A versatile CRISPR-Cas13d platform for multiplexed transcriptomic regulation and metabolic engineering in primary human T cells

Summary

CRISPR technologies have begun to revolutionize T cell therapies; however, conventional CRISPR-Cas9 genome-editing tools are limited in their safety, efficacy, and scope. To address these challenges, we developed MEGA (Multiplexed Effector Guide Arrays), a platform for programmable and scalable regulation of the T cell transcriptome using the RNA-guided, RNA-targeting activity of CRISPR-Cas13d. MEGA enables quantitative, reversible, and massively-multiplexed gene knockdown in primary human T cells without targeting or cutting genomic DNA. Applying MEGA to a model of CAR T cell exhaustion, we robustly suppressed inhibitory receptor upregulation and uncovered paired regulators of T cell function through combinatorial CRISPR screening. We additionally implemented druggable regulation of MEGA to control CAR activation in a receptor-independent manner. Lastly, MEGA enabled multiplexed disruption of immunoregulatory metabolic pathways to enhance CAR T cell fitness and anti-tumor activity in vitro and in vivo. MEGA offers a versatile synthetic toolkit for applications in cancer immunotherapy and beyond.

In Brief

MEGA is a programmable RNA-targeting platform that enables safe and effective multiplexed genetic perturbation of the primary human T cell transcriptome. Multi-gene disruption broadly enhances the anti-tumor activity of CAR T cells and uncovers a role for aerobic glycolysis in driving T cell exhaustion.

Graphical Abstract

INTRODUCTION

Chimeric Antigen Receptor (CAR) T cell therapies have revolutionized the treatment of refractory B cell and plasma cell malignancies and have demonstrated curative potential in aggressive preclinical cancer models. Despite these successes, however, significant barriers to progress remain, including primary or acquired resistance and limited efficacy against solid cancers^1–5. Recent studies have implicated T cell intrinsic mechanisms of dysfunction such as exhaustion and metabolic dysregulation in these failure modes^1,6–10. Born of concurrent advances in CRISPR-Cas9-based technologies, next-generation genetically engineered T cells offer a potential solution to these challenges^3,11. For example, specific genes can be deleted by Cas9 knockout to augment anti-tumor activity^12–22 or generate allogeneic “off-the-shelf” T cells^3,11,23–26.

Although the intersection of CRISPR engineering and CAR T cell therapy holds great promise, Cas9-based systems are limited in safety, efficacy, and potential for broader applications. Both Cas9 nuclease and base-editors permanently alter the T cell genome^27,28. While total gene ablation may be desired in certain contexts, a method to introduce graded perturbations in primary human T cells that can be dynamically regulated remains an unmet need for both biological discovery and therapeutic development²⁹. Furthermore, cutting genomic DNA with Cas9 nuclease can trigger unintended and potentially genotoxic side-effects: large-scale chromosomal translocation/truncation^25,30,31, chromosome loss/gain (aneuploidy)^32,33, and downregulation/loss of p53 activity^32,34,35 have been reported by multiple groups. Without the appropriate toolkit, our ability to engineer complex multi-gene programs and pathways for the optimization of next-generation T cell therapies³⁶ has remained limited.

To overcome the limitations posed by state-of-the-art T cell genome engineering technologies, we present MEGA (Multiplexed Effector Guide Arrays), a multi-functional platform for transcriptome regulation using CRISPR-Cas13d. Cas13d is a compact CRISPR effector with RNA-guided RNA endonuclease activity^37,38. Unlike Cas9, Cas13d does not bind or cut DNA³⁷. Rather, it complexes with a CRISPR-associated RNA (crRNA) containing a programmable ~23 nucleotide spacer that guides the ribonucleoprotein to RNA transcripts for sequence-specific degradation without PAM restriction. Additionally, Cas13d has the ability to process poly-crRNA guide arrays into individual crRNAs to facilitate efficient simultaneous targeting of multiple RNA transcripts in single cells^37,39. Finally, Cas13d is roughly two-thirds the size of wild-type Cas9, making it highly amenable to T cell manufacturing.

Here, we develop a Cas13d-based method for massively-multiplexed, quantitative, and reversible perturbation of the primary human T cell transcriptome. Focusing on lessons learned from the clinic, we apply MEGA to study dysfunctional GD2-targeting HA-28ζ CAR T cells, a well-characterized model that recapitulates key molecular, phenotypic, and functional features of T cell exhaustion^7,40–42. We deploy MEGA across a broad swath of applications ranging from druggable regulation of CAR signaling to multiplexed knockdown of nearly 10 genes at the single-cell level. MEGA also enables pooled combinatorial CRISPR screening in primary human T cells – using this approach, we identify pairs of putative exhaustion-related genes that synergistically regulate T cell function in the setting of chronic stimulation. Finally, by leveraging multi-gene knockdown to disrupt core metabolic pathways, we uncover a role for aerobic glycolysis in regulating T cell dysfunction and utilize these findings to potentiate MEGA HA-28ζ CAR T cell activity in vitro and in vivo.

RESULTS

Development of Cas13d for optimal expression and activity in primary human T cells

The CRISPR-Cas13d system derived from Ruminococcus flavefaciens XPD3002 (RfxCas13d) exhibits superior catalytic activity and targeting specificity in mammalian cells^37,43. To confirm the functional activity of lentivirally-integrated RfxCas13d in primary human T cells, we decided to target the upregulation of LAG3, PD-1, and TIM3 in dysfunctional HA-28ζ CAR T cells⁷ (Figure S1A). Prior studies have shown that blocking these inhibitory receptors is of clinical significance⁴⁴. We hypothesized that MEGA HA-28ζ CAR T cells, which co-express RfxCas13d and a targeting guide RNA array, could simultaneously suppress the upregulation of LAG3, PD-1, and TIM3 (Figure 1A). We designed three guides per gene⁴⁵ and evaluated knockdown efficiency by measuring the corresponding surface expression of LAG3/PD-1/TIM3 on day 10 in comparison to a non-targeting (NT) control guide (Figure S1B–C). Using a bicistronic CRISPR system, we observed limited knockdown across three donors due to inefficient lentiviral transduction of RfxCas13d (Figure S1D). Interestingly, we discovered that all single-vector bicistronic configurations resulted in low functional viral titer, which we attribute to either RfxCas13d array processing (forward orientation) or crRNA-guided cleavage (reverse orientation) of lentiviral RNA during packaging in 293T cells (Figure S1E–F). Guided by these data, we established an optimized workflow to manufacture MEGA HA-28ζ CAR T cells, in which primary human T cells were co-transduced with separate RfxCas13d and HA-28ζ CAR lentiviruses prior to transduction with crRNA lentivirus (Figure 1B, S1G).

Figure 1. MEGA HA-28ζ CAR T cells robustly suppress inhibitory receptor upregulation.

(A) Schematic of lentiviral constructs. DR: direct repeat.

(B) Optimized workflow to generate MEGA CAR T cells.

(C) Violin and box plots depicting day 10 surface LAG3/PD-1/TIM3 expression for single/double guides from 1 representative donor. Targeting guides are colored (LAG3, yellow; PD-1, red; TIM3, blue); non-targeting guides are grey. Dashed lines indicate MFI of mock untransduced or exhausted non-targeting (NT) cells. Numbers indicate normalized expression values (mock = 0; NT = 1).

(D) Normalized surface LAG3 (L), PD-1 (P), and TIM3 (T) expression as described in (C), mean ± s.d. of n = 2–3 donors from independent experiments. Ordinary one-way ANOVA with Dunnett’s multiple comparisons test.

(E) Correlation plots between transcript and surface protein expression for LAG3, PD-1, and TIM3 from 1 representative donor. Colored lines represent best fit from linear regression. Error bars are s.e.m.

(F) Bulk RNA-seq of day 10 MEGA HA-28ζ CAR T cells. Transcript abundances for targeting guides are plotted against those of the non-targeting control, averaged over n = 2 donors from independent experiments. Each dot represents a single gene; gene-level expression was determined by aggregating TPM values and p-values across all detected transcripts for each gene. Differentially-expressed genes with adjusted p-value < .05 and absolute log2 fold-change > 0.5 are highlighted. The green line indicates where gene expression is equal across both conditions.

(G) Normalized surface expression as in (D) for triple guide arrays, mean ± s.d. of n = 2–3 donors from independent experiments. Ordinary one-way ANOVA with Dunnett’s multiple comparisons test.

(H) Correlation plots as in (E) for triple guide arrays from 1 representative donor. Error bars are s.e.m.

(I) Pie charts depicting relative percentages of LAG3^+/−PD-1^+/−TIM3^+/− MEGA HA-28ζ CAR T cells from 1 representative donor. Yellow: triple-positive cells; blue: triple-negative cells.

See also Figures S1 and S2.

MEGA CAR T cells robustly suppress upregulation of exhaustion markers driven by tonic signaling

To assess whether optimized MEGA HA-28ζ CAR T cells could suppress LAG3/PD-1/TIM3 upregulation, we isolated primary T cells from multiple healthy donors and co-transduced cells with RfxCas13d and HA-28ζ CAR constructs. One day later, we transduced cells with either 1) single guides targeting each receptor, 2) double guide arrays targeting all pairwise combinations of receptors, or 3) NT guide. This sequential transduction process ensured that expression of the HA-28ζ CAR, which drives exhaustion, was uniform across all experimental conditions (Figure S1H). We sorted mCherry⁺ cells on day 5 and measured receptor surface expression on day 10 (Figure 1B). We observed strong upregulation of LAG3/PD-1/TIM3 in the NT control when compared to mock untransduced T cells (Figure 1C). Single guides and double guide arrays specifically suppressed upregulation of each targeted receptor to near-baseline levels, which were consistent across independent experiments with multiple donors (Figure 1D). To verify that inhibitory receptor suppression at the protein level was due to RfxCas13d activity at the transcript level, we quantified the abundance of LAG3, PDCD1 (PD-1), and HAVCR2 (TIM3) transcripts relative to NT (Figure 1E). We observed a strong correlation between transcript abundance and surface protein expression across all samples. Interestingly, we noted spacer-dependent positional effects within double guide arrays: while the LAG3 targeting spacer exhibited efficient knockdown regardless of array position, PD-1 and TIM3 targeting spacers were generally more effective in the 3’-proximal position.

To evaluate potential transcriptome-wide off-target effects, we performed bulk RNA-seq on day 10 MEGA HA-28ζ CAR T cells and analyzed LAG3, TIM3, and LAG3+TIM3 targeting guides because they exhibited the greatest knockdown efficiency across donors. We identified significant on-target knockdown of LAG3 and/or HAVCR2 compared to NT in all samples (Figure 1F). We detected minimal significant off-target or downstream biological effects: 0 when targeting LAG3, 7 when targeting TIM3, and 4 when targeting LAG3+TIM3 together. Notably, we did not observe activation of interferon pathways in response to either crRNA expression, array processing, or on-target RNA cleavage⁴⁶.

We next generated MEGA HA-28ζ CAR T cells expressing triple guide arrays that encoded all unique permutations of LAG3/PD-1/TIM3-targeting spacers to evaluate simultaneous knockdown of all three receptors. We confirmed that CAR expression was uniform across all samples (Figure S1I). All triple guide arrays were able to suppress upregulation of exhaustion markers across 2–3 donors (Figure 1G, S1J), and transcript abundance was strongly correlated with surface protein expression (Figure 1H). Certain spacer configurations exhibited more robust triple knockdown efficiency, and this pattern was consistent with positional dependencies that we previously observed. To validate triple knockdown at the single-cell level, we quantified the relative abundance of LAG3^+/−PD-1^+/−TIM3^+/− populations in our flow cytometry data and observed an increase in the triple-negative population across all guide arrays, with LPT being the optimal configuration (Figure 1I, S1K–L).

MEGA does not exhibit collateral activity in primary human T cells

RfxCas13d can occasionally trigger non-specific degradation of RNA (collateral activity) upon on-target cleavage of highly-expressed transcripts in certain mammalian cells^47–50. To examine whether collateral activity was present in our system, we performed sample-matched flow cytometry and bulk RNA-seq on MEGA T cells generated from two healthy donors (Figure S2A). We chose to target B2M, one of the most abundant transcripts expressed in primary human T cells (Table S7).

MEGA exhibited efficient knockdown of B2M at the protein level and did not affect the expression of off-target proteins such as CD46 or CD3 (Figure S2B). We also did not detect collateral knockdown of the mCherry reporter, a phenomenon commonly observed by other groups^47–49 (Figure S2C). We next sorted the mCherry⁺ fraction and immediately isolated total RNA for transcriptome-wide analysis. To evaluate potential toxicity, we counted viable sorted cells and did not observe significant differences between groups (Figure S2D–E). We also did not observe significant differences in total RNA yield, suggesting a lack of nonspecific RNA degradation in cells with on-target RfxCas13d activity (Figure S2F).

Principal component analysis on bulk RNA-seq data showed that most of the sample variation arose from donor-donor differences (PC1) rather than differences in targeting vs. non-targeting conditions (PC2) (Figure S2G). To investigate PC2 further, we performed differential expression analysis between B2M and NT groups; we confirmed B2M knockdown and identified 37 differentially-expressed genes (Figure S2H, Table S7). A recent study used mitochondrial RNAs (mtRNAs) as an internal control to identify collateral effects in count-normalized data⁴⁹. We identified mtRNAs in our samples and did not observe any significant differences between targeting and non-targeting groups (Figure S2H, Table S7). mtRNA levels in our bulk RNA-seq data remained unchanged across all targeting and non-targeting conditions (Figure 1F). Taken altogether, MEGA does not show evidence of collateral activity in primary human T cells.

