GSK3 Inhibition Drives Maturation of NK Cells and Enhances Their Antitumor Activity

The effector functions of natural killer (NK) cells are triggered by engagement of germline-encoded activating receptors. Cytotoxicity by resting NK cells can be efficiently triggered through CD16 (FcγRIII), a low-affinity receptor for IgG. Many other activating receptors that are involved in natural cytotoxicity, including NKp46, 2B4, DNAM-1, and NKG2D, act synergistically in pairwise combinations to induce cytotoxicity and cytokine production (1). Two additional receptor families that recognize human leukocyte antigen (HLA) molecules and contribute to NK-cell development and function are killer immunoglobulin-like receptors (KIR) and killer cell lectin-like receptor, subfamily C (KLRC). Individual KIR can be stimulatory or inhibitory depending upon whether they have short or long cytoplasmic domains (2). Inhibitory KIR play a key role in the process of NK cell education whereby NK cells with inhibitory KIR that recognize self-HLA molecules are licensed for enhanced function (3). NKG2A and NKG2C belong to the KLRC family. Both receptors form heterodimers with CD94 and recognize HLA-E (4). NKG2A has two immunoreceptor tyrosine-based inhibitory motifs and recruits the tyrosine phosphatases SHP1 and SHP2 to inhibit NK-cell function (5). In contrast, NKG2C interacts with the tyrosine kinase DAP12 to potently trigger NK-cell activation (6).

About 30% to 60% of all CD56^dim NK cells in healthy adults express the terminally sulfated glycan carbohydrate CD57 on their surface (7, 8). The frequency of CD57 expression increases with age (9) and is higher in individuals who are seropositive for cytomegalovirus (CMV; ref. 10). CD57 is absent or dimly expressed on CD56^bright adult NK cells and on fetal and newborn NK cells (11). Therefore, CD57 is regarded as a marker of maturation. Importantly, high frequencies of CD57⁺ NK cells in the peripheral blood or tumor microenvironment in cancer patients have frequently been linked to less severe disease and better outcomes (12). The CD57⁺NKG2C⁺ NK-cell subset is considered to be a counterpart of mouse Ly49H⁺ NK cells that expand in response to MCMV infection and exhibit characteristics of immunologic memory including viral antigen specificity, clonal-like expansion, heightened effector function, persistence, and rapid recall response (13). Because CD57⁺NKG2C⁺ NK cells share most, if not all, of these properties, they are referred to as “adaptive” NK cells. The phenotype of adaptive NK cells that arise in response to CMV is remarkably heterogeneous and extends beyond NKG2C expression. Subsets of epigenetically distinct cells lacking expression of B cell- and myeloid cell-related signaling proteins (FcϵR1γ, EAT-2, and SYK) along with reduced expression of the promyelocytic leukemia zinc finger (PLZF) transcription factor have also been identified in CMV seropositive individuals (14).

Despite the potential clinical significance of both canonical CD57⁺ NK cells and mature adaptive NK-cell subsets that express CD57, relatively little is known with respect to the signals that drive late-stage NK-cell maturation. High-dose IL2 or IL15 can induce CD57 expression on a fraction of CD57⁻ NK cells in short-term cultures (15, 16), and IL21 can potentiate NK-cell maturation from CD34⁺ hematopoietic cell precursors in vitro (17). However, a role for other signaling pathways outside of class I cytokine receptor engagement in driving NK-cell maturation have not been explored in depth.

Glycogen synthase kinase (GSK) 3 is a constitutively active serine-threonine kinase that has two isoforms known as GSK3α and GSK3β. Within the nucleus, GSK3 can influence gene expression directly by targeting transcription factors or indirectly by phosphorylating histones, histone deacetylases, and histone acetyltransferases (18). Recent studies have shown that pharmacologic inhibition of GSK3 promotes the developmental progression of thymocytes through the β-selection stage in the absence of pre-TCR or Notch signaling (19). Additionally, inhibition of GSK3 through siRNA-mediated knockdown or pharmacologic inhibition has also been shown to enhance CD8⁺ cytotoxic T-cell responses (20). Based on these studies demonstrating a positive role for GSK3 inhibition in T-cell differentiation and function, we hypothesized that GSK3 inhibition could represent a novel approach to drive NK-cell maturation and effector function during ex vivo expansion.

