Using microRNA to control multiple genes in CAR T cell therapy

AsPC-1 cell line was obtained from American Type Culture Collection (ATCC) and was cultured in R20 culture media [RPMI 1640 (Gibco) + 20% FBS (Avantor Seradigm, 97068-086), 1% penicillin/streptomycin (50 IU/mL; Gibco), 1% GlutaMAX™ (Gibco), 2% 1M Hepes (Gibco)] at 37°C in 5% CO₂. AsPC-1-click beetle green (CBG) and green fluorescent protein (GFP) expressing cells (AsPC-1 CBG GFP cells) were generated by lentiviral transduction for cell killing assays and in vivo studies. The cell lines were selected by sorting on FACS Aria Sorter (BD Biosciences) and monitored for growth and stable gene expression for 8 weeks. All cell lines were authenticated by the University of Arizona Genetics Core and were tested for the presence of mycoplasma contamination (MycoAlert^® Mycoplasma Detection Kit, Lonza) every 6 months and before being used in vivo.

CRISPR single guide RNA (sgRNA) targeting CD3ζ were designed using multiple softwares: Benchling, CRISPick (Broad Institute), IDT, ChopChop. The most common sgRNAs were selected and chemically modified and synthesized by Integrated DNA Technologies (IDT) (). 1

One of five sgRNA targeting CD3ζ (sgRNA_4) was selected for further experiments after validation for highest deletion efficiency at the DNA and protein levels. β2M sgRNA sequence was taken from the product designed for NCT05037669, while the TRAC sgRNA sequence was taken from Ren et al. (16), both in Supplementary Table 1. A lentiviral vector for the expression of the second generation anti-CD19 CAR (CAR19) composed of the FMC63 scFv and CD8α hinge and transmembrane domains fused to 4-1BB and CD3-ζ cytoplasmic signaling domains expressed under control of the EF-1α promoter (17) was also employed for CD3ζ and/or β2M deletion evaluation on CAR T cells.

miRNA target sequences for CD3ζ and β2M (Supplementary Table 2) were previously designed using the methodology described in Myburgh et al. (12) as well as online resources for target sequence design based on sequence prioritization, and individual scoring of each target sequence to identify the potentially best performing candidates, while mitigating the risk of potential off-target gene silencing. Prioritized target sequences were synthesized within a mirGE backbone (12) by Thermo Fisher Scientific and subsequently cloned into a pENTR plasmid using basic restriction enzymes. The final lentivector plasmid was generated by an LR Clonase II (Invitrogen, Carlsbad, CA)-mediated recombination of a pENTR plasmid containing an elongation factor 1 short promoter (pENTR-L4-EFs-L1R), a pENTR plasmid containing the mirGE expression cassette (pENTR-L1-spacer-mirGE-L2), and a pCWX-R4-DEST-R2-PGK-mCherry lentivector destination plasmid. Successful cloning of all constructs was confirmed via restriction enzyme digestion pattern and DNA sequencing. Silencing lentiviral vectors were then used to transduce T cells derived from healthy donors. Transduction efficiency (based on mCherry reporter gene expression) and target silencing were evaluated by flow cytometry.

A bidirectional dual-promoter expression system was selected for optimal simultaneous gene silencing and CAR expression. Optimized miRNA expression constructs for silencing of CD3ζ and/or β2M, driven by the EFs promoter, were cloned into the pTRPG vector, which encodes M5CAR under the control of the human phosphoglycerate kinase (hPGK) promoter, in a bidirectional promoter configuration. The second generation M5CAR is composed of a human Msln-binding (M5) scFv and CD8α hinge and transmembrane domains fused to 4-1BB and CD3ζ cytoplasmic signaling domains expressed under control of the PGK promoter. The M5CAR expression construct was combined with a scrambled miRNA target sequence in scrambled M5 CAR T cells, miRNAs for single silencing of CD3ζ and β2M transcripts in S CD3ζ M5 CAR T cells and sβ2M M5 CAR T cells, respectively, and miRNAs for dual silencing of CD3ζ and β2M in S CD3ζ&β2M M5CAR T cells.

Replication defective lentiviral vectors were generated by transient transfection of HEK293T cells (ATCC ACS-4500) using Lipofectamine 2000 (ThermoFisher Scientific, Cat#11668500). Approximately 10 x 10⁶ cells were plated in T150 culture vessels in R10 culture media (RPMI 1640 + 10% FBS + 1% penicillin/streptomycin (50 IU/mL), 1% GlutaMAX™, 2% 1M Hepes) and incubated overnight (o/n) at 37°C. 18–24 hours later, cells were transfected using a combination of Lipofectamine 2000 (96mL, Invitrogen), pTRP gag/pol (Lot# RR24JUN21D-2) (18µg), pTRP RSV-Rev (Lot# RR24JUN21B) (18µg), pTRP Cocal (Lot# RR24JUN21A-2) (7µg) packaging plasmids and 15µg of expression plasmid (anti-Msln M5bbz scFv cloned in pTRPG). Lipofectamine and plasmid DNA were diluted in 3 mL Opti-MEM media (Gibco, Life Technologies) prior to transferring into lentiviral production flasks. At both 24 and 48 hours following transfection, culture media was isolated and concentrated using high-speed ultracentrifugation (8000g o/n and then 25,000g for 2.5 hours). To generate the lentiviral stocks, the resulting concentrated lentivirus batches were resuspended in cold R10 media and stored at -80C.

CAR T cells were manufactured from frozen human healthy donor (ND) bulk T cells or peripheral blood mononuclear cells (PBMCs). In both cases, PBMCs were acquired through leukapheresis from healthy volunteers by the University of Pennsylvania Human Immunology Core (HIC), with the approval of the University's Institutional Review Board. Study participants provided written informed consent, in compliance with the Declaration of Helsinki, the International Conference on Harmonization's Good Clinical Practice guidelines, and the United States Common Rule. Lymphoprep™ (Stemcell Technologies) followed by the RosetteSep™ Human T Cell Enrichment Cocktail (Stemcell Technologies) were used by the HIC for PBMC and bulk T cell isolation, respectively. From PBMCs, CD4 and CD8 T cells were separately isolated by the HIC using CD4 (Catalog# 15062; Stemcell Technologies) and CD8 (Catalog# 15063, Stemcell Technologies) selection kits, then pooled. Cells were cultured in T cell media composed by the OpTmizer™ CTS™ T-Cell Expansion SFM (Thermo Fisher), 5% human AB serum (Valley Biomedical), 1% penicillin/streptomycin (50 IU/mL), and 1% GlutaMAX™.

