Targeting ADAR1 p150 triggers tumor inhibition and antitumor immunity to overcome immunotherapy resistance Authors: Zhichao(Eric) Ai, Yuying Bian, Gang Li, Xiaobin S Wang, Sining Zhu, Yumeng Li, Xiangyu Yao, Zhipeng Yan Published: 2026 ABSTRACT SummaryImmunotherapy efficacy is limited by resistance. While ADAR1 regulates tumor immune evasion, targeting it is challenged by the essential role of the constitutive p110 isoform. Here, we identify the interferon-inducible p150 isoform and its Zα domain as critical therapeutic vulnerabilities. ADAR1 p150 is frequently overexpressed in human cancers and correlates with immunosuppressive environments. Genetic ablation of p150 inhibits tumor proliferation and triggers a type I interferon response, converting “cold” tumors to “hot.” Myeloid-specific ADAR1 deletion synergizes with anti-PD-1 therapy by enhancing antigen presentation. We establish the Zα domain as the essential determinant for dsRNA editing selectivity. Targeted disruption of this domain alone recapitulates the effects of complete ADAR1 ablation, leading to immunogenic dsRNA accumulation and potent anti-tumor immunity. This work provides a rationale for selectively targeting the ADAR1 p150-Zα axis to reverse malignant RNA editing and overcome immunotherapy resistance with a refined therapeutic index.Subject areas: Molecular biology, Immunology, Cancer Introduction Cancer immunotherapy, particularly immune checkpoint blockade (ICB) targeting the PD-1/PD-L1 axis, has fundamentally transformed oncology practice. However, therapeutic efficacy remains constrained by primary and acquired resistance mechanisms, predominantly orchestrated through immunosuppressive networks within the tumor microenvironment (TME).1 The conceptual framework of the “cancer-immunity cycle” delineates the requisite cascade for effective antitumor T cell responses while highlighting vulnerable nodes susceptible to dysregulation.2 Emerging evidence implicates nucleic acid sensing pathways as crucial determinants of therapeutic resistance, wherein malignant cells exploit intrinsic immunoregulatory mechanisms to evade detection. Central to this process is the adenosine deaminase acting on RNA 1 (ADAR1), an enzyme that catalyzes the adenosine-to-inosine (A-to-I) editing of double-stranded RNA (dsRNA).3,4 ADAR1 exists as two distinct isoforms: the constitutively expressed, nuclear-localized p110, and the interferon (IFN)-inducible p150. While both isoforms share a conserved catalytic deaminase domain and three dsRNA-binding domains (dsRBDs), p150 is uniquely characterized by an extended N-terminus containing a Zα domain. This domain serves as a specialized sensor that specifically recognizes and binds to left-handed Z-form nucleic acids (Z-NA), a high-energy conformer often associated with genomic instability and viral infection.4 Under physiological conditions, this editing serves as a fundamental mechanism of “self” versus “non-self” discrimination, marking endogenous RNAs to prevent their misrecognition by cytosolic dsRNA sensors like melanoma differentiation-associated protein 5 (MDA5) and protein kinase R(PKR), which modulate cell fate decision. However, in the context of cancer, chronic inflammation within TME drives sustained ADAR1 p150 overexpression, leading to hyper editing of immunogenic dsRNAs and subsequent immune evasion.4,5,6,7 Recent clinical and genetic evidence has solidified ADAR1 p150 as a promising therapeutic target. It is frequently overexpressed across a spectrum of human cancers, including leukemias, glioblastomas, and colorectal cancers, where its expression consistently correlates with poor prognosis and therapy resistance.4,6 Crucially, landmark studies have demonstrated that genetic disruption of ADAR1 in tumors can overcome resistance to ICB. This effect is mechanistically linked to the activation of not only MDA5 but also other dsRNA sensors like PKR, leading to growth inhibition and robust tumor inflammation.8,9,10,11 Consequently, by editing these endogenous dsRNAs, ADAR1 p150 acts as a key suppressor of endogenous dsRNA-mediated signaling, enabling tumor cells to evade immune-mediated clearance and maintain proliferative capacity. Notwithstanding compelling genetic validation of ADAR1 as an oncology target,12 clinical translation confronts a fundamental biological constraint: achieving tumor-specific efficacy without triggering systematic autoimmunity. While complete ADAR1 ablation demonstrates embryonic lethality, even isoform-specific germline knockout of p150 leads to early postnatal death in murine models, underscoring its essential role in maintaining immune homeostasis.13,14,15 This pathophysiological framework contraindicates broad-spectrum enzymatic inhibition strategies targeting both the immunomodulatory p150 isoform and the constitutively expressed p110 isoform responsible may possess a prohibitive toxicity profile for the therapeutic development of inhibitors that selectively disrupt ADAR1 p150. However, clinical observations in patients with Aicardi-Goutières syndrome (AGS) provide a critical nuance to this constraint.16 Many AGS-associated pathologies arise from missense mutations specifically within the Zα domain (e.g., N173S/P193A), which impair Z-RNA binding rather than abolishing the p150 protein itself.16 The survival of these individuals suggests that the Zα domain represents a functionally distinct and potentially druggable target. In the current study, we integrate clinical, genetic, and mechanistic evidence to establish the ADAR1 p150-Zα axis as a critical therapeutic vulnerability in cancer. We first provide clinical evidence of ADAR1 p150 overexpression correlating with immunosuppressive tumor microenvironments across malignancies. Genetic ablation studies then demonstrate that p150 depletion intrinsically inhibits tumor proliferation while extrinsically inducing inflammatory reprogramming of the tumor immune landscape. Furthermore, we establish that myeloid-specific ADAR1 p150 deletion synergizes with anti-PD-1 blockade by enhancing antigen presentation and myeloid cell activation. Mechanistically, we identify the Zα domain as the essential structural determinant governing dsRNA editing selectivity and innate immune recognition. Genetic disruption of this domain alone suffices to inhibit tumorigenesis and activate antitumor immunity, recapitulating the therapeutic effects of complete ADAR1 inhibition. By integrating these clinical correlations with functional genetics and structural insights, our work provides a compelling preclinical rationale for targeting the ADAR1 p150-Zα interface to reverse cancer immune evasion and overcome immunotherapy resistance. Results ADAR1 p150 isoform is overexpressed in cancers and correlates with poor prognosis Interrogation of multi-omic datasets from The Cancer Genome Atlas (TCGA) revealed significant overexpression of the ADAR1 p150 isoform across multiple human malignancies, including colorectal adenocarcinoma (CRC), pancreatic ductal adenocarcinoma (PDAC), lung squamous cell carcinoma (LUSC), and