Epigenetic silencing of CDR1as drives IGF2BP3-mediated melanoma invasion and metastasis

Summary

Metastasis is the primary cause of death of cancer patients. Dissecting mechanisms governing metastatic spread may uncover important tumor biology and/or yield promising therapeutic insights. Here we investigated the role of circular RNAs (circRNA) in metastasis, using melanoma as a model aggressive tumor. We identified silencing of Cerebellar Degeneration-Related 1 Antisense (CDR1as), a regulator of miR-7, as a hallmark of melanoma progression. CDR1as depletion results from epigenetic silencing of LINC00632, its originating long non-coding RNA (lncRNA) and promotes invasion in vitro and metastasis in vivo through a miR-7-independent, IGF2BP3-mediated mechanism. Moreover, CDR1as levels reflect cellular states associated with distinct therapeutic responses. Our study reveals functional, prognostic and predictive roles for CDR1as and expose circRNAs as key players in metastasis.

Graphical Abstract

In Brief

Investigating the role of circular RNAs (circRNA) in metastasis, Hanniford et al. observe epigenetic silencing of circRNA CDR1as in melanoma progression. Depletion of CDR1as promotes melanoma metastasis via an IGF2BP3-mediated mechanism that is independent of its well-characterized inhibition of miR-7.

Introduction

For most cancer types, metastatic spread is the pivotal determinant of poor outcomes for patients. Roughly 90% of cancer deaths are attributable to metastasis(Lambert et al., 2017). Uncovering the mechanisms responsible for metastatic spread may reveal previously unappreciated vulnerabilities of cancer cells. Melanoma is an excellent model to investigate the mechanistic bases of metastasis. Reflective of a highly aggressive nature, melanoma metastasis frequently occurs from primary tumors that are measured in millimeters. The cellular and molecular mechanisms responsible for this phenotype are incompletely understood.

At its core, tumorigenesis is a genetic disease, but mutations alone insufficiently explain metastasis(Vogelstein et al., 2013). In recent years, the expansion of knowledge of non-coding RNA (ncRNA) biology has revealed critical roles in tumorigenic processes(Arun et al., 2016; Gupta et al., 2010; He et al., 2005; Klein et al., 2010). Circular RNAs (circRNAs) are an emerging class of primarily ncRNAs, which arise by cis back-splicing and circularization of portions of protein-coding genes and other transcripts. During nascent RNA processing, a downstream splice donor is covalently ligated to an upstream splice acceptor, yielding a circularized ssRNA product (Salzman, 2016). Backsplicing is facilitated by pairing of repetitive sequences, such as Alu elements, flanking circularized regions (Ashwal-Fluss et al., 2014; Jeck et al., 2013; Liang and Wilusz, 2014), or by RNA binding proteins (RBPs), such as Quaking (QKI)(Conn et al., 2015), Muscleblind (MBNL1)(Ashwal-Fluss et al., 2014), and FUS(Errichelli et al., 2017). Functionally, although recent studies have reported translation of some circRNA species(Legnini et al., 2017; Pamudurti et al., 2017), most are exported and accumulate in the cytosol as untranslated ncRNAs(Guo et al., 2014; Huang et al., 2018a). Few circRNAs have well-characterized functions and their contributions to physiology or pathology are only starting to be elucidated (Guarnerio et al., 2016; Holdt et al., 2016; Zheng et al., 2016). Moreover, circRNAs have not been investigated in the context of melanoma biology or metastasis. Here, we characterize circRNA expression in melanoma and the melanocytic lineage and investigate the contribution of the circRNA CDR1as to metastatic progression of melanoma cells.

Results

circRNA profiling reveals CDR1as as a candidate melanoma suppressor

Using a previously developed splice junction detection algorithm (Mapsplice) (Wang et al., 2010), we analyzed stranded, rRNA-depleted RNA-seq data of a set of melanocytes (n = 4) and melanoma short-term cultures (STCs) (n = 10) (Figure S1A), which better reflect the intratumoral heterogeneity of human tumors than established cell lines(de Miera et al., 2012). Across all samples, Mapsplice annotated 13,725 circRNAs with at least 2 unique backsplice reads, associated to 4204 genes (Table S1). 65% were detected in one or two samples, while 35% were detected in three or more (Figure 1A). Of genes with circRNAs detected in only 1 or 2 samples, ~47% also yield more consistently detected circRNAs (Figure S1B), thus many detected circRNA may be rare species from genes that commonly yield circRNAs. We used a second algorithm (DCC, (Cheng et al., 2016)) to further test the reliability of our initial annotation. Restricting for circRNA with >1 read in at least 2 samples by both algorithms yielded 96.8% overlap of Mapsplice circRNA annotations with those of DCC (6849 of 7074) (Figure S1C and Table S1). Validating in silico predictions, we amplified putative backsplice junctions of 10 circRNA candidates by divergent primer PCR and Sanger sequencing (Figure S1D and data not shown). Comparing STCs with melanocytes, circRNA expression can robustly segregate tumor from normal in multidimensional scaling (MDS) analysis (Figure 1B). Moreover, differential expression (DE) analysis (logFC >1, FDR <0.05) revealed 116 and 572 up- and down-regulated circRNA, respectively, in melanoma STCs vs. melanocytes (Figure 1C and Table S1). Top upregulated candidates include circRNAs arising from LIFR, ZC3H13, and ARID1A, while top downregulated circRNAs arise from SIPA1L1, WDR37, and PAG1. In contrast to a recent study in prostate cancer(Chen et al., 2019), we observed that the vast majority of DE circRNA correlate strongly with their linear counterpart (Pearson, r > 0.4, Figure 1D and Table S1), suggesting change in host gene transcription is the underlying mechanism of altered expression for most circRNA in melanoma. Of DE circRNAs that correlate poorly with their linear counterpart, those arising from CORO1C, HIBADH, UBXN7, NDUFB2, and SMARCA5 are downregulated, while those from TMEM161B, GOLGA4, ERBB2IP, and NIPBL are upregulated. To further examine the annotated circRNAs, we sought to identify highly circularized transcripts, which we expect are more likely to harbor important biological functions. We calculated a circular-to-linear ratio by dividing backsplice-spanning read counts with the average total reads aligned to the 5’ and 3’ junctions of the backsplice. We examined the well-described circRNA, CDR1as (also known as CIRS-7), which was previously thought to be exclusively circular(Hansen et al., 2013; Memczak et al., 2013; Piwecka et al., 2017), as a control in this analysis. Surprisingly, CDR1as had a lower-than-expected circularization ratio in this analysis (Figure 1E). We also observed striking loss of CDR1as in the majority of STCs, with 6 of 10 lacking any detectable backsplice-spanning reads, while 2 additional had reductions compared with melanocyte cultures (Figure 1F). Moreover, we noted that the 4 STCs harboring backsplice spanning reads for CDR1as were derived from brain metastases; thus, in these early-passage sequencing studies, some or all of the detected CDR1as in these cultures could represent contamination of residual brain tissue, where CDR1as is highly enriched(Hansen et al., 2013; Memczak et al., 2013). In normal brain, CDR1as is thought to predominantly regulate a single microRNA, miR-7(Hansen et al., 2013; Hansen et al., 2011; Memczak et al., 2013). In cancer, miR-7 is typically considered a tumor suppressive microRNA, known to regulate pro-tumorigenic genes in the MAPK and PI3K pathways, such as EGFR, IGF1R, PIK3CD (Giles et al., 2013; Giles et al., 2016; Hanniford et al., 2015; Kefas et al., 2008; Kong et al., 2012; Webster et al., 2009; Zhang et al., 2013). It was thus surprising to observe loss of CDR1as in melanoma cultures. Knowledge of CDR1as’ function(s) in other lineages and in pathology is less clear, nor has it been studied in the context of melanoma biology. CDR1as’ role in neuronal lineages, which share a developmental origin with melanocytes, and its function regulating a microRNA previously linked to tumorigenesis and metastasis, including in melanoma(Giles et al., 2013; Giles et al., 2016; Hanniford et al., 2015), led us to study CDR1as biogenesis, expression, regulation and function in melanoma.

Figure 1. circRNA profiling of melanocytes and melanoma STCs reveals loss of CDR1as in melanoma.

(A) Pie chart of the fraction of circRNAs detected in the indicated number of melanocyte or melanoma STC samples. (B) MDS plot of melanocytes (n=4) and STCs (n=10) defined by circRNA expression. (C) Hierarchically-clustered heat map of circRNAs differentially expressed in melanoma STCs (n=10) vs. melanocytes (n=4) (EdgeR, logFC >1, FDR <0.05). (D) Average correlation coefficients (across samples, Pearson, r) between circular and linear RNAs, grouped by the number of samples in which the circRNA was detected. Mean ± SD per sample group are presented. (E) Cumulative distribution plot of the average circular/linear ratios across samples. (F) Backsplice read counts for CDR1as in melanocytes and individual melanoma STCs. Count or mean count ± SD. (G) FPKM or mean FPKM ± SD of CDR1as from cultured melanocytes (n = 6), and melanoma STCs (n = 13) and cell lines (n = 7). ND = not detected. Red hashed line set to mean FPKM of melanocyte samples. Color coding of cell lines indicating high (red circle), moderate (yellow circle) or low/absent (blue circle) CDR1as expression in subsequently used cell models. (H) Median normalized CDR1as expression (RT-qPCR) in cultured melanocytes (n = 3) and melanoma cell lines (n = 1 per cell line). I, Representative images of single molecule in situ hybridization (smISH) for CDR1as or PPIB (control) in high- and low-expressing melanoma cells. Scale bar is 100 μm. (J) Median normalized expression (log2) of CDR1as by RT-qPCR of melanoma patient samples, grouped by primary (n = 53) and metastatic (n = 52) samples. Primary samples grouped by Breslow thickness (<1 mm or >1 mm). Metastatic samples grouped as in-transit (ITM), lymph node (LN), or distal metastases (DM). (K-M) Median normalized CDR1as expression (RT-qPCR, log2) of primary vs. metastatic samples (K), stage I vs. II samples (L), and ulceration status (primary only) (M). Box plots depict median, 25^th and 75^th percentile, min-max whiskers. Statistical analyses by two-tailed student’s t test. * p<0.05, ** p<0.01, *** p<0.001 (N-P) Dot plots of median normalized CDR1as expression (log2) from primary melanoma patient samples with Breslow thickness (N), mitoses per mm² (O), and age at diagnosis (P). Pearson correlation coefficients (r) and p values are indicated. (Q and R) Kaplan-Meier curves of primary melanoma patients stratified by CDR1as expression (Bottom 25% (n = 13) vs. Top 75% (n = 40)) for metastasis-free survival (Q) or overall melanoma-specific survival (MSS) (R). Statistical analyses by Log Rank test.

See also Figure S1 and Tables S1 and S2.

CDR1as arises from upstream LINC00632

CDR1as is an atypical circRNA that arises antisense to a putative protein coding gene. With no known linear counterpart from the same strand, initial investigations of CDR1as did not characterize its biogenesis or source transcript, and proposed it to be an exclusively circular molecule(Hansen et al., 2013; Hansen et al., 2011; Memczak et al., 2013). A recent study documented CDR1as as embedded in an upstream lncRNA, LINC00632, and provided evidence of linear splicing products containing CDR1as (Barrett et al., 2017). We sought to elucidate the transcriptional origin of CDR1as in our melanoma cell models, which independently confirm these observations. LINC00632 harbors multiple isoforms and lies ~10 Kb upstream of CDR1as (Figure S1E). We observed correlations of CDR1as with LINC00632 in melanoma cell lines (r = 0.86, p<0.0001) and patient samples (TCGA, r = 0.61, p<0.0001)(Cancer Genome Atlas, 2015), in several other cancer types (TCGA)(Cerami et al., 2012; Gao et al., 2013), human tissues (GTEx), and in mouse neural crest-derived tissues(Minoux et al., 2017), the precursor lineage from which melanocytes arise (Figure S1F and Table S2). Moreover, depletion of LINC00632 in melanoma cells reduced expression of CDR1as (Figure S1G). Since many lncRNA control expression of neighboring genes in cis (Engreitz et al., 2016), we examined if CDR1as arises directly from LINC00632. Inspection of RNA-seq data revealed rare paired-end reads mapping to both LINC00632 and CDR1as (data not shown), indicative of linear splicing connecting these transcripts. Validating this observation, we PCR-amplified and sequenced spliced transcripts containing LINC00632 (ENST00000498732 and ENST00000602535) and CDR1as (Figures S1H-J), providing direct evidence that CDR1as arises from LINC00632. Additionally, disruption of CDR1as circularization using CRISPR-Cas9 targeting of the 3’ splice donor resulted in concomitant reduction of CDR1as with increased splicing between LINC00632 and CDR1as, denoted as LINC-CDR1as (Figure S1K). Finally, to further demonstrate that CDR1as can arise from multiple LINC00632 isoforms in our cells, we used SAM CRISPR/Cas9 activation (Konermann et al., 2015) to induce expression from the two annotated LINC00632 transcriptional start sites (TSS). SgRNAs to these TSSs induced expression of the targeted LINC00632 isoforms, yielding robust CDR1as expression (Figure S1L). In sum, our data supports that CDR1as arises directly from LINC00632 transcripts, independently confirming previously reported findings(Barrett et al., 2017).

