ATRA modulates mechanical activation of TGF-β by pancreatic stellate cells

The hallmark of pancreatic ductal adenocarcinoma (PDAC) is abundant desmoplasia, which is orchestrated by pancreatic stellate cells (PSCs) and accounts for the majority of the stroma surrounding the tumour. Healthy PSCs are quiescent, but upon activation during disease progression, they adopt a myofibroblast-contractile phenotype and secrete and concomitantly reorganise the stiff extracellular matrix (ECM). Transforming growth factor β (TGF-β) is a potent activator of PSCs, and its activation requires spatiotemporal organisation of cellular and extracellular cues to liberate it from an inactive complex with latent TGF-β binding protein (LTBP). Here we study the mechanical activation of TGF-β by PSCs in vitro by investigating LTBP-1 organisation with fibrillar fibronectin and show that all trans-retinoic acid (ATRA), which induces PSC quiescence, down-regulates the ability of PSCs to mechanically organise LTBP-1 and activate TGF-β through a mechanism involving myosin II dependent contractility. Therefore, ATRA inhibits the ability of PSCs to mechanically release active TGF-β, which might otherwise act in an autocrine manner to sustain PSCs in an active state and a tumour-favouring stiff microenvironment.


Fibronectin splice variants fibronectin-EDA and fibronectin-EDB are expressed by PSCs, and LTBP-1 alignment requires fibronectin but is unaffected by fibronectin levels in the ECM.
Fibronectin is alternatively spliced in wound healing and cancer stroma 29 ; for instance, the fibronectin extra domain-A (EDA) isoform is overexpressed in liver tumour vasculature 30 . To understand whether a similar alternative splicing mechanism occurs in PSCs, we examined the expression of fibronectin-EDA, which is the major isoform expressed in breast cancer, and fibronectin-EDB, which has been implicated in pancreatic cancer microenvironment but of an unknown cellular origin 31 . Quantitative PCR targeting fibronectin-EDA or fibronectin-EDB mRNA showed that both variants are expressed by PSCs and expression levels were significantly reduced upon ATRA treatment (p < 0.05, p < 0.01 respectively) (Fig. 2a).
To further dissect whether LTBP-1 fibril formation is dependent on the presence of fibrillar fibronectin, fibronectin expression was knocked down by siRNA (50 nM) in control PSCs prior to culturing the cells for 48 hours on LTBP-1 rich ECM. As a result of fibronectin knockdown, LTBP-1/fibronectin co-localisation and fibril thickness were significantly decreased (Fig. 2b,c) (p < 0.001 for both measurements). This demonstrates that LTBP-1 fibril formation by PSCs is dependent on fibronectin guidance, and in the absence of fibronectin, LTBP-1 is not organised in fibrils.
To exclude the possibility that the reduction in fibronectin/LTBP-1 co-localisation and fibril thickness with ATRA treatment (Fig. 1) might be simply due to differences in expression of extracellular fibronectin (Fig. 2a), control or ATRA treated PSCs were cultured on LTBP-1 rich matrices for 48 hours in the presence of excess fibronectin (40 μ g/cm 2 ). Despite excess fibronectin addition to culture media, there was a persistent significant decrease in fibronectin/LTBP-1 co-localisation with ATRA treatment (p < 0.001) as observed previously with endogenous fibronectin secretion (Fig. 2c). Therefore, PSC-mediated LTBP-1 fibril organisation onto fibrillar fibronectin depends on the presence of fibronectin but is insensitive to levels of fibronectin in the ECM. Taken together our results indicate that there is a threshold of fibronectin concentration that is needed for LTBP-1/ fibronectin alignment by PSCs, but provided this concentration is present, other cellular mechanisms in PSCs control LTBP-1/fibronectin alignment.

