Actin Cytoskeleton Remodeling Drives Breast Cancer Cell Escape from Natural Killer–Mediated Cytotoxicity

Natural killer (NK) cells are lymphocytes of the innate immune system with cytotoxic activity that contributes to ridding the organism of pathogen-induced infections and cancer cells. Unlike other cytotoxic cells of the adaptive immune system, such as CD8⁺ cytotoxic T lymphocytes (CTL), NK cells kill their targets on a short time scale without requiring preactivation via prior antigen exposure. Accordingly, they are frequently referred to as the first line of defense against cancer (1). NK cells recognize malignant cells by sensing a loss of inhibitory MHC class I molecules (missing-self) and/or an overexpression of activating stress-induced ligands (altered-self), such as the NKG2D receptor ligands ULBP1-6 and MICA/B (2, 3). The balance between inhibitory and activating signals critically determines the activation of NK-cell–mediated cytotoxicity toward their targets, although some complementary mechanisms have also been described (4). In addition to Ca2⁺-dependent exocytosis of cytotoxic enzymes (perforin and granzymes) leading to caspase-dependent and -independent apoptosis, NK cells can also promote slow, caspase-dependent, cancer cell death via the engagement of death receptors (e.g., Fas/CD95, DR4, and DR5) on target cells by their cognate ligands FasL and TRAIL (5). Besides their direct cytotoxicity and the production of proinflammatory cytokines, NK cells modulate the activity of other immune cells (6, 7) and can prime dendritic cells to activate antitumor-specific CTL responses (8, 9).

Although NK cells hold promising cytotoxic activity against transformed cells, their numbers in solid tumors are usually low. To overcome the poor infiltration, an autologous or allogeneic NK cell transfer followed by adjuvant high dose IL2 has been clinically evaluated (10–12). Both treatment regimens proved to be partially effective but failed to induce durable remissions. Major bottlenecks to clinical efficacy remain, including lack of consistent in vivo NK-cell expansion and tumor-induced immunosuppressive mechanisms. With the advent of efficient and FDA-approved autologous T-cell immunotherapy, interest in exploring the applicability of NK cells for use in cancer immunotherapies has been reawakened (13). Genetic modification of NK cells, as demonstrated for CAR T cells, could help to circumvent challenges posed by the tolerance to self-cells, helping establish NK cells as potent effectors against malignancy (14, 15). However, evasion mechanisms mounted by cancer cells and immunosuppressive tumor microenvironments remain major hurdles (13, 16, 17).

Upon recognition of tumor cells, NK cells undergo a series of specific steps leading to directed secretion of preformed lytic granules containing cytotoxic mediators, such as perforin and granzyme B (GzB). An early and key event in NK-mediated cell death is the formation of an immunologic synapse (IS) between the immune cell and its target, the ultimate function of which is to focus lytic granule secretion toward the target cell. Both the formation and activity of the IS critically rely on actin cytoskeleton rearrangements in NK cells (18–22). The initial steps of IS formation are associated with a prominent accumulation of actin filaments (AF), which facilitates the formation and stabilization of the so-called peripheral supramolecular activation cluster SMAC (23). Following actin reorganization, lytic granules converge on the microtubule-organizing center (MTOC) along microtubules. These lytic granules, together with the MTOC, subsequently polarize within the IS (24). High-resolution imaging studies have demonstrated that a dynamic meshwork of fine AFs is located at the center of the IS and this meshwork controls the delivery of myosin IIA-associated granules to specific membrane areas (25–28). The two major classes of actin nucleators (the ARP2/3 complex and formins) were found to play critical but distinct roles during NK-cell–mediated cytotoxicity (19). Recently, coronin 1A was reported to promote deconstruction of the synaptic cortical actin network (29), a process resulting in permissive “clearances” where granule secretion preferentially occurs (25).

In striking contrast to the well-characterized roles of the cell actin cytoskeleton of NK cells during target cell recognition and killing, little is known about actin configurations and functions in target cells during these processes. In the present study, we address this gap and provide for the first time direct and compelling evidence of the pivotal role of the tumor cell actin cytoskeleton in resistance to NK-cell–mediated cell lysis.

