ADAM12 induction by Twist1 promotes tumor invasion and metastasis via regulation of invadopodia and focal adhesions

During metastasis, carcinoma cells acquire the ability to invade surrounding tissues and intravasate through the endothelium to enter the systemic circulation. The basic helix-loop-helix transcription factor Twist1 is a potent promoter of tumor metastasis and its expression is associated with poor outcome in many human cancers. Mechanistically, Twist1 can induce epithelial–mesenchymal transition (EMT) to allow tumor cells to dissociate and promote dissemination (Yang et al., 2004; Tsai et al., 2012). Furthermore, Twist1 induces invadopodia formation and extracellular matrix (ECM) degradation to promote tumor invasion (Eckert et al., 2011). Invadopodia are specialized actin-based membrane protrusions found in cancer cells and degrade the ECM via localized proteolytic activity (Chen, 1989; Tarone et al., 1985). Their ability to mediate focal ECM degradation suggests that they have a critical role in tumor invasion and metastasis. As actin-based structures, invadopodia contain an F-actin core and actin regulatory proteins, such as cortactin, WASp (also known as WAS) and the Arp2/3 complex (Linder, 2007). The SH3 domain-rich proteins Tks4 (also known as SH3PXD2B) (Buschman et al., 2009) and Tks5 (also known as SH3PXD2A) (Seals et al., 2005) function as essential adaptor proteins in the clustering of structural and enzymatic components of invadopodia. The matrix degradation activity of invadopodia has been associated with a large number of proteases, including membrane-type matrix metalloproteinases, such as MT1-MMP (Linder, 2007). Invadopodia formation requires the tyrosine phosphorylation of several invadopodia components, including cortactin (Ayala et al., 2008), Tks4 (Buschman et al., 2009) and Tks5 (Seals et al., 2005), likely by Src family kinases. Our previous study showed that Twist1 induces expression and activation of platelet-derived growth factor receptor α (PDGFRα), which in turn activates Src kinase to promote the assembly of invadopodia (Eckert et al., 2011). However, it is unclear whether Twist1 also activates additional downstream genes to promote the assembly and function of the complex invadopodial structure.

ADAM12 is a member of the disintegrin and metalloproteinase family, the members of which possess both a metalloproteinase and a disintegrin domain (Kveiborg et al., 2008). ADAM12 is associated with disease stage in breast and bladder cancer (Frohlich et al., 2006; Narita et al., 2012). Interestingly, ADAM12 is induced during transforming growth factor β (TGFβ)-induced EMT (Ruff et al., 2015). In addition, expression of ADAM12 has been reported in liver, lung, stomach and colon cancer, as well as in glioblastoma (Kveiborg et al., 2008). Soluble ADAM12, either a splice variant without the transmembrane and cytoplasmic domains, denoted ADAM12S, or a cleaved form of full-length ADAM12, is found at high levels in the urine of breast cancer patients and is associated with cancer progression (Roy et al., 2004). Expression of ADAM12 under the control of a mammary-gland-specific promoter was not sufficient for tumorigenesis in a transgenic mouse model (Kveiborg et al., 2005). Induction of breast cancer by crossing this mouse into a mouse mammary tumor virus (MMTV)-polyoma middle T (PyMT) background, however, led to a significant increase in tumor growth rate and metastasis to the lung, with an additional expansion of the stromal compartment in the primary tumor (Kveiborg et al., 2005). Similarly, in a mouse model of prostate cancer, ADAM12 was found to be essential for prostate tumor growth (Peduto et al., 2006). Combined, these models demonstrate a clear role for ADAM12 in tumor growth and progression, but do not address potential roles in mediating local invasion or cell motility.

Interestingly, ADAM12 localizes to invadopodia in Hs578T cells (Courtneidge et al., 2005). In addition, both ADAM12 and the highly related ADAM19 directly interact with the vital invadopodia scaffolding protein Tks5 (Abram et al., 2003). Albrechtsen et al. developed an antibody directed against ADAM12 that induces clustering of ADAM12 and subsequent invadopodia formation at sites of clustering, a process that is dependent on the metalloproteinase activity of ADAM12 (Albrechtsen et al., 2011). Furthermore, ADAM12-mediated heparin-binding epidermal growth factor (HB-EGF) shedding is shown to potentiate invadopodia formation in response to hypoxia (Díaz et al., 2013). The ability of ADAM12 to interact with integrins through the disintegrin domain suggests a possible role for ADAM12 in modulating integrin interactions at invadopodia (Kawaguchi et al., 2003).

In this study, through a series of experiments in human cell cultures and in mouse tumor metastasis models, we explored the link between Twist1 and ADAM12 in tumor invasion and metastasis.

