Hippo Pathway Deregulation Drives Tissue Stiffness and Cancer Stem-like Cells in Lung Adenocarcinoma

Lung cancer remains the leading cause of cancer-related death due to poor treatment responses arising from tumor heterogeneity and epigenetic aberrations. Here we show that adverse prognosis associated with epigenetically silenced tumour suppressor RASSF1A is a consequence of increased extracellular matrix, tumour stiffness and metastatic dissemination in vivo and in vitro. We find that lung cancer cells with methylated RASSF1A display constitutive nuclear YAP1 and expression of prolyl4hydroxylase2 (P4HA2) into extracellular matrix that correlates with increases collagen deposition. Furthermore, we identify an epigenetic axis in tumour cells where elevated ECM impedes the intrinsic suppression of WNT signaling (via TPBG/5T4) triggering b-catenin-YAP1 activation and thus results in a cancer stem-like programming. As key drivers, we identified RASSF1A and P4HA2 mediating the ECM-dependent stemness and metastatic dissemination in vivo. Re-expression of RASSF1A or inhibition of P4HA2 activity reverse these effects and increase levels of lung differentiation markers (TTF-1, Mucin5B) in vivo and in vitro. Our study identifies an epigenetic program to cancer stemness and metastatic progression of lung adenocarcinoma and P4HA2 as potential target for uncoupling ECM signals towards cancer stemness.


Introduction
Cellular heterogeneity within the tumor microenvironment has been reported as a general property of solid cancers (Hanahan & Weinberg, 2011;Marusyk et al, 2014). The population of cells referred to as cancer stem cells (CSCs) exhibit extensive self-renewal abilities, multi-potent differentiation and increased metastatic tumor formation (Ponti et al, 2005;Al-Hajj et al, 2003;Sales et al, 2007;Wang et al, 2013). Although CSCs represent <1% of total population of a tumor, they represent a major contributing factor to radio-or conventional chemotherapy resistance and aggressive progression (Kaseb et al, 2016;Salcido et al, 2010). Dedifferentiation of cancer cells to a more pluripotent-like state results in the appearance of CSCs during malignant tumorigenesis (Codony-Servat et al, 2016), however, the mechanism behind this process in solid tumors remains to be elucidated. Recent evidence suggests that interaction with the surrounding microenvironmental niche contributes to the conversion of non-stem cancer cells to CSCs (Gupta et al, 2011;Chaffer et al, 2011). Both biochemical and physical signals from the tumor microenvironment can modulate properties of CSCs endowing them with the potential to adapt to the emerging cancer niche (Driessens et al, 2012;Chen et al, 2012;Schepers et al, 2012;Schwitalla et al, 2013). Mechanical properties provided by extracellular matrix (ECM), e.g. increased levels of collagen, enhanced deposition or crosslinking, have been described to influence CSCs plasticity and trigger cancer stemness in non-stem cancer cells (Ye et al, 2014;Wong & Rustgi, 2013). In line with promotion of CSCs, increased matrix stiffness within tumor tissue also correlates with elevated cancer invasion, migration and metastatic spreading (Lane et al, 2014;Turner & J. Dalby, 2014;Scadden, 2014;Morrison & Spradling, 2008). Mechanotransduction from the ECM serves as upstream regulator of the Hippo pathway transcription factors YAP/TAZ (Piccolo et al, 2014), recently reported as an essential component for formation of lung CSCs (Halder & Johnson, 2011;Tremblay & Camargo, 2012). Nuclear YAP1 is responsible for numerous oncogenic properties of tumor cells and is restricted by the Hippo pathway mediated phosphorylation on serine127 (YAP1-pS127) (Zhao et al, 2008;Zanconato et al, 2015). RASSF1A is a key regulator of Hippo signaling in humans and loss of expression has been correlated with reduced YAP1-pS127 across multiple clinical cohorts (Vlahov et al, 2015). Independently, RASSF1A has been extensively validated as a tumor suppressor in lung cancer where promoter methylation associated gene silencing correlates with poor progression and overall survival (Burbee et al, 2001;Lee et al, 2001;Neyaz et al, 2008;Pallarés et al, 2008;. Although RASSF1A has been studied in vitro, the precise consequence of loss of RASSF1A expression in vivo has been difficult to discern. Gene expression analysis of lung and breast cancers has recently provided insight as, in addition to YAP1 activation, embryonic stem cell signatures are significantly elevated in human tumors lacking RASSF1A (Pefani et al, 2016).
To address the precise consequence of RASSF1A loss in vivo we constructed H1299 isogenic human lung tumor cells in order to directly assess tumor characteristics in an orthotopic lung tumor model.
We found that lung cancer cells with methylated RASSF1A produce prolyl4hydroxylase2 (P4HA2) into the extracellular matrix, where is essential for maintaining extracellular matrix-dependent stemness.
Concomitantly we found that collagen deposition and tissue stiffness is directly associated with RASSF1A expression in vitro and in vivo. Moreover, this stiff-ECM displayed strong binding of the CSC marker and suppressor of WNT signaling, trophoblast glycoprotein 5T4 (also known as TPBG or WNT Associated Inhibitory Factor (WAIF1) (Hole & Stern, 1988). We found that 5T4/TPBG fails to suppress WNT in the presence of stiff-ECM and that beta-catenin activation triggers the appearance of a Nanog +ve stem-like cell population in RASSF1A null lung cancer cells. Together, our study provides evidence that widespread clinical prognostic value attributed to RASSF1A epigenetic silencing is due to ECM re-organization associated increase in Nanog +ve stem-like cells and offers new therapeutic opportunities to combat the heterogeneity underlying treatment failures in these cancers.

