The AML cell lines MOLM-14, NOMO-1, CMK, and THP-1 and the ALL cell lines REH and BV-173 were obtained from the DSMZ (Braunschweig, Germany) and cultured in RPMI-1640 GlutaMAX™ medium supplemented with 10% fetal bovine serum (FBS) and 1% penicillin and streptomycin (P/S; all from Thermo Fisher Scientific, Waltham, MA, USA), referred to as complete RPMI-1640. Human embryonic kidney cells (HEK293T, DSMZ) were cultured in Dulbecco’s Modified Eagle Medium (DMEM) GlutaMAX™ supplemented with 10% FBS and 1% P/S, referred to as complete DMEM.

Primary human NK cells were isolated from peripheral blood of healthy donors, for which they gave written informed consent according to the protocol (#2019-623) approved by the local ethics committee (Heinrich Heine University, Düsseldorf, Germany). Ficoll density gradient centrifugation (Ficoll-Paque Plus; Cytiva Europe, Freiburg, Germany) was performed to collect peripheral blood mononuclear cells (PBMCs). NK cells were then enriched by immunomagnetic negative selection using MojoSort™ Human NK Cell Isolation Kit (BioLegend, San Diego, CA, USA) according to the manufacturer’s instructions. NK cell purity and phenotype were determined by flow cytometry (MACSQuant^® Analyzer 10, Miltenyi Biotec, Bergisch Gladbach, Germany) on days 0, 7, and 14, using fluorochrome-conjugated antibodies against CD3, CD33, CD56, and CD94 (all REAfinity™ clones from Miltenyi Biotec). NK cells were activated and expanded in NK MACS^® Medium (Miltenyi Biotec) supplemented with 5% heat-inactivated human AB serum (Sigma-Aldrich, Darmstadt, Germany), 1% P/S, 500 IU/ml of IL2 (Proleukin, Novartis, Basel, Switzerland), and 10 ng/ml of IL15 (Miltenyi Biotec) at a concentration of 1–2 million cells/ml.

All CAR constructs were optimized for human codon usage and synthesized by GeneArt (Thermo Fisher Scientific) or BioCat (Heidelberg, Germany). The CARs were co-expressed via a T2A site with the monomeric tag blue fluorescent protein (BFP from Evrogen, Moscow, Russia) under the control of the myeloproliferative sarcoma virus (MPSV) promoter (27) and equipped with our CD34-derived hinge C6 (28), CD28 transmembrane, and co-stimulatory domains as well as the CD3 zeta-chain unit (27). The CD19, CD33, CD123, and epidermal growth factor receptor (EGFR) (clone cetuximab/C225, from now on referred to as Cetux) CARs were previously described (20, 27–29). In the ALL xenograft model, CD19 CARs were co-expressed with codon-optimized soluble human IL15 or human IL15 tethered to the IL15 receptor α-chain (IL15-IL15R). Both IL15 cDNAs used a codon-optimized CD8 leader peptide. The codon-optimized firefly luciferase/enhanced green fluorescent protein (EGFP) fusion (LucEG) (20) and the EGFP/neomycin resistance fusion (EGN) (27) lentiviral expression vectors were previously published. The MPSV, hPGK (human phosphoglycerate kinase 1), modified spleen focus-forming virus (SFFV), and short EF1α (elongation factor 1-α) promoter constructs were also previously published (27). The wild-type SFFV was newly introduced in the pCL6EGNwo vector (27). The optimized EF1α promoter was generated from the long EF1α construct with the wild-type SD and acceptor sites (27) by deleting ~600 bp of the natural intron and mutating an open reading frame (sequences are available upon request).

