|Home | About | Journals | Submit | Contact Us | Français|
Some solid tumors have reduced posttranscriptional RNA editing by adenosine deaminase acting on RNA (ADAR) enzymes, but the functional significance of this alteration has been unclear. Here, we found the primary RNA-editing enzyme ADAR1 is frequently reduced in metastatic melanomas. In situ analysis of melanoma samples using progression tissue microarrays indicated a substantial downregulation of ADAR1 during the metastatic transition. Further, ADAR1 knockdown altered cell morphology, promoted in vitro proliferation, and markedly enhanced the tumorigenicity in vivo. A comparative whole genome expression microarray analysis revealed that ADAR1 controls the expression of more than 100 microRNAs (miRNAs) that regulate many genes associated with the observed phenotypes. Importantly, we discovered that ADAR1 fundamentally regulates miRNA processing in an RNA binding–dependent, yet RNA editing–independent manner by regulating Dicer expression at the translational level via let-7. In addition, ADAR1 formed a complex with DGCR8 that was mutually exclusive with the DGCR8-Drosha complex that processes pri-miRNAs in the nucleus. We found that cancer cells silence ADAR1 by overexpressing miR-17 and miR-432, which both directly target the ADAR1 transcript. We further demonstrated that the genes encoding miR-17 and miR-432 are frequently amplified in melanoma and that aberrant hypomethylation of the imprinted DLK1-DIO3 region in chromosome 14 can also drive miR-432 overexpression.
Genetic alterations and dysregulated epigenetic modifications play a role in cancer development and progression (1). Recently, posttranscriptional events, such as perturbation of microRNA (miRNA) expression, are emerging as new players in the development of various human malignancies (2). A-to-I RNA editing is a posttranscriptional process mediated by the adenosine deaminase acting on RNA (ADAR) enzymes, which catalyze nucleotide conversion in RNA transcripts. It was previously reported that many solid tumors generally exhibit lower A-to-I RNA editing (3), but the mechanistic and functional significance has remained obscure.
The deamination of adenosines by ADAR enzymes may occur in coding regions, altering the amino acid sequence, codon reading frames, and splicing pattern (4), or in the noncoding regions, thus affecting the stability of RNA transcripts or their nuclear retention (5). There are 3 members of the ADAR family, ADAR1, ADAR2, and the brain-specific ADAR3 (5). The ubiquitously expressed ADAR1 has 2 isoforms: an interferon-inducible cytoplasmic protein (ADAR1-long, P150) and a constitutive nuclear protein (ADAR1-short, P110), which are synthesized from different translation-initiating methionines (6, 7).
Editing by ADAR1 has been implicated in different physiological processes (6–8), such as embryonic hematopoiesis (9, 10) and development of various non-nervous tissues and in host defense against viral infections by massive hyperediting of viral transcripts (4). In addition, editing of regulatory RNAs (pri- and pre-miRNAs) by ADAR1 has been reported (11). Editing of pri-miRNAs by ADAR1 may interfere with miRNA biogenesis at the precursor stage and thereby alter their homeostasis in the cell (12) or target binding (13). An RNA editing–independent regulation of miR-376a2 by ADAR2 was previously reported (13).
Here, we present what we believe is substantial new evidence that ADAR1 has a fundamental role in the regulation of cancer cell phenotype by controlling nuclear and cytoplasmic processing steps of miRNAs in an isoform-specific, RNA binding–dependent yet RNA editing–independent manner. Moreover, we unravel the underlying genomic and epigenetic events that are responsible for loss of ADAR1 expression in metastatic melanoma cells, which subsequently facilitate the acquisition of an aggressive phenotype. These findings provide insights into the process of cancer development, with potential implications for future translational medicine.
A significant hypoediting effect was reported mainly in brain tumors (3). Since melanocytes are derived from the neural crest, the expression of ADAR1 was investigated in situ in melanoma by using progression tissue microarray (TMA) obtained from National Cancer Institute Cancer Diagnosis Program (NCI CDP). ADAR1 was heterogeneously expressed in the epidermis of normal skin (control), nevi, primary melanomas, and metastasis (Supplemental Figure 1; supplemental material available online with this article; doi: 10.1172/JCI62980DS1). Staining was mainly observed in the nucleus, which is consistent with the constitutive short form of ADAR1 (ADAR1-P110) (5). There were no significant differences in the nuclear staining among normal epidermis, benign nevi, and primary melanoma specimens (Figure (Figure1A).1A). In contrast, a significantly weaker nuclear staining was observed in the metastatic specimens (Figure (Figure1A).1A). There was almost no cytoplasmic staining in the normal epidermis, as the long form of ADAR1 (ADAR1-P150) is inducible (5). Interestingly, nevi and primary melanoma specimens exhibited a moderate cytoplasmic staining (Figure (Figure1A).1A). While weak cytoplasmic staining could be observed in some metastatic specimens, it was not different in a statistically significant manner from the normal epidermis (Figure (Figure1A).1A). A significant trend toward downregulation of nuclear and cytoplasmic ADAR1 immunoreactivity in metastatic specimens was independently observed in another progression TMA obtained from Massachusetts General Hospital (Supplemental Figure 1, see Methods). There was no significant association between ADAR1 expression level and the BRAF (v-raf murine sarcoma viral oncogene homolog B1: ID 673) or NRAS (neuroblastoma RAS viral [v-ras] oncogene homolog: ID 4893) mutation status (data not shown).
In order to quantify the downregulation of ADAR1 in metastatic cells, 34 low-passage patient-derived metastatic melanoma cultures and 4 cultures of normal melanocytes were tested with real-time quantitative PCR (qPCR). ADAR1 was downregulated by at least 2-fold in 62.5% of the metastatic melanoma samples (Figure (Figure1B).1B). ADAR1 levels were almost identical among all cultures of normal melanocytes (Figure (Figure1B).1B). The ratio between ADAR1-P110 and ADAR1-P150 was between 4–12 times in favor of ADAR1-P110 in all melanoma cultures (data not shown). ADAR1 expression levels in selected melanoma cultures as tested by Western blot were in agreement with the qPCR results (Figure (Figure11C).
