Integrins synergise to induce expression of the MRTF-A–SRF target gene ISG15 for promoting cancer cell invasion

Integrin-mediated cell adhesion and signalling control numerous cellular processes, which are crucial for development and postnatal homeostasis, and influence diverse pathological processes such as cancer development and metastasis (Hynes, 2002; Winograd-Katz et al., 2014). Integrin signalling is a multistep process that is initiated with integrin activation and ligand binding, followed by integrin clustering and the progressive assembly of signalling platforms consisting of adaptor, signalling and/or catalytic and cytoskeletal proteins (Legate et al., 2009). The first integrin-based adhesion sites are small and short-lived and termed nascent adhesions. They eventually mature into large and stable focal adhesions that are connected to filamentous (F-) actin (Vicente-Manzanares et al., 2009).

Integrin signalling induces both short- and long-term effects in cells (Legate et al., 2009). The short-term effects mainly affect cytoskeletal rearrangements that allow cells to adopt their characteristic shape and initiate migration, and are to a large extent triggered by the activation of Rho family GTPases and actin-binding proteins (Danen et al., 2002). Long-term effects of integrin signalling result from changes in gene expression, which regulate numerous cellular processes including proliferation and differentiation (Legate et al., 2009). Integrin-dependent regulation of gene expression is primarily thought to arise from cross talk with growth factor receptors that increase the activity of mitogen-activated protein (MAP) kinase pathways. A recent study, however, reported that integrins can also control gene transcription by releasing the association of the transcriptional co-activator myocardin-related transcription factor-A (MRTF-A, also known as MKL1 or MAL) from monomeric or globular (G-) actin (Miralles et al., 2003; Plessner et al., 2015). MRTF-A belongs to the myocardin-related family of transcriptional co-activators, which consists of myocardin, MRTF-A and MRTF-B (also known as MKL2). MRTF-A and MRTF-B can bind monomeric G-actin, which retains them in the cytoplasm and prevents binding to SRF. Once G-actin is polymerised into different actin networks, MRTF-A/B are released from cytoplasmic and nuclear G-actin, associate with the transcription factor serum response factor (SRF) and regulate the expression of cytoskeletal proteins including actin and numerous focal adhesion proteins (Esnault et al., 2014; Posern and Treisman, 2006).

In the present paper we investigated whether fibronectin-binding integrins of the αV and β1 classes can regulate MRTF–SRF activity. To this end we used integrin-deficient fibroblasts (pan-knockouts, pKO cells) that were reconstituted with αV and/or β1 integrin cDNAs, giving rise to cells that express αVβ3 and αVβ5 (pKO-αV cells), cells that express α5β1 (pKO-β1), and cells that express both integrin classes (pKO-αV,β1 cells). We have shown in a previous paper that reconstituting cells with different fibronectin-binding integrins results in the formation of different types of adhesion sites and the induction of different actin structures (Schiller et al., 2013). Whereas the pKO-αV cells form large focal adhesions and primarily activate Rho–mDia-mediated actin polymerisation leading to thick, transversal stress fibres, pKO-β1 cells assemble small nascent adhesions along the cell edge, activate Rac, induce dense cortical actin networks and lamellipodia and link Rho primarily to myosin II activation. pKO-αV,β1 cells develop nascent adhesions, a cortical actin network at the cell edge and lamellipodia, and focal adhesions associated with stress fibres and high myosin II activity (Schiller et al., 2013). Using these cell lines we demonstrate that αV- and β1-class integrins synergise to induce high activation of MRTF–SRF and expression of target genes. We also identify the ubiquitin-like modifier interferon-stimulated gene 15 (ISG15) as a newly discovered integrin-induced MRTF–SRF target gene, and find that high levels of fibronectin-binding integrins and ISG15 promote the invasion of the malignant breast cancer line MDA-MB-231 and correlate with poor survival rates in a large cohort of breast cancer patients.

