Targeting ADAR1 with a Novel Small-Molecule for the Treatment of Prostate Cancer

Despite initial response to androgen signaling therapy, most prostate cancer (PCa) patients eventually relapse and remain incurable. ADAR1-mediated A-to-G editing plays oncogenic roles in various tumors. However, the specic function of ADAR1 and the global RNA edited targets governing PCa progression remain underexplored. Here, we demonstrate that highly expressed ADAR1 as a crucial oncogenic target in PCa, and develop a novel small-molecule ADAR1 inhibitor ZYS-1 with signicant anti-tumor ecacy and favorable safety prole. Either depletion or pharmacological inhibition of ADAR1 dramatically suppress PCa growth, inhibit metastasis, and potentiate immune response. We further reveal that the translation of MTDH is repressed by ADAR1 in an editing-dependent manner, which drives cell proliferation and invasion in PCa. Collectively, these results shed light on ADAR1 as a novel druggable target in PCa therapy and highlight the widespread applicability of ADAR1 inhibitors for a broad spectrum of malignancies.


Introduction
Prostate cancer (PCa) is the most frequently occurring malignancy and the second leading cause of cancer-related mortality among men in western countries (1)(2)(3), with an estimated over 1 million new cases and nearly 500,000 deaths expected to occur globally in 2020 (2). Primary PCa patients who are diagnosed with early precursor lesions are known as prostatic intraepithelial neoplasia (PIN), and treatments for these organ-con ned PCa include surgery and brachytherapy (4,5). PCa tumor growth is mainly dependent upon sustaining activated androgen receptor (AR) signaling. The backbone of PCa treatments includes androgen-deprivation therapy (ADT) and AR antagonists by blocking androgen production and/or function (6, 7). Despite the initial response, most patients eventually relapse to a therapy-resistant stage, which leads to metastatic castration-resistant prostate cancer (mCRPC) that remains incurable at the current time (7,8). In addition to the sustained role of AR, there has been an urgent need to identify speci c and novel drug targets. As such, the global understanding of the intricate mechanisms involved in these oncogenic processes is of critical importance for the development of novel therapeutic strategies (9)(10)(11).
Gene mutations are important in cancer onset and aggressiveness as well as in the generation of tumor heterogeneity. Besides alterations at the DNA level, RNA-based abnormalities have been emerging as important players in cancer pathogenesis (12,13). RNA editing is a common form of epitranscriptomic mechanism that modi es the sequence of RNA transcripts without a concomitant change in its genomic blueprints (12,14). High-throughput genome-wide studies suggest that editing occurs in more than 85% of human RNA noncoding and/or coding regions (15). As for physiological consequences, RNA editing events may lead to expanding diversity in gene products and gene regulatory mechanisms by nonsynonymous amino acid mutations, alternative RNA splicing, or modi cation of targeted RNAs (16, 17), thereby altering their stability, translational e ciency, and degradation (18). In humans, deamination of adenosine-to-inosine (A-to-I) within double-stranded RNA (dsRNA), mediated by adenosine deaminases acting on RNA (ADAR) family enzymes, is considered the most frequent type of RNA editing that impacts the bene t of cancer progress (15,19,20). Since inosine (I) is recognized as guanosine (G) by the translational machinery, this process may result in recoding transcripts and thus plays a vital role in cancer initiation and aggressiveness (21,22). Three ADAR genes are encoded in mammalian cells, ADAR1 and ADAR2 are the two catalytically active enzymes, whereas ADAR3 is enzymatically inactive (23,24).
The rst cloned and discovered A-to-I dsRNA deaminase was ADAR1, which resides at chromosome 1q21, a region that is frequently ampli ed in cancer. There are two isoforms of ADAR1, the constitutively expressed N-truncated-p110 is mainly localized to the nucleus, where the interferon (IFN)-inducible full length-p150 is mainly cytoplasmic (25,26). Increasing pieces of evidence have revealed a close association between majority RNA-editing events and elevated ADAR1 expression in various malignancies, thereby promoting tumorigenesis, therapeutic resistance and poor survival (27,28). Crucially, several articles have reported higher RNA editing levels in prostate cancer (29)(30)(31). In silico analyses demonstrated that AR transcripts within more RNA editing events in PCa cell lines, meanwhile these cells also expressed high levels of ADAR1 (32). In addition, a recent study has suggested that elevated ADAR1 expression causes aberrant editing of the dominant-negative oncogene prostate cancer antigen 3 (PCA3) to the pathogenesis of PCa, whereby higher A-to-I editing of lncRNA PCA3 could upregulate expression of itself, and then downregulates the antisense tumor suppressor gene prune homolog 2 (PRUNE2) level, thereby, promoting PCa cell growth and metastasis (33). Despite PCA3 is a novel target of ADAR1, the signi cance of ADAR1 and the landscape of RNA-edited targets in PCa remain unanswered. Therefore, further identi cation of these aberrant targeted transcripts and investigation of the regulatory mechanisms can better enhance our understanding of PCa carcinogenesis.
It has been reported that loss of ADAR1 compromises tumor growth and metastasis in many cancer types (27). More importantly, emerging studies have shown that inhibition of ADAR1 in tumors not only potentiates the e cacy of epigenetic therapy, but also sensitizes tumors to immunotherapy and overcomes resistance to immune checkpoint blockade, which is bene cial for cancer therapy (34)(35)(36). Taken together, these studies suggest that ADAR1 maybe a new potential PCa therapeutic target, and exploitation of innovative targeted therapeutics against ADAR1 is warranted. Although to date, 8azaadenosine (8-aza) has been considered as a potential ADAR1 inhibitor (37), there was not yet strong evidence suggesting its direct binding and inhibition with ADAR1 protein, and simultaneously, the action mode and therapeutic activity have not been adequately evaluated. Thus, there is still an urgent need to discover druggable inhibitors against ADAR1 to treat PCa and other cancers.
Herein, we carried out detailed investigation into RNA editing in PCa and the rst identi cation of landscape of A-to-G RNA edited targets through systematic assays. Our results demonstrated that ADAR1 but not ADAR2 is abundantly expressed in prostate cancer compared with adjacent non-tumor tissues and associated with poor clinical outcomes. Furthermore, ADAR1 depletion inhibited PCa cells growth and metastasis in vitro, whereas overexpression had the opposite effects. Most importantly, loss of ADAR1 increased abundance of IFN-γ and CD8 + T cells in ltration in vivo. In addition, we identi ed metadherin (MTDH) as a crucial edited target of ADAR1 in PCa and provided a new insight that this RNA epitranscriptomic modi cation regulates MTDH protein levels through the translation modulation mechanism, which relied on the catalytic function of ADAR1. Rescue results showed that ADAR1/MTDH axis mediates tumor initiation and aggressiveness, which indicated the existence of a new protumorigenesis pathway in PCa. To this end, using a series of structure-based virtual screening and validation assays, we further developed a novel small-molecule ADAR1 inhibitor, ZYS-1, which directly bind to ADAR1 and inhibit its deaminase activity with an IC 50 of 0.866 µM. Consistent with ADAR1 knockdown, ZYS-1 not only strongly inhibited PCa cells proliferation and metastasis in vitro, but also dramatically suppressed tumor growth in mice with favorable safety pro les, and enhanced anti-tumor immunity, hopefully holding synergetic e cacy combining with immunotherapy. Collectively, our study suggests profound therapeutic effects and novel mechanism insights of targeting ADAR1 in PCa therapy, highlighting ADAR1 as a novel druggable target with broad potential for cancer therapy.

