Adaptive single-KIR+NKG2C+ NK cells expanded from select superdonors show potent missing-self reactivity and efficiently control HLA-mismatched acute myeloid leukemia

Background Natural killer (NK) cells hold great promise as a source for allogeneic cell therapy against hematological malignancies, including acute myeloid leukemia (AML). Current treatments are hampered by variability in NK cell subset responses, a limitation which could be circumvented by specific expansion of highly potent single killer immunoglobulin-like receptor (KIR)+NKG2C+ adaptive NK cells to maximize missing-self reactivity. Methods We developed a GMP-compliant protocol to expand adaptive NK cells from cryopreserved cells derived from select third-party superdonors, that is, donors harboring large adaptive NK cell subsets with desired KIR specificities at baseline. We studied the adaptive state of the cell product (ADAPT-NK) by flow cytometry and mass cytometry as well as cellular indexing of transcriptomes and epitopes by sequencing (CITE-Seq). We investigated the functional responses of ADAPT-NK cells against a wide range of tumor target cell lines and primary AML samples using flow cytometry and IncuCyte as well as in a mouse model of AML. Results ADAPT-NK cells were >90% pure with a homogeneous expression of a single self-HLA specific KIR and expanded a median of 470-fold. The ADAPT-NK cells largely retained their adaptive transcriptional signature with activation of effector programs without signs of exhaustion. ADAPT-NK cells showed high degranulation capacity and efficient killing of HLA-C/KIR mismatched tumor cell lines as well as primary leukemic blasts from AML patients. Finally, the expanded adaptive NK cells had preserved robust antibody-dependent cellular cytotoxicity potential and combination of ADAPT-NK cells with an anti-CD16/IL-15/anti-CD33 tri-specific engager led to near-complete killing of resistant CD45dim blast subtypes. Conclusions These preclinical data demonstrate the feasibility of off-the-shelf therapy with a non-engineered, yet highly specific, NK cell population with full missing-self recognition capability.


INTRODUCTION
Natural killer (NK) cells can mediate strong antitumor immunity, which is being explored in immunotherapy strategies for human malignancies. Adoptive transfer of allogeneic WHAT IS ALREADY KNOWN ON THIS TOPIC ⇒ Adoptive natural killer (NK) cell therapy in combination with hematopoietic stem cell transplantation can provide cure for acute myeloid leukemia (AML) but is in part dependent on variable NK cell reactivity. Adaptive NK cells are a highly potent NK cell subset that can be used to maximize 'missing-self' reactivity against tumor cells.

WHAT THIS STUDY ADDS
⇒ It describes a novel GMP-compliant protocol to expand clinically relevant numbers of single self-killer immunoglobulin-like receptor + adaptive NK cells from third-party 'superdonors' that provide strong alloreactivity in a mouse model of AML as well as against primary AML blasts ex vivo.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
⇒ These preclinical data demonstrate the feasibility of cell therapy with a non-engineered and yet highly specific NK cell population with maximized missingself recognition.

Open access
NK cells has a good safety profile without cytokine release syndrome, neurotoxicity, or GvH disease and has shown efficacy for subgroups of refractory leukemia patients. [1][2][3][4][5] Current approaches to generate NK cells for therapy include sourcing and expanding NK cell lines, donorblood and cord-blood NK cells, and inducible pluripotent stem cell (iPSC)-derived NK cells using feeder cell lines and cytokines. 6 Despite promising results, the persistence, and variable antitumor responses of different NK cell subsets along with high diversity of NK cell repertoires in healthy donors, 7 represents a potential limitation of donor-derived NK cell-based therapy strategies. Identifying biology-driven approaches that can enhance these attributes prior to adoptive transfer and establishing selection procedures for optimal donors therefore represent key bottlenecks in the field. Early observations showed that NK cell cytotoxicity was triggered by tumor cells that lacked expression of some or all self-MHC class I molecules, referred to as 'missing-self' recognition. 8 Therefore, it was postulated that NK cells might be particularly effective when transferred across HLA barriers, 9 10 where NK cell alloreactivity would be unleashed through mismatch between donor NK clones bearing inhibitory killer immunoglobulin-like receptor (KIR) specific for self-HLA class I molecules ('self-KIR') and recipient cells lacking the cognate HLA class I ligands.