MEGA facilitates combinatorial CRISPR screening to identify paired regulators of CAR T cell proliferation

Dual knockout of candidate genes identified in CRISPR-Cas9 screens often results in greater enhancements to T cell function than single knockouts alone^13,51–53. However, strategies to unbiasedly discover these “synergistic” gene pairings have not been developed for primary human T cells. Combinatorial CRISPR screens, which perturb multiple genes per cell, offer a powerful method to systematically decode these interactions^54,55.

We applied MEGA in a proof-of-concept screening study to identify combinations of putative exhaustion-related genes that regulate the proliferation of dysfunctional CD8⁺ HA-28ζ CAR T cells in culture (Figure 2A). We curated a list of 24 genes based on prior CRISPR-Cas9 knockout studies in primary human CD8⁺ T cells (Table S1). We designed and assembled a custom library of 5,184 double guide arrays targeting all 576 pairwise combinations of these genes (Figure S3A, Table S1, see Methods). Each targeting guide was also paired with a set of 8 randomly generated non-targeting guides to simulate an additional 1,152 “single” guide perturbations. Finally, as a control we included 64 paired non-targeting guides for a total of 6,400 unique guide arrays.

Figure 2. Combinatorial screening identifies gene pairs that regulate HA-28ζ CAR T cell proliferation and anti-tumor activity.

(A) Overview of 2D CRISPR screen.

(B) 2D heatmap of guide array enrichment between early and late timepoints.

(C) Volcano plot depicting log2 fold-change (l2fc) of guide arrays between early and late timepoints, n=2 replicates. Significantly enriched/depleted arrays (adj. p < .05, |l2fc| > 0.5) are highlighted.

(D) Ranked guide arrays using likelihood-ratio test, n=2 replicates. Arrays with the most significant count variation across all screening timepoints are highlighted.

(E) Top: Histogram depicting the distribution of l2fcs between early and late timepoints for all 6,400 guide arrays. Bottom: Rug plots depicting l2fcs for all 9 guide arrays targeting either significantly enriched (red) or depleted (blue) gene pairs, overlaid on the l2fc distribution of all 64 non-targeting guides (grey).

(F) Fold-change expansion of FACS-sorted RfxCas13d⁺ CD8⁺ HA-28ζ CAR T cells compared to NT over 15 days in culture, mean ± s.e.m. of n=3–4 donors from independent experiments. Ordinary one-way ANOVA.

(G) Top: schematic detailing serial stimulation assay. Bottom: Kinetics of repeated tumor killing, mean ± s.e.m. of triplicate wells from 1 representative donor. Red: enriched guide arrays; black: NT control. Repeated measures one-way ANOVA.

See also Figures S3 and S4.

We transduced T cells isolated from a healthy donor with RfxCas13d, HA-28ζ CAR, and the pooled double guide array library (Figure 2A, S3B, see Methods). Cells were sorted for RfxCas13d expression on day 5 and split into two replicates to account for technical variation in culturing conditions and downstream processing. Cells were cultured in parallel over two weeks, with HA-28ζ tonic signaling driving both T cell expansion and subsequent dysfunction. CD8⁺ T cells were then magnetically isolated from the bulk population upon sample collection on days 11, 13, and 15 of culture. We quantified guide array abundance through PCR amplification and deep sequencing of the guide array cassette. Unlike multiplex Cas9 screens performed in cancer cell lines, which require multiple reads and complex barcoding schema to accurately identify guide pairs⁵⁴, our approach easily identified pairs without barcoding by directly sequencing the compact RfxCas13d guide array in a single read. Complete library coverage was supported by strong replicate correlation and non-zero counts for all guide arrays across all samples (Figure S3C–D).

We performed pairwise statistical testing to determine whether guide arrays were significantly enriched or depleted between early and late screening timepoints (Figure 2B). Numerous guide arrays exhibited a consistent effect on HA-28ζ CAR T cell proliferation across replicates. As expected, non-targeting control arrays were neither enriched nor depleted. Single perturbations (those paired with a non-targeting spacer) generally resulted in a weaker effect on proliferation than double perturbations (Figure 2C, S3E). We were initially surprised by the robust depletion of guides targeting IRF4 and JUNB; however, in addition to driving dysfunction, these transcription factors are also critical for the expansion and persistence of T cells^7,56–58. Highly-enriched double arrays frequently included FAS (such as FAS in combination with either ZC3H12A, CTLA4, or SOCS1), which is consistent with its role in activation-induced cell death^59,60. We also aggregated guide array enrichment data and observed comparable results at the broader gene level (Figure S3F–G). Lastly, we ranked guide arrays with the most significant count variation across all screening timepoints as determined by likelihood-ratio test and identified top hits consistent with pairwise analyses (Figure 2D).

We chose a set of top-ranking paired hits (7 enriched and 2 depleted) to functionally validate based on both guide- and gene-level data (Figure 2E). For each of these gene pairs, we cloned the corresponding guide array with the greatest screen effect size based on log2 fold-change. We then generated MEGA HA-28ζ CAR T cells from 3–4 donors across independent experiments and tracked T cell expansion relative to NT over 15 days in culture. Paired perturbations regulated T cell expansion in culture and generally recapitulated screening results in both CD8⁺ and bulk T cells (Figure 2F, S3H–I). Surprisingly, we also observed a significantly smaller RfxCas13d⁺ population in cells expressing depleted arrays by day 5 (despite uniform CAR expression), suggesting that dysregulation of proliferation occurs within 48 hours post-transduction (Figure S3J–K).

Paired transcriptomic perturbations enhance the anti-tumor activity of dysfunctional CAR T cells

To investigate whether our screen hits could improve HA-28ζ CAR T cell anti-tumor activity, we performed a small-scale secondary screen using in vitro cytokine secretion and tumor killing as functional readouts (Figure S4A). Top guide arrays robustly augmented IFNγ secretion and moderately improved IL-2 secretion of day 10 MEGA HA-28ζ CAR T cells upon stimulation with Nalm6-GD2 tumor (Figure S4B–D). We also observed markedly improved tumor killing across multiple donors and effector:target (E:T) ratios in comparison to NT (Figure 2G, S4E–G). This was especially apparent upon serial stimulation: paired knockdown of CBLB+FAS potentiated a robust and durable anti-tumor response across multiple tumor rechallenges unlike NT, which failed to control tumor outgrowth. We hypothesized that dual targeting of CBLB and FAS improved HA-28ζ CAR T cell activity through a synergistic disruption of two orthogonal pathways known to negatively regulate T cell activation^14,60,61 (Figure S4H). Indeed, when we repeated tumor co-culture assays with single targeting controls (CBLB alone and FAS alone), we observed that only the dual targeting array resulted in an enhancement of both tumor killing and cytokine secretion (Figure S4I–L).

MEGA enables rapid, tunable, and reversible perturbation of the T cell transcriptome

In earlier experiments, we noted that transcript knockdown was wholly dependent on RfxCas13d expression (Figure S5A). Informed by this data, we fused a destabilization domain from the Escherichia coli dihydrofolate reductase (DD) to the C-terminus of RfxCas13d to enable tunable and reversible control of gene knockdown (Figure 3A). At steady-state, regulatable RfxCas13d-DD is rapidly degraded by the proteasome⁶². In the presence of trimethoprim (TMP), an FDA-approved small molecule antibiotic, RfxCas13d-DD is stabilized and can target and cleave RNA.

Figure 3. MEGA enables rapid, quantitative, and reversible control of the T cell transcriptome.

(A) Regulation of RfxCas13d-DD expression by trimethoprim (TMP).

(B) Violin and box plot overlays depicting surface CD46 expression on day 5 from 1 representative donor. Purple: on-target; grey: off-target.

(C) Kinetics of CD46 knockdown after TMP addition (left) or removal (right) over 72 hours.

(D, E) Violin and box plot overlays

(D) and dose-response curve

(E) depicting TMP-dependent titration of CD46 knockdown. Ordinary one-way ANOVA.

(F) LAG3 (yellow), PD-1 (red), or TIM3 (blue) expression on day 10 MEGA HA-28ζ CAR T cells expressing LPT array in response to TMP dosage. Two-way ANOVA (row-factor).

(G) Top: TCR signal transduction pathway. Bottom: TMP-dependent degradation of LCK and ZAP70 transcripts tunes T cell activation strength.

(H) Top: PROX (4-plex) guide array. Bottom: Day 10 intracellular protein expression of ZAP70 and Lck in MEGA HA-28ζ CAR T cells expressing either the PROX array (right) or NT control (left).

(I) IL-2 secretion by day 10 MEGA HA-28ζ CAR T cells stimulated with Nalm6-GD2 in response to TMP dosage. Ordinary one-way ANOVA.

(J, K) CD69

(J) and CD3

(K) expression on day 10 MEGA HA-28ζ CAR T cells in response to TMP, mean±s.d. of triplicate wells from 1 representative donor.

(L) LAG3 (yellow), PD-1 (red), or TIM3 (blue) expression on day 10 MEGA HA-28ζ CAR T cells expressing PROX array in response to TMP dosage. Two-way ANOVA (row-factor).

(C-F) and (I-L) show mean ± s.d. of triplicate wells from 1 representative donor. See also Figure S5.

We initially confirmed that RfxCas13d-DD protein expression was regulatable by TMP in primary human T cells through intracellular staining (Figure S5B). To characterize the dynamic range of RfxCas13d-DD for endogenous gene repression, we generated MEGA T cells expressing RfxCas13d-DD and either a CD46 targeting guide or a non-targeting guide. Following co-transduction, we cultured MEGA T cells in the presence or absence of TMP for 48 hours and measured CD46 surface expression on day 5. MEGA T cells repressed CD46 in a TMP-dependent manner, retained full functional activity in the presence of drug, exhibited minimal leaky activity in the absence of drug, and did not alter guide targeting specificity (Figure 3B). To investigate whether we could reverse gene repression, we repeated the experiment but on day 5, we removed or added TMP, respectively. We then tracked surface CD46 expression over 72 hours using flow cytometry, whereby we observed rapid kinetics of both induction and reversal of CD46 downregulation, and further confirmed that gene repression was completely reversed to baseline (pre-induction) levels upon TMP removal (Figure 3C, S5C).

We next sought to quantitatively tune CD46 surface expression by varying TMP dosage. We performed a TMP titration experiment and observed a remarkably sensitive sigmoidal dose-response curve with an analog-like linear regime between approximately 1–100 nM (Figure 3D–E). Although increasing concentrations of TMP led to lower CD46 expression, no collateral mCherry knockdown or differences in cell viability were detected (Figure S5D–F).

The HEPN-2 domain located at the C-terminal end of RfxCas13d enables guide array processing³⁹. To evaluate whether our C-terminal DD fusion retained the ability to process guide arrays, we knocked down LAG3, PD-1, and TIM3 in RfxCas13d-DD MEGA HA-28ζ CAR T cells expanded with different TMP dosages over 10 days (Figure 3F). RfxCas13d-DD enabled multiplexed repression of all targeted markers in a dose-dependent manner, with highly comparable knockdown efficiencies to those of the constitutive RfxCas13d system (Figure 1G).

Dynamic regulation of proximal signaling modulates MEGA CAR T cell activity

The safety and efficacy of CAR T cells can be improved through synthetic control of CAR structure/function at the protein-level^41,63,64. As an alternative approach we utilized RfxCas13d-DD to target the proximal signaling genes LCK and ZAP70⁶⁵ to build a tunable regulator of CAR activation in a receptor-agnostic manner (Figure 3G, S5G). We assembled a multiplex targeting array (“PROX”) that was validated in RfxCas13d-DD MEGA HA-28ζ CAR T cells (Figure 3H, S5H). With increasing amounts of TMP in culture, PROX exhibited a dose-dependent decrease in IL-2 secretion and CD69 expression (Figure 3I–J). To examine off-target effects, we quantified CD3 expression, which did not change significantly with added TMP (Figure 3K). We measured levels of PD-1/TIM3/LAG3 and observed a drug-dependent reduction in the exhausted T cell surface phenotype (Figure 3L).

We next explored whether MEGA could regulate the activity of other CARs. We generated MEGA CD19-BBζ and ROR1–28ζ CAR T cells, which utilize different costimulatory domains and scFv domains. On day 10 of culture, we stimulated these cells with corresponding antigen-positive Nalm6 tumor cells with and without TMP (Figure S5I, S5L). Across both CAR–antigen pairs, we observed significantly decreased T cell activation with TMP as measured by IL-2 secretion and surface expression of CD25 (IL2RA) (Figure S5J–K, S5M–N). Our findings demonstrate that, by regulating steady-state levels of proximal signaling molecules, MEGA enables effective receptor-independent tuning of CAR T cell activation.

MEGA enables massively-multiplexed gene knockdown in primary human T cells

To explore potential technical limitations in the number of genes that could be targeted in parallel using MEGA, we designed a facile one-step cloning scheme to enable fast and accurate assembly of arbitrarily long RfxCas13d guide arrays (Figure 4A, see Methods). We built longer arrays targeting the exhausted CAR T cell surfaceome due to the abundance of unique proteins with detectable surface epitopes and demonstrated clinical relevance. We assembled a 5-plex array (“SURF1”) by adding two new spacers targeting FAS⁶⁰ and CTLA4⁶⁶ to our array targeting LAG3, PDCD1, and HAVCR2. To stress-test the multiplex capabilities of our system, we added five more genes to SURF1 for a total of 10 targeted genes (“SURF2”): LAG3, FAS, CD5⁶⁷, ENTPD1⁶⁸, CD46⁶⁹, TRAC, B2M, CTLA4, PDCD1, and HAVCR2.

