Systematic discovery of CRISPR-boosted CAR T cell immunotherapies

Abstract

Chimeric antigen receptor (CAR) T cell therapy has shown remarkable success in treating blood cancers, but CAR T cell dysfunction remains a common cause of treatment failure¹. Here we present CELLFIE, a CRISPR screening platform for enhancing CAR T cells across multiple clinical objectives. We performed genome-wide screens in human primary CAR T cells, with readouts capturing key aspects of T cell biology, including proliferation, target cell recognition, activation, apoptosis and fratricide, and exhaustion. Screening hits were prioritized using a new in vivo CROP-seq² method in a xenograft model of human leukaemia, establishing several gene knockouts that boost CAR T cell efficacy. Most notably, we discovered that RHOG knockout is a potent and unexpected CAR T cell enhancer, both individually and together with FAS knockout, which was validated across multiple in vivo models, CAR designs and sample donors, and in patient-derived cells. Demonstrating the versatility of the CELLFIE platform, we also conducted combinatorial CRISPR screens to identify synergistic gene pairs and saturation base-editing screens to characterize RHOG variants. In summary, we discovered, validated and biologically characterized CRISPR-boosted CAR T cells that outperform standard CAR T cells in widely used benchmarks, establishing a foundational resource for optimizing cell-based immunotherapies.

Fig. 1. Genome-wide fitness screens in human primary CAR T cells with the CELLFIE platform.

a, Overview of the CELLFIE platform for highly scalable CAR T cell engineering and CRISPR screening. Human primary T cells are isolated, activated and expanded, transduced with a lentivirus carrying sequences for the CAR and gRNA library and electroporated with synthetic mRNA delivering the CRISPR editor. The CRISPR-edited CAR T cells are functionally screened in vitro and in vivo using multiple readouts. hU6, human U6 promoter; LTR, long terminal repeat; puro, puromycin. b, Experimental timeline for genome-wide fitness screens. Human primary T cells are isolated from whole blood, activated, pre-expanded, activated again and transduced with the CROP-seq-CAR lentivirus for co-delivery of sequences for the CAR and the genome-wide gRNA library. Two days later, cells are electroporated with synthetic mRNA for Cas9 and blasticidin resistance (blasticidin-S deaminase, BSD), followed by antibiotic selection for successful lentiviral transduction (puromycin) and successful mRNA electroporation (blasticidin). CRISPR-edited CAR T cells are expanded under repeated TCR stimulation with anti-CD3/CD28 beads or repeated CAR stimulation with CD19⁺ K562 cells. Genomic DNA is collected, and gRNA representation is analysed by sequencing on days 0, 7, 14 and 21. c, Gene-level log₂[fold change (FC)] between day 14 and day 0 for the fitness screens (four donors in two independent screens), mapped onto a schematic of the TCR and CAR signalling pathways. For each protein shape, the colour in the top half corresponds to the TCR stimulation screens and the colour in the bottom half corresponds to the CAR stimulation screens. d, Effect sizes for fitness screens with TCR stimulation (x axis) and CAR stimulation (y axis). MAGeCK MLE β values comparing day 14 and day 0 are normalized to values for essential genes to account for the different proliferation rates upon TCR or CAR stimulation. Colours denote screening hits with increased fitness (green), known negative T cell regulators (magenta), known essential genes (purple) and neutral olfactory receptors (blue).

Fig. 2. Genome-wide FACS-based screens for clinical limitations of CAR T cells.