Results
Highly expressed ADAR1 is a favorable prognostic marker in prostate cancer Since the involvement and role of ADAR1 in prostate cancer have not been dissected systematically, we rst queried whether the expression of ADAR1 is high and associated with A-to-G editing level in PCa patients. We collected three pairs of matched tumor and normal tissues and performed RNA-sequencing to acquire their global RNA editing pro le. There were 1949 genes with A-to-G editing frequency difference > 0 (Tumor vs Normal group, labeled in red), and 1543 genes with difference < 0 (labeled in blue) ( Fig. 1A; Supplementary Table S4). ADAR1 expression level was signi cantly higher in tumor ( Fig. 1I; Supplementary Fig. S1D), while ADAR2 expression was low compared to ADAR1 (data not shown) and was in a promiscuous pattern between tumor and normal tissues ( Supplementary Fig. S1F). We next selected 16 patients from a previous reported TCGA PCa patients cohort (n = 31; ref. 30) according to their A-to-G editing level higher in tumor other than normal tissue ( Supplementary Fig. S1A).
This increase appears to be caused by higher ADAR1 expression in tumor rather than ADAR2 (Supplementary Fig. S1B and S1C). Higher ADAR1 expression in PCa was again found in another TCGA cohort (n = 51; Fig. 1B). Interestingly, aged, high-risk and high Gleason grade patients had a higher ADAR1 expression signature, which indicated worse clinical prognosis in patients with higher ADAR1 expression ( Fig. 1C-E). Notably, these patients' relapse-free survival (RFS) was signi cantly worse than those with low expression (Fig. 1F). Consistently, a tissue microarray from 38 PCa patients staining with ADAR1 antibody showed that ADAR1 protein intensity score in tumor was signi cantly (P = 0.004) higher than in adjacent normal tissue ( Fig. 1G and H), and ADAR1 intensity score was apparently stronger in high Gleason grade patients (Fig. 1G, bottom). Furthermore, ve prostate cancer cell lines were all detected to have a high level of ADAR1 protein ( Fig. 1J; Supplementary S1E). Previous study has indicated loss of ADAR1 regulates tumor microenvironment (TME) and enhances immune in ammation through a MDA5/MAVS-dependent manner, as evidenced by increase of immune cells and decrease of myeloidderived suppressor cells (MDSCs) (36). We used TIMER2.0 web tool to analyze correlation of ADAR1 and tumor immune in ltrating pro le, and con rmed negative correlation with CD4 + T cells, CD8 + cells, M1 macrophages, and NK cells, and positive correlation with cancer-associated broblast (CAFs), endothelial cells and inhibitory immune cells M2 macrophage ( Supplementary Fig. S1G). Collectively, these evidences suggest ADAR1 is aberrantly overexpressed in PCa tumors and associated with poor clinical outcomes.