Figure 4. MEGA enables massively-multiplexed gene knockdown in primary human T cells.

(A) Guide array assembly method.

(B) Top: SURF1 (5-plex) and SURF2 (10-plex) guide arrays. Bottom: Electropherogram traces (with 15 and 1500 bp markers) of genomic PCR amplicons. Expected amplicon lengths are highlighted in pink.

(C) Day 10 transcript levels of LAG3/FAS/CD5/ENTPD1/CD46/TRAC/B2M/CTLA4/PDCD1/HAVCR2 relative to NT, mean ± s.d. of n=3 technical replicates from one representative donor. Ordinary one-way ANOVA with Dunnett’s multiple comparisons test.

(D) Day 10 surface protein expression (MFI) relative to NT, mean ± s.d. of n=2 donors across independent experiments. Two-way ANOVA (row-factor) with Bonferroni’s multiple comparisons test.

(E) t-SNE plot depicting FlowSOM-clustered high-dimensional flow cytometry data representing n=2000 (donor 1) or n=600 (donor 2) day 10 MEGA HA-28ζ CAR T cells expressing either the SURF2 array or NT control. Each dot represents a cell cluster (k=100). Dot position: cluster centroid; dot size: number of cells, dot coloring: cell enrichment (l2fc count ratio) of SURF2 (dark-blue) or NT (grey) cells after de-barcoding. Contour lines: kernel density estimation.

(F) t-SNE plots as described in (E) colored according to total surface expression of targeted markers (summed z-scores) from n=2 donors across independent experiments.

(G) Differential analysis of total surface protein expression in cell clusters with more NT cells (grey dots, cell enrichment < 0) versus those with more SURF2 cells (blue dots, cell enrichment > 0) from n=2 donors across independent experiments. Each dot represents a cell cluster as described in (E). Pink lines: median values. Ordinary one-way ANOVA with Bonferroni’s multiple comparisons test.

(H) t-SNE plots as in (E) colored according to surface marker expression (z-score) for each targeted marker from one representative donor.

We generated MEGA HA-28ζ CAR T cells expressing either SURF1, SURF2, or NT for downstream characterization. First, we performed targeted amplification of the guide array cassette from isolated genomic DNA and did not observe a heterogeneous mixture of fragment lengths, indicating that SURF1 and SURF2 arrays do not recombine after genomic integration (Figure 4B). We next quantified transcript abundances for all targeted genes (Figure 4C). As expected for SURF1, we observed significant knockdown across all 5 targeted transcripts to baseline levels or lower. Remarkably, for SURF2 9 of 10 genes had significant transcript knockdown without prior optimization of spacer sequence or position. However, for all genes targeted in the SURF1 array, we observed slightly lower knockdown efficiency by SURF2, suggesting that guides may be competing for limited Cas13 protein as in previous Cas9 retroactivity studies^59,70. High-dimensional flow cytometry of SURF2 cells validated massively-multiplexed knockdown at the protein level (Figure 4D). We found that 8 of 10 markers targeted by SURF2 showed significantly decreased surface expression relative to NT across two donors.

To further examine multiplexed knockdown efficacy at the single-cell level, we performed automated unsupervised clustering on flow cytometry data of SURF2 and NT cells and visualized the output using t-SNE. We observed a significant overlap between cell clusters enriched for SURF2 and those with low total surface expression scores across two donors from independent experiments, while the inverse was true for NT clusters (Figure 4E–G). On a marker-by-marker basis, we also observed that SURF2 clusters had lower expression levels than NT clusters for most of the targets, and clusters that were low in one target were generally low for most others (Figure 4H). Collectively, our data highlights the ability of MEGA to knock down large gene sets and offers a powerful approach to investigate pathways that remain unexplored in primary human T cells.

Whole-pathway disruption of purine metabolism enhances CAR T cell effector function

The purinergic signaling pathway rapidly converts extracellular inflammatory ATP into immunosuppressive adenosine (ADO) through a cascade of enzymatic reactions that involves four key surface proteins: CD39, CD73, A2AR, and A2BR^71,72 (Figure 5A). We hypothesized that we could slow the generation of ADO while accumulating ATP to enhance the anti-tumor activity of dysfunctional CAR T cells through multiplexed knockdown with MEGA. We built a 4-plex guide array (“PURI”) targeting ENTPD1 (CD39), NT5E (CD73), ADORA2A (A2AR), and ADORA2B (A2BR) by screening three spacers per gene and concatenating top spacers together (Figure 5B, Figure S6A). To validate array functionality, we manufactured MEGA HA-28ζ CAR T cells expressing either the PURI array or an NT guide and observed significant knockdown across all four targeted transcripts on day 10 of culture (Figure 5C). We next performed an ATP spike-in experiment to measure concentrations of ATP, AMP, and ADO in culture (Figure 5D). MEGA HA-28ζ CAR T cells expressing the PURI array accumulated significantly more ATP and generated significantly less AMP and less ADO than NT (Figure 5E–G). We also measured effective rate constants and found that PURI significantly slowed ATP hydrolysis and AMP production (Figure 5H–I).

Figure 5. Whole-pathway disruption of purine metabolism boosts CAR T cell effector function.

(A) Schematic of baseline purinergic signaling (NT).

(B) Disrupted purinergic signaling (PURI) limits adenosine-mediated immunosuppression.

(C) Day 10 transcript levels of ENTPD1/NT5E/ADORA2A/ADORA2B relative to NT, mean ± s.d. of n=3 technical replicates from one representative donor. Welch’s t-test.

(D) Schematic for metabolite detection.

(E, F, G) Concentrations of ATP

(E), AMP

(F), and ADO

(G) in culture media of MEGA HA-28ζ CAR T cells, mean ± s.d. of n=2–3 replicate wells from 1 representative donor. Unpaired t-test.

(H, I) Concentrations of ATP

(H) and AMP

(I) in culture media of MEGA HA-28ζ CAR T cells over time after spike-in with 20 μM ATP, mean ± s.d. of triplicate wells from 1 representative donor. Repeated measures two-way ANOVA, column-factor.

(J) Schematic for Nalm6-GD2 co-culture assays.

(K, L) Secretion of IFNγ

(K) or IL-2

(L) by MEGA HA-28ζ CAR T cells after 24 hr culture with (stimulated) or without (baseline) tumor at 1:1 E:T, mean ± s.d. of triplicate wells from 1 representative donor. Two-way ANOVA with Bonferroni’s multiple comparisons test.

(M, N) Kinetics of tumor killing

(M) and T cell proliferation

(N), mean ± s.e.m. of triplicate wells from 1 representative donor. Repeated measures two-way ANOVA, column-factor.

See also Figure S6.

We next set up co-culture assays with Nalm6-GD2 cells to investigate whether disruption of the purinergic pathway could improve effector function in dysfunctional CAR T cells (Figure 5J). Prior to tumor stimulation, we confirmed that CAR expression was uniform across all samples (Figure S6B). PURI knockdown significantly enhanced IFNγ and IL-2 secretion (Figure 5K–L, Figure S6C–D) and boosted in vitro tumor killing and T cell activation and proliferation in response to antigen stimulation (Figure 5M–N, Figure S6E) in comparison to NT. We additionally confirmed that multiplexed perturbation by PURI resulted in greater enhancement of anti-tumor function than single-gene targeting controls (Figure S6F–J).

Multi-gene targeting of aerobic glycolysis improves CAR T cell fitness and limits exhaustion in tonically signaling CAR T cells

The metabolic switch from oxidative phosphorylation (OXPHOS) to aerobic glycolysis in activated T cells is coupled to the acquisition of effector function^73,74, which if sustained may lead to eventual exhaustion^10,75. Accordingly, we hypothesized that we could use MEGA to limit T cell exhaustion by suppressing the PI3K/Akt axis⁷⁶ and downstream glycolytic enzymes in a specific and cell-intrinsic manner. We targeted the two main isoforms of Akt (AKT1 and AKT2) and hexokinase (HK1 and HK2) expressed in human T cells using a 4-plex array (“GLY”) (Figure 6A, S7A). Tonically-signaling MEGA HA-28ζ CAR T cells expressing the NT control strongly upregulated AKT1, AKT2, and HK2 in comparison to mock untransduced cells (Figure 6B). Expression of the GLY array counteracted this upregulation as demonstrated by significant knockdown of all four targeted transcripts. To further explore the GLY vs. NT phenotype, we performed high-dimensional mass cytometry using an exhaustion-specific panel, ran automated unsupervised cell clustering, and visualized the output with t-SNE (Figure 6C). We observed strong overlap between clusters enriched for GLY and clusters with decreased expression of molecules associated with co-stimulation (4–1BB, OX40), Th1 differentiation (T-bet), and inhibition (LAG3, PD-1, TIM3, CD39, and CTLA4), while the inverse was true for NT clusters (Figure 6D).

Figure 6. Disruption of aerobic glycolysis modulates CAR T cell metabolism and improves T cell fitness.

(A) Multiplexed knockdown of GLY genes disrupts effector differentiation and exhaustion.

(B) Day 10 transcript levels of HK1/HK2/AKT1/AKT2 relative to NT, mean±s.d. of n=3 technical replicates from one representative donor. Ordinary one-way ANOVA with Dunnett’s multiple comparisons test.

(C) t-SNE plot depicting FlowSOM-clustered mass cytometry data representing n=8000 day-10 MEGA HA-28ζ CAR T cells expressing either the GLY array or NT control. Each dot represents a cell cluster (k = 100). Dot position: cluster centroid; dot size: number of cells, dot coloring: cell enrichment (l2fc count ratio) of GLY (dark-green) or NT (yellow) cells after de-barcoding. Contour lines: kernel density estimation.

(D) t-SNE plots as described in (C) colored according to scaled median marker expression.

(E) Bulk RNA-seq data from day 10 MEGA HA-28ζ CAR T cells (GLY vs. NT). Transcript abundances were averaged over n=2 donors from independent experiments. Differentially-expressed genes with adj. p-value<0.2 and |l2fc|>0.5 are highlighted.

(F) GSEA of transcriptomic data in (E). Top 5 significant gene sets are shown. Labeled genes contributed most to the normalized enrichment score (NES).

(G) Generation of Cas9 gene-edited HA-28ζ CAR T cells via RNP electroporation.

(H) Editing efficiency of four-gene knockout (4KO) vs. safe-harbor control (AAVS1) as determined by TIDE, mean ± s.d. of n=2 donors across independent experiments. Two-way ANOVA with Bonferroni’s multiple comparisons test.

(I) Fold-change expansion of HA-28ζ CAR T cells with four gene Cas13 knockdown (left) or Cas9 kockout (right) over 10 days in culture, mean±s.d. of n=3–4 donors across independent experiments. Unpaired t-test.

(J) Normalized Nalm6-GD2 tumor intensities after 48 hr co-culture with HA-28ζ CAR T cells, mean ± s.d. of triplicate wells from one representative donor. Ordinary one-way ANOVA with Bonferroni’s multiple comparisons test.

(K) Bulk RNA-seq as in (E) of day 10 HA-28ζ CAR T cells gene-edited with Cas9 4KO (y-axis) or Cas9 AAVS1 control (x-axis), averaged over n=2 donors from independent experiments.

(L) GSEA as in (F) for transcriptomic data in (K).

(M) Euler diagrams depicting significantly differentially-expressed genes for Cas13 (GLY vs. NT) vs. Cas9 (4KO vs. AAVS1). Top: downregulated; bottom: upregulated.

(N) Top 5 GSEA results for Cas13 vs. Cas9.

See also Figure S7.

We next performed bulk RNA-seq to investigate any global transcriptomic differences arising from GLY knockdown (Figure 6E). To our surprise, we detected wide transcriptional reprogramming, with 188 downregulated and 123 upregulated genes resulting from our 4-plex perturbation. These changes aligned with our protein-level findings, with significant downregulation of HAVCR2 (TIM3), PDCD1 (PD-1), and TNFRSF9 (4–1BB). We also observed consistent upregulation of histone genes (H1, H2A, H2B, H4) and downregulation of many effector- and exhaustion-related genes (GZMB, PRF1, IFNG, BATF3, CSF1, NR4A2/3). Gene set enrichment analysis (GSEA) indicated significant downregulation of gene-sets related to effector function (IL-2/STAT5 and NF-κB signaling) and NK-like dysfunction⁷⁷, suggesting a fitter and less exhausted phenotype (Figure 6F). Notably, we observed significant upregulation of OXPHOS, which suggests that our findings were linked to the successful reprogramming of T cell metabolism.

