In vivo imaging of emerging endocrine cells reveals a requirement for PI3K-regulated motility in pancreatic islet morphogenesis

ABSTRACT The three-dimensional architecture of the pancreatic islet is integral to beta cell function, but the process of islet formation remains poorly understood due to the difficulties of imaging internal organs with cellular resolution. Within transparent zebrafish larvae, the developing pancreas is relatively superficial and thus amenable to live imaging approaches. We performed in vivo time-lapse and longitudinal imaging studies to follow islet development, visualizing both naturally occurring islet cells and cells arising with an accelerated timecourse following an induction approach. These studies revealed previously unappreciated fine dynamic protrusions projecting between neighboring and distant endocrine cells. Using pharmacological compound and toxin interference approaches, and single-cell analysis of morphology and cell dynamics, we determined that endocrine cell motility is regulated by phosphoinositide 3-kinase (PI3K) and G-protein-coupled receptor (GPCR) signaling. Linking cell dynamics to islet formation, perturbation of protrusion formation disrupted endocrine cell coalescence, and correlated with decreased islet cell differentiation. These studies identified novel cell behaviors contributing to islet morphogenesis, and suggest a model in which dynamic exploratory filopodia establish cell-cell contacts that subsequently promote cell clustering.

directions, to account for potential sample movement. Data was not included for analysis in cases in which the cell shifted out of view. When possible, the sample was re-positioned and imaging re-started. This causes the occasional appearance of irregular time intervals in our time lapse image sequences. A proportion of image series could not be analyzed due to poor quality (low signal or loss of signal over time) or difficulty in distinguishing individual cells. In some samples, signal recovery permitted resumption of analysis at subsequent time points, with times of image acquisition as indicated in the figures.
As older fish (>5dpf) show decreased viability following continuous imaging for more than 2-3h, imaging studies extending >3h were performed by adapting a 'catch-and-release' approach (Mumm et al., 2006). In brief, samples were mounted in low melt agarose in glass bottom plates (maximum 2 per plate), overlaid with tricaine and directly imaged. The x,y coordinates of the imaged region were recorded relative to the primary islet to enable identification of the position at later times. After imaging, samples were carefully removed from the agarose using fine forceps, rinsed in egg water, and returned to the incubator in individual 3cm plates until the next imaging session.
To follow clustering of Notch inhibitor-induced endocrine cells from 6-9 dpf (Fig. S2), 20 samples were imaged in total. 9/9 that were followed to 8 or 9 dpf showed cluster formation. 5/8 samples imaged to 7 dpf had formed clusters, while the remaining 3 showed cell movements but had not formed clusters. 3 samples were not analyzed as they did not survive beyond the first time point. Regions for extended imaging in uninduced 13-15 dpf samples were selected to contain single cells within 2-3 cell diameters of a larger cluster or of other cells. Already formed clusters or isolated single cells were not followed.

Cell Tracking
For analysis of time lapse series, image stacks were first cropped and registered using ImageJ. Images for extended time series were manually aligned. Individual cells followed over time in z-projections were highlighted by a color overlay using Photoshop. Cells were manually tracked using 3D visualization in Imaris, with identity based on relative positions Development 145: doi:10.1242/dev.158477: Supplementary information and relation to nearby structures. In some series, cells could be followed based on GFP in addition to dsRed expression. A subset of dsRed + cells became GFP + over time, reflecting progressive differentiation of endocrine progenitor into beta cells. To quantitate cell clustering, cell coordinates (x,y,z) at each time point were determined using Imaris and exported to Matlab. Volume of a polygon that contains the tracked cells was calculated using the 'convex hull' function of Matlab. Polygon representations were generated in Matlab.

Filopodia Tracking
For filopodia tracking, the image stack was processed using ImageJ. The image was first cropped to contain a single cell. Contrast enhancement and background subtraction were performed, and gamma adjustment was applied to enhance the weak signal of fine protrusions. Filopodia length was measured in each frame using the 'Neuron Growth' plugin for ImageJ (Fanti et al., 2011). As recommended (Fanti et al., 2011), images were resized twofold to increase pixel density and improve detection of boundary features. Detection was performed using the automatic mode where possible, with manual correction applied as necessary. Cell morphology and motility were analyzed on 2D projections, recognizing that information along the z-axis will go undetected. From a lateral view, the pancreas extends primarily along the x-y dimension. Z-projections were necessary in order to have sufficient signal intensity to detect boundaries and fine protrusions in the x-y dimension. Of 13 time lapse movies acquired in 2 independent experiments, 5 had cells with robust membrane signal, and maintained signal intensity that was sufficient for filopodial tracking analysis. In all, 41 filopodia were analyzed.

