MHC I Expression Regulates Co-clustering and Mobility of Interleukin-2 and -15 Receptors in T Cells

Abstract

MHC glycoproteins form supramolecular clusters with interleukin-2 and -15 receptors in lipid rafts of T cells. The role of highly expressed MHC I in maintaining these clusters is unknown. We knocked down MHC I in FT7.10 human T cells, and studied protein clustering at two hierarchic levels: molecular aggregations and mobility by Förster resonance energy transfer and fluorescence correlation spectroscopy; and segregation into larger domains or superclusters by superresolution stimulated emission depletion microscopy. Fluorescence correlation spectroscopy-based molecular brightness analysis revealed that the studied molecules diffused as tight aggregates of several proteins of a kind. Knockdown reduced the number of MHC I containing molecular aggregates and their average MHC I content, and decreased the heteroassociation of MHC I with IL-2Rα/IL-15Rα. The mobility of not only MHC I but also that of IL-2Rα/IL-15Rα increased, corroborating the general size decrease of tight aggregates. A multifaceted analysis of stimulated emission depletion images revealed that the diameter of MHC I superclusters diminished from 400–600 to 200–300 nm, whereas those of IL-2Rα/IL-15Rα hardly changed. MHC I and IL-2Rα/IL-15Rα colocalized with GM₁ ganglioside-rich lipid rafts, but MHC I clusters retracted to smaller subsets of GM₁- and IL-2Rα/IL-15Rα-rich areas upon knockdown. Our results prove that changes in expression level may significantly alter the organization and mobility of interacting membrane proteins.

Introduction

Membrane receptors perform their roles in functional complexes, where some elements participate in signaling directly, others act as regulators, and the function of certain other elements is as-yet undefined. Earlier we have shown that MHC I and II were associated with IL-2R and IL-15R, and these proteins formed supramolecular clusters in lipid rafts (membrane domains rich in cholesterol and glycosphingolipids) of T cells (1, 2). However, the role of the highly expressed MHC glycoproteins in these complexes has not been elucidated. In this article we examined how the assembly, interactions, and mobility of IL-2 and -15 receptors are influenced by MHC I in T lymphoma cells.

MHC I is a heterodimer consisting of a transmembrane heavy chain and an extracellular light chain, i.e., the beta-2 microglobulin (β₂m), which has no transmembrane domain. MHC I is expressed in most nucleated cells, where it presents self or viral antigens to T cell receptors expressed by CD8⁺ T cells, initiating T cell activation upon antigen recognition (3). MHC I can form oligomers as indicated by Förster resonance energy transfer (FRET) and electron microscopy data from immunogold-labeled B and T cells (4, 5, 6, 7). In addition to IL-2Rα (8, 9) and IL-15Rα (1), it can also associate with a series of different membrane proteins including receptors in various cell types: IL-9Rα on T lymphoma cells (10), MHC II on B lymphoma cells (11), EGF receptors in human fibroblasts and A431 cells (12), ICAM-1 on T and B lymphoma, uveal melanoma, and colon carcinoma cells (4, 8, 9, 13, 14), insulin receptors on B lymphoblasts (15, 16) and fibroblasts (17), tetraspan molecules on B lymphoblasts (18), and transferrin receptors on T lymphoma cells (13, 19). Through its interactions with receptors, MHC I may interfere with signaling processes, such as in the case of the insulin receptor (20, 21). MHC I can mediate transmembrane signaling (22), and antibody binding to the light chain may induce apoptosis via the PI-3 kinase pathway (23). Thus, over the decades a multifaceted view of MHC I has been emerging—one that goes beyond its classical antigen-presenting function (see reviews, e.g., in Arosa et al. (24) and Bowness et al. (25)).

The cytokines IL-2 and -15 play important roles in the activation, proliferation, death, and survival of immune cells (26, 27). Their receptors are built of three subunits, two of which, the β- and γ_c-chains, are shared by the two cytokines. We have shown by FRET that IL-2 (28) and IL-15 receptors were preassembled on lymphoma/leukemia cells (1), and that they could form a heterotetrameric complex composed of IL-2Rα, IL-15Rα, and the β- and γ_c-chains (1). The β- and γ_c-subunits are the starting points of several signaling processes including the JAK/STAT pathway (27, 29, 30, 31). Because they share the same signaling receptor chains, IL-2 and IL-15 also have several functions in common: they promote the proliferation of CD4⁺ T lymphocytes and augment the cytotoxicity of T and NK cells in vitro (26, 32). However, they may also play opposing roles in T cells: IL-2 induces apoptosis in activation-induced cell death, whereas IL-15 preferentially promotes the survival of CD8⁺ memory T cells and inhibits apoptosis (32, 33). This divergence of cell fates is not fully understood. It may be caused by different interactions of the α-subunits with other signaling molecules or membrane proteins. Therefore, clustering and interactions of IL-2 and -15 receptors with surrounding membrane components is an important issue to investigate.

The first pieces of evidence for the close molecular proximity of MHC I and IL-2Rα were obtained from FRET experiments on a T lymphoma cell line (34) and later on activated peripheral T cells (35). Fluorescence recovery after photobleaching showed the slowing-down of IL-2Rα diffusion after ligating MHC I with an antibody (36). Transmission electron microscopy analysis indicated that the two molecules were partially co-clustered in Kit 225 K6 T lymphoma cells (37). Our FRET studies on Kit 225 FT7.10 T lymphoma cells overexpressing IL-15Rα revealed the close molecular proximity of IL-2Rα/IL15Rα subunits and MHC I/MHC II molecules in lipid rafts (1). Using fluorescence cross-correlation spectroscopy, we have also proven that IL-2Rα, IL-15Rα, and MHC I stably codiffused in the plasma membrane (1).