MEGA outperforms conventional Cas9 gene-editing for metabolic engineering in primary human T cells

To explore whether we could achieve the GLY knockdown phenotype in HA-28ζ CAR T cells using the conventional Cas9-based genome-editing approach, we generated quadruple AKT1/AKT2/HK1/HK2 knockout, or “4KO”, HA-28ζ CAR T cells through electroporation of Cas9 RNPs complexed with a mixture of validated sgRNAs (Figure 6G). We confirmed successful editing against an AAVS1 safe-harbor cutting control (Figure 6H). Notably, 4KO (Cas9 knockout) exhibited poor expansion in culture over 10 days, while GLY (Cas13 knockdown) augmented the proliferative capacity of HA-28ζ CAR T cells, consistent with the upregulation in histone genes that we observed in our RNA-seq data (Figure 6I). Furthermore, upon co-culture with Nalm6-GD2 cells, 4KO HA-28ζ CAR T cells failed to clear tumor unlike other CAR T cell groups (Figure 6J). We did not observe differences between Cas13 NT and Cas9 AAVS1 control groups.

To directly compare the phenotypes by Cas13 knockdown and Cas9 knockout, we performed bulk RNA-seq and GSEA on our Cas9 4KO vs. AAVS1 HA-28ζ CAR T cells (Figure 6K–L). 4KO resulted in 87 downregulated genes (25 shared with GLY) and 92 upregulated genes (18 shared with GLY) in comparison to AAVS1 (Figure 6M). While we observed similar effects on effector- and exhaustion-related genes and pathways between the two methods, only Cas9 4KO downregulated the p53 pathway and upregulated DNA damage and G2-M checkpoint pathways. Importantly, these changes – which we attribute to the documented genotoxicity of multiplexed Cas9 knockouts and more generally to NHEJ-mediated repair of genomic double-strand breaks^30–32,35 – were not detected using MEGA (Figure 6N). These results were also consistent with the decreased expansion of 4KO cells that we observed in culture (Figure 6I). Furthermore, only Cas13 GLY knockdown resulted in significant upregulation of OXPHOS. In summary, our findings suggest that 1) MEGA enables multiplexed perturbation without undesirable and potentially genotoxic side-effects, and 2) successful reprogramming of core metabolic processes such as aerobic glycolysis necessitates the use of knockdown-based tools like MEGA rather than knockout-based tools like Cas9.

Downregulation of glycolytic activity in dysfunctional CAR T cells limits terminal differentiation to potentiate tumor clearance in vitro and in vivo

We next evaluated whether the healthier and less exhausted GLY phenotype translated to enhancements in anti-tumor efficacy. We generated MEGA HA-28ζ CAR T cells and observed that CAR expression was uniform between GLY and NT groups across all donors (Figure S7B). Despite proliferating at a faster rate than NT cells, GLY cells maintained a higher steady-state pH level in culture media, which is consistent with lower extracellular acidification and metabolic activity (Figure 7A). GLY cells also expressed lower levels of the activation marker CD69 than NT cells, both at resting (due to tonic signaling) and when stimulated with tumor (Figure S7C).

Figure 7. Glycolytic disruption enhances MEGA HA-28ζ CAR T cell anti-tumor activity in vitro and in vivo.

(A) pH of MEGA HA-28ζ CAR T cell culture supernatant, mean ± s.d of triplicate wells representative of n=3 donors across independent experiments. Unpaired t-test.

(B, C) Secretion of IL-2

(B) or IFNγ

(C) by MEGA HA-28ζ CAR T cells after 24 hours of co-culture with Nalm6-GD2 at 1:1 E:T. Data are mean ± s.d. of triplicate wells from 1 representative donor. Unpaired t-test.

(D) Kinetics of repeated tumor killing. Data are mean ± s.e.m of triplicate wells from 1 representative donor. Repeated measures two-way ANOVA.

(E) In vivo assessment of MEGA HA-28ζ CAR T cells in Nalm6-GD2 tumor-bearing NSG mice (n=4 per group).

(H) Phenotyping of day 10 pre-infusion MEGA HA-28ζ CAR T cells. E: effector; SCM: stem cell memory; EM: effector memory; CM: central memory.

(G) Analysis of tumor clearance via BLI measurement. Grey: mock; yellow: MEGA HA-28ζ NT; dark green: MEGA HA-28ζ GLY. Dotted line indicates background signal (limit of detection). Two-way repeated measures ANOVA, column-factor; unpaired t-test of endpoint BLI.

(H) Tumor BLI at week 3 post-treatment, mean ±s .d. of n=4 mice per group. Ordinary one-way ANOVA with Dunnett’s multiple comparisons test.

(I) CAR T abundance at week 3 post-treatment, mean ± s.d. of n=4 mice per group. Unpaired t-test.

See also Figure S7.

Interestingly, GLY cells secreted lower amounts of both IL-2 and IFNγ than NT cells across multiple donors, suggesting restrained effector differentiation consistent with our RNA-seq data (Figure 7B–C, S7D–E). This phenotype was only partially achieved with Cas9 4KO, which had no effect on IFNγsecretion (Figure S7G–H). Intracellular cytokine staining revealed a decreased frequency of IFNγ⁺ cells resulting from GLY knockdown in both resting and stimulated conditions, while the proportion of IL-2⁺ and TNFα⁺ cells were comparable across all groups (Figure S7F). GLY cells initially cleared tumor at similar or slower rates than NT cells; however, given our cytokine data we were surprised to find that upon repeated stimulation with fresh tumor, GLY cells outperformed NT cells and significantly enhanced tumor killing (Figure 7D, S7I).

We proceeded to evaluate the in vivo anti-tumor efficacy of MEGA HA-28ζ CAR T cells using Nalm6-GD2 tumor-bearing mice. After one week of tumor outgrowth, we injected mice with either mock untransduced T cells, MEGA HA-28ζ CAR T cells expressing NT, or MEGA HA-28ζ CAR T cells expressing GLY (Figure 7E). We evaluated differentiation during the manufacturing phase and observed predominantly effector memory T cells in the NT group and central memory T cells in the GLY group (Figure 7F). Post-treatment, we monitored tumor growth over 40 days using bioluminescence imaging (BLI) and additionally bled mice after 3 weeks to evaluate CAR T cell persistence. GLY knockdown significantly enhanced HA-28ζ CAR T cell tumor clearance in vivo: 3 of 4 mice had undetectable levels of tumor by day 40 in the GLY group as opposed to 0 of 4 in the NT group (Figure 7G). Analysis of peripheral blood showed that the decrease in tumor burden evident in the GLY group was associated with a concomitant increase in CAR T cell abundance (Figure 7H–I). Taken altogether, our results demonstrate that MEGA-mediated multiplexed disruption of aerobic glycolysis limits HA-28ζ CAR T cell differentiation and dysfunction to potentiate CAR T cell anti-tumor efficacy in the setting of chronic stimulation.

DISCUSSION

Multiplexed gene silencing is a major focus in the development of next-generation T cell therapies^3,23–25. Unlike Cas9-based tools, MEGA does not incur genotoxicity and/or perturbation heterogeneity, with increasing plexity. Indeed, we demonstrated the capacity to simultaneously repress nearly 10 genes in primary human T cells from a single guide array without targeting or cutting genomic DNA. While higher-order multiplexing likely requires optimization of spacer position/sequence for significant repression of all targeted genes, we describe a rapid guide array assembly method to accelerate this process.

Our results indicate that MEGA does not trigger detectable collateral RfxCas13d activity in primary human T cells, which is consistent with prior work showing that collateral activity depends on cell type and spacer sequence, among other factors^47–50. We note that all mammalian Cas13 collateral activity studies have utilized plasmid transfection, where copy number can easily exceed that of lentiviral transduction (our method) by 100- to 1000-fold. Interestingly, a recent study focused on lentiviral transduction of Cas13 and crRNA in common cell lines determined that a two-vector approach was the optimal configuration to avoid undesirable “intrinsic” RNA targeting⁷⁸. MEGA, which also utilizes a two-vector system, corroborates these findings.

RfxCas13d activity does not disrupt CAR signaling and is compatible with endogenous T cell function. MEGA robustly suppressed LAG3/PD-1/TIM3 despite active transcriptional^7,41 and epigenetic⁴² upregulation of these genes. Our combinatorial screen identified paired hits that broadly enhanced CAR T cell function in the setting of chronic stimulation. Although our proof-of-concept study focused on a subset of 24 genes (576 pairings), MEGA is highly scalable and can be easily adapted to screen larger libraries with higher dimensionality. For such applications, pre-validated spacer sequences should improve signal-to-noise to allow for quantitative genetic interaction mapping⁵⁴, which can be aided by algorithms that generate highly-efficient RfxCas13d spacers43,45,79,80.

T cell metabolic engineering is promising to improve therapeutic outcomes. However, effective methods remain an unmet need^10,75, as illustrated by our results with the gold-standard Cas9 approach. In contrast, glycolytic disruption with MEGA resulted in enhanced T cell fitness and anti-tumor activity, which was initially surprising given that we observed diminished cytokine secretion in co-culture assays. However, our findings are consistent with prior studies linking T cell stemness to clinical benefit and suggest that exhaustion is driven in part by the PI3K/Akt axis and downstream glycolytic activity. We also observed persistent metabolic reprogramming during in vivo challenge. Overall, we believe that MEGA will significantly advance metabolic engineering efforts in T cells and anticipate that further applications may inform metabolic interventions in cancer immunity.

In summary, we developed MEGA for versatile transcriptomic regulation in primary human T cells using CRISPR-Cas13d. Our work highlights a variety of T cell engineering applications with MEGA and addresses key limitations posed by state-of-the-art CRISPR-Cas9 genome-editing technologies. As a synthetic immunology toolkit, MEGA equips T cells with capabilities spanning applications from basic biological discovery to enhanced immunotherapies for cancer and beyond.

Limitations of the study

Further studies are required to evaluate the feasibility of MEGA for clinical translation. Our study focused on improving anti-tumor activity in the setting of chronic stimulation using tonically signaling HA-28ζ CAR T cells. Follow-up studies should examine how multiplexed perturbations affect the anti-tumor function of other CAR and TCR T cells, and interrogate the underlying biological mechanisms behind paired screen hits such as CBLB+FAS. As with most synthetic modalities, RfxCas13d immunogenicity is a potential limitation (Figure S2I). However, whether immunogenicity is a barrier to successful outcomes remains an open question since all FDA-approved CAR T cell therapies currently utilize non-human scFv recognition domains. In addition, a clinical study using Cas9-edited T cells suggested that pre-existing immunodeficiency in cancer patients may preclude an immune response to exogenous peptides²⁴. Future strategies to address immunogenicity could include RfxCas13d protein engineering⁸¹, bioinformatic mining for smaller Cas13s, or including a B2M-targeting spacer in guide arrays to minimize antigen presentation by MEGA T cells.

STAR METHODS

RESOURCE AVAILABILITY

Lead contact

Requests for reagents, resources, and further information should be directed to and will be fulfilled by the lead contact, Lei S. Qi (stanley.qi@stanford.edu).

Materials availability

All resources and materials reported in this paper will be shared by the lead contact upon request.

Data and code availability

• The raw and processed RNA-seq data generated in this study have been deposited, and are publicly available, at GEO: GSE246823 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE246823]. Analysis details are provided in the STAR Methods section.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Animal models

Immunocompromised NSG mice (NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ, JAX #005557) mice were purchased from the Jackson Laboratory and bred in-house under sterile conditions compliant with Stanford University Laboratory Animal Care (APLAC) protocols. Mice were housed in a barrier facility at Stanford University with a 12-hour light/dark cycle and monitored daily by veterinary staff. Healthy male mice were used for in vivo experiments. Mice were 4 months old at tumor engraftment and had not been involved in previous procedures, drug treatments, or experiments.

Cell lines

Lenti-X 293T female human embryonic kidney cells (Takara) were cultured in 0.22 μm sterile-filtered DMEM (Gibco) supplemented with 10% fetal bovine serum (FBS). Nalm6-GD2, Nalm6-ROR1, and Nalm6-CD19 male acute lymphoblastic leukemia cells were cultured in 0.22 μm sterile-filtered complete medium (CM) [RPMI supplemented with 10% FBS, 100 U ml⁻¹ penicillin, and 100 ug ml⁻¹ streptomycin (Gibco)]. All cells were cultured in humidified incubators at 37 °C and 5% CO₂.

Primary human T cell isolation and expansion

Buffy coats derived from anonymous healthy blood donors were purchased from the Stanford Blood Center under an IRB-exempt protocol. Primary human T cells were isolated using the EasySep Human T cell Isolation kit (STEMCELL) according to the manufacturer’s protocol with Ficoll-Paque PLUS (GE Healthcare) and SepMate-50 tubes. Isolated T cells were immediately cryopreserved at 4–10 × 10⁶ cells per vial in either FBS supplemented with 10% DMSO (Sigma-Aldrich) or CryoStor (STEMCELL).

Cryopreserved T cells were thawed and activated same-day using Human T-Activator CD3/CD28 Dynabeads (Gibco) at a 3:1 bead-to-cell ratio. T cells were cultured in human T cell medium (HTCM) [0.22 μm sterile-filtered AIM V supplemented with 5% FBS, 10 mM HEPES, 2 mM GlutaMAX, 100 U ml⁻¹ penicillin, and 100 ug ml⁻¹ streptomycin (Gibco)]. Human recombinant IL-2 (STEMCELL) was provided at 100 U ml⁻¹. On day 3 of culture, Dynabeads were magnetically removed. Cells were expanded every other day by adding HTCM to maintain an overall cell concentration of 0.5–1 × 10⁶ cells per ml. Live cell counts were obtained following the manufacturer’s protocol for Trypan Blue exclusion using the Countess 3 automated cell counter (Invitrogen).

