Cooperative polarization of MCAM/CD146 and ERM family proteins in melanoma

Abstract

The WRAMP structure is a protein network associated with tail-end actomyosin contractility, membrane retraction, and directional persistence during cell migration. A marker of WRAMP structures is melanoma cell adhesion molecule (MCAM) which dynamically polarizes to the cell rear. However, factors that mediate MCAM polarization are still unknown. In this study, BioID using MCAM as bait identifies the ERM family proteins, moesin, ezrin, and radixin, as WRAMP structure components. We also present a novel image analysis pipeline, Protein Polarity by Percentile ("3P"), which classifies protein polarization using machine learning and facilitates quantitative analysis. Using 3P, we find that depletion of moesin, and to a lesser extent ezrin, decreases the proportion of cells with polarized MCAM. Furthermore, although copolarized MCAM and ERM proteins show high spatial overlap, 3P identifies subpopulations with ERM proteins closer to the cell periphery. Live-cell imaging confirms that MCAM and ERM protein polarization is tightly coordinated, but ERM proteins enrich at the cell edge first. Finally, deletion of a juxtamembrane segment in MCAM previously shown to promote ERM protein interactions impedes MCAM polarization. Our findings highlight the requirement for ERM proteins in recruitment of MCAM to WRAMP structures and an advanced computational tool to characterize protein polarization.

FIGURE 1:

Copolarization of MCAM, myosin, and F-actin. Confocal imaging of WRAMP structures, illustrating MCAM polarization with enrichment of cortical F-actin and myosin-IIB. For fluorescence imaging experiments, WM239a melanoma cells were plated on untreated glass at low confluency to minimize contacts between cells and serum-starved overnight prior to fixation. (A) Cells show IF signal from mouse anti-MCAM (AF488), rabbit anti-myosin-IIB heavy chain (AF594), and F-actin (phalloidin, AFP405). (B) Cells show IF signal from anti-MCAM, rabbit anti-phosphorylated-Ser19 myosin light chain 2 (pMLC, AF594), and phalloidin. Confocal Z-stacks were collected with step size 200 nm. For each image channel, the full cell is shown as a maximum intensity projection. An enlarged orthogonal view (cross-section) generated along the white dashed line is shown below. Images are representative of 20−30 cells per condition across two to three biological replicates. Scale bars (full images) = 10 μm. Scale bars (cross-section) = 3 μm in both dimensions. Imaging controls are shown in Supplemental Figure S1.

FIGURE 2:

BioID experimental design. (A) Constructs used for MCAM bait and controls. (B) Western blots probed with anti-HA and anti-MCAM antibodies show stable expression of BioID constructs (left panel), and comparable expression of MCAM-BirA*-HA to endogenous MCAM (right panel). (C) Detection of protein biotinylation with HRP-coupled streptavidin, in stable cell lines grown for 72 h in biotin-free RPMI + 10% FBS followed by treatment with or without 50 µM biotin for 16 h. Streptavidin-HRP blots showing biotinylation activity and silver stains of samples used in MS experiments are shown in Supplemental Figure S3, A and B. Blot images were cropped to the right and left of the region shown to remove protein ladders and extra lanes, and along the dashed lines to remove a lane with ladder. (D) Confocal fluorescence images showing close alignment between MCAM-BirA*-HA (anti-HA) and total MCAM (anti-MCAM), which is absent with BirA*-HA-NES and Lyn₁₁-BirA*-HA controls. White arrows indicate the location of polarized MCAM and MCAM-BirA*-HA. (E) Localization of biotinylated proteins detected using AlexaFluor 488 streptavidin (SAv) relative to HA-tagged BioID constructs. Cells were treated with 50 µM biotin for 16 h prior to fixation unless noted otherwise. Full X-Y images are shown as maximum intensity projections of confocal Z-stacks. Enlarged orthogonal views (cross-section) along the dashed white lines are shown below or at the right. Images are representative of approximately 20 cells per condition across two biological replicates for panel D and 10 cells in 1 replicate for panel (E). For (E), contrast was linearly scaled between the same lower and upper limits for each channel. Scale bars (full images) = 10 μm. Scale bars (cross-section) = 3 μm in both dimensions.

