cGMP-dependent protein kinase Iβ regulates breast cancer cell migration and invasion via interaction with the actin/myosin-associated protein caldesmon

The cGMP-dependent protein kinases (PKGs) have important physiological functions in the regulation of vascular tone, inhibition of platelet activation, modulation of neuronal functions, and maintenance of bone homeostasis (Hofmann et al., 2009; Rangaswami et al., 2010). Mammalian cells have the capacity to express three functional PKG proteins, transcribed from two separate genes. The type I gene produces two PKGI isoforms (PKGIα and PKGIβ) which are splice variants differing in their first ∼100 amino acids. The N-terminal regions give each isoform unique dimerization and autoinhibitory domains, with the dimerization domains forming leucine/isoleucine zippers that have been shown to mediate PKGI isoform-specific protein–protein interactions (Casteel et al., 2002; Schlossmann et al., 2000; Surks et al., 1999; Tang et al., 2003). Due to its dual role, we refer to this region as the dimerization/docking (D/D) domain. The type II gene produces one kinase, which has the same domain structure as type I PKGs, including an N-terminal leucine/isoleucine-rich dimerization region, but displays different tissue distribution and subcellular localization (Uhler, 1993).

Protein–protein interactions play an important role in mediating specificity in signal transduction, and identifying interaction partners results in a better understanding of a protein's function. A number of proteins have been found that interact exclusively with either PKGIα or PKGIβ. Specific interacting proteins for PKGIα include the myosin targeting subunit of myosin phosphatase (MYPT1), the regulator of G-protein signaling 2 (RGS2), formin homology domain protein 1 (FODH1), and PKG anchoring protein 42 kDa (GKAP42) (Surks et al., 1999; Tang et al., 2003; Wang et al., 2004; Yuasa et al., 2000). The PKGIα–MYPT1, PKGIα–RGS2 and PKGIα–FODH1 interactions are important for regulating vascular tone (Surks et al., 1999; Tang et al., 2003; Wang et al., 2004). GKAP42 is a substrate for PKGIα but its exact cellular function is unknown (Yuasa et al., 2000). Specific binding partners of PKGIβ are the transcription factor TFII-I and the inositol triphosphate (IP₃) receptor-associated PKG substrate (IRAG) (Casteel et al., 2002; Schlossmann et al., 2000). TFII-I is a multifunctional transcription factor involved in the regulation of a number of genes including c-fos (Casteel et al., 2002; Kim et al., 1998; Roy, 2012), and IRAG is essential for PKGIβ-mediated intracellular calcium regulation (Schlossmann et al., 2000). The interaction with IRAG can prevent PKGIβ nuclear localization and gene transactivation (Casteel et al., 2008).

We have previously identified residues that mediate interaction between PKGIβ and TFII-I or IRAG (Casteel et al., 2005). PKGIβ binds to both proteins through a common interaction motif consisting of acidic residues within the PKGIβ leucine/isoleucine zipper and basic residues within TFII-I and IRAG. PKGIβ containing D26K/E31R substitutions (i.e. changing the residues in the PKGIβ D/D domain to the corresponding residues in the PKGIα D/D domain) no longer interacts with either TFII-I or IRAG (Casteel et al., 2005). In this report we used affinity purification to find novel PKGIβ-interacting proteins. The screen was designed to identify proteins that differentially bound bacterially-produced affinity probes consisting of GST-tagged wild-type and D26K/E31R-mutant PKGIβ D/D domains. We found that PKGIβ specifically interacts with caldesmon (CaD), an actin-, myosin-, and calmodulin-binding protein that controls smooth muscle and non-muscle actin–myosin dynamics, and regulates cell migration and invasion (Mayanagi and Sobue, 2011).

To screen for the presence of novel PKGIβ-interacting proteins, we performed overlay assays using radioactively-labeled wild-type and D26K/E31R mutant PKGIβ D/D domains (amino acids 1–55) as probes. We detected a number of unique protein bands (numbered 2–8) with the wild-type probe but not with the mutant probe (Fig. 1A, compare lanes 1 and 2). Band 1 gave a strong signal with the wild-type probe and a much weaker signal with the mutant probe. The weaker band may be explained by dimer formation between the PKGIβ D/D domain probe and PKGIβ present in PAC1 cells, as the D26K/E31R mutation does not interfere with PKGIβ dimer formation (Casteel et al., 2005). The stronger signal seen with the wild-type probe suggests the presence of an additional protein interacting with the wild-type D/D domain. Pre-incubation of the membranes with 20-fold excess unlabeled wild-type probe disrupted binding of the radioactive wild-type probe to all but band 7 (Fig. 1A, compare lanes 1 and 3). These experiments provided evidence for the existence of proteins that directly interact with the PKGIβ D/D domain in a manner similar to TFII-I and IRAG.

