MMP14 is required for delamination of chick neural crest cells independently of its catalytic activity

ABSTRACT Matrix metalloproteinases have a broad spectrum of substrates ranging from extracellular matrix components and adhesion molecules to chemokines and growth factors. Despite being mostly secreted, MMPs have been detected in the cytosol, the mitochondria or the nucleus. Although most of the attention is focused on their role in matrix remodeling, the diversity of their substrates and their complex trafficking open the possibility for non-canonical functions. Yet in vivo examples and experimental demonstration of the physiological relevance of such activities are rare. Here, we have used chick neural crest (NC) cells, a highly migratory stem cell population likened to invasive cancer cells, as a model for physiological epithelial-mesenchymal transition (EMT). We demonstrate that MMP14 is required for NC delamination. Interestingly, this role is independent of its cytoplasmic tail and of its catalytic activity. Our in vivo data indicate that, in addition to being a late pro-invasive factor, MMP14 is also likely to be an early player, owing to its role in EMT. Summary: MMP14 is required for the delamination of chick neural crest cells by promoting the basal positioning of nuclei during EMT, an effect that is independent of its catalytic activity.


INTRODUCTION
Matrix metalloproteinases (MMPs) are widely expressed in cancer and inflammation (Bonnans et al., 2014;Khokha et al., 2013). Most of the attention has been focused on their role in matrix remodeling. However, we now know that MMPs can traffic to various cell compartments, including the nucleus, the mitochondria and the cytosol (Jobin et al., 2017). In addition, numerous non-canonical substrates, unrelated to matrix biology, have been found. It is therefore assumed that MMPs influence a broad range of biological functions, ranging from cell proliferation to differentiation to cell migration (Rodríguez et al., 2010).
MMP14 was the first membrane-bound MMP discovered in 1994 (Sato et al., 1994). MMP14 is expressed in numerous human cancers where it modulates tumor growth and cancer invasion (Turunen et al., 2017). MMP14 is also required in physiological events, such as mammary gland (Feinberg et al., 2016) and bone development (Holmbeck et al., 1999), and skeletal muscle regeneration (Kopinke et al., 2017). In most cases, MMP14 acts via activation of MMP2 and/or degradation of extracellular matrix. However, the range of the substrates of MMP14 extends beyond the matrix. MMP14 can cleave membrane proteins such as CD44 (Suenaga et al., 2005) and ADAM-9 (Chan et al., 2012), or secreted factors such as SDF1 (McQuibban et al., 2001). It can also interact with and process integrins (Gonzalo et al., 2010b). Furthermore, MMP14 can have non-catalytic functions. The cytoplasmic domain contains binding sites for regulators of Rho GTPases (Gonzalo et al., 2010a;Hoshino et al., 2009) and regulators of the transcription factor HIF (Sakamoto and Seiki, 2010;Sakamoto et al., 2014). Finally, MMP14 has been detected inside the nucleus, where it regulates gene expression (Shimizu-Hirota et al., 2012). Most of these non-canonical functions have not been tested in vivo. In addition, these observations suggest that MMP14 could be involved in early steps of epithelialmesenchymal transition (EMT) as Rho GTPases and HIF have been implicated in this process (Barriga et al., 2013;Joseph et al., 2018). During EMT, cells modulate cell-cell and cell-matrix adhesion, remodel their polarity from apico-basal to front-rear and degrade extracellular matrix (Nieto et al., 2016). In human prostate cancer cell lines, MMP14 has been shown to contribute to EMT in a Wnt5a-dependent manner (Cao et al., 2008) and MMP14 can trigger an E-to N-cadherin switch when experimentally overexpressed in esophageal squamous cell carcinoma in vitro (Pang et al., 2016). MMP14 was shown to be important for in vivo cell migration of Xenopus neural crest (NC) cells (Garmon et al., 2018) in the context of an MMP2-positive environment. However, Xenopus NC cells migrate as a pseudoepithelial sheet and do not perform the typical basal extrusion from an epithelium observed during carcinoma progression or during NC delamination in amniotes . Thus, a direct involvement of MMP14 in EMT in vivo remains to be tested.