The breast adenocarcinoma cell lines used as target cells in this study were purchased from ATCC and include MCF-7, MDA-MB-231, T47D, and Hs578T cells. Each of these cell lines was authenticated and checked for not being cross-contaminated through STR profiling analysis (Microsynth). All cell lines were maintained in DMEM high glucose with l-glutamine medium (Lonza). The NK-92MI cell line was purchased from ATCC and cultured in RPMI-1640 (Lonza) supplemented with 10% (v/v) fetal bovine serum (FBS, Life Technologies) and 10% (v/v) horse serum (ATCC). All other media were supplemented with 10% (v/v) FBS, 100 U/mL penicillin, and 0.1 mg/mL streptomycin (Sigma-Aldrich). Cell lines were cultured in a humidified atmosphere at 5% CO₂ and 37°C and routinely checked for Mycoplasma contamination using the MycoAlert Mycoplasma detection kit (Lonza).

Peripheral blood mononuclear cells (PBMC) were isolated from buffy coats provided by the Luxembourg Red Cross using Lymphoprep density gradient medium (StemCell). Briefly, samples were diluted 1:5 with PBS supplemented with 10% FBS. Thirty milliliters of diluted buffy coat was poured onto 15 mL of Lymphoprep medium in LeucoSep centrifuge tubes (Greiner Bio-One) and centrifuged for 30 minutes at 800 × g in a swinging-bucket rotor with slow acceleration and no brake. The enriched PBMC fraction was harvested and washed twice with PBS containing 10% FBS. Remaining erythrocytes were lysed with 1× ACK lysis buffer for 5 minutes before NK-cell isolation using the MACS NK cell negative isolation kit (Miltenyi Biotec) according to the manufacturer's instructions. Human NK cells were cultured in RPMI medium (Lonza) supplemented with 10% (v/v) FBS, 10% (v/v) horse serum, 100 U/mL penicillin, and 0.1 mg/mL streptomycin, and 100 U/mL recombinant human IL2 (PeproTech).

MCF-7 cells were transfected with wild-type SNAIL and SNAIL-6SA (a constitutively active mutant of SNAIL) expression vectors. These vectors were obtained from Addgene (gift from Mien-Chie Hung; Addgene plasmids #16128 and #16221). Parental cells were transfected 48 hours prior to subsequent analyses using Lipofectamine 2000 transfection reagent (Thermo Fisher Scientific). The expression of recombinant SNAIL/SNAIL-6SA was confirmed by Western blotting. N-WASP and CDC42 knockdown was achieved by transfecting MCF-7 and MDA-MB-231 cell lines with N-WASP and CDC42 siRNAs (siWASP#1 5′-GGUUUGUCGUAAUCCUCUATT-3′, siWASP#2 5′-CCUUAAUGUAAUUUACUUA-3′, siCdc42#1 5′-CAGCAAUGCAGACAUUAATT-3′, and siCdc42#2 5′-CGAUGGUGCUGUUGGUAAATT-3′). The nontargeting siCtrl was purchased from Qiagen. All cell lines were transfected 48 hours prior to each assay using DharmaFECT transfection reagent (GE Dharmacon), and knockdowns were confirmed by a Western blotting. The LifeAct-mEGFP-7 plasmid was obtained from Addgene (gift from Michael Davidson, Addgene plasmid #54610). The LifeAct-mEGFP fragment was subcloned into the lentiviral plasmid pCDH-EF1α-MCS-IRES-Puro (CD532A-2; System Biosciences) using the XbaI and BamHI restriction enzyme sites and NEBuilder HiFi DNA Assembly (NEB) to generate the pCDH-LifeAct-mEGFP expression plasmid used for lentiviral transduction. Transduced cells were selected with puromycin (0.5 μg/mL, Sigma-Aldrich).

In addition to SNAIL-induced epithelial-to-mesenchymal transition (EMT), MCF-7 cells were treated by either 5 ng/mL TGFβ (PeproTech) for 6 days or 10 ng/mL TNFα (BioLegend) for 3 days. Prior to TGFβ treatment, MCF-7 cells were starved for 24 hours (1% FBS).

For actin drug-based assays, target cells were treated with 0.5 μmol/L cytochalasin D or DMSO (Sigma; control) for 30 minutes and washed twice prior to presentation to NK cells.

