cytoNet: Spatiotemporal network analysis of cell communities

Abstract

We introduce cytoNet, a cloud-based tool to characterize cell populations from microscopy images. cytoNet quantifies spatial topology and functional relationships in cell communities using principles of network science. Capturing multicellular dynamics through graph features, cytoNet also evaluates the effect of cell-cell interactions on individual cell phenotypes. We demonstrate cytoNet's capabilities in four case studies: 1) characterizing the temporal dynamics of neural progenitor cell communities during neural differentiation, 2) identifying communities of pain-sensing neurons in vivo, 3) capturing the effect of cell community on endothelial cell morphology, and 4) investigating the effect of laminin α4 on perivascular niches in adipose tissue. The analytical framework introduced here can be used to study the dynamics of complex cell communities in a quantitative manner, leading to a deeper understanding of environmental effects on cellular behavior. The versatile, cloud-based format of cytoNet makes the image analysis framework accessible to researchers across domains.

Fig 1. cytoNet workflow.

(a) The cytoNet pipeline begins with masks and optionally microscope images, which can be static immunofluorescence images or calcium image sequences. (b) Spatial proximity is determined either by measuring shared pixels between cell pairs–type I networks, or by comparing the distance between cell centroids to a threshold distance–type II networks (right panel). Functional networks are estimated from correlations in calcium time series data. (c) Cell community descriptors provide information on local neighborhood characteristics of individual cells, like degree and centrality measures, and global neighborhood characteristics like modularity and path lengths.

Fig 2. Dynamics of spatial and functional topology in developing neural progenitor cells (NPCs).

(a) Spatial NPC networks at day 1, 3 and 5 of differentiation, overlaid on immunofluorescence images of nuclei stained with Hoescht dye; segmented cells are outlined in red, and spatial proximity edges are shown as yellow lines; scale bar = 50 μm. (b) Network heterogeneity of spatial NPC networks is lowest at day 3; red notches show mean and standard deviation; *p < 0.005 from two-sample t-test. (c) Functional networks obtained through calcium imaging with Fluo-4 in developing NPC networks at days 1, 3 and 5. Correlations between calcium traces from individual cells are shown as a network plot overlaid on the maximum intensity image from calcium image sequences. (d) Fraction of active cells in the network; *p < 0.005 from two-sample t-test. Active cells are defined as cells whose normalized fluorescence traces have three or more calcium transients. (e) Frames from time-lapse movies of differentiating NPCs transfected with FUCCI cell cycle reporters. Borders of mCherry+ nuclei (G1) are outlined in magenta, Venus+ nuclei (S/G2/M) are outlined in green, and mCherry-/Venus-nuclei (quiescent) are outlined in blue, spatial edges are overlaid in yellow; scale bar = 50μm. (f) Neighborhood similarity score for low-density culture across time. (g) Neighborhood similarity score across time for medium-density culture. Fig 2A–2D adapted from reference [12].

Fig 3. Dynamics of Coupled Functional & Spatial Analysis In Vivo.

cytoNet captures relationships between spatial proximity of neurons and functional features of multicellular modules in vivo. (a) Cells classified according to the time required to first reach their maximum ΔF/F0 values from 20% of that value (ramp-up) and the time required to return to 20% (ramp-down). Edges connect similarly classified cells that are within 10 cell diameters of each other. All cells reached their peak values at 20 seconds except for those circled which reached their peak values at 25 seconds. (b) Calcium time series (ΔF/F0) plotted for 6 categories of cells with unique combinations of ramp-up and ramp-down times. The blue braces indicate a cell’s ramp-up and ramp-down. Each inset image is a spatial pattern of cells with the same ramp-up and ramp-down times. RU = ramp-up time; RD = ramp-down.

Fig 4. Influence of local neighborhood density on primary human endothelial cell (HUVEC) morphology.

(a) Distribution of cell circularity values grouped under different levels of closeness centrality; sample size, n = 786 cells (group 1; c_n < 0.025), 741 cells (group 2; 0.025 < c_n < 0.05) and 782 cells (group 3; c_n > 0.05); Cohen’s d effect size: groups (1, 2) = 0.34, groups (1, 3) = 0.62 (b) Sample immunofluorescence image with graph representation overlaid; scale bar = 50 μm. (c) Heatmap depicting closeness centrality of each cell, with circularity values overlaid in text. (d) Representative cells from cluster analysis, highlighted in magenta. (e) Cell size, closeness centrality and circularity distribution plots for each cluster. (f) Bar plot of variance explained by growth factor treatment and local network metrics. (g) Box plot of cell size as a function of growth factor treatment. (h) Box plot of mean actin intensity as a function of growth factor treatment. Legends and axes in (f-h) contain information on treatment (BDNF, VEGF), concentration (50ng/ml, 100ng/ml) and time of treatment (6 hours and 12 hours). Cohen’s d effect size for (f-h) is shown in S2 Table.

Fig 5. Spatial Analysis of the Pericapillary Niche in Adipose Tissue.

Example confocal images of wild type (a) and knock out (b) adipose tissue and the corresponding output graph for the wild type image (e). Red = lectin (capillaries). Green = Bodipy (adipocytes). Yellow: integrin α7 positive cells. Violin plots of cell properties comparing wild-type and knockout (c, d, f-h). Distances are measured between the closest border pixels of pairs of objects. Fig 5F is adapted from reference (1). Values represent the results of analyzing confocal images from 17 samples (9 wild-type and 8 knockout). Error bars are mean +/- standard deviation. p-values were computed using the Wilcoxon rank sum test (*: p ≤ 0.05, ***: p ≤ 0.001).