFs activates NF-κB signalling and promotes cancer cell migration [15], while in ovarian cancer, TAGLN modulates the RhoA/ROCK pathway in a matrix stiffness-dependent manner [16]. Additionally, TAGLN may function as a transcriptional regulator through its interaction with transcription factors (TFs) such as HIF1α and promotion of their nuclear translocation in glioblastoma cells [17]. Despite its emerging roles, the functional significance of TAGLN expression in myCAFs, particularly in relation to αSMA, remains poorly understood.

Mutations in KRAS, CDKN2A, TP53, and SMAD4 are key drivers of PDAC tumorigenesis[18]. Genetically engineered mouse models that harbour these mutations, such as KPC (LSL-Kras^G12D/+; LSL-Trp53^R172H/+; Pdx-1-Cre)[18] and KPP (Pdx1^cre/+;LSL-Kras^G12D/+;p16/p19^flox/flox)[19], have been established to elucidate the molecular mechanisms underlying PDAC development.

In this study, we employed the single-cell assay for transposase-accessible chromatin with high-throughput sequencing (scATAC-seq) and single-cell RNA sequencing (scRNA-seq) to characterise the heterogeneity of myCAFs in PDAC using the KPC mouse model. We identified three distinct myCAF subpopulations, among which Tagln emerged as a key functional mediator. Tagln knockout mice showed reduced collagen production and immune cell recruitment, suggesting that TAGLN-expressing myCAFs drive PDAC progression. To our knowledge, this is the first study to reveal that TAGLN-expressing CAFs function as a critical determinant of PDAC behaviour, highlighting their potential as a therapeutic target.

Materials and Methods

Animal model

Animal protocols were approved by the Animal Care and Use Committee of Nagoya University Graduate School of Medicine (approval number M240289-002).

The KPC (LSL-Kras^G12D/+; LSL-Trp53^R172H/+; Pdx-1-Cre) mouse model was established as described previously [20]. Mouse pancreas tissues harvested from 3-month-old KPC mice and wild type (WT) mice were minced and dissociated using the tumour dissociation kit (130-096-730, MACS Miltenyi Biotec, Bergisch Gladbach, Germany). The dissociated cells were further used for scATAC-seq and scRNA-seq. The generation and characterisation of Tagln−/− mice (STOCK Tagln^tm1Liw/Mmnc) on a mixed C57BL/6 x SV129 genetic background were described previously[13, 14].

To establish an orthotopic pancreatic model, mT5 cells were resuspended in a 1:4 mixture of D-PBS (Fujifilm Wako Pure Chemical, Tokyo, Japan) and Matrigel (Corning, Corning, NY, USA) at a concentration of 1 × 10⁵ cells/ml. The peritoneum of the anaesthetised 6-week-old Tagln^+/+ mice and Tagln^−/−mice was surgically opened, and 50 µl of cell solution, containing 1000 cells, was injected into the pancreatic tail.

For the subcutaneous tumour model, a mixture of mT5 cells (3 × 10⁶ cells) and CAFs (9 × 10⁶ cells) in a mixture of 50 μl of D-PBS and 50 μl of Matrigel was injected subcutaneously into the back of anaesthetised 9-week-old male Tagln^−/− mice (n = 4). Tumour length (L) and width (W) were measured every other day, starting on day 1 after injection. The tumour volume was calculated using the following formula: L × W²/2. Twenty-eight days after injection, mice were euthanized, and the primary subcutaneous tumours were removed, paraffin-embedded, and sliced for staining. In this study, animals were not randomised. No blinding was performed in this study.

scATAC-seq analysis

scATAC-seq libraries were prepared utilising a SureCell ATAC-Seq Library Prep Kit (Bio-Rad, Pleasanton, CA, USA) and SureCell ddSEQ Index Kit (Bio-Rad) following the manufacturer’s instructions. The libraries were loaded at 1.5 pM on a NextSeq 550 (Illumina, Santa Clara, CA, USA) and sequencing was performed utilising the subsequent read protocol. Read 1: 118 cycles; i7 index read: 8 cycles; and Read 2: 40 cycles. The FASTQ files were processed utilising the ATAC-Seq Analysis Toolkit (Bio-Rad) to generate debarcoded and aligned read data.