Here, we demonstrate that addition of the GSK3 inhibitor CHIR99021 during ex vivo NK-cell expansion with IL15 significantly enhanced CD57 acquisition and maturation. NK cells expanded in the presence of CHIR99021 exhibited markedly increased TNF and IFNγ production in response to target cell recognition and greater natural cytotoxicity and antibody-dependent cellular cytotoxicity (ADCC) against a variety of solid tumor cell lines. In a xenogeneic model of ovarian cancer, adoptive transfer of NK cells expanded with CHIR99021 demonstrated better and more consistent antitumor efficacy. Thus, pharmacologic inhibition of GSK3 represents a novel strategy to enhance NK-cell maturation and function during ex vivo expansion.

Peripheral blood mononuclear cells (PBMC) from healthy CMV seropositive donors were obtained from Memorial Blood Bank, AllCells Inc., and Key Biologics.

All samples obtained with written consent and were de-identified before receipt. The University of Minnesota Institutional Review Board in accordance with the Declaration of Helsinki approved their use.

PBMCs were isolated from whole blood by density gradient centrifugation using Ficoll-Paque Premium (GE Healthcare). T-cell and B-cell depletions were performed using anti-CD3 and anti-CD19 microbeads (Miltenyi Biotech). Untouched CD3⁻CD56⁺ NK cells were isolated using negative selection kits (StemCell Technologies). Monocytes were isolated by positive selection using anti-CD14 microbeads (Miltenyi Biotech). For sorting experiments, cells were stained with flurochrome-conjugated antibodies and sorted to >95% purity using a FACS Aria.

CD3/19-depleted cells or sorted CD3⁻CD56^dim NK cells were cultured in 24-well plates at a concentration of 0.5 × 10⁶ cells per mL in B0 media (21) supplemented with 10 ng/mL IL15 (NCI or Miltenyi Biotech) and either dimethyl sulfoxide (DMSO) as a vehicle control (Sigma) or CHIR99021 (6-[[2-[[4-(2,4-Dichlorophenyl)-5-(5-methyl-1H-imidazol-2-yl)-2-pyrimidinyl]amino]ethyl]amino]-3-pyridinecarbonitrile (BioVision) solubilized in DMSO. In some experiments, sort-purified NK cells were cultured in 10 ng/mL IL15 ± CHIR99021 together with sort-purified, autologous CD14⁺ monocytes at a 1:1 ratio. Additional culture experiments were performed with CD3/19-depleted cells where bead-selected NK cells and monocytes were cocultured together at a 1:1 ratio or separated by transwells with 0.4-μm pores (Corning). In some experiments, NK cells were labeled with CellTrace proliferation dye (Thermo Fisher Scientific) prior to culture. The SKOV-3 cell line was purchased from the ATCC in March 2015. Cells were transduced at passage 5 and used for functional experiments within 5 additional passages (total passage number less than 10). The A549 cell line was purchased from Essen Bioscience in February 2016. Cells were used for functional experiments within 6 passages. The PANC-1 cell line was purchased from the ATCC in June 2016. Cells were transduced at passage 4 and used for functional experiments within 6 additional passages (total passage number less than 10). K562 cells were purchased from the ATCC in March 2015. Cells were used for functional experiments before passage 20. The SKOV-3, A549, and PANC-1 cell lines were analyzed regularly by FACS to ensure that they maintained the expected expression patterns of MHC class I, HER2, and EGFR. The K562 cell line was tested using the LegendScreen antibody array (BioLegend) and displayed the expected phenotype. All cell lines were tested for mycoplasma contamination in July and December 2016 using the q-PCR–based Universal Mycoplasma Detection Kit (ATCC). All lines tested negative on both dates.

Cells were stained with fluorochrome-conjugated antibodies against the following antigens: CD56 (clone HCD56), CD3 (clone OKT3), CD57 (clone NK-1), CD62L (clone DREG-56), CD16 (clone 3G8), CD158a (HP-MA4), CD158b (DX27), NKB1 (DX9), Perforin (dG9), granzyme B (GB11), DNAM-1 (11A8), NKG2D (1D11), NKp80 (5D12), 2B4 (C1.7), TNF (MAb11), IFNγ (clone B27; all from Biolegend), NKG2C (134591), BLIMP1 (646702; from R&D Systems), NKG2A (Z199; from Beckman Coulter), ZEB2 (from Novus Biologicals), EOMES (WD1928), PLZF (Mags.21F7), SYK (4D10.1; from eBiosciences), FcϵR1γ (from Millipore), and T-BET (04-46) and NKp30 (p30-15) (from BD Pharmingen). All staining was done in combination with Live/Dead Fixable Dead Cell Stain (Thermo Fisher Scientific). Detection of intracellular Perforin, granzyme B, T-BET, ZEB2, EOMES, BLIMP-1, PLZF, SYK, FcϵR1γ, TNF, and IFNγ was performed following fixation and permeabilization (eBioscience) according to the manufacturer's instructions. Cells were acquired on either an LSRII or Fortessa cytometer (BD Biosciences), and data were analyzed using FlowJo (TreeStar).

