Mapping mesoscale axonal projections in the mouse brain using a 3D convolutional network

Abstract

The projection targets of a neuronal population are a key feature of its anatomical characteristics. Historically, tissue sectioning, confocal microscopy, and manual scoring of specific regions of interest have been used to generate coarse summaries of mesoscale projectomes. We present here TrailMap, a three-dimensional (3D) convolutional network for extracting axonal projections from intact cleared mouse brains imaged by light-sheet microscopy. TrailMap allows region-based quantification of total axon content in large and complex 3D structures after registration to a standard reference atlas. The identification of axonal structures as thin as one voxel benefits from data augmentation but also requires a loss function that tolerates errors in annotation. A network trained with volumes of serotonergic axons in all major brain regions can be generalized to map and quantify axons from thalamocortical, deep cerebellar, and cortical projection neurons, validating transfer learning as a tool to adapt the model to novel categories of axonal morphology. Speed of training, ease of use, and accuracy improve over existing tools without a need for specialized computing hardware. Given the recent emphasis on genetically and functionally defining cell types in neural circuit analysis, TrailMap will facilitate automated extraction and quantification of axons from these specific cell types at the scale of the entire mouse brain, an essential component of deciphering their connectivity.

Single-cell transcriptomes and whole-brain projections of serotonin neurons in the mouse dorsal and median raphe nuclei

“…In short, we trained a 3D U-Net convolutional neural network (Çiçek et al, 2016) to identify axons in volumes and post-processed the resulting probability-based volumes as previously described (Ren et al, 2018). Python code for training is available at GitHub (https://github.com/AlbertPun/TRAILMAP; copy archived at https://github.com/elifesciences-publications/TRAILMAP)  (Friedmann et al, 2019). Using the autofluorescent channel, we aligned samples to the Allen Institute’s Common Coordinate Framework (Renier et al, 2016), applied the same transformation vectors to the volumetric projection of axons, and quantified total axon content in each brain region listed in Supplementary file 5.…”

Cerebellar nuclei evolved by repeatedly duplicating a conserved cell type set

“…To comprehensively characterize the differences between the individual nuclei, we began by comparing their projection patterns. We performed CNS-wide anterograde tracing of each nucleus using AAV8-CAG-tdTomato followed by brain and spinal cord clearing and light-sheet imaging (Chi et al 2018;Ren et al 2019;Friedmann et al 2020) (Figs. 1D-H, S1-S9).…”

“…Axon quantification. We classified axons in our whole-brain imaging datasets using a hybrid strategy using the 3D U-Net convolutional network TrailMap (Friedmann et al 2020) and an ilastik pixellevel classifier (Berg et al 2019). After adjusting pixel values by multiplication with a scalar to match the background levels seen in the TrailMap training dataset, TrailMap was very good at detecting most axons but missed axons and fiber bundles with the highest signal-to-noise ratio.…”

“…After expression, brains were cleared and imaged, and images were registered onto Allen CCF. Axons were quantified using a custom classification pipeline based on TrailMap (Friedmann et al 2020) and ilastik (Berg et al 2019). ( Hierarchical clustering of excitatory CN cell types in the space of differentially expressed genes, using a correlation-based distance metric.…”

New techniques for studying neurodevelopment

“…ClearMap is another algorithm that maps cells automatically in the mouse brain of LSFM datasets, which could be applied to neonate brains 24 . Another very promising avenue is the algorithm Trailmap, which was recently developed in the Luo lab to automatically identify and extract axonal projections in 3D image volumes 68 and may be easily implemented to improve the quantification of axon guidance studies. Adoption of such neural networks will probably require re-training and optimisation to the specific use-case scenario and dataset, which highlights the need for fast-training computational strategies in order to facilitate the broader use of these techniques.…”