a, Overview of CAR T cell surface proteins used as screening readouts. Antigen recognition by the CAR can result in CD19 transfer from target cells to the CAR T cell surface (trogocytosis), serving as a marker of target cell recognition. CD69 expression is a marker of CAR T cell activation, FAS expression mediates apoptosis, and the combined expression of PD-1, LAG3 and TIM3 serves as an early indicator of T cell exhaustion. b, Validation analysis showing strong enrichment of FACS marker gene knockouts in marker-negative cells. Significantly enriched genes (FDR < 0.01 and log₂[FC] > 1.5; thresholds indicated by dashed lines) are coloured, and FACS marker genes are labelled. c, Identification of knockouts that improve CAR T cell function by FACS-based screening. Scatterplots show gene-level log₂[FC] between sorted and unsorted cells (x axis) plotted against FDR-adjusted P values (y axis). Genes that passed stringent significance thresholds (FDR < 0.01 and log₂[FC] > 1.5; indicated by dashed lines) are highlighted. d, Summary of screening hits (top hits across all screens, n = 43) that improved CAR T cell properties, identified in genome-wide fitness screens (blue) and FACS-based screens (green). The bubble plot shows z-normalized log₂[FC] for significant gene knockouts. Validated immunotherapy target genes are highlighted in magenta.

Fig. 3. In vivo CROP-seq screening of putative CAR T cell boosters in leukaemic mice.

a, Experimental timeline for pooled in vivo screening. Immunodeficient NSG mice are injected with human NALM6 cells to induce B cell leukaemia and then treated with a deliberately noncurative dose of CAR T cells with CRISPR knockout for the in vitro screening hits. Spleen and bone marrow are collected on day 9 and 21, and gRNA representation is compared with that of the pre-injection sample. Luc, firefly luciferase. b, Vector design and technical outline of the in vivo CROP-seq method. The gRNA cassette, including the UMI and the Illumina read 2 sequence, is located in the 3′ LTR of the CROP-seq-CAR vector. The gRNA and UMI are detected from the polymerase II transcript encoding CAR-P2A-Puro using reverse transcription and nested PCR. The gRNA is sequenced with read 1, and the UMI is sequenced with read 2 by paired-end sequencing. c, gRNA detection performance using conventional gRNA amplification from genomic DNA or gRNA amplification from mRNA/cDNA (in vivo CROP-seq), both tested with single-PCR and nested PCR amplification. R2, read 2; WPRE, woodchuck hepatitis virus post-transcriptional regulatory element. d, In vivo CROP-seq data analysis with standard methods (left) or using UMIs to group reads into 16 internal replicates (right) for bone marrow (top) and spleen (bottom). RHOG, FAS, PRDM1 and CDKN2A were identified as significant (FDR < 0.01 and log₂[FC] > 1.5). The separation between depletion (left) and enrichment (right) is denoted by a dashed line at log₂[FC] = 0. e, Knee plot showing rank versus read number for each UMI (clones versus background). The inflection point separates distinct CAR T cell clones (coloured) from background noise (grey). f, Number of distinct CAR T cell clones detected for each organ, donor and time point. g, CAR T cells for the three top-ranked gene knockouts (FAS, PRDM1 and RHOG) dominate the CAR T cell response, collectively accounting for over 25% of gRNA reads on day 21 after injection.

Fig. 4. RHOG and FAS knockout CAR T cells enhance leukaemic cell control and survival in vivo.

a, Experimental timeline of in vivo validation experiments (leukaemia model). Immunodeficient NSG mice are injected with human NALM6 cells to induce B cell leukaemia and treated with a deliberately noncurative dose of standard (safe harbour locus-edited) CAR T cells or with knockout CAR T cells for RHOG, FAS or both. b, Leukaemic cell load over time measured by bioluminescence imaging (two donors) comparing standard CAR T cells (ten mice) with the RHOG knockout (ten mice), FAS knockout (nine mice) and RHOG-and-FAS double-knockout (nine mice) CAR T cells. IVIS, in vivo imaging system; Exp, experiment. c, Leukaemia-reducing effect of knockout CAR T cells compared with standard CAR T cells for the mice shown in b (tumour reduction over time in vivo). Colored lines were obtained by local regression (LOESS); shaded areas represent 95% confidence intervals. d, Survival analysis for the mice shown in b. e, Survival analysis for all mice treated with 0.6 million RHOG-knockout (42 mice) or standard (41 mice) CAR T cells from all eight donors; the P value is from the log-rank test comparing RHOG-knockout and standard CAR T cell treatment); shaded areas represent 95% confidence intervals. f, Percentage of central memory cells (CD45RO⁺CD62L⁺) among RHOG-knockout or standard CAR T cells after repeated in vitro CAR stimulation with K562-CD19 target cells (three donors). g, Fold increase in CD4⁺ or CD8⁺ CAR T cell numbers in mice treated with RHOG-knockout (KO) or standard CAR T cells, measured on day 15 after CAR T cell injection (four donors). The box plot’s centre line is the median, the box limits are the upper and lower quartiles, and the whiskers extend to 1.5 times the interquartile range. h, Percentage of RHOG-knockout CAR T cells positive for the T cell exhaustion markers PD-1, LAG3, TIM3 and TIGIT on day 15 after CAR T cell injection (three donors). NS, not significant. i, Conceptual summary of the biological effects underlying the enhanced efficacy of RHOG-knockout CAR T cells. P values in f–h are from paired one-tailed t-tests. Source Data

