Fig. 2

Methods for muscle fibre OXPHOS classification based on quadruple immunofluorescent staining. (a) Tissue sections were stained using quadruple immunofluorescence (QIF). Fibres were segmented by intensity thresholding of the membrane marker as previously described³⁷. Example microscopy image shows one OXPHOS protein (blue) overlaid with the mitochondrial marker (red, VDAC1); the corresponding segmentation mask is shown in white. Scale bar: 100 μm. Mean per-fibre QIF signal was extracted for classification. (b) Frequentist linear regression method: fibres classified using 95% predicted intervals (PI, grey lines) from linear regressions (LR, red lines) fitted to bootstrapped control data (n = 10,000 iterations). Colour scale represents per fibre classification certainty as the mean across models. (c) Visual 2D-mitoplot method: investigators manually lassoed fibres with relatively high or low OXPHOS signal compared to control-like or ‘normal’ patient fibres. The superimposed polygon outline illustrates how one investigator could visually classify fibres as OXPHOS-deficient. Colour scale represents per fibre classification certainty as the mean label across investigators. Control data was displayed to anchor classifications. (d) Ground truth: fibres classified by direct visual inspection of representative ¼ mm² regions from images. b-d) In all 2D-mitoplots²⁴, fibres are shown in OXPHOS–VDAC1 space: control fibres (black); patient fibres classified for OXPHOS status as high (green), normal (blue), low (red), or not classified (cyan). Under the 2D-mitoplot of each classification effort, OXPHOS-deficient fibres are highlighted on the segmentation mask shown in (a), with the colour scale representing classification certainty. (e) For both classification methods in (b) and (c), distributions of proportions of fibres high (overabundant) or low (deficient) in each OXPHOS protein per section were generated by bootstrapping patient fibres. Stripchart shows an example of these per-section distributions for patient 1 according to the frequentist method in (b), where each dot (black) represents one bootstrapped value of proportion of NDUFB8-deficient fibres. Dashes are per-distribution median (red), median absolute deviation (cyan) and full range (black).