ArchR [21] v1.0.2 was used for scATAC sequencing analysis. All analyses were conducted using the mm10 genome assembly, utilising the ArchR “addArchRGenome(‘mm10’)” function. Low-quality cells were excluded by applying a cut-off of 4 for transcription start site enrichment score and 1500 unique fragments. The ArchR “addDoubletScores()” function was utilised with “k = 10, knnMethod = ‘UMAP’, LSIMethod = 1” parameters to eliminate doublets. The ArchR “addIterativeLSI()” function, using a genome-wide 500-bp tile matrix was employed to calculate iterative LSI information. Cell clustering was performed using the ArchR “addClusters() function” along with the Seurat “FindClusters()” function utilising default parameters. To perform UMAP, the “addUMAP()” function was utilised. We visualised the gene activity scores in the UMAP overlay using the ArchR “plotEmbedding()” function. Gene activity scores were calculated based on chromatin accessibility within the gene body, promoter, and distal regulatory elements. These scores correlated with experimentally determined gene expression [21–24].

Isolation of mouse cancer-associated fibroblasts

The CAFs were isolated from the pancreata of orthotopic Tagln^+/+ and Tagln^−/− mouse models. Briefly, mouse pancreatic tissue was minced and dissociated using the tumour dissociation kit (130-096-730, MACS Miltenyi Biotec). Subsequently, the tumour-associated fibroblast isolation kit (130-116-474, MACS Miltenyi Biotec) protocol was employed, utilising magnetic beads with a depletion cocktail to remove non-tumour-associated fibroblasts including tumour cells, epidermal cells, and immune cells. After centrifuging the remaining cell suspensions at 700 x g for 10 min at 4°C, the pellets were resuspended in DMEM medium (Fujifilm Wako Pure Chemical) containing 10% Foetal Bovine Serum (FBS) (Nichirei Biosciences, Tokyo, Japan) and 1% antibiotic–antimycotic (Fujifilm Wako Pure Chemical) and plated in culture dishes. When most of the fibroblast cells adhered and changed shape, the culture medium was removed. The cells on the dishes were then gently washed twice with PBS to remove unattached cells; and then fresh DMEM medium was then added prior to cell culture. CAFs (2 × 10⁵ cells) were plated into 10 cm² culture dishes with fresh DMEM containing 10% FBS, and 1% penicillin–streptomycin and grown for 3 days. The supernatant was collected and centrifuged at 700 x g for 10 min at 4°C. The supernatant was stored as conditioned medium (CM). CM was mixed with fresh DMEM at a ratio of 2:1 and used to treat the cells.

Characterisation of stromal tissue by histological staining

Immunohistochemistry (IHC) was performed as described previously [25]. The antibodies used in the IHC analysis are listed in Supplementary Table S1. Masson’s trichrome staining was performed by using a Trichrome Stain Kit (Modified Masson’s) (TRM-IFU; SCYTEK Laboratories, Logan, UT, USA). Images were acquired using a KEYENCE BZ-X810 (KEYENCE, Osaka, Japan).

The number of immune cells (CD4, CD8 and CD206 cells) was quantified using ImageJ. The positive cell score was calculated as follows: (1 x the number of weak staining cells/μm²) + (2 x the number of moderate staining cells/μm²) + (3 x the number of strong staining cells/μm²) within the region of interest (ROI). Collagen-positive areas were quantified using the KEYENCE BZ-X800 Analyser software.

Transwell migration and invasion assay

To clarify the indirect (non-contact) interactions between different cell types, Transwell migration and invasion assays were conducted using cell culture Inserts with 8-μm pores (Falcon, Tewksbury, MA, USA) for migration and Matrigel-coated Transwell chambers (Corning) for invasion (Matrigel coated on the upper surface). In these assays, the two cell types were cultured in separate compartments- one in the upper insert and the other in the lower chamber- allowing only soluble factors to pass through the membrane without direct cell-cell contact. Cells that migrated or invaded and attached to the lower surface of the membrane were fixed with 4% paraformaldehyde and stained with 0.1% crystal violet. Stained cells were counted in randomly selected fields in each well using the ImageJ software.

Human tissue samples

Human PDAC surgical specimens were obtained at Nagoya University Hospital with the approval from the Ethics Committee of Nagoya University Graduate School of Medicine (approval number: 2017-0127). Informed consent was obtained from all subjects. This study was conducted in accordance with the Declaration of Helsinki. Background of samples is provided in Supplementary Table S2.

Statistical analysis

Data are presented as the mean ± SD or SEM. Statistical analysis and visualisation were performed with Microsoft Excel and GraphPad Prism 5 (GraphPad Software, San Diego, CA, USA). P-values < 0.05 were considered statistically significant. P values are indicated as follows. *P < 0.05; **P < 0.01; ***P < 0.001. Data were analysed using either an unpaired two-tailed Student’s t-test or one-way analysis of variance (ANOVA). All experiments were conducted in triplicate.