The production of lentiviral particles in HEK293T cells was performed on 10-cm dishes as previously described (20, 30). For pseudotyping, the following envelopes were used: the vesicular stomatitis virus glycoprotein (VSV-G, 6 µg per transfection), the codon-optimized baboon endogenous virus lacking the R-peptide (BaEV-Rless, 1 µg per transfection), the codon-optimized BaEV surface and transmembrane units fused to the cytoplasmic units of the amphotropic murine leukemia virus (aMLV; BaEV-TR, 1 µg per transfection), the codon-optimized gibbon ape leukemia virus (GALV) surface unit fused to the transmembrane and cytoplasmic units of aMLV (GALV-TM, 1 µg per transfection), or the previously published (31) feline endogenous virus envelope fused to the cytoplasmic units of aMLV (RD114-TR, 6 µg per transfection). Twenty-four hours after transfection, the culture medium was replaced with 10 ml of IMDM (Sigma-Aldrich) supplemented with 10% FBS, 1% P/S, and 1% l-glutamine (Thermo Fisher Scientific). Cell supernatants containing viral particles were harvested 48 h after transfection, filtered through a 0.45-µm filter, and either used directly or after concentration (5-fold) at 10,000 × g for 2 h at 4°C. Lentiviral supernatants were used fresh or stored at −80°C until usage.

After 7 to 10 days of expansion in a complete NK MACS medium, primary human NK cells were transduced with lentiviral particle-containing supernatants. For the envelope, promoter, and entry enhancer testing, 1 to 1.5 × 10⁵ NK cells per well were transduced in flat-bottom 96-well plates in working volumes of 100 µl with serially diluted lentiviral particles encoding EGN. When RetroNectin™ (TaKaRa Bio Inc., Otsu, Japan; from now on referred to as Retronectin) was used as transduction enhancer, non-treated 96-well plates were coated either overnight at 4°C or for 2 h at 37°C prior to use (32, 33). In contrast, Vectofusin^®-1 (Miltenyi Biotec, from now on referred to as Vectofusin) was added at a final concentration of 10 µg/ml to tissue culture-treated 96-well plates. For transductions of higher numbers of NK cells, 1 to 1.5 × 10⁶ NK cells were transduced on Retronectin-coated 12-well plates (1 ml final volume) or 3.5 to 4 × 10⁶ NK cells on Retronectin-coated 6-well plates (2.5 ml final volume) with 500 or 1,250 µl concentrated lentiviral particles, respectively, encoding CD19, CD33, CD123, or Cetux CARs. After 24 h, cells were supplemented with 2 ml (12-well) or 5 ml (6-well) fresh complete NK cell medium. Transduction efficiency was determined 3 to 4 days after transduction by flow cytometry, measuring BFP and/or CD34 (CAR) expression with phycoerythrin (PE)-conjugated QBend-10 CD34 antibody (Thermo Fisher Scientific).

Three to four days after transduction, CAR NK cells were enriched on an OctoMACS™ Separator using MS columns or on a QuadroMACS™ Separator using LS columns after staining with CD34 microbeads (clone QBend-10) according to the manufacturer’s instructions (Miltenyi Biotec) and similarly as described previously for CAR T cells (28). In order to investigate the MACS enrichment efficiency, the CAR NK cell purity and yield, and the expression of the different CAR constructs on the transduced NK cells, the three cell fractions—before MACS, flowthrough, and after MACS—were collected and analyzed by flow cytometry for BFP expression and/or CD34 positivity. Before any functional analyses were performed, the MACS-enriched CAR NK cells were further cultured for at least an additional 2 days in a complete NK MACS medium.

To block the activating receptors, non-transduced and CAR NK cells were incubated with monoclonal REAfinity™ antibodies against NKG2D, DNAM-1, NKp30, NKp44, NKp46, and NKp80 (Miltenyi Biotec) at concentrations of 10 µg/ml in complete RPMI-1640 at 4°C. These REAfinity™ antibodies harbor mutations in their Fc receptors, which prevents their binding to human Fc receptors. Control cells were incubated in parallel with REAfinity™ isotype controls (Miltenyi Biotec). After the 2 h of incubation, the NK cells were added at 3:1, 1:1, and 0.3:1 ratios (without washing) to MOLM-14, NOMO-1, CMK, THP-1, REH and BV-173 cells seeded at 2 × 10⁴ cells/well in a 96-well U-bottom plate in complete RPMI-1640 medium. After 16 h, the cultures were harvested, and leukemic cell lysis was analyzed by flow cytometry, using propidium iodide (Sigma-Aldrich) for live/dead cell discrimination. The leukemic cells in the culture were specifically recognized by either EGFP expression (REH, MOLM-14) or after staining with CD15-FITC or CD33-FITC REAfinity™ antibodies. The samples were analyzed on a MACSQuant^® Analyzer 10 (Miltenyi Biotec). The specific lysis in % was calculated as [1 − number of viable target cells (sample)/number of viable target cells (control)] × 100%.