For conjugate formation, NK cells were counted, stained with anti-human CD56-PE-Cy7 (BioLegend, clone: HCD56), and washed before the direct contact with LifeAct-mEGFP-target cells at an effector-to-target (E:T) ratio of 5:1. Cells were coincubated for 30 minutes prior to fixation with 2% paraformaldehyde (Thermo Fisher Scientific) and permeabilized with 0.1% Triton X-100 (Sigma-Aldrich). Afterward, cells were washed twice with PBS and then stained with anti-human GzB monoclonal antibody-APC (Thermo Fisher Scientific, clone: GB12), anti-γ-tubulin antibody-PE (Santa Cruz, clone: D-10), and DAPI (0.3 μg/mL, Sigma-Aldrich) for 20 minutes. The Amnis brand ImageStreamX Mark II (EMD Millipore) imaging flow cytometer with five built-in lasers (405, 488, 561, 640, and 785 nm) was used for acquisition. Using INSPIRE (EMD Millipore), 2 × 10⁴ events were collected per tube at 60× magnification on a low speed and high-sensitivity settings. Unstained, single-stained, and Fluorescence Minus One- (FMO-) stained samples were collected for each experiment as controls. For better visualization of lytic granules, the extended depth of field was activated during the acquisition of GzB-containing samples. Ideas software (IDEAS 6.2.64.0, EMD Millipore) was used for data analysis. The gating strategy, masks, and features to analyze conjugates were created and applied. Features are used to calculate, analyze, and measure specific intensities in the cell while masks are designed to define a specific area of the cell where features could be applied. Masks can be established on the basis of the bright field or florescence image. The mask overlaying the synaptic region was created by Boolean logics of mEGFP-LifeAct and CD56-PeCy7, Dilate (mEGFP-LifeAct, 1) and Dilate (CD56-PE-Cy7,2). For actin response measurements, the mean fluorescence intensity (MFI) of mEGFP-LifeAct was calculated at the IS mask. To better calculate the intensity of the MTOC and improve the detection of γ-tubulin, the feature of Bright Detail Intensity R7 (BDI) was used. The software's BDI feature can compute the intensity of the bright spot that have radii smaller than 7 pixels. For GzB quantifications in target cells, the mask was defined on the mEGFP-LifeAct after subtracting the region of the immune synapse in order to exclude the intensity of GzB derived from NK cells. In order to assess apoptosis in target cells upon the contact with NK cells, the intensity of the signal of Annexin V–APC signal and DAPI was measured at the mask covering the mEGFP-LifeAct. The intensity of the Annexin V–PE was plotted versus the intensity of DAPI to distinguish between viable (Annexin V⁻, DAPI⁻), early apoptotic (Annexin V⁺, DAPI⁻), late apoptotic (Annexin V⁺, DAPI⁺), and necrotic cells (Annexin V⁻, DAPI⁺).

Target cell surface antigens were stained for 30 minutes prior incubation with NK cells, using the following antibodies: anti-human HLA-A, -B, -C-Brilliant Violet 605 (BioLegend; clone: W6/32), anti-human MICA/MICB-APC (BioLegend; clone: 6D4) and anti-human PD-L1-Brilliant Violet 605 (BioLegend; B7-H1, CD274) antibodies. The data were acquired on our imaging flow cytometer, and MFI values for each ligand were determined at the synaptic region using the above described IS mask.

Overall ligand expression at the cell membrane of control and N-WASP- and CDC42-depleted target cells was analyzed using a BD FACSAria II flow cytometer (BD Biosciences).

The unpaired Student t test in Microsoft Excel 2016 and GraphPad Prism was used to determine the statistical significance of the results obtained. For Annexin V experiment, a Z-score test for two population proportions was used to determine the statistical significance between samples. *, P ≤ 0.05; **, P ≤ 0.01 and ***, P ≤ 0.001.

To examine AF configurations in tumor cells during NK-cell attack, two breast adenocarcinoma cell lines were transduced with lentivirus to achieve stable expression of the actin reporter LifeAct-mEGFP (Fig. 1A; Supplementary Fig. S1A; ref. 30). The epithelial-like MCF-7 cell line (Supplementary Fig. S1B) was chosen for its high susceptibility to NK-cell–mediated lysis, whereas the mesenchymal-like MDA-MB-231 cell line was chosen for its highly resistant phenotype. As shown in Fig. 1B, MDA-MB-231 cells were almost three times less susceptible to NK-mediated lysis when compared with MCF7 cells. Only 13% of MDA-MB-231 cells were killed by NK-92MI cells at an E:T ratio of 5:1 after 4 hours, while 42% of MCF-7 cells were killed under the same conditions. A confocal microscopy analysis conducted 30 minutes after tumor cells were exposed to NK cells revealed that most NK-cell–conjugated MDA-MB-231 cells exhibited a massive accumulation of actin near the IS, hereinafter referred to as the “actin response” (Fig. 1C; Supplementary Movie S1). In contrast, this actin response was rarely observed in NK-cell–conjugated MCF-7 cells. Instead, most conjugated MCF-7 cells exhibited a rather homogeneous distribution of actin at their cortex (Fig. 1D; Supplementary Movie S2).

