Animals
In a study involving male Uchl1-eGFP and Cd68-eGFP mice, aged eight weeks on a C57BL/6J background, and wild-type C57BL/6J mice, the subjects were allowed to eat either a regular chow diet or a high-fat diet (60% fat content, source D12492i from Research Diets) for a period ranging from 16 to 18 weeks. The mice were kept under a 12-hour light-dark cycle, with room temperatures set between 20 to 24 °C and humidity levels maintained at 45 to 65%. Body composition of the mice was assessed using an EchoMRI-100H system. For insulin tolerance tests, the mice were fasted for six hours and given an intraperitoneal injection of insulin (0.75 U kg−1). Blood glucose was monitored through tail vein samples using glucose test strips at specific time intervals. After the procedures, the mice were euthanized following deep anesthesia with a combination of ketamine and xylazine, then underwent intracardiac perfusion with heparinized PBS (10 U ml−1) followed by a perfusion with 4% paraformaldehyde (PFA). They were then post-fixed in 4% PFA overnight, washed multiple times with PBS, shaken at room temperature. This experiment followed the European Union directives and German animal welfare regulations, with approvals from the state ethics committee and Upper Bavaria government.
Human participants
Samples from trigeminal ganglia were collected post-mortem from body donors at the Institute of Anatomy in Leipzig, Germany, after obtaining informed and written consent for research and educational purposes (ethical approval number 129/21-ck). Participants were categorized based on BMI into lean (BMI < 30) and obese (BMI ≥ 30) groups. Additional details on age and sex can be reviewed in Supplementary Table 11. Proteomic profiling involved dissection of three regions of interest from each ganglion for each participant.
Whisker stimulation test
The methodology for the whisker stimulation test was adapted from earlier studies. Mice were kept in their original cages to eliminate variables. The test employed a cotton swab to stroke the whiskers of the mice, first allowing the swab to touch the mouse’s head, followed by four strokes to each side. Reactions were evaluated using a modified whisker score system. A score was given based on whether the mouse turned its head or initiated grooming, with a score of one for positive responses and zero for a lack of response. Each side received four stimulations, and scores were recorded by an evaluator unaware of the treatment condition. The total potential score was eight, with higher scores (3-4) indicating normal responses and lower scores (0-2) pointing to possible sensory deficits.
vDISCO nanobody labeling and clearing
The vDISCO method involved previously outlined procedures and combined active and passive GFP-Nanobooster labeling for enhanced signal clarity. Mice underwent DISCO clearing using a series of tetrahydrofuran (THF) solutions followed by benzyl alcohol/benzyl benzoate until tissue transparency was achieved.
WildDISCO antibody labeling and clearing
WildDISCO antibody labeling also adhered to established techniques, utilizing specific antibodies for Uchl1 and CGRP. Subsequently, mice were subjected to the DISCO clearing process as described earlier.
Fluorescence light-sheet imaging
For whole-body imaging of mice, light-sheet imaging was employed using an Ultramicroscope Blaze along with a dipping objective lens. The imaging involved tiling scans and stitching together images using specific software, capturing detailed cellular structures at various resolutions.
3D reconstruction
3D reconstructions combined dorsal and ventral scans, facilitating an organized view of the sample using designated analysis software. The compiled data in TIFF format underwent further image analysis for increased clarity.
VR data annotation
In virtual reality, data annotation concentrated on creating a reliable nerve segmentation model. A diverse dataset was curated from Uchl1-eGFP mouse scans, leading to extensive patches used in model assessment to ensure accuracy and variability in anatomical contexts.
Annotation for Cd68-eGFP+ cells involved representative regions selected for training and evaluation of 3D networks. This process enhanced the refinement of segmentation capabilities on the specified markers.
The Tissue-Module development involved annotating various organs across multiple mouse scans, effectively distinguishing tissues. A reference dataset was created based on vast volumes of tissue data, progressively improved through ongoing inference and manual corrections.
Peripheral nerve segmentation
The Nerve-Module utilized a fine-tuned foundation model, leveraging extensive training to adapt segmentation outputs for nerve-specific data. The fine-tuning strategy supported stable convergence while integrating previous knowledge from broader datasets.