Fig. 5. Combinatorial knockout and base-editing screening with the CELLFIE platform.

a, Experimental timeline of combinatorial fitness screens (four donors) using CROP-seq-CAR-multi vectors for three different CARs (19-BBz, 19-28z, GD2–BBz). The screening library comprises 238 combinations of gRNAs targeting six top hits from the in vivo pooled screening and control gRNAs targeting essential genes and a safe harbour locus. b, Vector design and technical outline of combinatorial screening with the CROP-seq-CAR-multi construct with three different CARs. The vector adapts the CROPseq-multi technology with a dual-gRNA cassette separated by a tRNA sequence that is cleaved by cellular enzymes. CMV, cytomegalovirus promoter. c, Ranking of fitness effects (MAGeCK MLE β value comparing day 12 versus day 0 gene interactions) for pairwise gene knockouts. d, Overview of saturation base-editing screening. The library includes all possible gRNAs tiling the gene bodies of RHOG and of the puromycin resistance gene PAC (for validation, encoding puromycin N-acetyltransferase) and gRNAs targeting essential genes (positive controls) and a safe harbour locus (negative controls). e, Number and type of base-editing mutations introduced by the gRNA library with each of the four CRISPR base editors (A to G, C to T, and their near-PAM-less variants), based on computational predictions (n = 3,755). UTR, untranslated region. f, Mutagenesis map for RHOG based on base-editing tiling screens (two donors), comparing gRNA distribution (introducing missense or nonsense mutations) between day 12 and day 0 after CAR restimulation. The plot shows z-normalized log₂[FC] (y axis). The top 15 most enriched gRNAs are labelled with their predicted amino acid changes. g, Amino acids with the strongest mutagenesis effects in the screen, mapped onto the RHOG protein structure (PDB 6UKA). The visualization is a magnified view of the GTP-binding site, with the full protein shown in the inset.

Extended Data Fig. 1. CELLFIE: a versatile platform for human CAR T cell engineering.

a, Design of the mRNA production plasmid with the open reading frame (ORF) of the CRISPR editor or selectable marker between untranslated regions (UTRs) to stabilize the mRNA. For mRNA production, the PCR template is amplified (top) and used for in vitro transcription (IVT) from the T7 promoter with co-transcriptional capping and transcription termination at a hardcoded A-tail of 55 bases. b, Representative size profiles of PCR templates and in vitro transcribed mRNA for CRISPR editors and selection markers. The leftmost sample is commercial Cas9 mRNA (Trilink), all other samples were produced with CELLFIE. c, Validated synthetic mRNAs for delivery of CRISPR editors and selection markers into human primary T cells and CAR T cells. d, Efficient knockout of IL2RA (CD25) at the DNA and protein level in CD4⁺ T cells (4 donors). e, Cost of custom-made synthetic Cas9 mRNA compared to commercial reagents. Left y-axis: mRNA cost for small-scale T cell editing. Right y-axis: Total cost for all mRNA produced for this study, enabling the delivery of CRISPR editors into 66 billion human primary CAR T cells, including assay development and optimization experiments. f, CRISPR base editing efficiency (top: A-to-G, bottom: C-to-T) in CD4⁺ T cells using standard and near-PAM-less base editors for multiple genomic loci (2 to 4 donors). g, CRISPR base editing specificity within the editing window for two genomic loci. h, CRISPR activation of CD34 using mRNA-based delivery of the SAM system into CD4⁺ T cells (4 donors).

