Ligation of the Fc receptor, CD16, on NK cells triggers cytolytic activity and the secretion of pro-inflammatory cytokines (5). Here, we used an afucosylated version of anti-CD20 mAb (anti-CD20-AF) that was afucosylated using POTTELIGENT technology (BioWa).

To investigate how afucosylation affected NK cell functional responses, NK cells were incubated with anti-CD20 or anti-CD20-AF mAb opsonized target cells; either the transformed B cell line, 721.221 (hereafter referred to as 221) or the Burkitt's lymphoma, Daudi, cell line (37, 38). It may be important to note that both cell lines tested do not express classical class I MHC proteins and are particularly susceptible to NK cell-mediated lysis. To assess opsonization, 221 cells were incubated with increasing concentrations of each mAb. Binding of the mAbs was detected using a fluorescent secondary antibody with maximal opsonization reached at 10 μg/mL (Supplementary Figure 1A). Opsonization of 221 target cells with anti-CD20 and anti-CD20-AF partially reduced surface CD20 availability to fluorescently labeled anti-CD20 added after opsonization, indicating that loading was equivalent between the two antibodies but that not all surface CD20 was bound by the mAbs when opsonizing targets (Supplementary Figure 1B).

Human primary NK cells, which had been rested for 6 days after stimulation with low dose IL-2, were co-incubated with opsonized 221 cells at an effector to target ratio (E:T) = 1:3, reflecting immune cells being outnumbered by cancerous cells (39). After 2 h, target cell viability was assessed by staining with Annexin V (apoptotic marker) and propidium iodide (PI), indicative of a permeable membrane (40, 41). The presence of NK cells induced specific lysis of non-opsonized 221 (19.2 ± 5.8%) and Daudi cells (10.5 ± 2.2%; Figure 1A). Opsonization with anti-CD20 mAb increased specific target cell lysis (42.4 ± 5.7% for 221 and 34.4 ± 9.3% for Daudi), whilst anti-CD20-AF opsonized targets were killed more effectively still (54.6 ± 10.5% for 221 and 40.8 ± 9.6% for Daudi; Figure 1A). To assess cytokine secretion, NK cells were co-incubated with 221 target cells opsonized with increasing concentrations of anti-CD20 or anti-CD20-AF mAbs. Non-opsonized target cells caused some secretion of IFNγ (52.7 ± 25.2 pg/mL) whilst 221 cells opsonized with anti-CD20 mAb induced significantly higher secretion of IFNγ (Figure 1B). Opsonization with anti-CD20-AF dramatically increased IFNγ secretion across all concentrations. In contrast, opsonization with either mAb did not trigger any significant level of TNFα secretion (Figure 1C). Thus, opsonization with anti-CD20-AF augments NK cell-mediated specific lysis and secretion of IFNγ relative to anti-CD20.

NK cell cytolytic responses can be assessed by quantification of surface CD107a, a marker of lytic granules, that transiently appears on the surface of degranulating NK cells (42). To assess how afucosylation affects NK cell degranulation, NK cells were co-incubated with target cells and the percentage of CD107a⁺ NK cells was measured. In the absence of target cells, a small fraction of NK cells exhibited some level of staining with an anti-CD107a mAb (4.4 ± 0.8%). However, in the presence of 221 and Daudi target cells 25.5 ± 9.3% and 15.4 ± 6.5% of NK cells degranulated, respectively (Figures 1D,E). Opsonization of 221 cells with anti-CD20 mAb significantly increased the amount of degranulating NK cells (32.4 ± 8.1%), while anti-CD20-AF opsonization further increased degranulating NK cells (50.0 ± 6.8%). Similarly, anti-CD20 opsonized Daudi cells triggered 23.5 ± 7.4% of NK cells to degranulate, while anti-CD20-AF triggered 36.9 ± 7.2% (Figure 1E). Together these data are consistent with the higher affinity anti-CD20-AF mAb triggering more NK cell degranulation.

