Malignant A-to-I RNA editing by ADAR1 drives T-cell acute lymphoblastic leukemia relapse via attenuating dsRNA sensing

Leukemia initiating cells (LICs) are regarded as the origin of leukemia relapse and therapeutic resistance. Identifying direct stemness determinants that fuel LIC self-renewal is critical for developing targeted approaches to eliminate LICs and prevent relapse. Here, we show that the RNA editing enzyme ADAR1 is a crucial stemness factor that promotes LIC self-renewal by attenuating aberrant double-stranded RNA (dsRNA) sensing. Elevated adenosine-to-inosine (A-to-I) editing is a common attribute of relapsed T-ALL regardless of molecular subtypes. Consequently, knockdown of ADAR1 severely inhibits LIC self-renewal capacity and prolongs survival in T-ALL PDX models. Mechanistically, ADAR1 directs hyper-editing of immunogenic dsRNA and retains unedited nuclear dsRNA to avoid detection by the innate immune sensor MDA5. Moreover, we uncovered that the cell intrinsic level of MDA5 dictates the dependency on ADAR1-MDA5 axis in T-ALL. Collectively, our results show that ADAR1 functions as a self-renewal factor that limits the sensing of endogenous dsRNA. Thus, targeting ADAR1 presents a safe and effective therapeutic strategy for eliminating T-ALL LICs.


INTRODUCTION
T-cell acute lymphoblastic leukemia (T-ALL) is an aggressive hematological malignancy that frequently occurs in children, adolescents, and young adults. Approximately 10-20% of T-ALL patients will experience relapse months or years following remission and will often become refractory to further treatments 1,2 . The survival of relapsed/refractory patients is very poor, with an overall survival rate of less than a 25% overall survival rate 3 . Relapsed patients often have enriched pools of leukemia initiating cells (LICs) with enhanced pro-survival and self-renewal capacity, suggesting a potential vulnerable population for effective targeted therapies with less toxicity 4-6 .
An emerging research topic in LIC biology is the identification of RNA modifying enzymes that may cooperate with genetic lesions to provide advantages in important LIC functions 7 . ADAR enzymes catalyze the transition of adenosine (A) to inosine (I) in precursor double-stranded RNA (dsRNA) that are extensively detected in the mammalian transcriptome [8][9][10] . Epitranscriptomic adenosine-to-inosine (A-to-I) RNA editing events are widespread in the cancer transcriptome and are critical for the transition from pre-leukemic cells to fully functional LICs 7,11-13 .
The best documented functional roles of ADAR1 are suppression of the interferon (IFN) response 18,19 and RNA editing of self-dsRNA to prevent abnormal activation of cytosolic self-dsRNA sensing 17,20 . Concurrent deletion of the cytosolic dsRNA sensors melanoma differentiationassociated protein 5 (MDA5) and protein kinase R (PKR) is able to completely rescue embryo death and reverse the IFN signatures [21][22][23] . ADAR1 has two isoforms, the inflammation-induced p150 that is expressed in the cytoplasm and the constitutively expressed nuclear p110, which have diverse cellular functions 15,[24][25][26] . The p150 isoform is thought to be the main regulator of the MDA5 pathway and is the major contributor to LIC generation in myeloid leukemia 15,24,25,27 . In addition, recent reports suggest that the Za-RNA binding region specific to the ADAR1 p150 isoform is responsible for the induction of IFN-stimulated genes in hematopoietic cells 23,28,29 , Compared to myeloid leukemia, the role of ADAR1 in lymphoid progenitor maintenance and malignant transformation is not well understood. In this study, we applied both a threedimensional human thymic organoid system 30 and a T-ALL patient-derived xenograft (PDX) model to examine the function of ADAR1 in the context of T-ALL LIC maintenance. We found that the inflammation-induced ADAR1 p150 isoform is highly expressed within the LIC compartment.
A thorough comparison of the A-to-I RNA editing landscape between non-relapsed and relapsed T-ALL patient cohorts revealed hyper-editing within interferon-stimulated genes (ISG). Moreover, depletion of ADAR1 inhibits LIC self-renewal and survival through both MDA5-dependent andindependent pathways based on the intrinsic expression of MDA5. Our findings indicate that deregulated RNA editing is a critical process in the generation of LICs, which has important implications for T-ALL chemoresistance and therapeutic outcomes.