We then used bacterially-produced GST-tagged PKGIβ D/D domains as bait to affinity-purify proteins from PAC1 cell lysates. Multiple proteins bound to the wild-type D/D domain of PKGIβ but not to the mutant D26K/E31R D/D domain (Fig. 1B, left panel); these bands were excised from the gel and subjected to tryptic-fragment mass-mapping. Some bands were seen to bind with higher affinity to the mutant probe. Since the mutations in the probe make the D/D domain more like PKGIα, it is possible that these bands represent PKGIα-interacting proteins. However, the D26K/E31R mutations do not completely change this part of the D/D domain to resemble the region in PKGIα, and it is likely that these interactions are non-physiological. Since we were looking for PKGIβ-specific interacting proteins, bands associated with both the wild-type and mutant probes or with higher affinity to the mutant probe were not analyzed. Table 1 shows the protein identifications and the number of peptides found for each protein. Several bands contained TFII-I, a known PKGIβ-interacting protein (Casteel et al., 2002), with bands B and C migrating at the expected molecular weight of TFII-I isoforms (∼135 kDa), while TFII-I in bands F and L may represent breakdown products. An overlay assay, done in parallel with the affinity purification (Fig. 1B, right panel), produced a similar banding pattern to that shown in Fig. 1A, with the addition of two high molecular weight bands (labeled 0 and 0′), which were only seen with the larger amount of cell lysate used in this experiment. Of note, the PKGIβ-interacting protein IRAG was not detected in this screen (Schlossmann et al., 2000); it is possible that IRAG protein abundance is very low in PAC1 cells and/or that other protein bands were masking IRAG.

Three bands seen on the overlay seem to match the TFII-I-containing bands on the Coomassie gel (0′, 0 and 6), consistent with direct PKGIβ binding. We also identified several cytoskeletal proteins. The prominent bands labeled A, G and J in Fig. 1B were found to contain myosin IIA, non-muscle CaD and β-actin, respectively. The overlay assay done in parallel suggests that CaD could be a direct interacting partner for PKGIβ (compare coomassie-stained band labeled G with overlay band labeled 1). Additionally, we identified gelsolin, coronin 1C and tropomyosin α4, which are known interaction partners of actin and/or CaD (Iizaka et al., 2000; Smith et al., 1987; Yin et al., 1980).

Since most of the identified proteins are known to bind together in large cytoskeletal complexes, we reasoned that PKGIβ binds directly to one protein common to these complexes, and that other proteins identified were co-purified with the direct PKGIβ interaction partner. In order to determine which protein interacted directly with PKGIβ, we transfected 293T cells with expression vectors encoding epitope-tagged versions of several candidate proteins, and performed small-scale affinity purification experiments using similar conditions as those in the initial screen. Since recombinant proteins are expressed at very high levels in 293T cells, a direct interacting protein would strongly associate with the wild-type probe, whereas proteins that are not direct binding partners for PKGIβ would not be detected, or would only be found in very small amounts, depending on the abundance of endogenous proteins mediating indirect interactions with PKGIβ. As can be seen in Fig. 2A, both CaD and TFII-I interacted with the PKGIβ wild-type probe, but not with the probe containing the D26K/E31R mutations. However, myosin IIA, tropomyosin α4, coronin 1C and β-actin were not detected in the pulldown of the wild-type probe under these conditions suggesting that these proteins are not direct binding partners of PKGIβ. Moreover, we found that FLAG-tagged CaD purified from 293T cells (supplementary material Fig. S1) bound directly to the recombinant wild-type GST-PKGIβ D/D domain in vitro (Fig. 2B).

To examine the association of PKGIβ with CaD in intact cells, we transfected 293T cells with expression vectors encoding full-length GST-PKGIα, GST-PKGIβ or GST alone, and analyzed CaD–PKG interaction using pulldown assays. (We used GST-tagged PKG constructs, because unlike untagged PKGIβ, they do not migrate with a similar apparent molecular weight as CaD and do not interfere with detection of CaD by western blotting.) We found that endogenous CaD associated with full-length PKGIβ but not with PKGIα (Fig. 2C). Next, we used immunofluorescence staining to examine the subcellular distribution of endogenous PKGIβ and caldesmon in human breast cancer MDA-MB-231 cells, which contain PKGIβ and isoform 4/5 caldesmon (described below). We found that PKGIβ and CaD colocalized in lamellipodial structures at the mobile edges of the cells (Fig. 3A, arrowheads). Activation of PKG by cGMP stimulation did not change this co-staining pattern (data not shown). To confirm isoform-specific colocalization, we performed immunofluorescence staining on MDA-MB-231 cells transfected with PKGIβ and FLAG-tagged isoform 5 CaD. A similar co-staining pattern of the two proteins in small membrane ruffles was detected at the leading edges of the cells (supplementary material Fig. S2). Endogenous PKGIβ and CaD colocalized with filamentous (F-)actin, mainly at the edges of lammellipodial structures (Fig. 3B,C). These results suggest that the kinase is tethered to the actin/myosin cytoskeleton through a direct interaction with CaD.

A number of previous studies have demonstrated that CaD is regulated by phosphorylation (Lin et al., 2009). Since PKGIβ binds to CaD, we asked if CaD was a direct substrate for PKGIβ in vitro. There are five CaD isoforms in mammalian cells (Fig. 4A,B), which are generated from a single gene by alternative transcriptional initiation and alternative splicing (Mayanagi and Sobue, 2011). We performed in vitro kinase assays with human isoform 2 and 5 CaD using purified PKGI catalytic domain (PKGI-CT), again to avoid co-migration of full-length PKGIβ and CaD. While isoform 2 CaD was minimally phosphorylated by PKGI-CT (Fig. 4C, lane 1), phosphorylation of isoform 5 CaD was much more robust (Fig. 4C, lane 2). Neither CaD isoform was phosphorylated in the absence of PKGI-CT (Fig. 4C, lanes 3 and 4).