Depletion of ADAR1 inhibits tumor cells growth in vitro and in vivo
To assess the effect of depletion of ADAR1 on PCa cells, we generated stable ADAR1-knockdown cells (shADAR1) and negative control cells (shNC) in DU-145 and VCaP cells using lentivirus-packaged shorthairpin RNA (shRNA). ADAR1 protein level was greatly decreased in knockdown cells relative to control without affecting ADAR2 ( Fig. 2A Supplementary Fig. S2C). Via ow cytometry, we observed ADAR1 knockdown results in cell cycle arrest at G0/G1 phase and increased apoptosis ( Fig. 2G and H). Next, to determine whether ADAR1 knockdown compromises tumor growth in vivo, we inoculated subcutaneously DU-145 shADAR1 and shNC cells into mice. Surprisingly, only one mouse grew little tumor while others (5/6) did not grow visible tumors, suggesting ADAR1 loss did abrogate tumor growth (Fig. 2J-L; Supplementary Fig S2D). Next, we overexpressed empty vector (EV), Flag-tagged ADAR1-wild type (ADAR1-WT), and catalytically inactive ADAR1 E912A mutant (ADAR1-Mut) in DU-145 cell (Fig. 2D). A large increase in editing level of GLI1 was found in ADAR1-WT cells relative to EV while ADAR1-Mut did not (Fig. 2E). Cell proliferation assay showed a higher growth rate of WT cells and comparable proliferative pro le between Mut and EV cells, which demonstrated ADAR1's pro-proliferation ability is deaminase activity-dependent (Fig. 2F). These ndings clearly implicate that loss of ADAR1 inhibits PCa tumor cell growth in vitro and in vivo via cell cycle arrest and apoptosis.

