MEK Inhibition Reverses Aberrant Signaling in Melanoma Cells through Reorganization of NRas and BRAF in Self Nanoclusters

The mitogenic pathway conveys growth signals from specific receptors (e.g., EGFR) into the cell and triggers its growth and division. This pathway is often described as a linear cascade of specific and synchronized molecular interactions (1). For instance, mitogenic signals are conveyed from the growth receptors to Ras, RAF, MEK, and ERK. ERK then translocates to the nucleus, activates transcription, and promotes cell growth (2). As a key step in this pathway, (H,K,N)-Ras and BRAF are recruited from the cell cytosol to its plasma membrane (PM), where these proteins get activated. Recent studies suggest that these proteins form transient complexes at the PM, together with additional proteins such as scaffolds, to convey the receptor signals (3). Indeed, NRas and BRAF have been described to form functional dimers and small (nanoscale) clusters at the PM (4). These proteins are then recycled in vesicles between the cell cytosol, its various organelles, and the PM (5).

Multiple proteins involved in the mitogenic pathway undergo mutations that drive a variety of cancers by aberrant signaling, independent of external growth signals (6, 7). Specifically, approximately 14%–20% of melanomas carry mutations in NRas (8, 9). Frequent mutations include G12D, G13D, and Q61R (10). BRAF often carries the V600E-activating mutation (11).

MEK inhibitors (MEKi) have become a prominent route for clinical treatment of patients with melanoma, with some promising success rates (12). Still, the detailed dynamic organization of signaling complexes, nucleated by NRas or its oncogenic mutants on the PM, and their potential role in cell transformation and cancer progression remain poorly understood (13). Such understanding is critically needed for optimal targeting of the aberrant activity of these proteins in various cancers.

Here, we hypothesized that molecular complexes and clusters of NRas and BRAF at the PM could play a critical role in regulation of the mitogenic signal in health and disease. Thus, we employed single-molecule localization microscopy (SMLM) to study NRas and BRAF at the PM of melanoma cells and in single molecule detail. Interestingly, NRas and BRAF resided in self-clusters that did not associate well in resting cells. In EGF-activated cells, these clusters became more diffused while overall protein levels at the PM increased; thus allowing enhanced association of NRas and BRAF and signaling downstream.

Mutant NRas (Q61R, G12D, G13D) resided in more pronounced self-clusters relative to wild-type (WT) NRas in multiple melanoma cell lines. Still, they could associate more with the clustered, and more abundant BRAF. Strikingly, treatment of such cells with trametinib, a clinical MEKi, reversed the effect of the mutant NRas expression. It restored the nonoverlapping self-clusters of NRas and BRAF and reduced their PM levels. Treatment with trametinib could also reduce the elevated pERK levels, caused by the NRas mutants, to the levels of resting cells. Our results indicate a new mechanism for signal regulation of NRas in melanoma through its nanoscale dynamic organization. They also show a new mechanism for MEKi function in melanoma cells carrying NRas mutations, but lacking MEK mutations.

This study includes the following patient-derived melanoma cell lines (background mutations are in parentheses): 108T (NF1-H1366Q), A375 (BRAF-V600E), 72T (NRas-G12D), 74T (NRas-Q61R), 83T (NRas-G13D), and 12T(NRas-Q61R). A subset of cell lines used in the study (108T, 72T, 74T, 83T, and 12T) were derived from a panel of pathology-confirmed metastatic melanoma tumor resections collected from patients enrolled in Institutional Review Board (IRB)-approved clinical trials at the Surgery Branch of the National Cancer Institute (Bethesda, MD). Pathology-confirmed melanoma cell lines were derived from mechanically or enzymatically dispersed tumor cells, which were then cultured in RPMI1640 + 10% FBS at 37°C in 5% CO₂ for 5–15 passages.

These cell lines were used in previous studies (14, 15). They were originally derived from a panel of pathology-confirmed metastatic melanoma tumor resections collected from patients enrolled in IRB-approved clinical trials (with written informed consent) at the Surgery Branch of the National Cancer Institute (Bethesda, MD). Pathology-confirmed melanoma cell lines were derived from mechanically or enzymatically dispersed tumor cells, which were then cultured in RPMI1640 + 10% FBS at 37°C in 5% CO₂ for 5–15 passages. Cell lines genotypes are indicated in Supplementary Table S1. All cell lines have been tested negative for Mycoplasma.

In this work, we constructed BRAF and NRas WT and three mutants G12D, G13D, and Q61R conjugated to the photoactivatable fluorophores PAGFP and PAmCherry, respectively, by the gateway cloning method (16). All constructs containing photoactivatable fluorescent proteins were cloned in pcDNA3 plasmids with a cytomegalovirus promoter. The cloning of the fluorescent tags was performed at the N-terminus to prevent localization disruption. Constructs were validated by DNA sequencing. The 108T melanoma cells were transfected with PAmCherry-NRas (WT or mutants) and PAGFP-BRAF using Lipofetamin 3000 (L3000008, Invitrogen) for 48 hours before EGF addition and fixation. Cell seeding and imaging was conducted on glass coverslips (#1.5 glass chambers, LabTek and Ibidi), and fixed with 2.4% paraformaldehyde for 30 minutes at 37°C. Trametinib (GSK1120212, Selleckchem, S2673) was used in a concentration of 10 nmol/L for MEKi treatment of cells over 16 hours.