All animal experiments were approved by the state animal research committee (LANUV, NRW, Germany). The use of leftover primary blasts from the initial AML diagnosis of children or adolescents treated within the AML-BFM study group was authorized by the ethics committee of the medical faculty at the University of Duisburg-Essen (application number 16-7069-BIO). Each AML-BFM trial was previously approved by their local institutional ethics committees: the Hannover Medical School for AML-BFM 2012 trial and registry (application number 13.03.12/La) and the University of Duisburg-Essen’s medical faculty for AML-BFM 2017 registry (application number 17-7462-BO). The primary blasts were injected into 8- to 16-week-old NOD.Cg-Prkdc^SCIDIl2rg^tm1Wjl/SzJ (NSG) mice (Charles River Laboratories, Sulzfeld, Germany) and then sacrificed upon the clinical manifestation of the AML (Füchtjohann, Hanenberg, in preparation, 2022).

For analyzing the in vivo persistence of NK cells, 8- to 10-week-old NSG mice were treated by i.p. injection of 25 mg/kg of busulfan (Busilvex^®, Busulfan, Medac, Darmstadt, Germany) and the next day intravenously transplanted with 10 × 10⁶ NK cells containing 50% transduced and 50% untransduced NK cells. Transduced NK cells co-expressed either soluble human IL15 and EGFP or human IL15 bound/tethered to the IL15 receptor α-chain (IL15-IL15R) and EGFP. Control mice were transplanted with 10 × 10⁶ untransduced NK cells. Each group consisted of 3 animals. Peripheral blood was analyzed at days 5, 10, 20, and 27 after lysis of erythrocytes with BD Pharm Lyse (BD Biosciences, San Jose, CA, USA) on a MACSQuant™ Analyzer 10 for murine CD18, human CD45, CD56, and EGFP-expressing cells (all antibodies were REA clones from Miltenyi Biotec) as described previously (28).

For the engraftment of REH cells expressing a firefly luciferase/EGFP fusion protein (REH^LucEG) (28), 7- to 10-week-old NSG mice were treated by i.p. injection of 25 mg/kg of busulfan and the next day intravenously transplanted with 3.5 × 10⁶ REH^LucEG cells via tail vein injections. Two days later, the animals were randomized into different treatment groups and injected with 3.5 × 10⁶ NK cells containing 50% positive CD19 CAR NK cells co-expressing BFP, IL15, or IL15-IL15R and 50% untransduced NK cells (6 mice per group, 7 mice as untreated control). Luminescence analysis was performed on days 7, 15, 22, and 27 as previously described (20) to assess the progression/proliferation of the ALL blasts. Peripheral blood was obtained at days 8, 15, and 22 via the tail vein and, after lysis of erythrocytes with BD Pharm Lyse, analyzed on a MACSQuant™ Analyzer 10 for EGFP positive REH^LucEG cells and BFP, CD34, CD56, and CD94 NK cells (CD34-PE from Thermo Fisher Scientific, all other antibodies from Miltenyi Biotec).

MACS-enriched CAR NK cells co-expressing BFP were stained with the QBend-10 CD34-PE antibody (as described above) and then co-cultured with EGFP-expressing REH cells in complete DMEM at an NK cell to target cell ratio of approximately 0.3:1. For these experiments, gelatin-treated 8-well slides with glass bottom (ibidi GmbH, Gräfelfing, Germany) were used. Bright-field and epifluorescence images were acquired at 37°C for 1–4 h after the seeding using an IX83 Inverted Microscope (Olympus, Hamburg, Germany), a 100× oil objective (NA 1.49, UAPON100XOTIRF, Olympus, Hamburg, Germany), and the cellSens Dimension software (Olympus, Hamburg, Germany). Images were processed using the open-source Fiji software (https://imagej.net/software/fiji/).

Statistical analyses were performed using GraphPad PRISM 9.0 using the log-rank test, one-way ANOVA with Tukey adjustment for multiple comparisons, and unpaired Student’s t-test for single comparison. p-Values ≤ 0.05 were considered significant and indicated with an asterisk.