ADAR1-controlled RNA epitranscriptome in relapsed T-ALL
Here, we investigate if RNA modifications by ADAR1 contribute to T-ALL relapse. The three isoforms within the ADAR family of RNA deaminases (ADAR1, ADAR2, and ADAR3) play different roles depending on the particular cancer type 31 . We analyzed the expression of ADAR family genes in the NCI TARGET T-ALL dataset and discovered that the most abundant RNA editase is ADAR1 ( Figures 1A). In contrast, ADAR2 is expressed at very low levels and ADAR3 expression is below detection, therefore both are unlikely to play any significant roles in T-ALL. By comparing the isoform expression between normal HSPCs and T-ALL samples, the ADAR1 p150 isoform is overexpressed in T-ALL, while the p110 expression remains constant ( Figure 1B). These data were confirmed in three patient samples by intracellular flow cytometry to detect ADAR1 protein expression ( Figure 1C). Interestingly, ADAR1 is expressed predominantly in the immature CD34 + Linpopulation which are enriched for T-ALL LICs, instead of the more differentiated CD34 -Lin + fractions. Together these findings raise the possibility that ADAR1 may play an important role in LIC maintenance 4,32 .
Next, we applied the A-to-I RNA editing bioinformatic pipeline to the TARGET T-ALL datasets by calculating the percentage of guanosine (G) reads at adenosine (A) at known RNA editing sites (inosines are read as guanosines) 12 . The editing events were restricted to those detected at minimum of 10% of samples and >10 reads per site to avoid false-positives. To understand if RNA editing contributes to T-ALL relapse, the overall RNA editing levels were compared between relapsed and non-relapsed patient groups ( Figures 1D-E). A significantly increased incidence of A-to-I RNA modifications is associated with both increased risk of relapse and leukemia-associated mortality. A total of 338 under-edited and 1,472 over-edited sites were found in relapsed patients compared to non-relapsed samples ( Figure 1F and Table S4). Similar to previously reported work 7,33,34 , A-to-I RNA editing occurs predominately in intronic regions, followed by 3'UTR, 5'UTR, and lastly missense or coding regions ( Figure 1G). The increased editing levels were observed in all locations, indicating no selective pressure on location-specific hyper-editing during relapse ( Figure 1H).
Certain molecular subtypes and early T-cell precursor (ETP) status have been associated with more aggressive disease and higher chance of developing relapse 35,36 . We also examined if RNA editing is distinctive among different T-ALL subtypes based on the specific genomic lesions [37][38][39] (Fig. S1A). However, there was very little difference in ADAR1 expression, the overall A-to-I RNA editing levels, total number of editing sites, and editing location among various molecular subtypes. Similarly, no difference in RNA editing level is associated with sex or ETP status ( Fig.   S1B-C). Together, these data indicate that ADAR1 expression and elevated A-to-I RNA editing is a common attribute of relapsed T-ALL regardless of the genetic mutation status, sex, and ETP status.

Reduction of ADAR1 impairs T-ALL LIC survival and self-renewal capacity.
The significant upregulation of ADAR1 and elevated A-to-I RNA editing levels in relapsed T-ALL cohorts suggest a potential role of ADAR1 in disease relapse and maintenance of LIC properties.
LICs exhibit characteristics comparable to those of normal stem cells, such as self-renewal capacity, that enables them to evade chemotherapy and induce relapse [40][41][42] . To examine ADAR1's function in T-ALL LICs, we adapted the ATO system for leukemic cell expansion and established in vivo PDX models with high human leukemic engraftment (Figures 2 and S2A).
Similar to previous reports of co-culture of T-ALL cells with MS5-DLL 43 , the MS5-DLL4 ATO system permits successful T-ALL LIC expansion in vitro ( Figure S2A). Primary T-ALL cells cultured in ATOs display a 2-fold expansion by week 6 and more than 20-fold by week 10 ( Figure   2B-C). In PDX models, abundant human CD45 + leukemic engraftment is usually observed in bone marrow, spleen, and thymic hematological niches within 6-10 weeks after intrahepatic transplant into neonatal Rag2 -/gc -/mice ( Figures 2D-H).
To study the effects of ADAR1 on self-renewal capacity in LICs, ADAR1 was knocked down by shRNA in patient-derived enriched LICs (CD34 + Lin -) followed by culture in ATOs or transplantation into PDX models (Figures 2A-B). ADAR1 reduction led to decreased leukemia cell propagation (>70% reduction) in both systems ( Figures 2C-E and S2B). However, the most striking effects were seen in serial transplant recipients. Equal numbers of bone marrow cells derived from lentiviral control or shADAR1 mice were transplanted to assess the self-renewal capacity of LICs. Spleen and thymus weights of mice injected with shADAR1 cells returned to the same level of non-transplanted litter controls ( Figures 2F-G). In addition, ADAR1 knockdown strongly impedes serial leukemic engraftments in all hematological niches ( Figure 2H). Because of the marked differences in leukemia burden between shADAR1 and control conditions, we evaluated the survival potential between these two groups. We observed significantly improved survival in shADAR1 mice (p<0.0076, Figure 2I). Together, these data suggest that ADAR1 contributes to self-renewal and survival in T-ALL LICs.