acute myeloid leukemia (AML) (Figure 1A). Quantitative histological assessment confirmed pronounced p150 overexpression in tumor tissues compared to matched normal counterparts, with particularly strong expression in gastrointestinal cancers and LUSC (Figure S1A). ADAR1 p150 isoform is overexpressed in human cancers and correlates with poor prognosis (A) Analysis of TCGA datasets showing ADAR1 p150 mRNA expression levels across cancer types. Data are presented as log2(TPM+1). Statistical significance was determined by unpaired, two-tailed Student’s t test (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001). (B) Correlation analysis between ADAR1 p150 expression and ESTIMATE immune scores in multiple cancer types. Pearson correlation coefficients and p values are shown. (C) Representative multiplex immunofluorescence images of a representative human colorectal cancer tissue microarray (TMA) core stained for ADAR1 p150 (red), CD8 (cyan), ARG1 (yellow), and PD-L1 (magenta). Scale bar, 50 μm. (D and E) Quantification of spatial associations from the TMA (MC2082d) showing correlations between tumor ADAR1 p150 expression and the abundance of stromal CD8+ T cell, ARG1+ macrophage, and tumor cell PD-L1 expression. Scale bar, 100 μm. Pearson correlation was shown. To delineate the relationship between ADAR1 p150 expression and tumor microenvironment composition, we performed systematic correlation analysis using ESTIMATE algorithm-derived immune and stromal scores. ADAR1 p150 expression showed significant positive correlations with immune cell infiltration scores in CRC (r = 0.4721, p = 0.0198), PDAC (r = 0.1648, p = 0.0275), and LUSC (r = 0.2151, p = 1.04E−05) (Figures 1B and 1C), in contrast to p110 isoform (Figure S1B). Multiplex immunofluorescence analysis of tumor microarrays demonstrated spatial associations between ADAR1 p150 expression and specific immune cell populations, revealing positive correlations with CD8+ cytotoxic T cells, ARG1+ M2-polarized macrophages, and tumor cell PD-L1 expression (Figures 1D and 1E). These findings establish ADAR1 p150 as frequently overexpressed across diverse malignancies and associated with distinct tumor microenvironment compositions characterized by the presence of both cytotoxic and immunosuppressive immune populations. p150 depletion inhibits tumor growth with augmented inflammation Initial assessment of ADAR1 dependency across tumor cell lines using DEPMAP datasets revealed a significant correlation between high ADAR1 dependency scores and sensitivity to ADAR1 loss in colony formation assays (Figure S2A). Then, to functionally characterize the oncogenic dependency on ADAR1 p150, we employed syngeneic transplant models utilizing isogenic wild-type and ADAR1-knockout B16-F10 melanoma cells and KPC pancreatic cancer cells engrafted in IFN-β/Luc reporter mice, which tracks IFN-β gene induction in vivo (14) (Figure 2A).Orthotopic implantation of ADAR1-knockout cells into immunocompetent syngeneic hosts led to profound suppression of tumor growth compared to wild-type controls in both B16-F10 melanoma (p < 0.0001) and KPC pancreatic cancer (p < 0.001) models (Figures 2B and S2B). p150 depletion is sufficient to inhibit tumor growth inhibition and induce tumor inflammation (A) Schematic of the experimental design using syngeneic transplant models in IFN-β/Luc reporter mice. (B) Tumor growth curves of wild-type versus ADAR1-knockout B16-F10 cells in C57BL/6 mice (n = 10 mice per group). Data are mean ± SEM. Statistical significance was determined by two-way ANOVA (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001). (C and D) In vivo bioluminescence imaging (D) and quantification of photon flux (E) from tumors in IFN-β/Luc reporter mice bearing WT or ADAR1-KO B16-F10 tumors (n = 5 mice per group). Data are mean ± SD. Unpaired t test (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001). (E) Flow cytometric analysis of tumor-infiltrating leukocytes in WT and ADAR1-KO B16-F10 tumors (n = 4 tumors per group from two independent experiments). (F) Cytokine levels of IFN-β and CXCL10 in tumor homogenates measured by MultiPlex (n = 5 tumors per group). Data are mean ± SD. Unpaired t test (∗∗p < 0.01, ∗∗∗p < 0.001). (G and H) Interferon-stimulated genes (ISGs) expression (H) and heatmap of DEGs with corresponding gene enrichment analysis (I) of RNA-seq data from WT and ADAR1-KO B16-F10 cells (n = 2 biological replicates per group), showing upregulation of ISGs and dsRNA sensing pathways. Concurrent with tumor growth inhibition, we observed robust activation of interferon signaling via bioluminescence imaging, with maximal photon fluxes were significantly increased in ADAR1-knockout tumors relative to wild-type controls across both tumor models (Figures 2C, 2D, S2C, and S2D). Profiling of tumor-infiltrating leukocytes by flow cytometry revealed substantial remodeling of the tumor immune landscape. ADAR1-deficient tumors exhibited significant increases in proinflammatory monocytes, dendritic cells(DCs), and NK cell frequencies, alongside elevated cytokine production, alongside elevated production of cytokines including IFN-β and CXCL10 (Figures 2E, 2F, S2E, an d S2F). Transcriptomic analysis of ADAR1 p150-depleted cells demonstrated significant upregulation of ISGs and dsRNA sensing pathways, confirming the induction of cell-intrinsic immune activation (Figures 2G, 2H, S2G, and S2H). These findings demonstrate that ADAR1 p150 serves as a critical checkpoint against cell fate decision, and its ablation unleashes a coordinated anti-tumor response through both direct effects on tumor cell viability and profound remodeling of the tumor immune microenvironment. Myeloid-specific ADAR1 p150 deletion synergizes with anti-PD-1 therapy Given the prominent role of myeloid cells in shaping antitumor immunity, we investigated the functional consequences of myeloid-specific ADAR1 p150 deletion alone and in combination with anti-PD-1 therapy. Longitudinal assessment of tumor growth kinetics in MC38 syngeneic models demonstrated that myeloid-specific ADAR1 p150 deletion (LyzM-Cre Adar1-p150-fl/fl) synergistically enhanced anti-PD-1 efficacy. Combination therapy achieved profound tumor growth inhibition compared to monotherapy arms, without significant alterations in body weight (Figures S3A–S3C). Flow cytometric quantification confirmed that this synergy was driven by a remodeled immune landscape, characterized by augmented frequencies of proinflammatory monocytes, CD8+ T cells, and NK cells within the tumor-infiltrating leukocyte compartment. Integrated single-cell RNA sequencing of the whole tumor cell compartments identified 20 transcriptionally distinct cellular clusters (Figures S3D–S3E) and observed increased immune cell activation with enhanced gene expression related to immune effector functions in myeloid-specific ADAR1 p150-deficient tumors (Figures S3F–S3H). Uniform manifold approximation and projection (UMAP) dimensional reduction demonstrated substantial remodeling of the myeloid compartment, with myeloid-specific ADAR1 p150-deficient