The actin response was induced remarkably fast, as evidenced by the many MDA-MB-231 cells showing synaptic actin accumulation, in as little as 2 minutes after presentation to NK cells. Target–effector cell conjugates were tracked over long periods of time using live-cell imaging, and time-lapse movies were assembled. In MDA-MB-231 cells, the actin response persisted throughout the whole duration of the interaction between the cancer and immune cells (Fig. 1E; Supplementary Movie S3).

After about 30 minutes, the NK cell detached without achieving lysis of its target, and the actin response in the escaped cancer cell rapidly ceased. As exemplified in Fig. 1E and Supplementary Movie S3, the escaped cells were still alive at the end of the recording (120 minutes; Fig. 1E; Supplementary Movie S3). In most MCF-7 cells, no actin response could be observed at any time during the interaction (Fig. 1F; Supplementary Movie S4). Multiple membrane blebs were detected at the target cell surface after 1 hour, indicating that the target cell entered apoptosis. A few minutes later, the target cell lysed, and the green fluorescence of the actin reporter disappeared.

Quantification of GFP fluorescence showed that, in MDA-MB-231 cells, the cortical region that included the IS contained on average about two times more F-actin as compared with the opposite side of the cell, whereas no statistically significant asymmetrical distribution of actin was found in MCF-7 cells (n = 25; Fig. 2A). Nevertheless, we noticed that a few MDA-MB-231 cells in conjugate with NK cells did not show the typical actin response; conversely, a number of MCF-7 cells accumulated F-actin at the IS. We accordingly assumed that each cell line comprised two cell subpopulations differing in their ability to remodel their actin cytoskeleton in response to NK-cell attack. To further characterize these subpopulations, we used high-throughput imaging flow cytometry and analyzed a large number of cell conjugates. Breast cancer cells were cocultured with NK cells for 30 minutes prior to fixation and permeabilization. Typically, 2 × 10⁴ events were acquired on the imaging flow cytometer, and a gating strategy was applied to select in-focus, live, and interacting effector–target cell pairs (Fig. 2B). For analysis, a mask was manually generated using the IDEAS software to define the region of interest corresponding to the IS and its surrounding intracellular areas, hereinafter referred to as the “synaptic region” (Supplementary Fig. S1C). Such a mask was automatically applied to the data sets generated by imaging flow cytometry and confirmed to efficiently capture the actin response as exemplified in Fig. 2C. For each cell line, the subpopulations of cells with or without an actin response were discriminated by comparing the relative intensity of LifeAct-mEGFP in the synaptic region and in the entire cell (Supplementary Fig. S1D). From three independent experiments, including a total of at least 500 conjugates, we calculated that 62% of MDA-MB-231 cells conjugated with NK cells exhibited an actin response, whereas the remaining 38% did not (Fig. 2D). The opposite results were obtained for the MCF-7 cell line, in which only 22% of conjugated cells showed an actin response.

To extend the above data and further evaluate the relationship between tumor cell resistance to NK-mediated cell death and the actin response, we analyzed two additional breast cancer cell lines, the epithelial-like T47D cell line, and the mesenchymal-like Hs578T cell line (Supplementary Fig. S1B). As previously described, these cell lines were transduced to stably express LifeAct-mEGFP in order to image actin in target cells only, and to exclude any potential spillover effects from the actin cytoskeleton of NK cells (Supplementary Figs. S1A and S2A). Cytotoxic assays revealed that, similar to MDA-MB-231 cells, Hs578T cells were highly resistant to NK-cell lysis, displaying an NK-cell–mediated cytotoxicity of 13% at an E:T ratio of 5:1 (Fig. 3A). In addition, Hs578T cells were highly competent for the actin response, and more than 60% of conjugated tumor cells exhibited F-actin accumulation in the synaptic region (Fig. 3B; Supplementary Fig. S2B). In comparison, T47D cells were significantly more susceptible to NK-cell–induced lysis, and 36% displayed immune cell–mediated lysis at an E:T ratio of 5:1 (Fig. 3A). These cells also exhibited a reduced ability to respond to NK-cell attack by an actin response (<30% of conjugated cells; Fig. 3B; Supplementary Fig. S2B). Interestingly, the amplitude of the actin response was relatively similar across cell lines with 30% to 45% of total cell F-actin accumulated at the synapse (Fig. 3C).

