Targeting cancer matrix mechanics: fibronectin fiber relaxation correlates with ADAM8 expression in PDAC and is amenable to in vitro modulation

Pancreatic ductal adenocarcinoma (PDAC) is among the deadliest cancers, characterized by a dense desmoplastic stroma rich in extracellular matrix (ECM) that impairs therapy and promotes tumor progression. Fibronectin (FN), a major ECM component, provides binding sites for growth factors and integrins, and changes in its mechanical signature have been associated with the progression of several pathologies, including the invasiveness of breast cancer. Using the mechano-sensitive peptide probe FnBPA5, which binds relaxed FN fibers, we mapped FN fiber tension in cryosections from PDAC patient biopsies by comparing tumor regions to matched non-cancerous tissue. Even though counterintuitive, tumor stroma was strongly enriched in untensed FN fibers, whereas non-cancerous tissue showed minimal FnBPA5 binding. Untensed FN fibers localized near alpha smooth muscle actin (αSMA)-positive myofibroblastic cancer-associated fibroblasts (myCAFs) and collagen triple helix repeat containing 1 (CTHRC1)-positive stromal cells and correlated strongly with a disintegrin and metalloproteinase domain-containing protein 8 (ADAM8), a FN-cleaving protease. In 2D culture, healthy pancreatic stellate cells (PSCs) adopted a myCAF-like phenotype when exposed to PANC-1 cancer cells, as shown by αSMA and CTHRC1 upregulation, and assembled an ECM enriched in relaxed FN fibers. Supplementation with standard PDAC chemotherapeutics (paclitaxel or gemcitabine) and addition of ADAM8 inhibitors to PSCs partially reversed loss of FN fiber tension in vitro. Together, our findings establish untensed FN fibers as a hallmark of PDAC stroma and present a minimal in vitro model to probe and modulate ECM fiber mechanics for disease-specific drug screening.

Pancreatic ductal adenocarcinoma (PDAC) accounts for over 90% of pancreatic cancers and represents a major global health burden, with approximately half a million new cases in 2022 (Stoffel et al., 2023; Halbrook et al., 2023; Sung et al., 2021; Bray et al., 2024). Despite its lower relative incidence, PDAC is the 6 th leading cause of cancer deaths worldwide due to its extreme lethality: fewer than 25% of patients survive 1 year and under 10% reach 5 years post-diagnosis (Li X. et al., 2024; Ushio et al., 2021; Pizot et al., 2019). Incidence is highest in industrial nations (Bray et al., 2024). In the United States, PDAC is the 3rd leading cause of cancer mortality and is projected to become the 2nd by 2040 (Rahib et al., 2021). This trend is driven not only by rising incidence but also by improved clinical management of other previously lethal cancers, such as breast and colorectal cancer, which topped mortality statistics in the past (Ushio et al., 2021; Ferlay et al., 2016), underscoring the urgent need for finding more effective treatment options also for PDAC.

PDAC often develops after chronic pancreatitis, an inflammatory condition that disrupts acinar cell homeostasis (the exocrine cells of the pancreas responsible for digestive enzyme secretion) leading to cell death, inflammation, and fibrosis (Tao et al., 2021). Risk factors for pancreatitis include untreated diabetes, obesity, higher age, smoking, and frequent alcohol consumption (Pang et al., 2019). Such persistent stress promotes acinar-to-ductal metaplasia and genetic dysregulation (Tao et al., 2021; Hessmann et al., 2020), allowing pancreatic cancer cells (PCCs) to emerge through Kirsten rat sarcoma viral oncogene homolog (KRAS) mutation found in approximately 90% of cases (Halbrook et al., 2023; Alvarez Fallas et al., 2021; Than et al., 2024). KRAS also enhances secretion of profibrotic cytokines such as transforming growth factor beta-1 (TGF-β1) into the tumor microenvironment (Halbrook et al., 2023; Bräutigam et al., 2025), and chronic exposure to TGF-β1 and interleukin-6 (IL-6) can further accelerate mutagenesis (Stoffel et al., 2023; Tao et al., 2021; Choi et al., 2020).

PDAC progression involves close interaction between PCCs and stromal cells, disrupting extracellular matrix (ECM) homeostasis and creating a tumor microenvironment (TME) supportive for disease progression (Hessmann et al., 2020; Winkler et al., 2020; Cox, 2021). The TME in PDAC is marked by a dense desmoplasia, stroma rich in collagens and fibronectin (FN), forming a physical barrier and creating an immunosuppressive environment that drastically reduces the efficacy of common chemo- and immunotherapies (Hessmann et al., 2020; Ho et al., 2020; Zhang et al., 2022; Zhu et al., 2025; Gross et al., 2025; Wartenberg et al., 2018). Cancer-associated fibroblasts (CAFs), such as myofibroblastic CAFs (myCAFs), are the main drivers of this aberrant ECM deposition (Halbrook et al., 2023; Zhang et al., 2022) and originate from the main pancreatic fibroblasts (so-called pancreatic stellate cells, PSCs) found in close proximity to cancer cells (Hessmann et al., 2020; Richards et al., 2017). PSC activation into myCAFs is characterized by (i) loss of lipid droplets which would contain Vitamin A to regulate glucose metabolism (Brun et al., 2015), (ii) assembly of contractile alpha smooth muscle actin (αSMA) stress fibers and (iii) collagen I deposition (Halbrook et al., 2023; Hessmann et al., 2020), as well as (iv) expression of collagen triple helix repeat containing 1 (CTHRC1) protein (Li E. et al., 2024; Liu et al., 2023; Kang et al., 2023). Beyond ECM remodeling, myCAFs support PCC survival and migration, fostering treatment resistance and creating an immunosuppressive niche (Winkler et al., 2020; Zhang et al., 2022; Gross et al., 2025; Wartenberg et al., 2018; Richards et al., 2017). Attempts to disrupt the desmoplastic barrier as a treatment strategy unexpectedly worsened outcomes by increasing metastasis (Halbrook et al., 2023; Li E. et al., 2024), suggesting that tumor encapsulation also restrains dissemination and highlighting the need to better understand the interactions between PCCs, myCAFs and the TME to develop better therapies (Ho et al., 2020; Zhang et al., 2022).

The highly desmoplastic TME of PDAC, which can comprise up to 90% of the tumor volume (Efthymiou et al., 2025; Neesse et al., 2011), is enriched in deposited ECM that can account up to 60% of the tumor mass (Henke et al., 2019) thereby profoundly altering tissue mechanics, increasing stiffness and viscoelasticity (Zhu et al., 2025; Cambria et al., 2024; Safaei et al., 2023; Piersma et al., 2020; Chaudhuri et al., 2020; Barrett et al., 2025). These changes create a reciprocal feedback loop between the ECM and myCAFs, where altered mechanics further activate fibroblasts through mechanosensitive pathways (Panciera et al., 2020), ultimately promoting malignancy in solid tumors (Piersma et al., 2020). Importantly, the pancreas is an inherently soft tissue with an elastic modulus of around 1 kPa, whereas PDAC tissue is about ten times stiffer, reaching up to 10 kPa (Sa et al., 2020; Rice et al., 2017). Increased tumor stiffness has been directly linked to chemoresistance, such as reduced paclitaxel efficacy in 3D pancreatic cancer models (Rice et al., 2017).

Given their profound influence on tumor progression, it is not surprising that biophysical traits are increasingly exploited for diagnostics (Sigrist et al., 2017) and for therapy (Subrahmanyam and Ghandehari, 2021; Lee et al., 2017; Frazier et al., 2016). While early research focused mainly on global tissue mechanics such as increased stiffness (Cambria et al., 2024; Kaushik et al., 2016; Liu et al., 2024), far less is known regarding the tension of individual ECM fibers within the TME. As tissue stiffness has multifactorial origins, including enhanced ECM deposition and crosslinking (Barrett et al., 2025), tissue tension does not necessarily correlate with the tension of individual ECM fibers. Mapping FN fiber tension and how it is altered with disease progression is particularly important as FN fibers provide numerous binding sites for growth factors, cytokines, and other matrix molecules (Vogel, 2018; Hynes, 2009; Vega and Schwarzbauer, 2016). Mechanical forces can further expand this binding repertoire (Kubow et al., 2015; Ortiz Franyuti et al., 2018; Cao et al., 2017; Hubbard et al., 2016), as stretching of FN fibers can expose cryptic epitopes or disrupt multivalent binding sites (Vogel, 2018; Ortiz Franyuti et al., 2018; Cao et al., 2017; Selcuk et al., 2024; Krammer et al., 1999).

The development of the mechano-sensitive FnBPA5 peptide probe has enabled the visualization of FN fiber strain at the molecular and tissue levels, both in vitro and in vivo (Arnoldini et al., 2017; Crestani et al., 2025; Miéville et al., 2025a; Miéville et al., 2025b), making it a key tool for mapping FN fiber tension in health and disease. FnBPA5 binds the FnI 2-5 modules of relaxed FN fibers with nanomolar affinity (Arnoldini et al., 2017; Chabria et al., 2010; Hertig et al., 2012), but fiber stretching disrupts this multivalent epitope, reducing binding (Chabria et al., 2010; Hertig et al., 2012). Steered molecular dynamics simulations showed that tensile forces can extend the FnI 1-5 fragment via linker residue stretching (Hertig et al., 2012), and FN fiber stretch assays have confirmed stronger binding to relaxed fibers with gradually decreasing affinity upon fiber stretching (Arnoldini et al., 2017).