tumors exhibiting expansion of monocyte-derived macrophage and concomitant contraction of myeloid-derived suppressor cells (MDSCs), and immunosuppressive macrophage (M2-macrophage) (Figures 3A–3E). Feature plot visualization revealed enhanced expression of activation markers (Cd84, Adgre1, and Irf7) and suppression of immunoregulatory genes (Arg1, Il1b, and Srgn) across ADAR1 p150-deficient myeloid populations (Figure 3B). Myeloid-specific ADAR1 deletion reprograms the tumor microenvironment and synergizes with anti-PD-1 therapy (A) UMAP visualization of single-cell RNA sequencing (scRNA-seq) data from total tumor cells of ∗LyzM-Cre;Adar1fl/fl∗ and control tumors (n = 3 tumors per group, pooled cells). (B) Feature plots showing expression of activation markers (Cd84, Adgre1, and Irf7) and immunoregulatory genes (Arg1, Il1, and Srgn) across the UMAP space. (C) Pseudotemporal trajectory analysis of myeloid cells from control and ADAR1-deficient tumors. (D) Proportion of cells from control and ADAR1-deficient tumors along the pseudotime trajectory. (E) Expression patterns of antigen presentation and relevant genes (H2-Aa, H2-Ab1, and Cd74) along the pseudotime trajectory. (F) Schematic of the allogeneic mixed leukocyte reaction (MLR) assay. (G and H) Flow cytometric analysis of CD80 and CD68 expression on ADAR1-deficient monocytes (G) and subsequent T cell proliferation (H) in the MLR (n = 4 independent experiments). Data are mean ± SD. Unpaired t test (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001). (I) Levels of inflammatory cytokines and chemokine in MLR supernatants measured by cytokine array (n = 2). Unpaired t test (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001). To understand the drivers of this shift, we performed pseudotemporal trajectory analysis, which reconstructed the differentiation continuum from monocytic precursors to terminal myeloid states (Figure 3C). This analysis demonstrated that ADAR1 p150 deletion steers myeloid commitment toward immunostimulatory fates, bypassing the suppressive differentiation programs typically enforced by TME (Figure 3D). Analysis of trajectory-correlated genes revealed a progressive upregulation of antigen presentation machinery (H2-Aa, H2-Ab1, and Cd74) and co-stimulatory molecules along this differentiation axis (Figure 3E). Differential expression analysis identified 1,487 significantly regulated genes (adj.p < 0.05, |log2FC| > 1), with notable upregulation of genes central to interferon responses and antigen presentation pathways in ADAR1 p150-deficient myeloid populations (Figures S3F–S3G). Cell-type-specific ontology enrichment confirmed distinct transcriptional reprogramming: while myeloid cells showed enhanced capacity for T cell priming, tumor-infiltrating T cells exhibited elevated cytotoxic programs, indicating that the p150-deficient myeloid environment provides essential “second signals” for T cell activation (Figure S3H). To validate our computational findings from sc-RNA sequencing, we performed histological analysis on excised tumor sections. Immunofluorescence staining demonstrated that myeloid-specific ADAR1 deletion (Lyz-Cre), when combined with anti-PD1 therapy, led to a significantly higher accumulation of CD8+ T cells and a marked reduction in Gr-1+ MDSCs within TME compared to the WT+anti-PD1 group(Figure S4A). Functional validation through allogeneic mixed leukocyte reactions(MLR, Figure 3F) demonstrated enhanced myeloid cell activation in ADAR1-deficient monocytes, resulting in increased levels of CD80 and CD68, markers of macrophage activation, and enhanced T cell stimulatory capacity with significant increases in T cell proliferation (Figures 3G and 3H). Furthermore, ADAR1p150-deficient antigen-presenting cells triggered elevated production of inflammatory cytokines including CCL2 and TNFA (Figure 3I). Collectively, these results demonstrate that myeloid-specific ADAR1 p150 serves as a non-redundant regulator of the tumor immune microenvironment. Its deletion orchestrates a comprehensive multicellular reprogramming, shifting myeloid differentiation trajectories to establish a permissive niche for T cell priming. This transition from a suppressive to an immunostimulatory compartment underlies the observed synergy with ICB, providing a strong rationale for targeting the ADAR1-p150-Zα axis in myeloid lineages. The Zα domain governs dsRNA editing and innate immune sensing To delineate the molecular mechanisms underlying ADAR1 p150 function, focusing on the essential role of the Zα domain in dsRNA editing. We introduced a specific point mutation (N173S) that disrupts Z-RNA binding capacity while preserving the overall protein structure.17 Similarly, the Y179A mutation was employed as an additional loss-of-function variant, as this residue is critical for the architectural recognition of the Z-conformation.11,15,18 Reconstitution experiments in ADAR1-responsive luciferase reporter cells demonstrated that wild-type ADAR1 p150 efficiently suppressed dsRNA-mediated reporter activation in a dose-dependent manner, whereas the Zα domain mutant (N173S) exhibited significantly impaired suppression capacity, indicating compromised RNA editing function (Figures 4A and 4B). Zα domain governs dsRNA editing and innate immune sensing (A) Schematic of the ADAR1 p150 protein domain structure and the location of the Zα domain point mutation (N173S). (B) Reconstitution assay in ADAR1-KO reporter cells showing dose-dependent suppression of dsRNA-mediated luciferase activation by WT p150, but not the Zα (N173S) mutant (n = 4 independent experiments). two-way ANOVA (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001). (C) Representative immunofluorescence images of WT and Zα (N173S) mutant cells stained with the J2 anti-dsRNA antibody (green). Scale bar, 20 μm. (D) HTRF analysis of phosphorylated PKR (p-PKR) in WT and Zα mutant cells under basal and IFN-β-stimulated conditions. Data are representative of three independent experiments. (E) Heatmap of ISG expression from RNA-seq of WT and Zα mutant cells (n = 2 biological replicates). (F) Quantification of A-to-I editing efficiency at specific sites in the AZIN1 and miR-378 transcripts from RNA-seq data. (G) Colony formation assays of WT and Zα mutant cells treated with IFN-β (100 pg/mL). Scale bar, 10 mm. We subsequently utilized IFN treatment to simulate the inflammatory tumor microenvironment and to maximize ADAR1 p150 expression, thereby intensifying the dependence of the cell on Zα-mediated RNA masking. Immunofluorescence imaging using the J2 monoclonal antibody, which specifically recognizes unedited dsRNA, revealed substantial accumulation of dsRNA species in Zα mutant cells compared to wild-type counterparts (Figure 4C). This accumulation of unedited dsRNAs resulted in increased phosphorylation of dsRNA sensor PKR under both basal and IFN-stimulating conditions (Figure 4D). Transcriptomic profiling confirmed significant upregulation of ISGs including MX1, OAS1, and ISG15 in Zα mutant cells, establishing a cell-intrinsic interferon