To unravel the possible role of MHC I in clustering of IL-2/IL-15 receptors, we knocked down MHC I expression at the cell surface. We used siRNA against β₂m, because the transport of the MHC I heavy chain to the cell surface is impaired in the absence of the light chain (38). FRET was applied to investigate changes of protein-protein interactions at the nanometer scale, fluorescence correlation spectroscopy (FCS) was used to assess protein mobility and homoaggregation state, and confocal microscopy was applied to analyze compartmentation in lipid rafts. The size of IL-2Rα, IL-15Rα, and MHC I protein clusters and the extent of co-clustering at 40-nm resolution were determined by combining stimulated emission depletion (STED) superresolution microscopy with image analysis methods.

Materials and Methods

Cells

FT7.10 is a CD4⁺, IL-2-dependent human adult T lymphoma cell line derived from Kit 225 cells (39). It expresses all three subunits of IL-2R constitutively, and is stably transfected with IL-15Rα carrying an N-terminal FLAG-tag. Cell culturing and fluorescence labeling of cells with monoclonal antibodies is described in the Supporting Material.

RNA interference

While the heavy chain of MHC I has A, B, and C alleles with many sequence variants each, the sequence of the light chain (β₂m) is conserved. Association with β₂m is indispensable for MHC I to reach the cell surface (40), thus, we knocked down β₂m. Absolute numbers of MHC I, MHC II, IL-2Rα, and IL-15Rα molecules per cell were estimated by flow cytometry using QIFIKIT beads (Dako North America, Carpinteria, CA; Figs. 1 and S1). For details on these procedures, see the Supporting Material.

Figure 1.

Protein expression in control and MHC I knockdown cells determined by flow cytometry. Quantification of IL-2Rα molecules per cell was performed by using QIFIKIT beads (Dako North America). Expression of MHC I, MHC II, and IL-15Rα proteins was assessed by comparing the mean fluorescence intensities of directly labeled epitopes (targeted by Alexa 546-tagged W6/32, L243, and anti-FLAG mAbs) to that of directly labeled IL-2Rα (Alexa 546-anti-Tac mAb). Fluorescence intensities were normalized by the dye/protein antibody labeling ratios determined by absorption photometry. Error bars are SD values of averages of triplicate samples (n ∼ 10⁴ cells per sample).

FCS

To measure diffusion properties of membrane proteins, we used FCS (41, 42). The FCS microscope is described in Beck et al. (43) and Mocsár et al. (44). Autocorrelation curves were fitted to a model assuming lateral diffusion of a single component and triplet state formation. The lateral e⁻² radius of the detection volume was calibrated with a solution of Alexa 488 having a known diffusion coefficient (45). Correlation curves were fitted by using the QuickFit 3.0 software (46). Molecular brightness (photon count rate per molecule) was calculated as the ratio of the average, background-corrected fluorescence count rate, F, and the mean (background-corrected) number of molecules, N, in the detection volume. This parameter gives an estimate of the number of fluorescent molecules in a jointly moving complex if normalized with the F/N of freely diffusing Fab fragments. The expression levels of MHC I, IL-2Rα, and IL-15Rα in the subpopulation sampled by FCS were similar to those measured by flow cytometry (Fig. S1). Bleached fractions of proteins were assessed from count traces of FCS measurements. Further details on FCS are given in the Supporting Material.

Confocal microscopic colocalization

Colocalization of membrane proteins with lipid raft markers at the few-hundred-nm scale was determined from confocal images of doubly labeled cells by confocal microscopy. Colocalization was assessed by calculating the Pearson’s correlation coefficients between the pixel intensities of the two detection channels as described in Vereb et al. (37). Further details are given in the Supporting Material.

Determination of FRET efficiency by flow cytometry and confocal microscopy

Molecular proximity between proteins was detected by Förster resonance energy transfer. Cells were doubly labeled with mAbs tagged with donor (Alexa 546) or acceptor (Alexa 647) dyes. Flow cytometric FRET measurements were carried out on a FACSAria III instrument (Becton Dickinson, Franklin Lakes, NJ) and analyzed as described in Sebestyén et al. (47) and Szentesi et al. (48). Acceptor photobleaching FRET measurements were carried out and analyzed on a pixel-by-pixel basis as described earlier in Vámosi et al. (1) and Nizsaloczki et al. (10). For details, see the Supporting Material.

STED microscopy

Two-color superresolution image acquisition was carried out using a custom-built STED microscope described in Wildanger et al. (49), Bückers et al. (50), and Rönnlund et al. (51). Before STED imaging, regular confocal images were also recorded by the microscope. The resolution of the system in STED mode was ∼40 nm for both colors (52). The distance between two neighboring pixels was 20 nm for STED and 50 nm for confocal images; the image size was 15 × 15 μm². For excitation 570 ± 5 and 647 nm, STED beams 710 ± 10- and 750 ± 10-nm wavelengths were used. Fluorescence emissions were collected at 615 ± 15 and 675 ± 15 nm for the Alexa-594 and ATTO-647N dyes.

Estimating the size of protein clusters from analysis of STED images

The sizes of IL-2/15R and MHC I protein clusters in control and MHC I knockdown (MHC I KD) cells were determined by using three different methods.

Full-width at half-maximum method

In the first procedure, the full width at half-maximum (FWHM) values of the intensity peaks detected on the cell surface were measured. Before image analysis, all images were deconvolved using 20 iterations of the Richardson-Lucy algorithm (51) and assuming a 40-nm wide Lorentzian-shaped point spread function (PSF). The size of each peak was calculated as the mean FWHM value measured over 20 different angles.