To assess whether fibronectin-binding integrins differentially regulate MRTF-A–SRF activity, we seeded pKO-αV, pKO-β1 and pKO-αV,β1 cells (Schiller et al., 2013) on fibronectin and analysed the levels and distribution of MRTF-A, G-actin and F-actin. Western blots of whole cell lysates revealed significantly higher levels of β-actin in pKO-αV cells compared with pKO-β1 and pKO-αV,β1 cells (Fig. 1A,B), and staining with fluorescently labelled phalloidin showed that pKO-αV and pKO-αV,β1 cells developed thick stress fibres, whereas pKO-β1 contained thin stress fibres on fibronectin-coated culture dishes and micropatterns (Fig. 1C; Fig. S1A). Furthermore, pKO-β1 and pKO-αV,β1 cells formed multiple lamellipodia with subcortical actin networks, which were rarely seen in pKO-αV cells (Fig. 1C). Visualisation of cytoplasmic G-actin with fluorescently labelled DNaseI revealed high cytoplasmic signals in pKO-αV and lower signals in pKO-β1 and pKO-αV,β1 cells (Fig. 1D). To quantify the G- and F-actin levels in our cell lines, we performed cell fractionation assays followed by western blotting and found that the soluble fractions (G, containing G-actin and positive for GAPDH) contained higher β-actin levels in pKO-αV cells compared with pKO-β1 and pKO-αV,β1 cells (Fig. 1E), whereas the cytoskeletal fractions (F, containing F-actin and positive for Vimentin) contained highest β-actin levels in pKO-αV,β1, intermediate in pKO-β1 and lowest in pKO-αV cells (Fig. 1E). The calculated G-/F-actin ratio from nine different western blot analyses was significantly higher in pKO-αV cells compared with pKO-β1 and pKO-αV,β1 cells (Fig. 1F). To determine monomeric G-actin levels in the cytoplasm and nucleus, we prepared soluble and nuclear cell fractions, respectively, and performed sepharose-coupled DNaseI pull-downs followed by western blotting with anti-β-actin antibodies. The experiments revealed that in both compartments the G-actin levels were highest in pKO-αV, intermediate in pKO-β1 and lowest in pKO-αV,β1 cells (Fig. 1G–I). Importantly, BSA-coupled control beads did not pull down β-actin from either subcellular fraction (Fig. 1G), and FACS analysis excluded nuclear size differences between our cell lines as a possible cause for the different nuclear G-actin contents (Fig. S1B).

Re-analysis of our previously published whole-cell- and phospho-proteome of pKO-αV, pKO-β1 and pKO-αV,β1 cells (Schiller et al., 2013) revealed significantly elevated levels of thymosin β4 (Tβ4) and phospho-serine-3 (pSer3)-cofilin in β1-expressing cells (Fig. S1C–E), indicating that G-actin is sequestered significantly more by the high Tβ4 levels and F-actin severed less by the pronounced inactivation of cofilin in pKO-β1 and pKO-αV,β1 cells compared with pKO-αV cells, which further decreases the free G-actin contents in pKO-β1 and pKO-αV,β1 cells. Taken together, these results show that fibronectin-binding integrin classes differentially regulate cellular G- and F-actin contents.

MRTF-A is a transcriptional co-activator that is sequestered in the cytoplasm through binding G-actin and released upon actin polymerisation (Posern and Treisman, 2006). Therefore, the different levels of monomeric G-actin in pKO-αV, pKO-β1 and pKO-αV,β1 cells might correlate with different MRTF-A distributions. To test this hypothesis, we stained the cells with specific rabbit polyclonal anti-MRTF-A antibodies (Descot et al., 2009). Indeed, nuclear MRTF-A levels were lowest in pKO-αV, intermediate in pKO-β1 and highest in pKO-αV,β1 cells (Fig. 2A). Treatment with Jasplakinolide, which polymerises free G-actin and stabilises actin filaments (Fig. 2B; Fig. S1F), increased nuclear MRTF-A localisation, whereas treatment with Latrunculin A, which increases monomeric G-actin by depolymerising F-actin, decreased nuclear MRTF-A levels in all three cell lines (Fig. 2C; Fig. S1F). To quantify MRTF-A levels in the cytoplasm and nucleus, the lysates of the respective cell fractions were analysed by immunoblotting using anti-MRTF-A antibodies. Western blot analysis revealed weak nuclear MRTF-A signals in pKO-αV cells, intermediate signals in pKO-β1 cells and highest signals in pKO-αV,β1 cells (Fig. 2D). Similarly, the nuclear/cytoplasmic ratios of MRTF-A were significantly higher in pKO-αV,β1 cells compared with pKO-β1 and especially pKO-αV cells (Fig. 2E). These results show that fibronectin-binding integrin classes differentially regulate nuclear MRTF-A levels.

To test the functional consequences of the different nuclear MRTF-A contents in our three cell lines, we first measured SRF transcriptional activity using luciferase reporter assays (Busche et al., 2008; Sotiropoulos et al., 1999). The assays revealed low SRF reporter activity in pKO-αV cells, 5-fold higher activity in pKO-β1 cells and ∼10-fold higher activities in pKO-αV,β1 cells (Fig. 3A). Serum treatment, which is known to induce nuclear translocation of MRTF-A, increased SRF reporter activities in pKO-β1 and pKO-αV,β1 cells to a much higher extent compared with pKO-αV cells (Fig. S2A), suggesting that β1-integrin-mediated adhesion in cooperation with serum or growth factor-signalling facilitates maximal SRF activity. Overexpression of a dominant-negative version of MRTF-A reduced SRF reporter activity to basal levels in the three cell lines (Fig. S2B), whereas overexpression of a G-actin-binding-deficient MRTF-A (ΔN-MRTF-A) or a constitutive active mDia, which catalyses actin polymerisation, increased MRTF- and/or SRF-driven luciferase activities to similar extents in all three cell lines (Fig. 3A).