Taken together, our data reveal that a subpopulation of tumor cells responds to NK-cell attack by fast remodeling of their actin cytoskeleton, leading to a massive F-actin accumulation in the synaptic region. This is a process we termed the actin response. Remarkably, this subpopulation was significantly larger in NK cell-resistant breast cancer cell lines than in more susceptible cell lines, suggesting that the actin response contributes to the resistant phenotype. In addition, our data point to a potential link between the EMT status of tumor cells and the actin response. Indeed, mesenchymal-like breast cancer cell lines, including MDA-MB-231 and Hs578T, proved to be significantly more competent for the actin response when compared with the epithelial-like ones, including MCF-7 and T47D (Figs. 2D and 3B; Supplementary Fig. S1B).

To further evaluate the relationship between EMT, actin response, and resistance to NK-cell lysis, EMT was experimentally induced in MCF-7 cells through transient expression of the wild-type EMT inducer SNAIL, or its degradation-resistant variant SNAIL-6SA variant. The mesenchymal transition of SNAIL- and SNAIL-6SA–expressing cell lines was evident by the adaption of a spindle-like morphology and reduced cell–cell contacts, as well as by the loss of E-cadherin (Fig. 3D). Both SNAIL- and SNAIL-6SA–overexpressing cells exhibited a robust increase (>2-fold) in the subpopulation of cells responding to NK-cell attack via synaptic F-actin accumulation as compared with control, transfected only, MCF-7 cells (Fig. 3E; Supplementary Fig. S2B). Consistent with its enhanced stability, SNAIL-6SA had a stronger effect than wild-type SNAIL, resulting in 82% and 74% of conjugated cells exhibiting an actin response, respectively (Fig. 3E). Forced expression of SNAIL or SNAIL-6SA also dramatically decreased MCF-7 cell susceptibility to NK-cell–mediated lysis. At an E:T ratio of 5:1, SNAIL- and SNAIL-6SA–expressing MCF-7 cells were 2- and 3-fold more resistant to NK-cell lysis when compared with the control cells, respectively (Fig. 3F).

In addition to SNAIL-induced EMT, we treated MCF-7 cells with either TGFβ or TNFα, and subsequently analyzed their competency for the actin response and their susceptibility to NK-cell–mediated cell lysis. Treatment efficacy was validated by E-cadherin downregulation and changes in cell morphology (Fig. 3G). Consistent with the above data, both treatments strongly increased the rate of actin response in MCF-7 cells as well as MCF-7 cell resistance to NK cells (Fig. 3H and I).

Together, our data indicate that the EMT status is a critical determinant of the capacity of breast cancer cells to remodel their actin cytoskeleton upon NK-cell attack, and to escape from NK-cell–mediated lysis.

We next aimed to evaluate the causal link between the actin response and breast cancer cell resistance to NK-cell–mediated lysis. First, we examined whether inhibition of synaptic F-actin accumulation affected breast cancer cell survival in cytotoxicity assays. Actin polymerization was impaired in target cells by knocking down either N-WASP (a key regulator of the ARP2/3 complex) or the Rho-subfamily small GTPase CDC42 (its upstream regulator; ref. 31). Resistant MDA-MB-231 and susceptible MCF-7 cell lines were transfected with small interfering RNA constructs (siRNAs) targeting N-WASP or CDC42 transcripts (siN-WASP and siCDC42, respectively). The two independent siRNAs used to target each transcript significantly decreased the corresponding protein expression levels in the two cell lines (Fig. 4A). Imaging flow cytometry analysis showed that N-WASP- and CDC42 depletion reduced the subpopulation of actin response-competent cells in MDA-MB-231 cells by 2- to 3-fold (Fig. 4B). This inhibition of the actin response was associated with a significant increase in target cell susceptibility to NK-cell–mediated lysis, ranging from 50% to 150%. Conversely, N-WASP or CDC42 knockdown did not significantly affect the already reduced size of this subpopulation (around 20%) of MCF-7 cells that produce an actin response upon NK-cell attack (Fig. 4C), nor did it increase the susceptibility of MCF-7 cells to NK-cell lysis. The depletion of N-WASP or CDC42 was verified not to modify the overall expression of important ligands of NK-cell receptors at the target cell surface, including HLA-A, -B, -C, MICA/B, and program cell death-ligand 1 (PD-L1; Supplementary Fig. S3A–S3C), supporting our hypothesis that the increase of MDA-MB-231 cell susceptibility following N-WASP or CDC42 depletion is indeed the result of actin response inhibition.

