Tethering of Intercellular Adhesion Molecule on Target Cells Is Required for LFA-1–Dependent NK Cell Adhesion and Granule Polarization

The α_Lβ₂ integrin LFA-1 (a heterodimer of CD11a/CD18) binds to ICAMs and mediates arrest of rolling leukocytes in blood vessels (1) and tight adhesion of cytotoxic lymphocytes to target cells, which is essential for the cytotoxic activity of T cells and NK cells (2–4). LFA-1 plays a central role in the organization of immunological synapses formed by T cells and NK cells (5–10). LFA-1 on resting T cells is kept in an inactive, low-affinity conformation and changes into a higher affinity conformation through “inside-out” signals delivered by other receptors (11). In contrast, resting NK cells bind directly to ICAM-1 in a signaling dependent way (12, 13). Furthermore, LFA-1 binding to ICAM-1 and ICAM-2 is sufficient to induce polarization of cytolytic granules in NK cells (12, 14, 15). In mouse NK cells, talin is required for outside-in signaling by LFA-1, which, together with signaling by NKG2D, induces granule polarization (16).

Whereas polarization of cytolytic granules is induced by LFA-1 in NK cells, degranulation is triggered by the low-affinity FcγR CD16 or by synergistic combinations of coactivation receptors, independently of LFA-1 (10, 13, 14, 17). This uncoupling of signals for granule polarization and degranulation observed in NK cells is very different from cytotoxic T cells, in which these two functions are controlled centrally by the TCR and are costimulated by other receptors including CD28 and LFA-1 (9, 18, 19). The ability of LFA-1 in NK cells to signal autonomously make NK cells an excellent tool to study LFA-1–dependent functions.

Lateral segregation of proteins within cell membranes leads to functional subcompartementalization of the lipid bilayer (20, 21). How receptor distribution at the plasma membrane and receptor binding to the cytoskeleton determines signaling output has been examined (16, 22–28). In contrast, little is known about how receptor function is controlled by the distribution of ligands on target cells. A few studies have suggested that ligand distribution on target cells may be important for proper recognition by T cells and NK cells. A striking finding was that the cytoplasmic tail of CD80 (B7-1), a ligand of CD28 expressed on APCs, is required for the proper segregation of CD28 at immunological synapses and for full T cell activation, highlighting an unexpected contribution of the intracellular portion of a ligand to synapse organization and to signaling in another cell (29). A mutation in the acylation site of MICA, an NKG2D receptor ligand on target cells, resulted in diminished NKG2D-dependent killing by NK cells, suggesting that sorting of NKG2D ligands into cholesterol-enriched membrane domains of target cells improves signaling by NKG2D in NK cells (30). An earlier study suggested that polarization of ICAM-2 to the uropod of a target cell through expression of the ezrin/radixi/moesin protein ezrin rendered the cells more sensitive to lysis by NK cells (31). However, whether it is distribution, clustering, or mobility of ICAM at the surface of target cells that is critical for LFA-1–dependent responses is not known.

In this study, we tested how the distribution of ICAM on target cells affects NK cell function. We focused on LFA-1–dependent functions, namely adhesion to target cells and polarization of cytolytic granules. We have used several approaches to manipulate the attachment of ICAM-1 and ICAM-2 to the cytoskeleton or to artificial surfaces. Our results show that it is the immobilization of ICAM, rather than polarization to one end of the cell or clustering, that is required for functional interaction with LFA-1.

Human NK cells were isolated from peripheral blood cells by negative selection using an NK cell isolation kit (Miltenyi Biotec, Auburn, CA). Resting NK cells (95–99% CD3⁻CD56⁺) were resuspended in IMDM (Invitrogen, Carlsbad, CA) supplemented with 10% human serum (Valley Biomedical, Winchester, VA) and used 1–2 d after isolation. Polyclonal IL-2–activated NK cells were expanded in the presence of feeder cells (0.5 × 10⁶ PBLs/ml, gamma-irradiated with 4.5K) with IMDM supplemented with 10% human serum, 100 IU/ml rIL-2 (Hoffman-La Roche, Basel, Switzerland), and 10% purified human IL-2 (Hemagen Diagnostics, Columbia, MD). IL-2–activated cells were used 2–3 wk after isolation. The B cell lymphoma line 721.221 was cultivated in IMDM supplemented with 10% heat-inactivated FBS (Hyclone, Logan, UT). The mouse thymoma cell line BW5147 (a gift from T. Kamala, National Institute of Allergy and Infectious Diseases, National Institutes of Health, Bethesda, MD) was cultivated in IMDM supplemented with 10% heat-inactivated FBS.