Spatial correlation function method

In the second procedure, the spatial correlation curves of superresolution images of MHC I, IL-2Rα, and IL-15Rα were calculated. Empirical curves were fitted with a model function, which allowed estimation of the average cluster diameter (53, 54). Spatial correlation can be defined as follows:

$$G\left(r\right)=\langle I\left(r^{\prime }\right)\times I\left(r^{\prime }+r\right)\rangle /{\langle I\left(r^{\prime }\right)\rangle }^2-1,$$ (1)

where $I\left(r\right)$ is the fluorescence intensity at position r, and brackets $\langle \rangle$ denote averaging over positions $r^{\prime }$. Then the spatial correlation function was calculated with the following formula:

$$G\left(r\right)=N\left(r\right)\times \sum _{\begin{array}{l}i,j\in \text{ROI}\\ k,l\in \text{ROI}\end{array}}I\left(i,j\right)\times I\left(k,l\right)/{\left(\sum _{i,j\in \text{ROI}}I\left(i,j\right)\right)}^2-1,$$ (2)

where $I\left(i,j\right)$ is the intensity of the (i,j)th pixel in the region of interest, r is the distance between two pixels in the calculation of correlation, and N(r) is the number of pixels that are r pixels apart. If multiple molecular clusters are present, the total correlation function can be broken down into three terms:

$$G\left(r\right)=G^{\text{PSF}}\left(r\right)+G^{\text{within}\text{clusters}}\left(r\right)+G^{\text{between}\text{clusters}}\left(r\right).$$ (3)

The first term (G^PSF) describes the shape of the correlation function of the PSF, whereas the other two terms $\left(G^{\text{within}\text{clusters}}\text{and}{G}^{\text{between}\text{clusters}}\right)$ describe the pair correlation of particles within and between clusters. To fit empirical correlation functions and estimate cluster sizes, we used the sum of three Gaussians:

$$G\left(r\right)={\rho }_1\times e^{-r^2/w^2}+{\rho }_2\times e^{-r^2/d^2}+{\rho }_3\times e^{-r^2/g^2},$$ (4)

where w is the lateral radius of the effective PSF, parameter d was interpreted as a quantity proportional to the average radius of clusters, and g is the average separation distance between clusters. (Coefficient ρ₁ is proportional to the number of fluorophores, ρ₂ to the number of particle pairs within a cluster, and ρ₃ to the number of particle pairs from different clusters.) The formula was tested by computer simulations.

Cluster segmentation method

The third procedure, the cluster segmentation method (55), is based on the classification of pixels of STED images of cell surface protein distributions. Regions of interest were constructed for each cell, then images were intensity-thresholded and pixels above the threshold were selected. The threshold was set halfway between the background and the average of the half-maximal intensities of fluorescent peaks for each channel. A pixel cluster is defined as a contiguous region of neighboring pixels above the threshold. The number of pixels (characterizing cluster size) and the total intensity within the clusters (measuring protein content) were collected for all clusters and both channels. Applying this segmentation algorithm to dual-color images allowed the identification of either common or unique clusters of IL-2Rα or IL-15Rα and MHC I. Data were presented as probability density histograms. For statistical comparisons, the Kolmogorov-Smirnov test was used. To characterize the distribution of protein content within clusters, the total intensities of the clusters were calculated and plotted as a function of cluster size.

Calculation of molecular fractions

Fractions of IL-2Rα and IL-15Rα in common clusters with MHC I were calculated by dividing the total intensity from receptor subunits in common clusters with MHC I by the total intensity of receptor subunits in the whole region of interest, e.g.:

$$\frac{\left[IL-2R\alpha \right]\cap \left[MHCI\right]}{\left[IL-2R\alpha \right]}=\frac{\sum _{\begin{array}{c}i,j\in \text{ROI}\\ \text{red}>\text{threshold}\\ \text{green}>\text{threshold}\end{array}}{I}^{\text{red},IL-2R\alpha }\left(i,j\right)}{\sum _{\begin{array}{c}i,j\in \text{ROI}\\ \text{red}>\text{threshold}\end{array}}{I}^{\text{red},IL-2R\alpha }\left(i,j\right)},$$ (5)

where $I^{\text{red},IL-2R}\left(i,j\right)$ is the intensity of a pixel of IL-2Rα measured in the red channel. Summation in the numerator includes double positive pixels, where both the IL-2Rα (red) and MHC I (green) signals are above the threshold, whereas the denominator includes all pixels where the IL-2Rα signal is above the threshold. Molecular colocalization was also assessed by calculating the Pearson’s correlation coefficients between the pixel intensities of the two detection channels. Associated fractions and Pearson’s coefficients were plotted as bar diagrams. Image analysis methods and fitting routines were implemented in custom-written MATLAB scripts (The MathWorks, Natick, MA).

Results

Expression of IL-2/15R and MHC glycoproteins before and after MHC I knockdown

IL-2R, IL-15R, MHC I, and MHC II form superclusters in lipid rafts of Kit 225 FT7.10 T lymphoma cells. To define the role of the most abundant element, MHC I, in the organization of these clusters, we knocked down MHC I expression by anti-β₂m siRNA. Fig. 1 shows the number of antibody binding sites per cell for each molecular species before and after KD; histograms of expression distributions are presented in Fig. S1. Because of the high abundance of MHC I, it was essential to maximize silencing efficiency. MHC I expression at the cell surface decreased with increasing siRNA concentration, reaching a maximal effect at 100 μg/mL where expression was reduced to 5–10%. KD was specific: expression levels of IL-2Rα, IL-15Rα, or MHC II did not change. Irrelevant siRNA (anti-GFP or anti-erbB1) used as control caused no variation in the expression of MHC I, IL-2Rα, or IL-15Rα.

Increased mobility of IL-2Rα, IL-15Rα, and MHC I upon MHC I KD

We were interested how the expression level of MHC I at the cell surface affects the dynamic properties and organization of the elements of the superclusters, including those of MHC I itself. We assessed the location of cells measured by FCS in the expression range of the total cell population to make sure that the sample represents the whole population well. Fig. S1 shows the histograms of the expression distributions of the total cell population determined by flow cytometry, and those of the FCS-sampled population as overlaid box-and-whiskers plots (converted to number of receptors in the confocal spot). The good overlap between the graphs proves that cells picked for FCS measurements had similar protein expression distributions as the original population measured by flow cytometry.

