Microchip-Based Single-Cell Imaging Reveals That CD56^dimCD57⁻KIR⁻NKG2A⁺ NK Cells Have More Dynamic Migration Associated with Increased Target Cell Conjugation and Probability of Killing Compared to CD56^dimCD57⁻KIR⁻NKG2A⁻ NK Cells

Natural killer cells are lymphocytes capable of exerting direct cytotoxicity against transformed, virus-infected, and stressed cells (1, 2). NK cells also produce proinflammatory cytokines and chemokines and aid the recruitment of other immune cells (2, 3). The cytotoxic response of NK cells is regulated by integration of signals received through inhibitory and activating receptors. Under physiological conditions, NK cell tolerance is maintained primarily by inhibitory receptors (IRs) that recognize MHC class I molecules. The human IRs include the family of inhibitory killer cell Ig-like receptors (KIRs) and the lectin-like CD94/NKG2A heterodimer. The quality of these inhibitory receptor–ligand interactions also tune NK cell responsiveness in a process called education or licensing (4–6), determining the strength of the missing self response in situations where MHC class I expression is lost (4–7). One previous study showed that NK cells educated via NKG2A are less responsive than NK cells educated through KIRs (8). In contrast, two other studies reported that KIR⁻NKG2A⁺ NK cells are fully functional both in terms of cytotoxicity and cytokine secretion, thus advocating the role of NKG2A as an educating receptor (9, 10). Uneducated KIR⁻NKG2A⁻ NK cells are present in human peripheral blood and have been found to be hyporesponsive to various stimuli while displaying a mature phenotype (9–11). The phenotypic stability of hyporesponsive NK cells may, however, be debated as cytokine stimulation can induce uneducated KIR⁻NKG2A⁻ NK cells to acquire effector functions and expression of KIR and/or NKG2A (9, 12). In addition, uneducated NK cells have been found to be equally functional as educated NK cells in secretion of IFN-γ during Listeria monocytogenes infection and are the primary mediators in suppression of mouse CMV infection (13, 14). A recent study revealed that the diminished cytotoxicity of uneducated NK cells is not due to compromised granule polarization but can be partially explained by reduced inside-out signaling leading to weaker LFA-1–dependent adhesion and failure to form stable conjugates with target cells (15). Still, little is known about how education of NK cells influences different aspects of NK cell cytotoxicity such as migration and efficacy of activation and killing when encountering target cells. Moreover, the distributions of killing and migration potential within both uneducated and educated NK cell populations are yet to be resolved. Using previously described microwell assays designed for efficient single-cell analysis (16, 17), in this study, we compare the migration behavior, interactions with target cells, and cytotoxicity of IR⁻ and NKG2A⁺ NK cells. We show that NKG2A⁺ NK cells displayed a more dynamic migration behavior compared with IR⁻ NK cells. NKG2A⁺ NK cells also formed more conjugates and were more effective killers of encountered target cells. Interestingly, although the fraction of killers was significantly higher among NKG2A⁺ NK cells, killing dynamics in terms of duration of lytic conjugation periods and activation-induced spreading of the NK cell membrane at the immune synapse (18) were similar in both subsets.

This study was performed in accordance with local ethics regulations. Human NK cells were freshly isolated from PBMCs of anonymous healthy donors by negative selection using an NK cell isolation kit (MACS; Miltenyi Biotec). The enriched NK cells were stained with Abs against CD57 (clone NK-1; BD Biosciences), CD56 and NKG2A (clone N901 and Z199; Beckman Coulter), CD3 (clone UCHT1; DakoCytomation), and panKIRs (clones NKVFS1 and 5.133; Miltenyi Biotec), and two subsets were sorted in purity mode using a FACSAria instrument (BD Biosciences): 1) CD56^dimCD3⁻CD57⁻KIR⁻NKG2A⁻ (referred to as IR⁻) and 2) CD56^dimCD3⁻CD57⁻KIR⁻NKG2A⁺ (referred to as NKG2A⁺). Sorted IR⁻ and NKG2A⁺ NK cell populations were maintained in RPMI 1640 medium supplemented with 10% human serum (Swedish blood bank), 2 mM l-glutamine, 50 U/ml penicillin, 50 μg/ml streptomycin, 1 mM sodium pyruvate, 1× nonessential amino acids (all Sigma-Aldrich), with or without 200 U/ml IL-2 (PeproTech) for 2 d past sorting.

