Identification of nuclear pores and variations in their membrane pore diameter. A nuclear pore model was generated by training on ground truth annotations of nuclear pores from cell 13a SVG-A prepared by HPFS and imaged at 5 nm isotropic resolution. (A) Orthogonal views of a representative nuclear pore not used for training show ground truth annotations and model prediction. Scale bar, 50 nm. (B) Nuclear pore predictions for all the pores on the nuclear envelope of naïve cell 19 (Hela-2) prepared by HPFS and visualized during interphase at 4 × 4 × 5.3 nm isotropic resolution (left panel); the inset highlights the characteristic doughnut shape of the nuclear pore. Scale bar, 2 μm. Representative orthogonal views (right panels) of a nuclear pore and model prediction. Scale bar, 50 nm. (C) Histogram of nuclear membrane pore diameters measured in naïve cells 15, 17, and 19 (N = 934) identified by the nuclear pore model. Each membrane pore diameter determined in the raw image represents the average value from 18 measurements spaced 10° apart (see inset and Materials and methods). The Gaussian fit (blue) shows the expected size distribution if the data had come from membrane pores of a single diameter centered on the experimentally determined median (d = 100 nm, most abundant species); a standard deviation of 6 nm corresponds to the expected error of the measurements (see Materials and methods). (D) Three-dimensional distribution of nuclear pores on the nuclear envelope of cell 19, color-coded as a function of membrane pore diameter.