FMRP ligand circZNF609 destabilizes RAC1 mRNA to reduce metastasis in acral melanoma and cutaneous melanoma

The detailed information of human cutaneous melanoma cell lines and acral melanoma cells, growth media, and their metastatic state are in Table S1. All cells were grown in a humidified incubator at 37 °C with 5% CO2 and authenticated by STR analysis. Cells were maintained in culture for no more than 20 passages, excluding passaging prior to receipt in our lab. All cell lines were routinely tested negative for mycoplasma contamination.

Human melanoma samples were obtained with informed consent, and the use of clinical samples was approved by the Medical Ethics Committee of the Beijing Cancer Hospital & Institute. All samples were diagnosed via histopathological examination, including haematoxylin and eosin (H&E) staining and immunohistochemistry (IHC) for melanoma markers (S-100, MART-1 and HMB-45).

Transwell chambers with 8 μm pores in 6-well plates (Corning, USA) were used to obtain highly invasive HMY-1 cell sublines. First, 7 × 10⁵ HMY-1 cells in 1 mL serum-free DMEM were seeded into the upper chamber, which was precoated with Matrigel (BD Biosciences, USA), and 2 mL of DMEM supplemented with 25% foetal bovine serum was placed into the lower well. After incubation for 48 h, the invasive cells on the bottom of the chambers were harvested. Then, the harvested invasive cells were cultured and filtered for another 6 rounds using the Transwell chambers. After 7 continuous invasion assays, a highly invasive HMY-1 subline derived from the HMY-1 cell line was obtained. The invasive ability of these two cell lines was then confirmed by Transwell assays and lung metastasis models.

Five hundred highly invasive HMY-1 cells were evenly seeded onto a 10-cm dish. When a single cell grew to a visible cell cluster, a sterilized clone ring was used to get the clones. Different clones were separately cultured in a 24-well plate and passed on to 6-well plate when the plate was full.

Total RNA was extracted from FFPE samples and cell lines using an RNeasy FFPE Kit (Qiagen, Germany) and TRIzol reagent (Life Technologies, USA), respectively. Then, cDNA was synthesized using a PrimeScript RT reagent kit (Takara, Japan) according to the manufacturer’s instructions. Quantitative reverse transcription polymerase chain reaction (qRT–PCR) was performed using SYBR Green Real-time PCR Master Mix (TOYOBO, Japan). The sequence for each primer is listed in Table S2.

For circRNA sequencing, total RNA was isolated from HMY-1 cells and highly invasive HMY-1 cells using TRIzol reagent (Life Technologies, USA) and purified with an RNease Mini kit (Qiagen, Germany). Then, circRNAs were sequenced using a human ceRNA Array V3.0 (4x180K, designed by Shanghai Biotechnology Corporation, and made by Agilent Technologies), which contains 105,509 circRNA probes.

Actinomycin D and RNase R treatment experiments were performed to assess the stability of circZNF609 and ZNF609 mRNA. In brief, cells were treated with 2 mg/ml actinomycin D in a 6-well plate. After incubation for several hours, the cells were collected. For RNase R treatment, total RNA (2 μg) was incubated for 15 min at 37 °C with 3 U/μg RNase R (Epicentre, WI, USA). After treatment with actinomycin D or RNase R, the circZNF609, ZNF609 and RAC1 mRNA expression was measured via qRT–PCR.

siRNA duplexes for knockdown of circZNF609, ZNF609 and both were synthesized by GenePharma (Suzhou, China). Cells were transfected with the siRNAs using Lipofectamine RNAiMAX (Invitrogen, USA) according to the manufacturer’s instructions. The siRNA sequences are listed in Table S2.

The circZNF609 overexpression plasmid and vector plasmid were obtained from Geneseed (Guangzhou, China). 293 T cells were transfected with the plasmids to package the lentivirus using Lipofectamine 3000 (Invitrogen, USA) according to the manufacturer’s instructions. Human melanoma cells were infected with the packaged lentivirus and selected with 0.5–2 μg/ml puromycin for 4 days. Surviving melanoma cells exhibiting GFP expression were then selected for subsequent culture.