Extended Data Fig. 2. Development and optimization of the CELLFIE platform.

a, Titration of synthetic mRNA concentration for efficient CRISPR knockout of CD44 in human primary CD4⁺ and CD8⁺ T cells (2 donors), with pan-CD3⁺ T cells as starting material. Comparable editing efficiencies were observed for custom-made and commercial Cas9 mRNA. b, Consistent editing efficiencies in CD4⁺ T cells for custom-made Cas9 mRNA from multiple production rounds. c, Cell proliferation of CD4⁺ and CD8⁺ CAR T cells made from pan-CD3⁺ T cells (green), from isolated CD4⁺ T cells (blue), or from isolated CD8⁺ T cells (purple) as starting material. d, Declining representation of CD4⁺ CAR T cells in co-culture with CD8⁺ CAR T cells when using pan-CD3⁺ T cells as starting material. e, Puromycin titration to determine an optimal concentration (0.5 µg ml⁻¹, grey dotted line) to select for T cells that have been successfully transduced with the CROP-seq-CAR lentivirus. f, Blasticidin titration to determine an optimal concentration (50 µg ml⁻¹, grey dotted line) to select for T cells that were successfully electroporated with blasticidin resistance mRNA co-delivered with the CRISPR editor mRNA. g, CD44 knockout efficiency using the optimized electroporation programs for small-scale experiments (up to 1.5 million cells per cuvette, Amaxa electroporator) and large screens (up to 100 million cells per cuvette, MaxCyte electroporator). Results are shown for CD4⁺ and CD8⁺ T cells from pan-CD3⁺ T cells (1 donor). h, T cell expansion after electroporation of synthetic Cas9 mRNA and blasticidin resistance mRNA (1 donor). i, Effect of T cell stimulation reagents on lentiviral transduction rates (4 donors). j, Effect of common transduction supplements on lentiviral transduction rates (4 donors). k, Effect of common transduction supplements on T cell viability (4 donors). l, Quantification of lentivirus titers using RT-qPCR of lentiviral RNA (mean ± s.e.m. for 3 technical replicates). m, Percent transduced CD4⁺ and CD8⁺ T cells for a titration of lentivirus amounts. The chosen amounts (CD4⁺: 247 copies per cell; CD8⁺: 432 copies per cell) are indicated by dotted lines (2 donors). n, Detection of gRNAs (y-axis: gRNA read counts) in clonally expanded human primary T cells transduced with CROP-seq-CAR lentivirus carrying the genome-wide Brunello gRNA library. Representative examples of T cell clones with 1, 2, or 3 gRNA integrations are shown, and a total of 62 clonally expanded T cell clones were profiled. o, Barplot showing the frequency of 1, 2, or 3 independent lentiviral integrations into the same cell across 62 clonally expanded T cell clones. The average number of gRNA integrations per cell was 1.5. p, CAR expression in human primary CAR T cells prepared with the CROP-seq-CAR lentivirus, using PE-labelled recombinant CD19 antigen for labeling. q, Specific killing of CD19⁺ cancer cells by CAR T cells prepared with the CROP-seq-CAR (anti-CD19) lentivirus. For all boxplots (panels i-l), the center line is the median, the box limits are the upper and lower quartiles, and the whiskers extend to 1.5 times the interquartile range.