Using FnBPA5, untensed FN fiber mapping progressed from mouse xenograft models (Arnoldini et al., 2017; Fonta et al., 2020) and orthotopic models (Fonta et al., 2023) to human breast cancer (Miéville et al., 2025a) and glioblastoma (Crestani et al., 2025) tissues. Elevated FnBPA5 binding is also documented across various infections (Fonta et al., 2020; Rappold et al., 2025), as well as in vascular (Miéville et al., 2025b; Fonta et al., 2020), and benign (Trachsel et al., 2024) proliferative pathologies. In cancer, these untensed FN fibers frequently co-localize with myCAFs and dense, aligned collagen bundles (Arnoldini et al., 2017; Crestani et al., 2025; Miéville et al., 2025a; Fonta et al., 2020), which can shield FN fibers from cellular traction forces (Kubow et al., 2015; Miéville et al., 2025a; Fonta et al., 2020; Fonta et al., 2023). Alternatively, proteolytic cleavage by matrix metalloproteinases (MMPs) near FnBPA5 signals provides a potential mechanism for FN fiber relaxation (Trachsel et al., 2024; Rappold et al., 2024; Sridhar et al., 2025).

Besides MMPs, ADAM8 is also known to cleave FN in the extracellular space (Zack et al., 2009). Disintegrins are soluble proteins from snake venom that bind integrins via RGD motifs (Marcinkiewicz, 2005), often with high affinity, thereby competitively inhibiting binding of ECM proteins like FN and preventing processes such as platelet aggregation during blood coagulation (Kuo et al., 2019). While venom-derived disintegrins antagonize integrin–ECM interactions, the disintegrin domain of ADAM8 mediates β 1 -integrin binding as part of its targeting mechanism instead (Schlomann et al., 2015). ADAM8 multimerization is required for β 1 -integrin binding and is also essential for its proteolytic activity (Schlomann et al., 2015). By interacting with β 1 -integrins, ADAM8 may localize to sites of cell–matrix adhesion, positioning its metalloproteinase domain in proximity to ECM proteins like FN for cleavage (Zack et al., 2009). In malignancy, ADAM8-mediated ECM remodeling has been implicated in cancer cell detachment and directed migration during metastasis (Mierke, 2023). Consistently, inhibition of ADAM8 multimerization using cyclic peptide inhibitors significantly reduces tumor growth and metastatic progression in PDAC models (Schlomann et al., 2015). Clinically, high ADAM8 expression levels are associated with poor clinical outcomes due to enhanced metastatic spread (Valkovskaya et al., 2007).

In this study, we mapped the tensional state of FN fibers in cryosections from PDAC patient biopsies by comparing classified tumor regions with matched non-cancerous regions. Using the mechanosensitive probe FnBPA5, we found that the highly fibrotic PDAC TME (Myo Min et al., 2023) is significantly enriched in relaxed FN fibers. ADAM8 was strongly upregulated in PDAC tissue and showed high spatial correlation with FnBPA5 signal, whereas FnBPA5 correlated far less with collagen fiber bundles. FnBPA5 signal also consistently localized near αSMA-positive stromal cells and, to a lesser extent, CTHRC1-positive cells. Epithelial cell masses (positive for E-Cadherin) were closely surrounded by FnBPA5-positive stroma, creating a striking compartmentalization. CD31-positive endothelial vessels were more frequent in tumors than in non-cancerous tissues, interspersing FnBPA5-positive stroma and suggesting angiogenesis. To study how PCCs shape the stromal niche through myCAFs as the source of diseased ECM, we modeled this interaction more closely in vitro by co-culturing healthy PSCs with PANC-1 cancer cells. Healthy PSCs responded to the presence of PANC-1 cancer cells by upregulating αSMA and CTHRC1 and losing lipid droplets, key features commonly used to confirm myCAF identity in vitro (Kang et al., 2023; Yin et al., 2025; Feng et al., 2021; Wang et al., 2019). Concomitant with myCAF activation, FN fibers became more untensed and showed increased FnBPA5 binding, which we partially reversed by adding paclitaxel and gemcitabine, chemotherapeutics actively used against PDAC in the clinic, or by addition of ADAM8 inhibitors to PSCs. Together, our tissue data identify the presence of untensed FN fibers as a hallmark of PDAC stroma and, complemented by our reductionist in vitro model, demonstrate how this system can report drug effects on ECM fiber tension in a disease-specific context.

Ex vivo tissue staining

Human pancreatic ductal adenocarcinoma sampling and characterization

Non-cancerous and tumor pancreatic tissues were obtained from the Tissue Bank Bern under amended ethical approval from the Kantonale Ethikkommission (BASEC 2017-01828). Cryosections were prepared and cryobanked by Tissue Bank Bern technicians, characterized by a certified pathologist with expertise in PDAC (MW), and stored at −80 °C until further use. Both non-cancerous and tumor samples originated from surgical tumor resections, with non-cancerous tissue collected adjacent to the tumor site and confirmed as non-malignant by pathological assessment. Additional pathological characterization of the resected specimens, as summarized in Supplementary Table S1 and as annotated in Figure 1A; Supplementary Figures S1, S3A, was performed by a pathologist (MW).

Hematoxylin and eosin staining

Hematoxylin and Eosin (H&E) staining was performed by the Diagnostic Histology Laboratory of the Institute of Tissue Medicine and Pathology (ITMP). Frozen tissue sections were fixed in Delaunay’s Fixative for 2 min, followed by standard staining with hematoxylin and eosin to visualize cellular and tissue morphology.

Immunohistochemistry staining on human patient tissue sections

Cryopreserved sections from non-cancerous and tumor pancreatic tissues were stained for biomarkers and the Cy5-labelled FnBPA5 tension probe following previously established protocols (Arnoldini et al., 2017; Miéville et al., 2025a; Fonta et al., 2020). Briefly, cryosections were thawed at room temperature, rehydrated in 1x PBS for 5 min, and encircled with a hydrophobic barrier pen (Vector Laboratories, #H-4000) to prevent liquid spreading beyond the tissue section. Sections were then blocked for 30 min with 4% bovine serum albumin (BSA, Sigma-Aldrich, #05470-5G) and incubated for 1 h at room temperature with 5 µg mL −1 Cy5-labeled FnBPA5 or scrambled-FnBPA5 (control). After incubation, sections were washed three times in 1x PBS for 5 min each (standard washing procedure) and fixed with 4% paraformaldehyde (PFA; ROTI®Histofix, RotCarl Roth, #P087.6) in 1x PBS for 10 min, followed by standard washing. Permeabilization was performed with 0.1% Triton X-100 (VWR, #437002A) in 1x PBS for 30 min, followed by standard washing. Subsequently, sections were blocked for 1 h with a solution containing 5% goat serum (Sigma-Aldrich, #G9023-10ML), 0.3 M glycine (Sigma-Aldrich, #G7126-1KG), and 0.01% Triton X-100. Primary antibodies (Table 1) diluted 1:100 in the same blocking solution were applied overnight at 4 °C in a humidified chamber to minimize evaporation. The next day, sections were washed (standard procedure) and incubated for 1 h at room temperature with secondary antibodies (Table 1) diluted 1:200 in the same solution, followed by standard washing. Some sections were additionally stained with 4′,6-diamidino-2-phenylindole (DAPI; Sigma-Aldrich, #D9564-10MG; 2 µg mL −1, 10 min), followed by standard washing, and mounted using ProLong™ Gold Antifade Mountant (Thermo Fisher Scientific, #P36930). Mounted slides were allowed to harden for at least 24 h at room temperature before imaging on a confocal microscope (Leica SP8 multiphoton system or Nipkow spinning disk confocal microscope).

Microscopic examination of biopsy tissue sections

Mounted slides were imaged using either confocal or multiphoton microscopy. A Nipkow spinning disk confocal microscope was used when only fluorescent stains were imaged, whereas a Leica SP8 multiphoton system was employed for simultaneous visualization of fluorescent markers and collagen fibrils via second harmonic generation (SHG), a nonlinear optical two-photon process.

The spinning disk setup consisted of a Nikon Eclipse Ti2 inverted microscope equipped with a Confocal Scanner Unit CSU-W1-T2 SoRa (50 µm pinholes at 500 µm spacing) and a Hamamatsu Orca Fusion BT sCMOS camera (2304 × 2304 pixels, 6.5 × 6.5 µm pixel size). Images were acquired as z-stacks in either 12-bit sensitive mode or 16-bit default mode. For stitched overview images, a ×10 air objective (0.45 NA, CFI Plan Apo λ) was used, and stitching was performed in NIS Elements software using blending mode; only single z-planes were acquired for these overviews. For higher-resolution imaging, z-stacks of three slices spaced 1 µm apart were captured using a 20× air objective (0.75 NA, CFI Plan Apo λ).

When collagen fibrils were also imaged, we used a line-scanning Leica TCS SP8 MP inverted multiphoton laser scanning microscope equipped with a ×25 water immersion objective (0.95 NA, HCX IRAPO L). Fluorophores were excited in single-photon mode for FnBPA5-Cy5 and, where indicated, also for αSMA (AF555-tagged goat anti-mouse antibody). Collagen fibrils were visualized using SHG in multiphoton mode with an open pinhole. Sequential excitation was applied. For two-photon imaging, a Mai Tai XF (Spectra-Physics, USA) femtosecond Ti-sapphire pulsed laser, tuned to 880 nm, generated SHG signals detected at 440 nm sensitive to the helical nature of for collagen I and III containing fibrils. Images were acquired at 1024 × 1024 pixel resolution as z-stacks (three slices, 1 µm step size).

Imaging parameters and laser intensities were kept constant across imaging batches (patients 1 & 2, 3 & 4) and non-cancerous/tumor samples to enable quantitative comparisons. Image processing was performed in FIJI, adjusting LUTs and brightness/contrast (minimum and maximum values) identically for all samples within an imaging batch to ensure valid comparisons.

Image analysis - ratiometric heatmaps and FnBPA5/FN scores

To generate pixel-resolution ratiometric heatmaps from tissue overview images, FnBPA5 and FN channels were loaded into FIJI. Images were scaled by a factor of 0.4-0.6 depending on tissue sizes. A pixel-by-pixel matrix division was performed using FIJI’s built-in Image Calculator plugin. The resulting ratiometric image was visualized using FIJI’s “royal” LUT.