RASSF1A impairs metastases in lung adenocarcinoma.
DNA methylation of the CpG island spanning the RASSF1A promoter has been widely appreciated to associate with poor clinical outcome of non-small cell lung cancer (Kim et al, 2003;Fischer et al, 2007).
Unfortunately, to date RASSF1A mRNA levels have not been correlated with epigenetic silencing in NSCLC cohorts, nor have they provided the same prognostic value as methylation. Here we have explored expression levels in The Cancer Genome Atlas (TCGA) and demonstrate that RASSF1A mRNA expression does correlate with a good prognosis for lung adenocarcinomas but not squamous cell carcinomas ( Fig 1A and Fig EV1A). To address the physiological role of RASSF1A we constructed isogenic RASSF1A-methylated H1299 lung adenocarcinoma cells to stably express RASSF1A or pcDNA3 (control) (Fig 1B). H1299 RASSF1A and H1299 con cells were orthotopically injected the left lung of mice and examined for tumor formation at day17 or day30 ( Fig 1C and Fig EV1B, C). At day 30, 85% of mice (n=5/6) injected with H1299 con cells developed clear evidence of primary tumors in left lung and 100% developed contra-lateral metastases in of the right lung (Fig 1C,D,E and Fig EV1C). This process was accompanied by production of liquid oedema around lungs, likely induced by inflammatory response to cancer progression (Movie EV1, 2). Surprisingly, 50% of mice (n=3/6) injected with H1299 RASSF1A cells also developed primary tumors similar in size to controls, however, there was a suppression of metastatic events in the both the ipsilateral and contralateral lungs (Fig 1C,D,E and Fig EV1C). This suggests that while RASSF1A can suppress primary tumor growth many, the primary consequence of loss in human tumors may be the failure to suppress metastatic events. Similar results were obtained for primary tumors when sacrificed earlier (day 17) but no metastases were apparent at this time, suggesting that metastases of H1299 con are unlikely to be an early event and are a consequence of dissemination from a mature primary tumor ( Fig EV1B).

