Membrane cholesterol interferes with tyrosine phosphorylation but facilitates the clustering and signal transduction of EGFR

Epidermal growth factor receptor (EGFR) activates major cell signaling pathways that regulate various cell responses. Its dimerization and clustering coupled with its lateral mobility are critical for EGFR function, but the contribution of the plasma membrane environment to EGFR function is unknown. Here we show, using single-molecule analysis, that EGFR mobility and clustering are altered by the depletion of cholesterol or sphingomyelin, major lipids of membrane subdomains, causing significant changes in EGFR signaling. When cholesterol was depleted, the subdomain boundary in EGFR diffusion disappeared, the fraction of EGFR pre-dimers was increased, and the ligand-induced phosphorylation of EGFR was enhanced. In addition, the depletion of either lipid prevented the formation of immobile clusters after EGF association and decreased the phosphorylation of downstream proteins. Our results revealed that cholesterol plays dichotomous roles in the signaling pathway of EGFR and that clustering in the membrane subdomains is critical for EGFR signal transduction.


Introduction 1
Epidermal growth factor receptor (EGFR), a receptor tyrosine kinase, is a major 2 regulator of several intracellular signaling cascades by receiving extracellular ligands at 3 the plasma membrane. EGFR signaling transfers information to the well-known RAS-4 MAPK pathway, inducing essential cell responses such as proliferation, differentiation, 5 migration, apoptosis, and others. Defects in EGFR function affect cellular responses, 6 often inducing hyperactivated signaling to causes carcinoma and other diseases 7 (Carpenter et al., 1978;Lemmon and Schlessinger, 2010). Ligand-bound EGFR is auto-8 phosphorylated when EGFR takes a dimer structure to activate downstream signaling. 9 Even before the ligand binding, dimer formation (pre-dimer) occurs (Hiroshima et  shown to transit between three motional states; namely, immobile, slow-, and fast-26 mobile, which were determined in terms of the size of the diffusion coefficient 27 Yasui et al., 2018). We previously found that EGFR molecules 28 in the slow-mobile state showed a confined diffusion surrounding the trajectory of the 29 immobile state in which the position of the molecules fluctuated within a confined area 30 (~60 nm) only two times larger than the localization accuracy. Fast-mobile EGFR 31 molecules moved in space between the slow-mobile compartment with simple diffusion. 32 The clustering states, which correspond to the number of EGFR molecules moving 33 have been reported to disperse and affect the physical properties of lipid rafts, 54 respectively (Cremesti et al., 2002;Smith et al., 2010), indicating they alter EGFR 55 behavior. These perturbations offer information on how cellular signaling is affected by 56 the plasma membrane environment through the dynamics of EGFR behavior at the 57 molecular level. 58 To understand the dependency of EGFR behavior, including the localization, 59 mobility, clustering, and their coupling, on the membrane structure, the present study 60 employed single-molecule analysis while depleting cholesterol or sphingomyelin. 61 Furthermore, assessments of the receptor and its downstream activity were carried out 62 to reveal the correlation between EGFR behavior and cellular signaling. 63 phosphorylation 66 CHO-K1 cells were transfected with EGFR-GFP and treated with MCD or 67 sphingomyelinase for the lipid depletion. Cholesterol was reduced to 33% and 16% with 68 5 and 10 mM MCD treatment, respectively, according to GC-FID or GC-MS 69 measurements (Fig. 1a). Similar depletion was also confirmed by the exogenous 70 addition of fluorescent EGFP-labeled  toxin, a probe of free cholesterol (Fig. 1b), that 71 binds to the cells. Sphingomyelin was reduced to 18% by sphingomyelinase treatment 72 ( Fig. 1b) based on the fluorescence of GFP-labeled lysenin, a specific probe of 73 sphingomyelin (Fig. 1b). The observed fluorescence of  toxin-GFP in 74 sphingomyelinase-treated cells and lysenin-GFP in MCD-treated cells were the same 75 as in non-treated cells (Fig. S1), indicating that the specific depletion of one lipid 76 neither affected the content of the other lipid in cells. 77 The time-course of EGFR phosphorylated at Y1173 in cholesterol-depleted and 78 control cells reached a maximum at 1-3 minutes after EGF stimulation according to 79 Western blotting results ( Fig. 1c and S2a). The phosphorylation level under cholesterol-80 depletion was higher than that in the control condition and was dependent on the MCD 81 concentration. The phosphorylation of EGFR at Y1173 and Y1068 was increased by the 82 cholesterol depletion two minutes after the stimulation (Fig. 1d). The half-maximal 83 effective concentrations (EC50) of EGF was almost the same between control and 84 cholesterol-depleted conditions: 1.9 nM and 2.1 nM for pY1173, and 1.5 nM and 1.3 85 nM for pY1068, respectively. Hill coefficients indicating no cooperativity (0.6-1.0) 86 were not changed by the depletion. After 30 nM EGF stimulation, the cholesterol-87 depletion condition induced 1.8-fold and 2.7-fold higher phosphorylation of Y1068 and 88 significantly (Fig. 1e). 