RNA was extracted from cells using the RNeasy Mini Kit (Qiagen), and DNA was digested using DNase I (Thermo Fisher Scientific). cDNA was synthesized using SuperScript III reverse transcriptase (Thermo Fisher Scientific) and used as a template for qRT-PCR using SYBER green master mix (Applied Biosystems). The following primers were used (5′-3′): ACTB fwd; CCCAGCACAATGAAGATCAA, ACTB rev; ACATCTGCTGGAAGGTGGAC, TBX21 fwd; AGGATTCCGGGAGAACTTTG, TBX21 rev; CCCAAGGAATTGACAGTTGG, ZEB2 fwd; AGGAGCTGTCTCGCCTTG, ZEB2 rev; GGCAAAAGCATCTGGAGTTC, EOMES fwd; CAACCTGGGACCAACAAACT, EOMES rev; GCTGCCATCTTCCTCTGGTA, PRDM1A fwd; TCAAACTCAGCCCTCTGTCCA, PRDM1A rev; TCCAGCACTGTGAGGTTTCA, PRDM1B fwd; CCCGAACATGAAAAGACGAT, PRDM1B rev; ATAGCGCATCCAGTTGCTTT. All PCR reactions were run on a 7500 Real Time PCR System (Applied Biosystems).

K562 cells were prestained with eFluor670 proliferation dye (eBiosciences) at a final concentration of 5 μmol/L for 15 minutes at 37°C, followed by washing in complete media prior to mixture with NK cells at indicated effector to target (E:T) ratios in a final volume of 100 μL. Cells were then incubated at 37°C for 3.5 hours followed by the addition of CellEvent Caspase-3/7 Green Detection Reagent (Thermo Fisher Scientific). Cells were incubated for an additional 30 minutes, followed by FACS analysis. Specific killing was calculated based on caspase-3/7 activities. For determination of TNF and IFNγ production, NK cells were incubated at 37°C for 4 hours alone or with K562 target cells at a 1:1 ratio. GolgiPlug and GolgiStop (BD Biosciences) were added to the culture media during incubation. Intracellular staining and FACS analyses were performed after fixation.

SKOV-3 and PANC-1 cells were transduced with red fluorescent protein using the NucLight Red lentiviral reagent (NLR; Essen BioScience). A549 cells expressing NLR were purchased from Essen Biosciences. For kinetic analysis of tumor cell killing, A549-NLR, SKOV-3-NLR, and PANC-1-NLR cells were plated at a concentration of 4 × 10³ cells per well in 96-well flat bottom plates. The following day, NK cells were added at a 3:1 ratio. All cocultures were performed in B0 media. In some wells, 1 μg/mL anti-EGFR (Invivogen) or anti-HER-2 (Herceptin; Roche) antibodies were added. Anti-CD20 (Rituxan; Roche) was used as an isotype control for human IgG₁. The number of viable target cells was monitored by hourly fluorescence imaging over 48 hours using an IncuCyte Live Cell Analysis System (Essen BioScience). Live cell numbers were quantified using IncuCyte Zoom software and normalized to the number of live cells remaining in the target cell only control group.

Six- to 8-week-old NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG) mice were purchased from Jackson Laboratories. Mice were conditioned with 300 cGray radiation. The following day, 1 × 10⁵ luciferase-expressing SKOV-5 cells were injected intraperitoneally (i.p.). Five days later (allowing for establishment of the tumor), 2.5 × 10⁶ NK cells from overnight or 7-day cultures were injected i.p. along with 5 μg IL2 (Life Technologies). In some groups, 100 μg Herceptin antibody (Roche) was administered i.p. All groups of mice that received NK cells continued to receive IL2 injections twice per week at 5 μg per dose for an additional 3 weeks. Tumor progression was monitored by weekly IVIS imaging. Mice were euthanized when total luminescence signal reached 1 × 10¹⁰ p/s. All animal studies were conducted in accordance with the University of Minnesota Institutional Animal Care and Use Committee (IACUC).

