Cyclic AMP Signaling as a Mediator of Vasculogenic Mimicry in Aggressive Human Melanoma Cells In vitro

Melanoma originates from the malignant transformation of melanocytes that reside in the basal layer of the epidermis and in other anatomic regions such as the uvea of the eye. The current approach to treating melanoma relies on prevention, early diagnosis, and local management of the primary tumor (1, 2). However, for later stages of the disease, therapeutic strategies are quite inefficient at targeting the elusive metastatic phenotype. Genomic analyses of highly aggressive and less aggressive human cutaneous and uveal melanoma cell lines have revealed the complexity of this phenotype, with a plasticity close to that of embryonic cells (3, 4). In particular, aggressive melanoma cells, but not their poorly aggressive counterparts, can engage in a process termed “vasculogenic mimicry” (VM). VM describes the ability of these cells to express endothelium- and epithelium-associated genes and form extracellular matrix–rich tubular networks that mimic the pattern of embryonic vasculogenic networks (5, 6). Also occurring in nonmelanoma tumors, VM coincides with poor clinical outcome. Conveniently for biological investigation, VM can be observed in vitro in three-dimensional cultures on Matrigel or collagen matrices (7). This has allowed the identification of some mediators involved in VM promotion (8), including ephrin-A2, vascular endothelial cadherin (VE-cadherin), focal adhesion kinase, extracellular signal-regulated kinase 1 and 2 (ERK1/2), phosphatidyl inositol-3-kinase (PI3K), and Nodal (4, 8, 9). In this study, we wondered whether cyclic AMP, a second messenger controlling many cellular processes with idiosyncratic responses depending on cell type, could be a mediator of VM. The binding of many hormones to cells, such as those involving G protein–coupled receptors (GPCR), induces the activation of adenylyl cyclases that produce cyclic AMP from ATP (10). Cyclic AMP binds to PKA, which then phosphorylates several substrates involved in signal transduction pathways. Among these, the mitogen-activated protein kinase (MAPK) cascade consisting of small GTP-binding protein Ras/B- and/or C-Raf kinases/MAPK kinase 1/2 (MEK1/2) and ERK1/2 (11). More recently, additional cyclic AMP targets, the (guanine nucleotide) Exchange proteins directly activated by cyclic AMP (Epac) Epac1 and Epac2, which mediate PKA-independent cell responses, have been discovered (12, 13). Epac1 and Epac2 are unique exchange factors of the small GTPases Ras–associated proteins (Rap) 1 and 2, which become activated upon releasing GDP and binding GTP. Initially Rap1 was considered an antagonist of Ras signaling, but follow-up studies have shown that it induces ERK1/2 stimulation or inhibition depending on cell type, mode of activation, and cellular compartmentalization (11, 14). Epac responses can, however, cooperate with PKA responses (12, 15), The complex relationships between PKA and Epac effectors explain why the cyclic AMP signal may lead to apparent contradictory responses in different cell types. Recent lines of evidence indicate a link between PI3K-dependent protein kinase Akt (PKB) activation and both Epac and Rap1. Akt is a major effector of the PI3K signal and the PI3K/Akt pathway is frequently altered in cancers (16). Phosphatidylinositol 3,3,5-triphosphate [PI(3,4,5)P3] produced by PI3K upon activation by receptor tyrosine kinases (RTK) and GPCR binds Akt. This then allows the Akt-activating phosphorylation at Thr308 by phosphatidylinositol-dependent kinase 1 (PDK1) and at Ser473 by a putative PDK2. Modulation of PI3K-dependent Akt signaling by Epac and Epac/Rap1 has been observed for various cell responses and may be opposite to Akt signaling through PKA (13, 17, 18). Finally, recent studies have proven that cyclic AMP could mediate a physical association between Epac, Rap1, and phosphorylated Akt (19, 20). The in vitro data presented in this study define a novel role for cyclic AMP in the regulation of the plasticity of aggressive melanoma cells shown by the VM phenotype.

