P-cadherin-induced decorin secretion is required for collagen fiber alignment and directional collective cell migration

Directional collective cell migration (DCCM) is characterized by groups of cells that migrate in a coordinated manner, and is a key process during morphogenesis, regeneration and cancer invasion (Friedl and Gilmour, 2009; Friedl et al., 2012; Haeger et al., 2014). DCCM of mesenchymal and epithelial cells is observed during embryo development and also during carcinoma and sarcoma cell invasion (Scarpa and Mayor, 2016; Friedl et al., 2012; Friedl and Gilmour, 2009; Gaggioli et al., 2007). The extracellular matrix (ECM) provides a physical scaffold for cell adhesion and migration by acting on cell tension and by activating signaling pathways through ECM receptors (Egeblad et al., 2010).

During mesenchymal and epithelial DCCM, cadherin-mediated cell–cell adhesion plays an essential role in maintaining cell–cell cohesion and also in allowing mechanosignaling. This is a complex but crucial signaling process elicited at cell–cell contact sites and leading to the generation of traction forces that promote migration. P-cadherin (also known as CDH3), which is overexpressed in epithelial (breast, ovarian, prostate, endometrial, skin, gastric, pancreas and colon) and mesenchymal tumors (rhabdomyosarcomas) with important tumor-promoting effects (van Roy, 2014; Thuault et al., 2013; Vieira and Paredes, 2015), induces DCCM through activation of the β-Pix (also known as ARHGEF7)/CDC42 polarity axis (Halbleib and Nelson, 2006; Plutoni et al., 2016). This polarity regulatory pathway is also involved in DCCM of astrocytes (Cau and Hall, 2005; Osmani et al., 2006) and of anterior visceral endoderm cells in early mouse embryos (Omelchenko et al., 2014). P-cadherin expression predicts the level of intercellular tension in epithelial and mesenchymal cells, and increases intercellular stress anisotropy (Bazellières et al., 2015; Plutoni et al., 2016). This mode of local cell guidance, called plithotaxis, allows the efficient translocation of the entire cell layer, because cells are aligned and migrate along the direction of transmitted stresses. Importantly, this is associated with the development of traction forces that counterbalance the intercellular forces (Maruthamuthu et al., 2011) and drive the cell layer movement. Traction forces are the cell migration engine, and cadherin-based intercellular adhesion stimulates cell–ECM traction forces (Jasaitis et al., 2012; Plutoni et al., 2016; Mertz et al., 2013). However, it is largely unknown how cadherin-mediated mechanical coupling between migrating cells influences force transmission to the ECM. Cadherin mechanosignaling-dependent processes could induce changes in ECM organization that modify ECM–cell interactions in favor of efficient traction force generation. Here, to determine the mechanism of P-cadherin role in DCCM, first we performed a transcriptomic analysis of mesenchymal C2C12 cells that overexpress P-cadherin and the parental line. We found that the matrisome was the main affected group and that Dcn, the gene encoding the collagen remodeling molecule decorin, was among the most upregulated genes in mesenchymal C2C12 cells that overexpress P-cadherin compared with the parental line. We then confirmed that P-cadherin expression in C2C12 myoblasts induces decorin expression and secretion, and that decorin is required for P-cadherin-induced DCCM. By performing 2D and 3D migration assays, we found that P-cadherin expression promotes type I collagen fiber alignment in the direction of migration, in a decorin-dependent manner. We then showed that, in breast cancer specimens, collagen fiber alignment is a prominent feature of P-cadherin/decorin-expressing cells at invasion sites. Moreover, decorin is required for P-cadherin-mediated β1 integrin and CDC42 activation, via activating threonine dephosphorylation of β-Pix. Finally, through micro-patterned cell monolayers and traction forces microscopy, we demonstrated that decorin is required for P-cadherin-mediated traction force increase and anisotropy.

Our results show that P-cadherin acts a key mediator of DCCM through a new signaling pathway that induces decorin upregulation leading to collagen fiber orientation in the direction of cell migration, and activation of β1 integrin and of the β-Pix/CDC42 axis.

To identify the underlying molecular mechanisms of P-cadherin-induced DCCM, we compared gene expression, by performing a transcriptome analysis, in C2C12 myoblasts that express P-cadherin (C2C12 Pcad) and control C2C12 cells that express only the LZRS vector (C2C12 LZRS; herein denoted C2C12 CTL). We identified 176 genes that were significantly up- or down-regulated (fold change >3) upon P-cadherin expression (Table S1). Gene ontology (GO) cluster analysis of these genes using PANTHER and GORILLA revealed that P-cadherin expression was associated with a transcriptomic signature of ECM, and particularly of collagen (Fig. 1A; Tables S2 to S5). We also compared the list of genes differentially modified between C2C12 CTL and C2C12 Pcad cell samples to the mouse matrisome signature established in a meta-analysis combining ECM proteomics data from 14 different tissues and tumors (Naba et al., 2016). One third of the modified genes upon P-cadherin expression belonged to the mouse matrisome signature (Fig. 1B). Interestingly, the gene encoding decorin, a collagen-remodeling protein, was one of the five most upregulated genes upon P-cadherin expression (Fig. 1C). We confirmed that decorin mRNA and protein levels were increased in C2C12 Pcad cells compared with C2C12 CTL cells (Fig. 1D,E). Similarly, decorin secretion into the medium was higher in C2C12 Pcad than in C2C12 CTL cells (Fig. 1F), particularly the core protein (38 kDa) that binds to fibrillar collagen (Orgel et al., 2009). This effect was P-cadherin specific because decorin was not upregulated in E-cadherin- (C2C12 Ecad) and R-cadherin-expressing cells (C2C12 Rcad) (Fig. 1D,E). Moreover, P-cadherin-mediated upregulation of decorin required P-cadherin homotypic interaction, because decorin expression was not induced in isolated (non-confluent) C2C12 Pcad cells (Fig. S1A). These results indicate that P-cadherin expression promotes decorin expression and secretion.