The human leukemia K562 cell line and the adherent human embryonic kidney (HEK) 293T cell line were maintained in RPMI 1640 medium supplemented with 10% FBS, 2 mM l-glutamine, 50 U/ml penicillin, and 50 μg/ml streptomycin.

For flow cytometry experiments, magnetic-bead purified NK cells or sorted IR⁻ and NKG2A⁺ NK cells were cultured in complete medium with or without 250 U/ml IL-2 for 2 d. This was followed by Ab labeling of CD3 (UHCT1; Beckman Coulter), CD19 (HIB19; BD Biosciences), CD56 (NHK-1; Beckman Coulter), CD57 (NC1; Beckman Coulter), panKIR (NKVFS1; Miltenyi Biotec), NKG2A (z199; Beckman Coulter), DNAM-1 (Dx11; BD Biosciences), NKp30 (p30-15; BD Biosciences), NKp46 (9E2; BioLegend), and NKG2D (1D11; BD Biosciences). The expression levels of activating ligands on K562 and HEK293T target cells were assessed by Ab labeling of PVR (SKIL4; BioLegend), Nectin-2 (Tx31; BioLegend), ULBP-1 (170818; R&D Systems), ULBP-2 (165903; R&D Systems), ULBP-3 (166510; R&D Systems), ULBP-4 (M476; Amgen), MIC/A (236511; R&D Systems), and MIC/B (159227; R&D Systems) as well as labeling with NKp30 and NKp46 fusion proteins (R&D Systems). K562 and HEK293T cells were also stained with anti–HLA-E (3D12HLAE; eBioscience) to evaluate expression levels. A dead cell marker (Dead Cell Marker Aqua; Life Technologies) was included in all stainings. In degranulation experiments, NK cells were incubated at a 1:1 ratio with K562 or HEK293T target cells for 4 h with anti-CD107a (H4A3; BD Biosciences) present in the medium and subsequently labeled for the cell surface markers listed above. In some degranulation experiments, activating receptors were blocked with Abs against NKp46, NKp30, NKG2D, or DNAM-1. All results were analyzed using an LSR Fortessa (BD Biosciences) and FlowJo software (Tree Star).

Sorted NK cells were stained with 0.3 μM Calcein red-orange and K562 or adherent HEK293T target cells were stained with 1 μM Calcein green AM and 5 μM Far Red DDAO-SE (all from Life Technologies). The stained NK cells were coincubated with stained target cells in microwells at 37°C and 5% CO₂ and imaged using a Zeiss 510 Meta microscope. The IR⁻ and NKG2A⁺ subsets were imaged in parallel by using two separate basins, each covering multiple wells, of the microchip. In the migration assay, HEK293T target cells and NK cells were imaged using a 20× objective every 2 min for 12 h in wells of dimensions 450 × 450 × 300 μm³ that were precoated with 25 μg/ml fibronectin (Sigma-Aldrich). Killing of K562 target cells were studied by imaging NK and target cells confined in wells of dimensions 50 × 50 × 300 μm³. Images, each covering 81 individual wells, were acquired with a 10× objective every 3 min for 12 h. In both assays, the number of target cells per microwell as well as the NK to target cell ratios were similar in microwells containing IR⁻ and NKG2A⁺ NK cells.

NK cells were tracked with Volocity software (PerkinElmer) using a manual tracking tool. Further analysis of migration and target cell interactions was performed using Matlab (Mathworks). NK-target cell contacts lasting at least two time points were scored as conjugates. Target cell death was determined manually based on the fluorescence intensity of Calcein, which leaks out when the plasma membrane is ruptured, as well as on visible signs of death observed in the transmitted light channel (i.e., changes in morphology such as membrane blebbing or abnormal swelling or shrinking). Each NK-target cell interaction outcome was scored as kill/no kill and the duration of conjugation and attachment periods were determined. Spreading response of individual NK cells in conjugations was assessed by estimating the fraction of NK cell plasma membrane engaged in the intercellular contact. The NK cell perimeter and intercellular contact were outlined and the ratio interface/perimeter calculated.