For such an NK cell-mediated graft-versus leukemia (GvL) effect to take place, there must be a mismatch in any of the three major KIR-KIR ligand (KIR-L) settings (KIR2DL1 missing HLA-C2, KIR2DL3 missing HLA-C1, or KIR3DL1 missing HLA-Bw4). The importance of NK cell alloreactivity was elegantly explored in landmark studies by the Velardi group, where the size of the alloreactive NK subset was linked to improved outcomes in patients, both in hematopoietic stem cell transplantation (HSCT) and later in adoptive NK cell transfer in acute myeloid leukemia (AML). 2 10 11 Studies investigating a KIR-L mismatch effect have not always been conclusive though, which may in part be related to the variable frequency of the alloreactive NK cell subset, 12 in turn due to the stochastic expression of KIR. 13 Therefore, selectively expanding the alloreactive subset prior to adoptive NK cell transfer to increase missing-self reactivity 14 would be greatly desired and could lead to clinical benefits.
NK cell differentiation spans from naive CD56 bright cells to highly differentiated CD56 dim adaptive NK cells induced by cytomegalovirus (CMV) infection. 15 16 Adaptive NK cells are characterized by the expression of the activating receptor CD94/NKG2C (NKG2C), whose cognate ligand HLA-E also binds to the inhibitory receptor CD94/NKG2A (NKG2A). Engagement of NKG2C by peptide-stabilized HLA-E can elicit effector functions and support adaptive NK cell expansion in vitro. [17][18][19][20] Apart from NKG2C, the adaptive NK cell signature is characterized by a unique surface phenotype, distinct patterns of transcription factors and signaling molecules as well as altered DNA methylation patterns. [21][22][23] Adaptive NK cells have a high effector molecule content and an increased capacity for IFN-γ production and antibody-dependent cellular cytotoxicity (ADCC). 18 21 24 25 Intriguingly, their unique and preferential expression of a single self-KIR 18 make them especially attractive for strategies aiming at exploiting missing-self recognition since they will demonstrate predictable alloreactivity in HLA-mismatched settings. In allogeneic HSCT, reactivation of CMV and concurrent expansion of adaptive NK cells has been associated with remarkable reductions in AML relapse rates, [26][27][28][29] highlighting their clinical relevance and potential.
We have previously developed an HLA-E feeder-based platform for selective expansion of adaptive NK cells, leading to highly pure populations with potent alloreactivity against primary acute lymphoid leukemia. 30 Here, we used CD3/CD19-depleted cryopreserved cell material from 'superdonors' that harbored large NKG2C + adaptive NK cell subsets with single self-KIR expression, to generate high expansion rates of adaptive NK cells. ADAPT-NK cells largely maintained their adaptive signature, and demonstrated strong alloreactivity against tumor cell lines, in an in vivo AML model and against primary AML samples. Finally, we show that a combination of ADAPT-NK cells with an immune engager enhances reactivity against more resistant AML blast subtypes, broadening the therapeutic utility of the ADAPT-NK cell platform.

METHODS
Additional Methods can be found in online supplemental materials.
Human participants and cell processing Healthy CMV-seropositive blood donors with large preexisting adaptive NK cell subpopulations (>20% of NKG2C + with single self-KIR2DL1 or KIR2DL3 expression, denoted as 'superdonors') were included in the study. Peripheral blood monunuclear cells (PBMCs) were isolated from buffy coats using Lymphoprep and CD3/CD19-depleted as per manufacturer's instructions (Miltenyi Biotec). PBMC were also obtained from leukapheresis products that were subsequently CD3/CD19depleted by CliniMacs (Miltenyi). Depleted cell products were cryopreserved in 90% heat-inactivated fetal bovine serum (FBS,Sigma-Aldrich) or 90% autologous plasma with 10% DMSO until use. To determine donor HLA-C type, DNA was isolated using a commercial kit (DNeasy Blood & Tissue, Qiagen) and the KIR HLA ligand kit was used (Olerup SSP). CMV serology was determined using an ELISA-based assay on plasma obtained during sample preparation. Purified nuclear CMV antigen (AD 169) was used, and the cut-off level for seropositivity was an absorbance of ≥0.2 at a dilution of 1/100.
ADAPT-NK cell expansion protocol CD3/CD19-depleted PBMC were co-cultured with 100 or 200 Gy irradiated K562 feeder cells transfected with lentiviral construct to express high levels of HLA-E with an HLA-G-leader-derived peptide (Fate Therapeutics, details Open access in 'Genetic cell engineering') at a 1:2 ratio in G-Rex24 plates (Wilson Wolf) at 0.5×10 6 total cells/cm 2 . Cells were cultured in GMP-grade Stem Cell Growth Medium (Cell-Genix) supplemented with 100 IU/mL human recombinant IL-2 (Proleukin) for 11 days with 60% medium exchange on day 7, and IL-2 addition days 4, 7 and 10.
