Alu-dependent RNA editing of GLI1 promotes malignant regeneration in multiple myeloma

Despite novel therapies, relapse of multiple myeloma (MM) is virtually inevitable. Amplification of chromosome 1q, which harbors the inflammation-responsive RNA editase adenosine deaminase acting on RNA (ADAR)1 gene, occurs in 30–50% of MM patients and portends a poor prognosis. Since adenosine-to-inosine RNA editing has recently emerged as a driver of cancer progression, genomic amplification combined with inflammatory cytokine activation of ADAR1 could stimulate MM progression and therapeutic resistance. Here, we report that high ADAR1 RNA expression correlates with reduced patient survival rates in the MMRF CoMMpass data set. Expression of wild-type, but not mutant, ADAR1 enhances Alu-dependent editing and transcriptional activity of GLI1, a Hedgehog (Hh) pathway transcriptional activator and self-renewal agonist, and promotes immunomodulatory drug resistance in vitro. Finally, ADAR1 knockdown reduces regeneration of high-risk MM in serially transplantable patient-derived xenografts. These data demonstrate that ADAR1 promotes malignant regeneration of MM and if selectively inhibited may obviate progression and relapse.

M ultiple myeloma (MM) is a plasma-cell neoplasm that represents the second most common blood cancer in the United States. High-risk forms of this disease have been associated with amplifications at chromosome 1q21, which occurs in over 30% of MM patients and is associated with poor clinical outcomes 1,2 . Notably, the inflammation-responsive RNA editase gene adenosine deaminase acting on RNA-1 (ADAR1) as well as interleukin-6 receptor (IL-6R) localize to this unique IL6R ADAR 100 kb amplifications, compared to no 1q (1q stage I n = 29, stage II n = 39, stage III n = 32; no 1q stage I n = 146, stage II n = 150, stage III n = 133). The error bars represent ± S.E.M. of the mean; **p < 0.01 by two-tailed, Mann-Whitney U test. c IL6R relative expression in primary CD138 + cells from patients harboring 1q amplifications, compared to no 1q. The error bars represent ±S.E.M. of the mean; **p < 0.01 by two-tailed, Mann-Whitney U test. d ADAR1 isoform relative expression in 1q-amplified (1q) versus no 1q patients. The error bars represent ±S.E.M. of the mean; *p < 0.05, **p < 0.01 by two-tailed, Mann-Whitney U test. e Total ADAR1 mRNA levels assessed in total MNCs from primary samples (n = 19; see Table 1). Age-matched (n = 3, mean age = 60.6 ± 16.8 years old) BM collected from patients undergoing hip replacement therapy for reasons other than cancer were used as normal healthy controls. HPRT gene expression was used for normalization. Dots represent mean values for individual patients ± S.E.M (ctrl n = 3; smoldering MM n = 4; newly diagnosed MM n = 4; relapsed MM n = 7; PCL n = 4). **p < 0.01 compared to normal controls by unpaired, two-tailed Student's t-test. f Kaplan-Meier curves for overall survival (OS) of high (n = 162) versus low ADAR1 (n = 159) expressing cohorts in the CoMMpass (IA8) study (n = 643). **p < 0.001 by Cox regression. g Kaplan-Meier curves for progression-free survival (PFS) of high (top 25% ADAR1) versus low ADAR1 (bottom 25% ADAR1) expressing cohorts in the CoMMpass (IA8) study (n = 643), further stratified into 1q or no 1q cohorts. **p < 0.001 by Cox regression. Statistical significance was indicated when p < 0.05. See also Supplementary Fig. 1 chromosome region. ADAR1 edits adenosine-to-inosine (A-to-I) nucleotides, primarily within double-stranded (ds) RNA loops formed by primate-specific Alu repeat sequences 3 . Although ADAR1 p150 is expressed in response to inflammatory cytokine signaling, ADAR1 p110 is constitutively expressed 4 . Proinflammatory cytokine signals derived from the bone marrow (BM) microenvironment have a key role in MM progression and have been correlated with symptom severity and clonal evolution 5 . Moreover, induction of vital microenvironment-responsive stem cell self-renewal pathways, such as NOTCH1 6 , promotes progression of MM. Interestingly, widespread deregulation of inflammation-responsive epitranscriptomic events also contribute to cancer stem cell (CSC) generation and maintenance, which governs cancer progression and drug resistance 7,8 . In addition to recurrent DNA mutations and epigenetic deregulation 9 , RNA editing mediated by cytokine-responsive ADAR1 has emerged as a vital contributor to transcriptome remodeling leading to cancer relapse and progression 7,10-12 , however the contribution of proinflammatory signaling leading to ADAR1-dependent RNA editing in MM pathogenesis has not been previously explored.
