Human peripheral blood mononuclear cells (PBMCs) were isolated from peripheral blood products by density gradient centrifugation (Histopaque^®-1077 HybriMax™). Depletion of CD3+ cells was achieved using CD3 MicroBeads and LD columns, and enrichment of CD56+ cells with CD56 MicroBeads and LS columns. MACS^® MultiStand and QuadroMACS™ Separator was used for magnetic separation (all from Miltenyi Biotec).

Three selection methods [Depletion of CD3+ cells (CD3-), enrichment of CD56+ cells (CD56+), and their combination (CD3-/CD56+)], four activation protocols (continuous, pre-activation, re-activation, and boost) and four cytokine cocktails [IL-21 (25 ng/ml) + IL-15 (10 ng/ml); IL-15 (10 ng/ml) + IL-18 (50 ng/ml) + IL-12 (10 ng/ml); IL-15 (10 ng/ml) + IL-18 (50 ng/ml) + IL-27 (10 ng/ml); IL-2 (100 IU/ml) + IL-15 (10 ng/ml)] were used for the expansions ( Supplementary Figure 1 ). Concentrations were based on the following publications and preliminary studies: IL-2 (14), IL-15 (24, 27), IL-21 (27), IL-12 and IL-18 (25, 28), and IL-27 (24). The following cytokines were used: premium grade recombinant human (rh) IL-2 Improved Sequence (IS), rhIL-12, rhIL-15; research grade rhIL-21, rhIL-27 (all from Miltenyi Biotec), and rhIL-18 (InvivoGen). NK cells were cultured in RPMI-1640 medium (Lonza) supplemented with 5% human AB serum (HABS; Sigma-Aldrich), 2 mM GlutaMAX™ (Gibco), 100 U/ml penicillin and 100 µg/ml streptomycin (P/S, Gibco) (hereinafter referred to as complete medium). Cells were cultured in a humidified atmosphere at 37°C, with 5% CO₂ for 16 days. Day 16 was selected based on previous studies (14, 29, 30) and preliminary experiments (unpublished data). Expansions were started in T25 flasks, cells were counted and complete medium added, or partly changed, every 2-3 days and the cells were kept approximately at 1x10⁶ cells/ml during expansion. Cell count and viability were determined using NucleoCounter^® NC 200™ automated cell counter (ChemoMetec), or Guava MUSE^® cell analyzer using Muse Count & Viability Kit (Luminex).

Used activation protocols are illustrated in Supplementary Figure 1 : Pre-activation (Pre) was done as previously in (25) with 2x10⁶ cells/ml density for 16-18 hours with IL-21/15, IL-27/18/15 or IL-12/18/15, after which the medium was changed to complete medium with 10 ng/mL IL-15, and the cell concentration was adjusted to 1x10⁶/ml. Re-activation (Re) started on day 15 (16-18 hours before harvest), with same cytokines used for pre-activation. Boosted cells (Boost) were expanded in complete medium with 10 ng/ml IL-15 and the boosting cytokines (IL-21/15, IL-27/18/15, or IL-12/18/15) were added 16–18 hours before harvest on day 16. For the continuous group (Cont), the cytokines were added every 2-3 days with the fresh complete medium.

DK-MG (ACC277, glioblastoma multiforme) was obtained from German Collection of Microorganisms and Cell Cultures (DSMZ), U-87 MG human EGFR vIII (CL01004-CLTH, epithelial, likely glioblastoma, transduced with epidermal growth factor receptor variant III (EGFR vIII), hereinafter referred to as U87vIII) from Amsbio, U-87 MG (HTB-14, epithelial, likely glioblastoma, hereinafter referred to as U87-wt), LN-229 (CRL-2611, epithelial, glioblastoma) and U-118 MG (HTB-15, mixed, grade IV glioblastoma, hereinafter referred to as U-118) both from ATCC. U87-wt, U87vIII, LN-229, and U-118 were maintained in DMEM (high glucose and pyruvate, Gibco) supplemented with 10% fetal bovine serum (FBS, gamma irradiated, Gibco) and 100 U/ml penicillin and 100 µg/ml streptomycin (P/S). In addition to this, U87vIII medium was supplemented with 200 µg/ml of the selection antibiotic Geneticin (Gibco). DK-MG were maintained with RPMI-1640 supplemented with 10% FBS, P/S, and 2 nM GlutaMax. Cells were passaged when they reached 80-90% confluency.