response (Figure 4E). Mechanistically, these results suggest that the defective Zα domain fails to sequester the heightened burden of immunogenic Z-RNA induced by IFN, triggering an amplified feedforward loop that overwhelms cellular homeostasis and potentiates apoptosis. Deep transcriptional sequencing analysis of WT and Zα mutant cells revealed that the N173S mutation significantly reduced A-to-I editing efficiency at known ADAR1 target loci, such as AZIN1 and miR-378 (Figure 4F). To evaluate global editing activity, we calculated the global Alu editing Index across human cell lines. While IFNβ treatment markedly induced global editing, Zα mutant cells consistently exhibited a reduced global editing index compared to WT p150 (Figure S4B).These findings indicate that an intact Zα domain is required for the efficient recruitment of ADAR1 p150 to its Z-RNA substrates. Interestingly, while the N173S mutation reduced the Alu editing index (AEI) under basal conditions, we observed an elevation in AEI following IFN-b treatment (Figure S4B). This suggests that while the Za domain is essential for standard p150 function, its disruption in a high-IFN environment may interfere with the normal feedback inhibition of ADAR1, potentially leading to dysregulated editing levels. This phenomenon in KYSE70 cells highlights the complex interplay between IFN-induced ADAR1p150 expression and Z-RNA-mediated structural regulation. Functional assessment through colony formation assays demonstrated that Zα disruption sensitized tumor cells to exogenous interferon signaling, with mutant cells exhibiting significantly reduced clonogenic capacity following IFN treatment compared to wild-type controls (Figure 4G). Consistent with our molecular observations, this suggests that the failure of the defective Zα domain to sequester and resolve the increased burden of immunogenic Z-RNAs induced by IFN creates a terminal stress signal, effectively suppressing tumor cell proliferation, and survival. Genetic disruption of the Zα domain inhibits tumor growth and activates immunity To evaluate the therapeutic potential of selectively targeting the Zα domain, we engineered B16-F10 cells expressing wild-type ADAR1 p150 or isogenic Zα domain point mutants (N175S and Y179A) that disrupt Z-RNA binding (Figure 5A). Transcriptomic profiling revealed substantial differential gene expression between wild-type and Zα mutant cells under both basal and IFN-β-stimulated conditions (Figures 5A and 5B). Gene ontology enrichment analysis demonstrated a significant shift in the cellular landscape. Zα mutant cells exhibited a robust enrichment of innate immune activation pathways, coupled with a coordinated downregulation of metabolic and biosynthetic genes (Figure 5B). Genetic disruption of the Zα domain inhibits tumor growth and activates anti-tumor immunity (A) Schematic of the B16-F10 cells engineered to express WT ADAR1 p150 or Zα domain mutants (N175S, Y179A). (B) Heatmap of DEGs between WT and Zα (N175S) mutant cells under basal conditions (n = 3 biological replicates per group). (C) Gene ontology (GO) enrichment analysis of upregulated pathways in Zα mutant cells compared to WT. (D) Heatmap of ISG upregulation in Zα mutant cells compared to WT under PBS and IFNb condition. (E) Colony formation assay of WT and Zα mutant cells with or without IFN-β treatment. Scale bar, 10 mm. (F) Tumor growth curves of syngeneic mice implanted with B16-F10 cells expressing WT, heterozygous N175S, homozygous Y179A, or homozygous N175S ADAR1 p150 (n = 8 mice per group). two-way ANOVA ([∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001] for mutants vs. WT). (G) Flow cytometric quantification of tumor-infiltrating leukocytes in WT and Zα (N175S) mutant tumors at endpoint (n = 6 tumors per group). Data are mean ± SD. two-way ANOVA ([∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001] for mutants vs. WT). Comparative analysis of differentially expressed genes (DEGs) between wild-type and mutant cells (N175S and Y1719A) confirmed significant upregulation of ISGs including IFIT1, ISG15, and MX1 in Zα domain mutants (Figures 5C and 5D).Consistent with our earlier observations that the defective Zα domain fails to sequester the increased burden of IFN-induced Z-RNA, functional assessment through colony formation assays showed that Zα mutant cells were hypersensitive to IFN-treatment, exhibiting a profound loss of clonogenic capacity compared to wild-type controls (Figure 5E). In syngeneic mouse models, both N175S and Y179A mutations significantly inhibited tumor growth, with tumor growth inhibition (TGI) rates of 55.2%, 92.8%, and 99.7% for heterozygous N175S, homozygous Y179A, and homozygous N175S mutants, respectively (Figure 5F). This growth arrest was accompanied by a transformation of the tumor immune microenvironment. Immunophenotyping of tumor-infiltrating leukocytes revealed substantial increases in CD8+ T cells, natural killer cells, and activated dendritic cells in Zα mutant tumors compared to wild-type controls (Figure 5G), indicating that Zα disruption effectively transforms immuno-suppression to provoke anti-tumor inflammatory responses. Complementary rescue experiments in ADAR1-knockout B16-F10 cells confirmed the functional necessity of an intact Zα domain (Figure S5). Western blot analysis validated successful reconstitution of p150, p110, and Zα mutant p150 protein expression in ADAR1-knockout cells (Figure S5A). Colony formation assays demonstrated that while wild-type p150 fully rescued the interferon-sensitive phenotype of ADAR1-knockout cells, Zα mutant p150 failed to restore interferon resistance (Figure S5B). Similarly, in vivo tumor growth studies and immune profiling revealed that Zα mutant p150 could not rescue the tumor-suppressive phenotype observed in ADAR1-knockout models, with sustained tumor growth inhibition and enhanced immune infiltration persisting in Zα mutant-reconstituted tumors (Figures S5C and S5D). These rescue data underscore the functional necessity of Zα-mediated RNA masking for tumor survival. These findings demonstrate that targeted disruption of the ADAR1 Zα domain is sufficient to recapitulate the potent anti-tumor effects of complete ADAR1 ablation. By establishing this domain as a critical gatekeeper for innate immune evasion, our results provide a robust rationale for Zα-specific therapeutic interventions to activate intrinsic interferon responses and enhance anti-tumor immunity. Discussion The current study integrates clinical, genetic, and mechanistic evidence to establish the ADAR1 p150 isoform and its Zα domain as critical determinants of tumor immunogenicity and therapeutic targets. Our findings support a model wherein malignant cells co-opt the evolutionarily conserved immunoregulatory function of ADAR1, primarily through interferon-driven overexpression of the p150 isoform, to suppress the immunogenicity of endogenous dsRNAs derived from repetitive genetic elements. This enzymatic activity subverts the activation of cytosolic nucleic acid sensors, including MDA5 and