Cell cultures, treatments, and observations. The uveal MUM-2B and cutaneous C8161 aggressive human melanoma cell lines (a generous gift from Dr. M. Hendrix, Children's Memorial Research Center, Chicago, IL) have been described previously (7). VM was initiated by seeding 2.5 × 10⁵ cells on dried matrices produced with 13 μL of Matrigel (BD Biosciences) diluted one-half on 15-mm glass coverslips. When indicated, 15 μL of rat tail collagen I (4.1 mg/mL; BD Biosciences) were also used (21). For phenotype observations, treatment with pharmacologic agents was carried out 24 h after cell seeding, when cells were just confluent and then renewed daily. Cells were stained with periodic acid Schiff (PAS; ref. 7) and VM patterns were observed with an Olympus IX-70 microscope equipped with the AnalySIS FIVE software (Olympus). For semiquantitation of VM pattern, images were converted into gray levels and threshold of lighter, PAS-negative zones, determined and used to calculate the PAS-negative area. The PAS-positive area, accounting for tubular networks and isolated PAS-positive cells, was calculated by subtracting the PAS-negative area from the total microscopic field and expressed as a percentage of the latter. For the same total field, the number of tube connections was also determined. Student's test was performed on at least five measures for each variable collected from six different experiments.

For biochemical investigations, ∼5 × 10⁶ cells at preconfluency were switched to serum-free medium overnight before treatment for 15 to 20 min with pharmacologic agents in serum-free medium. For double exposure to agents, inhibitors H8, U0126, and LY294002 were incubated for 45 to 60 min before and throughout the follow-up 15- to 20-min treatment with other pharmacologic agents. All compounds were from Sigma.

Rap1 activation assay (“Pull down”). Cells at preconfluency were serum-starved overnight, treated or not (control) for 15 min with pharmacologic agents in serum-free medium, and lysed in 500 μL cell lysis buffer as described previously (22). To pull down activated GTP-bound Rap1, lysates were centrifuged and 740 μg (∼350 μL) of supernatant proteins were incubated for 1 h at 4°C with 25 μL of glutathione-agarose beads (Sigma) preloaded with glutathione S-transferase–tagged Rap-binding domain of RalGDS (GST-RBD-RalGDS) and then processed as already described (22) using an antibody against Rap1 (sc-65; diluted 1:200; Santa Cruz Biotechnology) and an Alexa fluor 488-conjugated secondary antibody (Invitrogen). Total Rap1 amounts in clarified lysates were also determined. The pGEX plasmid vector containing the GST-RBD-RalGDS sequence was kindly provided by Dr. JL Bos and W Pellis, University Medical Center, Utrecht, the Netherlands.

SDS-PAGE and immunoblot analysis. Protein concentration was determined with the Quanti-Pro BCA kit (Sigma) on 10,000 g supernatants from cells extracted in 50 mmol/L Tris-HCl (pH 7.5) containing 120 mmol/L NaCl, 2.5 mmol/L MgCl₂, 10% glycerol, 1% NP40, 2 mmol/L activated Na₃VO₄, 5 mmol/L NaF, and protease inhibitors. Proteins (100 μg) were analyzed by 10% reducing SDS-PAGE and immunoblotting using antibodies as indicated in figure legends. Blots were then reacted with secondary antibodies conjugated to Alexa fluor 488 (Invitrogen) for signal detection with a phosphofluoro-imager (FLA 2000; Fujifilm).