Results

scATAC-seq identified four distinct subpopulations of CAFs in KPC mice

To delineate the epigenetic landscape and heterogeneity of CAFs in PDAC, we began by performing scATAC-seq analysis on pancreatic tissues from two 3-month-old male KPC mice and one WT mouse (Supplementary Fig. S1a). This analysis yielded 2,880 high-quality chromatin accessibility profiles (Fig. 1a). We observed a high activity score, indicating open chromatin, for Onecut1 in C4 (normal ductal cells) and Car2 in C7 (tumour cells) (Fig. 1b). Furthermore, we clustered the cells and identified seven distinct cell types based on marker gene activity scores: ductal cells or tumour cells (Krt18, Sox9 and Tspan8), T cells (Cd3e), B cells (Cd19), macrophages (C1qc) and fibroblasts (Col1a1, Pdpn) [5, 26, 27] (Fig. 1a, b, See Methods).

Fig. 1. Single-cell ATAC-seq of mouse PDAC reveals heterogeneity of myCAF.

a UMAP plot of primary scATAC-seq cells (2880 cells) from two KPC and one wild-type mouse. Different colours represent different cell types. The red circle indicates fibroblast clusters. b UMAP plot showing cells coloured by the gene score of marker genes for each cell type. Blue indicates low gene activity scores, while purple represents high gene activity scores for the given gene. c UMAP plot of 533 fibroblast-derived cells from Fig. 2a. Different colours represent four clusters (F1-F4). The pink circle highlights myCAFs. d UMAP plot showing cells coloured by the gene score of marker genes for the different fibroblast subclusters. ChromVAR TF motif bias-corrected deviations overlaid on the UMAP projection of scATAC-seq-based myCAF, as showed in Fig. 2c. e iCAF regulator (Stat3 motif). f myCAF regulator motifs.

Fibroblasts (C2 and C5) were further subclustered, resulting in the identification of four distinct subclusters (hereafter F1–F4; Fig. 1c, Supplementary Fig. S1b). High activity scores of the myCAF marker genes (Acta2 and/or Tagln) were observed in F2, and F3, and F4, while high activity scores of the iCAF marker genes (Il6 and Il1r) were observed in F1 (Fig. 1d, Supplementary Fig. S1c). However, CAFs displaying typical apCAF markers did not form a distinct subcluster, suggesting that apCAFs are epigenetically similar to other CAF subtypes, such as myCAF and iCAF (Supplementary Fig. S1c). Notably, open chromatin structures at the Tagln locus were predominantly observed in the F2 and F3 subpopulations, whereas those at the Acta2 locus were detected across F2, F3, and F4, with marked enrichment in F4. Based on the epigenetic profiles obtained from scATAC-seq, the myCAF population could be classified into the following clusters: F2 and F3, Acta2-open/Tagln-open myCAF; F4, Acta2-open/Tagln-close myCAF (Fig. 1d).

Next, enrichment of transcription factor (TF) binding motifs was analysed in the open chromatin regions of these CAF clusters. The Stat3 motif was enriched in the iCAF cluster (F1, Fig. 1e), consistent with previous reports identifying STAT3 as a well-established regulator of iCAF differentiation [3]. Regarding myCAF-associated TF motifs, Smad4 binding motifs were enriched in the Acta2-open/Tagln-open myCAF cluster F3 (Fig. 1f). Prrx1, a TF critical for CAF activation [28], showed motif enrichment in a distinct Acta2-open/Tagln-open myCAF cluster, F2 (Fig. 1f). SRF (serum response factor), an important TF for fibroblast activation and pro-desmoplastic transcription [29], showed motif enrichment in the Acta2-open/Tagln-close myCAF cluster (F4) (Fig. 1f). The distinct distribution of TF motifs across the open chromatin landscapes of CAF subtypes underscores the complexity and context-dependent nature of the regulatory networks that may govern the functional diversity of CAFs in cancer progression.

Analysis of transcriptional programs in CAFs from KPC mouse models

Next, we examined the transcriptional status in the pancreatic tissues obtained from two 3-month-old male KPC mice and two WT mice using scRNA-seq analysis (Supplementary Fig. S1a). After quality filtering, we obtained 17,843 high-quality transcriptome profiles, which were then subjected to unsupervised, graph-based clustering, revealing 11 distinct cell clusters (Fig. 2a, Supplementary Materials and Methods). Based on canonical marker gene expression profiles [5, 26, 27], these clusters were annotated as ductal, tumour, acinar, islet, endothelial, fibroblast, monocyte, granulocyte, macrophage, T/ Natural killer (NK), and B cells.