For direct Stochastic Optical Reconstruction Microscopy (dSTORM) imaging, cells were seeded and fixed for 15 minutes after adding EGF. For permeabilization, we added 0.4 mL 0.1% Triton X-100 in PBS per well and incubated for 3–4 minutes. The cells were blocked by 2% normal goat serum in PFN buffer (phosphate-buffered saline, 10% fetal calf serum, and 0.02% NaN₃) for 30 minutes. For 0.5 million cells, we added 0.5 μg rabbit αhuman phsopho-BRAF (ab19283, Abcam) and mouse αhuman NRas (SC-31, L1115) as primary antibodies diluted in 2% normal goat serum in PFN, incubated for 60 minutes at room temperature and washed three times with PFN. Alexa647 (αrabbit, Invitrogen, A21244) and Atto488 (αmouse, Sigma, 62197) was added as secondary antibodies diluted (1/3000) in 2% normal goat serum in PFN, incubated for 45 minutes at room temperature and washed three times with PFN.

Proteins were extracted with RIPA buffer (R0278, Sigma) and Protease phosphatase inhibitor cocktail (PPC1010, Sigma). SDS-PAGE electrophoresis was performed using Novex WedgWell 4%–12% gels (XP04120BOX, Invitrogen) and blotted on iBlot2 membranes (IB23001, Invitrogen). Western blot detection was performed with following antibodies: Anti-NRas (F155, Santa Cruz Biotechnology), Anti-ERK-P (ab201015, Abcam), Anti-ERK (4695, Cell Signaling Technology), anti-tubulin (T6074, Sigma), and Anti-ABCB5 (ab140667, Abcam).

In our study, we aimed to incorporate relatively low levels (∼30%) of exogenous proteins relative to endogenous proteins. Individual expression levels of proteins were monitored by Western blots (Supplementary Fig. S1A) and imaging (throughout the study). Higher protein levels bear the risk of artificially promoting the mitogenic signaling of the cell. At these moderate levels of expression, exogenous proteins that are similar in sequence to the endogenous proteins (e.g., WT NRas and WT NRas-PAmCherry) are expected to represent the behavior of the endogenous proteins by distributing randomly among the endogenous proteins (either as monomers or in clusters). Exogenous proteins that are different in sequence (e.g., NRas mutants carrying a fluorescent tag) could differ in their self-organization at the PM, and in their mutual interactions with other proteins (esp. BRAF, as in Figs. 3 and 5). Notably, in Fig. 7A and B, the moderate expression of mutant NRas was sufficient to cause upregulation of ERK, as measured by pERK, relative to the effect of expression of WT NRas.

See further details in the Supplementary Information regarding the expression levels of ABCB5 and the cell proliferation assay.

For dSTORM imaging, cells expressing endogenous WT and mutant NRas were seeded and fixed on coverslips, as described above. For cell permeabilization, we added 0.4 mL 0.1% Triton X-100 in PBS per 1 × 1 cm² well and incubated for 3–4 minutes. The cells were blocked by 2% normal goat serum in PFN (PBS + 10% serum + 0.02% sodium azide) for 30 minutes. Cells were next stained using a primary NRas antibody (L1115, Santa Cruz Biotechnology) and a secondary antibody stained with Alexa647 (company). This fluorophore was imaged using 647 nm laser illumination at up to 100% (170 mW in maximum). Direct-STORM buffer included GLOX (0.5 mg/mL glucose oxidase, 40 μg/mL catalase, 10% glucose), 10 mmol/L 2-aminoethanethiol, in buffer that included 50 mmol/L Tris and 10 mmol/L NaCl with a pH of 8 (17). For the phosphorylation experiment, WT endogenous NRas and pBRAF were tagged by mouse anti-NRas (SC-31, L1115) and rabbit anti-pBRAF (ab19283, Abcam) as primary and Atto488 anti-mouse (αmouse, Sigma, 62197) and Alexa647 anti-rabbit (αrabbit, Invitrogen, A21244) as secondary, respectively.

We conducted two-color photoactivated localization microscopy (PALM) imaging similar to previous studies (18), using a total internal reflection (TIRF) Nikon microscope with a CFI Apo TIRF X100 oil objective (NA 1.49, WD 0.12 mm). PAGFP and PAmCherry were photoactivated using variable intensity (0.5%–10% of 30 mW in maximum) of 405 nm laser illumination that was changed gradually from beginning of imaging to the end and alternate excitation using (50% of 90 mW in maximum) 488 nm laser excitation for PAGFP and (50% of 90 mW in maximum) 561 nm for PAmCherry unchanged during the imaging. Laser illumination at all wavelengths covered a circular area with a diameter of 80 μm at the sample. Movies of fixed cell imaging were acquired for up to 3,000 frames at 13.1 fps of an EMCCD Ixon⁺ camera. The focus of the microscope was maintained throughout the imaging using the PerfectFocus system of the microscope.