Ratiometric quantification was performed for all four patients on stitched 10× overview images for non-cancerous and PDAC tissues. Tissue areas were identified based on FN signal, where FN-positive pixels were defined as all pixels with intensities exceeding the background threshold determined from the secondary-antibody-only control (Supplementary Figure S3). Corresponding masks were loaded into the ROI Manager in FIJI, and only pixels within these regions were included in the analysis. For non-cancerous samples, the entire FN-positive tissue area was quantified. In PDAC samples, analysis was restricted to regions that were both FN-positive and annotated as “tumor” by a pathologist. To account for technical variability between independent imaging batches (patients 1 & 2 vs. patients 3 & 4), ratio values were normalized within each batch to their respective intra-patient non-cancerous controls, setting the average control value to 1. The mean ratios were exported to Prism for statistical analysis.

Image analysis - signal vicinity analysis

To quantify spatial relationships between FnBPA5 signal and other markers such as αSMA, CTHRC1, ADAM8, or collagen fibrils via SHG, we performed pixel-level proximity analysis on paired images read into RStudio as numeric matrices representing channel 1 (FnBPA5) and channel 2 (marker of interest), similarly done before (Fonta et al., 2020). Pixels were classified as positive if their intensity exceeded a background threshold defined per patient and tissue type using two controls: (i) an adjacent z-slice stained with scrambled FnBPA5 peptide (scr-Cy5) and (ii) adjacent sections stained with secondary antibodies only to account for nonspecific binding. After thresholding, binary masks were generated for both channels. For each pixel, we assessed whether it was located within a 5 × 5 pixel neighborhood, corresponding to approximately 1.6 µm × 1.6 µm at 0.325 µm per pixel, meaning any pixel within about 0.65 µm in any direction from an FnBPA5-positive pixel was considered proximal. Two complementary metrics were calculated: (1) the fraction of all image pixels that are channel 2-positive and within FnBPA5 vicinity, indicating how much of the tissue area occupied by the marker of interest is spatially associated with regions where FN fibers had lost their tension (FnBPA5-positive pixels), and (2) the fraction of FnBPA5-adjacent pixels that are channel 2-positive, indicating among all pixels near FnBPA5 what proportion also contains the marker of interest. Analyses were implemented in RStudio using custom code for neighborhood computation via a 5 × 5 moving window. Per-patient thresholds were determined manually in FIJI and imported in RStudio. Scores were computed for higher-resolution images acquired with the 20× objective on the spinning disk setup or the 25× objective for SHG on the two-photon microscope across two to three regions of interest per condition (tumor or non-cancerous). Outputs included percentages per patient, per condition, and per region, exported as CSV for downstream visualization as box plots in RStudio.

Image analysis - signal distance analysis

The analysis strategy described here is recapitulating key steps as previously used (Miéville et al., 2025a). To quantify distances from channel 2-positive pixels (αSMA, CTHRC1, ADAM8, or SHG for collagen) to the nearest channel 1-positive pixel (FnBPA5), we analyzed paired grayscale images in RStudio. Images were imported as matrices and thresholded to identify positive pixels as described above. After thresholding, the coordinates of all FnBPA5-positive and marker-positive pixels were extracted and converted to micrometer scale (0.325 µm per pixel). Nearest-neighbor distances from every channel 2-positive pixel to the closest channel 1-positive pixel were then computed using kd-tree search implemented in the RANN package (function nn2). Distances greater than 20 µm were excluded from analysis. The resulting distance distributions were summarized as histograms with 1 µm bins. For global summaries, all imaged regions from the same patient and condition (tumor or non-cancerous) were combined, and variability across regions was represented as error bars on the histogram line. For patient-level comparisons of tumor versus non-cancerous tissue, histograms were normalized within each region to fractions and then aggregated per patient and tissue as mean fraction per bin. The pipeline returned summary tables containing per-bin fractions and vectorized distance lists. Plots showing the fraction of channel 2-positive pixels (y-axis) as a function of distance to FnBPA5 in micrometers (x-axis) were generated using ggplot2 in RStudio.

In vitro culture model

Fibronectin isolation and labelling

FN was purified from human plasma obtained from the Blood Bank Zurich using gelatin sepharose affinity chromatography as described by Speziale et al. (2008). Briefly, thawed plasma was first desalted on a PD-10 column (#17-0851-01, GE Healthcare), which uses size exclusion chromatography to remove small molecules while allowing larger proteins such as FN to pass through. The eluate was then loaded onto a gelatin sepharose column, exploiting FN’s gelatin-binding domain for selective capture. After extensive washing with buffers containing gradually increasing concentrations of arginine-HCl (pH 7.4) to displace contaminants, FN was eluted using 1.4 M arginine-HCl (pH 7.4). Aliquots were stored at −80 °C until further use. For fluorescent labeling, FN was exchanged into 0.2 M carbonate buffer (pH 8.5) prepared from Na 2 CO 3 (#13418-1KG-R, Sigma-Aldrich) and NaHCO 3 (#S5761-500G, Sigma-Aldrich). Labeling was performed by reacting FN with Alexa Fluor 488 NHS ester (#A20000, Thermo Fisher) at a 25-fold molar excess for 1 h at room temperature, following the manufacturer’s instructions. Unreacted dye was removed by passing the sample through a PD-10 column, where free dye is retained while labeled FN elutes, and the protein was re-buffered into PBS. For unlabeled FN, the same PD-10 desalting step was used to exchange arginine-containing buffers for PBS.

Healthy human pancreatic stellate cell culture

Normal human pancreatic stellate cells (HPaSteC or PSCs; #3830) were purchased from ScienCell. These primary cells were isolated from the pancreas of a healthy 22-week-old male fetal donor (Lenggenhager et al., 2019). Cells were cultured in Stellate Cell Medium (SteCM; #5301; ScienCell), supplemented with 2% Fetal Bovine Serum (FBS; #S181M; Biowest or #0010; ScienCell), 1% Penicillin-Streptomycin (#15140122; Gibco or #5352; ScienCell), and 1% Stellate Cell Growth Supplement (#5352; ScienCell). T75 cell culture flasks (75 cm 2 culture area) with 10 mL of 20 μg mL −1 FN in PBS for 1 h at 37 °C and rinsed once with PBS before seeding. Cells were seeded in 10 mL complete SteCM medium. Medium was changed the day after seeding and refreshed as needed. Cells were passaged at 90% confluency for subculturing or for experiments. For passaging, cells were washed with PBS once, detached with 2 mL 0.05% Trypsin (#25300054; Gibco) for ∼ 2 min at 37 °C, and neutralized with 4 mL complete medium for counting.

PANC-1 cancer cell culture

The pancreatic cancer cell line PANC-1 was supplied by CLS (#300228; Cell Lines Service GmbH). Cells were cultured in T75 flasks with 20 mL complete DMEM, consisting of high-glucose DMEM (#41966029; Gibco) supplemented with 10% FBS (#S181M; Biowest) and 1% Penicillin-Streptomycin (#15140122; Gibco). Cell passaging for seeding experiments was performed as for stellate cells, except that 2 mL accutase (#A6964; Sigma-Aldrich) replaced Trypsin. After detachment, cells were neutralized with complete DMEM, passed through a 70 μm cell strainer (#431751; Corning) before counting and further use.

Co-culture assay (with unlabeled FN)

Cell culture

This section provides the detailed methodology corresponding to the simplified outline shown in Figure 5A. Four days before treatment start (Day −4), wells of a 24-well ibiTreat plate (#82426, ibidi) or a 12-well polystyrene plate (#92012, TPP) were coated with 20 μg mL −1 FN in PBS to fully cover the surface and incubated overnight at 4 °C. The 12-well plate condition was used for generating cell lysates for subsequent western blotting, while cells seeded in 24-well plates were used for αSMA immunostaining and BODIPY 493/503 lipid droplet staining. On Day −3, plates were sterilized for 20 min in a biosafety cabinet equipped with UV light, followed by seeding of stellate cells at 55,000 cells mL −1 in 800 μL for 24-well plates and 1600 μL for 12-well plates in complete stellate cell medium supplemented with 100 μM ascorbic acid (#A8960, Sigma-Aldrich) and 5 μg mL −1 FN in PBS. On Day −2, stellate cells were primed with 2.5 mM N-acetyl-L-cysteine (#A9165, Sigma-Aldrich) for 48 h. On Day 0, either treatments were applied by replacing the entire medium volume (800 μL for 24-well plates or 1600 μL for 12-well plates) with fresh medium supplemented with 1 ng mL −1 TGF-β1 (#100-21C, PeproTech) or 85 ng mL −1 CTHRC1 (#11647-H02H, Sino Biological), or co-cultures with PANC-1 cells were initiated as a separate condition by refreshing the well medium and adding additional medium to the inserts, which contained PANC-1 cells seeded at 73,000 cells ml -1 in 600 μL for inserts in 24-well plates and 1200 μL for 12-well plates. For co-culture, PANC-1 cell medium was changed from DMEM to stellate cell medium by resuspension after centrifugation. Inserts had a 1 μm pore size and were sourced from Sarstedt (#83.3931.101) for 12-well plates and Steriltech (#9321012) for 24-well plates. Samples were collected after three or 9 days of treatment or co-culture. During this period, all conditions received medium changes every 3 days; treatments were re-applied at each refresh, and for co-culture conditions, PANC-1 cells were re-seeded into fresh inserts at the initial starting concentration.

