Ex Vivo Expanded Adaptive NK Cells Effectively Kill Primary Acute Lymphoblastic Leukemia Cells

Natural killer (NK) cells are innate lymphoid cells characterized by a potent intrinsic cytotoxic potential and an ability to kill virus-infected and transformed cells (1–4). In humans, the functional potential of NK cells is regulated through expression of germline-encoded activating and inhibitory receptors (5). Activating receptors bind to stress-induced ligands, which are highly expressed on tumor cells (6, 7). Inhibitory receptors, including killer cell immunoglobulin-like receptors (KIR) and CD94-NKG2A heterodimers, bind to groups of HLA class I alleles and the nonclassical HLA class I molecule HLA-E, respectively (8, 9). The variegated and random distribution of KIR in the NK-cell compartment endows the innate immune system with a broad and highly diverse range of responsiveness without receptor rearrangement (10, 11). The functional potential of a given NK cell is quantitatively tuned by the net input obtained through interactions with ligands for activating receptors and self-MHC molecules (12, 13). This functional calibration process is termed NK-cell education and allows the cell to sense a sudden decrease of normal HLA expression as a result of virus infection or tumor transformation (14).

In the context of allogeneic adoptive NK-cell therapy or stem cell transplantation (SCT), such reactivity due to a lack of recognized MHC-encoded inhibitory ligands is triggered when NK cells are transferred across HLA barriers (15). There is compelling evidence that NK cells play an important role as antileukemic effector cells in T-cell–depleted haploidentical, KIR ligand–mismatched, SCT for patients with acute myeloid leukemia (AML; refs. 16–18). NK cell–mediated graft-versus-leukemia (GVL) effects have been less well documented in the context of acute lymphoblastic leukemia (ALL), in particular in adult ALL (19, 20). Being derived from lymphoid progenitors, ALL blasts express more HLA class I than myeloid cells, potentially contributing to its increased resistance to NK-cell killing (21). Other suggested mechanisms of resistance include insufficient binding to LFA-1 and lower expression of ligands for activating NK-cell receptors, including MICA/B on target cells (15, 19).

The KIR repertoire in any given donor is determined by KIR gene content, copy number variation, and a stochastic epigenetic regulation at the promoter level (22, 23). Thus, the frequency of alloreactive NK cells varies, ranging from a few percent to 60% to 70% of the NK-cell population in exceptional cases (24). Therefore, expression of HLA class I on acute leukemia blasts remains an important limiting factor for the efficacy of NK-cell therapy despite the use of allogeneic NK cells. ALL cells are resistant to the NK92 cell line (19). NK92 lacks KIRs, but expresses NKG2A, an inhibitory receptor that binds to HLA-E, which may contribute to the NK-cell resistance. However, pediatric ALL cells express little HLA-E (25). Thus, it remains unclear whether NK-cell recognition of ALL cells is functionally regulated through activating and inhibitory input delivered through NKG2C and NKG2A receptors, respectively, binding to their shared ligand, HLA-E.

Although several GMP-compliant NK-cell expansion protocols yield large numbers of NK cells for therapy, it has proven difficult to selectively expand specific subsets of NK cells expressing a single self-specific KIR, which would allow more effective targeting of allogeneic leukemia cells (26, 27). Insights into the natural immune response to human cytomegalovirus (HCMV), manifested as expansion and differentiation of highly cytotoxic NK cells with unique and predictable KIR expression profiles, may hold keys to the design of more selective in vitro expansion protocols. Numerous reports have shown in vivo expansion of so called “adaptive” NKG2C⁺ NK cells expressing a single self-KIR in HCMV seropositive individuals (28–30). HCMV reactivation and the associated expansion of adaptive NK cells have been linked to a lower relapse rate in AML patients after cord blood allogeneic SCT, a type of transplant in which one or more HLA mismatches are common (31). This suggests that spontaneously occurring adaptive NK cells may have beneficial effects mediated through eradication of residual leukemic blasts. In vitro expansion of NKG2C⁺ adaptive NK cells can be achieved by coculturing naïve NK cells with CMV-infected fibroblasts or HLA-E transfected cell lines (28, 32), but the robustness and efficacy of such expansion protocols remains to be determined. Herein, we have examined the use of a feeder-based platform for selective expansion and differentiation of educated NK cells aiming to target primary ALL blasts.