FIGURE 3:

BioID identifies ERM proteins in the MCAM proximity proteome. (A) Proteins isolated from cells expressing MCAM-BirA*-HA were compared with cytosolic BioID (BirA*-HA-NES) and no-construct controls. SAINT scores were plotted against the log2 fold change of MCAM-BirA*-HA over 2 virtual controls (two highest spectral counts). Points in red identify 416 protein groups with SAINT scores corresponding to FDR < 0.05 (1,815 total protein groups). The protein group labeled “MCAM(BirA*)” combines sequences for endogenous MCAM and MCAM-BirA*-HA. (B) Volcano plot showing enrichment of MCAM-BirA*-HA samples over the plasma membrane BioID (Lyn₁₁-BirA*-HA) control. P values were calculated for 1,502 high-confidence protein groups using a two-tailed Student’s t test for log2-transformed normalized iBAQ protein intensities as described in Supplemental Methods. Points in red correspond to those enriched in MCAM-BirA*-HA over the cytosolic BioID and no-construct controls, shown in panel (A). Log2 fold changes for proteins only detected in 1 Lyn11-BirA*-HA sample, which could not be compared by t test, are arbitrarily plotted at -0.5. (C) Log-ratio/log-average (MA) plot corresponding to data in panel B, showing log2 fold change vs. averaged mean iBAQ intensities for MCAM-BirA* and Lyn₁₁-BirA*-HA. Points in orange are statistically significant after adjusting p-values using permutation-based correction (FDR < 0.05, s0 = 0 or s0 = 0.5). Proteins in blue highlight ERM proteins and a previously identified WRAMP component, NIBAN2. (D) Normalized iBAQ intensity values for ERM proteins detected with each BioID construct. (* indicates adjusted p value < 0.025, and ** indicates adjusted p-value < 0.005, for s0 = 0). (E) Gene symbols for proteins statistically significant in MCAM-BirA*-HA over Lyn₁₁-BirA*-HA. Proteins are listed vertically by decreasing averaged mean iBAQ intensities for MCAM-BirA* and Lyn₁₁-BirA*-HA. Those in bold were also significant in MCAM-BirA*-HA compared with BirA*-HA-NES and no-construct controls. Data in panels (A)−(E) represent three biological replicates of MCAM-BirA*-HA, Lyn11-BirA*-HA, and no construct samples, and two replicates of BirA*-HA-NES.

FIGURE 4:

Copolarization of ERM proteins and MCAM. (A−H) Widefield and (I−L) confocal immunofluorescence images showing the localization of ERM proteins relative to endogenous MCAM in WM239a cells. (A−C) Immunostaining of endogenous (A) MSN, (B) EZR, and (C) RDX against endogenous MCAM and F-actin/phalloidin. (D) Detection of phosphorylated ERM (pERM) using a phosphospecific antibody (T567/T564/T558 for EZR/RDX/MSN). For A−D, representative polarized cells were selected from several thousand cell images that were quantitatively analyzed in Figures 7 and 9. (E−G) Confirmation of ERM protein co-polarization with MCAM in WM239a cells expressing (E) MSN-mCherry, (F) EZR-mCherry, or (G) RDX-mCherry. Cells were fixed 48 h after transient transfection and stained with anti-MCAM antibody and phalloidin. Constructs were detected based on mCherry fluorescence which was preserved during fixation. Images are representative of at least 12 cells for each construct across two biological replicates. (H) Controls showing signal from cells transfected with mCherry empty vector or no plasmid (“mock” transfection, inset). White arrows denote regions with polarized MCAM. The mock transfection image is shown using the same linear contrast adjustment as the highest contrast mCherry construct. (I−L) Confocal Z-stacks shown as maximum intensity projections, with orthogonal views (cross-section) along dashed white lines. Cells were immunostained for endogenous MCAM and (I) MSN, (J) EZR, (K) RDX, (L) pERM, and F-actin/phalloidin. Images are representative of 16–24 cells per condition across 2 (I–K) or 3 (L) biological replicates. Scale bars (full images) = 10 μm. Scale bars (cross-section) = 3 μm in both dimensions.

FIGURE 5:

Quantitative image analysis pipeline for classification and characterization of protein polarization. (A) Overview of the 3P image analysis pipeline. Cells are segmented based on MCAM IF signal, and regions with bright signal (> 80^th percentile pixel intensity) are isolated as individual objects. The object with the greatest integrated intensity is selected as the brightest object. A subset of manually labeled cells is then used to train and test a ML classifier, based on selected features that are either standard from CellProfiler and MATLAB or custom-defined based on the brightest object. An iterative approach (gold box) is used for classifier training, by adding test images that are misclassified to the training set and repeating the cycle twice (see Supplemental Methods). (B) Violin plot showing the distributions and 25%, 50%, and 75% quantiles for the example of “median distance”, a custom feature that measures the median distance between the cell center and pixels in the brightest object. The results show overlapping distributions for polarized (N = 939) and nonpolarized (N = 1,238) cells based on manual scoring. (C) Examples of cells from three classes used to describe protein polarization. “Both Ends” and “None” classes were combined into a single “Nonpolarized” cell category after ML classification. Scale bars = 10 μm. (D) Confusion matrices after the second and final iteration using gbm (left) and extraTrees (center) classification models. Classifications from gbm and extraTrees were then combined to yield greater specificity, by requiring both classifiers to identify a cell as polarized (right). Classifications are shown for cells with a clear polarized or nonpolarized phenotype from a test set of cells separate from those used in iterative training. Cells with an ambiguous manual classification were omitted (Supplemental Methods). (B and D) The data are for experiments shown in Figure 6 and Supplemental Figures S7D−S9.

