The Prolyl Isomerase PIN1 Impacts Fibroblast Differentiation States and Crosstalk in Pancreatic Cancer

Abstract

The prolyl isomerase PIN1 is overexpressed in cancer and contributes to cancer cell-intrinsic phenotypes including proliferation and migration. However, PIN1 may also function in stromal cells within the tumor microenvironment (TME). Here, we showed that PIN1 is a critical regulator of pancreatic stellate cell (PSC) state at baseline and in response to the myofibroblast activating factor TGF-β. Loss or inhibition of PIN1 altered the epigenetic and transcriptional response of PSCs to TGF-β, preventing PSC differentiation to a myofibroblast state and altering expression of secreted matrix proteins and signaling molecules. Consistent with inhibition of the TGF-β response, low fibroblast PIN1 expression in mouse and human pancreatic ductal adenocarcinoma (PDAC) correlated with low expression of α-SMA, a marker of myofibroblast activation. Decreased PIN1 expression at baseline also altered paracrine HGF signaling from fibroblasts to tumor cells. PSCs with low PIN1 expression displayed reduced expression and secretion of HGF, resulting in an attenuation of c-MET receptor phosphorylation and signaling in nearby cancer cells. In allograft models, host PIN1 was critical for normal growth of a subset of pancreatic cancer cell lines that are responsive to HGF signaling. Through the identification of changes to fibroblast activation state and crosstalk following PIN1 loss or inhibition, these data suggest that systemic targeting of PIN1 will suppress the pro-tumorigenic PDAC microenvironment and may differentially affect heterogeneous patient populations.

Introduction

The 5-year survival rate for pancreatic cancer patients is only 13%, the lowest of all major cancers, [1] indicating a critical need for increased mechanistic understanding of pancreatic cancer in order to identify new therapeutic targets. Pancreatic ductal adenocarcinoma (PDAC) arises due to epithelial cell transformation, which drives cellular proliferation and tumor progression [2]. Concurrent with hyperplasia of the epithelial compartment during PDAC progression, activated stromal fibroblasts proliferate and secrete extracellular matrix (ECM) proteins, growth factors, and cytokines that influence the behavior of tumor cells and other cells in the tumor microenvironment [3]. Pancreatic cancer associated fibroblasts (CAFs) have been described to play multiple, often conflicting, roles to support or restrain tumor growth [2, 3]. Recent work suggests that heterogeneous differentiation states of fibroblasts contribute to their diverse and variable functions. These CAF states are temporally and spatially controlled during tumor development, ultimately leading to the co-existence of multiple subtypes of fibroblasts in PDAC [3–10].

There are multiple cells of origin for CAFs, with recent work showing that the majority of CAFs are derived from the splanchnic mesenchyme [11]. Specific cell types including pancreatic stellate cells (PSCs), mesothelial cells, endothelial, and bone marrow derived cells have been shown to contribute to PDAC CAFs [3, 12, 13]. PSCs are resident fibroblasts within the pancreas that exist in a quiescent state in the normal pancreas, but can be activated in response to injury or tumor development and contribute to the dense stromal reaction [14]. Lineage tracing of PSC fate in in vivo models showed that while PSCs contribute to only a portion of the total CAFs, they predominately differentiate into an aggressive myofibroblast-like CAF state [12]. The specific signals driving CAF activation states have begun to be elucidated, with TGF-β shown to be a strong driver of the PSC to myofibroblast (myCAF) transition [5, 9]. TGF-β-driven CAFs make up a large portion of CAFs in mouse models and human tumors [9], and these CAFs impact the tumor environment through secretion of extracellular matrix proteins as well as paracrine signaling factors. Additional signals, such as IL-1 and IFNγ contribute to inflammatory CAF (iCAF) and antigen presenting CAF (apCAF) states, respectively [5, 15]. In vitro, PSCs display plasticity, with the ability to shift between quiescent, myCAF, or inflammatory CAF (iCAF) subtypes [4, 5]. Evidence of in vivo plasticity between CAF states suggests that reprogramming tumor supporting CAFs into a normal or tumor-restraining CAF state may suppress tumorigenic phenotypes and improve patient outcomes [16–18]. A better understanding of the mechanisms controlling plasticity between fibroblast differentiation states and the resulting impact on tumor phenotypes is needed.

PIN1 is a prolyl isomerase that is upregulated in many cancer types and can regulate multiple oncogenes and tumor suppressors, including signaling pathway mediators and transcription factors [19]. PIN1 binds phosphorylated targets at pSerine (pS)/pThreonine(pT)-Proline motifs to facilitate cis-trans isomerization of the proline adjacent to the phosphorylated pS or pT. This structural change affects the function of target proteins by altering their stability, localization, or binding partners [20]. In general, PIN1 has been described to stabilize or increase the activity of oncogenes and destabilize or decrease the activity of tumor suppressors [19]. Furthermore, PIN1 null mice are viable and display resistance to oncogene-induced tumorigenesis [21–24]. Together with the findings that PIN1 is upregulated in PDAC, and high PIN1 expression in either pancreatic tumor cells or CAFs has been shown to correlate with poor patient survival, this suggests that therapeutic targeting of PIN1 may improve patient outcomes [25–28]. Consistent with this notion, PIN1 inhibitors in combination with gemcitabine and immune checkpoint blockade were recently reported to have efficacy in preclinical models of PDAC [28]. In addition, targeting PIN1 specifically in CAFs was shown to decrease growth of a tumor cell line in vitro and synergize with an immune aptamer in vivo to eradicate its growth [29]. The mechanisms, however, by which PIN1 regulates CAF state and subsequent crosstalk to cancer cells are incompletely understood.

Here, we demonstrate that PIN1 plays a critical role in fibroblast plasticity to mediate a tumor supportive microenvironment. Full body loss of PIN1 in a highly aggressive PDAC mouse model (KPC) reduced tumor growth in vivo, and the Pin1^−/− microenvironment showed a marked loss of αSMA, a marker of fibroblast activation to a myCAF state. Similarly, in human PDAC patients, low fibroblast expression of PIN1 correlated with low αSMA expression. We tested the direct effects of loss or inhibition of PIN1 on TGF-β-driven activation of human primary PSCs and CAFs in vitro. We found that low PIN1 expression or activity mitigated TGF-β-driven activation toward a myofibroblast state, resulting in substantial changes to the expression of myofibroblast matrix proteins and signaling factors. Single cell ATAC-sequencing indicated that loss of PIN1 impaired chromatin accessibility changes in response to TGF-β treatment. In addition, at baseline, suppression of PIN1 in PSCs altered expression of secreted paracrine factors known to impact tumor progression. For example, expression and secretion of HGF was markedly decreased following loss of PIN1, resulting in decreased paracrine HGF-MET signaling in cancer cells in both in vitro and in vivo experiments. Finally, we show in vivo a direct role for microenvironmental PIN1 in controlling the growth of PDAC cell lines that are responsive to HGF signaling. Together, our data support a model whereby loss or inhibition of PIN1 affects fibroblast activation states in the PDAC tumor microenvironment, altering crosstalk between cell populations and negatively affecting growth of a subset of pancreatic cancer cell lines. Understanding the dependencies of distinct populations of PDAC cells on microenvironmental PIN1 will help determine how PIN1 targeting therapeutics might be most clinically effective in heterogeneous patient populations.

Materials and Methods:

Reagents:

Recombinant human TGF-β1 (Cat# 100–21), human HGF (Cat# 100–39H), and murine HGF (Cat# 315–23) were purchased from Peprotech (Rocky Hill, NJ). TGF-β was used at a concentration of 5 ng/mL, human HGF was used at 100 ng/mL, and mouse HGF at 5, 12.5, 25, or 100 ng/mL. KPT-6566 (MedChem Express, Monmouth Junction, NJ), ATRA (Millipore Sigma, Burlington, MA), Sulfopin (Selleck Chemicals, Houston, TX), and PHA-665752 (Selleck Chemicals or Medchem Express) were used at the indicated concentrations.

Mouse models:

All animal studies were conducted in compliance with Oregon Health & Science University (OHSU) animal use guidelines and were approved by the OHSU Institutional Animal Care and Use Committee (IACUC protocol numbers TR1_IP00001014 and TR1_IP00004194). Pin1^+/− [22], LSL-Kras^G12D [30], P53^LSL-R172H [31], and Ptf1a-Cre [32] mice have all been previously described. Pin1^−/− mice were maintained on a pure C57/Bl6 background; the KPC and KPC;Pin1 mice were on a mixed background. SHO™ (Prkdc^scidHr^hr, RRID IMSR_CRL:474) mice were obtained from Charles River (Wilmington, MA). Both male and female mice were used.

Pancreatic orthotopic allografts and xenografts were performed as previously described [33]. For allografts, the host mice were Pin1^+/+ or Pin1^−/− mice and for xenografts, host mice were SHO™ mice. Briefly, mice were anesthetized with isoflurane, and meloxicam was used as an analgesic. A small abdominal flank incision was made, and the spleen was exteriorized. Cells were resuspended in 50% DMEM / 50% Matrigel and injected into the tail of the pancreas using a 28-gauge syringe needle. The needle was held in place for 30 seconds and the injection site swabbed with sterile gauze to prevent leaking of tumor suspension. Tissue was placed back into the body cavity and the incision was sutured. Animals were monitored daily for 1-week post injection for overall health and wound healing. Mice were then imaged by ultrasound with a FujiFilm Visualsonics Vevo 2100 high frequency ultrasound (Toronto, Ontario, Canada).

