k-means clustering identifies highly different populations contralaterally vs ipsilaterally in an ischemic stroke model but offers a heterogeneous cluster composition.

A T2-weighted MRI acquisition of the tMCAO rat brain 24 hours after reperfusion (top). Z maximum projections of IBA1 staining of microglia on the ipsilateral side to the lesion (bottom left) and contralateral side (bottom right). Scale bar: 20 µm. B Scatter plot with each dot representing one microglia from the complete dataset (ipsilateral and contralateral), plotted in function of their PC1 against PC2 values from the PCA conducted on all 20 initial morphological parameters. The colors represent the clusters identified by the k-means method, and the ‘x’ symbols are the centers of each cluster. C All microglia from the dataset with colors associated to their affiliated cluster: green corresponds to cluster 1 (ramified microglia), blue corresponds to cluster 2 (activated microglia), yellow corresponds to cluster 3 (amoeboid microglia) and pink to cluster 4 (rod-like microglia). The white crosses indicate cells that exhibit morophological characteristics that should categorize them in another cluster. D Pie chart illustrating the proportion of each cluster constituting the contralateral and ipsilateral sides.

MorphoCellSorter approach to generate a ranking of the cells based on their morphology applied to the ischemic stroke model on fixed tissue.

A Summary of the fully automated MorphoCellSorter approach: morphological indexes of the individualized and binarized microglia are computed on a first script (MorphoCellMeter), and the resulting data table is used as an entry for MorphoCellSorter to generate the ranking of the cells based on their morphological characteristics. B Cascade of normalized eigenvalues. The blue bars represent the eigenvalues for each principal component (PCs), and the red curve is the cumulative sum of the values. The first two eigenvalues corresponding to the first two PCs that account for over 70% of the total sum. C Correlation circle in the PC1 and PC2 planes. D Parameters ranked according to their projection on PC1. Values are normalized by the highest projection value. To determine the number of morphological parameters selected, here we consider the median number of the normalized cumulative projection (pink curve), leading to a threshold at 0.5 (red line), and thus 7 kept parameters in this case. E Andrews plots generated with the 7 parameters weighted according to their contribution onto PC1. Each curve represents one microglia in the high-dimensional space of the 7 strongest parameters’ projections onto PC1. The maximum variance between curves is obtained at 359°.

MorphoCellSorter offers a reliable ranking of microglia in an ischemic stroke model based on their morphologies.

A Automatic ranking generated by MorphoCellSorter of the 347 microglia constituting the ischemic stroke model dataset. B Distribution of the rank difference between expert 1 manual and MorphoCellSorter rankings. C Distribution of the rank difference between expert 2 manual and MorphoCellSorter rankings. D Distribution of the rank difference between expert 1 and expert 2 manual rankings. E Correlation between expert 1 and MorphoCellSorter rankings. Spearman’s correlation coefficient Rs = 95968, p<0.0001. F Correlation between expert 2 and MorphoCellSorter rankings. Spearman’s correlation coefficient Rs = 97188, p<0.0001. G Correlation between expert 1 and expert 2 rankings. Spearman’s correlation coefficient Rs = 93498, p<0.0001.

Evaluation of MorphoCellSorter rankings on highly heterogeneous datasets of microglia.

The evaluation of the rankings is conducted by comparing the Spearman’s correlation coefficients. w: weeks, m: months, E: embryonic day, P: postnatal day, DIV: day in vitro, Ex: expert.

MorphoCellSorter identifies morphological alterations in the visual cortex of APPxPS1-KI mice.

A Z maximum projections of IBA1 (microglia) and 4G8 (Aβ plaques) stainings of control and APPxPS1-KI mice in the visual cortex. The white arrowheads point at Aβ plaques positive for 4G8, green arrowheads designate ramified microglia far from Aβ plaques and red arrowheads point to highly morphologically altered microglia close to Aβ plaques. Scale bar: 20µm. B MorphoCellSorter automated ranking. Control microglia are represented in green and APPxPS1-KI microglia in orange. C Distribution of the Andrews values at the maximum variance (Andrews scores) for control and APPxPS1-KI microglia. D-WMorphological indexes computed by MorphoCellSorter. The data shown are the mean ± standard deviation (std). Linear mixed models were applied when the application conditions were respected and Wilcoxon tests were performed otherwise (for the Span Ratio) (N=3 mice for each condition, nAPP-PS1xKI and nControl = 180). *p<0.05; **p<0.01; ***p<0.001.

Accurate sorting of embryonic microglia reveals no morphological difference between CCHS and WT embryos’ microglia.

A IBA1 stainings in WT and CCHS embryos’ brainstems. B MorphoCellSorter automated ranking. WT microglia are displayed in light purple and mutated microglia are represented in pink. Only every 3 microglia composing the dataset are displayed for readability purposes. C Distribution of the Andrews scores for CCHS and WT microglia. D-W Morphological indexes computed by MorphoCellSorter. The data’s distribution is represented as well as the mean ± std. Linear mixed models were applied (NWT=5, Nmutant=4; nWT=498 and nmutant=444).