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Regional Ethical Review Board in Stockholm, Sweden. Healthy blood donors were used as a source for NK-cell expansion experiments. For all donors, peripheral blood mononuclear cells (PBMC) were separated from buffy coats by density gravity centrifugation (Ficoll–Hypaque; GE Healthcare) and were cryopreserved in 10% DMSO and 90% heat-inactivated FBS for later use. Bone marrow samples from patients with pediatric ALL and adult Myelodysplastic Syndromes (MDS) were processed like the healthy PBMCs. When required, genomic DNA was isolated from 200 μL of whole blood using DNeasy Blood and Tissue Kit (Qiagen).

Thawed PBMCs from healthy blood donors were subjected to NK-cell negative selection (NK Cell Isolation Kit, Miltenyi Biotec). NK cells were then cocultured with irradiated (40 Gy) 721.221.AEH cells (33) in complete medium (RPMI, glutamine, 10% FCS) in the presence of IL15 (final concentration 100 ng/mL, R&D Systems) at a final concentration of 5 × 10⁶/mL in a 96-well U-bottom plate (E:T ratio 10:1). Cells were labeled with CellTrace violet (1 μmol/L, Life Technologies) prior to coculture.

Isolated NK cells, either rested overnight or expanded after 14 days, were labeled with CellTrace violet and used as effector cells at a final concentration of 1.25 × 10⁶/mL. Blasts were thawed and added at a final concentration of 2.5 × 10⁵, and caspase-3 substrate was added at the beginning of the assay. After 4 hours, blasts were stained with the live/dead cell marker aqua (DCM) and relevant extracellular receptors. The percentage of cytotoxic activity was calculated using the following formula: percent specific killing = (C3⁺ – spontaneous C3⁺) + (C3⁺DCM⁺ − spontaneous C3⁺DCM⁺)/[live targets + (C3⁺ − spontaneus C3⁺) + (C3⁺DCM⁺ − spontaneous C3⁺DCM⁺)].

KIR ligands were determined using the KIR HLA Ligand Kit (Olerup SSP; Qiagen) for the detection of, HLA-C1, HLA-C2, and four HLA-Bw4 motifs. KIR genotyping was performed using a high-throughput technology called quantitative KIR automated typing (qKAT; ref. 34).

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.

Stainings were performed using an appropriate mix of the following antibodies (clone names are given in brackets): CD14-V500 (M5E2), CD19-V500 (HIB19), CD3-PE-Cy5.5 or PE.Cy5 (UCHT1), CD56-ECD or PE-Cy5.5 (N901), CD57 purified (TB01) or BV605 (NK-1), anti-mouse-IgM-EF650 (II/41), FCER1γ-FITC (rabbit polyclonal), CD7-PE.Cy7 (8H8.1), ILT2-PE (HP-F1), CD161-BV605 (HP-3G10), NKG2A-APC or APC.AF750 or PE-Cy7 (Z199), NKG2C-PE (FAB138P, REA205), DNAM-1-PE.vio770 (Dx11), NKG2D-BV711 (1D11), CD16-AF700 or BV785 (3G8), 2B4-PE (C1.7), NKp46-BV786 (9E2), CD2-PB (TS1/8), KIR2DL3-FITC (180701), KIR2DL1-APC (143211), KIR3DL1-AF700 (Dx9), KIR2DS4-QD585 (179315), KIR3DL2-biotin (Dx31), KIR2DL2/L3/S2-PE.Cy5.5 (GL183), KIR2DL1/S1-PE.Cy7 (EB6), CD10 (HI10a), granzyme A (CB9), granzyme B (GB11), perforin (B-D48), and TRAIL (RIK-1). Dead cells were labeled using live/dead aqua (Life Technologies). Biotin-conjugated antibodies were visualized using streptavidin-Qdot 585 or 605 (Life Technologies). After staining, cells were fixed and permeabilized using a Fixation/Permeabilization Kit (eBioscience) prior to intracellular staining. Samples were acquired using an LSR-Fortessa 18-color flow cytometer (Becton Dickinson), and data were analyzed with FlowJo software version 9 (Tree Star, Inc.). The BD-LSR Fortessa instrument was equipped with a 100 mW 405 nm laser, a 100 mW 488 nm laser, a 50 mW 561 nm laser, and a 40 mW 639 nm laser.