Extended Data Fig. 3. Optimization of genome-wide fitness screening in human primary CAR T cells.

a, Proof-of-concept fitness screen with a focused gRNA library of 100 gRNAs. gRNA-level log₂ fold changes are shown for positive control gRNAs (targeting the puromycin resistance gene PAC) and control gRNAs (targeting a safe harbor locus) at days 7, 14, or 21 relative to day 0. Cells were electroporated with custom-made (top) or commercial (bottom) Cas9 mRNA. b, gRNA representation after cloning the genome-wide Brunello library into the CROP-seq-CAR vector, plotted as a cumulative distribution function based on amplicon sequencing of the plasmid pool, with highlighted fold difference between the 10^th and 90^th percentiles as a measure of gRNA library balance. c, Detailed experimental timeline for the genome-wide fitness screens (4 donors, 2 independent experiments). Key steps in the CELLFIE workflow are highlighted. Samples for gRNA sequencing were collected at day 0 (before Cas9 electroporation), and at days 7, 14, and 21 after electroporation. d, Expansion of human primary CAR T cells upon repeated CAR or TCR stimulation (2 donors). e, CD19 accumulation on the surface of CAR T cells during co-culture with target cells as the result of trogocytosis. f, Flow cytometry profiling of the T cell exhaustion markers PD1, LAG3, TIM3, and TIGIT during co-culture of CAR T cells with K562-CD19 target cells. g, T cell subset profiling by flow cytometry at the isolation and readout time points of the genome-wide screens. h, Essential genes in human primary CAR T cells under CAR and TCR stimulation (4 donors, details in Supplementary Table 2). The scatterplot shows gene-level log₂ fold changes comparing day 14 and day 0 of the screen (x-axis) plotted against FDR-adjusted p-values (y-axis). Genes that passed stringent significance thresholds (FDR < 0.01 and log₂ FC < − 1.5 based on MAGeCK RRA) are highlighted. i, Gene set enrichment analysis for essential genes in human primary CAR T cells under CAR and TCR stimulation. Clustering of the top-100 most enriched Gene Ontology (GO) terms from the Biological Process category (left) is shown together with one cluster related to T cell functions visualized as a tree plot (right).

Extended Data Fig. 4. Comparison of data quality and gene hits between CELLFIE and published T cell or CAR T cell screening datasets.

a, Published CRISPR screens in human T cells or CAR T cells included in this comparison. b, Screening timelines for the studies included in this comparison. c, Dropout of essential genes across studies based on log₂ fold changes for known essential genes (n = 684, based on BAGEL2 predictions) relative to non-essential genes (n = 927) in genome-wide proliferation screens. d, Dropout of essential genes across studies based on MAGeCK MLE beta values. e, Comparison of PD1 (PDCD1) knockout enrichment in FACS-based screens sorting for PD1-negative cells. f, Strength of the separation between essential and non-essential genes, quantified by the strictly standardized mean difference (SSMD). CRISPR screens in the Jurkat cell line from the DepMap project are included for reference. g, Significant screening hits across all high-quality genome-wide screens in human T cells or CAR T cells (z-normalized log₂ fold change > 2 and FDR < 0.05 for MAGeCK RRA results, normalized beta > 0.99 for MAGeCK MLE results). Genes highlighted by each study are shown in green; established clinical targets are shown in magenta.

Extended Data Fig. 5. Optimization of FACS-based readouts for genome-wide screens in CAR T cells.

