Box 2. Imaging

Serial optical sections should be collected through focus at intervals sufficient to allow adequate deconvolution during post-processing, and saved to disk.

To optimize resolution of the 3D structure recovered by Rayburst sampling, we recommend using the highest available numerical aperture (n.a.), the highest pixel resolution in the image plane and Z-step intervals sufficient to approximate cubic voxels as closely as possible^{19, 30}. In recent tests to assess the minimum image quality necessary to recover accurate neuronal tree geometry, we found that a minimum of 60 magnification (1.4 n.a.) with 8-bit grayscale range and voxel sizes of approximately 0.2 0.2 0.2 m (field size 1,024 1,024) was required for adequate automated reconstruction of both branching topology and spine morphology of typical layer III neocortical pyramidal neurons, without significant manual intervention¹⁹. Nonetheless, even using a 40 (0.6 n.a.) objective lens with 8-bit grayscale range, we have successfully reconstructed the dendritic trees of such neurons, provided that disconnected dendrites were visually reconnected during a final manual tree-editing phase (see Wearne et al.¹⁹ for details).

For optimal 3D shape analysis, adequate deconvolution is important, using either an experimentally measured point spread function or a theoretical point spread function calculated from the data, as is used in blind deconvolution. All data shown in this protocol were processed with the montaged blind deconvolution system AutoDeblur (Media Cybernetics), which we find produces excellent results for data obtained by CLSM or MPLSM.