Here, we have used chick NC cells as in vivo model of EMT. NC cells are a multipotent stem cell population that leaves the neuroepithelium by a series of EMT-driven basal extrusions (Gouignard et al., 2018). We found that MMP14 is expressed by trunk NC cells during EMT. We demonstrate that MMP14 is required for NC delamination. MMP14 does so independently of its catalytic activity and does not affect cadherins or laminin distribution. Instead, MMP14 acts by promoting the occurrence of basal mitoses. Our in vivo data place MMP14 as an early player in the EMT cascade.

MMP14 expression prior to and during NC delamination
First, we analyzed the expression of MMP14 by in situ hybridization in the trunk of chicken embryos. At stage HH13 (19 somites), MMP14 mRNA is detected in the neural tube, the somites, the notochord and the lateral mesoderm, as seen in whole mount (Fig. 1A) and on transverse sections ( Fig. 1B-D). In the NC domain, MMP14 is weakly expressed before EMT (Fig. 1B,C, asterisks). Its expression increases during EMT (Fig. 1D). Owing to the strong expression of MMP14 in somites, it is difficult to identify NC cells expressing MMP14 while they migrate through the paraxial mesoderm. To overcome this issue, we cultured NC cells. In culture, expression of MMP14 is clearly seen in migratory NC cells that have left the neural tube explant (Fig. S1).
MMP14 is required for trunk neural crest delamination independently of its catalytic activity To assess whether MMP14 is required for NC delamination, we performed loss of function by electroporation with a siRNA against MMP14 or a scrambled version of the siRNA as control ( Fig. 2A,B). The siRNA-MMP14 efficiency was tested by western blot (Fig. S2). Twenty-four hours post electroporation (24 hpe), the scrambled control did not affect the size of the NC domain ( Fig. 2A,F). However, siRNA-MMP14 led to an accumulation of Snail2positive NC cells in the dorsal neural tube, shifting the limit of the NC region ventrally (Fig. 2B,bracket;2F). This effect is not specific to Snail2 and can be observed with other NC markers such as Sox9 (Fig. S3) or AP2 (see Fig. 3 hereafter). We then attempted to rescue MMP14 knockdown by co-expressing GFP-tagged versions of wild-type MMP14 or a point mutant version (MMP14-EA) in which a glutamate essential for the catalytic activity has been replaced by an alanine (Rozanov et al., 2001). Catalytic inactivity of MMP14-EA was confirmed by acceptor photobleaching FRET and by monitoring the cleavage pattern of the FRET reporter by western blot (Fig. S4, see Materials and Methods). Importantly, MMP14-GFP and MMP14-EA-GFP were generated by codon optimization, making them resistant to the siRNA. MMP14 (Fig. 2C) and MMP14-EA (Fig. 2D) were equally capable of preventing accumulation of NC in the dorsal neural tube when co-electroporated with siRNA-MMP14 (Fig. 2F). Such accumulation of NC cells in the dorsal neural tube could be due to an expansion of the NC domain or to a lack of delamination. To distinguish between these two possibilities, we performed a lossof-function with an alternative method by using an inhibitor of MMP14 (NSC405020) that binds to the hemopexin domain of MMP14 (Remacle et al., 2012). We cultured neural tube explants with NSC405020 or DMSO as control. Exposure to NSC405020 reduced the total number of NC cells outside the neural tube compared with the DMSO (Fig. S5) indicating that blocking MMP14 prevents delamination. Furthermore, we found no effect on proliferation after electroporation with si-RNA-MMP14 or its scrambled control (Fig. S6). Altogether, these results indicate that MMP14 is required for NC delamination from the neural tube and that it acts independently of its catalytic activity. As NC delamination is achieved by EMT in the chicken embryo, these data indicate that MMP14 is involved in EMT.