In parallel, CD16 surface expression was quantified relative to unstimulated NK cells. Co-culture with non-opsonized 221 and Daudi target cells induced significant decreases in CD16 surface expression (Figures 1F,G). This is consistent with previous observations that NK cell activation leads to a loss of CD16 from the cell surface, even in the absence of specific ligation of CD16 (17). Opsonization of 221 target cells with anti-CD20 mAb significantly reduced surface CD16 (54.2 ± 12.7% reduction), whereas targets opsonized with anti-CD20-AF reduced surface CD16 expression to an even greater extent (78.8 ± 5.5% reduction; Figure 1F). Similarly, opsonization of Daudi target cells with anti-CD20 significantly reduced surface expression of CD16 by 59.6 ± 8.8% and anti-CD20-AF reduced this by 73.9 ± 6.8% (Figure 1G). Across both cell lines, degranulation and CD16 surface expression were inversely correlated (221, R² = 0.55 and Daudi, R² = 0.57; Figures 1H,I) implying that both happen concurrently with NK cell activation. The expression of other NK cell receptors, and integrin LFA-1, were also compared on NK cells co-incubated with Daudi or Daudi cells opsonized with anti-CD20 or anti-CD20-AF. We found that surface expression of the activating receptors NKG2D, NKp30, NKp44, and NKp46, or the integrin LFA-1 were not altered (Supplementary Figure 2). However, NKG2D and NKp30 expression was reduced when NK cells were co-incubated with target cells expressing their respective ligands, Daudi-MICA (43) or K562 (44). Overall, anti-CD20-AF mAb triggers potent NK cell activation beyond that seen with anti-CD20 mAb, resulting in strongly reduced expression of surface CD16 with no effect on other activating receptors.

To investigate how mAb afucosylation impacts the killing kinetics of NK cells, we performed live-cell time-lapse microscopy in fibronectin-coated microwells (450 × 450 μm) (45). 221 or Daudi cells were used as target cells and stained with one dye, Calcein Green, before co-incubation with NK cells stained with another dye, Calcein Red-Orange, at an E:T ratio of 1:3. Target cells were also stained with another dye, To-Pro-3, and cell death was indicated by its release (Figures 2A,B). This system enabled detailed quantification of NK cell-target cell interactions, with total contact time, time to kill and time to detachment after kill all assessed (Figure 2C). Example interactions are shown in Figure 2B and Supplementary Videos 1–5. In some cases, target cells were killed very quickly, within 12 min (Supplementary Video 1), while other targets required interactions of almost 3 h to induce lysis (Supplementary Video 2). Detachment is a crucial step in NK cell serial killing but occasionally NK cells can kill a second target without prior detachment from the previous target (Supplementary Video 3). Interestingly, some NK cells killed two targets within a short time span (Supplementary Video 4), and some killed several targets sequentially (Figure 2B and Supplementary Video 5).

Detailed analysis of NK cell-target interactions was performed, using either opsonized 221 or Daudi cells. First contacts between NK cells and anti-CD20 mAb opsonized 221 cells that led to kills resulted in less frequent detachment (53.3 ± 3.9%) than anti-CD20-AF-opsonized 221 cells (71.9 ± 2.6%; Figure 2D), whilst NK cells which contacted and killed a second opsonized target cell were equally likely to detach with either mAb. We then focused on contacts between NK cells and opsonized 221 cells that resulted in kills and examined the duration of lysis and total contact time. Opsonization with anti-CD20-AF significantly reduced the time required to kill target cells, 24 ± 10.2 min, in comparison to anti-CD20, 49.8 ± 3.3 min (Figure 2E). The time taken to kill a second cell was also reduced for 221 targets opsonized with anti-CD20-AF (27.3 ± 7.9 min) compared to anti-CD20 mAb (54 ± 26.1 min). Importantly, detachment time following the first kill was significantly longer with anti-CD20-opsonized 221 cells (28.1 ± 12.8 min compared to anti-CD20-AF, 11.6 ± 2.5 min; Figure 2F). Similarly, detachment after a second kill was faster in the presence of anti-CD20-AF (10.8 ± 3.3 min compared to anti-CD20 mAb, 22.5 ± 2.7 min; Figure 2F). Taken together, both first and subsequent contacts between NK cells and anti-CD20-AF opsonized 221 cells that resulted in a kill were significantly shorter (1st, 42.7 ± 6.6 min and 2nd, 40.8 ± 13.4 min) than with anti-CD20-opsonized 221 cells (1st, 60 ± 8.8 min and 2nd, 81.7 ± 26.4 min; Figure 2G). Altogether, these data indicate that anti-CD20-AF facilitated faster killing and more efficient detachment than anti-CD20 mAb.