Altered mechanical properties of ECM in RASSF1A null tumors.
Cancer cell invasion into surrounding tissue is associated with dynamic processes that require ECM and tissue remodeling (Hynes, 2009;Frantz et al, 2010;Back et al, 2005;Barry-Hamilton et al, 2010).
In line with reduced metastatic potential, H1299 RASSF1A cancer cells had a dramatically reduced ability to invade through 3D-Matrigel matrix compared to H1299 con controls (Fig 2A). To further explore processes associated with tissue remodeling during invasion, cells were assessed for their ability to contract a 3D-collagen matrix in vitro. After four days, extensive matrix remodeling and contraction of collagen was apparent in the presence of control cancer cells, whereas collagen surrounding H1299 RASSF1A cells remained unaltered ( Fig 2B). Moreover, substantially greater collagen deposition and organization was apparent around H1299 con tumors in vivo as visualized by second harmonic generation (SHG) (Fig 2C). YAP is a mechanical sensor of ECM stiffness that facilities growth and expansion of stem cells and tumor cells and intriguingly, it is regulated by RASSF1A. Nuclear localization of YAP occurs on stiff matrix and indeed we observed this in H1299 con cells in threedimensional 2% collagen, however in line with its role in Hippo pathway activation, re-expression of RASSF1A prevented nuclear accumulation ( Fig 2D). To determine if this was cell intrinsic or due to the "specific" ECM, we extracted ECM from H1299 con or H1299 RASSF1A cells and used these as a substrate for naïve H1299 RASSF1A cancer cells (where YAP is more cytoplasmic due to activated Hippo pathway).
YAP transduced to the nucleus in H1299 RASSF1A cancer cells grown on the ECM con but not on ECM RASSF1A , suggesting that the H1299 control cancer cells with epigenetically silenced RASSF1A produce stiffer ECM that is key for cancer cell re-programming ( Fig 2E). Taken together, the lack of 3D-collagen contractility and producing less stiff ECM of H1299 RASSF1A cancer cells is consistent with reduction of invasion in vitro and in vivo, and decreased potential to form primary tumors and metastases.
We next hypothesized that methylation of RASSF1A in H1299 con lung cancer cells may be associated with expression of specific ECM components that are responsible for increased invasion and metastatic dissemination in vivo. To identify candidate elements, we performed mass spectrometry analysis of isolated ECM from both H1299 con and H1299 RASSF1A lung cancer cells. As expected a variety of ECM components were produced by both cells, e.g. collagens, laminins and fibronectin, however, H1299 con ECM exclusively contained laminin-beta2 (LAMB2) and prolyl 4-hydroxylase subunit alpha-2 (P4HA2) which contribute to the deposition and stability of ECM ( Fig 3A). Interestingly, also exclusive to H1299 con ECM was ADAMTSL5 and TPBG (also known as WAIF1 or 5T4; a stem cell receptor that suppresses WNT) which together with P4HA2 all displayed adverse prognostic value for lung adenocarcinoma conversely to RASSF1A expression ( Fig 3B). ECM components specifically identified from H1299 RASSF1A ; MRX7A, SDF2, TFPI2 and TIMP1 also had expression profiles that significantly correlated with a good prognosis in lung adenocarcinoma or breast cancer, in keeping with clinical observations for RASSF1A ( Fig 3B and Fig EV1D). Interestingly, we showed that high expression of P4HA2 displayed worse prognoses with widespread clinical significance across a range of solid malignancies ( Fig EV2A).
We next confirmed the proteomic data by immunofluorescence staining and did not observe expression of P4HA2 in extracellular space in RASSF1A expressing cells compared to control H1299 cells where RASSF1A is methylated (Fig 3C and Movie EV3,4). P4HA2 catalyzes the formation of 4hydroxyproline in collagens and it is essential for proper three-dimensional folding of newly synthesized procollagen fibers (Myllyharju, 2003(Myllyharju, , 2008. Absence of P4HA2 results in reduced collagen deposition and severe ECM abnormalities (Myllyharju & Kivirikko, 2004). In agreement with P4HA2 expression, H1299 con cells display higher levels of stable collagen I compared to H1299 RASSF1A cells ( Fig   3D, E). To test if RASSF1A is responsible for collagen I switch, we selected a lung adenocarcinoma HOP92 and human cervical HeLa cancer cells that retain endogenous RASSF1A expression and find that these cells displayed low levels of P4HA2 and collagen I, both of which could be upregulated upon silencing of RASSF1A expression with siRNAs ( Fig 3F and Fig EV2B).

P4HA2 expression associates with organized ECM and stiffness in vitro and in vivo
To investigate if methylation of RASSF1A is associated with the quality of collagen deposition we embedded 3D-spheroids into existing ECM to examine collagen structure by second harmonic generation microscopy (SHG). We found local production of collagen around H1299 con spheroids with long, organized fibers radiating from 3-D aggregates which were not visible around H1299 RASSF1A spheroids and where appeared as a short, disorganized fibers ( Fig 4A). Intriguingly, collagen fibers produced by H1299 con cells seem to serve as tracks for invading cells and initiate spread from the central spheroid ( Fig 4A, red arrows). P4HA2 is known to have a major effect on physical properties of tumor-associated ECM, which in turn leads to increased stiffness during cancer progression (Provenzano et al, 2006;Levental et al, 2009). To assess whether our observations correlated in vivo, we determined P4HA2 levels in primary tumors by immunohistochemistry (IHC) and, as expected, controls expressed high levels (peri-nuclear due to processing of collagen in the ER; Human Protein Atlas), and reduced staining was evident in H1299 RASSF1A tumors ( Fig 4B). Correspondingly, SHG microscopy revealed that compared to controls, H1299 RASSF1A primary tumors showed significantly lower level of collagen fibers and remaining signal indicated disperse organization with no unifying pattern ( Fig 4C), suggesting the reduced levels of P4HA2 affected collagen deposition and may have either prevented metastasis emanating from H1299 RASSF1A tumors or restricted their ability to colonize the contralateral lung. Interestingly, H1299 con micrometastases mirrored similar collagen deposition surrounding tumorous tissue (Fig 4C, lower) as we observed in 3D spheroids in vitro ( Fig 4A). We also observed that pre-metastatic control mice (day17) displayed more collagen with organized deposition in the ipsilateral lung ( Fig.2C) but not in the contralateral site destined to succumb to metastasis ( Fig   EV2C). Thus, the variation in collagen level and its organization of the primary tumors is likely to be associated with dissemination rather than colonization. To evaluate how these collagen expression and organization influence the mechanical properties of tumor microenvironment, we next employed atomic force microscopy (AFM) analysis to determine nanoscale variations in stiffness. Consistent with SHG data, detailed topographic analyses of lung tumor stroma by AFM displayed organization of very compact extracellular network mesh with multiple fibers produced by H1299 con in vitro and in vivo, in contrast to H1299 RASSF1A (Fig 5A, B). As a consequence, AFM measurements demonstrate H1299 con lung tumors tissue were substantially stiffer (16 x kPa vs 1.1 x kPa, p<0.001) compared to tumors generated by H1299 RASSF1A cells (Fig 5C,D,bars). As the collagen enrichment in tumor ECM is often associated with appearance of fibrotic tissue, contributing to cancer progression (Liu et al, 2010), we found that lung tumors from H1299 con cells displayed an extended fibrosis that was not apparent in H1299 RASSF1A tumors ( Fig 5E).