Loss of ADAR1 reduces hyper-editing events
Since the TARGET dataset is based on bulk cell sequencing, LIC specific events could be masked.
To gain better insights into LIC-specific molecular targets and pathways regulated by ADAR1, we performed RNA-seq studies on enriched T-ALL LICs (CD34 + Lin -) with ADAR1 knockdown (Figures 3 and S3). Since loss of ADAR1 leads to reduced cell survival (Figure 2), the lentivirus to cell ratio was carefully titrated to obtain approximately 50% reduction of ADAR1 ( Figure S5A).
This allows for sufficient cell recovery after transduction for sequencing analysis. A total of 661 genes are differentially expressed upon ADAR1 knockdown, including 56 downregulated and 605 upregulated genes ( Figure 3A and Table S3). A close examination of the "lymphoblastic leukemia" and the "acute undifferentiation leukemia" pathways revealed several critical self-renewal genes (e.g. CD34, CD44, LMO2, JAK3, and TAL1) were downregulated in ADAR1-deficient LICs (Figures 3B-C and S3B) [44][45][46] . Interestingly, A-to-I RNA editing is rarely detected in these transcripts regardless of direction of differential expression, except for three editing sites within the LMO2 intronic region in scramble control cells (Table S4). Similarly, the most extensively edited genes are often not differentially expressed (e.g. IL17RA and EIF2AK2), suggesting indirect regulation of ADAR1 on gene expression ( Figure 3D). These data indicate that ADAR1 promotes LIC stemness by indirectly targeting self-renewal genes.
We also profiled the RNA editome landscape in FACS-sorted LICs of two T-ALL patients prior to and after ADAR1 knockdown (Figures S3C-F). Reduced ADAR1 led to a small but significant decrease in overall editing levels ( Figure S3C). However, the most profound effect was the reduction in the number of editing events (~ 50%) ( Figure S3D). The total number of edits decreased from 1,698 in the control to 901 in the shADAR1 condition with a predominant drop in Alu-enriched intronic editing sites ( Figure S3E). However, the A-to-I editing level within intronic regions is not altered ( Figure S3F). Since ADAR1 has a tendency to edit in clusters, a phenotype termed hyper-editing 15,47,48 , we calculated the number of edits and changes in editing level between control and shADAR1 per each transcript. Hyper-editing is widespread in intronic (e.g. However, dsRNA sensing of immunostimulatory transcripts as a mechanism in ADAR1-regulated LIC self-renewal has never been fully characterized. To investigate this functionally, we performed concurrent knockdown of MDA5 (mCherrylabeled) and ADAR1 (EGFP-labeled) in PDX T-ALL LICs. The successfully dual-transduced cells (mCherry + EFGP + ) were transplanted into immunocompromised Rag2 -/gc -/mice and then serial transplanted to quantify self-renewal capacity of LICs ( Figure 4A). Surprisingly, co-knockdown of ADAR1 and MDA5 exhibits diverse rescue effects on self-renewal in the three PDX models tested.
A partial rescue of serial leukemia engraftment was detected in co-knockdown compared to the ADAR1 deficient alone condition in PDX-070 ( Figure 4B). In PDX-081, a complete rescue was observed in all hematopoietic niches ( Figure 4C). In contrast, no differences in serial leukemia engraftments or spleen weight were observed in PDX-076 ( Figures 4D-E). These data indicate that ADAR1-directed RNA editing controls LIC self-renewal through dsRNA sensing in at least a portion of T-ALL patients.