CDR1as expression is downregulated during melanoma progression

To expand on our initial findings, we calculated fragments per kilobase of transcript per million mapped reads (FPKMs) for the full CDR1as sequence to provide a more quantitative measure of its expression in melanoma samples. While we note considerable heterogeneity of CDR1as expression in different donor melanocyte cultures, we observed limited expression of CDR1as in the majority of STCs and melanoma cell lines (Figure 1G, color coding indicates high (red), moderate (yellow) and low/absent (blue) CDR1as expression in subsequently used cell models). Moreover, consistent with previous findings(Piwecka et al., 2017), the protein-coding gene CDR1 was undetectable in all analyzed samples (Figure S1M). Similarly, loss of CDR1as expression was observed in a panel of melanoma cell lines from publicly-available RNA-seq (Figure S1N)(Kaufman et al., 2016). We confirmed these results by RT-qPCR with divergent primers in additional melanoma cell panels (Figures 1H and S1O,P) and by single molecule RNA in situ hybridization (smISH) (Figures 1I and S1Q). Moreover, CDR1as was more abundant in primary tumor-derived WM115 cells than matched metastatic-derived cells WM239A/WM266-4, and was nearly undetectable in primary-derived WM983A cells and absent in matched metastatic-derived cells WM983B/C (Figures S1R,S). In addition to CDR1as (Figure S1T), we analyzed LINC00632 isoform expression for a subset of samples. Interestingly, we observed a switch of LINC00632 isoform expression in melanoma. While melanocytes predominantly express the longest LINC00632 isoform (ENST00000498732), melanoma cells typically only express an alternative LINC00632 isoform (ENST00000602535) from a different TSS (Figures S1U,V). Intriguingly, in these analyses, we observed that loss of LINC00632/CDR1as expression was more common in metastatic-derived compared with primary-derived melanoma cell lines, suggesting CDR1as loss may be involved in melanoma progression. Consistent with this possibility, in a cohort of fresh-frozen primary and metastatic melanoma patient samples (n = 53 and 52, respectively), we observed progressive loss of CDR1as expression (Figures 1J,K). Within the primary melanoma samples, CDR1as loss associated with prognostic clinical factors, including stage, ulceration, thickness, and mitotic index (Figures 1L-O), but not with age at diagnosis, sex, lymphocyte infiltration, or anatomic location (Figures 1P and S1W). Moreover, low abundance of CDR1as in these samples associated with shorter progression-free and overall survival (OS) (Figures 1Q,R). Collectively, our data show that CDR1as expression is reduced in melanoma, and its loss associates with melanoma progression and patient outcomes. These results suggest that CDR1as loss may contribute to melanoma progression, and its expression may have prognostic value for primary melanoma patients. Next, we examined the association of CDR1as expression with OS in other cancers. In low-grade glioma (LGG), but not other tumor types, low expression of CDR1as was also significantly associated with poor OS (Figures S1X,Y)(Anaya, 2016).

The LINC00632/CDR1as locus is epigenetically silenced by PRC2/H3K27me3 in melanoma

We sought to further define the molecular mechanisms driving CDR1as silencing in melanoma. Previous reports have shown miR-671-5p to be a negative regulator of CDR1as abundance, through AGO2-mediated cleavage and degradation(Hansen et al., 2011). Given the high level of correlation of CDR1as and LINC00632, which is not a known miR-671 target, miR-671-5p is unlikely to be responsible for CDR1as loss in melanoma. Consistently, we observed only a modest inverse correlation of miR-671-5p and CDR1as expression in melanoma patient samples, and no correlation in cell lines (Figures S2A,B). Moreover, while miR-671-5p overexpression depleted CDR1as in melanoma cells, LINC00632 expression was not affected (Figure S2C). Conversely, inhibition of miR-671-5p did not restore CDR1as expression in CDR1as-deficient cells (Figure S2D). These data show that miR-671-5p is not a primary mechanism of CDR1as silencing in melanoma.

Given the biogenesis of CDR1as, its silencing in melanoma may be dependent on disrupted LINC00632 expression. Analysis of publicly available data did not reveal genetic deletion(Cerami et al., 2012) of LINC00632/CDR1as, nor changes in locus-specific DNA methylation as common features of melanoma progression (Figures S2E-G). Previous evidence suggests LINC00632 expression may be dependent on chromatin state (Barrett et al., 2017), which is consistent with the dynamic regulation of LINC00632 isoforms (Figure S1U and S1V). We hypothesized that an epigenetic mechanism may dictate CDR1as expression in melanoma. In publicly-available ChIP-seq data(Verfaillie et al., 2015), H3K27 trimethylation (H3K27me3, a repressive histone mark) is abundant in distinct regions of the LINC00632/CDR1as locus in melanoma cells (Figure 2A), and the balance between H3K27me3 and H3K27 acetylation (H3K27ac, an activating histone mark) in these regions associates with LINC00632 and CDR1as expression (Figure S2H). EZH2, the enzymatic component of the PRC2 complex (Margueron and Reinberg, 2011), catalyzes H3K27me3 at target loci; it is mutated and reported to be pro-tumorigenic and pro-metastatic in melanoma (Barsotti et al., 2015; Fisher et al., 2016; Manning et al., 2015; Souroullas et al., 2016; Tiffen et al., 2015; Zingg et al., 2015). We tested if H3K27me3 is responsible for silencing the LINC00632/CDR1as locus in melanoma. Since the LINC00632-CDR1as locus is on the X chromosome, to eliminate confounding X-inactivation, we used cell lines derived from male patients for these experiments. By ChIP-PCR, we observed EZH2-dependent enrichment of H3K27me3 at two loci of LINC00632 in 501MEL cells (Figures 2B,C), which lack expression of LINC00632 and CDR1as. Moreover, treatment of CDR1as-deficient melanoma cells with EZH2 inhibitor GSK126 induced robust re-expression of all tested isoforms of LINC00632 (ENST00000498732, ENST00000602535, and ENST00000370535) and CDR1as (Figures 2D,E, and S2I,J). Collectively, these data demonstrate that PRC2-mediated H3K27me3 is a major mechanism of LINC00632/CDR1as silencing in melanoma cells.

Figure 2. LINC00632/CDR1as are epigenetically silenced by PRC2/H3K27me3 in melanoma.

(A) Density tracks of H3K27me3 from melanoma STCs/cell lines (Verfaillie et al., 2015) (GSE60666) in the LINC00632/CDR1as locus. (B) Representative immunoblot of H3K27me3 and total Histone H3 of EZH2 inhibitor-treated 501MEL cells. (C) Normalized abundance (%ChIP/Input, qPCR) of the indicated region from IgG, H3, or H3K27me3 ChIP samples isolated from DMSO- or EZH2 inhibitor-treated (GSK126, 2 mM for 10 days) 501MEL cells. Mean ± SD of a representative experiment. (D and E) Relative expression (RT-qPCR) of LINC00632 transcript variants (ENST00000498732, ENST00000602535, and ENST00000370535, left panel) or CDR1as (right panel) from 501 MEL (CDR1as^Lo) (D) or SK-MEL-28 (CDR1as^Lo) (E) cells treated with DMSO or GSK126 (2 μM) for the indicated number of days. Data normalized to DMSO-treated control (first time point). Mean ± SD of a representative time course (n ≥ 2 replicate experiments).

See also Figure S2.

CDR1as modulation alters melanoma cell invasion in vitro and metastasis in vivo.

To model CDR1as loss in melanoma, we generated doxycycline-inducible shRNAs, which efficiently deplete CDR1as, without reducing LINC00632 (Figure 3A,B). CDR1as depletion in slow growing, poorly invasive primary tumor-derived WM278 and WM115 cells, which retain high basal CDR1as expression, had no effect on cell growth, but significantly enhanced in vitro invasion (Figures 3C-G). Moreover, these shRNA had no effect on cell invasion in melanoma cells lacking LINC00632/CDR1as expression (Figures 3H,I), supporting that CDR1as depletion is required for this phenotype. Conversely, ectopic expression of CDR1as (without LINC00632), from a construct that uses circZK-SCAN1 flanking sequences to drive circularization (Kramer et al., 2015), suppressed the invasive capacity of metastatic-derived melanoma cell lines with moderate or absent basal CDR1as (Figures 3J,K, and S3A,B). To test the effect of CDR1as loss on tumor growth and metastasis in vivo, we selected 451Lu cells, which express moderate amounts of CDR1as and metastasize to lungs after implantation in immunocompromised mice(Hanniford et al., 2015; Juhasz et al., 1993). We implanted 451Lu cells expressing GFP, luciferase, and an inducible non-targeting (shNTC) or CDR1as-targeting shRNA (shCDR1as), subcutaneously in the flanks of immunocompromised mice (NOD/Shi-scid/IL-2Rγ^null). We allowed palpable tumors to develop to approximately 100 mm³ prior to initiating doxycycline-induced CDR1as depletion, which was subsequently assessed in explanted tumor tissue (Figure S3C). CDR1as silencing had no effect on tumor growth in the flank or tumor mass at resection (Figures 3L and S3D). Mice were further monitored for development of lung metastasis by IVIS imaging, and analyzed by ex vivo fluorescence imaging at euthanasia. We observed enhanced lung metastasis in mice bearing tumors depleted of CDR1as (Figures 3M,N, and S3E). Finally, gene sets associated to invasion and metastasis were enriched in CDR1as-depleted WM278 cells by gene set enrichment analysis (GSEA) of RNA-seq (Figure S3F). Collectively, these data demonstrate that CDR1as modulation in cell models can alter melanoma invasion in vitro and metastasis in vivo.

Figure 3. CDR1as depletion drives melanoma invasion and metastasis.

(A and B) Relative expression (RT-qPCR) of CDR1as and LINC00632 in WM278 (CDR1as^Hi) (A) or WM115 (CDR1as^Hi) (B) cells expressing doxycycline-inducible control (shNTC) or CDR1as-targeting (shA/B) shRNA. Cells harvested 48-72 h post-induction with doxycycline. Mean ± SD of a representative experiment. Statistical analyses by paired student’s t test of replicate experiments (n ≥ 3). (C and D) In vitro cell proliferation of doxycycline-treated WM278 (C) or WM115 (D) cells of (A and B). Cell abundance measured by absorbance at 595 nm from crystal violet staining. Data normalized to day 0. Mean ± SD of a representative experiment. Replicate experiments (n ≥ 2) were performed. (E and F) Normalized cell invasion of WM278 (E) or WM115 (F) cells expressing dox-inducible control (shNTC) or CDR1as-targeting (shA/B) inducible shRNA. (G) Representative fluorescence images of WM278 cell invasion of (E). Scale bars are 300 μm. (H and I) Normalized cell invasion of SK-MEL-28 (CDR1as^Lo) (H) or 501MEL (CDR1as^Lo) (I) melanoma cells expressing dox-inducible control (shNTC) or CDR1as-targeting (shA/shB) inducible shRNA. (J and K) Normalized invasion of 501MEL (J) or SK-MEL-147 (CDR1as^Mod) (K) cells transiently transfected with control (Empty) or CDR1as-overexpression (ZK-SCAN-CDR1as) constructs. (E,F,H-K) Invasion assay data normalized to cell input controls and scaled to the mean of all conditions within experiments. Normalized, scaled means ± SD of replicate experiments (n ≥ 3) are plotted. Statistical comparisons between shA/B and shNTC or ZK-SCAN-CDR1as and Empty vector by two-tailed, paired student’s t test, * p<0.05, ** p<0.01, *** p<0.001. (L) Tumor growth curves of 451Lu (CDR1as^Mod) xenografts expressing inducible non-targeting (shNTC, n = 11) or CDR1as-targeting (shCDR1as, n = 11) shRNA. (M) Lung metastasis burden at experimental endpoint as assessed by in vivo luciferase imaging (IVIS) of animals in (L). Radiance (photons/s/cm²/sr) from individual animals and mean ± 95% CI are plotted. Statistical analysis by Mann Whitney test. (N) Representative ex vivo fluorescence images of whole lungs of animals in (L). Scale bars are 2 mm.

See also Figure S3.

miR-7 is not a critical mediator of CDR1as depletion in melanoma cells

Next, we sought to investigate the molecular mechanism(s) underlying CDR1as’ role as a metastasis suppressor in melanoma. CDR1as interacts with miR-7 through a series of conserved binding sites. Initial reports suggested CDR1as functions as an endogenous inhibitor of miR-7 (Hansen et al., 2013; Memczak et al., 2013). Recently, CDR1as knockout (KO) mice revealed a more complex interaction(Piwecka et al., 2017), as CDR1as also stabilizes miR-7 in brain tissues. We examined the effects of CDR1as silencing on miR-7 expression and activity, and the potential contribution of miR-7 to the effects of CDR1as loss in melanoma. CDR1as depletion did not impact miR-7-5p abundance in melanoma cells (Figure 4A), nor does their expression correlate in melanoma cell lines or tissues (Figures S4A,B). Additionally, CDR1as knockdown did not modulate the signal from a fluorescent miR-7 reporter (Figure 4B)(Mullokandov et al., 2012), and inconsistently alters the abundance of known miR-7 targets (Figures 4C,D, and S4C). Moreover, analysis of predicted or experimentally defined miR-7 targets in RNA-seq of CDR1as-depleted melanoma cells revealed only minimal effects on the miR-7 target landscape (Figure S4D). Finally, to directly assess if miR-7 activity is required in this context, we tested effects of CRISPR-Cas9-mediated disruption of miR-7-1 processing, which strongly reduces mature miR-7 expression (Figure 4E), and transient transfection of oligonucleotide inhibitors of miR-7 in CDR1as-depleted cells. Critically, neither miR-7 depletion nor inhibition were sufficient to revert effects of CDR1as depletion on melanoma cell invasion (Figures 4F and S4E-G). Analysis of known miR-7 targets by immunoblot in these experiments revealed both increased (IRS1, IRS2, RAF1) and decreased (EGFR) abundance in CDR1as-depleted cells, with limited additional modulation of these targets by miR-7 depletion (Figures 4G and S4H). Collectively, these data demonstrate that CDR1as modulation has limited effects on miR-7 abundance and activity in melanoma cells, and miR-7 is insufficient to explain effects of CDR1as depletion on melanoma cell invasion.

Figure 4. miR-7 does not mediate effects of CDR1as depletion in melanoma cells.

(A) miR-7-5p expression by RT-qPCR from the indicated melanoma cells expressing dox-inducible non-targeting (shNTC) or CDR1as-targeting shRNA (shA or shB). (B) Average median fluorescence intensity (MFI) of melanoma cells stably expressing a perfectly matched (WT) or seed-mutant (MUT) miR-7 GFP reporter and the indicated dox-inducible non-targeting (shNTC) or CDR1as-targeting (shA/shB) shRNA. (C) Relative expression by RT-qPCR of the indicated miR-7 target genes from WM278 (CDR1as^Hi) cells expressing dox-inducible non-targeting (shNTC) or CDR1as-targeting (shA/shB) shRNA. (A-C) Mean ± SD of replicate experiments (n ≥ 3) are plotted. Statistical analyses by two-tailed, paired student’s t test. NS = not significant, ** p<0.01 (D) Representative immunoblot of the indicated proteins from WM278 cells expressing dox-inducible non-targeting (shNTC) or CDR1as-targeting (shA/shB) shRNA. Band intensity relative to shNTC condition is shown. (E and F) miR-7-5p expression (E) and normalized invasion (F) of dox-inducible control (shNTC) or CDR1as-targeting (shB) WM278 cells co-expressing Cas9 and the indicated miR-7 sgRNA. Data normalized to cell input controls and scaled to the mean of all conditions within experiments. Normalized, scaled means ± SD of replicate experiments (n ≥ 3). Statistical analyses by Sidak’s multiple comparison test. NS = not significant, * p<0.05. (G) Representative immunoblot images of the indicated proteins from the cells of (F). Band intensity relative to the shNTC/sgNTC condition is shown.