For HPAFII cell co-xenografts with PSCs, 5×10⁵ cancer cells alone or 5×10⁵ cancer cells + 1×10⁶ shSCR or shPIN1 PSCs in 50 μL were injected into SHO™ mice, n=5 for each condition. For KPC8060, KPC8069, or KPC7107 injections into wildtype vs knockout hosts, 5,000 cells in 50 μL were injected into Pin1^+/+ or Pin1^−/− hosts on a pure Bl6 background.

For allografts for the PHA-665752 treatment studies, for KPC8060 25,000 cells were injected and for KPC8069 and KPC7107 5,000 cells were injected into wildtype B6 mice (The Jackson Laboratory, RRID: IMSR_JAX:000664). Seven days post injection, tumors were measured by ultrasound on a FujiFilm Visualsonics Vevo 3100, and mice randomized into two cohorts of the same average size to begin treatments. PHA-665752 was resuspended in DMSO to make a 150 mg/mL stock, which was further diluted in sesame oil for a final concentration of 6 mg/mL. Mice were treated with 30 mg/kg PHA or vehicle control (4% DMSO in sesame oil) i.p. every other day for a total of 14 days, at which point mice were euthanized and final tumor weights assessed.

Patient samples:

Human tissues, including the patient tissue for the tissue microarray (TMA) and the tissue used for generation of the CAF line (see below under cell culture) was acquired through the Oregon Pancreas Tissue Registry under OHSU Institutional Review Board (IRB) protocol #3609. All tissues were obtained with written informed consent, and studies were conducted in accordance with the Declaration of Helsinki. For TMA generation, one or two 1.5 mm cores from 34 primary PDAC samples in FFPE were arrayed on a TMA along with other metastatic and control normal tissues as described [34].

Cell culture:

ASPC1 (RRID: CVCL_0152), CFPAC1 (RRID: CVCL_1119), HPAFII (RRID: CVCL_0313), and PANC1 (RRID: CVCL_0480) cell lines were obtained from Joe W. Gray (Oregon Health & Science University) and were maintained in DMEM supplemented with 10% FBS. Cell lines were authenticated by STR profiling and tested to ensure they were free of mycoplasma contamination. In general, cells were tested for mycoplasma when banked, and then used for fewer than 10 passages or 2 months in culture, depending on growth characteristics. KPC8060, KPC8069, and KPC7107 cell lines were provided by Michael A. Hollingsworth (University of Nebraska Medical Center) and cultured in DMEM supplemented with 10% FBS. Primary human pancreatic stellate cells (PSCs; Cat# 3830) were obtained from ScienCell (Carlsbad, CA) and grown in complete Stellate Cell Medium. Three separate lots, 10295, 27648, and 36362, were used throughout this manuscript. Immortalized mouse PSCs were obtained from Mara H. Sherman (Oregon Health & Science University) and grown in DMEM supplemented with 10% FBS.

CAFs were propagated from dissociated primary PDAC tissue (patient public ID: ST-00020107) and cultured in PDCL media (F-12:DMEM (1:3) media supplemented with 5% FBS, 0.4 μg/ml hydrocortisone, 5 μg/ml human insulin, 8.4 ng/ml cholera toxin, 10 ng/ml EGF, 24 μg/ml Adenine, 10 μM ROCK inhibitor Y-27632, 0.25 μg/ml Amphotericin B, and 1X Primocin). Hallmark fibroblast gene and protein expression, as well as minimal KRT gene and protein expression, were confirmed by RNA-seq and IF. STR profiling (Genetica, Burlington, NC) matched that of the respective patient’s peripheral blood leukocytes, and while mutant KRAS was confirmed in the patient tumor, wild-type KRAS codons 12 and 13 was confirmed in this CAF line.

Cells with stable knockdown of PIN1 were generated by lentiviral infection of pLKO-Pin1 shRNA #1 (TRCN0000001033) or #5 (TRCN0000010577) or an shSCR control (SHC-002) (MilliporeSigma), followed by selection with 5 μg/mL puromycin. Stable knockdown pools were assessed to confirm knockdown by Western blot and then used in further experiments. Overexpression of PIN1 was achieved with adenoviral infection using Ad-Pin1, and control infections with Ad-GFP as described [35].

Immunohistochemistry, Immunofluorescence, and Imaging:

H&E staining was performed with standard methods with Hematoxylin (Vector Laboratories, Burlingame, CA) and Eosin Y solution (MilliporeSigma). Masson’s Trichrome staining was performed with a Trichrome Stain Kit (MilliporeSigma) or with Richard-Allan Scientific Masson Trichrome Kit (Thermo Fisher) according to the manufacturer’s protocol.

For IHC, formalin-fixed, paraffin embedded (FFPE) sections were de-paraffinized, rehydrated, and antigen retrieval performed with pH9 Tris-EDTA buffer (Dako) in a pressure cooker for 10 min. After cooling for 20 min, slides were quenched in 3% hydrogen peroxide for 10 min and blocked with 5% goat serum and 2.5% BSA for 1 hour at room temperature. Primary antibodies pMET (D26; Cell Signaling Technology Cat#3077, RRID AB_2143884, 1:200) or tMET (D1C2; Cell Signaling Technology Cat# 8198, RRID AB_10858224, 1:150) were diluted in blocking buffer and tissues incubated overnight at 4°C. Sections were incubated with secondary antibodies for 1 hour, the Vectastain ABC kit (Vector Laboratories) for 1 hour, and then exposed to the DAB substrate for 5–10 minutes for color development. Slides were counterstained with hematoxylin for 5 min and then mounted using Vectamount mounting media (Vector Laboratories).

For IF on FFPE mouse tissue sections, slides were de-paraffinized, rehydrated, and antigen retrieval performed in citrate buffer pH6 (Sigma) in a pressure cooker for 15 minutes, pressure was released, and then slides were incubated in pre-heated pH9 Tris-EDTA (Dako) for 10 minutes. Slides were then allowed to cool to room temperature. Slides were placed in a 3% H₂O₂ bleaching solution under direct light for 20 minutes and then were blocked with 5% goat serum, 3% BSA and 0.2% Triton X. Primary antibodies KRT8/18 (Fitzgerald Cat# 20R-CP004, RRID AB_1284055, 1:100), αSMA (EPR5368, AF-488 conjugate; Abcam Cat#202295, RRID: AB_2890884, 1:100), PDPN (8.1.1; Thermo Fisher Cat# 14–5381, RRID: AB_1210505, 1:100), pSTAT3 (D387; Cell Signaling Technology Cat #9145, RRID: AB_2491009, 1:100), or CD74 (LN-2; Santa Cruz Biotechnology sc-6262, RRID: AB_627176, 1:100) were diluted in blocking buffer and tissues incubated overnight at 4°C. Secondary antibodies were diluted 1:250 in blocking buffer and applied for 1 hour at room temperature. Slides were stained with DAPI (1:1000) and mounted with antifade mounting reagent.

For IF on the human TMA, a single section of the TMA was iteratively stained for PIN1, αSMA, KRT19, CDH1, CD31, and CD45 expression using the cyclic immunofluorescence protocol at dx.doi.org/10.17504/protocols.io.23vggn6. Briefly, slides were de-paraffinized, rehydrated, and antigen retrieval performed in citrate buffer pH6 (Sigma) followed by pH9 Tris-EDTA (Dako). Slides were blocked with 10% normal goat serum and 1% BSA. Primary antibodies PIN1 (G8; Santa Cruz Cat# sc-46660 AF488, RRID: AB_3697814, 1:50), αSMA (1A4; Santa Cruz Cat# sc-32251 AF488, RRID: AB_3697813, 1:200), KRT19 (A53-B/A2; Biolegend Cat# 628502, RRID: AB_439773, 1:200), CDH1 (EP700Y; Abcam Cat# ab201499, RRID: AB_2910587, 1:200), CD31 (EPR3094; Abcam Cat# ab218582, RRID: AB_2857973, 1:100), and CD45 (EP322Y; Abcam Cat# ab200317, RRID:AB_3697812, 1:50) were all directly conjugated to AlexaFluor dyes. They were diluted in 5% goat serum and 3% BSA and tissues incubated overnight at 4°C. Slides were mounted with antifade mounting reagent with DAPI.

For IF on cultured cells, cells were plated into black-walled 96-well plates, cultured and treated as described, and then fixed at endpoint for 15 minutes in 2% paraformaldehyde in PBS. Cells were permeabilized in 0.3% TritonX-100, washed 3X in PBS, and then incubated overnight at 4°C with anti-E-cadherin (36/E-cadherin; BD Biosciences Cat# 610181, RRID: AB_397580) diluted 1:200 in blocking buffer (2% BSA in PBS), followed by secondary antibody staining. For BODIPY staining, after permeabilization, cells were treated with 20 μg/mL BODIPY 493/503 (Thermo Fisher, Cat# D-3922) diluted in PBS. For both approaches, DAPI was used at 1 μg/mL.

H&E, Trichrome, and immunostaining images were acquired on a Zeiss AxioScan with Zeiss Zen software (Zeiss Microscopy, Thornwood, NY). Tiled images were digitally stitched by Zen software to generate full scan images. Some individual images were also taken with a Hamamatsu digital camera mounted on a Leica fluorescence microscope. Phase contrast images of cells and BODIPY images were taken on an EVOS-FL microscope (Thermo Fisher). IF images of cells in 2D culture were acquired on an INCell Analyzer 6000 (GE Healthcare Life Sciences, Pittsburgh, PA).