MMP14 is required for the remodeling of cell polarity during NC delamination EMT is a complex process that involves a qualitative and quantitative change in cell adhesion, extracellular matrix remodeling, a loss of apicobasal polarity and the acquisition of migratory capabilities (Nieto et al., 2016). Thus, we aimed to identify which of these steps is failing after MMP14 loss-offunction in NC cells. First, we checked the distributions of E and N cadherins, and that of laminin, a component of the basal membrane. E-cadherin expression in the trunk neuroepithelium is lost several hours prior to EMT of NC cells (Dady et al., 2012;Dady and Duband, 2017). N-cadherin is downregulated at transcriptional and post-translational levels at the time of trunk NC delamination (Shoval et al., 2007). Whereas laminin distribution is not complete over the dorsal neural tube during NC EMT (Leonard and Taneyhill, 2019). Thus, a maintenance of any of these three molecules might account for the defect in delamination. However, the loss of MMP14 had no effect on their distributions (Fig. S7). Next, we assessed cell polarity. In the chicken embryo, delamination of trunk NC cells occurs after neural tube closure and, contrary to what is observed at cephalic levels, the epithelial integrity of the dorsal neural tube is maintained throughout the whole delamination process. That is due to the fact that trunk NC cells undergo EMT one cell at a time. Thus, any given delaminating cell is surrounded by non-delaminating cells that retain their normal adherence junctions and polarity markers. Therefore, apical markers such as aPKC or ZO1 are always present in the dorsal neural tube. In addition, such apical markers are no longer detectable in cells undergoing EMT and accumulate relatively late during epithelialization (Fig. S8). Thus, they are not informative when it comes to analyzing an experimental condition in which cells are prevented from leaving the neural tube, as observed with siRNA-MMP14 (Fig. 3A,B). By contrast, we found that pericentriolar material 1 (PCM1) subcellular localization clearly changes when cells toggle between epithelial and mesenchymal phenotypes, e.g. during secondary neurulation or mesoderm development (somite epithelialization, EMT of the sclerotome) (Fig. S8). More precisely, PCM1 is localized apically early on during epithelialization and remains detectable in cells that have undergone EMT. This is likely due to the fact that PCM1 is not a strict apical marker but is a protein found throughout the cytoplasm that preferentially associates with centrioles and microtubules, and that modulates the centrosomal actin network (Farina et al., 2016). Therefore, we assessed PCM1 distribution in the neural tube at the time of NC delamination. We found PCM1 to be apically biased in neuroepithelial cells. However, in NC cells undergoing EMT, PCM1 is relocated randomly (Fig. 3C-D′, arrowheads) and this change is not affected by the scrambled control (Fig. 3C). On the contrary, siRNA-MMP14 prevented the relocalization of PCM1 in AP2-positive NC cells that were retained in the dorsal neural tube (Fig. 3D,D′, arrows). In the neuroepithelium, mitoses are strictly apical and one can force some cells to divide basally by interfering with apicobasal polarity (Afonso and Henrique, 2006). During NC EMT, basal mitoses have been observed and can be either explained by apical detachment or the loss of apicobasal polarity (Ahlstrom and Erickson, 2009). Thus, we decided to analyze the distribution of mitoses after MMP14 loss of function. We labeled mitotic nuclei using phospho-histone H3 (Fig. 4, pH3). The scrambled control had no effect on the position of mitoses (Fig. 4A) and we found the same percentage of non-apical mitoses on non-electroporated and electroporated sides (Fig. 4C-E). However, the siRNA-MMP14 (Fig. 4B) decreased the proportion of basal mitoses in the NC domain from 37% to 13% (Fig. 4C-E). Despite having a strong effect, MMP14 knockdown does not completely abolish the occurrence of basal mitoses. As we found no effect of siRNA-MMP14 on the normal reduction of N-cadherin occurring in the NC domain ( Fig. S7), the remaining 13% of basal mitoses likely correspond to NC cells that had already detached apically. Nuclei are counterstained with DAPI (grey); the electroporated side is visualized using GFP (green). Brackets indicate neural crest accumulation (B). Scale bars: 20 µm. (E) The method for measuring the area of the neural crest domain. The neural crest area on the electroporated side was divided by the neural crest area in the control side. (F) Plot of neural crest area (snail 2) with scrambled (n embryos =11, n sections =132 from three independent experiments), siRNA-MMP14 alone (n embryos =31, n sections =445 from eight independent experiments), siRNA-MMP14 with wild-type MMP14 (n embryos =7, n sections =134 from two independent experiments) and siRNA-MMP14 with MMP14-EA (n embryos =4, n sections =90 from one experiment). ANOVA with a Kruskal-Wallis Test and Dunn's multiple comparisons: ****P<0.0001 (scrambled versus siRNA), *P=0.0441 (siRNA versus siRNA+MMP14), **P=0.0030 (siRNA versus siRNA+EA) and P>0.9999 (not significant; scrambled versus siRNA+MMP14 or siRNA+EA).