Fig. 2. Single-cell RNA-seq profiles of the mouse PDAC.

a Uniform Manifold Approximation and Projection (UMAP) plot of 17,843 cells from two KPC and two wild-type mice. Different colours represent different cell types. The purple circle highlights fibroblast clusters. b UMAP plots showing the expression of pan-fibroblast marker genes (Col1a1 and Pdpn) across all clusters. c UMAP plot of fibroblasts (derived from Fig. 2A), colour-coded by different fibroblast subclusters. d Violin plots displaying the normalised expression of marker genes across different fibroblast subclusters. e UMAP plots of myCAFs (derived from Fig. 2C) colour-coded by distinct myCAF subclusters. f Violin plots depicting the normalised expression level of transcription factors, cytokines and their corresponding target genes across myCAF subclusters. g Scatter plot showing the expression levels of ACTA2 and TAGLN in myCAFs identified from scRAN-seq data of human PDAC samples (GSE155698). The Pearson correlation coefficient (r) and corresponding P-value are indicated in the graph.

Fibroblast clusters were identified based on the expression of the marker genes Col1a1, Col5a2, and Pdpn (Fig. 2b–d). The majority of cells within Cluster 1 (C1) originated from WT mice, accounting for over 70% of the population. Therefore, C1 was designated as the normal fibroblast cluster. The remaining fibroblast clusters were further classified into three subpopulations according to marker genes [4, 5]: iCAF (C2), myCAF (C4), and apCAF (C3) (Fig. 2c, d). Furthermore, myCAF cells were classified into three distinct subclusters: C4_1, C4_2, and C4_3 (Fig. 2e). Interestingly, each subcluster was enriched for different TFs, related pathway genes and cytokines (Fig. 2f). For instance, Tgfbr2 and Smad2 were predominantly expressed in the C4_2 subcluster, which corresponds to the Acta2-open/Tagln-open cluster (F3) in scATAC-seq. Similarly, Tgfb3, a gene upregulated by PRRX1[30], was highly expressed in the C4_3 subcluster, which corresponds to the Acta2-open/Tagln-open (F2). In line with the widespread enrichment of the Srf motif across all myCAF subclusters in the scATAC-seq analysis (Fig. 1f), downstream targets of SRF, such as Myh9 and Itga1 [31], were upregulated in all subclusters, with the highest expression observed in the C4_1 cluster (corresponding to the Acta2-open/Tagln-close, F4 cluster) compared to the other clusters. These patterns highlight a strong concordance between chromatin accessibility and the transcriptional programs defining each myCAF subcluster (Fig. 1f). Consistent with this relationship, Acta2 and Tagln also exhibit distinct expression patterns across clusters, with particularly high expression in C4_3 as shown by FeaturePlot analysis (Supplementary Fig. S1d).

To validate the findings in KPC mice, we analysed a publicly available scRNA-seq data from human PDAC samples (GSE155698). myCAFs were identified based on established marker genes (Supplementary Fig. S2a–S2d). The myCAF cluster was further subdivided into three subclusters, each enriched with distinct TFs and pathway-associated genes, consistent with findings in KPC mice (Fig. 2e, and f). For instance, TGFB1 and the SMAD3 target gene SERPINE1 were enriched in myCAF_2 (corresponds to C4_2 in KPC mice)[32]. ZEB1, a TF regulated by SMAD4 [33] and PRRX1 [34], was highly expressed in myCAF_2 and myCAF_3 (C4_2 and C4_3 in mice). CEBPB, known to regulate ACTA2[35], was upregulated across all subclusters, with particularly high expression in myCAF_1. TNC, a target of both CEBPB [36] and PRRX1 [37], was found in both myCAF_1 (C4_1 in mice) and myCAF_3. In addition, we found that the majority of myCAFs co-expressed both ACTA2 and TAGLN (r = 0.75, P < 2.2e-16), whereas a small subset expressed only one of the two (Fig. 2g and Supplementary Fig. S2G).