αSMA readout by immunofluorescence and imaging readout

To visualize αSMA stress fibers in stellate cells, samples were first rinsed three times with pre-warmed PBS and fixed with 4% PFA for 10 min, followed by three PBS washes. Cells were permeabilized for 20 min with 1% Triton X-100 in PBS, rinsed three times with PBS, and blocked for one hour in a solution containing 2% (w/v) BSA and 5% (v/v) Normal Donkey Serum (ab7475; Abcam). The primary antibody, AF594-conjugated anti-αSMA (ab202368; Abcam), was diluted 1:200 in blocking solution and incubated for one hour in the dark. After three PBS washes, nuclei were counterstained with 2 μg mL −1 DAPI for 5 min, followed by three final PBS washes before storage at 4 °C until imaging. Images were acquired using a Zeiss Axiovert microscope equipped with an Apotome module and an air Plan-Apochromat 20× objective (NA 0.8, M27-Air). The microscope was configured using the Smart Setup in ZEN 3.8 software. Excitation was provided by a X-Cite Xylis LED lamp, using the green/yellow peak at approximately 545 nm for AF594 and the UV peak at 385 nm for DAPI. Five z-planes were captured at 0.49 μm intervals with an Axiocam 807 camera. Deconvolution was applied at medium strength without Fourier transformation, and image analysis was performed in FIJI.

Lipid droplet stain by BODIPY 493/503 and imaging readout

To stain stellate cells for intracellular lipid droplets, cells were first rinsed three times with pre-warmed PBS, then fixed and washed as described above. BODIPY 493/503 (#D3922; Thermo Fisher Scientific) was prepared at 1 mg mL −1 in PBS, and cells were incubated with this solution for one hour at room temperature in the dark. After incubation, samples were washed three times with PBS, and nuclei were counterstained with Hoechst (62249; Thermo Fisher Scientific) diluted 1:1000 in PBS for 10 min. A final three PBS washes were performed before imaging. Images were acquired using the same Zeiss Axiovert microscope used for αSMA staining. Excitation was provided by the X-Cite Xylis LED lamp, using the ∼430 nm peak for BODIPY 493/503 and the 385 nm UV peak for Hoechst. Z-stack acquisition, deconvolution, and image analysis were performed identically as for the αSMA immunostaining and analysis was completed in FIJI.

Western blotting for determining CTHRC1 and vinculin protein levels

Harvest and Lysis – At day 3 of the experimental timeline (Figure 5A), stellate cells were washed once with PBS and lysed directly in the well with 50 μL lysis buffer (1% Triton X-100, #437002A, VWR; cOmplete™ Protease Inhibitor Cocktail, #11836170001, Sigma-Aldrich; 40 mM HEPES, #3724, Applichem; 5 mM MgCl 2, #63068-250G, Sigma-Aldrich; 10 mM KCl, #60130-250G, Sigma-Aldrich). Cells were mechanically detached, homogenized by pipetting, sonicated for 15 min in an ice-cooled bath, and rotated end-over-end for 30 min at 4 °C. Lysates were cleared by centrifugation (14,000 g, 15 min, 4 °C), and supernatants were collected. Protein Quantification and SDS-PAGE - Protein concentrations were determined by Bradford assay (#5000006, Bio-Rad) using reagent diluted 1:5 in water and lysates diluted 1:500. Absorbance was measured at 595 nm, and samples normalized to the lowest concentration. Lysates were denatured by addition of 4X loading dye (#1610747, Bio-Rad) and boiling at 95 °C for 5 min. Proteins were then separated by SDS-PAGE using precast gels (#4561086, Bio-Rad) and running buffer prepared from 10X concentrate (1 L H 2 O + 30 g Trizma® base, #T1503-1KG; +144 g glycine, #G7126-1KG; +10 g SDS, #A4159). Transfer to Immun-Blot® PVDF membranes (#162-0177, Bio-Rad) was performed by wet transfer in CAPS buffer (1.8 L H 2 O + 200 mL ethanol + 4.44 g CAPS, #C2632-100G; +10 NaOH tablets, #28244.295) at 250 mA for 2 h. Antibody Incubation and Detection - Membranes were blocked for 15 min in 5% milk (Rapilait, #L30195806) in TBST (10X TBS diluted 1:10 with 0.1% Tween-20, #P1379-500ML). Primary antibodies (CTHRC1, #E6W2E, Cell Signaling; Vinculin, #C117938, LSBio) were diluted 1:1000 in 5% BSA (#05470-5G) in water with 0.02% sodium azide (#71290-100G) and incubated on cut membranes overnight at 4 °C. Membranes were washed three times with TBST and incubated with secondary antibody (Peroxidase AffiniPure Donkey Anti-Rabbit IgG, #711-035-152) diluted 1:5000 in 5% milk in TBST buffer for 1 h at room temperature. After washing, blots were developed using WesternBright™ ECL substrate (#K-12045-C20, Advansta) and imaged on a Fusion FX chemiluminescence system (Vilber).

Co-culture assay (with labeled FN for final FnBPA5 binding readout)

Cell culture – chemotherapeutics

This section provides the detailed methodology corresponding to the two outlines shown in Figure 6A. The experimental timeline was identical to the previous setup for unlabeled FN, with the following modifications. Only 24-well plates were used for this peptide-binding assay. Wells were coated with unlabeled FN as before, but starting from the first stellate cell seeding, the stellate cell medium was supplemented with AF488-labeled FN at 5 μg mL −1, which remained present throughout all subsequent medium refreshes. For the left timeline (Figure 6Ai), these were the only changes. For the right timeline (Figure 6Aii), which included paclitaxel and gemcitabine treatments, an additional pre-step was introduced. Up to Day 0, the procedure was identical, but then the medium was refreshed and co-culture with PANC-1 cells was initiated for 1 day. After this period, inserts containing PANC-1 cells were removed, the medium in the wells was replaced with fresh volumes as before, and the inserts were also refreshed. For designated treatment conditions, paclitaxel (HY-B0015-10MG, Lucerna-Chem) and gemcitabine (HY-17026R-10MG, Lucerna-Chem) were added at the concentrations indicated, while control conditions (mono- and co-culture) received an equivalent amount of DMSO solvent. Cultures were maintained for 3 days after treatment before harvest.

Cell culture – ADAM8 inhibitors

For testing the effect of ADAM8 inhibition on FnBPA5 binding, PSCs were seeded as described before in the “ Cell culture–chemotherapeutics ” section, following the timeline shown in Figure 6Aii. No co-culture with PANC-1 cells was performed. Instead, ADAM8 inhibitors were added on Day 1 and cells were treated for 3 days with either 1.82 µM cyclo-RLsKDK (HY-P1676-1MG, Lucerna-Chem), 120 nM JG26 (HY-120852-1MG, Lucerna-Chem), 5 µg mL −1 propofol (HY-B0649, Lucerna-Chem), or vehicle control (DMSO; CTRL). Concentrations of cyclo-RLsKDK (Schlomann et al., 2015) and JG26 (Cuffaro et al., 2021) correspond to 10× their reported IC 50 values (182 nM and 12 nM, respectively) to ensure effective ADAM8 inhibition. The concentration of propofol (Yu et al., 2019) was selected based on published literature, using a subtoxic yet effective dose. All subsequent steps followed the same protocol as described below.

Peptide binding

To read out the tension of FN fibers with FnBPA5 binding, cells were first washed with pre-warmed PBS and incubated for 1 h at 37 °C in stellate cell medium lacking FBS but containing all other supplements and additionally supplemented with 5 μg mL −1 FnBPA5-Cy5 (synthesized by the Peptide Specialty Laboratory). After incubation, cells were washed three times with PBS, fixed with 4% PFA for 10 min, washed again three times with PBS, and stored at 4 °C until imaging.

Microscopy

Images were acquired on the spinning disk (SD) setup described above using the 20× air objective in 16-bit standard camera mode. Z-stacks of three slices were captured with a step size of 1 μm. FN-AF488 was excited with the SD GFP laser at 488 nm, and FnBPA5-Cy5 was excited with the SD Cy5 laser at 640 nm. The TIFF images were then imported as matrices in RStudio (version 2025.09.2) running on R (version 4.5.2). Each z-stack slice from FnBPA5 and FN channels was imported individually. A 3 × 3 low pass averaging filter was applied by computing the mean of each 3 × 3-pixel neighborhood and assigning it to the central pixel in a new image file. Pixel-by-pixel ratios were then calculated by dividing FnBPA5 intensity by FN intensity for each slice, followed by averaging across all slices in the stack. This produced a per-pixel mean ratio over three z-slices. Processing steps were parallelized using mclapply() in R to leverage multiple CPUs for faster computation. The resulting ratiometric heatmaps were saved as TIFF files and imported into FIJI, where the “royal” LUT was applied and brightness/contrast limits were set to 0 to 95. For quantitative analysis of each imaged region, the ratiometric matrices were processed in RStudio by computing the mean ratio of all pixels within the region. These values were then plotted as bar graphs in Prism for plotting and statistical analysis.

The stroma of pancreatic ductal adenocarcinoma is enriched in relaxed fibronectin fibers, in contrast to normal pancreatic tissues

Tissue cryosections from PDAC tissue samples and matched non-cancerous pancreas, obtained from the Tissue Bank Bern were analyzed. All patients had not received chemotherapy or radiotherapy prior to tissue donation. As FnBPA5 staining is incompatible with formalin-fixed, paraffin-embedded (FFPE) sections due to protein denaturation during embedding, only cryosections were used, limiting access to patient samples. Tumors were classified by size and grade according to routine histopathologic criteria (Table 2), and full pathological reports are provided in Supplementary Table S1.