It has previously been shown that some NC cells delaminate by forming a protrusion, grabbing onto the extracellular matrix and extracting their cell body, sometimes leaving the apical tail within the neuroepithelium (Ahlstrom and Erickson, 2009). Therefore, impairing migratory capabilities could contribute to a delamination defect and NC accumulation in the dorsal neural tube. Thus, we wondered whether the loss of MMP14 might also impair the migratory capabilities of NC cells. To this end, we analyzed the distribution of electroporated cells in the anterior trunk region of embryos expressing the scrambled control or siRNA-MMP14 at 24 hpe. To detect the full extent of NC migration, we counterstained for HNK1 (see Materials and Methods). In control conditions, GFP + cells migrate as well as non-electroporated HNK1-positive migratory NC cells, whereas siRNA-MMP14 cells were mostly detected close to the dorsal region of the neural tube (Fig. S9). On the one hand, this migration defect could be due the aforementioned defect of delamination, causing cells to depart with a delay. On the other hand, it could indicate a defect of cell motility per se. To address this, we cultured electroporated neural tube explants on fibronectin (Fig. S10). The siRNA-MMP14 led to fewer GFP-positive NC cells outside the neural tube compared with the scrambled control, confirming that MMP14 knocked-down cells have problems delaminating. However, the mean area explored by the electroporated cells was similar in both conditions. This indicates that, once they have left the neural tube, NC cells electroporated by the scrambled control and the siRNA-MMP14 seem to migrate as efficiently. Altogether, these data indicate that: (1) MMP14 has no effect on E or N cadherins or on laminin distribution in the dorsal neural tube; (2) MMP14 does not seem to affect the intrinsic motility of NC cells; and (3) MMP14 is required for basal positioning of mitoses during NC EMT.
Overexpression of MMP14 is sufficient to promote basal mitosis in the neuroepithelium We next wondered whether the endogenous expression of MMP14 in the neuroepithelium is required for the normal apical positioning of mitoses and whether upregulating MMP14 might be sufficient to trigger basal mitosis. We first analyzed the effect of the scrambled control and the siRNA-MMP14 in the neuroepithelium. They had no effect on the distribution of mitoses (Fig. 5A,B), indicating that neuroepithelial cells do not require MMP14 to maintain normal apicobasal polarity and normal interkinetic nuclear movements. Next, to assess whether MMP14 is sufficient to promote non-apical mitosis, we overexpressed different GFP-tagged forms of MMP14 in the neuroepithelium: wild-type MMP14, the inactive point mutant (MMP14-EA), a delta-catalytic form (MMP14-ΔCAT) or a delta-cytoplasmic form (MMP14-ΔCyto) (Fig. 5C). All forms of MMP14 distributed homogenously along the apicobasal axis ( Fig. 5D) and were able to promote non-apical mitosis (Fig. 5D, arrows; 5E,F). As neuroepithelial cells robustly divide with their nuclei in an apical position, we determined that non-apical mitoses occurred more than two nuclei diameters away from the apical domain. In non-electroporated halves of the neural tubes, used as internal controls, only 1.7% of non-apical mitoses were observed (Fig. 5F), whereas overexpression of MMP14 or MMP14-EA led to a threefold increase in the number of non-apical mitoses to 5% and 5.2%, respectively (Fig. 5F). MMP14-ΔCAT had a weaker effect, producing 3.3% of non-apical mitoses, whereas MMP14-ΔCyto had a stronger effect, producing 10.2% of non-apical mitoses (Fig. 5F). Altogether, these results indicate that increasing MMP14 expression is sufficient to promote basal mitoses independently of the catalytic activity or the cytoplasmic tail. Importantly, overexpression of The distance between the mitoses and the apical part of neural tube (d) was divided by the neural tube width (D). A ratio of 0 means that the mitoses were close to the apical region. A ratio of 1 means that the mitoses were close to the basal region. (D) Plot of the position of the mitoses along the apico-basal axis in the neural crest in embryos electroporated with the scrambled control (n embryos =3, n mitoses non-elec side =33, n mitoses elec side =36 from one experiment) or siRNA-MMP14 (n embryos =8, n mitoses non-elec side =51, n mitoses elec side =69, from three independent experiments). Dashed line indicates the border between the apical half and basal half. ANOVA followed by Dunn's multiple comparisons **P=0.0014 (electroporated scrambled versus electroporated siRNA), **P=0.0027 (non-electroporated siRNA versus electroporated siRNA). In the scatter dot plot, the lines indicate the median and the error bars show the extent of the whole dataset. (E) Percentage of basal mitoses in neural crest domain. Basal mitoses were defined as mitoses located within the basal half of the neural crest domain (corresponding to values between 0.5 and 1 on the graph in D). Statistics were performed according to Taillard et al. (2008), **T>6.635 (non-electroporated siRNA versus electroporated siRNA) the null hypothesis is rejected with 99% confidence. No-elec, non-electroporated side; elec, electroporated side.
