Formation of an invasion-permissive matrix requires TGFβ/SNAIL1-regulated alternative splicing of fibronectin

Cells were grown in standard medium and conditions. MDA-MB-231, MCF7, HT-29 M6 and NIH3T3 cells were acquired from the repository stock at the Institut Hospital del Mar d’Investigacions Mèdiques (IMIM). Control and Snai1 KO mouse embryonic fibroblasts (MEF), mesenchymal stem cells (MSC) [36] and control and Snai1 KO breast cancer-associated fibroblasts (CAF) [43] were previously established in our laboratory. MSC are isolated non-transformed fibroblasts that are recruited by tumors in vivo. The MEF lines expressing EDA- or EDA+ fibronectin or control unmodified MEF were prepared at the ICGEB (Trieste, Italy). BJ human fibroblasts were kindly gifted by Dr. Cristina Peña (Hospital Universitario Puerta de Hierro). The EpRas tumor cell line was kindly provided by Dr. Antoni Celià lab (IMIM) and were originally generated by Dr. Robert Weinberg (Whitehead Institute for Biomedical Research). The AT-3 tumor cell line was kindly gifted by the Dr. José Yelamos Lab (IMIM). Patient-derived xenografts (PDX) were provided by Dr. Joaquín Arribas (IMIM) and previously analyzed in our laboratory [37]. See Additional file 1 Materials and Methods for treatment reagents and transfected siRNAs.

Fibronectin RNA expression and splicing data were obtained from TSVdb [38], and the SNAIL1 protein levels for the same cohort of patients, from cBioPortal. The characteristics of each patient were also downloaded from the cBioPortal database. Thus, all procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.

For each cancer type studied (lung adenocarcinomas, skin melanomas, breast adenocarcinomas and kidney renal cell carcinomas), tumor data were analyzed in two groups: i) stages I and II, and ii) stages III and IV. The number of patients per group ranged from 69 to 590 (Fig. 1). Each tumor was classified according to its relative SNAIL1 protein levels and the percentage of EDA inclusion. Only full-length fibronectin isoforms, including or excluding exon 33, were considered. The cutoff value to discriminate between low and high expression was set in relation with the average value for the EDA percentage in each cancer type, ranging between 68 and 96%; the value was set at 80% for kidney cancer, 85% for breast cancer, 90% for lung cancer, and 99% for skin cancer. Normal tissue data were analyzed with these values, and only 3–8% of specimens expressed high EDA (Fig. 1). Normal tissue data for skin cutaneous melanoma were not available. As SNAIL1 protein stability is tightly regulated post-translationally [39], protein but not RNA levels were analyzed. A cutoff value of 1 was set by SNAIL1 expression data from Reverse-Phase Protein Array, normalized by z-score. For lung cancer, no normalized data were available, and the cutoff was set at 0.25 as the average value of the collection was –0.22. With these settings, advanced tumors (stages III and IV) concomitantly expressing high levels of SNAIL1 and fibronectin EDA represented 11.1%, 2.9%, 2.7% and 2.1% of the lung, skin, breast and kidney cancers, respectively. For PDXs, the SNAIL1 and EDA+ fibronectin cutoff values corresponded to 80% of the band intensity obtained in a positive control sample from activated EDA+ MEFs. Seven of 29 PDXs (24%) expressed high levels of SNAIL1 and EDA+ fibronectin. For statistical analysis, percentages of elevated EDA inclusion in high versus low SNAIL1 tumors were compared.