The nuclear and cytoplasmic fractions were extracted using a Minute™ Cytoplasmic and Nuclear Extraction Kit (Invent Biotechnologies, USA) according to the manufacturer’s instructions. In brief, cells were washed with cold PBS for 2 min. Then, cell lysis buffer was added to the cells on ice. After incubation for 5 min, the cell lysate was transferred to a prechilled 1.5 ml microcentrifuge tube. The tube was vigorously vortexed for 15 s. After centrifugation of the tube for 5 min at top speed at 4 °C, the supernatant was collected as the cytosolic fraction. Then, an appropriate amount of nuclear extraction buffer was added to the pellet, and the tube was vigorously vortexed for 15 s and incubated on ice for 1 min. The 15 s of vortexing and 1 min incubation were repeated 4 times. After that, the nuclear fraction was transferred to a prechilled filter cartridge with a collection tube and centrifuged at top speed in a microcentrifuge for 30 s.

A Cy3-labelled specific probe for circZNF609 was designed and synthesized by RiboBio. The FISH experiment was performed using a FISH kit (RiboBio, China) according to the manufacturer’s instructions.

Human melanoma cells were fixed with 4% paraformaldehyde for 15 min and then blocked with 5% BSA for 30 min. The melanoma cells were incubated with primary antibodies at 4 °C overnight and fluorescence-conjugated secondary antibodies at room temperature for 1 h. After washing with PBS, the cells were counterstained with DAPI for 10 min and imaged under a microscope (Olympus Corp).

Wound healing assays were performed using Culture–Insert (Ibidi, Germany) according to the manufacturer’s instructions. In brief, 1 × 10⁵ human melanoma cells in 70 μl complete medium were seeded into each culture well and incubated for 24 h, and then, the culture insert was gently removed using sterile tweezers. After washing with PBS, the cells were cultured in culture medium with 1% FBS. Images were taken at several time points.

Human melanoma cells were cultured in serum-free culture medium 24 h before performing Transwell assays. Cell culture medium supplemented with 25% foetal bovine serum was added to the bottom chambers of a 24-well plate. For the invasion and migration assays, 1 × 10⁵ melanoma cells in serum-free medium were seeded into the top chamber with or without Matrigel (BD Biosciences, USA), respectively. After incubation for several hours, the Transwell inserts were fixed in 4% paraformaldehyde for 10 min and then stained with 1% crystal violet for 10 min. Cells that did not migrate through the 8 μm pores of the Transwell chambers were removed using cotton swabs. Migratory or invasive melanoma cells located on the bottom of the chamber were counted using an inverted phase-contrast microscope.

3D spheroid-based Matrigel invasion assays were performed as previously described [25‐27]. In brief, 1 × 10⁴ LM-MEL-45 cells in 200 μl complete culture medium were grown in ultralow attachment (ULA) 96-well round-bottom plates (Corning, USA) for 3 days. After tumor spheroids formed, 100 μl of growth medium was gently removed from the spheroid plates. Then, 100 μl Matrigel (BD Biosciences, USA) was gently dispensed into the U-bottom well. After incubation in an incubator at 37 °C for 1 h, 100 μl complete growth medium was gently added to each well. Three days later, images of invasive tumor spheroids were taken with an inverted phase-contrast microscope.

All animal experiments were approved by the Medical Ethics Committee of the Beijing Cancer Hospital & Institute. For in vivo cutaneous melanoma lung metastasis assays, 4-week-old female BALB/c nude mice were intravenously injected with 5 × 10⁵ A2058 cells in 100 μl PBS via the tail vein. The mice were sacrificed when they became weak, hunched and emaciated. For in vivo acral melanoma lung metastasis assays, 5-week-old female NOD/SCID mice were intravenously injected with 2 × 10⁶ HMY-1 cells in 200 μl PBS via the tail vein. The mice were sacrificed when they became weak, hunched and emaciated. For in vivo acral melanoma liver metastasis assays, 1 × 10⁶ HMY-1 cells suspended in 50 μl PBS were injected into the inferior hemispleen of 4-week-old female NOG mice. The mice were euthanized after 4 weeks. The lungs and livers were excised, and then, the number of metastatic tumor nodules in the lungs and livers was carefully counted. Subsequently, all the samples were fixed with 4% paraformaldehyde and analysed via H&E staining. To detect tumors in live animals, mice were injected with 100 μl D-luciferin (15 mg/ml) and then anaesthetized with isoflurane. Ten minutes later, bioluminescence images were acquired and analysed using an IVIS Spectrum In Vivo Imager.