To test the effect of GSK3 inhibition during ex vivo NK-cell expansion, we isolated PBMCs from the peripheral blood of healthy CMV seropositive donors. CMV seropositive donors were selected due to the higher frequencies of both canonical CD3⁻CD56^dimCD57⁺ and adaptive CD3⁻CD56^dimCD57⁺NKG2C⁺ NK cells in these donors. T and B cells were depleted, leaving a mixed population of NK cells and autologous monocytes. This isolation strategy was chosen due to the fact that human NK-cell expansion ex vivo is markedly enhanced by the presence of autologous monocytes (22). CD3/19-depleted cells were cultured for 7 days with 10 ng/mL IL15 and either a vehicle control (DMSO) or 5 μmol/L CHIR99021, an aminopyrimidine derivative that potently inhibits both isoforms of GSK3 without showing off-target inhibitory effects on other closely related kinases. CHIR99021 is the most selective inhibitor of GSK3 reported to date (23, 24).

We observed similar overall NK-cell expansion when comparing 7-day cultures with DMSO (8.38-fold ± 1.77) to those with CHIR99021 (7.66-fold ± 1.71 Fig. 1A). However, there was a profound decrease in the frequency of CD57⁺ NK cells after ex vivo expansion with IL15 and the vehicle control (23.47% ± 3.28%) relative to cells preculture (52.81% ± 3.51%; P < 0.0001). This result is most likely due to the low proliferative response of CD57⁺ NK cells to common γ chain cytokines including IL2 and IL15 (15, 16). Intriguingly, NK cells that were expanded with IL15 and CHIR99021 exhibited a significantly higher frequency of CD57⁺ NK cells (44.18% ± 4.83%) relative to control cultures (P < 0.0001). The mean fluorescence intensity of CD57 expression on CD57⁺ NK cells from CHIR99021 cultures (74,293 ± 3,813) was also higher relative to CD57⁺ NK cells from control cultures (58,020 ± 3,725; P = 0.007). In addition, we observed an increase in the percentage of adaptive CD57⁺NKG2C⁺ NK cells in CHIR99021 cultures (9.34% ± 2.39%) relative to DMSO controls (6.08 ± 1.63; P = 0.038; Fig. 1B). Adaptive NK cells expressing CD57 and lacking the transcription factor PLZF and the signaling molecules SYK and FcϵR1γ were also enriched in CHIR99021 cultures (Supplementary Fig. S1).

To more extensively characterize the effect of GSK3 inhibition on the maturation state of ex vivo expanded NK cells, we performed FACS analysis for additional surface receptors as well as intracellular analysis of perforin and granzyme B expression on NK cells expanded for 7 days with either DMSO or CHIR99021. We observed significantly higher KIR (P = 0.008), and 2B4 (P = 0.008) expression on NK cells cultured with CHIR99021 relative to DMSO. These cells also expressed higher levels of both perforin (P = 0.041) and granzyme B (P = 0.012). NK cells cultured with CHIR99021 expressed markedly lower levels of NKG2A (P = 0.002). No statistically significant differences between the DMSO and CHIR99021 conditions were observed for the expression of CD16, NKp30, NKp80, DNAM-1, or NKG2D. However, culture in IL15 significantly increased the expression of NKp30 (P = 0.014), DNAM-1 (P = 0.004), and NKG2D (P = 0.011) relative to resting cells preculture (Fig. 1C). Modulation of surface receptor expression and levels of cytotoxic granule components was evident in both sorted CD3⁻CD56^dimCD57⁻ and CD3⁻CD56^dimCD57⁺ NK cells cultured with IL15 and CHIR99021 (Supplementary Fig. S2).

In mouse CD8⁺ T cells, ex vivo culture with GSK3 inhibitors has been shown to increase Tbx21 (the gene encoding T-bet) transcription (20). T-BET is a key transcription factor for antigen-driven effector function in CD8⁺ T cells (25) and terminal maturation, homeostasis, and IFNG transcription in NK and Vα14i NKT cells (26). T-BET induces the expression of two other transcription factors, zinc finger E-box-binding protein (ZEB) 2 and PR domain-containing protein (PRDM) 1 (BLIMP-1) that promote NK-cell maturation, homeostasis, and function (27, 28). To determine whether GSK3 inhibition increases TBX21 transcription in NK cells, CD3/19-depleted cells were cultured for 7 days with 10 ng/mL IL15 and either DMSO or CHIR99021. NK-cell isolations were then performed, and TBX21 mRNA levels were analyzed by quantitative RT-PCR. We observed higher expression of TBX21 transcripts in NK cells cultured with the GSK3 inhibitor (1.58-fold ± 0.13, P = 0.0023). In addition, we observed higher expression of ZEB2 transcripts (1.94-fold ± 0.23, P = 0.0062) and transcripts for two isoforms of PRDM1: PRDM1A (2.62-fold ± 0.47, P = 0.0135) and PRDM1B (1.91-fold ± 0.19, P = 0.0013). No statistically significant difference in expression was observed for Eomesodermin (EOMES; 0.95-fold ± 0.08, P = 0.5345), a T-box transcription factor with homology to T-BET (Fig. 2A; ref. 29). We also analyzed transcription factor expression at the protein level by intracellular FACS. Again, we observed consistently higher expression of T-BET (1.16-fold ± 0.03, P = 0.0005), ZEB2 (1.13-fold ± 0.02, P = 0.0001), and BLIMP-1 (1.31-fold ± 0.05, P = 0.0002) but not EOMES (0.99-fold ± 0.04, P = 0.8223; (Fig. 2B). Together, these results show that inhibition of GSK3 during ex vivo expansion leads to an enrichment of NK cells with a mature phenotype and is associated with increased transcription of key transcription factors including T-BET, ZEB2, and BLIMP-1.