To determine whether FnBPA5 binding, indicative of untensed FN fibers, is a hallmark of PDAC stroma, paired tumor and non-cancerous cryosections were stained with a polyclonal anti-FN antibody and Cy5-labeled FnBPA5. Tumor regions were first identified by hematoxylin and eosin staining and confirmed by a pathologist (Figure 1A; Supplementary Figures S1, S2A). Confocal microscopy at low resolution was used to image entire cryosections by stitching multiple fields. Total FN content did not visibly differ between tumor and non-cancerous regions when imaged under identical conditions and processed uniformly (Figure 1B; Supplementary Figure S2B). In contrast, FnBPA5 signal was virtually absent in non-cancerous regions but strongly present in tumor stroma, particularly in patients 1, 2 & 3, but also to certain degree in patient 4 (Figure 1B; Supplementary Figure S2B). This difference became more evident when the FnBPA5/FN ratio was calculated and visualized as heatmaps at pixel resolution (Figure 1C; Supplementary Figure S2C). To control for autofluorescence and non-specific binding, adjacent sections were incubated either with secondary antibodies alone or with a Cy5-labeled scrambled peptide (scr-Cy5) of identical amino acid composition but altered sequence; representative background images for all imaging settings used throughout this study, including those in Figures 2–4; Supplementary Figures S5, S6, are provided in Supplementary Figure S3.

To establish the presence of relaxed FN fibers as a hallmark of PDAC stroma, we quantified FnBPA5/FN ratios across tissue cryosections from four patients, enabling a statistically robust comparison between non-cancerous and PDAC regions. Analysis was restricted to tissue areas, defined as FN-positive pixels with intensities exceeding the secondary-antibody-only background threshold. Pixel-wise FnBPA5/FN ratios were computed for all FN-positive pixels and averaged per condition. For PDAC samples, quantification was restricted to pixels that were both FN-positive and located within tumor regions as identified by a pathologist on corresponding H&E sections (Figure 1A, Supplementary Figures S1, S2A). To account for batch effects (patients 1–2 vs. 3–4), ratios were normalized within each batch to the mean of the respective non-cancerous controls (set to 1). Statistical comparison of non-cancerous versus PDAC regions revealed a significantly higher normalized FnBPA5/FN ratio in PDAC tissue (Figure 1D), indicating an increased prevalence of relaxed FN fibers in PDAC tissue.

To examine FN fiber organization at higher resolution, representative regions with elevated FnBPA5 signal were imaged at higher magnification (Figure 1E, Supplementary Figure S4; patients 1–2). Non-cancerous tissue showed minimal FnBPA5 signal, whereas PDAC regions displayed strong binding also on the fiber level. FN signal appeared more fibrillar in PDAC tissue, while exhibiting a more diffuse and discontinuous pattern in non-cancerous regions (Supplementary Figure S4).

Relaxed fibronectin fibers in pancreatic ductal adenocarcinoma localize near myCAFs expressing αSMA and CTHRC1

To explain the increased FnBPA5 signal in tumor regions, we considered previous reports that untensed FN fibers often locate adjacent to αSMA-positive stromal cells (Arnoldini et al., 2017; Miéville et al., 2025a; Fonta et al., 2020). This is counterintuitive at first glance as αSMA forms contractile stress fibers, and one would expect that these cells exert tension on the ECM. However, myCAFs not only express αSMA but also assemble abundant collagen fibrils. In vitro studies have shown that mechanically more stable collagen fibrils become load-bearing, thereby leading to the relaxation of FN fibrils (Kubow et al., 2015).

To determine whether FN fiber relaxation spatially coincides with myofibroblasts (myCAFs) in PDAC, cryosections were co-stained for αSMA and FnBPA5, with DAPI as a nuclear counterstain (Figure 2Ai; Supplementary Figure S5Ai). Across all patients, tumor regions exhibited extensive stromal αSMA signal that was largely absent in non-cancerous tissue, and FnBPA5 signals were frequently observed in close proximity to these αSMA-positive regions.

Notably, in patient 4, the non-cancerous cryosection used in this analysis already displayed detectable FnBPA5 signal. This section corresponds to a different z-plane than the one shown in Supplementary Figure S2B, where FnBPA5 binding appeared lower, suggesting local variability within the tissue. Although classified as non-cancerous by pathological assessment, this observation may indicate that loss of FN fiber tension can occur prior to overt tumor development; however, this remains an isolated observation and should be interpreted with caution.

In addition to stromal αSMA, vessel-like αSMA-positive structures were observed in both PDAC and non-cancerous tissues. These likely correspond to vascular smooth muscle cells (Petsophonsakul et al., 2019), as indicated by their characteristic ring-like morphology in high-resolution images. This morphology is clearly distinct from the elongated, fiber-like αSMA structures observed within the tumor stroma, further supporting the stromal origin of the αSMA signal that spatially coincided with FnBPA5.

As αSMA alone does not fully define myCAFs, we additionally stained for CTHRC1, a myCAF marker reported to co-localize with tumor-associated macrophages to promote an immunosuppressive tumor microenvironment associated with poor patient prognosis (Li E. et al., 2024). CTHRC1 was visibly upregulated in PDAC cryosections but also detectable in non-cancerous tissues, indicating partial overlap with normal stroma (Figure 2Aii; Supplementary Figure S5Aii). The increase in CTHRC1 levels in PDAC compared to non-cancerous tissue was more pronounced in patients 1 & 2, whereas patients 3 & 4 showed less distinct differences. In particular, patient 3 displayed appreciable stromal CTHRC1 signal in non-cancerous tissue.

Background controls revealed elevated signal in ductal structures (Supplementary Figure S3), likely due to non-specific binding of the secondary antibody. In contrast, stromal fiber-like CTHRC1 patterns were clearly distinguishable from this background and represent a genuine signal. At higher magnification, these fiber-like CTHRC1 structures spatially aligned with FnBPA5-positive FN fibers, supporting their association with regions of relaxed FN fibers.

To analyze these spatial relationships more quantitatively, we performed signal vicinity analyses between FnBPA5 and αSMA (Figure 2Bi) or CTHRC1 (Figure 2Bii), as described previously (Fonta et al., 2020). For each patient and condition, three high-resolution images (20× objective) were acquired. Pixels were classified as positive if their intensity exceeded background controls as described earlier (Supplementary Figure S4). The metric calculated was the “proportion of αSMA- or CTHRC1-positive pixels within a FnBPA5-proximal region.” For each FnBPA5-positive pixel, a 5 × 5 pixel neighborhood was defined, and the presence of αSMA- or CTHRC1-positive pixels within this neighborhood was counted. Proximity was defined by this 5 × 5 pixel window, where the distance from the center pixel to the outer boundary corresponds to 2 pixels (≈ 0.65 µm). For statistical analysis, a linear mixed-effects model (LLM) was used, treating the patient as the biological replicate and modeling ROIs as nested replicates within each patient and condition. Data are presented as box plots, with statistical results shown in Figure 2Bi (αSMA) and Figure 2Bii (CTHRC1). For both myCAF markers, PDAC tissues exhibited a significantly higher fraction of αSMA- and CTHRC1-positive pixels within FnBPA5-proximal regions compared to non-cancerous tissue. For αSMA, more than 80% of positive pixels in tumor regions were FnBPA5-proximal, whereas this fraction remained below 20% in non-cancerous tissue. A similar but less pronounced trend was observed for CTHRC1, where approximately 5–10% of positive pixels in non-cancerous tissues were FnBPA5-proximal, compared to ∼50% in PDAC regions.

Next, we queried the opposite metric: how many of all FnBPA5-positive pixels are actually αSMA- or CTHRC1-positive (“percentage of FnBPA5-proximal pixels that are αSMA/CTHRC1-positive”). These metrics are shown in Supplementary Figure S5Bii for αSMA and S5Bii for CTHRC1. Notably, the trend for αSMA between non-cancerous and PDAC tissues was inverted. In non-cancerous tissue, around 50% of FnBPA5-positive pixels were αSMA-positive, whereas in PDAC this fraction was reduced to approximately 20%, a difference that was statistically significant. This likely reflects the markedly higher abundance of FnBPA5-positive stromal areas in PDAC, such that the available αSMA-positive signal cannot proportionally match the expanded FnBPA5-positive regions. In contrast, no significant differences were observed for CTHRC1, likely due to higher inter-patient variability in this marker.

Finally, we assessed spatial proximity using distance mapping, following an approach similar to that described previously (Miéville et al., 2025a). For each FnBPA5-positive pixel, we calculated its distance to all positive pixels of the second marker (Supplementary Figure S5Ci for αSMA and S5Cii for CTHRC1) within a maximum cutoff of 20 µm. Distances were binned in 1-µm intervals, and for each bin the fraction of positive pixel pairs was calculated relative to the total, enabling quantification of how frequently marker-positive pixels occur near FnBPA5. In PDAC tissues, αSMA- and CTHRC1-positive pixels were predominantly located within 0–2 µm of FnBPA5, whereas in non-cancerous tissue these markers were distributed at larger distances. For αSMA, the majority of pixels in tumors were concentrated within 0–1 µm of FnBPA5 across all patients, while in non-cancerous tissue αSMA-positive pixels were shifted to greater distances, with peaks observed at ∼5 µm (patient 1) and ∼10–15 µm (patient 2). A similar pattern was observed for CTHRC1, with most tumor-associated pixels located within 0–2 µm of FnBPA5 and a more dispersed distribution in non-cancerous tissue. This trend was particularly pronounced in patients 1 and 2, while in patients 3 and 4 the distinction between PDAC and non-cancerous tissue was less evident.

These findings underscore that untensed FN fibers in PDAC seem to be closely associated with myCAF markers, reinforcing the synergy between FN fiber relaxation and myCAF-rich stroma.

Relaxed fibronectin fibers show limited co-localization with SHG-positive collagen fibrils, while a disintegrin and metalloproteinase domain-containing protein 8 (ADAM8) is strongly associated with FnBPA53

We aimed to investigate more closely why the relaxation of FN fibers are increasingly seen in PDAC tumor sections. Based on cell culture experiments (Kubow et al., 2015) and tissue stainings (Miéville et al., 2025a; Fonta et al., 2020), it has been hypothesized that collagen fibrils, which require relaxed FN fibers to initiate nucleation of their polymerization, subsequently bare the main mechanical load in the tissue. This transfer of tension would relieve FN fibers from cell-generated traction forces, allowing the FN fibers to relax. This concept is supported by consistently high second harmonic generation (SHG) signals, which arise from the non-centrosymmetric, helical repetitive structure of collagen fibrils, which was observed in regions where FnBPA5 signal was elevated (Miéville et al., 2025a; Fonta et al., 2020).