If a protein is in a jointly moving cluster/aggregate, its mobility in the membrane is inversely related to the size of the aggregate. To assess changes of mobility, the lateral diffusion coefficients of IL-2Rα, IL-15Rα, MHC I, and MHC II were determined by FCS. Fig. 2, A and B, shows representative autocorrelation functions (ACFs); in Fig. S2, raw intensity traces and ACFs of individual runs, fits, and residuals are presented. Upon MHC I KD, ACFs shifted toward shorter diffusion times, indicating an increase of mobility (Fig. 2, A and B). Diffusion coefficients (D) increased significantly for IL-2Rα, IL-15Rα (p < 0.001), and MHC I (p < 0.01). Conversely, we detected no statistically significant change for MHC II (Fig. 2 C). We separately analyzed diffusion coefficients of MHC I and IL-2Rα residing in large patches and outside patches. The mobility of these proteins was lower inside patches than outside them (Fig. S4). The value D decreased with increasing expression density (increasing local F) of MHC I, in line with the observed increasing tendency of D upon KD (Fig. S5, A and B).

Figure 2.

FCS analysis of molecular mobility, concentration (N), aggregation (molecular brightness), and bleached fraction upon MHC I knockdown. (A and B) Normalized autocorrelation functions of MHC I and IL-2Rα. (C) Average diffusion coefficients. (D) Average particle numbers in the FCS detection volume. (E) Specific molecular brightness, F/N, normalized by the F/N values of Fab fragments in solution. (F) Median bleached fractions of molecules calculated from the intensities after 5, 20, and 100 s (after the first, fourth, and 20th runs) of the 100-s illumination period. The following Alexa 488-conjugated Fab fragments were used for labeling: W6/32 for MHC I, anti-Tac for IL-2Rα, anti-FLAG for IL-15Rα, and L243 for MHC II. Distributions were compared with Kolmogorov-Smirnov tests, ^∗∗∗p < 0.001, ^∗∗p < 0.01, ^∗p < 0.1; boxes represent 25 and 75%; whiskers, 10 and 90%. n = 25–35 cells per sample.

Decreased molecular brightness of MHC I indicates reduction of protein aggregate size

The mean number of molecules in the detection volume (N) is a measure of the local concentration of diffusing particles (single molecules or aggregates), whereas the molecular brightness (the fluorescence intensity per particle, F/N, measured in photon counts per second per particle) is proportional to the number of labeled proteins forming an aggregate. To assess changes in the number and size of aggregates, we measured these parameters before and after MHC I KD. For MHC I, the average value of N decreased 10-fold (Fig. 2 D, p < 0.001). We found a strong positive correlation between the expression density of MHC I and the number of aggregates (Fig. S5, C and D), corroborating the result of KD. In contrast, the N values for IL-2Rα, IL-15Rα, and MHC II were not largely affected by MHC I KD; small variations of N might be due to variations in the expression levels of the randomly selected cells.

To calibrate F/N as a measure of the degree homoaggregation (number of proteins of interest in the aggregate), we first measured the brightness of the fluorescent Fab fragments freely diffusing in buffer solution, which yields the brightness of a single Fab. Next, we measured F/N values of Fab-s bound to their target proteins in the cell membrane, and normalized them by the F/N values measured in solution (Fig. 2 E). On control cells, the median normalized F/N values were ∼2.3 for MHC I, 1.3 for IL-2Rα, 1.0 for IL-15Rα, and 1.3 for MHC II, suggesting protein aggregation. Because the dye/protein labeling ratios (D/p) of the Fab fragments were <1 (see the Supporting Material), these numbers may underestimate the extent of aggregation. MHC I KD caused a 40% reduction of the median specific brightness of MHC I (p < 0.001), indicating that the number of MHC I molecules jointly diffusing in a cluster decreased significantly. The brightness of IL-2Rα, IL-15Rα, and MHC II showed only a slight decrease, which was statistically not significant. For MHC I, the dependence of the specific brightness had an increasing tendency with increasing local expression density (Fig. S5, E and F), in line with our observations upon MHC I KD.

Finally, bleached fractions of molecules were estimated from intensity traces (Fig. S2) of FCS experiments. Bleached fractions were calculated after 5, 20, and 100 s of illumination (Fig. 2 F). After 20 s of illumination, the fraction of bleached proteins was in the range of 50–70%. For MHC I, the bleached fractions decreased upon KD, corroborating the overall increase of mobility.

FRET implies partial disassembly of MHC I/IL-2Rα/IL-15Rα clusters upon MHC I KD: decrease of the extent of homo- and heteroassociations of MHC I

To dissect interactions between individual protein species at the nanometer scale, we applied FRET. Homo- and heteroassociations of IL-2R and IL-15R subunits, MHC I and II glycoproteins, were analyzed on a cell-by-cell basis using flow cytometry (Fig. 3 A), or on a pixel-by-pixel basis with the acceptor photobleaching technique using confocal microscopy (Fig. 3 B). As a negative control, the FRET efficiency between coated-pit-located transferrin receptors and lipid-raft-located GPI-anchored CD48 proteins was measured, yielding E = 3 ± 1%. As positive control, we measured FRET between the light and heavy chains of MHC I, which decreased from 52 ± 1% to 28 ± 1% upon MHC I KD. Here, the FRET process is only partially intramolecular: FRET can also occur between the light and heavy chains of distinct MHC I molecules residing next to each other in an aggregate. Thus, the decrease of FRET efficiency suggests diminishing of MHC I homoassociation. We measured homoclustering of MHC I heavy chains, which decreased significantly upon MHC I knockdown: the FRET efficiency dropped from 35 ± 1% to 11 ± 1% (Fig. 3 A). Here, MHC I was targeted using donor- and acceptor-tagged antibodies specific for the heavy chain of intact MHC I, containing both β₂m and the heavy chain. The heteroassociation of donor-labeled IL-2Rα and acceptor-labeled MHC I also decreased significantly upon MHC I KD, from 47 ± 1% to 9 ± 1%. FRET between IL-15Rα and MHC I decreased similarly, from 44 ± 1% to 7 ± 1%. This is not surprising because the acceptor/donor ratio decreased, thereby the probability that a given donor-labeled protein had an acceptor-labeled partner within FRET-range (2–10 nm), diminished. The heteroassociation of IL-15Rα with IL-2Rα, or IL-2Rβ with IL-2Rα did not change significantly, suggesting that interleukin receptor assembly was not influenced by interactions with MHC I. We also analyzed FRET efficiencies as a function of expression levels of the interacting proteins shown in Fig. S6. Cells were classified into three classes according to the expression (i.e., low, medium, high) of the donor-tagged proteins. For the heteroassociation of donor-tagged IL-15Rα or IL-2Rα with acceptor-tagged MHC I, there is a clear tendency of increasing FRET efficiency versus increasing N_A/N_D acceptor/donor molecular ratio, and a faster saturation of the graph for higher donor expressors, just as expected for association equilibria. For MHC I homoassociation, the FRET efficiency increases with an increasing expression level of MHC I as expected.