MMP14 in the neural tube is not sufficient to promote ectopic EMT, as we observed neither a loss of N-cadherin in electroporated cells nor a local degradation of laminin next to them (Fig. S11). These data demonstrate that the non-apical mitoses induced by MMP14 are not caused by a local induction of EMT and confirm a specific role for MMP14 in nuclear positioning that is independent of its catalytic function.
We next wondered whether destabilizing apicobasal polarity in the NC domain would be sufficient to compensate for the lack of MMP14. To investigate this, we made use of Par3 overexpression, previously described as sufficient to cause non-apical mitoses in the chick neuroepithelium without causing ectopic delamination of neuroepithelial cells (Afonso and Henrique, 2006). Importantly, the accumulation of NC in the dorsal neural tube induced by siRNA-MMP14 can be rescued by Par3 overexpression (Fig. 6A,B).
Par3 has been implicated in the regulation of cell migration (e.g. Moore et al., 2013;Nishimura and Kaibuchi, 2007;Pegtel et al., 2007). Thus, one could suggest that overexpressing Par3 might compensate for the lack of MMP14 by promoting delamination independently of its effect on cell polarity. To circumvent this, we Control side (n embryos =26, n mitoses =1908), MMP14 (n embryos =7, n mitoses =716), MMP14-ΔCAT (n embryos =7, n mitoses =406) and MMP14-EA (n embryos =9, n mitoses =707) from three independent experiments, and MMP14-ΔCyto (n embryos =3, n mitoses =431) from one experiment. (F) Percentages of nonapical mitoses for each condition depicted in D. Non-apical mitoses were determined as mitoses occurring more than two nuclei diameters away from the apical domain. Statistics were performed according to Taillard et al. (2008), Thresholds for T were such that, at *T>3.841, the null hypothesis is rejected with a 95% confidence, at **T>6.635, the null hypothesis is rejected with 99% confidence and at ***T>10.83, the null hypothesis is rejected with 99.9% confidence. In B,E, the lines indicate the median and the error bars show the extent of the whole dataset. Electroporated side is indicated by GFP (A,D, green). Scale bars: 35 µm. attempted to rescue MMP14 knockdown by performing Par3 loss of function experiments. We designed a translation-blocking morpholino against chicken Par3 (Par3MO). Electroporation of Par3MO in the neural tube has only mild effects, leading to occasional ectopic mitoses, basal accumulation of nuclei and some basal rosette-like organization of nuclei, all indicative of mild apico-basal polarity defects (Fig. S12). Importantly, the overall structure of the neural tube is intact and no ectopic delaminate or EMT-like phenotype is observed. Coelectroporation of siRNA-MMP14 and Par3MO was sufficient to prevent the accumulation of snail2-positive NC cells in the dorsal neural tube (Fig. 6A,B). However, these data should be taken with caution. It is important to note that we were unable to check for the Par3MO efficiency by immunostaining or western blots as the two anti-Par3 antibodies we have tested did not produce specific signals (see Materials and Methods). In addition, assessment for specificity by a rescue strategy of the Par3MO is not possible because expression of Par3 alone also leads to apicobasal defects (Afonso and Henrique, 2006). The mild effects observed after Par3MO electroporation are in line with Par3 lossof-function data published elsewhere (Chuykin et al., 2018), suggesting that the knockdown is efficient; however, readers may decide to ignore these data due to the lack of formal validation of Par3MO.

