Melanomas require HEDGEHOG-GLI signaling regulated by interactions between GLI1 and the RAS-MEK/AKT pathways

SHH, PTCH1, and GLI1 were expressed in the matrix of anagen-stage human scalp hair follicles surrounding the dermal papilla (blue in Fig. 1 A and B), including pigmented melanocytes (brown in Fig. 1B Center) and those expressing the melanocyte lineage regulator microphthalmia-associated transcription factor (MITF) (11) (red in Fig. 1B Right). These mRNAs were not detected in MITF+ upper hair follicles or in epidermal basal layer melanocytes (Fig. 1A and data not shown). Proliferating foreskin melanocytes expressed GLI1 and other SHH pathway components [supporting information (SI) Fig. 5].

Addition of SHH ligand for 48 h increased melanocyte BrdU incorporation, as a measure of proliferation, by ±2-fold, whereas treatment with cyclopamine (cyc), a selective inhibitor of SMOH, decreased proliferation ± 2-fold (Fig. 1C). These changes are similar to those seen with cerebellar granule neuron precursors, the proliferation of which is regulated by SHH signaling (12). Cyc did not induce β-gal+ senescence (not shown). SHH also induced its targets GLI1 and hedgehog-interacting protein (HIP) by 2- to 3-fold and by 6-fold that of MITF-M (Fig. 1D), the melanocyte-specific transcript of MITF. Conversely, cyc reduced GLI1, PTCH1, and HIP levels. It also decreased those of MITF-M and of its targets TBX2 and BCL2 (Fig. 1E; refs. 13 and 14), consistent with the presence of putative GLI-binding sites near MITF (not shown), identical to others shown to bind GLI1 (15), and with the finding that misexpression of human GLI1 in frog embryo epidermis (9) induced the expression of endogenous Gli1, Ptch1, and Mitf (SI Fig. 6).

SHH-GLI pathway components were detected in human melanoma and nevi (blue in Fig. 1F Center; SI Fig. 7 and SI Tables 1 and 2), coincident with MITF-M and MELAN-A, two melanoma markers (red in Fig. 1F Right, SI Fig. 7). PTCH1, GLI1, GLI2, and GLI3 expression levels varied but were similar to those in sporadic basal cell carcinomas (9) (BCCs; SI Fig. 5). The GLI1 targets PDGFRα, FOXM1, and BMI1 (10, 16–18) were also expressed in melanomas and nevi (SI Fig. 5).

Four skin (Me-1 and -3–5) and one lymph node (Me-2) metastases (SI Table 3) grown as primary cultures had pigmented elongated cells with prominent nucleoli. All expressed SHH-GLI pathway components and the melanoma markers NESTIN and GP100 (refs. 19 and 20; Fig. 1G). Most also expressed MELAN-A and the melanoma stem/precursor markers SOX10 and PAX3 (21) (Fig. 1G). Single-cell clones expressed GLI1, and Me-3 was tumorigenic in nude mice, producing a heavily pigmented tumor type after 75 days (not shown).

Cyc inhibited PTCH1, BMI1, FOXM1, and MITF-M expression in three primary cultures tested (Fig. 1H), although MITF-M was not detected in Me-3 or -4. MITF is commonly repressed in metastatic melanomas (22). All primary cultures, the primary cutaneous melanoma cell line WM-115 (BRAF^V600E), and the two metastatic melanoma cell lines SK-Mel2 (NRAS^Q61R) and MeWo responded to 48 h cyc treatment in a dose-dependent manner (Fig. 1I; SI Fig. 8A) and decreased GLI1 expression (data not shown), as compared with sibling cultures treated with tomatidine (tom) as control. Cyc showed cumulative effects over time (SI Fig. 8B), leading to complete proliferative arrest after 10 d for WM-115 and ≈80% for MeWo. Twenty days of treatment were sufficient to kill the cells and prevent the recovery of the cultures as determined 10 d after cessation of treatment (Fig. 1J, arrows). Cyc also increased apoptosis by 1.2- to 11.3-fold, as revealed by activated Caspase-3 labeling after 48 h (SI Fig. 9). In contrast, addition of anti-SHH blocking antibody for 48 h reduced proliferation of Me-1 by only ≈50% (data not shown). Addition of exogenous SHH for 48 h produced a modest but significant increase in BrdU incorporation in three of five primary cultures and in two of three cell lines (SI Fig. 10).