Recently, two constructs derived from the envelope protein of the baboon endogenous virus, BaEV-Rless and BaEV-TR (14) have been published that are optimally suited for genetically modifying NK cells (21, 23). We additionally wanted to assess the transduction efficiency achieved in human NK cells with an alternative envelope pseudotype, GALV-TM, a chimera generated from the gibbon ape leukemia virus and the aMLV envelope proteins (34), that we previously used for the transduction of human CD34+ stem cells (35). For efficient virus production, our lentiviral vector plasmid pCL6EGNwo, expressing a fusion protein of EGFP and the neomycin resistance gene under the control of a modified SFFV promoter (27), was pseudotyped with the three envelopes. As additional controls, we also pseudotyped the lentiviral vector with the feline endogenous virus chimeric construct RD114-TR (31) and VSV-G (20).

Primary human NK cells were isolated and expanded as previously described (26) and then transduced with serially diluted (1:10 to 1:10,000) non-concentrated supernatants of freshly produced lentiviral vector particles on Retronectin-coated 96-well plates. Three to four days after transduction, NK cells were analyzed by flow cytometry for EGFP expression. As shown in Figure 1A , BaEV-Rless was the most efficient pseudotype, as 28% of NK cells were EGFP positive using 1:10-diluted viral supernatant; even a 1:1,000 dilution of this pseudotype was associated with a higher transduction efficiency than VSV-G at a 1:10 dilution. The lentiviral vectors pseudotyped with BaEV-TR or GALV-TM transduced approximately 9% (at 1:10) to 0.3% (at 1:1,000) of the NK cells. While being more efficient than VSV-G, the RD114-TR pseudotype was still rather inefficient with gene transfer rates ranging from 2.2% to 0.1%.

To promote the binding of lentiviral particles on the surface of target cells and thereby increase the likelihood of viral entry, cationic culture additives such as polybrene or protamine sulfate have been used during the transduction procedure and neutralize the electrostatic repulsion between the opposing negatively charged membrane bilayers (36, 37). Vectofusin is such a histidine-rich cationic amphipathic peptide that was developed to promote efficient lentiviral transduction of primary human cells (38), including T and NK cells (39). Alternatively, a recombinant chimeric fragment derived from human fibronectin, CH-296 or Retronectin, is used since 1996 in research settings and in clinical trials to promote colocalization of viral particles and mammalian target cells, thereby increasing the efficiency of the genetic modification for both adherent and non-adherent cells (32, 33). Although the effects of Vectofusin and Retronectin on the transduction of different non-adherent cell types including human CD34+ hematopoietic stem cells (38) and also primary human NK cells (23, 39) have been reported, this type of comparison has not been performed for the lentiviral BaEV pseudotype that mediates the most efficient gene transfer into primary human NK cells ( Figure 1A ).

Therefore, we transduced primary human NK cells with the pCL6EGNwo vector (27) pseudotyped with BaEV-Rless, BaEV-TR, or GALV-TM using serial dilutions, ranging from 1:2 to 1:1,000. Transduction was performed in flat-bottom 96-well plates, either coated with Retronectin or after supplementing Vectofusin to the culture medium. Control NK cells were transduced in 96-well cell culture plates without any additional transduction enhancer. The gene transfer efficiencies into the human NK cells were analyzed 3 to 4 days later by flow cytometry, and the values were normalized by assigning 100% to the highest transduction efficiency achieved within the cell series of each donor. Remarkably, the BaEV-Rless pseudotype on Retronectin mediated the highest gene transfer in each of the four donors and was therefore set at 100%, while the other two pseudotypes BaEV-TR and GALV-TM did not even achieve 50% of the BaEV-Rless transduction efficiency ( Figure 1B ). Using Vectofusin with the BaEV-Rless-pseudotyped lentiviral particles resulted in roughly 80% of the Retronectin-assisted gene delivery and was clearly better than the gene transfer efficiencies observed with the other two pseudotypes BaEV-TR and GALV-TM. The gene transfer without any transduction enhancer was very low regardless of the glycoprotein used. Therefore, for the remaining experiments in this study, we chose to perform the lentiviral transductions of primary human NK cells with the BaEV-Rless pseudotype on Retronectin-coated plates.