Figure 3.

FRET analysis of the effect of MHC I knockdown on molecular proximities. (A) Means of average FRET efficiencies of duplicate flow cytometric measurements (N > 10,000 cells) between IL-2/15Rα subunits and MHC I. Proteins were labeled with different pairs of donor (Alexa 546)- and acceptor (Alexa 647)-tagged antibodies (IL-2Rα, IL-15Rα, and MHC I were labeled with Alexa 546-anti-Tac, Alexa 546-7A4 24. and Alexa 647-W6/32 mAbs). Positive control: FRET between the light chain (β2m, labeled with Alexa546-L368) and heavy chain of MHC I. Negative control: transferrin receptor (TfR, located in coated pits, labeled with Alexa 546-MEM75 mAb) and GPI-anchored CD48 (enriched in lipid rafts, targeted by Alexa647-MEM102 mAb). Error bars are SD values. (B) Distribution of pixelwise FRET efficiencies between IL-2Rβ + IL-2Rα measured by the acceptor photobleaching method using confocal microscopy (n > 25 cells per sample; IL-2Rβ and IL-2Rα were labeled with Alexa 546-Mikβ3 and Alexa 647-7G7-B6). Mean FRET efficiency values were 9 and 10% in control and KD cells. Cells were fixed with formaldehyde.

Analysis of STED images shows a reduction of MHC I protein supercluster size upon KD

Our FCS and FRET data suggested that the size of small-scale protein aggregates diminished upon MHC I KD. We used superresolution STED microscopy to get a direct insight into the effect of MHC I KD on the higher-order clustering of the studied proteins. STED (Figs. 4 and S7) and confocal images (Fig. S8) of the (co)distributions of the fluorescently labeled proteins were taken in KD and control cells. The optical resolution of the STED microscope was ∼40 nm, and the pixel size was 20 nm. To estimate the size of protein superclusters, we used three different image analysis methods (Fig. 5, A1–C1).

Figure 4.

Two-color STED images of cell surface protein codistributions on FT7.10 T lymphoma cells. (A) MHC I and IL-2Rα and (B) MHC I and IL-15Rα. MHC I was labeled with Alexa 594-W6/32 mAb (green, top row), IL-2Rα and IL-15Rα with ATTO 647N-anti-Tac and ATTO 647N-anti-FLAG mAbs (red, middle row), and fixed with formaldehyde. In the overlay images (bottom row), the yellow color shows the overlapping regions of MHC I and IL-R distributions. Because of the large differences in gray values between raw intensity data of MHC knockdown and control samples, intensities were rescaled. (Yellow squares) Areas displayed in the high-magnification images next to the low-magnification image. Scale bar in the low- and high-magnification images represent 7 μm and 0.7 μm, respectively.

Figure 5.

Cluster-size distributions of MHC I, IL-2Rα, and IL-15Rα determined from STED microscopy images. The schemes in the top row illustrate the calculation of each parameter shown below. (A) Effect of MHC I knockdown (KD) on cluster size determined from Gaussian fits to intensity peaks in STED images. Pixel size is 20 nm. (A1) Direction-averaged FWHM of intensity peaks. Each peak represents a receptor cluster (see Materials and Methods); FWHM values are measures of average cluster size. (A2–A4) Histograms display the probability distributions of peak widths of MHC I, IL-2Rα, and IL-15Rα clusters in control and MHC I knockdown cells. (A5) Mean values of FWHMs. ^∗∗∗∗p < 0.0005, ^∗p < 0.05 (Kolmogorov-Smirnov test), n = 10–15 cells per sample. (B) Spatial autocorrelation analysis of protein cluster sizes. (B1) Scheme of calculating radial autocorrelation functions for radii of 1–100 pixels. (B2–B4) Average radial autocorrelation functions of MHC I, IL-2Rα, and IL-15Rα in control and MHC KD cells. Radii were measured in pixels. (B5) Estimated mean cluster sizes of MHC I, IL-2Rα, and IL-15Rα based on a triexponential fit of the correlation curves. The shortest correlation radius is related to the PSF; the medium one (shown in the figure) corresponds to the average cluster size; and the largest one probably to intercluster distances. ^∗p < 0.05 (unpaired t-test). (C) Image segmentation analysis of protein clusters. (C1) Images were intensity-thresholded and the number of pixels in contiguous clusters was determined. (C2–C4) Probability distributions of cluster areas (number of pixels in contiguous areas above threshold) in control or MHC I KD cells. (C5) Mean values of cluster areas. ^∗∗∗∗p < 0.0005 (Kolmogorov-Smirnov test). (D) Analysis of relative protein content of clusters. (D1) Total fluorescence intensities of clusters defined in (C) were calculated and sorted according to cluster size. (D2–D4) Histograms represent probability distributions of protein content versus cluster area. (D5) Most populated cluster sizes for MHC I, IL-2Rα, and IL-15Rα. ^∗∗∗∗p < 0.0005 (Kolmogorov-Smirnov test).

