Modular strategies for spatial mapping of diverse cell type data of the mouse brain

James C. Gee, Nicholas J. Tustison, Min Chen, Fae N. Kronman, Jeffrey Duda, Clare Gamlin, Mia G. Tustison, Michael Kunst, Rachel Dalley, Staci A. Sorensen, Quanxin Wang, Lydia Ng, Yongsoo KimResearch Square 2025 · DOI

TL;DR

Mapping diverse mouse brain cell type data into a common reference space is crucial for understanding brain function, but presents significant challenges due to variations in data acquisition and processing. This paper introduces modular, open-source mapping strategies within the ANTsX framework, enabling researchers to integrate various data types (spatial transcriptomics, morphology, developmental) into existing atlases, and offers novel tools for continuous developmental mapping and automated brain parcellation.

What Problem Does This Paper Address?

The BRAIN Initiative Cell Census Network (BICCN) is building an unprecedentedly comprehensive database of mouse brain cell types, utilizing over 40 different profiling techniques from numerous research groups. A major hurdle is integrating these diverse datasets into common reference spaces (atlases) for joint analysis of spatial, structural, and functional information. The problem stems from the inherent variability in how data is acquired, processed, and imaged, leading to unique tissue distortions and signal characteristics for each data type. This heterogeneity prevents a one-size-fits-all mapping solution, necessitating tailored approaches for each modality to overcome specific barriers and accurately align data within a common coordinate framework (CCF).

Methods

The core of this work lies in the development and application of modular atlas mapping strategies within the Advanced Normalization Tools Ecosystem (ANTsX). ANTsX is a powerful, open-source framework for image analysis and registration. The paper highlights three key contributions:

  • MERFISH and fMOST Data Mapping: They developed pipelines to map spatial transcriptomic (MERFISH) and high-resolution morphology (fMOST) data into the Allen Common Coordinate Framework (AllenCCFv3). These pipelines leverage ANTsX's registration capabilities, including preprocessing steps like bias field correction and denoising, and employ specific strategies tailored to the challenges of each modality (e.g., section matching for MERFISH, an intermediate fMOST atlas for fMOST).
  • Continuous Developmental Trajectory Mapping: A novel velocity flow-based approach was developed to map developmental trajectories within the Developmental Common Coordinate Framework (DevCCF). This method addresses temporal gaps in discrete developmental atlases by generating a continuous, time-parameterized velocity field, allowing for the inference of transformations between any two time points and the creation of virtual templates for intermediate developmental stages.
  • Automated Structural Parcellation: An automated framework for determining structural morphology was created, leveraging deep learning components. This pipeline integrates brain extraction and parcellation for multi-modal MRI data, utilizing few-shot and single-shot learning techniques to overcome the typical requirement for large training datasets in mouse brain research.

Datasets

The study utilizes several key datasets:

  • MERFISH data: Acquired using a previously detailed procedure, involving fresh frozen C57BL/6 mouse brains sectioned at 10µm and imaged for 500 genes.
  • fMOST data: High-throughput, high-resolution fluorescence micro-optical sectioning tomography data from 55 mouse brains.
  • DevCCF data: A public resource comprising multi-modal MRI and LSFM symmetric ANTsX-generated templates sampling mouse brain development from embryonic (E11.5) to postnatal (P56) stages.
  • CAMRI dataset: Used for training the mouse brain extraction network, consisting of 16 T2-w MRI volumes.
  • High-resolution dataset: Another dataset used for training, comprising 88 specimens with three spatially aligned canonical views.
  • Independent evaluation dataset: A publicly available dataset of 12 specimens imaged at seven time points, used to evaluate the brain extraction and parcellation networks. This dataset has anisotropic spacing.

Key Insights and Ideas

  • Modularity is Key: The ANTsX framework's modularity allows for the development of tailored pipelines that can address the unique challenges of diverse neuroimaging modalities, rather than relying on a single, monolithic approach.
  • Bridging Temporal Gaps: The velocity flow model offers a powerful way to create a continuous representation of developmental processes, enabling more nuanced analysis of intermediate stages that are not explicitly captured in discrete atlases.
  • Deep Learning for Mouse Brain Morphology: The application of few-shot and single-shot deep learning for brain extraction and parcellation significantly lowers the barrier to entry for mouse brain morphological analysis, reducing reliance on extensive manual annotation.
  • Leveraging Existing Resources: The work emphasizes the power of building upon existing open-source tools and datasets (like the AllenCCFv3 and DevCCF) to accelerate research and promote reproducibility.