DISCUSSION
The first MMP was identified as a collagenolytic enzyme in tadpoles (Gross and Lapiere, 1962). Since then, more than 20 MMPs have been discovered in vertebrates. The vast majority of publications on MMPs has focused on their roles in extracellular matrix remodeling, most of these works being conducted in vitro either from a purely biochemical perspective or in cell culture. It is unfortunate, because MMPs have a much broader range of capabilities. MMPs have a secretion peptide in their N terminus but are not restricted to the extracellular space. MMPs have been found in the cytoplasm, the mitochondria and the nucleus, indicating that we still know very little about how their trafficking is controlled. In addition, several nonproteolytic activities, including signaling or transcription, have been found. (For reviews on the non-canonical roles and subcellular localizations of MMPs, see Iyer et al., 2012;Jobin et al., 2017;Mannello and Medda, 2012.) Here, our data provide one of the very few in vivo examples of a physiologically relevant non-catalytic MMP activity by showing that MMP14 is essential for NC delamination.
MMP14 expression is often associated with poor prognosis in cancers such as colorectal, breast and stomach (Cui et al., 2019;Dong et al., 2015;Yao et al., 2013). Here, we found MMP14 mRNA in numerous epithelial structures, such as the neuroepithelium, the notochord and the epithelial somites. These data indicate that, contrary to other MMPs, such as MMP2 (Duong and Erickson, 2004), MMP14 gene expression is not systematically linked to EMT or to cell migration during early development.
Our in vivo and in vitro experiments show that knocking down MMP14 reduces the number of delaminating NC cells but does not seem to affect their intrinsic motility. In addition, adjacent cells that were not electroporated with siRNA-MMP14 were not affected. These data indicate that MMP14 is specifically required for delamination and acts cell autonomously. Our in vitro results do not completely rule out the possibility that MMP14 may play a role in NC migration in vivo as cells are known to adopt different migratory behaviors in 3D and 2D environments (Caswell and Zech, 2018). In addition, proteolytic enzymes are extremely versatile, and demonstrating a non-proteolytic role in one phenomenon does preclude a catalytic function of the same enzyme once cells are in a different context (e.g. located within the extracellular matrix after delamination). Interestingly, trunk NC cells in chick migrate through an environment rich in stromal cell-derived factor 1 (Sdf1/CXCL12) (Saito et al., 2012) and fibronectin (Perris and Perissinotto, 2000), and NC cells need Sdf1 to reach the anlagen of the sympathetic ganglia (Kasemeier-Kulesa et al., 2010). Interestingly, Sdf1 displays a high affinity for fibronectin (Pelletier et al., 2000), suggesting that most of the secreted Sdf1 might be trapped in fibronectin-rich matrices in vivo. Both fibronectin and Sdf1 are substrates of MMP14 (Barbolina and Stack, 2008;Shi and Sottile, 2011). More specifically, MMP14 can render Sdf1 inactive by removing the first four amino acids in its N terminus (McQuibban et al., 2001). Therefore, one possibility is that, despite being involved in NC delamination in a non-proteolytic manner, MMP14 might later contribute to the normal dorsoventral migration of trunk NC cells in vivo by first promoting the release of Sdf1 from fibronectin and subsequently inactivating it by removing its N terminus. The shorter migration observed in siRNA-MMP14 transfected cells would be the result of a delayed delamination due to polarity defects and additional migration defects. Such migration defects would not be detectable in our 2D in vitro assays, as they are performed on ready-to-use fibronectin and in the absence of Sdf1. This possibility, which is well beyond the scope of our study, remains to be explored.
Delamination of trunk NC cells is known to be bound to the G1/S transition of the cell cycle (Burstyn-Cohen and Kalcheim, 2002). Owing to interkinetic movements, nuclei are located basally when cells are in S phase and apically in mitosis (Spear and Erickson, 2012). It has been proposed that blocking cells from entering in S phase hinders NC delamination by retaining nuclei in an apical position (Burstyn-Cohen and Kalcheim, 2002). However, the fact that NC cells lose their apicobasal polarity and display a high rate of basal mitosis indicates that basal positioning of NC nuclei is no longer tightly linked to interkinetic movements during NC EMT. Therefore, part of the previously described effect of blocking the G1/S transition may be due to blocking proliferation (Burstyn-Cohen and Kalcheim, 2002). Our data show that the percentage of NC cells undergoing basal mitosis depends on the presence of MMP14. Importantly, this happens in absence of effects on the overall proportion of pH3 + cells ruling out the possibility that MMP14 acts on proliferation itself.
