CRISPR/Cas9 editing of NKG2A improves the efficacy of primary CD33-directed chimeric antigen receptor natural killer cells

Abstract

Chimeric antigen receptor (CAR)-modified natural killer (NK) cells show antileukemic activity against acute myeloid leukemia (AML) in vivo. However, NK cell-mediated tumor killing is often impaired by the interaction between human leukocyte antigen (HLA)-E and the inhibitory receptor, NKG2A. Here, we describe a strategy that overcomes CAR-NK cell inhibition mediated by the HLA-E-NKG2A immune checkpoint. We generate CD33-specific, AML-targeted CAR-NK cells (CAR33) combined with CRISPR/Cas9-based gene disruption of the NKG2A-encoding KLRC1 gene. Using single-cell multi-omics analyses, we identified transcriptional features of activation and maturation in CAR33-KLRC1^ko-NK cells, which are preserved following exposure to AML cells. Moreover, CAR33-KLRC1^ko-NK cells demonstrate potent antileukemic killing activity against AML cell lines and primary blasts in vitro and in vivo. We thus conclude that NKG2A-deficient CAR-NK cells have the potential to bypass immune suppression in AML.

Fig. 1. Generation process and evaluation of CRISPR/Cas9 knockout for CAR33-KLRC1^ko-NK cells.

a Scheme of the in vitro generation of primary CAR33-KLRC1^ko-NK cells. b Exemplary flow cytometry plots of CAR33 and NKG2A expression on modified NK cells. c Flow cytometry-based CAR33 surface expression analysis over time (n = 6–9). Mean ± SD. CAR33 vs CAR33-KLRC1^ko: d14 (n = 0.1158), d21 (n = 0.1296), d28 (n = 0.8095). d Flow cytometric analysis of NKG2A expression following CRISPR/Cas9 knockout of the KLRC1 gene (n = 7). Mean ± SD. e Frequency of KLRC1 disruption on genomic level was evaluated by Inference of CRISPR Edits (ICE) and T7E1 assay for CAR-transduced (+CAR) and control NK cells (−CAR) (n = 3). Mean ± SD. f Insertion/deletion (indel) distribution profiles were shown for ICE analysis from three different donors (D1, D2, D3). g Expansion-fold and workflow of non-transduced (NT)-NK, KLRC1^ko-NK, CAR33-NK and CAR33-KLRC1^ko-NK cells generation in the presence of IL-2 (500 U/mL) and IL-15 (10 ng/ml Miltenyi Biotec or 50 ng/mL CellGenix) (n = 7). Mean ± SD (ns = 0.9991). Statistical analysis was performed using two-way ANOVA (c, g) and paired Student’s t test (d, e). The entity of n is biological replicates (from different healthy donors) (c–e, g).

Fig. 2. CAR33 expression and KLRC1 knockout induce changes in NK cell gene and but not in surface marker expression profiles.

a Scheme of the sorting strategy for NK single-cell analysis of the four different NK cells preparations. CAR-NK cells were sorted on CAR+, KLRC1^ko-NK cells on NKG2A- and CAR33-KLRC1^ko-NK cells on CD33+/NKG2A− NK cells prior to CITE-Seq analysis. b-d Overall transcriptional (RNA) changes in NK cells upon introduction of CAR33 (b), knockout of KLRC1 (c) and synergistic effects of CAR33 and knockout of KLRC1 (d). Statistical test: quasi-likelihood (QL) F-test against threshold (b–d). Volcano plots show combined analysis of NK cells from donors D1 and D2 and indicate adjusted p-value (y-axis) and inferred logFC (x-axis) for each gene feature and for each genetic modification (CAR33 (b), KLRC1^ko (c), synergistic effect of CAR33 and KLRC1^ko (d)). Up- and down-regulated features are highlighted in color. The horizontal line indicated an FDR of 5% and the vertical lines delimit the range of logFC between −log2 (1.2) and log2 (1.2). Selected genes of interest are labeled. Statistical test: quasi-likelihood (QL) F-test against threshold. e Distribution of surface marker expression from CITE-seq data for the four cell preparations. Distributions are displayed as density, with an area under the curve normalized to 1. f Surface expression of different receptors of NK cells measured by flow cytometry. Representative histograms are shown.