To study the effect of ADAR1 silencing, 624mel melanoma cells were stably transduced with shRNA that targets both the short and long forms of ADAR1. The transductants were then stably transfected with ADAR1-P110 or ADAR1-P150 bearing silent mutations conferring resistance to the ADAR1-selective shRNA to restore specifically their expression or with an empty pcDNA3 vector (rescue-P110, rescue-P150, and ADAR1-knockdown [ADAR1-KD] and control, respectively) (Figure (Figure1D).1D). Accordingly, reduced editing rate of bladder cancer–associated protein (BLCAP; ID:10904) transcripts, a verified editing target of ADAR1 (14), was observed in the ADAR1-KD cells (9%) as compared with the rescue-P150, rescue-P110, and control cells (18%, 18%, and 19%, respectively) (Figure (Figure1E).1E). The expression level of ADAR2 was similar among the various cells, while the brain-specific ADAR3 was undetected in all melanoma samples tested (data not shown).
Light microscopy revealed distinct morphology of the KD cells (elongated cells with dendrites) when compared with the rescue-P110, rescue-P150, and control cells (short and wide cells) (Figure (Figure2A).2A). Digital image analysis confirmed that ADAR1-KD cells exhibit significantly longer and narrower cell bodies as compared with the rescue and control cells (Supplemental Figure 2).
A remarkably enhanced net proliferation rate following ADAR1-KD was observed in the 624mel cells (Figure (Figure2B)2B) and in 6 other melanoma cell lines tested (Supplemental Figure 3). This effect was corroborated by a significant increase in the population of proliferating cells (G2-M) at the expense of resting cells (GO-G1) in the ADAR1-KD cells as compared with rescue and control cells (Figure (Figure2C).2C). On the other hand, loss of ADAR1 expression had no effect on spontaneous apoptosis (Figure (Figure2C,2C, Sub-G1). Further, there was no effect on the rate of apoptosis induced by different mechanisms, such as interference to microtubule breakdown (Taxol) or DNA crosslinking (Cisplatin) (Supplemental Figure 4). Similar effects on morphology, RNA-editing rate, and proliferation were observed with another system, independently established in another melanoma cell line, 526mel (Supplemental Figure 5). The combined effects suggest that ADAR1 has a fundamental role in the regulation of cancerous features and phenotype.
The various transfectants were injected subcutaneously into SCID-NOD mice. All developed into tumor masses, but ADAR1-KD tumors exhibited a dramatically enhanced growth rate in vivo (Figure (Figure2D).2D). After 45 days, tumors were excised and analyzed macroscopically, histologically, and at the molecular level. Macroscopically, the ADAR1-KD tumors were substantially larger (Figure (Figure2E)2E) and heavier (data not shown), supporting the volume measurements (Figure (Figure2D).2D). Immunostaining for the proliferation marker Ki-67 confirmed a considerable increase in cell proliferation in the ADAR1-KD tumors (Figure (Figure2E).2E). Most proliferating cells were located at the periphery, with extensive central necrosis that was considerably larger in the ADAR1-KD tumors (Figure (Figure2E).2E). Explanted tumors maintained their original ADAR1 expression level (Figure (Figure2F).2F). There were no statistically significant differences between the rescue-P110 and rescue-P150 cells. The in vitro and in vivo observations (Figure (Figure2)2) concur and strongly suggest that the downregulation of ADAR1 observed in metastatic melanoma promotes tumor growth in vivo by facilitation of cell division and proliferation and not by enhancing resistance to apoptosis.
A comparative whole-genome expression microarray analysis of ADAR1-KD and control cells revealed 702 differentially expressed genes (DEGs) (Supplemental Table 1). Ten DEGs with key roles in cancer development were validated at the RNA level, and 2 were further verified at the protein level (Supplemental Table 2 and Supplemental Figure 6A). Analysis of the gene clusters showed that “cell growth and proliferation”, “morphology”, and “cell cycle” were among the top 5 most significantly altered clusters (Figure (Figure3A).3A). This concurs with the phenotype demonstrated above (Figure (Figure2).2). However, RNA-editing site analysis (15) in the DEGs revealed that only 5% (35/702) include putative A-to-I editing sites (Supplemental Table 1), similar to the predictions for the total genome (16). We therefore hypothesized that ADAR1-mediated regulation of the cancerous phenotype is mostly independent of RNA editing.
Several His-tagged ADAR1-modified constructs were generated: ADAR1-P150 and ADAR1-P110 bearing the previously described point mutations in the catalytic site H910Y-E912A (17) (CAT-MUT-P150 and CAT-MUT-P110) or devoid of the deamination domain (ΔCAT-P150 and ΔCAT-P110) and isolated catalytic domain (CAT), dsRNA binding domains (dsRBD), or Z-DNA binding domains (Z-DBD-P150 and Z-DBD-P110) (Figure (Figure3B).3B). 624mel cells were stably transfected with all the various constructs or with an empty vector (mock). All His-tagged constructs were expressed at similar levels, as confirmed by intracellular staining with anti-His antibodies (Figure (Figure3C)3C) or with Western blot (data not shown). In addition, all transfectants were tested for endogenous ADAR1 expression level, RNA-editing capacity, and proliferation rate. Mock cells served as negative control and the ADAR1-P150 and ADAR1-110 cells served as positive controls. Importantly, endogenous ADAR1 expression and RNA-editing rate of BLCAP remained similar in all transfectants, except for positive controls (Figure (Figure3,3, D and E). Concurring with the KD experiments, overexpression of ADAR1-P150 and ADAR1-P110 inhibited melanoma cell proliferation (Figure (Figure3F).3F). Supporting our hypothesis, a similar inhibitory effect was indeed observed in the CAT-MUT-P150, CAT-MUT-P110, ΔCAT-P150, and ΔCAT-P110 cells (Figure (Figure3F).3F). Inhibition of proliferation was corroborated by accumulation of cells with G1 DNA content in cell-cycle analysis, indicative of reduced proliferation (Figure (Figure3G).3G). In addition, ADAR1-P150, ADAR1-P110, CAT-MUT-P150, CAT-MUT-P110, ΔCAT-P150, and ΔCAT-P110 transfectants exhibited similar expression of selected genes (Supplemental Figure 6B). In conclusion, these experiments show that ADAR1 regulates gene expression and proliferation of melanoma cells independently of RNA editing.