Several explanations could be posited to account for the increased frequency of CD57⁺ NK cells in ex vivo expansion cultures with CHIR99021. One possibility is that GSK3 inhibition differentially affects the proliferation or survival of particular subsets of NK cells. Another possibility is that CD57 expression is not stable, and NK cells downregulate CD57 in culture. A third possibility is that GSK3 inhibition potentiates the maturation of less mature CD56^dimCD57⁻ NK cells and enhances CD57 acquisition. To distinguish between these possibilities, we sorted CD3⁻CD56^dimCD57⁻ NK cells and CD3⁻CD56^dimCD57⁺ NK cells from freshly isolated PBMCs, labeled the cells with CellTrace proliferation dye, and cultured these cells with autologous CD14⁺ monocytes for 7 days with 10 ng/mL IL15 in the presence or absence of CHIR99021. CD3⁻CD56^dimCD57⁻ NK cells proliferated to a similar extent in control and CHIR99021 cultures. However, CD57 expression was induced to a significantly greater extent in CHIR99021 cultures relative to controls (3.66 ± 0.71-fold, P = 0.033). We also observed a marked downregulation of CD62L expression in 7-day CHIR99021 cultures (1.76 ± 0.24-fold, P = 0.05). CD57 expression remained stable on sorted CD57⁺ NK cells in culture (Fig. 3). Thus, our results show that GSK3 inhibition drives NK-cell maturation.

Culture experiments described above were performed with the inclusion of autologous monocytes to support NK-cell expansion. Therefore, we next wanted to determine whether CHIR99021 acts directly on NK cells to induce maturation or whether the effect is indirect and monocyte-dependent. To test this, we isolated CD3⁻CD56⁺ NK cells and cocultured together with CD14⁺ monocytes at a 1:1 ratio for 7 days with IL15 and either DMSO or CHIR99021 and compared these cultures with ones where NK cells and monocytes were separated by transwells. We observed significantly higher frequencies of CD57⁺ NK cells and a greater intensity of CD57 expression on CD57⁺ NK cells in CHIR99021 cultures regardless of whether NK cells were in contact with monocytes (Fig. 4A), suggestive of a contact-independent effect.

To determine whether GSK3 inhibition could enhance the maturation of NK cells cultured in isolation, we sorted CD3⁻CD56^dim NK cells from freshly isolated PBMCs, labeled sorted cells with CellTrace proliferation dye, and cultured cells for 7 days with 10 ng/mL IL15 and either DMSO or 5 μmol/L CHIR99021. Similar to results obtained with CD3/19-depleted cultures, a substantial decrease in the frequency of CD57⁺ NK cells was observed after ex vivo expansion with IL15 and DMSO (20.96% ± 5.53) relative to cells preculture (48.95% ± 8.46) (P = 0.0057). The frequency of CD57⁺ NK cells was significantly increased in the presence of CHIR99021 (37.49% ± 8.19, P = 0.0047), and the intensity of CD57 expression on CD57⁺ NK cells was higher (18,934 ± 1,770) relative to the DMSO control (14,239 ± 2,720; P = 0.0306; Fig. 4B).

Next, we analyzed the proliferation of sorted CD3⁻CD56^dim NK cells in control and CHIR99021 cultures by CellTrace dye dilution. As expected, CD3⁻CD56⁺CD57⁻ NK cells proliferated to a much greater extent during culture with IL15, but no differences in proliferation were observed for either subset when comparing the DMSO and CHIR99021 culture conditions (Fig. 4C). Similar to proliferation kinetics, CHIR99021 did not adversely affect NK-cell viability (Supplementary Fig. S3).