Isolated NK cells were rested overnight in complete medium and distributed at a final concentration of 2.5 × 10⁵ cells/mL in a 96-well U-bottom plate. All target cells were added to NK cells at a final concentration of 2.5 × 10⁵ cells/mL. For conventional functional assays, ALL blasts, PHA blasts, 721.221.wt, or 721.221.AEH cells were used as target cells. The cells were incubated for 6 hours (37°C, 5% CO₂) after addition of monensin (GolgiStop, BD Biosciences, 1/1,500 final concentration), brefeldin A (GolgiPlug, BD Biosciences, 1/1,000 final concentration), and CD107a-BV421 (H4A3, 1/100 final concentration). After incubation, cells were washed and stained for extracellular receptors, permeabilized (fixation/permeabilization buffer, eBioscience) and stained for intracellular TNF-APC (MAb11) and IFNγ-AF700 (B27) prior to analysis with flow cytometry.

For comparisons of independent groups, the Student t test or the Mann–Whitney test was performed. For comparisons of matched groups, the paired Student t test or the Wilcoxon matched test was performed depending on the sample size. For comparisons of qualitative variables, the Fisher exact t test was performed. In the relevant figures, n.s. indicates not significant; *** indicates P < 0.001; ** indicates P < 0.01; and * indicates P < 0.05. Analyses were performed using GraphPad software.

Here, we set out to establish a platform for selective expansion of NKG2C⁺ NK cells expressing a single self-specific KIR (self-KIR). NK cells were cocultured together with 721.221 cells transfected with HLA-E (221.AEH) in the presence of IL15 for 14 days and stained for inhibitory KIRs, NKG2A, and NKG2C. As previously reported (28), a profound skewing toward NKG2C⁺ NK cells expressing self-KIR was seen after 14 days in culture (Fig. 1A). Stratification of NK-cell subsets during proliferation revealed that CD57⁺ NK cells dominate among nondividing NK cells (generation 0), whereas self-KIR⁺NKG2C⁺ NK cells dominated in the fraction of NK cells reaching later generations (Fig. 1B). To explore the robustness of this platform, we generated adaptive NK cells from 14 HCMV seropositive donors, with or without a preexisting NKG2C⁺CD57⁺ adaptive NK-cell populations. The 2-week culture protocol successfully expanded NKG2C⁺ NK cells from all donors tested. The skewing of the KIR repertoire with a bias toward expression of self-KIRs was noted both at the bulk NK-cell level and in the NKG2C⁺NKG2A⁻ NK-cell subset. In donors with preexisting expansions of adaptive NK cells, the skewing of the repertoire was further enhanced (Fig. 1C). This was also observed in C1/C2 donors where the expanded phenotype was determined by the preexisting CMV-induced imprint in the KIR repertoire (Supplementary Fig. S1). Although it cannot be excluded that the enrichment of the adaptive NK cells was partly a consequence of selective survival, the protocol also led to a 2.4-fold expansion of the adaptive NK-cell subset (Fig. 1D). Culture of adaptive NK cells in IL15 alone or in combination with 221.wt cells did not induce a similar skewing toward an end product enriched for NKG2A⁻ NK cells (Supplementary Fig. S2). These results demonstrate the feasibility and robustness of the expansion of self-KIR⁺ NKG2C⁺ NK cells from HCMV-seropositive donors using HLA-E–overexpressing feeder cells.