Cells undergoing miRNA-mediated gene silencing were activated using CTS™ Dynabeads™ CD3/CD28 (Catalog# 40203D; Thermo Fisher) at a 3:1 bead-to-cell ratio and resuspended at 1 x 10⁶ cell/mL in T cell media (day 0). Single lentiviral vectors coding for the M5CAR (3) and CD3ζ and/or β2M-targeting miRNAs were used for transduction approximately 24 hours post bead stimulation (day 1) at an MOI of 3. For each silencing condition, 2 x 10⁶ cells were transduced, as in the control M5CAR T cell group.

Cells undergoing the CRISPR/Cas9-genome engineering were incubated o/n at 5 x 10⁶ cells/mL with 10 ng/mL IL-7 and IL-15 (Catalog# 200–07 and 200-15, respectively; Peprotech) (day -1). CRISPR/Cas9-gene deletion, T cell activation, transduction, expansion, and KO validation were performed following an optimized protocol previously described (18). Approximately 8–12 x 10⁶ cells were used as the starting amount for each condition. Briefly, primary T cells were nucleofected (day 0) using the Lonza 4D-Nucleofector Core/X Unit and the P3 Primary cell 4D-nucleofactor X Kit (Lonza). For Cas9 and sgRNA delivery, the ribonucleoprotein (RNP) complex was formed by incubating 10mg of Spy Fi Cas9 (Catalog# 9214; Aldeveron) with 5mg of sgRNA (IDT) for 10 minutes at room temperature (RT). Cells were resuspended at a concentration of 8–12 x 10⁶ cells/100µL in the specified kit buffer (P3 Solution with Supplement). The RNP complex, 100µL of resuspended cells, and 4.2 µL of 4 µM IDT Electroporation Enhancer (a non-homologous ssDNA oligonucleotide) were combined and nucleofected in a cuvette. Pulse code EH115 was used for primary T cells. After nucleofection (NF), the cells were incubated in T cell media at 37°C in 12-well plates. Four hours later, nucleofected cells were then activated using CTS™ Dynabeads™ CD3/CD28 at a 3:1 bead-to-cell ratio and resuspended in T cell media at a concentration of 2 x 10⁶ cells/mL. 24 hours later, for each NF condition, cells were counted and the M5CAR lentiviral vector was used for transduction at an MOI of 3 (day 1). For each KO condition, all counted cells were transduced.

Over both protocols, beads were removed on day 5 after stimulation. Cells were then monitored daily using the Multisizer 4 Coulter Counter (Beckman Coulter) until growth kinetics and cell size demonstrated that they had rested from stimulation. On average, T cells were grown for 9–10 days in the presence of 10 ng/mL of IL-7 and IL-15 and maintained at 8 x 10⁵ cells/mL prior to cryopreservation. A proportion of cells in groups with either CD3ζ S or KO, or dual CD3ζ and β2M S or KO, were depleted of TCRαβ before cryopreservation using the EasySep™ Human TCR Alpha/Beta Depletion Kit (Stemcell Technologies). Fold expansion was expressed as the ratio of the final number of cells to the initial number of cells, while population doublings (PDs) were calculated as the log₂ of this ratio. The percentage reduction in T cell numbers after NF was calculated as the difference between pre- and post-NF cell counts, divided by the pre-NF count and multiplied by 100. At the end of expansion, cells were collected, and flow cytometry analysis was performed for evaluation of M5CAR+ T cell percentage, CD3ζ and/or β2M silencing or KO on T cells, memory phenotype by CD45RO and CCR7, and activation marker expression (4-1BB, TIGIT, PD-1, TIM-3, LAG-3, and OX40). Prior to all experiments, T cells were thawed and rested at 2 x 10⁶ cells/mL in T cell media at 37°C for 16 hours.

From each T cell culture engineered with CRISPR/Cas9, 3 x 10^6–5 x 10⁶ cells were flash frozen as dry cell pellets at the end of expansion. At time of DNA extraction, cell pellets were thawed and resuspended in 200µl Phosphate Buffer Saline (PBS). Genomic DNA from nucleofected cells was isolated using the DNeasy Blood & Tissue Kit (Catalog# 69504; Qiagen) and 200 ng DNA was PCR amplified using AccuPrime™ Pfx SuperMix (Catalog# 12344040; Thermo Fisher) and 10µM forward and reverse primers flanking the region of interest. Primers were designed such that the amplicon was at a target size of 600-700bp. PCR products were gel purified, Sanger sequenced, and trace files were analyzed using both TIDE (Tracking of Indels by Decomposition; Netherlands Cancer Institute) (19) and ICE (Inference of CRISPR Edits; Synthego) (20) analyses to detect deletion efficiency at the genomic level. PCR primers and sequencing primers were designed to detect each target locus. The following primers for target sequence amplification ('PCR') and Sanger sequencing ('Seq') were used for validation of deletion efficiency: CD3ζ.PCR.F 5' CCATTGCCCCAGGTTCTTTG 3', CD3ζ.PCR.R 5' AAGCCCCTTTGTCACCAGTA 3', CD3ζ.Seq.F 5' CCACATCTGCCGTTGGTG 3', CD3ζ.Seq.R 5' TTCCCTGGGACACTCTGATG 3'; β2M.PCR.F 5' TCCAGCCTGAAGTCCTAGAATG 3', β2M.PCR.R 5' TTATCGACGCCCTAAACTTTGT 3', β2M.Seq.F 5' ACAGACAGCAAACTCACCCA 3', and β2M.Seq.R 5' CCAAAGGTCTCCCCTGCTC 3'.