Isoforms 2 and 5 CaD have unique N-termini (Fig. 4B); isoform 5 CaD contains serine 12 with basic residues at the −1 and −2 positions, indicating a potential PKGI recognition site. When serine 12 was changed to alanine (S12A), in vitro phosphorylation of isoform 5 CaD by PKGI-CT was almost abolished (Fig. 4D). To examine if CaD serine 12 is phosphorylated by PKGIβ in intact cells, we transfected MDA-MB-231 cells with PKGIβ and wild-type or S12A mutant CaD, and treated ³²PO₄-labeled cells with 8-pCPT-cGMP. cGMP-induced phosphorylation of isoform 5 CaD S12A was reduced by 51±3.6% compared to the wild-type protein (Fig. 4E,F). Most of the residual PKGIβ-mediated phosphorylation of the S12A mutant seen in intact cells is likely indirect, as the second direct in vitro phosphorylation site is very weak (seen in human isoform 2 CaD, Fig. 4C, lane 1).

Myosin has been shown to interact with the extreme N-terminus of CaD (Li et al., 2000). Since PKGIβ phosphorylates serine 12 of human isoform 5 CaD, we reasoned that phosphorylation of CaD might affect its interaction with myosin. To test this, we transfected 293T cells with expression vectors for GFP-tagged myosin IIA, PKGIβ and FLAG-tagged wild-type or S12A human isoform 5 CaD, and treated some cells with cGMP (Fig. 5A,B). Of note, exogenous PKGIβ was added to achieve robust phosphorylation of transfected CaD since 293T cells express relatively low levels of endogenous PKGI (data not shown). In the presence of PKGIβ, cGMP treatment significantly decreased the amount of myosin co-immunoprecipitated with CaD (Fig. 5A, compare lanes 3 and 4; the graph in Fig. 5B summarizes four independent experiments). However, in the absence of exogenous PKGIβ, cGMP did not affect the CaD–myosin interaction (Fig. 5A, compare lanes 1 and 2), indicating that the effect is mediated by PKG, and not via cross-activation of PKA or activation of another cGMP-regulated effector. Like wild-type CaD, mutant CaD S12A interacted with myosin, but its interaction was not altered by cGMP/PKGIβ (Fig. 5A, compare lanes 5 and 6). Next, we examined the effects of a phospho-mimetic (S12E) mutation on the CaD–myosin interaction. When compared to wild-type and S12A CaD, significantly less myosin co-immunoprecipitated with CaD containing a phospho-mimetic S12E mutation (Fig. 5C, compare lane 4 to lanes 2 and 3; the graph in Fig. 5D summarizes three independent experiments). Thus, adding a negative charge to CaD serine 12, either through PKGIβ-mediated phosphorylation or mutation to glutamate, inhibits the CaD–myosin interaction. Since CaD regulates myosin-dependent actin filament movement, we reasoned that PKGIβ phosphorylation of CaD may affect cell motility (Mayanagi and Sobue, 2011).

Using RT-PCR, we found that MDA-MB-231 human breast cancer cells express PKGIβ, and isoforms 2/3 and 4/5 CaD (supplementary material Fig. S3). To examine the effect of PKG signaling on cell migration, we performed scratch wounding assays on confluent MDA-MB-231 cells; some cells were pre-treated with L-Name (to inhibit nitric oxide synthase), ODQ (to inhibit soluble guanylyl cyclase), or with Rp-PET-CPT-cGMPS (Rp) or DT3 (two structurally-unrelated PKG inhibitors). Pharmacological inhibition of the NO/cGMP/PKG pathway significantly decreased the number of cells that migrated into the wound, indicating that basal NO/cGMP/PKG activity is necessary for migration of these cells, while activation of PKG with 8-pCPT-cGMP significantly increased migration (Fig. 6A). In addition, using time-lapse cell monitoring we found that the cGMP-stimulated increase in migration was due to an increase in cell velocity (supplementary material Fig. S4). Thus, the NO/cGMP/PKG pathway has a pro-migratory function in MDA-MB-231 breast cancer cells.

CaD is an inhibitor of cell motility, and manipulation of CaD expression affects migration of several cell types (Mirzapoiazova et al., 2005; Yokouchi et al., 2006; Zheng et al., 2007). For example, increased CaD expression in human lung carcinoma A549 cells enhances stress fiber formation and reduces cell motility, whereas CaD depletion increases motility (Mayanagi et al., 2008). When we transfected MDA-MB-231 with a CaD-specific siRNA to deplete all CaD isoforms, we observed an increase in migration, consistent with the anti-migratory function of CaD seen in other cell types (Fig. 6B,C; >95% CaD depletion by the siRNA is shown in the lower panel of C). Adenovirus-mediated reconstitution with an siRNA-resistant wild-type human isoform 5 CaD reversed the phenotype and cell migration returned to control levels; similar results were observed with reconstitution using phospho-deficient (S12A) mutant CaD. In contrast, reintroduction of isoform 5 CaD containing the phospho-mimetic S12E mutation was unable to reduce the rate of migration. Since CaD S12E mimics PKGIβ phosphorylation, and PKG activation increases MDA-MB-231 migration, we infer that the cGMP-induced increase in migration was due, at least in part, to CaD phosphorylation by PKG.