To test whether or not these results were generalizable to other target cells, we used Daudi as an alternative CD20-expressing cell. For Daudi, there was a significant difference in the number of cells which could detach after first (56.8 ± 3.1% for anti-CD20-AF compared with 44 ± 3.9% for anti-CD20 mAb) or second contacts (59.8 ± 3.3% for anti-CD20-AF compared with 45.2 ± 8.4% for anti-CD20; Figure 2H). The time taken for NK cells to kill the first target cell was significantly shorter for cells opsonized with anti-CD20-AF (21.7 ± 13.1 min compared to 76.5 ± 31.3 min for anti-CD20 mAb; Figure 2I). While the detachment time was unaffected for first or second kills (Figure 2J), the total duration of the second contact of anti-CD20-AF-opsonized Daudi was significantly shorter (43.1 ± 14.4 min compared to 92.6 ± 27 min for cells opsonized with anti-CD20 mAb; Figure 2K). Collectively, these data indicate that the higher affinity anti-CD20-AF mAb altered the dynamics of intercellular interactions and induced more efficient killing of multiple opsonized target cells.

The total number of target cells killed by an individual NK cell was also quantified. A higher proportion of NK cells did not kill any 221 cells opsonized with anti-CD20 mAb (33.3 ± 4.2%) compared to anti-CD20-AF opsonized targets (24 ± 1.5%; Figure 2L). NK cells incubated with 221 targets opsonized with anti-CD20 mAb predominantly killed a single target (75.5 ± 3.7%). However, more NK cells made sequential kills with 221 target cells opsonized with anti-CD20-AF than with anti-CD20 mAb (32.9 ± 6.7% compared to 20.3 ± 8.1%; Figure 2M). Similarly, more NK cells did not kill any Daudi target cells opsonized with anti-CD20 mAb (28.7 ± 6.3%) compared to anti-CD20-AF (19.4 ± 4.5%; Figure 2N). When NK cell-mediated killing was triggered, anti-CD20 mAb induced mostly single kills (76.7 ± 8.1%) whilst anti-CD20-AF triggered two sequential kills significantly more frequently (35.2 ± 1.5% for anti-CD20-AF compared to 20.7 ± 7% for anti-CD20 mAb) or more than three kills (5.3 ± 4.3%; Figure 2O). Collectively, these data indicate that afucosylated anti-CD20 mAb triggers faster kills, shorter contacts and more efficient detachment which boosts serial killing.

Intercellular dynamics are known to vary in 3D environments (46). Thus, to analyze NK cell killing kinetics in a more physiological environment, we next sought to investigate whether afucosylation affects NK cell serial killing in a 3D environment. NK cells were labeled with Calcein Red-Orange and co-incubated with 221 target cells stained with Calcein Green opsonized with anti-CD20 mAb or anti-CD20-AF in a 3D Matrigel for 6 h, while dead cells were indicated by To-Pro-3 (Figure 3A). The density of target cells here was higher, compared to microwells tested above, therefore more NK cells were serial killers. In some cases, NK cells killed up to 5 targets sequentially by properly detaching (Figure 3B and Supplementary Video 6), whereas in other cases NK cells killed multiple available neighboring targets without prior detachment (Supplementary Videos 7, 8). Occasionally, NK cells could kill multiple targets within a particularly short time span (Supplementary Video 9).