BW5147 cells were transfected with hEzrin-EGFP (pHJ421) human ezrin coding sequence subcloned into pEGFP-N1 eukaryotic expression vector [a gift from J.-J. Hao and S. Shaw, National Cancer Institute, National Institutes of Health (32)], using the BTX machine (ECM830, settings: 230 V, 10 ms, 1 pulse; Harvard Apparatus, Holliston, MA). Cells were recovered for 16 h at 37°C and 5% CO₂, and hEzrin-EGFP–positive cells were selected by FACS. Stable transfectants were generated from the FACS-sorted cells by cultivating them in IMDM/10% FBS/1 mg/ml Geneticin (Life Technologies, Grand Island, NY) and subcloning.

All inhibitors were dissolved in DMSO (Sigma-Aldrich, St. Louis, MO). For latrunculin A (LatA) or jasplakinolide (Jasp) (both Calbiochem, San Diego, CA) treatment, 10 × 10⁶ 721.221 or K562 target cells were resuspended in 1 ml IMDM supplemented with 10% FBS containing 2% DMSO as a negative control to match the final concentration of carrier in treated cells. For the experiments, cells were treated 0.5, 3, or 20 μM LatA or 0.5, 1, or 2 μM Jasp. Staining F-actin with phalloidin revealed that 3 μM LatA was required to disrupt F-actin completely and that the effect of Jasp was evident at a concentration as low as 0.5 μM (data not shown). Cells were incubated for 40 min at 37°C. Inhibitors were washed away prior mixing with NK cells for the different assays.

Conjugate formation between NK cells and target cells was determined as previously described (12) with minor modifications. Briefly, NK cells were labeled with 1 μg/ml Cell Tracker Green CMFDA (Invitrogen) for 30 min at 37°C and 5% CO₂, washed, and incubated for another 30 min at 37°C and 5% CO₂. If BW5147 cells were used as targets, NK cells were labeled with 20 μl/ml anti-CD56 allophycocyanin (BD Biosciences, San Jose, CA) for 15 min at 37°C and 5% CO₂. Target cells were labeled with PKH26 Red (721.221, K562) or PKH67 Green (BW5147 cells) (both Sigma-Aldrich) for 5 min at room temperature, washed extensively, and recovered for 30 min at 37°C and 5% CO₂. NK cells and target cells were mixed at a 1:2 E:T cell ratio with 1 × 10⁵ NK cells and 2 × 10⁵ target cells at 4°C. Cells were spun down at 20 × g for 3 min, and conjugate formation was stopped by vortexing and fixation of cells using 0.5% paraformaldehyde solution (Electron Microscopy Sciences, Hatfield, PA) after 0-, 5-, 10-, 20-, 30-, or 60-min incubation at 37°C. Conjugate formation was determined by flow cytometry (FACSCalibur; BD Biosciences) and is represented as the fraction of NK cells that shifted into two-color conjugates. For blocking of LFA-1 on NK cells, 1 × 10⁶ Cell Tracker Green-labeled NK cells were pretreated with 20 μg/ml IgG2a (HOPC-1; Sigma-Aldrich) or anti-human CD18 (Calbiochem) for 15 min at 4°C.

The degranulation assay was performed as described previously (14). Briefly, 2 × 10⁵ NK cells were added to 4 × 10⁵ 721.221 or K562 cells in a total volume of 200 μl IMDM supplemented with 10% heat-inactivated FBS. Cells were mixed and incubated for 1 h at 37°C and 5% CO₂. Afterward, the cells were spun down and stained with PE-conjugated anti-CD56 (BD Biosciences) and FITC-conjugated anti-CD107a (BD Biosciences) Ab for 45 min at 4°C. Degranulation of CD56-positive NK cells was analyzed by flow cytometry.