a, Target antigen on CAR T cells (as the result of trogocytosis) for two anti-CD19 CARs (FMC63 scFv with 41BB or CD28 co-stimulatory domains) and one anti-GD2 CAR (14g2a scFv with 41BB) following stimulation with K562-CD19 or NALM6-GD2 cells, respectively (2 donors). b, Antigen expression on the cell lines used for CAR T cell stimulation (K562-CD19 and NALM6-GD2). c, Expression of the activation markers CD69 and CD25 on CAR T cells (3 donors). d, Expression of the T cell exhaustion markers PD1, LAG3, and TIM3 on CAR T cells (2 donors). e, Overview of the FACS gating strategy. Viable T cells were selected based on physical gates, live/dead stain, and CD4 or CD8 marker expression, while excluding cell doublets. For screens involving co-culture with cancer cells, target cell engaging CAR T cells were selected based on trogocytosis-acquired CD19 (except in the CD19 screen). In the focused validation screens, cells were sorted for the top and bottom 50% of marker expression. For the genome-wide screens, cells were sorted using stringent thresholds of marker expression (panel k). f, Marker protein expression over time, comparing six methods for cell fixation with antibody staining either before or after fixation. Mean fluorescent intensities (MFI) were normalized to the staining of fresh unfixed cells. g, Recovery of fixed cells removed from storage at different days after cell fixation, for two alternative experimental workflows: staining after fixation (top) or staining before fixation (bottom). h, Genomic DNA yield from fresh and fixed cells for different de-crosslinking conditions. i, qPCR quantification of gRNA amplification from fresh and fixed cells for different de-crosslinking conditions. j, Optimization of marker protein gating and screening time points by flow cytometry (≥ 3 donors), with a representative histogram of marker-negative and marker-positive CAR T cells (left) and violin plots of log₂ fold change distributions (right) for gRNAs targeting the sorting markers (colored) or a safe harbor locus (grey). k, Sorted cell populations for all genome-wide FACS-based screens (n = 45, details in Supplementary Table 3). Within each sorted population, bars represent individual T cell donors. l, Scatterplot showing the percentage of reads aligning to the gRNA library as a function of the number of cells that were used as input for the gRNA amplification. Good results were obtained starting with as few as 1000 sorted cells. m, Barplot showing the percentage of all sequenced samples that achieved sufficient gRNA alignment rates. n, RNA expression of top hits from the genome-wide screens (Fig. 2d) based on RNA-seq for the CAR T cells.

Extended Data Fig. 6. Leukemia xenograft model and in vivo CROP-seq optimization.

a, Experimental timeline of the in vivo validation experiments with CRISPR-boosted CAR T cells, which were genetically engineered either by lentiviral co-delivery of the CAR and a pool of 8 gRNAs followed by mRNA delivery of Cas9 (as in the in vitro screens, top) or by electroporation of a pre-assembled RNP complex of Cas9 protein and one top-performing gRNA (as is common practice in CRISPR-edited cell therapy, bottom). b, CAR T cell titration in a xenograft mouse model of human leukemia. Immunodeficient NSG mice were injected with 0.5 million NALM6 cells engineered to express firefly luciferase. On day 5, mice were treated with different doses of CAR T cells. Leukemic cell load was monitored using live bioluminescence imaging. When left untreated, mice succumb to the leukemia around day 21. To make the model most informative, we selected a low (and deliberately non-curative) dose of CAR T cells that leads to initial leukemic control followed by a quick relapse. c, Survival analysis for the mice shown in panel b. d, Percentage of UMI reads perfectly matching the reference sequence, comparing the established method (gRNA amplification from genomic DNA) with in vivo CROP-seq (gRNA amplification from mRNA), both tested with single PCR and nested PCR amplification. e, Optimal number of UMI-based internal replicates for data analysis based on screening controls. Given that each UMI base can be A, C, G, or T, using UMI bases 1 to 5 results in 4, 16, 64, 256, and 1024 random internal replicates. Standard analysis is labeled as 0 internal replicates (left). Negative, neutral, and positive controls are color-coded. f, Dropout of neutral control gRNAs targeting a safe harbor locus when the read number in the internal replicates gets too small for large numbers of internal replicates. The box plot’s center line indicates the median, the box limits represent the upper and lower quartiles, and the whiskers extend to 1.5 times the interquartile range. g, Recall of negative and positive controls for different numbers of internal replicates. The grey line represents the optimal number of 16 internal replicates chosen for the analysis. h, gRNAs selected for in vivo screens (8 gRNAs per gene) or individual validation (as pools of 8 or single gRNAs). i, Individual effects of the 8 gRNAs per gene in the focused validation screen, with single gRNAs used for individual validation highlighted. The box plot’s center line indicates the median, the box limits are the upper and lower quartiles, and the whiskers extend to 1.5 times the interquartile range. j, Number of distinct T cell clones detected based on unique molecular identifiers (UMIs) in the in vivo screens. k, Estimation of CELLFIE’s scalability to large discovery screens in vivo, extrapolating the number of screenable perturbations from empirical measurements for different screening configurations.