Depletion of ADAR1 signi cantly modulates transcriptional activity involved in tumor survival
To comprehensively understand the signaling modulation of ADAR1, we performed whole transcriptome sequencing (RNA-seq) on DU-145 shADAR1 and shNC cells with good concordance between two biological replicates ( Supplementary Fig. S3A). After ADAR1 knockdown, 139 and 48 genes were upregulated and downregulated, respectively, such as ADAR, CDK6, and CDH1, as shown in volcano plot and heat map ( Fig. 3A; Supplementary Fig. S3B). In accordance with cell proliferation arrest phenotype observed in ADAR1 loss, gene set enrichment analysis (GSEA) using the Hallmarks gene set from the Molecular Signatures Database (MSigDB) showed substantial suppression of cell proliferation and cellcycle related pathways, such as MYC targets, G2M checkpoint, and E2F targets (Fig. 3B). In addition, ADAR1 loss increased gene enrichment in P53 and apoptosis pathway, though not signi cant (Fig. 3D). It is worth noting that cell adhesion molecules (CAMs) pathway in KEGG (Kyoto Encyclopedia of Genes and Genomes) was dramatically upregulated (Fig. 3C), which may be attributed to suppressed epithelialmesenchymal transition (EMT) signaling as evidenced by increased expression of epithelial marker (Ecadherin) and decreased expression of mesenchymal markers (N-cadherin) (Fig. 3F). We further con rmed reduced expression of cell proliferation and cell cycle-regulated protein CDK6 and c-MYC, and increased P53 protein and cleaved PARP upon ADAR1 KD ( Fig. 3E and F). As an RNA editing enzyme, ADAR1 is responsible for avoiding hyper-activated IFN-I responses induced by endogenous dsRNA. Consistent with this, IFN-α response, IFN-γ response, and TNFα signaling via NF-κB pathways were found to be positively enriched in shADAR1 group, which could be attributed to elevated expression of IFNstimulated genes (ISGs) (Fig. 3C). We then used enzyme-linked immunosorbent assay (ELISA) to detect IFN-γ content in peripheral blood in shADAR1 and shNC xenograft mice since shADAR1 mice barely grew tumors. As expected, IFN-γ in PB of shADAR1 mice showed an approximate 2-fold increase relative to shNC mice (Fig. 3G). To assess immune in ltration, we performed ow cytometry to determine the abundance of cytotoxicity CD8 + T cells. In line with above bioinformatics results ( Supplementary Fig.   S1G), CD8 + T cells in shADAR1 mice spleen were signi cantly more than shNC ( Fig. 3H; Supplementary   Fig. S3C). All these results show the critical roles of ADAR1 in regulating signaling involved in tumor cell proliferation, apoptosis, immune response, and metastasis.
Characterize MTDH as a novel target of ADAR1 governing PCa carcinogenesis and metastasis Subsequently, to investigate the mechanism of signaling modulation of ADAR1, we identi ed the direct targets of ADAR1 in PCa through multi-omics sequencing. In our RNA-seq data, a large decline of A-to-G editing events occurred in shADAR1 cells ( Supplementary Fig. S4A), and these editing sites were mainly enriched in intron and 3′ UTR ( Supplementary Fig. S4B). To acquire ADAR1 bound transcripts, we performed RNA-immunoprecipitation sequencing (RIP-seq) using an ADAR1 antibody ( Supplementary  Fig. S5C). In all bound approximate 2000 RNA peaks, introns and exons account for 49.72% and 27.34% of peak region, respectively ( Supplementary Fig. S4D). In addition, based on our RNA-seq data, we screened A-to-G editing frequency difference of more than 0.1 between shNC and shADAR1 group (shNC -shADAR1 > 0.1) to acquire 487 genes ( Supplementary Fig. S4C). Subsequently, according to 487 editing genes, all bound 1993 peaks, and total genes (expression count > 200), we obtained 114 potential targets of ADAR1 (listed in Supplementary Table S3), as shown in the Venn diagram ( Supplementary Fig. S4E).
GO and KEGG analysis showed that these 114 genes were mainly enriched in processes such as translation, viral process, cell cycle, ribosome, and apoptosis (Supplementary Fig. S5A and S5B). Based on their functions in literature, we further selected 28 genes and validated their bindings with ADAR1 using qPCR assay ( Supplementary Fig. S4F). We found numbers of targets were positively accumulated in ADAR1 relative to lgG, such as MTDH, EIF2AK2, and IFNAR1 ( Fig. 4A; Supplementary Fig. S5D). Sanger sequencing was performed to validate editing sites indicated in RNA-seq data. RBBP4, EIF2AK2, DFFA, IFNAR1, and MTDH were con rmed with a decline of editing levels upon ADAR1 loss ( Fig. 4B; Supplementary Fig. S5E).
To investigate the crucial targets governing tumor survival in PCa, we rstly examined their correlations with ADAR1. Bioinformatics analysis data suggested that ADAR1 and MTDH have a strong positive correlation in PCa ( Supplementary Fig. S6D). Our RNA-seq data also showed a large decline editing levels of MTDH in normal PCa tissues compared to tumors (Supplementary Table S4). Previous studies demonstrated that MTDH is an oncogenic protein associated with tumors growth and metastasis via modulation of cell cycle arrest, apoptosis, and EMT signaling, furthermore, MTDH loss also inducing cytotoxic T cell in ltration in colorectal and lung cancers (42,43). These phenotypes are consistent with ADAR1 loss in PCa. Therefore, we speculated that MTDH might be the downstream target of ADAR1 in promoting PCa cell growth and metastasis. Firstly, RIP-qPCR and Sanger sequencing assay had con rmed that MTDH is a direct editing target of ADAR1 ( Fig. 4A and B). Then we decided to elucidate the mechanism through which ADAR1 interacts and regulates MTDH. We examined expression of MTDH in RNA-seq data which did not show any difference between shNC and shADAR1 groups and con rmed using qPCR (Fig. 4C). However, the protein level of MTDH showed a pronounced reduction upon ADAR1 knockdown (Fig. 4C). Then, we overexpressed ADAR1-WT and -Mutant in shADAR1 cells, and found that wild-type ADAR1 could completely rescue MTDH protein level and cell viability while mutant-type could not even partially, demonstrating maintenance of MTDH protein is dependent on ADAR1's deaminase activity, which is also essential for cancer cells survival ( Fig. 4D and E). Next, we decided to dissect why MTDH would dramatically reduce upon ADAR1 knockdown. Unexpectedly, knockdown of ADAR1 did not alter MTDH mRNA stability after transcription was inhibited by actinomycin D treatment (Fig. 4F). Therefore, we questioned whether ADAR1 would affect accumulation of MTDH mRNA in ribosome machinery. We conducted ribosome immunoprecipitation assay using anti-RPL22 (constituent of 60S ribosome protein) antibody, and results showed that MTDH mRNA enrichment signi cantly decreases in ribosome machinery upon ADAR1 knockdown ( Fig. 4G and H). We speculated less degree of editing in 3′ UTR of MTDH mRNA resulted in reduced ribosome occupancy. ADAR1 editing of endogenous RNA plays a role in preventing PKR activation, which allows PKR-eIF2α axis downregulating translation during IFN response (44). Moreover, we observed PKR and eIF2α activation following ADAR1 knockdown in DU-145 cell line, consistent with previous studies ( Fig. 4I; ref. [44][45][46]. This may be one reason for reduction of MTDH translation upon ADAR1 knockdown. We then overexpressed MTDH in ADAR1 knockdown cells and found MTDH could rescue compromised proliferation and invasion phenotype induced by ADAR1 loss (Fig. 4J-L), which demonstrated that suppressed proliferation and invasion upon ADAR1 loss in PCs is MTDH-dependent.
Finally, we were interested in clinical signi cance of MTDH in prostate cancer. Above mentioned three pairs of PCa tissues were subjected to Sanger sequencing and WB analysis to evaluate editing level and protein level of MTDH, respectively. We observed elevated editing level of MTDH in tumors relative to normal tissues ( Supplementary Fig. S6A). Expression of MTDH was also signi cantly higher in tumors in both protein and mRNA levels (Supplementary Fig. S6B and S6C). Collectively, these ndings demonstrate MTDH as a novel and critical downstream target of ADAR1 is responsible for cell proliferation and invasion, and may be a clinical biomarker for prostate cancer.
Discover a novel and potent small molecule inhibitor of ADAR1 Though others have reported 8-aza as a potential inhibitor of ADAR1, there has not yet been strong evidence suggesting its direct binding and inhibition with ADAR1 protein. 8-aza has been reported to slightly inhibit ADAR2 deaminase activity with an IC 50 of 15 mM, and has only obscure evidence to show ADAR1 inhibition (37,47).
To discover potential ADAR1 inhibitors, a structure-based virtual screening was conducted by LibDock and CDOCKER of Discovery Studio 2021 (Fig. 5A). According to a recent study of ADAR1 protein structure (48), homology modeling of ADAR1 was applied to this procedure and the pocket was de ned as a catalytic pocket. Through two rounds of screening of SPECS and in-house libraries, we obtained the hit  Supplementary Fig. S7D). Docking analysis suggested that in addition to hydrogen bonds with crucial residues, some extra weak interactions were formed between ZYS-1 and ADAR1 catalytic pocket, possibly helping better binding a nity (Fig. 5B). To detect direct binding of ZYS-1 to ADAR1 protein, we performed microscale thermophoresis assay (MST) and con rmed their interaction with a K D value of 7.24 ± 2.88 µM (Fig. 5F). Moreover, cellular thermal shift assay (CETSA) validated that ZYS-1 interacted with ADAR1 protein in cellular conditions (Fig. 5E). Next, we assessed the direct inhibitory function of ZYS-1 on ADAR1 in PCa cells. A comparable cell proliferation pro le was observed in shADAR1 and 50 nM treated cells (Fig. 5G). We simultaneously treated shNC and shADAR1 cells with or without ZYS-1, and the difference of proliferation ratio (treated/untreated) started to show after 24 h and was more aggravated after 48 h (Fig. 5H). Furthermore, ADAR1 loss resulted in a about 2-fold increase of IC 50 value of ZYS-1 in DU-145 cell ( Supplementary Fig.   S7E), meanwhile, forced expression of ADAR1 signi cantly attenuated growth inhibition by ZYS-1 ( Supplementary Fig. S7F). As expected, ZYS-1 dose-dependently decreased editing level of GLI1 in DU-145 and VCaP cells (Fig. 5I). This dose-dependent decrease was also con rmed using RNA-editing site speci c quantitative PCR (RESSqPCR) (Fig. 5J;