91 Time-course of the EGF-induced EGFR phosphorylation (pY1173) (n = 3 trials). Fold-97 changes relative to phosphorylation at 0 min are indicated. d. Dose-response curves for 98 EGF-induced tyrosine phosphorylation in EGFR. e. Comparison between the 99 phosphorylation levels at 2 min after 30 nM EGF stimulation. ** p < 0.01 (t-test). a-e, 100 Error bars: SE. All data points are shown in Fig. S2. 101 cholesterol-depleted conditions with consideration of EGFR behavior, we applied 105 single-molecule imaging of EGFR-EGFP on the plasma membrane of living cells. We 106 analyzed the trajectories of individual fluorescent spots using an HMM-based machine 107 learning method to assign every step along the trajectories with specific motional and 108 clustering states. The movements consisted of immobile, slow-, and fast-mobile states in 109 all lipid conditions ( Fig. S3 and Table S1). The depletion of either lipid increased the 110 diffusion coefficient in the immobile state and the fraction of the slow-mobile state 111 while decreasing the fraction of the fast-mobile state. Sphingomyelin depletion also 112 increased the diffusion coefficient of the fast-mobile state. The observed changes in the 113 diffusion coefficients were consistent with previous reports indicating that membrane 114 fluidity is reduced by the addition of cholesterol (Tabas, 2002)  where D1 and D2 are the diffusion coefficients, L is the confinement length, and t is the 125 diffusion time. The suitable diffusion mode for each MSD profile was determined using 126 Akaike's information criterion (AIC), which was calculated using the equation below, 127 where RSS is the residual sum of squares between the data and the model, N is the 129 number of data points, and k is the number of parameters. The model with the higher 130 AIC was selected. In the control condition, confined diffusion was observed in the 131 immobile and slow-mobile states, but the fast-mobile state showed simple diffusion. 132 The confinement lengths (L) of the immobile state were 60 nm for all cluster sizes 133 (monomer, dimer, and higher-order clusters). L for the slow-mobile state was equivalent 134 to the size of well-known membrane subdomains (including lipid rafts). Finally, the 135 mobility of the monomer (~310 nm) was more confined than the mobility of the other 136 clusters (~570 nm). 137 When cholesterol was depleted, the diffusion mode in the slow-mobile state was 138 altered from confined to simple diffusion for all cluster sizes. The cholesterol depletion 139 had little effect on the MSD of both the immobile and fast-mobile states or on the 140 distance between the centers of the immobile state regions (Fig. S5). When 141 sphingomyelin was depleted, the slow mobile EGFR still showed confined diffusion, 142 but the confinement was less in comparison with the control condition (~720 nm for ≥ 143 dimer). In Fig. 2c, individual trajectories were superimposed to exhibit the expanding 144 diffusion area of EGFR, for which the center of the first immobile state in each 145 trajectory was shifted to the origin. The second and later immobile states are seen as 146 small islands separated from the first region, and the slow-and fast-mobile states were 147 distributed around the immobile states (Fig. 2c, left). Typical trajectories (Fig. 2c, right) 148 rarely showed a direct transition between the immobile and fast-mobile states, as 149 observed in the transition probability (Table S1). Transitions between the immobile and 150 slow-mobile states often occurred at the surrounding boundary of the region for an 151 immobile state, suggesting that an EGFR particle in the immobile state was trapped 152 within a 60-nm membrane subdomain that was stable during the observation time.  The cluster size distribution, which was obtained from the HMM analysis, showed that 174 EGFR dimer and higher-order clusters existed even without ligand stimulation and thus 175 could be called pre-dimer and pre-clusters, respectively. The pre-dimer is responsible 176 for increasing the sensitivity of EGF signaling ( phosphorylation. We found that when cholesterol was depleted, the fraction of pre-180 dimer in the slow-mobile state was significantly increased (1.4-fold), but the fractions of 181 monomers and higher-order clusters (≥ trimer) were unchanged ( Fig. 3a and S6), 182 suggesting an upshift in dimerization affinity between EGFR monomers and the 183 destabilization of clusters to the dimer. The fractions from monomer to tetramer were 184 reduced in the fast-mobile state, but no change was observed in the immobile state 185 (Table S2). These changes increased the total slow-mobile fraction 1.3-fold ( Fig. S3b  186 and Table S1).