Key Results

  • Successful Mapping: The developed pipelines successfully mapped MERFISH and fMOST data into the AllenCCFv3, with qualitative assessment by expert anatomists and quantitative metrics (e.g., Dice values for fMOST structures ranging from 0.63 to 0.99).
  • Continuous Development Mapping: The velocity flow model was successfully implemented and demonstrated to generate transformations between developmental stages, allowing for the creation of virtual templates.
  • High Accuracy in Brain Extraction: The automated brain extraction network achieved good agreement with user-generated brain masks on an independent dataset, with a mean Dice overlap of 0.94 for whole brain extraction.
  • Improved Parcellation Accuracy: The parcellation scheme, when used in conjunction with intensity information, significantly improved regional alignment compared to intensity-only registration.

Limitations and Future Work

  • MERFISH Data Challenges: While the MERFISH pipeline addresses specific challenges, the paper notes that the 3D orientation of MERFISH sections may not be fully accounted for during volume reconstruction, leading to potential obliqueness artifacts.
  • fMOST Data Resolution: Mapping high-resolution fMOST data to the lower-resolution AllenCCFv3 inevitably leads to some loss of high-resolution information.
  • Deep Learning Refinements: The authors acknowledge that while their deep learning approaches show initial success, refinements will likely be necessary, and they plan to manually correct failing cases for future training.
  • Future Work: The paper suggests that the velocity flow model has potential applications beyond brain development, such as modeling cardiac and respiratory cycles. They also welcome community contributions to improve their deep learning approaches and plan to extend these efforts to human brain mapping.

Summary updated

Abstract

Large-scale, international collaborative efforts by members of the BRAIN Initiative Cell Census Network (BICCN) consortium are aggregating the most comprehensive reference database to date for diverse cell type profiling of the mouse brain, which encompasses over 40 different multi-modal profiling techniques from more than 30 research groups. One central challenge for this integrative effort has been the need to map these unique datasets into common reference spaces such that the spatial, structural, and functional information from different cell types can be jointly analyzed. However, significant variation in the acquisition, tissue processing, and imaging techniques across data types makes mapping such diverse data a multifarious problem. Different data types exhibit unique tissue distortion and signal characteristics that precludes a single mapping strategy from being generally applicable across all cell type data. Tailored mapping approaches are often needed to address the unique barriers present in each modality. This work highlights modular atlas mapping strategies developed across separate BICCN studies using the Advanced Normalization Tools Ecosystem (ANTsX) to map spatial transcriptomic (MERFISH) and high-resolution morphology (fMOST) mouse brain data into the Allen Common Coordinate Framework (AllenCCFv3), and developmental (MRI and LSFM) data into the Developmental Common Coordinate Framework (DevCCF). We discuss common mapping strategies that can be shared across modalities and driven by specific challenges from each data type. These mapping strategies include novel open-source contributions that are made publicly available through ANTSX. These include 1) a velocity flow-based approach for continuously mapping developmental trajectories such as that characterizing the DevCCF and 2) an automated framework for determining structural morphology solely through the leveraging of publicly resources. Finally, we provide general guidance to aid investigators to tailor these strategies to address unique data challenges without the need to develop additional specialized software.
BibTeX
@inproceedings{gee2025modular,
  author = {James C. Gee and Nicholas J. Tustison and Min Chen and Fae N. Kronman and Jeffrey Duda and Clare Gamlin and Mia G. Tustison and Michael Kunst and Rachel Dalley and Staci A. Sorensen and Quanxin Wang and Lydia Ng and Yongsoo Kim},
  title = {{Modular strategies for spatial mapping of diverse cell type data of the mouse brain}},
  booktitle = {Research Square},
  year = {2025},
  doi = {10.21203/rs.3.rs-6289741/v1},
}
Figure