PKR, thereby facilitating immune evasion. Furthermore, the results presented herein specifically elucidate the non-redundant role of the Zα domain in governing dsRNA editing selectivity and innate immune sensing, offering a refined mechanistic framework for therapeutic intervention. The functional uniqueness of the p150 isoform is inextricably linked to its Zα domain. This domain acts as a critical molecular switch governing the balance between immune tolerance and activation.19,20,21 Our data, consistent with structural studies,22,23,24,25 confirm that the Zα domain is essential for efficient dsRNA editing. When this domain is disrupted, the accumulation of unedited immunogenic dsRNAs leads to dsRNA sensor activation, triggering a robust IFN-I response and converting immunologically “cold” tumors into “hot” ones. This aligns with the concept that tumors with high intratumoral heterogeneity (ITH), including genetic and epigenetic diversity, can evolve mechanisms to suppress such immunogenic responses, and ADAR1 is a key player in this suppression.26 The potent anti-tumor effects and immune activation observed upon Zα domain disruption highlight that targeting this specific interface, rather than the catalytic domain, could effectively unmask tumor cells without completely abolishing ADAR1’s essential physiological functions. Our results demonstrate that myeloid-specific ADAR1 p150 deletion synergizes with anti-PD-1 therapy can be contextualized within the conceptual framework of the cancer-immunity cycle and its regulatory checkpoints.2 The synergy arises from the coordinated targeting of two complementary barriers to effective anti-tumor immunity. Myeloid cells, including dendritic cells and macrophages, are crucial for initiating and executing the cancer-immunity cycle.2 Myeloid-specific ADAR1p150 ablation augments antigen presentation and fosters a pro-inflammatory tumor microenvironment, thereby potentiating the initial phases of the cancer-immunity cycle—specifically, antigen release and presentation. ADAR1 functions as a key immunoregulatory node across multiple cellular compartments within TME. Our findings align with and extend the observations by Ishizuka et al.,10 who demonstrated that tumor-intrinsic ADAR1 loss induces a repolarization of the myeloid compartment, leading to improved T cell responses. Crucially, our results also resonate with the findings of Zhang et al.,11 who revealed that ADAR1 p150 specifically masks the immunotherapeutic potential of ZBP1-driven necroptosis. Our study moves beyond correlation and directly demonstrated, through myeloid-specific genetic deletion, that myeloid ADAR1 is a non-redundant regulator of cellular immunogenicity. Additionally, while others implicated ADAR1 in shaping myeloid cell function,17,27,28,29 our results confirm that ADAR1 p150 deficiency enhances antigen-presenting cell activity, and establish the Zα domain as a requisite mediator of the cell-intrinsic tumor-suppressive effects induced by ADAR1 disruption. By demonstrating that targeted disruption of ADAR1 Zα domain is sufficient to recapitulate the anti-tumor efficacy of complete ADAR1 inhibition, we validate a therapeutic path to selectively inhibit the ADAR1-mediated immune-silencing function while potentially preserving essential physiological editing. The compelling preclinical efficacy of targeting the ADAR1 pathway has catalyzed the development of several preclinical and clinical stage modalities.12 Our data provides a robust rationale for therapeutically targeting ADAR1 p150 isoform without disrupting the physiological functions of the p110 isoform. It is evidenced by Rebecsinib, a small-molecule splice-switching modulator that selectively suppresses p150 expression and re-sensitizes tumors to immune attack.30 Recent strategies have focused specifically on the p150-unique Zα domain. These include AVA-ADR-001, a potent Zα domain inhibitor, and its derivative AVA-ADR-703, a first-in-class PROTAC (proteolysis targeting chimera) designed to induce the targeted degradation of the p150 protein.31 Additionally, indirect approaches like CBL0137 mimic ADAR1 deficiency by inducing Z-DNA formation, which triggers ZBP1-dependent necroptosis.32 Together, these diverse strategies form the foundation for an emerging class of “viral mimicry” therapies, which trigger innate sensing of endogenous nucleic acids as foreign, effectively tricking the immune system into recognizing and attacking the tumor as if it were a viral infection. Future research must focus on several key areas to fully realize the potential of ADAR1-directed therapy. First, a critical frontier lies in the discovery of potent and selective Zα targeting molecules that can bind to this specialized domain and interfere with its essential functions. By disrupting the ability of the p150 isoform to recognize and bind Z-nucleic acids, these inhibitors could prevent the suppression of innate immune sensing. Second, identifying robust biomarkers for patient selection—such as ADAR1 editing signatures, dsRNA accumulation, or specific tumor immune phenotypes—will be crucial for clinical success. Furthermore, the inflammatory consequences of ADAR1 inhibition, while therapeutically beneficial, require careful clinical management to balance anti-tumor efficacy with potential autoinflammatory toxicity. Overall, targeting the ADAR1 p150-Zα axis provides a mechanism-based strategy to reverse aberrant RNA editing and activate intrinsic anti-tumor immunity. By modulating these pathways, this approach addresses a key driver of resistance to current immunotherapies, offering a specific means to restore immune recognition in non-responsive tumors. Limitations of the study While our study establishes the ADAR1 p150-Zα axis as a compelling therapeutic target for overcoming immunotherapy resistance, several limitations should be considered. First, our functional genetic studies and syngeneic models were primarily conducted in murine systems. While these models are essential for assessing systemic anti-tumor immunity, they may not fully recapitulate the complexity of the human immune system or the diverse genomic landscapes of clinical cancers. Second, although we demonstrated that point mutations in the Zα domain effectively inhibit tumor growth, our work relies on genetic disruption rather than pharmacological inhibition. The development of small-molecule inhibitors that can achieve similar selectivity and potency remains a critical future challenge. Furthermore, while we identified p150-mediated RNA editing as a driver of immune evasion, the precise identity of the endogenous dsRNA substrates that trigger the sensing response in specific tumor types remains to be fully characterized. Finally, the potential for autoinflammatory toxicity following long-term ADAR1 p150 inhibition requires further evaluation in diverse physiological contexts. Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Zhipeng Yan (zhipeng.yan@roche.com). Materials availability All unique reagents generated in this study are available from the lead contact with a completed materials transfer agreement. Data and code availability •RNA-seq and scRNA-seq data have been deposited at GEO and are publicly available as of the date of publication, with accession number GSE314521, GSE316054. RNA-seq and scRNA-seq data have been deposited at GEO and are publicly available as of the date of publication, with accession number GSE314521, GSE316054. •All codes used have been stored at DOI: https://zenodo.org/doi/10.5281/zenodo.19756546. All codes used have been stored at DOI: https://zenodo.org/doi/10.5281/zenodo.19756546. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Acknowledgments This work was supported by China Innovation Center of Roche (CICoR) R&D Center (China) Ltd. This study was funded and supported by CICoR. The authors would like to thank all the researchers in the CICoR and the external collaborators who provided their expertise and assistance in this study. Author contributions E.A., X.Y., and Z.Y. conceived and planned the experiments; E.A., X.S.W., S.Z., Y.B., and G.L. carried out the experiments; Y.L., G.L., X.S.W., S.Z., and E.A. contributed to the analysis and interpretation of the results; E.A. took the lead in writing the manuscript in consultation with X.Y. and Z.Y. All authors reviewed the results and approved the final version of the manuscript. The data that support the findings of this study are available from the corresponding authors, X.Y. and Z.Y., upon reasonable request. Declaration of interests Z.A., X.S.W., S.Z., Y.B., G.L., Y.L., X.Y., and Z.Y. are employees of China Innovation Center of Roche (CICoR) and shareholders of Roche. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesAnti-mouse CD45-BV785 (Clone 30-F11)BioLegendCat# 103149; RRID: AB_2564590Anti-mouse CD3e-APC/Cy7 (Clone 145-2C11)BioLegendCat# 100330; RRID: AB_1877170Anti-mouse CD4-PerCP/Cy5.5 (Clone GK1.5)BioLegendCat# 100434; RRID: AB_893324Anti-mouse CD8a-BV711 (Clone 53-6.7)BioLegendCat# 100759; RRID: AB_2563510Anti-mouse NK-1.1-PE/Cy7 (Clone PK136)BioLegendCat# 108714; RRID: AB_389364Anti-mouse CD11b-BV605 (Clone M1/70)BioLegendCat# 101257; RRID: AB_2565431Anti-mouse F4/80-APC (Clone BM8)BioLegendCat# 123116; RRID: AB_893481Anti-mouse Ly-6G/Ly-6C (Gr-1)-FITC (Clone RB6-8C5)BioLegendCat# 108406; RRID: AB_313371Anti-mouse CD11c-PE (Clone N418)BioLegendCat# 117308; RRID: AB_313777Anti-mouse MHC Class II (I-A/I-E)-BV421BioLegendCat# 107632; RRID: AB_2650896Anti-mouse ARG1-PE (Clone A1exF5)Thermo FisherCat# 12-3697-82; RRID: AB_2734839Anti-mouse PD-L1-APC (Clone 10F.9G2)BioLegendCat# 124312; RRID: AB_10612741Anti-mouse IFN-γ-BV421 (Clone XMG1.2)BioLegendCat# 505830; RRID: AB_2563105Anti-mouse TNF-α-PE (Clone MP6-XT22)BioLegendCat# 506306; RRID: AB_315427Anti-dsRNA J2 monoclonal antibodySCICONSCat# 10010200; RRID: AB_2651015Anti-mouse PD-1 blocking antibody (RMP1-14)Bio X CellCat# BE0146; RRID: AB_10949053Rat IgG2a Isotype Control (Clone 2A3)Bio X CellCat# BE0089; RRID: AB_1107769Anti-ADAR1 p150Cell Signaling TechnologyCat# 70257; RRID: AB_3750741Biological samplesHuman peripheral blood mononuclear cells (PBMC)STEMCELL TechnologiesN/AChemicals, peptides, and recombinant proteinsRecombinant murine GM-CSFPeproTechCat# 315-03Recombinant murine IL-4PeproTechCat# 214-14Critical commercial assaysPan B cell isolation kit, humanMiltenyi BiotecCat# 130-101-638Pan Monocyte isolation kit, humanMiltenyi BiotecCat# 130-096-537Pan T Cell isolation kit, humanMiltenyi BiotecCat# 130-096-535Cell Activation Cocktail (with Brefeldin A)BioLegendCat# 423304Foxp3/Transcription Factor Staining Buffer SetThermo FisherCat# 00-5523-00LEGENDplex™ Mouse Anti-Virus Response PanelBioLegendCat# 740622Zombie Aqua™ Fixable Viability KitBioLegendCat# 423102Fixable Viability Stain 620BD HorizonCat# 564996Precision Count BeadsBioLegendCat# 424902R&D Systems Mouse Cytokine Array Panel AR&D SystemsCat# ARY006Deposited dataSingle-cell RNA-seq dataThis paperGSE314521Buk RNA-seq dataThis paperGSE316054Experimental models: cell linesMouse: B16-F10 (Melanoma)ATCCCRL-6475Mouse: MC38 (Colon adenocarcinoma)Shanghai Model Organisms CenterN/AMouse: KPC (Pancreatic cancer)Shanghai Model Organisms CenterN/AHuman: OE33 (Esophageal adenocarcinoma)ATCCCRL-2110Human: BJATCCCRL-2522Human: DLD-1ATCCCCL-221Human: AsPC-1ATCCCRL-1682Human: CAL 27ATCCCRL-2095Human: HT-29ATCCHTB-38Human: PLC/PRF/5ATCCCRL-8024Human: CFPAC-1ATCCCRL-1918Human: OE33ATCCCRL-2110Human: SW1990ATCCCRL-2172Human: HPAF-IIATCCCRL-1997Experimental models: organisms/strainsMouse: Adar1ˆfl/fl: B6.129S-Adar1ˆtm1.1Dfg/JShanghai Model Organisms CenterN/AMouse: Lyz2ˆCre: B6.129P2-Lyz2ˆtm1(cre)Ifo/JShanghai Model Organisms CenterN/AMouse: IFN-β/Luciferase reporterShanghai Model Organisms CenterN/AOligonucleotidessgRNA targeting ADAR Zα: 5′-actcgattgatttctttctt-3′This paperN/ARecombinant DNAPlasmid: pSpCas9(BB)-2A-GFPAddgeneAddgene Plasmid #48138Software and algorithmsFlowJo (Version 10.8)BD Life Scienceshttps://www.flowjo.comGraphPad Prism (Version 10.0)GraphPad Softwarehttps://www.graphpad.comSeurat (Version 2.3.4)Satija Labhttps://satijalab.org/seurat/STAR aligner (v2.7.9a)Dobin et al.33https://github.com/alexdobin/STAROtherPannoramic SCAN Image system3DHistechhttps://www.3dhistech.comHALO Image Analysis platformIndica Labshttps://indicalab.com/halo/R codesThis paperhttps://doi.org/10.5281/zenodo.19756547 Experimental model and study participant details Mice All mouse experiments were conducted in accordance with protocols approved by the Institutional Animal Care and Use Committee (IACUC) of the China Innovation Center of Roche(ONC-AS2024-042). Adar1fl/fl mice (Strain: B6.129S-Adar1tm1.1Dfg/J) and Lyz2Cre mice (B6.129P2-Lyz2tm1(cre)Ifo/J) were obtained from Shanghai Model Organisms Center, Inc. The IFN-β/Luciferase reporter mice (on a C57BL/6J background) were also sourced from Shanghai Model Organisms Center, Inc. To generate mice with myeloid-specific ADAR1 deletion, Adar1fl/fl mice were crossed with Lyz2Cre mice to produce Lyz2Cre;Adar1fl/fl (hereafter ∗LyzM-Cre;Adar1∗fl/fl) and control Adar1fl/fl littermates. All mice were maintained on a C57BL/6J genetic background. Age-matched (8-12 weeks old) female and male mice were used for all syngeneic tumor studies and were randomly assigned to experimental groups. Mice were housed under specific pathogen-free conditions with a 12-hour light/dark cycle and provided with autoclaved food and water ad libitum. Cell lines The following cell lines were utilized for in vitro assays and in vivo syngeneic models: MC38 (mouse colon adenocarcinoma) and KPC (mouse pancreatic ductal adenocarcinoma) cell lines were obtained from the Shanghai Model Organisms Center. KPC cells were originally derived from spontaneous tumor models carrying Pdx1-cre/LSL-Kras G12D/P53(R172H) transgenes. B16-F10 (mouse melanoma) cells were used for genetic ablation and reconstitution studies. Murine cancer cell lines MC38, KPC, B16-F10 cells were purchased from Shanghai Model Organism. Human tumor cell lines, including BJ (CRL-2522) and DLD-1 (CCL-221), were purchased from ATCC. Additional human