FIGURE 6:

Depletion of ERM proteins decreases MCAM polarization. (A) Western blots confirm specific knockdown of targeted ERM proteins. These blots are shown without vertical cropping in Supplemental Figure S9F. (B) Effects of siRNA knockdown of MSN, EZR, and RDX on the percentage of WM239a cells with polarized MCAM. The 3P pipeline was used to classify cells with polarized MCAM after treatment with each siRNA. For each replicate, the percentages of cells with polarized MCAM in each condition were normalized to cells treated with scrambled control siRNA by dividing by the percentage of polarized cells in the control siRNA condition. The average percentage of polarized cells for control replicates was 36%. Cells examined were from three biological replicates with at least two replicate coverslips, and totaled to 6,585 (scrambled siRNA control), 3,189 (siRNA-MSN), 3,227 (siRNA-EZR), and 4,316 (siRNA-RDX) cells (** indicates p-value < 0.025, * p-value < 0.05 for pairwise two-tailed Welch’s t tests after Holm-Bonferroni correction). For one replicate, one condition was an upper outlier based on Dixon’s Q test, and samples for all siRNAs were excluded. An additional MSN knockdown replicate was excluded due to differences in IF methods and a RDX knockdown with N < 200 cells (described further in Supplemental Methods). (C and D) The percentages of cells with polarized MCAM that also showed copolarization of (C) MSN or (D) EZR, based on coimmunostaining. Copolarization was scored manually by visible overlap of enriched MSN or EZR with MCAM, after classifying cells for polarized MCAM using 3P, as described in Supplemental Methods. Cells examined were gathered from two to three biological replicates and totaled 437 (control) and 147 (siRNA-MSN) cells in panel (C), and 258 (control) and 249 (siRNA-EZR) in panel (D). (ns indicates p-value > 0.05 for two-sample Wilcoxon test in panel (C); significance was not calculated in panel (D), because the control condition had only two replicates). Bars and errors show mean values and standard deviations for biological replicates.

FIGURE 7:

Quantitative colocalization analysis identifies a cell subpopulation with ERM proteins closer to the cell periphery. Three-color IF widefield images of WM239a cells were labeled with anti-MCAM, phalloidin, and endogenous MSN, EZR, RDX, or pERM. The data are from experiments shown in Figure 4, A−D. (A) Venn diagrams showing the intersection of cells classified with polarized MCAM or ERM proteins using 3P. Percentages are shown relative to the total number of cells with polarized MCAM. (B) The percentages of cells with polarized MCAM that also show copolarized ERM proteins. Cells were first classified for polarized MCAM using 3P, and then manually assessed for copolarization of MSN, EZR, RDX or pERM, as described in Supplemental Methods. Numbers of cells examined totaled 1,291 cells costained for MCAM + MSN (5 replicates), 856 costained for MCAM + EZR (4 replicates), 419 costained for MCAM + RDX (2 replicates), and 1,096 costained for MCAM + pERM (4 replicates). (* indicates p-value < 0.05 for Kruskal-Wallis test followed by Dunn test with Holm-Bonferroni p-value adjustment). (C and D) Modified Manders’ Colocalization Coefficients for MCAM and MSN, EZR, or pERM in manually curated copolarized cells (see Supplemental Methods). Calculations of (C) MCC1 and (D) MCC2 defined MCAM as the first channel. The graphic illustrates MCC1 and MCC2 calculations in cells costained for MCAM + MSN. Boxplots show the distribution of single-cell values for each biological replicate, and individual data points are offset along the X-axis to show the distribution of cells across all replicates, where different colors correspond to different experiments. (E) Distributions of the distance between the weighted centroid of the brightest objects in MCAM and ERM channels. The distance was normalized by dividing by the maximum Feret diameter of each cell. Positive values indicate cells where ERM proteins were closer to the cell periphery than MCAM, relative to the cell center. Dashed lines indicate a normalized distance of ± 0.04, used to select outlier cells for manual evaluation. (F−H) Widefield images of cells showing separations between brightest object centroids, with (F) MSN, (G) EZR, or (H) pERM located closer to the cell periphery than MCAM. Scale bars = 10 µm.