Quantitation of cell morphology and membrane dynamics
For analysis of morphology and membrane dynamics in time series, PI3K-inhibitor treated and control samples at 6 dpf were imaged at 3-minute intervals with 1024x1024 pixel resolution. For analysis of cell morphology in hsp:LifeActTom-PTX transgenics and controls, heat-shocked samples at 7 dpf were imaged at 4-minute intervals with 512x512 pixel resolution to minimize bleaching, and images were resized to double the pixel density.
Images were smoothed by a median filter, and gamma adjustment was applied. Cell morphology analysis was performed using the ImageJ plugin ADAPT (Barry et al., 2015) on image series cropped to contain a single cell. Parameters were adjusted empirically for optimized detection. Fine protrusions not recognized automatically were outlined manually.
Cell circularity is calculated as 4π(area/perimeter 2 ). Solidity is the area divided by the area of the convex hull (Fig. S8A). Membrane dynamics were determined using CellGeo (Tsygankov et al., 2014), using data sets of cell boundaries computed by the ADAPT PlugIn.
Total cell membrane dynamic activity (protrusion + retraction) between successive video frames was normalized to cell perimeter. The boundary velocity threshold was set to 5.

Secondary islet quantitation
In control experiments, islet formation was most robust and consistent when more cells were induced. In a previous report, maximal islet cell induction was achieved with 50µM Ly411575 for up to 3 days (Ninov et al., 2012). For our islet assembly assay, we applied an induction method using lower doses of notch inhibitor and retinoic acid inhibitior for 24h, as samples are subjected to an additional 48h of inhibitor treatments. Combining low-dose inducing treatments acting on 2 different pathways minimized systemic toxicity, and represents the milieu that controls physiological islet cell differentiation (Huang et al., 2014;Kimmel and Meyer, 2016). Results shown for islet assembly assays are representative of at least 2 independent experiments.
To quantitate inhibitor effects on secondary islet formation, 3D object analysis was performed using the Particle Analyzer of ImageJ (Doube et al., 2010). Image stacks were prepared for analysis by preprocessing steps using ImageJ, in which the pancreas is manually outlined based on mCherry expression in the exocrine pancreas, to define the location of the secondary islets, and exclude other GFP + cells in the region (Fig. S10B-C, for details see Supplemental Protocol 1, below). For experiments using heat-shock inducible PTX expression, in preprocessing steps the pancreas is manually outlined based on a corresponding brightfield image, followed by analysis using the Particle Analyzer (see Supplemental Protocol 2, below).

Automated islet quantitation
The process for automated analysis of islet volumes involved the following steps: (1) The user is presented with an image combining the middle slice of the red channel (exocrine pancreas) and a maximum-intensity projection of the green channel (endocrine islets), and required to mark a line separating the pancreatic head from the tail region. (2) Both channels are blended with a histogram equalized version of themselves, to achieve contrast enhancement. (3) The red channel (pancreas) is segmented using the Chan-Vese model  (Chan and Vese, 2001), with an additional term added for boundary avoidance. (4) The image for volume detection is prepared by first excluding points outside the segmented pancreas or anterior to the pancreatic tail, then the intensity and gradient of the green image are combined. To reduce the sensitivity to noise, the image gradients are computed by applying an anisotropic Gaussian filter. (5) Object boundaries are determined by thresholding the image of the previous step and triangulating with standard level set techniques (Sethian, 1999). (6) An edge connectivity measure is applied to split minimally connected objects, and hidden components (ie, completely contained inside others) are discarded. (7) Finally, the volume and surface area of each component is computed.
Minimum object size is 50µm 3 . In addition, we apply the assumption that islets have a certain regularity of shape, and components with isoperimetric ratio (Area/Volume^(2/3)) above a defined threshold were considered noise and discarded. (Details of program code used for analysis are available upon request.)   (For sample details see Tables S1, S2, S3.)   Table S1).     Samples with transgenes indicating endocrine-(pax6b-promoter) and early beta-cells (mnx1-promoter) were examined by confocal microscopy and the number of transgenepositive cells in the pancreatic tail were counted. Samples transgenic for pax6b:GFP (n=53) or pax6b:dsRed (n=42) were examined by confocal microscopy at 2 weeks (14-15dpf). Secondary islets in the pancreatic tail were counted and categorized as small (1-3 cells), medium (4-6 cells) or large (>7 cells). Samples transgenic for pax6b:dsRed or pax6b:GFP were imaged by confocal microscopy at 13-15 dpf (as shown in Fig. 1H, S1). Configurations selected for imaging contained single isolated cells in proximity to each other or close to larger clusters (see Materials and Methods). Of 106 samples examined, 42 were imaged and followed as possible. "Static" refers to observations seen at a single time point, dynamic features occurred over >1 time point. Samples may be entered into >1 category.

CTL WORTMANNIN LY294002
Cell # # Frames Cell # # Frames Cell # # Frames Table S7. Toxicity assays for compounds used in this study.
Concentrations used are shown in bold. For details refer to Materials and Methods.