We have previously shown that the U3 region of the MPSV as an internal promoter in our self-inactivating (SIN) lentiviral vectors is the best choice for achieving high-level gene expression in primary human T cells (27). Here, in order to identify an optimal internal promoter for stable high-level CAR expression in primary human NK cells, we tested the MPSV and two additional viral promoters as internal promoters in our lentiviral pCL6EGNwo vector (27): the U3 regions from SFFV (40) with one (mod. SFFV) or with two enhancer regions (SFFV). We also included two promoters of human cellular housekeeping genes, namely, the human phosphoglycerate kinase-1 (hPGK) (27) and the human elongation factor 1α (EF1α) promoter with (opt. EF1α) or without (short EF1α) an optimized splicing unit as internal promoters.

Primary human NK cells were transduced with serially diluted BaEV-Rless-pseudotyped lentiviral vectors expressing the EGFP-neomycin fusion as marker gene under the control of the different promoters. Three to four days after transduction, NK cells were analyzed for EGFP expression by flow cytometry. We compared the mean fluorescence intensity (MFI) of the different promoters in samples with a gene transfer of approximately 5%–10%, as cells with this gene transfer rate most likely carry only one copy of the vector integrated into their genome (27), thereby eliminating the bias of multiple integrations on the transgene expression levels. As shown in Figure 1C , the hPGK and short EF1α promoters with MFIs of 19.0 and 24.9, respectively, were associated with the lowest transgene expression levels. Optimizing the splice unit of EF1α by shortening the intron and correcting an open reading frame in the intron improved the transgene expression levels of the optimized EF1α promoter to those of the SFFV promoters with MFIs ranging from 35.7 to 38.9. However, the MPSV U3 promoter with an MFI of 47.8 still provided the highest level of transgene expression, and thus it was chosen for the expression of CAR constructs in primary human NK cells for our subsequent experiments.

After the establishment of an optimized transduction protocol, primary human NK cells with low CD33 expression were transduced with 5-fold concentrated BaEV-Rless-pseudotyped lentiviral particles encoding BFP in cis with CD19, CD33, CD123, or EGFR CARs, harboring our CD34-derived hinge (20, 29), on Retronectin ( Figures 2A, B ). Three to four days after transduction, the expression of the CARs was analyzed after staining of the C6 hinge region with the QBend-10 antibody and detecting BFP expression by flow cytometry. As shown in Figure 2C , CD34/CAR expression strongly correlated with BFP expression for all four CAR constructs. To purify the CAR NK cells, we then performed MACS enrichment on an OctoMACS™ separator using MS columns and CD34 microbeads binding to the C6 hinge. As an indicator of the content of transduced NK cells, BFP expression was analyzed in three fractions: before MACS, the flowthrough, and after MACS (representative samples in Figure 2D ). Before MACS enrichment, we obtained on average of between 38.8% ± 3.4% (Cetux) and 47.0% ± 3.7% (CD123) CAR-positive NK cells. Enrichment of CAR NK cells on MACS columns led to purification rates of between 96.2% ± 1.2% (CD123) and 97.7% ± 0.8% (CD19). However, the flowthrough contained relatively high percentages of CAR NK cells (between 24.4% ± 2.2% for CD123 and 29.0% ± 4.2% for Cetux), albeit with much lower MFIs than the CAR NK cells before or after MACS ( Figure 2E ).

For more than 40 years, it is well established that after stimulation, primary human NK cells can exhibit high cytotoxic activities against human leukemic cells (41). This cytotoxicity is often mediated by the expression of antigens on the leukemic blasts that are ligands for activating receptors on NK cells (42). As in vitro expanded NK cells are highly activated (26), they often recognize and thus kill leukemia cells independently of CAR antigen binding (42). In order to eliminate this “background” killing for follow-up assays, we planned to systematically block activating receptors of the NK cells in cytotoxic experimental settings. To this end, we co-cultured cells of four AML cell lines and two B-cell precursor ALL cell lines overnight with three ratios of PB-derived 7- to 14-day-old NK cells with the effector-to-target cell ratios of 3:1, 1:1, and 0.3:1. In parallel, overnight co-cultures were set up with the same NK cells, albeit preincubated with monoclonal antibodies that recognized and block either three (NKG2D, NKp30, and NKp46) or six (NKG2D, DNAM-1, NKp30, NKp44, NKp46, and NKp80) activating receptors on NK cells (11, 43). Although these recombinant antibodies were all mutated in their Fc domains and therefore should not bind to human Fc receptors, we still included a co-culture condition with NK cells that were preincubated with the appropriate isotype controls.