Interestingly, detachment from first contacts that led to kills was not clearly affected by the mAb used for opsonization but a significantly greater number of NK cells detached from target cells opsonized with anti-CD20-AF and went on to kill a second time (62.4 ± 9.5% compared to 39.8 ± 7.5% for anti-CD20 mAb; Figure 3C). The speed of first kills was slightly, but not significantly faster in the presence of anti-CD20-AF opsonized 221 target cells (25.1 ± 2.8 min compared to 39.3 ± 15.8 min for anti-CD20; Figure 3D). However, there was no difference in the time taken for a second kill. Strikingly, anti-CD20-AF caused significantly faster detachment after second kills (15.3 ± 4.3 min for anti-CD20-AF compared to 28.8 ± 6.1 min for anti-CD20 mAb; Figure 3E). Although the total contact time for first kills was not significantly affected by the afucosylated mAb, second kills of anti-CD20-AF-opsonized target cells were significantly shorter, 39.7 ± 8.7 min, compared to anti-CD20 mAb, 64.5 ± 9.4 min (Figure 3F). Thus, mAb afucosylation boosted NK cell killing kinetics in a 3D microenvironment.

Notably, in a 3D environment, a greater fraction of NK cells engaged in serial interactions with target cells, irrespective of the mAb they were opsonized with. In microwells, 75.5 ± 3.7% and 60.9 ± 7.5% of NK cells were restricted to a single kill for target cells opsonized with anti-CD20 mAb or anti-CD20-AF, respectively. However, this reduced to 65.8 ± 9.1% and 43.1 ± 9.6% in a 3D environment (Figure 3G). In addition, there was a significant increase in the number of NK cells killing three or more target cells opsonized with anti-CD20-AF (18.9 ± 5.9% compared to 7.4 ± 3.6% with anti-CD20; Figure 3G). Collectively, these data establish that the higher affinity anti-CD20-AF mAb triggered more rapid and efficient killing of opsonized targets, and augmented serial killing of multiple targets in microwells and a 3D environment.

Shedding of CD16 on activated NK cells via the metalloproteinase ADAM17 has been described to aid serial engagement of multiple opsonized targets (20). Using TAPI-0, we investigated how inhibition of CD16 shedding affected NK cell function. NK cells were incubated alone or with 221 targets or 221 targets opsonized with anti-CD20 or anti-CD20-AF for 5 h in the presence of TAPI-0 inhibitor after which we measured surface expression of CD16 and the marker of NK cell degranulation, CD107a. While CD16 expression was not affected on unstimulated NK cells with varying doses of TAPI-0 (Figure 4A), CD16 shedding following co-incubation with 221 cells was inhibited in all conditions in the presence of TAPI-0 (5 μM). TAPI-0 completely inhibited CD16 shedding and even resulted in increased CD16 expression on NK cells co-incubated with non-opsonized or anti-CD20 opsonized 221 cells. However, CD16 levels were still reduced by 20.2 ± 12% in the presence of anti-CD20-AF opsonized 221 cells despite inhibition with TAPI-0 (Figure 4B and Supplementary Figure 3). TAPI-0 did not affect the fraction of NK cells expressing CD107a, indicating that degranulation was unaffected (Figure 4C and Supplementary Figure 3). Accordingly, TAPI-0 caused a significant increase in the number of CD107a⁺CD16⁺ NK cells, indicating that degranulating NK cells retained surface CD16 (Figures 4D,E). In addition, TAPI-0 increased the secretion of IFNγ across all conditions, although due to donor variability this difference was not statistically significant except for anti-CD20-AF opsonized target cells, where the presence of TAPI-0 inhibitor increased IFNγ secretion from 161 ± 199 to 524 ± 378 pg/mL (Figure 4F and Supplementary Figure 3). In contrast, TAPI-0 did not have any significant effect on the secretion of TNFα (Figure 4G). Thus, inhibition of CD16 shedding with TAPI-0 sustained expression of CD16 and boosted IFNγ secretion especially in the presence of 221 target cells opsonized with anti-CD20-AF.

