Buffy coats from healthy donors (Klinikum Dortmund) were used for PBMC isolation. Daudi JP cells were a kind gift of Dr. R. Seggewiss-Bernhardt, University Hospital of Wuerzburg, who obtained them from Prof. P. Fisch, University of Freiburg (21), and K562 cells were obtained from the German Collection of Microorganisms and Cell Cultures (DSMZ). Both cell lines were maintained in complete RPMI medium, RPMI 1640 (Biowest) supplemented with 10% fetal bovine serum (Biochrom), and 2 mmol/L L-glutamine (PAA), in a humidified atmosphere with 5% CO₂ at 37°C.

Peripheral blood mononuclear cells were isolated from buffy coats by density gradient centrifugation using Pancoll (PAN-Biotec). To obtain different feeder cell fractions, PBMCs were depleted first of CD56⁺ cells by MACS sort using anti-CD56 microbeads (Miltenyi Biotec), and when indicated they were further depleted of CD19⁺ and CD3⁺ or CD14⁺ cells using the corresponding microbeads (Miltenyi Biotec). The differently depleted PBMCs fractions were X-ray irradiated (20 Gy) (RS 2000 Biological Research Irradiator, Radsource) and used as autologous feeder cells for co-culture with NK cells within 1–1.5 h post-irradiation. The irradiated autologous CD56-depleted PBMCs (“IAPs”) were used as the major feeder cells. NK cells were purified from PBMCs using the human NK cell isolation kit (Miltenyi Biotec), and expanded in 24-well plates, either in co-culture with autologous feeder cells at a 20:1 feeder-NK cell ratio, based on protocol from Ahn et al. (16), or without feeder cells, in complete culture medium; TexMACS medium (Miltenyi Biotec) supplemented with 5% human AB serum (Life Technologies) and 1,000 U/mL of Proleukin S (rhIL-2) (Novartis). Co-cultures were additionally supplemented with 10 ng/mL of anti-CD3 mAb (functional grade OKT3, Miltenyi Biotec). When indicated, NK cells were labeled prior cultivation with CellTrace™ Violet Cell Proliferation Kit [cell trace violet (“CTV”) dye] (Life Technologies). Initial total cell densities of cultures were 1 × 10⁶ cells/mL. Culture plates were incubated in a humidified atmosphere with 5% CO₂ at 37°C. To determine NK cell fold expansions, NK cell densities were checked at different indicated time points by volumetric counting and detection of viable CD3⁻CD56⁺ cells using MACSQuant Analyzer 10 (Miltenyi Biotec). For 12-day expansions, NK cells were harvested from 24-well plates on day 7 and transferred to 25 or 75 T-flasks with replenishment of fresh medium without OKT3. Fresh medium was additionally replenished on day 9 or 10.

Purified NK cells were cultured for up to 5 days with or without IAPs in 12-well polystyrene plates equipped with Transwell^® inserts (Costar). The insert system consisted of a 500 µL upper well (12 mm diameter), separated from the bottom well (1.5 mL) by a 0.4 µm microporous tissue culture-treated polycarbonated membrane. All cells were resuspended in complete culture medium and IAPs suspensions were supplemented with OKT3 as described. Four different culture conditions were established, all of them containing 1 mL of NK cells (5 × 10⁵ cells/mL) at the bottom wells. In two conditions, either 1 mL of complete medium or 1 mL of IAP suspension were added to the NK cells on the bottom wells, and upper wells were left empty. In two other conditions, either 1 mL of IAP suspension (10 × 10⁶ cells/mL) was inoculated through the insert, or a combination of 0.5 mL of IAPs (10 × 10⁶ cells/mL) and 0.5 mL of purified NK cells (5 × 10⁵ cells/mL) was inoculated, remaining the inoculated cells in both situations on the upper well of the insert. Transwell plates were incubated in a humidified atmosphere with 5% CO₂ at 37°C.

Natural killer cell frequencies were determined using the following panel of mAbs from Miltenyi Biotec: CD45-VioGreen (5B1), CD3-FITC (BW264/56), BDCA1-PE (AD5-8E7), CD14-PE (TÜK4), CD19-PEVio770 (LT19), CD16-APC (VEP13), and CD56-APCVio770 (REA196). In experiments with feeder cell fractions containing additional depletions of CD19⁺ and CD3⁺ or CD14⁺ cells, the mAb BDCA-2-PE (AC144) was included instead of CD14-PE, and CD16-APC replaced by CD14-APC. The human FcR Blocking Reagent from Miltenyi was also used to block unspecific antibody binding to Fc receptors on CD14⁺ enriched feeder cell fractions. Propidium iodide (Miltenyi Biotec) was used at a final concentration of 1 µg/mL to exclude dead cells from the analysis.