For culture conditions requiring trimethoprim (TMP, Sigma-Aldrich), a 1000 μM concentrated stock solution of TMP (in DMSO) was freshly thawed and added to cells starting on day 3 of culture to a final concentration of 1 μM unless otherwise noted. To maintain culture conditions, cells were expanded every other day in HTCM supplemented with 1 μM TMP unless otherwise noted. To remove TMP, cells were washed twice with FACS buffer [DPBS supplemented with 2% FBS and 1 mM EDTA (Gibco)] and resuspended in fresh prewarmed HTCM.

METHOD DETAILS

Cloning of lentiviral constructs

All oligonucleotides were synthesized by IDT or by the Stanford Protein and Nucleic Acid (PAN) Facility. To facilitate flexible cloning of whole guide arrays and avoid hairpin formation near the Esp31 digest site, pLentiRNAGuide_002 (Addgene #138151) was modified to remove the direct repeat region (pSLQ4419). Sequences for the high-affinity 14G2a-GD2(E101K) chimeric antigen receptor (HA-28ζ CAR) and mCherry-P2A-Ruminococcus flavefaciens Cas13d (RfxCas13d) (Addgene #155305) sequences were previously described^7,43. The HA-28ζ CAR sequence was cloned into a pHR lentiviral backbone with a constitutive SFFV promoter (pSLQ5263). Regulatable RfxCas13d-DD was constructed by fusing the destabilizing domain (DD) from Escherichia coli dihydrofolate reductase (DHFR)⁶² to the C-terminus of RfxCas13d with a short glycine-serine linker.

RfxCas13d spacer sequences were generated using the previously-described cas13design web tool (https://cas13design.nygenome.org/)⁴⁵ and can be found in Table S1. Guide array constructs were cloned using the following methods: For traditional restriction-ligation cloning, forward and reverse oligonucleotides for each spacer were annealed together, phosphorylated using T4 PNK (NEB), and ligated into backbone using T4 DNA ligase (NEB). For In-Fusion assembly (Takara), whole guide arrays were amplified using PCR primers containing 15–20 nt long 5’ and 3’ homology regions. The PCR amplicon was then directly inserted into backbone. For NEBuilder HiFi assembly (NEB), single-stranded oligonucleotides with 20–30 nt long 5’ and 3’ homology regions were synthesized and directly inserted into backbone.

Longer (5- and 10-plex) guide arrays were cloned using a two-step overlap extension PCR method. Arrays were synthesized as single-stranded oligo fragments encoding a direct repeat flanked by two overlapping spacer regions (see Figure 4A and Table S1). In the first PCR reaction, oligo pairs were annealed and extended. In the second PCR reaction, double-stranded fragments were pooled together and amplified without primers for 15 cycles. Then, forward and reverse primers containing 5’ and 3’ homology regions were spiked in before continuing for another 15–20 cycles. PCR amplicons were run on a 2% agarose gel, purified, and inserted into backbone using NEBuilder HiFi assembly.

Lentiviral preparation

7.5 × 10⁵ Lenti-X cells were seeded in 6-well plates overnight containing 2 ml DMEM supplemented with 10% FBS. The next morning, 850 μl culture medium was removed and cells were transfected with 0.55 μg pMD2.G (Addgene plasmid #12259), 1.28 μg psPAX2 (Addgene plasmid #12260), and 1.79 μg transfer plasmid in 426 μl Opti-MEM (Gibco) using 10.88 μl TransIT-LT1 (Mirus Bio). Six hours later, culture medium was completely removed and replaced with fresh DMEM supplemented with 10% FBS and 1× ViralBoost (Alstem Bio). 24 hours after transfection, viral supernatant was harvested and filtered through a 0.45 μm syringe filter (Millipore). The supernatant was mixed with lentivirus precipitation solution (Alstem Bio), incubated at 4 °C for a minimum of 4 hours, and concentrated 10–100× in Opti-MEM following the manufacturer’s protocol. Concentrated virus was either used fresh or kept frozen at −80 °C for future use. Lentiviral production in 10- and 15-cm dishes were scaled up proportionally to culture vessel surface area.

Transduction of primary human T cells

Primary human T cells were transduced with concentrated lentivirus at 1–10% v/v (unless otherwise noted) 24–48 hours after bead activation. Construct expression was verified on day 5 of culture via flow cytometry as described above. For constructs requiring puromycin selection, a concentrated stock solution of 10 μg μl⁻¹ puromycin was thawed and added to cells on day 3 of culture (24–48 hours after transduction) to a final concentration of 1 μg ml⁻¹. Puromycin selection was complete by day 5 of culture.

CRISPR-Cas9 genome-editing of primary human T cells

On day 3 of culture, 1 × 10⁶ primary human HA-28ζ CAR T cells were de-beaded, spun down, and resuspended in supplemented P3 buffer (Lonza). For multiplexed knockouts, 20 μg recombinant Cas9 nuclease (IDT) was complexed with 4 μl of sgRNA mixture, e.g. 1 μl each sgRNA diluted to 100 μM in TE (Twist Bioscience) for 30 minutes at room temperature. For single knockout, 5 μg Cas9 was complexed with 1 μl sgRNA. Following incubation, T cells were mixed with Cas9 RNPs and immediately electroporated using the P3 Primary Cell 4D-Nucleofector kit (Lonza) according to the manufacturer’s protocol. Cells were recovered in fresh prewarmed HTCM.

Cell sorting

On day 5 of culture, primary human T cells were spun down and resuspended in ice-cold FACS buffer at a concentration of 1–2 × 10⁷ cells per ml. Cells were sorted on a Sony SH800 sorter (Sony Biotechnology) using a 130 μm chip at low sample pressure on “semi-purity” or “purity” mode. Mock untransduced T cells were used to determine RfxCas13d^+/– gating. Live RfxCas13d⁺ cells were sorted into collection tubes containing cold HTCM. After sorting, cells were spun down and resuspended in fresh prewarmed HTCM supplemented with 100 U ml⁻¹ IL-2 to a final concentration of approximately 3–5 × 10⁵ cells per ml. After sorting, cells were expanded as described above.

Immunostaining and flow cytometry

mCherry fluorescence was used as a quantitative measure of RfxCas13d expression in all flow cytometry experiments. The 1A7 anti-14G2a idiotype antibody used to detect HA-28ζ CAR surface expression was conjugated in-house with the DyLight 650 antibody labeling kit (Thermo Fisher). Details on specific antibody clones and conjugated fluorophores used can be found in the Key Resources table.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
anti-14G2a idiotype antibody (clone 1A7)	National Cancer Institute	N/A
AF405 anti-human CD8 (clone 3B5)	Invitrogen	Cat#MHCD0826; RRID:AB_10372951
AF405 anti-human LCK (clone LCK-01)	Novus Biologicals	Cat#NB500-336AF405; RRID:AB_3073774
AF488 anti-human ZAP70 (clone A16043B)	Biolegend	Cat#693505; RRID:AB_2715832
APC anti-human CD46 (clone TRA-2-10)	Biolegend	Cat#352405; RRID:AB_2564356
APC anti-human CD69 (clone FN50)	Biolegend	Cat#310910; RRID:AB_314845
APC anti-human PD-1 (clone A17188B)	Biolegend	Cat#621609; RRID:AB_2832829
APC-Cy7 anti-human CD39 (clone A1)	Biolegend	Cat#328225; RRID:AB_2571980
BUV395 anti-human LAG3 (clone T47-530)	BD Biosciences	Cat#569247; RRID:AB_3073773
BV421 anti-FLAG tag (clone L5)	Biolegend	Cat#637321; RRID:AB_2750051
BV421 anti-human CD3 (clone OKT3)	Biolegend	Cat#317343; RRID:AB_2565848
BV421 anti-human CD45RA (clone HI100)	Biolegend	Cat#304129; RRID:AB_10900421
BV421 anti-human CD5 (clone L17F12)	Biolegend	Cat#364029; RRID:AB_2734407
BV421 anti-human TIM3 (clone F38-2E2)	Biolegend	Cat#345007; RRID:AB_10900073
BV510 anti-human PD-1 (clone EH12.2H7)	Biolegend	Cat#329931; RRID:AB_2562255
BV605 anti-human FAS (clone DX2)	Biolegend	Cat#305627; RRID:AB_2562444
BV711 anti-human CTLA4 (clone BNI3)	Biolegend	Cat#369631; RRID:AB_2892450
FITC anti-human CD3 (clone OKT3)	Biolegend	Cat#317305; RRID:AB_571906
FITC anti-human LAG3 (clone 11C3C65)	Biolegend	Cat#369307; RRID:AB_2629750
FITC anti-human β2-microglobulin (clone 2M2)	Biolegend	Cat#316304; RRID:AB_492837
PE anti-human α/β T Cell Receptor (clone IP26)	Biolegend	Cat#306707; RRID:AB_314645
PE-Cy7 anti-human CD25 (clone 2A3)	BD Biosciences	Cat#335789; RRID:AB_399968
PerCP-Cy5.5 anti-human CD62L (clone DREG-56)	Biolegend	Cat#304823; RRID:AB_893396
PerCP-Cy5.5 anti-human TIM3 (clone F38-2E2)	Biolegend	Cat#345015; RRID:AB_2561933
Bacterial and virus strains
Stellar Competent Cells	Takara	Cat#636766
Biological samples
Buffy coats from healthy human subjects	Stanford Blood Center	N/A
Chemicals, peptides, and recombinant proteins
Adenosine 5’-triphosphate disodium salt hydrate (ATP)	Sigma-Aldrich	Cat#A6419
Alt-R S.p. Cas9 Nuclease V3	IDT	Cat#1081058
Human recombinant IL-2	STEMCELL	Cat#78036
Lentivirus precipitation solution	Alstem Bio	Cat#VC100
TransIT-LT1	Mirus	Cat#MIR2306
Trimethoprim	Sigma-Aldrich	Cat#92131
Trypan blue stain	Invitrogen	Cat#T10282
ViralBoost	Alstem Bio	Cat#VB100
Critical commercial assays
Adenosine Assay Kit (Fluorometric)	Abcam	Cat#ab211094
AMP-Glo Assay	Promega	Cat#V5011
CellTiter-Glo 2.0 Cell Viability Assay	Promega	Cat#G9241
Cyto-Fast Fix/Perm Buffer Set	Biolegend	Cat#426803
DNeasy Blood and Tissue kit	QIAGEN	Cat#69506
DyLight 650 Microscale Antibody Labeling Kit	Thermo Fisher	Cat#84536
EasySep Human CD8 Positive Selection Kit II	STEMCELL	Cat#17853
EasySep Human T cell Isolation kit	STEMCELL	Cat#17951
Human IFNγ ELISA MAX Deluxe	Biolegend	Cat#430104
Human IL-2 ELISA MAX Deluxe	Biolegend	Cat#431804
Human T-Activator CD3/CD28 Dynabeads	Gibco	Cat#11132D
In-Fusion Snap Assembly Master Mix	Takara	Cat#638947
iScript cDNA Synthesis kit	Bio-Rad	Cat#1708891
iTaq Universal SYBR Green Supermix	Bio-Rad	Cat#1725124
KAPA HiFi HotStart	Roche	Cat#07958897001
MiSeq Reagent Kit v3 (600-cycle)	Illumina	Cat#MS-102-3003
NEBuilder HiFi assembly	NEB	Cat#E2621S
NucleoSpin PCR Clean-up kit	Macherey-Nagel	Cat#740609.50
P3 Primary Cell 4D-Nucleofector X kit S	Lonza	Cat#V4XP-3032
RNeasy Plus Mini kit	QIAGEN	Cat#74136
SepMate-50 tubes	STEMCELL	Cat#85450
Deposited data
Raw and processed data	This paper	GEO: GSE246823
Experimental models: Cell lines
Lenti-X 293T	Takara	Cat#632180
Nalm6-CD19	Lynn et al.7	N/A
Nalm6-GD2	Lynn et al.7	N/A
Nalm6-ROR1	Labanieh et al.63	N/A
Experimental models: Organisms/strains
Mouse: NSG (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ)	Jackson Laboratory	Cat#005557; RRID:IMSR_JAX:005557
Oligonucleotides
See Table S1 for key oligonucleotide sequences.	This paper	N/A
Recombinant DNA
ecDHFR	Banaszynski et al.62	N/A
HA-28ζ CAR	Lynn et al.7	N/A
pLentiRNAGuide_002	Wessels et al.45	Cat#138151; RRID:Addgene_138151
pMD2.G	Didier Trono	Cat#12259; RRID:Addgene_12259
pSLQ4419 pHR (U6-BsmBI-BsmBI-EFS-PuroR-WPRE)	This paper	N/A
pSLQ4428 pHR (EF1a-mCherry-P2A-RfxCas13d-2xNLS-3xFLAG-ecDHFR-WPRE)	This paper	N/A
pSLQ5428 pHR (EF1a-mCherry-P2A-RfxCas13d-2xNLS-3xFLAG-WPRE)	Abbott et al.43	Cat#155305; RRID: Addgene_155305
psPAX2	Didier Trono	Cat#12260; RRID:Addgene_12260
Software and algorithms
bowtie2	Langmead et al.92	N/A
cas13designtool	Wessels et al.45	https://cas13design.nygenome.org/
CATALYST	Crowell et al.82	N/A
cutadapt	Marcel Martin91	N/A
CytExpert Software	Beckman Coulter	https://www.beckman.com/
DESeq2	Love et al.94	N/A
fgsea	Korotkevich et al.88	N/A
FlowJo v10	FlowJo	https://www.flowjo.com/
FlowSOM	Quintelier et al.83	N/A
ggplot2	Wickham et al.87	https://ggplot2.tidyverse.org/
Graphpad Prism 9	GraphPad	https://www.graphpad.com/
Incucyte Base Analysis Software	Sartorius	https://www.sartorius.com/
kallisto	Bray et al.85	https://pachterlab.github.io/kallisto/
Living Image v4.7.4	PerkinElmer	https://www.perkinelmer.com/
MAGeCK	Li et al.96	N/A
RStudio	RStudio	https://posit.co/products/open-source/rstudio/
samtools	Li et al.93	N/A
sleuth	Pimentel et al.86	https://pachterlab.github.io/sleuth/
SnapGene	SnapGene	https://www.snapgene.com/
Other
2100 Bioanalyzer	Agilent	N/A
Countess 3 automated cell counter	Invitrogen	N/A
CytoFLEX S	Beckman Coulter	N/A
Helios	Fluidigm	N/A
IncuCyte S3	Sartorius	N/A
IVIS Spectrum imaging system	PerkinElmer	N/A
MiSeq System	Illumina	N/A
SH800 sorter	Sony Biotechnology	N/A
Synthetic sgRNA Kit	Synthego	N/A
Twist Oligo Pools	Twist Bioscience	N/A

Approximately 1–2 × 10⁵ cells were resuspended in 100 μl FACS buffer and stained at room temperature for 20 minutes. Samples probing for CAR surface expression were stained on ice for 10 minutes to minimize receptor internalization. When required, cells were fixed and permeabilized using the CytoFast Buffer set (Biolegend) following the manufacturer’s protocol. For intracellular cytokine staining, cells were stimulated with tumor for 6 hours prior to fixing. Cells were washed twice with FACS buffer prior to flow cytometry, which was conducted on a CytoFLEX S (Beckman Coulter). For all samples, a minimum of approximately 1 × 10⁴ events was collected within the final gated population-of-interest.