Flow cytometry and mass cytometry NK cell phenotype and function was evaluated using standard protocols for flow-and mass cytometry. Samples were acquired on a BD LSRII, or LSR-Fortessa 18-color flow cytometer (BD Biosciences) or on a Helios CyTOF instrument (Fluidigm), and data were analyzed with FlowJo software V.10 (BD Biosciences) and R (R Core Team, 2019). Details are available in online supplemental materials including flow cytometry antibodies (online supplemental sTable 1) and mass cytometry antibodies (online supplemental sTable 2). Cellular indexing of transcriptomes and epitopes by sequencing CITE-Seq CITE-seq was performed as outlined in Biolegend 'Total-SeqTM-A Antibodies and Cell Hashing with 10X Single Cell 3' Reagent Kit V.3 3.1 Protocol' with minor modifications. Briefly, cells were stained with CD56-biotin (REA 196), followed by TotalSeq antibodies and TotalSeq streptavidin-PE and Live/Dead Aqua (Invitrogen) and subsequently sorted for viable CD56 + cells by flow cytometry (online supplemental sFigure 2A,B). To distinguish day 0 and day 11 cells, streptavidin-PE conjugated to different oligos were used, and equal number of sorted viable CD56 + cells from the two cell preparations were pooled as one sample. This was followed by standard 10X Genomics library preparation and sequencing workflow (Genomics Core Facility OUH/UiO, Oslo, Norway). Sequencing was performed with the recommended read lengths on the NextSeq500 sequencer (Illumina) and detailed description of the analysis of CITE-seq data is available in online supplemental materials. For network analyses, the DifferentialNet database 31 (integrated data from experimentally detected protein-protein interactions (PPIs), and RNA sequencing gathered by the Genotype-Tissue Expression consortium) was used to generate tissue-specific (whole blood) PPI networks that were visualized with NetworkAnalyst V.3.0. Enrichment analyses were performed in NWA3.0 with the Gene Ontology (Biological Processes) database.
AML patient samples EDTA blood samples were collected at diagnosis from AML patients (Dep. of Hematology, Oslo University Hospital, Oslo, Norway). PBMC were isolated by Lymphoprep and cryopreserved in RPMI 1640 with 60% FBS and 10% DMSO. Patients were genotyped by Illumina NGS platform (Dep. of Immunology, Oslo University Hospital, Oslo, Norway) and for this study, samples were selected from patients of either HLA-C1/C1 or HLA-C2/ C2 haplotypes.
Flow cytometry-based NK cell functional assays NK cell degranulation and cytokine production was evaluated by mixing ADAPT-NK products at different effector to target (E:T) ratios with tumor targets in 6 hour co-cultures. To measure ADCC, anti-CD20 (MabThera, 1 µg/ mL) was added to co-cultures with 721.221 cells. Phorbol-12-myristate-13-acetate (50 ng/mL) + ionomycin (1 µg/ mL) (Sigma) was used as positive control. In cytotoxicity assays, target cells were pre-stained with CellTrace Violet (Invitrogen), and FITC-DEVD-FMK (Abcam) to detect active Caspase-3, was added at the start of the incubation. Dead/dying cells were defined as CellTrace + Caspase-3 + and/or dead cell marker + and specific cytotoxicity was calculated as: (% dead experimental -% dead target only ) ÷ (100% % dead target only ) x 100%. For competitive cytotoxicity assays, K562 expressing an HLA-C1-or HLA-C2-dimer were stained with two different concentrations of Cell-Trace Violet prior to being mixed at a 1:1 ratio and subsequently seeded at different E:T ratios with ADAPT-NK cells. Cytokine stimulations with IL-12 (10 ng/mL) and IL-18 (10 ng/mL) (both Biotechne) were performed for 25 hours. Further experimental details are provided in online supplemental methods.