Given the induction of invadopodia following Twist1 expression, we first sought to identify candidate genes downstream of Twist1 involved in regulating invadopodia-mediated ECM degradation. Using a human breast cancer RNA-Seq data set (817 samples) from The Cancer Genome Atlas (TCGA), we examined genes that co-expressed with Twist1 and found that ADAM12 mRNA expression was highly correlated with Twist1 mRNA expression in human breast cancers (Fig. 1A). ADAM12 is ranked 61st among 16,494 genes (99.6 percentile) for association with Twist1 (Fig. 1B). This correlation was further validated in four published large human breast tumor gene expression data sets summarizing 860 primary breast cancers (Pawitan et al., 2005; Sotiriou et al., 2006; Wang et al., 2005; Miller et al., 2005). In each data set, we calculated the rank-based Spearman correlation coefficient between Twist1 and all 22,282 genes in the array, including ADAM12. Expression of Twist1 and ADAM12 positively correlated with correlation coefficients ranging from 0.49 to 0.58 (Fig. S1A–D). Furthermore, ADAM12 was consistently among the top-ranked genes associated with TWIST1 (99.2 percentile in GSE2990, 99.1 percentile in GSE2034, 99.3 percentile in GSE3494, and 98.4 percentile in GSE1456) in all four breast cancer data sets (Fig. S1E–H).

To determine whether ADAM12 was in fact upregulated following Twist1 activation, we first utilized a Twist1-estrogen receptor (Twist1-ER) fusion protein construct that can be activated to enter the nucleus to drive Twist1-dependent transcription upon addition of 20 nM 4-hydroxytamoxifen (4-OHT) (Mani et al., 2008). Upon Twist1-ER activation at time points from day 0 to 14, substantial induction of ADAM12 expression was observed (Fig. 1C). In HMLE cells overexpressing Twist1 (HMLE-Twist cells), we also observed strong upregulation of ADAM12 protein; the two bands likely correspond to long and short isoforms of ADAM12 resulting from alternative splicing (Fig. 1D) (Wewer et al., 2006). To determine whether endogenous ADAM12 expression is regulated by Twist1 in human breast cancer cells, we knocked down Twist1 in Hs578T cells through lentiviral-mediated shRNA delivery (Yang et al., 2004). Knocking down Twist1 also significantly reduced the expression of ADAM12 (Fig. S2A). We therefore concluded that Twist1 is both necessary and sufficient for the induction of ADAM12 expression.

Because our previous studies show that Twist1 plays a critical role in tumor invasion and metastasis (Yang et al., 2004; Eckert et al., 2011), we set out to investigate the effects of ADAM12 on Twist1-induced invasion and metastasis. We first virally transduced HMLE-Twist1 cells with two shRNAs against ADAM12 [shADAM12.1 (denoted as shA12.1 in figures) and shADAM12.4 (denoted as shA12.4 in figures)]. Both constructs were efficient at stably knocking down ADAM12 expression at the protein level in HMLE-Twist1 cells (Fig. 1D). It is important to note that knockdown of ADAM12 did not affect cell proliferation, nor did it result in the EMT phenotype, in HMLE-Twist1 cells (Fig. S2B–D). HMLE-Twist1 cells expressing shADAM12 constructs still downregulated epithelial markers, such as E-cadherin (also known as CDH1), and upregulated mesenchymal markers, such as neuronal cadherin (N-cadherin, also known as CDH2) (Fig. 1D). Importantly, knockdown of ADAM12 did not affect levels of MT1-MMP (also known as MMP14) (Fig. 1D). Therefore, ADAM12 is not required for the ability of Twist1 to induce EMT in 2D culture.

Next, we tested whether ADAM12 was required for invasion in a 3D culture system by embedding HMLE-Twist1 cells expressing the control or shADAM12 constructs in a Matrigel–collagen culture system. After 1 week in 3D Matrigel–collagen culture, cells expressing the control knockdown had projected into the surrounding matrix; by contrast, HMLE-Twist cells expressing shADAM12 (HMLE-Twist1-shADAM12 cells) remained largely rounded with few projections into the surrounding ECM (Fig. 1E). Further quantification showed a significant reduction in the 3D invasiveness of HMLE-Twist1-shADAM12 cells (Fig. 1F), suggesting a critical role for ADAM12 in promoting invasion in the 3D culture.

