Hyperspectral imaging and adaptive thresholding to identify agonist-induced cAMP signals in pulmonary microvascular endothelial cells

Abstract

Cyclic AMP (cAMP) is a second messenger that regulates a wide variety of cellular functions. There is increasing evidence suggesting that signaling specificity is due in part to cAMP compartmentalization. In the last 15 years, development of cAMP-specific Förster resonance energy transfer (FRET) probes have allowed us to visualize spatial distributions of intracellular cAMP signals. The use of FRET-based sensors is not without its limitations, as FRET probes display low signal to noise ratio (SNR). Hyperspectral imaging and analysis approaches have, in part, allowed us to overcome these limitations by improving the SNR of FRET measurements. Here we demonstrate that the combination of hyperspectral imaging approaches, linear unmixing, and adaptive thresholding allow us to visualize regions of elevated cAMP (regions of interest – ROIs) in an unbiased manner. We transfected cDNA encoding the H188 FRET-based cAMP probe into pulmonary microvascular endothelial cells. Application of isoproterenol and prostaglandin E1 (PGE₁) triggered complex cAMP responses. Spatial and temporal aspects of cAMP responses were quantified using an adaptive thresholding approach and compared between agonist treatment groups. Our data indicate that both the origination sites and spatial/temporal distributions of cAMP signals are agonist dependent in PMVECs. We are currently analyzing the data in order to better quantify the distribution of cAMP signals triggered by different agonists.

1. INTRODUCTION

Cyclic AMP is a second messenger and is known to differentially regulate a variety of cellular processes depending on where the cAMP signal is originated or distributed. For example, cAMP produced by membrane bound adenylyl cyclase protects and enhances the pulmonary microvascular endothelial barrier integrity. Whereas, the cAMP produced by soluble adenylyl cyclase such as ExoY or heterologously expressed soluble adenylyl cyclase disrupts the endothelial barrier integrity in PMVECs.¹ These and other studies^2,3 suggest that the cAMP signals are highly compartmentalized and that the compartmentalized cAMP signals dictate the signaling specificity and cellular or physiological outcome. However, how different agonist-induced cAMP signals encode for discrete physiological responses through the cAMP signaling pathway is not yet clear.

Fӧrster resonance energy (FRET) based imaging has become a standard approach for the measurement of second messenger signals in cells.^4,5 However, FRET-based sensors have inherently low signal strength, and corresponding low signal to noise ratio than their individual parent fluorophores, thus limiting most FRET studies to two spatial dimensions. In addition, not considering axial dependence of signaling agent distribution can result in misinterpretation of data. Using the improved signal-to-noise ratios offered by spectral imaging FRET microscopy, our lab developed a novel 5D ((x, y, z, lambda, and time) hyperspectral FRET imaging approach for the measurement of cAMP signals in three spatial dimensions in single living cells.

In these studies, in need of an approach to quantify the segmented/localized cAMP signals in cells, we combined hyperspectral imaging and analysis approaches along with an in-house developed automated regions of interest-based feature analysis algorithm, called S8. The potential of this combined approach is illustrated by investigating agonist-induced cAMP responses in pulmonary microvascular endothelial cells. This approach will be specifically helpful to determine the spatial and temporal distribution of second messenger signaling events and the impact of localized cAMP signals in physiological outcomes.

2. METHODS

2.1. Sample Preparation

Pulmonary microvascular endothelial cells (PMVECs) were cultured and prepared as described previously⁴. Briefly, PMVECs were seeded onto round glass coverslips and were maintained in Dulbecco Modified Eagle’s Medium (DMEM) supplemented with 10% FBS and antibiotics and incubated at 80⁰F for 24 hours. Cells were then transfected with cAMP biosensor encoding Turquoise (donor fluorophore), cAMP binding domain isolated from Epac, and Venus (an acceptor fluorophore)⁶ and incubated for 48 hours. Prior to imaging, coverslips were transferred to a cell chamber and covered with buffer of interest. Cells were labeled with 25 μM DRAQ5 nuclear label for 10 minutes. Single labeled controls and sample blanks were prepared to construct the spectral library as described previously^7,8.

2.2. Spectral Imaging and Analysis

Spectral z-stacks were acquired using Nikon A1R hyperspectral confocal fluorescence microscope equipped with 60X (Plan Apo VC 60x DIC N2 WI NA-1.2, Nikon Instruments, Melville, New York) water immersion objective and a 32 array PMT detector. Samples were excited using 405 nm, 2% (for donor excitation) laser and 561 nm, 10% (for DRAQ5 excitation) laser lines and emission was collected from 414 nm to 724 nm at 10 nm increments. Spectral z-stacks were acquired every 30s for 20 minutes. After 1 minute of baseline acquisition, PMVECs were treated with either 0.1 μM isoproterenol or 0.1 μM PGE₁ or vehicle control. Similar acquisition settings were utilized to acquire spectral images of single label controls and were then utilized to construct spectral library.

Spectral data were unmixed to its individual endmembers (fluorophores/labels/background) using the spectral library containing pure spectra of each endmember. FRET efficiency was calculated and was then mapped to cAMP concentration as explained previously.⁷ Spectral unmixing, FRET efficiency measurements, and mapping FRET to cAMP concentration were performed in MATLAB programming environment using custom written scripts. Agonist-induced changes in cAMP signals were then quantified using the regions of interest (ROI) – based image analysis approaches.