Extended Data Fig. 7. Validation of CRISPR-boosted CAR T cells in in vivo models.

a, Bioluminescence images showing initial leukemic cell control and subsequent relapse for leukemic mice treated with standard (safe harbor locus-edited) CAR T cells or with PRDM1 or RHOG knockout CAR T cells (1 representative donor out of 9 tested). b, Overview of the leukemia xenograft models and CAR T cells used in this study, including 2 target antigens, 3 CAR designs, and 3 CAR T cell doses. c, Leukemic cell load (top: absolute values; bottom: fold reduction) for NALM6 xenograft mice treated with 0.6 million 19-BBz CAR T cells produced from six additional healthy donors (on top of those shown in Fig. 4b). d, Leukemic cell load (as in panel c) for CAR T cells produced from a leukapheresis product of one patient with multiple myeloma scheduled to receive CAR T cell therapy. e, CD19⁺ (NALM6) leukemic cells observed in mouse spleens at relapse following treatment with standard (safe harbor locus-edited) and RHOG knockout CAR T cells. NALM6 cells cultured in vitro are shown as reference. No downregulation of CD19 antigen was observed at relapse. f, Leukemic cell load (left) and survival analysis (right) for leukemic mice treated with a higher dose of 1.2 million 19-BBz CAR T cells (3 donors). g, Percentage of CAR T cells (CD4⁺ or CD8⁺) and NALM6 (CD19⁺) cells in long-term survivor mice for the NALM6 xenograft model treated with RHOG knockout or RHOG-and-FAS double-knockout CAR T cells. Blood (tail bleeds) and organs (at the end point) were analyzed by flow cytometry following erythrocyte lysis. h, Fold increase of RHOG knockout over standard (safe harbor locus-edited) CAR T cells with two different CARs (19-BBz or 19-28z) on day 6 of expansion following restimulation with anti-CD3/CD28 5 days after Cas9 RNP electroporation (3 donors). i, Leukemic cell load (left: absolute values; center: fold reduction) and survival analysis (right) for NALM6 xenograft mice treated with 0.6 million 19-28z CAR T cells. j, Leukemic cell load (left: absolute values; center: fold reduction) and survival analysis (right) for NALM6-GD2 xenograft mice treated with 0.3 million GD2-BBz CAR T cells. k, Overview (left), caliper-measured tumor size (top right), and survival analysis (bottom right) for the Huh7 solid tumor xenograft model treated with 0.5 million GPC3-BBz CAR T cells. P-values for survival analysis in panels f and i-k used the log-rank test comparing mice treated with knockout versus standard (safe harbor locus-edited) CAR T cells. Source Data

Extended Data Fig. 8. Molecular characterization of RHOG knockout CAR T cells.

a, Percentage of successful RHOG editing determined by sequencing of the gene locus. b, Killing of luciferase-expressing NALM6 cancer cells by standard (safe harbor locus-edited) or RHOG knockout CAR T cells, measured by luciferase assays at 18 h of co-culture for different CAR T cell versus target cell ratios. c, CAR expression on standard (safe harbor locus-edited) CAR T cells and RHOG knockout CAR T cells, determined by flow cytometry the day before injection into mice. d, CD4⁺ and CD8⁺ cell composition among standard (safe harbor locus-edited) CAR T cells and RHOG knockout CAR T cells, determined by flow cytometry the day before injection into mice (6 donors). e, T cell differentiation status after 10 days of co-culture of CAR T cells and cancer cells. Representative flow cytometry data are shown (1 out of 3 donors; complete data are provided in Fig. 4f). f, Cell surface marker expression after 16 h of co-culture of standard (safe harbor locus-edited) CAR T cells or RHOG knockout CAR T cells with K562-CD19 target cells (3 donors). P-values used paired one-tailed t-tests. g, Cell viability and apoptosis marker expression after 16 h of co-culture of standard (safe harbor locus-edited) CAR T cells or RHOG knockout CAR T cells with K562-CD19 target cells (3 donors). P-values used paired one-tailed t-tests. h, Principal component analysis (PCA) of normalized RNA-seq data, with samples color-coded by cell type, T cell donor, time point, and RHOG knockout versus standard CAR T cells. i, Number of differentially expressed genes (|log₂FC| > 0.5, adjusted p-value < 0.05) in RNA-seq data comparing RHOG knockout and standard CAR T cells, shown separately for CD4⁺  and CD8⁺ CAR T cells. j, Number of differential GO Biological Process terms (FDR < 0.05) in RNA-seq data comparing RHOG knockout and standard CAR T cells, shown separately for CD4⁺ and CD8⁺ CAR T cells. k, Network visualization of upregulated GO Biological Process terms in RNA-seq data comparing RHOG knockout and standard CAR T cells, shown separately for CD4⁺ (top) and CD8⁺ (bottom) CAR T cells. l, Gene set enrichment analysis of GO Biological Process terms for the three largest clusters from panel k. m, Differential expression analysis for genes associated with cell proliferation (S phase marker genes) comparing RHOG knockout and standard CAR T cells.