Genetic cell engineering Engineered cell lines used in this study are described in detail in online supplemental material. Briefly, for the expansion of ADAPT-NK cells, K562 cells (ATCC Cat: CCL-243) were engineered by Fate Therapeutics using a third-generation lentiviral transfer plasmid designed to contain a chimeric protein with the HLA-G leader peptide (1-24) and the mature HLA-E*0103  driven by an EF1α promoter. For additional HLA-C and HLA-E variants, K562 were engineered in-house using VSV-G-pseudotyped lentiviral particles with a mammalian LeGO-G2 expression vector, to express the following synthetic proteins: β2m-HLA-C1 (HLA-C*07:01) single chain dimer (HLA-C1-dimer); β2m-HLA-C2 (HLA-C*04:01) single chain dimer (HLA-C2-dimer) and HLA-G 3-11 -β2m-HLA-E*01:01 single chain trimer (HLA-E-trimer). For HLA-E knock-out in NALM-6 cells, Cas9 and a pool of synthetic guide RNAs (sgRNA) (CRISPRevolution sgRNA EZ Kit, Synthego) were used. Single cells were sorted on a FACS Aria II (BD Biosciences) to establish multiple clones. DNA sequencing and flow cytometry confirmed HLA-E depletion in the selected clones used Open access in the study. HLA-E sgRNAs further targeted the HLA-C locus resulting in a combined KO for NALM-6.

Serial killing assays
The cytotoxic potential of individual NK cells was evaluated using a previously described microwell chip screening assay. 32 Briefly, labeled NK cells and labeled target cells were seeded in a microwell chip containing complete RPMI supplemented with Sytox Green (Thermo Fisher). Cells stochastically distributed in the 8064 60-µm-wide wells at an average E:T ratio of 1:5. The co-cultures were imaged every 3 hours for 15 hours using a LSM 880 (Carl Zeiss AG) inverted confocal microscope. The images were processed using a custom-built MATLAB script whereby live and dead cells were quantified at all time points, and for each condition a minimum of 500 wells containing a single NK cell at start along with at least 4 live target cells, were included for analysis. Further details are available in online supplemental material.

IncuCyte measurement of tumor killing
Tumor killing was measured in real-time using the Incu-Cyte S3 platform. Detailed description of the protocol is available in online supplemental material. Briefly, target cells stably expressing NucLight Red (Essen Biosciences) were overnight rested and subsequently co-cultured with ADAPT-NK cells at different E:T ratios. Images (3/well) from at least two technical replicates for each condition were acquired every 90 min for 48 hours, using a ×10 objective lens and analyzed by IncuCyte Controller v2020A (Essen Biosciences). Graphed readouts represent percentage live target cells.
In vivo AML tumor model Detailed description of the protocol is available in online supplemental material. Briefly, NOD.Cg-PrkdcscidIl2rgt-m1Wjl/SzJ mice (NSG, Jackson Laboratories) mice were injected intravenously with HL-60 cells stably expressing firefly luciferase (1.5×10 6 /mouse). Four control mice on the same background received no injections. After allowing tumors to engraft for 4 days, bioluminescence imaging was performed, and mice were randomized into three groups: tumor alone, HLA-C/KIR matched ADAPT-NK cells (flat-dose 5×10 6 /mouse) or HLA-C/ KIR mismatched ADAPT-NK cells (flat-dose 5×10 6 / mouse). Mice receiving NK cell injections were also injected intraperitoneally with IL-15 (National Cancer Institute, 6 ug/mouse) twice weekly for 3 weeks. Bioluminescence imaging was performed weekly to track tumor burden using the IVIS Spectrum In Vivo Imaging System (Perkin-Elmer).

Statistical analyses
Statistical analyses and visualization were performed using Prism V.9 (GraphPad) and R with the ggplot2 and dunn. test packages. Paired Student's t-tests were used for paired comparisons. Repeated measures one-way analysis of variance (ANOVA) with Tukey's correction was used to analyze three groups (flow, mass-cytometry and CITE-Seq). One-way ANOVA with Šidák's multiple comparison correction was performed for NK cytotoxicity assays evaluated by flow cytometry and in the IncuCyte platform. Two-way ANOVA with Tukey's multiple comparison correction was performed to analyze the in vivo AML tumor model. Nonparametric Spearman-rank correlation was performed for evaluation of HLA-E in AML patient samples and NK cell cytotoxicity against these.