Decorin is a class I member of the small leucine-rich proteoglycan family and has a single glycosaminoglycan chain composed of dermatan or chondroitin sulfate with 12 leucine-rich tandem repeats (Iozzo and Schaefer, 2015). Decorin core protein binds to type 1 collagen fibrils (Scott and Orford, 1981; Pogány and Vogel, 1992), and its GAG chain extends laterally from adjacent collagen fibrils to maintain interfibrillar spacing (Scott, 1988; Weber et al., 1996). To understand decorin role in P-cadherin-induced DCCM, we generated stable C2C12 Pcad cell lines that express anti-Dcn shRNA (C2C12 Pcad Dcn shRNA) by retroviral infection. This reduced decorin protein and mRNA expression to the same level as in control C2C12 CTL cells (Fig. 1D,E). By combining time-lapse microscopy, cell tracking and computational analyses, we monitored and quantified cell movements in a 2D migration assay. These experiments showed that after removal of the physical barrier, Dcn silencing (C2C12 Pcad Dcn shRNA cells) hindered the P-cadherin-dependent increase in cell velocity and migration persistence observed in C2C12 Pcad cells (Fig. 2A–C; Movie 1). Moreover, analysis of the trajectory angle distribution revealed that decorin expression was needed for P-cadherin-induced directionality increase of individual migrating cells (Fig. 2D). To map cell velocities within entire layers of cells, we used particle image velocimetry analysis (Petitjean et al., 2010). We found that Dcn silencing (C2C12 Pcad Dcn shRNA cells) decreased the orientation in the direction of migration of velocity fields and their angle distribution compared with C2C12 Pcad cells (Fig. 2E). As P-cadherin expression increases intercellular stress and plithotaxis within the monolayer during cell migration (Plutoni et al., 2016), we investigated whether decorin contributes to this mechanism. Dcn silencing prevented P-cadherin-dependent increase of intercellular stress (Fig. 2F) and plithotaxis (Fig. 2G). Moreover, in C2C12 Pcad Dcn shRNA cells that stably expressed human decorin (insensitive to the mouse shRNA; C2C12 Pcad Dcn shRNA Rescue cells), all the examined parameters were similar to those of C2C12 Pcad cells. Decorin protein level reached the level of C2C12 Pcad cells (Fig. 1D), and P-cadherin accumulated at cell–cell contacts in both C2C12 Pcad Dcn shRNA and C2C12 Pcad Dcn shRNA Rescue cells (Fig. S1B). As decorin binds to growth factors (Järvinen and Prince, 2015), we also analyzed cell proliferation and global tyrosine phosphorylation levels in C2C12 CTL, Pcad and Pcad Dcn shRNA cells (Fig. S1C), but did not find any differences.

We next analyzed cell organization and cell polarization in the different cell lines. P-cadherin expression induced the formation of a large and oriented protrusion in the direction of migration. The protrusion orientation, but not length was impaired by Dcn silencing and was restored in C2C12 Pcad Dcn shRNA Rescue cells (Fig. 3A). Next, we monitored cell polarity by quantifying the position of the centrosome and Golgi relative to the nucleus (Yadav et al., 2009), and found that decorin was required for P-cadherin-induced cell polarization. Indeed, the orientation of the centrosome and Golgi in front of the nucleus toward the protruding edges was perturbed in C2C12 Pcad Dcn shRNA cells, but not in C2C12 Pcad Dcn shRNA Rescue cells (Fig. 3B). Finally, we analyzed focal adhesion (FA) organization by paxillin immunostaining (Zaidel-Bar et al., 2003). The FA orientation in the direction of migration observed in C2C12 Pcad cells was completely abolished by Dcn silencing and restored by human decorin expression (Fig. 3C). Analysis of global CDC42 activation using a pulldown assay revealed that P-cadherin-induced CDC42 activation was lost upon Dcn silencing (Fig. 3D). Taken together, these results indicate that decorin is required for P-cadherin-mediated CDC42 activation, and orientation of the membrane protrusion, cell organelles and FAs in the migration direction.

Dcn^−/− mice show defects in collagen fibril formation and organization (Danielson et al., 1997), and it has been suggested that decorin could regulate FA reorganization and endothelial cell motility on collagen I (Fiedler et al., 2008). Therefore, we investigated whether P-cadherin-induced decorin expression could affect type I collagen fiber organization.

First, we embedded C2C12 cell spheroids in type I collagen matrix, and monitored 3D cell invasion through the matrix by time-lapse phase-contrast microscopy. After 3 days, C2C12 CTL and C2C12 Pcad cells evaded from the spheroid to colonize the surrounding matrix (Fig. S2A,B). The precise morphological analysis indicated that C2C12 CTL spheroids migrated as an elongated group of cells that detached one from the other during migration (Fig. S2D, and quantification of the number of cells that detached from the spheroids during the 72 h migration in Fig. S2C). Conversely, C2C12 Pcad cells migrated as a collective group of cells that maintained intercellular contacts and did not detach (Fig. S2C,D). Dcn silencing strongly inhibited the invasive potential of C2C12 Pcad cells in the 3D collagen matrix (Fig. S2A,B). This effect was specific to P-cadherin expression because C2C12 Ecad and Rcad cells did not migrate in the 3D collagen matrix.

Then, we analyzed type I collagen fiber organization in the 3D matrix surrounding the cell spheroids using second harmonic generation (SHG) microscopy (Fig. 4A). Cells of the spheroid were revealed by Rhodamine–phalloidin staining. After 72 h, type I collagen fibers assembled as a complex network of fibers that formed intricate meshes. When cells started to invade the 3D collagen matrix, collagen fibers were re-organized in C2C12 CTL cells, and particularly in C2C12 Pcad cells, where cell migration was associated with massive collagen fiber alignment radially from the spheroid in the direction of cell invasion. Quantification of directional collagen fiber orientation, as determined through a Fourier component analysis of directionality on images obtained by second-harmonic generation (SHG) microscopy, showed that P-cadherin, but not E- or R-cadherin, expression promoted collagen fiber orientation towards the direction of migration (Fig. 4B,C). This effect was impaired by Dcn silencing.