The polarization assay was performed as described previously (15). Briefly, NK cells and target cells (721.221, K562, or BW5147) were mixed at a 1:1 E:T ratio, with 1 × 10⁶ of each cell type per sample. Cells were incubated for 20 min at 37°C and 5% CO₂. Cells were resuspended and put onto a poly-d-lysine–coated 2-well Culture Slide (BD Biosciences). Cells were allowed to settle down for 1 h at room temperature and fixed using 4% paraformaldehyde (Electron Microscopy Sciences). Cells were permeabilized using 0.5% Triton X-100 and stained using an anti-human perforin Ab (Pierce, Rockford, IL) and an Alexa 488-conjugated secondary goat anti-mouse IgG Ab (Molecular Probes, Eugene, OR). Cells were imaged by confocal microscopy (LSM510; Zeiss, Oberkochen, Germany) with a ×40, 1.3 NA Plan Neofluar (Zeiss) oil immersion objective lens. Fifty to 200 NK cells in contact with target cells were analyzed for polarization of perforin containing granules.

ICAM-1 on 721.221 cells was stained using a PE-conjugated anti–ICAM-1 Ab (BD Biosciences). Fluorescence recovery after photobleaching (FRAP) was performed using a confocal microscope (LSM510; Zeiss). A region of the cell surface was bleached using the 488-nm laser. The recovery of the bleached region was captured at one frame per 30 s for 10 frames. During the assay, cells were maintained at 37°C. The fluorescence intensities of the bleached region and an unbleached region of the same cell over time were analyzed using the LSM510 software (Zeiss). The medians of the relative intensities of different cells were plotted over time.

Endogenous cell surface ICAM-1 or ICAM-2 on 721.221 or BW5147 cells was fluorescently labeled with PE-conjugated anti-human ICAM-1 (BD Biosciences), anti-human ICAM-2 (Beckman Coulter, Fullerton, CA), or anti-mouse ICAM-2 (BD Pharmingen) Ab. Total internal reflection fluorescence (TIRF) imaging was performed using an Olympus IX81 TIRF microscope equipped with a 561-nm diode-pumped laser (Cobolt, Stockholm, Sweden), ×100 1.45 NA Olympus TIRF microscopy lens, and a Cascade II 1024B EM-CCD camera (Photometrics, Tucson, AZ). Cells were maintained at 37°C using a LiveCell environmental chamber (Pathology Devices, Westminster, MD). Images were captured at 32 frames/s using MetaMorph software (Molecular Devices, Sunnyvale, CA) for 250 frames. The movement of labeled ICAM molecules was automatically tracked using code developed for MatLab software [http://physics.georgetown.edu/matlab (33)]. The code was modified to refine particle positioning with a two-dimensional Gaussian fit (34) and to determine intensities of individual particles. Mean square displacements were determined from positional coordinates to calculate short-range diffusion coefficients of individual particles for five time intervals. The resulting diffusion coefficients were plotted as a cumulative distribution function (CDF). A CDF plots the same information as a histogram for a given data set; however, where the histogram plots the frequency distribution (y-axis) as a function of the binned data set (x-axis), the data are not binned for a CDF plot. The CDF plot is generated as follows: the data set is sorted from the smallest to greatest value. Next, the number of data points is summed, and the value of the position of a data point on the sorted list is divided by the sum to yield the probability value (y-axis). Finally, based on its position on sorted list, the probability is plotted against the value of the data point. The CDF displays the distribution of the data set from the smallest (from left on the x-axis) to greatest value (at the right of the x-axis) and provides the probability (y-axis) of whether a particular value will occur at or less than a specified point on the x-axis. Moreover, a CDF plot allows for the rapid identification of the median value of the data set by interpolation at 50% on the y-axis, it is not prone to binning artifacts, and the separation of the data set by log scale is more apparent for visual inspection. In addition, graphical comparisons between data sets are generally more easily interpretable relative to overlaying histograms.