Fig. 3. CAR33- and CAR33-KLRC1^ko-NK cell pools contain an increased fraction of cells with activated, mature NK cell phenotype.

a t-SNE representation of NK cells at single-cell level according to their combined RNA and protein expression profile. Cells are colored and faceted according to the different conditions (NT-NK, KLRC1^ko-NK, CAR33-NK and CAR33-KLRC1^ko-NK cells). b Principal component analysis of differently modified NK cell expression profiles following CITE-seq of two different donors. c Pooled subclustering of NK cell expression profiles from all conditions (t-SNE representation as in (a)). Louvain clustering on PCA results. d Relative abundance of cells belonging to the six clusters as percentage of all sequenced single cells. e Relative abundance of cells belonging to the 6 subclusters as fraction of the NK cell pools. For each NK cell pool (NT-NK, KLRC1^ko-NK, CAR33-NK and CAR33-KLRC1^ko-NK), the relative abundances over all clusters sum up to one. f Differential expression score of selected gene or protein features per cluster. The gene and protein features are grouped according to function (Checkpoint & Suppressive Regulators, Downstream TCR signaling, Lymphocyte Activation, Mature NK Cells). Each dot represents one gene or protein feature. The differential expression score of feature X in cluster Y is the mean expression of feature X over all cells in cluster Y divided by its mean over all cells in all other clusters (excluding cluster Y, displayed as log10). A score above zero indicates upregulation of a particular feature in a given cluster compared to the other clusters. Black squares indicate median differential expression score of its features (statistics are medians over genes). g Grouping of genes and proteins according to their function. A description of each group and their associated features is shown. a, c–f Representative analysis of cells from donor D1.

Fig. 4. KO of inhibitory receptor NKG2A (KLRC1) boosts anti-AML activity of CAR33-KLRC1^ko-NK cells in vitro.

a Short-term (4 h) and long-term (24 h) flow cytometry-based killing assay of NK cells against CD33⁺/HLA-E⁺ OCI-AML2 cells at indicated E:T ratios (n = 5). Mean ± SD. b Dynamic monitoring of CAR33-KLRC1^ko-NK-mediated killing using an IncuCyte-S3 imager. CAR33-KLRC1^ko-NK cells were co-cultured with GFP⁺ OCI-AML2 cells for 26.5 h and the fluorescence emission was measured over time (OCI-AML2 only condition n = 3; rest n = 4). Median + range. c Representative images taken after 24.5 h of IncuCyte analysis of NT-NK, KLRC1^ko-NK, CAR33-NK and CAR33-KLRC1^ko-NK cells co-cultured with GFP⁺ OCI-AML2 cells (E:T = 0.5:1). Viable tumor cells are shown in green based on their GFP expression. Apoptotic tumor cells are labeled red by Annexin V staining. For imaging, the IncuCyteS3 platform was used. d Caspase-cleavage in survived and sorted OCI-AML2 cells following 2 h NK cell co-culture was analyzed using western blot (one representative experiment of 3 is shown; “clvd” = “cleaved”, “ctrl” = “control”). e qPCR gene expression analysis of NT-, KLRC1^ko-, CAR33- and CAR33-KLRC1^ko-NK cells following 2 h co-cultured with OCI-AML2 cells (E:T = 3:1) (n = 5). Mean of technical triplicates ± SD. Statistical analysis was performed by two-way ANOVA (a), paired Student’s t test (b), paired Wilcoxon (e). The entity of n is biological replicates (from different healthy donors) (a, b, e).

Fig. 5. HLA-E is a homogenously expressed target in AML-patients and CAR33-KLRC1^ko-NK cells demonstrated superior killing capacity against patient-derived blasts cells ex vivo.

a Scheme of workflow for AML patient preparation and co-incubation with healthy donor NK cells followed by functional read out of lysed patient-derived bone marrow (BM) cells. b Rank plot-visualization of coefficients of variation (CV). Every point is a protein’s CV ranked by increasing order. HLA-E is highlighted. c PCA scatter plot of the first two principal components (PCs) colored by HLA-E intensity. Every point is a sample. d Heatmap of adjusted -values from a non-parametric Kruska–Wallis test comparing HLA-E intensity over categories: de novo AML (de novo vs. secondary vs. therapy-related, adj. p-value = 1.0), NPM1 (mutated vs. wild type, adj.p-value = 1.0), FLT3 (ITD vs. TKD vs. wild type, adj.p-value = 1.0), FAB (MO to M7, adj.p-value = 1.0), Age group (<50 vs. 50-65 vs. >65, adj.p-value = 1.0), ELN 2017 (favorable vs. intermediate vs. adverse, adj.p-value = 1.0), Sex (female vs. male, adj.p-value = 0.37). e CD33 expression, HLA-E expression and CD45^dim AML-blast expression of one representative primary BMC sample of AML patient day one post-thawing (one day before co-culture with NK cells) were analyzed by flow cytometry. f, g 4 h flow-cytometry-based killing assay of NK cells against primary AML patient material (f) and material from patients with high-risk molecular subsets (g). Due to partially high rates of spontaneous lysis of patient material post-thawing, the specific killing of viable AML cells is displayed. Mean of technical replicates of AML cell lysis by different NK cell donors (Donor 1-20) are shown for each AML patient #1–10 (f, g).