LTBP-1 fibril alignment onto fibrillar fibronectin by PSCs is an actomyosin dependent mechanism.
Because fibronectin/LTBP-1 co-localisation was still significantly down-regulated in ATRA treated PSCs compared to control PSCs even when excess fibronectin is added to culture media, we investigated the role of biomechanical reorganisation of ECM by PSCs in vitro. Actomyosin contraction-mediated cellular tension applied to ECM is required for LTBP-1 alignment 23,24,32 . ATRA has been previously shown to reduce cell stiffness (unpublished data) and cell contractility markers α -SMA and vimentin 25 . Similarly, ATRA reduces expression of cell contractility marker vimentin in PSCs seeded on LTBP-1 rich matrices (Fig. S2). Therefore, to understand the association between PSC contractility and LTBP-1 organisation in our system, control or ATRA treated PSCs were seeded on LTBP-1 rich matrices as previous. PSCs were incubated for 16 hours to allow cell attachment and spreading after which, 20 μ M blebbistatin (BBI), which blocks myosin II ATPase activity 33 , or DMSO vehicle was added. Both fibronectin/LTBP-1 co-localisation and the thickness of co-localised fibrils were significantly reduced in untreated PSCs when exposed to BBI (p < 0.001, p < 0.05 respectively) (Fig. 3), indicating that LTBP-1 incorporation onto fibrillar fibronectin is dependent on actomyosin contraction. No significant differences were observed when PSCs were treated with both ATRA and BBI in comparison to ATRA alone (Fig. 3), which indicates that ATRA mainly acts through a reduction of actomyosin contraction. Therefore, inhibition of cell contractility by ATRA or BBI inhibits biomechanical organisation of LTBP-1 and fibronectin.
PSCs act on LTBP-1 through a β1-integrin mediated mechanism. LTBP-1 carries an arginyl-glycylaspartic acid (RGD) consensus motif, which could act as a recognition site for RGD binding integrins such as the β 1 integrin that is responsible for cell-fibronectin adhesion 19,34,35 . Moreover, cell adhesion to LTBP-1 can be reduced by blocking β 1-integrin in myofibroblasts 24 . To determine the importance of β 1-integrin activity in LTBP-1 fibril organisation by PSCs, prior to culturing on ECM, the control cell group was incubated with 1 μ g/ mL β 1-integrin function blocking antibody (BV7) for 30 minutes. Control PSCs were then seeded onto LTBP-1  rich matrices and incubated for 48 hours. When β 1-integrin was blocked in control PSCs, the fibronectin/LTBP-1 alignment was significantly decreased in comparison to untreated group (p < 0.001) (Fig. 4a). Furthermore, phalloidin staining showed that the cell area occupied by control PSCs was also significantly reduced after β 1-integrin inhibition (p < 0.001) (Fig. 4b). These results show that β 1-integrin expression by PSCs is a prerequisite for actomyosin-mediated alignment of LTBP-1 and fibronectin.

ATRA inhibits PSC-mediated liberation of bioactive TGF-β from its latent LTBP-1 complex.
Given that ATRA prevented LTBP-1/fibronectin organisation, we investigated whether this was sufficient to inhibit activation of TGF-β from its LTBP-1-bound latent form. To measure the level of TGF-β activated by

Figure 3. Effect of actomyosin cytoskeleton inhibition on LTBP-1 and fibronectin fibril alignment.
Fibronectin/LTBP-1 co-localisation and fibril density analysis of immunofluorescent images of BBI or DMSO only treatment of PSCs showed a significant difference between control and ATRA group in vehicle control condition; however, under BBI treatment control and ATRA groups establish similar fibronectin/LTBP-1 colocalisation levels and fibre thickness. Representative for 3 independent experiments, *** p < 0.001, one-way ANOVA with Tukey's post-hoc comparison test, error bars show means ± SEM, n.s. not significant. PSCs, HEK-293 cells were transfected with whole TGF-β coding vector, left to deposit ECM on glass coverslips, then removed as previous. PSCs were then cultured on the ECM for 48 hours in serum-free media. In order to assess baseline active TGF-β production by PSCs during 48 hours, control or ATRA treated PSCs were cultured on non-transfected HEK-293 derived ECM or glass coverslips in serum free media. Conditioned media was collected after the incubation period to quantify the levels of active TGF-β released from the ECM by PSC activity (shown schematically in Fig. 5a). We used a TGF-β reporter cell that carries a luminescent vector fused with TGF-β downstream target PAI-promoter (Fig. 5b); this assay allows quantification of bioactive TGF-β to distinguish it from latent TGF-β and is more physiologically relevant. Reporter cells were incubated for 3 hours in the presence of conditioned media collected from control or ATRA PSCs that were cultured on whole TGF-β matrices. The TGF-β bioassay showed that ATRA treatment significantly reduced active TGF-β levels in the media by 50% when compared to control (p < 0.05) (Fig. 5b), which is consistent with the decrease in fibronectin/LTBP-1 fibril alignment with ATRA treatment. Baseline active TGF-β derived from either control or ATRA treated PSCs was negligible during 48 hours given that media from PSCs grown on non-transfected HEK-293 derived ECM or on glass coverslips produced minimal active TGF-β signal (Fig. 5b). In addition, there was no significant difference in baseline active TGF-β between control or ATRA treated PSCs. Taken together, these findings show that ATRA reduced LTBP-1 alignment and fibronectin/LTBP-1 fibril co-localisation by PSCs which indicates a reduction in ECM remodelling. Concurrently, we showed that ATRA reduces active free TGF-β which could be a result of reduced ECM reorganisation.

