Macromolecular connections of active zone material to docked synaptic vesicles and presynaptic membrane at neuromuscular junctions of mouse.

Preparation

The bilaterally paired levator auris muscles ( Angaut-Petit et al., 1987 ), which lie just beneath the skin, are broad but only a few muscle fibers thick. They were exposed in four terminally anesthetized (50 mg/kg pentobarbital, IP; Sigma, St. Louis, Missouri) mice (C57/BL6), under a dissecting microscope. Mouse Ringer’s solution (150 mM NaCl, 5.6 mM KCl, 2.25 mM CaCl 2 , 2.4 mM NaHCO 3 , 0.56 mM glucose, pH adjusted to 7.2 with NaOH) containing tetrodotoxin (10 mg/ml; Sigma, St. Louis, Missouri) was dripped onto the muscles’ superficial surface and injected next to their deep surface for 10 min to inhibit impulse evoked synaptic activity that might be caused by the application of fixative. The fixative, a solution of 1% glutaraldehyde (Ted Pella, Inc., Redding, California) in Millonig’s phosphate buffer, which was isosmolar to the Ringer’s solution (320 mOsM total, pH 7.2), was applied similarly for 30 min. The muscles were then removed from the animal and pinned out flat in Sylgard 184 (Dow Corning, Midland, Michigan) coated petri dishes containing the same fixative for 30 min. They were washed for 1 hr in 320 mOsM phosphate buffer (pH 7.2), fixed and stained for 1 hr in 1% OsO 4 in phosphate buffer (pH 7.2), washed 1 hr in H 2 O, stained 1 hr in saturated aqueous uranyl acetate, dehydrated in increasing concentrations of ethanol and embedded flat in a wafer of Eponate 12 (Ted Pella, Inc., Redding, California) less than 1 mm thick. Regions of the muscles containing NMJs were identified in the wafers at x400 magnification with a light microscope, and blocks containing them were cut out and mounted for sectioning. The sections varied from 50 nm to 250 nm in thickness. They were stained with uranyl acetate in methanol and with aqueous lead citrate. In ET studies on active zones in axon terminals of frog NMJs fixed either with an isosmolar solution of 1% glutaraldehyde in Millonig’s phosphate buffer (220 mOsM total) or by ultra-rapid freezing ( Heuser and Reese, 1981 ), which is preferable for characterizing the structure of certain cellular components, we observed no significant difference in the diameter of synaptic vesicles, in the presence of beams, ribs and pegs in the AZM, and in the spacing between the pegs (unpublished observations; see also Sosinsky et al., 2008 ). We chose to use glutaraldehyde fixation for the experiments presented here because of its convenience. The animal experimentation described here was approved by Stanford University’s Administrative Panel on Laboratory Animal Care (IACUC), which oversees the use of animals according to U.S. federal law.

