Mesenchymal cells reactivate Snail1 expression to drive three-dimensional invasion programs

Snail1, a zinc finger–type transcriptional repressor, initiates an epithelial–mesenchymal transition (EMT) that is critical for the morphogenetic events that characterize developmental programs such as gastrulation (Carver et al., 2001; Nieto, 2002; Murray et al., 2007). Snail1 triggers this transdifferentiation program, in part, by repressing epithelial markers and related cell–cell junction proteins while coordinately acting as a major cytoskeletal regulator (Batlle et al., 2000; Cano et al., 2000; Moreno-Bueno et al., 2006; Peinado et al., 2007). The aberrant postnatal expression of Snail1 is sufficient to confer a mesenchymal, fibroblast-like phenotype in differentiated epithelial cells during pathological states associated with cancer and fibrosis (Yook et al., 2005, 2006; Boutet et al., 2006; Moreno-Bueno et al., 2006; Olmeda et al., 2007a,b; Peinado et al., 2007).

At sites of active tissue remodeling, changes in vascular permeability disperse serum-derived soluble growth factors within the interstitial compartment, which serve to activate signal transduction cascades in resident fibroblasts (Martin, 1997; Bhowmick et al., 2004; Dong et al., 2004; Orimo et al., 2005; Klapholz-Brown et al., 2007). Accordingly, these agonists trigger changes in gene expression programs that shift the fibroblast phenotype from a quiescent status to an “activated” state characterized by increased proliferation, tissue-invasive activity, and the induction of angiogenesis (Martin, 1997; Iyer et al., 1999; Bhowmick et al., 2004; Sabeh et al., 2004; Klapholz-Brown et al., 2007). Growth factors capable of promoting the activated fibroblast phenotype, such as PDGF-BB, have been identified (Dong et al., 2004; Gao et al., 2005), but key transcription factors that regulate downstream gene programs remain largely uncharacterized. Herein, we identify Snail1 as a critical regulator of both fibroblast gene expression programs and fibroblast function in vitro as well as in vivo. The results demonstrate that Snail1, a master EMT inducer, continues to subserve vital cellular functions following mesenchymal cell terminal differentiation.

Under serum-free conditions, fibroblasts do not express detectable levels of Snail1 mRNA or protein (Fig. 1, A and B). In contrast, in the presence of 10% serum or PDGF-BB, both Snail1 mRNA and intranuclear protein levels are strongly induced in mouse as well as human fibroblasts (Fig. 1, A–C). In epithelial cells, Snail1 protein half-life is controlled by GSK3-β–dependent and –independent ubiquitination pathways that lead to proteasome-mediated Snail1 destruction (Zhou et al., 2004; Yook et al., 2005, 2006; Vernon and LaBonne, 2006). As expected, blockade of fibroblast proteasome activity with the inhibitor, MG132, results in a marked accumulation of the Snail1 protein (Fig. 1 B). In the GSK3-β–dependent pathway, Snail1 is marked for ubiquitination after phosphorylation of its N-terminal domain (Zhou et al., 2004; Vernon and LaBonne, 2006; Yook et al., 2006). As PDGF-BB signaling can inhibit GSK3-β activity via the phosphatidylinositol 3-kinase (PI3K)/Akt-dependent phosphorylation of GSK3-β serine 9 (Ser⁹; Julien et al., 2007), Akt phosphorylation, Ser⁹ phosphorylation, and Snail1 protein levels were monitored in fibroblasts in the absence or presence of the PI3K inhibitor, LY 294002. As predicted, treatment of serum-starved fibroblasts with PDGF-BB induces an increase in phospho-Akt and Ser⁹ GSK3-β levels in tandem with an increase in Snail1 protein (Fig. 1, D and E). In the presence of LY 294002, however, both Akt and Ser⁹ GSK3-β phosphorylation are blocked, and Snail1 levels fall to undetectable levels (Fig. 1, D and E).

Snai1-deficient mice die early in development before the differentiation of mesodermal lineages (Carver et al., 2001; Peinado et al., 2007). Hence, we generated mice in which Snai1 could be inactivated in selected tissues by Cre/loxp-mediated recombination (Fig. 2, A–C). Fibroblasts isolated from a Snai1^+/fl mouse were treated with an adenoviral Cre recombinase construct (adeno-Cre) or a control adenovirus (β-galactosidase [β-gal]), and recombination at the Snai1 locus was verified by PCR. As shown in Fig. 2 D, although adeno–β-gal–infected fibroblasts yield P1/P2 amplicons corresponding to both the wild-type and loxp alleles of Snai1, adeno-Cre–infected fibroblasts yielded a single amplicon corresponding to the wild-type Snai1 allele with P1 and P2, as well as a P3/P4 amplicon representing the Snai1^- allele. Fibroblasts isolated from Snai1^fl/fl mice and infected with adeno-Cre display a 95% reduction in Snail1 mRNA, whereas Snail1 protein expression is undetectable by Western blotting or immunocytochemistry (Fig. 2, F and G).

