Target cell-type dependence of NK cell cytotoxicity
To investigate the mechanisms underlying the differential activity of NK cells on distinct targets, we chose a panel of human epithelial cell lines that exhibited variable sensitivity to primary NK cell killing, including one normal cell line, LO2 (immortalized normal hepatic cell line), and four cancer cell lines, i.e., HeLa (cervix cancer), SMMC-7721 (liver cancer), MCF7 (breast cancer), and U-2 OS (bone cancer). To visualize the dynamic response to NK cells, we engineered each epithelial cell line with a previously established fluorescent FRET reporter specific to granzyme B activity (GrzmB-FRET) [20] as well as a mitochondria reporter of apoptosis (IMS-RP) [21,22,23]. The GrzmB-FRET reporter consists of a cyan (CFP, donor) and yellow (YFP, receptor) fluorescent protein linked by a peptide substrate specific to granzyme B, i.e., VGPDFGR. Upon lytic granule transfer and release of granzyme B into the target cell, granzyme B cleaves the peptide linker of the FRET reporter, and energy transfer from CFP to YFP is thus lost, resulting in a decrease of YFP fluorescence and increase of CFP fluorescence. All NK-target cell co-culture assays were conducted with human primary NK cells purified from healthy blood donors and pre-activated by IL-2 for 3 days, at an NK-to-target cell ratio of 3:1. Target cell death was scored morphologically by blebbing followed by cell lysis, and kinetics of target cell death from NK cell addition to morphological cell death was plotted as cumulative survival curves. As shown in Fig. 1a, the non-cancer cell line, LO2, was the most resistant to NK cell killing, consistent with NK cell’s function in eliminating abnormal and cancerous cells. Among the cancer cell lines, SMMC-7721 and MCF7 were the most sensitive, with MCF7 exhibiting the fastest kinetics to cell death.
In addition to variable sensitivity to overall NK cell killing, we also observed a striking difference in the cytotoxic modes used to kill the different target cell lines. The majority of cell death events (about 60%) seen in SMMC-7721 were preceded by a loss of the granzyme-B FRET (i.e., an increase of CFP (donor) signal and decrease of YFP (acceptor) signal) followed by mitochondrial outer membrane permeabilization (MOMP), suggesting that NK cells killed SMMC-7721 mainly by the lytic granule and granzyme B-mediated cytotoxic pathway and intrinsic apoptosis (Fig. 1b). In contrast, NK cell cytotoxicity towards U-2 OS cells mostly did not associate with a change in the FRET signal of granzyme B. We have previously characterized this mode of target cell death to be mediated by the death ligand, e.g., FasL, and subsequent extrinsic apoptosis [22]. Interestingly, we found necrosis, a much less explored NK cell cytotoxic mechanism, was the primary cytotoxic mode that triggered cell death in MCF7 cells. As shown in the bottom panel of Fig. 1b, MCF7 cell death was not preceded by MOMP, indicating it was not classic apoptosis. Moreover, the signal from the granzyme B-FRET reporter was abruptly lost upon extensive MCF7 cell blebbing, pointing to a large-scale leakage of the intracellular content likely due to membrane ruptures. Such dynamic features were consistent with necrotic cell death. The variable extent of target cell death induced by NK cells was confirmed by flow cytometry analysis of Annexin V staining that measured total cell death (Additional file 1: Figure S1). However, such an ensemble method cannot distinguish the three distinct NK cell killing modes as revealed by our live-cell imaging assays.
