Deep Learning and 3D Imaging Reveal Whole-Body Alterations in Obesity

Doris Kaltenecker, Izabela Horvath, Rami Al-Maskari, Zeynep Ilgin Kolabas, Ying Chen, Luciano Hoeher, Mihail Ivilinov Todorov, Saketh Kapoor, Mayar Ali, Florian Kofler, Pauline Morigny, Julia Geppert, Denise Jeridi, Carolina Cigankova, Victor Miro Kolenic, Nilsu Gur, Chenchen Pan, Marie Piraud, Daniel Rueckert, Maria Rohm, Farida Hellal, Markus Elsner, Harsharan S. Bhatia, Bjoern Menze, Stephan Herzig, Johannes C. Paetzold, Mauricio Berriel Díaz, Ali ErtürkbioRxiv 2024 · DOI

Abstract

Systemic diseases such as obesity have profound effects that impact multiple organ systems throughout the body. However, tools for comprehensive, high-resolution analysis of these changes at a whole-body scale have been lacking. We developed MouseMapper, a suite of deep learning-based image analysis algorithms integrated with tissue clearing and light-sheet microscopy, to enable holistic quantification of cellular and structural changes. Applying MouseMapper to high-fat diet-induced obesity, we uncovered widespread alterations in the nervous and immune systems, including a 67% reduction in facial nerve endings and significant heterogeneity in systemic inflammation. This study demonstrates a powerful approach for investigating the systemic impacts of complex diseases.

Introduction: Beyond the "Keyhole" View

Traditional biological analysis often relies on 2D histology—slicing small pieces of tissue to examine them under a microscope. While informative, this "keyhole" view misses the big picture of systemic diseases like obesity, which affect the entire body's interconnected networks. The challenge in moving to 3D whole-body analysis is the sheer scale of the data; a single mouse scan can exceed 10 terabytes, making manual annotation impossible. MouseMapper addresses this by providing a scalable, automated framework for quantifying these changes holistically.

The MouseMapper Pipeline

The authors integrated a specialized imaging workflow with a new software suite. The process begins with vDISCO clearing, which renders mouse bodies transparent while labeling specific targets with nanobodies. These are then imaged using Light-Sheet Fluorescence Microscopy (LSFM).

Whole-body vDISCO clearing and imaging comparing chow-fed (left) and obese (right) mice.

The MouseMapper software consists of three primary deep-learning modules based on 3D UNet architectures:

  • Nerve-Module: Segments the peripheral nervous system and converts it into mathematical graphs to quantify connectivity.
  • Immune-Module: Detects and clusters CD68+ macrophages, categorizing them by size to assess inflammatory states.
  • Tissue-Module: Automatically segments 20 different organs and tissues to provide anatomical context.

Systemic "Rewiring" in Obesity

Applying MouseMapper to obesity revealed that high-fat diets trigger a dramatic structural pruning of the nervous system. A striking finding was the degradation of the infraorbital nerve (a branch of the trigeminal nerve). While the main nerve trunk remained intact, the fine endings near the skin surface were reduced by over 67%. This structural loss correlated with a measurable decrease in whisker sensitivity, suggesting that obesity-induced neuropathy directly impairs sensory perception.

Furthermore, the pipeline revealed that inflammation is not uniform across the body. While visceral fat showed a massive shift toward large, pro-inflammatory immune clusters, other tissues like skeletal muscle remained relatively stable, highlighting the heterogeneity of the disease's impact.

Quantitative Results and Performance

The study generated a massive dataset of 12 whole-mouse scans totaling approximately 10.35 Terabytes. Key quantitative findings include:

  • Nerve Degradation: Obese mice showed a 67.1% reduction in nerve endings and a 69.5% drop in connections (edges).
  • Systemic Inflammation: Significant increases in large immune cell clusters (>500 cells) were found in visceral fat, the peritoneum, and the colon.
  • Model Accuracy: The 3D UNet achieved high volumetric Dice scores (0.79 for nerves, 0.88 for tissues), outperforming alternative architectures.
Visualization of systemic immune cell distribution and clustering across the mouse body.

Conclusion and Future Outlook

MouseMapper demonstrates the capability to discover and quantify pathological alterations at the whole-body level. Although computationally intensive—requiring over 7,000 GPU hours—it offers a transformative approach for investigating systemic diseases. Future work will adapt the pipeline to study other networks, such as the lymphatic or vascular systems, and apply it to conditions like aging and metastatic cancer spread. The authors have made their data and code publicly available to encourage further community development.

Summary updated

Abstract

Many diseases, such as obesity, have systemic effects that impact multiple organ systems throughout the body. However, tools for comprehensive, high-resolution analysis of disease-associated changes at the whole-body scale have been lacking. Here, we developed a suite of deep learning-based image analysis algorithms (MouseMapper) and integrated it with tissue clearing and light-sheet microscopy to enable a comprehensive analysis of diseases impacting diverse systems across the mouse body. This approach enables the quantitative analysis of cellular and structural changes across the entire mouse body at unprecedented resolution and scale, including tracking nerves over several centimeters in whole animal bodies. To demonstrate its power, we applied MouseMapper to study nervous and immune systems in high-fat diet induced obesity. We uncovered widespread changes in both immune cell distribution and nerve structures, including alterations in the trigeminal nerve characterized by a reduced number of nerve endings in obese mice. These structural abnormalities were associated with functional deficits of whisker sensing and proteomic changes in the trigeminal ganglion, primarily affecting pathways related to axon growth and the complement system. Additionally, we found heterogeneity in obesity-induced whole-body inflammation across different tissues and organs. Our study demonstrates MouseMapper's capability to discover and quantify pathological alterations at the whole-body level, offering a powerful approach for investigating the systemic impacts of various diseases.
BibTeX
@misc{kaltenecker2024deep,
  author = {Doris Kaltenecker and Izabela Horvath and Rami Al-Maskari and Zeynep Ilgin Kolabas and Ying Chen and Luciano Hoeher and Mihail Ivilinov Todorov and Saketh Kapoor and Mayar Ali and Florian Kofler and Pauline Morigny and Julia Geppert and Denise Jeridi and Carolina Cigankova and Victor Miro Kolenic and Nilsu Gur and Chenchen Pan and Marie Piraud and Daniel Rueckert and Maria Rohm and Farida Hellal and Markus Elsner and Harsharan S. Bhatia and Bjoern Menze and Stephan Herzig and Johannes C. Paetzold and Mauricio Berriel Díaz and Ali Ertürk},
  title = {{Deep Learning and 3D Imaging Reveal Whole-Body Alterations in Obesity}},
  year = {2024},
  doi = {10.1101/2024.08.18.608300},
  howpublished = {bioRxiv},
}
Figure