(Tables S1 and S2). In the control condition, a significant net efflux of monomers (0.27 190 ± 0.05% s -1 ) was observed from the slow-mobile state, but there was no significant We also measured the reaction rate constants of dimerization and dimer 204 decomposition. The 1st-order dimerization rate constants in the slow-mobile state, were 205 calculated from the frequency of dimerization events and found to be 7.3 ± 0.5 s -1 (123 206 cells) in the control condition and 7.4 ± 0.4 s -1 (69 cells) in the cholesterol-depleted 207 condition. The difference was not statistically significant. The decomposition rate 208 constants were 7.6 ± 0.05 s -1 under the control condition and 6.1 ± 0.06 s -1 in the 209 cholesterol-depleted condition, which was a significant difference (Table S3). The total 210 density of fluorescent particles on the cell surface (1.4 ± 0.7 m -2 ) was not affected by 211 the cholesterol depletion. When this particle density was applied to the region for slow-212 mobile motion, the fractions of monomers and dimers in the slow-mobile state were 213 converted into particle densities of 0.32 ± 0.02 m -2 and 0.67 ± 0.04 m -2 in the control 214 in the stability of pre-dimers, mainly contributed to the increased pre-dimer formation 218 under the cholesterol depletion. Furthermore, the destabilization of the EGFR clusters 219 suggested above may have another cause to increase the pre-dimer fraction. 220 Next, we checked locations of the pre-dimer formation relative to the center of 221 the first immobile region along the single-molecule trajectories. The frequency of 222 dimerization events per area in the slow-and fast-mobile states was mapped in two 223 dimensions and averaged over the circumference (Fig. 3b). Reflecting the release from 224 confinement (Fig. 2b), the pre-dimerization locations spread further away when 225 cholesterol was depleted ("MCD" in Fig. 3b). Most of the molecules traveled to the 226 next immobile region (long-traveled; Fig. 3c) and formed pre-dimers regardless of the 227 lipid condition (Fig. 3d). In the absence of EGF, the long-traveled fraction was largest 228 upon cholesterol-depletion (Fig. 3e). Dimerization and higher-order clustering also 229 occurred more frequently in the long-traveled fraction (Fig. 3f). These results suggest 230 that cholesterol-depletion spreads the pre-dimers and pre-clusters (Fig. S7a) of EGFR 231 over a large region of the plasma membrane. 232 On the other hand, sphingomyelin-depletion caused no obvious change 233 regarding the EGFR clustering, such as the fraction distribution ( Fig. 3a and S6), 234 reaction rate constants (Tables S1 and S2), location of the dimerization (Fig 3b, d, and  235 f), or fraction of long-traveled molecules (Fig. 3d). 236  (Fig. 4b). However, the area of the EGF-268 induced dimerization was similar under all lipid conditions (Fig. S7b). The EGF-269 induced dimerization during long traveling occurred at the same frequency between all 270 lipid conditions (Fig. 4c), whereas higher-order clustering was facilitated by EGF only 271 in the control condition (Fig. 4d) in parallel with the increase in long-traveled molecules 272 (Fig. 3e). 273 Changes in the cluster size distributions (Fig. 4a and S8) Fig. 4e and S4) and reduced the confinement lengths to ~160 (monomer) and ~340 nm 283 (≥ dimer) and ~160 (monomer) and ~530 nm (≥ dimer), respectively (Table S1) (trajectory -1 s -1 ) relative to the center of the first immobile region. Bottom, ratios of 295 clustering (after:before the EGF addition). Scale bars, 500 nm. c and d. Dimerization (c) 296 and higher-order clustering (≥ trimer; d) events as the relative fraction among the total 297 long-traveled molecules. The fractions are normalized to control cells before the EGF 298 stimulation. * p < 0.05 (t-test) between control and lipid-depleted conditions; ### p < 299 0.001 (t-test) between before and after EGF stimulation. e. MSD-t plots of the slow-300 mobile state. Error bars: SE. All single-cell data points are shown in Fig. S4. 301 302

phosphorylation. 320