AsPC-1 CBG GFP target cells were seeded in white 96 well plates at 4000 cells/well. After 24 hours, M5CAR+ T cells were added at different effector:target (E:T) ratios (1:3, 1:1, 3:1, 10:1 E:T ratios). M5CAR T cell groups were matched for total T-cell dose added in co-culture by not transduced (NTD) normalization. Target cell survival was measured using bioluminescent quantification. After 24, 48 hours, and 4 days, D-luciferin potassium salt (Perkin-Elmer) was added to cell cultures at a final concentration of 150μg/mL and incubated at 37°C for 5 minutes. Bioluminescent signal was detected using the BioTek Synergy H4 plate reader, and the signal was analyzed using the BioTek Gen5 software. Tumor cells containing the medium alone or treated with cell lysis buffer (Qiagen) were used as controls, representing 100% and 0% cell viability, respectively. Triplicate wells were averaged, and cell viability was calculated with the following formula: % cell viability = 100 × (experimental bioluminescence − 0% bioluminescence)/(100% bioluminescence − 0% bioluminescence).

Using the same plate settings as for the bioluminescence-based co-culture assays, AsPC-1 CBG GFP viability was monitored in real-time using the IncuCyte^® SX5 (Sartorius) over 6 days. The amount of live GFP+ tumor cells for each condition was quantified by the IncuCyte SX5 live imaging system. For data analysis, the fluorescence integrated intensity per well were normalized to the time point of CAR T cell addition (0 hours) and triplicate wells were averaged. On day 6, cells were collected, and flow cytometry analysis was performed for evaluation of M5CAR+ T cell absolute number and percentage over total T cells, CD3ζ and/or β2M silencing or KO on T cells, memory phenotype by CD45RO and CCR7, and activation markers (4-1BB, TIGIT, PD-1, TIM-3, LAG-3, and OX40). CountBright™ Absolute Counting Beads (Catalog# C36950; ThermoFisher) were used as an internal standard according to the manufacturer's instructions to calculate absolute cell counts in cell suspensions.

Primary non-matched healthy donor NK cells were enriched by the HIC from PBMCs using the RosetteSep™ Human NK Cell Enrichment Cocktail (Stemcell Technologies). Purified NK cells were cultured o/n in T cell media supplemented with IL-2 cytokine. TCRαβ-depleted M5CAR T cells were stained with 1 μM CalceinAM (BioLegend) and used as target cells. These NK cells were then mixed with the M5CAR T cells at NK:T ratios of 12:1 and 6:1. After 4 hours of co-culture, cells were harvested and stained with 3 µM propidium iodide (PI; ThermoFisher). The percentage of viable target cells (PI-negative/CalceinAM^high-positive) was assessed via flow cytometry. For baseline determination, target cells stained with CalceinAM and maintained in culture for 4 hours were mixed with NK cells at a 1:1 NK:T ratio just before final staining and subsequent flow cytometry determination of the percentage of viable target cells (baseline). This baseline value was used to normalize the co-culture values. The data were expressed as the ratio of the percentage of viable target cells in the co-culture condition to the percentage of viable target cells at baseline.

Donor TCRαβ-depleted M5CAR T cells (stimulator) were co-cultured with non-matched healthy donor PBMCs (responders) at a 1:1 E:T ratio for six days. NK cells were kept or removed from PBMCs using human CD56 MicroBeads and LD columns (Miltenyi Biotech) or the EasySep™ Human NK Cell Isolation Kit (Stemcell Technologies), as per manufacturer's instruction. Stimulator TCRαβ-depleted M5CAR T cells were irradiated at 2500 cGy. T cell media was added after three days. PBMC proliferation was determined by tritiated thymidine (³H-TdR, 100µCi/mL, PerkinElmer) incorporation for the last 8 hours of culture. Data were analyzed by the Microbeta2 Lumijet microplate reader (PerkinElmer).

For flow cytometry and sorting assays cells were stained in fluorescence-activated cell sorting (FACS) buffer consisting of PBS (Gibco) and 2% fetal bovine serum (ThermoFisher). In some assays, CountBright™ Absolute Counting Beads (ThermoFisher) were used to calculate absolute cell counts. A specific cell population was considered detectable and reliably quantifiable when at least 30 events were recorded under the established experimental acquisition conditions. Antibodies specific for human CD45 (Catalog# 368524, Clone 2D1), CD4 (317414, OKT4), CD8 (344721, SK1), TCRαβ (306714, IP26), CD3ϵ (300400/300429 UCHT1), MHC-I (311436, W6/32), β2M (316316/316304, 2M2), CCR7 (353244/353244, G043H7), CD45RO (304231, UCHLI), OX40 (350021, Ber-ACT35), 4-1BB (309827, 4B4-1), PD-1 (329927, EH12.2H7), TIM-3 (119721, RMT3-23), and TIGIT (372735, A15153G), were purchased from BioLegend. The anti-human CD8 antibody was obtained from BD Pharmigen (Catalog# 560179) The anti-human LAG-3 antibody (Catalog# 61-2239-42) was purchased from eBioscience. M5CAR was detected using the Biotin-SP (long spacer) AffiniPure™ F(ab')₂ Fragment Goat Anti-Human IgG, F(ab')₂ fragment specific (Catalog# 109-066-006; Jackson ImmunoResearch), followed by a PE-streptavidin (Catalog# 554061, BD Biosciences). Cell viability was measured using LIVE/DEAD™ Fixable Violet Dead Cell Stain Kit (Catalog# L34964; Thermo Fisher) according to the manufacturer's instructions. Samples were acquired on an LSRII Fortessa Cytometer (BD Biosciences) or FACSymphony A5 SE Cell Analyzer (BD Biosciences). All data analysis was performed using FlowJo 9.0 software (FlowJo, LLC). Sorting assays were performed using a FACS Aria Cytometer (BD Bioscience). At the end of expansion, levels of target KO and silencing were assessed via flow cytometry. Of note, CD3ϵ, TCRαβ, and β2M expression was gated based on the control M5CAR T cells. For KO conditions, we determined the percentage of cells still expressing the targets, as well as of the KO populations. For the silencing conditions, the relative expression percentage was derived as follows: first we subtracted the median fluorescence intensity (MFI) of the unstained control and then we normalized the MFI of the transduced M5CAR+ T cells to the MFI of the negative cells within each sample. Relative expression levels were then obtained by normalizing to the scrambled control, where present, or to control M5CART cells. Silencing efficiency was finally calculated as 100% minus the relative expression percentage.