Next, we explored potential mechanisms guiding the difference in response to coknockdown of MDA5 and ADAR1. Curiously, differential gene expressions in the Rig-I-Like signaling and cytosolic sensing pathways were detected between the PDX-070 and -076 samples, which could suggest differences in intrinsic signaling properties and dependency on dsRNAsensing pathways among patients ( Figure S4). Moreover, the ADAR1 p150 isoform is thought of as the main regulator of RNA editing in the cytoplasm and therefore is responsible for regulating dsRNA sensing by MDA5, while ADAR1 p110 is largely dispensable for MDA5 signaling 27 . Thus, the intrinsic expression of p150 and MDA5 dsRNA sensor in T-ALL patients might dictate the level of dependency on the MDA5 pathway. To test this hypothesis, we measured the expression of ADAR1 isoforms and MDA5 in FACS-enriched LICs in the three T-ALL samples ( Figure 5). Patient 070 did not yield enough LICs therefore we could not complete the analysis. Patient 081 has significantly elevated expression of ADAR1 p150, p110, as well as MDA5, compared to patient 076 ( Figures 5A-C). The level of another dsRNA sensor PKR was also determined and showed no difference between patients. Coupled with the differential rescue effects of MDA5 and ADAR1 co-knockdown ( Figure 4C), our data support that patient 081 relies on the ADAR1 p150-MDA5 axis for promoting self-renewal, while patient 076 likely depends on both p150-MDA5 axis and p110 dependent mechanisms.
We next sought to validate whether this isoform-specific dependency of ADAR1 is also presented in T-ALL cell lines. We first evaluated the endogenous expression of the p150 and p110 isoforms, and dsRNA sensors MDA5 and PKR, in CUTTL1, SUP-T1, and Jurkat cells. SUP-T1 has the highest expression of p150 isoform, followed by CUTTL1 and then Jurkat cells ( Figures   5D-E). The protein expression of p110 is comparable among the three cell lines as shown in western blot analysis ( Figure 5E). MDA5 is expressed at the highest level in Jurkat and lowly expressed in CUTTL1 and SUP-T1 ( Figure 5F-G). Consistent with T-ALL patient LICs, the expression of PKR does not vary significantly among the cell lines ( Figure 5F). Next, we performed ADAR1 knockdown alone, MDA5 knockdown alone, or MDA5 and ADAR1 co- ADAR1 RNA editing-independent activity promotes nuclear localization of dsRNA ADAR1 can operate as a dsRNA binding protein with functions independent of editing activity to promote cancer progression 15,34,53,54 . To explore RNA editing dependent and independent mechanisms of ADAR1, we first generated an ADAR1 knockout Jurkat T-ALL cell line and a wildtype "add-back" cell line by re-expressing wildtype ADAR1 p150 isoform (ADAR1 WT ) in knockout cells with a lentiviral overexpression vector ( Figure 6A). The p150 mRNA expressed both p150 and p110 isoforms due to leaky ribosome scanning 26 . To activate the IFN response, these cells were treated with various doses of IFNa, b, and g for 48 hours to examine changes in cell viability and ADAR1 expression ( Figures 6B and S6A). IFN treatment upregulates ADAR1 in wildtype and ADAR1 WT add-back cells, while knockout cells had little response. As expected, exposure to IFNb predominately upregulates the p150 isoform rather than p110 ( Figure 6C).