The Mann–Whitney U test was used to evaluate statistical significance between the two subsets when collected data groups were non-Gaussian continuous data. The χ² test was used to evaluate statistical significance between the subsets when collected data groups were nominal data. All statistical analysis was performed using Matlab software, and p < 0.05 was considered as significant differences between the two subsets. In figures, *p < 0.05, **p < 0.01, and ***p < 0.001.

Mature CD56^dimCD57⁻KIR⁻ NK cells were sorted on basis of expression of the educating self-receptor NKG2A into “IR⁻” and “NKG2A⁺” populations (Fig. 1). To collect subsets at similar levels of developmental maturity, CD57⁺ NK cells were excluded in the sorting process because CD57 expression has been associated with highly differentiated NK cells (19). We observed very low cytotoxic activity in FACS-sorted resting NK cell populations, which appeared to be coupled to the sorting procedure because NK cells negatively selected by MACS microbeads showed cytotoxicity under similar conditions (Supplemental Fig. 1A). Therefore, sorted NK cells were left to recover for 2 d in medium supplemented with IL-2 that restored the weakened cytotoxic response imposed by the sorting process. During culture, a few percentages of the IR⁻ NK cells acquired NKG2A expression, and a small fraction of NK cells from both subsets upregulated KIR and CD57 expression (Supplemental Fig. 1B, 1C). However, the small fraction of cells that shifted phenotype in culture did not contribute significantly to the degranulation response of the IR⁻ NK cell population (Supplemental Fig. 2). Thus, the contribution from any cytotoxic IR⁻ NK cells that have acquired effector functions by in vitro education or differentiation should be negligible, corroborating the use of this experimental setup in downstream single-cell imaging assays. Characterizing the expression of activating receptors on IR⁻ and NKG2A⁺ NK cells, we found that NKp46, NKp30, NKG2D, and DNAM-1 were all expressed on >90% of the cells from both subsets with no significant influence of IL-2 treatment (data not shown). Gating for expression of either of the receptors showed that >99.9% of the NK cells from both subsets expressed at least one of the activating receptors studied. When looking at relative expression levels, we found, consistent with our recent results (20), that DNAM-1 expression was higher on NKG2A⁺ compared with IR⁻ NK cells, and the expression levels were similar on resting and IL-2–stimulated NK cells (Fig. 2A). The expression of NKp30 and NKG2D was heightened after 2 d of IL-2 stimulation, especially on the NKG2A⁺ NK cells (Fig. 2A). In contrast, the expression of NKp46 decreased with IL-2 treatment in both subsets (Fig. 2A). A corresponding characterization of the ligand expression on the K562 and HEK293T cells showed that both target cell lines expressed ligands for NKp30, NKG2D, and DNAM-1, whereas the HEK293T target cells also expressed low levels of NKp46 ligands (Fig. 2B). To elucidate the contribution of specific activating receptors on the cytolytic activity of IR⁻ and NKG2A⁺ NK cells, activating receptors were blocked with Abs during a CD107a degranulation assay with either K562 or HEK293T target cells. The results were similar for experiments with both target cell lines, showing only small decreases in the frequency of degranulating IR⁻ and NKG2A⁺ NK cells when either of the activating receptors NKp30, NKp46, NKG2D, or DNAM-1 was blocked (Fig. 2C). In contrast, blockade of all activating receptors together lead to decreased frequencies of degranulating IR⁻ and NKG2A⁺ NK cells (Fig. 2C). Thus, it was not possible to directly associate the increased degranulation of the NKG2A⁺ subset with elevated expression levels of activating receptors and no specific receptor could be identified as dominating the cytotoxicity for either of the two target cell lines, supporting the use of this model system to explore qualitative differences in the cytolytic behavior between the two subsets at the single-cell level.