To address this, here we performed gene expression analysis in a cohort of primary samples with 1q amplification from the Multiple Myeloma Research Foundation (MMRF) CoMMpass study, coupled with sensitive RNA editing analyses in high-risk primary patient samples that establish robust in vivo disease models and in the setting of acquired drug resistance in vitro. Analyses revealed widespread RNA editing activation in high-risk MM patients with poor overall and progression-free survival, and robust induction of ADAR1 expression and activity in the setting of acquired immunomodulatory drug (IMiD) resistance. In contrast to ADAR1 shRNA knockdown and overexpression of an editase defective ADAR1 mutant, wild-type ADAR1 expression enhances Alu-dependent editing and transcriptional activity of GLI1, a Hedgehog (Hh) pathway transcriptional activator and self-renewal agonist, and promotes lenalidomide resistance in vitro. Finally, lentiviral shRNA ADAR1 knockdown reduces regeneration of high-risk MM in humanized serial transplantation mouse models. Thus, aberrant RNA editing represents a compelling prognostic and therapeutic target that may be exploited to obviate drug resistance in patients with MM and other therapeutically recalcitrant human malignancies.

Results
High ADAR1 expression in multiple myeloma predicts outcomes. Approximately one-third of MM patients harbor amplification of chromosome 1q21 1,2 , which contains ADAR1 and IL-6R loci (Fig. 1a) and is associated with a poor prognosis 13 . Although ADAR1 activation promotes progression in a broad array of malignancies 3,10,14 , its role in MM pathogenesis has not been explored. To investigate if ADAR1 has a role in MM progression and relapse, we analyzed the MMRF CoMMpass database 15,16 . In 1q-amplified patients, ADAR1 levels were higher than in non-1q-amplified MM patients with the highest levels in 1q-amplified, stage III patients (Fig. 1b). In contrast to genes such as IL-6ST and ADAR2 that reside on other chromosomes, IL-6R transcripts were elevated in 1q-amplified stage I MM patients but showed no change with advancing stage ( Fig. 1c; Supplementary  Fig. 1a, b). Isoform-specific analyses revealed that both ADAR1 p150 and p110 were upregulated in 1q-amplified patient samples ( Fig. 1d; Supplementary Fig. 1c) and in advanced-stage MM ( Table 1; Fig. 1e; Supplementary Fig. 1d). Of particular prognostic relevance, a subset of samples expressing the highest levels of ADAR1 (n = 159) had significantly lower overall and progression-free survival rates than the low ADAR1 cohort (n = 162) (Fig. 1f, g; Supplementary Fig. 1e, f). Thus, in MM patients, 1q21 amplification promotes ADAR1 overexpression, and during cancer progression, as shown in other malignancies, inflammatory cytokine signaling may promote ADAR1 expression leading to poorer clinical outcomes 7,11 . ADAR1-dependent RNA editing of GLI1 typifies high-risk myeloma. Deregulation of ADAR1 has been linked to malignant reprogramming of progenitors into self-renewing cancer stem cells that promote progression and relapse 11,12 . Thus, we employed RNA editing site-specific quantitative real time PCR (RESSq-PCR) 17 to detect A-to-I editing, which is subsequently read as guanosine (A-to-G), in self-renewal transcripts including the Hh pathway transcription factor GLI1. Deregulation of Hh self-renewal pathway signaling, through GLI1 and GLI2 activation, has been linked to cancer stem cell generation and therapeutic resistance in MM [18][19][20] and other hematopoietic malignancies [21][22][23]    pathway negative regulator, SUFU 24 (Fig. 2a). In a Vienna software-based prediction analysis of GLI1 pre-mRNA structure, the ADAR1 edited site in exon 12 was detected in a dsRNA hairpin adjacent to a larger dsRNA structure formed by inverted AluY and AluSx, thereby altering the exonic hairpin ( Fig. 2b; Supplementary Fig. 2a) and underscoring the importance of primate-specific secondary RNA structure for ADAR-mediated RNA editing 3,25 . Notably, GLI1 transcript editing rates were significantly higher in relapsed MM and PCL than age-matched controls (Fig. 2c, d), and correlated with ADAR1 expression levels ( Fig. 2e). Analysis of a larger RNA-sequencing CoMMpass data set showed increased A-to-G editing of GLI1 transcripts in 1qamplified compared with non-1q-amplified MM samples (Supplementary Fig. 2b). Gene set enrichment analyses (GSEA) of CoMMpass data revealed that 