See also Figure S4.

IGF2BP3 interacts with CDR1as and mediates invasion induced by CDR1as depletion

As observed for other non-coding RNA(Abdelmohsen et al., 2017; Lee et al., 2016), we reasoned that CDR1as may interact with and modulate the function of an RNA-binding protein(s). Examination of publicly available CLIP-seq (cross-linking immunoprecipitation followed by sequencing) data(Yang et al., 2015) revealed candidate interactors of CDR1as, including the IGF2BP family (Figures S5A,B). IGF2BPs are oncofetal proteins that regulate a large repertoire of target mRNA transcripts through diverse mechanisms (Bell et al., 2013; Degrauwe et al., 2016; Pryor et al., 2008; Sheen et al., 2015; Yu et al., 2010), and are reported to be pro-tumorigenic in numerous cancers (Bell et al., 2013; Lederer et al., 2014). Confirming these interactions, we observed significant enrichment of CDR1as in IGF2BP2 and 3 RNA immunoprecipitation (RIP-PCR) experiments, particularly for IGF2BP3 (Figure 5A,B). Strikingly, silencing of IGF2BP3 completely abolished the induction of invasion elicited by CDR1as depletion (Figures 5C and S5C-E), suggesting IGF2BP3 is a critical mediator of this phenotype. To further investigate this possibility, we performed IGF2BP3 RIP-seq in control and CDR1as-depleted melanoma cells (Figure S5F). In control samples, CDR1as was the top enriched transcript in IGF2BP3 RIPs (Figures 5D,E), further validating their interaction. 1658 protein-coding genes were bound by IGF2BP3 (logFC>1, FDR < 0.05, RIP/Input), including known targets HMGA2(Jonson et al., 2014), CD44(Vikesaa et al., 2006), and CDK6(Palanichamy et al., 2016). Comparison of genes bound by IGF2BP3 in melanoma cells with published RIP- and CLIP-seq targets revealed from ~20-40% overlap (Table S3)(Conway et al., 2016; Ennajdaoui et al., 2016; Hafner et al., 2010; Huang et al., 2018b), suggesting conserved and context-dependent targets. Moreover, in our IGF2BP3 RIP-seq, we observed significant enrichment by GSEA of IGF2BP, but not negative control RBFOX2, CLIP targets extracted from POSTAR(Hu et al., 2017) or previous studies (Conway et al., 2016; Ennajdaoui et al., 2016) (Table S3 and Figures S5G-I). Finally, previously published IGF2BP binding motifs, which are largely CA-rich sequences, are scattered across the CDR1as sequence, and frequently coincide with IGF2BP3 CLIP peaks from HEK293 cells (Figure 5F and Table S3). These data suggest that CDR1as is densely occupied by IGF2BP proteins when concurrently expressed.

Figure 5. IGF2BP3 interacts with CDR1as and mediates invasion induced by CDR1as depletion.

(A) Representative immunoblot of IGF2BP proteins from Input, IgG, or the indicated IGF2BP RIP samples. (B) Enrichment of the indicated genes by RT-qPCR of IGF2BP RIP. Mean % of input ± SD from replicate experiments (n = 2) are plotted. Statistical comparisons between IGF2BP and IgG RIPs for each gene were performed by Two-stage linear step-up procedure of Benjamini, Krieger and Yekutieli, with Q = 1%. * indicate FDR < 0.01. (C) Normalized cell invasion of control (shNTC) or CDR1as-depleted WM278 melanoma cells co-transfected with control (siNTC) or IGF2BP3-targeting siRNA (siIGF2BP3). Data normalized to cell input controls and scaled to the mean of all conditions within experiments. Normalized, scaled means ± SD of replicate experiments (n ≥ 3). Statistical analyses by two-tailed, paired student’s t test. ** p<0.01. (D) Cumulative distribution plot of RIP/Input ratios (log2FC) from IGF2BP3 RIP-seq, with CDR1as and (+) control transcripts indicated. (E) Snapshot of LINC00632/CDR1as locus with tracks from Input and IGF2BP3 RIP-seq samples. (F) Illustration depicting IGF2BP binding motifs, IGF2BP3 CLIP-seq peaks, and miR-7 seed sites on the CDR1as sequence. (G) Venn diagram of genes up- or down-regulated by CDR1as depletion (red circles) overlapped with IGF2BP3-bound genes (blue circle). ES = enrichment score. Statistical analyses by Fisher’s Exact test. (H) Normalized cell invasion of control (shNTC) or CDR1as-depleted WM278 melanoma cells co-transfected with control (siNTC) or siRNA targeting the indicated gene. Data normalized to the mean of all conditions per experiment (n = 2). Scaled means ± SD of independent transfections (n = 8) are plotted. Statistical comparisons by Mann Whitney test between shCDR1as/siNTC group with shCDR1as/siTargets. * p<0.05, ** p<0.01 (I) Representative immunoblot of the indicated proteins in control (shNTC) or CDR1as-depleted (shCDR1as) WM278 cells co-transfected with control (siNTC) or IGF2BP3-targeting siRNA. Band intensity relative to the shNTC/siNTC condition is shown.

See also Figure S5 and Table S3.

Next, we sought to examine the effects of CDR1as depletion on IGF2BP3 function. We did not observe a general change in IGF2BP3-bound targets (logFC > 1, FDR <0.05 RIP/Input) between control and CDR1as depleted cells (Table S3). Moreover, IGF2BP3-bound transcripts were not more likely to be modulated by CDR1as depletion than non-targets (Figure S5J). These results show that CDR1as modulation does not alter the landscape of IGF2BP3-bound targets or, broadly, their expression, consistent with a recent report which found that even direct modulation of RBPs only impacts transcript abundance of some targets(Van Nostrand et al., 2018). Consistent with this notion, genes that were up- or down-regulated by CDR1as depletion were significantly enriched for IGF2BP3 targets (Figures 5G and S5K). To test the contribution of modulated IGF2BP3 targets in our models, we performed an RNAi-based mini-screen of targets up-regulated by CDR1as silencing, which were selected based on reported functions, such as SNAI2 (SLUG, EMT), NTRK2 (neuronal migration), SEMA6D (axon guidance), and MEF2C (neural crest development). Depletion of 6 of 9 candidates significantly reverted the increased invasion elicited by CDR1as depletion (Figures 5H and S5L). Importantly, immunoblot analysis revealed that up-regulation of SNAI2 and MEF2C in CDR1as-depleted cells is IGF2BP3-dependent (Figure 5I). Overall, our results uncover an interaction between CDR1as and IGF2BP3, and provide evidence that IGF2BP3 and a subset of its targets mediate invasion effects of CDR1as depletion in melanoma cells. These data warrant future studies to define the precise mechanism(s) regulating CDR1as-modulated IGF2BP3 targets and to dissect their relative contributions to phenotypes elicited by the CDR1as-IGF2BP3 axis.

CDR1as expression is reflective of distinct cell states and response to GPX4 inhibitors

Finally, because melanoma cell lines have highly variable CDR1as abundance, we investigated associations of CDR1as expression with sensitivity to therapeutic agents or genetic dependencies in melanoma cells. We leveraged large-scale cancer profiling studies: Depmap (Doench et al., 2016; Meyers et al., 2017; Tsherniak et al., 2017), which catalogues genetic dependencies in cancer cells by pooled RNAi or CRISPR screening, and the Cancer Therapeutics Response Portal (CTRP) (Basu et al., 2013; Rees et al., 2016; Seashore-Ludlow et al., 2015), which profiles the sensitivity of cancer cell lines to a collection of small-molecule compounds. For both CTRPv2 and Depmap Avana 1.0 sgRNA library screening, we extracted data for all melanoma cell lines available. Next, we extracted expression from the CDR1as locus for each cell line available from Cancer Cell Line Encyclopedia (CCLE) (Barretina et al., 2012; Ghandi et al., 2019). We categorized melanoma cell lines by high (CDR1as^Hi), moderate (CDR1as^Med), or low (CDR1as^Lo) CDR1as expression (Figures S6A,B), and compared CDR1as^Hi vs. CDR1as^Lo groups in CTRP and Depmap datasets (Table S4). Interestingly, CDR1as^Lo melanoma cell lines were more sensitive to multiple MAPK pathway inhibitors (Figure 6A). However, CDR1as depletion had no effect on response to BRAF inhibitor Dabrafenib in CDR1as^Hi cells (Figures S6C,D). Subsequent analysis of BRAF mutation status of the CCLE melanoma lines exposed an imbalance of BRAF mutations (4/10 CDR1as^Hi vs. 18/21 CDR1as^Lo) that may be responsible for the initial observation.

Figure 6. CDR1as expression is reflective of MITF/AXL cell states and response to GPX4 inhibitors.

(A) Heat map depicting differential drug sensitivity of the indicated drugs in CDR1as^Hi (n = 10) vs. CDR1as^Lo (n = 21) melanoma cells. (B) Area under the curve (AUC) of dose response curves of the indicated agent for CDR1as^Hi, CDR1as^Med, and CDR1as^Lo groups. (C) Average dependency score (CERES) for GPX4 sgRNA in CDR1as^Hi, CDR1as^Med, and CDR1as^Lo groups. (D) MITF or AXL expression (RPKM) in CDR1as^Hi, CDR1as^Med, and CDR1as^Lo groups. (B-D) Box plots in display median, 25^th and 75^th percentile, and min-max whiskers. Statistical analyses by two-tailed student’s t test comparing CDR1as^Lo with CDR1as^Hi groups. ** p<0.01, *** p<0.001, **** p<0.0001 (E) Mean normalized CDR1as expression ± SD (log2) by RT-qPCR in 5 selected melanoma cell lines. (F) Representative immunoblot of AXL and MITF for cells in (E). (G) Representative RSL3 dose response curves of cells in (E). Viability assessed by Cell Titer Glo (RLU) at t = 24 h post-treatment. Mean ± SD of min-max normalized relative light units (RLU) are plotted. Curves fit by log(inhibitor) vs. normalized response with variable slope and IC50s estimated (μm). (H) Individual IC50s (μm) and group means ± SD of replicate dose response curves of (G) (n = 3). Statistical comparisons between IGR1 or 501MEL with WM278, WM1361a, or WM115 by Sidak’s multiple comparisons test. NS = not significant, *** p<0.001.

See also Figure S6 and Table S4.

CDR1as^Hi melanoma cell lines were strikingly more sensitive to three different GPX4 inhibitors (Figures 6A,B, and S6E), which are known to elicit ferroptotic cell death(Dixon et al., 2012; Yang et al., 2014). Moreover, in genetic dependency data, GPX4 sgRNAs were the top differentially depleted in CDR1as^Hi vs. CDR1as^Lo melanoma cells (Figure 6C), providing independent evidence that CDR1as expression associates with response to GPX4 blockade. Recent studies have documented sensitivity to GPX4 inhibitors as a feature of distinct cell states and drug-tolerant persister cells in a range of tumor types(Hangauer et al., 2017; Viswanathan et al., 2017). In melanoma, GPX4 inhibitor sensitivity associates with cells states characterized by inverse MITF and AXL expression. The MITF^Lo/AXL^Hi cell state is associated with increased NF-κB activity and GPX4 inhibitor sensitivity, but resistance to other drugs, such as MAPK inhibitors(Konieczkowski et al., 2014; Muller et al., 2014; Viswanathan et al., 2017). In contrast, the MITF^Hi/AXL^Lo cell state is associated with MAPK inhibitor sensitivity(Konieczkowski et al., 2014). We observed that high CDR1as expression associates with the MITF^Lo/AXL^Hi cell state and low CDR1as with the MITF^Hi/AXL^Lo cell state in CCLE melanoma cells (Figure 6D). Validating these findings, we observed a similar association in 5 characterized cell lines (Figure 6E,F), which exhibited clear differential sensitivity to the GPX4 inhibitor RSL3 (1S,3R-RSL-3) (Figures 6G,H). However, similar to MAPKi, CDR1as modulation did not impact response to GPX4 inhibition in melanoma cells (Figure S6F). Overall, these data support a model in which CDR1as expression reflects, but is insufficient to dictate, transcriptional states in melanoma, which have implications for treatment responses to clinically-relevant and experimental therapies.

Discussion

During the multi-step process of metastasis, tumor cells encounter an array of barriers requiring diverse cellular properties to successfully colonize distant organs(Hanahan and Weinberg, 2011). Cellular phenotypes regulated by transcriptional and epigenetic mechanisms are important drivers of metastatic processes(Chatterjee et al., 2017). The role(s) of emerging molecular classes, such as circRNAs, to cancer cell metastasis are only beginning to be characterized. Our study reveals loss of CDR1as expression as a critical feature of melanoma progression, expands our understanding of CDR1as biogenesis, and uncovers pathways regulated by CDR1as.

Recently, the connection between LINC00632 and CDR1as was reported(Barrett et al., 2017). We show that LINC00632 transcripts are the source of CDR1as in melanoma cells, providing independent confirmation of their findings. Whether LINC00632 transcripts also harbor important functions is not known. LINC00632 is substantially less abundant than CDR1as, predominantly nuclear, and unlikely to encode for protein(s)(Barrett et al., 2017). Our data are consistent with substantially lower LINC00632 transcript abundance compared with CDR1as, suggesting that LINC00632 may exist solely to generate CDR1as. However, we also observed a clear switch in TSS use and isoform expression of LINC00632, which is surprising if its sole function is CDR1as production. Future studies investigating transcriptional regulation, processing, turnover, and possible functions of LINC00632 in various physiological settings are warranted to expand our understanding of the transcriptional output of this locus.

CDR1as was originally described as an inhibitor of miR-7 in brain tissues(Hansen et al., 2013; Memczak et al., 2013). Surprisingly, analyses of Cdr1as KO mice paradoxically revealed loss of mature miR-7-5p, with increased miR-7 target abundance(Piwecka et al., 2017). These observations suggest that CDR1as stabilizes mature miR-7, in addition to limiting its access to mRNA targets. Further revealing the complexity of this network, a recent study implicated the lncRNA Cyrano (OIP5-AS1 in humans) in miR-7 turnover. In its absence, miR-7 accumulates and facilitates miR-671-5p-mediated destruction of CDR1as (Kleaveland et al., 2018). In melanoma cells, short-term depletion of CDR1as did not affect miR-7 expression, nor is miR-7 abundance different in CDR1as^Lo vs. CDR1as^Hi melanoma cells, despite expression of OIP5-AS1 (data not shown). Moreover, in our models, we failed to detect clear or consistent changes in miR-7 activity after CDR1as depletion. More importantly, inhibition of miR-7 did not revert the effects of CDR1as depletion on melanoma cell invasion, providing evidence that miR-7 deregulation is not a critical mediator of CDR1as silencing effects in melanoma.