Image quantification:

Quantification of IHC staining for pMET or tMET in autochthonous, allograft, or xenograft tumors, was performed by deconvolution of the IHC image using ImageJ Fiji software (RRID: SCR_002285) [36] version 2.1.0/1.53c. Color Deconvolution with the “H DAB” vector option to separate the Hematoxylin staining (blue/purple) from the DAB staining (brown). The threshold was adjusted for the DAB staining image, and the % positive area and intensity were calculated. To quantify % Trichrome positive staining, a binary image was created in ImageJ, and the “analyze particles” function was used to calculate the area of collagen (blue) staining as a percent of the total pancreas area. All tumor staining quantified is shown as an average value for each mouse, with n indicated on the plot or in the figure legends. At least 4 ROIs were analyzed in each tumor.

For quantification of IF staining for CAF phenotypes in KPC tumor tissues, we used Zeiss arivis Pro 4.3.0 analysis software. Following slide scanning, 4 ROIs were taken from representative areas of the tissue. ROIs were input into arivis and the DAPI channel was used in the ‘Cellpose-based Segmenter’ module to identify and count nuclei based on area (minimum of 2 μm²). These nuclei were then dilated using the ‘Watershed method; Region Growing’ module using size and thresholding (Maximum distance: 1.5 μm, threshold 490). This module was then used as the input for filtering using the ‘Object Feature Filter’ that allowed us to identify single cells positive for PDPN staining and negative for KRT8/18. Nuclear staining for pSTAT3 or cellular staining for αSMA or CD74 was then determined, and results are shown as percent of cells with specific staining patterns. All quantitation is shown as an average of the ROIs from each mouse tumor.

For quantitation of the TMAs iteratively stained for PIN1, αSMA, KRT19, CDH1, CD31, and CD45 expression, images were registered and aligned by DAPI [37]. Segmentation was performed using the Mesmer algorithm [bioRxiv 2023.01.31.525753]. Cells were binned into epithelial (KRT19+ CD31- CD45-), endothelial (KRT19- CD31+ CD45-), immune (KRT19- CD31- CD45+), and fibroblast (KRT19- CD31- CD45-) groups. Fibroblast staining for PIN1 and αSMA was determined for each of the primary PDAC samples, shown as an average of two spots for any patient with replicate spots. Fibroblast PIN1 levels for each patient were also correlated with the percent of cells labeled as fibroblast, epithelial, and endothelial.

BODIPY was quantified in CellProfiler. Briefly, two groups of matched images, BODIPY and DAPI stained, were loaded to the CellProfiler software and were converted to grayscale images using ‘ColorToGray’ modules. In the DAPI image, the ‘IdentifyPrimaryObjects’ module was used to count the number of nuclei based on size, shape, and pixel intensity (minimum diameter: 15; maximum diameter: 80-pixel units). In the BODIPY image, the ‘IdentifyPrimaryObjects’ module was used to count the number of lipid droplets in each image also based on the size, shape, and pixel intensity of the lipid droplets. The average number of BODIPY lipid droplets per nuclei was averaged for each condition for four biological replicates.

To calculate cell area from conditioned media experiments, single cells were segmented from bright field images using the Cellpose-SAM model [bioRxiv 2025.04.28.651001] and stored as labeled images. The number of pixels contained within each cytoplasmic mask (label) was calculated within scikit-image and celltraj [38] software packages to quantify the attachment area of each cell to the plate with at least 400 individual cells counted per condition in each of three independent experiments.

Western Blots:

Cells were lysed in AB Lysis buffer, RIPA buffer, or SDS loading buffer and Western blot analysis was performed. Blots were visualized and bands quantified on an Odyssey imaging system (LI-COR Biosciences, Lincoln, NE). Primary antibodies to PIN1 (G8, Santa Cruz Cat# sc-46660, RRID: AB_628132, 1:500), αSMA (1A4; Thermo Scientific Cat# 53–9760-82, RRID: AB_10979529, 1:500), PALLD (1E6; Novus Biologicals Cat# NBP1–25959, RRID: AB_1726016, 1:1000), pMET (D26; Cell Signaling Technology Cat# 3077, RRID: AB_2143884, 1:1000), tMET (25H2; Cell Signaling Technology Cat# 3127, RRID: AB_331361, 1:1000), PDGFRA (D1E1E; Cell Signaling Cat# 3174S, RRID: AB_2162345, 1:1000), PDPN (D9D7; Cell Signaling Cat# 9047S, RRID: AB_2797694, 1:1000), KRT8/18 (C51; Cell Signaling Cat# 4564S, RRID: AB_2134843, 1:1000), and GAPDH (6C5; Thermo Fisher Cat# AM4300, RRID: AB_2536381, 1:5000) were diluted in blocking buffer (either 1:1 Odyssey Blocking Buffer (LI-COR Biosciences):PBS or 1X Blocker FL Fluorescent blocking buffer (Thermo Fisher)). Primary antibodies were detected with secondary antibodies labeled with the near-infrared fluorescent dyes IRDye800 (Rockland, Philadelphia, PA) and Alexa Fluor 680 (Molecular Probes, Eugene, OR) diluted 1:20,000 in blocking buffer.

Quantitative RT-PCR (qRT-PCR):

RNA was isolated using Trizol (Life Technologies) or an RNeasy kit with on-column DNase treatment (Qiagen). cDNA was generated using the Multiscribe Reverse Transcriptase kit (Thermo Fisher). qPCR analysis was performed with Fast SYBR Green reagent (Thermo Fisher) on a StepOne machine (Applied Biosystems) or PowerUp SYBR Green master mix (Thermo Fisher) on a QuantStudio 6 machine (Applied Biosystems). Primers were validated by performing a standard melt curve analysis. Primers for human HGF were 5’-CCCTGTAGCCTTCTCCTTGA-3’ and 5’-CGAGGCCATGGTGCTATACT-3’ and for human TBP were 5’-TGCACAGGAGCCAAGAGTGAA-3’ and 5’-CACATCACAGCTCCCCACCA-3’.

Cell growth assays:

To assess population growth, 1500 cells were seeded into each well of a 96 well plate and allowed to attach overnight. Some cells (shSCR, shPIN1 PSCs) were left untreated while in other experiments, hPSCs were treated with varying doses of PIN1 inhibitors or vehicle controls the following day. Plates were then imaged on an IncuCyte ZOOM or Incucyte SX5 (Essen Bioscience) at regular intervals, and the Incucyte software was used to calculate the cell confluence at each time point. Shown is the average of three independent experiments.

Spheroid assays:

To assess spheroid growth, 2000 KPC cells were seeded into a low attachment U-bottom plate in either DMEM or Advanced DMEM with 2% serum. Cells were allowed to form spheroids for three days, after which time they were treated in 1% FBS +/− HGF at the concentration stated in the figure or figure legend. Spheroid growth was monitored on an Incucyte SX5 for six days, with media +/− HGF replenished after 72 hours. Three independent experiments with at least three technical replicates per experiment were performed.

ELISA and Ligand Binding Assays for CM:

For ELISA assays, hPSCs were seeded in 6-well plates at a density of 300,000 cells/well. After 40 hours, media was collected, centrifuged to remove debris, and stored at −80C until assayed. An HGF ELISA (R&D Systems, Cat# DGG00B) was performed per the manufacturer’s protocol. Briefly, 50 μl of sample or standard were added to precoated wells containing 100 μl of assay diluent RD1–38. Plates were incubated with agitation for 2 hours at room temperature and then washed 4 times. Plates were then incubated with 200 μl/well human HGF conjugate for 1 hour, with agitation. After washing to remove unbound conjugate, the plates were incubated in 200 μl/well streptavidin-HRP for 30 min with agitation. Following a final wash, 200 μl substrate solution was added to each well and incubated for 30 minutes, and then color development was stopped by adding 50 μl stop solution. Plates were scanned at 450 nm followed by a scan at 540 nm. Readings at 540 nm were subtracted from readings at 450 nm. All samples were tested in duplicate, and the assay was repeated for three biological replicate experiments.

For the Luminex assay, conditioned media (DMEM 10% media conditioned for 24 hours on confluent shSCR or shPIN1 PSCs) was collected, and 50 μL used in a Luminex Multiplex magnetic beads 30-plex Assay (Thermo Fisher, Cat# LHC6003M) according to manufacturer’s instructions. Two independent experiments were performed.

RNA-sequencing:

For RNA-sequencing, samples included shSCR or shPIN1 pools of hPSCs with and without 5 ng/mL TGF-β1 treatment for 24 hrs (n=5 for each condition) or parental PSCs treated with DMSO, 10 μM ATRA, or 10 μM KPT-6566 for 48 hours, prior to treatment with 5 ng/mL TGF-β1 for 24 hrs (n=3 for each condition). RNA was isolated from cells using an RNeasy kit with on-column DNase treatment (Qiagen).

Sequencing libraries were constructed in two batches. One included two sets of the shSCR and shPIN1 conditions (reps 4 and 5 in the GEO submission; 8 samples total). These libraries were constructed using a TruSeq Stranded mRNA kit (Illumina, San Diego, CA). Briefly, poly(A)+ RNA was isolated from 400 ng of total RNA (per sample) using oligo-dT-coated magnetic beads. The recovered RNA was then chemically fragmented. First strand cDNA was generated using random hexamers as primers for reverse transcriptase. Strand specific second strand cDNA synthesis was performed, and a single ‘A’ nucleotide was then added to each end. Illumina adaptors were ligated to the cDNAs. Limited-round (15-cycle) PCR was used to amplify the material to yield the final libraries. Library concentration was determined using real-time PCR with primers complementary to the Illumina adaptors. Sample libraries were diluted and applied to an IlluminaNextSeq500 HiOutput flow cell at a concentration appropriate to generate about 400 million reads. All libraries were prepared with indexing barcodes to permit multiplexing in a single lane. The 75-cycle single read sequencing was used to assemble the reads into standard fastq formatted data.