First, the FWHM values of intensity peaks of STED images were calculated in control and MHC I KD cells. The concept of the method is shown in Fig. 5 A1. The distributions of the FWHM of peaks are depicted in Fig. 5, A2–A4. Interestingly, the mean diameter of MHC I clusters decreased starkly after KD, from 640 ± 120 nm to 260 ± 80 nm (31 ± 6 to 13 ± 4 pixels, mean ± SE, p < 0.0005). In contrast, the average size of IL-2Rα clusters did not change significantly, and the FWHM of IL-15Rα peaks decreased only slightly from 360 ± 60 nm to 300 ± 60 nm (18 ± 3 to 15 ± 3 pixels, p < 0.05; Fig. 5 A5).

Second, we applied spatial correlation analysis of intensity distributions to assess cluster sizes (Fig. 5, B1–B5). The angle-averaged radial correlation functions of fluorescence images were calculated according to Eq. 2 (see schematic drawing in Fig. 5 B1). Fig. 5, B2–B4, shows average radial correlation curves of protein distributions in cells after MHC I KD and for controls. Correlation curves were fitted to a triple exponential model function, where the first term is assigned to the PSF, the second one to the mean diameter of clusters on the cell surface, and the third one to intercluster distances. This analysis confirmed the impact of KD on cluster sizes: the mean radius of MHC I clusters (correlation length of the second term) was reduced from 340 ± 120 nm to 200 ± 120 nm (17 ± 2 to 10 ± 2 pixels, means ± SE) after MHC I KD. The average radii of IL-2Rα and IL-15Rα clusters were in the range of 280–320 nm (14–16 pixels), and did not change significantly (Fig. 5 B5). The third component of the correlation function was well separable from the second one; its characteristic radius was in the range of 1.04–1.32 μm (52–66 pixels) for all the studied proteins.

Third, we measured the mean occupied areas of MHC I, IL-2Rα, and IL-15Rα clusters by an image segmentation method (see Fig. 5 C1 for scheme). The previous methods estimate the average diameter of clusters, whereas this procedure assesses the area of contiguous high-intensity pixels. The probability distributions of above-threshold contiguous areas are shown in Fig. 5, C2–C4. The mean area of MHC I clusters decreased significantly upon KD (p > 0.0005) from (7.2 ± 0.9) × 10⁴ nm² to (1.8 ± 0.2) × 10⁴ nm² (181 ± 23 to 44 ± 5 pixels, mean ± SE). The areas occupied by interleukin receptors did not change significantly (Fig. 5 C5). It is interesting to note that the area occupied by MHC I and IL-2Rα in control cells was approximately twice the area occupied by IL-15Rα.

The use of superresolution microscopy made it possible to investigate the decrease of MHC I cluster size; confocal imaging could not resolve changes at this length scale. Taking together the results of the three distinct analysis methods of the STED images, a significant reduction of occupied areas and diameters of MHC I clusters was observed upon MHC I KD, whereas the cluster sizes of IL-2 and -15 receptors changed only slightly. Thus, changes seen by molecular brightness analysis and FRET at the scale of small molecular aggregates were paralleled by similar changes at the scale of submicrometer-sized superclusters.

The previous analyses dealt with the spatial dimensions of large protein clusters, but not with their protein content, which also depends on the density of proteins in the clusters. To quantify the amount of receptors in clusters of different sizes, the summed fluorescence intensities of clusters of a given size were calculated. Intensities were normalized to the total intensity of all clusters. The relative protein content of MHC I, IL-2Rα, and IL-15Rα clusters versus cluster size are shown in Fig. 5, D2–D5, for control and KD cells. The most populated cluster size of MHC I decreased from (10 ± 0.4) × 10⁴ nm² to (3.2 ± 0.1) × 10⁴ nm² (2500 ± 100 to 800 ± 30 pixels, mean ± SE) after KD. The most preferred cluster sizes of IL-2Rα and IL-15Rα receptors did not change significantly (Fig. 5 D5).

Colocalized fraction of IL-2Rα/IL-15Rα with MHC I is reduced upon MHC I KD

FRET analysis of molecular interactions at the 2–10-nm scale showed that heteroassociations of MHC I with IL-2Rα and IL-15Rα diminished when the MHC I content of the cell membrane decreased. To assess changes of colocalization with a resolution of 40 nm, molecular fractions of colocalized proteins were calculated from STED images (Fig. 6 A).

Figure 6.

Colocalization analysis of MHC I with IL-2Rα and IL-15Rα in STED images. (A) Fractions of IL-2Rα and IL-15Rα molecules residing in common clusters with MHC I (and vice versa) derived from STED images of doubly labeled samples (referred to, e.g., as IL-2Rα near MHC I). (B) Pearson’s correlation coefficients of MHC I-IL-2Rα and MHC I-IL-15Rα codistributions from STED images (n = 10–15 cells per sample). Samples were compared with two-tailed t-tests, ^∗∗∗p < 0.001, ^∗∗p < 0.01.

Our first observation is that in control cells, the fractions of IL-2Rα and IL-15Rα in the vicinity of MHC I, as well as the fraction of MHC I in the vicinity of IL-2Rα, are high, as opposed to the scarcity of MHC I near IL-15Rα, which may be due to the lower expression level of IL-15Rα. Remarkably, the mutual vicinities of MHC I and interleukin receptors showed an asymmetric behavior after MHC I KD. The fraction of IL-2Rα found near MHC I in STED images was reduced from 75 ± 3% (control) to 5.4 ± 3% (KD). Similarly, the proportion of IL-15Rα near MHC I was significantly reduced, from 91 ± 4% to 20 ± 6% (Fig. 6 A). Conversely, the fraction of MHC I colocalized with IL-2Rα decreased only slightly, from 61 ± 8% to 52 ± 7%, and its fraction near IL-15Rα even increased from 12 ± 2% to 19 ± 5% upon KD (Fig. 6 A). MHC I molecules probably retracted to a smaller total area, but remained in the vicinity of interleukin receptors, suggesting that their association with IL-2Rα and IL-15Rα was stable.