We used the ThunderSTORM software (19) to analyze PALM and dSTORM movies and generate images. Briefly, this software served to identify individual peaks and to assign them to individual molecules for rendering of the PALM or dSTORM images. The localization uncertainties of the different fluorophores used in our study peaked at approximately 20–30 nm (Supplementary Fig. S1B). A distance threshold and a temporal gap were employed for peak grouping to account for possible molecular blinking (20). Temporal gaps were tested for each fluorophore separately to minimize possible overcounting of molecules.

In SMLM, single fluorophores can be typically detected over multiple frames till their photobleaching. Fluorophore blinking may cause absence of some detections from specific frames in the detection sequence. Such blinking and “off-frames” may then lead to fluorophore overcounting because single fluorophores could then be counted more than once (e.g., ref. 21). A standard approach to overcome this issue is to first characterize the statistics of “off-times” for each fluorophore. Then, a threshold for maximal “allowed” temporal gap between detections can be assigned (21–23).

Another potential issue that can lead to fluorophore overcounting is due to the spreading of the fluorophores’ localizations across their detected area [the point-spread function (PSF)] on the detector. To avoid both issues, a standard procedure in SMLM reconstruction is to group the detected peaks over space and time, and assign them to single fluorophores. We provide further details on this procedure in the Supplementary Information, in Supplementary Fig. S2A–S2H and grouping parameter values in Supplementary Table S2. These parameters were provided through dedicated filters during SMLM reconstruction using ThunderSTORM (19). The effect of selected grouping parameters on our results is shown in Supplementary Fig. S3A–S3D.

Drift compensation and channel registration were performed using dedicated algorithms in the ThunderSTORM software. Individual molecules were presented in corresponding to the probability density values of their fitted Gaussian. This scale was set for each species separately, according to its maximal probability density in the field (note that it should not be interpreted as the density of molecules in clusters).

To classify points into clusters, we used the published algorithm DBSCAN using the Matlab function "rangesearch" (MathWorks; ref. 24), with a distance threshold of <30 nm between molecules in same cluster. This algorithm served to generate the size distribution of molecular clusters in copy numbers (25). Histograms are shown in Fig. 1O as cumulative distributions to allow clearer comparison between results under various cell conditions. See further details in the Supplementary Information on coclustering analysis of NRas and BRAF (Fig. 6G).

Modeling and simulations (Fig. 2)—see details in the Supplementary Materials and Methods and in Supplementary Table S3.

A detailed description of the second-order statistics used in this study has been published and extensively described previously (26–28). This description includes the univariate pair correlation function (PCF), the bivariate PCF (BPCF), and extent of mixing (EOM). Briefly, PCF [denoted here also as g(r)] describes and quantifies in a point pattern how density varies as a function of distance from a reference particle/point. Usually PCF is normalized by the density of the sample, and was used in this normalized form throughout the study. The univariate PCF is used to explore a point pattern of a single species. It is further useful to compare the PCF results with a Poisson model that describes random placement of points across the field. This model results in a flat PCF, for which g(r) = 1. Thus, higher values of the PSF [i.e., g(r)] indicate self-clustering.

For patterns with two species, BPCF statistics quantify the density of pattern 2 at distance r from an arbitrary point of pattern 1. To investigate whether or not two species (in a joint point pattern) are significantly interacting, we used the random labeling model. In this model, points of pattern 1 (n₁) and points of pattern 2 (n₂) distribute randomly at n₁ + n₂ fixed locations. Multiple Monte Carlo simulations replicate 19 times the point patterns while randomly relabeling the points (with the number of points from each species). The BPCF of the original point pattern g₁₂(r) is then compared with the BPCFs of the simulations. We used the lowest and highest g₁₂(r) of the different simulations as a 95% confidence interval for the acceptance or rejection of the model as a null hypothesis. Agreement of the data with the random labeling (RL) model indicates homogeneous mixing, and hence strong interaction (in a statistical sense) of the two species under study. Alternatively, a model of no interaction would result in a flat curve where g₁₂(r) = 1. The EOM represents the average of BPCF over multiple cells using 95% confidence interval due to the RL model. EOM is normalized between the values of +1 for perfect mixing and 0 for the model of no interaction.

To quantify the statistical significance of two-dimensional [2D; density-g(r)] distributions, we performed two-sample Kolmogorov–Smirnov using the function “kstest_2s_2d” in Matlab. This function evaluates the difference between two vectors of 2-by-n given as an input. The result is a value between 0 and 1 (for none to perfect fit, respectively; ref. 19). All significance values of the [density-g(r)] plots and of the EOM graphs are summarized in Supplementary Table S4.

In the rest of the analyses, P values were calculated using Student t test, assuming two-tailed distribution and samples with unequal variances.