Immunofluorescence and confocal microscopy. For “in situ Rap1 pull down,” MUM-2B cells cultured in 8-chambers Labtek slides (D. Dutscher) coated with collagen I (0.6 μg/mL) were serum-starved overnight and treated for 75 s with pharmacologic agents in serum-free medium as indicated in the Results section. After fixation with 3.7% paraformaldehyde and incubation with 50 mmol/L NH₄Cl for 5 min, they were permeabilized for 5 min with 0.1% Triton-X100 and saturated for 10 min with 1% bovine serum albumin. GST-RBD-RalGDS activation-specific probe (diluted 1:4 in PBS) was then incubated for 1 h at room temperature with cells. Specificity of the GST-RBD-RalGDS/Rap1-GTP complex formation was ascertained by omitting the activation-specific probe in some chambers. Cells were incubated simultaneously for 2 h with a goat antibody against GST (sc-34072; Santa Cruz Biotechnology) and the rabbit antibody against Rap1. Alexa fluor 594–conjugated anti-rabbit secondary antibody (Invitrogen) and Alexa fluor 488–conjugated anti-goat secondary antibody (Invitrogen) were then added for 45 min. Nuclei were labeled with a 4′,6-diamidino-2-phenylindole solution. Slides were observed by epifluorescence with the Olympus IX-70 microscope. Some cells were also observed with the TCS 4D confocal laser scanning microscope (Leica).

Cellular cyclic AMP elevation inhibits VM in vitro through the contribution of an Epac-dependent signaling pathway. We first raised the cellular cyclic AMP level with a mix of the direct activator of most adenylyl cyclases, forskolin (10 μmol/L), and the nonspecific inhibitor of phosphodiesterases, 3-isobutyl-1-methyl-xanthine (IBMX; 100 μmol/L; F/I; Fig. 1A). The aggressive uveal MUM-2B and cutaneous C8161 melanoma cells cultured on Matrigel matrices were either untreated (control) or treated with the mix for 2 days. Intracellular elevation of cyclic AMP clearly inhibited the formation of VM in both cell lines. The effect was reversible as the subsequent removal of F/I during 2 additional days allowed VM to commence (Fig. 1A , third column).

In an attempt to semiquantify the morphologic observations, we performed image analyses of MUM-2B VM networks by selecting two variables. First, the surface of PAS-stained elements accounted for both tubular network importance and the amount of cells producing an intense extracellular matrix, a feature of commitment to the expression of the VM multipotent phenotype (5, 7). However, as this variable overestimates the amount of VM tubes, we introduced a second one: the tube connection number. Because multipotent phenotype-expressing cells can appear among an aggressive cell population without necessarily being organized into typical VM tubes, this variable accounts for a full-structured VM network. As shown in Fig. 1D, such analysis revealed that F/I induced a 73% inhibition of PAS-stained tubular structures and cells and an 88% loss of connections, the net result being a total impairment of tube formation (Fig. 1A).

To confirm that elevated intracellular cyclic AMP levels inhibit VM, we used a cell permeable cyclic AMP analogue, the dibutyryl-cyclic AMP (dbcA; 0.5 mmol/L), and GPCR physiologic ligands, namely adrenaline (ADR, 0.1 μmol/L) and vasoactive intestinal peptide (VIP; 0.01 μmol/L). All these agents led to a reduction and/or inhibition of patterned vasculogenic-like network formation to different extents in MUM-2B cells (Fig. 1B,, Matrigel; Fig. 1D, 52.5–92% inhibition of PAS-stained structures, 78–94% inhibition of connections). Similar effects occurred with C8161 cells (data not shown). Cyclic AMP–elevating agents (CAEA) also impaired VM when the cells were cultured on collagen I matrix, as shown for MUM-2B cells treated with F/I, ADR, and VIP (Fig. 1B , collagen). However, on collagen, the VM pattern displayed much fewer tubes and connections and was more heterogeneous between experiments, owing to an apparent greater dependence on local variations of matrix thickness. Additionally, the effects of GPCR ligands were less substantial. In subsequent experiments, we therefore preferred to use Matrigel. Altogether, these morphologic data provide evidence that cyclic AMP can regulate VM in vitro.