We next quantified the extent of target cell apoptosis associated with the actin response. Briefly, target cells were incubated for 30 minutes with NK cells, labeled with Annexin V–APC and DAPI, and subsequently analyzed by imaging flow cytometry. For the analysis, a new mask was designed to capture Annexin V–APC and DAPI signals only in the LifeAct-mEGFP–expressing target cell from conjugates (Fig. 4D). For both MDA-MB-231 and MCF-7 cell lines, the number of apoptotic cells was considerably higher in the subpopulation of cells without an actin response when compared with the subpopulation of cells with an actin response. Consistent with the reduced subpopulation of actin response-competent cells in the MCF-7 cell line, this cell line contained a much higher total number of apoptotic cells when compared with the MDA-MB-231 cell line. Altogether, the above data provide further evidence that the actin response mediates breast cancer cell resistance to NK-cell–mediated lysis. The data also show that the inhibition of this process is sufficient to restore susceptibility in otherwise resistant breast cancer cells.

The possibility that the actin response was associated with local changes in the density of HLA-A, -B, -C, MICA/B, and PD-L1 on target cell surface was evaluated using imaging flow cytometry. Very interestingly, both HLA-A, -B, -C ligands and PD-L1 levels were increased (by 100% and 50%, respectively), while MICA/B ligands were modestly reduced (30%), at the synaptic region of MDA-MB-231 cells with an actin response as compared with MDA-MB-231 cells without an actin response (Supplementary Fig. S3D–S3F). Likewise, HLA-A, -B, -C ligands were significantly enriched at the synapse of MCF-7 cells with an actin response as compared with MCF-7 cells without an actin response (Supplementary Fig. S3D). However, synaptic MICA/B levels did not significantly differ in the two MCF-7 cells subpopulations (Supplementary Fig. S3E). PD-L1 was not amenable to quantification in the synaptic region of MCF-7 cells by imaging flow cytometry in our experimental setup. Together, these data suggest that the actin response increases inhibitory ligands, and possibly decreases activating ligands, at the synaptic region of target cells, respectively, which in turn may alter NK-cell function.

We then evaluated if the actin response was associated with a defect in NK-cell activation. To do so, we quantified MTOC polarization in NK cells conjugated with MCF-7 or MDA-MB-231 cells. The two cell lines induced a similar overall polarization of the MTOC in NK cells (Fig. 5A). Moreover, MTOC polarization was not statistically different in NK cells conjugated with target cells with or without an actin response (Fig. 5B). In addition, MCF-7 and MDA-MB-231 cell lines induced similar NK-cell degranulation as evaluated by the degranulation marker lysosome-associated membrane protein-1 (LAMP-1)/CD107a (Fig. 5C; refs. 32, 33). However, these results do not totally rule out a role for the actin response in blocking NK-cell activation through modifying the ligand density at the synaptic region. Noticeably, NK92MI cells are a highly cytotoxic NK cell line lacking most of the killer inhibitory receptors (KIR) and, accordingly, are insensitive to an increase of HLA-A, -B, -C ligands (Supplementary Fig. S3D; refs. 34–36).

Following IS formation, NK cells directionally exocytose specialized secretory lysosomes containing cytotoxic proteins. Among these, the serine protease GzB is one of the central mediators of target cell death. We assessed the possibility that the actin response may alter GzB levels in target cells. First, we quantified GzB levels in NK-cell–conjugated target cells with or without an actin response using imaging flow cytometry. To exclude the bulk of GzB contained in effector cells and exclusively focus on the GzB fraction that is transferred to target cells, an appropriate mask was designed for data analysis (Fig. 5D). The poorly susceptible MDA-MB-231 cells were used as targets. We found that GzB levels were markedly reduced (by about 50%) in the cell subpopulation with an actin response compared with the cell subpopulation lacking an actin response. Next, we analyzed the effects of pharmacologic disruption of the actin response on GzB levels in target cells. The concentration of the actin-depolymerizing drug cytochalasin D (CD) and the treatment duration were optimized to disrupt most filamentous actin in MDA-MB-231 cells, without compromising cell viability (Supplementary Fig. S4A and S4B). In addition, CD was carefully washed out prior to target cell incubation with NK cells in order not to impair their function, leaving a reduced, but sufficient, time window for analysis before substantial actin cytoskeleton recovery (Supplementary Fig. S4C). As shown in Fig. 5E, CD treatment strongly reduced the subpopulation of actin response–competent target cells by 3-fold. Along with this effect, CD treatment significantly increased the overall average level of GzB in conjugated target cells, indicating that the actin response was causally linked to GzB level reduction (Fig. 5F). To rule out any potential effect of residual amounts of CD on NK-cell activity, we also quantified GzB levels in target cells in which the actin response was genetically impaired via N-WASP or CDC42 knockdown (see Fig. 4). Consistent with the above data, N-WASP and CDC42 depletion significantly increased the overall average level of GzB in target cells by about 2-fold (Fig. 5G). Stratifying the data according to the presence or absence of an actin response revealed that the fraction of cells that resisted F-actin disruption and responded to NK attack by synaptic actin accumulation maintained reduced levels of GzB (Supplementary Fig. S5). Collectively, our data indicate that the actin response protects target cells from lysis by limiting GzB accumulation.