To assess migration behavior NKG2A⁺ (Supplemental Video 1) and IR⁻ (Supplemental Video 2), NK cells were coincubated with HEK293T target cells in 450-μm wells and imaged for 12 h. HEK293T cells were chosen as target cells because they are susceptible to NK cell killing and adherent, thus facilitating the evaluation of NK cell migration behavior. Individual NK cells were tracked, and from the trajectories (Fig. 3A), the fractions of time spent in three different modes of migration: transient migration arrest periods (TMAPs), random movement, and directed migration were assessed for each NK cell as described previously (21, 22). There was no significant difference in the distributions of mean migration speeds between IR⁻ (median, 1.4 μm/min) and NKG2A⁺ (median, 1.6 μm/min) NK cells (Fig. 3B). However, a more comprehensive single-cell analysis showed that IR⁻ NK cells spent more time in TMAPs than NKG2A⁺ NK cells (Fig. 3C). NKG2A⁺ NK cells spent significantly more time migrating, both in a random and directed manner as compared with IR⁻ NK cells (Fig. 3D, 3E). Furthermore, NKG2A⁺ NK cells made transitions between different modes of migration on average 3.3 times, which was more often than IR⁻ NK cells (2.4 times; Fig. 3F). Thus, NKG2A⁺ NK cells displayed a more dynamic behavior involving more frequent alternations between different modes of migration.

Next, the contact and killing dynamics of IR⁻ and NKG2A⁺ NK cells interacting with HEK293T target cells were compared. NK-target cell contacts were divided into periods of “conjugation,” where the NK cell assumes a rounded morphology and forms a tight intercellular contact, and periods of “attachment,” which follows conjugation, where the NK cell remains in contact but assumes a more migratory morphology (Fig. 4A) (17). NKG2A⁺ NK cells interacted with more target cells, with an average of 2.6 contacts per NK cell compared with an average of 1.5 contacts made by IR⁻ NK cells (Fig. 4B). NKG2A⁺ NK cells also spent on average approximately twice as much time interacting with target cells than IR⁻ NK cells did (Fig. 4C, 4D). To complement the histograms in Figs. 2 and 3 that are based on pooled data, the results are also given as means ± SE indicating the variation between experiments (Supplemental Fig. 3).

The qualitative differences in conjugate formation between IR⁻ and NKG2A⁺ NK cells prompted us to examine the outcome of the conjugates in terms of killing efficiency. The frequencies of NK cells cytotoxic against HEK293T cells were 53% of NKG2A⁺ and 15% of IR⁻ NK cells (Fig. 5A). Serial killers (in this study defined as killing ≥ 3 target cells) were more common among NKG2A⁺ NK cells (17%), although a few (4%) serial killers were also observed among IR⁻ NK cells. We have previously shown that NK cells can be grouped into five different categories based on their cytotoxic response (17). Applying the same framework in this study, we identified IR⁻ and NKG2A⁺ NK cells displaying the five classes of cytotoxic response: 1) NK cells that never interacted with target cells, 2) NK cells that never killed encountered target cells, 3) NK cells that killed all encountered target cells, 4) exhausted NK cells that first killed but later failed to kill all subsequent target cells, and 5) NK cells that killed in a stochastic manner (Fig. 5B). This analysis showed that the nonkillers among the IR⁻ NK cells (representing 85% of the population) contained approximately equal fractions of cells that never interacted with targets (41% of the total population) or never killed the targets they encountered (44% of the total population). The distribution appeared different for the NKG2A⁺ NK cells where most nonkillers interacted with target cells (31% of the total population) and a smaller fraction (17% of the total population) of nonkillers never interacted with target cells. As previously observed, the populations of NK cells killing in a stochastic manner were relatively small (2% of IR⁻ and 9% for NKG2A⁺, respectively), indicating that individual NK cells have a consistency in their cytotoxic response through individual encounters with target cells (17).

Although suitable for the migration assay, the HEK293T cell line expresses low levels of HLA-E (Supplemental Fig. 4) that could possibly influence the outcome of NK-target cell conjugates. Therefore, we performed microwell killing assays where sorted IR⁻ and NKG2A⁺ NK cells were coincubated with K562 cells, a widely used HLA-E–negative target cell line, in arrays of 50-μm wells and imaged for 12 h. NK cells from both subsets made nonlytic and lytic interactions with target cells (Fig. 6A), but there was a substantial difference in the fraction of cytotoxic NK cells in the two populations; 30% of NKG2A⁺ versus only 5% of IR⁻ NK cells were killing K562 cells (Fig. 6B). Furthermore, we evaluated serial killing; examples of time-lapse sequences of serial killers from both the NKG2A⁺ and IR⁻ subsets are shown in Supplemental Videos 3 and 4. The frequencies of serial killers against K562 target cells were 10% of all NKG2A⁺ NK cells and 1% of all IR⁻ NK cells (Fig. 6B). Thus, the relative differences in cytotoxicity and serial killing between NKG2A⁺ NK cells and IR⁻ NK cells were even more pronounced with HLA-E–deficient K562 cells corroborating the data obtained with HEK293T target cells.

