Rotating a petavoxel reconstruction exposes the viewing-angle bias inherent to Golgi-Cox and confocal dendritic-spine classification

Dendritic spines are the principal postsynaptic sites of excitatory transmission. For over a century, their shape has been sorted into discrete categories such as filopodia, thin, long thin, stubby, mushroom, and branched, largely by Golgi-Cox impregnation and, more recently, confocal microscopy. Yet both approaches share a fundamental limitation. Here, using the human H01 petavoxel reconstruction imaged by electron microscopy, we show that spine-shape classification depends strongly on viewing angle. A total of 445 spines on layer-4 basal dendrites of five pyramidal neurons were classified from an arbitrary initial viewpoint (Angle 1), then reclassified after rotation in Neuroglancer (Angle 2). Only 20.9% kept their category, so chance-corrected agreement was negligible (Cohen's kappa = 0.027), and a McNemar-Bowker test showed the reclassifications were directional rather than random. Comparing morphometric measurements against three-dimensional ground-truth shapes further showed that spine length is systematically distorted by rotation across spine types, whereas caliber parameters (neck diameter, head diameter, and head area) generally remain stable except in mushroom spines, which are highly sensitive to viewing angles in both length and caliber. These observations demonstrate that the rigid Golgi-Cox and confocal taxonomies conflate true spine morphology with the arbitrary angle of view, cautioning against treating spine categories from single-viewpoint imaging as reliable descriptors of shape. Our results therefore support, as best practice, the use of continuous and coarse morphometric descriptors. More broadly, they mark the limits of optical resolution for dendritic-spine analysis and signal a shift toward three-dimensional, electron-microscopy-based reconstruction as the reference for studying spine morphology.