The biotin-labelled circZNF609 probe was synthesized by RiboBio, and the RNA pull-down assay was performed using a Pierce™ Magnetic RNA-Protein Pull-Down Kit (Thermo Fisher, USA) according to the manufacturer’s instructions. First, the biotin-labelled circZNF609 probe or control probe was incubated with streptavidin magnetic beads at room temperature for 30 min. Then, the melanoma cell lysates were incubated with probe-bead complexes at 4 °C for 1 h to allow binding of RBPs to RNAs. Subsequently, the RNA-protein complexes were washed and eluted from beads via incubation for 30 min at 37 °C with agitation. The eluted proteins were finally analysed via silver staining and Western blotting. The circZNF609 probe and control probe sequences used are listed in Table S2.

Silver staining was performed using a Fast Silver Stain Kit (Beyotime, China) according to the manufacturer’s instructions.

RIP assays were performed using a Magna RIP RNA-Binding Protein Immunoprecipitation Kit (Millipore) as described in the protocol. Briefly, human melanoma cells were lysed in RIP lysis buffer on ice for 5 min. After washing, 50 μl magnetic beads were incubated with 5 μg antibodies with rotation for 30 min at room temperature. After incubation, the melanoma cell lysates were incubated with bead-antibody complexes at 4 °C overnight. Subsequently, the immune complexes were washed six times with washing buffer, and the remaining RNA was purified and extracted for later qRT–PCR analysis and RIP sequencing.

All experiments were carried out at least three times, and the results are presented as the mean ± standard deviation (S.D.). Statistical analysis was performed using Student’s two-tailed t test or one-way analysis of variance (ANOVA). The expression correlations in melanoma samples were assessed using Pearson correlation analysis. Survival curves were assessed using the Kaplan–Meier method and compared with log-rank tests. Statistical significance was defined as a P value of less than 0.05 (*P < 0.05, **P < 0.01, ***P < 0.001); ns: not significant (P > 0.05).

To investigate the key driver molecule in the metastasis of melanoma, we first applied cell invasion models to mimic the invasion process in vitro (Fig. 1a). After 7 continuous selections, a highly invasive HMY-1 subline was established, and metastatic ability was further verified by in vitro Transwell assays (Fig. S1a-c) and in vivo lung metastasis experiments (Fig. 1c). Subsequently, we conducted circRNA transcriptome analysis of HMY-1 cells and the highly invasive HMY-1 cells using a human ceRNA array. A total of 13,378 dysregulated circRNAs were identified in these cells harbouring different metastatic abilities. Considering that circRNAs are tissue-specific and cell-specific and that most circRNAs may result from splicing errors [8, 28], we conducted further screening to select high-abundance circRNAs that truly participate in the progression of melanoma (Fig. 1c). Among these circRNAs, we identified 4 dysregulated circRNAs that met the following requirements (Fig. S2d, Table S2): (1) fold change > 1.3 or < 0.7 in the array; (2) detected in 3 cutaneous melanoma cell lines and 4 normal skin tissues [29] (Fig. S2a-c); and (3) circRNAs originating from highly expressed genes with exceptionally high back-splicing rates [28]. Next, we analysed the expression of 4 circular RNAs in 4 melanocytes and 10 cutaneous melanoma short-term cultures from GSE138711 [30]. We found that only circZNF609 was downregulated in cutaneous melanoma, which was consistent with the trend in acral melanoma (Fig. 1d). Therefore, these findings led us to study circZNF609 in acral melanoma and cutaneous melanoma.

According to the CircBase database [31], circZNF609 is an exonic circRNA that is formed by exon 2 of the ZNF609 gene. Then, we verified the existence and circular characteristics of circZNF609 through several experiments. The back-spliced junction of circZNF609 was amplified using divergent primers and confirmed by Sanger sequencing (Fig. 1e). Since circRNAs are single-stranded circular transcripts without a 3′ poly(A) tail and oligo dT can only amplify linear transcripts containing a poly(A) tail, circZNF609 could not be amplified when only oligo dT and divergent primers were used (Fig. 1f). We also observed that circZNF609 had a longer half-life than ZNF609 mRNA (Fig. 1g) and was resistant to digestion with RNase R (Fig. 1h), further demonstrating its circular characteristic. In addition, using a nucleoplasmic separation assay and FISH, we confirmed that circZNF609 was predominantly localized in the cytoplasm of melanoma cells (Fig. 1i, j). Taken together, these results indicate that circZNF609 is an abundant, stable, cytoplasmic and exonic circRNA in human melanoma cells.

