Allogeneic CD33-directed CAR-NKT cells for the treatment of bone marrow-resident myeloid malignancies

Abstract

Chimeric antigen receptor (CAR)-engineered T cell therapy holds promise for treating myeloid malignancies, but challenges remain in bone marrow (BM) infiltration and targeting BM-resident malignant cells. Current autologous CAR-T therapies also face manufacturing and patient selection issues, underscoring the need for off-the-shelf products. In this study, we characterize primary patient samples and identify a unique therapeutic opportunity for CAR-engineered invariant natural killer T (CAR-NKT) cells. Using stem cell gene engineering and a clinically guided culture method, we generate allogeneic CD33-directed CAR-NKT cells with high yield, purity, and robustness. In preclinical mouse models, CAR-NKT cells exhibit strong BM homing and effectively target BM-resident malignant blast cells, including CD33-low/negative leukemia stem and progenitor cells. Furthermore, CAR-NKT cells synergize with hypomethylating agents, enhancing tumor-killing efficacy. These cells also show minimal off-tumor toxicity, reduced graft-versus-host disease and cytokine release syndrome risks, and resistance to allorejection, highlighting their substantial therapeutic potential for treating myeloid malignancies.

Fig. 1. Biomarker and transcriptome profiling of primary AML and MDS patient samples.

a Experiment design to profile primary AML and MDS patient bone marrow (BM) samples using flow cytometry and single cell RNA sequencing (scRNA-seq). 8 AML and MDS primary samples were included for flow cytometry analyses. Data from Gene Expression Omnibus database (GSE235923) and NCBI Sequence Read Archive (PRJNA720840) were included for scRNA-seq analyses. Created in BioRender. LI, Y. (2025) https://BioRender.com/o37h997. b Diagram showing the progression from leukemia stem cells (LSCs) to myeloblast cells (MBCs) in myeloid malignancies, along with associated biomarkers. Created in BioRender. FANG, Y. (2025) https://BioRender.com/f57l159c–e Profiling AML and MDS blast cells using flow cytometry. c FACS detection of the subpopulations of AML blast cells, and their expression of CAR target (CD33), NKT TCR target (CD1d), and NKR ligands (i.e., CD112, MICA/B, and ULBP-1). Two representative data sets from AML samples #1 and #2 are presented. d Quantification of the proportions of the three subpopulations of AML and MDS blast cells. The combined percentage of these subpopulations totals 100%. e Quantification of the expression of CAR target, NKT TCR target, and NKR ligands on the three subpopulations of AML and MDS blast cells (n = 7 for LSPC, and n = 8 for CMP and MBC; n represents different patient samples). f–k Profiling AML blast cells using scRNA-seq. Data from Gene Expression Omnibus database (GSE235923) were analyzed. f Combined UMAP plot showing the formation of four major cell clusters. 19 primary AML blast samples were analyzed. g UMAP plots showing the expression distribution of the CD34, CD38, and stem genes SOX4 and CD99. h Bar graphs showing the cell cluster proportions of the 19 primary AML blast samples. Expression of cancer stem cell (CSC) gene signature (i) and NKR ligand gene signature (j) in the indicated cell clusters. UMAP plots showing the gene expression distributions and violin plots showing the gene expression levels are presented. Data from the 19 primary AML blast samples are shown. k Violin plots showing the expression distribution of NKR ligand gene signature in the 19 primary AML blast samples. Representative of 1 (d–k) and 8 (c) experiments. In the violin plots (i, j), box and whisker plots exhibit the minimum, lower quartile, median, upper quartile and maximum expression levels of each type of cell. Source data and exact p values are provided as a Source Data file.

Fig. 2. Generation and characterization of HSPC-engineered allogeneic IL-15-enhanced CD33-directed CAR-NKT (^Allo15CAR33-NKT) cells.