In addition to its well-defined role in promoting EMT, Snail1 can regulate cell cycle progression and sensitivity to proapoptotic stresses (Vega et al., 2004; Barrallo-Gimeno and Nieto, 2005; Escriva et al., 2008). Snail1-deficient fibroblasts proliferate, however, at normal rates, with no observed changes in apoptosis under serum-free conditions (Fig. S1, A and B). Furthermore, though Snail1 can promote a motile phenotype in epithelial cells (Barrallo-Gimeno and Nieto, 2005; Peinado et al., 2007), Snail1-deleted fibroblasts migrate at rates comparable to wild-type fibroblasts in a two-dimensional wound assay (Fig. S1, C and D). Likewise, whereas increased fibronectin synthesis and matrix assembly are characteristic features of EMT programs (Barrallo-Gimeno and Nieto, 2005; Peinado et al., 2007), Snail1-deficient fibroblasts deposit a fibronectin matrix at rates comparable to control fibroblasts (Fig. S1 E). Consequently, insights into Snail1 function were alternatively sought by interrogating the gene expression patterns of Snail1-deleted fibroblasts. Recent studies have demonstrated that cell behavior in vitro more closely recapitulates that observed in vivo when cells are cultured within a 3D ECM (Hotary et al., 2003; Yamada and Cukierman, 2007; Zhou et al., 2008). Hence, Snail1 wild-type and deficient cells were suspended in type I collagen matrices, the dominant matrix component of interstitial tissues (Grinnell, 2003; Sabeh et al., 2004), and subjected to transcriptional profiling. Using cutoffs of P ≤ 0.005 and a minimum fold change of 1.5, Snail1 deficiency in fibroblasts exerts a global effect on transcription, with >1,000 significant changes in gene expression detected (Fig. 3 A and Table S1). Gene ontology (GO) analysis further demonstrates that Snail1 governs multiple processes critical to fibroblast motile behavior, including adhesion, migration, and proteolysis (Fig. 3 B). Snail1 deletion did not trigger a mesenchymal-to-epithelial transdifferentiation process, as assessed by transcriptional analysis, which suggests that Snail1 is required for the induction, but not maintenance, of the mesenchymal phenotype during development.

To assess the consequences of Snail1 loss on 3D ECM invasion, a critical component of fibroblast wound and tumor responses (Martin, 1997; Grinnell, 2003; Bhowmick et al., 2004; Sabeh et al., 2004; Orimo et al., 2005), cells were embedded within a 3D bed of type I collagen. Snail1-deleted, but not Snail1-rescued, fibroblasts displayed a significant defect in their ability to negotiate the type I collagen barrier (Fig. 4, A and B; and Fig. S2, A and B). Focusing on candidate genes implicated previously in 3D cell motility and invasion (Yamaguchi and Condeelis, 2007; Olson and Sahai, 2008; Sakurai-Yageta et al., 2008), probe sets corresponding to transcripts for cortactin (Cttn), enabled homologue (Enah), ezrin, moesin, rhoA, ROCK1, myosin light chain kinase, tropomyosin, and membrane type-1 matrix metalloproteinase (MT1-MMP; Mmp14) are significantly altered in Snail1-deficient fibroblasts (Fig. 3 C). Consistent with the altered patterns of gene expression revealed by the microarray data and confirmed by quantitative PCR (Fig. S2C), Snail1-deficient fibroblasts exhibit a significant reduction in the cortactin-rich membrane protrusions that mark invadopodia, the actin-rich, cellular processes that focus proteolytic activity at sites of cell–ECM contact (i.e., 83.3 ± 8.3% cortactin-positive processes in wild-type fibroblasts vs. 31.2 ± 13.2% in Snail1-deleted fibroblasts; P < 0.01; Fig. 4 C; Gimona et al., 2008).

