Mapping the Cellular Landscape of the Brain: A Scalable Approach to Comprehensive Microscopy Data Analysis

Abstract

Recent advances in intact tissue processing and imaging have enabled the generation of whole brain microscopy data at subcellular resolution, revealing intricate morphological details of cells at unprecedented scales. Given that cellular morphology is strongly linked to distinct functional states of cells, in-depth morphological analysis of such data offers immense potential for understanding their roles in brain development and disease. However, the lack of scalable computational techniques poses a substantial challenge in achieving comprehensive morphological characterization. To efficiently and accurately analyze cellular morphology, we need to process terabyte-scale three-dimensional (3D) data, which inevitably complicates downstream analysis workflows with existing methods.

To address the challenge, we developed an end-to-end scalable framework that seamlessly strings each step of the analysis pipeline together, enabling comprehensive fluorescence microscopy data analysis. The framework, termed MorPheT (Morphology Phenotyping Tool), serves as an all-in-one solution, offering a suite of analysis modules spanning from image pre-processing to precise cell detection, atlas alignment, morphological phenotyping, and interactive visualizations. MorPheT employs an ensemble method using both supervised and unsupervised approaches to maximize feature learning for unbiased morphological characterization. A novel deep neural network (ALNet) was designed to capture the long-range contextual dependencies inherent in 3D training data during supervised learning. Unsupervised learning leverages complementary features from the supervised approach, demonstrating the powerful synergy of this ensemble method.

We applied MorPheT to two main projects. First, we profiled brain-resident macrophages (BRMs) and created the first fetal mouse brain atlases across multiple developmental stages, revealing distinct regional growth patterns of BRMs throughout development. We also demonstrated MorPheT’s effectiveness in characterizing microglia distribution patterns and morphological properties brain-wide in both control and neurodegeneration mouse brains. In the second project, we investigated cFos+ cells in a memory engram study, showcasing MorPheT’s utility for brain-wide analysis of engram cells. By examining regions hypothesized to hold memory engrams for contextual fear conditioning memory, we identified brain regions where engrams for a specific memory are distributed. Taken together, MorPheT is a powerful tool for cell profiling and mapping across the brain, and we anticipate it will help democratize computational analysis for large-scale microscopy datasets, making advanced analytical approaches more accessible to the broader scientific community.