1q-amplified samples were significantly enriched in cancer and stem cell pluripotency 22 pathways, along with cell-cell adhesion 26 and drug metabolism 27 genes ( Fig. 2f; Supplementary Fig. 2c). Among the KEGGannotated pathways regulating stem cell pluripotency and pathways in cancer, several stem cell regulatory transcripts modulated by ADAR1 11 , including AXIN2, MAPK3, and FGFR3 (Fig. 2g), were enriched. Moreover, there was significant Hh pathway enrichment (NES = 2.11, false discovery rate (FDR) < 0.0001) in high ADAR1 (n = 177) compared to low ADAR1 (n = 152) patients, consistent with ADAR1-associated Hh pathway modulation (Fig. 2h). In addition, IL-6 signaling and JAK/STAT signaling were enriched in the high ADAR1 group, suggesting that inflammatory cytokine signaling may further enhance ADAR1 expression in high-risk MM ( Supplementary Fig. 2d, e).
To further explore ADAR1-driven RNA editing events, we performed RESSq-PCR analysis on other cancer-associated loci, including the DNA cytidine deaminase APOBEC3D, antizyme inhibitor 1 (AZIN1), and murine double minute 2 E3 ubiquitin protein ligase (MDM2, Supplementary Fig. 2g-j). Consistent with previous findings in leukemia progression 11 , APOBEC3D editing was significantly increased in late-stage MM patients, whereas AZIN1 and MDM2 loci showed heterogeneous editing levels. These data highlight the cell-type and context specific effects of ADAR1 editing.

ADAR1 silencing reduces engraftment of myeloma in vivo.
Considering that increased ADAR1 expression was shown to enhance cancer stem cell generation in leukemia 11,12 , we analyzed the serial transplantation potential of high-risk MM in RAG2 −/− γc −/− mice, as a gold standard in vivo cancer stem cell selfrenewal assay 28,29 , as well as the plasmacytoma-forming capacity of ADAR1-enriched MM samples ( Fig. 3a; Supplementary  Fig. 3a). Bioluminescent imaging demonstrated that lentiviral luciferase-expressing MM cells homed to the BM and spleen (SP) (Fig. 3b). Consistent with maintenance of malignant plasma-cell clones in vivo, patient-specific light chains were detected in the serum of the transplanted mice, within 7 to 21 weeks after transplantation (Fig. 3c). Robust engraftment of MM patientderived cells was observed by FACS analysis in all hematopoietic tissues ( Supplementary Fig. 3b). Specifically, CD138 + /CD319 + cells 30 were detected in the BM, SP, peripheral blood (PB), and liver (L, initial site of transplantation) of RAG2 −/− γc −/− mice. Notably, MM9-engrafted mice developed plasmacytomas (PC) (Supplementary Fig. 3b). Consistent with previous reports 31 , FACS analysis revealed that MM cells were CD38 high and CD45 dim (Fig. 3d; Supplementary Fig. 3c, d). In primagraft models, Alu repeat qPCR 32 detection of human-specific RNA confirmed the high rate of primary engraftment of human malignant plasma cells ( Supplementary Fig. 3e, f). Moreover, high ADAR1 expression and Alu-dependent GLI1 RNA editing were also maintained in vivo ( Supplementary Fig. 3e-h). Furthermore, this primagraft model propagated MM in serially transplanted recipient mice that recapitulated the engraftment phenotype observed in primary MM9-transplanted mice, as assessed by clonal light chain production in serum, and human CD138 + /CD319 + /CD38 high /CD45 dim cell engraftment in BM, SP, L, and PC (Fig. 3e, f; Supplementary Fig. 3i). In serially transplanted mice, the tissues with the highest GLI1 editing also had the highest levels of human cell (Alu) engraftment and human ADAR1 expression ( Fig. 3g; Supplementary Fig. 3i, j). Perhaps most notably, ADAR1 silencing by lenti-shADAR1 transduction significantly reduced serial transplantation (Fig. 3h- MM cells. These data are consistent with ADAR1's role in malignant self-renewal 11,12 , and support a vital functional role for ADAR1 activity in malignant regeneration of MM. Activation of ADAR1 in myeloma cell lines promotes drug resistance. To determine if inflammatory cytokine signaling contributes to ADAR1 activation in MM, as reported in other contexts 12,33 , we treated the 1q-amplified human myeloma cell line H929 with increasing concentrations of IL-6 (0-10 ng/ml) in vitro. We observed increased ADAR1 protein accumulation and a significant increase in GLI1 RNA editing rates (Fig. 4a, b; Supplementary Fig. 4a-d). These results highlight the importance of inflammatory signaling pathways in amplification of malignant RNA editing.

