Probing the Heterogeneity of Protein Kinase Activation in Cells by Super-resolution Microscopy

Abstract

Heterogeneity of mitogen-activated protein kinase (MAPK) activation in genetically identical cells, which occurs in response to epidermal growth factor receptor (EGFR) signaling, remains poorly understood. MAPK cascades integrate signals emanating from different EGFR spatial locations, including the plasma membrane and endocytic compartment. We previously hypothesized that in EGF-stimulated cells the MAPK phosphorylation (pMAPK) level and activity are largely determined by the spatial organization of the EGFR clusters within the cell. For experimental testing of this hypothesis, we used super-resolution microscopy to define EGFR clusters by receptor numbers (N) and average intracluster distances (d). From these data, we predicted the extent of pMAPK with 85% accuracy on a cell-to-cell basis with control data returning 54% accuracy (P < 0.001). For comparison, the prediction accuracy was only 61% (P = 0.382) when the diffraction-limited averaged fluorescence intensity/cluster was used. Large clusters (N ≥ 3) with d > 50 nm were most predictive for pMAPK level in cells. Electron microscopy revealed that these large clusters were primarily localized to the limiting membrane of multivesicular bodies (MVB). Many tighter packed dimers/multimers (d < 50 nm) were found on intraluminal vesicles within MVBs, where they were unlikely to activate MAPK because of the physical separation. Our results suggest that cell-to-cell differences in N and d contain crucial information to predict EGFR-activated cellular pMAPK levels and explain pMAPK heterogeneity in isogenic cells.

Keywords: Bayesian modeling; EGFR; MAPK; cell-to-cell heterogeneity; super-resolution microscopy.

Figure 1

Cell-to-cell heterogeneity of HCC1143 breast cancer cells in their MAPK signaling response subsequent to EGF stimulation. Each histogram was built from single-cell phospho-MAPK intensity levels (arbitrary units (a.u.)) from three independent experiments (9 confocal tiles per image, N ≥ 350 cells per condition for each experiment). The full width half-maximum (fwhm) of the anti-phospho-MAPK intensity is a measure of cellular heterogeneity. Typical micrographs are shown.

Figure 2

Super-resolution microscopy of EGFR clusters. EBSQ bound to EGFR was imaged and super-resolved by gSHRImP based on QD blinking. A receptor dimer (A–C) or trimer (D–F) appears as a blurry spot expanding about 4–5 actual camera pixels (100 nm/pixel) in diameter when imaged by diffraction-limited microscopy (purple). For visual guidance, we show the mean fluorescence intensity levels corresponding to the individual QDs by red dotted lines in QD blinking traces (A or D). Both traces have the background subtracted. The white overlay images in (B) and (E) represent the corresponding Gaussian point-spread-functions (PSFs) as determined via the gSHRImP algorithm. Please note that gSHRImP PSFs are not intensity-normalized. The final super-resolved images are generated by determining the centers of the single-molecule PSFs and are shown in zoomed-in micrographs in (C) for the dimer (EGFR molecules are 24 nm apart) and (F) for the trimer (EGFR distances are 92, 107, and 116 nm, respectively). Yellow circles indicate individual EGFR molecule positions with the circle centers positioned at the calculated centroid positions of each EGFR molecule. Scale bars are 50 nm.

Figure 3

Cell-by-cell analysis of EGFR clusters. (A) Number of resolved EGFR clusters under different treatment conditions. Clusters were categorized into EGFR monomers, dimers, trimers, and oligomers as determined by counting the number of receptors per resolved cluster. Oligomers formed of more than 5 or 6 receptors could not always be resolved. (B) Trimers and fully resolvable multimers (i.e., clusters with N = 4–6; from here onward referred to as “N > 3” multimers) were analyzed for their intracluster distances between EGFR molecules. We categorized them as either d ≤ 50 nm or d > 50 nm. The relative numbers in those cluster categories change upon treatment. (C) Binary classification of each HCC1143 cell in each treatment condition into a “low” and “high” pMAPK intensity class. Data shown in all panels were obtained from 46 analyzed cells and 2164 clusters acquired in two independent experiments. The classes are significantly different; P < 0.0001 (unpaired two-tailed t-test with Welch’s correction assuming unequal standard deviation).

Figure 4

BLC-based prediction of cellular MAPK phosphorylation. Prediction uses the number of EGFR molecules per cluster N as the only input parameter (A and B) or both N and the intracluster distances d as input parameters (C and D). (A, C) BLC prediction performance of the training and validation sets as a function of the number of covariates used. (B, D) Graphical representation of the regression weights assigned to each covariate. The magnitude of each weight gives a measure of how informative each covariate is in predicting the class membership. A positive weight implies that large values of that covariate are associated with the “high” class, whereas a negative weight indicates that large values of that covariate are associated with the “low” class. The weights are obtained by averaging over the weights obtained during leave-one-out cross-validation with all covariates. The 95% confidence intervals of the regression weights are plotted. The Bayesian analysis has been performed on imaging data acquired in two independent experiments.

Figure 5

TEM analysis of EGFR clusters. (A) Direct visualization and subcellular localization of EGF-induced EGFR cluster formation in HCC1143 cells. Black arrows point toward the limiting membrane of the MVB, while white arrows point toward ILVs; scale bar is 100 nm. (B) Partitioning between MVB limiting membranes or ILV membranes for dimers, trimers, and multimers across all measured average intracluster distances; a box and whisker plot is shown with whiskers indicating minimum/maximum values. Localization on the limiting membrane of MVB becomes dominant with increasing average intracluster distance and number of receptors per cluster. Localization on the limiting membrane was found significantly different than localization on ILVs for trimers (P = 0.0247) using an unpaired two-tailed t-test with Welch’s correction assuming unequal standard deviation.