Fig. 6. CAR33- and CAR33-KLRC1^ko-NK cells show preserved activation state and distinct gene expression profiles following AML cell contact.

a Scheme of the sorting strategy for NK single-cell analysis post-co-culture is shown. CAR-NK cells were sorted on CAR⁺, KLRC1^ko-NK cells on NKG2A^- and CAR33-KLRC1^ko-NK cells on CD33⁺/NKG2A^- NK cells following 2 h co-culture with OCI-AML2 cells. b Overall transcriptional (RNA) changes in all NK cell pools after co-culture with OCI-AML2 cells. Volcano plot as in Fig. 2b–d. c Heatmap showing the logarithmic fold change of selected differentially expressed genes and proteins after contact with OCI-AML2 cells compared to before contact in each NK cell pool. A positive logFC represents an upregulation of the gene after co-culture when compared to its expression before co-culture. d Pooled subclustering of single cells from all NK cell pools after co-culture (t-SNE representation). Louvain clustering on PCA results. e Dot plots illustrating expression of genes and surface markers of interest across the 6 subclusters of NK cells after co-culture with OCI-AML2 cells. Color scale: the unit of measurement is average of log-transformed normalized expression values. f Relative abundance of cells belonging to the 6 subclusters as fraction of the NK cell pools after co-culture. For each NK cell pool (NT-NK, KLRC1^ko-NK, CAR33-NK and CAR33-KLRC1^ko -NK), the relative abundances over all clusters sum up to one. g Expression values of IFNG, TOP2A and AURKB are compared between the 4 conditions (NT-NK, KLRC1^ko-NK, CAR33-NK and CAR33-KLRC1^ko-NK) of different NK cell pools before and after co-culture with OCI-AML2 cells. h Heatmap showing the logarithmic fold change of IFNG, TOP2A and AURKB and proteins after contact with OCI-AML2 cells compared to before NK cell contact in each NK cell pool (as in (c)). i INF-γ levels in supernatants of 4 h co-cultures of NK cells with OCI-AML2 (n = 3, biological replicates from different healthy donors). j Analyses of NK cell receptor expression after 24 h of co-culture with OCI-AML tumor cells was measured by flow cytometry (n = 3 donors). Mean ± SD. (b) shows combined data on donors D1 and D2. (c–g) show the representative analysis of cells from donor D1. Statistical analysis was performed by paired Student’s t test (i) and two-way ANOVA (j).

Fig. 7. Low doses CAR33-KLRC1^ko-NK cells in an OCI-AML2-xenograft NSG-SGM3 mouse model show improved efficacy compared to KLRC1^ko-NK or CAR33-NK cells.

a Scheme of the in vivo evaluation of a triple injection treatment with CAR33-KLRC1^ko-NK cells (three doses with 5 × 10⁶ cells intravenously) together with a subcutaneous treatment with IL-2 in OCI-AML2 (Luc⁺) xenograft NSG-SGM3 mouse model. b BLI images of differently treated OCI-AML2 (Luc⁺) engrafted NSG-SGM3 mice over time (n = 4–9). Mice received three doses of 5 × 10⁶ NK cells on day 3, day 7 and day 10 post-AML cell injection. c Total flux analysis (photons/second) mice (n = 4–9). Mean ± SD. d Survival and leukemic burden of OCI-AML2 engrafted mice was observed over 36 days. e Scheme of the in vivo evaluation of a double injection treatment with CAR33-KLRC1^ko-NK cells (two doses with 3 × 10⁶ cells intravenously) together with a subcutaneous treatment with IL-2 in OCI-AML2 (Luc⁺) xenograft NSG-SGM3 mouse model. f BLI images of differently treated OCI-AML2 (Luc⁺) engrafted NSG-SGM3 mice over time (n = 3–4). Mice received two doses of 3 × 10⁶ NK cells on day 3 and day 10 post-AML cell injection. g Total flux analysis (photons/s) mice (n = 3–4). Mean ± SD. h Scheme of in vivo bone marrow (BM) re-engraftment experiment. BM cells of NK treated animals (previously untreated (UT), non-transduced (NT)-NK, KLRC1^ko-NK, CAR33-NK and CAR33-KLRC1^ko-NK treated animals) were isolated at day 18, pooled, and injected into new NSG-SGM3 mice (5 × 10⁶ cells/animal; n = 2–6). i Survival and leukemic burden of BM re-engrafted mice was observed over 120 days (animal protocol: endpoint of the experiment). j Representative confocal microscopy analysis of GFP⁺ AML cells in BM histology d18 post-AML cell injection. Statistical analysis was performed two-way ANOVA (c, g) and Kaplan–Meier (Log-rank (Mantel–Cox) test) (d, i).