Figure 1c summarized the percentage of live target cells and the cell death population via the three distinct cytotoxic modes of primary NK cells exhibited by the five epithelial target cell lines after 12 h of NK-target cell co-culture. The two most sensitive target cell lines, SMMC-7721 and MCF7, which showed rapid kinetics of cell death induction, were killed mainly (around 60%) through granzyme B activity or necrosis, while cytotoxicity mediated by death ligand was more dominant in the less sensitive target cell types, U-2 OS, HeLa, and LO2. All target cell lines showed substantial cell death triggered by death ligands of NK cells, ranging from 25 to 57% of the total target cell population, suggesting the death receptor pathway is widespread. In contrast, the extent of cell death activated by the granzyme B and necrosis pathways varied more significantly, ranging from 4 to 56% for the lytic granule mode and 1 to 57% for the necrotic mode. Such large variability in the sensitivity to NK cell killing via granzyme B and necrosis indicated that activation of these two cytotoxic modes may depend on epithelial features that are more cell type specific. All the cell death data analyzed in our study were averaged from primary NK cells from at least 3 different healthy blood donors. Variability between NK cells from the different donors could be estimated by the standard deviations shown as error bars in the data plots. Such inter-donor variability is much smaller than the phenotypic difference that we characterized in terms of the variable sensitivity of different epithelial cell targets both to overall NK cell killing and to the distinct NK cell cytotoxic modes.
Extent of cytotoxicity, but not cytotoxic mode, correlated with MHCI expression
To investigate the potential molecular determinants underlying the observed variable sensitivity of epithelial cell lines to overall NK cell killing as well as to the distinct NK cell cytotoxic modes, we first profiled the expressions of NK cell-interacting surface molecules known to be involved in cancer-associated NK cell cytotoxicity for the selected target cell panel (Fig. 2a). These molecules include the human MHCI molecules (HLA-A,B,C and HLA-E), NKG2D ligands (MICA, MICB, and ULBP-2,5,6), and DNAM-1 ligands (PVR/CD155 and Nectin-2/CD112) [24]. We also measured the expression of the death receptor Fas and an integrin ligand key for NK cell conjugation, ICAM-1 (CD54), in our analysis by flow cytometry (Fig. 2b). The quantified FACS results showed highly variable expressions of all surface molecules that we profiled between the five target cell lines. The sensitivity of target cell lines to overall NK cell killing correlated relatively well with the expression levels of the inhibitory molecules, HLA-A,B,C and HLA-E, and to a much lesser extent with MICB, but did not correlate with the expression of the other NKG2D activating ligands, MICA and ULBPs, or the activating ligands for DNAM-1. Surprisingly, the expression level of Fas did not correlate with the differential sensitivity to the death ligand-mediated NK cell cytotoxicity. Our data thus indicated that the inhibitory strength of KIRs-HLA-A,B,C and NKG2A-HLA-E may exert the primary control over target cell recognition and overall NK cell cytotoxicity against a particular epithelial target, while control by the activating ligands is less universal and more context-dependent. However, our data did not show any significant correlative feature that was specific to the three distinct NK cell cytotoxic modes. The only differential ligand expression that may potentially render specificity to necrotic killing is MICB, as the MICB expression level in MCF7, the cell line that was particularly prone to necrotic death, was 5–10-folds higher than the other four cell lines. But we found RNAi knockdown of MICB in MCF7 cells did not attenuate necrotic death induced by NK cells (Additional file 1: Figure S2), indicating MICB expression was not the molecular determinant for the necrotic killing.
As we did not observe a correlation between the expressions of well-known NK cell-interacting surface ligands with the variable sensitivity to different NK cell cytotoxic modes, we went on to investigate the involvement of key inhibitory and activating receptors on NK cell surface, as the target specificity could be conferred by surface ligands beyond the profiling panel that we selected. The NK cell receptors that we examined included two types of inhibitory receptors, KIRs and NKG2A (CD94), and three activating receptors important for cancer target recognition, including NKG2D (CD314), DNAM-1 (CD226), and NKp46 (CD335) [11, 21]. Specifically, we used neutralizing antibodies to block individual receptors or a combination of the receptors and then compared the cytotoxic response via the three cytotoxic modes with those under the control condition (i.e., no receptor blocking). We used a broad-spectrum neutralizing antibody against HLA-A, HLA-B, and HLA-C to block the interaction of all KIRs and their MHCI binding partners as a whole, instead of examining the individual KIR, to simplify the analysis.