As demonstrated in Figure 3 , the degree of target cell lysis by activated NK cells seemed to be cell line-specific. On the other hand, three AML cell lines [MOLM-14 (AML-M5), THP-1 (AML-M4), and CMK (AML-M7)] and two B-cell precursor leukemic lines [REH and BV-173] were killed efficaciously by the activated NK cells in the absence of blocking antibodies and also in the presence of isotype controls; NOMO-1 (AML-M5) cells were resistant to NK cell killing. Importantly, preincubation with all six antibodies ameliorated the cytotoxicity of the NK cells against the five susceptible cell lines quite efficiently; however, using the three antibodies against NKG2D, NKp30, and NKp46 only partially reduced the NK cell-mediated killing for two AML cell lines, MOLM-14 and CMK ( Figure 3 ).

After successfully establishing a blocking protocol, we specifically tested the functionality of CAR NK cells against one ALL and two AML cell lines. To this end, MACS-enriched CAR NK cells of at least five healthy donors with low CD33 expression (data not shown) were blocked with all six monoclonal antibodies described above and then co-cultured overnight with NOMO-1, MOLM-14 (both CD19− CD33+ CD123+), and REH (CD19+ CD33− CD123-) cells at various effector-to-target cell ratios. Flow cytometric analysis revealed the expression patterns of CD19, CD33, and CD123 for each of the target cell lines ( Figure 4A ). After 16 h of co-culture, CD33 and CD123 CAR NK cells efficaciously killed NOMO-1 and MOLM-14 cells with approximately 90% specific lysis at an effector-to-target cell ratio of 1:1 ( Figure 4B ). Interestingly, although CD123 is expressed much lower on NOMO-1 cells than CD33 ( Figure 4A ), CD123 CAR NK cells performed as well as CD33 CAR NK cells in the NOMO-1 co-cultures. REH cells were highly efficaciously (>95% at 1:1) eradicated by CD19 CAR NK cells, but not by the other CAR constructs ( Figure 4B ). Importantly, the lysis was strongly dependent on the specific antigen expression, and we did not observe off-target toxicities in these experiments.

Next, we hypothesized that the CAR NK cells would also specifically kill primary AML blasts obtained at diagnosis and expanded in NSG mice (Füchtjohann, Hanenberg, manuscript in preparation 2022). As AML blasts can express low levels of CD19, we used the EGFR-recognizing Cetux CAR NK cells as negative controls. Flow cytometric analysis revealed the expression patterns of EGFR, CD33, and CD123 for the primary AML blasts used ( Figure 4C ). The results in Figure 4D demonstrated that the CD33+ CD123+ primary AML-M4 blasts were specifically killed (up to 90% at 1:1), while only 60% of CD33+ CD123+ primary AML-M5 blasts were eliminated by both CD33 and CD123 CAR NK cells at 3:1 and 1:1 ratios, respectively.

Finally, we took advantage of the fact that our CARs contained the CD34-derived C6 hinge to visualize the specific interaction of CD19 CAR NK cells with the CD19-positive REH cells in the presence of blocking antibodies. For the fluorescence microscopy analysis, the PE-conjugated QBend-10 antibody was employed to label the hinge region of CD19 or CD33 CAR NK cells, which both co-expressed BFP, while the target REH cells expressed EGFP. As indicated by the white arrow in Figure 5A , the CD19 CAR constructs on the transduced NK cells accumulated in immunological synapses formed between 1 and 4 h between NK and REH cells. In contrast, the CD33 CAR constructs were equally dispersed on the cell surfaces of CD33 CAR NK cells ( Figure 5B ), suggesting that no specific interaction with the CD33-negative REH cells occurred.