Cell spreading defects can also lead to reduced nuclear MRTF-A localisation and SRF reporter activities (Connelly et al., 2010). To exclude that spreading defects of pKO-αV cells (Schiller et al., 2013) were responsible for reduced MRTF-A activities we seeded the three cell lines on fibronectin-coated circular micropatterns with 28 μm or 40 μm diameters (that have twice the surface area of 28 μm sized patterns), and performed SRF luciferase reporter assays and immunostainings. The experiments revealed that pKO-αV cells displayed similarly low SRF reporter activities and nuclear MRTF-A levels on small as well as large fibronectin-coated circular micropatterns, whereas pKO-β1 and pKO-αV,β1 cells exhibited increased SRF activities and nuclear MRTF-A levels on 40-µm-sized micropatterns compared with 28-µm-sized micropatterns (Fig. 3B,C). Taken together these results indicate that reduced spreading of pKO-αV cells is not responsible for their low SRF activity.

Next we investigated whether changes in the activities of fibronectin-binding integrin classes affect SRF reporter activity. First, we activated integrins by treating the cells with Mn²⁺, which significantly elevated SRF reporter activity in all three cell lines, most prominently in pKO-β1 and pKO-αV,β1 cells (Fig. 3D). Second, we inhibited individual integrin classes in pKO-αV,β1 cells, either with antibodies or peptidomimetics that block specific classes of fibronectin-binding integrins. After seeding pKO-αV,β1 cells on fibronectin (bound by αV- and β1-class integrins), vitronectin (VTN) or gelatin (both ligands bound by αV-class integrins only), they were treated with either anti-α5β1-integrin-blocking antibodies or the αV-class-specific small molecule inhibitor cilengitide and immunostained for MRTF-A. The experiments showed that inhibition of either α5β1 or αVβ3 integrins on fibronectin-adherent pKO-αV,β1 cells significantly reduced nuclear MRTF-A levels and thus phenocopied the situation in pKO-αV or pKO-β1 cells (Fig. 3E; Fig. S2C), and confirmed the synergistic role of both fibronectin-binding integrin classes for nuclear MRTF-A accumulation. Similarly, when pKO-αV,β1 cells were seeded on VTN and gelatin, they showed weak nuclear MRTF-A signals, which were not further decreased by blocking the function of α5β1 integrin (Fig. S2C). Treatment of VTN- or gelatin-seeded pKO-αV,β1 cells with cilengitide blocked cell adhesion (Fig. S2C). Taken together these data indicate that αV- and β1-class integrins cooperate to maximise nuclear accumulation of MRTF-A and SRF activity.

The high nuclear MRTF-A content and SRF activity in pKO-αV,β1 cells suggest that MRTF-A–SRF-induced gene transcription is controlled, at least in part, by a cooperation of αV- and β1-class integrins. To test this possibility, we compared published MRTF and SRF transcriptomes (Balza and Misra, 2006; Cooper et al., 2007; Descot et al., 2009; Philippar et al., 2004; Selvaraj and Prywes, 2004; Sun et al., 2006; Zhang et al., 2005) with the whole cell proteome data of pKO-αV, pKO-β1 and pKO-αV,β1 cells (Schiller et al., 2013) and found a large number of MRTF–SRF targets, including SRF itself, and focal adhesion proteins such as vinculin and talin-1, enriched in pKO-β1 and pKO-αV,β1 cells (Table 1). The increased levels of SRF, talin-1 and vinculin mRNA and protein in pKO-β1 and pKO-αV,β1 cells were confirmed by qRT-PCR and western blotting (Fig. 4A,B).