By using micro-patterned cell monolayers and scanning electron microscopy (SEM), we analyzed collagen fiber organization during 2D migration on type I collagen for 6 h (Fig. 4D). Again, P-cadherin expression was associated with collagen fiber organization and alignment in the migration direction. In control (C2C12 CTL) cells, collagen fibers overlapped and interweaved. Conversely, in C2C12 P-cad cells, collagen fibrils were organized and aligned in the direction of migration. Dcn silencing impaired P-cadherin-mediated collagen fiber orientation during migration, while human decorin expression restored collagen fiber orientation in the migration direction. Quantification of directional collagen fiber orientation, through Fourier component analysis of directionality on SEM images, revealed that P-cadherin, but not E- or R-cadherin, expression promoted collagen fiber orientation towards the direction of migration in a decorin-dependent manner (Fig. 4E).

Finally, we assessed whether decorin is expressed in invasive breast tumors in which microdensity increase has been associated with higher collagen density (Skandalis et al., 2011). We did not observe DCN mRNA upregulation in a cohort of 527 breast tumor samples, most probably because decorin is expressed both by normal and tumoral cells (Fig. 4F, left panel, the asterisk shows a normal acinus surrounded by tumoral cells). However, low DCN mRNA level was significantly associated with a better prognosis in two breast tumor types [triple negative and hormone receptor (HR)⁻, ERBB2⁺] (Fig. S3A). The analysis of CDH3 and DCN expression in different morphological structures of breast tumors revealed that their highest levels were found in cells presenting a morphological organization characteristic of invasive tumor cells (Fig. S4D). In human breast tumor samples, decorin protein was expressed by cytokeratin 7 (CK7, also known as KRT7)-positive tumor cells and, to a lower extent, by α-smooth muscle actin (α-SMA)-positive stromal cells (Fig. 4F,G; Table S6). Moreover, immunohistochemistry of serial breast tumor tissue sections allowed identifying migrating tumor cells that co-expressed P-cadherin and decorin and that were surrounded by tracks of aligned collagen fibers (Fig. 4H; Table S6). We also found that specific CDH3:DCN ratios were associated with poor disease-free survival in patients with HR⁺/ERBB2⁻ and HR⁻/ERBB2⁺ breast tumors (Fig. S3B).

Altogether, these results obtained in 2D and 3D migration assays in type I collagen matrix indicate that P-cadherin expression promotes collagen fiber alignment in the direction of migration and that decorin is required for this process. Moreover, collagen fibers form parallel fascicles that surround P-cadherin and decorin-expressing breast tumor cells in invasive regions.

Fibrillar collagen activates the β-Pix/CDC42 axis through β1 integrin (Kutys and Yamada, 2014), and P-cadherin regulates DCCM through β-Pix-mediated CDC42 activation (Plutoni et al., 2016). Therefore, we assessed whether P-cadherin-mediated decorin expression regulates β1 integrin activation during DCCM. To this aim, we plated cells on a 2D collagen matrix because P-cadherin induces collagen fiber alignment through decorin both in 2D and 3D (Fig. 4), and because 2D cell migration along fibers recapitulates migration in a 3D matrix (Doyle et al., 2009). The level of active β1 integrin was high in C2C12 Pcad cells, but not in C2C12 Ecad and Rcad cells, whereas the total β1 integrin level was comparable in all tested cell lines (Fig. 5A; Fig. S5A). Decorin addition, known to increase C2C12 myoblast migration persistence (Goetsch et al., 2011), was associated with an increase of active β1 integrin. Conversely, the level of active β1 integrin was strongly decreased in C2C12 Pcad Dcn shRNA cells, and was restored by human decorin expression (Fig. 5A). We then used the soluble arginine-glycine-aspartic acid (RGD) peptide, which inhibits β1 integrin in our experimental setting (Fig. S5B). We showed that RGD peptide addition affected P-cadherin-induced persistence of migration and cell polarization to the same extent as Dcn silencing (Fig. 5B,C; compare with Figs 2A and 3B). Moreover, RGD peptide addition to the collagen type I impaired 3D matrix invasion by C2C12 Pcad cells (Fig. 5D). Nevertheless, in this condition, we still observed collagen reorganization in the areas around the spheroid, but to a lower extent than with C2C12 Pcad migrating cells (Fig. 5E). This indicates that efficient collagen fiber reorganization requires both decorin secretion and integrin-mediated adhesion. We then asked whether P-cadherin-mediated decorin expression was involved in the activation of the β1 integrin/β-Pix/CDC42 signaling axis. Incubation of C2C12 Pcad cells with the RGD peptide decreased CDC42 activation, indicating that integrins are activated downstream of P-cadherin to allow CDC42 activation (Fig. 5F). As β-Pix must be dephosphorylated at threonine 526 to induce CDC42 activation (Kutys and Yamada, 2014), we monitored β-Pix threonine phosphorylation during migration after DCCM in C2C12 Pcad cells, control C2C12 CTL and C2C12 Pcad Dcn shRNA cells. P-cadherin expression was associated with lower β-Pix threonine phosphorylation (Fig. 5G) and concomitant increase of CDC42 activity (Fig. 3D). Conversely, in C2C12 Pcad Dcn shRNA cells, the β-Pix threonine phosphorylation level was comparable to that of control C2C12 CTL cells (Fig. 5G), while CDC42 activation was markedly reduced (Fig. 3D).