As cortactin-rich invadopodia play a critical role in recruiting MT1-MMP, a membrane-anchored collagenase critical for cell invasion, to zones of pericellular proteolysis (Artym et al., 2006; Clark et al., 2007; Li et al., 2008), wild-type and Snail1-deficient fibroblasts were cultured with fibrillar gels of Alexa 594–labeled type I collagen, and collagenolysis was monitored (Sabeh et al., 2004). Whereas wild-type fibroblasts generate collagenolytic zones that are associated with adhesive sites enriched for actin spikes and cortactin, Snail1-deficient fibroblasts exhibit a significantly diminished ability to degrade collagen or mobilize invadopodia-like structures (Fig. 4, D and E). In accordance with these collagenolytic defects, invadopodial clusters of MT1-MMP and cortactin localized at the fibroblast–collagen interface are reduced by ∼80% in Snail1-deficient cells (Fig. 4, F and G). Reconstitution of Snail1-deficient fibroblasts with full-length human Snail1 normalizes expression of cortactin and MT1-MMP (Fig. S2 D). Furthermore, consistent with GO enrichment scores that did not detect changes in cell cycle or apoptosis regulation, wild-type or Snail1-deleted fibroblasts embedded within 3D collagen gels proliferate at indistinguishable rates (7.8 ± 3.2% Ki67-positive for Snail1 wild-type cells vs. 8.1 ± 1.4% Ki67-positive for Snail1-null cells; n = 3) and display similar low levels of apoptosis (Snail1 wild-type, 1.6 ± 0.8%; Snail1-null, 1.4 ± 1.4%; assessed by TUNEL; n = 3).

Though Snail1-deficient cells display defects in the pericellular proteolysis and invasion of homogeneous collagenous barriers in vitro, connective tissue barriers in vivo are more complex, multimolecular composites of ECM macromolecules (Grinnell, 2003; Hotary et al., 2003; Yamada and Cukierman, 2007; Zhou et al., 2008). As such, wild-type and Snail1-deleted fibroblasts were cultured atop the chorioallantoic membrane (CAM) of live chick embryos (Sabeh et al., 2004), a tissue characterized by a type IV collagen-rich basement membrane and an underlying interstitium containing both type I and type III collagens (the stroma also contains blood vessels circumscribed by type IV collagen-positive basement membranes; Fig. 5 A). Although wild-type fibroblasts efficiently breach the CAM basement membrane and invade into the underlying stroma, Snail1-deficient fibroblasts exhibit a complete defect in invasion and fail to penetrate the CAM surface (Fig. 5, B and C), a phenotype identical to that described previously for MT1-MMP–null fibroblasts (Sabeh et al., 2004). In vivo, fibroblasts can initiate neovascularization during wound healing (Martin, 1997), but Snail1-deficient fibroblasts also demonstrate a significantly attenuated ability to induce neovessel formation (Fig. 5, B and D). Neither proliferative nor apoptotic indices of the fibroblasts are affected in the CAM model (Fig. 5, E and F). Collectively, the data identify Snail1 as a master regulator of activated fibroblast function in vivo by controlling tissue-invasive as well as proangiogenic functions.

Snail1 exerts global effects on epithelial cell gene expression by binding consensus sequences within the promoter regions of target genes while recruiting histone deacetylases, arginine methyltransferase, and DNA methyltransferases to chromatin remodeling complexes (Peinado et al., 2007; Herranz et al., 2008; Hou et al., 2008). Despite the remarkable range of Snail1's impact on epithelial cell fate determination, a functional role for Snail1 in terminally differentiated mesenchymal cells has not been explored previously. Unexpectedly, under 3D culture conditions, GO analyses revealed that major shifts had occurred in fibroblast behavior in the absence of Snail1 expression, with changes concentrated in functional programs tightly linked to cell adhesion, migration, proteolysis, and morphogenesis. Among Snail1-regulated targets, cortactin has been found to regulate MT1-MMP–dependent proteolysis, an activity critical for mesenchymal cell trafficking through ECM barriers (Chun et al., 2004; Sabeh et al., 2004; Filippov et al., 2005; Artym et al., 2006; Hotary et al., 2006; Clark et al., 2007). As such, the defects in cortactin and MT1-MMP expression and function observed in Snail1-deficient fibroblasts, in tandem with predicted changes in accessory molecules such as rhoA, ROCK, myosin light chain kinase, and tropomyosin, correlated with a marked loss in collagenolytic potential as well tissue-invasive activity in vitro and in vivo. Snail1-deleted fibroblasts were also unable to initiate an angiogenic response, a result likely consistent with the ability of MT1-MMP to induce angiogenesis by generating bioactive collagen fragments, regulating VEGF expression, or mediating semaphorin 4D shedding (Sounni et al., 2004; Weathington et al., 2006; Basile et al., 2007).