To study the phenotypic differences between resting, proliferating, and non-proliferating NK cells, cells were labeled prior cultivation with or without IAPs, with the cell trace proliferation dye eFluor^®670 (eBioscience), in order to differentiate proliferating and non-proliferating cells. Different candidate molecules were analyzed using several antibody panels designed based on previous work (22). These panels shared a backbone of mAbs: CD45-VioGreen, CD3-VioBlue, TCRγδ-VioBlue (11F2), CD14-VioBlue, CD19-VioBlue, and CD56-APCVio770, and SYTOX^® Blue (Life Technologies) was used to exclude dead cells from the analysis. The backbone was combined with groups of the following mAbs (Miltenyi Biotec unless otherwise indicated): KLRB1-FITC (191B8), CTLA-4-PE (BNI3), Siglec-9-PEVio770 (REA492), Siglec-7-PerCP700 (REA214), LILRB1-FITC (GHI/75), KLRG1-PE (REA261), CD16-PerCP700 (VEP13), RANKL-PE (DN254), B7-H3-PEVio770 (FM276), NKp44-PEVio770 (2.29), NKp80-FITC (4A4.D10), 4-1BB-PE (4b4-1), 2B4-PEVio770 (REA112), NKG2D-PerCPCy5.5 (1D11) (Biolegend), HLA-DP/DQ/DR-FITC (REA332), ALCAM-PEVio770 (REA442), HLA-A/B/C-PerCPVio700 (REA230), and HLA-DR-FITC (AC122). The corresponding mouse immunoglobulin (Ig) G1, IgG2A, IgG2B, IgM, or REAs conjugated with the respective dyes were used as isotype controls. Cells were acquired using MACSQuant Analyzer 10 and analyzed using MACSQuantify 2.8 software (Miltenyi Biotec).

Supernatants from co-cultures of NK cells with IAPs further depleted of CD19⁺, and CD3⁺ or CD14⁺ cells were collected after 5 days and cytokine production was detected using the flow cytometry bead-based array MACSPlex Cytokine 12 kit, human according to manufacturer’s instructions (Miltenyi Biotec). The MACSPlex Cytokine 12 kit allows for the detection of human GM-CSF, IFN-α, IFN-γ, IL-2, IL-4, IL-5, IL-6, IL-9, IL-10, IL-12p70, IL-17A, and TNF-α. Samples were acquired using a MACSQuant Analyzer 10 and analyzed using the Express Mode option of the MACSQuantify 2.8 software (Miltenyi Biotec).

Target cell killing of K562 and Daudi JP cell lines was analyzed using a flow cytometry-based assay (23). Briefly, target cells were labeled with CTV dye and seeded in 96 well U-bottom plates at a cell density of 2 × 10⁵ cells/mL, and incubated alone, or with expanded NK cells (2 × 10⁶ cells/mL) at different effector-to-target (E:T) ratios, for 4 h in a humidified atmosphere with 5% CO₂ and 37°C. Antibody-dependent cellular cytotoxicity (ADCC) of expanded NK cells was further analyzed against the CD20⁺ Daudi cells, by adding 5 µg/mL of the anti-CD20 mAb rituximab (Hoffman-La Roche). After 4 h incubation, plates were transferred to 4°C for at least 30 min to stop cell killing before quantifying viable CTV-positive target cells using the MACSQuant Analyzer 10. The frequency of killed target cells was calculated by the difference between the number of viable target cells in samples with effector NK cells and samples with targets cells containing no effector cells. Representative flow cytometry data are included in Figure S1 in Supplementary Material.

Natural killer cells from five different donors were labeled with the cell trace proliferation dye eFluor^®670 (eBioscience) and co-cultured for 5 days with IAPs. On day 5, cells were harvested and proliferating and non-proliferating NK cells were sorted according to the brightness of the cell trace dye in a FACSAria III cytometer (BD Bioscience, cell sorting facility of the Center for Molecular Medicine of Cologne, Cologne, Germany). 7-AAD (BD Pharmigen™) was used to exclude dead cells. To ease the sorting process, prior to sort, samples were enriched on NK cell content using the human NK cell isolation kit from Miltenyi. Sorted fractions containing the proliferating and non-proliferating (≈0.8 × 10⁵ cell in each fraction) were lysed in RA1 buffer (Macherey-Nagel) for total RNA isolation, and stored at −20°C until use. For comparison, resting NK cells (1 × 10⁶ cells) were lysed and stored under identical conditions.