Cells were harvested and washed with FACS Buffer [phosphate buffered saline (PBS, Gibco) + 0.02% Sodium Azide (NaN₃, Sigma) + 2 mM ethylenediaminetetraacetic acid (EDTA, Sigma) + 1% bovine serum albumin (BSA, Biomol)]. Before staining, samples were blocked with 10% human FcR Blocking Reagent (Miltenyi Biotec) in FACS buffer. Antibody panels were added to the cells for 15 min at RT then washed with FACS Buffer (centrifugation 300 x g, 6 min). The following anti-human mAbs were used (indicated as target and fluorochrome): CD45-APC Fire750, CD3-BV510, and CD56-BV421 to evaluate the NK cell purity; CD25-PE, CD16-VB515, NKp46-PECy7, and CD69-APC as NK cell phenotype markers; ITGAL-FITC, TRAIL-APC, NKG2D-BV605, and FasL-PE as markers for functionality; MIC-A/B-PE, ULPB-1-PE, ICAM-1-APC, and EGFR-APC for target phenotyping. Used clones are listed in Supplementary Table 1 . All fluorescent mAbs were obtained from Biolegend, except the CD16 mAb from Miltenyi Biotech and EGFR from Absolute Antibody. 7-amino-actinomycin D (7-AAD) was used in all panels to identify the live and dead cells (BD Biosciences). Cells were run on CytoFLEX S (Beckman Coulter) and analyzed with CytExpert v2 (Beckman Coulter) and FlowJo v10 (BD Life Sciences, TreeStar).

Cytotoxicity was determined from the co-cultures with Cell Counting Kit-8 (CCK-8; CK04-13, Dojindo). Target glioma cells (T) were seeded into flat 96-well plates at a density of 2 x10⁴ cells per well, in cell type-specific medium (see above). Cells were allowed to settle in wells at room temperature (RT) for 15 minutes before 24 hours incubation at 37°C, 5% CO₂. After 24 hours, the medium was discarded and NK cells (Effectors; E) were added in RPMI-1640 medium supplemented with 5% HABS, 2 mM GlutaMAX and 100 U/ml penicillin 100 µg/ml streptomycin, without cytokines. Three effector to target (E:T) cell ratios (0.5:1, 1:1, and 2:1) were used. E:T ratios were calculated based on the amount total viable effector cells. In some experiments, additional cytokines, 5 ng/ml IL-15 or 20 IU/ml IL-2 were added to the co-cultures. After 24 hours co-culture the target cell viability was determined with CCK-8 [10% CCK-8 in RPMI-1640 without phenol red (Lonza)] following the manufacturer’s instructions, with 1 hour incubation. Absorbance was measured at 450 nm with Cytation5 (BioTek) multi-mode reader. After blank subtraction the relative cytotoxicity was calculated against the control (targets only) wells as follows:

To analyze NK cell degranulation, CD107a expression was measured before (baseline expression) and after co-culturing with target cells by flow cytometry. Staining was done as in flow cytometric characterization. For the co-culturing, 1.5x10⁵ target glioma cells were seeded per well into 24-well plates. After 24 hours, the NK cells were added in 1:1 E:T ratio, and after 3 hours co-culturing the cells were collected and stained with CD45-APC-Fire750, CD107a-PE (Miltenyi Biotec) and 7-AAD. CD107a expression was measured from viable CD45 positive population with CytoFLEX S and analyzed with CytExpert v2 and FlowJo v10.

Statistical analyses were performed using GraphPad Prism 9. Comparisons between groups were performed using two-way ANOVA with Tukey’s or Dunnett’s post hoc test. P < 0.05 was defined as statistically significant. Results are shown as mean ± SEM.

NK cells were isolated from peripheral blood mononuclear cells (PBMCs) from voluntary blood donors with informed consent at the Finnish Red Cross Blood Service (Helsinki, Finland) and handled under an approval by the Ethics Committee, Hospital District of Northern Savo (1480/13.02.00/2019). These cells were isolated from buffy coats not required for treatment of patients. All in all, the buffy coats of 25 donors were used during the studies. Donor information is shown in Supplementary Tables 5 , 6 .

We studied 39 culture conditions with the aim of identifying an optimal expansion method for pure and cytotoxic NK cells. These culturing conditions ( Supplementary Figure 1 ) comprised of three selection methods [Depletion of CD3+ cells (CD3-), enrichment of CD56+ cells (CD56+), and their combination (CD3-/CD56+)]; four activation protocols [continuous (Cont), 16-18 h pre-activation (Pre), pre-activation followed by re-activation on day 15 (Re), and culturing in IL-15 followed by short activation on day 15 (Boost)], and four cytokine combinations (IL-2/15, IL-21/15, IL-27/18/15 and IL-12/18/15). At the end of the culture (day 16), the cells were analyzed for their viability, expansion rate, purity, phenotype, and cytotoxicity against the U87-wt glioma cell line.