To determine whether ADAM12 promotes Twist1-induced metastatic dissemination in vivo, we first transformed the HMLE-Twist1 cells expressing control shRNA (shCtrl) or shADAM12 constructs with oncogenic H-Ras and labeled the cells with GFP. Then, we subcutaneously injected the cells, along with Matrigel, into the flanks of nude mice. Approximately 40 days later, when tumors reached 1.5 cm in diameter, we euthanized the mice and collected lung and primary tumor tissue for further analysis. Knocking down ADAM12 did not affect cell proliferation in culture (Fig. S2B) or primary tumor growth in mice (Fig. 2D). To examine local invasion, we stained tumor cells for large T antigen as the HMLE cells were immortalized with SV40 large T antigen. Tumors compromising HMLE cells overexpressing H-Ras (HMLER) expressing a control shRNA presented poorly defined borders, and single tumor cells often projected into the surrounding tissue (Fig. 2A); by contrast, the periphery of HMLER tumors expressing shADAM12 constructs remained largely encapsulated in a layer of connective tissue. When we examined the lungs for GFP-positive disseminated tumor cells, we observed a significant reduction in disseminated cells in the HMLER-Twist1 tumors expressing the shADAM12 construct compared to in control tumors (Fig. 2B,C). We thus conclude that ADAM12 is necessary for local invasion and distant metastasis downstream of Twist1.

Next, we sought to understand the cellular function of ADAM12 in tumor invasion and metastasis. ADAM12 localizes at invadopodia and directly interacts with the invadopodia-specific scaffolding protein Tks5 (Abram et al., 2003). Given the crucial role of Twist1 in promoting invadopodia-mediated ECM degradation, we suspected that ADAM12 might be a key downstream mediator of Twist1 in promoting invadopodia formation. First, we verified that endogenous ADAM12 protein localized to invadopodia in Hs578T cells grown on collagen by staining for cortactin, F-actin and ADAM12. We observed extensive colocalization of ADAM12 with puncta of F-actin and cortactin corresponding to invadopodia (Fig. 3A,B). When ADAM12 was knocked down in both Hs578T and HMLE-Twist1 cells (Fig. 3C and Fig. 1C), we observed a significant reduction in F-actin- and cortactin-positive invadopodia formation. Interestingly, we observed that most of the cortactin was associated with cortical actin at the edge of the HMLE-Twist1 cells expressing shADAM12 constructs (Fig. 3D-G). Taken together, these data demonstrate that ADAM12 is required for TWIST1-induced invadopodia formation.

We then went on to test whether ADAM12 affects the ability of the cells to degrade ECM by performing the fluorescein isothiocyanate (FITC)–gelatin degradation assay. In these assays, areas of gelatin degradation or proteolysis are visualized as dark areas beneath the cell because the fluorescent FITC-labeled gelatin is degraded by the action of invadopodia. Indeed, a significant decrease in gelatin degradation was observed upon knockdown of ADAM12, indicating that ADAM12 is required for gelatin degradation (Fig. 4A,B). Knocking down ADAM12 also significantly reduced the ability of these cells to invade through Matrigel-coated Boyden chambers (Fig. 4C). Taken together, these data demonstrate that ADAM12 functions downstream of Twist1 to promote invadopodia-mediated matrix degradation and tumor invasion.

While characterizing the role of ADAM12 in regulating cell invasion, we tested whether ADAM12 regulates cell migration. Surprisingly, in both HMLE-Twist1 and Hs578T cells, knockdown of ADAM12 drastically reduced cell migration in Transwell migration assays (Fig. 5A,B). To further assay defects in cell motility, we also performed scratch assays with the same cell lines as above. A significant decrease in motility upon knockdown of ADAM12 was observed over all time points examined (Fig. S3A,B). Taken together, these results show that, in addition to promoting invasion, ADAM12 also regulates cell motility. To understand the underlying mechanism, we first noticed that upon knocking down ADAM12, these cells appeared to be significantly larger and more spread out than the control cells (Fig. 5C). When we quantified by measuring the relative surface area of each cell, we noted a significant difference in apparent cell size (Fig. 5D). Increased cell size in two dimensions can be caused by either an intrinsic increase in cell volume or an increase in cell spreading (Xiong et al., 2010). We therefore measured the volume of HMLE-Twist1 and Hs578T cells expressing shCtrl and shADAM12 constructs in suspension. There was no significant difference in cell volume upon knockdown of ADAM12 (Fig. S3C,D), implying that the apparent increase in cell size was caused by an increase in cell spreading rather than an actual increase in cell volume.