To investigate whether inhibition of CD16 shedding impacts NK cell-mediated lysis, TAPI-0 inhibitor was added to NK cells co-incubated with target cells (E:T = 1:3) for 2 h. In the presence of non-opsonized Daudi and 221 cells, specific lysis was slightly increased by TAPI-0 for all donors (221; 23.0 ± 10.4% vs. 28.0 ± 8.6% and Daudi; 21.2 ± 15.4% vs. 23.7 ± 15.5%). TAPI-0 had no effect on specific lysis of anti-CD20-opsonized 221 target cells, but caused a slight but significant reduction in lysis of Daudi cells (221; 41.1 ± 16.3% vs. 38.9 ± 14.9% and Daudi; 39.8 ± 15.1% vs. 40.0 ± 15.1%). Interestingly, TAPI-0 inhibitor caused a small and significant reduction in the killing of Daudi and 221 cells opsonized with anti-CD20-AF (221; 46.6 ± 12.0% vs. 41.6 ± 13.4% and Daudi; 45.7 ± 15.6% vs. 42.2 ± 15.6%; Figures 4H–J). As before, there was a significant increase in specific lysis of both target cell lines when opsonized with anti-CD20-AF, compared with anti-CD20. However, this difference was eliminated in the presence of TAPI-0 suggesting that CD16 cleavage plays a role in the increased cytotoxicity observed with the afucosylated mAb. These changes in cytotoxicity are small, but measuring the effects of serial killing in bulk assays is challenging, so we next assessed the impact of TAPI-0 on the killing of anti-CD20-AF opsonized targets using time-lapse microscopy.

Shedding of CD16 has been shown to be one factor which can control the disassembly of immune synapses (20), and we next investigated the effect of TAPI-0 on the dynamics of interactions between NK cells and opsonized 221 target cells. As previously, NK cells and 221 target cells were stained with differently colored Calcein dyes, co-incubated in fibronectin-coated microwells, and imaged by time-lapse microscopy (Figures 5A,B). Cells were either left alone or treated during acquisition with 5 μM TAPI-0 inhibitor to block CD16 shedding. TAPI-0 dramatically reduced the percentage of NK cells which detached following both the first (63.4 ± 10% without TAPI-0 and 38.8 ± 12.8% with TAPI-0) and second contacts (60.6 ± 17.2% without TAPI-0 and 26.9 ± 15% with TAPI-0; Figure 5C) that resulted in kills of 221 cells opsonized with anti-CD20-AF. It is possible that sustained interaction with target cells in the presence of TAPI-0 is directly responsible for increasing IFNγ secretion, although this is hard to test directly. An example of TAPI-0 preventing an NK cell from detaching is shown in Supplementary Video 10. In the absence of TAPI-0, NK cells normally detach as shown in Supplementary Video 11. Of the contacts which resulted in kills and detachment, the presence of TAPI-0 did significantly impact the time for the first kill but not significantly for the second kill (Figure 5D), the time taken to detach after killing (Figure 5E), nor the total time of contact (Figure 5F). It's important to emphasize that the data shown in Figures 5E,F only represents interactions where cells detached. As shown by Figure 5C, most cells remained attached in the presence of TAPI-0 and therefore the overall times for cells to detach after killing and the total contact times are understated by the values shown. Accordingly, the number of NK cells able to kill three or more anti-CD20-AF opsonized target cells (18.9 ± 9.1%) was dramatically reduced in the presence of TAPI-0 (5 ± 3.5%; Figure 5G and Supplementary Video 12). Thus, inhibition of CD16 shedding by TAPI-0 causes less efficient NK cell detachment inhibits serial killing of target cells opsonized with anti-CD20-AF.