Regulation of RNA binding proteins is a plausible alternative role for CDR1as, and interactions of circRNAs with RBPs have been reported(Abdelmohsen et al., 2017; Schneider et al., 2016). We provide evidence that IGF2BP3 interacts with CDR1as and is a key downstream effector of CDR1as depletion in our models, suggesting that CDR1as loss in melanoma may be required to unleash pro-metastatic functions of IGF2BP3. Consistently, our data identify several IGF2BP3 targets that may contribute to this phenotype. While our findings illustrate the importance of individual targets, we theorize that modulated IGF2BP3 targets likely contribute collectively to CDR1as phenotypes in melanoma cells. Moreover, we suspect that we are not fully assessing the effects of CDR1as depletion on many IGF2BP3 targets, which may be impacted translationally or by localization within the cell. Elucidation of the full set of targets and pathways regulated by the CDR1as-IGF2BP3 axis and the interplay of their likely positive and negative effects on melanoma progression await future study.

Finally, for melanoma patients, there remains a critical need to identify those patients diagnosed with localized primary tumors who are at greatest risk for metastatic dissemination of their tumors. Moreover, for patients with advanced disease, markers predictive of response to therapy could aid selection of treatment options for individual patients. CDR1as may represent a valuable prognostic biomarker for high-risk primary melanoma patients, as we observed a clear association between CDR1as loss and poor patient outcomes. Moreover, we find that CDR1as abundance might be a useful marker to predict response to current (MAPK inhibitors) and possible future treatments (GPX4 inhibitors). Expanded analyses examining these associations to determine CDR1as potential utility as a prognostic and/or predictive biomarker, and investigating the contribution of CDR1as to the establishment or maintenance of these cell states, are warranted.

Overall, our work, revealing CDR1as as a robust metastasis-suppressive circRNA in melanoma, highlights an important role for this quintessential circRNA in cancer pathogenesis. Moreover, our data support alternative functions of CDR1as beyond its well-described regulation of miR-7 in brain tissues. We provide evidence that CDR1as modulates the activity of IGF2BP3, an interacting partner of CDR1as. Our findings may have prognostic and/or therapeutic applications in melanoma or other tumor types with LINC00632/CDR1as silencing.

STAR Methods

Lead Contact and Materials Availability

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Eva Hernando (Eva.HernandoMonqe@nvulanqone.org). Plasmids and short-term cultures generated during this study are available upon request and are restricted only by standard MTA.

Experimental Model and Subject Details

Cell Lines

501MEL cells (gift of Dr. Ruth Halaban, Yale Medical School), SK-MEL-2, SK-MEL-28, SK-MEL-147, A375 (ATCC), SK-MEL-173 (gift of Dr. Alan Houghton, MSKCC), SK-MEL-239 (Gift of Poulikos Poulikakos, Mount Sinai), and 113/6-4L (4L, gift of Dr. Robert Kerbel, Sunnybrook Health) were cultured in DMEM (Invitrogen) with 10% (v/v) fetal bovine serum (FBS, Corning), and 1% (v/v) penicillin/streptomycin. 451Lu (ATCC), WM278, WM1361a, WM793b, WM35, WM902b, WM115, WM266-4, and WM1552c (gift of Dr. Meenhard Herlyn, Wistar Institute, or obtained from Coriell Biorepository or Rockland Scientific) cells were cultured in medium containing 80% (v/v) MCDB153 (Sigma Aldrich), 20% (v/v) L15 (Invitrogen), and supplemented with1.2g/L NaHCO₃ (Sigma Aldrich), 2% heat-inactivated FBS or heat-inactivated tet-free FBS (Clontech/Takara), 1.68 mM CaCl₂ (Sigma Aldrich), 5 μg/mL bovine insulin (Sigma Aldrich), and 1% (v/v) penicillin/streptomycin. Melanocytes purchased from Sciencell and Promocell were cultured in Melanocyte growth media (Sciencell or Promocell) and dissociated for harvest using Detach Kit (Promocell). All cells were grown in a humidified incubator at 37°3 and 5% CO₂. Cell line authentication by ATCC’s STR profiling service has been performed for and authenticated the following cell models: 451Lu, A375, 113/6-4L (4L), WM278, and WM793b. STR profiling of SK-MEL-147 and 501MEL confirmed human origin and lack of contamination with cell lines present in the ATCC STR database, but a reference profile was unavailable for these cultures. Other cell lines have not been subjected to STR profiling. Cells were maintained in culture for no more than 25 passages, excluding passaging prior to receipt in the Hernando lab. All cells routinely tested negative for Mycoplasma contamination.

Short-term cultures (STCs) were established at NYU Langone Medical Center (I.O. lab) from surgically resected patient tissues, as previously described (de Miera et al., 2012). The use of patient materials was approved by the New York University Institutional Review Board (number 10362), and all patients provided written informed consent prior to tissue collection. Briefly, after surgical removal, tissue samples were placed immediately in cold sterile RPMI (Mediatech Cellgro) + penicillin (1000,000 IU/l) streptomycin (100 mg/l), and then transferred to research laboratory. The time between tissue harvesting and arrival in the lab was recorded (30 - 60 minutes). The tissue was then placed in a 60mm Petri dish with 4ml of RPMI 1640 with 10 % FBS (Mediatech Cellgro), supplemented with 2 mM L glutamine, 1 mM sodium pyruvate, 1x MEM nonessential amino acid solution (Sigma-Aldrich) and penicillin (1000,000 IU/l) streptomycin (100 mg/l)). Tissue was cut into 1.5 to 2 mm cubes, transferred to a 15 ml conical tube, and shaken vigorously for 10 seconds to release cells. Fragments were allowed to settle to the bottom for 3 min, the supernatant was removed and transferred to a sterile 60 mm dish for culture. The fragments were re-suspended in fresh media and transferred to a 25 cm² tissue culture flask, attached to the bottom of the flask using a scalpel, then cultured at 37°C (5% CO₂). STCs were passaged once the dish or the flask reached 90-100% confluence. All melanoma STCs were fully established by passage 7, approximately two months after initial cell isolation and culture. By passage 10 to 14, the cell lines reached 100% purity, containing only melanoma cells. RNA-sequencing of STCs was performed on RNA-extracted from cultures at passage 12 or fewer.

Clinical Specimens

The study included 105 fresh-frozen tumor samples from patients with cutaneous malignant melanoma: 53 primary melanomas and 52 melanoma metastases, including 22 in-transit, 18 regional lymph node metastases and 12 distal metastases. All tumor specimens were collected at the time of surgery at the Department of Anatomic Pathology, Hospital Clínico Universitario of Valencia, from July 2003 to October 2013. Melanoma patients were categorized by gender, age at diagnosis, and primary tumor localization. Additionally, primary tumor parameters relevant to this study were collected, including Breslow thickness, ulceration, mitotic rate, histological type, among others. Clinical follow-up ranged from 6.6 to 158.6 months (mean 71.6 months, median 68 months). All tumors were classified according to the 2009 American Joint Committee on Cancer (AJCC) staging system.

Primary and metastatic tumor specimens were manually macro-dissected to obtain samples with greater than 90% tumor tissue content. In particular, for primary melanomas, a tumor slice immediately adjacent to the thickest area of the tumor was selected for RNA extraction, being immediately frozen in liquid nitrogen and stored at −80°C. The remaining tumor tissue from each case was formalin-fixed and paraffin-embedded for routine diagnosis. This protocol was approved by the Ethical and Scientific Committees of the Hospital Clínico Universitario of Valencia, Spain, and all patients provided written informed consent before tissue collection. Medical records from all patients were reviewed and clinical follow-up for this study was locked in December 2016.

Mice

In vivo mouse experiments were performed in compliance with a referenced protocol (120405-03) approved by the NYU Institutional Animal Care and Use Committee (IACUC). 4-6 weeks old NOD/Shi-scid/IL-2Rgamma null (NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ (NSG)) mice (female) were purchased from Jackson Laboratory and maintained under standard pathogen free conditions. Experimental sample size was based on our previous experience using this xenograft model system.

Bacterial Strains

All plasmid constructs were propagated in STBL3 or DH5α (Invitrogen) or XL1 Blue (Agilent) on LB plates or in liquid LB media with 100 μg/mL Ampicillin at 37°C. Liquid cultures were grown with orbital shaking at >200 RPM.

Methods Details

Plasmids

Tet-pLKO-puro was a gift from Dmitri Wiederschain (Addgene plasmid # 21915; http://n2t.net/addgene:21915; RRID:Addgene_21915)(Wiederschain et al., 2009).

pLKO-sgRNA-GFP and bdLV-GFP MCS-NGFR (for miR-7 sensor cloning) were gifts of Brian Brown (Mount Sinai School of Medicine).

LentiCas9-Blast (Addgene plasmid # 52962; http://n2t.net/addgene:52962; RRID: Addgene_52962), lentiGuide-Puro (Addgene plasmid # 52963; http://n2t.net/addgene:52963; RRID:Addgene_52963), lenti dCAS-VP64_Blast (Addgene plasmid # 61425; http://n2t.net/addgene:61425; RRID:Addgene_61425), lenti MS2-P65-HSF1_Hygro (Addgene plasmid # 61426; http://n2t.net/addgene:61426; RRID:Addgene_61426), and lenti sgRNA(MS2)_puro (Addgene plasmid # 73795; http://n2t.net/addgene:73795; RRID:Addgene_73795) were gifts from Feng Zhang(Konermann et al., 2015; Sanjana et al., 2014).

psPAX2 (Addgene plasmid # 12260; http://n2t.net/addgene:12260; RRID:Addgene_12260) and pMD2.G (Addgene plasmid # 12259; http://n2t.net/addgene:12259; RRID:Addgene_12259) were gifts from Didier Trono.

pcDNA3.1(+) ZKSCAN1 MCS Exon Vector (Addgene plasmid # 69901; http://n2t.net/addgene:69901; RRID:Addgene_69901) and pcDNA3.1(+) ZKSCAN1 MCS-ciRS7 Exon (Addgene plasmid # 69906; http://n2t.net/addgene:69906; RRID:Addgene_69906) were gifts from Jeremy Wilusz(Kramer et al., 2015).

pLenti PGK V5-LUC Neo (w623-2) (Addgene plasmid # 21471; http://n2t.net/addgene:21471; RRID:Addgene_21471) and pLenti CMV GFP Puro (658-5) (Addgene plasmid # 17448; http://n2t.net/addgene:17448; RRID:Addgene_17448) were gifts from Eric Campeau & Paul Kaufman(Campeau et al., 2009).

Plasmid preparation

Plasmids were extracted by mini- or maxi-prep (Qiagen) following manufacturer’s recommendations. All constructs were verified by analytical digest and/or Sanger sequencing.

Cloning

sgRNA

Non-targeting and miR-7-1-targeting sgRNA sequences were extracted from the GECKO sgRNA library file. sgRNA targeting the 3’ splice donor site of CDR1as or the TSS’s of LINC00632 isoforms were generated using Benchling’s built-in tool. Briefly, CDR1as genomic sequence and >1Kb of flanking sequence was downloaded directly within Benchling. The 100 bases surrounding the 3’ splice were selected as a target region for guide design using base parameters. 3 guides were selected with predicted cut sites within 15 bases of the splice site. LINC00632 TSS-targeting sgRNA were designed in Benchling to target within 100 bases proximal to the annotated TSS for ENST00000498732 or ENST00000602535. All sgRNA sequences were cloned into pLKO-sgRNA-GFP (gift of Dr. Brian Brown, Mount Sinai School of Medicine), pLentiguide-Puro (Addgene # 52963), or Lenti-sgRNA(MS2)-Puro (Addgene #73795) by golden gate cloning of in vitro hybridized and phosphorylated oligos into BbsI or BsmBI cloning sites. sgRNA sequences are available in Supplementary Information: Oligos. pLentiguide-Puro and pLKO-sgRNA-GFP constructs were used in conjunction with pLentiCas9-Blast. Lenti-sgRNA(MS2)-Puro constructs were used in conjuction with Lenti-MS2-P65-HSF1-Hygro and Lenti-dCas9-VP64-Blast.

shRNA

shRNA sequences were cloned into Tet-pLKO-Puro (Addgene # 21915) by standard cloning methods. Briefly complementary DNA oligos were annealed and phosphorylated with T4 PNK, leaving overhangs compatible with EcoRI/AgeI. Vector backbone was digested with EcoRI/AgeI, releasing a stuffer fragment, and gel purified. Purified cut vector was dephosphorylated using Fast Alkaline Phosphatase (Roche) following manufacturer’s directions. Annealed oligos were cloned into dephosphorylated vector using T4 DNA ligase following standard methods.

Bacterial Transformation

Ligation products were transformed in STBL3 (Thermo Fisher), DH5alpha (Invitrogen) or XL1-Blue (Agilent) following standard methods. Colonies were screened by colony PCR and verified by Sanger sequencing.

Transfection

The indicated cells were seeded 16-24 hr prior to transfection to yield a density of 60-80% confluence at the time of transfection. Liposomal cocktails with siRNA (Dharmacon, smart pools) (25 nM final), microRNA mimics or inhibitors (Dharmacon, miRIDIAN, 25-50 nM final), or GapmeRs (Exiqon) were generated with Lipofectamine 2000 or 3000 (Invitrogen) in Opti-MEM (Invitrogen) following manufacturer’s recommendations. Cells were incubated with transfection complexes for at least 8 hr prior to media change. Transfected cells were incubated for 48-72 hr prior to use.

Viral Production

4x10⁶ HEK293T cells were seeded per 10 cm tissue culture dish coated with gelatin solution and incubated overnight at 37°C and 5% CO2. 16-20 hr after seeding, HEK293T were co-transfected with lentiviral construct (15 μg), viral packaging plasmid (psPAX2, 10 μg), and viral envelope plasmid (pMD2.G, 5 μg) using Lipofectamine 2000 or 3000 (Invitrogen), following manufacturer’s recommendations. Viral supernatant was collected, filtered through 0.45 μm filters at 36 hr post-transfection, and stored at 4°C for short-term use (1-5 days) or −20°C for long-term storage (5-30 days).

Viral Transduction

Target cells were seeded (1e6 cells per 10 cm dish) and incubated overnight prior to infection. Medium was replaced with 1:2 or 1:4 diluted viral supernatant, and incubated for 16-20 hr, followed by replacement with growth medium. Cells were checked for fluorescent protein expression and/or drug selection (puromycin (500 ng/mL – 1 μg/mL), blasticidin (5-10 μg/mL), hygromycin (100-200 μg/mL), and/or G418 (200-400 μg/mL)) performed beginning at 48 hr post-infection to ensure pure populations of transduced cells.