Increased cell spreading directly correlates with increased focal adhesion number (Tamura et al., 1998). This was particularly interesting in light of the fact that ADAM12 has been shown to interact with integrins involved in focal adhesions (Kveiborg et al., 2008). Indeed, when we stained Hs578T cells for ADAM12 and the focal adhesion protein vinculin, ADAM12 localized to focal adhesions on the periphery of the cell (Fig. 5E,F). This is consistent with previous reports that ADAM12 localizes to both invadopodia and peripheral areas of the cell (Abram et al., 2003; Kawaguchi et al., 2003). We therefore assayed focal adhesion formation by staining both Hs578T and HMLE-Twist1 cells expressing control and shADAM12 vectors for vinculin. A clear increase in focal adhesion formation (elongated ellipsoid vinculin staining adjacent to F-actin stress fibers) was observed upon ADAM12 knockdown (Fig. 6A,B). Stress fiber formation requires focal adhesion formation and is a surrogate marker for focal adhesion formation in many cell lines. Indeed, F-actin staining showed that stress fiber formation also significantly increased upon knockdown of ADAM12 (Fig. 6A,C).

Increased attachment strength is associated with focal adhesions and increased interactions with the surrounding matrix (Elineni and Gallant, 2011). To determine whether the increase in focal adhesion number upon ADAM12 knockdown affected the adhesive properties of the cells, we performed a cell detachment assay. Briefly, sub-confluent Hs578T cells expressing shCtrl or shADAM12 constructs were treated with 0.05% trypsin for the indicated amount of time before washing, fixing, and staining with Crystal Violet to quantify the percentage of cells remaining attached. Cells expressing shADAM12 were more resistant to trypsin-mediated detachment than those expressing shCtrl (Fig. 6D), indicating that ADAM12 expression reduces cell–ECM attachment.

Focal adhesion kinase (FAK, also known as PTK2) is a downstream effector of focal adhesion activation (Schlaepfer et al., 1999). Autophosphorylation of FAK at tyrosine 397 (Y³⁹⁷FAK) is an early event in focal adhesion formation and is required for subsequent signaling through Src kinase and phosphoinositide 3-kinase (PI3K) (Schlaepfer and Hunter, 1997). We therefore investigated the status of FAK phosphorylation upon knockdown of ADAM12 in Hs578T cells. Knocking down ADAM12 with both shRNAs induced a >4-fold increase in Y³⁹⁷FAK phosphorylation (Fig. 6E), supporting a role for ADAM12 in reducing cell–matrix interaction in tumor cells.

Finally, we analyzed the turnover of adhesions in Hs578T ADAM12-knockdown cells. Cells transfected with paxillin–mCherry were imaged using live-cell confocal microscopy for 1 h time-lapse acquisitions (Fig. 6F and Movie 1). We observed that ADAM12-knockdown cells formed focal adhesions that were significantly more stable than those formed in control cells (Fig. 6F,G). ADAM12-knockdown cells also formed significantly fewer dynamic adhesions compared to control cells (Fig. 6G). This effect on focal adhesion lifetime was also evident in the dynamic adhesion heat maps (Fig. S4), in which the color of the focal adhesion is a function of its lifetime and stability (Santiago-Medina et al., 2012). Taken together, these findings support a role for ADAM12 in maintaining adhesion turnover as well as a population of dynamic focal adhesions.

ADAM12 has multiple functional domains, including a metalloproteinase domain that cleaves other proteins, a disintegrin domain that interacts with integrins, and a cytoplasmic tail that might act as a scaffold for signaling or a physical linkage to the cytoskeleton (Kveiborg et al., 2008). To determine which domains of ADAM12 are necessary for its roles in focal adhesion and invadopodia regulation, we performed a series of rescue experiments with two mutant ADAM12 proteins, including a metalloproteinase-null ADAM12 (E351Q or A12ΔMMP, denoted as ΔMMP in Fig. 7A) and a disintegrin domain mutant ADAM12 (D488A or A12ΔDis, denoted as ΔDis in Fig. 7A). The metalloproteinase mutation disrupts a critical catalytic glutamate, whereas the disintegrin mutation replaces a critical charged aspartate with an uncharged alanine residue (Huang et al., 2005; Jacobsen et al., 2008).