Primary human NK cells were isolated from the peripheral blood of healthy donors from the National Blood Service under ethics license REC 05/Q0401/108 (University of Manchester, Manchester, UK). In brief, peripheral blood monocyte cells (PBMC) were purified by density gradient centrifugation using Ficoll-Paque (GE Healthcare, Life Sciences). Primary human NK cells were subsequently isolated from PBMCs by negative magnetic selection, using the human NK cell isolation kit (Miltenyi Biotec). Isolated cells were cultured at 37°C and 5% CO₂ at 10⁶ cells/mL in clone media [Dulbecco's Modified Eagle Medium (DMEM) medium containing 30% Ham's F12 nutrient mixture, 10% human serum, 1% Non-essential amino acids, 1 mM sodium pyruvate (all Sigma-Aldrich), 2 mM L-glutamine, 50 U/mL penicillin/streptomycin, 50 μM 2-mercaptoethanol (all Gibco)]. Primary NK cells were also stimulated with 200 U/mL IL-2 (Roche) and rested for 6 days prior to experiments. NK cells rested in low-dose IL-2 had comparable CD16 expression to freshly isolated NK cells (Supplementary Figure 4).

Daudi (Burkitt's lymphoma) and 721.221 (B lymphoblastoid) cell lines were used as NK cell susceptible targets and purchased from ATCC. Cells were cultured at 37°C and 5% CO₂ in media comprising RPMI 1640 medium supplemented with 10% FBS, 1% Non-essential amino acids, 1 mM sodium pyruvate (1%), 2 mM L-glutamine (1%) and 50 U/mL penicillin/streptomycin (1%). All cell lines were routinely tested for mycoplasma infection using a PCR-based kit (PromoCell) that utilizes specific primers designed from DNA sequences coding for highly conserved ribosomal RNAs (16S-rRNA). All cells tested were negative.

Human IgG1 anti-CD20 monoclonal antibodies were generated based on publicly available sequences (Drugbank Accession: DB00073). Anti-CD20 was produced using HEK293 6E cells and anti-CD20-AF was produced with POTELLIGENT Technology licensed from BioWa, Inc. This technology involves the reduction of the amount of fucose in the carbohydrate structure of an antibody using a proprietary fucosyl transferase-knockout CHO cell line as a production cell. Evidence shows that this technology can enhance ADCC activity of an antibody in vitro (58, 59). Opsonization was based on incubation of target cells with mAbs for 45 min at 37°C. To assess target cell opsonization, cells were washed with 1% FBS/PBS, fixed with 4% PFA at room temperature for 10 min and blocked with 2% human serum/PBS (Sigma-Aldrich) for 10 min at 4°C. Following target cell opsonization mAbs were labeled with Alexa Fluor 488-conjugated to goat anti-human IgG F(ab')2 fragment specific (Jackson ImmunoResearch) or a second anti-CD20 mAb was added (anti-CD20 PE; clone 2H7, Biolegend) for 20 min at 4°C. Unless indicated otherwise, all isotype mAbs were used at 10 μg/mL based on experiments that defined the saturating concertation of mAb target opsonization.

Primary NK cells were co-incubated with opsonized target cells, and TAPI-0 (5 μM) where indicated, in polystyrene flat bottom 48-well plates (Nunc) for 24 h at 37°C. Where indicated, TAPI-0 in DMSO (stock solution concentration 10 mg/mL) was added such that the final concentration of the inhibitor was 5 μM. Cell supernatants were collected and centrifuged at 500 × g for 5 min at 4°C to remove cell debris. Secretion of IFNγ and TNFα were quantified in the supernatants using ELISA (DuoSet, R&D Systems) according to manufacturer's instructions.

In each co-culture, NK cells were labeled with cell proliferation dye (Cell Trace Violet, CTV, Thermo Fisher). Briefly, cells were washed three times in 5 mL of cold RPMI and stained with 5 μM Cell Trace Violet dye in warm RPMI for 15 min at 37°C (E:T = 1 (10⁵):3 (3 × 10⁵). NK cells were co-incubated with opsonized 221 and Daudi cells at E:T ratio 1:3 in a 48-well plate for 2 h at 37°C with TAPI-0 added where indicated. The cells were aspirated and washed once with cold Annexin V binding buffer (25 mM CaCl₂, 1.4 M NaCl, 0.1 M HEPES). Each condition was then stained with 2 μl of Annexin V-APC (BioLegend), incubated for 15 min at room temperature, and washed with Annexin V binding buffer. 20 μl of PI (1 μg/mL) was then added to each tube and samples were immediately analyzed with BD FACS LSRFortessa™ X-20 flow cytometer (BD Biosciences) and analyzed by FlowJo V10 software (BD).