To determine whether this relationship also applies to PDAC, we stained an adjacent cryosection to those shown in Figures 1, 2 for FnBPA5 and performed two-photon SHG imaging using a line-scanning microscope to visualize collagen fibrils. As line scanning is considerably slower, we first acquired full-tissue stitched overviews for FnBPA5 and αSMA using a spinning disk microscope to provide orientation within the specimen (Figure 3Ai; Supplementary Figure S6Ai). In the higher-magnification insets, we display FnBPA5 together with SHG. Consistent with observations in other tumors characterized by highly fibrotic microenvironments (Miéville et al., 2025a; Fonta et al., 2020), the SHG signal was also prominent in PDAC tumor regions. By visual inspection, SHG signal appeared in regions also enriched in FnBPA5 signal at a measurable distance rather than directly overlapping with FnBPA5-positive pixels.

We next performed the same proximity analysis as introduced in Figure 2, including statistical evaluation across all four patients. The proportion of SHG-positive pixels within FnBPA5-proximal regions was relatively low in PDAC tissue, averaging around 20%, but still significantly higher than in non-cancerous tissue, where values were less than 5% (Figure 3Bi). The inverse metric (’percentage of FnBPA5 proximal pixels that are SHG positive’) did not show a significant difference between PDAC and non-cancerous tissue (Supplementary Figure S6Bi). Finally, distance-based analysis was performed by calculating the fraction of pixel pair distances within 1 µm bins (Supplementary Figure S6Ci). Across all patients, SHG-positive pixels in PDAC tissue were consistently located closer to FnBPA5-positive regions than in non-cancerous tissue. However, the overall fractions remained low, as a large portion of SHG-positive pixels were located outside the 20 µm cutoff window used for the analysis.

The limited proximity of FnBPA5 and SHG suggests critical roles of additional mechanisms: FnBPA5 binding may also result from proteolytic FN degradation, which can disrupt FN fiber connections to the surrounding ECM and cells (Trachsel et al., 2024; Rappold et al., 2024), in addition or rather than from mechanical off-loading of strain from FN fibers onto collagen fribils. Matrix metalloproteinases (MMPs) are strong candidates for this, as they cleave FN, remodel ECM, and are frequently upregulated in malignancies (Duffy and McCarthy, 1998). Importantly, their net activity depends not only on enzyme levels but also on their inhibitors, the tissue inhibitors of metalloproteinases (TIMPs). An excess of TIMPs has been associated with tensed FN fibers and absence of FnBPA5 signal, as inhibition of proteolytic activity would prevent FN fiber degradation, thereby blocking their relaxation (Rappold et al., 2024). In PDAC, another FN-cleaving protease, ADAM8 has been reported to be upregulated and associated with malignancy (Schlomann et al., 2015). We therefore examined ADAM8 expression and its spatial colocalization with FnBPA5.

PDAC tumor tissue showed higher ADAM8 signal overall, most prominently in patients 1 & 2, while the difference between PDAC and non-cancerous tissue was less pronounced in patients 3 & 4, as observed in both stitched low-magnification overviews and higher-resolution insets (Figure 3Aii; Supplementary Figure S6Aii). Notably, background controls for the ADAM8 acquisition settings (Supplementary Figure S3) were elevated for patients 3 & 4, with particularly strong signal observed in ductal structures. Despite this, stromal regions in PDAC tissue still exhibited higher ADAM8 signal compared to non-cancerous tissue. ADAM8-positive features showed strong spatial overlap with FnBPA5-positive regions, especially in patients 1 & 2, and frequently extended even beyond them. Proximity analysis revealed that in non-cancerous tissue in average 5–10% of ADAM8-positive pixels were FnBPA5-proximal, whereas in PDAC this fraction increased significantly to in average ∼80% across all patients (Figure 3Bii), placing ADAM8 among the markers with the strongest spatial correspondence to FnBPA5, comparable to αSMA. The inverse metric (“percentage of FnBPA5 proximal pixels that are ADAM8-positive”) did not differ significantly between PDAC and non-cancerous tissue (Supplementary Figure S6Bii). Distance mapping further showed that ADAM8-positive pixels were located closer to FnBPA5-positive regions in PDAC compared to non-cancerous tissue in patients 1, 3, and 4, whereas no clear difference was observed for patient 2 (Supplementary Figure S6Cii).

In summary, these findings suggest that the enrichment of untensed FN fibers in PDAC might correlate with proteolytic remodeling, for example through ADAM8. In contrast, the spatial redistribution of SHG-positive collagen fibrils shows weaker spatial correlation with FnBPA5 compared to markers such as αSMA or ADAM8. However, it is important to note that these observations are based on correlational analyses at the tissue level and do not establish a causal mechanistic relationship. In particular, the proposed involvement of ADAM8-mediated proteolytic activity remains observational and cannot be interpreted as a direct mechanistic explanation for now.

PDAC stroma enriched in relaxed fibronectin fibers is interspersed with cancerous ducts and vascular structures

As PDAC arises from pancreatic epithelial cells but develops into a highly heterogeneous tumor composed of epithelial, mesenchymal-like, and stromal components, we sought to determine how tumor cells with epithelial phenotype are spatially organized relative to the stroma, which, as established in Figure 1, is characterized by massive enrichment of untensed FN fibers. Tumor cells are a major source of cytokines and growth factors that drive stromal activation and fibrotic remodeling (Than et al., 2024; Wang et al., 2019), raising the question of whether relaxed FN fibers are preferentially enriched at the tumor–stroma interface. Co-staining with E-cadherin was performed to identify tumor cells with epithelial phenotype derived from ductal cells (Sommariva and Gagliano, 2020). Notably, as in many other cancers, a subset of PDAC cells undergoes an epithelial-to-mesenchymal transition (EMT), often resulting in partial or hybrid phenotypes rather than complete mesenchymal conversion (Jolly et al., 2015). Accordingly, epithelial features are commonly preserved, particularly in well- and moderately differentiated tumors, and even in poorly differentiated PDAC a proportion of cells remains E-cadherin-positive (den Hollander et al., 2024), supporting its use as a suitable marker for tumor cell localization in this context.

To explore the spatial relationship between untensed FN fibers as detected by FnBPA5 and E-cadherin, we co-stained these markers on cryosections from patient 1 & 2 for PDAC and non-cancerous regions (adjacent to those in Figures 1–3) and examined the tissue architecture (Figure 4A). In non-cancerous tissue, E-cadherin signal appeared diffuse overall, with occasional ductal structures visible in higher-resolution insets. In contrast, tumor tissue displayed globally stronger E-cadherin staining in stitched low-resolution overview images, and magnified regions revealed highly E-cadherin-positive structures, most of which retained ductal morphology characterized by a hollow lumen. These E-cadherin-positive clusters likely represent cancer cells, as they were observed almost exclusively in tumor regions of both investigated patients (1 & 2) and were absent in non-cancerous tissue. Interestingly, we detected high FnBPA5 signal closely surrounding these ductal structures but not within or overlapping with them. The FnBPA5-positive stroma appeared confined to regions lacking E-cadherin, giving the impression that E-cadherin-positive structures intersect the FnBPA5-rich stroma. Although E-cadherin and FnBPA5 signals did not overlap, FnBPA5 was consistently found visually close to the duct-like structures.

Lastly, we examined the vascularization of FnBPA5-positive tumor stroma and its spatial correlation with relaxed FN fibers. For this purpose, we co-stained cryosections of PDAC and non-cancerous regions (patients 1 & 2) with FnBPA5 and CD31, a well-established marker for endothelial cells, including those lining blood vessels (Figure 4B). Recent work using FnBPA5 has demonstrated that relaxed FN fibers can localize next to defined vascular niches in tumors such as glioblastoma (Crestani et al., 2025), where they are enriched within aberrant microvascular proliferations (MVPs). These MVPs are complex, multilayered vascular structures composed of endothelial and stromal cells. This prompted us to investigate whether a comparable spatial association between relaxed FN fibers and the vasculature exists also in PDAC tissue.

Similar to E-cadherin, the overall CD31 signal in stitched low-resolution overview images was higher in tumor tissue compared to non-cancerous tissue, indicating increased vascularization, which coincides with the observed higher FnBPA5 signal in tumor regions. At higher-resolution insets, CD31-positive structures were evident in both tumor and non-cancerous areas, some displaying a hollow center indicative of vessel-like morphology of varying sizes. For patient 1, FnBPA5-positive stroma appeared interspersed with CD31-positive structures, including vessel-like formations, whereas for patient 2 this spatial correlation was less pronounced, with CD31-positive structures distributed across the entire region of interest and only localized FnBPA5 signal, rather than the more pervasive association observed in patient 1.

In contrast to observations in glioblastoma, where relaxed FN fibers are detected within ECM-rich MVPs, CD31-positive vessels in PDAC were not associated with intravascular or intra-structural FnBPA5 signal. Instead, relaxed FN fibers were predominantly localized in the surrounding stroma, arranged perivascularly around CD31-positive vessel structures.

In summary, we observed a clear spatial relationship between E-cadherin-positive epithelial structures and FnBPA5-positive stroma, with FnBPA5 confined to regions surrounding E-cadherin-positive ductal formations. This indicates that relaxed FN fibers are enriched at the interface with epithelial tumor cells. Furthermore, although CD31-positive structures were slightly more abundant in tumor tissue, indicating increased vascularization, loss of FN fiber tension was not localized within vascular structures themselves but instead accumulated in the surrounding stromal compartment, pointing to a distinct spatial organization of ECM remodeling in PDAC compared to glioblastoma.