To date, analyses of Snail1 function in mammalian cells have focused on the ability of the transcription factor to initiate the transdifferentiation of normal or neoplastic epithelial cells. The findings presented herein, coupled with the fact that Snail1 protein is expressed in fibroblasts localized at damaged or carcinomatous tissues in vivo (Franci et al., 2006; Rosivatz et al., 2006), demonstrate that Snail1 activity is not confined to epithelial cells alone. Although our studies have focused on the role of Snail1 in regulating fibroblast function, it is intriguing to note that Snail1 may also be expressed in the neoplastic mesenchyme (Franci et al., 2006). Indeed, large T antigen/Ras-transformed fibroblasts are similarly reliant on Snail1 for the expression of a tissue-invasive phenotype (Fig. S3). Hence, in addition to its essential roles in EMT, we propose that Snail1 now be considered as a transcription factor capable of exerting key regulatory effects in the mesenchyme during development as well as disease.

To generate the Snai1 conditional knockout mouse, a targeting vector was constructed consisting of a flippase recognition target (FRT)-flanked phosphoglycerine kinase (PGK)-neo cassette 3′ to the loxp-flanked exon 3 of mouse Snai1, predicted to encode the two C-terminal zinc finger domains as well as the polyadenylation sequence for the Snai1 mRNA. Approximately 4 kb of flanking genomic sequence was then inserted at the 5′ and 3′ ends of the loxp-flanked exon and FRT-flanked neomycin cassette to promote homologous recombination. The linearized targeting vector was electroporated into W4 embryonic stem cells (Auerbach et al., 2000), and stable transfected clones were selected with G418. Clones were screened for targeting of the Snai1 by Southern blotting, and recombination was verified at both the 5′ and 3′ ends of the construct. Of 100 clones screened, 3 were identified with correct targeting and used for injection into C57BL/6NCrl × (C57BL/6J × DBA/2J)F1 blastocysts to produce chimeric mice. Out of three chimeric lines produced, two lines transmitted the targeted Snai1 allele (Snai1^tm1Stjw) through the germ line. β-actin FLPe mice (stock No. 003800; Jackson ImmunoResearch Laboratories) were backcrossed to C57/6J mice to generate a congenic strain before mating with chimeras for excision of the FRT-flanked PGK-neo cassette to generate the Snai1^fl (Snai1^tm2Stjw) allele. Snai1^fl/fl homozygous conditional knockout mice were born in the expected Mendelian ratios, which indicates that the Snai1^fl allele functions equivalently to the wild-type Snai1 allele.

The 173EC2, 173EC3, and Sn9H2 anti-Snail1, and anti–MT1-MMP mAb1 antibodies were prepared and characterized as described previously (Franci et al., 2006; Rosivatz et al., 2006; Ingvarsen et al., 2008). The anti–GSK3-β phospho-serine 9 and anti-Akt phospho-serine 473 antibodies were obtained from Cell Signaling Technology. The anti-cortactin, anti-actin, and anti-Ki67 antibodies were obtained from Santa Cruz Biotechnology, Inc., Sigma-Aldrich, and Abcam, respectively. Adeno–β-gal and Adeno-Cre (transgenes driven by a cytomegalovirus promoter) were obtained from the University of Michigan Vector Core. LY-294002 and MG132 were obtained from EMD and Sigma-Aldrich, respectively. Apoptotic cell death was measured with an in situ apoptosis detection kit (ApopTag Red) according to the manufacturer's instructions (Millipore).

For Western blotting, the following primary antibody dilutions were used: 173EC2 hybridoma supernatant (1:40), 173EC3 affinity-purified antibody (1:10,000), anti–GSK3-β phospho-serine 9 and anti-Akt phospho-serine 473 antibodies (1:1,000), and anti-actin (1:4,000).

Quantitative PCR was performed using the SYBR green PCR master mix (Applied Biosystems) according to the manufacturer's instructions. Primers for mouse cortactin were: forward, 5′-GCAGCCATCCCAGGTGTTTTAGTT-3′, and reverse, 5′-CTTTGGTCCCCTTTCCTCCTCTTC-3′; mouse MT1-MMP primers were: forward, 5′-TGATTCTGCCGAGCCCTGGACTGT-3′, and reverse, 5′-TGAGGGGGCATCTTTGTGGGTGAC-3′; mouse Snail primers were: forward, 5′-CTGCTTCGAGCCATAGAACTAAAG-3′, and reverse, 5′-GAGGGGAACTATTGCATAGTCTGT-3′; and glyceraldehyde 3-phosphate dehydrogenase primers were: forward, 5′-CCAAGGTCATCCATGACAACT-3′, and reverse, 5′-GTCATACCAGGAAATGAGCTTGACA-3′.