Hs578T cells expressing shADAM12 (Hs578T-shADAM12 cells) transiently transfected with ADAM12 constructs and plated on 0.1% collagen were stained for vinculin and F-actin to visualize focal adhesions. These cells were transfected along with Lifeact–mCherry to identify the transfected cells. The metalloproteinase-null ADAM12 mutant constructs reduced focal adhesion number similar to the wild-type ADAM12, whereas the disintegrin domain mutant ADAM12 failed to do so (Fig. 7B), suggesting a key role for the disintegrin domain in regulating cell–matrix adhesion. Consistently, the metalloproteinase-null ADAM12 mutant could also rescue the focal adhesion defects owing to knockdown of ADAM12 in HMLE-Twist1 cells (Fig. S5). Next, invadopodia formation was assayed by quantifying the percentage of cells positive for punctate colocalization of F-actin and cortactin. The full-length ADAM12 was able to increase invadopodia formation to levels comparable to those in Hs578T-shCtrl cells, whereas both A12ΔMMP and A12ΔDis mutants failed to significantly rescue the invadopodia defects (Fig. 7C). Interestingly, co-transfection of both A12ΔMMP and A12ΔDis could significantly rescue invadopodia formation, suggesting that both mutant forms of ADAM12 can function in trans to support invadopodia formation. To assess whether restoration of invadopodia formation resulted in functional ECM degradation, Hs578T-shADAM12 cells transfected with various mutant ADAM12 constructs were assayed for their matrix degradation abilities. Indeed, A12ΔMMP and A12ΔDis together could significantly rescue matrix degradation (Fig. 7D). Taken together, these results indicate that ADAM12 utilizes multiple domains to mediate its role in ECM adhesion and degradation: the disintegrin domain of ADAM12 is required for focal adhesion turnover, while both the MMP domain and the disintegrin domains regulate invadopodia formation and function.

In this study, we identify ADAM12 as a novel downstream effector of Twist1 in promoting tumor cell migration, invasion and metastasis. We show that ADAM12 is required for invadopodia formation and matrix degradation to promote tumor invasion; furthermore, ADAM12 weakens focal adhesions to promote cell migration. Together, both functions of ADAM12 contribute to its crucial role in mediating Twist1-induced tumor metastasis in vivo.

Although ADAM12 is strongly upregulated in response to Twist1 activation, it is unlikely to be a direct transcriptional target of Twist1. The ADAM12 promoter lacks Twist1-responsive E-box elements, and it is induced relatively slowly in response to Twist1 to be a direct target. Interestingly, transfection of cells with Akt1 or inhibition of mammalian target of rapamycin (mTOR) led to increased transcription of ADAM12 in head and neck cancer cell lines (Le Pabic et al., 2005). Twist1 has been reported to increase Akt2 activation in breast cancer cell lines, suggesting a possible link between ADAM12 and Twist1 through regulation of Akt activity (Cheng et al., 2007). Furthermore, hypoxia-induced Notch activation has been shown to induce ADAM12 expression (Díaz et al., 2013). Hypoxia-inducible factor 1-α (HIF1α) also directly induces Twist1 (Yang et al., 2008), suggesting that hypoxia in the tumor microenvironment could induce ADAM12 via two distinct mechanisms, Notch activation and Twist1.

A potential role for ADAM12 in invadopodia has been suspected since the discovery that ADAM12 directly interacts with the invadopodia-specific scaffolding protein Tks5 (Abram et al., 2003). Characterization of the function of ADAM12 at invadopodia has been limited to the observation that inducing clustering of ADAM12 with a monoclonal antibody induces localized EGF shedding and invadopodia formation (Albrechtsen et al., 2011). In rescue experiments, we observed that both the disintegrin domain and metalloproteinase domain of ADAM12 were necessary for rescuing the invadopodia formation defect, and could function in trans to do so. Adhesion rings containing integrins are found at invadopodia and have been shown to be crucial for invadopodia maturation (Branch et al., 2012). Therefore, the disintegrin domain of ADAM12 could potentially regulate adhesion ring dynamics around invadopodia to affect invadopodia formation.

Several previous reports have also suggested that focal adhesions inhibit invadopodia formation. It has been hypothesized that focal adhesions sequester Src to areas at the periphery of the cell, reducing the pool of Src kinase available to phosphorylate invadopodia components (Chan et al., 2009). A similar phenotype was observed in melanoma, where suppression of focal adhesions led to decreased migration, but increased invadopodia (Kolli-Bouhafs et al., 2014). We show that ADAM12 knockdown significantly inhibited focal adhesion turnover. Disruption of the kinetics of focal adhesion turnover and maturation leads to decreased invasion in other model systems (Fukumoto et al., 2015). Therefore, it is also likely that ADAM12 regulates focal adhesion turnover to affect invadopodia formation and activity.

It is worth noting that ADAM12 has two mRNA splicing isoforms: ADAM12-L (transmembrane) and ADAM12S (secreted). Several studies have reported important roles of ADAM12S in promoting invasion in breast tumor cells (Roy and Moses, 2012) and during trophoblast fusion (Aghababaei et al., 2014), for which the proteolytic activity of ADAM12S is required. Although it is possible that ADAM12S could also regulate invadopodia function to affect invasion, the ability of ADAM12-L to rescue invadopodia and focal adhesion defects caused by ADAM12 gene knockdown observed in our study suggests that ADAM12-L plays a cell autonomous role in regulating invadopodia and focal adhesion dynamics, independent of ADAM12S.