We asked whether the protection of breast cancer cells by the actin response was restricted to cytotoxicity induced by the NK-92MI cell line used in this study. To address this question, primary NK cells were isolated from buffy coats of five human donors by negative selection (>99% purity; Supplementary Fig. S6), and they were used in the subsequent analyses as effector cells. First, we evaluated primary NK-cell cytotoxicity against the MCF-7 and MDA-MB-231 cell lines (Fig. 6A). As expected, donor-derived NK cells exhibited reduced cytotoxic activity compared with the highly cytotoxic NK-92MI cell line, and only modest target cell killing (1%–10%) was achieved at an E:T ratio of 1:1. However, significant target cell death (5%–33%) was induced at an E:T ratio of 5:1, confirming that isolated primary NK cells were functional. Despite interindividual variability in the cytotoxic potential, NK cells from all five donors more effectively killed MCF-7 cells when compared with MDA-MB-231 cells. Consistent with our previous data, confocal microscopy revealed that a majority of MDA-MB-231 cells conjugated with primary NK cells exhibited an actin response (Fig. 6B), whereas most MCF-7 cells did not. Quantitative analysis using imaging flow cytometry analyses established that the actin response was at least twice as frequent in MDA-MB-231 cells (60%–76%) as compared with MCF-7 cells (26%–37%; Fig. 6C). In addition, for each of the combinations of the five sources of primary NK cells and the two target cell lines, we found significantly lower GzB levels in target cells exhibiting an actin response when compared with those without an actin response (Fig. 6D and E). In conclusion, the actin response and its protective effects were fully recapitulated using donor-derived primary NK cells as effector cells.

Tumor escape from cytotoxic immune cells is a major hurdle for achieving efficacious immunotherapies. Here, we demonstrate a critical role for the actin cytoskeleton in driving breast cancer cell resistance to NK-cell–mediated lysis. Using high-throughput imaging flow cytometry, we found that common tumor cell lines contain two subpopulations of cells differing in their capacity to respond to NK-cell attack via fast and prominent accumulation of AFs near the IS. Remarkably, the rate of the “actin response” in a given cell line is inversely correlated with the overall susceptibility of this cell line to NK-cell–mediated lysis. Live-cell imaging provided direct evidence that tumor cells exhibiting an actin response survive NK-cell attack and remain alive after immune cell detachment, while tumor cells without an actin response are efficiently lysed. Accordingly, apoptosis in NK-cell–conjugated target cells is markedly reduced in the actin response–competent cell subpopulation. Moreover, inhibition of the actin response is sufficient to convert the initially resistant MDA-MB-231 cell line into a highly susceptible phenotype. Altogether, these findings demonstrate a causal relationship between the actin response and resistance to NK cells.

The actin response is a remarkably fast and localized process that takes place almost immediately after physical contact between the target and effector cells. Tracking individual cell conjugates over time revealed that the actin response lasts throughout the entire period of interaction and rapidly stops after effector cell detachment. This suggests a model in which a signal from the IS is transmitted to the proximal tumor cell cortex where it induces sustained actin polymerization. Although the upstream components of the signaling pathway remain to be identified, the robust inhibition of the actin response induced by N-WASP or CDC42 knockdown suggests that the ARP2/3 complex is a key downstream effector responsible for the burst of actin polymerization following NK-cell attack. The role of N-WASP and CDC42 in driving ARP2/3 complex-dependent actin polymerization at the membrane has been extensively documented (37, 38). Notably, CDC42 is a key regulator of membrane protrusions (such as filopodia) and polarity (39, 40). Consistent with this, the actin response is associated with spike-like projections (Fig. 1C and E). Similar to filopodia or invadopodia, actin nucleators of the formin family (41) and AF crosslinking proteins, such as fascin (42) or CSRP2 (43), are likely to be required for the extension of actin response-associated protrusions.