Finally, using micro-patterned cell monolayers and traction-force microscopy, we investigated whether decorin participates in P-cadherin-mediated traction force strength and polarization exerted by the cells on the underlying substrate that drive the cell layer movement. We measured the traction forces parallel (Tx, Movie 2) and perpendicular (Ty) to the direction of migration and calculated the Tx:Ty ratio. As previously shown (Plutoni et al., 2016), P-cadherin expression increased the Tx:Ty ratio. The Tx:Ty ratio was reduced in C2C12 Pcad Dcn shRNA cells (Fig. 5H) and restored in C2C12 Pcad Dcn shRNA Rescue cells. Analysis of the traction force orientation from 0 to 40 µm and from 60 to 140 µm from the leading multicellular row toward the center of the cell layer showed that P-cadherin-induced traction force orientation towards the direction of migration was impaired in C2C12 Pcad Dcn shRNA cells (Fig. 5I; Fig. S5C) and restored in C2C12 Pcad Dcn shRNA Rescue cells.

Taken together, these data demonstrate that P-cadherin-mediated decorin expression specifically activates β1 integrin and decreases β-Pix threonine phosphorylation. This promotes CDC42 activation, leading to an increase in traction force strength and anisotropy towards the migration direction, which are processes required for DCCM.

In this study, we identified an original mechanism through which cadherin-mediated mechanical coupling between migrating cells results in force transmission to the ECM to promote cell migration. P-cadherin, a major DCCM inducer, increases the traction force anisotropy and strength that pull the cell layer (Bazellières et al., 2015; Ng et al., 2012; Plutoni et al., 2016). We demonstrated that P-cadherin, but not E- or R-cadherin expression, is associated with collagen fiber organization and alignment in the migration direction and identified decorin, a secreted proteoglycan, as a key player in this process. Collagen fiber alignment facilitates cancer cell invasion by providing tracks on which cells migrate and has a bad prognostic value (Conklin et al., 2011; Drifka et al., 2016; Gjorevski et al., 2015; Levental et al., 2009; Lyons et al., 2011; Provenzano et al., 2008; Riching et al., 2014). Conversely, inhibition of collagen fibril alignment, without affecting ECM composition, blocks tumor cell invasion (Grossman et al., 2016). Here, we experimentally showed that P-cadherin expression specifically promotes collagen fiber alignment concomitantly with polarization of the front protrusion and of FAs in the migration direction and with β1 integrin activation. This is consistent with the observation that FAs are localized along collagen fibers (Wolf et al., 2007). The reorganization of the collagen ECM leads to local stiffening, which enhances integrin-dependent mechanotransduction. Using a mathematical model that takes into account matrix ligand density, stiffness and alignment, it was previously proposed that the increased migration persistence is due to ligand presentation along an axis allowing cells to form and stabilize FAs in a given direction, leading to more efficient migration (Riching et al., 2014). A recent report showed that cadherin-11-mediated cell–cell adhesion regulates the synthesis of ECM proteins, such as collagen and elastin (Row et al., 2016). Our identification of an ECM signature associated with P-cadherin expression also supports the hypothesis that cadherin-mediated adhesion could regulate both ECM production and organization.

We also showed that P-cadherin-induced collagen fiber alignment is mediated by decorin, a secreted proteoglycan that binds to type I collagen fibers and promotes and regulates collagen fibril formation and stability (Chen and Birk, 2013; Kalamajski and Oldberg, 2010; Neame et al., 2000). Dcn knock-out in mice leads to perturbed collagen fiber morphology and skin fragility (Danielson et al., 1997). Decorin is known to be involved in several cellular functions, such as morphological changes, cell migration, cell signaling and autophagy (Gubbiotti et al., 2015, 2016; Tufvesson, 2003; Goetsch et al., 2011). Decorin is detected in the tumor microenvironment and can either suppress tumor growth (Bozoky et al., 2014; Csordás et al., 2000; Santra et al., 2000) or promote tumor invasiveness, metastasis and angiogenesis (Dil and Banerjee, 2012, 2011; Banerjee et al., 2003; El Behi et al., 2013; Cawthorn et al., 2012). In two breast tumor types (triple negative and HR⁻/ERBB2⁺), we found that low DCN mRNA expression level is associated with better prognosis. Decorin expression has also been associated with poor prognosis in mammary and ovarian tumors (Cawthorn et al., 2012; Newton et al., 2006) and shown to have the highest expression in the most invasive breast tumor cells (Denisov et al., 2017). It is also correlated with the formation of a collagenous-rich stroma associated with microdensity, and is detected in the tumor stroma of non-palpable breast carcinoma specimens with high microdensity and micro-calcifications (Skandalis et al., 2011). Our transcriptomic analysis showed that Dcn is strongly upregulated upon P-cadherin expression. We confirmed that only P-cadherin (and not E- or R-cadherin) promotes decorin expression and secretion. This P-cadherin–decorin link is also observed in the MCF10A mammary cell line (Fig. S4A–C) and in discrete groups of invasive breast tumor cells (Fig. S4D) and is also supported by the observation that disease-free survival is poor in patients with HR⁺/ERBB2⁻ and HR⁻/ERBB2⁺ breast tumors with specific CDH3:DCN ratios. Moreover, by analyzing transcriptomic and proteomic data, we identified and validated interesting candidate signaling pathways activated in P-cadherin-expressing cells (i.e. the ERK, p38 and JNK MAPK signaling pathways, and the NFκB pathway; Fig. S6). Their inhibition led to a strong decrease of P-cadherin-induced persistence of migration and of Dcn mRNA expression in P-cadherin expressing cells (Fig. S6). Dcn silencing inhibited P-cadherin-induced collagen alignment and DCCM, polarization of the front protrusion and of FAs, as well as β1 integrin activation and traction force generation (Fig. 6). In clinical mammary invasive carcinoma samples, we observed aligned collagen fibers at invasion sites that were adjacent to P-cadherin- and decorin-expressing cells. This suggests that collagen fiber alignment could contribute to P-cadherin/decorin-positive cancer cell dissemination. It has been often reported that cells in the tumor microenvironment, such as cancer-associated fibroblasts, contribute to ECM molecule secretion and ECM fiber alignment (Erdogan et al., 2017). In the invasive mammary tumor samples we analyzed, decorin was expressed by tumor cells and, to a lesser extent, by cells in the tumor microenvironment (Table S6). Decorin binds not only to collagen I, but also to collagens II–VI, XII and XIV and other proteins present in the ECM, such as growth factors (Gubbiotti et al., 2016). Therefore, besides promoting collagen fiber alignment, decorin could also sequester these growth factors along the fibers to limit ligand binding in the direction of migration or to contribute to integrin stabilization for efficient migration.