We monitored the phenotypic changes and granular load in the NK cells pre- and post in vitro expansion and found that both conventional and adaptive NK cells displayed lower expression of CD7, Siglec-7, and PLZF and upregulation of CD2 and DNAM-1 (Fig. 2). Whereas the frequency of CD57⁺ NK cells decreased somewhat, due to preferential proliferation of CD57^– NKG2C⁺ single-KIR⁺ NK cells (Fig. 1B; ref. 28), the expression intensity of CD57 increased over the course of the 14-day culture. Loss of FcϵR1γ and Syk are additional hallmarks of adaptive NK cells (35, 36). In contrast to the surface receptors, these adaptor molecules were either stable or normalized during the 14-day culture. Expression of the inhibitory receptor LILRB1 did not show any consistent change.

Adaptive NK cells can express slightly more CD16 and perform better in antibody-dependent cytotoxicity assays assays (37). However, activation of NK cells with target cells and/or cytokines leads to loss of CD16 through a metalloprotease called ADAM17 (38). Indeed, the feeder-based expansion protocol led to a relative loss in CD16 expression down to approximately 50% positivity, in line with previously reported data (Fig. 2). Notably, CD16 was more stable in adaptive NK cells, leading to a relatively bigger difference between adaptive and conventional NK cells after coculture with HLA-E–expressing feeder cells. In addition to the observed changes in surface receptors and adaptor molecules, the levels of granzyme A and perforin increased or remained stable in all NK-cell subsets (Supplementary Fig. S3). Together, these data suggest that the ligation of NKG2C under the influence of IL15 leads to a gradual shift toward an adaptive phenotype.

The dynamic change toward a more adaptive phenotype prompted us to test the specificity and functional potential of in vitro generated NK cells. To this end, expanded NK cells expressing either KIR2DL3 or KIR2DL1, derived from HLA-C1/C1 (C1/C1) or HLA-C2/C2 (C2/C2) donors, respectively, were tested against C1/C1⁺ and C2/C2⁺ PHA blasts in a flow cytometry–based cytotoxicity assay. Target cell death was measured with caspase-3 (C3) activity and a live/dead cell marker (DCM). The expanded NK cells displayed effective and specific killing of mismatched PHA blasts, reaching nearly 100% killing of C1/C1 blasts by KIR2DL1 single-positive NK cells (Fig. 3B). Despite the 2-week culture in IL15, negligible killing was seen in the KIR-HLA–matched setting (Fig. 3A and B). To evaluate the impact of the selective expansion protocol, the cytotoxicity of the expanded NK cells was compared with resting NK cells before expansion. All NK-cell expansions displayed significantly more potent cytotoxicity against HLA-C–mismatched PHA blasts on a per cell basis compared to the resting NK cells day 0 (Fig. 3B). In the presence of HLA class I–blocking antibodies, expanded NK cells were also able to kill PHA blasts in KIR-HLA–matched settings. In contrast, HLA class I blockade had no effect on the killing of HLA-mismatched PHA blasts, suggesting that the reprogrammed NK-cell cultures displayed a maximized alloreactive response (Fig. 3C).

Previous studies have reported that ALL blasts are relatively resistant to lysis by allogeneic NK cells and the NK cell line NK92 (19, 39). Therefore, KIR2DL1- and KIR2DL3 single-positive, in vitro expanded adaptive NK cells were tested against a panel of blasts from 23 pediatric, primary ALL bone marrow samples of three different subgroups; T-cell ALL and B-cell precursor cells with either high hyperdiploidy 47-50 chromosomes (Heh) or (t12;21) (Table 1). All samples were taken at the time of diagnosis and were >80% blasts. We established a FACS-based killing assay using an individualized gating strategy for each ALL sample, based on the immune phenotyping developed by the Nordic society of pediatric hematology and oncology (Fig. 4A). Expanded NK cells displayed high killing activity against HLA-mismatched primary ALL blasts (Fig. 4B and C). The average killing was approximately 70% at an effector to target ratio of 5:1 for ALL blasts in the mismatched setting, independent of disease subgroup (Fig. 4B and C). A similar efficiency was noted when selectively expanded NK cells were tested against primary MDS and AML blasts (Supplementary Fig. S4), suggesting the possibility of a potentially broader clinical implementation of in vitro expanded adaptive NK cells.