Figure 2c, d shows the receptor neutralizing results for HeLa cells in co-culture with primary NK cells. As expected, blocking the inhibitory NK-HeLa cell interaction via KIRs-HLA-A,B,C accelerated and enhanced HeLa cell killing, resulting in a degree of cell death similar to that observed in the sensitive target cell lines, such as SMMC-7721 (Fig. 2c). In contrast, blocking the other inhibitory receptor, NKG2A, did not alter cell death significantly. Neutralizing the activating receptors exerted a less prominent effect in attenuating the cell death response of HeLa, possibly because HeLa cells under the control condition were already relatively resistant to primary NK cell killing. Inhibition of NKG2D activity exhibited a stronger effect in attenuating cell death than neutralizing DNAM-1 or NKp46 (Fig. 2c). Double blocking of NKG2D plus DNAM-1 or NKp46 further rescued HeLa cell death, confirming that the NK cell cytotoxicity is regulated by collective, rather than individual, signaling receptors.
We next investigated how blocking individual receptors altered induction of the three cytotoxic modes, expecting individual receptors to selectively regulate particular modes. However, this was not the result. Although blockage of the inhibitory KIRs-HLA-A,B,C mainly enhanced death ligand-mediated HeLa cell death, the fraction of NK cell killing mediated by granzyme B, death ligand, and necrosis all decreased in parallel upon neutralizing the activating receptor NKG2D, either alone or in combination with DNAM-1 or NKp46 (except for the apoptotic population via death ligand under NKG2D inhibition alone) (Fig. 2d). Therefore, in HeLa, receptor modulation appears to largely tune the overall cytotoxic activity of NK cells and/or receptivity of targets, but not the specific death pathway.
Figure 2e summarized the receptor neutralization results for all three pathways across the five epithelial target cell lines (detailed data can be found in Additional file 1: Figure S3). Here, we plotted the ratio of perturbation versus control. Intuitively, data points along the diagonal indicated no change relative to the control condition, and the further away the data points were from the diagonal, the larger the effect of the respective receptor inhibition in diverting the cell death response into, or away from, one of the three cytotoxic modes. Similar to HeLa cells, neutralization of the inhibitory KIRs-HLA-A,B,C interaction exerted a strong effect in enhancing the cell death response of the other four target cell lines, and loss of NKG2D activity exerted the strongest effect in attenuating target cell death. However, as of HeLa cells, the three cytotoxic modes were altered approximately in parallel, again showing that receptor modulation tuned the overall activity of NK cells, but not the activity of one specific death pathway. We also noted that except for MCF7, no other cell line showed significant cell death via necrosis under all receptor perturbation conditions, suggesting that MCF7 cells may have unique cellular features that promote the induction of necrotic killing, which we further investigated below.
It is possible that the receptor neutralization treatment may trigger another type of cell death, i.e., NK cell-mediated antibody-dependent cell-mediated cytotoxicity (ADCC). To examine the involvement of ADCC, we blocked CD16, the key Fcγ receptor on NK cells that mediates ADCC, and then co-cultured the CD16-neutralized NK cells with HeLa cells treated with HLA-A,B,C, neutralizing antibody, as HLA-A,B,C blocking showed the most significant effect in sensitizing HeLa cells to NK cell killing. We found blocking CD16 did not substantially attenuate NK cell killing under the HLA-A,B,C neutralizing antibody treatment, suggesting that ADCC did not contribute significantly to the NK cell killing that we observed (Additional file 1: Figure S4).