Discussion
Seminal studies demonstrate that MM is typified by widespread genetic heterogeneity and subclonal diversity as well as Hh pathway activation that contribute to therapeutic resistance 18,35,36 . Compared with ADAR2 24,37 , ADAR1 is the most abundantly expressed and active RNA editase in MM, consistent with potentiation of malignant regeneration by ADAR1. In MM, here we demonstrate that malignant regenerative capacity is associated with ADAR1-mediated recoding of the self-renewal agonist GLI1, a pathway which has therapeutic potential in cancer while sparing normal stem cell maintenance 23,38 . Although the Hh signaling pathway has been associated with cancer stem cell generation, and both the CD138 +19,20 and CD138 -18 cell populations secrete Hh ligands that activate autocrine signaling in MM, we have identified a primate-specific mechanism of Hh pathway activation in MM through enhanced transcriptional activation of GLI1 following Alu-dependent ADAR1-mediated RNA editing. Importantly, ADAR1-dependent RNA recoding of GLI1 and regeneration of MM cells harboring 1q21 amplification could be reversed through genetic knockdown of ADAR1. Overall, these findings demonstrate that both genetic (1q21 amplification) and microenvironmental (inflammatory cytokines, IMiDs) factors modulate A-to-I editing activity, which drives GLI1-dependent malignant regeneration in MM (Fig. 4k). Thus, ADAR1 activation represents both a vital prognostic biomarker and therapeutic vulnerability in MM.
Primary sample processing. MM patient samples and normal age-matched (n = 3, mean age = 60.6 ± 16.8 years old) BM samples were obtained from consenting patients in accordance with Institutional Review Board approved protocols at University of California-San Diego (UCSD or the Princess Margaret Hospital (Toronto, Ontario, Canada). Peripheral blood (PB) or BM samples were processed by Ficoll density centrifugation in a SepMate conical tube (StemCell Technologies). Viable total mononuclear cells (MNC) were collected for further analyses and stored in liquid nitrogen.
RNA editing site-specific quantitative PCR. RESSq-PCR assay primer design 17 was carried out for specific cancer and stem cell-associated loci. For these assays, allele-specific primers were designed using Primer1, generating two outer and two inner primers for each editing site with melting temperatures ranging from 60 to 68°C. The forward (FW) outer and reverse (REV) outer primers flank the editing site and can be used for Sanger sequencing validation of each editing site, and also as a qPCR positive control to ensure that the editing region is detectable in cDNA. The   3′ ends of the FW inner and REV inner primers match either the WT A or edited G nucleotide, and include an additional mismatch two nucleotides upstream of the 3′ primer end to enhance allelic discrimination. Relative RNA editing rates (Relative Edit/WT RNA) were calculated using the following calculation: 2 −(Ct Edit − Ct WT) .