ZYS-1 strongly suppresses PCa cell growth and invasion via ADAR1 inhibition
Consistent with the effect of ADAR1 knockdown, ZYS-1 not only inhibited cell proliferation ( Fig. 5C and G), but also inhibited cell colony formation ability, and induced cell cycle arrest at G0/G1 phase and substantial apoptosis in a dose-dependent manner ( Fig. 6A-C; Supplementary Fig. S8A). In addition, ZYS-1 signi cantly impaired DU-145 cell migration and invasion ability (Fig. 6D). Interestingly, we further found that ZYS-1 could reduce ADAR1 protein levels in DU-145 and VCaP cells, while did not alter ADAR1 mRNA levels (Fig. 6E). These data suggest that ADAR1 protein degradation induced by ZYS-1 is in a posttranscriptional manner. However, ADAR1 degradation induced by ZYS-1 was independent of proteasome degradation (Supplementary Fig. S8C and S8D). Precise mechanism needs further investigation.

ZYS-1 displays signi cant anti-tumor e cacy and favorable safety pro le in vivo
As ZYS-1 exhibits a well anti-tumor effect in vitro, we then assessed the therapeutic effects of ZYS-1 in vivo using a DU-145-derived xenotransplantation mice model. Intraperitoneal administration of low dose (20 mg/kg/day) and high dose (40 mg/kg/day) of ZYS-1 signi cantly suppressed tumor growth relative to vehicle group in a dose-dependent manner, with tumor growth inhibition (TGI) rate achieving 66.68% and 81.83%, respectively ( Fig. 7A-C). Especially, tumor growth almost stagnated after treatment for 20 days. Meanwhile, no differences in body weight or behavior (data not shown) were observed between dosing and control group ( Supplementary Fig. S10A). Also, H&E staining of key organs of dosed mice did not exhibit apparent necrosis or in ammation ( Supplementary Fig. S10B). Both results prove ZYS-1 is well tolerated and safe in vivo even after continuous administration. Next, H&E, Ki-67, and tunel staining were performed to evaluate tumor proliferation and apoptosis. Necrosis area of dosing group tumor was markedly larger as shown in H&E staining (Fig. 7D). Proliferation level of tumor cells was lower after ZYS-1 treatment as indicated by Ki-67 staining ( Fig. 7D and E). Tunel staining was utilized to determine apoptosis and suggested ZYS-1 induced tumor cell apoptosis in a dose-dependent manner ( Fig. 7D and F). We also performed immunohistochemical staining of ADAR1 and showed ZYS-1 decreased ADAR1 protein level dose-dependently in tumor, recapitulating the effects in vitro ( Fig. 7D and G). Also, apoptosis-, cell cycle-, and EMT-associated protein level changes had a comparable tendency in tumors upon ZYS-1 treatment (Supplementary Fig. S10D). Subsequently, we assessed the acute toxicity of ZYS-1 in mice by treatment with various doses of ZYS-1 via intraperitoneal or oral administration. Lethal dose 50% (LD 50 ) of i.p. and p.o. was 208 and 687 mg/kg, respectively. Body and key organ weights and their H&E staining analysis were not found aberrant between dosed and control groups (Supplementary Fig.  S9A and S9B). These evidences show that ZYS-1 has strong anti-tumor e cacy and is relatively safe in vivo.
Previous studies have demonstrated that loss of ADAR1 in tumor promotes IFN release and e cacy of immunotherapy such as PD-1 blockade (36). This sensitivity of tumor to immunotherapy is mediated by IFN release induced by ADAR1 loss. We thus determined the concentration of IFN-γ in tumor of xenograft mice. ELISA and WB analysis both validated increased human IFN-γ release in tumor in a dose-dependent manner ( Fig. 7H; Supplementary Fig. S10D). Consistent with the increase of CD8 + T cells upon ADAR1 loss, we found that ZYS-1 treatment signi cantly elevated proportion of CD8 + T cells in tumor and spleen ( Fig. 7I-K). In addition, immunosuppressive myeloid-derived suppressor cells (MDSCs) CD11b + Gr1 + were decreased in tumors after ZYS-1 treatment (Supplementary Fig. S10C). These results together suggest ZYS-1 not only exerts anti-proliferation and pro-apoptosis effects on tumor, but also ameliorate TME and enhance anti-tumor immunity by induction of IFN-γ release, CD8 + T cell in ltration, and decrease of MDSCs. Taken together, targeting ADAR1 is a promising therapy for prostate cancer and may synergize with immunotherapy such as PD-1 blockade deserving further studies.
Particularly, ZYS-1 also potentiated ATRA-induced myeloid differentiation in these primary AML cells in a dose-dependent manner, possibly through an unknown mechanism by which ADAR1 inhibition on AML myeloid differentiation (Supplementary Fig. S11H). Collectively, these results suggest targeting ADAR1 with small molecule inhibitor ZYS-1 has broad therapeutic potential against a wide range of tumor types.