Next, we tested whether NO/cGMP/PKG signaling affects the invasive properties of human breast cancer MDA-MB-231 cells. Using Matrigel invasion chambers, we found that inhibition of nitric oxide synthase or PKG lead to decreased invasion, whereas cGMP-mediated stimulation of PKG increased the invasive capacity of MDA-MB-231 cells (Fig. 7A; supplementary material Fig. S5A). siRNA-mediated knockdown of CaD caused a pro-invasive effect in human breast cancer cells, which was reversed by reintroduction of wild-type CaD but not by CaD containing the phosphomimetic S12E mutation (Fig. 7B; supplementary material Fig. S5B). Similar results were seen when we examined migration/invasion in normal human mammary epithelial (HMLE) cells (supplementary material Fig. S6), indicating that this mechanism appears to be more generally important in normal and transformed cells of the human mammary gland. Taken together, these results reveal a novel signaling pathway through which PKGIβ regulates the migratory and invasive capacities of human breast cancer cells (Fig. 7C).

We used a combination of overlay assays and affinity purification/mass-spectrometric analysis to identify novel PKGIβ-interacting proteins, and found that the kinase is tethered to the actin/myosin cytoskeleton through a direct interaction with CaD. In a previous study, both actin and myosin IIA were found to co-purify with endogenous PKGI isolated from bovine trachea (Koller et al., 2003). CaD was not identified in these studies, but its presence may have been masked by the large amount of isolated PKGI, since PKGI and CaD migrate with the same apparent molecular weight.

The cytoskeletal functions of CaD are still not completely understood. It binds both F-actin and myosin, and acts as an inhibitor of myosin ATPase activity, thus blocking cellular functions that require dynamic myosin–actin interactions, such as cell contraction and migration (Chalovich et al., 1987; Mayanagi and Sobue, 2011). Presently, the physiological roles of the various CaD isoforms are not known. While expression of heavy h-CaD (isoform 1) is restricted to smooth muscle cells, expression of the four light l-CaD isoforms (non-muscle) is ubiquitous. The tissue distribution of l-CaD isoforms has not been studied in detail, but based upon the cells they were isolated from, isoforms 2/3 have been referred to as WI-38-type whereas isoforms 4/5 have been referred to as HeLa-type (Hayashi et al., 1992). Interestingly, CaD isoforms 4/5 are aberrantly expressed in endothelial cells during the early stages of tumor neovascularization (Zheng et al., 2005). Our findings that i) human isoform 5 CaD is a PKG substrate, ii) PKGIβ-mediated phosphorylation of serine 12 disrupts CaD–myosin interactions, and iii) modification of this site prevents CaD's ability to inhibit breast cancer cell migration and invasion, may represent identification of the first isoform-specific l-CaD function.

Although PKGIβ phosphorylation of CaD serine 12 disrupted the CaD–myosin interaction, it is unknown how this phosphorylation affects myosin–actin dynamics. The CaD C-terminus binds F-actin, and is sufficient to inhibit actin-activated myosin ATPase activity, most likely by masking myosin-binding sites on F-actin (Chalovich et al., 1987). However, both CaD N- and C-terminal domains are necessary to inhibit actin filament velocity in motility assays performed in vitro (Wang et al., 1997); the underlying mechanism is thought to involve ‘tethering’, in which CaD's N- and C-termini prevent movement by stably binding myosin and F-actin, respectively. In contrast, a recent report demonstrates that the interaction between an N-terminal CaD fragment (amino acids 1–263) and myosin is necessary and sufficient for proper axon extension in hippocampal neurons (Morita et al., 2012). Since axon extension does not require the CaD C-terminus, the underlying mechanism is most likely not due to myosin–actin tethering and remains unknown.

A number of protein kinases phosphorylate and regulate CaD, including: Ca²⁺-calmodulin kinase II; cdc2 kinase; cyclin dependent protein kinase 2 (Cdk2); casein kinase II; p21-activated kinase (PAK); and extracellular-signal-regulated kinase (Erk) (Childs et al., 1992; Ikebe and Reardon, 1990; Van Eyk et al., 1998; Yamashiro et al., 1991). In most cases, phosphorylation of CaD occurs in the C-terminus and disrupts its interaction with actin (reviewed by Lin et al., 2009). For example, cdc2 kinase phosphorylation triggers dissociation of CaD from actin filaments, and is thought to be responsible for regulating dynamic changes in actomyosin during mitosis (Yamashiro et al., 1991), and PAK phosphorylation of CaD is necessary for migration of CHO-K1 cells (Eppinga et al., 2006). Our finding that PKGIβ phosphorylates human isotype 4/5 CaD on serine 12 and disrupts CaD–myosin interactions is consistent with the binding of the N-terminus of CaD to myosin. Like PKG, casein kinase II phosphorylates CaD's N-terminus (on serines 26 and 73) leading to a decrease in the interaction between CaD and myosin (Wang and Yang, 2000). It is important to note that CaD phosphorylation at serine 12 has been identified by a number of phospho-proteome screens in HeLa cells, indicating that phosphorylation at this site is relatively common under basal growth conditions (Dephoure et al., 2008; Olsen et al., 2010; Perkins et al., 1999). The interaction site for myosin in CaD has been mapped to amino acids 24–53 (Wang et al., 1997). Since serine 12 is outside of the known myosin interaction interface, we hypothesize that CaD phosphorylation on serine 12 disrupts CaD–myosin interactions through a conformational change in CaD's N-terminus that alters the myosin binding site. Alternatively, the phosphorylation-induced negative charge at serine 12 may alter post-translational modification at another site in CaD, which leads to disruption of CaD–myosin binding; however, we found that the overall residual phosphorylation levels of CaD S12A and S12E are equal (unpublished data), suggesting that a negative charge at serine 12 does not affect phosphorylation at another location. Further work will be required to determine in detail how PKG phosphorylation of CaD S12 disrupts the interaction between CaD and myosin.