Previous reports have implied a link between ADAM12 and tumor growth in vivo. In both mouse mammary cancer and prostate cancer models, transgenic expression of ADAM12 induces increased proliferation or reduced apoptosis in the primary tumor (Kveiborg et al., 2005; Peduto et al., 2006). Deletion of ADAM12 in MMTV-PyMT tumors reduced tumor cell proliferation caused by defective Akt protein activation (Frohlich et al., 2011). Similarly, knocking down Tks5, another critical invadopodia protein, in the same HMLER-Twist1 cells did not affect growth rate in vitro or in vivo, whereas knockdown of Tks5 in MDA-MB-231 cells resulted in reduced cell growth in 3D culture and in mice, without changing the cell proliferation rate on 2D plastic dishes (Blouw et al., 2015). By contrast, we observed no overt growth defects in vitro or in vivo. This is likely to be caused by differences in cooperative genetic and epigenetic changes in individual systems, and our use of constitutively active H-Ras in our xenotransplantation model of metastasis, which override the requirement for Akt protein activation. It is important to note that in all of these cases, invadopodia are essential for the invasive behavior of tumor cells, regardless of their possible requirements in tumor growth.

Combining these data, we propose a model in which ADAM12 increases focal adhesion dynamics and turnover. These changes increase both the migratory ability of cells and invadopodia formation. This combined increase in migration and invadopodia formation promotes Twist1-induced local invasion and eventual tumor metastasis.

HMLE cells were generated in the laboratory of Robert Weinberg (Whitehead Institute, Cambridge, MA). Hs578T cells were provided by S.A.C. HMLE cells were cultured as previously described (Yang et al., 2004). Hs578T breast carcinoma cells were maintained in a Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum and insulin (Sigma-Aldrich). Cell authentication was performed according to American Type Culture Collection (ATCC) guidelines by performing morphology analysis, growth curves and Mycoplasma testing (MycoAlert, Lonza) within 6 months of use.

Antibodies were used at the following concentrations: ADAM12 [GeneTex, GTX11536, 1:2000 in western blotting (WB), 1:500 in immunofluorescence (IF)], β-actin (Abcam, ab8226, 1:20,000 in WB), β1 integrin (P4C10, EMD Millipore, 1:500 in IF), cortactin (Upstate, 05-180, 1:1000 in IF), cortactin (Santa Cruz Biotechnology, sc-11408, 1:2000 in WB), E-cadherin (BD Biosciences, 610182, 1:1000 in WB, 1:200 in IF), MT1-MMP (GeneTex, EP1264Y, 1:500), N-cadherin (Santa Cruz Biotechnology, H-63, 1:1000 in WB), N-cadherin (BD Biosciences, 610921, 1:200 IF), SV40 large T [Santa Cruz Biotechnology, sc-147, 1:100 in immunohistochemistry (IHC)], Twist1 (gift from Dr Inna Gitelman, Ben-Gurion University of the Negev, Israel, 1:500 in WB), vinculin (GeneTex, SPM227, 1:500 in IF).

Stable shADAM12 cell lines were created via infection of target cells with either lentiviruses or Moloney viruses. 293T cells were seeded at 1×10⁶ cells per 6 cm dish. After 18 h, cells were transfected as follows: 6 μl TransIT-LT1 (Mirus Bio) was added to 150 μl DMEM and incubated for 20 min. Viral vector (1 μg) along with 0.9 μg of the appropriate gag-pol expression vector (pUMCV3 for pBabe or pWZL or pCMVΔ8.2R for lentiviral vectors) and 0.1 μg VSV-G expression vector were then added to the mixture. The mixture was incubated for 30 min and then added to 293T cells overnight. The subsequent day, fresh medium was added to the transfected 293T cells. Viral supernatant was harvested at 48 and 72 h post-transfection, passed through a 0.45 μm syringe filter, and added to the recipient cell lines with 6 μg/ml protamine sulfate for a 4 h infection. HMLE and Hs578T cells were then selected with 2 μg/ml puromycin (EMD Biosciences) or 10 μg/ml blasticidin (Invitrogen).

The three shRNA lentiviral constructs against Twist1 in the pSP108 vector have been described previously (Yang et al., 2004). ADAM12 was subcloned into a pWb expression construct from pcDNA3.1 by ligation into pWZL-Blast mammalian expression vector after cutting with XhoI/SalI restriction enzymes. The shRNA constructs against ADAM12 were created using the Invitrogen Blockit RNAi Designer. Candidate shRNAs were screened with NCBI BLAST to ensure specificity. Oligonucleotides were synthesized by Integrated DNA Technologies and were cloned into pSp81 vector with BstBI and BamHI restriction digestion followed by ligation. The shRNA constructs with associated U6 lentiviral promoter cassette and construct were cloned into pSp108 transfer vector with BamHI and SalI cuts followed by ligation.