Discussion
Prostate cancer (PCa) remains a leading cause of cancer death in men (52). However, it will inevitably progress into incurable metastatic CRPC in most patients following ADT and AR antagonist therapy (53). Therefore, identifying novel and e cacious drug targets is urgently needed to expand therapeutic options for lethal PCa. ADAR1-mediated deamination of adenosine-to-inosine (A-to-I) within double-stranded RNA (dsRNA) is the most prevalent form of RNA editing that modi es the sequence of RNA transcripts without changing its genomic blueprints, and its aberration has revealed a close association with autoimmune disease Aicardi-Goutieres Syndrome (AGS), neurodegenerative disease Autism and many kinds of cancer (14,54). Although previous study has reported that PCA3/PRUNE2 axis is mediated by ADAR1 dependent RNA editing effects on tumor growth in AR-dependent prostate cancer cell lines (33), little is known about the role of ADAR1 in AR-negative prostate cancer, and the landscape of A-to-G editing targets in PCa remains underexplored. In the present study, we not only demonstrated that ADAR1-mediated A-to-G editing is a crucial carcinogenesis event in PCa, but also developed a novel small-molecule ADAR1 inhibitor, ZYS-1, that exhibits profound anti-proliferation activities against multiple cancer cells, not limited to PCa. Our data showed that ADAR1 is highly expressed in prostate cancer and associated with higher A-to-G editing levels. ADAR1 depletion and pharmacological inhibition dramatically suppressed PCa cells growth and metastasis, induced apoptosis, reduced the tumor burden and ameliorated immune response in vivo, We further identi ed oncogenic protein MTDH as a novel crucial edited target of ADAR1 driving tumor proliferation and invasion in PCa, whose translation was repressed by ADAR1 in an editing dependent manner. These data establish the critical regulatory role of ADAR1 in PCa initiation and progression, thereby expanding remedies for lethal PCa treatment.
Several studies have analyzed RNA editing pro le in cancer from high-throughput whole transcriptome RNA-seq data in public databases, showing that A-to-G editing levels are higher in most tumors rather than in the matched normal tissues (28-31), including prostate cancer. Recently, Chan and colleagues reported that ve types of cancers have hyperediting patterns in the mesenchymal (M) phenotypes rather than in the epithelial (E) phenotypes, including prostate cancer (55). However, the precise molecular mechanism through which ADAR1-mediated A-to-G editing leading to prostate carcinogenesis and metastasis was not fully explored in these studies. Here combined with the TCGA public data and the RNA-seq data of three pairs of matched PCa and normal tissues, we demonstrate that a higher A-to-G editing level in PCa appears to be caused by higher expression of ADAR1 rather than ADAR2. Subsequently, we identi ed 114 potential targets of ADAR1 through multi-omics analysis, which were mainly enriched in processes such as translation, viral process, cell cycle, ribosome and apoptosis. Based on their functions in literature, we further experimentally validated their binding with ADAR1. Importantly, we identi ed MTDH as a crucial edited target of ADAR1 in PCa. MTDH as an oncogenic protein has been implicated in various cancers, such as breast cancer, prostate cancer, hepatocellular carcinoma, colorectal cancer, gastric cancer, and lung cancer, which drove tumor development and drug resistance via modulation of AKT/PI3K, NF-κB, MAPK, and Wnt/β-catenin signaling (42,43,56). Previous studies showed higher A-to-G editing levels of MTDH in breast cancer and hepatocellular carcinoma (57, 58), however, the connection of ADAR1 and MTDH has not been reported in PCa. In this study, we found elevated A-to-G editing level and higher expression level of MTDH in the three PCa specimens rather than normal tissues, and loss of ADAR1 reduced MTDH protein level, but not altered mRNA stability and transcript abundance. Rescue results showed that ADAR1/MTDH axis is regulated in an editingdependent manner and mediates PCa cells proliferation and invasion. Indeed, we have shown that ADAR1 editing in 3′ UTR of MTDH affects the accumulation of MTDH mRNA in ribosome machinery, and ADAR1 knockdown resulted in upregulation of p-PKR and p-eIF2α, which trigger comprehensive translation shutdown (44). Rather, other studies suggested that ADAR1 regulates translation e ciency by affecting transcript spatial distribution and ribosome stalling (59,60). More precise mechanisms towards MTDH post-transcriptional regulation still need to be illuminated future.
In addition, several recent articles reported loss of ADAR1 in tumors not only potentiates the e cacy of epigenetic therapy in colorectal cancer, but also confers susceptibility to the immunotherapy and overcomes resistance to immune checkpoint blockade in melanoma (35,36). Therefore, development of small-molecule drugs against ADAR1 is of therapeutic signi cance. However, hitherto, due to the lack of crystal structural information for the deaminase domain of ADAR1, no drug has been successfully discovered against ADAR1. Here, using the homology model created in a recent study (48), by structurebased virtual screening, we obtained the hit compound 8-Br-Ado which moderately inhibited the deaminase activity of ADAR1 and cell proliferation. Next, through further structural optimization and subsequent validation assays, we developed a novel small-molecule ADAR1 inhibitor, ZYS-1, which could bind to ADAR1 and inhibited its deaminase activity with an IC 50 of 0.866 µM. Recapitulating ADAR1 knockdown effect, ZYS-1 dramatically suppressed PCa cells growth and metastasis, induced apoptosis, reduced tumor burden and ameliorated immune response with a favorable safety pro le in vivo, furthermore, modulating the certain signaling targets, e.g., CDK6, MYC, CDH1, CDH2, TP53, and EIF2AK2. Importantly, ZYS-1 impaired the A-to-G editing levels of the novel target MTDH and IFNAR1, and reduced MTDH protein level but not changed its mRNA level. Rescue assays showed that ZYS-1's contribution to anti-tumor activity in PCa was through inhibition of ADAR1/MTDH axis.
We rst speculated ADAR1 protein degradation induced by ZYS-1 is in a post-transcriptional manner as ZYS-1 did not alter ADAR1 mRNA levels. A previous study has indicated that IFN signaling activation would induce ADAR1 degradation in an ubiquitination-dependent manner (61). We thus assumed that ZYS-1-induced ADAR1 degradation is ubiquitination/proteasome-dependent. At rst, we used cycloheximide (CHX) to inhibit new protein synthesis. However, ZYS-1 only slightly induced ADAR1 protein degradation after 6 and 12 hours treatment ( Supplementary Fig. S8C). Next, we used a proteasome inhibitor MG132 to inhibit proteasome-mediated protein degradation, however, MG132 treatment did not rescue the dramatic loss of ADAR1 upon ZYS-1 inhibition, suggesting independent of proteasome degradation (Supplementary Fig. S8D). Coincidently, overexpressed m6A methyltransferase METTL3/14 has been suggested to regulate N 6 -methyladenosine (m6A) on ADAR1 mRNA in glioblastoma. The m6A binding protein YTHDF1 was demonstrated to recognize m6A modi cation on ADAR1 mRNAs and promote its translation, as evidenced by ADAR1 protein reduction without affecting its mRNA level upon YTHDF1 knockdown (62). Therefore, we postulated ADAR1 degradation upon ZYS-1 treatment maybe through an m6A-YTHDF1 dependent mechanism as ZYS-1 led to a signi cant reduction of YTHDF1 protein level in PCa cells (Supplementary Fig. S8E). Though this may be one reason for ADAR1 protein reduction upon ZYS-1 treatment, the precise mechanism still needs further exploration.
Finally, due to the lack of crystal structural information for the deaminase domain of ADAR1, we performed in silico docking to elucidate the binding mode between ZYS-1 and ADAR1 homology model. Docking analysis suggested that some crucial hydrogen bonds interactions (V894, K1003, V1004 and G1009) formed between ADAR1 and 8-Br-Ado or ZYS-1, which are located closely with catalytic metalbinding sites (C966, H988) and deaminase domain (E912, F972, D973, K996). However, in the interaction between ZYS-1 and ADAR1, halogens such as 2-F and 8-Cl formed some weak σ-hole interactions with C893, L1031 or T1033, similar to hydrogen bonds, which play an important role in the process of protein and ligand recognition and binding. This may be the reason why compound ZYS-1 exhibits a higher a nity to ADAR1 compared to 8-Br-Ado. Further study to obtain the crystal structure of ADAR1 has been undertaken in our collaborated laboratory. Expectantly, the co-crystal structure would elucidate the binding mode and more precisely guide to optimize ZYS-1 to enhance the inhibitory activity, pharmacokinetic pro les, safety properties or anti-tumor e cacy.
In summary, our study carries out a model in which the disturbing of ADAR1-mediated A-to-G editing in MTDH triggers its translational shutdown, modulating tumor cells proliferation and invasion in prostate cancer. Encouragingly, we discover and develop a novel small-molecule ADAR1 inhibitor ZYS-1 with strong anti-tumor e cacy and favorable safety pro le. Meanwhile, we highlight the widespread applicability of ADAR1 inhibitors in various cancers. All these ndings may represent a new therapeutic strategy for prostate cancer and more malignancies by targeting epitranscriptomic RNA editing with single agent or combination with other therapies.