Serine 12 in isoform 4/5 is not evolutionarily conserved. While it is found in several species, including humans, chimpanzees, orangutans, horses and pigs, it is absent in others, like mice and rats (supplementary material Fig. S7). Uncovering the differences in kinomes and phospho-proteomes between various organisms is important for understanding species-specific differences in protein functions and limitations of animal models for the study of human diseases (Caenepeel et al., 2004; Jalal et al., 2009). For example, the myosin phosphatase-regulating zipper interacting protein kinase (ZIPK) contains a relatively conserved phosphorylation site (T299) that is found in human ZIPK, but not in the murine or rat orthologs (Weitzel et al., 2011). Lack of this phosphorylation site leads to nuclear localization of ZIPK in mice and rats, whereas ZIPK is predominately cytoplasmic in human cells (Haystead, 2005; Weitzel et al., 2011). While the nuclear functions of ZIPK are unknown, cytoplasmic ZIPK increases Ca²⁺ sensitivity of the contractile apparatus in smooth muscle cells by directly phosphorylating myosin light chain and myosin phosphatase (Haystead, 2005).

Previous studies have examined the role of NO/cGMP/PKG signaling and CaD in cell migration/invasion. PKG is required for plasmin- and α-elastin-stimulated migration of monocytes and macrophages, respectively (Kamisato et al., 1997; Uemura and Okamoto, 1997), and PKG mediates neutrophil motility induced by carbon monoxide and endothelin-1/endothelin-3 (Elferink and De Koster, 1998; VanUffelen et al., 1996). Yet, the mechanisms underlying PKG-dependent increase in migration of immune cells are unknown. In smooth muscle and endothelial cells, PKG signaling has been reported to both stimulate and inhibit motility (Brown et al., 1999; Dubey et al., 1995; Kawasaki et al., 2003; Rolli-Derkinderen et al., 2010; Smolenski et al., 2000; Zhang et al., 2003); these conflicting results may be explained by differences in primary cell cultures and/or variations in cell migration assays. While the inhibitory effects are mediated, at least partly, through PKG phosphorylation of VASP or induction of MAP kinase phosphatase-1 (Jacob et al., 2002; Smolenski et al., 2000), pro-migratory effects of PKG depend on activation of phosphatidylinositol 3-kinase and RhoA phosphorylation (Kawasaki et al., 2003; Rolli-Derkinderen et al., 2010).

In agreement with our findings, a number of reports have demonstrated that decreased CaD expression promotes, and enhanced CaD expression inhibits migration and invasion of tumor cells [reviewed in (Mayanagi and Sobue, 2011)]. CaD is thought to inhibit cell motility and invasion through reorganization the cytoskeleton and/or by inhibiting myosin/actin force generation (Mayanagi et al., 2008; Mirzapoiazova et al., 2005).

We found that depleting CaD isoforms lead to enhanced motility and invasion in human breast cancer MDA-MB-231 cells; reintroduction of wild-type isoform 5 CaD reversed these effects, but the phospho-mimetic S12E CaD mutant failed to recover CaD's anti-migratory and anti-invasive activity, suggesting that NO/cGMP/PKGI signaling increases migration and invasion of human breast cancer cells, at least in part, by PKGIβ phosphorylation of serine 12 in CaD. Little is known about a direct role of PKG in cancer cell motility and invasion. Deguchi et. al. reported that PKG inhibits migration of SW480 human colon cancer cells and suggested that the effect was mediated in part by VASP phosphorylation, but the detailed mechanism was not studied (Deguchi et al., 2004). In contrast, NO/cGMP signaling has been implicated in stimulating migration of mouse mammary carcinoma cells (Jadeski et al., 2003; Jadeski et al., 2000; Punathil et al., 2008). However, since these studies were performed in murine breast cancer cells, the increased motility cannot be through CaD serine 12 phosphorylation, and the underlying mechanism(s) remains to be determined. NO/cGMP/PKG-mediated effects on tumors are widespread, and affect not only the migratory and invasive capacities of cancer cells. It was shown that PKG has both pro- and anti-proliferative effects on various tumor cell types. In some colon and breast cancer cells, PKG activation inhibits proliferation and increases apoptosis (Deguchi et al., 2004; Tinsley et al., 2011; Whitt et al., 2012), and these effects may be mediated by decreasing β-catenin levels (Tinsley et al., 2011; Whitt et al., 2012). In stark contrast, PKG facilitates pro-proliferative/anti-apoptotic signaling in neuronal, ovarian and lung tumor cells lines through inhibition of caspase activity and increasing expression of anti-apoptotic genes (Kim et al., 1999; Leung et al., 2010; Wong et al., 2012). An exact understanding of how PKG's effects on cell proliferation, survival, migration and invasion promote or inhibit tumor growth in vivo will require further studies.