From a functional standpoint, the actin response was associated with a significant reduction in GzB levels in target cells. Moreover, inhibiting the actin response through pharmacologic impairment of actin dynamics or genetic ablation of CDC42 or N-WASP restored high GzB levels in target cells. Interestingly, our data show that the actin response is associated with modifications in the density of NK-cell receptor ligands in the region of the synapse. Most noticeably, both MCF-7 cells and MDA-MB-231 cells with an actin response exhibited a significant increase (+100%) of HLA-A, -B, -C inhibitory ligands at the synapse as compared with the respective cells without an actin response. Such modifications provide a mechanistic insight into how the actin response mediates target cell resistance to NK-cell–mediated lysis. However, we found no association between the actin response and abnormal MTOC polarization or degranulation activity in the effector cells. These results can be explained by the lack of expression of most KIR receptors in the NK cell line (NK92MI) used in our study, which renders this cell line insensitive to HLA-A, -B, -C inhibitory ligands (34–36). Consequently, the decrease in GzB levels observed in target cells with an actin response must originate from an additional mechanism, e.g., actin response-mediated obstruction of GzB uptake into target cells or actin response-mediated GzB degradation inside target cells.

In this regard, we recently reported that autophagy promotes NK-cell–derived GzB degradation in hypoxic MCF-7 cells, thereby reducing target cell susceptibility to NK-cell–mediated lysis (44, 45). In keeping with this, inhibition of autophagy in target cells improves tumor elimination by NK cells in in vivo mouse models of breast cancer and melanoma (45). Considering the multiple and critical roles of the actin cytoskeleton during various steps of autophagy (46), follow-up studies should elucidate the link between the actin response and autophagy-mediated GzB degradation in tumor cells.

There is mounting evidence that EMT promotes tumor cells' escape from cytotoxic immune cells, such as CTL- and NK cells (47–50). In support of this, our data show that the mesenchymal-like breast cancer cell lines have a much higher capacity to generate an actin response as compared with epithelial-like breast cancer cell lines. Furthermore, both genetic (SNAIL or SNAIL-6A overexpression) and pharmacologic (TGFβ or TNFα treatment) induction of EMT increases tumor cell competency for the actin response and resistance to NK-cell–mediated cell death. During EMT, extensive actin cytoskeleton remodeling is required to drive morphologic and functional adaptations, such as the acquisition of migratory and invasive properties (51). Thus, EMT-associated cytoskeletal changes likely confer tumor cells with an increased capacity to undergo rapid actin remodeling in response to immune attack.

An intriguing finding of our study is that breast cancer cell lines contain two main subpopulations of cells differing in their capacity to mobilize the actin cytoskeleton in response to NK cell attack and to survive this attack. A direct and important implication of this previously unknown facet of intra-cancer cell line heterogeneity is that the apparently high susceptibility of a given cell line, such as MCF-7, may actually mask the existence of a minor subpopulation of immune-resistant (and actin response–competent) cells. This knowledge should be taken into consideration in cytotoxicity assay-based studies, and efforts should be made to target the actin response–competent cell subpopulation. The characterization of the signaling pathway(s) controlling the actin response and the identification of druggable molecular targets to impair this process are promising future directions. These efforts might help sensitize intrinsically resistant cancer cells to immune cell–mediated cytotoxicity and improve the efficacy of immunotherapies.

J.P. Thiery is CSO at Biocheetah Singapore, has ownership interest (including stock, patents, etc.) in CNRS Paris, is a consultant/advisory board member for AIM Biotech Singapore, Biosyngen Singapore, and ACT Genomics Taiwan. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Al Absi, B. Janji, C. Thomas

Development of methodology: A. Al Absi, C. Guerin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Al Absi, H. Wurzer, C. Guerin, C. Hoffmann, F. Moreau, X. Mao, J. Brown-Clay, R. Petrolli, C. Pou Casellas

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Al Absi, C. Guerin, C. Hoffmann, J.-P. Thiery, G. Berchem, C. Thomas

Writing, review, and/or revision of the manuscript: A. Al Absi, H. Wurzer, J. Brown-Clay, J.-P. Thiery, S. Chouaib, G. Berchem, B. Janji, C. Thomas

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Dieterle

Study supervision: A. Al Absi, B. Janji, C. Thomas

The authors are grateful to Tony Kaoma (bioinformatics and modeling, Department of Oncology) for support in statistical analyses and to Vanessa Marani (LECR) for technical assistance.

A. Al Absi and H. Wurzer are recipients of PhD fellowships from the National Research Fund (FNR), Luxembourg (AFR7892325 and PRIDE15/10675146/CANBIO, respectively). J. Brown-Clay is recipient of a Postdoctoral fellowship from "Fonds De La Recherche Scientifique", FNRS "Télévie" (7.4512.16). Work in C. Thomas's team is supported by National Research Fund, Luxembourg (C16/BM/11297905) and the Cancer Foundation (FC/2016/02).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.