Mass cytometry

On day 10 of culture, 1 × 10⁶ primary human HA-28ζ CAR T cells were spun down, washed 2× with PBS (Gibco), and stained with 250 nM cisplatin (Fluidigm) for 3 minutes at room temperature. Cells were immediately quenched with cell staining media (CSM, PBS supplemented with 0.05% BSA and 0.02% sodium azide), spun down, and fixed with 1.6% paraformaldehyde for 10 minutes at room temperature. Fixed cells were washed with PBS, resuspended in CryoStor (STEMCELL), flash-frozen, and stored at −80°C. Cells were thawed, stained, washed, prepared, and run on a Helios mass cytometer (Fluidigm) as previously described⁶³.

High-dimensional cytometry analysis

Marker expression was analyzed in a custom R (version 4.1.2) script using the CATALYST workflow⁸². Cells were de-barcoded and filtered for live RfxCas13d⁺ singlets. Cells were downsampled (see figure legends for exact numbers), pooled across conditions, clustered using FlowSOM unsupervised clustering (k = 100)⁸³, and visualized using t-distributed stochastic neighbor embedding (t-SNE). Cell enrichment scores were determined for each FlowSOM cluster by calculating log2 fold-change of the ratio between targeting and non-targeting cell counts. Where described, total surface expression scores were calculated by summing the z-scored expression level of each targeted surface marker.

Electropherogram analysis

Genomic DNA was isolated from approximately 1 × 10⁶ MEGA HA-28ζ CAR T cells on day 10 of culture. Genomic PCR primers were designed to bind upstream of the hU6 promoter and downstream of the poly-T terminator, resulting in targeted amplification of the integrated guide array cassette. PCR amplicons were run on the Agilent 2100 Bioanalyzer.

RNA extraction and quantitative reverse transcription PCR (RT-qPCR)

Total RNA was extracted from approximately 1–10 × 10⁶ primary human T cells using the RNeasy Plus Mini kit (QIAGEN) according to the manufacturer’s protocol using additional 2-mercaptoethanol. RNA concentration was measured using a NanoDrop One (Thermo Fisher) and normalized across all cell samples prior to reverse transcription using the iScript cDNA Synthesis kit (Bio-Rad). Diluted cDNA was added to iTaq Universal SYBR Green Supermix (Bio-Rad) and run on a CFX384 Touch (Bio-Rad) with the following PCR conditions: 50 °C for 10 min, 95 °C for 30 s, 40 × (95 °C for 10 s, 60 °C for 30 s). Gene expression was quantified using the 2^–ΔΔCt method using SDHA as an internal control⁸⁴. Primer sequences can be found in Table S1.

Bulk RNA-seq and transcriptomic analyses

Total RNA was extracted from approximately 5 × 10⁶ primary human T cells on day 10 of culture, as described above. Bulk RNA-seq was performed by Novogene (Davis, CA) on an Illumina NovaSeq 6000 with 20–30 × 10⁶ 150-bp paired-end reads per sample. Transcript abundance was quantified from raw reads via pseudo-alignment to the human transcriptome using kallisto⁸⁵ with n = 100 bootstraps (index built with Homo sapiens GRCh38.p14 cDNA). Custom scripts in R (version 4.1.2) were used to perform differential expression analysis: quants were filtered, batch-corrected, aggregated, and analyzed using sleuth⁸⁶, and resulting outputs were visualized using ggplot2⁸⁷. The output transcripts per million (TPM) datasets are provided for all bulk RNA-seq experiments (Table S3, S5, S7). Raw FASTQ and processed (kallisto quants) data are available at GEO with accession number GSE246823. Gene set enrichment analysis (GSEA) was performed in R using the fgsea package⁸⁸. GSEA ranking metric = -log10(adj. p-value) * sign(l2fc treated/control). All HALLMARK gene sets⁸⁹ were included for analysis, as well as a published list of 30 genes associated with NK-like T cell dysfunction⁷⁷. Full GSEA results are provided in Table S6.

Design and assembly of pooled double guide array library

A custom, curated library of 24 putative exhaustion-related genes was designed based on results from prior CRISPR-Cas9 proliferation and functional knockout screens in primary human T cells (see Table S1). Three top-ranking 23-nt guide spacer sequences targeting the CDS of the broadest range of human transcript isoforms were generated per gene as described above using the cas13design tool⁴⁵, yielding 72 targeting guide spacers. In addition, eight 23-nt random non-targeting guide spacers that did not align to the human transcriptome (as confirmed by nucleotide BLAST) were generated. In total, 80 guide spacers were generated. A custom Python script was used to concatenate RfxCas13d direct repeats to spacer sequences and to output all pairwise guide array combinations, resulting in 80 × 80 = 6,400 double guide array sequences (see Table S1). Flanking 5’ and 3’ homology regions were added to facilitate insertion into the pSLQ4419 backbone as described above. The guide array library was synthesized as a single-stranded oligo pool (Twist Bioscience) and 10 ng template was amplified using KAPA HiFi HotStart (Roche) with the following PCR conditions: 95 °C for 3 min, 8 × (98 °C for 20 s, 55 °C for 15 s, 72 °C for 15 s), 72 °C for 1 min. The reaction product was recovered using a NucleoSpin PCR Clean-up kit (Macherey-Nagel) and run on a 2100 Bioanalyzer (Agilent) by the Stanford PAN Facility for sizing and quantification (85% double guide arrays) and quality control (minimal PCR overamplification). The amplified library was efficiently cloned, with less than 5% background, into pSLQ4419(Esp31) using NEBuilder HiFi DNA assembly and variable spacer regions matching the nucleotide distribution of the library were confirmed by Sanger sequencing. Complete library representation with minimal bias⁹⁰ (Gini coefficient: 0.146, 90^th percentile/10^th percentile of guide array reads: 1.96, see Figure S3A) was further verified by Miseq v3 600-cycle paired-end sequencing (Illumina).

CD8⁺ CAR T cell proliferation screen

8 × 10⁷ primary human T cells (adjusted 1:1 CD4⁺-to-CD8⁺ ratio) were thawed and activated 3:1 with CD3/CD28 Dynabeads same-day, as described above. Throughout the screen, CD8⁺ T cells were cultured together with CD4⁺ T cells in HTCM supplemented with 100 U ml⁻¹ IL-2. On day 1 of culture, cells were co-transduced with the following freshly harvested and 100× concentrated lentiviruses (prepared as described above with endotoxin-free plasmid DNA): 2% v/v mCherry-P2A-RfxCas13d, 1.5% v/v HA-28ζ CAR, and 0.2% v/v pooled double guide array library. On day 3 of culture, cells were selected with 1 μg ml⁻¹ puromycin over 48 hours. Approximately 28.9% of cells survived puromycin selection, corresponding to ~83.9% single lentiviral integrations (assuming independent transduction events following the Poisson distribution). On day 5, a portion of cells was stained and flowed for HA-28ζ CAR and CD8 expression to determine the overall percentage of CAR⁺CD8⁺ T cells. This percentage informed the subsequent FACS of RfxCas13d⁺ cells and ensured that a sufficient number of cells was sorted to achieve library coverage of approximately 1000× within the CAR⁺CD8⁺ T cell subpopulation. Post-sort, cells were split into two replicates at ~1000× representation and expanded every other day as described above. On days 11, 13, and 15 of culture, a portion of cells was stained and flowed to determine the percentage of CD8⁺, RfxCas13d⁺, and CAR⁺ cells. Following flow cytometry, CD8⁺ T cells were magnetically isolated from the total T cell culture and collected at ~1000× representation using the EasySep human CD8 Positive Selection kit (STEMCELL). The proliferation screen was terminated after 15 days of ex vivo expansion.

Genomic DNA extraction and library preparation

Genomic DNA (gDNA) was extracted from magnetically isolated CD8⁺ T cells using the DNeasy Blood and Tissue kit (QIAGEN). Cells were lysed using Buffer AL and spun down at maximum speed for three minutes to pellet magnetic positive selection beads. The cell lysate supernatant was then carefully transferred into a clean tube before proceeding with the manufacturer’s protocol for gDNA isolation.

The double guide array library was prepped for next-generation sequencing through a two-step PCR protocol. Briefly, PCR1 primers targeted and amplified lentivirally-integrated guide array sequences, appended Illumina sequencing primer binding sites, and increased the overall sequence diversity of the amplicon with eight custom stagger sequences to improve cluster identification. PCR2 primers appended paired-end indexes for sample identification and contained Illumina P5/P7 sequences that bound to the flow cell. Primer sequences for both PCR1 and PCR2 can be found in Table S1. For each collected sample, a total of ~40 μg gDNA was PCR1 amplified (6 μg gDNA roughly corresponds to 1 × 10⁶ cells) as follows: a maximum of 2 μg gDNA template was amplified per 50 μl reaction with the following cycling conditions: 95 °C for 3 min, 24 × (98 °C for 20 sec, 70 °C for 15 sec, 72 °C for 15 sec), 72 °C for 1 min. 200 ng of plasmid DNA (pDNA) library was also PCR1 amplified. The PCR1 products were pooled together by sample and column purified. Approximately 10 ng of each purified PCR1 product was loaded as template DNA for PCR2 with the following cycling conditions: 95 °C for 3 min, 12 × (98 °C for 20 sec, 65 °C for 15 sec, 72 °C for 15 sec), 72 °C for 1 min. All reactions were set up with KAPA HiFi HotStart (Roche). PCR2 products were run on a 2% ultra-pure agarose gel (Invitrogen), bands were gel-extracted using the Nucleospin Gel Clean-up kit (Macherey-Nagel), and DNA was quantified on a Qubit fluorometer (Invitrogen).

Screen readout and analysis

CRISPR guide array enrichment was read out via pooled Miseq v3 600-cycle paired-end sequencing (Illumina). Samples were accurately identified according to paired-end indexes. Raw fastq files were 5’ and 3’ adapter-trimmed and filtered by quality (-q 30) and length using cutadapt⁹¹. Processed reads were aligned to a custom index built from a reference library of all double guide array combinations using bowtie2 end-to-end alignment⁹². Guide array counts were generated from alignment data using samtools⁹³ (Table S4).

Screening data was analyzed using DESeq2⁹⁴ within a custom R (version 4.1.2) script. Technical replicate pair correlation was evaluated by linear regression of normalized replicate count data. Significant guide array enrichment and depletion were analyzed via pairwise comparisons using the Wald test to calculate log2 fold-changes and adjusted p values for each guide array. Based on previously-described Methods⁹⁵, log2 fold-changes were determined by comparing array abundance in collected cell samples (collected on days 11, 13, or 15 of culture) to array abundance in the original plasmid DNA prep as a starting timepoint. The likelihood ratio test was executed by comparing full (~timepoint) and reduced (~1) models. These results were used to rank guide arrays based on how significantly attributed their count variation across samples was to the (reduced) timepoint variable. Significant gene-level enrichment and depletion were determined using robust rank aggregation⁹⁶ of guide-level Wald test results. Data was visualized using ggplot2 and gene-level hierarchical clustering was performed using pheatmap.

Cytokine secretion

On day 10 of culture, primary human T cells and GFP⁺ Nalm6-GD2 cells were spun down and resuspended in fresh prewarmed CM. 1 × 10⁵ GFP⁺ Nalm6-GD2 cells were seeded in clear flat-bottom 96-well plates and co-cultured with 1 × 10⁵ primary human HA-28ζ CAR T cells in a total volume of 200 μl CM. Triplicate wells were plated per condition. After 24 hours, the plate was spun down and the co-culture supernatant was collected for immediate analysis or frozen at −80 °C. Cytokine concentration was measured using IFNγ or IL-2 ELISA Max kits (Biolegend) following the manufacturer’s protocol.