2.3. Adaptive Thresholding and ROI-based Image Analysis:

Agonist induced changes in cAMP signals were quantified using in-house developed software package called S8 which is based on adaptive thresholding and ROI-based feature analysis. cAMP data was spatially and temporally smoothed using Savitzky-golay smoothing algorithms. A threshold was then calculated using smoothed image data and was then applied to timelapse data to identify regions(s) of interest. Identified region (s) of interest was then tracked over time. The timelapse ROI – based image data were then utilized to extract quantitative data including, but not limited to, area of ROIs, number of events happened at a given timepoint, intensity, and baseline/treatment areas of ROI.

3. RESULTS AND DISCUSSION

The combination of hyperspectral imaging, linear spectral unmixing, adaptive thresholding, and automated image analysis to track the regions of interest allowed us to visualize the agonist -induced changes in cAMP signals in PMVECs. PMVECs were cultured and transfected with cAMP biosensor as described in section 2.1. Spectral images were acquired using Nikon A1R hyperspectral confocal microscope and cAMP signaling events were measured using automated image analysis that was developed in-house called S8 as explained above in sections 2.2. and 2.3. Isoproterenol induced an increase in intracellular cAMP levels from apical to basal side of the cell as shown in Figure 1. S8 was then implements on the cAMP data shown in Figure 1 to quantify and track regions of interest (cAMP signaling events) in PMVECs.

Figure 1:

Isoproterenol induced changes in cAMP signals axially in PMVECs. FRET efficiency calculated using unmixed Turquoise and Venus image data is mapped into cAMP concentration. Left panel represents the distribution of cAMP signals axially at baseline conditions. Middle and right panels represent the distribution of cAMP signals axially at 4 minutes and at 10 minutes after addition of the 0.1 μM isoproterenol. It is important to note that the axial gradients are readily observable upon treatment (cAMP concentration is higher in apical slices when compared to that of basal slices).

We implemented an automated regions of interest - based analysis approach called S8 to investigate the spatial and temporal distribution of agonist-induced cAMP signaling events in PMVECs. Figure 2 demonstrate a representative experiment where cells were treated with 0.1 μM isoproterenol and data were analyzed using S8. We found that S8 (adaptive thresholding along with automated tracking of regions of interest) allowed us to automatically segment the cAMP signaling events (regions) that are significantly above the baseline/noise levels (compare middle and right panels in Figures 1 and 2).

Figure 2:

cAMP signals quantified using regions of interest - based feature analysis algorithms developed in-house. Left column represents the distribution of cAMP signals calculated using S8 algorithm at baseline (using the cAMP images from left panel in figure 1). Middle and right panels represent the distribution of cAMP signals calculated at 4-minute and 10-minute timepoints using S8 algorithm. Signals in each image panel represents the masked outputs from S8 that are significantly above the noise.

We then investigated the PGE₁ – induced changes in cAMP signaling events (Figure 3) in PMVECs using similar approaches explained in Figure 1 and Figure 2. Unlike isoproterenol, PGE₁ induced cAMP production at the apical regions and also in discrete localized intracellular regions which are more noticeable (compare 3^rd and 4^th panels in Figure 3) when implemented S8 on the cAMP data.

Figure 3:

PGE₁ – induced cAMP distributions quantified using regions of interest - based feature analysis algorithms developed in-house. First and third columns represent cAMP concentration calculated by mapping FRET efficiency to cAMP concentration at baseline and 15 minutes after treatment with 0.1 μM PGE₁, respectively. Second and third columns represent the distribution of PGE₁-induced cAMP signals at baseline and 15 minutes after treatment using S8 algorithm, respectively.

These masked output data from S8 (Figures 2 and 3) were then utilized to quantify the area size of ROIs, estimate the number of events or regions of interests identified at the beginning of each time point (starting frame) (Figurer 4). We observed a trend in number of events identified over time when cells were treated with either isoproterenol or PGE₁. Whereas, the number of events identified remained constant over time when cells were treated with vehicle control. Note that the elevated number of events seen in the first timepoint is an analysis artifact.

Figure 4:

Distribution of number of events identifies at the beginning of each timepoint (each frame represents a timepoint in the timelapse data). Top left panel represents isoproterenol induced changes in number of events over time. Top right panel represents PGE₁ induced changes in number of events identifies at each frame over time. Similarly, bottom panel represents the distribution of events when cells were treated with vehicle control. Each event corresponds to a ROI in the cell mask identified at that specified frame/timepoint.

4. CONCLUSION

In this paper, we demonstrated the use of hyperspectral imaging, linear spectral unmixing, and automated ROI-based image analysis (S8) approaches to quantify the agonist-induced localized cAMP signals in PMVECs. We are currently investigating the isoproterenol and PGE₁ - induced changed in cAMP signals in PMVECs. This approach will be extremely helpful to determine the localized cAMP events during physiological and patho-physiological conditions of the cell (such as cells infected with ExoY, soluble adenylyl cyclase known to increase intracellular cAMP and alter normal cell physiology).