RNA Extraction

Cultured Cells and Explanted Tumors

RNA was extracted using miRNeasy mini kit (Qiagen) following manufacturer’s recommendations. Briefly, 700 μL of Qiazol per sample were added to pelleted cells or tumor tissue pieces (10-20 mg). To resuspend, cells were vortexed or tumor tissue chopped with scissors followed by repeatedly passing through a 22.5 gauge needle. Samples were incubated at RT for 5 min. 140 μL chloroform were added and tubes were shaken for 15 s, followed by 2 min incubation at RT. Tubes were centrifuged at 12,000 xg at 4°C for 15 min. Aqueous phase (~350 μL) was transferred to a fresh microcentrifuge tube. 1.5X volumes of 100% EtOH were added and mixed by vortex. 700 μL at a time were transferred to RNeasy mini spin columns and centrifuged at 13,000rpm for 30 s. Repeat with remainder of sample, discarding flow-through. 350 μL of buffer RWT were added per column and centrifuged at 13,000 rpm for 30 s. Flowthrough was discarded. 80 μL of RNAse-free DNAse (Qiagen) were added to each column and incubated at RT for 15 min. 350 μL RWT buffer was added to column, followed by centrifugation at 13,000 rpm for 30 s. Two washes with 500 μL buffer RPE were performed discarding flow-through each time, followed by centrifugation at 13,000 rpm for 2 min to remove all traces of ethanol from RPE buffer. Columns were transferred to 1.5 mL provided RNA collection tubes and 30 to 50 μL RNase-free H₂O were added per column for RNA elution. After 1 min incubation at RT, columns were centrifuged at 13,000 rpm for 1 min. Eluted RNA was quantified by Nanodrop 8000 or Qubit (Thermo Fisher) following manufacturer’s recommendations, and stored at −80°C.

Clinical Specimens

Total RNA extraction from patient melanoma tissues was performed following manufacturer’s recommendations using the mirVana miRNA Isolation Kit (Life Technologies, Ambion). Melanoma tissues were immediately disaggregated into very small portions with a pre-chilled scalpel and were mechanically homogenized in 600μl of Lysis/Binding buffer with the bead-based disruption system TissueLyser LT (Qiagen). Total RNA was eluted in 60 μl (2 X 30 μl) RNase-free H₂O. RNA quality and quantity were determined using Nanodrop ND-1000 Spectrophotometer (Thermo Fisher), and stored at −80°C.

RNA-Seq of melanocytes, STCs and cell lines

RNA was extracted using QIAGEN miRNeasy minikit. RNA was processed with Ribo-Zero rRNA Removal Kit (Illumina), and further processed into sequencing libraries using the Illumina ScriptSeq Complete Gold or TruSeq kits following manufacturer’s recommendations. Samples were barcoded and sequenced using standard Illumina chemistry. Original 10 melanoma STCs used for circRNA annotation were sequenced on Illumina HiSeq2500 (~150M reads, PE101). All other melanoma STCs and cell lines and cultured melanocytes were sequenced on Illumina HiSeq 2500 or 4000 (~40-100M reads, PE50). Indexed sample data was demultiplexed and individual FASTQ files generated, followed by quality control assessment with FASTQC.

RNA-seq Analyses

circRNA Profiling

Raw RNA-seq data for cultured human melanocytes were downloaded from ENCODE (Biosamples: ENCBS477ENC, ENCBS478ENC, ENCBS479ENC, and ENCBS480ENC). FASTQ files were analyzed and quality checked with FASTQC.

MapSplice

Reads were aligned to the human reference genome (hg38) using the MapSplice (Wang et al., 2010) alignment algorithm (version 2.1.8, using parameters –bam –fusion –fusion-non-canonical –filtering 1 –min_fusion_distance 200 –gene-gtf Ensembl). The number of unique reads for each circRNA junction was obtained from the resulting splice junction files. A post-hoc filtering step using custom python scripts was applied to eliminate chromosomal translocations, +/− fusions and < 2 reads. Differential expression analysis was performed using edgeR. Heatmap was generated in R using the code: Heatmap(log2(counts[de.genes, ]+1).

BEDtools multicov (Quinlan and Hall, 2010) was used to calculate the circularization ratio using the following formula: Circularization Ratio = Unique Backsplice Reads / (Unique Backsplice Reads + Average 5' - 3' Linear Reads).

DCC

All paired-end sequencing reads were processed with the DCC pipeline (version 0.4.7) in order to identify circular RNAs. Therefore, reads were aligned against hg38 human reference sequence using STAR aligner (version 2.5.0c) (Dobin et al., 2013) with parameters for paired-end reads defined by DCC instructions (see https://github.com/dieterich-lab/DCC for details). To adjust for differences in read-length across the various datasets, we used less stringent alignment criteria (--outSJfilterOverhangMin 10 10 10 10; --alignSJoverhangMin 10; --chimSegmentMin 10; --chimScoreMin 10; --chimScoreSeparation 10; --chimJunctionOverhangMin 10). The “chimeric.out.junction” output file of the STAR aligner (Dobin et al., 2013) was then used as input for the DCC algorithm with default parameters (-D option for circular RNA detection mode). Output was annotated with gene annotations from Ensembl Genes V85 and normalized to total sequencing depth. We further filtered for canonical junctions (GT/AG and CT/AC) and excluding all junctions annotated as “intergenic”.

Overlap of Mapsplice and DCC

Raw circRNA counts file obtained from DCC was pre-processed to remove circRNA annotations detected in only 1 sample. Mapslice and DCC raw circRNA counts files were pre-processed to harmonize unique identifiers for each annotated circRNA species, defined as Chromosome:Start-Stop:Strand. Pre-processed, data tables were uploaded to UseGalaxy.org and merged using the Join feature, with retention of circRNA annotations unique to each algorithm (Afgan et al., 2018). A venn diagram plot was generated in powerpoint to illustrate the percentage overlap of Mapsplice with DCC annotations.

CDR1-AS expression

Primary and cell line RNA-Seq datasets were aligned against hg19 human reference sequence using STAR aligner with default parameters. Read counts for CDR1-AS were counted in a stranded manner using bamutils count of the ngsutils package (version 0.5.7) (Breese and Liu, 2013), and FPKM was calculated using CDR1-AS gene length and total sequencing depth.

LINC00632 Isoform expression

FASTQ files were aligned against hg38 human reference sequence using Tophat (version 2.0.9)(Trapnell et al., 2009). Transcript abundances were calculated using Cufflinks (version 2.2.1)(Trapnell et al., 2010).

Validation of putative circRNAs

Cell pellets were resuspended in RLT buffer and snap frozen until extraction. All RNA samples were obtained using RNeasy kits (Qiagen) following the manufacturer’s instructions. Quality and concentration were routinely estimated by NanoDrop (Thermo) or Tapestation (Agilent) and RNA samples were kept at −80°C until analysis. One microgram of RNA was used for reverse transfection using random hexamers for priming. The resulting cDNA was used as template for divergent PCR using primers designed with Primer3 by restricting the regions of interest to the exonic sequences forming the backsplice and then manually selecting divergent primers with similar Tm. Proper amplicon size was determined by 2% agarose gel electrophoresis and reactions with unique bands of correct size were TOPO cloned (Invitrogen) and Sanger sequenced.

miRNA quantification by RT-qPCR

Clinical samples

Reverse transcription (RT) was carried out using TaqMan MicroRNA Reverse Transcription Kit in presence of RNase inhibitor (Applied Biosystems). 200 ng of total RNA was converted to cDNA in 45 μl reactions. RT reaction was multiplexed by creating a customized RT primer pool with miRNA-specific RT primers of interest following manufacturer’s recommendations. In brief, miRNA-specific primers were pooled and diluted in 1X Tris-EDTA (TE) buffer to obtain a final dilution of 0.05X each. The miRNA RT primer pool consisted of 5 miRNA stem-loop specific primer pairs: hsa-miR-7-5p ID: 000386, hsa-miR-574-3p ID: 002349, hsa-miR-671-5p ID: 197646_mat, RNU6B ID: 001093 and RNU48 ID: 001006.

To determine the levels of mature miRNAs miR-7-5p and miR-671-5p, quantitative Real time PCR (qPCR) was performed following cDNA generation.

Briefly, 1 μl of cDNA was used as template in a 10-μl qPCR reaction by adding TaqMan Universal Master Mix II, no UNG (Applied Biosystems) and predesigned TaqMan MicroRNA Assays PCR primers and probes (FAM dye-labeled) for target miRNAs. RNU6B and RNU48 small RNAs were used as endogenous references. All reactions were performed in triplicate on a 7900 HT Fast Real- Time PCR system (Applied Biosystems) in 384-well plates. RT and qPCR negative controls were included for each assay.

For the relative quantification analysis of miRNA expression, data was normalized to the geometric mean of RNU6B and RNU48, the comparative Ct method was used and the results were analyzed using Expression Suite software (Applied Biosystems).

Cultured Cells

Reverse transcription (RT) was carried out using TaqMan MicroRNA Reverse Transcription Kit in presence of RNase inhibitor (Applied Biosystems). Briefly, 25ng of input RNA was reverse transcribed by stem-loop method following manufacturer’s recommendations. To determine the levels of mature miRNAs miR-7-5p and miR-671-5p, quantitative Real time PCR (qPCR) was performed following cDNA generation. Briefly, 1μl of cDNA was used as template in a 10-μl qPCR reaction by adding TaqMan Universal Master Mix II, no UNG (Applied Biosystems) and predesigned TaqMan MicroRNA Assays PCR primers and probes (FAM dye-labeled) for target miRNAs. RNU44 small RNA was used as an endogenous reference gene. All reactions were performed in triplicate using Biorad CFX 384 or ABI StepOne Plus real-time cyclers, following manufacturer’s recommendations.

mRNA/circRNA quantification by RT-qPCR

All primers are listed in Table S5.

Clinical samples

200 ng of RNA extracted from human melanoma patient samples were reverse transcribed using Taqman RT kit (Applied Biosciences) following manufacturer’s recommendations. cDNA was diluted 1:4 with nuclease-free, pico-pure H₂O prior to use in qPCR reactions. Diluted cDNA was assessed for expression of CDR1as and 3 reference genes (GAPDH, PPIA, and SRCAP) by RT-qPCR. Further cDNA dilution (1:50) was used to assess 18S expression as a 4^th reference gene. CDR1as was measured using circular-specific divergent primers and primers that would measure both circular and linear products arising from the CDR1 locus (Pearson r >0.95). Briefly, 1uL (references) or 2 μL (CDR1as) of cDNA product was used in 10 μL qPCR reactions in 384-well plates using Power SYBR Green qPCR Master Mix and run on a Biorad CFX 384 quantitative PCR system with the following 2-step cycling parameters: 10 min at 95°C, 40 cycles of 95°C for 15 s followed by 60°C for 30 s, followed by melt curve analysis. Technical triplicates of PCR reactions were performed.

Cultured Cells and Explanted Tumors

250 to 1000 ng of RNA were reverse transcribed using Applied Biosciences Taqman RT kit (Applied Biosystems, Thermo Fisher) following manufacturer’s recommendations. cDNA was diluted with nuclease-free H2O prior to use in qPCR reactions. GAPDH, PPIA, and/or MALAT were used as endogenous reference genes in RT-qPCR experiments. Briefly, 15 μL or 25 μL PCR reactions were performed in 384-well or 96-well plates, respectively. Biorad CFX 384 or ABI StepOne Plus real-time cyclers were used with the following 2-step cycling parameters: 10 min at 95°C, 40 cycles of 95°C for 15 s followed by 58-60°C for 30 s, followed by melt curve analysis. Technical triplicates of PCR reactions were performed.

RT-qPCR Data Analysis

Median Ct values of technical triplicates were used for quantification of expression. When possible, delta Ct values of biological replicates were pooled. For clinical samples, expression of test genes was analyzed by comparative Ct method using the geometric means of the indicated four reference genes (delta Ct), followed by centering to the median across all samples (deltadelta Ct). Data are plotted as normalized log2 expression or as linear fold change (2^ddCt). For cultured cells, expression of test genes was analyzed by comparative Ct method using a single reference (GAPDH, MALAT, or PPIA) (deltaCt), followed by normalization to a control condition/treatment (deltadelta Ct). Data are plotted as normalized log2 expression or as linear fold change (2^ddCt) with standard deviation. Correlations between CDR1as and miR-7-5p or miR-671-5p were performed by Pearson comparing log2-transformed, mean-normalized expression. Correlations and associations of CDR1as with clinical parameters were performed by Student’s T Test (clinical associations) or Pearson (clinical correlations).

Single Molecule in situ hybridization

Target probe sequences, preamplifier, amplifier, wash buffer and target retrieval buffers are proprietary (Advanced Cell Diagnostics, CA). Cells were harvested by standard methods, counted, and seeded in 8-well chamber slides (LabTek II) at 25,000 cells per well. Cells were fixed in 10% neutral buffered formalin for 30 min at RT, washed 3X with PBS, and dehydrated in 50% then 70% then 100% EtOH. Fixed, dehydrated cells were stored in 100% ethanol at −20C until ready to use. The slides were rehydrated by submerging slides in 70% ethanol, followed by 50% ethanol and finally in 1XPBS. Hydrophobic barriers drawn around the edges of the chambers with Immedge Pen and submerging the slides back in 1X PBS. The slides were incubated with hydrogen peroxide (RTU from ACD) for 10 min at RT. After rinsing the slides twice in 1xPBS, slides were subjected to hybridization and staining following manufacturer’s recommendations. Briefly, slides were treated with Protease III (ACD Bio) at 1:15 in 1X PBS for 10 min at RT followed by rinsing in 1X PBS twice. Hybridization cassette (ACD Bio) and HybEZ oven (ACD Bio) were used for subsequent hybridization, preamplification and amplification process. CDR1as or PPIB probes (ACD Bio) were hybridized in the HybEZ oven set at 40°C for 2 hr, followed by preamplification and amplification steps based on ACD protocol. Chromogenic detection was performed using DAB/Fast Red (ACD). The slides were then counterstained with hematoxylin (Richard Allen 7211).

LINC00632-CDR1as splice product analysis

cDNA was generated of RNA from human cerebellum or WM278 melanoma cells, as previously described. LINC00632-CDR1as splice variants were amplified by PCR using forward primers binding in LINC00632 exons and a reverse primer binding near the 5’ splice acceptor of CDR1as. Amplified products were TOPO cloned (Invitrogen), following manufacturer’s recommendations, and analyzed by Sanger sequencing.