We also noted that the ISG15 mRNA and protein was significantly higher in pKO-αV,β1 cells, compared with pKO-αV cells (Fig. 4A, Table 1) and that its mRNA was previously shown to be downregulated upon SRF deletion in mouse embryonic stem cells (ESCs) (Philippar et al., 2004). Western blotting confirmed high ISG15 levels in pKO-αV,β1, intermediate levels in pKO-β1 and low levels in pKO-αV cells (Fig. 4B,C). Although it has been reported that ISG15 can be secreted we did not detect ISG15 in the supernatant of either cell line (Fig. S3A). Immunostainings also detected high ISG15 levels in pKO-αV,β1 and low levels in pKO-αV cells. The immunostaining also demonstrated that ISG15 colocalised with F-actin fibres, with the actin cortex in membrane protrusions (Fig. 4D) and with paxillin in focal adhesions of unroofed cells (Fig. 4E). ISG15 is an ubiquitin-like modifier, whose expression is induced by type I (α and β) interferons (Farrell et al., 1979; Haas et al., 1987). Whereas ELISA and qRT-PCR excluded expression of interferon α and β by pKO-αV,β1 cells as cause for the high ISG15 levels (Fig. S3B–E), Poly(I:C) treatment induced interferon α and β expression (Fig. S3B–E), which in turn triggered a strong Isg15 expression (Fig. S3F).

To demonstrate that Isg15 is an MRTF–SRF target gene, we searched for SRF consensus binding sites (CArG box) in the mouse Isg15 locus and performed chromatin immunoprecipitation (ChIP) assays. The mouse Isg15 gene contains a CArG box in the first intron (Fig. 4F). The MRTF–SRF targets vinculin (Vcl) and talin-1 (Tln1) were used as controls and gapdh as a negative control (Vartiainen et al., 2007). ChIP samples were obtained using either anti-MRTF-A, anti-SRF or control IgG antibodies and the resulting chromatin DNA samples were PCR amplified with primers flanking the indicated CArG boxes of Isg15, Tln1 and Vcl (Fig. 4F). As expected, the gapdh gene was not amplified and control rabbit IgG precipitates did not produce PCR products (Fig. 4G,H).

MRTF-A–SRF and ISG15 are upregulated in cancer and promote tumour cell invasion (Desai et al., 2012; Kressner et al., 2013). To test whether the invasive properties of MRTF-A and ISG15 are associated with and triggered by β1-integrin-mediated adhesion we compared the integrin expression profile, nuclear MRTF-A and ISG15 levels between the non-invasive MCF-7 (luminal type) and the low invasive MDA-MB-468 (basal-A type) and highly invasive MDA-MB-231 (basal-B type) breast carcinoma cell lines (Neve et al., 2006). FACS analysis revealed significantly higher α5 and β1 integrin levels and higher β1 integrin activity (shown with the activation-epitope binding 9EG7 antibody) on MDA-MB-231 cells compared with MCF-7 cells (Fig. 5A). Interestingly, the low invasive MDA-MB-468 cells demonstrated no detectable α5 integrin and low total and active β1 integrin surface levels (Fig. 5A). The levels of αV integrin were significantly elevated in MDA-MB-468 and MCF-7 cells compared with MDA-MB-231 cells (Fig. 5A). To test whether the high β1 integrin levels and activity in MDA-MB-231 cells are associated with high nuclear MRTF-A activity we stained for F-actin and MRTF-A, and investigated SRF activity with luciferase reporter assays. We found that MDA-MB-468 cells spread poorly, contained cortical F-actin rings and showed high cytoplasmic and low nuclear MRTF-A signals, whereas MCF7 cells were well spread, formed tight, F-actin-rich cell–cell contacts and contained MRTF-A in the cytoplasm as well as in the nucleus (Fig. 5B). By contrast, MDA-MB-231 cells were not adhering to each other, contained numerous F-actin stress fibres throughout the cytoplasm and almost the entire pool of MRTF-A protein was detected in the nucleus (Fig. 5B,C). In line with high nuclear MRTF-A levels, MDA-MB-231 cells also displayed significantly higher SRF reporter activity (Fig. 5D) and expressed significantly higher levels of MRTF–SRF target gene transcripts (Fig. 5E). The free levels of ISG15 protein were highest in MDA-MB-231, lowest in MDA-MB-468 and intermediate in MCF7 cells (Fig. 5F). Interestingly, the ISGylation of target proteins was highest in MDA-MB-231, intermediate in MDA-MB-468 and lowest in MCF-7 cells (Fig. S4A) indicating that MDA-MB-468 cells use their entire low ISG15 pool to ISGylate proteins.