Next, we examined the expression of circZNF609 in human melanoma cell lines. qRT–PCR revealed that circZNF609 was downregulated in 12 cutaneous melanoma cell lines and 3 acral melanoma cell lines compared with the normal keratinocyte cell line HaCaT (Fig. 2a). Besides, circZNF609 was downregulated in highly invasive HMY-1, which was consistent with the results of circRNA sequencing (Fig. S1d). Moreover, we found that circZNF609 was also downregulated in metastatic tissue than that in paired primary melanoma tissues (Fig. 2b). We further investigated the clinical significance of circZNF609 in cutaneous melanoma patients from our hospital. In a cohort of 37 FFPE primary cutaneous melanoma samples, we observed progressive loss of circZNF609 expression (Fig. 2b). Cutaneous melanoma patients with distant metastasis and deeper Breslow depth had lower circZNF609 expression (Fig. 2c-e). Intriguingly, we also found that the abundance of circZNF609 was significantly negatively correlated with the depth of melanoma invasion (Fig. 2f). Meanwhile, a low abundance of circZNF609 was also associated with shorter overall survival (OS) (Fig. 2g). Furthermore, correlation analysis demonstrated that low expression of circZNF609 was positively correlated with aggressive clinicopathological.

characteristics, especially distant metastasis and Breslow depth (Table 1). Collectively, these data show that circZNF609 was downregulated in cutaneous melanoma and acral melanoma, and its abundance was associated with cutaneous melanoma progression and patient prognosis, indicating that circZNF609 expression might have prognostic value for melanoma patients.

We verified that circZNF609 was downregulated in the highly invasive HMY-1 subline and that its abundance was negatively correlated with the depth of melanoma invasion and distant metastasis state. These results suggest that circZNF609 loss might contribute to melanoma progression, especially melanoma metastasis. To test whether circZNF609 can affect melanoma metastasis, loss-of-function and gain-of-function assays were performed. We first designed 3 siRNAs: one targeting the back-splice region of circZNF609, another targeting a sequence only in ZNF609 mRNA and a third targeting a sequence present in both the linear and circular transcripts (Fig. 3a). After transfection of LM-MEL-45 cells with the siRNAs for 48 h, the cells were harvested, and the expression of circZNF609 and ZNF609 was analysed via qRT–PCR (Fig. 3b). As expected, siRNA directed against the back-splice region only knocked down circZNF609 and did not affect the expression of ZNF609 mRNA (Fig. S3a-c). siRNA targeting sequences present in both the linear and circular transcripts effectively knocked down both transcripts. Simultaneously, circZNF609-overexpressing cells were successfully constructed, and no significant change in ZNF609 mRNA was observed (Fig. 3b, Fig. S3a-c).

To investigate the impact of circZNF609 on the metastatic phenotypes of melanoma cells, we selected 2 cutaneous melanoma cell lines (A2058 and A875) and 2 acral melanoma cell lines (HMY-1 and LM-MEL-45); the A875 and LM-MEL-45 lines retained high basal circZNF609 expression, and the A2058 and HMY-1 lines exhibited low basal circZNF609 expression. Transwell assays and wound healing assays showed that the migration and invasion capacities of the 2 cutaneous melanoma cell lines and 2 acral melanoma cell lines were significantly enhanced in response to circZNF609 silencing (Fig. 3c, d, Fig. S4c). Similarly, ectopic expression of circZNF609 in the 4 melanoma cell lines led to a significant decrease in cell migration and invasion abilities (Fig. 3d, Fig. S4a, b). Moreover, in 3D invasion assays, we observed that circZNF609 loss in LM-MEL-45 cells resulted in an enhanced invasion area and an increased number of protrusions (Fig. 3e). Conversely, stable circZNF609 overexpression remarkably reduced the invasion area and the number of protrusions (Fig. 3e). Taken together, these findings suggest that circZNF609 modulation in cell models altered cutaneous melanoma and acral melanoma migration and invasion in vitro.

