Whole-organism 3D quantitative characterization of zebrafish melanin by silver deposition micro-CT

Melanin-rich zebrafish melanophores are used to study pigment development, human skin color, and as a large-scale screening phenotype. To facilitate more detailed whole-body, computational analyses of melanin content and morphology, we have combined X-ray microtomography (micro-CT), a non-destructive, full-volume imaging modality, with a novel application of ionic silver staining to characterize melanin distribution in whole zebrafish larvae. Normalized micro-CT reconstructions of silver-stained fish consistently reproduced pigment patterns seen by light microscopy, and allowed direct quantitative comparisons of melanin content across wild-type and mutant samples, for both dramatic and subtle phenotypes not previously described. Silver staining of melanin for micro-CT provides proof-of-principle for whole-body, three-dimensional computational phenomic analysis of a particular cell type at cellular resolution, with potential applications in other model organisms and human melanoma biopsies. Whole-organism, high-resolution phenotyping is a challenging ideal, but provides superior context for functional studies of mutations, diseases, and environmental influences.

Methods
Micro-CT imaging was performed on Beamline 8.3.2, Tomography (micro-CT), at the Advanced Light Source (ALS) at the Lawrence Berkeley National Laboratory (LBNL, Berkeley, CA) in February 2020. A double-multilayer monochromator was used to select an X-ray energy of 26 keV to optimize silver-based contrast (elemental silver K edge = 25.5 keV) (Hubbell and Seltzer, 2004). Acquisition was performed using a custom detector system based on a 6x objective lens with 0.6 numerical aperture and a 101 megapixel thermo-electrically cooled CMOS camera outputting a 11,648-pixel x 8,742-pixel image (Vision Systems Technology, Vista, CA). Horizontal FOV was approximately 5 mm, but projections were cropped to cover the width of the sample tube as they were acquired. Whole 5 dpf larval zebrafish were imaged over two scans, covering head and tail regions, respectively. Each constant motion scan resulted in 1017 projections over 180 degrees with an exposure time of 175 ms per projection. A sample to scintillator distance of 33 mm was chosen to optimize phase effect. Two flat field images were acquired per sample, one before the head segment and one after the tail segment, to be used for image normalization.

Flat field correction, ring artifact reduction, and image reconstruction were performed using the open source TomoPy toolkit (Gürsoy et al., 2014). Flat field correction was performed using the flat field image taken either before or after sample scanning, depending on which produced the best contrast on test projections. Images were reconstructed using Gridrec (Dowd et al., 1999; Rivers, 2012) with ring artifact reduction (Miqueles et al., 2014; Münch et al., 2009) and a 2nd order Butterworth filter with a cutoff of 0.2 to reduce noise, resulting in a nominal isotropic voxel size of 0.52 μm, as estimated by the number of reconstructed voxels spanning the 1.03 mm outer diameter of the reconstructed polyimide sample tube.

32-bit reconstructed images were further processed using Fiji (Schindelin et al., 2012). Reconstructions were cropped and rotated, and regions containing the zebrafish sample were segmented on a per slice basis to remove the plastic tube and any bubbles which had formed near the sample. This “cleaned up” data was normalized to the LR White resin present in every sample to allow direct comparisons between scans as follows: An average attenuation coefficient for the resin in each reconstruction (μresin) was determined as the mean intensity value over all slices of a 100-pixel x 100-pixel square of empty resin, distant from the sample or sides of the tube. Subsequently each reconstruction was processed using the formula, μnormalized = (μ - μresin)/μresin, where μ is the attenuation value at any given pixel in a reconstruction and μnormalized is the normalized intensity. Normalized reconstructions were converted to 16-bit for visualization and analysis using the same minimum (−11) and maximum (155) μnormalized for all reconstructions.

3D rendering and analysis, including registration of head and tail segments, were performed using Avizo versions 2020.1 and 2020.2 (Thermo Fisher Scientific). A combination of thresholding and manual segmentation was used to assign voxels in merged datasets to established larval pigmented regions, including left and right RPE, dorsal stripe, ventral stripe, yolk sac stripe, and left and right lateral stripes. Regions of staining intensity not associated with one of these regions was assigned to an “other pigment” category. To clean up the segmentations by closing small holes and reducing noise, the morphological operator 3D Closing (size = 5 px), followed by 3D Opening by Reconstruction (size = 1 px), was used on the label file for each dataset.

This research used resources of the Advanced Light Source, a U.S. DOE Office of Science User Facility under contract no. DE-AC02-05CH11231. The Pennsylvania Department of Health specifically disclaims responsibility for any analysis, interpretations, or conclusions.