cell lines used for global editing assessment included ASPC, CAL27, KYSE70, OE21, HT29, SW1990, and OE33, were purchased from ATCC. Cell Culture Conditions: All cell lines were maintained at 37°C with 5% CO2 condition, and cells were cultured under the recommended condition by ATCC. All cells were harvested during the exponential growth phase, authenticated via STR profiling, and tested negative for mycoplasma contamination for experimental use. Human peripheral blood mononuclear cells (PBMCs) were purchased from STEMCELL Technologies. The use of human peripheral blood mononuclear cells (PBMCs) in this study was in accordance with the ethical principles outlined in the Declaration of Helsinki and adhered to the guidelines set forth by the U.S. Food and Drug Administration (FDA), the U.S. Department of Health and Human Services (HHS), and the International Conference on Harmonization (ICH), as approved by the Institutional Review Board (IRB) (IRB Tracking Number: 20152869, Panel: 4). Material acquisition was conducted by StemExpress, LLC (Sponsor Protocol Number: 401−01), and all procedures were performed at approved research locations under the supervision of qualified personnel. Ten tubes of PBMC samples from two independent healthy donors were used for the current study. Method details Cell culture and cellular assays Mouse tumor tissues and spleens were homogenized using the gentleMACS Dissociator (Miltenyi Biotec). The homogenate was filtered through 40 μm cell strainers (BD Falcon, 352340) into single-cell suspension. Splenocytes were treated with ACK lysis buffer (Gibco, A1049201) to remove blood cells for cell culture assay. The frozen PBMCs were thawed and allowed to recover overnight. Monocyte and T cell and B cell isolation were performed using the MACS isolation kits following the manufacturer’s protocols (Miltenyi Biotec, Pan B cell isolation kit, human 130−101−638; pan monocyte isolation kit, human 130−096−537; pan T cell isolation kit, human 130−096−535). Flow cytometry and instrumentation For immunophenotyping of tumor-infiltrating leukocytes, single-cell suspensions were prepared by mechanical dissociation of tumors followed by centrifugation on a density gradient. Cells were stained with a live/dead viability dye (Zombie Aqua™ Fixable Viability Kit, BioLegend, 423102) followed by surface antigen staining. For intracellular transcription factor and cytokine staining, the Foxp3/Transcription Factor Staining Buffer Set was used according to the manufacturer’s instructions. Data were acquired on a Fortassa flow cytometer (BD) equipped with 405 nm, 488 nm, 561 nm, and 638 nm lasers. Data analysis was performed using FlowJo software (Version 10.8, BD Life Sciences). Cytokine concentrations (IFN-β, CXCL10) in tumor homogenates were quantified using the LEGENDplex™ Mouse Anti-Virus Response Panel (13-plex, BioLegend, 740622) according to the manufacturer’s protocol, and data were acquired on the CytoFLEX LX and analyzed using LEGENDplex™ Data Analysis Software. Single cell suspensions were incubated with Fixable Viability Stain 620 (BD Horizon, 564996) for 15 minutes at 4°C. The Fc blocking (mouse BD Biosciences, 553142; human BD Biosciences, 564220) treatment was performed for 10 minutes at 4°C prior to fluorescent antibody staining. Primary antibodies were diluted in 1X Brilliant stain buffer (BD Biosciences, 563794) and cells were stained for 30 minutes at 4°C. For intracellular staining, cells were treated with Foxp3/transcription factor staining buffer set (eBioscience, 00-5523-00) following manufacturer’s protocol. Precision Count Beads (Biolegend, 424902) were used to determine the absolute number of each cell subpopulation following the manufacturer’s protocol. Data was acquired on a Cytek Aurora flow cytometer (Cytek Biosciences) using SpectroFlo software (Cytek Biosciences) or LSRFortessa (BD Biosciences) with FACSDiva Software (BD Biosciences) and analyzed with FlowJo v.10 (FlowJo LLC). Immunohistology The tumor tissues were fixed in neutral-buffered 10% formalin solution (Sigma-Aldrich, HT501128), embedded in paraffin(Leica Biosystems, 3801320) and sectioned with Semi-automated rotary microtome (Leica, HistoCore Multicut). Histology staining was performed and graded by Crown Bioscience (Taicang, China) with optimized protocols. Antibodies ADAR1 p150(70257), CD8, ARG1 pan-CK, and PD-L1 from Cell Signaling were used with optimized protocols. All stained sections were scanned with the Pannoramic SCAN Image system (3DHistech) for 40x magnification. To determine the sub-nuclear distribution of Z-DNA, a biotinylated Z22 anti-Z-DNA monoclonal antibody (AbAb, anti-z-dna-z22) was employed. Specifically, Z-DNA localization was assessed in permeabilized, microbead-encapsulated nuclei following the addition of radioactive streptavidin. These results were compared against encapsulated permeabilized nuclei stained with DAPI for nuclear visualization. All the images were analyzed with the HALOTM Image Analysis platform. The intensity of cytoplasm positive staining was scored at four levels, 0 (negative), 1+ (weak staining), 2+ (medium staining), 3+ (strong staining). The percentages of cells at different intensity levels were used to determine H-score. H-Score = (% at 0) ×0 + (% at 1) ×1 + (% at 2) ×2 + (% at 3) ×3 (H-Score range is between 0 to 300). Cytokine release detection Bone marrow cells, tumour cells and splenocytes were seeded at adjusted cell density for co-culture. After 24 hours incubation, supernatant was collected for cytokines release test with R&D Systems Mouse Cytokine Array, Panel A (R&D Systems, catalog ARY006) to simultaneously detect the levels of 40 different cytokines and chemokines, following the manufacturer’s instruction. Signals were detected by chemical luminescence and subsequently quantitated with Image Studio (LI-COR, Version 5.2.5). Generation of ADAR1 mutant cell lines To generate the N173S and Y179A mutant cell lines, we employed CRISPR/Cas9-mediated knock-in technology in forementioned cells. Briefly, a single guide RNA (sgRNA) targeting the Za domain of the ADAR gene (5′- actcgattgatttctttctt-3′]) was designed and cloned into the pSpCas9(BB)-2A-GFP plasmid. A donor repair template ssODN containing the desired point mutation (N173S or Y179A) and a silent PAM-site mutation to prevent re-cutting was co-transfected into the cells. Following transfection, GFP-positive cells were single-cell sorted into 96-well plates using Fluorescence-Activated Cell Sorting (FACS). Potential clones were expanded and screened via genomic DNA PCR and Sanger sequencing to identify homozygous knock-in clones. All constructs were verified by Sanger sequencing to confirm the presence of the intended mutations and the absence of secondary mutations. RNA extraction and library preparation Total RNA was extracted using the RNeasy Mini Kit (Qiagen, Cat# 74104) following the manufacturer’s instructions, including an on-column DNase I treatment (Qiagen) to remove genomic DNA contamination. RNA integrity and concentration were assessed using the Agilent 