Cells on cell culture dishes were grown to 80% confluence, washed with warm PBS, trypsinized and recovered by centrifugation. Cells were then resuspended in 300 µL lysis buffer (100 mM KCl, 5 mM MgCl₂, 10 mM HEPES pH 7, 0.5% NP-40, 1 mM DTT, 1 × RNase inhibitor, 1 × protease inhibitor) for each 15-cm-diameter plate, incubated 5 min on ice and snap-frozen in liquid nitrogen. Thawed samples were sonicated in a Bioruptor Pico Sonicator (Diagenode) for 15 cycles of 30 s ON/OFF at 4 °C. Samples were centrifuged at 16,000 g for 10 min at 4 °C, the supernatants were recovered and the protein content was quantified. In parallel, antibody-complexed beads were prepared. Per sample, 30 μL of Gammabind G Sepharose beads was washed in NET buffer (50 mM Tris pH 7.5, 150 mM NaCl, 0.1% NP-40, 1 mM EDTA); samples were blocked with 20 μg tRNA, washed again and incubated for 2 h at 4 °C with rotation after adding the primary antibody or Irrelevant IgG. All centrifugation steps were done at 350 g and 4 °C. Protein samples were precleared with unblocked beads for 30 min at 4 °C with rotation and centrifuged for 2 min at 350 g, and the supernatant was recovered. A total of 6 mg of protein was mixed with the previously blocked and antibody-complexed beads and incubated for 2 h at 4 °C with rotation. The mix was centrifuged, and beads were washed 4 times with NET buffer. Finally, beads were resuspended in 50 µL NET buffer with 2 µL of blue glycogen and 150 µL TRIzol, and RNA extraction was carried out as described.

Micromechanics of the matrices were measured using a custom-built atomic force microscope mounted on an inverted optical microscope (TE2000; Nikon, Tokyo, Japan). All experiments were performed in PBS buffer with a pH of 7.4 at 37 °C. Measurements were performed by using force–displacement curves on the surface of the sample with V-shaped silicon nitride microfabricated cantilevers (0.012 N/m of nominal spring constant) ended with a 2.5 μm radius spherical glass bead (Novascan Technologies, Ames, IA, US). The vertical position of the cantilever was controlled with a piezoelectric actuator and measured with a strain gauge sensor (Physik Instrumente, Karlsruhe, Germany). A four-quadrant photodiode (S4349, Hamamatsu, Japan) was used to measure the deflection of the cantilever. The elastic modulus was calculated from the force–displacement curves by fitting them to the Hertz model for sphere-plane contact, as described in [40], and computed at an indentation of 0.5 μm. Samples were probed in five randomly selected zones, with five different points probed in each zone (separated by at least 10 µm in the XY plane), for a total of twenty-five elastic modulus measurements in each sample.

All results shown are representative of at least three independent experiments. Data are represented as the mean ± SEM. When appropriate, statistical analyses were conducted using GraphPad Prism software (GraphPad, La Jolla, CA, USA), and data were analyzed for significance using unpaired t-tests and chi-squared tests. Values of p < 0.05 are marked with one asterisk and of p < 0.01 with two asterisks.

See Additional file 1 Materials and Methods for standard protocols and specific reagents for RNA extraction, reverse transcription and PCR, RNA-seq and alternative splicing analysis, Western blotting, immunofluorescence analysis, immunohistochemistry, collagen imaging, immunoprecipitation assay, chromatin immunoprecipitation, fibroblast activity, deposition of three-dimensional extracellular matrices, invasion assays and statistical analysis. Uncropped gel and blot images are provided with the Additional file 1.

All animal procedures were approved by the Animal Research Ethical Committee from the Parc de Recerca Biomèdica de Barcelona (Barcelona, Spain) and by the Generalitat de Catalunya.

Expression of both the transcription factor SNAIL1 and the fibronectin isoform including the EDA domain in myofibroblasts has been described [20, 39]. To determine whether SNAIL1 is involved in the controlling EDA exon inclusion, we first used RT-PCR to visualize the relative amount of skipping/inclusion isoforms in RNA from control or Snai1-deficient fibroblasts, including CAFs, MEFs and human BJ fibroblasts. Oligonucleotide primers annealing to the flanking EDA exons were used at not saturating cycles. We observed a qualitative switch toward EDA-skipped RNA, confirmed by densitometric quantification of the bands, in Snai1 KO relative to control CAFs (Fig. 1A). As described, we detected that the cytokine TGFβ promoted EDA+ fibronectin RNA expression and that the increase was SNAIL1 dependent (Fig. 1B; Additional file 1: Fig. S1). By analyzing EDA+ fibronectin expression at the protein level using a specific monoclonal antibody, we found that it was only faintly detected in Snai1 KO fibroblasts as compared to control cells (Fig. 1C). In human BJ fibroblasts, depletion of SNAIL1 using a specific siRNA downregulated EDA+ fibronectin levels in both untreated and TGFβ-treated cells (Fig. 1D) indicates that the Snai1-depletion effect was not limited to murine MEFs and CAFs or to the constitutive removal of the factor.