Synthesis of ZYS-1
To a solution of 2-uoroadenosine (5.70 g, 20 mmol) and pyridine (3.22 mL, 40 mmol) in acetonitrile (80 mL) was added drop-wise SOCl 2 (7.25 mL, 100 mmol) at 0 ºC. The reaction mixture was continuously stirred for 4 h, and then subsequently warmed to room temperature for overnight. The resulting suspension was concentrated in vacuo. To the reaction mixture was added methanol (120 mL), water (12 mL), and NH 4 OH (24 mL), followed by stirring for 0.5 h at r.t. The reaction mixture was concentrated and the solid was precipitated in the water phase, and the lter cake was obtained by ltration. The lter cake was re-dissolved in a small amount of methanol at 60 ºC, and dichloromethane was added dropwise. The solid was precipitated by cooling, ltered and washed with cold methanol to obtain the product ZYS-1 as a white solid (5.45 g, 90%). 1

Western Blotting
Western blotting assay was performed similarly as described previously (63). Cells were lysed in RIPA lysis buffer (Thermo Fisher Scienti c, 89901) at 4°C, and the lysate supernatant were added with protease inhibitor (Roche, 04693132001) and phosphatase inhibitor (Roche, 04906845001). Protein concentration of lysates was quanti ed using a BCA protein assay kit (Thermo Fisher Scienti c, 23227).
Equal amounts of protein were loaded onto 8-12% SDS-PAGE gel and transferred onto a polyvinylidene di uoride (PVDF) membrane, which was blocked with 0.5% skim milk in Tris-buffered saline supplemented with 0.5% Tween 20 (TBS-T) and then incubated with indicated primary antibodies.
Flowing washing in TBS-T, membranes was incubated with horseradish peroxidase-conjugated (HRP) secondary antibody for 1-2 h. The blots were visualized by a chemiluminscent detection system (Tanon, Shanghai, China).