Data supporting the findings of this study are available from the corresponding author upon reasonable request.

To study the nanoscale organization of NRas and BRAF at the PM of melanoma cells, we turned to two-color PALM imaging. For that, NRas was tagged with the photoactivatable red fluorescent protein PAmCherry (Supplementary Fig. S1A; ref. 29) and BRAF was tagged with the photoactivatable GFP (PAGFP; ref. 30). Two color SMLM using these proteins has been verified previously (29), yielded localization uncertainties of approximately 20–30 nm (Supplementary Fig. S1B), and used to study molecular organization for a variety of systems (e.g., refs. 28, 31). In our current experiments, 108T melanoma cells (expressing WT BRAF and WT NRas) were transfected with these labeled proteins and seeded on coverslips, and fixed for imaging. The cells, in their resting state, were imaged in total internal reflection (TIRF) mode, to study the PM, while rejecting much of the background from the cell cytosol (Fig. 1A). Two color PALM images of NRas and BRAF showed pronounced self-clusters of each of these proteins. We quantified the extent of relative self-clustering using univariate PCF [or g(r); ref. 26; see details in the Analyses section of the Materials and Methods]. For BRAF, the PCF curve averaged for multiple cells (Fig. 1B) indicated a level of significant self-clustering over 340 nm [g(0–340nm)>1] and up to ×36 higher than randomness [g(0–20) = 36.3 ± 6.4]; compare green curve with flat black line of g(r) = 1). NRas also showed significant self-clustering (Fig. 1C); however, it was more diffused in comparison with BRAF clustering [g(0–20) = 2.64 ± 0.23 and g(0–600 nm)>1].

Protein abundance (i.e., density) is a second key characteristic of protein organization at PM, and is presented in Fig. 1D for NRas and BRAF. BRAF had a density of 6.4 ± 1.3 molecules/μm², while NRas had a significantly higher density, of 58.2 ± 5.2 molecules/μm² (P << 0.001).

To summarize these results here, and in the rest of the article, we constructed a 2D map (Fig. 1E). In this map, the y-axis represents the average values of g(<100), namely, the extent of molecular self-clustering beyond randomness. The x-axis represents the protein density. These values are shown for individual cells as colored discs, either for BRAF (green) or NRas (red). Note that both axes are logarithmic. The grids are guidelines that were defined for later comparison with additional measurements, as detailed in the rest of the study.

The colocalization between BRAF and NRas was quantified using the BPCF statistics [BPCF (27); Fig. 1F; see Analyses]. Briefly, through this statistics, the data are compared with two useful null hypotheses. The first is a model of no interaction between the proteins, resulting in a flat curve where g₁₂(r) = 1 (26). The second model is known as the “random labeling model” and it indicates strong association of the interacting proteins in mutual clusters (i.e., molecular placement of both proteins is independent of their type). Our data indicate only partial mixing of NRas and BRAF, as experimental curve falls between the boarders of the two models. Normalizing the BPCF curves is useful for averaging the curves across multiple cells and results in our defined “Extent of mixing” (EOM; Fig. 1G). Here, the model of no interaction is represented by EOM(r) = 0, while homogeneous mixing (i.e., close association) is represented by EOM(r) = 1. Again, the results where EOM(r) ranged between 0.2 and 0.4 indicated a relatively low colocalization of BRAF and NRAS.

Upon cell activation with EGF, NRas and BRAF get recruited to the PM (32). There, BRAF gets activated by active NRas, and further conveys the mitogenic signal downstream to MEK. To study the effect of MAPK pathway activation on the nanoscale organization of NRas and BRAF at the PM, we imaged melanoma (108T) cells using two-color PALM after the addition of EGF to their medium (see Materials and Methods). As expected, the imaged cells demonstrated higher densities of NRas (by ∼5-fold) and BRAF (by ∼7-fold) as compared with resting cells (compare Fig. 1D and H). Both proteins seemed more diffused on average (yet not significantly), as found by PCF analyses for BRAF [compare g(0–20) = 27.93 ± 6.9 in Fig. 1I with g(0–20) = 36.3 ± 6.4 in Fig. 1B; P value N.S.], and more so for NRas [compare g(0–20) = 1.36 ± 0.05 in Fig. 1J with g(0–20) = 2.64 ± 0.23 in Fig. 1C; P = 4.81E-05]. These changes resulted in a significant shift of the data for BRAF and NRas in the clustering-density map (Fig. 1L; bold colored lines originate from the coordinates of the averaged results of resting cells in Fig. 1E). Strikingly, BRAF and NRas now showed significantly increased mutual association in clusters, which approached homogeneous mixing in (BRAF) clusters (Fig. 1M and N; compare with the gray arrowhead that represent the EOM for resting cells expressing WT NRas in Fig. 1G).

Taken together, our results suggest a novel mechanism of signaling by NRas and BRAF in clusters. In resting cells, most of these proteins present at the PM reside in segregated clusters. Upon cell activation, NRas and BRAF clusters disintegrate, as their extent of relative self-clustering is reduced. The levels of these proteins increase. Thus, the more diffused organization of these proteins allows their close interaction and effective signaling downstream.