The DEGs could not be ascribed to single specific pathways, thus implying that ADAR1 may operate via broad cellular regulators, such as miRNAs. Comparative miRNA expression profiles of ADAR1-KD and control cells demonstrated that the expression of 131 miRNAs (Supplemental Table 3) was significantly changed (defined as >2- or <0.5-fold change), of which 53% (70/131) were already known as oncogenic or tumor-suppressive miRNAs (Supplemental Table 3). Remarkably, KD of ADAR1 increased the expression of 89% (25/28) of the known oncogenic miRNAs and reduced 86% (36/42) of the tumor suppressor miRNAs (Figure (Figure4A).4A). Bioinformatics did not reveal statistically significant differences in the predicted structure of these groups of miRNAs (18, 19). Targets for the 131 miRNAs were predicted with TargetScan 5.1 (20) and crossed with the 702 DEGs. Since miRNAs are expected to suppress gene expression, we disregarded all predicted target genes that were not expressed inversely to the corresponding miRNA. This analysis suggested that the 131 miRNAs target at least 38% of the DEGs (264/702) (Supplemental Table 1).
In order to test the effects of ADAR1-regulated miRNAs on cell features, we focused on 2 representative miRNAs in ADAR1-KD cells: miR-21, an oncogenic miRNA (21) that was upregulated (Figure (Figure4B),4B), and miR-34a, a tumor-suppressive miRNA (22) that was downregulated (Figure (Figure4B).4B). ADAR1-KD cells were transiently transfected with anti–miR-21 oligonucleotide or transfected with miR-34a–encoding vector. The appropriate control oligonucleotides or empty vector, respectively, served as negative controls. Reduction in miR-21 expression or enhanced miR-34a expression caused a remarkable decrease in proliferation rate of ADAR1-KD cells (Figure (Figure4,4, C and D). The expression of ADAR1 remained unchanged following these manipulations (Figure (Figure4,4, E and F). In conclusion, ADAR1-controlled expression of miRNAs regulates melanoma cell proliferation.
ADAR1 alters the expression of many miRNAs (Supplemental Table 3), which might shape the metastatic phenotype. It was previously shown that RNA editing of specific pri-miRNA transcripts by ADAR1 affects their biogenesis (11). Here, we hypothesized that ADAR1 regulates miRNA biogenesis independently of RNA editing. So far, such a mechanism was known specifically for miR-376a2 by ADAR2 (13).
We evaluated the expression of pri-miRNA, pre-miRNA, and mature miRNA of miR-21 and miR-34a. The expression of the mature miR-21 was significantly increased in ADAR1-KD cells, while the expression of the pre- and pri-miRNA transcripts was extremely low, as compared with control cells (Figure (Figure5A).5A). The rescue-P110 and rescue-P150 cells displayed an accumulation of either the pri-miRNA or the pre-miRNA transcripts, respectively, with subsequent decrease in the mature miR-21 expression (Figure (Figure5A).5A). miR-34a presented the exact opposite response, as the expression of the mature miRNA decreased in response to KD of ADAR1 expression (Figure (Figure5B).5B). These results strongly suggest that ADAR1 isoforms affect the processing of miRNAs at different steps that coincide with their previously reported subcellular localization. Indeed, ADAR1-110 and Drosha are expressed in the nucleus, while ADAR1-P150 and Dicer are expressed in the cytoplasm (23).
The effects of ADAR1 on miRNA biogenesis were evaluated in the ADAR1 “truncations” system (Figure (Figure3).3). As expected, overexpression of full-length ADAR1-P110 or ADAR1-P150 similarly decreased the expression of mature miR-21 and increased the expression of mature miR-34a (Figure (Figure5,5, C and D). However, in line with the results described above, overexpression of ADAR1-P110 resulted in markedly increased pri-miR-21 and decreased pri-miR-34a, while overexpression of ADAR1-P150 affected the corresponding pre-miRNA forms, accordingly (Figure (Figure5,5, C and D). Importantly, the ΔCAT-P110, ΔCAT-150, CAT-MUT-P110, and CAT-MUT-P150 cells exerted exactly the same effects as their wild-type counterparts (Figure (Figure5,5, C and D), thus attesting that the regulation of miRNA biogenesis by ADAR1 is mediated mostly via an RNA editing–independent mechanism. The isolated dsRNA-binding domain or Z-DNA–binding domain did not have any significant effect (Figure (Figure5,5, C and D). These observations were confirmed in Northern blot analyses for mature miRNAs 21 and 34a, as well as for the pre-miR-21 (Figure (Figure5,5, E and F). The pre-miR-34a was only moderately altered in Northern blot analyses, probably owing to the lesser sensitivity of this method.
These observations suggest that ADAR1 controls the miRNA processing machinery itself. The effect of ADAR1 on 3 key proteins involved in miRNA processing, DGCR8 (DiGeorge syndrome critical region gene 8; ID 54487), Drosha and Dicer, was tested. KD of ADAR1 had no effect on these components at the mRNA level (Figure (Figure6A).6A). However, while Drosha and DGCR8 were unaffected also at the protein level, Dicer protein expression was increased in the ADAR1-KD cells compared with the control and the rescue cells (Figure (Figure6B).6B). This implies on regulation of Dicer at the translation level. Indeed, let-7a, -b, -d, and -e (Figure (Figure6C),6C), which were significantly downregulated following ADAR1-KD (Supplemental Table 3), are known to cause translational repression of Dicer (24–28).