YAP pathway regulation of P4HA2
The Hippo pathway mediator YAP1 is both a key regulator of P4HA2 levels and has been well documented to be regulated by RASSF1A (Piersma et al, 2015;Matallanas et al, 2007). Nuclear YAP1 is a readout of transcriptional activation and in agreement with previous data and the expression of P4HA2, YAP1 is significantly more nuclear in H1299 con cells compared to H1299 RASSF1A ( Fig 2D). YAP1 is a pleiotropic transcription factor that drives stemness through activation of the pluripotency cassette (OCT4, SOX2 and Nanog). Disruption of Hippo pathway leads to constitutive nuclear levels of YAP that in turn induces activation of lung cancer stem cells (Noto et al, 2017). Despite constitutive nuclear YAP1, H1299 con cells grown in 2D did not display upregulation of NANOG compared to H1299 RASSF1A and levels remain weak in both cell lines ( Fig EV3A). To address whether mechanical properties of ECM can induce stemness, we embedded H1299 con and H1299 RASSF1A three-dimensional spheroids into the 3D collagen matrix. Strikingly, H1299 con spheroids display a substantial level of bright Nanog +ve cells, whereas H1299 RASSF1A did not presumably due to reduced nuclear YAP ( Fig 6A and Fig EV3B). YAP also regulates Hif1α transcription, an additional activator of P4HA2 under hypoxic conditions, but our spheroid formation method does not result in a hypoxic core, nor do we see Hif1-alpha expression ( Fig EV3C). We also show that the appearance of the Nanog +ve cells in the spheroids embedded in three-dimensional collagen matrix was dependent on P4HA2, as the P4HA2 inhibitor 1,4-DPCA abolished nuclear staining of NANOG and reduced nuclear YAP (Fig 6B). To determine whether this effect was a property of 3D-spheroids or cells intrinsic, we next embedded single cells in 3-dimensional collagen matrix and subsequently exposed these cells to P4HA2i. As expected, we found that YAP and NANOG levels in single cells are not affected by acute exposure, but surprisingly we did find that the intrinsically high levels of b-catenin in H1299 con cells are immediately sensitive to P4HA2i (Fig 6C).
5T4/TPBG is a transmembrane glycoprotein and stem-cell marker known to restrict WNT signaling and in agreement with Figure 6A, WNT signaling levels are higher in H1299 con cells. We interpret this data, together with the isolation of 5T4/TPBG in H1299 con ECM to suggest that reduced binding of 5T4/TPBG to the ECM upon P4HA2i allows 5T4/TPBG to suppress WNT signaling and destabilize b-catenin. We found that Nanog +ve cells were similarly dependent on YAP as restriction of YAP expression with siRNA also abolished the stem-like cells population ( Fig 6D) and P4HA2 expression in H1299 con cells ( Fig 6E).
So far, our model would suggest that RASSF1A expression suppress YAP nuclear localization, thus fail to express P4HA2 and without the increased collagen deposition and stiffness the ECM cannot impede 5T4/TPBG. As YAP itself is mechanosensitive with ability to translocate into the nucleus independently on Hippo signaling, we were curious whether increased collagen stiffness can also induce cancer stemlike programing. Interestingly, we showed that elevated stiffness of ECM induces production of Nanog +ve cells independently on Hippo pathway ( Fig 6F and Fig EV3D). Taken together this data suggests that production of P4HA2 and stiffness is required for maintenance of cancer stem-like phenotype in a YAP and WNT dependent manner.

