Image based Machine Learning for identification of macrophage subsets

Abstract

Macrophages play a crucial rule in orchestrating immune responses against pathogens and foreign materials. Macrophages have remarkable plasticity in response to environmental cues and are able to acquire a spectrum of activation status, best exemplified by pro-inflammatory (M1) and anti-inflammatory (M2) phenotypes at the two ends of the spectrum. Characterisation of M1 and M2 subsets is usually carried out by quantification of multiple cell surface markers, transcription factors and cytokine profiles. These approaches are time-consuming, require large numbers of cells and are resource intensive. In this study, we used machine learning algorithms to develop a simple and fast imaging-based approach that enables automated identification of different macrophage functional phenotypes using their cell size and morphology. Fluorescent microscopy was used to assess cell morphology of different cell types which were stained for nucleus and actin distribution using DAPI and phalloidin respectively. By only analysing their morphology we were able to identify M1 and M2 phenotypes effectively and could distinguish them from naïve macrophages and monocytes with an average accuracy of 90%. Thus we suggest high-content and automated image analysis can be used for fast phenotyping of functionally diverse cell populations with reasonable accuracy and without the need for using multiple markers.

Figure 1

Fluorescent images of monocytes and macrophages stained for calprotectin (27E10 antigen, red, A–D), and mannose receptor (MR, green, E–H). Scale bar = 25 µm. Representative images from 6 independent biological samples (donors) are shown.

Figure 2

Comparison of cytokine profiles of M1 and M2 macrophages. Cytokines in the supernatants of human monocyte-derived M1 and M2 macrophages cultured for 6 days were measured by means of a bead-based flow cytometric system (for (A) IL-6, (B) IL-1β, (C) TNF-α), ELISA (for (D) CCL-18), and a bead-based luminex system (for (E) IL-10 and (F) IL-1RA). Data presented are the mean ± SD of at least 5 independent experiments using blood samples from different donors. Statistical significance was assessed using paired Student’s t-test (*P < 0.05, **p < 0.01, ***p < 0.001).

Figure 3

Comparison of transcription factor mRNA expression in M1 and M2 macrophages. qRT-PCR analysis of (A) STAT1, (B) SOCS1, (C) IRF5, (D) STAT6, (E) SOCS3, (F) IRF4 relative mRNA expression in M1 and M2 macrophages after 6 days of culture. All values are reported relative to the house-keeping gene GAPDH. Data show mean values ± SEM of 3 independent experiments using macrophages generated from 3 different donors. Statistical significance was assessed using student’s t-test (**p < 0.01, ***p < 0.001).

Figure 4

Immunofluorescent staining of monocytes and macrophages. (A) Monocytes cultured for 2 h; (B) monocytes cultured for 6 days; (C) naïve macrophages; (D) M1 macrophages; (E) M2 macrophages. F-actin and cell nuclei were stained with Phalloidin Alexa Fluor 488 (green) and DAPI (blue), respectively. The detected nucleus and cell outlines are shown as green and red lines respectively. Scale bar = 10 µm. Representative images from n = 6 are presented.

Figure 5

Comparison of key morphometric identifiers. Cell area (A), nuclear area (B), sum edge actin intensity normalised to cell area (C) and sum nuclear DNA intensity normalised to nuclear area (D) are shown as histograms for at least 500 cells of each type. The uniformity in size and shape is evident for monocytes observed at day 0, after which the characteristics of each cell type separate. M2 macrophages appear to have the largest cell area, whilst monocytes cultured for 6 days exhibit the largest nuclei. It follows that their nuclear DNA intensity is the highest. Actin edge intensity and cell area appear to correlate, suggesting a well-defined cytoskeleton in larger immune cells – more so in naïve macrophages than in monocytes.

Figure 6

Comparing classifier accuracy in determining M1 and M2 phenotype. (A) Classification accuracy of 5 immune cell types using 5 classifiers assessed by 10-fold cross validation. Logistic regression and random forest classifiers showed the highest accuracy for all cell types whilst support vector machine (SVM) and k-nearest neighbours (kNN) frequently misclassified cells. (B) ROC curves are plotted for each cell type as classifier sensitivity versus 1-specificity show the variation in classifier performance for each cell type. Dashed grey lines indicate a random guess. (C) The influence of training set size on the performance each classifier – accuracy is presented as an average of 10-fold cross validation across each cell type.