It was previously reported that ADAR1 can bind to other RNA-binding proteins (29–31). Thus, HEK 293T cells were transfected with ADAR-P110 or with an irrelevant protein, CEACAM1. Immunoprecipitation of ADAR1 or CEACAM1 was performed, with both proteins efficiently pulled down (Figure (Figure6D).6D). Remarkably, only DGCR8 was coimmunoprecipitated with ADAR1, but not Drosha or Dicer (Figure (Figure6D).6D). Since DGCR8 and Drosha form the active pri-miRNA processing complex (microprocessor) in the nucleus (32), this observation suggests that the complex DGCR8-ADAR1 is mutually exclusive with DGCR8-Drosha, thus unraveling a potentially novel layer of regulation of the Drosha complex. None of these proteins were coimmunoprecipitated with CEACAM1 (Figure (Figure66D).
We further tested the ability of ADAR1 to directly bind pri-miRNAs. HEK 293T cells were transiently transfected with ADAR1 or CEACAM1, along with pri-miR-21 or pri-miR-34a. 293T cells were used in these experiments due to the relatively low transfection efficiency into melanoma cells. Western blot confirmed strong expression of ADAR1 or CEACAM1 in the total cell lysates as well as in the immunoprecipitates following specific protein pull down (Figure (Figure6E).6E). Fold expression of pri-miR-34a and pri-miR-21 was quantified with qPCR, normalized to GAPDH, and determined relative to the parental cells. pri-miR-34a was strongly overexpressed with the cotransfected ADAR1 and similarly with the cotransfected CEACAM1 (Figure (Figure6E).6E). pri-mir-21 was enhanced in a mild manner with both proteins (Figure (Figure6E).6E). The presence of pri-miR-21 and pri-miR-34a in RNA extracted from the immunoprecipitate of ADAR1 or CEACAM1 was tested using qPCR. A specific signal was obtained for pri-miR-34a in the ADAR1 but not CEACAM1 immunoprecipitate, strongly implying on direct binding (Figure (Figure6E).6E). A similar observation for pri-miR-21 could not be made, suggesting that there is no direct binding in this case (Figure (Figure6E).6E). Alternatively, this might be due to significantly weaker expression of pri-miR-21 than pri-miR-34a (Figure (Figure6E).6E). When these experiments were performed on untransfected cells, no signal could be observed (data not shown), probably owing to the minute amounts of RNA extracted from the immunoprecipitates and the overall low sensitivity of the assay. Quantitative analysis is impossible due to the lack of a validated normalizer.
Taken together, our data suggest that ADAR1 affects the biogenesis of miRNAs at various points along the process directly by potentially affecting Drosha complex and indirectly by regulating Dicer via let-7 or by binding to various pri-miRNAs.
miR-17-5p is part of the miR-17-92 cluster, which plays an oncogenic role in other malignancies (33), but has not been studied in melanoma. miR-432 has never been investigated before. A putative binding site for miR-17-5p and 2 sites for miR-432 were identified in the 3′ UTR of ADAR1 (Supplemental Figure 7A). Preliminary analysis of 2 melanoma cell lines, C8161 highly aggressive (HAG) and the C81-61 poorly aggressive (PAG) cell lines (22), showed that ADAR1 was expressed 8-fold higher in the PAG cells, while miR-17-5p and miR-432 were expressed 5.5-fold and 20-fold higher in HAG cells (Supplemental Figure 7B).
A portion of ADAR1 3′ UTR containing the putative binding site for miR-17-5p (UTR) was cloned downstream to Renilla luciferase in a dual luciferase reporting system. The putative binding site was altered with 3 point mutations (UTR-MUT). miR-17-5p was cloned into the pQCXIP expression vector. Empty psiCheck2 (NO-UTR) and pQCXIP (mock) served as negative controls. The various constructs were cotransfected into C81-61 cells (PAG), which express moderate endogenous levels of miR-17-5p (Supplemental Figure 7). The luciferase signal of cells cotransfected with both empty vectors served as point of reference. Forced expression of miR-17-5p with the UTR construct significantly inhibited the luciferase signal, while the inhibitory effect was abolished when the UTR-MUT construct was tested (Figure (Figure7A).7A). This suggests that miR-17-5p binds directly to the 3′ UTR of ADAR1. In order to study the role of endogenous miR-17-5p, C8161 (HAG) cells, which express high levels of miR-17-5p (Supplemental Figure 7), were transiently transfected with anti–miR-17-5p oligonucleotides or with control oligonucleotides. miR-17-5p was efficiently silenced (Figure (Figure7B),7B), while the expression of ADAR1 was significantly enhanced both at RNA (Figure (Figure7C)7C) and protein (Figure (Figure7D)7D) levels. Both ADAR1-P150 and ADAR1-P110 were enhanced (Figure (Figure7D).7D). Functionally, cell proliferation rate (Figure (Figure7E)7E) as well as RNA-editing rate (data not shown) were inhibited significantly.
A similar approach was implemented for miR-432. An ADAR1 3′ UTR fragment that includes both putative sites for miR-432 was cloned. Constructs containing separately mutated sites (UTR-mutA and UTR-mutB) or both (UTR-mutAB) were tested as described above in PAG cells, which exhibit moderate miR-432 expression (Supplemental Figure 7). Transfection of miR-432 with UTR demonstrated a clear decrease in luciferase signal (Figure (Figure7F),7F), attesting for the direct binding of miR-432 to 3′ UTR of ADAR1. The presence of both intact binding sites is required for effective inhibition (Figure (Figure7F).7F). miR-432 was expressed in similar levels with all reporting constructs (data not shown). Effective silencing of miR-432 in HAG cells (Figure (Figure7G)7G) enhanced ADAR1 expression both at RNA (Figure (Figure7H)7H) and protein levels (Figure (Figure7I)7I) and accordingly inhibited cell proliferation (Figure (Figure77J).
The cofunction between miR-17-5p and miR-432 was studied by transfecting PAG cells with each miRNA separately or simultaneously, with half the amount of each miRNA (Figure (Figure7K).7K). The simultaneously transfected cells indeed overexpressed the miRNAs at lower levels than the cells transfected with each individual miRNA (Figure (Figure7K).7K). Nevertheless, ADAR1 expression was similarly downregulated at the RNA (Figure (Figure7L)7L) and protein (Figure (Figure7M)7M) levels in all transfectants. Accordingly, a similar enhancement in net proliferation was observed (Figure (Figure7N).7N). Taken together, it is shown that miR-17-5p and miR-432 are direct endogenous regulators of ADAR1, which cofunction in an additive manner.