In conclusion, we have presented a novel mechanism by which cGMP/PKGIβ regulates human breast cancer cell migration and invasion through a critical phosphorylation of CaD. cGMP-elevating agents are in widespread clinical use for the treatment of angina pectoris, pulmonary hypertension and erectile dysfunction (Francis et al., 2010). The results presented in this report suggest that PKG-activating drugs might increase the malignant potential of some tumors, and their use could be contra-indicated for some cancer patients. Our studies also suggest that pharmacological inhibition of the NO/cGMP/PKGIβ signaling pathway may be beneficial in breast cancer treatment.

Anti-GST, anti-HA, anti-Myc, anti-GFP and pan-anti-CaD antibodies were obtained from Santa Cruz Biotechnology (Santa Cruz, CA). Anti-FLAG antibodies, M2 affinity resin, 3×FLAG peptide and PKA C-subunit were from Sigma–Aldrich (St. Louis, MO). Glutathione Sepharose and thrombin were from GE Healthcare (Pittsburg, PA). siRNAs targeting human CaD (5′-CAGATAGGTATCAATATGTTT-3′) (Yoshio et al., 2007), green fluorescent protein (GFP) (5′-AAGCTGACCCTGAAGTTCATC-3′), and PKGIα (5′-AAGAGGAAACUCCACAAAUGC-3′) (Zhang et al., 2007) were obtained from Qiagen (Hilden, Germany). 8-pCPT-cGMP (referred to as cGMP) and the PKG inhibitors Rp-8-CPT-PET-cGMPS (referred to as Rp) and DT3 were purchased from Biolog (Bremen, Germany). The nitric oxide synthase inhibitor N-nitro-L-arginine methyl ester (L-Name) and the soluble guanylyl cyclase inhibitor 1H-[1,2,4]oxadiazolo[4,3-a]quinoxalin-1-one (ODQ) were from Cayman (Ann Arbor, MI).

EGFP-tagged myosin IIA (Wei and Adelstein, 2000) was obtained from Addgene [(Plasmid 11347), Cambridge, MA]. PKGIβ D/D wild-type (amino acids 1–55) and PKGIβ D/D D26K/E31R mutant were constructed by polymerase chain reaction (PCR) using previously constructed vectors as templates. PCR products were digested with BamHI/EcoRI and ligated into pGEX-2T and pGEX-2TK (GE Healthcare, Pittsburgh, PA). FLAG-tagged PKGI catalytic domain was generated by PCR (PKGIβ amino acids 347–686) and ligated into pFLAG-D. GST-tagged PKGIα, GST-tagged PKGIβ and Myc-tagged TFII-I have been described (Casteel et al., 2005). Coronin 1C and β-actin were cloned from total rat pulmonary artery smooth muscle cells (PAC1) RNA by reverse transcription PCR. cDNA for tropomyosin α4 and human isoform 2 CaD was obtained from the American Type Cell Culture (ATCC, Manassas, VA). cDNA for the N-termini of human isoform 4/5 was obtained from Open Biosystems (Lafayette, CO). Wild-type isoform 5 CaD and the S12A and S12E CaD mutants were made using overlapping PCR and ligated into pFLAG-D. Adenoviral vectors expressing wild-type isoform 5 CaD, and S12A and S12E isoform 5 CaD mutants, were produced using the ViraPower Adenoviral Expression System (Life Technologies, Carlsbad, CA), as per manufacturer's instructions. All constructs and mutants involving PCR were sequenced to ensure the absence of PCR induced mutations.

PAC1 (Rothman et al., 1992) and MDA-MB-231 cells were maintained in DMEM with 10% FBS. 293T cells were maintained in Iscove's MEM with 10% FBS. HMLE cells were grown as described (Eckert et al., 2011). Cells were grown at 37°C in a 5% CO₂ atmosphere.

GST-PKGIβ D/D wild-type and GST-PKGIβ D/D D26K/E31R mutants were produced in DH5α E.coli as follows: 50 ml overnight cultures were diluted 1∶10, allowed to grow for 1 hour and then induced with 1 mM IPTG. The cultures were grown for an additional 4 hours and the bacteria were harvested by centrifugation. The bacterial pellets were resuspended in 5 ml bacterial lysis buffer [phosphate buffered saline (PBS), 1% Triton X-100, 1 mM DTT, 0.4 µM PMSF, 10 µg/ml leupeptin and 10 µg/ml aprotinin]. Cleared cell lysates were incubated with 1 ml glutathione Sepharose for 1 hour at 4°C with agitation. Beads were washed 3× with bacterial lysis buffer, 2× with PBS containing 500 mM NaCl and 2× with PBS with 0.1% NP-40. Probes for overlay assays were produced and purified using the same conditions.