Discussion
In this study we showed that PSC contractility controls concomitant ECM remodelling and active TGF-β release from the ECM. Furthermore, ATRA treatment significantly suppresses the ECM remodelling capacity of PSCs needed to activate TGF-β . We also showed that PSC-mediated LTBP-1 alignment was dependent on fibronectin fibril guidance: in the absence of fibronectin, LTBP-1 fibril assembly was diminished. Yet, the effects of ATRA were not simply a reflection of fibronectin expression levels by PSCs (as shown by the reduction in fibronectin-EDA and fibronectin-EDB expression) because when the system was saturated with excess fibronectin, control PSCs were still able to organise significantly higher levels of LTBP-1 fibrils than the ATRA treated PSC group. Instead, Figure 5. The effect of ATRA treatment on liberation of TGF-β from its LTBP-1 latent form (a) Whole TGFβ transfected or non-transfected HEK-293 cells were cultured for 7 days and removed (first arrow). Treated or control PSCs were incubated on matrices for 48 hours during which they can release TGF-β from LTBP-1 into the culture media (second arrow). Active TGF-β containing conditioned culture media was collected. FN, fibronectin. (b) Left panel; schematic representation of TGF-β bioassay in which TGF-β reporter cells with a luminescence vector were incubated with active TGF-β containing control or ATRA treated PSC culture media. After ATRA treatment, TGF-β activation by PSCs was significantly reduced (right panel). There were no significant differences in luciferase signal between control and ATRA treated PSCs when cells were seeded on non-transfected HEK-293 ECM or on glass coverslips. The active TGF-β signals were negligible in the following cases: HEK-293 cells or PSCs seeded on glass and PSCs seeded on matrices previously deposited by nontransfected HEK-293. Representative of 3 independent experiments; * p < 0.05, two-tailed Student's t-test, error bars show means ± SEM, A.U. Arbitrary units. (c) Summary schematic illustrating the effects PSC deactivation on ECM remodeling capacity.
Scientific RepoRts | 6:27639 | DOI: 10.1038/srep27639 we found that actomyosin cytoskeletal activity was central for LTBP-1/fibronectin alignment because when actomyosin contraction was blocked by BBI, PSCs were unable to organise LTBP-1 into fibronectin fibrils. Treatment of ATRA pre-treated PSCs with BBI did not further reduce LTBP-1/fibronectin co-localisation, thereby confirming that ATRA acts through an actomyosin-dependent mechanism. Therefore, LTBP-1 fibril alignment is a function of cellular contractility and requires fibronectin secretion during TGF-β priming.
The role of α v -integrin on LAP is a well-known mechanism of TGF-β activation 9,36 ; however, the importance of the RGD sequence in human LTBP-1 is yet to be elucidated. Here we found that, in PSCs, β 1-integrin activity was needed for LTBP-1 fibril alignment and for cell spreading on LTBP-1 rich ECM in concert with fibronectin secretion. However, blocking β 1-integrin activity has a widespread effect, and we cannot exclude the possibility that blocking β 1-integrin might affect other adhesion-dependent processes and ECM remodelling behaviour.
Given the paramount relevance of the desmoplastic reaction in the stroma of pancreatic tumours, stromal modulatory drugs have been shown to be a robust strategy which benefits pancreatic cancer treatment by targeting the tumour promoting ability of PSCs, increasing drug delivery to tumour bed, and reducing cancer metastasis 4,25,37,38 . Stromal reprogramming therefore acts by restoring homeostatic conditions in cancer-related microenvironment rather than direct ablation of stromal matrix. ATRA has been shown to restore quiescent phenotype in PSCs by inducing transcriptional reprogramming 25,39 . Here we showed that ATRA treatment weakens the capacity of PSCs to activate TGF-β , and TGF-β biological activity is therefore reduced in vitro (Fig. 5c). Our findings contribute to previous findings that show ATRA hampers the active myofibroblast PSC phenotype, which is strongly associated with tumour growth and metastasis 25,40 . Further work is required to determine if ATRA abrogation of TGF-β activation is sufficient to inhibit secondary effects in a tumour environment such as cancer cell proliferation.