The invasive properties of MDA-MB-231 cells have been shown to depend on the expression of MRTF-A (Hu et al., 2011; Medjkane et al., 2009). To test whether ISG15 is required for SRF–MRTF-A-induced cell invasion, we depleted ISG15 or overexpressed the de-ISGylase UBP43 in MCF-7, MDA-MB-468 and MDA-MB-231 cells and performed trans-well migration assays through fibronectin-coated membranes and 3D-Matrigel invasion assays. MCF-7 cells were unable to invade through fibronectin-coated membranes and invade 3D-Matrigels (not shown). MDA-MB-231 and -468 cells, however, efficiently migrated through fibronectin-coated membranes, which significantly diminished upon shRNA-mediated depletion of ISG15 or overexpression of the de-ISGylase UBP43 (Fig. 5G; Fig. S4B). Furthermore, MDA-MB-231 cells, and to a lesser extent MDA-MB-468 cells, invaded 3D-Matrigels, which was further enhanced upon overexpression of ISG15 (Fig. 5H; Fig. S4C). Interestingly, overexpression of ISG15 significantly increased basal SRF activities in MCF-7, MDA-MB-468 and MDA-MB-231 cells (Fig. 5I), pointing to a positive feed-forward loop between integrin-mediated adhesion signalling, MRTF−SRF and ISG15. In line with a potential feed-forward loop, the decreased migration of MRTF-A-depleted MDA-MB-231 cells through fibronectin-coated membranes could not be improved by simultaneously overexpressing ISG15 (Fig. S4D).

Our results suggest that high levels of β1 integrin and ISG15 represent prognostic markers for breast cancer patients. To evaluate this possibility, we consulted microarray databases of breast cancer samples and correlated αV integrin, β1 integrin and ISG15 transcript levels with patient survival. The analyses clearly revealed that individuals with high expression of β1 integrin or ISG15 displayed a significant increase in the hazardous ratio (HR) compared with individuals with low levels of either transcript (Fig. 6A,B). Importantly, however, individuals with high levels of αVβ1 integrin and ISG15 transcripts in their tumour samples showed a pronounced reduction in disease-free survival rate compared with individuals with low ISG15 levels (Fig. 6C). These findings demonstrate that integrin-mediated MRTF-A–SRF–ISG15 signalling is elevated in breast cancer and associated with a poor prognosis.

Metastasis is initiated with the detachment of individual cancer cells from the tumour cell aggregate followed by invasion into the surrounding tissue, entry into the circulation and finally settlement in a distant organ. The process of tumour cell invasion is crucially dependent on the selection of tumour cells that are able to survive without tumour stroma, on the release of proteases that degrade and remodel the tumour- and tissue-derived extracellular matrix (ECM), and on cell adhesion molecules such as integrins (Hood and Cheresh, 2002). Integrins constitute the core components of the invasive machinery of cancer cells. They regulate the activity of small GTPases, actin binding, bundling and modifying proteins to enable cytoskeletal dynamics, membrane protrusions and invadopodia formation (Bishop and Hall, 2000; Hall, 2012; Hoshino et al., 2013; Murphy and Courtneidge, 2011). Interestingly, a recent paper showed that integrins also regulate the nuclear translocation of MRTF-A and induce MRTF-A–SRF target gene expression (Plessner et al., 2015). Because tumour cells express different types of integrins, it is unclear whether and how they co-operate to achieve maximal efficiency in tissue invasion and migration. The aim of the present study was to define integrin subfamilies and the signalling pathways involved in tumour cell invasion.

We have recently engineered a fibroblast-like cell line that allows expression of the fibronectin-binding integrins αVβ3 and/or α5β1. These cell lines were used to demonstrate that αVβ3 and α5β1 integrins control actomyosin-based cell contractility in a cooperative manner (Schiller et al., 2013). In the present study we also observed that the cells expressing αVβ3 and α5β1 integrins (pKO-αV,β1 cells) most markedly decrease their G-actin pool, which results in the nuclear translocation of the G-actin-binding transcriptional transactivator MRTF-A. The nuclear accumulation of MRTF-A allows binding to the transcription factor SRF and transcriptional activation of MRTF-A–SRF target genes, which comprise known target genes such as actin, vinculin and filamins (Esnault et al., 2014), as well as newly discovered targets such as talin-1 and ISG15, which contain functional CArG boxes known to bind the MRTF-A–SRF complex (Olson and Nordheim, 2010).

Our findings show that ISG15 is particularly highly elevated when αVβ3 and α5β1 integrins are co-expressed and bind fibronectin. ISG15 is a ubiquitin like modifier protein that is conjugated to specific lysine residues of target proteins (Kerscher et al., 2006). ISG15 expression, which is elevated in almost all tumours investigated so far (www.oncomine.org), is thought to be induced by type I interferons (Farrell et al., 1979; Haas et al., 1987). Because the tumour stroma is infiltrated by immune cells, it is believed that they are the source of interferon α and β production and hence the trigger for the high ISG15 expression in cancer cells (van der Veen and Ploegh, 2012). Although immune-cell-derived interferon α and/or β strongly promotes expression of ISG15 in tumours, it remained unknown why high ISG15 levels persisted in tumour cells isolated and cultured ex vivo in the absence of immune cells and interferons (Han et al., 2002; Hermeking et al., 1997; Lock et al., 2002). Based on our findings we propose that synergistic αVβ3- and α5β1-integrin-induced nuclear translocation of MRTF-A and activation of SRF are responsible for the high ISG15 transcription ex vivo.