Proliferation Assays

Cells harboring dox-inducible shNTC or shCDR1as constructs were treated with 1ug/mL Doxycyline for 48-72 hr prior to assay seeding. Cells were harvested, counted, and seeded in growth media with 1ug/mL Doxycycline at 2500 cells/well in replicate 96-well plates. After cells adhered, time 0 was fixed with 1% glutaraldehyde in PBS. Each 24 hr after, one replicate plate was fixed with 1% glutaraldehyde in PBS. 5-6 time points were collected, then stained with crystal violet staining solution (0.5% crystal violet in 10% Methanol and 90% water. Stained plates were washed extensively with water, dried, and destained with 15% acetic acid for measuring absorbance at 595nm.

In Vitro Invasion

Cell invasion was measured using 24-well Fluoroblok transwell inserts (Becton Dickinson, 8 μM pore. Briefly, inserts were coated for 2 hr at 37C with Matrigel (Becton Dickinson/Corning) diluted in coating buffer (0.01M pH8 Tris-HCl, 0.7% NaCl). Residual matrigel solution was aspirated immediately prior to cell seeding into inserts. Cells were harvested, counted in triplicate, washed, and resuspended in serum- and insulin-free growth medium. 40,000 cells (WM278 and WM115) per condition were seeded per coated Fluoroblok insert and corresponding control wells in cell input plate. Cells were allowed to settle for 10 min, followed by addition of growth media (700 μL) to lower chamber. 20-40 hr after seeding, invading cells were post-stained with Calcein AM diluted to 1-2 μg/mL in pre-warmed HBSS for 10 min at 37C. For each independent experiment, 3-4 inserts per condition were used. 5 randomly selected fields per insert were imaged using a 10X objective or 1 representative field per insert at 4X on an inverted fluorescence microscope. For IGF2BP3 siRNA ‘rescue’ experiments, WM278 cells harboring control (shNTC) or CDR1as-targeting (shCDR1as) shRNA were treated with 1 μg/mL doxycycline for 48 hr prior to transfection with siRNA. 48 hr post-siRNA transfection cells were harvested and invasion performed as above. For siRNA epistasis miniscreen, WM278 shNTC or shCDR1as cells were treated for 48 hr with 1 μg/mL doxycycline, harvested, and seeded at 20,000 cells per well in a standard 96-well tissue culture plate. 16-20 hr after seeding, transfection complexes were generated for each tested siRNA and added to a minimum of 4-replicate wells per condition. 48 hr after transfection, cells were washed 1X with 100 μL/well PBS and detached with 25 μL enzyme-free dissociation buffer at 37C for 8 min. 75 μL serum-free, insulin-free growth media was added per well. Single cell suspensions were generated by repeated pipetting (10x) and 50 μL transferred per well to 96-well Fluoroblok transwell plates precoated with 20 μL Matrigel. ~40 hr after seeding, cells were stained with Calcein and imaged with a 5X objective, as above.

Quantification of invading cells

Invading cells were counted using the following automated macro in ImageJ (FIJI):

 macro "Batch Convert to Binary" {
  dir = getDirectory("Choose a Directory ");
  list = getFileList(dir);
  setBatchMode(true);
  for (i=0; i<list.length; i++) {
    path = dir+list[i];
    open(path);
    run("8-bit");
    setAutoThreshold();
    run("Threshold…");
    setThreshold(30, 255);
    run("Convert to Mask");
    setThreshold(255, 255);
    run("Watershed");
    run("Analyze Particles…", "size=400-Infinity circularity=0.00-1.00 show=Outlines display clear summarize");
    dotIndex = lastIndexOf(path, ".");
    if (dotIndex!=−1)
        path = substring(path, 0, dotIndex); // remove extension
    save(path+"-bin.tif");
    close();
   }
}
saveAs("Text");
     dir = getDirectory("Choose a Directory ");


     list = getFileList(dir);

Changes to minimum size of Analyze Particles function were determined empirically for each cell line. Cell input control wells were fixed for 15 min with 1% Glutaraldehyde in PBS, washed 1X with PBS, stained with a 0.5% Crystal Violet solution for 30 min to 2 hr at room temp, followed by extensive washing with diH₂O. Cells were destained with 15% acetic acid and quantified by absorbance at 595 nm. Counts of invading cells for each well were normalized to the mean absorbance of the corresponding condition from the cell input plate to control for variations in cell seeding. Replicate experiments were combined by scaling the average of each condition to the replicate mean. Statistical analysis was performed by two-sided, paired Student’s T test or Mann-Whitney test (miniscreen)

Animal Studies

451Lu cells were infected with a lentiviral construct for GFP expression. Cells were selected by cell sorting to isolate a pure population of GFP-expressing cells (451Lu^GFP). 451Lu^GFP cells were subsequently infected with a Renilla luciferase expression construct carrying a neomycin resistance and were fully selected using G418 at 1 mg/mL. Finally, 451Lu^GFP/Neo-Luc cells were infected with pLKO Puro TetOn inducible shRNA constructs (negative control or shCDR1as_B) and fully selected with Puromycin at 1 μg/mL. Efficacy of CDR1as depletion in these cells was tested in vitro prior to use in animal experiments.

Cells trypsinized, washed 1X with PBS, resuspended at 10e6 cells/mL in growth media, and kept on ice prior to implantation in mice. 150 μL of cell suspension (1.5e6 cells) was mixed with 150 μL of undiluted Matrigel (Becton Dickinson/Corning). 200 μL (1e6 cells) were subsequently injected subcutaneously into the right flank of NSG mice (n = 12 per group). Palpable tumors were allowed to form prior to placing animals on DOX-containing food (200 mg/kg, ad libitum, starting 10 days post-implantation) to drive expression of shRNAs. When tumors were palpable, primary flank tumors were measured twice weekly by caliper (length (l) and width (w)) until resected. Tumor volumes were estimated by the formula: (w² * l)/2. Survival surgery to remove primary flank tumors was performed between day 35 and 42 post-implantation in accordance with the previously referenced protocol. Two animals were lost to follow up (1 did not survive surgery and 1 was subsequently sacrificed due to complications from surgery). Tumor volumes ranged between 200-600mm³. Small pieces of pure primary tumor tissue were cut and flash-frozen on dry-ice. Remaining primary tumor tissue was fixed in 10% buffered formalin for 48 hr, wash 2X with PBS, dehydrated through graded EtOH, and embedded in paraffin following standard conditions. Following removal of primary flank tumors, animals were monitored by IVIS imaging weekly to monitor development of lung metastasis. Briefly, luciferin substrate was injected IP at a dose of 150 mg/kg body weight (25 mg/ml luciferin), 15 min before imaging. Mice were anesthetized with isoflurane/oxygen and placed on the imaging stage. Dorsal images were collected by automatic exposure (.5 s to 2 min) using the IVIS (Xenogen Corp., Alameda, CA). Analysis was performed using LivingImage software (Xenogen) by measurement of photon flux (measured in photons/s/cm2/steradian) with a region of interest (ROI) drawn around the bioluminescence signals to be measured. All animals were euthanized in accordance with the referenced IACUC-approved protocol 106 days post-implantation. Organs (lungs, liver, kidney, spleen, and brain) were removed, rinsed briefly in Ca- and Mg-free PBS, fixed in 10% buffered formalin for 48 hr, and embedded in paraffin following standard conditions. For lungs, prior to fixation, ventral and dorsal macroscopic images were taken with a dissecting microscope equipped with a fluorescent lamp and digital camera. Primary tumors and lung tissues were sectioned (5 μM) and stained by Hematoxylin & Eosin following standard methods.

Protein Isolation

Protein lysates were generated using RIPA buffer (Thermo Fisher) supplemented with protease inhibitors (Complete EDTA-free, Roche) and phosphatase inhibitors (PhosStop, Roche) for 20 min on ice, followed by centrifugation for 15 min at 13,000 rpm at 4°C. Protein-containing supernatant was transferred to fresh microcentrifuge tubes and stored at −20 or −80°C until further use. Protein was quantified using DC Protein Assay (Biorad) following manufacturer’s recommendations, with standard curves generated with bovine serum albumin (Sigma Aldrich).

Histone Isolation

Histones were isolated by acid extraction. Briefly, adherent cells were washed 2X with PBS, removed from tissues culture dishes by scraping, and pelleted. Cell pellets were used immediately or snap-frozen and stored at −80°C until use. Cell pellets were resuspended in hypotonic lysis buffer (10 mM Tris-HCl pH 8, 1 mM KCl, 1.5 mM MgCl₂, 0.5% NP40. Add DTT fresh to 1 mM, protease inhibitor cocktail tablet (Sigma complete Mini, EDTA-free) added fresh), and incubated for 20-30 min on ice. Nuclei were pelleted at 4000 rpm for 5 min at 4°C, followed by resuspension in 400 μL of 0.4N H₂SO₄ and vortexing. Samples were gently rocked overnight at 4°C. Acid-insoluble material was pelleted 10,000 rpm for 10 min at 4°C, and supernatant transferred to a fresh tube, followed by addition of 100% TCA (w/v) to a final concentration of 20% (v/v). Histones were precipitated on ice for 60 min followed by pelleting at max speed for 10 min at 4°C. Pellets were washed 2X with ice-cold acidified-acetone (0.1% HCl in acetone) and 2X with ice-cold straight acetone to remove all residual acid. Pellets were air-dried (at least 20 min), followed by solubilization in H₂O at RT for 1-2 hr or 4°C overnight. Insoluble material was removed by centrifugation and protein quantified by DC Protein Assay (Biorad) following manufacturer’s recommendations, with standard curves generated with bovine serum albumin (Sigma Aldrich).