GLI RNA interference with siRNAs (23, 24) targeting GLI1 or GLI2 reduced BrdU incorporation after 48 h in all primary cultures and cell lines with lipofection efficiencies >30% (Fig. 1K; SI Fig. 11), whereas siRNAs against GLI3, and an unrelated control siRNA (siC) did not. An independent set of siRNAs gave the same result (SI Fig. 11C and D). siGLI1 led to complete growth arrest after 10 d for Me-3 and after 6 days for WM-115, whereas MeWo cells were more refractory, but the number of viable cells decreased in all cases (Fig. 1L; SI Fig. 11E). Cotransfection of GLI1- and GFP-expressing plasmids enhanced the proliferation of GFP+ SK-Mel2 cells over that of GFP+ cells cotransfected with an empty vector, and GLI1 rescued the inhibitory effects of cyc (Fig. 1M).

Three cultures transfected with siGLI3 or siC and then treated with cyc for 48 h showed that, whereas siC had no effect, GLI3 knockdown rescued the effects of cyc (±80% rescue for Me-3 and full rescue for WM–115 and SK-Mel2; Fig. 1N). Consistently, a C-terminally truncated GLI3 repressor (GLI3R; ref. 25), but not full-length GLI3, reduced BrdU incorporation to a similar extent as cyc treatment (Fig. 1M).

s.c. injection of metastatic melanoma MeWo cells, transduced with a lacZ-expressing lentiviral vector into nude mice, resulted in the rapid growth (7 d) of pigmented tumors (Fig. 2A, C, and F; n = 10). Injection of cyclodextrin-complexed cyc, but not cyclodextrin alone as carrier, intra- and peritumorally resulted in the rapid regression and disappearance of small (±10 mm³; n = 10) or larger (±65 mm³; n = 7) tumors (Fig. 2 A and C–K) and the appearance of a scar (Fig. 2 G and H). The length of treatment to achieve macroscopically tumor loss varied depending on the initial size of the tumors (Fig. 2A). Cyc induced a 2-fold reduction in BrdU incorporation and a 6-fold increase in activated caspase-3 labeling (not shown). WM-115 cells also induced s.c. tumors after ± 90 days, and cyc treatment led to their disappearance (n = two control carrier treated with tumors and n = two cyc treated without tumors; data not shown).

Systemic treatment of large s.c. tumors (±65 mm³) was less effective than direct intra- and peritumoral injection, showing a 33% decrease in tumor volume after 16 days of cyc treatment (n = 20) as compared with carrier-only treated tumors (n = 20; data not shown). The incomplete effects are likely because of inefficient delivery, because in vitro culture showed that tumor-derived cells maintained sensitivity to cyc (not shown).

X-Gal staining of histological sections of five overtly cured tumors treated by direct injection, taken 1–2 days after macroscopic disappearance, showed that four of five tumors contained small pockets of β-gal+ tumor cells or groups of single cells (Fig. 2J and data not shown), posing the threat of recurrence. MeWo/LacZ-melanoma-bearing mice were then treated with cyc intra- and peritumorally until tumor disappearance, split into three groups, treated for an additional 3, 10, or 20 days by injection into the site where the tumor had been, and left untreated thereafter. Tumors reappeared within 2–7 d from the end of 3 d (n = two mice) and 10 d (n = two mice) cyc treatments. However, there were no recurrences an additional 1.5 mo after 20 d treatment following tumor disappearance (n = 3; Fig. 2B).

To ascertain that the tumor-inhibitory effect of interference with HH pathway activity directly affects melanoma cells and to control for any hypothetical nonspecific effects of cyc, we used a lentiviral vector expressing a short hairpin RNA against SMOH (24), the target of cyc. Transduction of MeWo cells (72 h; ±80–90% efficiency) with LV-shSMOH but not with a control lentivector (LV-control) at similar multiplicity of infection, resulted in the specific decrease of SMOH mRNA by 55% and of GLI1 mRNA by 45%, as assessed by quantitative PCR. Grafting showed the normal growth of s.c. melanomas from LV-control-transduced cells (Fig. 2 L and M), whereas transduction with LV-shSMOH resulted in a large decrease in tumor volume (Fig. 2 L and N). Because the lentiviral particles used were noninfectious, the effect is cell-autonomous.

MeWo/LacZ melanoma cells were injected into the tail vein of adult nude mice, and the animals were allowed to rest for 2 weeks before starting i.p. injections of cyclodextrin-coupled cyc or carrier alone for 45 d. Carrier-only treated mice had metastases in all lung lobes (n = six of seven; Fig. 2 O and P). In contrast, the lungs of only one cyc-treated mouse showed very few small metastases (n = one of seven). The other cyc-treated mice (n = six of seven) did not have a single blue cell in their lungs (Fig. 2 O and Q).