Planar lipid bilayers carrying human ICAM-1 tagged with six histidines were formed between a coverslip and the microaqueduct slide of a Bioptechs parallel plate flow chamber-FCS2 (Bioptechs, Butler, PA) as described previously (10, 11). A human ICAM-1-Fc fusion protein was coated on coverslips in 100 mM sodium bicarbonate (pH 9.2) at 10 ng/ml, 100 ng/ml, 1 μg/ml, and 10 μg/ml at 4°C overnight. Five-percent FBS-containing culture medium was used to block nonspecific binding. The ICAM-1-Fc fusion protein cloned in the CD5lneg1 vector (35) was produced by transfection into 293T cells, as described previously (36). To avoid stimulation of NK cells via CD16 by the human IgG1 Fc portion of the ICAM-1-Fc fusion protein, the CD16 binding site was mutated in the CD5lneg1 vector, as described previously (14). Briefly, aa 235 and 236 of human IgG1 within the Fc domain were mutated as Leu²³⁵ to Gly²³⁵ and Gly²³⁶ to Leu²³⁶. Whereas unmutated ICAM-1-Fc fusion protein that was attached to plates triggered degranulation in NK cells, the ICAM-1-Fc protein carrying the mutated CD16 binding site did not. The movement of NK cells was measured 20 min after injection and tracked for 25 frames with a 10-s interval between frames by Dynamic Image Analysis System (37).

Two functions of NK cells for which LFA-1 signaling is sufficient are adhesion and granule polarization. The B cell lymphoma cell line 721.221 (38), which does not express HLA-A, HLA-B, and HLA-C, is sensitive to lysis by NK cells. To test whether conjugate formation of NK cells is dependent on an intact cytoskeleton in target cells, actin filaments (F-actin) were disrupted in 721.221 cells with different concentrations of LatA (0.5, 3, and 20 μM) prior to washing and mixing those cells with NK cells in a conjugation assay. Conjugate formation of freshly isolated, primary NK cells was inhibited by LatA treatment of 721.221 cells (Fig. 1A, left panel). Moreover, conjugate formation of IL-2–activated NK cells with 721.221 cells was also prevented by pretreatment of target cells with LatA (Supplemental Fig. 1). These results suggest that either an intact actin cytoskeleton or actin cytoskeleton remodeling is required for proper conjugate formation.

To distinguish between these two possibilities, F-actin filaments in target cells were stabilized by treatment with Jasp (Fig. 1A, right panel). Jasp stabilizes existing F-actin but prevents actin cytoskeleton remodeling. Because Jasp had no effect on conjugate formation with NK cells (Fig. 1A, right panel), cytoskeleton dynamics in target cells do not seem to be important for conjugate formation of NK cells. In addition, disruption of microtubules in target cells with Nocodazole had no effect on conjugate formation with resting NK cells (data not shown). We conclude that an intact cytoskeleton, but not cytoskeleton dynamics, in target cells is important for NK cell conjugate formation.

To test whether disruption of F-actin in target cells was preventing conjugate formation specifically, or whether it had more global inhibitory effects, the ability of NK cells to degranulate in response to target cells was evaluated. Whereas conjugate formation of NK cells with ICAM-expressing target cells is strictly dependent on LFA-1 (Supplemental Fig. 2) (39), degranulation is LFA-1 independent (13, 14). 721.221 cells treated with lower doses of LatA induced about twice as much degranulation by NK cells as untreated 721.221 cells (Fig. 1B, left panel). Our laboratory has observed that, under certain conditions, engagement of LFA-1 by ICAM-1 reduces the amount and delays the onset of degranulation induced by activation receptors (10, 14). The enhanced degranulation seen in this study after latrunculin treatment of target cells may be due to a reversal of LFA-1–mediated inhibition of degranulation. In agreement with this hypothesis, blocking of LFA-1 with an Ab resulted in a 2-fold increase of degranulation by NK cells (Supplemental Fig. 3). At the highest dose (20 μM), however, LatA blocked degranulation by NK cells (Fig. 1B, left panel). Stabilization of F-actin in 721.221 cells by Jasp (Fig. 1B, right panel), or disruption of microtubules with Nocodazole (data not shown), had no effect on degranulation by NK cells. We conclude that LFA-1–dependent conjugate formation, but not LFA-1–independent degranulation, requires an intact cytoskeleton on target cells.