ALL blasts from pediatric patients express little HLA-E (25). Thus, it remained an open question whether NKG2C contributed to the recognition of the mismatched targets. In line with previous reports, we observed lower expression of HLA-E on B-ALL blasts compared with healthy controls (Fig. 4D). To address the relative contribution of the NKG2C–HLA-E interaction, degranulation was monitored in NKG2C⁺NKG2A^– and NKG2C^–NKG2A⁺ NK-cell subsets following stimulation with ALL blasts and 221.AEH cells in the presence or absence of anti-CD94 blockade (Fig. 4F). Although predictable recognition patterns were observed for NKG2A^–NKG2C⁺ NK cells and NKG2A⁺NKG2C^– NK cells against 221.AEH cells overexpressing HLA-E (Fig. 4E), no significant difference in degranulation was seen with or without anti-CD94 mAb in either NKG2A^–NKG2C⁺ or NKG2A⁺NKG2C^– NK cells when stimulated with ALL blasts (Fig. 4F). Taken together, these results suggest that although NKG2C is functional on the in vitro expanded adaptive NK cells, its ligation is not required for the potent killing of ALL blasts in an HLA-mismatched setting.

The finding that NKG2C appeared redundant for the ability of the expanded adaptive NK cells to kill ALL blasts led us to examine the impact of the size of the alloreactive subset in more detail. To this end, we generated parallel cultures and monitored the ability of resting, short-term IL15 stimulated and expanded NK cells to kill mismatched ALL blasts (Fig. 5A). The killing of primary ALL blasts correlated with the generation of a large alloreactive subset (Fig. 5B). To further explore the impact of allorecognition and the intrinsic cytotoxic potential of adaptive NK cells, we sorted NKG2C⁺NKG2A^–KIR2DL1⁺ NK cells from C2/C2 donors to 100% purity (Supplementary Fig. S5) and compared their ability to kill HLA-mismatched PHA blasts with unsorted, expanded adaptive NK cells from the same donor. Notably, the sorted adaptive NK cells were highly potent against mismatched C1/C1⁺ PHA blasts, although expanded NK cells showed slightly higher killing against the same targets (Fig. 5C–D). These results show that adaptive NK cells, both resting and expanded, represent a unique subset with enhanced alloreactive capacity due to the near complete lack of inhibitory MHC-binding receptors when transferred across HLA-C barriers. The selective expansion and repertoire skewing using HLA-E–expressing feeder cells leads to further potentiation of the adaptive NK cells and holds promise for the next-generation NK cell–based cancer immunotherapy.

The survival rate for pediatric ALL currently exceeds 90%, but the remaining patients who relapse remain a major clinical challenge requiring effective treatment modalities. Allogeneic SCT has become the established procedure for treating children with high-risk and relapsed leukemia. Over the past decade, accumulating evidence indicates that NK cells may contribute to the GVL effect associated with allogeneic SCT. However, the contribution of NK cells in the GVL effect against ALL appears limited, possibly reflecting the poor ability of heterogeneous NK-cell populations to kill ALL blasts in vitro (39, 40). A large number of clinical trials are currently exploring the potential of resting and cytokine-activated NK cells as a therapeutic modality against various cancer types. Most of the existing protocols involve transfer of highly diverse populations of NK cells. Although currently available clinical NK-cell products contain cells with the desired alloreactive specificity, it is obvious that further refinement of the procedures hold promise to improve the efficacy of NK cell–based immunotherapy (26). Indeed, the size of the alloreactive NK-cell subset and induction of molecular remission and prolonged disease-free survival are correlated (41). In particular, it represents an exciting possibility to selectively expand and/or guide differentiation of adaptive NK cells showing high cytolytic potential and unique KIR specificities (42–44). In this study, we have shown the robustness of a preclinical platform for selective expansion of adaptive NKG2C⁺ NK cells and its potential in targeting primary ALL blasts.