How can MMP14 modulate apicobasal polarity? One possibility is via its cytoplasmic tail. The cytoplasmic tail of mammalian MMP14 is known to harbor binding sites for regulators of Rac1 and RhoA pathways (Gonzalo et al., 2010a;Hoshino et al., 2009), which could link MMP14 to the regulation of apicobasal polarity. However, several of these sites are absent in the sequence of chicken MMP14 (Yang et al., 2008). In addition, the truncated form MMP14-ΔCyto was sufficient to promote non-apical mitosis in the neuroepithelium, indicating that MMP14 can act on polarity without direct intracellular signaling events. In particular, the effect of MMP14-ΔCyto strongly suggests that a direct interaction between MMP14 and the Par complex is unlikely. MMP14-ΔCyto provoked twice as many non-apical mitoses as wild-type MMP14. The cytoplasmic tail of MMP14 is required for normal turnover and endocytosis of MMP14 (Jiang et al., 2001;Uekita et al., 2001). Thus, the MMP14-ΔCyto form is likely to accumulate in the membrane more than the other forms, which may account for its stronger effect. Interestingly, as MMP14 is capable of autocatalysis, cells that have a high level of MMP14 at the cell surface do not exhibit a significant MMP14-dependent activity. High MMP14 activity is instead linked to a rapid turnover and vesicular localization.
This, for example, has been shown in human endothelial cells in which membrane colocalization of MMP14 with β1-integrin is linked to low proteolytic activity and an epithelial phenotype, whereas vesicular MMP14 colocalizing with β3-integrin is associated with high catalytic activity and a migratory phenotype (Gálvez et al., 2002). In this context, a higher membrane accumulation of the ΔCyto form would lead to low catalytic activity due to auto-catalysis. Therefore, the fact that this truncated form is a potent inducer of nonapical mitoses further reinforces the notion that MMP14 acts independently of its catalytic activity, which is demonstrated by the ability of the MMP14 point mutant to rescue MMP14 knockdown and to promote non-apical mitoses as efficiently as wild-type MMP14. Interestingly, the ΔCAT form was less potent at inducing non-apical mitoses than all the other forms of MMP14, indicating that, even if it acts independently of its proteolytic activity, MMP14 requires its complete extracellular domain. This suggests that MMP14 most likely needs to interact with extracellular factors to impact cell polarity and nuclear positioning.
In conclusion, our data show that MMP14 mRNA expression does not correlate systematically with EMT and cell migration, contrary to what has been proposed for other MMPs such as MMP2 (Duong and Erickson, 2004). Importantly, our in vivo data demonstrate a direct role of MMP14 in the basal positioning of nuclei during EMT, which is independent of its catalytic activity and independent of cell-cell or cell-matrix adhesion remodeling. MMP14 is often expressed in human cancers (Castro-Castro et al., 2016) and, being an MMP, it is assumed to be a downstream effector of the EMT pathway involved in the matrix degradation that favors tissue invasion. However, our data indicate that MMP14 is specifically required for EMT and that, via this effect, MMP14 plays a crucial role in basal extrusion and thus may act from the onset of invasion. Therefore, it may be worth assessing the putative roles of MMP14 in the early, rather than the late, steps of cancer progression.

Chicken eggs
Fertilized chicken eggs were purchased from S.C.A.L. (Société commerciale avicole du Languedoc) and incubated at 38°C until the desired stage (Hamburger and Hamilton, 1992).

Electroporation
Embryos at stage HH12 were windowed. Using a glass capillary, a solution of 6% sucrose, 0.05% Fast Green and the expression vector was injected into the lumen of the posterior neural tube. A drop of PBS with antibiotic is added before closing the egg and incubating it overnight. Scrambled and siRNA vectors were electroporated at 1.5 µg/µl. All forms of MMP14 were electroporated at 1-1.5 µg/µl. Par3 vector (Afonso and Henrique, 2006) was electroporated at 0.6 µg/µl. The Par3MO was electroporated at 500 µM.

Antibody against MMP14
Full-length chick MMP14-Flag was expressed in human HEK293T cells using Extreme gene HP (Sigma, 6366244001) and purified using anti-Flagagarose beads (Sigma, M8823). Balbc mice (Jackson Laboratories) were immunized with 100 µl of purified MMP14-Flag with beads using complete and incomplete Freund's adjuvant. Tail bleeds were collected after two boosts and were tested against Hek293T cells transfected with Chick MMP14 and RFP-Flag (negative control).