To elucidate whether SNAIL1 extensively controls alternative splicing, we evaluated splicing events by deep sequencing of mRNA from control and Snai1 KO MEFs treated for 3 h with TGFβ. The SUPPA2 and SANJUAN pipelines revealed 674 and 299 significantly different splicing events, respectively, that affected more than 500 genes (Additional file 1: Fig. S2A, Tables S1 and S2). Among these, we found several genes involved in actin cytoskeleton regulation, such as Myl6, AnIn, Macf1, Tpm1, Tpm2, PP1R12A, FlnC and FlnB, as well as ECM genes, such as Col5α1 and FN1; we confirmed some of these genes by RT-PCR (Additional file 1: Fig. S2B). For fibronectin, the number of RNA-seq reads for the events including and excluding EDA allowed us to quantify the relative amount of EDA+ fibronectin in each sample, which was over 75% in control MEFs but was reduced by approximately half in Snai1 KO cells (Fig. 1E).

In primary breast tumors, the presence of myofibroblatic CAFs has been associated with tumor progression [5, 6, 10, 41], and the expression of fibroblastic SNAIL1, with a poor prognosis [10]. Here, we analyzed whether the simultaneous expression of SNAIL1 and EDA+ fibronectin correlates with advanced breast tumors. We collected available SNAIL1 protein data (cBioportal; TCGA, Firehose Legacy) and calculated the percentage of EDA + fibronectin RNA (using the web tool TSVdb for the TCGA splicing variants) from 809 invasive breast carcinomas. We categorized tumors according to their stage (I-II or III-IV) and calculated the relative levels for each molecule using a cutoff value (Material and Methods). Overall, 55% of advanced tumors (stages III and IV) expressing high levels of SNAIL1 also expressed high levels of EDA+ FN1. This percentage decreased to 33–37% in the remaining categories (Fig. 2A). In protein extracts from patient-derived xenografts (PDX) corresponding to HER2 + or triple-negative breast neoplasms, the percentage of tumors with high EDA+ fibronectin expression was increased in the PDXs expressing high levels of SNAIL1 (Fig. 2B). Remarkably, we found similar associations when we analyzed available data from other solid tumors in which the stromal component has been shown to be relevant for tumor progression, such as skin cutaneous melanoma [42], lung adenocarcinoma [43, 44] and kidney renal clear cell carcinoma [45] (Fig. 2B–D). Even though available data do not discriminate between tumor and stroma expression, our analysis shows that the expression of two myofibroblast-associated proteins mostly present in the stroma of colon and breast tumors was associated with advanced tumors.

To further analyze how SNAIL1 controls the EDA inclusion, we focused on SRSF1, a splicing factor involved in controlling this event [34, 35]. Although SRSF1-mediated splicing can be regulated by factor availability [34], we did not detect decreased SRSF1 protein levels or changes in its subcellular localization in TGFβ-activated Snai1 KO relative to control MEFs (Fig. 3A,B). In contrast, we observed that SRSF1 and SNAIL1 colocalized in nuclear granules (Fig. 3B) and co-immunoprecipitated (Fig. 3C), suggesting a more direct molecular connection.

As SRSF1 was described to bind EDA RNA [46], we tested whether SNAIL1 influences SRSF1 binding to exon 33 RNA using RNA immunoprecipitation (RIP) assays in Snai1 KO MEFs. SRSF1 or SNAIL1 was immunoprecipitated with specific antibodies, and the co-precipitating RNA was analyzed. Fibronectin RNA did not significantly co-precipitate with anti-SNAIL1, even in MEFs overexpressing ectopic SNAIL1-HA treated with TGFβ (Fig. 3D). RIP using anti-SRSF1 in these cells confirmed that SRSF1 interacts with EDA RNA, as previously reported (Fig. 3E). Remarkably, the binding of SRSF1 to the EDA RNA region (but not to a control irrelevant region, HPRT) was undetected in Snai1 KO MEFs (Fig. 3E). This result indicates that SNAIL1 is required for the specific binding of SRSF1 to the exon 33 RNA.