Data collection

Data sets were collected at x23,000 to x59,000 magnification with an FEI Polara G2 TEM electron microscope (FEI Company, Hillsboro, Oregon) equipped with a 2048×2048 CCD (Tietz F224HD; Tietz Video and Imaging Processing Systems, Gauting, Germany) in this laboratory or at x31,000 with a Phillips Tecnai T20 electron microscope (FEI Company Hillsboro, Oregon) equipped with a 1024×1024 CCD (Gatan, Inc., Pleasanton, California) in the laboratory of Dr. David Agard at the University of California, San Francisco, using an automatic data acquisition procedure (UCSF Tomography; Zheng et al., 2004 ). The stage was cooled to liquid nitrogen temperature to reduce specimen shrinkage. Eight of the thirteen data sets used for this study consisted of images taken at 1-degree tilt intervals to ±60 or ±70 degrees along a single tilt axis. The remaining five data sets consisted of images taken at 1-degree intervals to ±60 degrees along each of two orthogonal tilt axes. Reconstructions from the dual axis data sets have less noise than those made from single axis data sets ( Penczek et al., 1995 ; Cristina et al., 2005 ). However, both types of data sets yielded qualitatively similar structural models of AZM components, and measurements from both types rose to statistical significance. For generating reconstructions the images were first aligned automatically using 5 or 10 nm gold colloid (British Biocell International, Cardiff, U.K.) deposited on one or both sides of the sections before data collection. The average alignment error was 1.3 pixels (range, 0.9 to 2.2) root mean square. The reconstructions were made by a weighted back-projection method. Both the alignment and reconstruction algorithms are in the unified software package EM3D ( Ress et al., 1999 ; Ress et al., 2004 ; http://em3d.stanford.edu ). The spatial resolution in reconstructions generated from data sets collected at x31,000 to x62,000 magnification, which were used for making the surface models, was 2–3 nm for high contrast structures such as the cytoplasmic and extracellular layers of the plasma membrane ( Ress et al., 1999 ). Virtual slices, segmentation and rendering surface models Virtual slices through the reconstructed tissue sections were 1 voxel thick. Depending on the magnification of the images in a data set, the virtual slice thickness represented 0.58 nm to 1.5 nm of the tissue section’s thickness. When necessary, the angular orientation of the slice plane was adjusted to maximize contrast boundary discrimination of the structures under study. Structures were segmented from the high-resolution reconstructions (1 voxel = 0.58 nm to 1.16 nm) by using a combination of manual and semi-automatic methods in EM3D to define individual volumes-of-interest (VOIs; Ress et al. 2004 ). For the presynaptic membrane and synaptic vesicles, which were heavily stained and had a simple geometry, a semi-automatic scheme was used. For structures that had a complex geometry and light to moderate stain, VOIs were defined by manually marking a closed path on a series of slices. The VOIs were slightly larger than the structures that they enclosed to allow accurate and complete isodensity-surface calculations for the surface models. We used EM3D to calculate and render a surface model for each VOI. The calculation was done using a gray scale value that minimized the mean spatial uncertainty averaged across the whole area of the model. Surface models generated in this way had a spatial resolution equal to the resolution of the reconstructed volumes ( Ress et al., 2003 ; Ress et al., 2004 ). Surface models were generated from nine reconstructions (four single tilt-axis and five double tilt-axis data sets) from both of the levator auris muscles in one mouse. Serial virtual slices through these reconstructions were also examined in detail. Structural relationships quantitatively characterized in the surface models and serial slices from the one mouse were checked by eye in serial virtual slices from reconstructions (four single tilt-axis data sets) made from the muscles taken from three other mice. In all cases the relationships of primary docked vesicles, secondary docked vesicles, and invaginations in the presynaptic membrane to AZM, in general, and to the AZM’s ribs, beams and pegs, in particular, was similar. Contact areas between docked vesicles and presynaptic membrane In order to measure the size of the area of contact between the membrane of docked synaptic vesicles and the presynaptic plasma membrane at an active zone, surface models of each were first generated using EM3D ( Ress et al., 2004 ). The distance between each polygon vertex on the presynaptic membrane model and synaptic vesicle model was then computed. The area where the distance between them was within a spatial tolerance of 1.5 nm, which was below the resolution of the reconstructed volumes, was considered the contact area. To determine the size of the contact area it was projected onto a best-fit plane using the Pearson’s eigenvalues method ( Pearson, 1901 ). Briefly, a matrix containing the coordinates of vertices in the contact region was generated and the centroid of the contact region was subtracted from the matrix. The matrix was transposed and multiplied by the original producing a 3 × 3 covariant matrix. After obtaining the covariance matrix of vertices comprising the contact region, eigenvectors and eigenvalues were computed ( Arfken, 1985 ). The eigenvector corresponding to the smallest eigenvalue of the covariance matrix was considered the normal vector to the best-fit plane passing through the centroid of the contact region. Subsequently the area of the projected contact region on the best-fit plane was calculated. Diameter of docked vesicles Serial virtual slices made through reconstructed volumes in their x–y plane were used to identify the slice for each vesicle that passed through its equator. Because the vesicles were not perfect spheres, diameters to the outer surface of the vesicle membrane were measured for each vesicle along four separate axes (~45° increments) and expressed as an average.

Figure preparation

Figure layouts were prepared using Adobe Photoshop CS3 (Adobe Systems Incorporated, San Jose, CA).