To investigate a requirement of HH-GLI signaling in tumors directly induced by oncogenic RAS, 14 skin primary and lymph node metastatic melanomas from a tyrosinase→NRAS^Q61K; Ink4a^−/− mouse model (26) were individually tested. All expressed Gli1 and Ptch1 but not Shh (Fig. 3A and data not shown), indicating the ubiquitous presence of an active Shh-Gli pathway downstream of ligand. The expression of Gli1 and Ptch1 but not of the NRAS^Q61K transgene was significantly higher in lymph node metastases vs. skin tumors (paired t test; Fig. 3A). Wild-type and Ink4a^−/− axillary, cervical, and inguinal lymph nodes were negative (Fig. 3A). In situ hybridization of primary and metastatic tumors confirmed expression of Gli1 and Ptch1 in pigmented tumor cells (Fig. 3 B–G).

RAS-induced melanomas in primary culture variously expressed Nestin and S100 (26), and all maintained expression of NRAS, Gli1, and Ptch1 (not shown). Their proliferation, but not that of mouse embryonic fibroblasts, was inhibited by cyc (Fig. 3H). Transfection of GLI1 enhanced proliferation and rescued the effects of 10 μM cyc (Fig. 3I). Longer treatments with 5 μM cyc (Fig. 3J) inhibited proliferation and increased apoptosis before the death of all melanoma cells. Similarly, lipofection of GLI1 siRNA inhibited BrdU incorporation after 48 h (Fig. 3K).

Individual RAS-induced melanomas of 6- to 8-mo-old mice were then treated in vivo soon after tumor appearance. Cyclodextrin-complexed cyc treatment for 5 d by intra- and peritumoral injection eradicated overt tumors, leaving a depression and a scar, whereas treatment with equivalent doses of carrier alone had no effect (Fig. 3 L–R). Histology revealed a large decrease in tumor volume after cyc treatment (Fig. 3 S and T), although there were tumor cells still present under the scar (Fig. 3T), but their numbers were not greatly dissimilar to those seen in adjacent areas of the skin in these animals (Fig. 3 U and V). This result parallels the treatment of medulloblastomas from Ptch1^+/−;p53^−/− animals with cyc (27), which also leads to a drastic decrease in tumor volume, but tumor cells are always present due to the ubiquitous genetic lesions.

The requirement of Hh-Gli function in oncogenic-RAS-induced mouse melanomas suggested a possible interaction between RAS and HH-GLI signaling in cancer. There is evidence for interactions between EGF and SHH in neural stem cell lineages (28, 29), FGF and GLI in the embryonic mesoderm (30) and SHH and phosphatidylinositol 3-kinase/AKT and SHH and PKCδ/MAPK in 3T3 cells (31, 32). MAPK/AKT signaling also can enhance GLI reporter activity (31, 32), and EGF signaling can affect the kinds of GLI targets induced in keratinocytes (33).

Treatment with specific MEK (U0126) or AKT (SH6) inhibitors decreased BrdU incorporation after 48 h in a concentration-dependent manner in SK-Mel2 and WM-115 cells (SI Fig. 12), eliciting different early gene responses (not shown). Treatment with low doses of MEK or AKT inhibitors or cyc resulted in limited effects, but their combination yielded additive and synergistic inhibition, suggesting cooperation (Fig. 4A).

Transfection of GLI1 into SK-Mel2 (Fig. 4B) and WM-115 (SI Fig. 13A) increased BrdU incorporation by >5-fold, but this was blocked by inhibition of endogenous MEK or AKT. GLI1 also rescued cyc inhibition (data not shown; see Fig. 1M). Conversely, enhanced function of oncogenic HRAS^V12G or NRAS^Q61K, both of which induce melanoma in mice (26, 34), increased BrdU incorporation by ±5- to 8.5-fold, and this was inhibited by treatment with cyc (Fig. 4B; SI Fig. 13A). In this assay, oncogenic RAS thus requires an active SHH-GLI pathway to enhance proliferation, possibly converging with it on the regulation of GLI function, because the PKA agonist forskolin, which inhibits GLI1 activity (35), also blocked oncogenic RAS-induced proliferation (Fig. 4B; SI Fig. 13A). Using an alternative assay in which transfection of oncogenic RAS enhances colony formation in cells already harboring mutant RAS (36), we found that colony formation enhancement by oncogenic RAS in SK-Mel2 cells was inhibited by coexpression of GLI3R (SI Fig. 13 B and C).