Total RNA from sorted samples was isolated using the NucleoSpin^® RNA kit (Macherey-Nagel), amplified and labeled using the Agilent Low Input Quick Amp Labeling Kit (Agilent Technologies), prior to hybridization to Agilent Whole Human Genome Oligo Microarrays 8 × 60K V2 chips (Agilent Technologies) according to manufacturer instructions. A detailed description of microarray processing can be found in Supplementary Material.

Raw intensity data from feature extraction output files (FES 10.7.3.1, Agilent Technologies) were analyzed using the Rosetta Resolver^® software (Rosetta Biosoftware). The following calculations were performed with software packages within R/Bioconductor (24, 25). Intensity values were corrected with background subtraction and normalized by quantile normalization (26). Reliable signal intensities were considered significant when p ≤ 0.01, using the Rosetta error model (27). Subsequent statistical analysis was performed on normalized Log2-transformed intensity values. The data set can be found at the NCBI GEO public database with the accession number GSE92512.

Conventional statistics including parametric un-paired Student’s t-test and one-way ANOVA with Tukey’s post hoc test, or the non-parametric Wilcoxon test and Kruskal–Wallis test with post hoc Dunn’s test were performed with Graph Pad Prism 7 software (GraphPad). All statistical analysis were two-sided and p < 0.05 considered statistically significant, and indicated as *p < 0.05, **p < 0.01, and ***p < 0.001. Only statistical differences are shown.

Description of the statistical analysis of the microarray data can be found in Supplementary Material.

We first established an NK cell expansion method using irradiated autologous CD56-depleted PBMCs (IAPs) as feeder cells to achieve high expansion rates compared with culture with rhIL-2 only. We compared the proliferation kinetics over 7 days of NK cells cultured with IAPs or with rhIL-2 only. NK cells cultured with IAPs showed significantly higher cell counts visible already at day 5 (mean fold expansion: 3.2, range: 1.4–5.7) compared with cultivation with rhIL-2 only (mean fold expansion: 1.3, range: 0.4–1.9). This differential NK cell expansion rate between the two cultivation methods became much more pronounced after prolonged cultivation, as observed in a set of NK cell expansions over 12 days. Here, cells cultured with IAPs reached particularly high cell numbers (mean fold expansion: 212, range: 80–419) compared with rhIL-2 cultivation only (mean fold expansion: 22.5, range: 6.2–57.3) (Figure 1A). The killing capacity of the differently expanded NK cells was tested against two different leukemia tumor cell lines, the K562 cell line to analyze natural cytotoxicity, and the CD20⁺ Daudi cells to also determine ADCC after opsonization of these cells with anti-CD20 mAb rituximab. The killing capacity of the differently expanded cells was preserved in all tested settings. NK cells displayed comparable natural killing rates toward K562 and Daudi targets (Figures 1B,C; Figure S1 in Supplementary Material), as well as comparable ADCC responses upon rituximab opsonization of Daudi cells. In vitro ADCC responses can be monitored best at low E:T ratios, where read-out of natural killing activity is lower. Independent of the expansion method, NK cells showed a clearer ADCC effect of rituximab-coated target cells at low E:T ratios (Figure 1C; Figure S1 in Supplementary Material). Taken together, co-culture with IAPs significantly enhanced ex vivo NK cell proliferation preserving their cytotoxic capacity.

Next, we sought to determine whether the enhancement of NK cell ex vivo proliferation with our established protocol is a consequence of direct cell-to-cell interactions between NK and feeder cells, or of exposure to soluble factors released by feeder cells. To address this question, we cultured NK cells for 5 days at the bottom of Transwell^® plates, alone or with IAPs either in close contact or separated through a permeable membrane, allowing only traffic of soluble factors. NK cell fold expansions were significantly decreased when IAPs were separated through the membrane, reaching similar fold expansions as NK cells cultured alone. On the other hand, the addition of NK cells to IAPs separated by the membrane, rescued the expansion of the NK cells at the bottom of the Transwells (Figure 2A). These data indicate that cell-to-cell interactions between IAPs and NK cells are necessary to produce soluble factors beneficial for NK cell proliferation.