a Schematics showing the generation of ^Allo15CAR33-NKT cells. HSPC, hematopoietic stem and progenitor cells; Lenti/iNKT-CAR33-IL-15, lentiviral vector encoding a pair of iNKT TCR α and β chains, a CD33-directed CAR, and a human soluble IL-15. Created in BioRender. LI, Y. (2025) https://BioRender.com/o37h997. b Schematics showing the design of Lenti/iNKT-CAR33-IL-15 lentivector. ΔLTR, self-inactivating long terminal repeats; MNDU3, internal promoter derived from the MND retroviral LTR U3 region; φ, packaging sequence; RRE, rev-responsive element; cPPT, central polypurine tract; WPRE, woodchuck hepatitis virus posttranscriptional regulatory element; F2A, foot-and-mouth disease virus 2 A; P2A, porcine teschovirus-1 2A; T2A, thosea asigna virus 2A. c FACS and immunofluorescence (IF) monitoring of the generation of ^Allo15CAR33-NKT cells during the 6-week culture. iNKT TCR was stained using a 6B11 monoclonal antibody. d Percentage of ^Allo15CAR33-NKT cells in total live cells during the 6-week culture (n = 4; n indicates different CB donors). e Yield of ^Allo15CAR33-NKT cells (n = 4; n indicates different CB donors). f FACS detection of surface markers on ^Allo15CAR33-NKT cells. Healthy donor peripheral blood mononuclear cell (PBMC)-derived conventional CD33-directed CAR-engineered T (CAR33-T) cells were included as a control. DN double-negative, DP double-positive. g Comparison of the indicated subpopulation percentages between ^Allo15CAR33-NKT and conventional CAR33-T cells (n = 5; n indicates different cell batches) h Single cell TCR sequencing analyses of ^Allo15CAR33-NKT and conventional CAR33-T cells. i FACS detection of NK marker and NK receptor (NKR) expression, as well as intracellular cytokine and cytotoxic molecule production of ^Allo15CAR33-NKT and conventional CAR33-T cells. j Violin plots showing the expression distribution of the indicated gene signatures in ^Allo15CAR33-NKT and conventional CAR33-T cells. TF, transcription factor. k Pathway analyses of differentiated expressed genes comparing ^Allo15CAR33-NKT with conventional CAR33-T cells. GO, Gene ontology ID. Representative of 1 (h, j, k) and >5 (a–g, i) experiments. For the scTCR-seq (h) and scRNA-seq analyses (j, k), one ^Allo15CAR33-NKT sample (containing 12,006 cells) and one CAR33-T sample (containing 9122 cells) were analyzed. Source data and exact p values are provided as a Source Data file.

Fig. 3. In vitro tumor targeting efficacy and mechanisms of ^Allo15CAR33-NKT cells.

a–e Studying the in vitro antitumor efficacy of ^Allo15CAR33-NKT cells against human AML cell lines. CAR33-T cells and non-CAR33-engineered PBMC-T cells were included as therapeutic cell controls. a Experimental design. b Schematics showing the indicated human AML cell lines. THP1-FG, THP1 cell line engineered to overexpress the firefly luciferase and green fluorescence protein dual reporters (FG); KG1-FG, KG1 cell line engineered to overexpress FG; HL60-FG, HL60 cell line engineered to overexpress FG; THP1-FG^CD33-/-, THP1-FG cell line further engineered to knockout the CD33 gene; THP1-FG^CD1d-/-, THP1-FG cell line further engineered to knockout the CD1d gene; THP1-FG^CD33/CD1d-/-, THP1-FG cell line further engineered to knockout the CD33 and CD1d genes. c FACS detection of CD33 and CD1d expressions on the indicated AML cells. d Heatmap showing the NKR ligand expressions on the indicated AML cells. The number represents the percentage of NKR ligand-positive tumor cells out of the total tumor cells. Three independent tumor cell samples were analyzed, and the average numbers are presented. e Tumor cell killing data at 24 h (n = 4 from four different cell product donors). f, g Studying the tumor cell killing mechanisms of ^Allo15CAR33-NKT cells mediated by NKRs (i.e., NKG2D and DNAM-1). f Experimental design. g Tumor cell killing data at 24 h (E:T ratio = 10:1; n = 4 from four different cell product donors). h Diagram showing the CAR/TCR/NKR triple tumor-targeting mechanisms of ^Allo15CAR33-NKT cells, and the CAR single tumor-targeting mechanism of CAR33-T cells. GrzB, Granzyme B. Created in BioRender. FANG, Y. (2025) https://BioRender.com/j50y057. i, j Studying the expression of effector molecules of ^Allo15CAR33-NKT cells. i FACS detection of surface CD69 as well as intracellular Perforin and Granzyme B in ^Allo15CAR33-NKT cells. j Quantification of (i) (n = 3 from three different cell product donors). Representative of 3 experiments. Data are presented as the mean ± SEM. ns not significant, *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001, by one-way ANOVA (g, j), or two-way ANOVA (e). Source data and exact p values are provided as a Source Data file.