To investigate the mechanisms underlying the cyclic AMP effect, we examined the participation of its primary targets PKA and Epac. We cultured MUM-2B and C8161 cells during 3 d in the presence of, respectively: an activator of PKA, N⁶-Benzoyladenosine-3′,5′-cyclic monophosphate (6Bnz; 50 μmol/L), an inhibitor of PKA, N-[2-(Methylamino)ethyl]-5-isoquinolinesulfonamide, (H8; 20 μmol/L), and with the only known specific chemical modulator of Epac, the activator 8-(4-Chlorophenylthio)-2-O-methyladenosine 3,5-cyclic monophosphate (8CPT; 50 μmol/L; ref. 12). The concentrations stated above were used throughout our investigations. As shown in Fig. 1C and D, 6Bnz and H8 had no significant effect on the promotion of VM in MUM-2B cells, whereas 8CPT led to the disorganization and partial inhibition of VM tube formation (43% of PAS-stained structures and 60% of connections). In the latter case, impairment of VM was not as strong as that observed with F/I, indicating that the Epac signal only partially contributes to the VM process. 8CPT frequently induced a lightening of the PAS stain (as did F/I), suggesting a lower extracellular matrix synthesis (Fig. 1C,, 8CPT). Although PAS-stained cell aggregates were still observed, they were flat with reduced organization into tubes. In addition, PKA modulators 6Bnz and H8 were ineffective at reversing or modifying the inhibitory effect of F/I on VM (Fig. 1C,, second row). 8CPT added concomitantly with F/I induced a possible additive or F/I-dominant inhibition of VM (Fig. 1C). Similar observations were made with C8161 cells (data not shown). These results suggest that an Epac-dependent but PKA-independent pathway partially contributes to the cyclic AMP-induced VM inhibition observed in forskolin-stimulated cells.

Cyclic AMP induces an Epac-Rap1 signal in MUM-2B melanoma cells. Because Epac is a Rap1-specific guanine nucleotide exchange factor, we hypothesized that Epac-Rap1 signaling may be involved in the impairment of VM by cyclic AMP. Focusing on MUM-2B cells, we examined the levels of activated, GTP-bound Rap1 using a pull-down assay with GST-RBD-RalGDS activation-specific probe. Kinetics experiments showed that F/I induced a 3-fold increase in cellular basal level of cyclic AMP within 10 minutes that remained constant over 40 minutes of stimulation (Supplementary Fig. S1). We then decided to determine Rap1-GTP levels in cells stimulated for 20 min with, respectively, F/I and 8CPT, in the presence or absence of the PKA inhibitor H8. As shown by the representative experiment in Fig. 2A, although H8 did not affect levels of Rap1-GTP, 8CPT induced an increase, thus demonstrating a Epac-linked Rap1 activation via a PKA-independent modality. The same results were obtained with F/I, confirming that cyclic AMP-induced Rap1 activation was mediated by Epac and not PKA. The PKA-independent nature of Rap1 activation was additionally supported by the absence of modulation of the Rap1-GTP basal level by H8 or 6Bnz alone (Fig. 2A). As expected, dbcA increased Rap1-GTP (Fig. 2B). The GPCR ligands α-melanocyte stimulating hormone (αMSH; 1 μmol/L) and VIP also activated Rap1, although the signal detected was less intense than with dbcA (Fig. 2B).

Direct visualization of the activated state of MUM-2B Rap1 was next achieved by carrying out an in situ pull-down assay (see Materials and Methods). Untreated control cells were characterized by a (total) Rap1 fluorescent red staining throughout the cytoplasm with a stronger localization close to the nucleus, likely within the Golgi (Fig. 3, first row, left). Basal activated Rap1-GTP was scattered all over the cytoplasm, as shown by the faint green labeling (Fig. 3, first row, middle). Compared with unstimulated control cells, much stronger fluorescent staining of Rap1-GTP pools could be observed after 75 seconds after either Epac activation by 8CPT or increased levels of cyclic AMP by F/I or dbcA (Fig. 3, second column). Confirming the double-label microscopic observations (Fig. 3, third column), confocal microscopy revealed that the increase in Rap1-GTP mediated by Epac occurred for preexisting cytoplasmic pools of Rap1 located outside the Golgi area (Fig. 3, insert).