During DCCM, activation of the CDC42-mediated polarity pathway is crucial for the polarization of cells, trajectories, membrane protrusions and FAs in the migration direction (Cau and Hall, 2005; Osmani et al., 2006; Plutoni et al., 2016). FA polarization in the migration direction allows force-vector orientation, resulting in higher Tx:Ty ratios and efficient DCCM. β-Pix, a guanine nucleotide exchange factor for CDC42, is involved in P-cadherin-mediated DCCM (Plutoni et al., 2016), and is activated only during migration on collagen following threonine dephosphorylation (Kutys and Yamada, 2014). We demonstrated that decorin-mediated collagen fiber remodeling is important for the activation of the β-Pix/CDC42 axis, and is required for the regulation of the polarity axis and force anisotropy that drive P-cadherin-mediated DCCM. As CDC42 activation in P-cadherin-expressing cells is integrin dependent, and β1 integrin is specifically activated by P-cadherin in a decorin-dependent manner, one could hypothesize that collagen fiber alignment is an outside-in signal for β1 integrin activation. Through this mechanism, collagen fiber alignment increases traction forces that could participate in a feedback loop to regulate collagen fiber alignment in the direction of migration.

In conclusion, our data demonstrate that collagen fiber alignment mediated by P-cadherin-induced decorin directly controls DCCM driven by the orientation and directionality of collagen fibers to promote contact guidance. This is important because in vivo data indicate that, in breast cancer, aligned collagen fibers in the tumor microenvironment are associated with poor prognosis. At the molecular level, collagen fiber alignment allows β1 integrin and β-Pix/CDC42 axis activation, leading to traction force generation and anisotropy that promote DCCM.

Our data suggest that decorin promotes cell migration and invasion through its role on collagen fiber alignment. This implies that anti-tumor strategies that increase decorin expression should be carefully evaluated before their transfer to the clinic.

All cell lines were authenticated and tested for contamination. C2C12 mouse myoblasts (American Type Culture Collection, ATCC CRL-1772) cell lines expressing the LZRS empty vector (C2C12 CTL), P-cadherin (C2C12 Pcad), E-cadherin (C2C12 Ecad) or R-cadherin (C2C12 Rcad) were grown in DMEM supplemented with 10% fetal calf serum as described in Plutoni et al. (2016) and Thuault et al. (2013). C2C12 CTL, C2C12 Pcad, C2C12 Ecad, and C2C12 Rcad cells that stably express Dcn shRNAs were produced using the retroviral vector pSIREN-RetroQ. The pSIREN-RetroQ vector that targets the firefly luciferase (Luci shRNA) was used as control. Exponentially growing C2C12 myoblasts (5×10⁵ cells per 60-mm dish) were infected with 5 ml viral supernatant. Stably transfected cells were selected with 1 mg/ml G418 or 1 mg/ml puromycin (shRNAs), and different clones were isolated by limiting dilution. For rescue experiments, C2C12 Pcad Dcn shRNA cells were infected with the retroviral vector pBABE-Hygro-hDecorin ‘shRNAresist’ (human decorin; NCBI Reference Sequence accession no. NM_001920.4) and different clones were isolated by limiting dilution after selection with hygromycin (200 µg/ml).

shRNA constructs were generated using the retroviral vector pSIREN-RetroQ according to the manufacturer's protocol (BD Biosciences, San Diego, CA). To suppress endogenous Dcn expression, the annealed double-strand oligonucleotides 5′GATCCGGCATCTCAGACACCAACATAATTCAAGAGATTATGTTGGTGTCTGAGATGC CTTTTTTACGCGTG 3′ (forward) and 5′-AATTCACGCGTAAAAAAG GCATCTCAGACACCAACATAATCTCTTGAATTATGTTGGTGTCTGAGATGCCG-3′ (reverse) (Dcn shRNA2) and 5′-GATCCGGCCGAACTCTAGCAATGTATTTTCAAGAGAAATACATTGCTAGAGTTCGGCCTTTTTTACGCGTG-3′ (forward) and 5′-AATTCACGCGTAAAAAAG GCCGAACTCTAGCAATGTATTTCTCTTGAAAATACATTGCTAGAGTTCGGCCG-3′ (reverse) (Dcn shRNA3) were inserted in the RNAi-Ready pSIREN-RetroQ vector (Clontech, Mountain View, CA). Bold letters correspond to positions 902–923 and 1725–1746 of the mouse decorin cDNA sequence, NM_001190451.2). The control luciferase shRNA was made by inserting the oligonucleotide 5′-GATCCGTGCGTTGCTAGTACCAACTTCAAGAGAGTTGGTACTAGCAACGCACTTTTTTGCTAGCGAATTC-3′. Bold letters correspond to oligonucleotide 1310–1328 of the Photinus pyralis (firefly) luciferase sequence (GenBank accession no. M15077.1).

Affymetrix microarrays were processed at the Institute in Regenerative Medicine and Biotherapy Microarray Core Facility, CHU-INSERM-UM Montpellier (http://www.chu-montpellier.fr/fr/irmb/). Briefly, biotinylated cRNAs from control C2C12 CTL cells and C2C12 Pcad cells (clones #2 and #9; Thuault et al., 2013) were prepared according to the Affymetrix GeneAtlas IVT Express protocol from 100 ng of total RNA (GeneAtlas 3'IVT ExpressTechnical Manual, 2010, P/N702833 Rev4 Affymetrix). After fragmentation, 10 µg of cRNA were hybridized at 45°C on the Affymetrix^® MG-430PM Array for 16 h. StripChips were washed and stained on the fluidic station of the Affymetrix GeneAtlas system with the HWS kit. GeneChips were scanned using the Affymetrix GeneAtlas scanner and data were generated with the Affymetrix Expression Console v 1.2.1 and the GCRMA algorithm. Microarray data has been deposited in ArrayExpress under accession no. E-MTAB-8479.