Fig. 4. Distinct in vivo bone marrow homing of ^Allo15CAR33-NKT cells mediated by CXCR4/CCR5 expression.

a Experimental design to study the in vivo PK/PD of ^Allo15CAR33-NKT cells in a xenograft NSG mouse model. The therapeutic cells were labeled with FG. Created in BioRender. LI, Y. (2025) https://BioRender.com/u95q769. b BLI images showing the presence of therapeutic cells in experimental mice over time. Ventral and left-side views are shown. c Quantification of (b) (n = 3 from three experimental mice). TBL, total body luminescence; p/s, photons per second. d BLI images showing the biodistribution of ^Allo15CAR33-NKT and CAR33-T cells in representative experimental mice. Ventral and left-side views are shown. e Quantification of therapeutic cell tissue (i.e., bone marrow and other tissues) distribution (n = 3 from three experimental mice). f FACS detection of therapeutic cells in the bone marrow of experimental mice 30 days after cell injection. g Quantification of (f) (n = 5 from five experimental mice). h FACS measurement of surface CXCR4 and CCR5 expressions in the indicated therapeutic cells. i Quantification of (h) (n = 5 from five experimental mice). j Schematic illustrating the distinct in vivo bone marrow homing capacity of ^Allo15CAR33-NKT cells mediated by CXCR4/CCR5 expression. Created in BioRender. LI, Y. (2025) https://BioRender.com/g04r306. Representative of 2 experiments. Data are presented as the mean ± SEM. ***p < 0.001; ****p < 0.0001, by two-tailed Student’s t test (g, i). Source data and exact p values are provided as a Source Data file.

Fig. 5. Targeting bone marrow-resident leukemia stem and progenitor cells by ^Allo15CAR33-NKT cells using AML xenograft mouse models and primary patient samples.

a–g Studying the in vivo antitumor efficacy of ^Allo15CAR33-NKT cells in a THP1-FG human AML xenograft NSG mouse model. a Experimental design. b BLI images. c Quantification of (b) (n = 5). d Kaplan–Meier survival curves (n = 5). e BLI images showing the presence of residual tumor cells in mouse tissues at the termination day. GI tract, gastrointestinal tract. FACS analyses of surface CD33 (f) and intranuclear cancer stem cell (CSC) marker (g) expression in THP1-FG tumor cells, collected from mouse bone marrow at the termination day (n = 5). h–n Studying the in vivo antitumor efficacy of ^Allo15CAR33-NKT cells in a KG1-FG human AML xenograft NSG mouse model. h Experimental design. i BLI images. j Quantification of (i) (n = 5). k Kaplan–Meier survival curves (n = 5). l BLI images showing the presence of residual tumor cells in mouse tissues at the termination day. FACS analyses of surface CD33 (m) and intranuclear CSC marker (n) expression in KG1-FG tumor cells collected from mouse bone marrow at the termination day (n = 5). o–r Studying the antitumor efficacy of ^Allo15CAR33-NKT cells against primary patient samples. o Experimental design. Created in BioRender. LI, Y. (2025) https://BioRender.com/o37h997p Blast cell killing data at 24 h (n = 4 from four different therapeutic cell donors). Data from three AML and one MDS patient samples are presented. FACS analyses of surface CD33 (q) and intranuclear CSC marker r expression in the remaining blast cells collected after the 24-h in vitro tumor cell killing assay (n = 4 from four different therapeutic cell donors). s Diagram illustrating the ability of ^Allo15CAR33-NKT cells to target CD33-low/negative LSPCs in myeloid malignancies. Created in BioRender. LI, Y. (2025) https://BioRender.com/m06k581. Representative of 2 (a–n) and 3 (o–r) experiments. Data are presented as the mean ± SEM. ns, not significant; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001, by two-tailed Student’s t test (f, g, m, n), one-way ANOVA (c, j, p, q), or log rank (Mantel-Cox) text adjusted for multiple comparisons (d, k). Source data and exact p values are provided as a Source Data file.

Fig. 6. In vivo tumor targeting efficacy of ^Allo15CAR33-NKT cells in a human AML patient-derived xenograft (PDX) NSG mouse model.

a Experimental design. Created in BioRender. LI, Y. (2025) https://BioRender.com/o37h997. b FACS detection of AML blast cells (gated as CD33⁺CD45⁺ cells) in various tissues collected from experimental mice at the termination day. c FACS analyses showing the AML blast cell loads in various tissues of experimental mice (n = 3 for Vehicle and ^Allo15CAR33-NKT; n = 4 for CAR33-T). Tissues from both the Vehicle and CAR33-T cell groups were collected from experimental mice at the termination day. Tissues from the ^Allo15CAR33-NKT cell group were collected from experimental mice at day 100. d H&E-stained tissue sections. Tissues were collected from experimental mice at day 20 post therapeutic cell injection. Scale bars, 100 µm for bone marrow samples, 200 µm for liver and kidney samples. Three independent tissue samples from each group were analyzed, and one representative data are presented. e Kaplan–Meier survival curves of experimental mice over time (n = 3 for Vehicle and ^Allo15CAR33-NKT; n = 4 for CAR33-T). Representative of 2 experiments. Data are presented as the mean ± SEM. ns not significant, *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001, by one-way ANOVA (c), or log rank (Mantel-Cox) text adjusted for multiple comparisons (e). Source data and exact p values are provided as a Source Data file.