The phosphorylation of the downstream protein ERK in the EGFR signaling was 321 measured by time-course Western blotting and maximized 5 minutes after the EGF 322 stimulation, which is later than the time of EGFR phosphorylation (Figs. 5d and S10a). 323 EGF-induced ERK phosphorylation was observed in all conditions, but its level was 324 lower with cholesterol-depletion, which is expected when considering the reduced 325 GRB2 translocation to the plasma membrane (Fig. 5a) and SHC phosphorylation ( The effects of the lipid depletions on the phosphorylation of AKT and ERK were 333 evaluated 2 minutes after 30 nM EGF stimulation, which is almost the saturation 334 condition (Figs. 5f and S10b). Although cholesterol-depletion increased the level of 335 EGFR phosphorylation (Fig. 1e), the levels of AKT and ERK phosphorylation were 336 significantly decreased (Fig. 5f). Sphingomyelin-depletion also lowered the levels of 337 ERK and AKT phosphorylation, but differently, likely reflecting the specificities of the 338 signaling pathways. 339 on the increased amount of EGFR pre-dimer, which hardly undergoes auto-366 phosphorylation but is primed for a rapid response upon EGF stimulation (Teramura et 367 al., 2006). EGFR has three motional modes in its lateral diffusion coefficient. After 368 cholesterol-depletion, the amount of pre-dimer increased approximately 1.4-fold 369 primarily in the slow-mobile state. The diffusion mode of the slow-mobile state was 370 altered from confined to simple diffusion (Fig. 2b) without a significant change in the 371 diffusion coefficient (Fig. S3a). This observation indicates that cholesterol-depletion 372 enabled molecules to go freely through some barrier and move long distance (~1.8 fold 373 longer than the control during the observation time). This barrier might correspond to a 374 physical factor that maintains spatial phase separation in the membrane to impede 375 EGFR from moving over the subdomain border composed of cholesterol or some 376 component interacting with cholesterol around EGFR (e.g. shell model) (Anderson, 377 2002). Our results suggest that EGFR molecules in the slow-mobile state prefer to exist 378 in the subdomains. Since EGFR pre-dimers were mainly present in the slow-mobile 379 state, the disappearance of the barrier allowed them to spread over the cell surface. 380 The effect of cholesterol depletion on the affinity between EGFR protomers in 381 the pre-dimer was considered from the rate constants of dimerization and 382 decomposition. In the slow-mobile state, the rate constant of decomposition was 383 significantly decreased, but we did not detect a significant change in the dimerization 384 rate constant. As a result, the affinity was increased by the cholesterol depletion. In 385 addition, the fraction of the slow-mobile state was increased due to increased and 386 decreased of the transition probabilities (rate constants) from the fast-to slow-mobile 387 states and the slow-to fast-mobile states, respectively (Table S1). These two effects 388 induced the increase in the number of slow-mobile pre-dimers under the cholesterol-389 depleted condition and possibly resulted in the upregulation of EGFR phosphorylation. 390 The disappearance of the diffusion barrier for the slow-mobile state of EGFR molecules 391 may be related to the increase of the slow-mobile fraction. It also seems likely that the 392 stimulative effect of cholesterol-depletion on the EGFR phosphorylation ( Fig. 1)  Sphingomyelin-depletion, which did not affect cholesterol, also caused 401 significant effects on EGFR in the slow-mobile state. The confinement length for the 402 slow-mobile state was increased, though the confinement did not disappear (Fig. 2b). 403 The fraction of the slow-mobile state was increased (Fig. S3b), reflecting the rise in the 404 transition probability from the fast-to slow-mobile states (Table S1). The fractions of 405 the monomer and other clusters in the slow-mobile state were unchanged. pre-dimer fraction despite the increase in the slow-mobile state. 412 EGFR clusters larger than dimers were also formed before the EGF stimulation 413 (pre-clusters). Different from pre-dimers, cholesterol-depletion did not increase the pre-414 cluster fraction (Fig. 3a), although the confinement disappeared (Fig. 2b) to enlarge the 415 regions of the slow-mobile motions for the pre-clusters (Fig. 2c). On the contrary, the 416 EGF-induced formation of higher-order clusters, which was observed in the control 417 condition, was suppressed under cholesterol-depletion ( Fig. 4a and 4b) in a cholesterol 418 dose-dependent manner (Fig. S8b). Sphingomyelin-depletion also suppressed the EGF-419 induced clustering of EGFR. These lipid dependencies suggest that the clustering of 420 EGFR is caused by a mechanism different from that for EGF-induced dimerization. 