Activation of Hippo pathway leads to lung cancer cells differentiation.
To address whether the difference in tissue stiffness translated into increased cancer stem-like cells in vivo, we stained mouse lung primary tumors and found YAP exclusively nuclear in H1299 con tumors compared to more cytoplasmic staining in H1299 RASSF1A tumors, and correlated with NANOG expression ( Fig 7A). This data supports the model where P4HA2 is required for maintenance of cancer stem-like phenotype and cell pluripotency during early stages of tumor development. Expression pattern of P4HA2 was also highly up-regulated in H1299 con metastases but not in the few small metastatic tumors that did appear in mice bearing H1299 RASSF1A tumors, suggesting its importance in lung cancer progression ( Fig EV4A). We next hypothesized that reduced cancer stem-like cells involvement may imply that tumors expressing RASSF1A are well-differentiated as is observed clinically for RASSF1 methylation . To address the question, we evaluated immunohistochemical expression of TTF-1 and Mucin 5B as markers for terminal lung differentiation . TTF-1 overexpression in lung tumors tissue has been used as prognostic marker correlated with better prognosis and survival in lung cancer patients (Saad et al, 2004;Boggaram, 2009). We first determined levels of Mucin 5B and TTF-1 in vitro and found in higher levels in H1299 RASSF1A cells compared to controls ( Fig EV4B) supporting the idea that the cells were less stemlike and well differentiated. Interestingly, while H1299 con spheroids cultured on Matrigel maintained rounded morphology, H1299 RASSF1A spheroids collapsed and migrated into branched structures reminiscent of a differentiated epithelium ( Fig 7B and Movie EV5). We demonstrated, that high expression levels of mRNA TTF-1 and Mucin 5B correlates with better clinical outcome in patients with lung adenocarcinoma (Fig 7C). To simulate differentiation in vitro, we embedded 3D spheroids in three-dimensional collagen matrix and again observed greater expression of TTF1 transcription factor and Mucin 5B in H1299 RASSF1A spheroids ( Fig 7D). In line with this, primary tumors generated by H1299 RASSF1A cells in mice, similarly displayed significantly higher levels of human specific Mucin 5B and TTF-1 compared to H1299 con primary tumors ( Fig 7E and Fig EV4C). We also observed significant elevation of TTF-1 with its prominent nuclear localization in contralateral metastases generated by H1299 RASSF1A cells (Fig EV4D), but failed to observe changes in Mucin 5B expression suggesting this may be specific for primary tumors ( Fig EV4D). These data are consistent with a model whereby activation of the Hippo pathway supports differentiation and prevents cancer stem-like cells, ultimately leading to less aggressive forms of lung cancer. Altogether, our results identifies the epigenetic program demonstrating that P4HA2 together with biophysical properties of extracellular matrix drive cancer stem -like programming and metastatic progression in lung cancer and contribute to poor outcome in lung adenocarcinoma patients ( Fig 8A).

Discussion
Lung cancer has a dismal prognosis with only 10% of patients surviving greater than 5 years after diagnosis. As surgery is limited to patients with localized disease, greater confidence in determining which patients are more likely to recur with metastasis is vital to provide appropriate treatment regimens and increase overall survival (Yang et al, 2005). RASSF1A is a bone fide tumor suppressor in non-small cell lung cancer, the epigenetic silencing of which correlates with advanced disease, metastatic potential and adverse prognosis (Donninger et al, 2007;Dammann et al, 2000;Morrissey et al, 2001;Dammann et al, 2001;Dreijerink et al, 2001). However, a precise reason for these clinical associations has been lacking. Here we advance the original epigenetic association of RASSF1A levels with poor survival by utilizing mRNA data from The Cancer Genome Atlas (TCGA). We demonstrate for the first time that expression levels can similarly correlate with outcome, but strikingly now show this is specific to adenocarcinomas and no correlation is seen with squamous cell carcinomas of the lung, clarifying from which population the prognostic value is being derived. In line with observations in human tumors, our in vivo experiments highlight that RASSF1A only partially restricts tumor formation while the major consequence of loss is increased metastatic spread. Interestingly, metastatic cells are reported to be migrating phenotypically-plastic cells which have the attributes to survive dissemination in the blood stream and colonize distant sites (Van Engeland et al, 2002;Folkman, 2002;Woodhouse et al, 1997;Fidler, 1999;Chambers et al, 2001;Wyckoff et al, 2000). We found that lung adenocarcinoma cells lacking RASSF1A expression contain subpopulation of Nanog +ve stem-like cells, and that this population was mostly evident in the context of a tumor or in vitro simulated 3D microenvironment.
The primary tumor microenvironment is a heterogeneous and dynamic biological niche which can be composed of distinct populations of cancer cells, stromal fibroblasts and infiltrating immune components (Hanahan & Weinberg, 2011). The mechanical properties also play an important role that can provide a physical barrier and we find that the physical stiffness of the extracellular matrix directly influences the tumor cell heterogeneity. Mechanotransduction allows transcriptional responses to stiffness, in particular the Hippo pathway transcriptional regulator YAP, however how this functions in a 3D-microenvironment or in tumors has not been clear. RASSF1A activates the Hippo pathway to prevent YAP mediated transcription of pluripotency genes such as OCT4, but this requires cooperation from WNT mediated stabilization of beta-catenin for stable promoter binding (Papaspyropoulos et al, 2018). OCT4 is a key regulator of Nanog expression and together with SOX2 form the pluripotency cassette activation of which represents true stem cell behavior. Loss of RASSF1A is sufficient to activate YAP transcription that is beta-catenin independent and we find the YAP-target P4HA2 to be elevated in RASSF1-methylated H1299 cells compared to RASSF1A expressing controls. P4HA2 is an intracellular enzyme that modifies procollagen I to allow efficient export for contribution to the ECM and can associate with collagen fibers during export. We found that levels of P4HA2 in the extracted ECM were RASSF1A dependent, indicating that collagen processing is highly enriched and concomitantly, we find mature collagen fibers and a much stiffer ECM in control tumors and in vitro spheroids.
We show that stiffness of the tumor niche modulates expression of Nanog which, like RASSF1A, predicts tumor progression and poor prognosis, and is important for maintenance of cancer stem cells (Bussolati et al, 2008;Ibrahim et al, 2012;Zbinden et al, 2010). Similarly, P4HA2 not only gives a clear prognostic value for lung adenocarcinoma but across a range of solid tissues where RASSF1A silencing is also predictive and even in lung squamous cell carcinoma where RASSF1A does not -implying potential alternative routes to activating YAP in this setting. Interestingly, we also find an integral membrane protein associated with stem cells and poor prognosis, 5T4/TPBG, in the ECM of H1299 cells. The identification of a membrane protein in the ECM implies a stronger interaction with the ECM in controls as it is maintained during whole cell extraction and the levels of 5T4/TBPG (although higher in controls) do not appear to be RASSF1A dependent. 5T4/TBPG is a repressor of WNT signaling but in control cells WNT is not suppressed and beta-catenin levels are maintained. Inhibition of P4HA2, appears to weaken the ECM sufficiently to allow TPBG to suppress WNT and the reduce beta-catenin stability.
We suggest that mechanism whereby loss of RASSF1A allows YAP mediated expression of P4HA2, collagen deposition and a stiff ECM. The stiffness reinforces nuclear YAP but this is insufficient to drive cancer stem-like cells behavior alone. The ECM binding of 5T4/TPBG restricts its ability to suppress WNT, which then allows beta-catenin and YAP to coordinate activation of the pluripotency cassette via OCT4. The spheroids containing Nanog +ve cancer stem-like cell create highly organized long, very strong collagen branches serve as "tracks" for direct facilitated cell invasion from tumors and encapsulate micrometastases.
Our findings support the idea that the Hippo pathway is important for maintaining differentiation as re-expression of RASSF1A not only suppress cancer stem-like cells but and upregulates lung differentiation markers TTF-1 and Mucin 5B in vivo and in vitro. TTF-1 (thyroid transcription factor 1), also known as NKX1.2, has been useful marker for differentiating primary lung adenocarcinoma (Moldvay et al, 2004) and described as a prognostic marker where moderate or strong TTF-1 expression is associated with better survival (Saad et al, 2004;Boggaram, 2009). Overall our data indicates that the tumor microenvironment promotes dedifferentiation and an increase in stem-like behavior, thus targeting P4HA2 in tumor extracellular niche may be an important new strategy of eradicating chemotherapy resistant cancer stem cells during conventional tumor therapy. The work also indicates that RASSF1A levels or promoter methylation can serve as a prognostic biomarker and would be predictive for anti-ECM targeting companion diagnostics.