For in vivo application of activated NK cells in humans, it is paramount that NK cells do not possess autocrine stimulation loops that would ensure their survival beyond a few days in patients (44, 45). Therefore, allogeneic NK cell therapies are usually accompanied by the subcutaneous application of IL2 or intravenous infusions of lentiviral IL15; both treatments are actually associated with severe side effects in patients (44, 45). In order to evaluate the impact of IL15 signaling on the persistence of NK cells in vivo, we generated two lentiviral IL15 vectors ( Figure 6A ): the first expressed soluble human IL15 and EGFP, and the second human IL15 tethered to the IL15 receptor α-chain (IL15-IL15R) and EGFP. Importantly, both IL15 constructs were able to mediate the expansion of primary NK cells in in vitro experiments in the absence of any other growth factors (data not shown). Next, NSG mice were injected with 5 × 10⁶ EGFP+ NK cells expressing either soluble IL15 or IL15-IL15R and 5 × 10⁶ untransduced NK cells. We used this 1:1 ratio in vivo to investigate potential survival advantages of IL15 expressing NK cells over untransduced NK cells in each mouse and also to understand the impact of IL15-tethering for both populations. The persistence of human NK cells was assessed for 27 days ( Figure 6B ), using the gating strategy shown in Figure 6C for the peripheral blood analysis on day 10. Figure 6D reveals that the NK cell survival was highly dependent on IL15 signaling, as the percentages of NK cells in the blood of the mice transplanted with untransduced NK cells progressively decreased over time. In contrast, both strategies to express IL15 in vivo drastically improved the initial NK cell engraftment and also the persistence of NK cells at all sampling time points. Remarkably, the ratio of EGFP positive to negative cells stayed rather constant at 1:1 for the soluble IL15, while the IL15-IL15R expressing NK cells appeared to slowly outgrow the untransduced cells over time: after 27 days, approximately 70% of all NK cells in the blood of the animals were EGFP positive.

Finally, we used an ALL xenograft model in NSG mice to evaluate the impact of IL15 signaling for the survival and leukemia control of CAR NK cells in vivo. For this experiment, three CD19 CAR vectors were generated ( Figure 7A ) co-expressing via a T2A site either BFP, soluble human IL15 (17), or human IL15 tethered to the IL15 receptor α-chain (IL15-IL15R) (46). For the in vivo experiments, NSG mice were transplanted intravenously with 3.5 × 10⁶ REH cells expressing a firefly luciferase/EGFP fusion protein (REH^LucEG). Two days later, 3.5 × 10⁶ total primary human NK cells with 50% non-transduced and 50% transduced cells were injected via the tail vein ( Figure 7B ). Untreated NSG mice showed rapid leukemia progression and had to be sacrificed between days 17 and 19 ( Figures 7C, D ). Surprisingly, treatment of the mice by infusion of BFP/CD19 CAR NK cells did not improve the survival at all ( Figures 7C, D ). For these two groups, the REH cells made up between 0.3% and 0.4% of peripheral blood cells on day 7 and between 3.3% and 7% on day 15 ( Figure 7E ). Interestingly, the BFP/CD19 CAR NK cells were still detectable at day 8 (2.3% of peripheral blood cells) but did not persist in the blood beyond this time point ( Figure 7F ). When the CD19 CAR NK cells additionally expressed IL15 or IL15-IL15R instead of BFP, the leukemia progression was markedly reduced as seen by luminescence imaging ( Figure 7C ) and peripheral blood analysis ( Figure 7E ). We attributed the improved survival of the animals to the robust increase in the persistence of CAR NK cells, as these immune effector cells were detected at high levels on day 15 (4% for IL15/CD19 and 5.1% for IL15-IL15R/CD19 CAR NK cells) and were even present at lower levels on day 22 in the peripheral blood of the animals ( Figure 7F ). Importantly, however, the animals in the IL15 and IL15-IL15R groups all showed ALL persistence in central nervous system (CNS) lesions ( Figure 7C ). These lesions partly led to neurological deficits/abnormalities, and the majority of these animals had to be sacrificed between days 22 and 27. At the termination of the experiment, three NSG mice of the IL15-IL15R/CD19 CAR NK cell group were still alive, albeit all affected by CNS leukemia.