We noted that NRas and BRAF resided at the PM in nanoscale clusters, both before and after EGF stimulation (Figs. 1B, C, I, and J). Such clusters have been described before using immunogold EM (4) and PALM (33). Thus, we wanted to compare the size of these clusters in our data, in copy numbers of NRas and BRAF molecules. We show the cumulative probability of these proteins to reside in clusters of size 2–30 mer. The distribution for both proteins is skewed (having no characteristic size). The majority (∼70%–80%) of the clusters were in dimers, another approximately 5%–10% were trimers, with decreasing fractions of larger clusters (Fig. 1L). The size-distributions shifted to larger clusters upon EGF stimulation (darker green and red lines in Fig. 1O).

The recruitment of NRas and BRAF to the PM, upon EGF stimulation could follow various dynamic mechanisms. For instance, molecules could be delivered to the PM as monomers, or in vesicles. They could enrich preexisting clusters or form new ones. The monomers and clusters could diffuse and interact at the PM, or remain stationary. Thus, we modeled such representative mechanisms (Fig. 2A; models 1–6), and simulated each model using Monte Carlo simulation (with 20 repetitions for each model; see Materials and Methods). Initial conditions were chosen according to the measured densities in our PALM imaging experiments (Fig. 1). Representative initial and final patterns may serve for qualitative evaluation of the models (Fig. 2A; top and bottom rows, respectively). For quantitative evaluation, we analyzed the clustering-density 2D plots, as for the experimental data (Figs. 1E, I, and 2B). Clearly, models 1, 2, and 6 could be ruled out as their g(r)-density curves could not account for the experimental data of neither BRAF, nor NRas. Models 1 and 2 consider the arrival of proteins as monomers to the PM, while model 6 considers the arrival of the molecules into preexisting, static clusters. Model 3 describes best the data for NRas, while model 4 describes best the data for BRAF. These models consider the arrival of molecules in clusters, and either allow their diffusion and aggregation at the PM (model 3) or does not allow diffusion and aggregation of the clusters (model 4). Still, model 5 cannot be ruled out for BRAF. This model considers arrival of molecules to the PM into new clusters that remain static. Thus, we conclude that self-aggregation in either dynamic or static clusters (model 3 for NRas and model 4 or 5 for BRAF) seem to be needed to account for our data.

We next aimed to study the organization of mutant NRas and BRAF at PM. Specifically, we studied the molecular organization in 108T cells overexpressing the melanoma-driving NRas mutants G12D (Fig. 3A–C; Supplementary Fig. S1A), G13D (Fig. 3D–F; Supplementary Fig. S1A) or Q61R (Fig. 3G–I; Supplementary Fig. S1A). In resting cells, NRas levels at the PM were somewhat reduced, while BRAF levels were significantly elevated. Likewise, the extent of NRas self-clustering increased for G12D and G13D (Fig. 3A, B, D, and E). Importantly, for all mutants, the colocalization of NRas and BRAF was increased in these resting cells (Fig. 3C, F, and I; Supplementary Fig. S4A–S4F; clear significance was observed for the G13D and Q61R, while for G12D the increase was not significant). Thus, expression of NRas mutants lead to BRAF organization that resembles those of EGF-activated cells (compare results with Fig. 1E, L, and N). Our results indicate that although mutant NRas becomes more clustered relative to WT NRas at the PM (Fig. 3J), the increased levels of BRAF at the PM could allow for its efficient mixing with the now more clustered NRas. Our results further suggest an unexplored mechanism of mutant NRas-enhanced signaling, by more efficient recruitment of BRAF.

To validate that NRas mutants leads to increased self-clustering, we further imaged endogenous WT and mutant NRas proteins at the PM of multiple resting melanoma cell lines, A375 (WT NRas), 108T (WT NRas), 72T (G12D), 83T (G13R), 12T (Q61R), and 74T (Q61R). Imaging was conducted via dSTORM (34) of NRas, labeled with mouse anti-NRas primary Ab and an anti-mouse secondary labelled with Alexa647. dSTORM imaging has a limited ability to quantify absolute molecular counts due to molecular blinking, antibody aggregation and the existence of multiple fluorophores on each antibody. Thus, we restrict our dSTORM-related analyses here (and later in the text) to a relative comparison of molecular clustering between the different cell lines. As before, WT NRas showed relatively low levels of self-clustering (Fig. 4A and B), while cell lines containing mutated NRas, demonstrated a significantly higher levels of NRas self-clustering (Fig. 4C–F). Thus, our previous of enhanced NRas self-clustering upon expression of NRas mutants does not depend on its overexpression, as needed for the PALM experiments. Notably, in contrast to the dSTORM assay, by PALM imaging we could directly visualize the mutant NRas protein and not average them with the WT proteins.