The PANTHER (protein annotation through evolutionary relationship, version 14.0) classification system (http://www.pantherdb.org/) was used to perform a Gene Ontology (GO) enrichment analysis. Enriched terms were considered statistically significant when p_adj was <0.05, and a minimum of five genes were grouped for each significant term. GO terms were categorized in three major functional groups: biological process, molecular function and cellular component.

The Gene Ontology enRIchment anaLysis and visuaLizAtion tool (GORILLA) classification system (http://cbl-gorilla.cs.technion.ac.il/) was used to perform a GO term enrichment visualization. GO terms were categorized in three major functional groups: biological process, molecular function and cellular component. The relationship between GO terms is described by a directed acyclic graph (DAG).

Total RNA was extracted from the different cell lines using the RNeasy Mini Kit (Qiagen) according to the manufacturer's instructions. Then, RNA was reverse transcribed to cDNA using SuperScript II Reverse Transcriptase (Thermo Fisher Scientific) and real-time PCR was performed using a LightCycler 480 instrument (Roche) with the SYBR Green PCR Master Mix (Roche). The primer sequences were as follows: Dcn, forward, 5′-TTCCTACTCGGCTGTGAGTC-3′ and reverse, 5′-AAGTTGAATGGCAGAACGC-3′; and, Rpl32, forward, 5′-TTAAGCGAAACTGGCGGAAAC-3′ and reverse, 5′-TTGTTGCTCCCATAACCGATG-3′.

Mouse antibodies were against: P-cadherin (1:1000, 32-40000, Invitrogen), CDC42 (1:300, 610928, BD Transduction Laboratories), β-Pix (1:1000, 611648, BD Transduction Laboratories), paxillin (1:500, 610051, BD Transduction Laboratories) and β-actin (1:1000, 611648, BD Transduction Laboratories). A rabbit monoclonal antibody was used against phosphorylated threonine residues (1:100, 8954, Cell Signaling Technology). Rabbit polyclonal antibodies were used against pericentrin (1:150, PRB-432C, Eurogentec) and decorin (1:1000 for western blotting and 1:100 for immunofluorescence, LF-113 provided by Larry Fisher, NIH, Bethesda, MD). A rabbit polyclonal antibody against the SH3 domain of β-Pix (aa 1–65) was provided by Nathalie Morin (CRBM, 34293 Montpellier, France). F-actin was detected with Rhodamine–phalloidin (1:10,000, P1951, Sigma-Aldrich); nuclei were stained with Hoechst 33342 (1:2000, B2261, Sigma-Aldrich), The Golgi was detected using Alexa Fluor 488–lectin conjugates from Helix pomatia (L11271, Molecular Probes), activated β1 integrin was detected with the rat anti-mouse CD29 (1:100, 553715, Clone 9EG7, BD Transduction Laboratories) and β1 integrin with the clone MB1.2 antibody (1:100, mab1997, Millipore). Primary antibodies were detected with Alexa Fluor 488- or 546-conjugated goat anti-mouse or anti-rabbit IgG antibodies (Molecular Probes). The β1 integrin inhibitor (RGD peptide: N-CGPKGDRGDAGPKGA-C from GENECUST) was used at 100 µM and decorin (from bovine articular cartilage D8428 Sigma-Aldrich) at 10 µg/ml. After insert removal, cells were incubated with fresh medium containing the compounds and renewed every 6 h.

Dcn and P-cadherin mRNA levels in tumor samples from a cohort of patients with breast cancer were analyzed as described previously (Rosse et al., 2014).

Patients (n=15) with breast carcinoma were treated at the Cancer Research Institute, Tomsk NRMC (Tomsk, Russia) (Tables S6 and S7). The histological type was defined according to the World Health Organization recommendations (Lakhani et al., 2012). Tumor grade was determined using the Bloom and Richardson grading system (Bloom and Richardson, 1957). The expression of estrogen receptor (ER) and progesterone receptor (PR) was scored using the HSCORE method (Kinsel et al., 1989). HER2 expression was analyzed by immunohistochemistry and calculated on a scale 0–3+, according to the ASCO/CAP guidelines (Wolff et al., 2007). Ki-67 expression was calculated as the percentage of Ki-67-positive cells relative to all cells. Molecular subtypes were categorized on the basis of the primary tumor ER, PR, HER2, and Ki-67 status, according to the St Gallen recommendations (Wolff et al., 2007): luminal A (ER⁺ and/or PR⁺, HER2⁻, and Ki-67<20%), luminal B (ER⁺ and/or PR⁺, HER2^+/−, and Ki-67≥20%), HER2⁺ (ER⁻ and PR⁻, HER2⁺), and triple-negative (ER⁻, PR⁻, HER2⁻).

Frozen and formalin-fixed paraffin-embedded (FFPE) tumor tissue specimens obtained during surgery were used for immunohistochemical and immunofluorescence analyses.

The study procedures were in accordance with the Helsinki Declaration (1964, amended in 1975 and 1983). This study was approved by the institutional review board; all patients signed an informed consent for voluntary participation.

Immunohistochemical staining was used for the standard scorings (ER, PR, HER2 and Ki-67) and to assess P-cadherin, decorin, type I collagen and CK7 expression using primary mouse antibodies against ER (IR657, clone 1D5, ready-to-use, Dako, Denmark), PR (IR068, clone PgR636, ready-to-use, Dako, Denmark), Ki-67 (M306, clone SP6, ready-to-use, Spring Bioscience), and primary rabbit antibodies against HER2 (A0485, 1:800, Dako, Denmark), P-cadherin (1:50, in-house rabbit antibody; Thuault et al., 2013), decorin (1:100, LF113, from Larry Fisher) and type I collagen (1:100, 20111, Novotec). Immunohistochemical staining was performed as previously described (Zavyalova et al., 2013).