Splicing factors of the SRSF family are involved in coupling RNA Pol II transcription to pre-mRNA splicing [47] and precipitate genomic DNA in the presence of crosslinking agents [48]. Therefore, we used ChIP to test whether SRSF1 interacts with the genomic DNA at exon 33. We detected that TGFβ promoted precipitation of this region, but not of the fibronectin proximal promoter or exon 7 regions, with SRSF1 (Fig. 3F), indicating that the splicing factor is likely involved in a co-transcriptional complex. As SNAIL1 interacts with SRSF1, we tested whether SNAIL1 also interacts with the exon 33 genomic region. In contrast to the lack of its binding to RNA, we found that SNAIL1 interacted with the exon 33 in a TGFβ-dependent manner, as well as with the proximal FN1 promoter (as previously described; [49]) (Fig. 3G). Lack of binding to the fibronectin exon 7 region supported the specificity of these ChIP interactions.

Altogether, our data suggest that TGFβ induces localization of SNAIL1 at the alternative splicing region, which is required for the formation of a co-transcriptional complex that includes SRSF1 bound to the EDA RNA. In this case, we would expect that the DNA and RNA binding machineries (including SNAIL1 and SRSF1, respectively) connected by nascent RNA, would be separated by an RNase treatment. Indeed, co-immunoprecipitation of SNAIL1 and SRSF1 was disrupted when input lysates were treated with RNase (Fig. 3C).

As the reduction of the EDA-containing isoform in untreated Snai1 KO MEFs (Fig. 1A–C) could be independent of SRSF1, we tested the levels of other splicing factors involved in the regulation of EDA splicing, such as SFSR3, SRSF5 [20, 50] and QKI [51]. While the levels of the splicing silencer QKI increased in the KO MEFs (Additional file 1: Fig. S4A), no clear changes were observed for SRSF1 (Fig. 3A), SRSF3 or SRSF5 (Additional file 1: Fig. S4B). In contrast, in epithelial cells that did not express endogenous SNAIL1 or EDA+ fibronectin, exogenous expression of SNAIL1-HA promoted the increase in EDA+ fibronectin, as well as of SRSF5 and (to a lesser extent) SRSF3 (Additional file 1: Fig. S5A and B).

Our previous research has shown that the extracellular anisotropy generated by CAF lines correlates with their SNAIL1 levels and that TGFβ-treated fibroblasts produce anisotropic matrices in a SNAIL1-dependent manner [10]. Analogously, we found here that our CAF line but not the Snai1 KO one generates anisotropic three-dimensional extracellular matrices (3D ECMs). In detail, IF analysis showed aligned fibronectin fibers that decrease in Snai1 KO ECMs (Fig. 4A) and angle measurement of DAPI-stained nuclei indicated that the high proportion of oriented nuclei were decreased by half (Fig. 4B).

Moreover, we observed a reduction of the presence of EDA+ linear fibronectin fibers in ECMs from KO CAFs (Fig. 4A). Therefore, we analyzed the contribution of the EDA+ fibronectin on the extracellular topology using TGFβ-activated MEFs derived from two genetically engineered mouse strains that exclusively expressed either EDA-including (EDA+) or -excluding (EDA–) fibronectin isoforms [19]. RNA analysis confirmed the expression of the corresponding isoforms in these cell lines (Fig. 4C). In control MEFs under standard cell culture conditions, the expression of EDA+ fibronectin was predominant (Figs. 1A,B,E and 4C). Proteins corresponding to EDA+ fibronectin were detected only in the control and EDA+ MEFs (Fig. 4D). 3D ECMs from these MEF lines were produced and analyzed by immunofluorescence with an anti-fibronectin antibody recognizing both EDA+ and EDA– isoforms. The absence of EDA+ fibronectin in EDA– 3D ECMs was confirmed by immunofluorescence (Additional file 1: Fig. S6). In the absence of TGFβ, all lines produced randomly oriented fibronectin fibers (Fig. 4E). Fiber orientation was estimated to be over 33% with the OrientationJ plugin (Fig. 4F), as reported for unaligned fibers deposited by control fibroblasts [9, 10]. In the presence of TGFβ, the increase of oriented fibronectin fibers was significantly lower in matrices produced by EDA– MEFs (Fig. 4E,F), indicating that the presence of the EDA within the matrix favors TGFβ-induced aligned polymerization of fibronectin fibers. High-resolution images obtained by STED (stimulated emission depletion) microscopy illustrated the differences in the fiber nets between the three matrices (Fig. 4E).