Moreover, loss of ADAR1 induces apoptosis upon IFNb treatment as demonstrated by elevated levels of cleaved PARP1 and a reduced percentage of viable cells (Figures 6D-E).
In addition to MDA5-MAVS signaling, ADAR1 also suppresses cytoplasmic dsRNA sensing through RIG-I (retinoic acid-inducible gene I) and PKR pathways 17,23,55,56 . While ADAR1 KO or add-back do not change MDA5, PKR, and RIG-I levels, IFNb stimulated the expression of these dsRNA sensors in wildtype, ADAR1 KO, and ADAR1 WT cells, ( Figure 6E). This was confirmed by elevated PKR expression upon addition of IFNa and IFNg ( Figure S6B). Interestingly, knockdown of MDA5 in ADAR1 knockout cells abrogated the IFNb-induced RIG-I activation, but increased PKR activation, suggesting MDA5 may crosstalk with other dsRNA sensors in the combination of ADAR1 loss and IFNb treatment ( Figure 6E).
Next, we introduced a catalytically inactive mutant of the ADAR1 p150 isoform (ADAR1 E912A ) 15,18,50 in ADAR1 KO cells ( Figure 6F). The lentiviral construct produces both p150 and p110 isoforms as previously reported 26 . The expression of a selected panel of 790 ISGs using the Nanostring nCounter System was quantified in IFNb stimulated wildtype, ADAR1 KO, and ADAR1 E912A cells. The wildtype cells are relatively insensitive to IFNb treatment compared to ADAR1 KO cells. A total of 27 differentially expressed ISGs in ADAR1 E912A cells in comparison to 237 ISGs in ADAR1 knockout and 30 ISGs in wildtype cells was found (Table S5). Surprisingly, ADAR1 E912A overexpression was able to rescue all downregulated targets found in the IFNb treated KO condition as well as suppress the majority of upregulated ISGs. Approximately 70% (19 out of 27) of ADAR1 E912A regulated targets overlapped with those of ADAR1 knockout cells, while only 3.7% (1 out of 27) of targets overlapped with IFN-treated wildtype cells. Together with the cell lethality and functional rescue by MDA5 depletion, these data suggest that ADAR1's editing-independent function also contributes to suppress MDA5 signaling in T-ALL.
The activation of dsRNA sensing pathways depends on cellular localization and length of the accessible endogenous dsRNA 57 . The inosine containing dsRNA can be retained in the nucleus by p54 nrb , PSF, and MATR3, thus avoiding export to the cytoplasm and detection by MDA5 58 . Therefore, we hypothesized that ADAR1 E912A may retain dsRNA in the nucleus to limit the cytosolic dsRNA pool. Immunofluorescent staining using a J2 dsRNA antibody was applied to identify the cellular dsRNA localization prior to and after IFNb treatment ( Figures 6G-H). The IFNb treatment length was reduced to 24 hours to detect dsRNA localization prior to cell death. Wildtype cells respond to the addition of IFNb by increasing total dsRNA level (2.4 folds) and the percentage of nuclear dsRNA (average 56%). The same nuclear retention phenotype was also observed in ADAR1 E912A cells (average 57%). Interestingly, in the absence of IFNb, both ADAR1 knockout and ADAR1 E912A cells showed elevated total dsRNA levels compared to wildtype cells without triggering cell death, suggesting a certain level of unedited dsRNA is tolerated. However, the most striking effect was seen in IFNb treated knockout cells. We detected an abundant quantity of dsRNA accumulated in the cytoplasm. However, the level of nuclear dsRNA remained constant prior to and after IFNb treatment. To confirm this finding, we exposed CUTTL1 cell line to IFNb for 24 hours and analyzed dsRNA location ( Figure SC-D). While the total dsRNA level did not change, the nuclear dsRNA showed a significant increase from 35% to 62%, indicating this retention mechanism is commonly used by T-ALL cells in response to IFN stress. Together, these data reveal an important RNA editing-independent mechanism of ADAR1 in preventing MDA5directed dsRNA sensing.