The protocol developed in this study yielded NK cells with a differentiated phenotype displaying profound skewing toward expression of a single self-KIR. Consequently, the expanded adaptive NK cells displayed an elevated specificity for allogeneic targets lacking the cognate HLA ligand, demonstrating the maintained signaling potential of inhibitory KIRs, despite the 2-week culture in IL15. This is an important and unique feature of this particular strategy for expanding adaptive NK cells, because other ex vivo expansion protocols have been associated with loss of self-tolerance by overcoming KIR inhibition (45–48). It is well documented that the proliferative capacity of the CD57⁺ adaptive NK-cell subset is limited (49). This raises the question whether the observed enrichment of adaptive NK cells is merely a consequence of selective survival. We have previously shown that a proportion of CMV-seropositive patients with TAP deficiency also displays expansion of NKG2C⁺ NK cells with phenotypic hallmarks of adaptive NK cells (50). However, these expansions had polyclonal KIR repertoires, suggesting that the selective skewing in MHC-sufficient individuals depends on a selective advantage of inhibition through self-specific KIRs. This aligns with earlier studies in the mouse showing preferential proliferation and expression of Bcl-2 in NK cells expressing self-MHC Ly49⁺ in the mouse (51, 52). Notably, the current protocol yielded a small, but significant increase in absolute numbers of adaptive NK cells, reflecting their expansion in combination with differentiation from less differentiated NKG2C⁺ precursors, as previously suggested (28). NKG2C is unique among activating NK-cell receptors, and is the only NK-cell receptor in addition to CD16, that can activate resting NK cells (53, 54). Thus, we propose that the enrichment of adaptive NK cells under the current protocol is a combined effect of survival, NKG2C-driven expansion, and differentiation from earlier precursors.

The most appealing setting for clinical implementation of the current protocol is to transfer ex vivo expanded adaptive NK cells across HLA-C barriers to patients who are refractory to conventional treatment modalities. In such protocols, cells generated from a C1/C1 donor could be transferred to patients with C2/C2 genotypes and vice versa. Notably, C1/C2 donors could also be used pending on their CMV imprinted repertoire. In C1/C2 donors with a preexisting expansion of KIR2DL1⁺ adaptive NK cells, the current protocol expanded the same phenotype, possibly extending the number of possible donors for C1/C1 patients, which may be of clinical relevance given that the C2/C2 genotype is less common. Although not investigated here, it is possible to expand adaptive NK cells from CMV-naïve repertoires, albeit the robustness and efficiency is much lower (28).

Although no off-target effects have been noted in more than 200 patients that have so far received polyclonal conventional allogeneic NK cells, it is possible that the highly potent adaptive NK cells, lacking all inhibitory receptors to recipient HLA class I, may cause tissue damage from targeting mismatched normal cells. It will be essential to design phase I/II studies with transfer of low cell doses of expanded NK cells using gradually increasing doses of preconditioning with chemotherapy as a means to indirectly titrate the length of persistence of this highly cytotoxic subset in vivo. Paralleling the clinical testing of new chimeric antigen receptors by means of mRNA electroporation, any off-target effects will be transient in the nonconditioned host, as the allogeneic NK cells will be rapidly rejected by recipient T cells.