ERK1/2 can participate toward the inhibition of VM by forskolin but not by the GPCR ligands αMSH or VIP. To investigate the possible role of ERK1/2 in VM inhibition by cyclic AMP, MUM-2B and C8161 cells were challenged with U0126 (10 μmol/L), a compound that specifically inhibits MEK1/2, the immediate upstream kinase of ERK1/2. Within 2 days, the U0126 induced a complete inhibition of VM in the MUM-2B cells (Fig. 4A,, U0126; Fig. 1D, inhibition of 81% PAS-stained structures and 91% of connections). VM impairment by 8CPT was stronger in the presence of U0126 (Fig. 4A,, 8CPT+U0126), indicating either an additive or dominant effect of U0126. As expected, when ERK1/2 was nonoperative, the concomitant modulation of PKA was ineffective on the inhibited VM (Fig. 4A , 6Bnz+U0126, H8+U0126). Similar results were observed with C8161 cells (data not shown). These data show that if ERK1/2 is inhibited, VM is abolished.

To decipher the possible links between cyclic AMP-mediated activation of Epac/Rap1 and ERK1/2-mediated inhibition of VM, we monitored the phosphorylation (activated) status of ERK1/2. Serum-starved MUM-2B cells were challenged for 20 minutes with F/I, 8CPT, H8, 6Bnz, or a combination of H8 with F/I or 8CPT. Figure 4B shows that F/I treatment inhibited ERK1/2 basal phosphorylation, whereas 8CPT was ineffective. The concomitant presence of the PKA inhibitor H8 did not alter the phosphorylation status observed upon F/I or 8CPT treatment. Indeed, neither modulator of PKA activity, 6Bnz, or H8, modified the phosphorylation of ERK1/2 in these cells (Fig. 4B). In additional experiments, we observed no basal or F/I-induced phosphorylation of C-Raf at the Ser259 residue (data not shown). In many cells in which cyclic AMP inhibits ERK1/2, this site is phosphorylated by PKA, resulting in C-Raf inactivation (11).

In conclusion, forskolin induces an inhibition of ERK1/2, which is neither linked to Epac nor PKA (and likely not to C-Raf). Furthermore, the modulation of ERK1/2 activity does not transduce Epac/Rap1 activation.

To determine whether physiologic GPCR ligands could modulate ERK1/2 phosphorylation, we exposed MUM-2B and C8161 cells for 20 minutes to either VIP or αMSH. To check for a possible opposing effect of phosphodiesterases, we added IBMX in the set of experiments related to C8161 cells. As shown in Fig. 4C, neither the GPCR ligands nor dbcA modified ERK1/2 basal phosphorylation in either cell line (Fig. 4C).

Lastly, we checked for the cyclic AMP-dependent phosphorylation of Thr599/Ser602 residues in the B-Raf activation segment of the catalytic kinase domain. Indeed, B-Raf rather than C-Raf is operative in the ERK1/2 MAPK pathway of melanocytic cells (11, 23). Figure 4D shows that CAEA as well as Epac activation by 8CPT did not modify basal Thr599/Ser602 phosphorylation of 62- or 95-kDa B-Raf isoforms in MUM-2B cells. The PKA inhibitor H8 also showed no effect.

We first conclude that direct inactivation of ERK1/2 inhibits VM. Second, forskolin inhibits ERK1/2 phosphorylation in a PKA- and Epac-independent manner, whereas GPCR ligands and 8CPT-activated Epac are ineffective. Third, no PKA signal, involving or not C-Raf is transmitted to ERK1/2 in these cells. Lastly, neither CAEA nor activated Epac modify the basal phosphorylation of the B-Raf kinase activation site.