The sensitivity of conjugate formation but not degranulation of NK cells to the disruption of F-actin in target cells suggested that a functional LFA-1 interaction with ICAM is dependent on intact F-actin in target cells. To test this possibility, we evaluated the importance of an intact cytoskeleton in target cells for another LFA-1–dependent function in NK cells—granule polarization toward target cells (14, 15). Pretreatment of 721.221 cells with LatA resulted in diminished granule polarization in NK cells (Fig. 2). As expected, because of inhibition of tight conjugate formation, fewer NK cells were found in conjugates with 721.221 cells in the presence of LatA. However, the reduction of polarization was not a consequence of reduced conjugate formation as polarization was scored only in NK cells that had formed tight contact with target cells. Therefore, in addition to the LFA-1–dependent conjugate formation, another LFA-1–dependent process, polarization of cytolytic granules, is dependent on an intact cytoskeleton in target cells.

How could disruption of F-actin in target cells result in defective ICAM interactions with LFA-1 on NK cells? As ICAM is linked to the actin cytoskeleton by direct binding to α-actinin (α-actinin-1 and α-actinin-4) (40–42) and to the ezrin/radixi/moesin protein ezrin (31, 43, 44), we tested whether treatment of 721.221 cells with LatA had an effect on ICAM-1 mobility at the plasma membrane by FRAP (Fig. 3, Supplemental Videos 1, 2). ICAM-1 molecules on 721.221 cells were stained using a PE-conjugated anti–ICAM-1 Ab, and a region of the cell surface was bleached using a 488-nm laser (Fig. 3A, white arrow). The recovery of the bleached region of untreated (Fig. 3A, upper panel; 3B, ●, Supplemental Video 1) and LatA-treated (Fig. 3A, lower panel, 3B, ○, Supplemental Video 2) 721.221 cells was captured during 5 min. Photobleaching as a result of image acquisition was monitored in untreated and LatA-treated 721.221 cells to normalize recovery curves (Fig. 3B, ▪ and □, respectively). On untreated 721.221 cells, no recovery of the bleached region was observed (Fig. 3A, upper panel, 3B, ●, Supplemental Video 1), suggesting that ICAM-1 is immobilized on 721.221 cells. However, after disruption of F-actin with 20 μM LatA, recovery of the bleached region (Fig. 3A, lower panel, 3B, ○, Supplemental Video 2) occurred during 90 s, indicating a release of ICAM-1 from the cytoskeleton. Moreover, ICAM-1, which was polarized to one side of untreated 721.221 cells (Fig. 3A, upper panel, Supplemental Video 1), was evenly distributed after disruption of F-actin with LatA (Fig. 3A, lower panel, Supplemental Video 2).

To test whether polarization of ICAM-1 to one side of target cells, rather than or in addition to ICAM-1 tethering, was important for LFA-1–dependent adhesion of NK cells, we examined the cell line K562, which does not exhibit polarity of ICAM-1 at the cell surface (Fig. 4A). Treatment of K562 cells with LatA-released ICAM-1 form the cytoskeleton and increased its mobility (Fig. 4B). In accordance with the 721.221 data, the release of ICAM-1 from the cytoskeleton of K562 target cells prevented conjugate formation in freshly isolated primary NK cells (Fig. 4C) and polarization of cytolytic granules (data not shown), indicating that ICAM-1 tethering rather than polarization is important for LFA-1–dependent signaling in NK cells. The LFA-1–independent degranulation of NK cells was not inhibited by pretreatment of K562 target cells with different concentrations of LatA (Fig. 4D) excluding more global inhibitory effects of the drug.

We next used TIRF microscopy to visualize the distribution and mobility of ICAM molecules in the plasma membrane of target cells and to understand how the actin cytoskeleton affects ICAM mobility by treating target cells with either LatA or Jasp to disrupt or stabilize the actin cytoskeleton, respectively. TIRF microscopy is a spatially limited, high-contrast technique that eliminates interference from bulk fluorescence that may be present within cells to allow for the detection of fluorophores proximal to and within the plasma membrane of cells adhered to glass coverslips (45). Endogenous, cell surface ICAM-1 or ICAM-2 on 721.221 cells were fluorescently labeled with PE-conjugated anti-ICAM Abs. Although individual ICAM proteins were labeled with a single PE-fluorophore (see 1Materials and Methods), photobleaching characteristics (the presence of multiple-step bleaching events over long track lengths; data not shown) of fluorescent PE-labeled particles suggest that ICAM proteins were mostly observed as clusters and not single molecules (data not shown).