Western Blotting

10 μg or 20 μg of total protein lysate was loaded per lane of 4-20% Bis-tris polyacrylamide mini gels (Invitrogen). SDS-PAGE was run at 150V for 1.5 to 2 hr. Proteins were transferred to nitrocellulose or PVDF membranes by wet transfer for 60-75 min at 100V. Membranes were briefly washed 1X in diH2O followed by blocking with 5% non-fat dry milk (Bio-rad) in Tris-buffered saline supplemented with Tween20 (0.1%) (TBS-T) or Phosphate-buffered saline (PBS) for 45-60 min at RT. After blocking, membranes were washed briefly with TBS-, cut horizontally to examine multiple proteins of different sizes per gel. Membranes were incubated on a plate shaker overnight at 4°C with primary antibodies diluted in TBS-T. Membranes were washed extensively with TBS-T (minimum 4X for 5 min), followed by incubation with appropriate horseradish peroxidase-conjugated secondary antibodies diluted in TBS-T with 1-2% non-fat dry milk for 30-60 min at RT on a plate shaker. Membranes were washed extensively with TBS-T (minimum 4X for 5 min). Signal was detected using Luminata Crescendo detection system (EMD Millipore) following manufacturer’s recommendations. Multiple film (HyBlot CL, Denville) exposures ranging from 2 s to 2 min were performed for optimal images analysis. Primary antibodies used are listed in the Key Resource Table.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit mAB H3K27me3 (C36B11)	Cell Signaling	Cat. # 9733; RRID:AB_2616029
Rabbit mAB Histone H3 (D2B12)	Cell Signaling	Cat. # 4620; RRID:AB_1904005
Rabbit IgG	Bethyl Laboratories	Cat. # P120-101; RRID:AB_479829
Rabbit mAb EGFR (D38B1)	Cell Signaling	Cat. # 4267; RRID:AB_2246311
Rabbit mAb IRS1 (D23G12)	Cell Signaling	Cat. # 3407; RRID:AB_2127860
Rabbit pAb IRS2	Cell Signaling	Cat. # 4502; RRID:AB_2125774
Rabbit mAb c-RAF (RAF1) (D4B3J)	Cell Signaling	Cat. # 53745; RRID:AB_2799444
Rabbit mAb MEF2C (D80C1)	Cell Signaling	Cat. # 5030; RRID:AB_10548759
Rabbit mAb Slug (SNAI2) (C19G7)	Cell Signaling	Cat. # 9585; RRID:AB_2239535
Rabbit mAb AXL (C89E7)	Cell Signaling	Cat. # 8661; RRID:AB_11217435
Rabbit pAb TDP43 (G400)	Cell Signaling	Cat. # 3448; RRID:AB_2271509
Rabbit pAb IGF2BP1	MBL International	Cat. # RN007P; RRID:AB_1570640
Rabbit pAb IGF2BP2	MBL International	Cat. # RN008P; RRID:AB_1570641
Rabbit pAb IGF2BP3	MBL International	Cat. # RN009P; RRID:AB_1570642
Mouse mAb MITF (D5)	Thermo Fisher Scientific	Cat. # MA5-14154; RRID:AB_10982126
Mouse mAb α-Tubulin (DM1A)	Sigma-Aldrich	Cat. # T9026; RRID:AB_477593
Mouse mAb β-Actin-Peroxidase (AC-15)	Sigma-Aldrich	Cat. # A3854; RRID:AB_262011
Anti-Rabbit IgG (whole molecule)–Peroxidase antibody produced in goat	Sigma-Aldrich	Cat. # A0545; RRID:AB_257896
Anti-Mouse IgG (whole molecule)–Peroxidase antibody produced in rabbit	Sigma-Aldrich	Cat. # A9044; RRID:AB_258431
Donkey Anti-Mouse IgG IRDye 680RD	LI-COR Biosciences	Cat. # 926-68072; RRID:AB_10953628
Goat Anti-Rabbit IgG IRDye 800CW	LI-COR Biosciences	Cat. # 926-32211; RRID:AB_621843
Bacterial and Virus Strains
One Shot Stbl3 Chemically Competent E. coli	Thermo Fisher Scientific	Cat. # C737303
XL1-Blue Ultracompetent Cells	Agilent Technologies	Cat. # 200249
Subcloning efficiency DH5α competent cells	Invitrogen	Cat. # 18265017
Biological Samples
RNA from primary and metastatic melanoma tissues	Carlos Monteagudo, University of Valencia	http://www.uv.es
RNA from patient-derived short term cultures	Iman Osman, NYU Langone Health	(de Miera et al., 2012)
Chemicals, Peptides, and Recombinant Proteins
Calcein-AM	Fisher	Cat. # 08-774-501
GSK126, EZH2 Inhibitor	Cayman Chemical Company	Cat. # 15415
MCDB153	Sigma Aldrich	Cat. # M7403
Tet System Approved FBS	Clontech/Takara	Cat. # 631106
Insulin solution from bovine pancreas	Sigma Aldrich	Cat. # I0516
Opti-MEM	Thermo Fisher Scientific	Cat. # 31985-070
Melanocyte Growth Media kit	Promocell	Cat. # C-24110
Detach Kit	Promocell	Cat. # C-41200
Dynabeads Protein A/G	Thermo Fisher Scientific	Cat. # 10015D
Matrigel Basement Membrane Matrix	BD Biosciences	Cat. # CB-40234
Lipofectamine 2000	Thermo Fisher Scientific	Cat. # 11668027
Lipofectamine 3000	Thermo Fisher Scientific	Cat. # L3000015
1S,3R-RSL3	Tocris	Cat. # 6118
Luminata Crescendo western HRP substrate	EMD Millipore	Cat. # WBLUR0500
HyBlot CL Autoradiography film	Denville	Cat. # E3018
Cell Titer Glo 2.0	Promega	Cat. # G9242
Critical Commercial Assays
TruSeq Stranded Total RNA Library Prep Gold	Illumina	Cat. # 20020598
ScriptSeq Complete Gold kit	Illumina	Cat. # BEP1206
Ribo-Zero Gold Kit	Illumina	Cat. # MRZG12324
MiRNeasy mini RNA extraction kit	Qiagen	Cat. # 217004
MultiScribe Reverse Transcriptase (RT) kit	Applied Biosystems	Cat. # 4311235
Power SYBR Green qPCR MasterMix kit	Invitrogen	Cat. # 4368708
RIP Assay Kit	MBL International	Cat. # RN1001
SMARTer Stranded total RNA sample prep HI kit	Clontech (Takara)	Cat. # 634873
SimpleChIP Enzymatic Chromatin IP kit	Cell Signaling	Cat. # 9003
Taqman microRNA reverse transcription kit	Thermo Fisher Scientific	Cat. # 4366596
Taqman Universal Master Mix II, no UNG	Thermo Fisher Scientific	Cat. # 4440043
Universal cDNA synthesis kit II	Exiqon, Inc.	Cat. # 203301
RNA Spike-In Kit	Exiqon, Inc.	Cat. # 203203
TOPO TA Cloning Kit for Sequencing	Invitrogen	Cat. # 45-0030
RNAscope® Intro Pack 2.5 HD Reagent Kit Brown-Hs	ACD Bio	Cat. # 322370
RNAscope® 2.5 HD Reagent Kit-RED	ACD Bio	Cat. # 322350
Deposited Data
RNA-seq of cultured melanocytes	This manuscript	GSE138711
RNA-seq of melanoma short-term cultures	This manuscript	GSE138711
RNA-seq of melanoma cell lines	This manuscript	GSE138711
RNA-seq of WM278 control (shNTC) or CDR1as-depleted (shCDR1as) cells (2 biological replicates)	This manuscript	GSE138711
RIP-seq of WM278 control (shNTC) or CDR1as-depleted (shCDR1as) cells (3 biological replicates)	This manuscript	GSE138711
RNA-seq of cultured melanocytes (NHM_1 in this paper)	(Fontanals-Cirera et al., 2017)	GSE94488
RNA-seq of melanoma cell lines (SK-MEL-147 and 501MEL) treated with inactive enantiomer R-(-)-JQ1	(Fontanals-Cirera et al., 2017)	GSE94488
RNA-seq of human melanoma cells and cultured melanocytes	(Kaufman et al., 2016)	GSE75356
RNA-seq of human primary and metastatic melanoma	TCGA	The Cancer Genome Atlas, 2015, Level3
RNA-seq and ChIP-seq of melanoma STCs	(Verfaillie et al., 2015)	GSE60666
RNA-seq of mouse cranial neural crest tissues	(Minoux et al., 2017)	GSE89437
DNA methylation array data of human primary and metastatic melanoma	TCGA	The Cancer Genome Atlas, 2015, Level3
Experimental Models: Cell Lines
NHM_N4 (Melanocytes) NHEM PD#10 (Population Doubling)	Promocell	Cat. # C-12400; lot 429Z015.3
NHM_N3 (Melanocytes) HEM-adult P2 (Passage)	Sciencell	Cat. # 2230; lot 16333
NHM_N2 (Melanocytes) HEM-Dark P2 (Passage)	Sciencell	Cat. # 2220; lot 1446
NHM_N5 (Melanocytes) NHEM Juvenile, cell pellet in RNAlater	Promocell	Cat. # C-14040; lot 4070201.1
NHM_N6 (Melanocytes) NHEM-M2 Juvenile, cell pellet in RNAlater	Promocell	Cat. # C-14042; lot 5071302.2
451Lu	Rockland, Inc.	Cat. # 451Lu-01-0001; RRID:CVCL_6357
WM278	Rockland, Inc.	Cat. # WM278-01-0001; RRID:CVCL_6473
WM115	Rockland, Inc.	Cat. # WM115-01-0001; RRID:CVCL_0040
WM266-4	Rockland, Inc.	Cat. # WM266-4-01-0001; RRID:CVCL_2765
WM239A	Rockland, Inc.	Cat. # WM239A-01-0001; CVCL_6795
WM1552c	Rockland, Inc.	Cat. # WM1552C-01-0001; RRID:CVCL_6472
WM902b	Rockland, Inc.	Cat. # WM902B-01-0001; RRID:CVCL_6807
WM983A	Rockland, Inc.	Cat. # WM983A-01-0001; RRID:CVCL_6808
WM983B	Rockland, Inc.	Cat. # WM983B-01-0001; RRID:CVCL_6809
WM983C	Rockland, Inc.	Cat. # WM983C-01-0001; RRID:CVCL_A338
WM1361a	Rockland, Inc.	Cat. # WM1361A-01-0001; RRID:CVCL_6788
WM793b	Dr. Meenhard Herlyn (Wistar)	RRID:CVCL_8787
WM35	Dr. Meenhard Herlyn (Wistar)	RRID:CVCL_0580
IGR1	Dr. Danny Reinberg (NYU Langone Health)	RRID:CVCL_1303
501Mel	Dr. Ruth Halaban (Yale University)	RRID:CVCL_4633;
A375	ATCC	Cat. # CRL-1619; RRID:CVCL_0132;
SK-MEL-2	ATCC	Cat. # HTB-68; RRID:CVCL_0069
SK-MEL-28	ATCC	Cat. # HTB-72; RRID:CVCL_0526
SK-MEL-147	Dr. Alan Houghton (MSKCC, NY)	RRID:CVCL_3876
SK-MEL-173	Dr. Alan Houghton (MSKCC, NY)	RRID:CVCL_6090
SK-MEL-239	Dr. Poulikos Poulikakos (MSSM, NY)	RRID:CVCL_6122
113/6-4L	Dr. Robert Kerbel, (Sunnybrook Health Sciences Center, Toronto)	Cruz-Munoz et al., Cancer Research, 2008
HEK293T	ATCC	Cat. #CRL-3216, RRID:CVCL_0063
Experimental Models: Organisms/Strains
Mouse: female, 6 to 8 weeks old, NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ (NSG)	Jackson Laboratories	005557; RRID:IMSR_JAX:005557
Oligonucleotides
All primer, shRNA, and sgRNA sequences	Table S5	N/A
hsa-miR-7-5p Taqman primers	Thermo Fisher Scientific	Cat. # 4427975 Assay ID: 000386
hsa-miR-671-5p Taqman primers	Thermo Fisher Scientific	Cat. # 4427975 Assay ID: 197646_mat
RNU6B Taqman primers	Thermo Fisher Scientific	Cat. # 4427975; Assay ID: 001093
RNU48 Taqman primers	Thermo Fisher Scientific	Cat. # 4427975; Assay ID: 001006
RNU44 Taqman primers	Thermo Fisher Scientific	Cat. # 4427975 Assay ID: 001094
hsa-miR-7-5p LNA™ PCR primer set, UniRT	Exiqon, Inc.	Cat. # 205877
hsa-miR-671-5p LNA™ PCR primer set, UniRT	Exiqon, Inc.	Cat. # 205649
cel-miR-39-3p, LNA™ control primer set, UniRT	Exiqon, Inc.	Cat. # 203952
miRIDIAN microRNA Human hsa-miR-7-5p - Hairpin Inhibitor	Dharmacon, Inc.	Cat. # IH-300547-06-0005
miRIDIAN microRNA Human hsa-miR-7-5p - Mimic, 5	Dharmacon, Inc.	Cat. # C-300547-05-0005
Antisense LNA GapmeR Standard - Custom	Exiqon (Qiagen)	Cat. # 339511
Antisense LNA GapmeR Negative Control	Exiqon (Qiagen)	Cat. # 339515
siGENOME non-targeting control siRNA Pool #2	Horizon Discovery (Dharmacon)	Cat. # D-001206-14-05
siGENOME Human IGF2BP3 siRNA - SMARTpool	Horizon Discovery (Dharmacon)	Cat. # M-003977-02-0005
siGENOME Human HMGA2 siRNA – SMARTpool	Horizon Discovery (Dharmacon)	Cat. # M-013495-02-0005
siGENOME Human IGF2BP3 siRNA – Set of 4	Horizon Discovery (Dharmacon)	Cat. # MQ-003976-00-0002
siGENOME Human NTRK2 siRNA – SMARTpool	Horizon Discovery (Dharmacon)	Cat. # M-003160-02-0005
siGENOME Human SEMA6D siRNA – SMARTpool	Horizon Discovery (Dharmacon)	Cat. # M-013165-00-0005
siGENOME Human HAS2 siRNA – SMARTpool	Horizon Discovery (Dharmacon)	Cat. # M-012053-01-0005
siGENOME Human CD44 siRNA – SMARTpool	Horizon Discovery (Dharmacon)	Cat. # M-009999-03-0005
siGENOME Human GAS7 siRNA – SMARTpool	Horizon Discovery (Dharmacon)	Cat. # M-011492-00-0005
siGENOME Human SNAI2 siRNA – SMARTpool	Horizon Discovery (Dharmacon)	Cat. # M-017386-00-0005
siGENOME Human MEF2C siRNA – SMARTpool	Horizon Discovery (Dharmacon)	Cat. # M-009455-00-0005
siGENOME Human DAB1 siRNA – SMARTpool	Horizon Discovery (Dharmacon)	Cat. # M-008943-01-0005
RNAScope Probe- Hs-CDR1-AS-No-XMm	ACD Bio	Cat. # 510711
RNAScope Positive Control Probe- Hs-PPIB	ACD Bio	Cat. # 313901
Recombinant DNA
Tet-pLKO-Puro	(Wiederschain et al., 2009)	RRID:Addgene_21915
Tet-pLKO-Puro shNTC	This manuscript	N/A
Tet-pLKO-Puro shCDR1as_A (shA)	This manuscript	N/A
Tet-pLKO-Puro shCDR1as_B (shB)	This manuscript	N/A
pLenti-Cas9-Blast	(Sanjana et al., 2014)	RRID_Addgene_52962
pLentiguide-Puro	(Sanjana et al., 2014)	RRID_Addgene_52963
pLentiguide-Puro sgNTC	This manuscript	N/A
pLentiguide-Puro sgCDR1as_SD_A	This manuscript	N/A
pLentiguide-Puro sgCDR1as_SD_B	This manuscript	N/A
pLentiguide-Puro sgCDR1as_SD_C	This manuscript	N/A
pLKO-sgRNA-GFP	MSSM, NY	Gift from Dr. Brian Brown
pLKO-sgNTC-GFP	This manuscript	N/A
pLKO-sgMIR7-1_3-GFP	This manuscript	N/A
pLKO-sgMIR7-1_4-GFP	This manuscript	N/A
pLenti-CMV-GFP-Puro	(Campeau et al., 2009)	RRID_Addgene_17448
Lenti-dCas9-VP64-Blast	(Konermann et al., 2015)	RRID_Addgene_61425
Lenti-MS2-P65-HSF1-Hygro	(Konermann et al., 2015)	RRID_Addgene_61426
Lenti-sgRNA(MS2)-Puro	(Konermann et al., 2015)	RRID_Addgene_73795
Lenti-sgD (MS2)-Puro (LINC00632v602535 TSS)	This manuscript	N/A
Lenti-sgE (MS2)-Puro (LINC00632v602535 TSS)	This manuscript	N/A
Lenti-sgF (MS2)-Puro (LINC00632v498732 TSS)	This manuscript	N/A
Lenti-sgG (MS2)-Puro (LINC00632v498732 TSS)	This manuscript	N/A
Lenti-sgH (MS2)-Puro (Non-targeting control)	This manuscript	N/A
Lenti-sgI (MS2)-Puro (Non-targeting control)	This manuscript	N/A
pLenti PGK V5-LUC Neo (w623-2)	(Campeau et al., 2009)	RRID_Addgene_21471
psPAX2	Dr. Didier Trono (EPFL Institute)	RRID_Addgene_12260
pMD2.G	Dr. Didier Trono (EPFL Institute)	RRID_Addgene_12259
pcDNA 3.1(+) ZKSCAN MCS	(Kramer et al., 2015)	RRID_Addgene_69901
pcDNA3.1(+) ZKSCAN-ciRS7	(Kramer et al., 2015)	RRID_Addgene_69906
psiCHECK2 miR-7 perfect sensor	This manuscript	N/A
psiCHECK2 miR-7 mutant sensor	This manuscript	N/A
psiCHECK2 empty	Promega	Cat. # C8021
bdLV-GFP miR-7 WT Sensor-NGFR	This manuscript	N/A
bdLV-GFP miR-7 MUT Sensor-NGFR	This manuscript	N/A
Software and Algorithms
MapSplice	Version 2.1.8	http://www.netlab.uky.edu/p/bioinfo/MapSplice2
Tophat	Version 2.0.9	http://ccb.jhu.edu/software/tophat/index.shtml
Cufflinks	Version 2.2.1	https://github.com/cole-trapnell-lab/cufflinks
DCC	Version 0.4.7	https://github.com/dieterich-lab/DCC
ngsutils	Version 0.5.7	https://github.com/ngsutils/ngsutils
STAR	Version 2.5.0c	https://github.com/alexdobin/STAR
featureCounts	Version unknown	http://subread.sourceforge.net
DEseq2	Version unknown	https://github.com/mikelove/DESeq2
FASTQC	Version unknown	https://www.bioinformatics.babraham.ac.uk/projects/fastqc/
EdgeR	Version 3.9	https://bioconductor.org/packages/edgeR/
RStudio	Various	https://rstudio.com
BedTools	Version 2.22.0	https://bedtools.readthedocs.io/
Python	Version unknown	https://www.python.org
Integrative Genomics Viewer (IGV)	Version 2.4.14	http://software.broadinstitute.org/software/igv/home
Primer3	Version unknown	https://github.com/primer3-org
Benchling	Accessed 2016-19	http://www.benchling.com
Galaxy	Accessed 2019	https://usegalaxy.org
Living Image software	Xenogen Corp., Alameda, CA	http://www.perkinelmer.com/product/spectrum-200-living-image-v4series-1-128113
GSEA Preranked	Version 3.0	http://software.broadinstitute.org/gsea/index.jsp
MSigDB Gene Sets	Version 6.2	http://software.broadinstitute.org/gsea/index.jsp
Custom Gene Sets	This manuscript	N/A
ImageJ	Version 2.0.0-rc-68/1.52g	https://fiji.sc
PRISM	Version 7.0, 8.0	https://www.graphpad.com,
Other
Fluoroblok inserts,24-well inserts, 8uM pore	Thermo Fisher Scientific	Cat. # 351152
FluoroBlok Multiwell Insert Systems, 96-well plate, 8uM	Thermo Fisher Scientific	Cat. # 351164

RIP-PCR and RIP-sequencing

RIP was performed using the RIP Assay Kit (MBL International) following manufacturer’s recommendations with Dynabeads Protein A/G beads. Antibodies are listed in the Key Resource Table. Western blotting of a portion of washed RIP samples was performed to confirm specificity of IP. Following RNA isolation, RIP and input RNA was quantified by Qubit. Routinely, IGF2BP RIPs yielded >500ng of RNA while IgG controls were <50 ng. RT-qPCR was performed as previously described, using equal volumes of RIP eluate for IgG and IGF2BPs for cDNA production. For RIP-seq analyses, RIP was performed for IGF2BP3 as described from three biological replicates each of WM278 cells expressing control (shNTC) or CDR1as-targeting (shCDR1as) cells. Ribosomal RNA-depleted, indexed sequencing libraries for RIP (250 ng) and Input (1 μg) RNAs were prepared using SMARTer stranded total RNA sample prep HI kits (Takara) following manufacturer’s recommendations. Libraries were multiplexed and sequenced on a HiSeq 2500 (Illumina) with standard Illumina chemistry.

