BGPT: Paper Review: Loss of ADAR1 in tumours overcomes resistance to immune checkpoint blockade

Explore by Goal

Quick Explanation Copied

Core claim (mechanism-first)

Tumor-intrinsic ADAR1 loss reduces A-to-I editing of IFN-inducible dsRNA species, enabling dsRNA sensing via PKR (growth arrest) and MDA5 (IFN-driven inflammation/immune infiltration), which in turn sensitizes tumors to PD-1 blockade and can override checkpoint resistance even when antigen presentation is disabled (e.g., B2m loss).

Skeptical note: The mechanism is convincingly mapped with epistasis (IFN sensing + PKR/MDA5 dependencies), but translation to humans likely depends on tumor-context constraints (species, tumor models, and how ADAR1 loss is achieved safely).

Long Explanation

Paper Review (visual-first): ADAR1 loss overcomes PD-1 resistance

Target paper:

1) System map (what the paper argues)

The study proposes an ADAR1-controlled dsRNA-sensing checkpoint inside tumor cells. When ADAR1 is lost, IFN-inducible RNA species undergo less A-to-I editing, so they are more readily sensed as dsRNA. This engages PKR (IFN-triggered growth arrest / apoptosis) and MDA5 (driving IFNβ/IFN programs, tumor inflammation, and immune infiltration). Increased inflammation can bypass the need for CD8-mediated recognition of tumor cells and therefore can overcome PD-1 resistance due to antigen presentation defects.

Diagram is a structured visualization of the paper’s stated mechanism (no numeric claims).

2) Evidence ladder (what supports each claim)

Paper claim	Key experimental support described	What’s most uncertain
ADAR1 loss sensitizes tumors to anti–PD-1 therapy across models.	The paper reports ADAR1-null B16 tumors are equivalently viable in vitro but show profound anti–PD-1 sensitization in immunocompetent mice; similar results in additional transplantable models; and size-matched comparisons.	Magnitude/consistency in human tumors (and which ADAR1 isoform/context matters) is not directly resolved here; also, CRISPR off-target effects always warrant caution.
ADAR1 loss inflames the TME and reshapes immune composition.	The paper reports increased CD8+ T cells, increased overall immune infiltration, and decreases in suppressive myeloid populations; scRNA-seq supports myeloid repolarization and increased IFN signatures.	Inflammation can be induced by many stresses; while IFN/dsRNA sensing links are tested mechanistically, causal contributions from non-dsRNA effects of ADAR1 perturbation remain possible.
IFN sensing is required for the enhanced immunotherapy sensitivity.	The paper performs epistasis: double/triple knockouts in IFN signaling components (Ifnar2, Ifngr1, Stat1) abolish the sensitivity phenotype.	The exact cellular source/scope of IFN in vivo (tumor-intrinsic vs host) is partially mapped (e.g., IFNβ secretion assays), but broader network contributions could vary by model.
PKR and MDA5 mediate distinct phenotypes: growth arrest vs inflammation.	CRISPR screens + epistasis identify Eif2ak2 (PKR) for growth arrest after IFNβ; and in vivo dependencies show PKR loss does not eliminate inflammation while MDA5 loss largely abrogates inflammatory infiltration and IFN detection.	Distinct roles are compelling, but the full signaling decomposition (e.g., downstream ISGs vs other cytokines) is likely larger than depicted.
ADAR1 loss overcomes PD-1 resistance due to antigen presentation defects.	The paper generates B2m-null tumors (antigen presentation inactivation), showing resistance to immunotherapy, then shows ADAR1 loss restores sensitivity and eliminates many resistant tumors in vivo.	“Bypassing CD8 recognition” is supported by immune infiltration changes in specific models, but the generality across human MHC-loss contexts needs direct evaluation.

3) Mechanistic plausibility: dsRNA sensing ↔ IFN biology

ADAR1 is widely understood to regulate innate immune responses to endogenous dsRNA by limiting recognition by dsRNA sensors. The paper’s proposed model—loss of ADAR1 enables dsRNA sensing and IFN-program activation—is consistent with prior work.

Separate foundational work also links activation of MDA5 inflammation to recognition of self dsRNA-like ligands and supports the plausibility of a “self-tolerance to dsRNA” mechanism.

Additionally, dsRNA-triggered PKR activation is a known pathway for translational shutdown/growth inhibition, aligning with the paper’s PKR-dependent growth arrest phenotype after IFN exposure.

4) Critical appraisal (skeptical, bias-aware)

Strength: The paper uses multiple orthogonal mechanistic layers—genetic epistasis (IFN receptor/STAT1; dsRNA sensors), rescue/re-expression, and CRISPR screening—to argue causality rather than correlation.
Strength: It explicitly addresses a classic resistance axis (antigen presentation loss via B2m) and tests whether ADAR1 loss can overcome that resistance.
Red flag to monitor: CRISPR off-targets and sgRNA-specific artifacts are always possible, even with rescue attempts; the paper’s rescue suggests on-target effects, but broader off-target assessments are not detailed in the excerpt.
Translation uncertainty: The work is largely in mouse transplantable tumor systems (and uses immunodeficient hosts for specific comparisons). Human tumor microenvironments differ (baseline dsRNA landscapes, interferon thresholds, immune cell repertoires), so the mechanism may be context-dependent.
Human evidence is still limited by availability: The excerpt indicates a TCGA hyperediting correlation analysis, but correlation does not establish that ADAR1 loss causes the immune phenotypes in patients.

5) Data transparency & reproducibility checks

Data availability: Bulk tumor cell RNA-seq and single-cell RNA-seq are deposited in GEO with specific accession numbers (GSE110708; GSE110746).
Methods clarity (bioinformatics): The paper describes preprocessing/alignment/statistics and key pipelines (e.g., Cell Ranger, Seurat, DESeq2, GSEA; and RNA editing alignment with STAR and hyperediting approaches).
Competing interests: The excerpt reports funding from Calico Life Sciences and author links to patents/equity in Tango Therapeutics/consulting. These are not proof of bias, but they warrant heightened skepticism about therapeutic direction claims.

6) Two high-value “go deeper” queries (BGPT links)

Author-specific reviews (BGPT)

Click to open bespoke BGPT author reviews for this paper’s named authors.

Updated: March 28, 2026