Full-length GLI1, but not GLI3 or GLI3R, induced ≈6- to 20-fold higher activity from a multimerized GLI-binding site (GLI-BS)–luciferase reporter in SK-Mel2 and COS-7 as compared with controls. NRAS^Q61K or HRAS^V12G alone induced GLI reporter activity by ≈1.5- to 2-fold over background, likely potentiating endogenous GLI function (Fig. 4 C and D; SI Fig. 14), which then leads to an increase in GLI1 transcription as RAS transfection induced >2-fold the level of GLI1 mRNA (not shown). The activity of GLI1, but not of GLI3, was enhanced ≈2- to 5-fold by coexpression of NRAS^Q61K or HRAS^V12G, and they did not endow GLI3R with positive activity (Fig. 4 C and D; SI Fig. 14). Dominant active MEK-1 (p45MAPKK^S222D, MEK1*) and dominant active AKT1 (N-myristoylated AKT1; AKT1*) had similar effects to oncogenic RAS (RAS*) (Fig. 4D; SI Fig. 14A). The activity of GLI1 alone or its enhanced activity by RAS*, MEK*, or AKT* was suppressed by GLI3R in a concentration-dependent manner (Fig. 4D; SI Fig. 14B). RAS* also enhanced the activity of GLI1 but only by ≈30% from a PTCH1 reporter (Fig. 4 C and D), with a single GLI-binding site (37), suggesting cooperativity. GLI1 transcriptional activity depended on the presence of GLI-binding sites (GLI-BS-mut and PTCH-BS-mut series; Fig. 4C). Similar results were observed in MeWo cells (data not shown).

Oncogenic RAS* (NRAS^Q61K or HRAS^V12G) drove the nuclear localization of myc-tagged GLI1 but not GLI3 in SK-Mel2 cells (Fig. 4 E and F; SI Fig. 15). MEK1* or AKT1* mimicked oncogenic RAS*, enhancing nuclear accumulation of GLI1 (Fig. 4 E and F). Conversely, blocking endogenous MEK or AKT activity for 1 h increased cytoplasmic localization of GLI1 but not GLI3 and reversed the effect of oncogenic RAS (Fig. 4F; SI Fig. 15). None of these treatments affected the localization of the nuclear GLI3R (SI Fig. 15 and data not shown).

Suppressor of fused (SUFUH) is a negative regulator of SHH-GLI signaling known to inhibit the transcriptional activity of GLI1 and to sequester it in the cytoplasm (38, 39). Enhanced levels of SUFUH by transfection in SK-Mel2 cells prevented the localization of nuclear GLI1 driven by oncogenic RAS* (Fig. 4 E and F) and inhibited the transcriptional activity of GLI1 after enhancement by cotransfected AKT1* (not shown) or RAS* in a concentration-dependent manner (Fig. 4G).

RAS-MEK/AKT and GLI1 interactions were conserved across species and in different cancer types as seen in monkey COS7 cells (SI Fig. 16), human prostate cancer LNCaP cells (SI Fig. 17), and human glioma U87 cells (SI Fig. 18). Moreover, using these cells, we found that PTEN, an endogenous inhibitor of AKT, inhibited GLI1 transcriptional activity and nuclear localization (SI Fig. 18), and that the enhanced nuclear localization of GLI1 driven by oncogenic RAS* was independent of changes in nuclear export (SI Fig. 19).

Endogenous GLI1 protein in SK-Mel2 cells was mainly nuclear (Fig. 4 H and I), and treatment with AKT or MEK inhibitors decreased the number of cells with nuclear GLI1 and increased the number of cells with cytoplasmic labeling (Fig. 4 H and I). Inhibition of RAS-MEK or AKT function should therefore shut down the HH-GLI pathway. Indeed, inhibition of endogenous MEK and AKT function rapidly and strongly down-regulated the endogenous expression of GLI1 and PTCH1 but not of PAX3 and NESTIN used as controls, therefore shutting down HH-GLI signaling in SK-Mel2 and primary Me-3 melanoma cells (Fig. 4J).