The lateral movement of labeled ICAM-1 (Fig. 5A, upper panel, 5B, left panel, Supplemental Videos 3–5) and ICAM-2 (Fig. 5A, lower panel, 5B, right panel) particles recorded by TIRF microscopy was automatically tracked using an algorithm developed for MatLab software (33), which was further modified to refine particle positioning with a two-dimensional Gaussian fit (34). Short-range mean square displacements were determined from positional coordinates of particles tracked for five frames [>160 ms (34)] and were linearly dependent on time under all conditions measured, consistent with a simple diffusion model. Mean square displacement versus time plots clearly showed that diffusion increases, as indicated by the slope of the line, with LatA concentration (Supplemental Fig. 4). Short-range diffusion coefficients were then determined for thousands of particles in multiple cells and graphed as a cumulative distribution function (CDF) to represent the frequency of diffusion coefficients for the entire population of tracked particles (Fig. 5B) (34). Consistent with our FRAP data above, disrupting the actin cytoskeleton with LatA incurred a dose-dependent shift toward the mobile population for both ICAM-1 and ICAM-2 particles (Fig. 5B, Supplemental Videos 3, 4). The median diffusion coefficient for ICAM-1 and ICAM-2 in untreated control cells was calculated to be 0.013 and 0.030 μm²/s, respectively, and after the maximum dose of LatA treatment, mobility increased ∼10- and 5-fold, respectively (0.132 μm²/s for ICAM-1 and 0.146 μm²/s for ICAM-2) (Fig. 5B). In accordance with these results, binding of NK cells to 721.221 target cells was inhibited by disruption of F-actin filaments in 721.221 cells (Fig. 1A, left panel). Although LatA reduced the intensity of the brightest ICAM-1 clusters, it did not appreciably alter the median intensity of either ICAM-1 or ICAM-2 particles (Fig. 5C), suggesting that ICAM cluster size on target cells does not affect NK cell response.

Both our FRAP and TIRF microscopy results indicate that ICAM-1 and ICAM-2 are tethered to the actin cytoskeleton, which, in turn, slows their lateral mobility in the plasma membrane; if so, then one would predict that ICAM mobility would not increase, and may be reduced even further, after treatment with the actin-stabilizing drug Jasp. Indeed, the mobility of ICAM-1 (Fig. 5D, left panel, Supplemental Video 5) and ICAM-2 (Fig. 5D, right panel) was slightly reduced with Jasp and stabilization of F-actin in target cells had little effect on LFA-1–dependent signaling (Fig. 1A, right panel). In addition, disruption of the microtubule network with Nocodazole did not affect the mobility of ICAM-1 (data not shown).

Our results so far indicate that a higher mobility of ICAM in the plasma membrane hinders its interaction with LFA-1 on NK cells. However, the contribution of other changes induced by LatA in target cells that could impact on LFA-1–ICAM interactions cannot be ruled out. We therefore carried out gain-of-function experiments using a cellular system in which ICAM mobility could be reversed. A change in the distribution of ICAM-2 on the mouse BW5147 thymoma cell line from evenly distributed to polarized after transient expression of human ezrin has been reported (31). We generated BW5147 cells that stably express human ezrin tagged with GFP (Fig. 6A) and monitored the mobility of ICAM-2 by TIRF microcopy (Fig. 6B, 6C, Supplemental Videos 6, 7). As shown in Fig. 6B and Supplemental Video 6, ICAM-2 was evenly distributed and mobile on BW5147 cells. Expression of GFP-tagged ezrin (Fig. 6A, 150-kDa band) in BW5147 cells decreased the mobility of ICAM-2 (Fig. 6C, Supplemental Videos 6, 7) from 0.053 to 0.027 μm²/s. Moreover, ICAM-2 became polarized (Fig. 6B) and colocalized with the distribution of ezrin-GFP (data not shown). However, clustering of ICAM-2 molecules did not change (Fig. 6D).