ISG15 is conjugated to proteins (ISGylation) in a multistep process that requires E1, E2 and E3 enzymes and is comparable to the protein-ubiquitylation pathway (Haas et al., 1987; Lenschow et al., 2005; Loeb and Haas, 1992; Morales and Lenschow, 2013; Okumura et al., 2008; Yuan and Krug, 2001). Protein modifications with ISG15 are reversible and can be removed (or de-ISGylated) by the ISG15 de-conjugating peptidase Ubp43 (Malakhov et al., 2002). Protein ISGylation can alter protein function; for example, ISGylation of filamin-B can release filamin-B-bound Rac1, MKK4 and MEKK1, leading to the termination of JNK signalling and inhibition of apoptosis (Jeon et al., 2009). By contrast, it has also been shown that ISG15 modification can stabilise proteins by inhibiting ubiquitin-specific E2 enzymes (Desai et al., 2006; Malakhova and Zhang, 2008; Okumura et al., 2008) or by competing with ubiquitin for lysine residues on common target proteins (Liu et al., 2003). Finally, ISGylation has been shown to inhibit protein translation by either increasing the cap-structure-binding activity of the ISGylated translational suppressor 4EHP (Okumura et al., 2007) or by downregulating elF2α through ISGylation of dsRNA-dependent protein kinase (PKR) (Okumura et al., 2013).

The high expression of ISG15 in tumour cells in vivo and ex vivo indicates that ISGylation-dependent inhibition of apoptosis or stabilisation of certain oncogenic proteins represents a tumour-promoting function in primary tumours where interferon α and β levels are high, as well as during metastasis when interferon levels decrease. Hence, it is conceivable that during invasion and metastasis for example, cooperative signalling via αVβ3 and α5β1 integrins compensates for low interferon levels to sustain high ISG15 expression. Previously published mass spectrometry analyses of ISG15-modified proteins (ISGylome) in tumour cells revealed that actin, actin regulatory and focal adhesion proteins (filamin-B, vinculin, talin-1) can be ISGylated (Durfee et al., 2010). It is therefore possible that αVβ3- and α5β1-integrin-mediated signalling has two major consequences for tumour cell invasion; αVβ3 and α5β1 integrins activate GTPases leading to polymerisation of F-actin networks and stress fibres, which are essential for membrane protrusions, cell contractility and adhesion reinforcement. In addition, the consumption of G-actin for the formation of different F-actin networks results in liberation and nuclear translocation of MRTF-A, binding to SRF and transcription of cytoskeletal, focal adhesion proteins and ISG15 itself. ISG15 modifies various proteins, including focal adhesion proteins and actin-binding or -modifying proteins to increase their stability and/or improve their function (Fig. 7). Such a feed-forward circuit operates in mouse fibroblasts as well as in invading MDA-MB-231 breast cancer cells. The metastatic breast cancer cell line MDA-MB-231 expressed significantly more α5β1 integrin, contained higher levels of nuclear MRTF-A, showed an elevation of ISG15-conjugated proteins in whole-cell lysates and performed better in fibronectin-based trans-well migration assays and in 3D-Matrigel invasion assays than the less invasive MDA-MB-468 or non-metastatic MCF7 cells. We could also demonstrate that the ability of the invasive MDA-MB-231 and MDA-MB-468 cells to migrate relies on ISG15 expression and ISGylation and that ISG15 overexpression enhances 3D-Matrigel invasion of both, MDA-MB-231 and MDA-MB-468 cancer cells. Most importantly, transcriptome analyses of breast cancer samples from large cohorts of patients revealed a statistically highly significant association between disease free patient survival and high β1 integrin, αV-integrin and ISG15 transcript levels. These findings suggest that breast cancer cells with high integrin levels are selected in a primary tumour mass, which renders them independent of decreasing interferon and growth factor signals for sustaining nuclear MRTF-A activity, ISG15 expression and metastatic potential.

The generation of pKO-αV, pKO-β1 and pKO-αV,β1 cell lines was reported previously (Schiller et al., 2013). The breast cancer cell lines MDA-MB-231 and MCF-7 were purchased from ATCC (http://www.lgcstandards-atcc.org/Products/Cells_and_Microorganisms/Cell_Lines.aspx), MDA-MB-468 cells were obtained from Dr Axel Ullrich (MPI Biochemistry, Martinsried).