Cell migration and invasion assays
Each transwell (Falcon, 353097) was seeded with 5 × 10 4 cells which have been starved in FBS-free medium for 12 h. Then, cells were allowed to migrate or invade to the bottom chamber for at least 24 h. Cells in the bottom of the chamber were xed in 4% paraformaldehyde (PFA) for 10 min and stained with crystal violet for 10 min. For invasion assay, upper chambers were precoated with matrigel (Corning, 356234) and incubated for 12 h at 37°C. Images were taken using an inverted microscope (Nikon) and cell counts were quanti ed using ImageJ software.
Colony formation assay 500-1000 cells/well were plated to 24-well plate and allowed for growth for 10-14 days, during which medium in wells were replaced every three days. 4% PFA was used to x colony for 10 minutes, followed by staining with crystal violet for 10 minutes. Images were taken using an inverted microscope and colony numbers were quanti ed using ImageJ software.

RNA-seq and Data Analysis
Cells were harvested and total RNA was extracted using Beyozol Total RNA Extraction Reagent (Beyotime, R0011) following the instruction of the manufacturer. Total mRNA was enriched by Obligo(dT) beads, then fragmented into short fragments and reversely transcribed into cDNA with random primers. After the second-strand cDNA was synthesized, cDNA fragments were puri ed, end repaired, added poly(A), and ligated to Illumina sequencing adapters. The ligation products were size selected by agarose gel electrophoresis, ampli ed by PCR, and sequenced using Illumina HiSeq2500 by Genedenovo Biotechnology Co., Ltd (Guangzhou, China). Raw reads were cleaned to remove adapters or low quality reads and rRNA mapped reads. Clean reads were mapped to Homo sapiens genome. Gene expressions were quanti ed by FPKM (fragment per kilobase of transcript per million mapped reads) value and differential expression analysis was performed by DESeq2 software between two different groups with false discovery rate (FDR) below 0.05 and absolute fold change ≥ 2 considered differentially expressed

Cell Cycle and Apoptosis Assays
For cell cycle analysis, cells were collected and xed with 70% ethanol overnight at 4°C. Cells were washed with PBS and stained with propidium iodide (PI) using Cell Cycle Analysis Kit (Beyotime, C1052) for 30 min at 37°C. After staining, samples were subjected immediately to ow cytometry analysis. For cell apoptosis analysis, YF®488-Annexin V and PI Apoptosis Kit (US Everbright, Y6002) was used following the manufacturer's instructions. Cells were collected and stained with YF®488-Annexin V and PI for 20 min at room temperature away from light. To avoid disturbance from GFP uorescence in shNC and shADAR1 cells, we used Annexin V-APC and 7AAD doubling staining method (Annexin V-APC and 7AAD Apoptosis Kit, US Everbright, A6030) to determine their apoptosis. Samples were subjected immediately to ow cytometry analysis. Flow cytometry was conducted using FACS Celesta (BD) and FACS Calibur (BD) and data was analyzed using FlowJo V10 software.

ADAR1 deaminase activity determination
The ability of compounds to inhibit puri ed ADAR1 protein deaminase activity was determined by Adenosine Deaminase (ADA) Activity Assay Kit (Fluorometric) (Abcam, ab204695) following the manufacturer's instructions. Measure output on a uorescent microplate reader at Ex/Em = 535/587 nm in a kinetic mode, every 2 minutes, for at least 60 minutes at 37°C protected from light. The uorescent value in two time points in compounds treated group relative to control group was regarded as remaining ADAR1 deaminase activity.

Flow Cytometry
Tumors from xenograft mice were ground mechanically, digested, and ltered with 70-µm screen to and FACS Calibur (BD) and data was analyzed using FlowJo V10 software.

Immunohistochemistry (IHC)
Tumors and key organs from prostate cancer xenograft mice were xed in 4% PFA followed by embedded in para n wax, sectioned in slides. Sections were incubated with 3% H 2 O 2 for 10 min to quench endogenous peroxidase activity followed by antigen retrieval using unmasking solution. Nonspeci c binding was blocked with 2% goat serum for 30 min. These sections were then subjected to ADAR1, Ki-67,

Structure-Based Virtual Screening
The three-dimensional structure of ADAR1 protein referred to the previously reported literature (48), and Discovery Studio 2021 (Accelrys, CA, USA) was employed to screen the combined ligand database, which was ltered by Lipinski's rule of ve, Veber's rule and the 'pan-assay interference compounds'. The structure-based virtual screening procedure contains two programs, LibDock and CDOCKER, LibDock is a molecular docking program based on a high-throughput algorithm, while the CDOCKER is a more accurate docking method that employs ligands exibly to match the binding pocket. Based on the virtual screening procedure, we obtained 20 compounds from the ligand database for experimental study.

Statistical analysis
The unpaired, two-tailed Student's t test was used to compare difference between two groups carried on SPSS 22.0 software, if not speci cally indicated in gure legends. For bioinformatics analysis, Student's t test and Mann-Whitney Wilcoxon test were used to compare continuous variables between two groups.
Analysis of variance (ANOVA) was utilized to test the variance among multiple groups. Paired, two-tailed Student's t test was used to compare ADAR1, ADAR2, and editing levels between 16 pairs matched tumor and normal tissues in TCGA, carried on GraphPad Prism 8.0 software. Kaplan-Meier plots and log-rank tests were used for relapse-free survival analysis. The values were expressed as mean ± SD with the numbers of replicates listed in gure legends. P < 0.05 was considered statistically signi cant.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.