Tagln in CAFs drives migration, invasion, and proliferation of human PDAC cells

Previous studies analysing human and mouse PDAC samples have shown that TAGLN and ACTA2 are consistently co-expressed and upregulated in myCAF (Supplementary Fig. S3a) [4, 5]. Although TAGLN and ACTA2 were generally co-expressed in myCAFs, our single-cell analysis (Fig. 2g) revealed that a small subset of cells expressed only one of the two, suggesting functional heterogeneity within the myCAF population. This raises the question of whether TAGLN-positive CAFs may have distinct functional roles in the tumour microenvironment. TAGLN is an actin-binding protein involved in actin cytoskeleton remodelling [38]. To investigate the specific role of Tagln in fibroblasts during PDAC tumour progression, we orthotopically implanted the murine PDAC cell line mT5 was into the pancreas of Tagln^+/+ (wild type) and Tagln^−/− mice [14](Supplementary Fig. S3b). Two weeks after transplantation, CAFs were isolated from the pancreatic tumours for further analysis (Supplementary Fig. S3c, S3d; Materials and Methods).

A transwell assay assessing cell migration revealed no significant difference in migratory ability between Tagln^+/+ CAFs and Tagln^−/− CAFs under monoculture conditions (Fig. 3a). However, when co-cultured with PDAC cells (Panc-1 or BxPC-3), Tagln^+/+ CAFs exhibited significantly enhanced migration compared to Tagln^−/− CAFs (Fig. 3a, Supplementary Fig. S3e). In addition, PDAC cells co-cultured with Tagln^+/+ CAFs exhibited significantly greater migratory and invasive capabilities than those co-cultured with Tagln^−/− CAFs (Fig. 3b, Supplementary Fig. S3f). Notably, PDAC cells cultured in CM derived from Tagln^+/+ CAFs demonstrated significantly increased proliferation over a two-week period compared to those cultured in medium from Tagln^−/− CAFs (P = 0.04, Supplementary Fig. S3G).

Fig. 3. Tagln-positive fibroblasts promote cell migration and invasion of human PDAC cells.

a Transwell chamber migration assay of CAF cells. Left panel: A schematic illustration of the experimental setup, Middle panel: Representative images. Right panel: Quantification of migrated cells. CAF migration was assessed in co-culture with or without Panc-1 cells (n = 3, 7 images per group). The y-axis indicates the number of migrated cells. Scale bar: 50 µm. ***P < 0.001. b Panc-1 migration and invasion assay co-cultured with Tagln^+/+ or Tagln^-/- CAFs. Left panel: A schematic illustration of the experimental setup, Middle panel: Representative images. Right panel: Quantification of migrated cells or invaded cells (n = 3,7 images per group). The y-axis indicates the number of migrated and invaded cells. Scale bar: 50 µm. ***P   < 0.001. c Immunofluorescence staining for F-Actin (phalloidin) in Tagln^+/+ CAFs and Tagln^-/- CAF. Scale bar: 50 µm (left panel) and 10 µm (right panel).

Consistent with these functional differences, phalloidin staining of F-actin revealed distinct morphological differences between Tagln^−/− CAFs and Tagln^+/+ CAFs. In particular, the actin cytoskeleton in Tagln^−/− CAFs exhibited a disordered architecture, in contrast to the well-organised, directionally aligned fibres characteristic of Tagln^+/+ CAFs (Fig. 3c).

Under co-culture conditions with Panc-1 cells, Tagln+/+ and Tagln−/− CAFs exhibited distinct transcriptional responses. As shown in Supplementary Fig. S4, the baseline expression levels of Acta2 and Col1a1 were higher in Tagln+/+ CAFs than in Tagln−/− CAFs, suggesting that Tagln+/+ CAFs may have greater migratory potential even prior to co-culture. This observation may be related to the well-established role of TGFβ secreted by pancreatic cancer cells [39], which promotes the migratory capacity of CAFs through pathways involving Tagln, Acta2, and Col1a1, thereby enhancing their migratory and invasive phenotypes. Tgfbr1, an essential component of TGFβ signalling, was upregulated upon co-culture with Panc-1 cells only in Tagln+/+ CAFs.

Impact of Tagln-positive CAFs on PDAC progression in vivo

To evaluate the in vivo impact of Tagln-positive fibroblasts on PDAC progression, mT5 cells were subcutaneously co-transplanted with either Tagln^+/+ CAFs or Tagln^−/− CAFs at a 1:3 ratio into Tagln^−/− recipient mice (Fig. 4a). Tumours co-injected with Tagln^+/+ CAFs displayed significantly increased volumes compared to those co-injected with Tagln^−/− CAFs (P = 0.0297, Fig. 4b, 4c). Intriguingly, collagen deposition and the infiltration of CD206⁺ M2 macrophages were both significantly elevated in tumours containing Tagln^+/+ CAFs, relative to those with Tagln^−/− CAFs (P = 0.0065, and P < 0.0001, respectively; Fig. 4d).

