GoKawiilAll surgical and experimental procedures were in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and approved by the Animal Care and Use Committees of the Allen Institute or Johns Hopkins University. Allocations of mice to experiments were randomized and experimenter blind. Mice were housed at 20–22 °C and 38–42% humidity.
Definition of LC-NE neurons
We used Dbh-Cre mice80 (Dbhtm3.2(cre)Pjen, The Jackson Laboratory, 033951; RRID:IMSR_JAX:033951) backcrossed with C57BL/6J mice (The Jackson Laboratory, 000664; RRID:IMSR_JAX:000664). Dbh-Cre mice were crossed to RCL-H2B-GFP mice (The Jackson Laboratory, 036761; RRID:IMSR_JAX:036761), and the native fluorescence of the labelled nuclei was imaged on a SmartSPIM microscope after tissue clearing using LifeCanvas active delipidation, agar embedding, and refractive-index matching using EasyIndex (SmartSPIM, LifeCanvas Technologies). Images from 8 brains were acquired at 1.8 × 1.8 × 2.0 μm per voxel. Raw image data were passed through a customized 3D UNet (Trailmap)81 with a model trained for immunolabelled nuclei82. The resulting probability maps were convolved with a spherical kernel approximating a nucleus diameter and then centroids were estimated with 3D local maxima detection. Stitching and registration to the CCFv3 (RRID:SCR_020999) followed the SmartSPIM Pipeline (https://github.com/AllenNeuralDynamics/aind-smartspim-pipeline).
Points were subsequently transferred to the Allen atlas reference space via an ANTS transform (RRID:SCR_004757)81. After restricting analysis to caudal midbrain and dorsal pons, and after reflecting all points into a single hemisphere, local densities were estimated via nearest-neighbour analysis and points at uniform density levels were used to define meshes. In brief, normals for points were estimated and used to generate surfels which were wrapped in a watertight mesh83 by isosurface extraction (https://www.github.com/fwilliams/point-cloud-utils). A mesh representing LC-core at the 67th percentile represents the density threshold at which sub-coeruleus cleanly separates from the main portion of LC. Initial mesh estimates were made using dynamic radius and resolution parameters to compensate for point density and mesh size across the displayed percentiles.
Single-neuron reconstructions
Whole mouse brain specimens were prepared for single-neuron reconstructions using previously published methods33,84. In brief, 2 male and 2 female adult Dbh-Cre mice received systemic injections, via the retro-orbital sinus, of a 100 μl mixture of Cre-dependent Tet transactivator (AAV-PHP-eB_Syn-FlexTRE-2tTA, Addgene plasmid id: 191210; RRID:Addgene_191210; dosage range 1.0 × 108 genome copies (gc) ml−1 to 3.0 × 109 gc ml−1) and a reporter virus (either AAV-PHP-eB_7x-TRE-3xeGFP or AAV-PHP-eB_7x-TRE-tdTomato, typical dose 1.8 × 1011 gc ml−1; RRID:Addgene_191206 and RRID:Addgene_191207). Viruses were obtained from either the Allen Institute for Brain Science viral vector core, the University of North Carolina, or the BICCN-Neurotools core and were prepared in an adeno-associated virus (AAV) buffer consisting of 1× phosphate-buffered saline (PBS), 5% sorbitol, and 350 mM NaCl. Six-to-eight weeks after viral transfection, mice were anaesthetized with an overdose of isoflurane and transcardially perfused with 10 ml 0.9% saline at a flow rate of 9 ml min−1 followed by 50 ml 4% paraformaldehyde in PBS at a flow rate of 9 ml min−1. Brains were extracted and post-fixed in 4% paraformaldehyde at room temperature for 3–6 h and then left at 4 °C overnight (12–14 h). The following day, brains were washed in 1× PBS to remove all traces of excess fixative. Subsequent tissue processing including clearing, immunolabeling, and whole-brain expansion steps were carried out as previously described33,84. For immunolabeling, delipidated brains were equilibrated in a detergent buffer (PTxw) followed by incubation in PTxw containing 20 µg per brain in 4.5 ml of the primary antibody, either rabbit anti-GFP (ab290, Abcam) or goat anti-tdTomato (AB8181, SICGEN). After thorough washing in PTxw, the corresponding secondary antibody was applied at a dose of 40 µg per brain in 4.5 ml; donkey anti-rabbit Alexa Fluor 488 (A-21206, Invitrogen) for GFP-labelled brains or donkey anti-goat Alexa Fluor 568 (A-11057) for brains labelled with tdTomato. All other steps were carried out exactly as described previously33,84. Gelled brains were soaked in 0.05× saline sodium citrate (SSC) to achieve approximately 3× expansion, and 24 h prior to imaging the expanded brains were equilibrated in a solution of 0.05× SSC that contained 10 mM ascorbic acid included as an antifade.
Expanded brains were imaged on a custom SPIM microscope (ExA-SPIM)33,85. Brains were imaged with both 488 nm and 561 nm excitation and an 8× binned autofluorescence image volume was collected for the ‘non-signal’ channel for registration to the CCF. All subsequent data processing steps including image illumination correction, stitching and fusion into a coherent image volume were carried out via automated cloud pipelines33.
ExA-SPIM whole mouse brain image volumes were registered to CCFv3. A custom ExA-SPIM template was generated by aligning and averaging whole-brain images (8× binned, autofluorescence channel) from 7 brains, including flips (14 specimens total). Registration proceeded in two steps. First, an automated two-stage registration was performed using Advanced Normalization Tools (ANTs; https://github.com/ANTsX/ANTs), consisting of a per-sample affine and SyN deformable registration to the ExA-SPIM template, followed by a fixed ExA-SPIM-template-to-CCF registration. Second, local misalignments remaining after the automated step were manually refined via landmark-based registration, matching anatomical landmarks across the automatically aligned sample and the CCF reference, using 3D Slicer34. The resulting displacement field, computed by the automated sample-to-template and template-to-CCF transforms, was applied to the node coordinates of each neuronal reconstruction (SWC) via ANTs point transformations, thereby mapping reconstructed neurons from sample space into CCF space.
To generate whole-brain single-neuron reconstructions, we used HortaCloud86, an open-source streaming 3D annotation platform enabling fast visualization and collaborative proofreading of terabyte-scale image volumes. Using this platform, human annotators proofread individual neurons via a web browser on a personal workstation. Proofreading entailed starting from the soma and tracing out all axonal and dendritic segments using a depth first search tree traversal approach34,87. Neuronal trees generated by manual placement of control points were refined offline to generate dense (one node per voxel) and sparse (uniform sub-sampling of the dense tree) representations of the reconstruction data before use for subsequent quantitative analysis. Quantitative comparisons were made with neurons in the MouseLight database (RRID:SCR_016668).
Models of single-neuron axonal distributions
... continue reading