Histology
After fixation in 4% paraformaldehyde or immunostaining, embryos were washed in phosphate buffer (PB) and incubated in PB/15% sucrose overnight at 4°C. Embryos were transferred in PB/15% sucrose/7.5% gelatin (PBSG) for 2 h at 42°C. Small weighting boats were used as molds. A thin layer of PBSG was deposited at the bottom and left to set. Embryos were transferred onto the layer using a plastic 2.5 ml pipette. Each embryo was placed in a drop of PBSG. When drops were set, the weighting boat was filled with PBSG and left to set on the bench. The dish was incubated at 4°C for 1 h to harden the gelatin. Individual blocks were cut under a dissecting microscope to position the embryo in the desired orientation for sectioning. Blocks were frozen in isopentane (Sigma, 615838) at −70°C and stored at −70°C until sectioning.
The Par3 morpholino was designed to target the ATG codon and was produced by Genetools. The sequence of the Par3 morpholino is 5′gaagcacacggtgactttcatcctg-3′.

FRET analysis
Chick MMP14 FRET probe was made by replacing human pro-MMP2 MMP14 cleavage site with a chicken pro-MMP2 MMP14 cleavage site in the human MMP14 FRET reporter (Ouyang et al., 2008). FRET acceptor photobleaching was performed on cryosections using a Zeiss 710 confocal microscope. Individual migratory NC cells located in ventromedial pathway were imaged at 514 and 454 nm, with a 40× objective at zoom 8 to acquire fluorescence from YPet and CFP, respectively. The CFP image was acquired at 70% of the dynamic range. The bleaching module was then used to deliver a series of fast illuminations at 514 nm at 100% to bleach YPet. Bleaching automatically stops when the YPet intensity reaches 20% of its initial intensity (typically 5 to 10 illuminations, total bleaching time of ∼2-5 s). Then, YPet and CFP are imaged again with the same settings as the prebleach images. FRET index is calculated using FIJI/ImageJ. Each image was converted to 32-bit and thresholded, and the background was set to NaN. Mean gray level intensity are calculated on the whole image for CFP prebleaching and post-bleaching images. Then the following calculation was carried out in Excel [(CFPpost-CFPpre)/CFPpost]×100=FRET index.

Chick MMP14 expression vectors
Full-length MMP14 was synthetized by GenScript using codon optimization to reduce the amount of GCs. Deletions of the catalytic domain, the cytoplasmic tail and the point mutant forms were made by PCR strategies from the optimized sequence and subcloned into the pCAGGS vector for electroporation.

Image acquisition and analysis
Images were captured using a stereomicroscope Leica MZF10F equipped with a Leica DFC450C camera and LAS software, a Nikon Eclipse 80i equipped with a DXM1200C camera and the NIS-elements software, a Zeiss AxioImager 2 equipped with a Hamamatsu ORCA Flash 4 camera and the Zen2 software, and a confocal Zeiss 710. Images were then processed using FIJI.

Analysis of cell migration
On cryosections, we measured the net dorsoventral distance of the most dorsal GFP + cell (d) and the dorsoventral length of neural tube (D) on the corresponding section. We calculated the net distance of the most dorsal GFP + cells with the ratio r=d/D (Fig. S11). On NC cultures, we analyzed cell dispersion by dividing the area explored by the GFP + cell area by that explored by SiR-DNA + cells. To measure the area, a line was drawn to link all outer GFP + cells together. The same was done for siR-DNA + cells (Fig. S12).

Statistics
Statistical analyses were performed using Prism 6 (GraphPad). Datasets were tested for Gaussian distribution. Student's t-tests or ANOVA followed by multiple comparisons were used with the appropriate parameters depending on the Gaussian versus non-Gaussian characteristics of the data distribution. Significance threshold was set at P<0.05. Proportions were compared according to Taillard et al. (2008). For the box and whisker plots, the box extends from the 25th to the 75th percentile, and the whiskers show the extent of the whole dataset. The median is plotted as a line inside the box. C.A. was the recipient of a PhD fellowship from the Ministere de l'Enseignement supeŕieur, de la Recherche et de l'Innovation, a PhD Fellowship extension from the Association pour la Recherche sur le Cancer (ARC) and an international exchange grant from the UniversitéToulouse III -Paul Sabatier (ATUPS). Deposited in PMC for release after 12 months.