Membrane dynamics modulate target sensitivity to NK cell cytotoxic modes
Since receptor expression level did not predict which death pathway is preferred, and receptor inhibition did not modulate the fraction of death caused by a particular pathway, we sought other phenotypic properties that might predict and modulate individual NK cell cytotoxic pathways. We focused on MCF7 and SMMC-7721 because they had the most distinctive preferences, i.e., MCF7 cells were uniquely sensitive to necrosis, while SMMC-7721 cells were sensitive to granzyme B-mediated intrinsic apoptosis. A distinguishing feature of these cell lines that we observed in our movies was the difference in plasma membrane dynamics. MCF7 cells exhibited extensive dynamic protrusions driven by membrane blebs and lamellipodia (Fig. 3a). In contrast, SMMC-7721 cells exhibited only moderate membrane dynamics and the membrane of U-2 OS cells was largely quiet (Fig. 3a). This observation led us to examine whether differential membrane dynamics of the epithelial cell targets contributed to the activation of alternative NK cell cytotoxic modes.
To attenuate the formation of dynamic membrane protrusions, we used either NSC23766, an inhibitor of the small GTPase Rac1, to inhibit actin polymerization-driven lamellipodia, or Y27632, an inhibitor of the Rho Kinase (ROCK), to inhibit actomyosin contractility that promotes membrane blebs [25,26,27]. NSC23766 significantly reduced membrane protrusions in both MCF7 and SMMC-7721 cells, as evidenced by the kymograph of cell edge, which showed a relatively smooth edge under NSC23766 as compared to a jagged edge in the control cells (Fig. 3a). The effect of Y27632, however, diverged in MCF7 and SMMC-7721. MCF7 cells treated with Y27632 showed reduced membrane blebs but longer lamellipodia, a phenotype previously characterized as due to compensatory effects of membrane blebs and lamellipodia [25]. SMMC-7721 treated with Y27632 still showed evident membrane ruffling, indicating bleb did not contribute significantly to drive the membrane dynamics of SMMC-7721. Attenuation of membrane dynamics by both NSC23766 and Y27632 changed the primary cell death mode of MCF7 cells from necrosis to death ligand-mediated apoptosis, with NSC23766 exhibiting a slightly stronger effect (Fig. 3b). NSC23766 treatment of SMMC-7721 also resulted in a switch of NK cell killing from granzyme B-mediated apoptosis to death ligand-mediated apoptosis, while Y27632 did not show a significant effect on altering the NK cell cytotoxic mode (Fig. 3c). Moreover, we did not observe a distinctive change in the membrane distribution of NK ligand (e.g., HLA-A,B,C) on MCF7 cells or the NK receptor (e.g., KIR2D) on NK cell under either NSC23766 or Y27632 treatment, indicating that these two inhibitors did not alter NK cell killing modes by changing the receptor-ligand distribution in the plasma membrane (Additional file 1: Figure S5). Overall, our data suggested that inhibiting plasma membrane dynamics of epithelial targets promoted NK cell killing via death ligand-mediated apoptosis, while damping the other two cytotoxic pathways.
Next, we examined whether increasing the membrane dynamics, e.g., in SMMC-7721, would alter NK cell killing from granzyme B-mediated apoptosis to the necrosis mode, similar to MCF7. We noted that SMMC-7721 cells showed extensive membrane blebs when they started to adhere to the cell culture plate surface (Fig. 3d, left panel). And these blebs disappeared upon complete adherence (Fig. 3a, control). We therefore co-cultured primary NK cells with SMMC-7721 that were trypsinized and seeded onto the culture plate for only 4 h to investigate the effect of enhanced membrane dynamics, in this case, driven by blebs, on NK cell killing mode. Compared to control SMMC-7721 cells that had adhered to the culture plate for 24 h, which were mainly (~60%) killed by granzyme B-mediated apoptosis, the majority (~50%) of the short-adherence cells were killed via the necrotic mode (Fig. 3d, right panel). And the addition of the bleb inhibitor, Y27632, abrogated the increase in necrotic killing of the short-adherence cells (Fig. 3d). As U-2 OS cells adhered to the culture plate surface very quickly (in about 3–5 h) and did not exhibit prolonged blebs, we were unable to use this experimental strategy to study the effect of enhanced membrane dynamics on promoting granzyme B-mediated apoptosis over the death ligand-mediated mode. Nonetheless, overall, our results suggested that the extent of membrane dynamics exerted a key control over the sensitivity of different epithelial targets to the three distinct NK cell cytotoxic modes.