PAC1 cells were lysed in buffer A (PBS, 0.1% NP-40, 1 mM DTT, 0.4 mM PMSF, 10 µg/ml leupeptin and 10 µg/ml aprotinin) and lysates were separated by SDS-PAGE and transferred to Immobilon membranes. Membranes were blocked in PBS with 5% BSA and then cut into strips containing one lane for each probe condition. Glutathione Sepharose beads with 3.2 µg of GST-tagged PKGIβ D/D wild-type or D26K/E31R, containing a synthetic cAMP-dependent protein kinase (PKA) site, were phosphorylated in vitro using PKA C-subunit and [³²P]-γ-ATP. Excess radioactivity was removed by washing and radiolabeled probes were cleaved from the GST-tag with thrombin. The specific activity of the cleaved probes was determined by scintillation counting, and each probe was added to 20 ml PBS with 2% BSA and incubated with the membranes at room temperature for 1 hour. For some experiments, membranes were pre-incubated with 20× excess unlabeled wild-type or mutant probe for 30 minutes. After addition of radiolabeled probes, membranes were washed with PBS and analyzed by autoradiography.

10⁹ PAC1 cells were resuspended in 20 ml buffer A. The resuspended cells were lysed by dounce homogenization and the lysates were cleared by centrifugation: 2×20 minutes at 12,000 g at 4°C. After the second centrifugation, the lysate was filtered through a 0.45 µM syringe filter and equal amounts were added to the beads containing PKGIβ D/D domain wild-type or mutant affinity probes. The lysates were incubated with the beads for 1 hour at 4°C with agitation. The beads were washed 4× with 5 ml buffer A, and bound proteins were eluted with PBS containing 500 mM NaCl (4× 0.5 ml). The eluted proteins were concentrated using a YM-10 centricon and then boiled in 3× SDS sample buffer. The bound proteins were analyzed by SDS-PAGE and Coomassie staining, and stained bands were excised and identified by tryptic mass-mapping at the Stanford Protein and Nucleic Acid (PAN) Facility (http://pan.stanford.edu/index.html) using the Mascot search engine (Perkins et al., 1999).

293T cells were transfected with DNA constructs expressing epitope-tagged versions of the putative interacting proteins. Cells were lysed in buffer A and cleared lysates were incubated with 50 µg immobilized GST-PKGIβ wild-type or GST-PKGIβ D26K/E31R. Beads were washed and bound proteins were analyzed by SDS-PAGE and immunoblotting. In other experiments, 293T cells were transfected with expression vectors encoding GST, GST-PKGIα, or GST-PKGIβ. 24 hours later, cells were lysed in Buffer A supplemented with 1 mM EGTA, 0.1 mM EDTA, 0.1 mg/ml aprotinin, 0.1 mM leupeptin and 0.4 mM PMSF, and clarified lysates were incubated with glutathione Sepharose beads for 1 hour. The beads were washed and bound proteins were analyzed by SDS-PAGE and immunoblotting.

MDA-MB-231 cells were transfected with an siRNA targeting human PKGIα using Lipofectamine™ 2000 (Life Technologies) according to manufacturer's instructions. Cells were fixed in 3.8% para-formaldehyde and permeabilized in 0.5% Triton X-100. After blocking, immunofluorescence staining of endogenous proteins was carried out using anti-CaD (Santa Cruz Biotechnology, Santa Cruz, CA) and anti-PKGI (Calbiochem/EMD, Billerica, MA) antibodies, and fluorescent dye-coupled secondary antibodies [goat anti-mouse IgG (H⁺L), conjugated to FITC or goat anti-rabbit IgG (H⁺L), conjugated to TRITC (Jackson ImmunoResearch, West Grove, PA)]. For labeling F-actin, Rhodamine-phalloidin (Life Technologies) was used. To examine isoform-specific colocalization, MDA-MB-231 were transfected with FLAG-tagged isoform 5 CaD and PKGIβ using Lipofectamine™ 2000 and overexpressed proteins were stained with anti-FLAG (Sigma–Aldrich) and anti-PKGI (Cell Signaling Technology, Danvers, MA) antibodies, and the above mentioned fluorescent dye-coupled secondary antibodies. Nuclei were stained using Hoechst 33342. Cells were viewed with a confocal microscope (Olympus FV1000), a 40/1.3 oil-immersion objective, and 4× to 8× digital zoom. Images were analyzed with Fluoview (Olympus, Shinjuku, Japan) and Photoshop (Adobe, San Jose, CA).

For in vitro phosphorylation studies, 293T cells were transfected with FLAG-tagged CaD constructs using Lipofectamine™ 2000, and 24 hours later cells were lysed in Buffer A and CaD was isolated by incubation with anti-FLAG M2 agarose. The beads were washed and incubated in a kinase reaction performed as described in (Casteel et al., 2002), using 50 ng purified PKGI catalytic domain in the absence of 8-Br-cGMP. For studies examining phosphorylation in intact cells, 293T cells were transfected with PKGIβ and FLAG-tagged CaD expression constructs as described above. 20 hours later, cells were transferred to phosphate-free DMEM with 100 µCi/ml ³²PO₄ for 3 hours. Some cells were treated with 250 µM 8-pCPT-cGMP for 1 hour. Cells were lysed in Buffer A, and CaD was isolated using anti-FLAG M2 agarose and immunoprecipitates analyzed by SDS-PAGE and autoradiography. Blots were subsequently probed with anti-CaD antibodies to show equal expression of wild-type and mutant CaD constructs.