The University of Pennsylvania Institutional Animal Care and Use Committee approved all animal experiments, and all animal procedures were performed in the animal facility at the University of Pennsylvania in accordance with Federal and Institutional Animal Care and Use Committee requirements. Female 6–8-week-old NOD-scid IL2Rg^null (NSG) mice were purchased from Jackson laboratories and maintained in pathogen free conditions. All experimental protocols were approved by the Institutional Animal Care and Use Committee at the University of Pennsylvania. Mice were maintained in dedicated BSL-2 animal barrier spaces. Animals were injected subcutaneously (sc) with 2 x 10⁶ AsPC-1 CBG GFP cells in 50µl sterile PBS and 50µl Corning matrigel (Catalog# 356234). Once an average tumor volume of 150–200 mm³ was reached, 1 x 10⁶ M5CAR+ T cells were injected via tail vein in 200µl sterile PBS. The following T cell groups were tested, each matched for total T-cell dose injected by NTD normalization: NTD, M5CAR T cells, S CD3ζ&β2M M5CAR T cells and KO CD3ζ&β2M M5CAR T cells. No TCRαβ-depleted M5CAR T cells were used. We considered five different T cell donors in two independent experiments. In the first experiment, we employed ND587 and ND610 donors (n= 8–9 mice/group each donor), while in the second experiment, ND584, ND569, and ND627 were used (n= 6–7 mice/group each donor). Mice were evaluated by bi-weekly or weekly tumor volume, bioluminescent imaging (BLI) (Xenogen IVIS-200 spectrum camera), and weight measurements. Tumor volumes were calculated using the formula: volume (mm³) = (length x width²)/2. BLI data were analyzed using the Living Image software version 4.4 (Caliper LifeSciences, PerkinElmer). Animals were monitored for signs of clinical GVHD as evidenced by >10% loss in body weight, hunching, reduced activity, loss of fur, diarrhea, and conjunctivitis. Mice were euthanized if any tumor dimension exceeded 20 mm, if ulceration covered more than 90% of the tumor surface, if body weight loss was greater than 20%, in cases of disease-related hind limb paralysis, or for severe GVHD, following the approved protocol. A fixed endpoint of 100 days and 66 days after CAR T cell injection to terminate the first and second in vivo models, respectively, was defined for remaining mice. Mice that reached these endpoints were euthanized, and necropsy was performed to collect the primary tumor, assess treatment related toxicities and the presence of metastases. Peripheral blood was obtained by retro-orbital bleeding on days 16 (first and second experiment) and 26 (second experiment only) after CAR T cell injection and cell stained for M5CAR and human CD45, CD4, CD8, CD3ϵ, TCRαβ, and β2M markers. At the fixed final endpoint of 66 days after CAR T cell injection (for the second experiment only), three spleens from each donor were collected from the control M5CAR T cell group, nine spleens total, while 4 spleens from each donor were collected from both the S CD3ζ&β2M M5CAR T cell group and the KO CD3ζ&β2M M5CAR T cell group, twelve spleens total for each. Spleens were isolated, dissociated and cell stained for cell viability, M5CAR and for the previous markers. Cell subsets were quantified by flow cytometry using CountBright™ Absolute Counting Beads (BD). All experiments were performed in a randomized fashion.

Survival analysis was performed using the Kaplan Meier estimator and pairwise comparisons using the Log-Rank test. For pairwise comparisons, Wilcoxon Mann-Whitney test or Kruskall-Wallis test followed by Dunn's post-hoc test were used. If the assumptions for parametric tests were met, Student's T-test or linear regression models were employed. For longitudinal comparisons, linear mixed-effects models with a Gaussian distribution and an identity link were estimated. The model includes independent variables arm, time, and their interaction as fixed effects while subject-specific intercept and slope terms were included as random effects. All p-values were adjusted using the Bonferroni correction to account for multiple arms comparisons. Adjusted p-values were considered significant when below the alpha level 0.05.

For assays conducted on a limited number of T cell donors, non-parametric tests were preferred due to small sample sizes and the likely violation of parametric assumptions. However, given their limited statistical power, these tests could not reliably detect differences or establish statistical significance when biological replicates were limited (e.g., fewer than five in many datasets of this study). In such cases, descriptive statistics, including the mean and standard deviation (SD) or standard error of the mean (SEM), were reported in the graphs. The figure legends explicitly acknowledge the limitations of small sample sizes in accurately representing population variability and the associated challenges in achieving statistical significance.

Statistical analyses were carried out using R Statistical software (The R Foundation for Statistical Computing) version 4.2.3 and GraphPad Prism Version 10.1.1 (270) for macOS. Graphs were created by GraphPad Prism Version 10.1.1 (270) for macOS, BioRender (https://biorender.com↗) and Adobe Illustrator 2023, all under paid license.

To evaluate the effects of microRNA (miRNA)-based silencing versus CRISPR/Cas9-mediated deletion on the yield and functional performance of chimeric antigen receptor (CAR) T cells, we conducted proof-of-concept studies targeting the T-cell receptor (TCR) and major histocompatibility complex class I (MHC-I) complexes. These investigations aimed to develop non-alloreactive and hypoimmunogenic CAR T cells (Figure 1). We selected miRNAs and single guide RNAs (sgRNAs) that provide comparable repression of TCR expression and differential modulation of MHC-I for our experiments.