2100 Bioanalyzer (Agilent Technologies); only samples with an RNA Integrity Number (RIN) > 8.0 were used for library construction.For each sample, 1 mg of total RNA was used as input. Ribosomal RNA (rRNA) was depleted using the Illumina Ribo-Zero Plus rRNA Depletion Kit, and strand-specific libraries were prepared using the Illumina Stranded mRNA Prep Kit. This approach was chosen to ensure high-fidelity detection of antisense transcripts and repetitive elements, such as Alu sequences. High-throughput sequencing and quality control The prepared libraries were sequenced on the Illumina NovaSeq 6000 platform in a paired-end 150 bp (PE150) configuration. We targeted a sequencing depth of at least 50 million raw reads per sample to provide sufficient coverage for repetitive element analysis. Raw sequencing data (FASTQ files) were processed for quality control using FastQC (v0.11.9). Adapters and low-quality bases ($Q < 20$) were trimmed using Trimmomatic (v0.39). Bioinformatic analysis and Alu Editing Index (AEI) Global A-to-I editing levels were quantified using the Alu Editing Index (AEI) as previously described.34 Briefly. clean reads were aligned to the human reference genome (GRCh38/hg38) using the STAR aligner (v2.7.9a) with parameters optimized for multi-mapping reads in repetitive regions. PCR duplicates were marked and removed using Picard Tools. The Alu Editing Index (AEI), representing the global A-to-I editing activity, was calculated using the SAEI tool. Briefly, the index was defined as the weighted average of A-to-G transition frequencies across all known Alu repetitive elements annotated in the UCSC Genome Browser. To minimize noise, we excluded sites known to be common single-nucleotide polymorphisms (SNPs) as identified in dbSNP (v151). RNA-seq reads were aligned to the human genome, and the editing activity was calculated as the ratio of A-to-G mismatches to the total number of covered A positions within Alu repeats. Gene expression profiling Gel Beads-In-Emulsions and cDNA amplification from tumor cell suspension was performed with the 10x Genomics Chromium system (10x Genomics). The cDNA library was cleaned up and validated with Qsep100 (BiOptic) and quantified by Qubit 3.0 Fluorometer (Invitrogen). Transcriptome was analyzed using NovaSeq System (Illumina) according to the manufacturer’s instructions. Data analysis procedure was described in detail in the supplemental information. scRNA-seq data processing The quality of sequencing reads was evaluated using FastQC and MultiQC. Cell Ranger v2.2.0 was used to align the sequencing reads (fastq) to the mm10 mouse transcriptome and quantify the expression of transcripts in each cell. This pipeline resulted in a gene expression matrix for each sample, which records the number of UMIs for each gene associated with each cell barcode. For human data, sequenced reads were aligned to the hg38 human transcriptome, then quantify the expression of transcripts in each cell using BD™ Rhapsody Whole Transcriptome Assay Analysis Pipeline. Unless otherwise stated, all downstream analyses were implemented using R v3.5.2 and the package Seurat v2.3.456. Due to dissimilar data qualities, low-quality cells were filtered using sample-specific cutoffs. The NormalizeData function was performed using default parameters to remove the differences in sequencing depth across cells. scRNA-seq data: QC and defining major cell types Single-cell RNA-seq counts were obtained as the raw expression matrix of unique molecular identifier counts (UMI counts; indicative of the number of unique RNA molecules detected) for each gene in each cell group. UMI counts from each dataset were filtered on the following criteria: cells with >1000 UMI counts, >200 features, <7500 features and <25% mitochondrial gene expression, then normalized and log-transformed using the LogNormalize function in the Seurat package. We integrated all the datasets using the ‘scTransform’ package and ‘harmony’ package for batch correction and scRNA-seq data integration. It was then followed by a combined principal component analysis (PCA), computing 50 principal components, of which the first 20 (based on an elbow plot) were used to compute the Uniform Manifold Approximation and Projection (UMAP) analysis for dimensionality reduction. Next, we clustered cells, exploring a range of different resolution settings of k = 20/100/500 and resolution of 0.3/0.8/1.5. Cell clusters were then annotated following the annotation of the original research. Identification of DEGs To identify the DEGs, we ran a modified version of the Findmarkers by applying the average expression of all the genes across each cell types and setting the logfc.threshold=0.5(corresponding to a fold-change of 1.5) and p_val_adj<0.05 to identify DEGs. P value adjustment was performed using Bonferroni correction based on the total number of genes in the dataset. Gene Ontology analysis was performed by using the R package goTOP (25). Quantification and statistical analysis All statistical analyses were performed using GraphPad Prism software (Version 10.0). Data are presented as the mean ± standard deviation (SD) unless otherwise noted. Before statistical testing, the Shapiro-Wilk test was used to evaluate data normality, and the Brown-Forsythe test was applied to assess equality of variances. For correlations involving clinical datasets (e.g., TCGA), potential outliers were evaluated using Cook’s distance, and robust regression models (Huber and Theil-Sen estimators) were employed to ensure that findings were not disproportionately driven by single extreme values. For comparisons between two groups, an unpaired, two-tailed Student’s t-test was used for normally distributed data, while the Mann-Whitney U test was used for non-parametric data. For comparisons between two groups, an unpaired, two-tailed Student’s t-test was used. For comparisons among more than two groups, one-way or two-way analysis of variance (ANOVA) with appropriate post-hoc tests (Dunnett’s, Tukey’s) was applied as indicated in the figure legends. A p-value of less than 0.05 was considered statistically significant (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001). Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.isci.2026.115978. Contributor Information Xiangyu Yao, Email: xiangyu.yao@roche.com. Zhipeng Yan, Email: zhipeng.yan@roche.com. Supplemental information Document S1. Figures S1–S5 References Associated Data Supplementary Materials Document S1. Figures S1–S5 Data Availability Statement •RNA-seq and scRNA-seq data have been deposited at GEO and are publicly available as of the date of publication, with accession number GSE314521, GSE316054. RNA-seq and scRNA-seq data have been deposited at GEO and are publicly available as of the date of publication, with accession number GSE314521, GSE316054. •All codes used have been stored at DOI: https://zenodo.org/doi/10.5281/zenodo.19756546. All codes used have been stored at DOI: https://zenodo.org/doi/10.5281/zenodo.19756546. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.