ATRA (Sigma-Aldrich, USA) was prepared in ethanol and PSCs were treated with 1 μ M ATRA to limit cytotoxicity under dim light conditions for 10 days. ATRA treatment produced quiescence markers such as retinol droplets as described previously 25  ECM de-cellularisation and remodelling. Sodium deoxycholate (DOC) buffer is a water-soluble, ionic detergent commonly used in cell lysis buffers for isolation of cell membrane proteins and lipids. Here we used DOC buffer to disrupt the cell membrane and remove HEK-293 cells from 7 day old ECM. It has been shown previously that the ECM left behind is DOC resistant 24 . DOC was prepared with 150 mM NaCl, 50 mM Tris HCl, 1% NP-40 (v/v) substitute and 0.5% (w/v) sodium DOC (all reagents were purchased from Sigma-Aldrich, USA). Ice cold DOC buffer was added to 7 day HEK-293 cultures on coverslips and incubated at 4 °C for 10 minutes with gentle agitation. The HEK-293 cell debris was washed twice with PBS. Remaining DOC-resistant ECM was incubated with room temperature PBS for 15 minutes. Control or 10-day ATRA treated PSCs (3 × 10 4 cells) were then seeded on top of de-cellularised matrices and left to remodel ECM under pertinent treatment for 48 hours.
Gene knock-down. PSCs were reverse transfected with 50 nM fibronectin specific or non-targeting siRNA with JetPrime (Polypus, USA). Prior cell seeded matrices were coated with 200 μ l siRNA solution and incubated for 10 minutes, then PSCs were added and cultured.
TGF-β bioassay. Control or ATRA treated PSCs were starved for 48 hours on whole TGF-β matrices generated from whole TGF-β transfected HEK-293 or on non-transfected HEK-293 as described previously. Conditioned culture media was then collected and stored at − 80 °C until needed. The media was concentrated 4 times by centrifugation for 2 minutes at 13000 rpm using centrifugal filter units (Amicon Ultra 0.5 ml centrifugal filters, 10 K, UFC501024, USA). MDAMB231 cell line transfected with plasminogen activator inhibitor-1 (PAI-1) promoter fused with luciferase reporter gene (gifted from Dr Caroline Hill, London Research Institute, London) were used to determine the activity of TGF-β in conditioned media. When the reporter cells are incubated with active TGF-β containing culture media, downstream TGF-β proteins SMAD-2 and SMAD-3 are phosphorylated, form a complex with SMAD-4, and are transported to the nucleus. This complex targets the PAI-1 promoter, which in the reporter cells is fused with luciferase. MDAMB231 reporter cells were seeded in 96 well plates ( Image acquisition and quantitative analysis. Images were taken with a Motic AE31 trinocular inverted microscope by Motic Images Plus 2.0 software using 40x objective for fibronectin/LTBP-1 analysis and with 20x for phalloidin staining. Co-localisation and fibril density analysis were done with ImageJ (NIH) software by choosing 5 similar-sized (702 × 702 pixels) region of interests (ROI) per image in symmetry. Three images were analysed per condition per experiment. For 3 experiments, n = 45 per condition. The ROIs were then split into red and green channels. The amount of co-localisation of LTBP-1 (red channel) and fibronectin (green channel) was analysed with Coloc2 plug-in by obtaining Pearson's correlation coefficient (R value) which uses deviation from mean and is not affected from offset 41 . In order to analyse the fibril density co-localisation threshold images were created, converted into binary and analysed with BoneJ plug-in (NIH) by using thickness option. The intensity of vimentin staining was analysed by quantifying the mean fluorescence intensity (MFI) with ImageJ.

Statistical analysis. Control and ATRA treated conditions were compared by two-tailed Student's t-test
(GraphPad Prism, San Diego, CA). The p-values less than 0.05 were regarded as significant. For multi group comparisons a one-way ANOVA with Tukey's post-hoc comparison test was used. A single asterisk indicates * p < 0.05, a double asterisk indicates * * p < 0.01, and a triple asterisk indicates * * * p < 0.001. Error bars are standard errors of mean (SEM).