Utilizing methodologies described in Myburgh et al. (12) and online resources for target sequence design, we synthesized miRNAs within a mirGE backbone to target the transcripts of principal subunits of these protein complexes, specifically the CD3ζ and β2-microglobulin (β2M) genes. For CD3ζ, we designed miRNAs targeting mRNA regions outside the coding domain of CD3ζ in the CAR to avoid unintended effects on CAR expression. We considered five target sequences each for CD3ζ and β2M, cloning them individually into optimized gene silencing constructs (Figure 1b, Supplementary Table 2) (14), and delivering them to T cells via lentiviral transduction. Silencing efficiency was assessed in T cells by flow cytometry, using mCherry expression as a reporter for transduced cells.

Three constructs, CD3ζ_T1, CD3ζ_T2, and CD3ζ_T3, demonstrated highly efficient downregulation of TCRαβ expression (Figure 2a). Among these, CD3ζ_T2 achieved over 95% silencing of TCR expression, while CD3ζ_T4 and T5 were less effective. To achieve near-complete silencing of the TCR complex and facilitate purification of TCR-deficient cells, we combined the CD3ζ_T1 and CD3ζ_T2 miRNA coding sequences in the final expression construct (Figures 1b, 2b).

Regarding MHC-I, the β2M_T2 and T3 constructs resulted in mean (SD) silencing efficiencies of 47.5% (0.7) and 46.5% (0.7), respectively, while β2M_T1 and T4 had no apparent effect (Figure 2c). The β2M_T5 construct was the most efficient, achieving a 74.5% (3.5) reduction in MHC-I expression in T cells. To potentially protect modified T cells from a natural killer (NK) cell 'missing self' response, deeper silencing of β2M was not pursued. Consequently, the CD3ζ_T1, CD3ζ_T2, and β2M_T5 miRNA coding sequences were chosen for CAR T-cell engineering (Figure 1b) and cloned into the vector coding for the M5CAR in a bidirectional promoter configuration, resulting in dual CD3ζ and β2M silenced (S) M5CAR T cells (S CD3ζ&β2M M5CAR T cells) (Figure 1c).

For CRISPR/Cas9 knockout (KO) of MHC-I and TCR complexes, we utilized a sgRNA targeting the β2M gene, as designed for NCT05037669 (Supplementary Table 1). We also designed and screened multiple sgRNAs targeting regions outside the CD3ζ domain coding region in the CAR on T cells (Supplementary Table 1). Among them, sgRNA_4 was selected for its high KO efficiency, resulting in only 2.1% (0.4) CD3ϵ+ T cells with the lowest median fluorescence intensity (MFI) (Supplementary Figures 1A–D). Simultaneous deletions of CD3ζ and β2M in CAR-modified T cells targeting CD19 were efficient and comparable to single-target deletions (Supplementary Figures 1E–I). The selected sgRNAs were subsequently used to engineer dual CD3ζ and β2M KO M5CAR T cells (KO CD3ζ&β2M M5CAR T cells) and compared to miRNA-mediated silencing (Figure 1c).

The miRNA and M5CAR cassettes were expressed from the same lentiviral vector. Both single and dual CD3ζ and β2M S M5CAR T cells, including a scrambled control, were generated using the same protocol as control M5CAR T cells (Supplementary Figure 2A). In contrast, single and dual CD3ζ and β2M KO M5CAR T cells, along with control Mock M5CAR T cells, were produced using a protocol that included a CRISPR/Cas9 NF step prior to activation and transduction (Supplementary Figure 2B), as previously reported (18).

S M5CAR T cells, including S CD3ζ&β2M M5CAR T cells, exhibited expansion rates and cell size changes comparable to those of control M5CAR T cells (Supplementary Figures 3A, B). In contrast, the expansion of KO M5CAR T cells was negatively impacted by gene editing via NF (Supplementary Figures 3C, D). In all KO groups, including the KO CD3ζ&β2M M5CAR T cell group, NF on day 0 resulted in 40-50% reduction in cell numbers. Moreover, their expansion was impaired. Using the post-NF cell count as the baseline, KO cells showed significantly lower population doublings (PDs). This impairment was particularly evident before debeading on day 5, as KO cells displayed a longer and more pronounced lag phase compared to S cells. After day 5, KO and S cells expanded at comparable rates; however, KO cells failed to reach the same overall levels due to the initial deficit. The mean (SD) fold expansion of S CD3ζ&β2M M5CAR T cells from three and five T cell donors was 96 (30.4) (Figure 3a) and 90.5 (24.9) (Supplementary Figure 3E), respectively, comparable to those of single-S and control M5CAR T cell groups. The KO CD3ζ&β2M M5CAR T cell group showed markedly lower fold expansions of 14.8 (7.4) (Figure 3a) and 13.9 (5.4) (Supplementary Figure 3E), similar to the other KO groups. Together, these results indicate an approximately six fold difference in expansion yield between miRNA- and KO-modified M5CAR T cells. Starting with a higher number of cells for manufacturing in the KO groups, as specified in Materials and Methods and Supplementary Figure 2, compensated for these differences in expansion, resulting in a total number of cells after expansion comparable to that of the S groups.

The viability of S CD3ζ&β2M M5CAR T cells after bead removal on day 5 and at the end of expansion was similar to that of control M5CAR T cells (Figure 3b, Supplementary Figure 3F). In contrast, the viability of KO CD3ζ&β2M M5CAR T cells, as well as that of the single KO groups, was reduced due to NF and gene editing, leading to significant cell loss and a slightly higher rate of cell death compared to non-nucleofected groups (Figure 3b, Supplementary Figure 3F). The proportions of CD4+ and CD8+ cells were largely comparable across the M5CAR+ control, dual S, and dual KO groups, although overall CD4+ and CD8+ proportions varied noticeably between T cell donors, irrespective of the CAR T cell groups (Supplementary Table 3).