421 Cholesterol and sphingomyelin may pack and enclose the EGFR molecules in small 422 membrane subdomains or directly bind up the molecules. The oligomerization of TM 423 peptides of EGFR has been observed in liposomes containing cholesterol (Jones et al., 424 1998). Following our previous report that the EGF-induced EGFR clusters in the 425 immobile state are the primary interaction sites with the adaptor protein GRB2 426 , the deficient clustering by the lipid depletion correlated with 427 the reduction in the membrane translocation and in the phosphorylation of adaptor 428 proteins ( Fig. 5a and b). Indeed, the downstream proteins ERK and AKT showed 429 reduced phosphorylation (Fig. 5c, 5d, and 5e), suggesting that cholesterol and 430 sphingomyelin substantially contribute to the cellular signaling through the EGFR-431 immobile cluster formation. 432 Based on our observations (Fig. S11), we provide a scheme for EGFR-mediated 433 cell signaling (Fig. 6): First, the immobile and slow-mobile states of EGFR are confined 434 within a cholesterol-and sphingomyelin-enriched membrane subdomain (Fig. 2b). A 435 between the slow-mobile state; however, only cholesterol and not sphingomyelin 437 prevents the slow-mobile EGFR from freely passing over the border and interfering 438 with the pre-dimer formation (Fig. 3a). Then, EGF association quickly converts EGFR 439 from a pre-dimer to kinase active dimer. Moreover, EGF facilitates the formation of 440 clusters larger than dimers with an enlarged area of the clustering (Fig. 4b and   intensity acquired from regions with no cells was subtracted from the signal. 513

Microscopy and Image Analysis for Single-molecule Imaging and Tracking 514
Cell starvation was carried out by changing HAM F12 medium to modified Eagle's 515 medium minus phenol red and FBS 1 day before single-molecule imaging. Objective-516 type total internal reflection illumination was applied to observe EGFR-GFP in the basal 517 plasma membrane of the cells through a PlanApo 60 NA 1.49 objective (Nikon, 518 Tokyo, Japan) equipped on an inverted microscope (TE2000; Nikon). Lasers with 519 wavelengths of 488 nm (Sapphire 488; Coherent, Santa Clara, CA) and 561 nm 520 (Sapphire 561; Coherent) were used for the excitation of GFP and TMR, respectively. 521 The dichroic mirror and emission filter were Di02-R488 (Semrock) and FF01-525/45 522 (Semrock) for GFP, and Di02-R561 (Semrock) and BLP02-561R (Semrock) for TMR 523 imaging. An electron-multiplying CCD (EMCCD) camera (C9100-23; Hamamatsu, 524 Hamamatsu, Japan), which was controlled using HCImage software, acquired 525 fluorescence images at a frame rate of 33 s -1 . The imaging was done at 25°C. Image 526 processing was carried out with moving averages over two frames and background 527 subtraction using rolling ball filtering (radius: 25 pixels) of the ImageJ plugins. Single-528 molecule tracking was performed on the processed images with custom-made software. 529 The obtained data including positions and intensities of all fluorescent spots were 530 analyzed using the methods described below. which were used in the next VB-E step. Fourth, the lower bound of the evidence, Lq, 547 was calculated to evaluate its convergence (except for the first Lq value) by judging 548 whether the difference from the previous Lq was less than 0.001%. Fifth, if Lq was not 549 convergent, the next iteration was performed by repeating the third and fourth steps. 550 Finally, the state sequence was determined by choosing the state with the highest 551 probability at every frame. 552

MSD for Each Mobility and Clustering State 553
The MSD of a specific mobility and clustering state, which was attributed to steps along 554 the receptor trajectory, was calculated as 555 high efficiency, well plate-based measurements were performed with the automated 567 system that we developed (Yasui et al., 2018). Each of the automatically determined 5 568 fields of view, including 1-3 cells per field, was observed for 200 frames (6 sec) both 569 before and 2 minutes after the EGF stimulation. The acquired images were analyzed 570 using built-in software for tracking fluorescent spots. The spots observed in the 10th 571 frame were used for the analysis to exclude fluorescence debris, which was bleached 572 immediately after illumination. The number of translocated proteins on the plasma 573 membrane was reflected in the total brightness of the fluorescent spots, in which more 574 than one adaptor protein molecule might be included in a spot. The total brightness 575 before and after the EGF stimulation were compared by their ratio. 576