IAPs are composed of different immune cells, mainly T cells, B cells, monocytes, and dendritic cells. In this respect, we speculated whether any of these cells could have a predominant role in stimulating NK cell proliferation. NK cell fold expansions were compared after co-culture with a standard IAP feeder cell fraction and three other fractions that were depleted from CD19⁺ cells (B cells and a small subset of dendritic cells), or additionally further depleted from CD3⁺ (T cells) or CD14⁺ cells (monocytes), respectively (Figure 2B). Depletion of CD19⁺ cells from the standard IAP fraction, had no effect on NK cell proliferation, whereas further depletion of T cells or monocytes significantly reduced NK cell expansion, with a more dramatic effect upon the absence of monocytes (Figure 2C). Differences in NK cell fold expansions correlated with differences in number of dividing cells as observed with dilution of CTV dye labeling (Figure 2D).

We hypothesized that the differences in stimulating NK cell proliferation could be mirrored in the cytokine composition released by the different feeder cell fractions. GM-CSF, IFN-γ, IL-6, TNF-α, IL-5, and IL-9 could be detected at high levels (>100 pg/mL) and were produced at a similar extent in co-cultures of NK cells with IAPs or with CD19-depleted IAPs. A remarkable reduction in cytokine production was observed in co-cultures with further depletion of T cells or monocytes, respectively (Figure 2E). This, indeed, supported the differences observed in NK cell proliferation kinetics due to the different cellular composition of the feeder cell fractions. The presence of NK cells in co-culture with the different feeder cell fractions influenced the cytokine production, as observed when feeders were maintained in culture without NK cells (Figure S2 in Supplementary Material). Low detectable levels (<100 pg/mL) of IL-4, IL-10, IL-17A, IFN-α, and IL-12 were also found in all co-culture conditions, without prominent changes upon differential cell subsets depletions from IAPs (data not shown). IL-2 was detected at levels exceeding the maximum standard value (>10.000 pg/mL) in any co-culture condition, as a result of its supplementation at the beginning of the culture (data not shown); which was sufficient to support NK cell growth independently of the presence or absence of T cells and monocytes. Taken together, our data indicate that the cross-talk between NK cells, T cells, and monocytes, is crucial to enhance ex vivo NK cell proliferation when using IAPs, not only due to occurrence of receptor–ligand interactions among these cells but also possibly due to the production of soluble factors as consequence of this cross-talk.

The multitude of signals occurring during the co-culture of NK cells with IAPs helped to boost the ex vivo proliferation of NK cells. However, at early time points during culture (day 5) not all NK cells are actively proliferating, but some remain quiescent. We aimed at identifying molecules or pathways responsible for the boost of ex vivo NK cell proliferation under the co-culture conditions. For this purpose, we performed a whole genome microarray analysis, with five different donors, to study the transcriptome of resting (R0) NK cells, isolated on day 0, and compare it to proliferating (P), and non-proliferating (NP) NK cells co-cultured for 5 days with IAPs (Figure S3 in Supplementary Material). Principal components analysis (PCA) of the unfiltered transcriptomes separated the samples in three groups corresponding to resting, proliferating, and non-proliferating NK cells, respectively (Figure 3A). The expression differences between R0 and 5-day cultured samples appear to contribute most to the variation (principal component 1, 63.1%), while differences between P and NP can be ascribed to the second principal component (principal component 2, 11.5%). A total of 10,299 transcripts were differentially expressed among the three groups, only 2,902 of these transcripts were unique candidates (Figure 3B). Hierarchical clustering of the differentially expressed transcripts confirmed the PCA results of the unfiltered data by the formation of two main clusters according to the samples of resting and 5-day expanded NK cells, the latter separated in sub-clusters corresponding to P and NP (Figure 3C). A functional annotation analysis of all differentially expressed transcripts revealed to which biological processes and pathways the transcripts were related to. In addition to the anticipated associations with IL-2 pathway, the great majority of the identified transcripts were associated with nucleotide metabolism and categories principally related to regulation of cell proliferation and activation (Figure S4 in Supplementary Material). This demonstrates that the sorted fractions indeed represented proliferating and non-proliferating cells.

In this regard, among the top 20 most highly expressed transcripts in proliferating NK cells compared with resting cells, were several positive regulators of the cell cycle and mitosis, such as cyclin A2 (CCNA2) and centrosomal protein of 55 kDa (CEP55), or the zinc finger proteins ZBED2 and ZBTB32 (Figure 4A). Zinc finger proteins are structurally a heterogeneous group of molecules which regulate gene expression by binding to DNA and RNA (28). Particularly, Tramtrack bric á brac zinc finger proteins (BTB-ZF) have important roles in controlling development and functional activity of lymphocytes (29, 30). Our data show that in proliferating NK cells, ZBED2 was the zinc finger protein with highest transcript levels, and the transcription factor ZBTB32 was the family member of BTB-ZF proteins with highest transcript levels (Figure S5A in Supplementary Material). These results point to a potential role of zinc finger proteins in modulating human ex vivo NK cell proliferation.