Finally, S CD3ζ&β2M M5CAR T cells exhibited a memory profile similar to that of control M5CAR T cells, with only a slightly higher percentage of central memory (CM) T cells and a lower percentage of effector memory (EM) T cells at both day 5 and the end of expansion (Figure 3c, Supplementary Figure 3G). In contrast, KO CD3ζ&β2M M5CAR T cells displayed a slightly higher percentage of naïve-like T cells on day 5 and throughout expansion, along with a marginally increased proportion of CM and terminally differentiated (TD) T cells and a decreased percentage of EM T cells compared to control and S CD3ζ&β2M M5CAR T cells at both time points (Figure 3c, Supplementary Figure 3G). This pattern of memory T cell subtypes was also observed in KO CD3ζ M5CAR T cells but not in the other groups. The expression of activation markers (4-1BB, TIGIT, PD-1, TIM-3, LAG-3, and OX40) showed no significant differences between S and KO M5CAR+ T cells and remained comparable to control M5CAR+ T cells (Supplementary Figure 4).

Overall, S M5CAR T cells demonstrated expansion and viability comparable to control M5CAR T cells, with similar memory profiles, albeit with a slight shift toward higher CM and lower EM T cell levels. In contrast, KO M5CAR T cells exhibited reduced expansion, lower viability, and less favorable memory phenotypes, retaining lower percentages of naïve-like T cells while showing an increased proportion of TD and CM cells and a reduced proportion of EM cells by the end of expansion. These findings indicate that miRNA-based gene silencing preserves the production yield of phenotypically desirable CAR T cells, unlike CRISPR/Cas9 KO via NF.

Gene silencing and KO efficiency were assessed by evaluating TCR expression (via CD3ϵ and TCRαβ) and MHC-I expression (via β2M) (Figure 4a). The mean fluorescence intensities (MFIs) of CD3ϵ and TCRαβ in S CD3ζ&β2M M5CAR T cells were slightly higher than those observed in KO CD3ζ&β2M M5CAR T cells but remained lower than in control and scrambled M5CAR T cells. Silencing of β2M resulted in reduced β2M expression compared to control and scrambled M5CAR T cells, whereas β2M was almost entirely absent on the surface of KO CD3ζ&β2M M5CAR T cells (Figure 4a).

CD3ζ and β2M silencing in M5CAR T cells was evaluated before TCRαβ depletion, confirming that M5CAR+ transduced cells exhibited lower expression levels of both TCR and MHC-I compared to their non-transduced M5CAR- counterparts (Figure 4b, top row). Following TCRαβ depletion, the selected transduced M5CAR+ cells retained miRNA-mediated silencing (Figure 4b, bottom row). In S CD3ζ&β2M M5CAR T cells, gene silencing was efficient, achieving reductions in surface expression of 86.9% (3.6) and 91.1% (6.7) for TCR via CD3ϵ and TCRαβ, respectively, and 67.7% (9.3) for MHC-I via β2M (Figure 4c). These results were comparable to those of single-S CD3ζ or β2M M5CAR T cells. After TCRαβ depletion, S CD3ζ&β2M M5CAR T cells were nearly 100% M5CAR+ and lacked detectable TCR expression (Figure 4D). Additionally, MHC-I silencing was retained, with residual β2M expression at 31.7% (13.5) of control M5CAR T cell levels (Figure 4d).

Production of KO CD3ζ&β2M M5CAR T cells resulted in the KO of target genes in both CAR-transduced and non-transduced T cells (Figure 4e, top row). TCRαβ depletion yielded a mixed population of M5CAR+ T cells and non-transduced T cells, both carrying the KO, with a subset still expressing β2M (Figure 4e, bottom row). Prior to TCRαβ depletion, a low percentage of KO CD3ζ&β2M M5CAR T cells retained expression of CD3ϵ, TCRαβ, and β2M (Figure 4f). After TCRαβ depletion, KO CD3ζ&β2M M5CAR T cells maintained the same percentage of β2M+ and M5CAR+ cells as observed at the end of manufacturing (Figure 4g).

Analysis of additional T cell donors confirmed the low relative expression of CD3ϵ, TCRαβ, and β2M in S CD3ζ&β2M M5CAR T cells (), as well as the low frequency of these markers in KO CD3ζ&β2M M5CAR T cells (). After TCRαβ depletion, the percentage of cells still expressing CD3ϵ, TCRαβ, and β2M at levels comparable to control M5CAR T cells was similar between S CD3ζ&β2M and KO CD3ζ&β2M M5CAR T cells. However, only in TCRαβ-depleted S CD3ζ&β2M M5CAR T cells was a higher percentage of M5CAR+ cells achieved (). Among the T cell donors analyzed in, the mean (SD) yield of TCRαβ depletion was slightly lower for S CD3ζ&β2M M5CAR T cells than for KO CD3ζ&β2M M5CAR T cells, though the difference was not statistically significant. 1 1 1 1

At the end of T cell expansion, all gene-S groups had a lower percentage of transduced T cells expressing M5CAR compared to control M5CAR T cells and KO groups (). Additionally, CAR MFIs in S M5CAR T cells appeared lower than in control M5CAR T cells, whereas some KO donors exhibited higher CAR MFIs, although the difference was not statistically significant (). The reduced M5CAR+ T cell levels and M5CAR MFIs in S groups likely stem from lower gene transfer efficiency and/or expression levels inherent to the multiplexed constructs, suggesting that further optimization of miRNA-carrying vectors and experimental conditions may be required. 1 1

Several processing factors influence the final product, including cell loss after NF in ko M5CAR T cells, PDs, transduction efficiency, and TCRαβ depletion yield. Assuming an equal initial manufacturing count of T cells (e.g., 2 × 10⁶ cells) for both S CD3ζ&β2M and KO CD3ζ&β2M M5CAR T cells, distinct differences in the yield and composition of the final product can be illustrated using the average values of these factors from five unique T cell donors (Figure 5). Based on the donors and methods employed in this study, the fold expansion of M5CAR+ T cells is estimated to be approximately four times higher in S compared to KO M5CAR T cells. The final product for S CD3ζ&β2M M5CAR T cells consists of a highly pure population of TCRαβ-depleted M5CAR+ T cells with MHC-I silencing. In contrast, the final product for KO CD3ζ&β2M M5CAR T cells is a heterogeneous mixture of TCRαβ-depleted cells, including both M5CAR+ and M5CAR- T cells with β2M KO, as well as M5CAR+ and M5CAR- T cells still expressing β2M. The proportions of M5CAR+ and β2M+ T cells depend on transduction and KO efficiencies.