Gene set enrichment analyses and double strand structure prediction. Gene expression values (FPKM) downloaded from the MMRF CoMMpass study web portal (https://research.themmrf.org and study accession phs000748.v4.p3) were input into the GSEA software 11 , and through the MMRF GSEA algorithm with NCI-provided gene sets. Enriched gene sets were derived from the analysis with a FDR < 10% and p value < 0.05. GLI1 editing analysis was performed on whole transcriptome data publicly available from the CoMMpass study. A total of 97 samples were included. RNA-seq raw reads were aligned to reference genome GRCh37. Aligned reads were processed with SAMTools, and editing (A or G SNV) was evaluated in GLI1 17 . Putative A-to-G events with equal or higher than 2 reads were included in the final analysis (n = 14/17 with/without 1q21 amplifications, respectively). Consistent low coverage was observed among all samples for the site. Ratio of G over total reads is displayed in results. Secondary structure of GLI1 transcript were predicted by Vienna RNA software 40 .
Transient GLI1 and ADAR1 overexpression. HEK293T cells were plated in 12well plates (0.25 × 10 6 cells per well) 24 h prior transfection. pCDH backbone or pCDH-GLI1 plasmids were used for transient transcript overexpression, cotransfected with pDEST ADAR1-WT or mutant 11 ADAR1 E912A . 1 μg of total plasmid DNA was used for each condition and cells were collected 48 h posttransfection for further analyses.
Intrahepatic inoculation of tumor cells and tissue collection. All animal studies were performed in accordance with UCSD and NIH-equivalent ethical guidelines and were approved by the university institutional animal care and use committee (IACUC). Newborn (1-3 days) Balb/c Rag2−/−γc−/− mice (sample size depending on litter survival rates) were intrahepatically injected with a 30 gauge Hamilton syringe (Hamilton Company). Each animal received 1-2 × 10 6 MNCs isolated from primary samples. Animals were weaned at 3 weeks of age and monitored regularly by health status assessment; in vivo bioluminescence imaging (IVIS 200) and peripheral blood screening were regularly performed until signs of disease were observed, including significant loss of weight, limited mobility and presence of palpable tumors. Mice were killed by CO 2 inhalation. PB was collected by cardiac puncture immediately after sacrifice. Bones, SP, L, and PC were collected in ice cold HBSS containing 2% fetal bovine serum (FBS). For serial transplantation assays, hematopoietic tissues and PC were processed into single-cell suspensions, enriched for human cells using a mouse cell depletion kit (Miltenyi), and equal proportions of viable bone marrow and PC-derived human cells were mixed for transplantation into serial transplant recipient mice.
Western blot and nanoproteomic analysis. A total of 5-10 × 10 6 cells were harvested in RIPA buffer for total protein extraction. Overall, 10-20 μg of each sample were loaded onto 10% polyacrylamide gels for gel electrophoresis. Primary antibodies (polyclonal anti-ADAR1 ab88574 1:500, monoclonal anti-ADAR1 ab126745, 1:1000, and anti-actin ab8227, 1:1000, Abcam) were prepared in blocking buffer. Membranes were incubated overnight at 4°C with primary antibodies, followed by secondary antibody incubation (anti-rabbit HRP or anti-mouse HRP, 1:5000, Cell Signaling Technology; or anti-chicken HRP, 1:5000, Abcam) in blocking buffer for 1 h at room temperature. Blots were developed using enhanced chemiluminescence (Femto Detection kit, Promega) on a Chemidoc digital imaging machine. Representative images out of three independent Western blots are shown in main and supplementary figures, with uncropped images of key blots provided in Supplementary Fig. 5. Nanofluidic experiments were performed with the Nanopro 1000 instrument (Cell Biosciences). ADAR1 was detected using an ADAR1-specific antibody (ab168809, 1:500 Abcam). A β2-microglubulin-specific antibody (β2M; 1:500 Upstate) was used to normalize the amount of loaded protein.
Data availability. The RNA-Seq data that were analyzed in this study are publicly available through the Multiple Myeloma Genomics Initiative (https://research. themmrf.org) with the identifiers IA7, IA8, and IA9. These data were generated as part of the Multiple Myeloma Research Foundation CoMMpass (Relating Clinical Outcomes in MM to Personal Assessment of Genetic Profile) study (www. themmrf.org), and all linked genotype and phenotype data used have previously been deposited into the NCBI database of Genotypes and Phenotypes (dbGaP study accession phs000748.v4.p3, BioProject PRJNA248539, SRA study SRP047533). Data can be accessed via submission of a request to dbGaP at https:// www.ncbi.nlm.nih.gov/gap. All other remaining data are available within the article and supplementary files, or available from the authors upon request.