Cell culture and drug treatments
Human lung adenocarcinoma cells H1299TetON-pcDNA3 or H1299TetON-RASSF1A under tetracycline promoter were previously described (Van Der Weyden et al, 2012;Yee et al, 2012). RASSF1A expression was induced after treatment with 1μg/ml doxycycline for 24 hours before experiments.
H1299, HOP92 and HeLa cells were cultured in complete DMEM (Gipco) medium, supplemented with 10% foetal bovine serum (Sigma), penicillin/streptomycin (Sigma), and L-glutamine (Sigma) in 5% CO2 humidified atmosphere at 37°C. Cells used in experiments had been passaged fewer than 15 times. For transfections, cells were transfected with siRNA (100nM) using Lipofectamine RNAiMAX (Invitrogen) regarding to manufacturer's instructions. For generation stable H1299 cell lines, we used human RASSF1A in pBABE system and pBABE-pcDNA3 control. H1299 cells were infected with lentiviruses carrying RASSF1A or control pcDNA3 packaged in 293T cells using GP and VSVG plasmids for 48 hours and then selected with puromycin (10µg/ml) for two weeks. Activity of P4HA2 in cells was inhibited by using of 4 µM, 1, 4-DPCA, a prolyl-4-hydroxylase inhibitor (Cayman Chemical company) for 24 hours.
In brief, H1299TetON-pcDNA3 and H1299TetON-RASSF1A cells were induced with 1 μg/ml doxycycline for 24 hours before spheroids formation with or without 1,4-DPCA (Cayman Chemical company). Cells were detached with 2mM EDTA and re-suspended in medium supplemented with methylcellulose (20%, Sigma-Aldrich) and Matrigel matrix (1%, Corning, Growth factor reduced) and incubated as hanging droplets (25 μl °C. Spheroids-collagen gels were washed with PBS and mounted with DAPI. For single cells immunofluorescence staining in three-dimensional collagen, cells were trypsinized, washed in complete medium, counted and (10 5 /ml) cells were mixed with solution containing 1.5 mg/ml; 2mg/ml; 2.5 mg/ml or 3 mg/ml Collagen R (Serva), 10xPBS, 1M NaOH and complete medium. The suspension of cells-gel solution was loaded into 8 wells (Labtech). The gels were polymerized at 37 °C for 30 min and replaced with complete medium. After 48 hours the cells in 3D collagen were stained by protocol described above. Images were captured by using Nikon 10×/0.30 Ph1 objectives.