Colocalization between MHC I and IL-2Rα/IL-15Rα subunits was also analyzed by calculating Pearson’s correlation coefficients (C) of their pixelwise intensity distributions. We calculated C values with three thresholding conditions: for pixels where the intensity in one, the other, or both channels was above the threshold (Fig. 6 B). The C values were significantly reduced for both pairs (MHC I + IL-2Rα as well as MHC I + IL-15Rα) upon KD of MHC I. Our results on Pearson’s correlation coefficients are in line with the findings that, after KD, the majority of interleukin receptors were situated in clusters free of MHC I. As a negative control, pixels within both images were randomly mixed (>10,000 iterations), thereby making the intensity distributions in the two channels independent of one another. Correlation coefficients were calculated for these randomized images, resulting in C values of 0.007 ± 0.0001 (SD) for STED images. We applied the same analyses to confocal images acquired from the same cells, and found similar tendencies; however, numerical values of colocalized fractions and Pearson’s coefficients were different due to the inability of confocal microscopy to resolve finer details (see Fig. S8).

To better understand our results, we carried out simulations to model the effect of KD on the value of C. In one channel the expression density was set to 0.2 particle/pixel, corresponding to the density of IL-2Rα derived from flow cytometry (Fig. S9, right column). In the other channel (left column), the particle density was reduced from 0.8 particle/pixel (our estimate for MHC I in control cells) to 0.05 particle/pixel (the value after KD), while the fraction of particles participating in co-clusters was set to 20% throughout. Decreasing the number of MHC I particles resulted in a decrease of C from 0.8 to 0.5, showing that the experimentally observed decrease of C could partially be explained by the change of the S/N upon KD, even though the extent of co-clustering was maintained.

Membrane microdomain localization of IL-2Rα and IL-15Rα subunits remains unchanged upon MHC I knockdown

Earlier studies found that IL-2Rα, IL-15Rα, and MHC I molecules were enriched in lipid rafts of Kit 225 cells (1, 13, 37). The signaling efficiency of IL-2Rα was shown to depend on the cholesterol content of the membrane and the integrity of lipid rafts (13). We were interested in whether knocking down MHC I affects the localization of IL-receptors with respect to lipid raft domains. Codistributions of IL-2Rα, IL-15Rα, or MHC I with the lipid raft marker GM₁ ganglioside were recorded by confocal microscopy (Fig. S10) followed by intensity correlation analysis. The distribution of both receptor subunits and MHC I displayed a substantial overlap with GM₁ in control (mock-transfected) and MHC I KD cells as indicated by the yellow regions in the overlay images. Correlation coefficients of the images of the receptor subunits and GM₁ did not change significantly upon gene silencing (Fig. S10 D). These results imply that the enrichment of IL-2Rα and IL-15Rα in lipid rafts does not depend on the presence or absence of MHC I. On the other hand, the C values between MHC I and GM₁ decreased from 0.4 ± 0.1 to 0.28 ± 0.08 (mean ± SD), which may be explained by our simulations (see Fig. S9).

Discussion

Membrane proteins form dynamic clusters at different hierarchical levels (56, 57), which may be regulated by several factors. These include protein-protein interactions between transmembrane helices or extra/intracellular domains, the lipid composition of the membrane, the cytoskeleton, or cross-linking of glycoproteins by galectin lattices (58, 59). In T cells, disruption of lipid rafts by cholesterol depletion resulted in blurring of IL-2R clusters (37) and impairment of IL-2 signaling (13). Actin was shown to induce coconfinement of FcεRI receptors, thereby influencing their long-range mobility and response to ligand binding (60). In this article we give a detailed view of the effects of knocking down a highly expressed member of a membrane protein cluster on the lateral organization and dynamics of the remaining elements of the cluster. The different methods used sample different molecular populations. FCS is sensitive to the mobile fraction of molecules, whereas STED microscopy and FRET report on the total molecular population.

The diffusion coefficient of MHC I, IL-2Rα, and IL-15Rα was in the range of 0.03–0.15 μm²/s for control cells according to our FCS analyses. The mobility of IL-2Rα, IL-15Rα, and MHC I increased significantly (∼1.2–1.8-fold for the different proteins) upon KD of MHC I. This is compatible with the molecular interaction between MHC I and IL-2R/IL-15R suggested previously by our FRET and FCCS experiments (1, 13, 37), and implies a decrease of the size of protein aggregates accommodating these molecules. Molecular brightness data indicated that the extent of aggregation was highest for the MHC I molecules in control cells. FCS also revealed that the number of MHC I-containing aggregates decreased 10-fold and their specific brightness ∼1.7-fold. This clearly indicates that homoaggregation of MHC I depends strongly on its expression level. Altogether, brightness and mobility data suggest an overall decrease of the size of IL-2Rα/IL-15Rα/MHC I aggregates. Bleaching observed during FCS measurements suggested that a large fraction of the studied proteins had little or no mobility; ∼50–70% of proteins were bleached over 20 s. The bleached fractions decreased upon MHC KD for the case of MHC I and IL-2Rα. Mobility measurements of IL-2/IL-15 receptors highlight the sensitivity of these receptors to their micromilieu: the change in the number of interacting partners can considerably affect their dynamic behavior. Because the lifetime of MHC I clusters was found to be controlled by the cytoskeleton (61), and a direct or indirect link between the cytoskeleton and IL-2/IL-15 receptors was also found (62), our data might suggest that MHC I could mediate interaction between the cytoskeleton and IL-2R/IL-15R. Even the D values measured after MHC KD are lower than those reported for related single-spanning membrane proteins (0.3–0.4 μm²/s for IL-13Rα and 0.1–0.2 μm²/s for IL-2Rγ on HEK293T cells (63)), corroborating the clusterization and/or cytoskeletal interaction of MHC I and IL-2R/IL-15R proteins on FT7.10 T cells.

