Machine learning-based classification of mitochondrial morphology in primary neurons and brain

Abstract

The mitochondrial network continually undergoes events of fission and fusion. Under physiologic conditions, the network is in equilibrium and is characterized by the presence of both elongated and punctate mitochondria. However, this balanced, homeostatic mitochondrial profile can change morphologic distribution in response to various stressors. Therefore, it is imperative to develop a method that robustly measures mitochondrial morphology with high accuracy. Here, we developed a semi-automated image analysis pipeline for the quantitation of mitochondrial morphology for both in vitro and in vivo applications. The image analysis pipeline was generated and validated utilizing images of primary cortical neurons from transgenic mice, allowing genetic ablation of key components of mitochondrial dynamics. This analysis pipeline was further extended to evaluate mitochondrial morphology in vivo through immunolabeling of brain sections as well as serial block-face scanning electron microscopy. These data demonstrate a highly specific and sensitive method that accurately classifies distinct physiological and pathological mitochondrial morphologies. Furthermore, this workflow employs the use of readily available, free open-source software designed for high throughput image processing, segmentation, and analysis that is customizable to various biological models.

Figure 1

Identification of distinct mitochondrial morphologies in primary neurons. (A) Representative immunofluorescent image of mitochondria labeled for ATP synthase (green) and TOM20 (red). Insert: mitochondrial objects segmented and color coded by classified morphology. (B) Principal component analysis (PCA) of hand-classified mitochondrial objects in the training and test sets combined. Individual objects are color coded by morphology (n = 1091 mitochondrial objects). (C) Heatmap displaying the scaled 32 size and shape measurements by morphology. (D) Confusion matrix displaying the results of model evaluation using the hand-classified test set (217 objects).

Figure 2

Morphological analysis of mitochondria in Opa1 conditional knockout. (A) Representative images of primary Opa1^flx/flx neurons (top) and Opa1^flx/flx neurons with Cre (bottom). Panels (from left to right): ATP synthase (green) and TOM20 (red) merged, MAP2 (magenta), DAPI (blue), all channels merged. Scale bars = 10 µm. (B) Percentage of mitochondrial area per morphology in Opa1^flx/flx cells (black) and Opa1^flx/flx cells with Cre (gray). Mean ± SD (n = 9 per group). * indicates p < .05. (C) Representative Western blot displaying Opa1 and GAPDH protein levels across Cre-Lentivirus concentrations. Full-length blots are presented in Supplementary Fig. 3. (D) Western blot quantification of Opa1 protein levels after Cre-Lentivirus treatment (n = 3). For all Opa1 cKO experiments, 1 × MOI Cre-Lentivirus was used.

Figure 3

Morphological analysis of mitochondria in Drp1 conditional knockout (cKO) and glucose deprivation (GD). (A) Representative images of primary Drp1^flx/flx neurons. Rows (top to bottom): control Drp1^flx/flx, Drp1^flx/flx after 18 hr GD, Drp1^flx/flx with Cre, Drp1^flx/flx with Cre after 18 hr GD. Untreated control groups are shown on rows 1 and 3, 18 h GD groups are shown on rows 2 and 4. Panels (from left to right): ATP synthase (green) and TOM20 (red) merged, MAP2 (magenta), DAPI (blue), all channels merged. Scale bars = 10 µm. (B) Percentage of mitochondrial area per morphology in Drp1 cKO and GD experiments. Mean ± SD (n = 5–8 per group). * indicates p < .05 versus Drp1^flx/flx. # indicates p < .05 versus Drp1^flx/flx + GD. $ indicates p < .05 versus Drp1^flx/flx + Cre. (C) Representative Western blot displaying Opa1 and GAPDH protein levels across Cre-Lentivirus concentrations. Full-length blots are presented in Supplementary Fig. 3. (D) Western blot quantification of Drp1 protein levels after Cre-Lentivirus treatment (n = 3–4). For all Drp1 cKO experiments, 2 × MOI Cre-Lentivirus was used.

Figure 4

Morphological analysis of mitochondria in glutamate and cyclosporin A (CsA) + Ru360 experiments. (A) Representative images of primary cortical neurons. Rows (top to bottom): vehicle-treated control group, control cells treated with CsA + Ru360 only, vehicle-treated group with 30 min 100 µm glutamate exposure, glutamate challenged group pre-treated and co-incubated with CsA + Ru360 Panels (from left to right): ATP synthase (green) and TOM20 (red) merged, MAP2 (magenta), DAPI (blue), all channels merged. Scale bars = 10 µm. (B) Percentage of mitochondrial area per morphology in glutamate and CsA + Ru360 experiments. Mean ± SD (n = 6–8 per group). * indicates p < .05 versus Control. # indicates p < .05 versus Control + CsA + Ru360. $ indicates p < .05 versus Glutamate.

Figure 5

3D morphological classification of mitochondria using confocal (A–C) and electron microscopy (D–F). (A) Representative 3D renderings of individual mitochondrial objects from mouse brain tissue immuno-labeled for ATP synthase and imaged using confocal microscopy. (B) Heatmap displaying the scaled 8 size and shape measurements by morphology for confocal microscopy. (C) Representative confocal image of ATP synthase immunofluorescence from mouse hippocampus. (D) 3D rendering of mitochondria from the full z-series (50 slices, 5 µm) of the image shown in (C), acquired via confocal microscopy. (E) Representative 3D renderings of individual mitochondrial objects from rat brain tissue acquired via serial block-face scanning electron microscopy (SBF-SEM). (F) Heatmap displaying the scaled 8 size and shape measurements by morphology for SBF-SEM. (G) Representative single-plane image from SBF-SEM z-series. Insert: Enlarged area of representative image with mitochondria traced (green). (H) 3D rendering of mitochondria from the full z-series (100 slices, 7 µm) of the image shown in (G), acquired via SBF-SEM.