To assess degranulation, NK cells were co-incubated with opsonized target cell lines in the presence of GolgiPlug (1/1000 dilution, BD Biosciences), monensin (1/1000 dilution, Biolegend) and anti-CD107a PE mAb (clone H4A3, BD biosciences) or an isotype-matched control (PE-IgG1κ; clone MOPC-21, BD) for 4 h at 37°C. After incubation, cells were washed and stained with Zombie NIR viability dye (Biolegend), anti-CD56-AF488 mAb (clone HCD56, Biolegend), anti-CD107a-PE (clone H4A3) and anti-CD16-AF647 (clone; B.731, Biolegend) (42). To assess expression of other proteins, NK cells were co-incubated with opsonized target cell lines for 4 h, then washed and stained with Zombie NIR viability dye (Biolegend) and LFA-1-PE (Clone M24), NKp46-PercP-Cy5.5 (Clone 9E2) and NKp44-APC (Clone P44-8) or NKp30-AF647 (Clone P30-15) and NKG2D-PE (Clone 1D11, all from Biolegend). Isotype-matched control mAbs were also used accordingly (mouse IgG1 isotype control; clone MOPC-21, Biolegend). Finally, cells were fixed in 2% PFA/PBS and staining was assessed using a BD FACS LSRFortessa™ X-20 flow cytometer and analyzed by FlowJo V10 software (BD Biosciences). The gating strategy used for all flow cytometry assays is shown in Supplementary Figure 5.

Sterile microwells were coated with 10 μg/mL fibronectin (Sigma). NK cells were stained with 1 μM Calcein Red-Orange (Invitrogen) and opsonized targets were stained with 1 μM Calcein Green (Invitrogen). To stain with calcein dyes, cells were first washed three times with 10 mL RPMI and then incubated with the dye solution (1 μM) in RPMI for 15 min at 37°C. Labeled NK cells were then added to opsonized target cells at E:T = 1:3 in clone media supplemented with 1 μl To-Pro-3 per well (Thermo Fisher) to discriminate dead cells. Where indicated, 5 μM TAPI-0 (Santa Cruz Biotechnology) was added just before acquisition, the absence of the inhibitor was used as a control for Figure 4, while a vehicle only, DMSO, control was used in Supplementary Figure 3. For imaging in microwells (450 × 450 × 300 μm), stained cells were mixed and immediately added. For imaging in 3D Matrigel (Corning Matrigel Matrix), cells were resuspended in a cold Matrigel solution, diluted 1:4 with clone media and added to 8-well chamber slides (1.5 Lab-Tek II; Nunc). Time-lapse imaging was performed at 37°C and 5% CO₂ for 6 or 8 h with an image acquired every 3 min. Imaging was performed using a Leica TCS SP8 inverted confocal microscope with a 20x/0.75NA objective (SP8, Leica Biosystems) using a white light laser. Fluorescent and bright-field images were merged and cell-cell interactions were analyzed manually using ImageJ. Microwells were produced in-house and used as previously described (10).

For each dataset, a D'Agostino and Pearson omnibus test or Shapiro-Wilk normality test was used to evaluate whether the obtained values were normally distributed. The statistical significance between two groups of data with normal distribution was examined using a two-tailed Student's t-test. The statistical significance between three or more conditions was assessed by one-way ANOVA. Differences were defined as non-significant where P ≥ 0.05 and were statistically significant where ^*P < 0.05; ^**P < 0.01; ^***P < 0.001; and ^****P < 0.0001. All statistical analyses were performed using Prism v8.0 (GraphPad).