Fig. 4. Tagln-positive CAFs promote tumour growth in a xenograft mouse model.

a Schematic representation of the subcutaneous tumour implantation mouse model. b Tumour growth in xenograft mouse model transplanted with mT5 cells and either Tagln^+/+ CAFs or Tagln^-/- CAFs (n = 4). Results are presented as the mean ± SEM. *P < 0.05. c Tumour sections from Tagln^+/+ or Tagln^-/- orthotopic mouse models. Representative images of HE staining (scale bar: 500 µm) and IHC for Tagln. The magnified images (scale bar, 10 µm) in the upper left corner of the IHC are derived from the corresponding green boxes. d IHC images for αSMA, CD206 and Masson’s trichrome staining (collagen) (scale bar: 50 µm). The magnified images (scale bar, 10 µm) in the upper left corner of the IHC and Masson stains are derived from the corresponding green boxes. Areas of collagen, CD206-positive cells were quantified in the right panel (n = 4). Results are shown as the mean ± SEM and compared by an unpaired t-test. ***P < 0.001, **P < 0.01.

We next assessed the impact of Tagln-positive fibroblasts on PDAC progression using an orthotopic mouse model, in which mT5 cells were implanted into both Tagln^−/− and Tagln^+/+ mice (Fig. 5a). Three weeks post-implantation, tumour weights were significantly higher in Tagln^+/+ mice than in Tagln^−/− mice (P = 0.0022; Fig. 5b, c). Consistently, PDAC tissues from Tagln^+/+ mice exhibited significantly greater collagen deposition compared to those from Tagln^−/− mice (P < 0.0001; Fig. 5d). Notably, Tagln^−/− mice showed markedly increased infiltration of CD4⁺ and CD8⁺ T cells, accompanied by a reduction in CD206⁺ M2 macrophages within the tumour stroma (Fig. 5d). Consistently, the expression of M2 macrophage-associated cytokines, Ccl11 and Cxcl12, was significantly suppressed in Tagln^−/− CAFs (P = 0.0029 and 0.0041, respectively; Supplementary Fig. S5a) [40, 41]. These findings indicate that Tagln-positive CAFs contribute substantially to the establishment of an immunosuppressive tumour microenvironment. Although Tagln is also expressed in endothelial cells[42], tumour vasculature, as assessed by CD31 immunostaining, did not differ significantly between Tagln^+/+ and Tagln^−/− mice (P = 0.335; Supplementary Fig. S5b).

Fig. 5. Loss of Tagln reduces tumour growth and alters the tumour microenvironment in an orthotopic mouse model.

a Schematic representation of the orthotopic tumour implantation mouse model. b Establishment of an orthotopic PDAC mouse model using mT5 cells in Tagln^+/+ and Tagln^-/- mice. Mice were sacrificed three weeks after tumour cell inoculation, and tumour weight was measured (n = 8). Results are presented as mean ± SEM. **P < 0.01. c Tumour sections from the Tagln^+/+ or Tagln^-/- orthotopic mouse models. Representative images of H&E staining (scale bar: 500 μm) and IHC for Tagln (scale bar: 50 μm). The magnified images (scale bar, 10 µm) in the upper left corner of the IHC are derived from the corresponding green boxes. d Left panel: Representative images of Masson’s trichrome staining (collagen) and IHC for αSMA, CD4, CD8, and CD206. Scale bar: 50 µm. The magnified images (scale bar, 10 µm) in the upper left corners correspond to the green boxed area in the main images. Right panel: Quantifications of collagen area and CD4-, CD8-, and CD206-positive cells (n = 4). Results are shown as mean ± SEM and analysed using unpaired t-test. ***P  < 0.001, **P < 0.01.

Diversity of myCAFs in human PDAC tissues

To further investigate the role of TAGLN in human PDAC, we performed immunofluorescence analysis for TAGLN and ACTA2. Intriguingly, the analysis of human PDAC tissue (Case 4, Supplementary Table S2) and KPC PDAC tissue revealed variable staining patterns for these markers: double-positive for αSMA and TAGLN, αSMA-positive only, and TAGLN-positive only (Fig. 6a; Supplementary Fig. S6a). The heterogeneous expression of Acta2 and Tagln in fibroblasts from 3-month-old KPC mice was further confirmed by single-cell digital droplet PCR (ddPCR). Among fibroblasts sorted by Thy1 expression, approximately 70% co-expressed both Acta2 and Tagln, whereas a subset expressed only one of the markers (Supplementary Fig. S6b).