To examine the interaction between CaD and myosin, 293T cells were transfected with expression constructs for GFP-tagged myosin IIA, FLAG-tagged CaD variants (wild-type, S12A and S12E) and PKGIβ, as indicated. Some wells were treated with 250 µM 8-pCPT-cGMP for 1 hour and cells were lysed in Triton X-100-buffer [30 mM HEPES (pH 7.4), 0.05% Triton X-100, 50 mM NaCl, 5 mM MgCl₂, 5 mM EGTA, 1 mM ATP, 0.1 mg/ml aprotinin, 0.1 mM leupeptin and 0.4 mM PMSF]. Lysates were incubated with anti-FLAG M2 agarose and bound proteins were analyzed by SDS-PAGE and immunoblotting.

MDA-MB-231 cells were seeded into a 12-well plate and serum-starved (0.1% FBS) for 24 hours. Confluent cells were scratch-wounded with a 200 µl pipette tip and treated with 4 mM L-Name, 10 µM ODQ, 100 µM Rp-8-CPT-PET-cGMPS or 3 µM DT3 (in 2% FBS), or 100 µM 8-pCPT-cGMP (in 0.1% FBS). Cell migration was monitored for 24–48 hours after scratching and migrating cells were counted. For siRNA-mediated knockdown of CaD, MDA-MB-231 cells were transfected with 100 pmol of siRNA targeting the 3′ untranslated region of human CaD or control (GFP) siRNA using Lipofectamine™ 2000 (we consistently see >95% knockdown under these conditions). 24 hours after transfection, cells were reconstituted with siRNA-resistant human CaD isoform 5 or treated with control (LacZ) adenovirus (MOI of ∼10) for 18 hours (infection efficiency is >80%, as determined by LacZ staining). About 72 hours after transfection, confluent cells were scratch-wounded and cell migration was monitored for 24–36 hours. Pictures were taken at the time of wound closure of the CaD-depleted and control virus-infected cells. Numbers of migrated cells were determined by comparing the photographs to their corresponding pictures taken directly after scratch wounding; markings reflecting the edges of the fresh scratch (as shown in Fig. 6B) were used to count cells that have migrated into the gap (using brightfield microscopy).

To measure cell velocity, MDA-MB-231 cells were seeded into a 35mm fluorodish cell culture dish (World Precision Instruments, Sarasota, FL) and cells were treated as described above. Migration was monitored for 20 hours after scratching using time-lapse microscopy (IX8I; Olympus) and an interval of 10 minutes image acquisition was used. The microscope was equipped with a heated incubation chamber that maintains a temperature of 37°C and a 5% CO₂ atmosphere. The centroid of individual cells was tracked in acquired images using Fiji software and a manual cell tracking plug-in. The instantaneous velocity (µm/minute) was calculated between each point of centroid movement as the distance divided by time. These instantaneous velocities were then averaged to determine an overall average velocity for each track. Each track was then combined for both experimental conditions giving the total average velocity.

To assess migration of HMLE cells, cells were starved and pre-treated with 4 mM L-Name, 100 µM Rp-8-CPT-PET-cGMPS or 100 µM 8-pCPT-cGMP for 1 hour. Cells were trypsinized and 5×10⁴ cells in starvation media containing the appropriate drug were added to the upper chamber of 8 µm transwell inserts (BD Biosciences, Bedford, MA). The lower chamber contained growth media with 10% FBS as a chemo-attractant. After 24 hours, non-migrating cells on the upper side of the filter were removed using a cotton swab. The filters were fixed, stained with haematoxylin, and cells on the underside of the filter were counted under high power magnification, and photomicrographs were taken using an inverted microscope.

Cell invasion was assayed using Matrigel-coated invasion chambers (BD Biosciences). MDA-MB-231 or HMLE cells were starved and pre-treated with 4 mM L-Name, 100 µM Rp-8-CPT-PET-cGMPS or 100 µM 8-pCPT-cGMP for 1 hour. To study the effect of CaD mutants on cell invasion, MDA-MB-231 cells were transfected with 100 pmol siRNA targeting CaD or GFP using Lipofectamine™ 2000. At 24 hours post transfection, cells were starved and reconstituted with adenovirus expressing siRNA-resistant CaD wild-type, S12A or S12E human isoform 5 CaD for 18 hours. Control cells were treated with adenovirus expressing (LacZ). Trypsinized cells (1.5×10⁵ cells/well) were resuspended in DMEM containing 0.1% FBS and added to the upper compartment of the pre-hydrated Matrigel-coated chamber. The lower chamber was filled with DMEM containing 10% FBS, which acted as a chemo-attractant. Cells were incubated at 37°C for 24 hours and non-invading cells on the upper side of the filter were removed using a cotton swab. The filters were fixed, stained and cells were counted as described for transwell migration assays.

Western blots and autoradiographs are from representative experiments that were performed at least three times with similar results. Cell migration assays and experiments examining phosphorylation in 293T cells were performed at least three times; statistical analysis was done by analysis of variance (ANOVA) with Bonferroni's post-test.

We thank Dr Jing Yang for providing the HMLE cell line, Dr Michael W. Berns for help with cell velocity measurements, and the members of the laboratories of Drs Renate Pilz and Gerry Boss for helpful discussions about this project.