RESULTS
Efficient expansion of single-self-KIR + NKG2C + adaptive NK cells (ADAPT-NK) NKG2C + adaptive NK cells tend to express only a single self-KIR, a fact that translates to high alloreactive potential if transferred to recipients with KIR-L (HLA-C) mismatch. 18 30 Re-examining data from our previously published cohort of healthy blood donors, 18 48/202 donors had a large (over 20%) NKG2C + adaptive NK cell subset, and of those, 26 had a clear single-self KIR2DL1 + KIR2DL3or KIR2DL1 -KIR2DL3 + expression profile (figure 1A,B) and were subsequently denoted as 'superdonors'. Based on CD3/CD19-depleted material from these select superdonors, a GMP-compliant protocol to expand adaptive NK cells was developed (figure 1C). Monocytes and feeder cells were undetectable in the co-cultures by day 7, concomitant with a rise in NK cell numbers (figure 1D,E). On day 11, a median of 91% of NK cells were NKG2C + and 81% had single-self-KIR expression (either KIR2DL1 + KIR2DL3or KIR2DL1 -KIR2DL3 + , figure 1F,G). Hereafter, we refer to this expanded adaptive NK cell product as 'ADAPT-NK' cells. While CD57 expression declined slightly, less than 20% of cells expressed NKG2A post-expansion (figure 1F,G). Replicate expansions of cell material obtained from the same donor and from several donations, gave stable results (median 88% NKG2C + , SD 3.64% figure 1H). We achieved a median of 470-fold expansion of adaptive NK cells with some donors displaying over 1000-fold expansion rates (figure 1I). Thus, the developed expansion protocol produced high fold changes of single self-KIR + NKG2C + adaptive NK cells in 11 days, opening an avenue for clinical development.
ADAPT-NK cells retain an adaptive transcriptional signature with activation of effector programs and without signs of exhaustion To assess the adaptive state in NKG2C + NK cells preexpansion and post-expansion, we performed extended phenotyping by flow cytometr and mass cytometry (figure 2 and online supplemental sFigure 1) as well as cellular indexing of transcriptomes and epitopes by sequencing (CITE-Seq) (figure 2 and online supplemental sFigure 2). For mass cytometry, gating was performed on viable CD56 + NK cells and visualized in t-SNE plots, with distinct topological regions corresponding to day 0 and day 11 NK cells (figure 2A and online supplemental sFigure 1A). On day 0, NKG2C + cells displayed low expression of NKp30, Siglec-7, PLZF, CD38

Open access
and high expression of CD2 as compared with NKG2Cconventional NK (cNK) cells in accordance with previously published phenotypes. 22 33 In addition to NKG2C and single-KIR on ADAPT-NK cells, key adaptive markers such as CD16, CD2 and Siglec-7 were similarly expressed post-expansion whereas CD38, NKp30, PLZF and FcεRIγ were upregulated as compared with resting NKG2C + adaptive NK cells (  Open access compared with KLRC2 -cNK ( figure 2D). Post-expansion, KLRC2 + ADAPT-NK had 176 downregulated and 221 upregulated genes as compared with baseline KLRC2 + adaptive NK (figure 2E). Consistent with published transcriptomes for adaptive NK cells, 23 34 baseline KLRC2 + NK cells had higher expression of LAG3, PROK2 and MCOLN2 and lower expression of FCER1G, IL12RB2, XCL2, CCNJL and SOX4 as compared with KLRC2 -cNK. Postexpansion, KLRC2 + ADAPT-NK cells showed upregulation of FCER1G and IL12RB2 whereas expression of other adaptive signature genes was similar ( figure 2F).
Next, the transcriptional signature with 397 differentially expressed (DE) genes for KLRC2 + ADAPT-NK cells postexpansion (figure 2E) was interrogated. PPI analysis with the upregulated DE genes as a base resulted in a network of 208 elements (nodes; figure 2G), which was queried for pathway enrichment revealing lymphocyte activation and differentiation as highly upregulated processes (figure 2H, online supplemental sFigure 2C and online supplemental sTable 3). Using the downregulated DE genes as a base resulted in a network of 173 elements (nodes; figure 2I) where regulation of cell metabolism and proliferation dominated (figure 2J, online supplemental sFigure 2D and online supplemental sTable 4). Several DE genes annotated to cytokine-mediated signaling pathway and regulation of apoptosis were discovered among the observed downregulated and upregulated processes. Notably, despite a phenotypic resemblance to exhausted T cells, 35 cell exhaustion or senescence were not among the enriched processes and the overlap to transcriptional T cell exhaustion signatures was minimal (online supplemental sFigure 2E,F). Taken together, this reveals that expanded ADAPT-NK cells largely retained their adaptive state with evidence of metabolic and effector cell activation without signs of exhaustion.