For immunofluorescence microscopy, cells were seeded on ECM-coated micropatterns or glass surfaces (coating: 5 µg/ml fibronectin (Calbiochem) or 1% gelatin (Sigma) or 10 µg/ml vitronectin (STEMCELL Technologies) in PBS) in DMEM (GIBCO by Life Technologies) containing 10% FCS at 37°C, 5% CO₂. To culture cells on micropatterns the culture medium contained 0.5% FCS. After indicated time points the medium was soaked off, and cells were fixed with 3% PFA in PBS for 10 min at room temperature, washed with PBS, blocked with 1% BSA in PBS for 1 h at room temperature and then incubated with antibodies in a solution of 0.1% Triton X-100, 3% BSA in PBS. The fluorescent images were collected with a laser scanning confocal microscope (Leica SP5). For visualization of G- and F-actin structures, we followed a previously published protocol (Small et al., 1999).

All antibodies are listed in Table S2.

Enrichment for focal-adhesion-associated proteins was achieved by briefly fixing the ventral cell cortex using DSP crosslinker [DTSP; dithiobis(succinimidyl propionate)], followed by removal of non-crosslinked proteins and big organelles by stringent cell lysis and hydrodynamic sheer flow washing.

For cell fractionation the ProteoExtract^® Cytoskeleton Enrichment and Isolation Kit (Millipore), and for G-actin and F-actin analyses the In Vivo Assay Biochem Kit (Cytoskeleton, Inc.) were used. For the centrifugation-based method to isolate the nuclei, cells were washed with PBS and harvested in a buffer containing 250 mM sucrose, 10 mM HEPES and 1.5 mM EDTA. With the help of a syringe and a 26 Ga needle (Terumo) cells were disrupted further. After centrifugation the pellet (nuclei fraction) was washed five times with a buffer containing 20 mM Tris, 0.1 mM EDTA and 2 mM MgCl₂. The pellet was resuspended in FACS buffer or staining solution for immunostaining.

DNaseI (Sigma-Aldrich) was covalently linked to CNBr-activated Sepharose beads (Sigma-Aldrich) at a concentration of 1 mg/ml according to the manufacturer's protocol. BSA–sepharose beads, which served as control, were prepared in the same way. For G-actin pulldown, nuclear and cytosolic fractions were incubated with 35 µl of DNaseI- or BSA-coupled Sepharose beads in a volume of 200 µl overnight at 4°C in an end-over-end mixer. Next day, the beads were washed five times with cold wash buffer (1% NP-40, 0.1% SDS, 1 mM DTT, 1 mM PMSF in PBS). Beads were dried with a syringe and needle and SDS sample buffer was added for subsequent western blot analysis.

Constitutively active MRTF-A (ΔNMRTF-A), dominant negative MRTF-A (DN MRTF-A: ΔN1B1 and ΔNΔC) and MRTF-A–SRF reporter constructs have been previously characterised (Leitner et al., 2011). Constitutively active myc–mDia1 expression construct (myc-mDia1 FH1FH2) was amplified from an existing plasmid with forward primer 5′-gccaagaatgaaatggcttc-3′ and reverse 5′-tgcagagcttctagaagact, then the PCR product cloned into the pCRII-TOPO vector and sequenced. The integrin αV–mCherry and integrin β1–mCherry were generated in our laboratory (MPI Biochemistry, Martinsried). The knockdown constructs for stable knockdown of human ISG15 were purchased from Sigma and the UBP43 overexpression construct was obtained from Addgene. All transfections were carried out with Lipofectamine 2000 (Invitrogen through Life Technologies) according to the manufacturer's instructions.

Cells were incubated for 20 min with 50 µM of N-ethylmaleimide (NEM) and lysed on ice in RIPA buffer supplemented with 25 mM NEM and protease and phosphatase inhibitors. Protein concentration was measured by Bradford assay, samples were boiled for 5 min in Laemmli buffer containing 2% 2-mercaptoethanol.

Cells were plated on fibronectin coated 12-well plates (6.0×10⁵ cells per well) before transient transfection with 0.5 µg of MRTF-A−SRF reporter (p3DA.luc) (Sotiropoulos et al., 1999), indicated expression plasmid and 25 ng thymidine-kinase-driven Renilla luciferase (Promega) for normalization of transfection efficiency. The total amount of transfected plasmid DNA was kept constant at 1.5 µg per well by using pEGFP-C1 expression vector (Clontech). After 24 h luciferase activity was analysed with a Dual Luciferase reporter assay kit (Promega) using a GloMax^® Microplate Luminometer (Promega). Jasplakinolide (100 nM, #420107-50UG from Merck Millipore) and Latrunculin A (500 nM, L5163 from Sigma) treatment was initiated 3 h prior to cell lysis.