To test whether the reduced mobility of ICAM-2 as a result of ezrin expression had an impact on functional interaction with LFA-1 on NK cells, conjugate assays were performed. Conjugate formation between human primary IL-2–activated NK cells and BW5147 cells was increased (Fig. 7A) when ICAM-2 mobility was reduced through ezrin expression (Fig. 6C). Moreover, expression of human ezrin in BW5147 target cells increased polarization of cytolytic granules in IL-2–activated NK cells (Fig. 7B). We conclude that tethering of ICAM-2 to the cytoskeleton of BW5147 cells by expression of human ezrin restores functional interaction with LFA-1.

We next tested the interaction of LFA-1 on NK cells with ICAM-1 in the absence of other receptor–ligand interactions and compared interactions with mobile versus immobile ICAM-1. To visualize binding of NK cells to mobile ICAM-1, a histidine-tagged form of ICAM-1 was inserted into artificial planar lipid bilayers at the physiological concentration of 250 molecules/μm². Under these conditions, ICAM-1 exhibits high lateral mobility (10, 46). Primary resting NK cells on ICAM-1–coated lipid bilayers displayed rapid movement and failed to stop (Fig. 8A, Supplemental Video 8). A large fraction of NK cells displayed directed movement (Fig. 8A, Supplemental Video 8), whereas some moved randomly. The moving NK cells have a distinguished morphology consisting of a uropod and a dominant pseudopod at the leading edge of the cell (Fig. 8A, Supplemental Video 8). NK cells did not contact lipid bilayers in the absence of ICAM-1 (data not shown). At concentrations of 200 and 500 ICAM-1 molecules/μm², NK cells showed a similar phenotype than at 250 molecules/μm². The inability of NK cells to stop over lipid bilayers carrying ICAM-1 is not a general feature of their interaction with diffusible ligands, because the CD16 ligand IgG1 Fc inserted into lipid bilayers caused NK cell arrest and accumulation of IgG1 Fc into tight clusters (10). To monitor interaction with immobile ICAM-1, an ICAM-1-Fc fusion protein, in which the CD16 binding site had been mutated, was attached to coverslips. NK cells did not contact the coverslips in the absence of ICAM-1 or coverslips that had been coated with ICAM-1 at a concentration of 10 and 100 ng/ml (data not shown). However, ICAM-1 coating at concentrations of 1 and 10 μg/ml resulted in NK cells that were either attached and spread on the coverslips (Fig. 8B, left panel, Supplemental Video 9) or attached and spread NK cells that moved a little (Fig. 8B, right panel, Supplemental Video 10). In contrast to the persistent movement over ICAM-1–coated lipid bilayers, movement of NK cells over ICAM-1–coated coverslips was nondirectional and slower (Fig. 8B, Supplemental Videos 9, 10). A recent study reported that the NK cell line NKL moved over plate-bound ICAM-1-Fc (47). The difference may be due to the cells, because in our hands, NKL cells do not behave like primary NK cells when placed over lipid bilayers carrying ligands of NK cell receptors (D. Liu, unpublished observations). Taken together, our results indicate that diffusible ICAM-1 interacts with receptors on NK cells but does not promote the arrest of cell movement, whereas immobile ICAM-1 promotes stable binding of primary NK cells.

Lateral mobility and clustering of transmembrane receptors and their segregation within membrane domains are often essential for proper signaling (26–28). Less is known about the importance of receptor ligand distribution at the surface of opposing cells (48). We addressed the question of how the distribution or mobility of ligands affect receptor signaling in the context of β₂ integrin LFA-1 in NK cells and its functional interaction with ligands ICAM-1 and ICAM-2 on target cells. Earlier work established that the binding of LFA-1 on NK cells to ICAM-1 on target cells was sufficient for the formation of tight conjugates and polarization of cytolytic granules toward the NK-target cell interface (4, 12, 14, 15, 39). Because LFA-1 signaling for adhesion and granule polarization in human NK cells is uncoupled from signaling by other activation receptors (13, 14), it was possible to focus on NK cell functions controlled uniquely by LFA-1.