PI3K/Akt signaling can participate toward the inhibition of VM by forskolin but not by the GPCR ligands αMSH or VIP. We wondered if the inhibition of VM by cyclic AMP could be mediated in part by a PI3K/Akt signal. To this end, we raised cyclic AMP levels in MUM-2B cells during 20 minutes in the presence of the PI3K inhibitor LY294002 (20 μmol/L). Akt activation status was assessed via the detection of phosphorylated Ser473 and Thr308 critical residues. In addition, we examined the Ser241 phosphorylation of PDK1 which is required for enzyme activity (24). As a control for cyclic AMP–mediated GPCR-dependent activation of PI3K, we also evaluated the Tyr508 phosphorylation status of the PI3K p85α subunit that depends on RTK activity (25). Figure 5A shows that F/I inhibited basal Akt phosphorylation at Ser473, in contrast to 8CPT. Although dbcA seemed to be ineffective, αMSH and, to a lesser extent, VIP, slightly increased Akt phosphorylation (as did ADR; data not shown; Fig. 5B). A positive control was performed with ADR+insulin (ADR/INS) treatment, known to induce a strong Akt Ser473 phosphorylation in other cell types (Fig. 5A; ref. 18). For all the treatments, the PI3K inhibitor LY294002 abolished Akt Ser473 basal and induced phosphorylation, indicating that Akt activity was PI3K dependent.

Akt phosphorylation at Thr308 could not be detected, although it was observed in cells challenged with ADR/INS (Fig. 5A). Indeed, the latter treatment induced a considerably stronger Ser473 phosphorylation signal than those obtained with CAEA. Because the magnitude of the Thr308 phosphorylation signal seemed to be comparatively much lower than that of Ser473, Thr308 phosphorylation may not have been detected under treatments with GPCR ligands. Finally, the basal phosphorylation of PDK1 at Ser241 was not modified, indicating that PDK1 retained its full constitutive activity throughout the experimental conditions. As expected, p85α Tyr508 phosphorylation remained unchanged upon treatments, showing that Akt activation was mediated by class IB PI3K. Similar results were obtained with C8161 cells (data not shown).

Altogether, these data show that in contrast to αMSH and VIP, forskolin inhibits the PI3K/Akt module.

In this study, we have found that cyclic AMP signaling mediates in vitro the ability of aggressive melanoma cells to engage into VM, an example of tumor cell plasticity. In uveal and cutaneous melanoma cells, an increase in cyclic AMP by either pharmacologic agents or physiologic GPCR ligands leads reversibly to VM inhibition. However, predictive of the complexity of underlying ligand-specific signal pathways, various GPCR ligands induce inhibition of VM to different extents. Whereas forskolin shows the strongest effect followed by ADR and VIP, αMSH causes a variable and weak inhibition of VM, even in the presence of the phosphodiesterase inhibitor IBMX (data not shown). We hypothesize that these differences may be linked to the various adenylyl cyclase isoforms (10) and to the large repertoire of the GPCR superfamily (26, 27) responsible for the tuning of cyclic AMP biological responses.

Modulators of PKA neither modify the overall pattern of VM nor the forskolin-induced VM inhibition. This suggests that PKA is not a cyclic AMP mediator of VM. On the other hand, activation of the cyclic AMP target Epac by 8CPT impairs VM to some extent. Furthermore, forskolin induces a PKA-independent activation of Rap1 as does 8CPT. All GPCR ligands as well as dbcA activate Rap1, in contrast to modulators of PKA. Epac therefore represents a target of cyclic AMP that contributes to the inhibition of VM signaling. While Epac is associated with an increasing number of functional responses (13), our data report for the first time an additional role as a mediator of tumor cell plasticity.