Micropatterns were generated on PEG-coated glass coverslips with deep-ultraviolet lithography (Azioune et al., 2010). Glass coverslips were incubated in a 1 mM solution of a linear PEG, CH₃–(O–CH₂–CH₂)₄₃–NH–CO–NH–CH₂–CH₂–CH₂–Si(OEt)₃ in dry toluene for 20 h at 80°C under a nitrogen atmosphere. The substrates were removed, rinsed intensively with ethyl acetate, methanol and water, and dried with nitrogen. A PEGylated glass coverslip and a chromium-coated quartz photomask (ML&C, Jena) were immobilised with vacuum onto a mask holder, which was immediately exposed to deep ultraviolet light using a low pressure mercury lamp (NIQ 60/35 XL longlife lamp, quartz tube, 60 W from Heraeus Noblelight) at 5 cm distance for 7 min. The patterned substrates were subsequently incubated overnight with 100 μl of fibronectin (20 μg ml⁻¹ in PBS) at 4°C and washed once with PBS.

For FACS analysis suspension of fibroblasts was incubated for 1 h with primary antibodies on ice and then washed twice with FCS-PBS (3 mM EDTA, 2% FCS in PBS). Cell viability was assessed by propidium iodide staining. FACS analysis was carried out using a FACSCalibur Cytometer (BD Biosciences) and cell sorting with an AriaFACSII high-speed sorter (BD Biosciences), both equipped with FACS DiVa software (BD Biosciences). Purity of sorted cells was determined by post-sort FACS analysis and typically exceeded 95%. Data analysis was conducted using the FlowJo program (Version 9.4.10).

Total RNA from cells was extracted with RNeasy Mini extraction kit (Qiagen) following manufacturer's instructions. cDNA was prepared with an iScript cDNA Synthesis Kit (Bio-Rad). Real-time PCR was performed with an iCycler (Bio-Rad). Each sample was measured in triplicate and the values were normalised to gapdh levels. PCR primers are listed in Table S1.

Cells were plated on fibronectin-coated tissue culture dishes for three days and the cell supernatant was analysed. To induce interferon production, cells were transfected with 100 µg/ml Poly(I:C). ELISA for interferon α and interferon β secretion was performed with the VeriKine™ Mouse IFN-α and Mouse IFN-β ELISA kits (PBL Interferon Source, product #42120 and #42400) according to manufacturer's instructions.

Invasion assay (Merck Millipore QCM™ Boyden chamber, 8 µm pore size) was performed according to manufacturer's instructions. MCF-7, MDA-MB-468 and MDA-MB-231 cells were seeded into polycarbonate trans-wells at a density of 3×10⁴ cells and in a volume of 300 µl medium. Migration occurred for 12 h from serum-free medium towards medium with 10% FCS. After removing the non-migrating cells and staining the transmigrated cells with Phalloidin-TRITC and DAPI, micrographs were taken at 200× magnification. For each condition, cell migration was deduced by counting DAPI+ cell nuclei within 10 random regions on the trans-well membrane.

For all ChIP experiments, pKO-αV,β1 cells were used. Cross-linking, nuclei preparation and nuclease digestion of chromatin was performed according to manufacturer's instructions [SimpleCHIP Enzymatic Chromatin IP Kit (Magnetic Beads), #9003, Cell Signaling Technology]. Then, 500 µl of chromatin was incubated overnight at 4°C with 10 µl anti-SRF antibodies (#5147; Cell Signaling Technology) or 30 µl anti-MRTF-A rabbit serum (#79, our laboratory). After washing the immunoprecipitated chromatin, the DNA–protein complexes were eluted with the ChIP elution buffer. Crosslinks were reversed overnight at 65°C, and DNA was purified using columns provided in the kit. Quantification was done by quantitative real-time PCR and is shown as the percentage of input chromatin. Gene-specific primers for amplification of immunoprecipitated DNA are listed in Table S1. Primers for gapdh and srf loci were published previously (Vartiainen et al., 2007).

Cells were infected with retroviral 29mer shISG15 expression constructs purchased from Sigma (#TG502956). Puromycin selection was used to select for cells carrying integrations.

To analyse the prognostic value of integrin αV, integrin β1 and ISG15 mRNA expression levels the Kaplan–Meier plotter was used (http://kmplot.com/analysis/; Gyorffy et al., 2010). The patient samples were split into two groups according to various quantile expressions of the proposed biomarker (low and high expression). The two patient cohorts were compared by a Kaplan–Meier survival plot, and the hazard ratio with 95% confidence intervals and logrank P-value were calculated.

Data are represented as the mean±s.e.m. of at least three independent experiments. Unless otherwise indicated, statistical significance was determined using unpaired two-tailed Student's t-test.

We thank Dr Julien Polleux for help with micropattern generation and Dr Assa Yeroslaviz for supporting the analyses of survival rates of breast cancer patients.