Bayesian total internal reflection fluorescence correlation spectroscopy reveals hIAPP-induced plasma membrane domain organization in live cells

Abstract

Amyloid fibril deposition of human islet amyloid polypeptide (hIAPP) in pancreatic islet cells is implicated in the pathogenesis of type II diabetes. A growing number of studies suggest that small peptide aggregates are cytotoxic via their interaction with the plasma membrane, which leads to membrane permeabilization or disruption. A recent study using imaging total internal reflection-fluorescence correlation spectroscopy (ITIR-FCS) showed that monomeric hIAPP induced the formation of cellular plasma membrane microdomains containing dense lipids, in addition to the modulation of membrane fluidity. However, the spatial organization of microdomains and their temporal evolution were only partially characterized due to limitations in the conventional analysis and interpretation of imaging FCS datasets. Here, we apply a previously developed Bayesian analysis procedure to ITIR-FCS data to resolve hIAPP-induced microdomain spatial organization and temporal dynamics. Our analysis enables the visualization of the temporal evolution of multiple diffusing species in the spatially heterogeneous cell membrane, lending support to the carpet model for the association mode of hIAPP aggregates with the plasma membrane. The presented Bayesian analysis procedure provides an automated and general approach to unbiased model-based interpretation of imaging FCS data, with broad applicability to resolving the heterogeneous spatial-temporal organization of biological membrane systems.

Figure 1

Bayesian ITIR-FCS approach. High temporal resolution fluorescence images of the cell membrane are acquired using TIRF microscopy. Automated blocking is used to compute TACFs and their independent associated noise estimates from the raw intensity traces recorded at each pixel, which serve as inputs to the Bayesian model selection framework. Resulting spatial-temporal maps of inferred models and associated parameters are produced, including diffusion coefficients and number densities, providing information on the spatial-temporal heterogeneity of the cellular process of interest. Time-lapse model and parameter spatial maps are produced by the procedure, where the temporal sequence of model probabilities and parameter estimates are obtained from sufficiently long acquisition times (T_aq) to ensure proper noise estimation and unbiased inference, whereas their temporal spacing (ΔT) is chosen to resolve the time-dependent cellular process of interest. To see this figure in color, go online.

Figure 2

Automated blocking yields accurate noise estimates for unbiased multiple hypothesis testing of spatially resolved FCS data. (A) (Top) Schematic showing definitions of block times required for automated application of the blocking procedure. (Middle and bottom) Sample blocking curves that respectively pass and fail the blocking test. (B) Bayesian analysis of simulated compartmentalized two-component, two-dimensional diffusion with respective diffusivities D₁ = 4 μm²/s and D₂ = 0.4 μm²/s outside and inside the domains, respectively, and decreasing noise levels (left to right; number of frames = 10 × 10³, 50 × 10³, 200 × 10³). Automated blocking is performed to identify optimal block times for model selection and parameter estimation. Scale bar: 1 μm. (C) Bayesian model selection and parameter estimation using noise estimates from raw intensity traces without blocking. (Light-blue circles) Domain boundaries and dots indicate pixels where multiple diffusing components (N_D = 2, 3) are detected. Corresponding distributions of estimated diffusion coefficients from the inferred models are shown below each image. To see this figure in color, go online.

Figure 3

Membrane heterogeneity in two-component, phase-separated SLBs is resolved by Bayesian ITIR-FCS. (A) TIRF image of RhoPE-labeled two-component supported lipid bilayers (DLPC/DSPC). (Dots) Pixels where multiple diffusing components (N_D = 2, 3) are detected. Scale bar: 1 μm. (B) Map of estimated diffusion coefficients from inferred models. Diffusion coefficients corresponding to the slow component are shown for the two-component model. (C) Distribution of diffusion coefficients from inferred models. (Inset) Mean distances of two-component pixels and all pixels from domain boundaries (p value < 0.01). (D) Distribution of diffusion coefficients obtained using conventional FCS analysis. To see this figure in color, go online.

Figure 4

Bayesian ITIR-FCS supports the carpet model for organization of hIAPP-induced domains forming on the plasma membrane. (A) (First row) TIRF images of DiI-labeled cell membrane as a function of time after addition of peptide. (Dots) Pixels at which multiple diffusing components (N_D = 2, 3) are detected. Scale bar: 1 μm. (Second row) Spatial maps of estimated diffusion coefficients from the inferred models. Diffusion coefficients corresponding to the slower component are shown for the two-component model. (Third row) Spatial maps of the slow component fraction. Fractions in one-component regions are set to 1 for pixels located inside domains and 0 for pixels outside domains. (Fourth row) Distributions of diffusion coefficients estimated from inferred models. (Arrows at time-points 40 and 60 min) Peak of the one-component model pixels inside the domain at high fraction of the slow component. (Fifth row) Distribution of diffusion coefficients estimated using conventional FCS analysis that assumes a single component is present throughout the spatial domain. (B) (Left) Temporal evolution of diffusion coefficients estimated using conventional analysis (gray) and Bayesian analysis capturing two-component regions (orange). Medians and quartiles are shown. (Middle) Temporal evolution of the slow component fraction in two-component regions as a function of time. Medians and quartiles are shown. (Right) Temporal evolution of fractions of pixels classified to each model using Bayesian model selection. Four phases can be defined based on the trends of these parameters: I, t = 1–5 min; II, t = 5–40 min; III, t = 40–60 min; and IV, t = 60–75 min. (C) Schematic showing the proposed carpet model of hIAPP aggregation on the surface of the plasma membrane. Aggregates increase in size as peptides extract lipids from the membrane. To see this figure in color, go online.