Cyclic AMP has long been recognized as a messenger for the ERK1/2 MAPK pathway (11). PKA, Epac, and their downstream-related targets Ras and Rap1 could all mediate ERK1/2 activity through multiple signaling networks. Our study confirms that inhibition of ERK1/2 abolishes VM as already mentioned by Hess and colleagues (28). More precisely, we have found that forskolin inhibits ERK1/2 in a PKA-independent manner in MUM-2B and C8161 cells. ERK1/2 shows no modulation with 8CPT-activated Epac, suggesting that forskolin-activated Epac/Rap1 cannot transduce to ERK1/2 the inhibitory signal that causes VM impairment. This observation is reminiscent of what occurs in B16 mouse melanoma cells in which cyclic AMP activation of Epac/Rap1 is not transduced into an activation of ERK1/2 (23). In these cells, cyclic AMP activates ERK1/2 independently of Epac and PKA, suggesting the participation of an as yet undefined exchange factor(s) for Ras. Additionally, in melanocytic cells, cyclic AMP cannot activate C-Raf, whereas Ras transduces a PKA-independent signal to ERK1/2 (11, 23). Although detailed examination of C-Raf activation was beyond the scope of this study, we observed an independency of PKA/C-Raf in the cyclic AMP response because no Ser259 phosphorylation could be detected in control or CAEA-treated MUM-2B cells (data not shown). Indeed, in B16 cells, B-Raf transduces the Ras signal induced by cyclic AMP to ERK1/2 (11, 23). In our cells, none of the CAEAs modified Thr599/Ser602 phosphorylations in the B-Raf activation segment. This might suggest that B-Raf does not transduce the forskolin-inhibiting signal to ERK1/2. Similarly, no modification of these phospho-sites was observed after activation of Epac, which might suggest the uncoupling of Epac/Rap1 to B-Raf. However, based on these experiments, we cannot definitely rule out the possible involvement of B-Raf through other regulatory phospho-sites such as the enzyme inhibitory residues Ser365, Ser429, and Thr440.

As has been observed in other cells (11), forskolin inhibited ERK1/2 in the cells tested here. This is in contrast however to that occurring in B16 cells or some metastatic human melanoma cells (29). However, in the latter cell types, the activation of ERK1/2 was PKA-independent but Epac/Rap1-dependent. Notably, such activation was obtained not by the direct activation of adenylyl cyclase and GPCR as in our study but upon stimulation of a RTK. Thus, ERK1/2 modulation also depends on the primary receptors involved. None of the physiologic CAEA used in our study was able to modulate ERK1/2 activity, showing that ERK1/2 does not mediate VM inhibition via GPCR. In contrast, ERK1/2 inhibition by forskolin impairs VM. The mechanism of such inhibition remains to be fully determined, although we have established that neither PKA nor Epac/Rap1 are involved. Our data also indicate that the mechanisms of VM impairment likely depend on the mode by which cyclic AMP is elevated, as is found in other cells for different biological responses. For instance, the adenylyl cyclase 9 isoform is insensitive to forskolin (30), which may account for cytokine production differences in Kupffer cells in response to either forskolin or GPCR ligands (31). In addition, the cell compartmentalization of effectors, especially Epac and Rap1, plays an important role. Wang and colleagues (14) have shown that upon elevation of cyclic AMP, Epac activates a perinuclear pool of Rap1 that does not result in ERK1/2 activation, whereas Epac activated by C3G guanine nucleotide exchange factor localizes to the plasma membrane, thereby stimulating ERK1/2. In this regard, our in situ Rap1 activation assay shows that forskolin and 8CPT rather activate cytoplasmic pools of Rap1 than focal pools at the plasma membrane. Using indirect immunofluorescence, we observed similar staining intensities of phosphorylated ERK1/2 in cells challenged with GPCR ligands and in untreated cells (data not shown). ERK1/2 was located in the cytoplasm and/or the nucleus but not at the plasma membrane. Forskolin induced an inhibition of basal pools of phosphorylated ERK1/2, similar to that obtained with U0126 inhibitor, without any particular relocation. These observations are in agreement with the lack of ERK1/2 activation by either of the compounds used in our experiments.