Expression analysis of both miRNAs in 26 patient-derived low-passage metastatic melanoma cultures and 4 normal melanocyte cultures showed that about 30% of melanoma cases significantly upregulated miR-17-5p expression, while more than 95% strongly upregulated miR-432 (Figure (Figure8A).8A). Both miRNAs exhibited a significant inverse correlation with ADAR1 expression (miR-17-5p: P = 0.0013, R = –0.58; miR-432: P = 0.0014, R = –0.55) (Figure (Figure8B).8B). There was no correlation between ADAR1 and 6 other miRNAs (miR-20a, -133a, -184, -185, -31, and -204) (Supplemental Figure 8). These observations hint of the physiological relevance of miR-17 and miR-432 in cancer development. Indeed, chromosomal copy number analysis in 12 metastatic melanoma cell lines and 3 normal cell samples (melanocytes, HUVEC, and keratinocytes), showed a very common and significant increase both in miR-17-92 cluster (60%, 3–6 copies) and in miR-432 (92%, 3–7 copies) (Figure (Figure8C).8C). Thus, genomic copy number amplification of ADAR1-repressive miRNAs may account for downregulation of ADAR1 in cancer cells. However, although genomic copy numbers correlated with miRNA expression (miR-17-5p: P = 0.015, R = 0.68; miR-432: P = 0.0008, R = 0.81), the significant differences between miR-17-5p and miR-432 (Figure (Figure8A)8A) both in percentage and intensity suggest the existence of another, differential mechanism.
miR-432 is encoded in chromosome 14 in the DLK1-DIO3 region (Supplemental Figure 9), which has been shown to be imprinted (34), and was therefore suspected to be regulated epigenetically. Quantification of genomic methylation in more than 10 sites in RTL1 comparing 12 melanoma cell lines and 3 normal cell samples confirmed a robust hypomethylation in this region in the cancer samples (Table (Table1).1). Moreover, a striking hypomethylation was observed within the sequence of miR-432 and upstream of its genomic location (Table (Table1).1). The effect is site specific, as some sites in both RTL and miR-432 remained hypermethylated among both normal and cancer samples (Table (Table1).1). Methylation indeed accounts for repression of miR-432 expression because it was increased in a time-dependent manner in 2 different melanoma cell lines tested, A375 and MeWo, when exposed to epigenetic modifiers 5′-aza-2-deoxycitidine and 4-phenylbutyric acid (PBA) (Figure (Figure8D).8D). Demethylation was clearly demonstrated in sites no. 1 and no. 8 in the miR-432 region (Table (Table2).2). There were no predicted methylation sites in the region of the miR-17-92 cluster, and indeed, the expression of miR-17-5p remained constant (Figure (Figure8E).8E). Finally, concurring with the robust upregulation in miR-432 expression, a drastic downregulation of ADAR1 expression was observed after treatment with demethylating agents (Figures (Figures8,8, F and G). These combined results highlight the imprinted RTL1 region as a target for hypomethylation in cancer, which in turn controls ADAR1 expression via hypomethylation of miR-432.
The role of ADAR1 as an RNA-editing enzyme is well established (12). Many studies emphasize the significance of A-to-I RNA editing in regulation of gene expression (35), viral host defense (36), embryonic development (37), miRNA processing (38), alternative splicing (1), and the function of several receptors located in the CNS (39). Lower RNA-editing rates were described in various human cancers (3, 40); however, whether reduced editing rate facilitates cancer or whether it is a consequence of another process remained obscure. Here, we present what we believe to be novel mechanistic evidence for the roles of ADAR1 in regulation of cancerous features and for how it is lost in metastatic melanoma cells to facilitate the acquisition of an aggressive phenotype.
Metastatic melanoma cells exhibit a common and significant downregulation of both ADAR1-P110 and ADAR1-P150 expression as compared with normal melanocytes, nevi, and primary melanoma tumors (Figure (Figure1).1). Therefore, it seems that loss of both isoforms is associated with the metastatic transition and could be explained by the upregulation of miR-17 and miR-432 in metastatic cells (Figure (Figure8),8), which target them both (Figure (Figure7).7). Series of experimental ADAR1 manipulations in melanoma cell lines demonstrate that ADAR1 controls the expression of hundreds of genes and fundamentally suppresses their malignant phenotype in vitro and in vivo (Figures (Figures22–4). It should be noted that a weak, selective induction of the inducible ADAR1-P150 was observed only in nevi and primary tumors as compared with normal epidermis (Figure (Figure1).1). Given the suppressive features of ADAR1-P150 (Figures (Figures22 and and3),3), this could hint of its potential role as an inherent resistance mechanism to melanocyte transformation. Another melanoma suppressive protein, IGFBP7, which promotes apoptosis and senescence, operates in nevi bearing the BRAFV600E mutation and is then silenced by methylation in metastatic cells (41). It should be noted that an opposite role for ADAR1-P150 was reported in HeLa cells (42) and that these differences might be due to cell-specific miRNA and target gene expression profiles. In addition, an upregulation of ADAR1 was observed in some other malignancies, such as lobular breast cancer (43) and B cell acute lymphoblastic leukemia (44).The robust dominance of ADAR1-P110 probably dictates the final cellular outcome. The corroboration between the robust cell regulatory roles of ADAR1 with its common low expression in melanoma metastasis strongly points to a central involvement of ADAR1 in melanoma progression.