For the rest of the samples (reps 1, 2, and 3 of the shSCR and shPIN1 conditions as well as all of the inhibitor conditions), libraries were constructed by Novogene, Inc, using the ABclonal Fast RNA-seq Lib Prep Kit V2. Briefly, mRNA was purified from total RNA using poly-T oligo-attached magnetic beads. After fragmentation, the first strand cDNA was synthesized using random hexamer primers followed by the second strand cDNA synthesis. The library then underwent end repair, A-tailing, adapter ligation, size selection, amplification, and purification. The library was checked with Qubit and RT-PCR for quantification and bioanalyzer for size distribution. Quantified libraries were pooled and sequenced on an Illumina NovoSeq X plus for paired end sequencing. RNA-seq data can be accessed at the NIH Gene Expression Omnibus (GEO), data accession GSE299852 for the shSCR and shPIN1 samples and GSE299851 for the inhibitor treated samples.

For RNA-seq analysis, adapter trimming and quality filtering were performed using Trimmomatic v0.39 (RRID:SCR_0011848). Reads shorter than 36 bases after trimming were discarded. Trimmed reads were aligned to the GRCh38 human reference genome using STAR v2.7.11b with default parameters. Aligned BAM files were subsequently processed with featureCounts v2.0.8 to generate gene-level count matrices. DESeq2 R package version 1.46.0 [39] was used for factored analyses. GSEA (Version 4.3.2; [40]) was run on the ranked list of shSCR vs shPIN1 to compare enrichment to genesets in the C6 (oncogenic signatures) and C2 (curated genesets) databases. All pipelines used for RNA-seq analyses are available on GitHub: https://github.com/kyra-lindley/Langer_pin1/tree/main.

sci-ATAC-seq:

For sci-ATAC-seq analysis, barcoded adaptor oligos were loaded on Tn5 transposase in accordance with previously published work [41, 42]. PSCs (shSCR and shPIN1 pools) were treated with 10 ng/mL TGF-β for 8 hours, trypsinized, and washed with 10 mL ice cold PBS. To isolate nuclei, the cells were resuspended in 1 mL ice cold Nuclei Isolation Buffer (NIB) and incubated on ice for 15 min. DAPI was added to the NIB-suspended nuclei to a final concentration of 5 μg/ml. Following DAPI staining and counting, 5000 nuclei were plated into each well of 96-well plates (Eppendorf) in 10 μl 1X TD buffer (Illumina) with 7 μM Pitstop2 (Abcam). Tagmentation was performed by adding 1 μL of 8 μM dual barcoded Tn5 to each well, followed by incubation at 55°C for 15 minutes. Plates were then put on ice to stop the reaction. After tagmentation, all wells were pooled. 22 tagmented nuclei were FAN sorted and deposited into each well of two 96-well PCR plates containing 2.5 μL of barcoded forward and reverse primer, 0.25 μL 20 mg/mL BSA, 0.5 μL 1% SDS, and 7.75 μL nuclease-free water. After sorting, plates were incubated at 55°C for 15 min to denature transposase. 12 μL PCR mix (7.5 μL Nextera PCR Mix (NPM), 4 μL nuclease-free water, 0.5 μL 100 x SYBR Green) was added to each well. PCR amplification was performed with a Bio-Rad CFX thermocycler under the following conditions: 72°C for 5 minutes, then 98°C for 30 seconds; followed by cycles of 98°C for 10 seconds, 63°C for 30 seconds, 72°C for 1 minute, plate read, and 72°C for 10 seconds. Reactions were pulled in mid-exponential, typically between 18–20 cycles. 10 μL of each reaction was cleaned after PCR amplification through a QIAquick PCR purification column. We assessed library quality by using an Agilent Bioanalyzer, then diluted and sequenced on a research use-only NextSeq 500. Sequencing was performed with custom primers and chemistry, as previously described [43].

Processing of sequence data and subsequent analysis were done primarily through the single-cell combinatorial indexing (sci-) software package scitools (https://github.com/adeylab/scitools) [41], which contains wrappers of external tools and commands for common steps in sci- data analysis. Raw sequencing reads in BCL format were first converted to FastQ format using bcl2fastq, and then demultiplexed into FastQ files with matched barcodes via the scitools command ‘fastq-dump’. Barcodes were considered matched if their constituent indexes were within two Hamming distances away from a match in a list of possible index combinations. FastQs were aligned to human reference genome hg38 by using scitools ‘fastq-align’, which maps reads via bwa-mem (0.7.15-r1140) [44]. For quality control, we used scitools ‘bam-rmdup’ and ‘bam-filter’ to filter for quality 10 aligned reads, perform barcode-based PCR duplicate removal, and exclude cell barcodes containing less than 6275 passing reads. Peak calling was performed on the filtered bam file via scitools ‘atac-callpeak’, which utilized MACS2 [45] for peak identification, filtering, and extension to 500 bp. All raw and processed data can be accessed at GEO, accession GSE300473.

For the Euclidean distance calculation to determine similarity of chromatin accessibility profiles by treatment group, we first filtered the cisTopic cell x topic matrix to remove any shPIN1 cells that likely failed to achieve PIN1 knockdown (discussed below), then split the matrix into two separate cell x topic matrices containing shSCR and shPIN1 cells respectively. Pairwise Euclidean distances between cells were calculated for each matrix using the R function dist with method set to “euclidean”. Distance values for each unique comparison between TGF-β-treated cells and untreated cells were collected for the shSCR distance matrix and the shPIN1 distance matrix, then plotted with ggboxplot from ggpubr v0.4.0. Significance was calculated through the t_test function provided by rstatix v0.6.0. To remove PIN1 knockdown-failed cells prior to the distance calculation, we used PhenoGraph [46] to perform Louvain clustering on cells by inputting the unfiltered cell x topic matrix into the scitools function ‘matrix-pg’ (with ‘-k’ nearest neighbors set to 50). This function assigned all cells to a PhenoGraph group and generated an annotation file recording these assignments. PhenoGraph groups that represented shPIN1 TGF-β+ and shSCR TGF-β+ clusters were identified by coloring UMAP cell projections by treatment group or PhenoGraph group and looking for overlap. Cells that were within these identified PhenoGraph groups but not likewise annotated as shSCR TGF-β or shSCR UN cells were removed from the cell x topic matrix prior to the distance calculation described above.

To generate the counts matrix and cisTopic model, the filtered bam file and bed file of called peaks were inputted to the scitools function ‘atac-count’ to create a read counts matrix of barcoded cells (columns) by peaks (rows). Scitools ‘matrix-cistopic3’, a wrapper function for cisTopic v3 [47], was then called to perform Latent Dirichlet Allocation-based dimensionality reduction on the counts matrix and identify groups of co-accessible genomic regions (topics). The cisTopic cell x topic matrix was visualized as a heatmap through scitools ‘matrix-bicluster’, which utilizes the R package ComplexHeatmap v1.20.0 (https://github.com/jokergoo/ComplexHeatmap) [48] to perform biclusterization.

Using topic weights in the cisTopic cell x topic matrix, we performed uniform manifold approximation and projection (UMAP) [49] to project cells into 2D space. UMAP coordinates were generated through the R package umap (https://github.com/tkonopka/umap) that was called with the scitools wrapper function ‘matrix-umap’. All UMAP plots were generated by entering the output of ‘matrix-umap’ into the scitools function ‘plot-dims’, along with a file that specifies an annotation for each cell. Example annotations include treatment group, topic weight, or chromVAR motif deviation score.

To select topics relevant to the cell populations in our experiment, we annotated each cell projected on the UMAP plot according to its topic weight value. If topic weight values were elevated in a UMAP cluster corresponding to one or more treatment groups, we selected the topic for further analysis. Further confirmation of topic-group association was performed by plotting topic weight distribution across the treatment groups for each topic. We defined Topics 2, 10, and 13 as shSCR TGF-β regulated topics, while Topics 4, 15, and 20 were defined as All TGF-β regulated topics in both cell types.

We assessed global motif accessibility through the R package chromVAR (http://www.github.com/GreenleafLab/chromVAR) [50] by executing the scitools wrapper function ‘atac-chromvar’. ChromVAR supplies motif collections from multiple sources; for our analysis we used human_pwms_v2, which contains motifs derived from the cisBP database (http://cisbp.ccbr.utoronto.ca) [51]. The output of the scitools chromVAR wrapper function includes motif accessibility deviation Z-scores for each cell, which we used to create violin plots and compare global accessibility of selected motifs between treatment groups.

For identification of enriched motifs within topics, we used the scitools ‘da-enrichment’ wrapper function with options ‘-r’ and ‘-B’ enabled. These settings run a HOMER [52] motif analysis program, findMotifsGenome (http://homer.ucsd.edu/homer/ngs/peakMotifs.html), on test sets of genomic regions versus a background set (specified in scitools by -B and -r, respectively). In our analysis, the test peak sets were set as the groups of topics defined above (shSCR TGF-β and All TGF-β) while the background was set to the bed file of all called peaks.

Violin plot distributions of topic weights and chromvar accessibility deviation scores were plotted with the ggviolin function from the ggplot2-based R package ggpubr v0.4.0 (https://rpkgs.datanovia.com/ggpubr/). Internal boxplots were added with geom_boxplot, where the center black line represents the median, the upper and lower hinges represent the 75% and 25% quantiles respectively, and the lower and upper whiskers represent 1.5 times the inner quartile range. Outliers are represented by the black colored data points directly above and below boxplot whiskers. Significance calculations for all violin plots were performed using the wilcox_test function from the rstatix R package (v0.6.0; https://rpkgs.datanovia.com/rstatix/). P values were adjusted via p.adjust from R stats v4.0.2, with correction method set to “bonferroni”. For topic weight violin plots, the number-of-comparisons parameter n in p.adjust was set to n=120, representing 30 topics and 4 Wilcoxon tests for each. For violin plots distributing chromVAR deviation scores, this parameter was set to n=3480 to accommodate 4 Wilcoxon tests for 870 motifs each.