Primers were generated for mutagenesis using PrimerX software (Bioinformatics.org) to target the described residues in ADAM12. Primers were synthesized by Integrated DNA Technologies. Mutagenesis was performed using the QuikChange PCR protocol with Phusion DNA Polymerase (New England Biolabs) using the ADAM12-Myc construct as the template for mutagenesis. Following 30 cycles of synthesis, the template plasmid was digested with DpnI. DH5α cells were directly transformed with the product of the reaction. Colonies were screened by ampicillin resistance. Mutagenesis of the final product was verified by sequencing. The cytoplasmic tail truncation mutant was generated by PCR cloning of the extracellular and transmembrane domains from the pWZL-Blast construct, followed by restriction digestion with XhoI/SalI, and relegation into pWZL-Blast vector.

Total RNAs were extracted from cells at 80–90% confluency using an RNeasy Mini Kit coupled with DNase treatment (QIAGEN) and reverse transcribed with a High-Capacity cDNA Reverse Transcription Kit (Life Technology). The resulting cDNAs were analyzed in triplicate using SYBR Green Master PCR Mix (Bio-Rad) and an Applied Biosystems 7500 Fast Real-Time PCR System with SDS software. Relative mRNA concentrations were determined by 2^−ΔΔCt, after normalizing to GAPDH values. Primer pairs used for real-time PCR are as follows: GAPDH, 5′-GAGAGACCCTCACTGCTG-3′ and 5′-GATGGTACATGACAAGGTGC-3′; and ADAM12, 5′-TCAAAAGCCCCTGTAGAGAA-3′ and 5′-TCTGTGTGCACGAGCAAAAG-3′.

This protocol is adapted from Artym et al. (2009). In brief, 12 mm coverslips were incubated in 20% nitric acid for 2 h and washed in H₂O for 4 h. Coverslips were incubated with 50 μg/ml poly-L-lysine diluted in PBS for 15 min, washed in PBS, and then 0.15% gluteraldehyde in PBS was added for 10 min, before washing in PBS. Coverslips were inverted onto 20 μl droplets of 1:9 0.1% FITC–gelatin (Invitrogen) with 0.2% porcine gelatin (Sigma-Aldrich) for 10 min. Coverslips were washed in PBS and then incubated for 15 min in 5 mg/ml NaBH₄. Coverslips were rinsed in PBS and incubated at 37° in 10% calf serum (HyClone) in DMEM for 2 h. A total of 20,000 cells were seeded on each coverslip, incubated for 8 h, and processed for IF. Each experiment was performed in triplicate. Images were collected at 10 fields per sample for a total of ∼150 cells per sample with an Olympus FV1000 confocal microscope. Gelatin degradation was quantified using ImageJ software. To measure the percentage degraded area in each field, identical signal thresholds for FITC–gelatin fluorescence were set for all images in an experiment, and the degraded area with an FITC signal below the set threshold was measured using ImageJ. The resulting percentage degradation area was further normalized to the total cell number (counted by DAPI staining nuclei) in each field. The final gel degradation index was the mean percentage degradation per cell obtained from all 10 fields. Each experiment was repeated at least three times.

Matrix substrates were prepared by coating glass coverslips with 0.1% rat tail collagen I in DMEM for 1 h. Cells (2×10⁴) were seeded onto coverslips in a 24-well plate and collected after 3 days of incubation. The cells were then fixed at 37°C in 4% PFA in PBS with 50 μM CaCl₂ for 15 min, permeabilized with 0.1% Triton X-100/PBS for 10 min, and blocked with 5% goat serum. Samples were incubated with primary antibodies overnight at 4°C and with secondary antibodies and/or phalloidin (Invitrogen) for 2 h. After washing, coverslips were mounted with VECTASHIELD HardSet Mounting Medium (Vector Laboratories). Images were collected with an Olympus FV1000 confocal microscope.

Images of cells on 0.1% collagen-coated 24-well plates or in suspension were collected and identical thresholds were applied to all images to isolate cell area. The total area of cells was calculated using ImageJ. Relative areas were calculated by normalization to cells expressing shCtrl constructs.

Plates (24 wells) were coated with 0.1% rat tail collagen I diluted in DMEM for 1 h. Cells were plated onto the 0.1% rat tail collagen I matrix in a plate and allowed to sit for 48 h. Then, 0.05% trypsin was added to each well for the indicated times. Trypsin was inactivated by adding an equal volume of 10% calf serum in DMEM, after which the wells were rinsed with PBS, fixed with 4% PFA, and stained with 0.1% Crystal Violet. Acetic acid was used to extract Crystal Violet from cells and the amount of Crystal Violet extracted was quantified by measuring absorbance at 590 nm. The relative cell number, normalized to untrypsinized samples, was quantified for each time point.