When assessing diffusion coefficients in the membrane by FCS, a general concern is that long diffusion times of the studied molecules would necessitate long FCS measurements; e.g., for τ_D = 50 ms, 1000 × 50 ms = 50 s recording time would be necessary to estimate the diffusion coefficient adequately (64). However, photobleaching prompts the investigator to record shorter runs; thus, diffusion coefficients may be underestimated. In our case, this means that the real mobilities are even lower than those estimated by FCS.

FRET measurements gave insight into protein-protein interactions at the nanometer scale. Associations involving MHC I diminished upon its KD as indicated by the decrease of FRET efficiencies. Importantly, not only the homoassociation of MHC I diminished, but also the heterotypic interactions between IL-2Rα/IL-15Rα, and MHC I became less abundant, as expected. On the other hand, the interaction between IL-2R/IL-15R subunits did not change significantly, suggesting that the interactions holding these subunits together were not critically dependent on the presence of MHC I.

Previously we found that IL-2/15 receptors were associated with lipid rafts even in the absence of specific ligands (1), whereas other results suggested that the migration of IL-2 receptors to rafts was induced by ligand binding (62). Our present colocalization analyses suggest that the enrichment of IL-2Rα/IL-15Rα in lipid rafts is constitutive, and it is not affected by removing MHC I molecules either. Other members of the common γ_c family of receptors were also found to be enriched in lipid rafts previously (10, 65, 66).

The application of superresolution STED microscopy allowed us to visualize and quantify larger protein clusters directly with a resolution of 40 nm. We assessed the distribution of supercluster sizes by three complementary image analysis methods: from the widths of intensity peaks, the correlation lengths of intensity distributions, and the number of pixels in contiguous high-intensity areas. These methods assess slightly different parameters of the superclusters, characterizing them at a length scale of a few tens to hundreds of nanometers. All three methods gave unequivocal results on KD-induced changes of cluster sizes. The mean supercluster size of MHC I was reduced by KD (from 400–600 to 200–300 nm). Conversely, the mean geometry of IL-2/15 receptor superclusters did not change drastically by removing their nearby MHC I partners. Colocalization analysis showed that the fraction of IL-2Rα/IL-15Rα residing in the vicinity of MHC I decreased after KD. This was paralleled by a decrease of the Pearson’s correlation coefficients between IL-2/15 receptors and MHC I probably due to the lower cell surface coverage of MHC I. On the other hand, the fraction of MHC I molecules residing in the vicinity of IL2-Rα/IL-15Rα did not change, suggesting a strong interaction between receptors and the remaining MHC I molecules.

Based on our data, we propose a model of co-clustering of IL-2/IL-15 receptors and MHC I and the effect of MHC I knockdown, as depicted in Fig. 7. We identified two levels of protein codistributions. STED microscopy revealed a supercluster size of 300–600 nm (15–30 pixels) for MHC I and IL-2/15 receptors (areas encircled by dashed or dotted lines in the scheme), similar in size to those found earlier by SNOM for IL-2Rα and IL-15Rα (∼450- and ∼360-nm average diameters) (67). These protein superclusters greatly overlap with GM₁-rich lipid domains (rafts). The size of protein superclusters is similar to the lateral diameter of the FCS confocal volume (∼400-nm diameter), which contains an average number of N = 10–35 diffusing units or tight aggregates based on our FCS data (this N is an underestimation because, in an inhomogeneous system, dim aggregates are underweighted). Each of these units may consist of one or several proteins of a kind according to FCS brightness analysis, and a large fraction of these contain multiple protein species (IL-2Rα, IL-15Rα, MHC I, and MHC II) according to FRET. Taken together, our view is that lipid raft domains confine IL-2Rα/IL-15Rα, and MHC glycoproteins in superclusters (1, 13, 37, 57), in which smaller interacting tight aggregates are formed as identified by FCS and FRET. Some of these aggregates are mobile, while others have little or no mobility. Upon MHC I KD the mobility of these tight aggregates is increased, which is partially due to their lower MHC I content. A further reason for the increased mobility can be a lesser degree of molecular crowding (68) due to the lower overall MHC I content of the superclusters. After KD, MHC I retracts to smaller areas, as suggested by a higher fraction of MHC I-free IL-2Rα/IL-15Rα domains and the smaller correlation coefficients of codistributions of MHC I with IL-2Rα/IL-15Rα/GM1.

Figure 7.

A possible visualization of the effects of MHC I knockdown on the clustering of the IL-2/IL-15 receptor complex and MHC I molecules. Encircled areas mark protein superclusters as discerned by STED microscopy (red dashed line for MHC I and gray dotted line for IL-R superclusters), which contain several distinct tight aggregates of codiffusing proteins. Some of these aggregates are mobile, while others have little or no mobility. Upon MHC I KD, the MHC I content of the aggregates decreased, leading to a decrease of their size and an increase of their mobility. MHC I supercluster size decreased upon KD from 400–600 to 200–300 nm (marked by the decreased area of MHC I-containing regions encircled by red dashed lines), whereas the mean geometry of IL-2/IL-15 receptor superclusters did not change significantly. To see this figure in color, go online.

In conclusion, MHC I knockdown significantly influenced cell-surface supramolecular patterns. Changes of clustering properties induced by changes of MHC I expression level may modulate signaling properties of receptor complexes associated with MHC I. Our data demonstrate that modifying the expression of one element of a protein cluster can alter protein associations and the dynamic properties of the cluster.

Author Contributions

G.M. and J.V. carried out most experiments with contributions from G.V., A.B., D.R., and K.T.; G.M. and J.V. analyzed most experiments with contributions from G.V. and D.R.; G.M. carried out simulations; P.N. designed siRNA; C.K.G. provided reagents and cells; G.M. and G.V. wrote the article with inputs from K.T., J.S., S.D., T.A.W., P.N., J.W.; G.V. and G.M. conceived the experiments.

Editor: Anne Kenworthy.

Supporting Citations

References (69, 70, 71, 72, 73, 74, 75, 76, 77) appear in the Supporting Material.