First, we studied the impact from the selection methods to the expansion, cytotoxicity, and end purity. These results were used to determine the optimal starting material for the subsequent studies. After selection, the cells were expanded with different culture conditions ( Supplementary Figure 1 ). CD3-/CD56+ selection resulted in the highest end purity and cytotoxicity with all activation protocols ( Figures 1A, B ; Supplementary Table 5 ), no significant differences between expansion ratios were detected between the selection methods ( Figure 1C ). Next, we compared purities ( Figures 1D, G, J ) cytotoxicity ( Figures 1E, H, K ) and expansion rate ( Figures 1F, I, L ) within the selection method. CD3-/CD56+ selection method produces most pure NK cells ( Figure 1D ) while CD3 depletion ( Figure 1G ) and CD56 enrichment ( Figure 1J ) had more variation between expansion methods. Highest overall cytotoxicities were gained with CD3-/CD56+ selection ( Figure 1E ) compared to CD3 depletion ( Figure 1H ) and CD56 enrichment ( Figure 1K ). Expansion rates were overall highest with the CD3 depletion ( Figure 1I ) and lowest with CD56 enrichment ( Figure 1L ). No significant differences were observed within the selection methods due to donor variation. Cytotoxicity was increased with pre-activation and boost with all conditions and was most prominent with CD3-/CD56+ selection ( Figure 1E ). Unlike the other cytokine combinations tested, which resulted in poor survival and expansion if continuously present, the IL-2/15 combination was tolerated continuously during the expansion ( Figures 1D–L , condition 10). Other continuous protocols were omitted due to low cell viability and expansion rates. At this stage, one selection method (CD3-/CD56+) and 10 culturing conditions ( Supplementary Figure 2 , Supplementary Table 3 ) from the initial conditions, were selected for subsequent experiments based on the purity and expansion rate.

NK cells were identified as CD3 negative cells and CD56 positive (CD3-CD56+) within the CD45+ leukocyte population ( Figure 2A ). We noticed that the CD56 expression shifts from dim to bright during the expansion ( Supplementary Figure 3 ). This was seen with all methods. To examine further how the used cytokine and activation methods alter the expression of known NK cell markers, we characterized the expression of CD69, CD25, CD16, and NKp46. The markers CD69 and CD25 represent NK cell activation and maturation, CD16 represents the ADCC activity and NKp46 the natural cytotoxicity ( Figures 2B–F ; Supplementary Figure 4 ).

From the measured NK markers the CD69 expression was induced during the expansion in all conditions being highest after boosting with all cytokine combinations (IL-2/15, IL-21/15, IL-27/18/15, IL-12/18/15) and lowest with IL-12/18/15 pre-activation ( Figure 2B ; Supplementary Figure 4 ). CD25 expression was upregulated with all methods except continuous IL-2/15 being lowest after pre-activation and highest after boosting ( Figure 2C ). IL-12/18/15 cytokine combination had the most profound impact on the phenotype by decreasing CD16 and NKp46 expression. The CD16 downregulation was most evident with IL-12/18/15 re-activation and boost ( Figure 2D , conditions 8 and 9). CD16 downregulation was also seen to some extent when NK cells were boosted with IL-27/18/15 ( Figure 2D , condition 6) compared to pre-activation with the same cytokines. NKp46 expression was lowest with IL-12/18/15 re-activation ( Figure 2E , condition 8) when compared to other combinations (conditions 1-7, 9-10).

For the next set of studies NK cells, after CD3-/CD56+ selection, were expanded with 10 different expansion methods (conditions (1-10); Supplementary Figure 2 ; Supplementary Table 3 ). The NK cell cytotoxicity was assessed against glioblastoma cell line U87wt. Donor variation in cytotoxicity was observed with most of the activation methods. This variation was lowest with IL-21/15 and IL-12/18/15 re-activations with higher effector to target (E:T) ratios (2:1) ( Figure 3A , conditions 2 and 8; Supplementary Table 4 ). The pre-activated (Pre, conditions 1, 4 and 7) NK cells showed lower while the boosted (Boost; conditions 3, 6 and 9) exhibited higher cytotoxicity in response to all cytokine combinations ( Figure 3A ; Supplementary Table 4 ). This finding correlated with the NK cell degranulation, as measured by lysosome associated membrane protein-1 (CD107a) expression, after expansion and co-culturing with tumor targets ( Figure 3B ). Our results show a trend that high expansion rates can decrease the NK cell cytotoxicity. This was seen particularly with IL-12/18/15 pre-activation ( Figure 3C , conditions 7). This phenomenon could be obscured in other conditions due to the low expansion rates or be specifically related to the IL-12/18/15 combination, or due to donor variation ( Figure 3C ).

As the expansion rate and method impacted the cytotoxicity we wanted to see if NK cells would respond to additional cytokines after ex vivo expansion. To study this, we added IL-2 or IL-15 with the NK cells to the co-cultures. Additional IL-15 (5 ng/ml) enhanced the cytotoxicity of all, except continuous IL-2/15 ( Figure 3D , condition 10) expanded NK cells, whereas IL-2 (20 IU/ml) increased the cytotoxicity only for the IL-12/18/15 re-activated NK cells ( Figure 3D , condition 8). As seen in the phenotype, the continuous IL-2/15 generated NK cells with low CD25 expression ( Figure 2C , condition 10), which could explain the differences in NK cell responsiveness to additional IL-2.