Cells cultured for 48 h on 0.1% rat tail collagen substrates were fixed and stained for F-actin and vinculin. Stress fibers were quantified by counting visually at 60× magnification on an Olympus FV1000 confocal microscope. Cortical actin was not counted; only stress fibers crossing the entire cell body were included in the count. Images of vinculin IF were processed by applying an identical threshold to all cells. Focal adhesions were quantified by setting minimum and maximum size limits for analysis in the Image J particle analysis tool. Settings were calibrated by quantification of positive and negative standards: Hs578T cells treated with MnCl₂ (increase in focal adhesions) or AIIB2 integrin-blocking antibody (Developmental Studies Hybridoma Bank, 1:1000 dilution) (decrease in focal adhesions) to ensure that changes in focal adhesion number were captured with the algorithm.

Cells cultured for 16 h on 0.1% rat tail collagen were transfected with paxillin–mCherry (gift from David Schlaepfer, University of California, San Diego, CA) and imaged 48 h after transfection by using a confocal microscope suited with an environmental control chamber.

Dynamic adhesion map images were prepared from image stacks as detailed previously (Santiago-Medina et al., 2012). Briefly, an image stabilization algorithm and an unsharp mask routine were applied, to improve edge detection, followed by thresholding to highlight paxillin adhesions. An 8-bit binary filter was applied to equalize adhesion intensities. Image stacks were then converted into 16-bit and user-defined subsets were summed so that intensity was representative of pixel lifetime. Final images were pseudocolorized using the lookup table in ImageJ.

Equal volumes and concentrations of NaOH-neutralized collagen I (Millipore) and Matrigel (BD Biosciences) were mixed on ice, and then 20 μl of the mixture was added to the bottom of each well of an eight-chamber coverglass slide. Cells of interest were mixed with the Matrigel–collagen mixture to a final concentration of 200,000 cells/ml, and 100 μl of the cell–matrix mixture was added to each well. The medium supplemented with 5% Matrigel was changed every other day throughout assays.

After 1 week in 3D culture (described above), images of spheroid structures were collected. Structures were counted as invasive if they had projections from the main spheroid body that extended further than half the diameter of the spheroid. Structures that did not have a well-defined spheroid structure were not quantified. Images were analyzed with ImageJ software.

All animal care and experiments were approved by the Institutional Animal Care and Use Committee (IACUC) of the University of California, San Diego, CA. Cells (1.5×10⁶) resuspended in 50% Matrigel were injected into the left and right flanks of 8- to 10-week-old female nude mice and allowed to grow to 1.5 cm in diameter before mice were killed. Primary tumor size was measured every 5 days. Lungs were harvested and imaged for GFP-positive tumor cells with a Leica MZ16F with ImagePro MC6.1 software. Tissues were embedded in paraffin, sectioned, stained with hematoxylin and eosin, and imaged to identify GFP-positive tumor cells.

Paraffin sections of mouse samples were rehydrated through xylene and graded alcohols. Antigen retrieval was accomplished in 10 mM sodium citrate with 0.05% Tween 20 in a pressure cooker. Samples were incubated with 3% H₂O₂ for 30 min, and then blocked in 20% goat serum in PBS for 5 h. Endogenous biotin and avidin were blocked using an Avidin/Biotin Blocking Kit (Vector Laboratories). Primary antibodies were incubated overnight at 4°C in 20% goat serum. Biotinylated secondary antibody and VECTASTAIN ABC Kit (Vector Laboratories) were used as indicated by the manufacturer. Samples were developed with diaminobenzidine (DAB) (Vector Laboratories) and samples counterstained with hematoxylin (Vector Laboratories), before mounting with Permount Mounting Medium (Fisher).

For invasion assays, 50 μg of Matrigel was overlaid on Transwell permeable supports, dried overnight and reconstituted with a mammary epithelial growth medium lacking recombinant EGF. Cells were plated onto wells (4×10⁴ cells/well) in triplicate and incubated for 72 h. Cells were fixed with 4% PFA in PBS, washed extensively with PBS, stained with 0.1% Crystal Violet, washed extensively with PBS, and dried. Crystal Violet was released with 50 μl 10% acetic acid and absorbency was measured at 590 nm. Identical protocols were used for migration assays using Transwells without Matrigel. All assays were performed in triplicate.

We thank members of the Yang laboratory for helpful discussions and suggestions on the manuscript.