ADAPT-NK cells are highly functional with predictable alloreactivity
To elucidate the functional capacity of the ADAPT-NK cells, we next explored responses to a range of stimuli. We observed strong ADCC against anti-CD20-coated 721.221 B-cell lymphoma targets but also robust IFN-γ production in response to IL-12/IL-18 stimulation ( figure 3A,B), suggesting that ADAPT-NK cells retain an adaptive NK cell functional profile and, in addition, become more responsive to cytokine stimulation. 22 26-28 To investigate the contribution of signaling mediated via NKG2C/HLA-E interactions for ADAPT-NK cell functionality, expanded cells were tested against targets without HLA-E or engineered to express high levels of a β2m HLA-E single-chain trimer with forced presentation of the HLA-G leader peptide. K562 and NALM-6 target cells with high levels of HLA-E trimer elicited strong responses from ADAPT-NK cells above those of K562 wildtype (K562wt) or NALM-6 HLA-E KO, showing that the NKG2C receptor is signaling and contributes to the functional response of ADAPT-NK cells ( figure 3C,D).
To test the HLA-Cspecificity of ADAPT-NK responses, expanded NKG2C + KIR2DL1 + KIR2DL3or KIR2DL3 + KIR2DL1 -ADAPT-NK cells were stimulated with K562wt cells or K562 cells engineered to express high levels of single chain β2m HLA-C1 or HLA-C2 dimers. Strong responses were observed in the HLA-C/ KIR mismatched setting but ADAPT-NK cells were inhibited in the matched setting (figure 3E,F and online supplemental sFigure 3A). In mixed target cell assays, ADAPT-NK cells preferentially killed HLA-C mismatched targets ( figure 3G,H). Assessing the cytotoxic potential of individual NK cells in a microwell screening assay, 32 there were significantly more kills per cytotoxic cell in the HLA-C/KIR mismatched setting and enhanced serial killing ability ( figure 3I-K). These results revealed that ADAPT-NK cells efficiently convert the HLA-E check point into an activation signal and show specific and predictable alloreactivity against K562 targets expressing high levels of mismatched HLA-C.

Efficient alloreactivity against tumor cell lines and in vivo efficacy of ADAPT-NK cells
The importance of alloreactivity for tumor targeting was apparent when assessing the ADAPT-NK cell product. We next tested the long-term killing ability against tumor cell lines of various HLA-C genotypes but overall low/negative for HLA-E (figure 4A-C and online supplemental sFigure 3B-F). A group of tumor lines displayed negligible expression of HLA-C and were subsequently killed equally well by HLA-C/KIR-matched and mismatched ADAPT-NK cells (online supplemental sFigure 3C-F). Two HLA-C2/ C2 tumor lines, BJAB (Burkitt-like lymphoma) and HL-60 (AML), displayed significant levels of HLA-C2 as assessed by KIR2DL1-Fc binding, and high rates of killing could be observed in the mismatched setting ( figure 4A-C).
To evaluate the in vivo impact of ADAPT-NK cell alloreactivity, NSG mice were injected with luciferase-tagged HL-60 cells followed by a flat dose of 5×10 6 HLA-C/KIRmatched or mismatched ADAPT-NK, under in vivo IL-15 support. The tumor burden was significantly reduced by ADAPT-NK cell transfer for the duration of the experiment, with substantially better leukemia control by HLA-C/KIR mismatched ADAPT-NK cells by day 35 ( figure 4D,E). Together this demonstrates the therapeutic efficacy of alloreactive ADAPT-NK against HLA-C mismatched tumor cells and underscores their ability to deliver a maximized missing-self response in settings with high levels of mismatched HLA-C.

ADAPT-NK cells show potent alloreactivity against mismatched primary AML blasts and combination with an CD33/IL-15/CD16 TriKE overcomes resistant blast subtypes
Thus far, expanded ADAPT-NK cells showed efficient killing of tumor cell lines in vitro and in vivo. We next tested the ability of NKG2C + KIR2DL1 + KIR2DL3or KIR2DL3 + KIR2DL1 -ADAPT-NK cells to recognize and eliminate primary AML blasts in PBMC samples derived from patients with distinct HLA-C genotypes. Patient Open access characteristics, including cytogenetics and HLA-C haplotypes, is displayed in online supplemental sFigure 4A. Corroborating the importance of NK cell alloreactivity, minimal NK cell cytotoxicity was observed in HLA-C/ KIR matched conditions, whereas mismatched CD45 dim AML blasts were efficiently killed at different E:T ratios (figure 5A,B and online supplemental sFigure 4B). The killing of the AML blasts by mismatched ADAPT-NK cells was variable, with some patient samples being more effectively killed and others more resistant. Since the blast compartment of the different samples was heterogeneous, we could monitor putative immune selection events and resistant subtypes. No relation was found between specific cytotoxicity by mismatched ADAPT-NK cells and HLA-DR, CD33, CD34, CD38, CD47, CD112, CD117 or CD155 expression on CD45 dim blasts (online supplemental sFigure 4C). Detected levels of HLA-E on blasts were overall low but correlated with sensitivity to killing (figure 5C).