The essential role for SHH-GLI signaling in melanoma is unexpected given the lack of previous data implicating this pathway (2, 3), and that Ptch1^+/− mice develop other tumors (40, 41) but not melanomas. The same was true, however, before SHH-GLI function was implicated in human prostate and pancreatic cancers (e.g., refs. 23, 42, and 43). Although differences between humans and mice could depend on species- and/or tissue-specific modifiers of HH-GLI1 function, the number of human tumors requiring HH-GLI signaling may thus be much greater than suspected.

In the normal skin hair follicle, GLI1 expression is found in the matrix, where restricted stem cells reside (44), but not in the bulge area, which contains resting melanocyte stem cells in mice (45). SHH-GLI function in the matrix could therefore regulate melanocyte numbers by acting on the expansion of incoming melanocyte stem cells from the bulge, resident matrix restricted stem cells, or derived precursors. Here, SHH-GLI signaling could regulate properties of these progenitor cells, much as it regulates proliferation and self-renewal in brain stem cell lineages (e.g., refs. 28 and 29).

This role of SHH-GLI signaling in the control of melanocyte numbers is deregulated in tumors. SHH-GLI activity could be a basal, yet essential, requirement in melanoma and other human cancers, with other events driving disease progression. However, it has been proposed that cell-intrinsic increases in GLI1 activity (23) or ligand-dependent increases in pathway function (42) may drive cancer progression. In this context, we have not found different levels of GLI1 or PTCH1 in a limited set of sporadic human melanomas vs. nevi. However, we have found higher levels of Gli1 and Ptch1 in lymph node metastases vs. skin melanomas in the same tyr→NRAS^Q61K, INK4a^−/− mice. The acquisition of oncogenic RTK-RAS-RAF-MEK and/or AKT signaling (e.g., refs. 3, 5, and 7) could thus result in cell-intrinsic enhancements of GLI1 transcriptional activity (Fig. 4K) driven by enhanced nuclear localization and transcriptional activity, and such increases may then participate in disease progression. These findings can be extended to other human cancers that similarly require HH-GLI signaling and often harbor oncogenic RAS and/or loss of the PTEN tumor suppressor, such as those of the pancreas and prostate (e.g., refs. 23 and 43). Moreover, the regulation of ligands and receptors, such as PDGFRA (10, 24), by SHH-GLI that activate the RAS/AKT cascades suggests positive autocrine/paracrine feedback on GLI1 function.

The expression of GLI1 in both nevi and melanomas is puzzling, since the first are largely arrested by senescence (46). Because feedback inhibition of the RAS-MEK/phosphatidylinositol 3-kinase-AKT cascades has been suggested as a mechanism for oncogene-induced senescence (47), one possibility is that senescence might modulate GLI1 function, and likely stability, through the down-regulation of the RAS/AKT cascades. The balance of the activities of oncogenic RAS/AKT vs. SUFUH, PKA, or other HH-GLI modifiers, may therefore normally regulate GLI1 activity in normal and pathological conditions.

Melanoma incidence and mortality are on the rise, and there are no treatments for metastatic disease. We show that systemic or local interference with HH-GLI signaling inhibits melanoma growth and prevents recurrence and metastases. Treated adult mice did not reveal obvious side effects, with the exception of diarrhea in the smallest siblings treated systemically for ±2 mo. A limited long-term treatment with cyc killed all melanoma cells, preventing recovery in vitro and recurrence of in vivo. Interference with HH-GLI signaling may thus be effective to inhibit not only the growth of the melanoma bulk but also that of any putative melanoma cancer stem cells (48) that could regenerate the tumor population, possibly paralleling the requirement of HH-GLI signaling for glioma cancer stem cells (24). The data therefore suggest that inhibition of GLI1 function directly or through a combinatorial strategy blocking the HH-GLI and RAS/AKT pathways represents a therapeutic goal for melanomas and many other human cancers.

We thank A. Pellicer (New York University, New York, NY), R.Toftgård (Karolinska Institute, Huddinge, Sweden), J. Pouyssegur (University of Nice, Nice, France), and S. Konig, R. Brown, and I. Radovanovic (University of Geneva, Geneva, Switzerland) for reagents and help in obtaining samples. V.P. and F.B. were recipients of grants from the Leenaards Foundation and the Geneva Cancer League, and from Oncosuisse and the Swiss National Science Foundation, respectively. This work was supported by grants to A.R.i.A. from the Jeantet and Leenaards Foundations, Oncosuisse, the Swiss National Science Foundation, and the National Institutes of Health. A.R.i.A. is a member of the National Center of Competence in Research Frontiers in Genetics.