FIGURE 8:

ERM polarization precedes MCAM at the cell periphery. (A−C) Live imaging was carried out using cells stably expressing MCAM-GFP and transiently transfected with LifeAct-mTagBFP + (A) MSN-mCherry, (B) EZR-mCherry, or (C) RDX-mCherry (Supplemental Movies S1−S3). Kymographs and selected movie frames show the dynamics of proteins polarizing to the cell periphery. Black arrowheads indicate frames shown below each kymograph. The cyan line in the full cell image shows the kymograph axis and the white box shows the region enlarged in frames below. White arrows indicate arrival of the ERM signal at the cell periphery, preceding the arrival of MCAM signal. Overlays of the MCAM and ERM signals for kymographs and selected movie frames are shown in Supplemental Figure S10. Movie frames were collected every two minutes, and vertical scale bars to the left of each kymograph indicate 6 min = 3 frames. Horizontal scale bars = 5 µm. A total of 22, 26, and 26 cells across two biological replicates were analyzed for MCAM-GFP + MSN-mCherry, EZR-mCherry, and RDX-mCherry, respectively.

FIGURE 9:

Deletion of the ERM binding motif in MCAM blocks copolarization of MCAM and ERM. Three-color IF widefield images of WM239a cells were labeled with anti-MCAM, phalloidin, and endogenous MSN, EZR, RDX or pERM. The data for the naïve cells are also shown in Figure 4, A−D and Figure 7. (A) Polarized MCAM quantified in naïve vs. clonal MCAM_ΔKKGK cell lines engineered using CRISPR/Cas9. The 3P pipeline was used to classify cells from five biological replicates, totaling 7,642 (Naïve), 6,154 (Clone 1), and 6,574 (Clone 2) cells. For each replicate, the percentages of cells with polarized MCAM in each cell line were normalized to the naïve control. The average percentage of polarized cells for naïve replicates was 50%. (B) Western blots showing expression of MCAM and ERM proteins in naïve and knock-in cell lines. The uncropped blots are shown in Supplemental Figure S11. (C) Quantification of Western blots show mean and standard deviations of four technical replicates each from two biological replicate lysates. Integrated band intensities were normalized to GAPDH loading controls and divided by the mean of the naïve condition. (D) Kernel density plot showing mean MCAM intensities in individual cells normalized by the median intensity of naïve cells. The results show reduced MCAM expression on a single-cell level. Data show results for one replicate, which are representative of 4 other replicates. Numbers of cells totaled 1,878 (Naïve), 1,253 (Clone 1), and 1,359 (Clone 2). The X-axis is cropped, and 56 cells with a normalized intensity > 3 are not shown. (E) Percentage of cells with polarized MCAM normalized to naïve cells vs. mean MCAM intensity. Each curve represents 1 biological replicate; colors correspond to cell lines as indicated in panel (D). Cells were binned by normalized mean MCAM intensities, and the percentage of MCAM-polarized cells in each bin was divided by the overall percentage of polarized cells in the naïve cells for that replicate. The result shows that polarization of MCAM_ΔKKGK is lower than wild type, independent of expression level. Cells examined total 5,663 (Naïve), 4,851 (Clone 1), and 5,594 (Clone 2). Bins with less than 50 cells were omitted from the analysis. One replicate was omitted from the plot because only 1 cell line had > 50 cells per bin. (F−H) Percentage of cells with polarization of (F) MSN, (G) EZR, and (H) pERM in each cell line, normalized to the naïve control. For naïve replicates, the average percentage of cells with polarized MSN, EZR, and pERM was 53%, 55%, and 54%, respectively. Protein polarization was classified by machine learning using the 3P pipeline. The results suggest that ERM polarization is stabilized by interactions with MCAM. Cells were examined in four to five biological replicates, totaling 2,936 (Naïve), 2,435 (Clone 1), and 2,476 (Clone 2), for MCAM + MSN; 1,688 (Naïve), 1,587 (Clone 1), 1,713 (Clone 2) for MCAM + EZR; and 2,224 (Naïve), 1,721 (Clone 1), 1,949 (Clone 2) for MCAM + pERM. Two replicates of the naïve condition in (G) and one replicate of Clones 1 and 2 were quantified from a single coverslip per condition. For panels (A) and (F−H), bars and errors show mean values and standard deviations for biological replicates, and ** indicates p-value < 0.002, and * indicates p-value < 0.05, for pairwise Welch’s two-tailed t-test with Holm-Bonferroni correction. (I) A model for copolarization of MCAM and ERM proteins, proposing ERM recruitment to the membrane followed by MCAM via interactions with the juxtamembrane (KKGK) ERM-binding motif.