For Snail1 immunocytochemistry, cells were fixed in 4% paraformaldehyde, permeabilized with 1% sodium dodecyl sulfate, denatured with 6M urea and 0.1% glycine, pH 3.5, blocked with 3% goat serum, and incubated with either 173EC2 (1:5) or 173EC3 (1:1,000) overnight, followed by detection with Alexa 488–labeled anti–mouse secondary antibody (Invitrogen). The Alexa 532–labeled anti–MT1-MMP mAb1 was used at 5 µg/ml, and the anti-cortactin antibody was used at a dilution of 1:40 after paraformaldehyde fixation and permeabilization with Triton X-100. The anti-cortactin antibody was detected with an Alexa 488–labeled, anti–rabbit secondary antibody (Invitrogen). Cells were counterstained with either 4′,6-diamidino-2-phenylindole or propidium iodide (Invitrogen). Confocal images of cells were acquired on a confocal microscope (FV500) using a 60× water immersion lens with a 1.20 numerical aperture using FluoView software (all from Olympus). All images comparing Snail1 wild-type and deficient cells were acquired with equal photomultiplier tube intensity and gain settings. Phase contrast images were acquired with a inverted microscope (DM-ILB; Leica) with a 20× objective and 0.40 numerical aperture, and CAM images were acquired on a microscope (DM-LB; Leica) with a 20× objective and 0.50 numerical aperture (Leica). Phase contrast and CAM images were acquired and analyzed with SPOT cameras and software (Diagnostic Instruments, Inc.).

To analyze MT1-MMP in invadopodia, confocal cross sections of invadopodia costained for MT1-MMP and cortactin were analyzed with MetaMorph software (MDS Analytical Technology). Collagen degradation was analyzed using ImageQuant 5.2 software (GE Healthcare). Invadopodial clusters were traced, and the areas containing MT1-MMP and cortactin colocalization were quantified.

To analyze 3D invasion, 50,000 fibroblasts were embedded in 100 µl of type I collagen gel (2.2 mg/ml) isolated from rat tail (Sabeh et al., 2004). After gelling, the plug was embedded in a cell-free, 500 µl collagen gel (2.2 mg/ml) cultured within a 24-well plate. After allowing the surrounding collagen matrix to gel (1 h at 37°C), fibroblast invasion was stimulated with serum and 10 ng/ml PDGF-BB (Millipore). Invasion distance from the inner collagen plug into the outer collagen gel was quantified. CAM invasion assays were conducted using 11-d-old chick embryos where fibroblasts labeled with Fluoresbrite-carboxylated nanospheres (Polysciences, Inc.) were cultured atop the CAM for 24 h. Invasion depth was defined as the leading front of at least three invading cells in 10 fields in frozen sections (Sabeh et al., 2004). The invasion area was defined as the area occupied by invading cells in at least 10 fields (Sabeh et al., 2004), whereas angiogenesis was quantified by type IV collagen staining (Bajou et al., 2001). Snail1 conditional knockout transformed fibroblasts were generated by isolating embryonic day 13.5 mouse embryonic fibroblasts followed by serial retroviral introduction of the polyoma large T and activated Ras oncogenes (Land et al., 1983).

Total RNA was isolated from fibroblast cultures in 3D collagen, then labeled and hybridized to mouse 430 2.0 cDNA microarrays (Affymetrix). Three replicates each of Snail1 wild-type and deficient cultures were analyzed by the University of Michigan Microarray Core. Differentially expressed probe sets were determined using a minimum fold change of 1.5 and a maximum p-value of 0.005. GO analysis was performed to identify biological processes transcriptionally regulated by Snail. GO coefficients were calculated as −log(p-value).

Fig. S1 shows analysis of fibroblast function under 2D culture conditions. Fig. S2 shows transcript analysis by quantitative PCR and rescue of the Snail1-deficient phenotype by reconstitution with Snail1. Fig. S3 characterizes and analyzes the tissue-invasive potential of wild-type and Snail1-deficient transformed mouse embryonic fibroblasts. Table S1 shows a list of genes differentially expressed between Snail1 wild-type and deficient fibroblasts.

This work was supported by the National Institutes of Health (grant CA116516 to S.J. Weiss) and funds from the Cancer Biology Training Program (grant T32-CA009676) and the Breast Cancer Research Foundation. We thank E. Hughes (Transgenics Core) for technical assistance and Steve Lentz of the Morphology and Image Analysis Core (DK20572 and DK34933) for valuable assistance with confocal microscopy.