Several studies have shown in various cell types that cyclic AMP activates Akt in an Epac- but not PKA-dependent manner (see for examples, refs. 17, 18). In addition, PI3K has been recognized as an important mediator of VM, because its inhibition leads to VM abolition (32). In our cells, forskolin inhibited Akt phosphorylation, whereas GPCR ligands acted as stimulators. In light of our data, Fig. 6 represents a likely scheme of events occurring for the inhibition of VM by cyclic AMP. Forskolin and GPCR ligands both activate an Epac-dependent but PKA-independent pathway that contributes directly to VM inhibition and whose downstream targets remain to be identified. Unlike GPCR ligands, however, forskolin also inhibits ERK1/2 independently of Epac and PKA, leading to VM abolition. Furthermore, contrary to GPCR ligands, forskolin mediates the inhibition of PI3K/Akt, which in turn contributes to VM impairment. In this regard, the faint PI3K/Akt signal stimulation obtained with 8CPT-activated Epac and GPCR ligands is in apparent contradiction with VM inhibition. One explanation could be that the slight PI3K/Akt signal is dominated by the main Epac/Rap1 inhibitory signal induced by all CAEA. Another, but not exclusive, possibility is that activation of PI3K/Akt relates to concomitant cell responses, such as proliferation and survival (see for example, ref. 33). On the whole, our results strengthen the concept that cyclic AMP can switch on simultaneously multiple pathways to mediate VM inhibition.

Aside its links with ERK1/2, PI3K and Epac, how might cyclic AMP signaling relate to other mediators of VM? To our knowledge, no experimental data are available, although some reports suggest relationships between cyclic AMP and both VE-cadherin and Nodal. Cyclic AMP, through an Epac/Rap1 signal, enhances endothelial barrier properties via the redistribution of VE-cadherin and a strengthened cell adhesion (see for instance, ref. 34). The process seems to depend on endothelial cell origin, with cyclic AMP even displaying possible opposite effects (see for instance, ref. 35). Concerning VM inhibition by cyclic AMP, a possible change in VE-cadherin properties remains to be established. On the other hand, Yurugi-Kobayashi and colleagues (36) reported that under vascular endothelial growth factor treatment, an activation of cyclic AMP signal specifies vascular progenitor commitment into arterial rather than venous endothelial cells, associated with an increase in Notch-1 and Notch-4 gene transcription. Interestingly, Hendrix's laboratory reported that Notch-4 may preferentially up-regulate Nodal expression in aggressive melanoma cells (37). Activation of Nodal signaling supports VM and expression of the VM plasticity marker VE-cadherin (38). It is enticing to hypothesize therefore that, similarly to what occurs in endothelial cells sharing phenotypic markers with melanoma cells committed to VM, cyclic AMP switches on Nodal signaling in these latter cells. However, according to our results, the cyclic AMP–dependent activation of a Nodal signal would result in the inhibition of VM, in apparent contradiction with Hendrix's findings. Several hypotheses can currently be devised. As reported by Hendrix's group, Nodal is expressed at a high basal level in MUM-2B and C8161 cells (9, 38). Therefore, an overstimulation of Nodal signal could be nonoperative, excluding Nodal from the cyclic AMP–dependent inhibition of VM. However, activation of Nodal signaling in the embryonic endothelium of transgenic mice expressing constitutively activated Notch-4 leads to disorganized vascular networks (39). The data might suggest either that Nodal regulation of vascular morphogenesis is different and even opposite to that of VM, or, similarly to this in vivo model, that a possible cyclic AMP-mediated activation of Nodal may contribute to the disapperance of VM networks. Future investigations should clarify this point.

Conclusively, our study represents a new implication of cyclic AMP signaling and Epac function in tumor cell plasticity. These findings may help to define potential targets for the therapeutic management of melanoma tumors.

No potential conflicts of interest were disclosed.

Grant support: “Société de Recherche Dermatologique” France.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Dr. Mary Hendrix and Elisabeth Seftor for their generosity in providing the MUM-2B and C8161 cell lines; Dr. Johannes Bos and Wendy Pellis for their kind gift of the GST-RBD-RalGDS plasmid construct and accompanying experimental protocol; and Joel Courageot for confocal microscopy.