RIP-seq Analyses

Alignment and quantification pipeline

Sequencing results were demultiplexed and converted to FASTQ format using Illumina bcl2fastq software. The sequencing reads were aligned to the human genome (build hg19/GRCh37) using the STAR aligner. The featureCounts program(Liao et al., 2014) was utilized to generate counts for each gene based on how many aligned reads overlap its exons. These counts were then normalized and used to test for differential expression using negative binomial generalized linear models implemented by the DESeq2 R package (Love et al., 2014).

Enrichment Analyses

To identify the differentially bound transcripts in shCDR1as vs. shNTC IGF2BP3 RIP-seq data, we compared enrichments over enrichments using a certain design pattern in edgeR R package as follows:

$$\text{design}<\text{-}\text{model}.\text{matrix}(\sim \text{condition+protocol+condition:protocol})$$

where condition is shCDR1as.vs.shNTC and protocol is RIP.vs.Input

Next, we carried out the differential expression analysis with default parameters using edgeR with this design matrix to find the differentially bound transcripts. FDR < 0.05 were considered significant.

To compare broad expression changes between shCDR1as and shNTC for IGF2BP3-bound vs. -unbound genes, the log2fc was generated using the DESeq2 R package for shCDR1as vs. shNTC input samples using the following formula:

$$\text{Log2FC}=\log 2(\text{mean}(\text{shCDR1as}))\text{-}\log 2(\text{mean}(\text{shNTC}))$$

For identifying IGF2BP3-bound and -unbound genes, any gene with log2FC > 1 and FDR < 0.05 in shCDR1as(RIP vs. Input) differential expression analysis, filtered for protein-coding genes and raw read counts greater than 1 are considered as IGF2BP3-bound genes and the rest are considered unbound.

For analysis of CDR1as-modulated genes (up and downregulated genes in shCDR1as vs. shNTC) for enrichment with IGF2BP3-bound genes, we took the set of bound transcripts (shCDR1as RIP.vs.Input, filtered for protein-coding genes, minimum number of read counts > 1 and log2FC >1) and calculated an Enrichment Score with the following formula and parameters:

ES = #overlap / ((#upregulated * #bound) / #all genes)

# all genes = Input(shCDR1as.vs.shNTC) filtered for protein-coding genes and minimum number of read counts > 1

bound = Input(shCDR1as.vs.shNTC) filtered for protein-coding genes and minimum number of read counts > 1 and bound genes from shCDR1as(RIP.vs.Input) with log2FC >1

# unbound = Input(shCDR1as.vs.shNTC) filtered for protein-coding genes and minimum number of read counts > 1 and unbound genes from shCDR1as(RIP.vs.Input)

# upregulated = Input(shCDR1as.vs.shNTC) filtered for protein-coding genes and minimum number of read counts > 1 and log2FC >1

# not-upregulated = Input(shCDR1as.vs.shNTC) filtered for protein-coding genes and minimum number of read counts > 1 and !(log2FC >1)

# overlap genes = #bound+#upregulated

#upregulated-overlap = #upregulated - #overlap

#bound-overlap = #bound - #overlap

#all-overlap = #all genes - #overlap

The significance of this enrichment score was assessed by p value that was calculated using Fishers exact test on the following matrix of parameters:

Contingency_table = matrix(c(#overlap genes, #upregulated-overlap, #bound-overlap, #all-overlap-(#bound-overlap)-(#upregulated-overlap)),nrow = 2, dimnames = list(binding = c("bound", "not-bound"), expression = c("upregulated", "not-upregulated")))

We fixed a pvalue < 0.05 as a threshold to determine whether the enrichment score is significant or not. If ES = 1 then there is no enrichment and overlap of bound and upregulated is exactly as expected by chance. If ES >>1 then it is a strong enrichment and ES << 1 means less enrichment (i.e. upregulated genes tend to be less bound than expected by chance).

GSEA Analyses

GSEA (Subramanian et al., 2005) preranked was used to compare custom gene sets of IGF2BP targets to IGF2BP3 RIP-seq data from melanoma cells generated here. Briefly, for genes with base count >10, log2FC of RIP/Input of control (shNTC) or CDR1as-depeleted WM278 cells were ranked from most to least enriched. Custom gene sets of IGF2BP or RBFOX2 targets were generated from published CLIP-seq studies, as indicated in results.

GSEA preranked was used to compare MSIGdB gene sets to RNA-seq data of WM278 melanoma cells depleted of CDR1as. Genes with base count >10 were ranked by log2FC of CDR1as-depeleted vs. control (shNTC) WM278 cells.

EZH2 inhibitor treatment

EZH2 inhibitor GSK126 (Cayman Chemical) was resuspended in DMSO (20mM stock solution) and stored at −20°C in aliquots to avoid repeated freeze-thaw. Cells were treated with 2μM GSK126 final or equivalent volume of vehicle (DMSO). Media changes with fresh GSK126 were performed every 2-3 days. Cells were harvested at various time points after initiation of GSK126 treatment (as early as 2 days and as long as 21 days), washed 1X with PBS, flash-frozen, and stored at −80°C prior to RNA or Histone extraction.

ChIP-qPCR

501MEL melanoma cells were grown to 80% confluency. ChIP was performed using the SimpleChIP Enzymatic Chromatin IP kit (Cell Signaling Technology) according to the manufacturer’s directions. Native chromatin immunoprecipitation was performed overnight with an anti-H3K27me3 antibody (CST #9733) or anti-H3 (CST #4620) and rabbit IgG (Bethyl #P210-101) as negative controls. The eluted DNAs were amplified by real-time PCR using primer pairs listed in supplementary information.

Drug dose response curves

Cells were seeded at 5000 cells/well in 96-well clear-bottom, white walled plates. 16-20 hr after seeding, cells were treated with half-log drug dilution series for the indicated treatment (Dabrafenib 1 μM to 333 pM, or RSL3 3.333 μM to 1 nM). Cells were grown in the presence of drug for 20-24 hr (RSL3) or 6 days (Dabrafenib). Cell viability was assessed by Cell Titer Glo, following manufacturer’s recommendations, and chemiluminescent output (integration time 1000 ms) was measured on a Perkin Elmer Flexstation 3.0. Data were normalized to max/min and plotted in Graphpad PRISM. IC50s were estimated, and IC50’s from replicate experiments were compared by Student’s T test.

Public Data Resources and Data Mining

Associations of CDR1as locus expression with overall survival in TCGA tumor types were accessed via www.oncolnc.org using CDR1 as the search term. Anonymized patient level data (overall survival in days, live/dead status, and RSEMv2 expression) for significant associations (LGG) were downloaded, and plotted using Graphpad Prism.

CDR1as (CDR1) expression (FPKM) data were extracted from human melanoma cell line RNA-seq from data matrix of GSE75356.

H3K27me3 ChIP-seq of melanoma STCs was accessed by UCSC custom track link-out from the online version of (Verfaillie et al., 2015). H3K27me3, H3K27ac, and RNA-seq BigWig files were downloaded from GSE60666 and loaded in IGV (version 2.4.14)(Robinson et al., 2011) to generate representative image tracks of the LINC00632/CDR1as locus. (Figure S2H)

Data for CDR1 and C230004F18Rik (mouse ortholog of LINC00632) were obtained from non-stranded RNA-seq in supplementary data of Minoux et al.(Minoux et al., 2017). (Figure S1F)

Data for CDR1 and LINC00632 expression in normal human tissues were obtained via GTEx data portal (https://gtexportal.org) on 06/19/17. (Figure S1F)

Data for CDR1 and LINC00632 expression correlations in tumor samples were obtained via cBioPortal (https://cbioportal.org). Genome-wide expression correlations with CDR1 for each listed tumor type were downloaded to excel, then sorted/ranked by Spearman correlation coefficient. (Table S2)

DNA methylation data from Illumina 450K human methylation arrays available for melanoma patients was extracted from TCGA HM450K level 3. Median β value for each sample was calculated for the specified CpG dinucleotides included in the Illumina 450K Human Methylation array. β value methylation score ranges from 0 (lack of methylation) to 1 (complete methylation).

Oncoprint of TCGA – SKCM data depicting genetic amplification/deletion of LINC00632/CDR1as locus was generated using cBioPortal (http://www.cbioportal.org) on June 16, 2017 (Figure S2E).

CLIP-seq data mining to identify putative CDR1as-interacting RBPs was performed manually in jBrowse data viewer from CLIPdb/POSTAR (p value cut-off < 0.01)(Hu et al., 2017). CLIP-seq reads for IGF2BP3 were extracted from CLIPdb/POSTAR and mapped in Benchling to the CDR1as sequence. For visual output, individual RBP tracks from multiple samples were merged. IGF2BP1/2/3 tracks were merged. In addition, hsa_circ_0001946 (CDR1as) was used as search term at https://circinteractome.nia.nih.gov to examine possible interacting RBPs. (Figure S5A,B)

Cancer Cell Line Encyclopedia (CCLE) RNA-seq data were downloaded from (https://portals.broadinstitute.org/ccle/)(Barretina et al., 2012). Non-stranded data mapped to CDR1 was extracted and used as CDR1as, as CDR1 is not detected in stranded RNA-seq of melanoma cells.

AUC data for compound profiling was downloaded from CTRPv2 (https://portals.broadinstitute.org/ctrp/) and melanoma cell line-specific data was extracted. CDR1as expression data obtained from CCLE was mapped to corresponding CTRPv2 cell line drug profiling data. Cell lines were grouped by CDR1as expression as per Figure S6A. AUC data for each tested compound were averaged by CDR1as expression group. Averages were compared between CDR1as^Lo vs. CDR1as^Hi by subtraction and statistical comparisons made by Student’s t testing without multiple testing correction.

Gene-level, CERES-modified Avana 1.0 sgRNA library screening data were downloaded from Depmap.org, and melanoma cell line-specific data were extracted. CDR1as expression data obtained from CCLE was mapped to corresponding Avana 1.0 cell line sgRNA depletion data. Cell lines were grouped by CDR1as expression as per Figure S6B. Gene-level depletion values were averaged for each CDR1as expression group. Averages were compared between CDR1as^Lo vs. CDR1as^Hi by subtraction and statistical comparisons made by Student’s t testing without multiple testing correction. Significant associations were considered when p<0.01.

Quantification and Statistical Analysis

RNA-seq and RIP-seq data analyses were conducted in R and statistical analyses performed as described. All other data plotting and statistical analyses were performed with GraphPad Prism version 7 or 8 (GraphPad Software, Inc.). Data are presented as the mean ± SD of replicate experiments, as indicated, and presented as individual values, scatter plots, heatmap, box plots, and bar graphs. Sample sizes or number of experimental replicates (n) is indicated in figure legends. Significance was determined using unpaired/paired Student’s t test, Mann-Whitney test, or Log-Rank test (Kaplan-Meier curves), where appropriate. Correlations between CDR1as and LINC00632 were performed by comparing Log2 transformed, mean normalized data and analyzed by Pearson and Spearman correlation in GraphPad Prism.

Data and Code Availability

RNA seq and RIP seq data generated for this publication have been deposited in NCBI’s Gene Expression Omnibus under the accession: GSE138711.

Code associated with circRNA annotation by MapSplice is available in GitHub: https://github.com/alchilito/circRNA.

Code associated with RNA-seq and RIP-seq pipelines is available in GitHub: https://github.com/igordot/sns

Supplementary Material

3

Table S1: CircRNA annotations and raw backsplice counts detected in melanocytes and melanoma STC RNA-seq, Related to Figure 1.

4

Table S3: Analyses of IGF2BP3 RIP-seq, Related to figure 5.

5

Table S4: Analyses of Depmap and CTRP Data, Related to figure 6.

6

Table S5: Primer and Oligos Sequences, Related to STAR methods.

Significance.

Our results show that the circular RNA CDR1as is actively silenced during melanoma progression and that loss of CDR1as expression functionally contributes to the metastatic cascade of melanoma cells. Moreover, CDR1as abundance associates with patient survival and therapeutic response and thus may have prognostic and/or predictive value for melanoma patients. Finally, our work uncovers CDR1as as an interactor and regulator of the RNA binding protein IGF2BP3, thus revealing additional functions of CDR1as, which was previously thought to exclusively regulate miR-7.

Highlights.

• Loss of CDR1as expression promotes melanoma invasion and metastasis

• CDR1as arises from a lncRNA that is epigenetically silenced by EZH2/PRC2

• CDR1as-IGF2BP3 interactions largely mediate the effects of CDR1as loss in melanoma

• CDR1as levels associate with cell states and therapeutic sensitivities