To determine how the measured markers contribute to the cytotoxicity, we conducted a multiple regression analysis and a principal component analysis (PCA) from the following parameters: activation protocol, expansion rate, phenotype markers (CD25, CD69, CD16, NKp46), and CD107a ( Figure 4A ). Continuous IL-2/15 protocol was used as a reference level for the multiple linear regression analysis. It can be concluded that by multiple regression, using least-squares, the expression of CD107a, CD69, NKp46, and CD16, the activation protocol (treatment), and the expansion rate are significant measures of the variability in cytotoxicity. Furthermore, the pre-activation with IL-21/15 (IL-21/15_Pre), re-activation with IL-21/15, IL-27/18/15, or IL-12/18/15 (IL-21/15_Re, IL-27/18/15_Re, IL-12/18/15_Re), and boosting with IL-27/18/15 or IL-12/18/15 (IL-27/18/15_Boost, IL-12/18/15_Boost) increase the cytotoxicity significantly when compared to continuous IL-2/15 (IL-2/15_Cont) ( Figure 4A ). The PCA plots ( Figures 4B–D ) reveal the correlations between these parameters. They show that cytotoxicity correlates positively with CD107a and the phenotype markers CD25, CD69, CD16, and NKp46, and negatively with the expansion rate (eigenvectors in Figure 4B ). The highest expansion rates were achieved with the IL-12/18/15 pre-activation ( Figure 4C , light blue) while re-activation (Re, blue) increased the cytotoxicity ( Figure 4D ). Highest cytotoxicity, as already shown in Figure 3 , was achieved with IL-21/15 and IL-12/18/15 boosting ( Figure 4D , color scale shows the cytotoxicity), however, with modest expansion rates ( Figure 4C , size indicates the expansion rate).

At the final part of our study, we focused on the activation timing (Pre, Re, Boost). These studies were performed with the combination of IL-12/18/15, which had highest expansion rate within the cytokine combinations ( Figure 3C , conditions 7-8), and continuous IL-2/15 was used as a control condition. We also measured the expression of additional NK markers ( Figures 5A–E ; Supplementary Figure 5 ) FasL, NKG2D, TRAIL, and ITGAL. Boosting induced the highest expression of TRAIL ( Figure 5C ) and FasL ( Figure 5A ), while IL-12/18/15 pre-activation and continuous IL-2/15 induced the highest expressions of NKG2D ( Figure 2B ) and ITGAL ( Figure 5D ). NK cell functionality was measured with cytotoxicity ( Figure 6A ) and degranulation (CD107a expression, Figure 6B ) against five glioma cell lines (U87-wt, U87vIII, DK-MG, LN229 and U-118). Also, the cytokine secretion profile was assessed against two cell lines (U-87-wt and U-118). Degranulation increased in response to re-activation and boosting, being highest after boosting ( Figures 6A, B , conditions 8 and 9). This finding was verified with degranulation related cytokine secretion profile ( Supplementary Figure 6 ; IFN-g, GM-CSF) against U87-wt and U-118 cell lines. Cytotoxicity increased after re-activation and boost, as did the degranulation, against all cell lines, except LN-229 ( Figure 6A , conditions 8-9). The cytotoxicity of IL-12/18/15 (re-activated; condition 8) and IL-2/15 (continuous; condition 10) expanded NK cells varied between the cell lines. U87-wt and U87vIII were more sensitive to IL-12/18/15 re-activated NK cells, while DK-MG, and U-118 cell lines responded better to IL-2/15 expanded NK cells ( Figure 6A , conditions 8 and 10).

Since the differences in the target cell killing were evident, we characterized the expression of certain cytotoxicity-associated ligands from the target cell lines in an attempt to explain the differences in their sensitivity. We measured expression of the NKG2D ligands (UL16 binding protein (ULPB1) and MHC class I chain-related glycoproteins A and B (MICA and B)), LFA-1 ligand (ICAM-1), TRAIL and TRAIL receptors 1 and 2 (TRAILR1/2) and Fas ligand (FasL), and expression of CD56, CD45, EGFR and EGFRvIII. All target cell lines were positive for Fas and TRAILR1/2 ( Supplementary Table 2 ). The expression of other markers varied between cell lines but could not explain the differences in the sensitivity. For example, the U-118 cell line, which was one of the most resistant targets, had the lowest expression of MICA/B and ULPB1. Another resistant cell line, LN229, had a similar expression profile to the most sensitive cell line, U87-wt.