Data and statistics:

Three or more independent biological replicate experiments were performed in all cases, except as indicated in the figure legend or when primary patient materials were used. For all work except sequencing analyses, GraphPad Prism software was used for statistical analysis. Individual tests and p values are described in the figure legends.

Data availability:

RNA-sequencing and ATAC-seq datasets and analysis are included in Supplementary Datasets 1-15. All raw and processed data are publicly available at the Gene Expression Omnibus (GEO). The accession numbers for RNA-seq data are GSE299852 for the shSCR and shPIN1 samples and GSE299851 for the inhibitor treated samples. The accession number for ATAC-seq data is GSE300473. All pipelines used to analyze RNA-sequencing are available at: https://github.com/kyra-lindley/Langer_pin1/tree/main and to analyze sciATAC-sequencing are available at: https://github.com/adeylab/scitools. All other raw data are available upon request from the corresponding author.

Results

Loss of PIN1 decreases tumor burden and affects αSMA expression in the TME

Prior studies in mouse models demonstrated a striking suppression of mammary tumorigenesis in a Pin1^−/− background driven by either mutant HRAS or mutant p53 [23, 24]. To determine how whole-body loss of PIN1 affects pancreatic tumor growth driven by both mutant KRAS and mutant p53, we crossed the KPC alleles into a germline Pin1^−/− background. PIN1 loss alone did not alter normal pancreas histology or relative size (Figure 1A, Supplementary Figure S1A), consistent with other studies demonstrating that Pin1^−/− mice are viable, with minimal deleterious phenotypes under normal conditions [21, 22]. KPC;Pin1^−/− pancreata at 5 months of age, however, were significantly smaller than the KPC;Pin1^+/+ pancreata, showing reduced tumor burden (Figure 1A, Supplementary Figure S1B, C). We assessed the histology of these tissues and found reduced Trichrome staining in the lesions of KPC;Pin1^−/− mice (Figure 1B, C), indicating changes to the tumor stroma. To further understand the altered stroma, we stained KPC;Pin1^+/+ and KPC;Pin1^−/− lesions for markers of CAF states, using αSMA, pSTAT3, and CD74 to detect myCAF, iCAF, and apCAF states, respectively. While we found the total fibroblast and epithelial cell numbers to be similar between KPC;Pin1^+/+ and KPC;Pin1^−/− lesions (Supplementary Figure S1D, E, F), the percent of αSMA+ CAFs (αSMA+ PDPN+ KRT8/18- cells) was significantly decreased (Figure 1D). There was no significant change in the percent of fibroblasts (PDPN+ KRT8/18- cells) positive for pSTAT3 or positive for CD74 (Figure 1E, F). To assess the association of fibroblast PIN1 and αSMA expression in human pancreatic cancer, we stained a tumor microarray of 34 human PDAC samples and found that fibroblast expression of PIN1 and αSMA were significantly correlated (Figure 1G, H, Supplementary Figure S1G, H). Again, in human tumors, we did not observe a correlation of fibroblast PIN1 expression with the total number of fibroblast, epithelial, or endothelial cells (Figure 1I). Fibroblast PIN1 expression was also not significantly correlated with epithelial CDH1 expression (Supplementary Figure S1I). Together, these data suggest that loss of PIN1 expression regulates fibroblast phenotypes in pancreatic cancer.

Figure 1: Loss of PIN1 decreases tumor burden and affects αSMA expression in the TME.

A. Graph showing pancreas weight for Pin1^+/+, Pin1^−/−, KPC;Pin1^+/+, and KPC;Pin1^−/− mice at 5 months of age. Mean +/− SD is shown. One-way ANOVA followed by Tukey’s test for multiple comparisons was performed. B, C. Representative images of pancreas sections from KPC;Pin1^+/+ and KPC;Pin1^−/− mice stained for H&E (B) or Trichrome (C). Scale bars represent 100 μm. Quantitation of Trichrome staining is shown to the right. Mean +/− SD is shown. An unpaired t-test was performed. D-F. Representative IF images of pancreas sections from KPC;Pin1^+/+ and KPC;Pin1^−/− mice. In D, sections were stained for KRT8/18 (red), PDPN (magenta), and αSMA (turquoise). Left images are shown without PDPN staining. Scale bars represent 100 μm. Quantification of the αSMA stain in PDPN+ KRT8/18- cells is shown below. Mean +/− SD is shown for n=10 or 8 for KPC;Pin1^+/+ or KPC;Pin1^−/− mice, respectively. In E, sections are stained for KRT8/18 (orange), PDPN (turquoise), and pSTAT3 (magenta). Scale bars represent 100 μm, boxed inset without PDPN staining is shown to the right. Quantification of the pSTAT3 stain in PDPN+ KRT8/18- cells is shown below. Mean +/− SD is shown for n=3 mice. In F, sections were stained for KRT8/18 (red), PDPN (turquoise), and CD74 (magenta). Scale bars represent 100 μm, boxed inset is shown to the right. Quantification of the CD74 stain in PDPN+ KRT8/18- cells is shown below. Mean +/− SD is shown for n=4 mice. For D, E. and F, unpaired t-tests were performed. G. Representative images of human PDAC tissue on a TMA stained for PIN1 (red), αSMA (cyan), KRT19 and CDH1 (both in green). Scale bars represent 244 μm in left panel and 65 μm in all others. H. Pearson correlation is shown between fibroblast PIN1 and αSMA expression, n = 34 patients. I. Pearson correlations between fibroblast PIN1 expression and the percent of cells classified as fibroblasts (KRT19^-, CD31^-, CD45^- ; left), epithelial cells (KRT19⁺, CD31^-, CD45^- ; middle panel), or endothelial cells (KRT19^-, CD31⁺, CD45^- ; right panel). Whole figure: * indicates p<0.05, ** indicates p<0.01.

Inhibition or loss of PIN1 in PSCs alters the cellular response to TGF-β

We hypothesized that PIN1 may play a direct role in pancreatic fibroblasts, controlling the cellular response to extrinsic signals that drive the myofibroblast state. TGF-β has been identified as a major regulator of myofibroblast differentiation in pancreatic cancer [5, 9]. We acquired primary human PSCs (hPSCs) from a commercial vendor and confirmed fibroblast marker expression in these cells (Supplementary Figure S2A). To test for a role for PIN1 in pancreatic stellate cells, we treated hPSCs with ATRA, KPT-6566, or Sulfopin, three inhibitors of PIN1 [53–55], and found that PIN1 inhibition slowed the growth of PSCs in a dose-dependent manner (Supplementary Figures S2B, C, D). We then compared the TGF-β response of cells pretreated for 48 hours with PIN1 inhibitors to that of control cells. PIN1 inhibition significantly reduced the TGF-β-driven increase in protein markers of the myofibroblast state, including αSMA and Palladin (PALLD), as detected with the 1E6 antibody that recognizes isoforms expressed in pro-tumor myofibroblasts in PDAC [56] (Figure 2A, B). This result was consistent across PSCs from distinct donors (Supplementary Figure S2E, F). Moreover, similar results were observed in an immortalized mouse PSC cell line treated with ATRA or KPT-6566 prior to TGF-β treatment (Supplementary Figure S2G, H).

Figure 2: Inhibition or loss of PIN1 alters the cellular response to TGFβ.

A. Western blots of lysates from hPSCs treated with PIN1 inhibitors or DMSO (vehicle control) +/− TGFβ. B. Graphs showing quantitation of αSMA, PALLD, and PIN1 from Western blots, normalized to GAPDH levels and shown as the fold change over control cells. Mean +/− SD is shown for 3–5 independent experiments. C. Western blots of lysates from control (shSCR) or PIN1 knockdown (shPIN1) PSCs treated +/− TGFβ. D. Graphs showing quantitation of αSMA, PALLD, and PIN1 from Western blots, normalized to GAPDH levels. Shown is mean +/− SD for 4 independent experiments. E. Western blots of lysates from control (shSCR) or PIN1 knockdown (shPIN1) CAFs treated +/− TGFβ. GAPDH is a loading control. F. Graphs showing quantitation of αSMA, PALLD, and PIN1 from Western blots, normalized to GAPDH levels. Shown is mean +/− SD for 3 independent experiments. Whole figure: Two-way ANOVA followed by Tukey’s multiple comparisons tests were performed, * indicates p<0.05, ** indicates p<0.01, *** indicates p<0.001; **** indicates p<0.0001.

To complement the inhibitor studies, we also used lentiviral shRNA to generate a pool of hPSCs with stable PIN1 knockdown (shPIN1) and compared these to control (shSCR) PSCs. Similar to the cells treated with PIN1 inhibitors, we observed slower growth of shPIN1 PSCs (Supplementary Figure S3A, B), as well as a reduced response to TGF-β treatment as measured by αSMA and PALLD expression (Figure 2C, D). BODIPY staining to assess loss of lipid droplets also revealed reduced response of the shPIN1 cells to TGF-β, and notably, PIN1 knockdown PSCs showed a markedly higher baseline level of BODIPY staining (Supplementary Figure S3C, D, E). In contrast to PIN1 knockdown, overexpression of PIN1 in hPSCs drove a correlated increase in αSMA expression, even in the absence of TGF-β treatment (Supplementary Figure S3F, G). Finally, we tested the impact of PIN1 knockdown on TGF-β response in a patient-derived CAF cell line. While the CAFs exhibited a higher baseline level of myofibroblast marker expression, they still responded to TGF-β and both the baseline levels and the TGF-β-dependent response were reduced with PIN1 knockdown (Figure 2E, F). Together, these results demonstrate that PIN1 contributes to TGF-β-mediated activation of PSCs into a myofibroblast differentiation state.