Three-dimensional Matrigel migration assay
H1299 multicellular tumour spheroids were generated by hanging drops methods described above.
H1299TetON-pcDNA3 and H1299TetON-RASSF1A multicellular aggregates were washed and cultivated on Matrigel matrix (Corning, Growth factor reduced) for 24 hours. Images were monitored at 37 °C using a motorized inverted Nikon Ti microscope (4×/0.10 NA air objective lens) connective to Nikon camera and captured every 30 minutes.

Immunoblotting
For protein analysis on 2D, cells were cultivated on 100-mm dishes and lysed in RIPA buffer as described previously (Palakurthy et al, 2009). For protein analyses from 3D H1299 cells expressing either control pcDNA3 or RASSF1A vector were embedded into 3D collagen type I matrix. After 72 hours cells were washed two times with cold PBS and isolated by incubation in collagenase B for 10 minutes at 37°C and lysed in RIPA buffer. Both 2D and 3D lysates were cleared by centrifugation at 15,000 rpm for 20 min and protein concentration was determined by the using the BSA assay, diluted in 2x NUPage sample buffer containing DTT, and incubated at 99°C for 10 min. and lysed in RIPA buffer.

Animal experiments
All animal experiments were performed after local ethical committee review under a project licence issued by the UK Home Office. Lung orthotopic xenografts were generated following a published protocol (Onn et al, 2003) with minor modification. In brief, BALB/c nude mice (Charles River Laboratories, U.K) were anaesthetised with 2% isoflurane, and placed in the right lateral decubitus position. 10 6 of stably transfected either pcDNA3 (control) or Rassf1A expressing H1299 cells in 50 µL of 50% Matrigel (BD Biosciences) were injected into the left lung. Mice were sacrificed on day 17 (n = 4/ group) or day 30 (n = 6/ group), and lungs were collected.

Immunofluorescence microscopy
Cells were seeded and cultivated on coverslips, fixed in 4% paraformaldehyde for 15 minutes at room temperature and then permeabilized with 0.5% Triton X-100 (Sigma) in PBS for 10 minutes at room temperature. Non-specific binding was blocked with 3% bovine serum albumin (Sigma) in PBS for 30 minutes before incubation with a primary antibody against type I collagen (Novusbio), Nanog (Cell signalling), YAP (Santa Cruz Biotechnology) and P4HA2 (Santa Cruz Biotechnology) for 2 hours at room temperature. Secondary antibodies were applied for 1 hour at room temperature, followed by staining with or without Phalloidin (Life technologies) for 15 minutes and after extensive washing between each step, coverslips were mounted onto microscopy slides with mounting medium containing DAPI.
Images were captured by using confocal Nikon 40×/2. objectives. For each condition a minimum of 300 cells were analysed.

Collagen contraction assay
Collagen gels for contraction assays were prepared by using of 8 parts of Rat tail collagen I (Serva, Germany), 1 part of 10X DMEM, 1 part of 1X DMEM, 1M NaOH (final pH 7.4) and mixed with cells (1 × 10 6 /ml). Gel-cells suspension (triplicates) was pipetted into 24 wells and allowed polymerized in 5% CO2 at 37 °C for at least 30 minutes. The collagen lattices containing cells were detached from the surface of the wells with the sterile needle, replaced with completed medium and allowed to contraction for 4 days. After day 4, the photographs were taken and contraction was calculated as an increase of gel diameters. Comparison of collagen gel contraction was performed by using Student's unpaired t-test and p-values <0.05 were considered statistically significant.

Picro-Sirius Red Staining
4 µm paraffin sections were collected on 3-aminopropyltriethoxysilane (AAS) slides and staining was performed in accordance to manufacturer's protocol (Abcam, ab150681). Briefly, slides were deparaffinised in graded ethanol solutions and then placed in distilled water for 5 minutes. Picro-Sirius Red solution was placed on slides for 60 minutes at RT and then rinsed quickly with Acetic Acid Solution. After staining, slides were passed through graded ethanol solutions, cleared in acetone and mounted with synthetic resin.

Generation of cell-free extracellular matrix
2 × 10 5 stably transfected H1299 cells either with pcDNA3 (control) or RASSF1A vector were seeded on coverslips in 12-wells to allow them to produce extracellular matrix for 7 days. Coverslips were washed with PBS and ECM was extracted two times for 5 minutes at 4°C with 0.5% DOC in immunoprecipitation buffer under gentle shaking (Unsöld et al, 2001) After discarding cell debris, ECM was washed tree times with PBS to clean ECM of any remaining debris. 5 × 10 4 H1299 RASSF1A cells were seeded on cell-free extracellular matrix extracted from either control or RASSF1A cells and grown for 3 days. H1299 RASSF1A cells were fixed and stained with YAP antibody.