By tracking the fate of individual NK-target cell conjugates with both K562 target cells in the killing assay and HEK293T target cells in the migration assay, we noticed that killing was 2–3 times more likely to occur in conjugates with NKG2A⁺ NK cells (Fig. 7A, 7B). Analysis of individual conjugation periods with HEK293T target cells showed that in nonlytic interactions, NKG2A⁺ NK cells remained conjugated to target cells longer than IR⁻ NK cells but there was no difference in the duration of lytic conjugation periods (Fig. 7C). These results suggest that NK cells of both IR⁻ and NKG2A⁺ subsets behave alike in conjugates that result in killing of the target cell. Nonetheless, the majority of interactions made by IR⁻ NK cells did not result in killing and these contacts were shorter, suggesting weaker adhesion and weaker activation of the NK cells. It is known that engagement of activating NK cell receptors results in spreading of the NK cell plasma membrane at the site of contact (18). To estimate the level of activation, we analyzed NK cell spreading by assessing the fraction of the NK cell plasma membrane engaged in intercellular contact in NK-target cell conjugates. NKG2A⁺ NK cells displayed an increased spreading response, compared with IR⁻ NK cells, but only in nonlytic conjugates (Fig. 7D). In lytic interactions, the spreading response was comparable for NK cells from the two subsets indicating that those IR⁻ NK cells that kill target cells become activated to the same extent as NKG2A⁺ NK cells (Fig. 7D).

In this study, we have used microwell-based assays with single-cell resolution to elucidate functional differences between NKG2A⁺ NK cells and uneducated IR⁻ NK cells. Although the impact of NKG2A expression on NK cell function is commonly interpreted as being a consequence of education through interactions with HLA-E class I (8, 10, 23), it is possible that the observed differences between IR⁻ and NKG2A⁺ NK cells are partly related to their states of differentiation (9, 19). In the absence of a defined mechanism behind NK cell education, it is impossible to fully assign the heightened functionality of NKG2A⁺ NK cells to either education or differentiation. It seems likely that both processes act in synergy to build functional potential in the NKG2A⁺ NK cell subset, similar to what has been observed for KIR-mediated education in self KIR⁺CD57⁺ NK cells (19). Notably, the strength of the NKG2A–HLA-E interaction is determined by dimorphism in the HLA-E leader sequence peptide at position 2 (P2) (methionine versus threonine) and significantly influences the missing self response of NKG2A⁺ NK cells (24, 25). The link between inhibitory receptor input (through NKG2A) and function suggests that education plays an important role in shaping the responsiveness of the NKG2A⁺ NK cell subset. Regardless of the relative contribution of education and differentiation, it is essential to decipher qualitative aspects of the cytolytic response and probe the heterogeneity within discretely defined NK cell subsets.

In this study, we found that NKG2A⁺ NK cells spent more time in a migratory state and were more prone to alternate between different modes of migration compared with IR⁻ NK cells, showing that NKG2A expression is associated with a more dynamic migration behavior. This is consistent with our observation that the fraction of NK cells that never interacted with HEK293T target cells was considerably higher in the IR⁻ population. The observed difference in migration behavior between the subsets could partly explain our observation that NKG2A⁺ NK cells form more conjugates because increased motility is likely to lead to more frequent encounters with target cells. An additional contribution to the increased conjugation for the NKG2A⁺ subset could come from elevated expression of the open, active form of LFA-1, which has been associated with NK cell education through increased inside-out signaling by activating receptors to LFA-1 (15) and coordinated expression with DNAM-1 in the immune synapse (20). This is supported by our observation that nonlytic conjugates formed by IR⁻ NK cells were shorter than those formed by NKG2A⁺ NK cells.