To identify the effect of circZNF609 on melanoma metastasis in vivo, we established lung and liver metastasis models by implanting melanoma cells transfected with circZNF609 or vector. Firefly luciferase-expressing A2058 cells transfected with circZNF609 or vector were inoculated into female nude mice via tail vein injection. The mice were further monitored and analysed for lung metastasis using in vivo luciferase imaging under anesthesia. We observed that after ectopic expression of circZNF609 in cutaneous melanoma A2058 cells, lung metastasis was significantly inhibited (Fig. 4a, c, d) and OS was prolonged compared to that of control cells (Fig. 4b). Additionally, the role of circZNF609 in inhibiting lung metastasis of acral melanoma was also confirmed in the acral melanoma cell line HMY-1. The results of the acral melanoma lung metastasis assay revealed that the number of lung nodules in the circZNF609 overexpression group was less than that in the control group (Fig. 4e-g). To explore the effect of circZNF609 on liver metastasis, we further injected firefly luciferase-expressing HMY-1 cells into the inferior hemispleen of female NOG mice. As expected, the luminescence in livers was weaker and there were fewer liver metastasis nodules in the circZNF609 overexpression group than in the corresponding control group (Fig. 4h-j), demonstrating that ectopic expression of circZNF609 inhibited liver metastasis of acral melanoma. Taken together, these findings suggest that circZNF609 inhibited the metastasis of cutaneous melanoma and acral melanoma in vivo.

Next, we sought to investigate the potential molecular mechanisms of circZNF609 as a metastasis suppressor in melanoma. We first performed bioinformatic analysis to screen circZNF609-interacting proteins (Table S4). Intriguingly, we obtained only 1 candidate circZNF609-interacting protein, FMRP, through overlapping the prediction results of RBP recognition elements in the circZNF609 sequence in starBase [32], CircRic [33], CircInteractome [34] and circAtlas [35] (Fig. 5a). FMRP is an RNA binding protein that regulates a large repertoire of target mRNAs through diverse mechanisms. It has been reported that FMRP levels in breast primary tumors influence metastasis formation [20]. A possible explanation for this finding is that FMRP binds mRNAs involved in epithelial mesenchymal transition (EMT), cytoskeleton remodelling and cell adhesion and regulates their mRNA stability or protein translation. Thus, we hypothesized that circZNF609 might interact with FMRP and modulate its function. By performing IF and FISH assays, we observed colocalization of endogenously expressed circZNF609 and FMRP protein in the cytoplasm of cutaneous melanoma and acral melanoma cells (Fig. 5b, c), suggesting that circZNF609 might associate with FMRP to perform its function. Additionally, we performed RNA pull-down assays, silver staining and western blot analysis in HMY-1 acral melanoma cells (Fig. 5d, e) and A875 cutaneous melanoma cells (Fig. 5g, h), and the results revealed that FMRP is a putative circZNF609-binding protein. Moreover, we observed significant enrichment of circZNF609 in FMRP RIP experiments (Fig. 5f, i). Taken together, these data indicate that FMRP interacts with circZNF609 in cutaneous melanoma and acral melanoma.