Fig. 7. Synergistic effect of ^Allo15CAR33-NKT cells with hypomethylating agent (HMA) in the treatment of myeloid malignancies.

a–f Studying the synergistic effect of ^Allo15CAR33-NKT cells with HMA Decitabine using an in vitro tumor cell killing assay. a Experiment design. THP1-D, THP1 tumor cells treated with decitabine. b FACS detection of CAR target (CD33), iNKT TCR target (CD1d), and NKR targets (i.e., CD112, MICA/B, ULBP-1, and ULBP-2/5/6) on the indicated AML cells. c Quantification of (b) (n = 4 from four different experimental batches). d Tumor cell killing data at 24 h (n = 4 from four different experimental batches). e FACS detection of intracellular Granzyme B production by ^Allo15CAR33-NKT cells. f Quantification of (e) (n = 4 from four different experimental batches). g Diagram showing the upregulation of CD1d and NK ligands on AML tumor cells following treatment with HMA. Created in BioRender. FANG, Y. (2025) https://BioRender.com/n85n160. h–k Studying the in vivo synergistic effect of ^Allo15CAR33-NKT cells with HMA Decitabine using a THP1-FG human AML xenograft NSG mouse model. h Experimental design. i BLI images showing the presence of tumor cells in experimental mice over time. j Quantification of (i) (n = 5). k Kaplan–Meier survival curves of experimental mice over time (n = 5). Representative of 2 (h–k) and 3 (a–g) experiments. Data are presented as the mean ± SEM. ns not significant, *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001, by two-tailed Student’s t test (c), one-way ANOVA (d, f, j), or log rank (Mantel-Cox) text adjusted for multiple comparisons (k). Source data and exact p values are provided as a Source Data file.

Fig. 8. On-target off-tumor effect of ^Allo15CAR33-NKT cells against hematopoietic precursors.

a–e Studying the HSPC targeting using an in vitro immune cell killing assay. a Experimental design. G-CSF, granulocyte colony-stimulating factor. Created in BioRender. LI, Y. (2025) https://BioRender.com/o37h997b FACS detection of CD33 expression on the indicated immune cells. c Immune cell killing data at 24 h (n = 3 for healthy donor 1, and n = 5 for healthy donors 2 and 3; n indicates different therapeutic cell batches). d Flow detection of CD1d expression on HSPCs. e HSPC killing data at 24 h (n = 3; n indicates different therapeutic cell batches). f–h Studying the hematopoietic precursor targeting using an in vitro HSPC colony formation assay. f Experimental design. g Images showing the formation of Burst-Forming Unit-Erythroid (BFU-E) and Colony-Forming Unit-Granulocyte/Macrophage (CFU-GM) colonies. h Quantification of (g) (n = 8). i–p Studying the hematopoietic precursor targeting using an in vivo bone marrow-liver-thymus (BLT) humanized mouse model. i Experimental design. Created in BioRender. FANG, Y. (2025) https://BioRender.com/g87v886j FACS detection of human immune cells in the bone marrow collected from BLT mice 8 weeks post HSPC injection and prior to therapeutic cell injection. My, myeloid cell; DC, dendritic cell. k FACS detection of CD33 expression on the indicated immune cells. For (j, k) data from three independent mice were analyzed, and one representative data are presented. l Body weight measured over time. m Clinical scores recorded over time. The score was calculated as the sum of individual scores of 5 categories (activity, posture, dehydration, diarrhea, and dishevelment; score 0–1 for each category). n FACS analyses of immune cell targeting in bone marrow on Day 15. The percentage of the indicated immune cells among total CD45⁺CAR^- immune cells from each experimental mouse was recorded, and the fold change was calculated by normalizing to the NT group. FACS analyses of CD33⁺ and CD33^- cell targeting in bone marrow (o), and immune cell targeting in spleen and liver (p). NA not available. In (l–p), n = 3 for NT and CAR33-T, and n = 4 for ^Allo15CAR33-NKT; n indicates different experimental mice. Representative of 2 (i–p) and 3 (a–h) experiments. Data are presented as the mean ± SEM. ns not significant, *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001, by one-way ANOVA (h, n, o, p). Source data and exact p values are provided as a Source Data file.