To enhance segmentation accuracy, normalization procedures before training and testing were implemented, ensuring improved model performance by emphasizing nerve regions.
Comparative analysis of the nerve segmentation model against various advanced 3D segmentation networks revealed significant insights, employing a strategy that included diverse external datasets for evaluation.
Immune cell segmentation
In developing the Immune-Module, the segmentation network for CD68 focused on fine-tuning while freezing the encoder to adapt the output for the specific immune dataset. Several baseline architectures were compared for performance metrics, leading to a selection of the top-performing model based on previous assessments.
Organ and tissue segmentation
Training networks for organ segmentation involved both Cd68-eGFP and Uchl1-eGFP samples, utilizing multiple models. Various network architectures contributed to effective segmentation of different tissue types, evaluated through a robust selection framework.
Subsequently, soft tissue segmentation focused on substantial volumes, incorporating a growing dataset through iterative learning. The results favored specific networks, particularly the 3D UNet, combining various annotated anatomical data from samples.
The inference pipeline for the Tissue-Module enabled clear sequential segmentation of organs followed by tissues, generating effective spatial maps for quantifying structures within the mouse anatomy.
Whole-body inference
For applying the segmentation networks across full-body scans, an efficient inference method was refined to accommodate various dataset requirements, ensuring rapid analysis without loss of detail.
Normalization techniques prior to inference allowed better signal differentiation, critical for comparing nerve and tissue identifications during analytical processes.
Cd68-eGFP segmentation quantification
Post-segmentation analyses focused on extracting components and assessing volume and structural data, facilitating a thorough evaluation of segmented cells against established anatomical references.
Validations included visual comparisons among automatic and manual annotations to appraise the network’s capabilities across varied tissues, utilizing metrics like the Dice score for strength comparisons.
Uchl1-eGFP segmentation quantification
Following inference on Uchl1-eGFP mice, connected component analyses were conducted to refine segmentation outcomes, disregarding potential artifacts for accurate quantification across full-body perspectives.
For nerve quantification, comprehensive structuring allowed delineation across regions and compared tissue-specific statistics to advance understanding of nerve density within various environments.
Graph extraction
Graph extraction employed prior techniques involving skeletonization and mapping, optimizing segmentations against image memory limitations through systematic sub-block methodologies.
Computational load of MouseMapper
The overall computational efforts involved processing a large cohort of mice, generating expansive datasets while necessitating substantial GPU resource hours for training and inference across multiple segmentation tasks.
The annotation workload reflected significant engagement in dataset management, emphasizing the need for high-performance computing solutions to facilitate efficient analysis.
Spatial proteomics sample preparation
In preparing spatial proteomics samples from trigeminal ganglia, protocols were followed for effective protein extraction utilizing multiple preparative techniques to ensure integrity for analytical processes.
Human sample preparations involved distinct lysis methods and clean-up protocols that were essential for the quality of resulting data, reflecting the need for rigorous preparatory standards.
Evotip PURE clean-up of human samples
Sample clean-up was executed with care, aligning with established methodologies to ensure the removal of contaminants while optimizing yield through refined processing steps.
LC–MS
Data acquisition for both mouse and human samples followed rigorous statistical methods aimed at providing comprehensive analysis through detailed mass spectrometry techniques.
Proteomics data processing
We implemented robust data processing pipelines that integrated modern analysis tools, managing extensive datasets to isolate relevant proteins while maintaining high fidelity in results.
Proteomics data analysis
Statistical analyses of protein expression considered key comparison metrics, focusing on significant differences across groups and pathways to inform understanding of biological implications.
Western blot
Protein lysate preparations utilized optimized conditions for subsequent analyses, emphasizing accuracy in dilution protocols and antibody applications within gel electrophoresis settings.
Multiplex antibody labelling and analysis
Employing advanced multiplexing technology, we performed thorough tissue analyses to generate rich informative insights through intricate processing and visualization strategies.
Statistical analysis
Results were expressed with appropriate statistical measures, focusing on defining comparisons across groups. Methods of analysis adhered to established standards within the field, ensuring credibility across conclusions drawn.
Reporting summary
Detailed information on research design can be found in the linked Nature Portfolio Reporting Summary associated with this study.