Co-culture of pancreatic PANC-1 cancer cells activates healthy pancreatic stellate cells into myofibroblastic cancer-associated fibroblasts

To learn more about factors that contribute to the loss of FN fiber tension, as observed in PDAC tumor tissue stains, we used a streamlined 2D tissue culture model of PDAC onset. In this setup, two cell types were co-cultured while remaining spatially separated: one was seeded on the bottom of a tissue culture plate and the other in an insert with a porous membrane suspended in the same medium. This configuration allowed communication between the two cell types through secreted factors.

To establish this model, healthy pancreatic stellate cells (PSCs) were seeded as outlined in Figure 5A on FN-coated wells and primed with N-acetylcysteine, known to induce a quiescent phenotype in PSCs (Feng et al., 2021). Co-culture was initiated by adding PANC-1 pancreatic cancer cells into the inserts. Inserts were re-seeded with cancer cells every 3 days (for controls only media was refreshed). For some readouts, cultures were harvested already after 3 days, while other cultures were maintained for up to 9 days before harvest and analysis. Both cell types (PSCs and PANC-1 cells) were maintained in medium optimized for PSCs, which was further supplemented with ascorbic acid, required for efficient collagen fiber assembly, and soluble human plasma-derived FN, known to be co-assembled into FN fibers by stromal cell types (Baneyx et al., 2002; Zhang et al., 2014).

Using this model, we first asked whether healthy PSCs activate into a diseased phenotype upon co-culture with PANC-1 cells, indicated by the expression of the myCAF marker αSMA. PSCs cultured beneath the inserts were stained after removing the PANC-1 cancer cells, using an AF594-labeled primary antibody against αSMA with DAPI counterstain, and examined microscopically (Figure 5B; Supplementary Figure S7). As expected, PSCs expressed low αSMA levels in mono-culture but showed increased αSMA upon co-culture with PANC-1 cells. PSCs treated with TGF-β1 served as a positive control and exhibited the highest αSMA levels. These results demonstrate that our separated 2D co-culture successfully reproduced the emergence of αSMA-positive cells, consistent with observations from PDAC cryosections stains.

Next, we examined another myCAF marker: CTHRC1 expression. PSCs were either (i) mono-cultured for three or 9 days with media refreshes, (ii) co-cultured with PANC-1 cancer cells reseeded every 3 days, (iii) treated with TGF-β1, or (iv) exposed to recombinant human CTHRC1 protein which we supplemented to the medium (Figure 5C). After treatment, PSC lysates were immunoblotted for CTHRC1 with Vinculin as a loading control. The condition with recombinant CTHRC1 added as a supplement confirmed that the detected protein originated from PSC synthesis rather than from secretions of PANC-1 cancer cells. An increase in CTHRC1 levels in PSC lysates was detected only upon PANC-1 co-culture, which even increased over time from three to 9 days, while in the other conditions CTHRC1 protein levels remained low and constant. This illustrates how the synergy between the two cell types generates a PDAC-specific metabolic environment.

Finally, we assessed a third marker commonly reported for PSC activation in vitro: lipid droplet loss (Feng et al., 2021; Jesnowski et al., 2005). Healthy PSCs store vitamin A in lipid droplets, which supports endocrine pancreas function associated with glucose metabolism (Brun et al., 2015; Bachem et al., 1998; Watari et al., 1982; Zhou et al., 2021; Trasino et al., 2015). PSCs were stained with Bodipy 493/503 and Hoechst after 3 days of co-culture or TGF-β1 treatment and examined microscopically (Figure 5Di). Untreated controls visually displayed abundant Bodipy-positive droplets, whereas co-cultured PSCs showed marked depletion. For quantification, droplets per frame were counted using FIJI’s particle analysis plugin and normalized to per-frame nuclei counts as detected by the FIJI plugin StarDist, generating droplet-to-nuclei ratios for statistical testing (Figure 5Dii). Lipid droplet numbers were significantly reduced in both TGF-β1-treated and co-cultured PSCs compared to mono-cultured controls, confirming the transition into a myCAF phenotype, consistent with our other data on αSMA and CTHRC1 expression.

Fibronectin fiber tension is lost in co-culture of pancreatic stellate and cancer cells, and this effect can be reversed by chemotherapeutics used against pancreatic ductal adenocarcinoma

After confirming the emergence of myCAFs from healthy PSCs upon PANC-1 co-culture in vitro, we next investigated how this transition affects the mechanics of the assembled FN matrix. To do this, we followed a timeline similar to Figure 5A but replaced unlabeled FN with FN covalently bound to Alexa Fluor 488 (AF488) dye, enabling PSCs to assemble a fluorescently labeled matrix for ratiometric analysis of FnBPA5-Cy5 over FN-AF488 signal (Figure 6Ai). After 3 days of incubation, the FN matrix was probed with FnBPA5-Cy5 to detect relaxed fibers and examined microscopically. In addition, we implemented an alternative timeline where, after the first day of co-culture (or media refresh for untreated controls), chemotherapeutics were added at varying concentrations (or a corresponding amount of DMSO) to assess their impact on FN fiber tension (Figure 6Aii).

Using the timeline without chemotherapeutic treatment (Figure 6Ai), ratiometric heatmaps of FnBPA5-Cy5 over FN-AF488 revealed visually higher FnBPA5 binding when PSCs were co-cultured with PANC-1 cells (Figure 6Bi). This difference was statistically significant when analyzing all imaged regions of interest from two independent experiments (Figure 6Bii). These findings indicate that the emergence of myCAFs, as suggested by biomarker readouts in Figure 5, also led to changes in FN fiber tension, consistent with our observations from PDAC biopsy cryosection stainings, where tumorous tissue exhibited increased FnBPA5 binding over non-cancerous tissue.

Building on this, we asked whether our simplified 2D co-culture assay could also simulate changes in FN fiber tension upon supplementation of two clinically used chemotherapeutics, paclitaxel and gemcitabine, which we either added to mono-cultured PSCs or to co-cultures of PSCs and PANC-1 cells. This experiment served as a proof-of-concept to evaluate whether the assay can capture ECM mechanical changes during PDAC onset and to determine its sensitivity. We treated cultures with paclitaxel (0.01, 0.1, and 1 µM) and gemcitabine (0.1, 1, and 10 µM) according to the timeline shown in Figure 6Aii and assessed FN fiber tension using FnBPA5 binding as before.

Interestingly, paclitaxel at 0.01 and 0.1 µM significantly reduced FnBPA5 binding in the PSC mono-culture, while in co-cultures all three paclitaxel concentrations, including 1 μM, led to significantly reduced FnBPA5 binding. Notably, when 0.1 µM paclitaxel was added, we observed some rounded cells under brightfield phase-contrast microscopy (Supplementary Figure S8), suggesting some minor toxicity (most cells were in spindle-shape and not rounded) which was absent for gemcitabine and other paclitaxel concentrations. It cannot be excluded that toxicity may impact FnBPA5 binding.

For gemcitabine, no significant changes in FnBPA5 binding were detected in mono-cultured PSCs. However, in co-cultures, gemcitabine at 1 and 10 µM significantly reduced FnBPA5 binding.

In summary, although the changes were moderate, both drugs significantly reduced FnBPA5-Cy5 binding relative to FN-AF488 when PSCs were co-cultured with PANC-1 cells in our simplified model.

To relate the observed spatial correlation between FnBPA5 and ADAM8 signals in tissue cryosections to a perturbation-based experimental setting, we modulated ADAM8 activity in our in vitro assay and quantified the resulting changes in FnBPA5 binding. PSCs were treated with the selective ADAM8 inhibitors cyclo-RLsKDK (Schlomann et al., 2015) and JG26 (Cuffaro et al., 2021), as well as modulators of ADAM8 gene expression (Yu et al., 2019). The tensional state of FN fibers was quantified by measuring the FnBPA5-Cy5/FN-AF488 ratio as before (Supplementary Figure S9). Treatment with cyclo-RLsKDK and propofol resulted in a modest yet significant reduction in relative FnBPA5 binding (from 1.0 to ∼0.7), whereas treatment with JG26 did not result in detectable changes in FnBPA5 binding.

While the tissue co-stainings of FnBPA5 and ADAM8 on PDAC cryosections remain correlative and cannot imply a mechanistic relationship alone, the in vitro data provide a first, limited indication that modulation of ADAM8 activity may influence FnBPA5 binding under controlled conditions.

Eliciting characteristic features of the TME is crucial to understand how the cell-ECM crosstalk drives the progression of diseases, and how diseased ECM can be targeted for cancer (immuno-)therapies. While it is well-established that tissue stiffening increases with the progression of PDAC (Barrett et al., 2025; Atkins et al., 2024), our study identifies – perhaps counterintuitively – relaxed FN fibers as a defining feature of the PDAC stroma. This clearly demonstrates that tissue stiffening does not necessarily correlate with enhanced ECM fiber tension. Using the mechano-sensitive probe FnBPA5, we demonstrated that PDAC tumor regions are significantly enriched in relaxed FN fibers compared to matched non-cancerous tissue, despite similar total FN content. Relaxed FN fibers localized predominantly near αSMA-positive myCAFs and, to a lesser extent, CTHRC1-positive cells, reinforcing the link between FN fiber relaxation and myCAF-rich niches. While never reported before, spatial analysis revealed a strong correlation between FnBPA5 signal and ADAM8, a FN-cleaving protease. In vitro, co-culture of healthy PSCs with PANC-1 cancer cells induced a myCAF phenotype characterized by αSMA and CTHRC1 upregulation and lipid droplet loss, accompanied by increased FnBPA5 binding in the assembled FN matrix. Importantly, supplementation with paclitaxel or gemcitabine partially restored FN fiber tension, indicating that our simplified assay can report drug-induced changes in ECM fiber mechanics.