FFPE tumor sections (7 µm thick) were deparaffinized, rehydrated and processed for heat-induced epitope retrieval with EDTA buffer (pH 8.0) using PT Link (Dako, Denmark), and blocked with 3% bovine serum albumin (BSA; Amresco) in PBS. Sections were incubated with primary antibodies against CK7 (clone N-20, 1:50, Santa Cruz Biotechnology), α-SMA (clone 1A4, 1:500, Dako, Denmark), P-cadherin (1:50, in-house rabbit antibody) and decorin (LF113, Larry Fischer). The secondary antibodies were: donkey anti-mouse-IgG conjugated to Alexa Fluor 488 (Abcam), donkey anti-rabbit-IgG conjugated to Cy3 (Abcam), and donkey anti-goat-IgG conjugated to Alexa Fluor 647 (Abcam). Sections were mounted with Vectashield mounting medium (Vector Laboratories, USA) containing DAPI (for nuclear staining). Samples were analyzed with a LSM 780 NLO confocal microscope (Carl Zeiss, Germany).

Alternatively, expression of P-cadherin (1:50, in-house rabbit antibody), decorin (1:100, LF113, from Larry Fisher), collagen I (1:100, 20111, Novotec) and cytokeratin 7 (CK7, clone N-20, 1:50, Santa Cruz Biotechnology) was assessed using the BondRXm immunostainer (Leica, Germany) and the Opal 7-Color Automation immunohistochemistry Kit (PerkinElmer). Samples were analyzed using the Vectra 3.0 Automated Quantitative Pathology Imaging System (PerkinElmer).

Protein extracts (20 to 60 µg) prepared as previously described (Thuault et al., 2013) were resolved on 10% or 15% polyacrylamide gels and transferred to Immobilon-P membranes (Millipore) that were incubated with the indicated antibodies, as described previously (Thuault et al., 2013). The Odyssey Infrared Imaging System (LI-COR Biosciences) was used for protein detection and quantification. Immunoblots were quantified with Odyssey V3.0 and ImageJ.

To remove O-linked GAG chains from decorin, cell culture medium or whole-cell lysates was digested with chondroitinase ABC (Amsbio, 0.1 unit/100 mg of protein) at 37°C for 24 h, and then supernatants were concentrated by ammonium sulfate precipitation (0.4 g/ml overnight).

For global GTPase activity measurements and β-Pix immunoprecipitation, DCCM was induced in confluent cell monolayers using a device designed to perform multiple calibrated and reproducible injuries with a spiral scarificator (Plutoni et al., 2016). After 6 h of migration, cells were processed to measure CDC42 activity as described previously (Plutoni et al., 2016); 1 mg of the supernatant of a cell lysate obtained in buffer (10 mM PIPES pH 7, 100 mM NaCl, 300 mM sucrose, 3 mM MgCl2, 0.5% IGEPAL CA-640, 1 mM EDTA, 1 mM Na3 VO4, protease inhibitor cocktail) was used for immunoprecipitation. The polyclonal anti-β-PIX antibody was incubated with protein G (Dynabeads; Invitrogen) at room temperature for 30 min. After washing, 1 mg protein extract was added at room temperature for 3 h.

Cells were plated in Ibidi culture inserts on 12-mm-diameter cover-slips until they reached the desired density level. At 5–10 h after insert removal, cells were fixed in 3.2% paraformaldehyde (PFA) in PBS for 15 min, followed by 2 min permeabilization with 0.1% Triton X-100 (in PBS) and saturation with 2% BSA (in PBS). Cells were incubated with primary and secondary antibodies in PBS containing 2% BSA. After washing with PBS with 2% BSA, primary antibodies were detected with Alexa Fluor 488-conjugated goat anti-rabbit-IgG or -mouse-IgG in PBS with 1% BSA. Images were taken using a confocal SP5-SMD (Leica) with 40×/1.3 NA or 63×/1.4 NA oil HCX PL APO CS objectives (Leica) and captured with a hybrid detector (Leica HyD) controlled using the C software. Images were processed using Adobe Photoshop and assembled using Adobe Illustrator.

Spheroids were prepared as described previously (Thuault et al., 2013) and embedded in neutralized 3D collagen (2 ml/ml, rat-tail Corning) matrix according to the manufacturer's instructions. The RDG (100 µM) was added to the collagen. Spheroid invasion was monitored for 72 h using an Olympus IX-83 microscope equipped with a motorized stage, and heated and CO₂-regulated incubator. Phase-contrast images were taken every hour, using a 10×/0.3 NA RC objective and captured with a sCMOS Zyla 4.2.1 camera controlled by the Metamorph software. The invaded area occupied by the cells was measured using ImageJ (http://dev.mri.cnrs.fr/projects/imagej-macros/wiki/Phase_Contrast_Cell_Analysis_Tool_%28Trainable_WEKA_Segmentation%29). Detached cells (characterized by loss of cell–cell adhesion) were counted at 72 h of invasion. Data are the mean±standard error of the mean (s.e.m.) of at least four independent experiments in which at least five spheroids were embedded per experimental condition.

Cell spheroids embedded in type I collagen matrix for 72 h were fixed in 4% PFA in PBS at 37°C for 2 h. Samples were blocked and permeabilized in a 1% BSA and 0.1% Triton X-100 PBS solution for 2 h, stained with Rhodamine–phalloidin in 1% BSA in PBS at 4°C overnight. For observation, each spheroid-containing collagen sample was mounted in a PDMS tube between a slide and a coverslip. Samples were observed using a multiphoton Zeiss LSM 7MP Optical Parametric Oscillator microscope using a 20×/1.0 NA Plan Apochromat water immersion objective and PMT NDD detectors, controlled by the Zeiss Zen software. The collagen network was imaged by SHG combined with imaging of Rhodamine–phalloidin-labeled cells using the 1079 nm line. Z-stacks with a spacing of 1 µm were collected to obtain an accurate visualization of the collagen fibers. Images of collagen fibers in contact with the spheroid were analyzed using ImageJ (directionality plug-in) to measure the 2D-projected angles of all fibers and to obtain the alignment coefficient (Grossman et al., 2016). The migration axis is the perpendicular to the spheroid.