DISCUSSION
T-ALL is an aggressive hematological malignancy which arises from transformation of lymphoid progenitors with the cooperation of tumor suppressors and oncogenes [59][60][61] . We now understand that RNA modifications such as A-to-I RNA editing and m6A are critical in promoting cancer progression and therapeutic resistance 7,62 . We have shown RNA editing by ADAR1 is an important regulatory mechanism required for HSC maintenance and transformation into myeloid leukemia 15,20,24,34,63 . However, the role of ADAR1 in lymphoid neoplasms such as T-ALL has never been explored. Here, we described a fundamental role of ADAR1 in maintenance of T-ALL LICs.
Increased RNA editing and ADAR1 expression was detected in relapsed T-ALL patients, regardless of the molecular genetic mutations. ADAR1 directs A-to-I hyper-editing of ISG dsRNA and retains unedited nuclear dsRNA to avoid detection by the innate immune sensor MDA5.
Consequently, loss of ADAR1 leads to impaired survival and self-renewal and improves overall survival in PDX models. Finally, we demonstrated that ADAR1 regulates LIC self-renewal through aberrant dsRNA sensing in a portion of T-ALL samples depending on the expression level of the dsRNA sensor MDA5.
We have previously discovered ADAR1's contribution to neoplastic transformation of myeloid LICs via several different pathways: 1) regulation of self-renewing microRNA biogenesis, 2) editing of 3'UTR of oncogenes to prevent miRNA-directed degradation, 3) editing of coding genes, and 4) induction of oncogenic RNA splicing events 15,24,25,64 . Whether editing of immunogenic dsRNA and suppression of aberrant dsRNA sensing could enhance LIC selfrenewal capacities is an important question that has not been extensively addressed. We provide the link between malignant A-to-I RNA editing and suppression of dsRNA sensing as a mechanism in promoting LIC self-renewal. Hyper-editing events are commonly observed in ISG genes within intronic and 3'UTR regions. We identified unexpected differences among three T-ALL models in response to concurrent knockdown of MAD5 and ADAR1. This novel phenomenon could stem from the intrinsic tolerance of unedited dsRNA or variable levels of dsRNA sensors among patients. We further reported that the intrinsic expression of MDA5, rather than p150 or p110 isoforms, dictates the level of response to co-knockdown. Since the ADAR1 p150 isoform specifically regulates the MDA5-MAVS pathway 27 , our study suggests that LIC self-renewal and survival rely entirely on the p150-MDA5 axis and the p110 isoform is likely dispensable in such scenarios. In contrast, other T-ALL models may depend on both p150 and p110 isoforms in a MDA5-independent manner. It is possible that LICs possess different levels of dependency on ADAR1 due to the diverse ISG signatures in T-ALL patients as previously reported 50 . Future studies are necessary to definitively decouple the isoform-specific function, RNA editing targets, and pathways regulated by a particular isoform in a large cohort of patients and potentially other tumor types.
Lastly, we report an RNA editing-independent role of ADAR1 in attenuating aberrant dsRNA sensing. To trigger dsRNA sensing by cytosolic MDA5, unedited endogenous dsRNA in the absence of ADAR1 must be in the cytoplasm. The deaminase deficient ADAR1 E912A was able to suppress the majority of ISGs in ADAR1 knockout cells, in part through nuclear dsRNA retention to limit the cytosolic dsRNA pool. However, we cannot rule out the possibility that ADAR1 E912A competes with MDA5 to prevent efficient dsRNA sensing. Surprisingly, we noticed that T-ALL cells can tolerate a certain level of unedited cytosolic dsRNA without triggering apoptosis. It is curious as to how cancer cells set this limit using the complex and diverse regulatory dsRNA sensing network, which may contribute to the difference in response to cytosolic dsRNA in T-ALL models. In conclusion, this work highlights the intrinsic differences in how ADAR1 regulates malignant dsRNA sensing and promotes self-renewal among patient samples, in addition to mechanistic details of LIC maintenance. This in turn opens the door for therapeutic targeting of these downstream processes to prevent relapse and therapeutic resistance.