We hypothesized that the diffuse DEG profile observed following experimental KD of ADAR1 suggests that ADAR1 regulates miRNAs, broad-acting cellular mediators. Indeed, KD of ADAR1 altered the expression of 131 miRNAs (Supplemental Table 3), many of which are known cancer regulators (Supplemental Table 3), and target 38% of the DEGs (Supplemental Table 1). The rest of the DEGs, which are not predicted to be targeted by the ADAR1-controlled miRNAs, might be secondarily altered. Correction of exemplar upregulated miR-21 (by silencing) or of downregulated miR-34a (by forced expression) partially reversed the enhanced proliferation of ADAR1-KD cells (Figure (Figure4).4). This attests for the mechanistic role of ADAR1-controlled miRNAs in determining the phenotype of cancer cells. Interestingly, most altered miRNAs with known oncogenic or tumor-suppressive properties were upregulated or downregulated, respectively, following ADAR1-KD (Figure (Figure4).4). However, there were no structural differences among the groups, based on bioinformatics predictions (18, 19), which could explain this phenomenon. Further, crossing of TargetScan 5.1 (20) predicted targets for miR-21 or miR-34a with downregulated or upregulated DEGs, respectively, indicated that the increase in miR-21 was potentially responsible for the downregulation of 7 genes, 6 of them by 3.3- to 10-fold, while the decrease in miR-34a was potentially responsible for the upregulation of 10 genes, 4 of them by 4- to 9-fold (Supplemental Table 5). In light of these observations, the differences between the herein-reported DEGs in human melanoma cells and the interferon-related DEG pattern in hematopoietic cells reported in ADAR1-knockout mice (45) are not surprising, as they can be attributed to the different cell types, which endogenously express different sets of miRNAs.
A-to-I RNA editing of miRNAs has been shown to either lead to recognition of new targets due to altered seed sequence or to maturation defects due to dysregulated processing (46–48). Recently, 2 comprehensive bioinformatics studies revealed only a limited number of editing events in mature miRNA sequences (49, 50). These reports, combined with the robust alteration of more than 130 miRNAs following ADAR1-KD, further support our argument that ADAR1 regulates both cell proliferation and miRNA processing independently of RNA editing. Indeed, the effects exerted by ADAR1 bearing a mutated or truncated catalytic deamination domain (CAT-MUT or ΔCAT, respectively) were similar to the full ADAR1 enzymes (Figures (Figures33 and and5).5). While the effects of ADAR1 on miRNA processing, expression, and thereby functions were independent of RNA editing, they did depend on RNA-binding capacity. We show that both the dsRBD, essential for binding to the target RNA molecule and/or other RNA-binding proteins (51), and the Z-DBD, necessary for ADAR1 effects on NF-90 activity (30), nuclear export, and binding to specific sequences based on their conformation (52, 53), are required. ADAR1 truncation mutants lacking the dsRBD or ZDBD failed to exert any impact either on cellular proliferation or miRNA biogenesis (Figures (Figures33 and and5).5). In addition, we demonstrated that ADAR1 is capable of directly binding pri-miRNAs (Figure (Figure6E).6E). RNA editing–independent roles of ADAR1 were previously reported, for example, by creating specific protein-protein complexes with NF-90 (30), decreasing PKR kinase activity at the translation level (31). In addition, it was reported that ADAR2 regulates the processing of miR-376a2 independently of RNA-editing activity (13). Our findings expand the known scope of ADAR1-mediated regulation of miRNA biogenesis by direct RNA editing of specific pri-miRNAs (23, 48), and postulate a broad regulatory scope for ADAR1, which operates via systematic control of the miRNA-processing machinery.
A systematic regulation of the miRNA processing machinery by ADAR1 is supported by the specific effect of ADAR1-P110 and ADAR1-P150 on the processing of the pri- and pre-miRNA precursors, respectively, which ultimately have the same end result on mature miRNA expression (Figure (Figure5).5). Several mechanisms might be involved here: (a) ADAR1 creates complexes with DGCR8 as they are coimmunoprecipitated (Figure (Figure6D);6D); this might occur via their dsRNA-binding domains, as previously reported for ADAR1 and Exportin-5 or NF-90 (29, 54); the lack of ADAR1-Drosha complexes (Figure (Figure6D)6D) suggests that the formation of DGCR8-ADAR1 and DGCR8-Drosha complexes are mutually exclusive; thus ADAR1 could affect the availability of DGCR8 for Drosha, thereby altering the efficiency of the microprocessor in general; (b) ADAR1 regulates Dicer expression at the translation level (Figure (Figure6B).6B). This probably occurs via Let-7 miRNAs, which have been recently shown to regulate Dicer at the translation level (24–28) and are strongly controlled by ADAR1 (Supplemental Table 3); (c) ADAR1 can directly bind some miRNA precursors and compete over the dsRNA substrates with the enzymatic processors; (d) the expression of SMAD1 and NF-90, which were recently shown to regulate microprocessor activity (23), is also controlled by ADAR1 (Supplemental Tables 1 and 2) in a yet to be defined mechanism. Collectively, these results imply that the previously described cancer-associated diminished RNA editing (3) probably reflects the mere downregulation of ADAR, although a contribution of RNA editing to the malignant phenotype of cancer cannot be entirely excluded.
Finally, we identified miR-17-5p and miR-432 as the direct, independent, endogenous cellular regulators of ADAR1. Together, these miRNAs regulate ADAR1 expression in an additive manner (Figure (Figure7).7). Our findings suggest that loss of ADAR1 in cancer cells and the subsequent increase in malignant features is mediated by overexpression of these miRNAs (Figure (Figure8).8). miR-17-5p, an miRNA with known oncogenic properties, is overexpressed in various cancers, mainly in hematopoietic malignancies (33), due to amplification of miR-17HG, a noncoding genomic segment that contains the miR-17-92 cluster (13q31.3) (55). In contrast, there are only scarce data on miR-432.
Genomic and epigenetic mechanisms account for the overexpression of miR-17-5p and miR-432 in melanoma cells. We show that amplification of the genomic segment encoding miR-17-5p frequently occurs in melanoma, thereby facilitating the malignant phenotype by directly targeting ADAR1 (Figures (Figures77 and and8).8). The overexpression of miR-432 in melanoma occurs due to frequent genomic amplification and aberrant hypomethylation patterns of the DLK1-DIO3 locus in chromosome 14 (Figure (Figure88 and Table Table1).1). Cancer development and progression involve aberrant methylation in general, with most studies focusing on hypermethylation of tumor-suppressive genes (56). Epigenetic regulation of miRNAs has started gaining the focus of cutting edge research very recently (57). The regulation of miR-432 by hypomethylation is novel for the following reasons: (a) the miR-432 gene is found within the DLK1-DIO3 imprinted domain (58), which has been poorly studied in cancer; (b) one of the methylation sites is found within the sequence of the miR-432, which represents the first such example. The mechanisms by which the imprinting patterns of the DLK1-DIO3 region are destroyed in cancer could include, for example, DNA methyl transferase dysregulation or acquisition of somatic mutations in methylation sites and should be the focus of future investigations.