Polyacrylamide gels were prepared as previously described to support coating with neutralized type I collagen (2 mg/ml, Corning) (Plutoni et al., 2016). After polymerization at 37°C for 1 h, a PDMS membrane with a rectangular opening was deposited on top of the collagen-coated polyacrylamide gel. Then, 20,000 cells were seeded within the rectangle defined by the PDMS membrane opening. Cells were left to adhere and proliferate for a few hours. Then, the PDMS membrane was carefully removed, to allow cells to migrate towards the available substrate. After 12 h of migration, cells were fixed with 2.5% glutaraldehyde and dehydrated with ethanol and then with hexamethyldisilazane (HMDS). Samples were shaded with sprayed gold and observed using a scanning electron microscope (Hitachi S4000). For quantification of collagen fiber orientation, the ImageJ directionality plugin was used (Grossman et al., 2016).

Cells were plated in Ibidi Culture-Inserts (BioValley; 20,000 cells per chamber) until they reached the desired density level. Then, inserts were removed to allow cells to migrate into the cell-free areas. Cells were imaged using an Axiovert inverted microscope (Carl Zeiss) equipped with a motorized stage, and heated and CO₂ regulated incubator. Phase-contrast images were taken every 5 min overnight using a 10×/0.3 NA PH1 DIC1 objective and captured with a CCD MicroMax 1300 Y/HS camera (Roper Scientific) controlled by the MetaMorph 7.0 software.

Manual cell tracking and x/y cell position recording (of each cell at each time point) were performed using MetaMorph software. The mean speed, persistence and directionality of individual cells were calculated and illustrated using the Ibidi Chemotaxis tool (https://ibidi.com/chemotaxis-analysis/171-chemotaxis-and-migration-tool.html). Angles were plotted using the Rose.net software. The velocity field in monolayers was mapped by particle imaging velocimetry (PIV) analysis as previously described (Plutoni et al., 2016).

At 5–10 h after insert removal, cells were fixed, and nuclei, centrosomes and the Golgi were stained. Cells in the first four rows of the 2D migrating cells were analyzed. Measurements of cell polarity toward a global monolayer direction (centrosomes located in front of the nucleus and behind the Golgi apparatus within the quadrant facing the wound) were scored as correctly oriented (Osmani et al., 2006).

Confocal images of paxillin or F-actin staining were taken 6 h after insert removal. FA orientation, membrane protrusion size and orientation were measured using ImageJ, and angle values were plotted using the Rose.Net software. Each orientation was determined using the monolayer migration direction as reference axis.

Briefly, polyacrylamide gels with a Young's modulus of 12 kPa were prepared as previously described (Plutoni et al., 2016). After polymerization, gels were incubated with 0.1 mg/ml collagen I overnight. Then, to pattern the cells on top of the polyacrylamide gels, a PDMS membrane with a rectangular opening was deposited on top of the polyacrylamide gel and cells were seeded and left to adhere and proliferate for a few hours. Then, the PDMS membrane was carefully removed and cells could migrate toward the available substrate. Time-lapse imaging was performed using an automated inverted microscope (Nikon Eclipse Ti, 10× objective lens) equipped with thermal, CO₂ and humidity controls operated by the MetaMorph software (Universal Imaging) software. Recording started ∼30 min after removal of the PDMS membrane and lasted for 15 h. Images were obtained every 3 min over a period of 1 to 15 h. As cells migrated, they exerted traction forces on the substrate, resulting in polyacrylamide gel deformations that could be observed by imaging the displacements of fluorescent beads embedded in the gel. Bead displacement between any experimental time point and a reference image obtained after cell trypsinization was computed using an in-house PIV algorithm (available upon request). Based on these displacements, traction forces were computed using Fourier transform traction microscopy with finite gel thickness, as described in Trepat et al. (2009).

To compute maps of the mechanical inter- and intra-cellular tension within monolayer sheets, monolayer stress microscopy was used as previously described (Tambe et al., 2011). According to Newton's law, traction forces applied at the cell–gel interface must be balanced by intra- and inter-cellular forces (Serra-Picamal et al., 2015; Tambe et al., 2011, 2013). The alignment angle between the major axis of the principal stress ellipse and the direction of the cellular motion was measured as previously described (Serra-Picamal et al., 2015).

To assess significant differences between experimental conditions, Student's t-tests (experiments with n>30) and non-parametric Mann–Whitney U-tests (experiments with n<30) were used. At least three independent experiments were performed.

Metastasis-free survival (MFS) was determined as the interval between initial diagnosis and detection of the first metastasis. Survival distributions were estimated by the Kaplan–Meier method and the significance of differences between survival rates were ascertained with the log-rank test.

We thank Virginie Georget, Hassan Boukhaddaoui, Sylvain de Rossi, Simon Lachambre, Chantal Cazevieille, Alicia Cabalerro-Medigo and Vicky Dialou from the Montpellier Imaging Facility (http://www.mri.cnrs.fr/); the Institute in Regenerative Medicine and Biotherapy Microarray Core Facility, CHU-INSERM-UM Montpellier (http://www.chu-montpellier.fr/fr/irmb/) for the transcriptomic arrays and analysis. Mass spectrometry experiments were carried out at of the Functional Proteomics Platform Facility of Montpellier. We acknowledge Larry W. Fisher for the anti-decorin antibody, and Damien Planchon, Mallory Genest, Eva Faurobert, Corinne Albigès-Rizo, Matthew Kutys, Didier Fesquet and Florence Ruggiero for helpful discussions. The Montpellier Imaging Facility is supported by the Agence Nationale de la Recherche (ANR-10-INBS-04, ‘Investments for the future’).