Our model for NRas signaling at the PM suggests a linkage between NRas and BRAF activity to their dynamic and mutual organization at the PM. We also observe that mutant NRas tends to enhance association with BRAF and thus, promotes mitogenic signaling. Multiple clinical trials currently employ MEKi for reducing such aberrant signaling (35). Thus, we wondered whether MEK inhibitor could affect the organization of NRas (either WT or mutant) and BRAF at the PM. If so, the inhibitor might affect both the WT and oncogenic mutant versions of these proteins.

We started by imaging 108T cells overexpressing mutant NRas (namely, G12D, D13D, and Q61R) and BRAF after treatment of the cells with trametinib, MEKi (Fig. 5). These cells demonstrated enhanced clustering of both NRas and BRAF, relative to resting cells expressing WT NRas (compare Fig. 5A, D, and G with Fig. 1A, E, and G; see also Supplementary Fig. S4B, S4D, and S4F). NRas levels at the PM were significantly reduced (Supplementary Table S4), as for the cells that expressed mutant NRas without MEKi treatment (compare with Fig. 3B, E, and H). In contrast, BRAF PM levels did not change significantly relative to the resting cells expressing WT NRas (compare Fig. 5B, E, and H; green discs and Fig. 1E). Thus, the significant increase in BRAF levels caused by expression of NRas mutants (Fig. 3B, E, and H) was canceled by MEKi treatment.

Importantly, mutant NRas and BRAF were segregated from each other in the MEKi-treated cells (Fig. 5C, F, and I), in contrast to the elevated colocalization that we saw in the absence of MEKi treatment (Fig. 3C, F, and I).

We were interested if MEKi could also affect the nanoscale organization of NRas and BRAF in melanoma cells expressing WT NRas (Supplementary Fig. S5A–S5F). Similar to the results for cells expressing mutant NRas, we observed that the clustering extent of both proteins was enhanced (Supplementary Fig. S5A, S5B, S5D, and S5E). NRas and BRAF also showed colocalization levels similar to those of resting cells (Supplementary Fig. S5C). Thus, MEKi could reverse the effect of the NRas mutants on the nanoscale organization of NRas and BRAF in cell expressing either WT or mutant NRas.

Taken together, these results suggest that trametinib is changing the organization and dynamics of NRas and BRAF on the PM. We suggest that this mechanism is the reduction of these proteins’ ability to interact and form mutual complexes at the PM. This mechanism seems to act in parallel to the proteins’ propensity for self-clustering (which seems to be their default state in resting cells).

The continuous treatment of melanoma with MEKi often results in incurred resistance of the cells and cancer recurrence (36). Thus, we wanted to further test the possible relation of the nanoscale organization of NRas and BRAF in nanoclusters to MEKi resistance. For that, we imaged a melanoma cell line, 109T, which is significantly less sensitive to MEKi than 108T. This cell line has an IC₅₀ of 804.7 nmol/L, compared with the IC₅₀ of 1.4 nmol/L of 108T cells, and expresses the WT forms of NRas and BRAF (just like 108T cells).

In resting 109T cells, both NRas and BRAF remained in pronounced self-clusters (Fig. 6A and B; Supplementary Fig. S5G and S5H) with little mutual association (Fig. 6C). The density of NRas was significantly lower than in resting 108T cells (Supplementary Fig. S5I, magenta line). However, upon EGF activation, NRas remained in self-clusters (Fig. 6D and E; Supplementary Fig. S5J and S5K)—much like in cells expressing NRas mutants (Fig. 3A, D, and G), and significantly different than in EGF-activated 108T cells (Fig. 1H). At the same time, the density of BRAF clusters grew significantly (Fig. 6E, green line and compare green bars in S5I and S5L). While the EOM between NRas and BRAF in these cells did not change relative to resting 109T cells (Fig. 6F), we could identify many more coclusters of NRas and BRAF at the PM of these cells (Fig. 6D, yellow clusters; compare with 6A). We quantified the fraction of the pixels containing BRAF and NRas detections (yellow pixels in Fig. 6D) relative to NRas-containing pixels (red pixels, respectively). We found that NRas and BRAF colocalization in mutual clusters was significantly higher in resting 109T cells than in 108T cells, and was comparable in extent with 108T cells expressing NRas mutants (Fig. 6G; note significance is relative to resting 109T cells in leftmost bar, and unless specified otherwise). Upon EGF activation, this fraction of coclustered molecules increased approximately 4-fold relative to resting 109T cells, and significantly exceeded that of EGF-activated 108T cells and of 108T cells expressing NRas mutants.

We conclude that resistance to MEKi might involve a dynamic organization of NRas and BRAF in pronounced clusters at the PM that can come together upon EGF activation, as in cells expressing mutant NRas. This further suggests the existence of a shared mechanism of signaling regulation in NRas and BRAF clusters under conditions that mediate aggressive and persistent melanoma. Our study sets the stage for finding and characterizing such a mechanism.