Because FMRP interacted with circZNF609 in melanoma, we wondered whether circZNF609 modulation regulated FMRP function. Strikingly, silencing of circZNF609 or ectopic expression of circZNF609 in melanoma cells did not alter the expression or cellular distribution of FMRP (Fig. S5a, b). Previous studies have revealed that FMRP plays a critical role in determining the fate of mRNAs in a tissue- and disease-dependent manner, especially in protein translation. In neurons, FMRP is thought to be a translation repressor by integrating with both polysomes and messenger ribonucleoprotein particles (mRNPs) [17]. Although previous studies have mainly focused on the translational regulation of FMRP, there have also been reports showing that FMRP functions as a direct modulator of mRNA stability [14, 36]. Therefore, we performed RNA-seq analyses using circZNF609-depleted cells and control cells and RIP-seq analyses using LM-MEL-45 FMRP-IP RNA and IgG-IP RNA (Fig. 6a). Among these dysregulated mRNAs, we identified RAC1 as the only dysregulated mRNA that met the following strict requirements (Table S5): (1) normalized fold change ≥1.5 in HMY-1 RNA-seq; (2) normalized fold change ≥1.5 in A875 RNA-seq; (3) normalized fold change ≥1.5 in LM-MEL-45 RIP-seq; (4) FMRP mRNA target predicted by the starBase database [32]; (5) FMRP mRNA target predicted by the POSTAR database [37]; and (6) FMRP experimental mRNA target according to a previous study [38]. Then, we performed RIP experiments in acral melanoma and cutaneous melanoma cells. As shown in Fig. 6b and c, RIP assays showed that RAC1 mRNA was enriched in FMRP-IP groups in acral melanoma and cutaneous melanoma cells compared with IgG-IP groups. To further investigate this interaction, we performed FISH and IF in melanoma cells (Fig. 6d, e). Fluorescence results further confirmed the colocalization of FMRP protein and RAC1 mRNA in the cytoplasm of acral melanoma and cutaneous melanoma cells.

Next, we sought to examine the effects of FMRP depletion on RAC1 expression. Strikingly, although previous studies have revealed that FMRP might participate in RAC1 translation in mouse hippocampal neurons and Drosophila [23, 24], our results confirmed that FMRP loss in acral melanoma and cutaneous melanoma could lead to increased RAC1 expression at the mRNA and protein levels (Fig. 6f, g). Moreover, we extracted the transcriptome data of 34 patients with acral melanoma from Liang’s cohort through the cBioPortal database [39, 40] and found a negative correlation between FMRP mRNA and RAC1 mRNA (Fig. 6h), further confirming that FMRP is involved in regulation of RAC1 nontranslational pathways in melanoma. Given that FMRP is known to destabilize mRNA stability, we then assessed RAC1 mRNA stability following FMRP silencing. Using actinomycin D to block de novo transcription, we found that silencing FMRP increased RAC1 mRNA stability (Fig. 6i). Additionally, circZNF609 deficiency or overexpression in acral melanoma (Fig. 6l) and cutaneous melanoma (Fig. 6m) altered RAC1 expression at both the mRNA and protein levels, which was consistent with the RNA-seq results. In line with this finding, we found that circZNF609 deficiency also increased RAC1 mRNA stability (Fig. 6j). We then examined the effects of silencing circZNF609 on the interaction of FMRP and RAC1. RIP analysis indicated that silencing circZNF609 reduced the association of FMRP with circZNF609 but also decreased the binding of FMRP to RAC1 mRNA (Fig. 6k). Collectively, these data demonstrate that circZNF609 affects the abundance of RAC1 by participating in the regulation of RAC1 mRNA stability by FMRP.

RAC1 is known to be pro-tumorigenic in numerous cancers [41] (Fig. S6a, b). Our data verified that circZNF609 controls RAC1 mRNA stability by binding with FMRP, thereby regulating the metastasis of melanoma. Therefore, we sought further validation that the circZNF609-FMRP-RAC1 axis governs the metastasis process in acral melanoma and cutaneous melanoma. To this end, we first evaluated the expression levels of RAC1 in melanoma cohorts. From Liang’s cohort in the cBioPortal database, we found that high expression of RAC1 suggested worse OS and disease-free survival (DFS) in acral melanoma patients (Fig. 7a, b). Notably, the high expression levels of RAC1 also predicted worse OS in our acral melanoma cohort (Fig. 7c). Moreover, we found that the expression of RAC1 mRNA was positively correlated with the copy number values of skin cutaneous melanoma patients from TCGA database (Fig. 7d, e), and the group with the highest RAC1 expression, named the RAC1 amplification group, had significantly worse OS and DFS than the nonamplification group (Fig. 7f, g). We then investigated whether the role of circZNF609 in the metastasis of melanomas was dependent on RAC1 modulation. In vitro Transwell and 3D invasion assays demonstrated that silencing RAC1 functionally abolished the induction of melanoma migration and invasion elicited by circZNF609 depletion (Fig. 7h-j). In addition, a lung metastasis model showed that the metastatic nodules in the lungs formed after injection with A2058 cells was reduced by silencing RAC1 (Fig. 7k).