Inhibition or loss of PIN1 in PSCs alters the transcriptional response to TGF-β

To better understand the global pathways affected by PIN1 loss in PSCs, we performed RNA-seq on shSCR and shPIN1 PSCs treated with or without TGF-β (Supplementary Dataset 1). We performed a factored DESeq2 analysis to identify the significantly differentially expressed genes (DEGs) in control shSCR PSCs between the untreated and TGF-β-treated conditions (Supplementary Dataset 2). We then assessed expression of those DEGs across all five replicates. Consistent with the above protein expression data, we found that loss of PIN1 mitigated the transcriptional response of hPSCs to TGF-β, with a decreased response for both upregulated and downregulated gene sets (Figure 3A, B, Supplementary Dataset 3). Expression levels of TGFBR1, TGFBR2, and TGFBR3 did not significantly change at baseline in response to knockdown of PIN1 (Supplementary Figure S4A). To determine how PIN1 loss specifically affects fibroblast activation states, we assessed expression of described marker genes for myCAF, iCAF, and apCAF states [5, 6, 9]. As expected, control cells exhibited strong myofibroblast gene expression in response to TGF-β, which was inhibited in the shPIN1 condition (Figure 3C, D, Supplementary Dataset 4). We found no consistent change in iCAF marker genes, with the exception of LIF, which was strongly upregulated by TGF-β in this system. At baseline, prior to TGF-β treatment, the shPIN1 PSCs did appear to have elevated expression of some apCAF marker genes, albeit without MHCII expression (Figure 3C, D). Analysis of RNA-seq following PIN1 inhibitor treatment of PSCs with and without TGF-β treatment showed more subtle changes overall as compared to knockdown of PIN1, suggesting that ATRA and KPT-6566 are impacting only a subset of PIN1 function (Supplementary Figure S4B and Supplementary Datasets 5, 6, and 7). Nevertheless, a consistent decrease in myCAF markers was observed in RNA-seq of hPSCs treated with the PIN1 inhibitors (Supplementary Figure S4C, D and Supplementary Dataset 8).

Figure 3. PIN1 is required for normal induction of TGFβ-driven transcriptional programs.

A. Heatmap showing expression across four conditions for 1220 genes differentially expressed (p<0.01) between shSCR untreated and TGFβ-treated samples. B. Graphs of z-scores in the four conditions for the genes up- or down-regulated by TGFβ. C. Heatmap showing expression across four conditions for marker genes of myCAF, iCAF, and apCAF states. D. Graphs of expression of candidate marker genes of myCAF (top row), iCAF (middle row), and apCAF (bottom row) states. E. Venn diagrams showing overlap of the genes up- or down-regulated by TGFB in PSCs with the NABA_MATRISOME, NABA_ECM_AFFILIATED, and CellChatDB Secreted Signals genesets. F-K. Heatmaps (F, H, J) and graphs of z-scores (G, I, K) of expression across four conditions for the overlapping genes from each of those three genesets as shown in (E). Whole figure: One-way ANOVA followed by Tukey’s multiple comparisons tests were performed, * indicates p<0.05, ** indicates p<0.01, *** indicates p<0.001; **** indicates p<0.0001.

To understand how PIN1 loss in PSCs influences the signals to surrounding cells in the microenvironment, we identified the TGF-β-dependent matrix proteins and secreted ligands that were regulated in hPSCs by TGF-β by overlapping the set of genes differentially expressed in shSCR cells with and without TGF-β treatment with published genesets of the matrisome and ECM affiliated genes, as well as a genelist of secreted signals from CellChatDB [57, 58] (Figure 3E). PSCs with PIN1 knockdown had a significantly decreased enrichment of the TGF-β-driven matrisome, ECM, and secreted signal genesets (Figure 3F-K and Supplementary Datasets 9, 10, and 11). Together, these results demonstrate that PIN1 is necessary for the expression of TGF-β-mediated gene programs controlling the myofibroblast state including the expression of secreted matrix and signaling factors that drive the function of those fibroblasts.

PIN1 loss inhibits epigenetic remodeling in response to TGF-β activation

Our previous work in mouse embryonic fibroblasts (MEFs), identified a critical role for PIN1 in the epigenetic and phenotypic response to serum stimulation. Specifically, we found that Pin1^−/− MEFs displayed abrogated chromatin accessibility changes in response to serum stimulation [59]. We asked whether PIN1 also regulated chromatin accessibility changes in response to TGF-β in PSCs. We performed single-cell combinatorial indexing (sci)ATAC-seq to analyze chromatin accessibility on a single-cell level [41] in shSCR and shPIN1 PSCs treated with TGF-β for 8 hours. We applied cisTopic [47] to identify co-regulated chromatin regions (Topics) that defined cell states. This produced a cell × topic matrix showing the enrichment of each topic in every cell (Figure 4A), which was then visualized in a two-dimensional UMAP (Figure 4B and Supplementary Figure S5A). To determine the global magnitude of chromatin changes that occurred upon TGF-β treatment with PIN1 intact or knocked down, we calculated the Euclidian distance within the cell × topic matrix between all untreated or TGF-β treated cells within each group. This revealed a significant reduction in the distribution of cell-cell distances between treatment groups in the shPIN1 cells, indicating a muted global chromatin response (Figure 4C). We observed that specific topics (Topics 2, 10, and 13) were more enriched (sites more accessible) upon TGF-β treatment in control shSCR cells as compared to shPIN1 cells (Figure 4D, E). Other groups of topics were found to be equally enriched in both control and shPIN1 cells upon TGF-β treatment (e.g. Topic 15), enriched in shSCR versus shPIN1 cells regardless of treatment (e.g. Topic 11), or enriched in shPIN1 versus shSCR cells regardless of treatment (e.g. Topic 23) (Figure 4A, Supplementary Figure S5B).

Figure 4. PIN1 loss inhibits epigenetic remodeling in response to TGFβ.

A. Biclustering of cisTopic cell by topic matrix from sciATAC-seq data for shSCR and shPIN1 PSCs treated +/− TGFβ for 8 hours. Heatmap color indicates topic weight for each cell (columns). B. Projection of single cells from (A) through UMAP. C. Distribution of Euclidean distances between untreated and TGFβ treated shSCR PSCs (gray) or shPIN1 PSCs (red). Distances were calculated with respect to topic weights (see Methods). Significance was calculated by a t-test. D. UMAP from (B) colored by Topic 2 weight values for each cell. E. Distribution of topic weights for shSCR TGFβ enriched topics (2, 10, and 13). F. Top unique motifs identified by HOMER as being overrepresented in topics associated with the shSCR TGFβ condition (topics shown in E). G, H. GSEA analysis of RNA-seq showing top gene sets (p<0.05) upregulated in either shSCR (G) or shPIN1 (H) PSCs when assessing the C6 (oncogenic signatures) database. I. Distribution of Chromvar global motif deviation scores for two representative top motifs (from F). Whole figure: For all violin plots, internal boxplots show the 25 −75 percentile, and significance was calculated with a Wilcoxon rank sum test with Bonferroni adjustment, **** indicates p<0.0001.

We assessed enrichment of transcription factor motifs within topic-defined peak sets to identify factors that may contribute to the altered epigenetic response. SMAD transcription factor motifs were enriched in all TGF-β-induced topics, and a global analysis of motif accessibility using ChromVAR [50] supported increased accessibility of SMAD motifs in both shSCR and shPIN1 PSCs in response to TGF-β treatment (Supplementary Figure S5C). In contrast, we identified 22 unique motifs that were enriched in the TGF-β-induced topics in shSCR PSCs (Topics 2, 10, and 13), but not in the topics induced equally by TGF-β in shSCR and shPIN1 PSCs (Topics 4, 15, and 20; see Methods and Supplementary Dataset 12). Motifs that were open in response to TGF-β stimulation in a manner dependent on PIN1 included those that bind TCF21, TCF12, MYC, E2F3, TWIST, and many FOX factors (Figure 4F). Gene Set Enrichment Analysis (GSEA) of RNA-seq from shSCR and shPIN1 PSCs showed that pathways regulated by some of these factors, such as MYC and E2F3, were also enriched in shSCR PSCs (Figure 4G, H, Supplementary Figure S5D, E, and Supplementary Datasets 13 and 14). Global analysis of the top unique motifs in Fig. 4F showed that some (e.g. TCF12 and TWIST2) exhibited a genome-wide accessibility difference between shSCR vs shPIN1 cells in response to TGF-β (Figure 4I), while others (e.g. MYC) had similar accessibility across samples genome-wide, despite differential enrichment in the topics (Supplementary Figure S5F). Together this suggests that specific subsets of TGF-β-induced chromatin regulatory networks are dependent on PIN1 for epigenetic remodeling.

Fibroblast PIN1 supports paracrine HGF signaling

To further understand how loss of PIN1 affected PSCs at baseline, we performed a paired analysis with DESeq2 comparing the expression profiles of the untreated shSCR and shPIN1 PSCs (Supplementary dataset 15) (Figure 5A). We identified multiple secreted factors that were differentially regulated in shPIN1 vs control cells. For example, COL15A1, TMEFF2, IGFBP5, and HGF were significantly decreased in shPIN1 PSCs (Figure 5A). COL15A1 is a non-fibrillar multiplexin collagen that associates with basement membranes and can help control matrix organization [60]. In contrast, PDGFB, SAA1, and SPINT2 were significantly increased in shPIN1 PSCs (Figure 5A, Supplementary Dataset 15).