Tumor killing and serial restimulation

In vitro tumor killing and serial restimulation assays were performed on an IncuCyte S3 (Sartorius). On day 10 of culture, primary human T cells and GFP⁺ Nalm6-GD2 cells were spun down and resuspended in fresh prewarmed CM. 5 × 10⁴ GFP⁺ Nalm6-GD2 cells were seeded in clear flat-bottom 96-well plates and co-cultured with primary human HA-28ζ CAR T cells at varying effector:target (E:T) ratios as indicated in figure legends in a total volume of 200 μl CM. Triplicate wells were plated per condition, and four images were acquired per well over a total of 48–72 hours. Total integrated GFP intensity per well was used as a metric to quantify live GFP⁺ Nalm6-GD2 cells. All values were normalized to the corresponding initial scan and plotted over time using the IncuCyte analysis software. For serial restimulation experiments, GFP⁺ Nalm6-GD2 and HA-28ζ CAR T cells were co-cultured as described above. After 48–72 hours of initial stimulation, the remaining cells were spun down, resuspended in fresh prewarmed CM, counted, and co-cultured with an additional 5 × 10⁴ GFP⁺ Nalm6-GD2 at varying E:T ratios as described above. This process was repeated twice for a total of three stimulations.

Detection of metabolites in cell culture media

On day 10 of culture, 5 × 10⁴ primary human HA-28ζ CAR T cells were seeded in flat-bottom 96-well plates in a total volume of 200 μl AIM V without phenol red (Gibco) supplemented with 100 U ml⁻¹ IL-2. ATP (Sigma-Aldrich) was spiked-in to a final concentration of 20 μM. Cells were incubated for one hour (unless otherwise noted) at 37 °C and 5% CO₂. Cells were then spun down and culture supernatant was collected for immediate analysis using ATP (Promega), AMP (Promega), and adenosine (Abcam) detection kits following the manufacturer’s protocol.

Mouse experiments

All mouse experiments were compliant with Stanford University Laboratory Animal Care (APLAC) protocols. NSG mice (JAX #005557) mice were engrafted with 1 × 10⁶ Luc⁺ Nalm6-GD2 tumor via intravenous injection in 200 μl DPBS. One week later, 2 × 10⁶ day 10 HA-28ζ CAR T cells were administered per mouse via intravenous injection in 200 μl DPBS. To measure tumor burden (BLI), mice were imaged twice weekly over 40 days using the IVIS Spectrum imaging system (PerkinElmer). Cages were randomized prior to T cell administration and researchers were blinded during T cell administration and BLI measurements. Bioluminescence was quantified using the Living Image software (PerkinElmer). Peripheral blood was collected by retro-orbital bleeding on day 21 post-treatment into EDTA-coated collection tubes and red blood cells were lysed with ACK buffer (Gibco). Samples were immediately washed twice with DPBS and resuspended in cold FACS buffer for downstream cell counting, immunostaining, and flow cytometry experiments.

QUANTIFICATION AND STATISTICAL ANALYSIS

Box and violin plots were created using ggplot2 and represent at least ~1 × 10⁴ cells per sample. Box plots indicate the median as well as first and third quartiles. Whiskers extend from the box to no more than 1.5× the distance between first and third quartiles. Unless otherwise noted, all bar graphs show the mean and standard deviation of replicates, and individual replicates are plotted as points.

Statistical tests for significance between groups were conducted in GraphPad Prism 9. Details on all statistical tests can be found in figure legends, including the specific test used, the numerical value and representation of n, and how significance was defined. All p-values were corrected/adjusted for multiple comparisons testing when applicable. In figures, all p-values were reported in GraphPad Prism style, with asterisks representing the following: *p < .05, **p < .01, ***p < .001, ****p < .0001. Exact p-values and further information on statistical tests are included in Table S2. Linear regression analyses were conducted in R version 4.1.2 or in GraphPad Prism 9.

Supplementary Material

1

Figure S1, related to Figure 1 Optimization of RfxCas13d expression and multiplexed gene targeting in primary human T cells.

(A) Schematic detailing the mechanism by which HA-28ζCAR T cells are driven to exhaustion. Antigen-independent clustering of HA-28ζ CAR leads to tonic signaling and eventual dysfunction after 10 days of ex vivo expansion.

(B) Overview of lentiviral constructs used for repression of exhaustion markers in HA-28ζ CAR T cells.

(C) Expression of LAG3, PD-1, and TIM3 relative to NT as measured by flow cytometry on day 10 across n = 3 donors from independent experiments. Colored bars: on-target guides (LAG3, yellow; PD-1, red; TIM3, blue); grey bars: off-target guides.

(D) Histogram depicting expression of RfxCas13d as measured by flow cytometry on day 10 from 1 representative donor. Black unfilled histogram: mock untransduced cells; magenta filled histogram: cells transduced with RfxCas13d:crNT.

(E) Violin and box plot overlays depicting RfxCas13d expression as measured by flow cytometry on day 5 from 1 representative donor. Black unfilled histogram: mock untransduced cells; magenta filled histogram: cells transduced with RfxCas13d constructs in various configurations.

(F) Model describing the context-dependent effects of RfxCas13d expression and activity on lentiviral titer during viral packaging in transfected LX293T cells.

(G) Violin plots depicting RfxCas13d (top, magenta trace) and HA-28ζ CAR (bottom, blue trace) expression of co-transduced primary human T cells as measured by flow cytometry on day 5 with increasing amounts of virus.

(H) Violin and box plot overlays depicting HA-28ζ CAR expression for single and double guides on day 5 as measured by flow cytometry from 1 representative donor. Blue histograms: transduced cells; grey histogram: mock untransduced cells.

(I) Violin and box plot overlays depicting HA-28ζ CAR expression for triple guides on day 5 as measured by flow cytometry from 1 representative donor. Blue histograms: transduced cells; grey histogram: mock untransduced cells.

(J) Violin and box plot overlays depicting surface LAG3, PD-1, and TIM3 expression for triple guides as measured by flow cytometry on day 10 from 1 representative donor (see Figure 1H). On-target guides are colored (LAG3, yellow; PD-1, red; TIM3, blue); controls are grey. Dashed lines indicate MFI values of either mock untransduced cells or exhausted non-targeting cells used for normalization. Labeled numbers indicate normalized values for each condition.

(K, L) Gating strategy used in Figure 1I to determine LAG3^+/−PD-1^+/−TIM3^+/− populations in unsorted MEGA HA-28ζ CAR T cells expressing (K) non-targeting control or (L) triple guide array. Data are measured by flow cytometry on day 10 from 1 representative donor. From left to right: live cell (lymphocyte) gate, singlet gate, RfxCas13d⁺ gate, LAG3^+/−PD-1^+/− gate, TIM3^+/− gate. Each histogram shown in the TIM3^+/− gate represents cells from the same quadrant in the LAG3^+/−PD-1^+/− gate, e.g., the top-right histogram depicts TIM3 expression on LAG3⁺PD-1⁺ cells.

2

Figure S2, related to Figure 1 Evaluation of RfxCas13d collateral activity and immunogenicity.

(A) Schematic detailing the matched flow cytometry and bulk RNA-seq experiment to evaluate collateral effects resulting from B2M knockdown in primary human T cells from n = 2 healthy donors.

(B) Dot plots representing expression of RfxCas13d and either on-target (B2M) or off-target (CD46, CD3) protein expression on day 5 of culture as measured by flow cytometry from 1 representative donor. Left column: targeting guide; right column: non-targeting control.

(C) mCherry MFI of day 5 MEGA T cells expressing a B2M-targeting guide (light blue bar) or a non-targeting (NT) control (grey bar) as measured by flow cytometry across n = 2 donors. n.s., not significant (unpaired t-test). Error bars are ± s.d.

(D) Cell viability of day 5 MEGA T cells expressing a B2M-targeting guide (light blue bar) or a non-targeting (NT) control (grey bar) as measured by trypan blue exclusion across n = 2 donors. n.s., not significant (unpaired t-test). Error bars are ± s.d.

(E) Cell yield after FACS of day 5 MEGA T cells expressing a B2M-targeting guide (light blue bar) or a non-targeting (NT) control (grey bar) across n = 2 donors. n.s., not significant (unpaired t-test). Error bars are ± s.d.

(F) Total RNA yield from sorted day 5 MEGA T cells expressing a B2M-targeting guide (light blue bar) or a non-targeting (NT) control (grey bar) across n = 2 donors. n.s., not significant (unpaired t-test). Error bars are ± s.d.

(G) PCA plot of bulk RNA-seq data generated from sorted day 5 MEGA T cells expressing a B2M-targeting guide (light blue dots) or a non-targeting (NT) control (grey dots) across n = 2 donors. The first two principal components are shown.

(H) Transcriptome-wide quantification of bulk RNA-seq data from day 5 MEGA T cells. Transcript abundances for B2M-targeting cells (y-axis: B2M) are plotted against those of non-targeting cells (x-axis: NT), averaged over n = 2 donors. Each dot represents a single gene; gene-level expression was determined by aggregating transcripts per million (TPM) values and p-values across all detected transcripts for each gene. Differentially expressed genes with adjusted p-value < 0.1 and absolute log2 fold-change > 1 are highlighted. Red: upregulated genes; dark blue: downregulated genes; light blue: targeted gene; orange: mitochondrial genes. Dot sizes scale with statistical significance (-log10 adjusted p-value). The green line indicates where gene expression is equal across both conditions (y = x).

(I) Immunogenicity scores (IEDB, Immune Epitope Database) for commonly-used synthetic biology tools. For each of the proteins listed, IEDB scores were predicted by sliding a 9-amino acid window over the entire protein sequence from start to end. The total score represents the sum of IEDB scores for all 9-mer peptides generated from a protein in this manner.

3

Figure S3, related to Figure 2 Combinatorial CRISPR screening in HA-28ζ CAR T cells identifies paired regulators of T cell proliferation.

(A) Plot representing the distribution of guide arrays in the plasmid DNA prep of the assembled custom library as determined by NGS. Guide array counts are ranked in order of abundance. Dashed lines represent 90^th and 10^th percentiles. Blue labelled numbers indicate common metrics to quantify library coverage and bias.

(B) Histograms depicting RfxCas13d (left) and HA-28ζ CAR (right) expression over the duration of the screen. Colored histograms: cells for screening (RfxCas13d: magenta; CAR: blue); unfilled histograms: mock untransduced cells.

(C) Plots representing the distribution of guide arrays over the duration of the screen across two replicates.

(D) Plots depicting guide array abundance correlation between two replicates. Left: early timepoint; right: late timepoint.

(E) 2D heatmap of paired-effect scores for all guide arrays. For any given guide array targeting genes A and B in that order, the paired-effect score = abs(A+B l2fc) – max[(abs(A+NT l2fc), abs(NT+B l2fc)]. Log2 fold-change values were calculated between early and late timepoints as in Figure 2B.

(F) Aggregated gene-level differential abundance analysis. Volcano plot depicts the average differential abundance of gene pairs (each consisting of 9 guide arrays) between early (plasmid DNA) and late (day 13) timepoints for n = 2 replicates. Blue dots: significantly depleted pairs; red dots: significantly enriched pairs; open black dot: non-targeting control. p values from robust rank aggregation as implemented in MAGeCK.

(G) 2D heatmap of gene pair enrichment between early and late timepoints. Genes are ordered by hierarchical clustering as implemented in pheatmap.

(H) Normalized cell counts of FACS-sorted RfxCas13d⁺ HA-28ζ CAR T cells over 15 days in culture relative to a non-targeting control, n = 3–4 donors from independent experiments (see Figure 2F). Each panel corresponds to a different guide array. Red lines: enriched arrays; blue lines: depleted arrays; grey line: non-targeting control. Top row: CD8⁺ T cells; bottom row: bulk T cells. Error bars are ± s.e.m.

(I) Fold-change expansion of FACS-sorted RfxCas13d⁺ HA-28ζ CAR T cells compared to a non-targeting control over 15 days of culture from n = 3–4 donors from independent experiments. Significantly enriched guide arrays in red; significantly depleted guide arrays in blue; non-targeting (NT) control guide in black. Dashed line represents expansion of non-targeting control. p < .0001, ordinary one-way ANOVA. Error bars are ± s.e.m.

(J) Violin and box plot overlays depicting HA-28ζ CAR expression (top) and RfxCas13d expression (bottom) for validation arrays on day 5 (before FACS) as measured by flow cytometry from 1 representative donor. Dashed line represents gate for RfxCas13d⁺ cells. Red histograms: enriched arrays; blue histograms: depleted arrays; grey histograms: controls.

(K) Aggregate % RfxCas13d⁺ as detailed in (J) across n = 3 donors from independent experiments. Red bars: enriched arrays; blue bars: depleted arrays; grey bar: non-targeting. **p < .01, ****p < .0001, ordinary one-way ANOVA with Dunnett’s multiple comparisons test. Error bars are ± s.d.

4

Figure S4, related to Figure 2 Paired transcriptomic perturbations broadly enhance the anti-tumor activity of dysfunctional MEGA CAR T cells.

(A) Overview of experimental workflow to generate dysfunctional MEGA CAR T cells expressing validation arrays.