Fig. 6. High TAGLN expression in PDAC stroma is associated with poor survival.

a Representative images of H&E staining and IF for αSMA and TAGLN in a human PDAC section. The pink arrowhead indicates αSMA-positive/TAGLN-positive CAF, the gray arrowhead denotes αSMA-positive/TAGLN-negative CAF, and the yellow arrowhead represents αSMA-negative/TAGLN-positive CAF. T: Tumour. b The expression of ACTA2 and TAGLN in PDAC and normal tissues (n = 179 and 171, respectively). Box plots derived from GEPIA gene expression data. c Scatter plot showing the correlation between TAGLN and ACTA2 expression in primary PDAC samples (n = 185), based on TCGA data analysed using UCSC Xena (http://xena.ucsc.edu/). The Pearson correlation coefficient (r) and corresponding P-value are indicated in the graph. d Kaplan-Meier analysis showing the overall survival of PDAC patients, stratified by high and low ACTA2 and TAGLN expression using the optimal cutoff calculated via http://www.kmplot.com for the TCGA datasets. The hazard ratio (HR) and the number of patients is shown in the panels.

Further immunohistochemical staining of 19 additional human PDAC samples further demonstrated that both TAGLN and αSMA were abundantly expressed in a substantial fraction of CAFs within the tumour stroma. Immunofluorescence analysis of these samples showed that the area double-positive for αSMA and TAGLN was increased in advanced-stage (stage III) compared to early-stage (stage I and II) PDAC (P = 0.0156, Supplementary Fig. S7a).

Finally, we analysed RNA-seq data from The Cancer Genome Atlas (TCGA) to assess the clinical relevance of TAGLN expression in human PDAC. ACTA2 and TAGLN expression were significantly higher in PDAC tissues compared to normal tissues (P < 0.05, Fig. 6b). The expression of ATCA2 and TAGLN is highly correlated (r = 0.9554, Fig. 6c). Moreover, high TAGLN expression was significantly associated with poorer overall survival (P = 0.0442, Fig. 6d).

Collectively, these findings indicate that myCAFs in human PDAC exhibit notable heterogeneity, and that TAGLN-positive CAFs may be functionally linked to disease progression and unfavourable clinical outcomes.

Discussion

In this study, we employed scATAC-seq and scRNA-seq to analyse pancreatic tissue from KPC mice and PDAC patients, dissecting the cellular and epigenomic landscape of CAFs. We identified TAGLN as a hallmark gene of the myCAF population, which plays a pivotal role in modulating the tumour microenvironment and promoting tumour progression. Given that the PDAC in 3-month-old KPC mice represents an evolving tumour stage, we expected myCAFs would be heterogeneous. Indeed, our scRNA-seq analysis identified three distinct CAF subtypes (myCAF, iCAF and apCAF) in line with previous reports [3–5]. In contrast, scATAC-seq identified two distinct epigenetic states within CAF populations, corresponding to the iCAF and myCAF clusters. The lack of a distinct apCAF cluster in scATAC-seq is likely due to the overlapping TF motifs between apCAFs and both iCAFs and myCAFs. apCAFs were initially identified in mouse PDAC models as mesothelial cell-derived CAFs [43]. In human PDAC, scRNA-seq data indicate that apCAF cells are interspersed among iCAF and myCAF clusters [6]. This distribution is consistent with the fact that apCAF differentiation is driven by both IL-1, which promotes the iCAF lineage, and TGF-β, which induces the myCAF lineage. Consequently, the TF motifs of apCAFs may be shared with those of both iCAFs and myCAFs [43]. This suggests that apCAFs may not constitute a transcriptionally autonomous lineage, but rather occupy a hybrid epigenomic state that defies clear separation in chromatin accessibility-based analyses.

In the scRNA-seq analysis of KPC tumours, the myCAF cluster was smaller compared with that of human PDAC samples. This difference may be attributed to the timing of tumour collection in the mouse model (3-month-old, representing early-stage tumours). In the mouse model, the proportion of myCAFs relative to iCAFs is approximately 21% in early-stage tumours and increases to around 69% during tumour progression [7]. In contrast, stromal regions are already more substantial in human PDAC samples at early stages [44]. These differences in tumour stage and stromal architecture likely account for the discrepancy in myCAF abundance between the mouse and human datasets.