Additionally, we sorted CD3⁻CD56^bright NK cells, CD3⁻CD56^dimCD57⁻ NK cells, and CD3⁻CD56^dimCD57⁺ NK cells from freshly isolated PBMCs and cultured these sorted NK-cell subsets for 7 days with 10 ng/mL IL15 and either DMSO or 5 μmol/L CHIR99021. In these experiments, no statistically significant difference in the frequency of CD57⁺ NK cells was observed in cultures containing sorted CD3⁻CD56^bright NK cells or CD3⁻CD56^dimCD57⁺ NK cells. However, the frequency of CD57⁺ NK cells was significantly higher on sorted CD3⁻CD56^dimCD57⁻ NK cells cultured in CHIR99021 (23.51% ± 3.85) relative to DMSO (9.65 ± 1.69; P = 0.0281; Fig. 4D). Together, these data show that CHIR99021 acts directly on NK cells to enhance maturation without affecting proliferation.

Having characterized the effects of GSK3 inhibition on NK-cell maturation, we next wanted to test function. CD3/19-depleted cells freshly isolated from PBMCs were cultured for 7 days with 10 ng/mL IL15 and either DMSO or 5 μmol/L CHIR99021. As an additional control, cells were cultured overnight with 10 ng/mL IL15. In 4-hour cytotoxicity assays using caspase-3/7–labeled K562 myeloid leukemia cells as targets at a 1:1 E:T ratio, we observed significantly higher cytotoxicity by NK cells cultured in either the DMSO control or CHIR99021 conditions for 7 days relative to NK cells cultured overnight with IL15. No statistically significant differences were seen between NK cells from the 7-day DMSO and CHIR99021 conditions (Fig. 5A). However, NK cells from CHIR99021 cultures exhibited markedly higher TNF production (24.3% ± 3.62) relative to NK cells from DMSO (14.3% ± 1.91, P = 0.0336) and overnight IL15 (10.5% ± 1.73, P = 0.0046) cultures (Fig. 5B). Similarly, NK cells from CHIR99021 cultures exhibited higher IFNγ production (21.27% ± 3.53) relative to NK cells from DMSO (12.08% ± 1.79, P = 0.0401) and overnight IL15 (13.23% ± 2.08, P = 0.0144) cultures (Fig. 5C). In a more detailed analysis of degranulation and IFNγ production by subsets of NK cells in response to K562 targets after culture with IL15 and DMSO or CHIR99021, we observed similar levels of degranulation regardless of CD57, KIR, or NKG2C expression. However, IFNγ production was highest in CD57⁺KIR⁺ and CD57⁺NKG2C⁺ NK cells cultured with CHIR99021 (Supplementary Figs. S4 and S5).

Because K562 cells are highly sensitive to NK cell-mediated killing, we hypothesized that differences in natural cytotoxicity might be more evident in longer-term assays with targets that are less susceptible to NK-cell killing. To test this, we incubated NK cells from the same three culture conditions described above at a 3:1 E:T ratio with three different solid tumor cell lines: A549 lung carcinoma cells, SKOV-3 ovarian adenocarcinoma cells, and PANC-1 epithelioid carcinoma cells transduced with red fluorescent protein. Target-cell killing was measured each hour over the course of 48 hours by imaging using the IncuCyte Live Cell Analysis System. An anti-CD20 antibody, which does not bind NK cells or any of the target cells used in this assay, was added as a control. NK cells expanded ex vivo with CHIR99021 exhibited statistically significant natural cytotoxicity against A549 cells (P = 0.007; Fig. 5D) and SKOV-3 cells (P = 0.0311; Fig. 5E), whereas killing by NK cells expanded ex vivo with DMSO or primed overnight with IL15 did not reach significance. NK cells expanded ex vivo with DMSO (P = 0.003) or CHIR99021 (P < 0.001) exhibited significant natural cytotoxicity against PANC-1 cells, and the difference in killing between the 7-day CHIR99021 and overnight IL15 conditions also reached statistical significance (P < 0.001; Fig. 5F).