Mass-spectrometry
H1299 cells expressing empty pcDNA3 or RASSF1A vector were growing for 2 weeks in presence of 100µg/ml L-ascorbic acid. ECM was extracted by 0.5% DOC (Unsöld et al, 2001)

Mechanical analyses by scanning probe microscopy
5 µm slides from frozen lung tissue were transferred on coverslips and analyzed for stiffness of primary tumors by AFM. For in-vitro three dimensional stiffness analyses, 2 × 10 5 H1299 cells expressing either pcDNA3 (control) or RASSF1A vector were embedded into 3D rat tail collagen type I matrix and allowed them modified ECM for 5 days in 8 wells (Labtech). 3D collagen gels were transferred onto slides and subjected to mechanical testing by scanning probe microscopy. Scanning probe microscopy was performed on a MFP-3D Atomic Force Microscope (Asylum Research, High Wycombe, UK), with an AC240TS probe (k = 2.0 Nm -1 , Olympus, Japan). AMFM nanomechanical mapping (Material property measurements using multiple frequency atomic force microscopy, 2008) and loss-tangent imaging (Proksch & Yablon, 2012) were applied, with measurements based on the shift in the probe's resonant frequencies dependent on the strength of the interaction, or tip-sample contact force (Giessibl, 1997).
A tip correction factor was calculated based on the known compressive Young's modulus of a polycaprolactone calibration sample (300 MPa). Three 5 µm x 5 µm areas were then randomly selected and measured from each material. 512 x 512 pixel maps of height, Young's modulus and loss tangent were recorded for each image.

Second harmonic generation using two-photon microscopy
Second harmonic generation signal from collagen were detected upon simultaneous excitation with a 920 nm laser. The signal from collagen was collected using a BP 460/50 filter. The images were acquired using the LSM 7 MP microscope (Carl Zeiss, Jena, Germany) using the Zeiss W Plan-Apochromat 20×/1.0 NA objective lens.

Statistical analysis
The results from all experiments represent the means ± SD of replicated samples or randomly imaged cells within the field. Numerical values of cell culture and mouse cohort data were analyzed using Student's t-test for significance in Prism 6 (GraphPad Software Inc.). Unpaired 2-tailed Student's t-test were used to compare the mean values of 2 groups. The difference was considered as statistical significant of P values < 0.05.

TCGA survival analysis.
The Cancer Genome Atlas (TCGA) project of Genomic Data Commons (GDC) collects and analyses multiple human cancer samples. The TCGA RNA-seq data was mapped using the Ensembl gene id available from TCGA, and the FPKMs (number Fragments per Kilobase of exon per Million reads) for each gene were subsequently used for quantification of expression with a detection threshold of 1 FPKM. Genes were categorized using the same classification as described above. Based on the FPKM value of each gene, patients were classified into two expression groups and the correlation between expression level and patient survival was examined. Genes with a median expression less than FPKM 1 were excluded. The prognosis of each group of patients was examined by Kaplan-Meier survival estimators, and the survival outcomes of the two groups were compared by log-rank tests. Maximally separated Kaplan-Meier plots are presented with log rank P values. Survival analysis was performed using SPSS version 24.0.
Here we demonstrate that RASSF1A methylation in lung adenocarcinomas is a consequence of triggering signals from ECM due to expression of P4HA2 into extracellular matrix, that in turn induces cancer stem-like cells programming and metastatic dissemination in vivo and in vitro. Re-expression of RASSF1A or inhibition of P4HA2 activity converses these processes and switch into lung cancer differentiation accompanied by better prognosis and clinical survival.

Impact
Our study demonstrates an importance of tumour microenvironment in epigenetic regulation of cancer stemness and metastatic progression of lung adenocarcinoma and targeting P4HA2 as a new anti-tumour therapy agent.    A. Mass-spectrometry analysis of proteins purified from extracellular matrix with summary of results. A qualitative analysis of proteins identified was performed using a Venn diagram and only proteins identified with two or more peptides were taken into consideration. The resulting list of proteins was restricted to ECM proteins (results were provided from triplicates).

B. Kaplan Meier plots depicting prognosis in TCGA_LUAD (lung adenocarcinoma) patients with
high and low expression of ECM proteins revealed by proteomics study. The P values were derived from a log-rank test.
C. Representative immunofluorescence images of P4HA2 intra and extracellular localization in H1299 con and H1299RASSF1A cells. Scale bars 5µm, ZOOM scale bars 2µm.

D. Representative immunofluorescence images of expression of Collagen I, co-stained with
Phalloidin Alexa-568 in H1299 cells stably expressing either empty lentiviral vector or RASSF1A cultivated on 2D. Scale bars 50µm.
E. Western blot analyses of Collagen I expressed in H1299 control and H1299 RASSF1A cells.
F. Immunofluorescence staining and western blot analyses of collagen I and P4HA2 in HOP92 cells upon RASSF1A restriction using siRNA. Scale bars 20µm. Student's 2-tailed t-test shows a significant difference in collagen I production between control and RASSF1A expressing H1299 generated tumours.