Figure 5. Fibroblast PIN1 supports paracrine HGF signaling.

A. Volcano plot showing log(2)FC and -log10(p value) for DEGs from DESeq2 analysis of RNA-seq on shSCR and shPIN1 PSCs. B, C. Graphs showing HGF mRNA expression in shSCR and shPIN1 hPSCs (B) or hPSCs treated with PIN1 inhibitors (C). Shown is mean +/− SD for three independent experiments; One-way ANOVA was performed with Dunnett’s multiple comparisons test. D. Graph showing HGF protein levels detected by ELISA in conditioned media (CM) from shSCR or shPIN1 PSCs. Shown is mean +/− SD of three independent replicates. An unpaired t-test with Welch’s correction was performed. E. Western blot for phospho- and total MET (pMET and tMET, respectively) in pancreatic cancer lines treated for 15 min with control media, 100 ng/mL HGF, shSCR CM, or shPIN1 CM. F. Graph showing the ratio of pMET/tMET from Western blots as in (E). Shown is mean +/− SD of 3 (ASPC1) or 4 (CFPAC1, HPAFII) independent replicates. Unpaired t-tests were performed to compare CM conditions in each cell line. G, H. Representative phase images (G) or IF images (H) of HPAFII cells treated for 48 hours with conditions described in (E). IF images are stained for E-cadherin (green), DAPI (blue). Scale bars represent 200 μm (phase images) or 50 μm (IF). I. Graph showing quantitation of cell area in HPAFII cells following treatments. Shown is mean +/− SD of 3 independent experiments. A one-way ANOVA was performed, with Tukey’s test for multiple comparisons. J. Graph showing tumor size (mean +/− SD) over time for orthotopic xenografts of HPAFII cells alone or co-xenografted with either shSCR or shPIN1 PSCs. Area under the curve was determined for individual tumors, and a One-way ANOVA, followed by Tukey’s test was performed. K. Graph of final tumor weights (mean +/− SD) from mice in (I). A One-way ANOVA, followed by Tukey’s test was performed. L, M. Representative IHC images (L) and quantitation (M) for pMET staining of tumors from mice in (J). Scale bars represent 50 μm. Mean +/− SD of expression shown for 5 tumors in each group. A One-way ANOVA was performed, with Tukey’s test for multiple comparisons. N, O. Representative IHC images (N) and quantitation (O) for pMET staining of KPC;Pin1^+/+ or KPC;Pin1^−/− tumors. Scale bars represent 100 μm. Mean +/− SD of n=8 (KPC;Pin1^+/+) or n=6 (KPC;Pin1^−/−) tumors. An unpaired t-test was performed. Whole figure: * indicates p<0.05, ** indicates p<0.01; *** indicates p<0.001; **** indicates p<0.0001.

We initially focused on HGF as a secreted factor that has been well-described to function in paracrine signaling to cancer cells [61]. In addition to HGF itself being consistently upregulated in shSCR vs. shPIN1 PSCs, a secreted serine protease inhibitor, SPINT2, that prevents extracellular activation of HGF [62] was upregulated in shPIN1 PSCs (Supplementary dataset 15). This indicated that shPIN1 PSCs not only express reduced HGF but may also prevent HGF signaling through upregulation of an endogenous inhibitor. We used qRT-PCR to confirm the changes in HGF expression in hPSCs with knockdown of PIN1 (Figure 5B) and found HGF expression to also be decreased following hPSC treatment with PIN1 inhibitors (Figure 5C). Finally, we validated that less HGF protein was present in conditioned media from shPIN1 PSCs as compared to control shSCR PSCs (Figure 5D and Supplementary Figure S6A). In the ELISA assay, we also observed less HGF in the control shSCR PSCs as compared to the parental PSCs, perhaps due to decreased HGF with additional passages in culture during selection of the stable pools (Supplementary Figure S6B).

HGF secreted from CAFs has previously been shown to contribute to proliferation and migration of pancreatic cancer cells in vitro and in vivo [61]. To test whether the observed changes in HGF expression affect paracrine signaling between PSCs and tumor cells, we transferred conditioned media (CM) from shSCR or shPIN1 hPSCs onto human pancreatic cancer cell lines ASPC1, CFPAC1, and HPAFII. We used recombinant HGF as a positive control. We found that shSCR CM, but not shPIN1 CM was sufficient to drive phosphorylation of the HGF receptor, cMET (Figure 5E, F). HPAFII and CFPAC1 cells treated with conditioned media showed a striking morphological change in the cells treated with shSCR CM, representative of an epithelial-mesenchymal transition (EMT) and mimicking the phenotype of HGF-treated cells (Figure 5G, Supplementary Figure S6C, D). We quantified HPAFII cell area as a measure of this altered morphology and found shSCR CM to mimic HGF in driving increased cell area, while shPIN1 CM did not drive this change (Figure 5I). We also observed by immunofluorescence (IF) a downregulation of E-cadherin and a change in its localization away from cell-cell contacts in cells treated with shSCR CM, again mimicking the phenotype of recombinant HGF (Figure 5H, Supplementary Figure S6E).

To determine whether HGF signaling was suppressed in the presence of fibroblasts with low PIN1 expression in vivo, we orthotopically xenografted HPAFII cells alone, or in combination with shSCR or shPIN1 hPSCs into immunocompromised mice. Inclusion of shSCR, but not shPIN1 PSCs, increased growth of the tumors in vivo (Figure 5J, K), and we observed significantly decreased pMET staining in tumors when the tumor cells were xenografted with shPIN1 PSCs as compared to control shSCR PSCs (Figure 5L, M). Consistent results were observed in the autochthonous KPC mouse model, where pMET staining was decreased in the epithelial cells of KPC;Pin1^−/− as compared to KPC;Pin1^+/+ tumors (Figure 5N, O). Total MET staining was not altered between these conditions (Supplementary Figure S6F). Together, these data demonstrate a critical role for PIN1 in fibroblasts to mediate HGF signaling to cancer cells in support of tumorigenic phenotypes.

A subset of pancreatic cancer cell lines shows reduced growth in Pin1^−/− hosts, correlating with the cellular response to HGF signaling

To directly test how loss of PIN1 in the microenvironment impacts tumor growth, we orthotopically transplanted KPC cell lines into syngeneic Pin1^+/+ or Pin1^−/− hosts and measured tumor volume by ultrasound. Two KPC cell lines, KPC8060 and KPC8069, had significantly reduced tumor growth over time when transplanted into Pin1^−/− vs. Pin1^+/+ hosts, resulting in significantly smaller tumors at endpoint (Figure 6A-F, Supplementary Figure S7A-C). Trichrome staining was decreased in the KPC8060 and KPC8069 allografts in Pin1^−/− vs. Pin1^+/+ hosts, consistent with decreased myofibroblast differentiation in the PIN1 low setting (Supplementary Figure S7D, E). In contrast, KPC7107 allograft growth was not altered in Pin1^+/+ vs. Pin1^−/− hosts (Figure 6G-I), and in this model we also did not observe a change in Trichrome staining (Supplementary Figure S7F).

Figure 6. A subset of pancreatic cancer cells have reduced growth in Pin1^−/− hosts.

A, D, G. Graphs showing the tumor volume over time for KPC8060 (A), KPC8069 (D), or KPC7107 (G) cancer cell orthotopic allografts in Pin1^+/+ or Pin1^−/− hosts. Shown is a representative experiment with mean +/− SEM. Unpaired t-tests were performed on the AUC for growth of individual tumors. B, E, H. Graphs showing the final relative pancreas weight for mice from A, D, and G, respectively. Shown is mean +/− SD; unpaired t-tests were performed. C, F, I. Representative H&E images of allografts of KPC8060 (C), KPC8069 (F) and KPC7107 (I). Scale bars represent 1000 μm (left panels) or 200 μm (right panels). Whole figure: * indicates p<0.05, ** indicates p<0.01, *** indicates p<0.001.

We hypothesized that changes in paracrine signaling, specifically changes in HGF levels, from the knockout host environment may differentially affect the KPC7107 vs KPC8060 and KPC8069 tumor cell lines. We grew spheroids of the three KPC cell lines in vitro with and without 100 ng/mL HGF treatment for 6 days and found that both KPC8060 and KPC8069 cell spheroids increased growth in response to HGF treatment. KPC7107 cell spheroids, in contrast, did not significantly increase in size with the addition of HGF (Figure 7A, B, C). We next tested whether lower doses of HGF would be sufficient to affect spheroid growth. We found doses of 5, 12.5, and 25 ng/mL HGF to be sufficient to increase growth of the KPC8060 and KPC8069 spheroids, but once again these did not alter growth of the KPC7107 spheroids (Figure 7D, Supplementary Figure S8A-F). We confirmed that all three cell lines expressed the c-MET receptor, and that it could be activated by HGF treatment. While KPC7107 cells did express a lower level of total c-MET, these cells exhibited similar ratios of pMET/tMET following HGF treatment (Supplementary Figure S8G-I). These data indicated that HGF signaling was sufficient to increase the growth of the lines that were dependent on microenvironmental PIN1. We next stained the KPC cell line allograft tissues from both Pin1^+/+ and Pin1^−/− hosts for pMET. Consistent with less HGF production in Pin1^−/− fibroblasts, we observed a significant reduction in pMET staining in the KPC8060 and KPC8069 allografts in Pin1^−/− hosts compared to Pin1^+/+ hosts (Figure 7E, F). Levels of pMET staining in the KPC7107 allografts were not significantly different (Figure 7G), perhaps indicating that, similar to the Trichrome staining in the KPC7107 allografts, the KPC7107 cells are able to differentially activate the Pin1^−/− microenvironment, allowing it to support tumor growth.