Natural killer cell receptors and co-receptors play key roles in controlling NK cell activation. We additionally screened for the most well-known activating and inhibitory receptors and co-receptors differentially expressed among the groups. In general, transcript levels of activating receptors were increased in proliferating cells, meanwhile transcripts of many inhibitory receptors, were decreased (Figures 4B,C). Interestingly, among activating receptors and co-receptors, we found that proliferating NK cells compared with resting or non-proliferating contained increased transcript levels of receptor activator of the nuclear factor kappa-B ligand, RANKL, and CD276 molecule known as B7-H3. On the contrary, they had decreased transcript levels of the activating receptors FCGR3A (CD16), KLRF1 (NKp80), or KLRB1 (CD161) compared with resting and non-proliferating cells (Figure 4B). Furthermore, we noticed that transcriptional changes in proliferating NK cells extended also to differences in expression of HLA class I and II molecules. Remarkably, proliferating NK cells expressed high transcript levels of HLA class II, that correlates with their activated status, whereas transcripts corresponding to HLA class I molecules were decreased in comparison with resting and non-proliferating NK cells, representing an unexpected finding (Figure S5B in Supplementary Material). Altogether these results suggest the existence of differential activation responses of NK cells toward the stimuli provided by IAPs, which may translate into a rapid proliferation of certain subsets of NK cells compared with others with delayed or no proliferative response to the same stimuli.

The changes observed in transcript levels of activating and inhibitory receptors and co-receptors are likely to extend at the protein level, and may reveal phenotypic characteristics of proliferating and non-proliferating NK cells. In addition, the detected phenotypic differences may depend on the stimuli used to promote ex vivo NK cell expansion. Therefore, we analyzed by flow cytometry, the expression of several selected molecules (Table S1 in Supplementary Material) in resting, proliferating, and non-proliferating NK cells expanded for 5 days with IAPs or rhIL-2 only. Most of the observed changes in protein content confirmed the results obtained by transcriptional profiling, with significant differences especially between resting and proliferating cells (Figure 5; Figure S6 in Supplementary Material). Among them, we identified that proliferating NK cells contained significantly decreased frequencies of cells expressing the inhibitory receptors Siglec-7 and -9, independently of the expansion protocol, compared with both non-proliferating and resting cells. Also, independently of the expansion method was a significant reduced expression of the activating receptors KLRB1, NKp80, CD16, and the inhibitory receptor KLRG1 within proliferating cells compared with resting cells (Figure S6 in Supplementary Material). All other analyzed activating receptors and co-receptors were expressed at higher levels in proliferating NK cells, except the NKG2D co-receptor 2B4. Despite almost all resting NK cells expressed 2B4 at the beginning of the culture, this could not be detected on day 5 of expansion with none of the expansion methods used (data not shown). This suggests an oscillating expression of 2B4 on NK cells during expansion, since expression of 2B4 has been reported after longer culture periods (23). Moreover, we could confirm RANKL and B7-H3 expression, absent in resting cells, appeared after 5 days of expansion. Their expression was mainly restricted to proliferating NK cells and reached higher levels when NK cells were expanded with IAPs instead of with rhIL-2 only (Figures 5A,B). Similarly, expression of HLA class II molecules and particularly HLA-DR, were increased after 5 days of expansion with IAPs compared with rhIL-2 only. Expression of HLA-DR persisted after longer term culture (data not shown). Interestingly, the reduction observed earlier in the expression of HLA class I molecules after 5 days of culture was verified here too, independently of the expansion protocol used.

Given that co-culture with IAPs for 5 days produced an enrichment of HLA-DR expressing cells and induced expression of RANKL and B7-H3, we investigated the correlation of the co-expression of these three molecules with the proliferative status of NK cells by flow cytometry. HLA-DR expressing NK cells contained a subpopulation of double positive RANKL and B7-H3 NK cells that corresponded to the most highly proliferating NK cells (Figure 6; Figure S7 in Supplementary Material). This was substantiated by NK cells co-cultured with IAPs in contrast to cultures with rhIL-2 only. The HLA-DR⁺RANKL⁺B7-H3⁺ NK cell population was minimally represented in cultures with rhIL-2 only, but enlarged in co-cultures with IAPs. In summary, we identified a subset of NK cells that preferentially expand in response to the stimuli provided by IAPs and is characterized by expression of HLA-DR, RANKL, and B7-H3.