Overall, miRNA-mediated gene silencing resulted in significantly higher fold expansion and a nearly pure population of modified M5CAR T cells compared to CRISPR/Cas9-mediated KO.

The cytotoxicity of miRNA- and CRISPR-modified CAR T cells was assessed in vitro by co-culturing with the human PDAC Msln+ cell line AsPC-1. Both single and dual CD3ζ and β2M S M5CAR T cells exhibited potent tumor-killing ability, with no significant differences observed between groups in endpoint assays (Figures 6a, b). In real-time co-culture, S M5CAR T cells effectively eliminated tumor cells, albeit at a slightly slower rate compared to KO or control M5CAR T cells (Figure 6c, Supplementary Figure 6A). After six days of co-culture, cells were collected and analyzed by flow cytometry to assess potential differences among groups across T cell donors. The memory phenotype of M5CAR+ T cells was similar across all groups, with the majority exhibiting an effector memory (EM) profile (Supplementary Figure 6B). Gene silencing and KO of target genes were confirmed at the end of the co-culture (Supplementary Figure 6C).

Expansion rates for M5CAR+ and M5CAR- T cells, M5CAR MFI (normalized to the control group), and the percentage of M5CAR+ T cells post-co-culture were also evaluated (). Despite donor variability, M5CAR+ T cells in both the S CD3ζ&β2M and KO CD3ζ&β2M M5CAR T cell groups generally expanded less than control M5CAR T cells. Some donors also exhibited increased M5CAR- T cell populations in the S and KO groups at various effector-to-target (E:T) ratios (). For the percentage of M5CAR+ T cells over total live cells in co-culture and the M5CAR MFI of the M5CAR+ T cell population, trends remained consistent across all E:T ratios, with the S CD3ζ&β2M M5CAR T cell group showing, on average, lower percentages and MFIs (). Data summarizing treatment group comparisons for each T cell donor are presented in. In donors where scrambled, S CD3ζ, and S β2M M5CAR T cell groups were evaluated (ND587, ND561, and ND610), similar trends were observed as in the S CD3ζ&β2M M5CAR T cell condition (), suggesting that the lower proportions of M5CAR+ T cells were independent of the specific silencing targets. 1 1 1 1 1

Overall, these findings indicate a deficiency in the expansion of both S CD3ζ&β2M and KO CD3ζ&β2M M5CAR T cells, with a more pronounced effect on M5CAR+ T cells than on M5CAR- T cells. This discrepancy, noting a higher tendency for CAR downregulation following antigen stimulation, has been reported previously (17). Consistent with observations at the end of T cell expansion, post-co-culture M5CAR MFIs in S CD3ζ&β2M M5CAR T cells remained lower than in control and KO M5CAR T cells, whereas M5CAR MFIs in KO CD3ζ&β2M M5CAR T cells were slightly lower, comparable, or, in some cases, higher than in control M5CAR T cells. Despite variability among donors, differences in expansion, and a slower tumor-killing rate, both endpoint and real-time assays demonstrated that S M5CAR T cells effectively controlled tumor growth.

To compare the antitumor activity of S CD3ζ&β2M and KO CD3ζ&β2M M5CAR T cells, we conducted in vivo experiments using a subcutaneous PDAC model in NSG mice (Figure 7a). The first study, involving two T cell donors, ran for 100 days post-CAR T cell injection (Figure 7b), while the second study, with three donors, lasted 66 days (Figures 7c, d). Both S CD3ζ&β2M and KO CD3ζ&β2M M5CAR T cells demonstrated similar efficacy in controlling primary tumor growth, though responses varied between donors (Figures 7b, c). In all groups, tumor growth was effectively controlled during the initial weeks. Control M5CAR T cells exhibited the most potent and sustained tumor suppression across all donors, except for ND610 (Figures 7b, c). Survival was generally poorest in the NTD group, particularly for donors ND587 and ND584 (Supplementary Figures 7A, B). In donor ND569, the lowest survival was observed in the M5CAR T cell group, attributed to graft-versus-host disease (GVHD), suggesting a stronger xenogeneic response. Survival mediated by S CD3ζ&β2M and KO CD3ζ&β2M M5CAR T cells was comparable in ND584, slightly improved with S cells in ND587, ND569, and ND627, and better with KO cells in ND610 (Supplementary Figures 7A, B). Symptoms of GVHD, including >10% weight loss, hunching, reduced activity, fur loss, and conjunctivitis, were almost exclusively observed in the control M5CAR T cell group, affecting 13 out of 34 mice (38.2%) across both experiments, particularly in donors ND587 and ND569. GVHD symptoms were detected in only one mouse treated with KO CD3ζ&β2M M5CAR T cells and were completely absent in the S CD3ζ&β2M M5CAR T cell group.

The causes of death for each treatment group and T cell donor are detailed in Supplementary Tables 6 and 7, corresponding to the first and second in vivo studies, respectively.

At the fixed endpoint of 66 days post-injection in the second study, necropsy and a semiquantitative assessment of metastases were performed (Figure 7d, Supplementary Table 5). Lung metastases were the most common, appearing as either isolated small masses (fewer than two), small tumors covering the surface of one or both lung lobes, or large masses. Additional metastases were observed in the spleen and in the same flank as the primary tumor, extending toward the upper arm in the subcutaneous tissue outside the peritoneum. S CD3ζ&β2M M5CAR T cells were more effective than KO CD3ζ&β2M M5CAR T cells in preventing metastasis formation. In the S CD3ζ&β2M M5CAR T cell group, small, isolated lung metastases in a single lung lobe were detected in only 3 out of 18 mice (16.7%) (Figure 7d). In contrast, mice treated with KO CD3ζ&β2M M5CAR T cells exhibited higher metastatic burden, with 6.25% showing diffuse small lung metastases in one lobe, 37.5% with small metastases in two lobes, 18.8% with large lung metastases, and 12.5% with spleen metastases (Figure 7d). No metastases were found in the control M5CAR T cell group, while the non-transduced T cell (NTD) group exhibited significantly higher metastatic rates than all other groups (Figure 7d). Overall, the S CD3ζ&β2M M5CAR T cell group had significantly fewer lung metastases than both the NTD and KO CD3ζ&β2M M5CAR T cell groups.