(B) Schematic detailing experimental conditions for cytokine secretion assays.

(C, D) Secretion of IFNγ (C) or IL-2 (D) after 24 hours of culture with (pink dots, stimulated) or without (black dots, baseline) antigen-positive tumor cells at 1:1 E:T. Data are mean values of n = 3 replicate wells from 1 representative donor. *p < .05, **p < .01, ***p < .001, ****p < .0001, ordinary one-way ANOVA with Dunnett’s multiple comparisons test. Error bars are ± s.d.

(E) Kinetics of tumor killing (1:1 E:T) as measured using Incucyte live-cell imaging, as in Figure 2G. Data are mean values of n = 3 replicate wells. Each row corresponds to additional donors from independent experiments. Red traces: enriched guide arrays; blue traces: depleted guide arrays; black traces: non-targeting control. p < .0001, repeated measures one-way ANOVA with Dunnett’s multiple comparisons test. Shaded regions are ± s.e.m.

(F) Kinetics of tumor killing at a lower E:T ratio (1:5) as measured using Incucyte live-cell imaging. Data are mean values of n = 3 replicate wells. Each row corresponds to a new donor from an independent experiment. Red traces: enriched guide arrays; black traces: non-targeting control. p < .0001, repeated measures one-way ANOVA with Dunnett’s multiple comparisons test. Shaded regions are ± s.e.m.

(G) Incucyte images from 0 hr and 48 hr timepoints after three rounds of tumor stimulation as in Figure 2G. Day 10 MEGA HA-28ζ CAR T cells expressing a non-targeting guide (top row) or CBLB + FAS guide array (bottom row) were co-cultured with Nalm6-GD2 tumor cells. Green signal: Nalm6-GD2 tumor cells (GFP intensity); red signal: MEGA HA-28ζ CAR T cells (mCherry intensity).

(H) Schematics for hypothesized mechanism of enhanced anti-tumor activity in dual CBLB + FAS knockdown CAR T cells (bottom) compared to a non-targeting (NT) control (top).

(I) Kinetics of tumor killing (1:1 E:T) as measured using Incucyte live-cell imaging. Data are mean values of n = 3 replicate wells. Red traces: CBLB+FAS array; blue traces: CBLB guide only; orange traces: FAS guide only; black traces: NT control; grey trace: mock untransduced T cells. p-values from repeated measures two-way ANOVA (column-factor). Shaded regions are ± s.e.m.

(J) Tumor intensity at 48 hour endpoint after second tumor stimulation. Data are mean values of n = 3 replicate wells. **p < .01, ordinary one-way ANOVA with Dunnett’s multiple comparisons test. Error bars are ± s.d.

(K, L) Secretion of IFNγ (K) or IL-2 (L) after 24 hours of culture with (pink dots, stimulated) or without (black dots, baseline) antigen-positive tumor cells at 1:1 E:T. Data are mean values of n = 3 replicate wells. *p < .05, ***p < .001, ****p < .0001, ordinary one-way ANOVA with Dunnett’s multiple comparisons test. Error bars are ± s.d.

5

Figure S5, related to Figure 3 Regulation of RfxCas13d activity in primary human T cells by trimethoprim.

(A) Density plots representing expression of RfxCas13d and LAG3/PD-1/TIM3 in day 10 MEGA HA-28ζ CAR T cells as measured by flow cytometry from 1 representative donor. Left column: targeting guides; right column: non-targeting control.

(B) Violin and box plot overlays depicting intracellular RfxCas13d expression on day 5 as measured by FLAG tag staining and flow cytometry from 1 representative donor. Blue histograms: high expression conditions; grey histogram: low expression conditions.

(C) Violin and box plot overlays depicting surface CD46 as measured by flow cytometry across 72 hours from 1 representative donor (see Figure 3C). Top: TMP addition; bottom: TMP removal. Solid filled histograms: TMP addition (yellow) or removal (dark blue); dotted unfilled histograms: non-targeting control.

(D, E, F) Bar plots depicting CD46 expression (D), mCherry expression (E), or cell viability (F) with varying concentrations of TMP. Colored bars: CD46 targeting guide (CD46 expression: purple; mCherry expression: magenta; cell viability: blue); grey bars: non-targeting guide. Numbers labelled above bars indicate fold-change over NT. Data are mean of n = 3 replicate wells from 1 representative donor. *p < .05, **p < .001, ****p < .0001, multiple unpaired t-tests with FDR correction. Error bars are ± s.d.

(G) Schematic illustrating transfer curves of (top) conventional CAR T cells or (bottom) MEGA CAR T cells with a TMP-regulatable amplitude limiter.

(H) Violin and box plot overlays depicting intracellular ZAP70 (top) or Lck (bottom) protein expression in day 10 MEGA HA-28ζ CAR T cells as measured by flow cytometry. Numbered labels represent three different crRNA spacer sequences that bind to different regions of the target transcript. Green histograms: targeting guides; grey histograms: non-targeting (NT) control.

(I) Schematic illustrating co-culture experiments between day 10 MEGA CD19-BBζ CAR T cells targeting CD19⁺ Nalm6 tumor cells, with and without TMP.

(J) Secretion of IL-2 by MEGA CD19-BBζ CAR T cells after 24 hours of stimulation with antigen-positive tumor in the presence (purple bar) or absence (grey bar) of 1 μM TMP. Data are mean values of n = 3 replicate wells. **p < .01, unpaired t-test. Error bars are ± s.d.

(K) CD25 (IL2RA) expression on day 10 MEGA CD19-BBζ CAR T cells in the presence (purple bars) or absence (grey bars) of 1 μM TMP as measured by flow cytometry. Left panel: unstimulated cells (baseline); right panel: stimulated cells (with antigen-positive tumor). Data are mean values of n = 3 replicate wells. **p < .01, ****p < .0001, unpaired t-test. Error bars are ± s.d.

(L) Schematic illustrating co-culture experiments between day 10 MEGA ROR1–28ζ CAR T cells targeting ROR1⁺ Nalm6 tumor cells, with and without TMP.

(M) Secretion of IL-2 by MEGA ROR1–28ζ CAR T cells after 24 hours of stimulation with antigen-positive tumor in the presence (purple bar) or absence (grey bar) of 1 μM TMP. Data are mean values of n = 3 replicate wells.**p < .01, unpaired t-test. Error bars are ± s.d.

(N) CD25 (IL2RA) expression on day 10 MEGA ROR1–28ζ CAR T cells in the presence (purple bars) or absence (grey bars) of 1 μM TMP as measured by flow cytometry. Left panel: unstimulated cells (baseline); right panel: stimulated cells (with antigen-positive tumor). Data are mean values of n = 3 replicate wells. **p < .01, ****p <.0001, unpaired t-test. Error bars are ± s.d.

6

Figure S6, related to Figure 5 Functional characterization of purinergic disruption in MEGA HA-28ζ CAR T cells.

(A) Heatmap depicting mRNA transcript levels of ENTPD1, NT5E, ADORA2A (A2A), and ADORA2B (A2B) relative to NT as measured by RT-qPCR on day 10. Labeled numbers within boxes represent mean values of n = 3 technical replicates. Three spacer sequences were screened for each gene. Two guide arrays were screened to determine the optimal spacer order: array 1 = ENTPD1, NT5E, ADORA2A, ADORA2B; array 2 = ADORA2B, ADORA2A, NT5E, ENTPD1. Pink arrow indicates the optimal array (array 2 = “PURI” array).

(B) Violin and box plot overlays depicting CAR expression of MEGA HA-28ζ CAR T cells across targeting (PURI) and non-targeting groups on day 5 as measured by flow cytometry from 1 representative donor. Blue histograms: transduced cells; grey histogram: mock untransduced cells.

(C, D) Secretion of IFNγ (C) or IL-2 (D) after 24 hours of culture with (stimulated) or without (baseline) Nalm6-GD2 tumor cells at 1:1 E:T. Data are mean values of n = 3 replicate wells from an additional donor (see Figure 5K–L). Purple bars: PURI array; grey bars: non-targeting control. **p < .01, ****p < .0001, two-way ANOVA with Bonferroni’s multiple comparisons test. Error bars are ± s.d.

(E) Kinetics of tumor killing (top) and T cell proliferation (bottom) as measured using Incucyte live-cell imaging at a 1:2 E:T ratio from an additional donor (see Figure 5M–N). Data are mean values of n = 3 replicate wells from 1 representative donor. Purple trace: PURI array; grey trace: NT control. **p < .01, repeated measures two-way ANOVA, column-factor. Shaded regions are ± s.e.m.

(F, G) Tumor intensity (F) and T cell intensity (G) after 48 hour co-culture of day 10 MEGA HA-28ζ CAR T cells with Nalm6-GD2 tumor cells at 1:2 E:T. Dark purple bars: PURI array; light purple bars: single targeting guides; grey bars: non-targeting guide. Data are mean values of n = 3 replicate wells. *p < .05, **p < .01, ****p < .0001, ordinary one-way ANOVA with Dunnett’s multiple comparisons test. Error bars are ± s.d.

(H) Correlation plot of Incucyte data from (F) and (G). Tumor intensity (y-axis) is plotted against T cell intensity (x-axis) for all samples. Dark purple dots: PURI array; light purple dots: single targeting guides; white dots: non-targeting guide. Blue line represents best fit from linear regression.

(I, J) Secretion of IFNγ (I) or IL-2 (J) after 24 hour co-culture with Nalm6-GD2 tumor cells at 1:1 E:T. Data are mean values of n = 3 replicate wells. Dark purple bars: PURI array; light purple bars: single targeting guides; grey bars: non-targeting guide. *p < .05, **p < .01, ****p < .0001, ordinary one-way ANOVA with Dunnett’s multiple comparisons test. Error bars are ± s.d.

7

Figure S7, related to Figures 6 and 7 Functional characterization of glycolytic disruption in MEGA HA-28ζ CAR T cells.

(A) Heatmap depicting mRNA transcript levels of AKT1, AKT2, HK1, and HK2 relative to NT as measured by RT-qPCR on day 10. Labeled numbers within boxes represent mean values of n = 3 technical replicates. Three spacer sequences were screened for each gene. Two guide arrays were screened to determine the optimal spacer order: array 1 = AKT1, AKT2, HK1, HK2; array 2 = HK1, HK2, AKT1, AKT2. Pink arrow indicates the optimal array (array 2 = “GLY” array).

(B) Violin and box plot overlays depicting CAR expression of MEGA HA-28ζ CAR T cells across targeting (GLY) and non-targeting groups on day 5 as measured by flow cytometry for n = 3 donors across independent experiments.

(C) Density plots depicting CAR expression and CD69 expression on resting (left) or tumor stimulated (right) day 10 MEGA HA-28ζ CAR T cells by flow cytometry. Top row: NT control; bottom row: GLY array. Labeled numbers indicate the proportion of CD69⁺ cells.

(D, E) Secretion of IFNγ (D) or IL-2 (E) by day 10 MEGA HA-28ζ CAR T cells after 24 hours of culture with (stimulated) or without (baseline) Nalm6-GD2 tumor cells at 1:1 E:T. Data are mean values of n = 3 replicate wells from additional donors (see Figure 7B–C). Dark green bars: GLY array; yellow bars: NT control. **p < .01, ***p < .001, ****p < .0001, two-way ANOVA with Bonferroni’s multiple comparisons test. Error bars are ± s.d.

(F) Euler diagrams depicting the proportion of day 10 MEGA HA-28ζ CAR T cells expressing IL-2, IFNγ, and/or TNFɑ as determined by intracellular cytokine staining and flow cytometry. Top row: baseline expression; bottom row: expression after 6 hr stimulation with Nalm6-GD2. Left column: NT control; right column: GLY array.

(G, H) Secretion of IFNγ (G) or IL-2 (H) by day 10 Cas9-edited HA-28ζ CAR T cells after 24 hours of stimulation with Nalm6-GD2 tumor cells at 1:1 E:T. Data are mean values of n = 3 replicate wells. Dark blue bars: 4KO; light blue bars: AAVS1 control. ****p < .0001, unpaired t-test. Error bars are ± s.d.

(I) Kinetics of tumor killing in a serial stimulation challenge against Nalm6-GD2 as measured using Incucyte live-cell imaging at a 1:1 E:T ratio from an additional donor (see Figure 7D). Data are mean values of n = 3 replicate wells from 1 representative donor. Dark green trace: GLY array; yellow trace: NT control. **p < .01, repeated measures two-way ANOVA, column-factor. Shaded regions are ± s.e.m.

8

Table S1 Key oligonucleotide sequences, related to STAR Methods.

9

Table S2 Detailed reporting of statistical analyses, related to STAR Methods.

10

Table S3 Summarized DE analysis and TPMs from bulk RNA-seq data, related to Figure 1.

11

Table S4 Guide array counts (screen readout), related to Figure 2.

12

Table S5 Summarized DE analysis and TPMs from bulk RNA-seq data, related to Figure 6.

13

Table S6 Summarized GSEA analysis from bulk RNA-seq data, related to Figure 6.

14

Table S7 Summarized DE analysis and TPMs from bulk RNA-seq data, related to Figures 1 and S2.

Highlights.

• MEGA allows massively-multiplexed RNA knockdown in primary human T cells

• Combinatorial knockdown screen identifies paired regulators of CAR T cell function

• Tunable and reversible multi-gene regulation with an FDA-approved drug

• Metabolic engineering enhances CAR T cell fitness and anti-tumor activity in vivo