Inducible, split base editors for in vivo cancer functional genomics

nature biotechnology Article https://doi.org/10.1038/s41587-026-03077-5 Inducible, split base editors for in vivo cancer functional genomics Received: 27 January 2025 Accepted: 3 March 2026 Published online: xx xx xxxx Check for updates Diqiu Ren1,2,3,15, Shangshang Wang3,4,15, Keisuke Yamada 1,2,3,5,6,7, Yuqiao Liu 1,2,3, Robert Hapke3,4,8, Aktan Alpsoy 9, Yugong Ho10, Canjing Zhang1,2,3, Yemin Lan2, Shuo Zhang2, Joseph P. Milazzo9, Ruchi Lohia 1,2,3, Kiara N. Berríos5, Yongjun Li 11, Evan W. Weber3,8,12,13, Qin Li10, Christopher R. Vakoc 9, Andy J. Minn3,4,13,14 , Rahul M. Kohli 2,5,6 & Junwei Shi 1,2,3,13 Cancer functional genomics using CRISPR base editors (BEs) holds great promise for molecular characterization and new target discovery. However, traditional BEs, using intact DNA deaminases as mutators, are often constrained by limited control and nonspecific toxicities. Here we developed a small-molecule-controllable system using split-engineered BEs (seBEs). By placing deaminase activity under small-molecule control, seBEs significantly reduced cellular toxicity and enabled robust and inducible in vivo functional genomics screens. High-density seBE genetic screens using ~11,000 single guide RNAs in vitro and ~3,700 single guide RNAs in vivo reveal known and previously unknown loss-of-function and dominant-negative mutations in cancer therapeutic targets. A deeper tiling seBE screen against Adar1, a key mediator in cancer immunotherapy, reveals critical residues within functional domains that show no phenotype in vitro but distinctively elicit non-cell-autonomous cancer dependencies in vivo. Overall, our seBE system offers a generalizable, controllable and highly efficient method to systematically identify key residues in cancer functional genomics. Functional genomics in cancer is a powerful approach for identifying the specific roles of genes critical for tumor development, maintenance and progression1–4. These genetic screens link genetic perturbations to the quantitative analysis of cancer phenotypes using deep sequencing, enabling high-throughput dissection of therapeutic target genes. CRISPR functional genetic screens, conducted either in vitro or in vivo, help reveal both cell-autonomous and non-cell-autonomous cancer dependencies 5–12 and have been widely used for target identification at scale11,12. Crucial regions or residues of the target gene are typically then identified with follow-up molecular biology or biochemical approaches13–17; however, such approaches pose technical challenges with in vivo disease models. Recently, CRISPR base editors (BEs) have expanded functional genomic screens by enabling efficient and high-throughput molecular and mechanistic studies. BEs combine programmable catalytically compromised Cas proteins with cytosine or adenine deaminases to enable targeted nucleotide conversion18–23. Department of Cancer Biology, University of Pennsylvania, Philadelphia, PA, USA. 2Epigenetics Institute, University of Pennsylvania, Philadelphia, PA, USA. 3Abramson Family Cancer Research Institute, University of Pennsylvania, Philadelphia, PA, USA. 4Department of Radiation Oncology, University of Pennsylvania, Philadelphia, PA, USA. 5Department of Medicine, University of Pennsylvania, Philadelphia, PA, USA. 6Department of Biochemistry and Biophysics, University of Pennsylvania, Philadelphia, PA, USA. 7Department of Bioengineering, University of Pennsylvania, Philadelphia, PA, USA. 8Center for Cellular and Molecular Therapeutics, Children’s Hospital of Philadelphia, Philadelphia, PA, USA. 9Cold Spring Harbor Laboratory, Cold Spring Harbor, New York, NY, USA. 10Department of Genetics, University of Pennsylvania, Philadelphia, PA, USA. 11Department of Biology, University of Pennsylvania, Philadelphia, PA, USA. 12Department of Pediatrics, University of Pennsylvania, Philadelphia, PA, USA. 13Mark Foundation Center for Immunotherapy, Immune Signaling, and Radiation, University of Pennsylvania, Philadelphia, PA, USA. 14Present address: Immuno-Oncology Program, Memorial Sloan Kettering Cancer Center, New York, NY, USA. 15These authors contributed equally: Diqiu Ren, Shangshang Wang. e-mail: ; ; 1 Nature Biotechnology Article https://doi.org/10.1038/s41587-026-03077-5 We hypothesized that intact DNA deaminases, acting as mutators, contribute to cellular toxicity and that controllable deaminases could overcome these limitations. In prior work, we demonstrated that the DNA deaminases in BEs can be split into two inactive portions that can be conditionally reconstituted with rapamycin (Rap)45. To first investigate BE toxicity, we performed proliferation competition assays with single-copy viral integration to evaluate various BE configurations relevant for functional genetic screening. RN2 cells46, a genetically engineered mouse model of leukemia, were transduced with one of three constructs: a Cas9 nickase (nCas9), traditional BE4max with an intact evoA1 deaminase or seBE derived from BE4max. Each construct was linked to mCherry, allowing longitudinal tracking of the BE+ population via flow cytometry relative to nontransduced cells (Fig. 1a,b). We observed that the nCas9+ population remained stable, whereas the intact BE+ population showed significant and continuous depletion (Fig. 1a), indicating persistent toxicity even under single-copy lentiviral expression. By contrast, the seBE+ population, with or without Rap treatment, remained stable over time (Fig. 1b), indicating that controlling deaminase activity could mitigate proliferation-related cellular toxicity. We confirmed these findings in an independent B16 mouse melanoma model (Supplementary Fig. 1). Cell cycle and cell death analyses further revealed increased apoptosis and reduced S-phase populations in intact BE compared to seBE conditions, consistent with growth-inhibitory toxicity (Fig. 1c and Supplementary Fig. 2). To further assess BE-associated transcriptional toxicity, we performed RNA sequencing (RNA-seq) on transduced cells in the absence of targeting sgRNAs. Principal component analysis revealed close clustering of nCas9, catalytically mutated intact BE and seBE, which were separated from the intact BE groups in the first principal component (Supplementary Fig. 3a). Differential gene expression analysis showed that intact BE perturbed 1,307 genes relative to nCas9 and 951 genes relative to the catalytic mutant at an absolute log2 (fold change) (LFC) of >0.5 and adjusted P value of <0.05 (Fig. 1d). By contrast, seBE caused minimal transcriptional perturbation, altering 271 genes relative to nCas9 and 4 genes relative to the catalytic mutant, and with no detectable changes following Rap addition (Fig. 1d and Supplementary Fig. 3b). Consistent with findings from a prior doxycycline-inducible BE system44, no significant differences in C>U or A<G RNA editing were observed between intact BE and seBE under low-transduction conditions (Supplementary Fig. 4). These findings suggest (...truncated)