Two techniques, FRAP and TIRF microscopy, were used to visualize and quantify the distribution and movement of ICAM-1 and ICAM-2 on target cells in which the cytoskeleton was disrupted by LatA. The two-dimensional mobility of ICAM-1 and ICAM-2 at the surface of target cells was greatly enhanced after depolymerization of F-actin and correlated with reduced LFA-1–dependent NK cell responses. ICAM mobility, rather than clustering or polarization, was the key parameter for adhesion and granule polarization induced by LFA-1. LatA had minimal effect on the intensity of ICAM-1 and ICAM-2 clusters. Polarization of ICAM to one end of the cell, such as the one induced by ezrin, is not necessary for functional LFA-1–dependent responses because the target cell K562, which is highly sensitive to lysis by NK cells, does not display polarized ICAM-1 and stimulates strong LFA-1–dependent responses, which were lost upon release of ICAM-1 from cytoskeletal constraints.

To address the possibility that the reduction in NK cell adhesion and granule polarization after LatA treatment of target cells was due to a change other than ICAM distribution, gain-of-function experiments were carried out with the NK-sensitive mouse thymoma cell line BW5147, which expresses ICAM-2. Transfection of human ezrin into BW5147 causes a redistribution of ICAM-2 into a uropod-like cell extension and an increase in sensitivity to killing by NK cells (31). As reported in this paper, the lateral mobility of ICAM-2, but not the intensity of ICAM-2 clusters, was reduced by transfection of ezrin, which correlated with increased NK cell adhesion and granule polarization. Although ICAM clustering and polarization to one end of target cells may well contribute to NK cell activation, our data show that tethering and reduced mobility of ICAM are required for proper signaling by LFA-1.

To test the interaction of NK cells with either mobile or immobile ICAM-1 in the complete absence of other receptor–ligand interactions, purified rICAM-1 was inserted into artificial planar lipid bilayers, on which it is freely diffusible, or attached to glass. Primary, resting NK cells readily bound and formed tight contact with ICAM-1 on the solid support but moved continuously on lipid bilayers carrying diffusible ICAM-1. Therefore, ICAM-1 tethering is required for primary NK cells to form stable contacts, such as those occurring during NK-target cell interactions.

Adhesion of NK cells to target cells is accompanied by the formation of a cytotoxic immunological synapse, in which ICAM-1 on target cells segregates into a ring that surrounds a central region (10, 49, 50). At inhibitory NK cell immunological synapses, LFA-1 on NK cells and ICAM-1 on target cells are excluded from a central region where inhibitory receptors accumulate (8, 51). These results suggest that ICAM-1 acquires mobility during synapse formation. Engagement of ICAM-1 initiates signaling, which leads to recruitment of lipid rafts and MHC molecules (48). Whether or not ICAM-1 signaling is required for its peripheral distribution at NK cell cytotoxic immunological synapses remains to be tested.

Cellular signaling in response to mechanical forces exerted through cell surface receptors and the cytoskeleton, referred to as “mechanotransduction,” has been gaining recognition (52–54). Cells sense force at points of attachment, such as β₁ integrin binding to fibronectin (55). Force contributes to conformational and affinity changes in integrins (56, 57). As force sensing by LFA-1 would depend on tethering of its ligands, it is possible that a loss of mechanotransduction as a result of release of ICAM from the cytoskeleton in target cells is at the base of adhesion and granule polarization defects in NK cells.

We have shown that low mobility of ICAM on target cells is important for conjugate formation and adhesion of NK cells, suggesting that changes in the cytoskeleton could render target cells less sensitive to killing by NK cells. Changes in ICAM-1 mobility, such as those occurring during cell senescence (58), could have a direct impact on the sensitivity of cells to killing by NK cells. The importance of the cytoskeleton in cells that activate lymphocytes was illustrated recently by the effect of Schistosoma mansoni T2 RNase on dendritic cells, in which cytoskeletal changes were accompanied by deficient interaction with CD4⁺ T cells (59).

We thank J.-J. Hao and S. Shaw for the hEzrin-EGFP construct, T. Kamala for the BW5147 cells, M. Peterson for mutagenesis of the Fc fusion vector, M. March, P. Sun, and S. Rajagopalan for advice and comments, and P. Tolar for the MatLab code and technical discussions.

Disclosures The authors have no financial conflicts of interest.