Pro-necrotic membrane dynamics involve actin depolymerization
To determine how a highly dynamic target membrane promotes pro-necrotic killing, we stained the lytic granules of NK cells with an acidic organelle marker, LysoBrite, and investigated the collective dynamics of NK and target cells upon formation of the immunological synapse. NK cells typically showed polarized morphology with the acidic lytic granules stored in the tail end, as they moved in the co-culture environment. Such localization of lytic granules likely prevents undesirable leakage of lytic granules during the constant transient contacts between NK and target cells. When NK cells recognized an abnormal MCF7 cell target, a sustained conjugation called immunological synapse (IS) was formed and subsequently triggered the reorientation of lytic granule from the tail to the IS and then lytic granule transfer (Fig. 4a, left panel; Additional file 2: Movie S1). We found localization and transfer of lytic granule at IS were accompanied by large membrane blebs of MCF7 cell, which, within 4–8 min, ruptured the MCF7 cell membrane and caused necrotic death. We also noted the large membrane blebs tended to form at sites previously showing large lamellipodia (Fig. 4a, frames #1 and #2), suggesting possible conversion of lamellipodia to bleb. SMMC-7721, in comparison, exhibited significantly less membrane bleb upon the localization and transfer of lytic granules at the IS that was followed by the gradual accumulation of granzyme B activity as indicated by the loss of GrzmB-FRET signal, and induced apoptotic death in 30–40 min (Fig. 4a, right panel; Additional file 3: Movie S2). Attenuating MCF7 membrane dynamics by inhibiting Rac1 or ROCK activity abrogated the induction of large membrane bleb at the IS and subsequent necrotic killing (Fig. 4b). And increasing the membrane dynamics of SMMC-7721 by short adherence induced a large bleb at the IS and led to necrotic death, similar to the dynamic phenotype that we observed for MCF7 cells (Fig. 4b, right panel).
As the large membrane bleb was a key feature preceding necrotic killing, we further investigated its mechanistic origin. The observation of possible conversion of lamellipodia to large bleb led us to examine the dynamics of cytoskeleton, in particular F-actin, by imaging a transiently expressed Utrophin-GFP reporter in the MCF7 cells [28]. We also added propidium iodide (PI) in the live-cell imaging assay to determine the onset of membrane leakage in the necrotic killing process. As shown in Fig. 4c, lamellipodia immediately stalled and retracted upon IS formation and the actin cytoskeleton largely depolymerized within 2 min. Membrane leakage indicated by PI diffusion into the MCF7 cell was generally observed 1–1.5 min subsequent to the change in actin dynamics (the PI fluorescent signal was detected in the same channel as LysoBrite). Confocal imaging of F-actin and LysoBrite confirmed the change in actin dynamics and cytoskeleton upon lytic granule localization to the IS at higher spatial resolution (Fig. 4d; Additional file 4: Movie S3). Distinct from the MCF7 cells, IS formation in SMMC-7721 led to the alignment of F-actin into long fibers, likely due to increased contractility known in apoptotic cells (Additional file 5: Movie S4). Together our data suggested that lytic granules of NK cells induced actin deploymerization and loss of lamellipodia that initiated bleb formation, especially at the highly dynamic membrane sites. Subsequent membrane leakage after cytoskeleton destruction likely also contributed to enhance the bleb by increasing osmotic pressure, eventually leading to necrotic membrane rupture.