Extended Data Fig. 9. Optimization of combinatorial screening in CAR T cells.

a, Percentage of reads perfectly matching the full reference sequence (spacer1-iBAR1-spacer2-iBAR2) or one of its elements in the CROP-seq-CAR-multi plasmid library and at days 0 and 12 of the combinatorial screens. b, Percentage of reads showing evidence of recombination between the gRNA-iBAR pairs. c, Dropout of essential genes as determined by log₂ fold changes between day 12 and day 0, aggregated across screens with all three CARs. Results are grouped by gRNA combinations targeting only the safe harbor locus (“SH-SH”), essential genes only in position 1 (“essential-SH”) or position 2 (“SH-essential”), or both positions (“essential-essential”). d, Log₂ fold changes between day 12 and day 0 for gRNA combinations targeting RHOG-RHOG, FAS-FAS, or RHOG-FAS, aggregated across screens with all three CARs. e, Heat map of fitness effects (MAGeCK MLE beta values comparing day 12 and day 0) for pairwise gene knockouts. f, Dot plots of fitness effects (MAGeCK MLE beta values comparing day 12 and day 0) for pairwise gene knockouts. Dotted lines indicate same-gene combinations.

Extended Data Fig. 10. Optimization of base editing screening in CAR T cells and validation of single gRNAs.

a-e, Cumulative distribution plots showing enrichment and depletion of gRNAs predicted to introduce different types of mutations for Cas9 and four base editors (NS: nonsense mutation; MS: missense mutation; Splice: splice-site mutation). f, Mutagenesis map for the puromycin resistance gene PAC derived from the base editing screens (2 donors), comparing the gRNA distribution between day 12 and day 0 after CAR restimulation. Log₂ fold changes of z-scores (y-axis) are shown, and the top-15 most depleted gRNAs are labeled with the predicted amino acid changes. g, Amino acids with the strongest mutagenesis effects in the screen, mapped onto the PAC protein structure. The visualization is zoomed in on the PAC catalytic domain, with the full protein shown in the inset. h, Log₂ fold changes for 323 base editing gRNAs included in the validation screens (Supplementary Table 8). Aggregated results are shown for gRNAs targeting RHOG (left) and PAC (right) across the original screen (2 donors) and the validation screens (5 additional donors). Top gRNAs per base editor and donor are labeled with the predicted amino acid changes. i, Enrichment of two RHOG missense mutations in CAR T cells comparing day 12 and day 2 in the focused base editing screens, as determined by next-generation sequencing of the RHOG locus (4 donors). j, Quantification of ABEmax base editing by single gRNAs targeting RHOG or PAC, analyzed by Sanger sequencing with EditR. Editing efficiency was assessed on day 3 and mutation enrichment/depletion on day 12. k, Detailed editing outcomes for single gRNAs targeting RHOG, analyzed with EditR. Highlighted in green are the base editing rates for target bases whose substitutions introduce missense mutations.