CONTACT for REAGENT AND RESOURCE SHARING
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Qingfei Jiang (q1jiang@health.ucsd.edu).

Animal Experiments
All mouse studies were conducted under protocols approved by the Institutional Animal Care and  Table S2.

Patient sample preparation.
Peripheral blood mononuclear cells (PBMC) were extracted by Ficoll density centrifugation.  (Table S1) and analyzed on a BD Aria Fusion and with FlowJo.
Cells were harvested in staining media, counted, and equal numbers of cells per condition were transplanted into recipient mice (5x10 4 -1x10 5 per pup). Transplanted mice were FACS screened for human engraftment in peripheral blood at 6-10 weeks. Once human engraftment was confirmed (>1% human CD45 + cells in peripheral blood), mice were euthanized, and single cell suspensions of hematopoietic tissues (bone marrow, spleen, and thymus) were analyzed by FACS and FlowJo.

Flow cytometry analysis and sorting
Flow staining was performed in staining media for 30 min on ice in the dark. Cells were blocked  Table S6.

Interferon stimulation assay
Cells were seeded at a density of 10 5 in a 12-well plate and treated with a single dose of IFNa, IFNb, or IFNg (R&D Systems) at 0.05-500 ng/mL. After 48 hours, cells were harvested and analyzed by western blot and RT-qPCR. The cell supernatants were collected and the S100A9 level was determined by an ELISA assay (Invitrogen). Cell viability was determined trypin blue assay.

Nanostring nCounter
Jurkat cells were collected after 24 hours of IFNb-stimulation and RNA was isolated (RNeasy Plus mini kit, Qiagen). The mRNA levels were directly measured using the Human CAR-T characterization panel kit with additional custom probes (Table S5) from NanoString nCounter gene expression system (NanoString). The differential expression analyses of mRNA were performed using nSolver software (NanoString) and visualized in Prism software.

Immunofluorescence staining
Cells (1-2x10 3 ) were harvested in ice-cold PBS and loaded on adhesion slides (Marienfeld Superior) by incubating for 10 min at room temperature. The slides were transferred into a coplin jar containing ice-cold PBS for 5 min and fixed with 4% paraformaldehyde in PBS for 10 min at room temperature. Immunofluorescence was performed by immersing slides in PBST (1x PBS with 0.05% Tween-20). Slides were overlaid with blocking solution (2% fetal bovine serum in PBST) for 1 hour at room temperature. After washing, primary antibody was added to the slides and incubated overnight at 4C. Secondary antibody was overlaid to spotted cells for 1 hour in the dark. DAPI was added and the slides were sealed with a coverslip. Imaging was performed using a Keyence confocal microscope. The intensity and numbers of dsRNA foci were caculated using ImageJ software.

Whole RNA-sequencing
Samples with RNA integrity numbers (RIN) ≥7 will be processed using SMART cDNA synthesis

Other Statistical Analysis and Reproducibility
All experiments were performed with at least two biological or experimental replicates, with specific number of replicates stated in the figure legends. Unless otherwise stated, the statistical analyses were performed using GraphPad Prism (v7.0) and statistical significance were determined at p value < 0.05, with specific statistical test stated in the figure legends.

DATA AND SOFTWARE AVAILABILITY
The RNA-sequencing dataset used in this study will be upload to the GEO with a special password for editors and reviewers. The data will be made publicly available upon acceptance for publication. Table S1. Cell surface markers used for FACS sorting, related to STAR Methods. Table S2. Patient characteristics, related to STAR Methods. Table S3. Gene lists of ADAR1-regulated gene signatures, related to Figure 2 -6. Table S4. RNA editome in T-ALL LICs, related to Figure 6. Table S5. Differentially expressed ISGs in IFNb-treated Jurkat wildtype, ADAR1 knockout and ADAR1 E912A cells, related to Figure 8. Table S6. Primers for RT-qPCR used in this study, related to STAR Methods. Figure 1 Relapsed T-ALL acquires a distinct RNA editome in contrast to non-relapsed T-ALL. A. Expression of ADAR1 and ADAR2 in T-ALL patient by RNA-seq (n = 256). B. Isoform expression of ADAR1 p150 and p110 between HSPC (n=3) and T-ALL (n= 256). C. Quanti cation of ADAR1 expression in HSPCs     Concurrent knockdown of MDA5 and ADAR1 rescues self-renewal to various degrees in different T-ALL models. A. Experimental setup. T-ALL CD34 + cells were transduced with shCTRL, shADAR1, or shADAR1 and shMDA5 lentivirus in combination. Transduced cells were sorted based on GFP + mCherry + (GFP for shADAR1, and mCherry for shMDA5) and serial transplant potential was measured in recipient Rag2 -/gc -/mice. B-D. Serial leukemia engraftment and representative bone marrow FACS plot of patient 070 (B), patient 080 (C), and patient 076 (D) was determined for shCTRL, shADAR1, and shADAR1 in combination with shMDA5 (3-8 mice/condition). E. Images of spleen (left) and spleen weights (right) in serial transplanted patient 076 were determined after an 8-week engraftment interval. *p<0.05, **p<0.01, ***p<0.001, unpaired Student t-test.