Another attractive and potentially safer clinical indication for expanded adaptive NK cells is for patients with tumors that display downregulation or loss of HLA class I (55, 56). Resistance to checkpoint inhibition therapy with anti-PD-1 for patients with melanoma is associated with point mutations causing decreased antigen presentation and in one case the complete loss of HLA class I (57). If this observation were common across more tumors, NK cells may find their therapeutic niche as a combination with checkpoint inhibition or a rescue therapy following relapse. In fact, total or partial loss of HLA class I has been reported at diagnosis in numerous cancer types (55, 58, 59). Previous observations show a correlation between HLA-I expression and NK-cell susceptibility in pediatric B-ALL (20). Extending those results, we found that adaptive NK cells efficiently lyse HLA-matched targets in the presence of HLA class I blockade, mimicking the selective loss of HLA class I in HLA-matched tumor cells. The fact that in vitro expanded NK cells retained their sensitivity to cognate KIR–HLA interactions despite culture in IL15 suggests that they will not be toxic to normal tissues when transferred to HLA-matched patients. Furthermore, targeting patients whose tumors displayed low or absent HLA class I will facilitate treatment of patients who are C1/C2, the most common KIR ligand genotype.

Reusing and colleagues reported low expression of HLA-E on primary ALL blasts (25). This raised the question whether NKG2C was involved in the recognition of the ALL blasts. Antibody-blocking experiments suggested a redundant role for NKG2C in the activation of the in vitro expanded adaptive NK cells. NK-cell recognition of ALL blasts depends on DNAM-1 (40). Adaptive NK cells highly express DNAM-1 and CD2 (28, 53, 60), of which CD2 acts in synergy with both CD16 and NKG2C (53). Both these receptors are stable or induced during the in vitro expansion and may contribute to the efficient killing of ALL cells.

Although our expansion protocol resulted in highly specific cytotoxicity against all targets tested, the frequency of cells that mobilized CD107a to the cell surface was relatively low, ranging between 5% and 30%. A previous report showed a subpopulation of NK cells, termed “serial killers” that can deliver multiple lytic hits in a row before exhaustion (61). It is tempting to speculate that the some of the expanded adaptive NK cells are “serial killers” and display different killing kinetics that give rise to the discrepancy between high specific cytolysis and measured CD107a mobilization.

In conclusion, we have demonstrated the robustness of a culture platform yielding expansion and differentiation of adaptive NK cells with high cytotoxic potential and with predictable specificity. The feeder cell–based expansion platform should be scalable into a GMP-compliant process, having immediate implications for harnessing adaptive NK cells in cancer immunotherapy. These results may also guide future attempts to develop induced pluripotent stem cell–derived off-the-shelf adaptive NK cells (26).

V. Beziat has ownership interest (including patents) in Patent WO2014037422 A1. K.-J. Malmberg is a visiting professor at Karolinska Institute, reports receiving a commercial research grant from, has ownership interest (including patents) in, and is a consultant/advisory board member for Fate Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L.L. Liu, V. Béziat, S. Söderhäll, K.-J. Malmberg

Development of methodology: L.L. Liu, V. Béziat, K.-J. Malmberg

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L.L. Liu, V.Y.S. Oei, A. Pfefferle, M. Schaffer, S. Lehmann, E. Hellström-Lindberg, S. Söderhäll, M. Heyman, D. Grandér

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L.L. Liu, V.Y.S. Oei, A. Pfefferle, M. Schaffer, D. Grandér, K.-J. Malmberg

Writing, review, and/or revision of the manuscript: L.L. Liu, V.Y.S. Oei, A. Pfefferle, S. Lehmann, S. Söderhäll, M. Heyman, D. Grandér, K.-J. Malmberg

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V.Y.S. Oei

Study supervision: K.-J. Malmberg

Other (Patient clinical subgroup identification during project planning): S. Söderhäll

This work was supported by grants from the Swedish Research Council, the Swedish Children's Cancer Society, the Swedish Cancer Society, the Tobias Foundation, the Karolinska Institutet, the Wenner-Gren Foundation, the Norwegian Cancer Society, the Norwegian Research Council, the South-Eastern Norway Regional Health Authority, the European Commission Horizon 2020 Programme 692180-STREAMH2020-TWINN-2015, grant from National Science Center, Poland 2014/13/N/NZ6/02081 (to M. Schaffer), and Stiftelsen KG Jebsen. V. Béziat was supported by the French National Research Agency (ANR; grant no. NKIR-ANR-13-PDOC-0025-01).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.