These findings demonstrate that S CD3ζ&β2M M5CAR T cells were as effective as KO CD3ζ&β2M M5CAR T cells in controlling primary tumor growth. However, S CD3ζ&β2M M5CAR T cells provided superior metastatic control, highlighting the potential advantage of gene silencing over CRISPR-mediated KO in preventing tumor spread.

Differences in antitumor effects between treatment groups may be attributed to variations in the in vivo persistence of CAR T cells and/or M5CAR expression levels. On day 16, M5CAR+ T cells were detectable in 94.1% (32/34) of mice in the control M5CAR T cell group, 61.1% (22/36) in the S CD3ζ&β2M M5CAR T cell group, and only 22.9% (8/35) in the KO CD3ζ&β2M M5CAR T cell group. By day 26 of the second experiment, all mice (18/18) in the control M5CAR T cell group, 84.2% (16/19) in the S CD3ζ&β2M M5CAR T cell group, and just 26.3% (5/19) in the KO CD3ζ&β2M M5CAR T cell group still had detectable M5CAR+ T cells in circulation. Across both time points, CD4+ and CD8+ M5CAR+ T cell levels were highest in the control group, with the S CD3ζ&β2M M5CAR T cell group generally maintaining higher levels than the KO CD3ζ&β2M M5CAR T cell group (Figures 8a, b).

On day 66 of the second experiment, M5CAR+ T cells remained detectable in the spleens of all mice in the control M5CAR T cell group. Consistent with blood samples, splenic levels of adoptively transferred T cells were lower in both the S CD3ζ&β2M and KO CD3ζ&β2M M5CAR T cell groups (Figure 8c). In the S CD3ζ&β2M M5CAR T cell group, 75.0% (9/12) of mice had appreciable levels of M5CAR+ T cells, whereas only 33.3% (4/12) in the KO CD3ζ&β2M M5CAR T cell group had detectable levels. Differences in M5CAR expression levels observed in vitro were confirmed in vivo. Among mice with detectable M5CAR+ T cells, the MFI of S CD3ζ&β2M M5CAR+ T cells was lower than that of KO CD3ζ&β2M and control M5CAR+ T cells in both blood (days 16 and 26) and spleen (day 66) (Figure 8d).

Gene silencing in S CD3ζ&β2M M5CAR+ T cell population remained stable in vivo. On day 26, complete silencing of both target genes was still observed in the blood (Figure 8e) while the percentage of dual KO cells in the KO CD3ζ&β2M M5CAR+ T cell population had significantly declined (Figure 8e), averaging only 21.8% (10.1). Silencing persisted in the spleen on day 66, with 88.9% (8/9) of mice in the S CD3ζ&β2M M5CAR T cell group retaining full silencing of both CD3ζ and β2M (Figure 8f) while the KO CD3ζ&β2M M5CAR+ T cell group had largely lost the KO (Figure 8g). Among the four mice in the KO CD3ζ&β2M M5CAR T cell group with detectable M5CAR+ T cells, the majority of M5CAR+ T cells were non-KO, 64.9% (33.1), while only 18.8% (20.1) were dual KO, and 14.5% (14.7) retained a single CD3ζ KO.

Our findings indicate that S CAR T cells exhibited superior in vivo persistence compared to KO CAR T cells, despite having a lower percentage of M5CAR+ T cells and lower CAR MFI. The S CD3ζ&β2M M5CAR+ T cells maintained stable gene silencing in both blood and spleen. In contrast, the proportion of KO CD3ζ&β2M M5CAR⁺ T cells markedly decreased in the blood, and this reduction was subsequently confirmed in the spleen. These results demonstrate that S M5CAR+ T cells persist as a relatively homogeneous population, whereas KO M5CAR+ T cells become more heterogeneous over time. Control M5CAR T cells had the highest persistence and antitumor efficacy of all groups, suggesting that TCR absence in CAR T cells may impact potency and warrants further investigation.

To evaluate the immunogenicity of S CD3ζ&β2M M5CAR T cells and their resistance to host immune responses, we conducted co-culture assays with alloreactive NK cells and peripheral blood mononuclear cells (PBMCs). TCRαβ-depleted M5CAR T cells were co-cultured with freshly isolated peripheral blood NK cells from non-matched healthy donors. S CD3ζ&β2M M5CAR T cells, which retain partial MHC-I expression, were significantly less immunogenic to allogeneic NK cells compared to KO CD3ζ&β2M M5CAR T cells, although not to the extent observed in M5CAR T cells fully expressing MHC-I (Figure 9a). The level of protection from allo-NK cells appeared to be inversely correlated with the degree of MHC-I silencing across different M5CAR T cell donors (Figure 9b).

Next, we assessed whether MHC-I silencing in S CD3ζ&β2M M5CAR T cells was sufficient to prevent recognition and activation of allogeneic PBMCs (allo-PBMCs). As expected, the strongest allo-PBMC response was observed when stimulated by control M5CAR T cells (Figure 9c). Both S CD3ζ&β2M and KO CD3ζ&β2M M5CAR T cells exhibited a similarly reduced stimulatory capacity of allo-PBMCs, regardless of the presence of NK cells (Figure 9C). These data suggest that the tuned low surface MHC-I expression provides better protection against NK cell-mediated killing for S CD3ζ&β2M M5CAR T cells compared to KO CD3ζ&β2M M5CAR T cells. Additionally, functional MHC-I silencing in S CD3ζ&β2M M5CAR T cells was sufficient to prevent alloreactive responses from allo-PBMCs, comparable to that observed in KO CD3ζ&β2M M5CAR T cells.