Lack of NKG2D ligand expression is associated with leukemic stem cells (LSC) and involved in mediating their immune evasion. 36 Although NKG2DL + blasts were eliminated to a higher extent, NKG2DL -LSC were still killed, even at low E:T ratios (figure 5D and online supplemental sFigure 4C). To increase killing of more resistant blast subtypes, we combined mismatched ADAPT-NK cells Open access with a CD16/IL-15/CD33 tri-specific engager (TriKE), previously used to target AML blasts. 37 The expression of CD33 on AML blasts was high and addition of the TriKE resulted in significantly improved killing of AML blasts, even for more resistant samples and NKG2DL -LSC (figure 5E,F and online supplemental sFigure 4C). Thus, ADAPT-NK cells showed high efficacy against primary AML blasts, and due to their strong ADCC capacity, such expanded adaptive NK cells can be combined with immune engagers to overcome partial NK cell resistance of LSC.

DISCUSSION
NK cell-based immunotherapy can be a safe and effective treatment for subgroups of leukemia patients with relapsed or refractory disease. Emerging therapeutic approaches focus on enhancing NK cell cytotoxicity and persistence in vivo and making the therapy more readily available in off-the-shelf strategies. One constituent of the NK antitumor response is alloreactive cells in administered NK cell products that relate to better clinical outcomes. 2 10 11 The preferential expression of a single self-KIR on most adaptive NK cells provides an opportunity to boost the alloreactive response through selective expansion of this NK subset. Adaptive NK cells also inherently display high ADCC capacity and enhanced IFN-γ production. 18 21 24 25 To harness these properties, we developed a single self-KIR + NKG2C + adaptive NK cell product termed ADAPT-NK, using cryopreserved third-party donor material in a fully GMP-compliant protocol. Transcriptional signatures and adaptive features were largely preserved in ADAPT-NK cells post-expansion and the cells exhibited three natural functional modalities without genetic editing: Alloreactivity (single KIR), HLA-E targeting (NKG2C + NKG2A -) and potent ADCC activity (CD16 expression and signaling).
Generation of homogeneous alloreactive NK populations can theoretically be achieved through (1) selection of single-KIR + NK cells and subsequent Open access expansion, 38 (2) guided differentiation of precursors with or without genetic editing or (3), as shown here, through selective outgrowth from a mixed NK cell population. 14 The selective expansion of adaptive NK cells from a mixed NK population avoids any need for GMP-grade selection or editing procedures and therefore reduces the complexity of culturing. Robust cell expansions were generated from all donors with large pre-existing NKG2C + adaptive NK cell subsets, recapitulating the NKG2C phenotype with only slight variation in replicate experiments. This included those donors with almost exclusively CD57 + NKG2C + adaptive NK cells pre-expansion, suggesting that proliferation could be induced in this subset despite the low proliferative capacity of CD57-expressing cells. 39 Of note, and in contrast to many cytokine-driven expansion protocols, there was no increase in the fraction of cells expressing NKG2A. In fact, NKG2A experiments. Statistical differences were tested using a two-way ANOVA followed by Sidak's multiple comparison correction (B, D), using Spearman's correlation (C) or using one-way ANOVA followed by Sidak's multiple comparison correction (F). *p<0.05, **p<0.01, ***p<0.001. AML, acute myeloid leukemia; ANOVA, analysis of variance; E:T, effector to target; KIR, killer immunoglobulin-like receptor; NK, natural killer; t-SNE, t-distributed stochastic neighbor embedding.
Open access expression on NKG2C + ADAPT-NK cells was negligible, resembling NK cell preparations generated by CRISPR editing 40 or through intracellular retention. 41 The NKG2A expression remained low also during the phase of intense proliferation coinciding with the disappearance of HLA-E expressing feeders around day 7, suggesting an inheritable adaptive state. Absence of NKG2A-mediated inhibition with concomitant homogeneous expression of NKG2C on ADAPT-NK cells allowed for efficient reactivity toward HLA-E-expressing targets. Hence, in addition to KIRdriven alloreactivity, the switched HLA-E recognition enables targeting of tumors with high HLA-E levels where NKG2A otherwise provides an inhibitory check point. 42 Adaptive NK cells have a distinct molecular signature including alterations in signaling molecules and surface receptors. 18 22 23 Transcriptional analysis revealed that although FCER1G and IL12RB were upregulated the adaptive imprint was largely intact post ADAPT-NK cell expansion. Notably, CD16 and co-stimulatory CD2 expression, critical for the superior ability of adaptive NK cells to respond to antibodycoated target cells, 33 remained high enabling potent ADCC responses with rituximab and the CD16-IL-15-CD33 TriKE.