Figure 7. Cellular response to HGF signaling correlates with in vivo dependence on microenvironmental PIN1.

A. Representative phase images of spheroids of KPC cell lines as indicated grown in the presence or absence of exogenous 100 ng/mL HGF. B. Graphs showing the spheroid area over time for KPC spheroids as shown in A. Shown is mean +/− SEM for at least three independent experiments with at least four technical replicates each. Unpaired t-tests were performed on AUCs of each independent experiment. C. Graphs showing fold change in spheroid area at endpoint for spheroids as in A,B. Shown are individual values and median. Unpaired t-tests were performed. D. Graphs showing fold change in spheroid area at endpoint for spheroids grown with lower concentrations of HGF as shown for three independent experiments. One-way ANOVA was performed for each cell line. E, F, G. Representative images of pMET staining (left) and quantification of staining (right) of KPC allografts in Pin1^+/+ or Pin1^−/− hosts. Percent of area that is pMET+ is shown for individual tumors. Unpaired t-tests with Welch’s correction were performed. H-M. Graphs showing tumor volume over time (H, J, L) and final tumor weights (I, K, M) of KPC8060, KPC8069, or KPC7107 orthotopic allografts into wildtype mice that were treated with vehicle or PHA-665752, a MET inhibitor. For tumor volumes, mean and SEM are shown; AUC was calculated for individual tumors and unpaired t-tests performed for significance. For tumor weights, medians and individual values are shown; unpaired t-tests were performed. Whole figure: * indicates p<0.05, ** indicates p<0.01, *** indicates p<0.001; **** indicates p<0.0001.

Finally, to further examine the requirement for HGF/MET signaling for in vivo growth of the KPC cell line allografts, we performed orthotopic injections of KPC8060, KPC8069, and KPC7107 cells into wildtype mice, and then treated mice with either the MET inhibitor PHA-665752 [63] or a vehicle control. While MET inhibition significantly decreased growth of the KPC8060 tumors (Figure 7H, I), it did not affect growth of KPC8069 or KPC7107 tumors in vivo (Figure 7J-M). These data indicate that while paracrine HGF signaling is sufficient to support in vitro growth of both of the pancreatic cancer cells that are sensitive to loss of PIN1 in the microenvironment, it is only necessary in vivo for the growth of one of these lines. This may indicate that there are changes to the Pin1^−/− tumor microenvironment in addition to reduced HGF that impact growth of the KPC8069 tumors. Together, our data show that loss of PIN1 in the microenvironment alters fibroblast state driving changes to the extracellular matrix and the paracrine signals. Microenvironmental PIN1 is necessary for normal growth of some pancreatic cancer cell lines, and understanding differences in response to its inhibition in the environment may help reveal vulnerabilities of heterogeneous tumor populations.

Discussion

PIN1 has previously been shown to be critical for tumor development and progression in mammary and other tumor types [23, 24]. Mechanistically, PIN1 can contribute to tumor cell intrinsic phenotypes including proliferation, migration, and metabolic changes through interaction with a broad range of target proteins [25–27]. This function, as well as the fact that the loss of PIN1 appears to be well-tolerated and associated with resistance to oncogene-induced tumorigenesis in vivo, has led to much interest in targeting PIN1 in cancer. While a cancer cell-intrinsic role of PIN1 is well-established, here we demonstrate that PIN1 also regulates fibroblast phenotypic state and function, and loss or inhibition of PIN1 in the TME can profoundly impact tumor phenotypes and tumor development. Our data suggest, though, that the influence of fibroblast PIN1 expression is different for distinct pancreatic cancer cell lines, suggesting that better understanding of the crosstalk required by heterogeneous cancer cells for growth may give insight into tumor vulnerabilities.

We show a substantial loss of αSMA in mouse and human tumors that have no or low expression of PIN1 in fibroblasts. This work is consistent with a recently published study that showed decreased αSMA expression in mouse PDAC models upon treatment with PIN1 inhibitors [28]. We have extended this work to study the impact of loss of PIN1 on the TGF-β-driven transition from PSCs to myofibroblasts. We interrogate the transcriptional and epigenetic responses to TGF-β. We show a broad suppression of TGF-β-regulated gene programs, including a specific loss of TGF-β-dependent myofibroblast, matrisome, ECM, and secreted signaling factor gene programs. Future work can characterize specific changes to the tumor ECM in the absence of PIN1 expression in terms of composition, including both fibrillar and non-fibrillar collagens that have altered expression, as well as organization.

We also identify sets of co-regulated gene regulatory elements (Topics) that are more accessible in shSCR as compared to shPIN1 hPSCs in response to TGF-β. The transcription factor motifs enriched in these Topics include motifs (e.g. TCF12 and TWIST2) that are observed in genome-wide analysis to be globally dependent on PIN1 for response to TGF-β whereas others, e.g. MYC motifs, are not. Previously, we identified PIN1 as a critical regulator of MYC that controls transcription of a subset of MYC target genes [35, 59]. Our data here suggests that there is a subset of MYC motifs that are TGF-β-modulated in a PIN1-dependent manner that are represented in Topics 2, 10, and 13. This is consistent with MYC activity, as assessed by GSEA, being significantly enriched in the control shSCR PSCs as well. Finally, many transcription factor motifs whose accessibility was driven by TGF-β in a PIN1-dependent manner have been previously linked, either positively or negatively, to fibrosis. For example, while GLI1 and FOXM1 have been shown to contribute to fibrotic phenotypes [64–66], TCF21, FOXF1, and NRF2 all appear to inhibit fibrotic programs [67–69]. Precisely how PIN1 alters accessibility and transcriptional activity at specific loci to coordinate global changes in fibroblast phenotype will require additional study.

Pancreatic stellate cells are resident fibroblasts in the pancreas and have been shown to be one of the cellular contributors to CAF development and the dense stromal reaction that characterizes PDAC [12, 14]. Most of the in vitro work shown here assesses a role for PIN1 in hPSCs, controlling baseline phenotypes and response to TGF-β in one cell of origin of PDAC CAFs. The fact that αSMA is reduced across human and mouse tumors with low PIN1 expression suggests that PIN1 likely controls myofibroblast state in CAFs from multiple distinct cells of origin, but future work will be needed to confirm this directly. TGF-β has been shown to be an almost universal driver of myofibroblast phenotypes in many types of tumors as well as in other fibrotic conditions [70]. Indeed, PIN1 has also previously been shown to contribute to pulmonary, renal, and cardiac fibrosis, with loss or inhibition of PIN1 attenuating the fibrotic response [71, 72].

Our work here suggests that PIN1 contributes to the establishment of the desmoplastic tumor microenvironment that characterizes aggressive PDAC. Intriguingly, the decreased ECM observed in PIN1 null tumors may be an indicator for better therapeutic delivery, suggesting PIN1 inhibitor combinations with standard of care combination chemotherapy may provide increased efficacy. Further work, however, will be needed to establish the differences in phenotype resulting from knockdown and inhibition of PIN1. While both approaches affected myCAF marker expression in response to TGF-β, the use of either of the two PIN1 inhibitors, ATRA or KPT-6566, showed less of a global influence on TGF-β gene programs as compared to PIN1 knockdown. This could be due to the dose or time point chosen for study here or to different mechanistic effects of loss versus inhibition of PIN1. In addition, the inhibitors also appear to differentially affect some baseline transcriptional profiles in PSCs, which could be due to differences in off-target effects between these methods. KPT-6566 and Sulfopin have been described as covalent inhibitors that bind to the catalytic site of PIN1 [54, 55]. PIN1 has been shown to be an important target of ATRA, but ATRA also activates RAR-dependent gene programs that can influence PSC state [53, 73]. While other reports have shown loss of PIN1 protein upon PIN1 inhibitor treatment [28, 53–55], this response was not consistently observed in our system.

Finally, our work highlights a role for PIN1 in regulating paracrine signaling from fibroblasts to cancer cells. We show that PIN1 deficient fibroblasts have decreased expression and secretion of HGF, a factor that contributes to tumor cell proliferation and migration. Previous therapeutic studies targeting HGF and its receptor, cMET, in pancreatic cancer together with gemcitabine treatment have shown promise in slowing tumor growth and decreasing metastasis [61]. Our work shows heterogeneity of KPC PDAC cell lines in response to HGF. Despite the fact that all three lines tested express cMET, only two of the lines, KPC8060 and KPC8069, responded to HGF treatment in vitro with increased proliferation. These same two lines were dependent on microenvironmental PIN1 for growth in vivo, while the KPC7107 line did not depend on PIN1 in the host for growth in vivo and did not respond to exogenous HGF in vitro. Finally, we show that MET inhibition slowed the growth of KPC8060 in vivo, while it did not affect KPC8069 or KPC7107 allograft growth. Further research into other factors that are altered in the Pin1^−/− setting that also contribute to KPC8069 differential growth alone, or in combination with HGF, will be needed.

Together, our work indicates that targeting of PIN1 alters fibroblast phenotypes, and this may provide anti-tumor effects for some tumor populations. Future work will focus on further delineating the mechanisms by which PIN1 regulates myofibroblast phenotypes and extending these studies to better understand the heterogeneity of cancer lines in response to microenvironmental PIN1 inhibition. Overall, our findings provide insight into the multifaceted effects of PIN1 inhibition that could provide a rational basis for combination therapies that would benefit from fibroblast reprogramming.

Statement of Significance:

PIN1 plays a critical role in the response of pancreatic stellate cells to TGF-β and can be targeted to attenuate myofibroblast activation and pro-tumor crosstalk to suppress pancreatic cancer progression.