To validate the signaling effects of EGF stimulation or mutant NRas overexpression on the mitogenic pathway in our experiments, we performed Western blots of pERK (and ERK) in the 108T cells (Fig. 7A). Indeed, the overexpression of mutant NRas (either G12D, G13D, or Q61R) resulted in enhanced pERK levels, as compared with WT NRas resting cells (Fig. 7A and B). Moreover, pERK levels were also comparable or higher relative to EGF-activated cells. As expected, MEKi treatment of the cells, transfected with mutant NRas, completely abrogated pERK levels. These results correlate between the nanoscale organization of BRAF and either WT or mutant NRas with the downstream activation of the mitogenic pathway.

Our results suggest a potentially new mechanism of MEKi function in melanoma cells carrying NRas mutations, but lacking MEK mutations. Thus, we further conducted a proliferation assay to make sure that the growth of these cells were affected by MEKi treatment. Indeed, for the tested cell lines, the number of cells that underwent MEKi treatment did not increase or even decreased over 96 hours while the number of cells without treatment increased by 2- or 3-fold during this time (Supplementary Fig. S6A). We further characterized the transformation state of our used cell lines, in comparison with a wide variety of other melanoma cells lines with various genetic backgrounds (Supplementary Table S1). We found that the 108T cell line showed elevated expression of the transformation marker ABCB5 (Supplementary Fig. S6B and S6C), which was above the average of 25 cell lines for the 108T cells, while 12T cells were slightly below average (Supplementary Fig. S6D). Thus, the cell lines used in this study (and esp. 108T) showed a transformed phenotype that was sensitive to MEKi treatment.

Next, we wanted to directly show that the clustered state of NRas can locally diminish its capability for signaling via phosphorylation of BRAF. For that, we performed two-color dSTORM of phosphorylated BRAF (pBRAF) and NRas in the same cells and in single molecule detail. 108T cells were stimulated with EGF for 15 minutes, fixed and stained for NRas and pBRAF using primary antibodies and secondary antibodies stained with Atto488 and Alexa647, respectively. We found that NRas and pBRAF molecules showed both in self-clusters and in a diffused pattern (Fig. 7C, top; Supplementary Fig. S7A). We next distinguished the NRas molecules in either monomers or clusters of various sizes using the DBSCAN algorithm (coded in Matlab; ref. 25). We then studied the abundance of pBRAF at NRas clusters, as a function of their size. Specifically, individual pBRAF molecules were assigned by their proximity (<70 nm) to NRas monomers and or to individual clusters (Fig. 7D). Next, we plotted the relative number of NRas-associated pBRAF in NRas clusters, as a function of their size (in copy number; Fig. 7E; absolute counts of NRas-proximal pBRAF are shown in Supplementary Fig. S7B). Strikingly, NRas monomers mixed significantly more with pBRAF molecules relative to NRas in self-clusters (Fig. 7E; bar for monomers is highest among all cluster sizes, and bars get shorter with cluster size). Importantly, our results were insensitive to the chosen threshold of NRAF-pBRAF proximity (tested for thresholds of 35, 50, 70, 85, and 105 nm; Supplementary Fig. S7C–S7F).

Our results indicate that clustered NRas can associate less with BRAF, and thus has a lower ability to signal downstream. On the other hand, diffusion and mixing of BRAF and NRas at the PM promote signaling and cell activation. Thus, our observed spatial redistribution of NRas and BRAF at the PM directly affects their signaling.

In this study, we applied two-color PALM to resolve the organization of NRas and BRAF at the PM of melanoma cells. Imaging showed that NRas and BRAF reside in pronounced clusters and with low interaction in resting 108T cells. EGF activation of the cells caused the increase of NRas and BRAF levels, along with their more diffused organization at the PM. This reorganization enabled their enhanced interaction and signaling. Expression of oncogenic mutants of NRas in resting cells led to a similar reorganization and enhanced signaling. MEKi treatment of cells led to reduced NRas protein levels at the PM. This treatment was also able to reverse the reorganization caused by NRas mutants upon their expression in such cells. We further found that NRas and BRAF colocalization in mutual clusters was significantly higher in resting MEKi-resistant (109T) cells relative to nonresistant (108T) cells, and was comparable in extent with such nonresistant cells expressing NRas mutants.

Taken together, our results indicate novel mechanisms for NRas-BRAF signaling at the PM through changes in their self and mutual organization (Fig. 7F). NRas and BRAF clusters should be more dispersed, along with the increase of protein levels, for enhanced NRas signaling upon EGF activation. While the expression of NRas oncogenic mutants increases NRas clustering, it also elevates the levels and dispersion of BRAF at the PM. This allows NRas mutants to efficiently recruit BRAF and productively signal downstream, as indicated by elevated pErk levels (Fig. 7A–E). Treatment with MEKi reverses BRAF levels and organization at the PM (the reverse shift in Fig. 7F). It results in NRas-segregated BRAF clusters, as in resting cells, and thus inhibits the signaling (Fig. 7A–E).