Necrotic NK cell cytotoxicity is granzyme B-induced necroptosis
NK cell-induced necrosis is poorly characterized at the molecular level. As MCF7 cells were killed mostly by this mechanism, it provided a model for probing the mechanistic origin of necrotic killing. To test the involvement of lytic granules, we used Concanamycin A (CMA), an inhibitor of vacuolar-type ATPase (V-ATPase) that increased the pH of lytic granules, to disrupt lytic granules in NK cells and inhibit their activities. As shown in Fig. 5a, b, treatment of 10 nM CMA reduced the necrotic killing of MCF7 cells from 53 to 4% after 6 h of co-culture with NK cells and also abrogated the lytic granule-mediated apoptosis, which confirmed the key role of lytic granules in activating these two cytotoxic pathways. The extent of death ligand-mediated apoptosis was not affected by CMA treatment, as expected. We also used EGTA to chelate Ca2+ flux that is crucial for lytic granule transfer. We found both the necrotic death and lytic granule-mediated apoptosis in MCF7 cells were significantly decreased (Fig. 5b), again pointing to the involvement of lytic granule transfer in mediating both necrotic and granzyme B-dependent NK cell killing. Moreover, a substantial loss of necrotic MCF7 cell death was observed by treating the NK cells with a pan-granzyme inhibitor, 3,4-Dichloroisocoumarin (DCI), as well as a granzyme B-specific inhibitor, Ac-IEPD-CHO, illustrating that it is the granzyme activity, in particular, granzyme B activity, from the lytic granules that induces the actin depolymerization and bleb formation upon lytic granule transfer to the target cells (Fig. 5b).
Large membrane blebs dependent on the actin cytoskeleton are characteristic of necroptosis, a type of programmed necrosis that is usually induced by cell surface receptors [29,30,31,32]. This pathway has not been previously implicated in the killing by cytotoxic lymphocytes. To test its possible involvement, we knocked down key necroptosis pathway components, including receptor-interacting protein 1 (RIP1), RIP3, and mixed lineage kinase domain-like (MLKL), in MCF7 cells by RNA interference (RNAi). Loss of these three regulators all caused a significant reduction in necrotic killing and a concomitant increase in apoptotic killing (Fig. 5c). We also treated MCF7 cells with commonly used necroptosis inhibitors. The RIP3 inhibitor, GSK872, decreased necrotic MCF7 cell death to a similar degree as the RIP3 knockdown treatment (Fig. 5c). MCF7 cells are known to have very low RIP3 expression, which was attributed to their resistance to conventional necroptosis activated by death receptors [30]. We indeed observed orders of magnitude lower expression of RIP3 in MCF7 by western blot than in HT29, a cell line with high RIP3 expression. But given that both RNAi knockdown of RIP3 and the RIP3 inhibitor significantly attenuated necroptotic killing by NK cells, our results indicate the granzyme-induced necroptosis that we observed could be mediated by a low level of RIP3. We also tested the inhibitory effects of the RIP1 inhibitor, necrostatin-1 (from 10 to 100 μM), and MLKL inhibitor, necrosulfonamide (from 1 to 10 μM). At low concentrations, these two inhibitors did not exhibit significant effects on attenuating necroptotic killing of NK cells, while high concentrations of both inhibitors caused substantial toxicity to the primary NK cells so we were unable to obtain definitive results regarding the inhibitory effects of these two inhibitors.
To further confirm the activation of necroptosis in MCF7 cells treated with primary NK cells, we measured the level and localization of phospho-MLKL (Ser358), a key activation signal of necroptosis, by both western blotting and immunostaining (Fig. 5d and Additional file 1: Figure S6). We found significant upregulation of phospho-MLKL in MCF7 cells upon 3-h co-culture with NK cells. The induction level of phospho-MLKL did not further increase significantly at 6 h of NK-MCF7 cell co-culture, agreeing with our observation from time-lapse imaging that most necrotic NK cell killing occurred early within the first 3 h, while apoptotic killing accounted for the majority of late MCF7 cell death after 3 h. Moreover, immunofluorescence analysis revealed NK cell treatment induced distinctive phospho-MLKL puncta that indicated pore formation. Together with the RNAi data above, our results demonstrated the necrotic NK cell killing that we observed was granzyme B-induced necroptosis.