To test the ADCC function, parallel killing assays were set up where anti-HER-2 and anti-EGFR antibodies were added. A549 and PANC-1 cells express high levels of EGFR and lower levels of HER-2, while SKOV-3 cells express high levels of both EGFR and HER-2. NK cells expanded ex vivo with CHIR99021 were the only cells in this assay to exhibit statistically significant ADCC against A549 cells in the presence of antibodies against both HER-2 (P = 0.007) and EGFR (P < 0.001; Fig. 5G) and against SKOV-3 cells in the presence of antibodies against Her-2 (P = 0.003) and EGFR (P = 0.009; Fig. 5H). Similar to results from natural cytotoxicity assays, PANC-1 cells were more susceptible to killing via ADCC. Significant killing of PANC-1 cells was observed for NK cells cultured overnight with IL15 in the presence of anti-EGFR antibodies (P = 0.003), NK cells expanded ex vivo with DMSO in the presence of antibodies against both HER-2 (P < 0.001) and EGFR (P < 0.001) and NK cells expanded ex vivo with CHIR99021 in the presence of antibodies against both HER-2 (P < 0.001) and EGFR (P < 0.001). The difference in ADCC-mediated killing of PANC-1 cells between NK cells cultured overnight with IL15 and NK cells expanded ex vivo with CHIR99021 in the presence of anti–HER-2 antibody also reached statistical significance (P = 0.010; Fig. 5I). Together, these results demonstrate that inhibition of GSK3 during ex vivo expansion primes NK cells for enhanced inflammatory cytokine production and elevated cytotoxicity.

We next wanted to test the function of NK cells expanded ex vivo with CHIR99021 in an in vivo tumor model. We used a xenogeneic model where luciferase-expressing SKOV-3 tumors were established in NSG mice. Groups of mice were then treated with Herceptin antibody alone, antibody plus adoptively transferred NK cells that were primed overnight with IL15, antibody plus adoptively transferred NK cells that were expanded ex vivo for 7 days with 10 ng/mL IL15 and DMSO or antibody plus adoptively transferred NK cells that were expanded ex vivo for 7 days with 10 ng/mL IL15 and 5 μmol/L CHIR99021. Mice were then imaged repeatedly for luciferase intensity over the course of 40 days. Representative images of mice from each group in one experiment are shown in Fig. 6A. Collectively, mice treated with antibody in combination with adoptively transferred NK cells from ex vivo expanded CHIR99021 cultures showed more consistent tumor control, and this treatment group was the only one that reached statistical significance with respect to reduced luminescence relative to the untreated group (P < 0.05; Fig. 6B).

To determine whether the mature phenotype of NK cells after culture with CHIR99021 was maintained in vivo, we adoptively transferred NK cells expanded ex vivo for 7 days with 10 ng/mL IL15 and 5 μmol/L CHIR99021 into the peritoneum of NSG mice. Mice were injected i.p. with IL2 twice per week for 2 weeks, and NK cells were harvested for phenotypic analysis 14 days after adoptive transfer. In these experiments, we observed similar frequencies of both total CD3⁻CD56⁺CD57⁺ NK cells and adaptive CD3⁻CD56⁺CD57⁺NKG2C⁺ NK cells before and after adoptive transfer (Supplementary Fig. S6). Together, these results show that inhibition of GSK3 enhances NK-cell antitumor function after adoptive transfer.

Here, we describe a novel approach for ex vivo expansion and maturation of highly functional mature NK cells for adoptive therapy. The goal in developing this approach was to achieve a higher functional activity that was superior to the NK-cell product we have used in over 100 patients with advanced leukemia and solid tumors with the expectation that such a product would increase clinical efficacy. We show that the GSK3 inhibitor CHIR99021 markedly enhances NK-cell maturation during expansion. To the best of our knowledge, no NK-cell expansion strategy that promotes both NK-cell expansion and maturation has been reported to date.

An enrichment of CD57⁺ NK cells may be particularly important from an immunotherapy standpoint given the results of several clinical studies that have reported an association between CD57⁺ NK cells and survival in cancer patients. Higher numbers of tumor infiltrating CD57⁺ NK cells are associated with increased survival and tumor regression in patients with esophageal squamous cell carcinoma, squamous cell lung carcinoma, and pulmonary adenocarcinoma (12). In addition, work from our group has shown that higher absolute counts of CD57⁺NKG2C⁺ NK cells at 6 months post-HCT are associated with lower relapse rates at 2 years posttransplant (30).

Functionally, NK cells expanded with CHIR99021 exhibited elevated TNF and IFNγ production and enhanced natural cytotoxicity and ADCC against multiple cancer lines with variations in HLA-I content and expression levels in vitro. Despite the increased inhibitory KIR expression on NK cells cultured with CHIR99021, we did not observe any overall inhibition of killing activity. This is likely due to the stimulatory effects of IL15 and marked induction of both perforin and granzyme B when IL15 and CHIR99021 and used together. Importantly, these NK cells also demonstrated more robust and consistent tumor control in vivo. Interestingly, the most profound differences with respect to cytotoxicity between NK cells expanded ex vivo with vehicle or CHIR99021 were evident at later time points (6 hours and beyond) during live imaging analyses of killing. Whether this effect was due to greater persistence, motility, or increased serial killing is not yet clear. Future studies analyzing the cytotoxic function of subsets of NK cells on a per cell basis may yield more mechanistic insights.