Highlighting the importance of synergistic paracrine signaling between different cell types, a highly reduced 2D co-culture system, that can be scaled for high-throughput screens, recapitulated key features of myCAF activation at PDAC onset. PSCs co-cultured with PANC-1 cells showed robust induction of αSMA and CTHRC1 together with pronounced lipid droplet depletion, consistent with myCAF differentiation. These changes occurred despite spatial separation of both cell types, demonstrating that cytokine-mediated paracrine signaling alone is sufficient to drive myCAF emergence, in agreement with previous studies (Feng et al., 2021; Wang et al., 2019), and without requiring additional cell types such as immune cells. At the matrix level, this minimal system reproduced mechanical signatures observed in PDAC patient tissue. FN fibers assembled by PSCs in the presence of PANC-1 cells exhibited reduced mechanical strain, evidenced by increased binding of the FnBPA5 probe. A plausible explanation is that PCC-driven myCAF emergence enhances ECM production, particularly collagens (Feng et al., 2021; Wang et al., 2019), which redistributes tensile forces and permits FN fiber relaxation (Kubow et al., 2015). Importantly, FN fiber strain remained reversible upon treatment with clinically used chemotherapeutics. In our assay, PANC-1 cells were seeded subconfluently to allow continued expansion over the 3-day culture period without overgrowth. Chemotherapeutic agents that preferentially target rapidly proliferating cells are therefore expected to limit this expansion and consequently reduce the cumulative secretion of PCC-derived cytokines that drive myCAF activation. In line with this interpretation, reduced cytokine signaling would attenuate myCAF emergence and limit FN fiber relaxation. Proteolytic mechanisms may additionally contribute to reduced FN fiber strain. In this context, addition of ADAM8 inhibitors to PSCs lowered FnBPA5 binding in vitro, indicating that FN cleavage may modulate fiber tension as well. Proteolytic degradation of FN and collagen-mediated force redistribution may act in synergy to promote FN fiber relaxation. Despite targeting fast-dividing PCCs, partial chemoresistance is expected in co-culture. PCCs exhibit increased resistance to chemotherapeutics when cultured with PSCs compared to monocultures (McCarroll et al., 2014), underscoring the protective effect of stromal interactions. Existing co-culture systems range from established cell line models (Wang et al., 2019) to complex patient-derived PCC–PSC pairings (Osuna de la Peña et al., 2021), all aiming to capture tumor–stroma crosstalk for improved disease modeling and therapeutic development.

High FnBPA5 binding in biopsy sections from other cancers was frequently associated with regions rich in SHG-positive collagen fibers. This correlation was evident in rapidly growing murine cancer models (Arnoldini et al., 2017; Fonta et al., 2023) and in human biopsy samples, including breast cancer (Miéville et al., 2025a), where the FnBPA5 signal preferentially localized near collagen fibrils. This observation is mechanistically plausible as collagen fibrils are mechanically more rigid than FN fibers and can bare substantially more load. Collagen fibrils can thus enable mechanical offloading from FN fibers, allowing FN fibers to relax, a phenomenon that has also been demonstrated in cell culture (Kubow et al., 2015). However, this relationship may not universally apply. In other pathological contexts, elevated FnBPA5 signals were reported in regions independent of the pathological built-up of SHG-positive collagen fibers, such as in glioblastoma tissue sections (Crestani et al., 2025). Both the healthy pancreas and brain are inherently soft tissues (Guimarães et al., 2020), and malignant transformations typically induce substantial tissue stiffening, with pancreatic tissue exhibiting an up to tenfold increase in elastic modulus during PDAC pathogenesis (Atkins et al., 2024). It is therefore intriguing that PDAC and glioblastoma samples (Crestani et al., 2025) showed limited correlation between FnBPA5 and SHG signal, suggesting that enhanced collagen fibrillogenesis may not be the sole determinant of FN fiber relaxation. Significantly in the context of eliciting complementary proteolytic mechanisms that lead to FN fiber relaxation, FN fibers were seen to relax also following acute inflammatory processes (Rappold et al., 2025). In human myocarditis, loci enriched in relaxed FN fibers colocalized with the presence of infiltrated macrophages (Sridhar et al., 2025). In Salmonella-infected murine guts, where the day of infection can be precisely defined, neutrophil infiltration preceded FN fiber relaxation (Rappold et al., 2025). While Salmonella infection upregulated the expression of various proteolytic enzymes, including MMPs, their inhibitors where highly upregulated too (Rappold et al., 2025; Rappold et al., 2024). We thus propose that various factors contribute to FN fiber tension loss, including immune cell infiltration and crosstalk of pancreatic cancer cells with stellate cells. The latter hypothesis was indeed confirmed in our 2D in vitro assay, free of immune cells, showing that the synergistic paracrine crosstalk between pancreatic stellate and cancer cells is sufficient to reduce FN fiber tension. This observation is highly relevant for the development of ECM targeting strategies, as we could show that treating with standard PDAC chemotherapeutics, paclitaxel or gemcitabine, can partially restore FN fiber tension. Whether the partial restoration of FN fiber tension reflects a reduced abundance of PCC-derived pro-myCAF factors resulting from chemotherapy-induced PCC death, and/or involves direct effects of these chemotherapeutics on ADAM8 expression or activity, remains an interesting question for future studies.

Besides MMP-2 and MMP-9, ADAM8 is a fibronectinase capable of cleaving FN (Sch et al., 2002). ADAM8 is upregulated in multiple malignancies, including pancreatic cancer (Schlomann et al., 2015), colon cancer (Jin et al., 2020), and glioblastoma (Cuffaro et al., 2021), and high expression correlates with poor overall survival according to The Cancer Genome Atlas and Genotype-Tissue Expression Project datasets (Jin et al., 2020). Consistent with this, we observed strong enrichment of ADAM8 in PDAC tissue compared to adjacent non-cancerous regions. Spatially, ADAM8 showed the highest correlation with FnBPA5, together with αSMA, exceeding that observed for CTHRC1 and markedly surpassing correlations with SHG-positive collagen fibrils. ADAM8 proteolytic activity depends on its multimerization, requiring at least homodimer formation at the cell surface (Schlomann et al., 2015; Sch et al., 2002). This enables autocatalytic removal of the pro-domain and activation of the protease (Mierke, 2023; Sch et al., 2002). At the membrane, ADAM8 interacts with integrin β1 and activates downstream signaling pathways, including STAT3, ERK1/2, and Akt, which promote tumor cell survival and motility, and regulates intercellular communication via extracellular vesicle release (Cook et al., 2024). Once active, ADAM8 cleaves FN primarily at the Ala 271 /Val 272 site within the linker between type I domains 5 and 6, generating characteristic VRAA 271 and 272 VYQP neoepitopes in fragments of ∼30 kDa and 50–85 kDa (Zack et al., 2009). Given the strong spatial correlation between ADAM8 and FnBPA5, mapping ADAM8-generated FN neoepitopes in FnBPA5-positive regions represents a logical next step. Although neoepitope-specific antibodies have been described (Zack et al., 2009), their limited availability currently restricts such analyses. ADAM8 appears dispensable for physiological homeostasis, making it an attractive therapeutic target in PDAC (Cuffaro et al., 2021; Li et al., 2021). Its dependence on multimerization provides a druggable vulnerability. Bivalent inhibitors that block both monomers impair tumor cell migration and invasion (Cuffaro et al., 2021), and the peptidomimetic cyclo-RLsKDK (BK-1361), which disrupts multimerization via the disintegrin domain, reduces tumor burden and metastasis in orthotopic PDAC models (Schlomann et al., 2015). Propofol has also been reported to suppress ADAM8 expression via miR-328 induction (Yu et al., 2019). In our system, ADAM8 inhibition with cyclo-RLsKDK and propofol reduced FnBPA5 binding in vitro, whereas JG26 had no significant effect. Mechanistically, propofol acts at the transcriptional level, while cyclo-RLsKDK and JG26 inhibit protease activity. The higher specificity of cyclo-RLsKDK for ADAM8, compared to JG26, which also targets ADAM17, ADAM10, and MMP-12 82, likely explains this difference. If ADAM8 is confirmed as a primary fibronectinase driving enhanced FnBPA5 binding in PDAC, targeted delivery strategies exploiting FnBPA5 for tumor localization could enhance therapeutic efficacy, particularly as both ADAM8 activity and FnBPA5 binding occur in the extracellular space.

In summary, our study identified paradoxical FN fiber relaxation as a potential hallmark of PDAC stroma and demonstrates its reproducibility in a minimal cell co-culture model. FnBPA5-based mapping revealed strong association of relaxed FN fibers with myCAF niches and ADAM8. Future work should validate proteolytic FN neoepitopes on PDAC cryosections, indicative of ADAM8 cleavage, and explore FnBPA5-guided drug delivery strategies to enhance treatment specificity.

Limitations of the study

Our analysis of cryosections revealed a strong spatial association between ADAM8, myCAF markers, and FnBPA5 binding; however, these observations remain correlative and do not establish a definitive mechanism by which ADAM8 modulates FN fiber tension. The in vitro inhibitor experiments provide limited mechanistic support, as reduced FnBPA5 binding upon ADAM8 inhibition was observed in a simplified system that does not fully recapitulate the tissue context. Direct evidence of ADAM8-mediated FN remodeling in situ is therefore still lacking. Mapping ADAM8-specific FN cleavage neoepitopes within FnBPA5-positive regions would provide an elegant and direct readout of proteolytic activity in tissue cryosections and strengthen this mechanistic link, alongside additional complementary validation approaches.

In addition, FnBPA5 staining is not compatible with FFPE sections, restricting analyses to cryosections and limiting integration with standard pathological workflows.