A higher fiber alignment in EDA+ relative to EDA– matrices (Fig. 4G), as well as a lower curvature and branch points (Fig. 4H), was confirmed with the TWOMBLI plugin for FIJI [52]. Additionally, differences between the control and EDA+ matrices affecting alignment, curvature, branch points, lacunarity and endpoints were also detected (Fig. 4H). As extracellular fibronectin acts as a template to guide the polymerization of other extracellular fibers, we examined whether collagen deposition was dependent on the fibronectin isoforms. At a glance, the collagen pattern was clearly different in the EDA– matrices (Fig. 4I). As collagen organization and crosslinking determine the rigidity of the ECM, we used atomic force microscopy (AFM) to measure the micromechanical properties of decellularized matrices generated by TGFβ-treated MEFs; this revealed a significantly lower elastic modulus in matrices derived from EDA– fibroblasts (Fig. 4J). Therefore, we conclude that the presence of the EDA in fibronectin is necessary for activated fibroblasts to generate matrices with high alignment and rigidity.

The ECM in tumors is deposited in the presence of tumor cells, and matrix-remodeling enzymes, such as metalloproteinases, are activated by signaling resulting from the interactions between tumor cells and fibroblasts [53]. Thus, we evaluated the fibronectin fiber organization in co-cultures of tumor cells and fibroblasts expressing different fibronectin isoforms. In the presence of pre-seeded, Ras-transformed EpH4 breast cells (EpRas), MEFs deposited fibronectin fiber around tumor cell patches visualized as a net of fibers in a fibronectin staining (Fig. 5A). Lacunas between fibers measured by TWOMBLI were larger in co-cultures with MEFs expressing EDA– fibronectin than with the other MEF lines (Fig. 5B). Similar differences were observed in co-cultures with other tumor cells, such as MCF7 (breast) or HepG2 (liver) (Additional file 1: Fig. S7A and S7B). TWOMBLI analyses also detected differences in fiber branchpoints between co-cultures (Additional file 1: Fig. S7C and S7D). Notably, addition of the metalloproteinase inhibitor GM6001 reverted the lacunarity and branchpoints parameters in co-cultures with EDA– fibronectin (Fig. 5A,B, Additional file 1: Fig. S7C).

We also analyzed co-cultures of the colon cancer cells, HT-29 M6, which (unlike other co-cultures) grow in highly compacted colonies that are isolated from fibroblasts. In co-cultures with EDA+ MEFs, colonies of tumor cells were mostly fully encircled by fibroblasts and their fibronectin fibers. In contrast, colonies in co-cultures with EDA– MEFs were surrounded by unoccupied spaces (Fig. 5C), which were estimated with ImageJ to be eight times larger than in colonies from EDA+ MEFs co-cultures (Fig. 5D). EDA– MEFs co-cultures had broken fibronectin fibers that were clearly visualized in three-dimensional reconstructions generated from confocal images (Additional file 1: Fig. S7E), and the empty area was strongly reduced when the metalloproteinase inhibitor GM6001 was included (Fig. 5E). Altogether, these observations indicate that the activity of metalloproteinases affects fibronectin fiber organization in an EDA-dependent manner. Correlating with the fact that Snai1 KO MEFs express lower proportions of EDA+ fibronectin isoforms (Fig. 1), these MEFs also left an empty space around M6 colonies that was rescued by the MMP inhibitor when co-cultured (Fig. 5F).