Notably, scATAC-seq further resolved three distinct subpopulations within the myCAF population: Acta2-open/Tagln-open (F2 and F3) and Acta2-open/Tagln-closed (F4). In combination with TF motif enrichment analysis [45] this revealed unique programs regulated by cell type–specific TFs in each subpopulation. Prrx1 motifs were enriched in F2 myCAFs, consistent with the role of PRRX1 as a master regulator of myofibroblastic CAF identity via super-enhancer remodelling [46]. F3 was enriched for Smad2/3/4 motifs, indicative of TGF-β/SMAD2/3-driven terminal differentiation—a hallmark of late-stage myCAFs [4, 5, 47]. In contrast, F4 cells exhibited strong enrichment for Srf and Cebpb motifs, which have been associated with ACTA2 expression and collagen production [48]. These findings suggest that the myCAF compartment in PDAC comprises at least three transcriptionally and functionally distinct subtypes, regulated by discrete transcriptional circuits. Moreover, scRNA-seq analysis of human PDAC samples suggested the presence of myCAF subclusters analogous to those identified in mice. Although the plasticity among these subtypes remains to be elucidated, their coexistence in early-stage tumours highlights the dynamic nature of CAF differentiation.

CAFs are well recognised for their role in modulating immune cell activity[49]. Consistent with this, we found that orthotopic tumours derived from Tagln-deficient mice exhibited significant alterations in the immune landscape, including increased infiltration of CD4⁺ and CD8⁺ T cells and a notable reduction in M2-polarised macrophages. While the precise mediators of these immunological changes remain to be identified, our findings are congruent with prior reports demonstrating that CAF-derived ECM acts as a physical barrier to immune infiltration [50]. In line with this, collagen-related genes (Col1a1, Col1a2, Col12a1) were downregulated in TAGLN-deficient CAFs as previously observed in human mesenchymal stromal cells [11].

Beyond ECM remodelling, TAGLN has been implicated in cytoskeletal dynamics and endocytic activity. In PDAC, αSMA-positive myCAFs engage in macropinocytosis to support metabolic needs in nutrient-deprived environments [51]. Given the function of TAGLN in F-actin–dependent processes, including clathrin-mediated endocytosis [52], it is plausible that TAGLN-positive myCAFs contribute to tumour progression not only through ECM deposition but also through enhanced nutrient scavenging. However, this hypothesis remains to be directly tested in the context of CAFs.

Furthermore, fibroblast-expressed TAGLN may contribute to immune modulation through cytokine and chemokine signalling. Conditioned media from CAFs can induce monocyte differentiation into CD206⁺ M2 macrophages [53], and TAGLN influences cytokine secretion [15] and macrophage recruitment [25]. In our analysis, we identified several cytokines, including Ccl11 and Cxcl12, which are associated with M2 macrophage recruitment, as being highly expressed in Tagln-positive CAFs compared to Tagln-negative CAFs. Our data suggest that TAGLN expression in CAFs contributes to the modulation of both the structural and immunological aspects of the PDAC stroma, thereby contributing to enhanced tumour growth and aggressiveness.

In conclusion, our study highlights the epigenomic and functional heterogeneity of myCAFs in PDAC and identifies TAGLN as a key regulator of fibroblast-mediated tumour promotion. These findings emphasise the importance of dissecting the lineage-specific regulatory networks that govern CAF behaviour. Given the robust expression of TAGLN in human PDAC and its impact on both stromal composition and immune infiltration, targeting stromal TAGLN may offer a novel strategy for therapeutic intervention in pancreatic cancer.

Author contributions

WX performed all cell and animal experiments and drafted the manuscript. KS designed the experiments, contributed to the mouse experiments, and wrote the manuscript. KK and RM performed scATAC-seq and analysed the data. AK and YS(uzuki) performed scRNA-seq and analysed the data. YS(hiraki) analysed human scRNA-seq data. SM and AE performed IHC staining of human PDAC samples. TN, YM, and MS contributed to manuscript writing. TK and MF performed IHC staining of mouse PDAC samples. YK supervised the study and edited the manuscript.

Funding

This study was supported in part by the Japan Society for the Promotion of Science KAKENHI, grant number 23K06633 (K. Shinjo), a research program of P-CREATE and SICORP from the Japan Agency for Medical Research and Development, and Takeda Science Foundation (Y. Kondo).

Data availability

Datasets related to this article have been submitted to an open-source data repository, Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/, accession numbers GSE271510 and GSE271509).

Competing interests

The authors declare no competing interests.

Ethics

The research protocol was approved by the Institutional Review Board of Nagoya University of Graduate School of Medicine (2017-0127). An animal study was performed under protocols approved by the Institutional Animal Care and Use Committee of Nagoya University of Graduate School of Medicine (M240289-002).

Supplementary information

The online version contains supplementary material available at 10.1038/s41416-025-03299-5.