The ADAR1-dependent and RNA editing–independent roles presented here as well as the miRNA-mediated mechanisms of ADAR1 downregulation are probably applicable to additional types of cancer. These findings could be translated into innovative lines of diagnosis and therapy in a broad range of malignancies.
The melanoma lines 624mel 526mel, C8161 (HAG) and C81-61 (PAG), SKmel2, SKmel24, MalMe3M, G361, WN115, MeWo, A375, WM266-4, HEK293T cell lines (ATCC), normal human epidermal melanocytes (PromoCell) and normal melanocyte cultures hmel-p-16, Nohm-4 and Hermes-2B were maintained as previously described (22, 59–61). The 34 primary cultures derived from surgically removed metastatic melanoma specimens were established and cultured as described (59). Stably transfected cell lines were cultured with 1 μg/ml puromycin (Calbiochem) or 2 mg/ml G418 (Alexis Biochemicals)
The following primary antibodies were used: PE-His (Miltenyi Biotec), annexin V–FITC (Bender Medsystems), Ki67 (DAKO), ADAR1 (Sigma-Aldrich), β-actin (MP Biochemicals), Dicer (Abcam), Drosha (Abcam), DGCR8 (Abcam), and HRP-conjugated secondary antibodies (Jackson Immunoresearch).
Total RNA was isolated with Tri Reagent (Sigma-Aldrich) according to the manufacturer’s instructions. cDNA was synthesized using high-capacity or Taq-Man miRNA reverse transcription kit (Applied Biosystems).
qPCR was performed using SYBR Green PCR Master Mix (Applied Biosystems) with gene-specific primers (listed in Supplemental Table 4) or Taq-Man Universal Master Mix (Applied Biosystems) with custom primer/probe mixtures (Applied Biosystems) according to the manufacturer’s instructions. The real-time PCR (qPCR) reactions were normalized to GAPDH, HPRT, or U6 endogenous control as reported previously (60).
Genomic DNA (gDNA) was extracted from melanoma using a Wizard genomic purification kit (Promega) according to the manufacturer’s instructions. Expression analysis was performed using TaqMan Copy Number Variation Assay (Applied Biosystems) along with custom ordered probes (Applied Biosystems). The qPCR results were normalized to 2 TaqMan copy number reference assays: TERT and RNase-P (Applied Biosystems).
The expression systems used in this work were pSuper.puro, pCDNA3.neo, pQCXIP.puro, and psiCheck2. Identification of the various primers that were designed for cloning and introduction of mutations is available in Supplemental Table 4. Transfections were performed with Turbofect (Fermentas) according to the manufacturer’s instructions. Retroviral transductions were performed as previously described (22). Site-directed mutagenesis was performed using QuikChange kit (Stratagene) according to the manufacturer’s instructions.
The LNA anti-miRNA oligos (unmodified) along with the proper control oligos (miRcury LNA, Exiqon) were used for miRNA silencing. The various oligos were transiently transfected with Turbofect (Fermentas) and the cells were tested for miRNA expression 48 hours after transfection.
Lysates of 5 × 106 cells were analyzed by SDS-PAGE. Western blot for ADAR1 and actin with specific antibodies, which was developed with standard ECL reaction, was performed using standard protocols.
PCR amplifications and analysis of editing levels were carried out as described (61).
Cells (5 × 103) were seeded in triplicate in 96-well flat-bottom microplates, and designated wells were harvested daily for 72 hours. Net proliferation was tested via a standardized XTT (Biological Industries) colorimetric assay, as previously described (22). Standard curve was performed individually for each treatment in all experiments.
Staining for extracellular and intracellular antigens was performed according to standard protocols, as reported previously (60). Cell-cycle analysis was performed on 1 × 106 cells with standard overnight 70% ethanol fixation, permeabilization with 0.05% Triton X-100 (Sigma-Aldrich),and PI staining. All experiments were performed using a FACSCalibur instrument (BD Biosciences), and data were analyzed using FlowJo software (Tree Star Inc.).
Eight-week-old male SCID-NOD mice were used for xenograft studies. Approximately 5 × 106 viable tumor cells were resuspended in 100 μl PBS solution and injected subcutaneously. Mice were monitored 3 times a week for tumor volume by caliper measurements (longest diameter × shortest diameter2 × 0.52). On termination of each experiment, the tumors were extracted, documented, weighed, and preserved. Immunostaining was performed on the formalin-preserved tumors, and total RNA was extracted from the cryopreserved tumors according to standard procedures.
Immunohistochemical staining was performed on 5-μm sections of paraffin-embedded tissues or progression TMA slides (see below) according to standard procedures. Tissues were stained for Ki-67 or ADAR1, followed by hematoxylin (Sigma-Aldrich) counterstaining. Isotype-matched or preimmune rabbit serum served as negative control. The percentage of Ki-67–positive cells was determined by an expert pathologist (I. Barshack), who was blinded to the experimental groups.
TMA slides were provided by the NCI CDP and included 66 benign nevi, 90 primary tumors, and 74 metastases. Another TMA set was provided by the NCI SPORE program and included 66 melanocytic nevi, 66 primary melanomas, and 75 metastases. For each sample, intensity of ADAR1 expression (nuclear and cytoplasmic) was scored from 0 (negative) to 3, and percentages of expression were defined as 0 to 3 for 0%–5%, 6%–25%, 26%–75%, and 76%–100%, respectively. Cells were examined over superficial and deep dermal foci in primary melanomas and nevi in order to detect possible “maturation” changes within the lesions.