We next took advantage that our 3D ECM can be decellularized to evaluate the activity of polymerized fibronectin fibers with or without EDA on tumor cell migration and invasion. We used MDA-MB-231 breast tumor cells, which move individually in culture (Additional file 1: Fig. S8A); we found that a higher percentage of these cells achieved oriented displacements on matrices generated by TGFβ-treated EDA+ MEFs as compared to matrices generated by TGFβ-treated EDA– MEFs or non-treated MEFs lines (Fig. 6A). Oriented movements recapitulate the alignment index of the matrices (Fig. 4G). Remarkably, an equivalent trend was obtained in invasion experiments through matrices deposited by these two TGFβ-treated MEF lines (Fig. 6B). Intermediate levels of oriented movements and invading MDA cells were obtained in matrices generated by TGFβ-treated control MEFs (Fig. 6A,B). The EDA requirement was challenged with irigenin, a small molecule that selectively binds to and blocks the EDA-integrin interaction [54]. We found that matrices generated in the presence of irigenin had a low tumor cell invasion, similar to that generated by EDA– MEFs (Fig. 6C). Irigenin treatment interfered with the orientation of the fibronectin fibers and the lacunarity of TGFβ-treated EDA+ matrices, as well with the oriented displacement of MDA cells on these matrices (Additional file 1: Fig. S8B).

We also tested EpRas cells, which formed colonies that moved collectively on decellularized matrices (Additional file 1: Fig. S8C). Of note, colonies on TGFβ-activated EDA+ matrices moved more compactly and coordinately than those on activated EDA– matrices (Additional files 2, 3, 4: S1, S2, and S3). Immunofluorescence analysis also illustrated the difference in compactness between EpRas colonies growing on each substrate (Additional file 1: Fig. S8D), and invasion assays showed that the coordinated collective movement of EpRas colonies on aligned and rigid fibronectin EDA+ matrices was accompanied by a higher capacity to invade the matrix (Fig. 6D).

Fibroblasts are activated by recombinant EDA fragments [31]. Here, we tested whether EDA in extracellular-deposited fibronectin fibers was also effective. Naïve fibroblasts were seeded on decellularized 3D ECMs, and the presence of cytosolic α-SMA stress fibers (a marker of fibroblast activation) was analyzed 16 h later. For both mouse mesenchymal stem cells (MSC) and NIH3T3 fibroblasts, the percentage of cells with α-SMA–positive fibers was higher on matrices derived from EDA+ MEFs than EDA– MEFs (Fig. 7A; Additional file 1: Fig. S9).

Matrices deposited by MEF EDA+ in either the presence or absence of TGFβ stimulated NIH3T3 (Fig. 7B); however, matrices deposited by control MEFs stimulated fibroblasts only if generated in the presence of the TGFβ cytokine (Fig. 7B). Given the elevated EDA+ fibronectin ratio generated and deposited by control MEFs (Figs. 1 and 4), our result suggests a TGFβ-induced conformational change that exposes polymerized EDA to fibroblasts.

To validate that stimulation of naïve fibroblasts depends on EDA in our experimental approach, fibroblast stimulation was challenged with irigenin. Addition of this compound prevented the formation of α-SMA–positive fibers on matrices deposited by TGFβ-treated EDA+ or control MEFs (Fig. 7C), further supporting the idea of an EDA-dependent stimulatory action on fibroblasts.

Our results indicated that the EDA+ fibronectin is a critical structural element for the assembly of an invasion-permissive extracellular matrix. Therefore, we evaluated the action of EDA+ fibronectin in an orthotopic model of metastatic breast cancer. We injected AT3 tumor cells and EDA– or EDA+ MEFs into the inguinal mammary fat pads of NOD-SCID gamma mice. Tumors were monitored and resected simultaneously when they reached 0.2–0.4 cm of diameter. The volume of extracted tumors was accurately measured, and the tumors were then fixed for immunohistological analyses (Additional file 1: Fig. S10A). Tumors generated in the presence of EDA+ fibronectin–expressing MEFs were larger than those generated in the presence of EDA– fibronectin–expressing MEFs (Fig. 8A; Additional file 1: Fig. S10B).

After surgical resection of the primary tumors, mice were kept alive for 7 weeks to allow for the growth of lung metastasis. The lungs were then extracted, and hematoxylin and eosin staining was performed to study the presence of metastatic foci (Additional file 1: Fig. S10B). While over 60% of the animals injected with AT3 plus EDA+ MEFs developed metastatic foci, lungs of mice injected with AT3 plus EDA– MEFs were free of metastasis (Fig. 8B), clearly demonstrating the critical role of the fibronectin EDA in tumor malignant progression.