Traditionally, studies have addressed Ras and BRAF activity using various biochemical assays. Specifically, Western blotting and immunoprecipitation are insensitive to the exact organization of these proteins in clusters. Moreover, in such assays, the results are averaged over many millions of cells and a staggering number of individual complexes. While diffraction-limited microscopy often achieves single-cell resolution, it still averages the signal of thousands of protein complexes in each pixel (36).

More recently, high-resolution imaging using immunogold-labeling TEM has revealed that Ras reside in small (5–8 mer) nanoclusters, with a size < 10 nm (37, 38). Notably, such labeling in TEM often suffers from poor labeling efficiency and multiple clustering artifacts (39). It also requires the ripping of the cells from a coverslip before labeling, which might alter the intact organization of the proteins under study (40). Such artifacts largely restrict the ability of this method to resolve molecular organization in intact cells.

In contrast, studies involving superresolution microscopy have emphasized the role of Ras (41, 42) and BRAF dimerization (ref. 6; following biochemical studies; ref. 43). Nan and colleagues employed PALM and molecular complementation in their study. Still, these studies have also reported on larger scale self-clusters of the multiple Ras and RAF isoforms. Likewise, van Lengerich and colleagues have shown EGF-induced clustering of the mitogenic receptors HER3 using dSTORM (44). Still, the mutual organization of NRas and BRAF has not been studied, nor were these proteins studied under the range of conditions studied here; esp. in human melanoma cells.

In contrast to these previous studies, our imaging approach has revealed relatively higher-order clusters of NRas and BRAF. Both proteins demonstrated a size distribution of clusters from dimers to approximately 30 proteins (Fig. 1O). Importantly, we could follow changes in this organization under various conditions, including EGF-induced cell activation, expression of oncogenic NRas mutations and after cell treatment with relevant clinical inhibitor. Mutant NRas resided in more pronounced self-clusters relative to WT NRas in multiple melanoma cell lines, yet could associate more with the clustered, and more abundant BRAF in 108T cells.

Importantly, cells treatment with the MEKi trametinib had a pronounced effect on NRas and BRAF self-clustering and their spatial overlap. It diminished the density of these proteins at the PM, and increased their extent of self-clustering (Fig. 5; Supplementary Fig. S3). MEKi treatment also restored the nonoverlapping self-clusters of NRas and BRAF. Thus, MEKi treatment could restore NRas and BRAF PM levels and organization to their extent as in resting cells that express WT NRas. Because MEK acts downstream of NRas, this effect of its inhibitor exceeds the canonical mechanistic model by which this drug interferes with the catalytic site of MEK.

Taken together, our results raise the possibility of clinical treatment with MEKi of melanoma cells carrying NRas mutations, but lacking MEK mutations. Still, the underlying mechanisms of melanoma in treated patients seem to be complex. For instance, recent clinical updates report on gained resistance of melanoma cells to BRAF and MEK inhibitors in patients (45). Resistance mechanisms involve both genetic and epigenetic mechanisms. Mechanisms related to BRAF include the appearance of BRAF V600 mutations, gene amplification, splice variants, RAF–RAF dimerization, etc. (see refs. 44 and 46 for review). The molecular mechanisms that facilitate drug resistance are of prominent clinical interest. Our results show that NRas and BRAF nanoscale patterning in MEKi-resistant cells generally resemble the organization of these molecules in cells that express oncogenic NRas mutations. This suggests a mechanism of NRas and BRAF patterning that is common to signal regulation in both cases of aggressive cancer progression due to either expression of melanoma-driver mutations or drug resistance.

Because MEKi treatment inhibits aberrant mitogenic signals downstream of BRAF, treatment of melanoma cells carrying NRas mutations, but lacking MEK mutations may still be effective in the face of such BRAF-dependent mechanisms (although such clinical treatment has not shown significant success so far; ref. 47). Additional complications that are involved in cancer treatment with chemotherapy include the existence of residual dormant cancer cells and micrometastases (46). Thus, our suggested MEKi treatment of melanoma cells carrying NRas mutations clearly requires careful testing and validation in model animals.

No disclosures were reported.

O. Yakovian: Conceptualization, software, formal analysis, investigation, visualization, methodology, writing-original draft. J. Sajman: Resources, data curation, investigation, project administration. R. Arafeh: Conceptualization, resources, methodology, writing-original draft, writing-review and editing. Y. Neve-Oz: Software. M. Alon: Resources, project administration. Y. Samuels: Supervision, writing-original draft, writing-review and editing. E. Sherman: Conceptualization, supervision, funding acquisition, methodology, writing-original draft, writing-review and editing.

This research was supported by grants no. 1417/13, 1761/17, and no. 1937/13 from the Israeli Science Foundation (to E. Shermon). Y. Samuels is supported by the Israel Science Foundation grant no. 696/17, the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement no. 754282), the ERC (CoG- 770854), MRA (#622106), Rising Tide Foundation, Henry Chanoch Krenter Institute for Biomedical Imaging and Genomics, Estate of Alice Schwarz-Gardos, Estate of John Hunter, Knell Family, Peter and Patricia Gruber Award, and the Hamburger Family.

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