Lack of inhibitory input from tumor targets led to ADAPT-NK cell attack in HLA-C/KIR mismatched settings whereas tolerance to self was maintained in matched settings with high HLA-C expression. In addition to NKG2C, the cells displayed high levels of NKG2D, upregulated NKp30 and maintained DNAM-1 levels. They also displayed high levels of effector enzymes such as Granzyme B and perforin, above those of conventional NK cells. Overall, the ADAPT-NK cells showed upregulation of receptors and enrichment of genes related to activation, differentiation, proliferation, cytokine signaling, and regulation of apoptosis and metabolism. Although ADAPT-NK cells expressed surface markers classically associated with T cell exhaustion such as TIM-3, LAG-1 and TIGIT, the transcriptional overlap to exhausted CAR T cells was minimal and did not include the signature marker TOX. 43 44 Exhaustion in the NK cell compartment is still not well defined. The role of PD-1 on NK cells is controversial and TIGIT and TIM-3 may correlate both with activation/acute stimulation and more exhausted states. 45 Most importantly, the ADAPT-NK cells showed potent proliferation and cytotoxicity, also in longer-term in vitro assays and displayed serial killing capacity, demonstrating that they were not functionally exhausted.
AML is a heterogeneous disease with varying blast presentations and subtypes. ADAPT-NK cells were highly efficient against an AML cell line in vitro and in vivo as well as against primary AML blasts. Following a single dose of KIR-L mismatched ADAPT-NK cells, tumor burden was significantly reduced in the HL-60 AML mouse model with superior early limitation of tumor growth. Although this in vivo model is extensively used to assess NK cell function, 37 46 a weakness of the model is the aggressive nature of the tumor with exponential growth as soon as the tumor escapes initial control, which precludes comprehensive survival analyses.
Against primary patient blasts, efficacy of mismatched ADAPT-NK killing was correlated to HLA-E expression on CD45 dim blasts as well as NKG2D-L expression and varied greatly among samples. Previous studies have identified the NKG2DLblasts as particularly NK resistant and able to give rise to new blast populations acting as LSCs. 36 Therefore, it was particularly important to observe the killing of NKG2DLblasts even without stimulatory NKG2D input, and an enhanced effect with the TriKE immune engager. This was evident also for blasts carrying poor-prognosis mutations such as FLT3-ITD. 47 More extensive studies are needed to elucidate resistant and susceptible AML subtypes, also considering the genetic makeup of the AML clone(s), to enable patient stratification in a precision immunotherapy pipeline.
Beneficial features of adaptive NK cells are currently under development from third-party donor cells and engineered iPSC NK cells, mainly focused on NKG2C, FcεRIγ, and CD38 negativity in combination with daratumumab for multiple myeloma. 48 49 Like the 'g-NK cells', 48 ADAPT-NK cells circumvent the need for genetic engineering while relying on preselection of donors harboring adaptive NK subpopulations. Our KIR-centered strategy meant sourcing donors with adaptive NK cells educated by self-KIR2DL1 (HLA-C2/C2 or HLA-C1/C2) or -KIR2DL3 (HLA-C1/C1 or HLA-C1/C2), intended for treating patients with homozygous HLA-C1 or -C2 haplotypes. In preparation for a phase I/II clinical trial for refractory leukemias, we have established a GMPgrade donor cell bank with cryopreserved CD3/ CD19 depleted apheresis products from so far nine superdonors (recruitment ongoing), sufficient to manufacture 350 ADAPT-NK doses in an off-the-shelf setting. Although a complete KIR-L mismatch would be the most appealing approach, there is a risk for off-target cytotoxic effects, necessitating precautions in initial dosing regimens. Another important point to consider is the risk for downtuning (or disarming) of NK cell function following adoptive transfer to an HLA disparate recipient, as has been shown in MHCdeficient mice. 50 In summary, by expanding adaptive single-self-KIR + NKG2C + NK cells in a fully scalable GMP-compliant protocol our study provides the basis to explore alloreactive NK cell therapy. We see potential implications for combination therapies where the ADAPT-NK cell therapy platform is combined with immune engagers or engineering with CARs. If successful, clinical trials based on the ADAPT-NK cells may be able to answer fundamental questions regarding the importance