As previously observed (9, 10), NKG2A⁺ NK cells were more cytotoxic compared to uneducated IR⁻ NK cells. This was not only a consequence of NKG2A⁺ NK cells forming more conjugates because our single-cell studies also revealed that this subpopulation of NK cells more frequently killed the conjugated target cell. We also observed increased activation-induced spreading across the target cells from NKG2A⁺ NK cells. Both these observations support that NKG2A⁺ NK cells, on average, are more easily activated. Both NKG2A⁺ and IR⁻ NK cells expressed receptors for activating ligands expressed on HEK293T and K562 target cells. In agreement with our recent report, the expression level of DNAM-1 correlated with the expression of educating receptors as it was higher on NKG2A⁺ NK cells compared with IR⁻ NK cells (20). Although the expression of NKp30 was increased by IL-2 treatment, especially on NKG2A⁺ NK cells, blocking of NKp30 alone had only a small effect on the degranulation response in both subsets. The largest effect was observed when NKG2D was blocked, suggesting its importance in NK-mediated cytotoxicity against HEK293T and K562 target cells. However, no individual activating receptor could be singled out as responsible for inducing killing of the target cells used in this study, and the difference in cytotoxicity between the two subsets could not be attributed to expression levels of activating receptors.

Importantly, conjugates that led to killing showed similar dynamics in terms of duration of conjugation periods and the NK cell spreading response, for both IR⁻ and NKG2A⁺ NK cells. Thus, these observations propose that the high killing capacity of NKG2A⁺ NK cells is linked to processes regulating events in the recognition phase of NK-target cell contact (26) where the threshold for lytic response is set, rather than events after cytotoxicity has been triggered.

As in previous studies, we observed populations of NK cells capable of killing several target cells in sequence (17, 27, 28). Although serial killers were considerably more common among NKG2A⁺ NK cells, we also observed a small group of very efficient killers in the IR⁻ subset. This is intriguing because it suggests that there may be additional mechanisms, independent of NKG2A or KIR, leading to efficient NK cell–mediated cytotoxicity. However, more work is required to fully elucidate this. In this study, we cannot rule out that the few serial killers observed in the IR⁻ subset were NK cells that had upregulated inhibitory receptors (Supplemental Fig. 1B, 1C) during the cytokine treatment that was included to restore the diminished cytotoxicity inflicted by the sorting process (Supplemental Fig. 1A). Cytokine stimulation has been shown to induce both enhanced effector functions and expression of NKG2A and/or KIRs in KIR⁻NKG2A⁻ NK cells and has been suggested to have a natural function in generating competent cells during an immune response (9, 12). Although some cells upregulated KIRs and/or NKG2A in the IR⁻ subset, we established that the majority of CD107a⁺ IR⁻ NK cells after coincubation with target cells were NK cells that had not acquired KIRs or NKG2A expression during the 2-d culture time (Supplemental Fig. 2C, 2D). Thus, we confirmed that the majority of cytotoxic NK cells in the sorted IR⁻ subpopulation were cells that had remained negative for KIR and NKG2A expression. However, NK cells that did acquire NKG2A expression as a result of IL-2 culture appeared to respond better in the CD107a degranulation assays (Supplemental Fig. 2E). Thus, we expect that there is a small contribution from NK cells with de novo cytokine-induced expression of educating receptors in our cytotoxicity data from microwell assays.

In summary, our results show that NKG2A expression is associated with increased cytotoxicity of NK cells by modulation of several important steps in the NK-target cell interaction process (i.e., migration dynamics, establishment of stable conjugates, spreading response, and probability of killing). This, together with the similar interaction dynamics displayed by cytotoxic NK cells from both NKG2A⁺ and IR⁻ subsets, suggests that the most important regulatory steps are early in the NK-target cell interaction process while the threshold for activation is set. Future studies will have to address whether education by KIRs follows similar pathways and to what extent tuning of molecular events of the cytotoxic process are involved.

We thank Thomas Frisk for manufacturing microchips, Iyadh Douagi for assistance with cell sorting and Klas Kärre for helpful discussions.

The authors have no financial conflicts of interest.