C81-61 (PAG) cells were cotransfected with 1 μg of psiCheck2-ADAR1 3′ UTR (UTR), psiCheck2-ADAR1 mutated 3′ UTR (UTR-MUT), or psiCheck2-empty vector (No-UTR) and 0.1 μg of the pQCXIP-miR-17 (miR-17-5p) or pQCXIP-empty vector (mock) as control. HEK 293T cells were cotransfected with 1 μg of psiCheck2-ADAR1 3′ UTR plasmids (UTR), different psiCheck2-ADAR1 mutated 3′ UTR (UTR-mutA, UTR-mutB, and UTR-mutAB), or psiCheck2–empty vector (No-UTR) and 0.1 μg of the pQCXIP-miR-432 (miR-432) or pQCXIP–empty vector (mock) as control. Cells were harvested 48 hours after transfection and assayed with Dual Luciferase Reporter Assay System (Promega) according to the manufacturer’s instructions.
Total RNA was extracted and used as template to generate cDNA and subsequent biotinylated target cRNA that was processed by an Affymetrix GeneChip Instrument System (Affymetrix) according to the manufacturer’s recommendations ( http://affymetrix.com/support/technical/manual.affx). The DEGs were analyzed by Ingenuity Pathway Analysis ( http://www.ingenuity.com), while the miRNAs were analyzed using Partek Genomic Suite (Partek Inc.) and TargetScan 5.1 (Whitehead Institute for Biomedical Research). Full microarray data are deposit in NCBI GEO archives (GSE31963).
MC-Fold (62) was applied for prediction of miRNA secondary structure and ΔG values.
gDNA of melanoma cell lines was extracted as described above. Bisulfate treatment was carried out on gDNA using EZ-DNA Methylation Gold Kit (Zymo) according to the manufacturer’s instructions. Next, PCR was performed using bisulfate-treated DNA as a template with specifically designed primers (Supplemental Table 4).
Following PCR amplification, a shrimp alkaline phosphate (SAP) treatment, in vitro transcription, and RNaseA cleavage (MASScleave) were performed. The samples were purified by resin treatment, spotted on 384-well spectroCHIP, and analyzed by a spectral acquisition on a MASSarray Analyser (Sequenom). The average methylation ratio for each CpG position was calculated by averaging the ratios of the melanoma versus normal samples. Non-template control (NTC) sample used as negative control (for details see Supplemental Methods).
Melanoma cell lines A375 and MeWo were treated for 2, 4, and 6 days with 3 μM anti-methylation reagent 5′-Aza aza-2-deoxycitidine (Sigma-Aldrich) dissolved in 50% acetic acid and 3 mM anti-acetylation agent (PBA) (Sigma-Aldrich) dissolved in 50% EtOH. Control samples were treated with medium containing the same volume of dissolvent. The growth medium containing these reagents or dissolvent only was changed daily. The treated cells and the control samples were harvested at days 2, 4, and 6. RNA, cell lysates, and gDNA were extracted from the cells and analyzed, as previously mentioned.
HEK 293T transfected cells were harvested and total cell lysates were extracted and assayed following incubation with either anti-ADAR1 or anti-CEACAM1 with Dynabeads Protein-G beads (Sigma-Aldrich) according to the manufacturer’s instructions. Both total cell lysates and immunoprecipitates were analyzed by Western blot according to standard protocols. Total RNA was isolated and analyzed by qPCR as described above (for details see Supplemental Methods).
RNA samples were electrophoresed, transferred, and hybridized as previously described (63) using P32-labeled probes. The oligonucleotides used as probes are the complementary sequences of the mature miRNA (miRbase Registry). U6 was used to normalize expression levels (for details see Supplemental Methods).
Melanoma ADAR1-KD, ADAR1-rescue-P110, ADAR1-rescue-P150, and control cells (104 cells/well) were seeded in 96-well plates and untreated (control) or treated with either 100 μM cisplatin (Abiplatin; ABIC) or 50 μM taxol (Ebetaxel; Ebewe Pharma). Cells were collected 48 hours after treatment, washed with PBS, and stained with both annexin V–FITC and PI according to the manufacturer’s instructions (eBioscience). Apoptosis rate was further determined by FACSCalibur instrument (BD Biosciences), and data analysis was performed using FlowJo software (Tree Star Inc.).
Data were analyzed using the unpaired 2-tailed Student’s t test, the 1-way ANOVA test, and Fisher’s exact test. Correlations were examined with Pearson’s correlation test. P ≤ 0.05 was considered significant.
Animal studies were approved and performed in accordance with the Institutional Review Board (no. 504/2009), Sheba Medical Center. The primary melanoma specimens were obtained according to Israel Ministry of Health approval no. 3518/2004 (ClinicalTrails.gov Identifier NCT00287131). NCI CDP TMA slides were provided by the NCI CDP. Other investigators may have received slides from these same array blocks. The NCI SPORE TMA used in these studies was generated with the support of M.D. Anderson SPORE in skin cancer (NIH p50CA093459).
G. Markel is supported by the Israel Science Foundation (grant 1489/10) and the Carol Fingerut Foundation for Cancer Research (grant 2008). D.M. Kallenberg is supported by Wellcome Trust (grant 078327). G. Rechavi is supported by the I-CORE Program of the Planning and Budgeting Committee and the Israel Science Foundation (grant no. 41/11.1), the Flight Attendants Medical Research Institute (FAMRI), the I-CORE Program of the Planning and Budgeting Committee and the Israel Science Foundation (grant no. 41/11.1), and the Israel Ministry for Science and Technology (Scientific Infrastructure Program, grant 2010). G. Rechavi holds the Djerassi Chair in Oncology, Tel Aviv University. E. Galun is supported by an I-CORE ISF grant (grant no. 41/11.1) and an SFB DFG grant (grant no. 841). The authors would like to thank Michael Aronson for his support in this work. This work was performed in partial fulfillment of the requirements for a Ph.D. degree for Yael Nemlich, Sackler Faculty of Medicine, Tel Aviv University.
Conflict of interest: The authors have declared that no conflict of interest exists.
Citation for this article: J Clin Invest. 2013;123(6):2703–2718. doi:10.1172/JCI62980.