The enhanced maturation observed for NK cells expanded with CHIR99021 correlated with increased expression of T-BET, ZEB2, and BLIMP-1. Both T-BET and its homolog EOMES control NK-cell transit through maturational checkpoints and promote NK-cell cytotoxicity and IFNγ production (31). T-BET controls late-stage NK-cell maturation by increasing transcription of ZEB2, which is essential for the survival of mature NK cells, and BLIMP-1, which is required for NK-cell maturation and homeostasis (27, 28). Our results are consistent with a recent study by Taylor and colleagues, who demonstrated increased Tbx21 expression in mouse CD8⁺ T cells cultured ex vivo with GSK3 inhibitors or with siRNA knockdown of GSK3. Tbx21 directly suppressed expression of Pdcd1 (PD1) transcription, which potentiated CD8⁺ cytolytic T-cell responses and viral clearance (20). Of note, we observed low-to-absent expression of PD1 on NK cells regardless of culture condition. The mechanism by which GSK3 inhibition increases TBX21 transcription is not yet known and will be challenging to determine definitively given the range of targets and multifarious roles of GSK3 in cellular signaling pathways.

Taken together, our results demonstrate that small molecule modulation using a specific GSK3 inhibitor potentiates functional NK-cell maturation and CD57 acquisition. Adoptive transfer of NK cells with enhanced effector function including natural cytotoxicity, ADCC, and cytokine responses to target cell exposure along with lower expression of NK-cell inhibitory checkpoint receptors should lead to better clinical efficacy. These findings are of immediate translational value and constitute the basis for expanding NK cells ex vivo with IL15 and CHIR99021 to treat patients with advanced cancer using adoptive NK-cell infusions. We have scaled up this process under cGMP conditions, and a first-in-human clinical trial is open for accrual (NCT03081780).

F. Cichocki is a consultant at, reports receiving other commercial research support from, has ownership interest (including patents) in, and is a consultant/advisory board member for Fate Therapeutics. B. Valamehr is the vice president at and has ownership interest (including patents) in Fate Therapeutics. S.E. Abbot has ownership interest (including patents) in Fate Therapeutics. D. Shoemaker is a CSO at and has ownership interest (including patents) in Fate Therapeutics. Y.T. Bryceson has ownership interest (including patents) in patent “Method for Expansion of NK Cells Exhibiting an ‘Adaptive’ Phenotype (Defined by Expression of NKG2C and CD57 on CD56dim NK Cells and Silencing of Proximal Signaling Proteins.” S. Wolchko has ownership interest (including patents) in Fate Therapeutics, Inc. J.S. Miller is a consultant/SAB at, reports receiving other commercial research support 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: F. Cichocki, B. Valamehr, D. Sarhan, S.E. Abbot, D. Shoemaker, Y.T. Bryceson, B.R. Blazar, S. Wolchko, J.S. Miller

Development of methodology: F. Cichocki, B. Valamehr, R. Bjordahl, B. Rezner, P. Rogers, S. Gaidarova, D. Sarhan, S.E. Abbot, S. Wolchko, J.S. Miller

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Cichocki, R. Bjordahl, B. Zhang, B. Rezner, P. Rogers, S.K. Moreno, K. Tuininga, P. Dougherty, V. McCullar, P. Howard, D. Sarhan, E. Taras, H. Schlums

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Cichocki, B. Valamehr, R. Bjordahl, P. Rogers, S. Gaidarova, S.K. Moreno, V. McCullar, P. Howard, D. Sarhan, D. Shoemaker, B.R. Blazar, S. Cooley, J.S. Miller

Writing, review, and/or revision of the manuscript: F. Cichocki, B. Valamehr, B. Rezner, P. Rogers, D. Sarhan, S.E. Abbot, Y.T. Bryceson, B.R. Blazar, S. Wolchko, S. Cooley, J.S. Miller

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases):F. Cichocki, B. Zhang

Study supervision: F. Cichocki, B. Valamehr, J.S. Miller

We thank the University of Minnesota Flow Cytometry Core and Biomedical Genomics Core for their assistance.

This work was supported by the following NIH grants: P01 CA111412 (F. Cichocki, S. Cooley, and J.S. Miller), P01 CA65493 (S. Cooley, B.R. Blazar, and J.S. Miller), R35 CA197292 (S. Cooley and J.S. Miller), R01 HL122216 (J.S. Miller), R01 AI34495 (B.R. Blazar), R01 HL56067 (B.R. Blazar), and K99HL123638 (F. Cichocki). Supplemental funds were provided from Fate Therapeutics.

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.