Fig. 9. Safety study of ^Allo15CAR33-NKT cells.

a,b Studying the graft-versus-host (GvH) response of ^Allo15CAR33-NKT cells using an in vitro mixed lymphocyte reaction (MLR) assay. CD33-negative PBMCs were pre-sorted using MACS or FACS and used as stimulator cells. Conventional CAR33-T cells were included as responder controls. a Experimental design. b ELISA analyses of IFN-γ production on day 4. N, no addition of stimulator cells (n = 4). c–h Studying the GvHD risk of ^Allo15CAR33-NKT cells using a human xenograft NSG mouse model. c Experimental design. d Clinical GvHD score recorded over time (n = 5). The score was calculated as the sum of individual scores of 6 categories (body weight, activity, posture, skin thickening, diarrhea, and dishevelment; score 0–2 for each category). p was calculated using Day 50 data. e Body weight measured over time (n = 5). p was calculated using Day 50 data. f Kaplan–Meier survival curves (n = 5). g H&E-stained tissue sections. Tissues were collected from experimental mice on day 50. Scale bar, 100 µm. h Quantification of (g) (n = 5). i–l Studying the CRS response induced by ^Allo15CAR33-NKT cells using a THP1-FG human AML xenograft NSG mouse model. i Experimental design. j Body weight of experimental mice over time (n = 4). ELISA analyses of mouse IL-6 and SAA3 in mouse serum (k) or peritoneal fluid (l) (n = 4). NT, samples collected from tumor-bearing mice receiving no therapeutic cell treatment. m Studying the long-term safety of ^Allo15CAR33-NKT cells using a human xenograft NSG mouse model. Tissues from experimental mice were collected 120 days after injection with ^Allo15CAR33-NKT cells. Data were presented as pathologist’s scores of individual mouse tissues (n = 5). Representative of 1 (k) and 3 (a–j) experiments. Data are presented as the mean ± SEM. ns not significant, *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001, by Student’s t test (d, e, h), one-way ANOVA (b, k, l), or log rank (Mantel-Cox) text adjusted for multiple comparisons (f). Source data and exact p values are provided as a Source Data file.

Fig. 10. Comparison of ^Allo15CAR33-NKT cells and IL-15-enhanced conventional CAR33-T (¹⁵CAR33-T) cells.

a–d Generation of ¹⁵CAR33-T cells. a Diagram showing the design of Lenti/CAR33-IL-15 lentivector, and the generation of ¹⁵CAR33-T cells from healthy donor PBMCs. Created in BioRender. LI, Y. (2025) https://BioRender.com/o37h997b FACS detection of the CAR33 and CD4/CD8 co-receptors on CAR33-T and ¹⁵CAR33-T cells. c ELISA analyses of IL-15 production by CAR33-T and ¹⁵CAR33-T cells cultured in vitro for 24 h (n = 3; n indicates different PBMC donors). Note the successful incorporation and expression pf IL-15 transgene in the ¹⁵CAR33-T cells. d Yield of CAR33-T and ¹⁵CAR33-T cells (n = 3; n indicates different PBMC donors). e,f Studying the antitumor efficacy of ¹⁵CAR33-T cells against human AML cell lines. e Experimental design. f Tumor cell killing data at 24 h (n = 4). g–j Studying the in vivo antitumor efficacy of ¹⁵CAR33-T cells using a THP1-FG human AML xenograft NSG mouse model. g Experimental design. h BLI images showing the presence of tumor cells in experimental mice over time. i Quantification of (h) (n = 5). j Kaplan–Meier survival curves of experimental mice over time (n = 5). k–n Studying the in vivo antitumor efficacy of ¹⁵CAR33-T cells using a KG1-FG human AML xenograft NSG mouse model. k Experimental design. l BLI images showing the presence of tumor cells in experimental mice over time. m Quantification of (l) (n = 5). n Kaplan–Meier survival curves of experimental mice over time (n = 5). Note that the data for the Vehicle, ^Allo15CAR33-NKT, and CAR33-T groups were also presented in the main Figs. 5h–k. Representative of 2 (g–n) and 3 (a–f) experiments. Data are presented as the mean ± SEM. ns, not significant; ****p < 0.0001, by Student’s t test (c, d), one-way ANOVA (i, m), or two-way ANOVA (f). Source data and exact p values are provided as a Source Data file.