Spatial single-cell Organellomics reveals nutrient dependent hepatocyte heterogeneity and predicts pathophysiological status in vivo

Alexander Hillsley, Raghabendra Adhikari, Alexander Johnson, Isabel Espinosa-Medina, Jan Funke, Daniel FelicianobioRxiv 2024 · DOI

TL;DR

This paper introduces "scOrganellomics," a novel imaging and computational pipeline that analyzes organelle signatures within individual liver cells to reveal functional heterogeneity. They demonstrate that these organelle signatures can accurately predict a cell's location within the liver, its nutritional status, and even the progression of metabolic disease (MASLD).

What Problem Does This Paper Address?

Understanding how cells within a complex organ like the liver adapt to their local environment and nutritional cues is crucial for comprehending tissue function and disease. While transcriptional analysis has provided insights, it often requires tissue dissociation, losing vital spatial information. The authors highlight that the real "machinery" for metabolic reactions lies within organelles, yet how their organization changes in vivo in response to microenvironments and nutritional shifts remains largely unexplored. This gap limits our ability to understand cellular adaptation and predict disease states at a fine-grained level.

Methods

The core innovation is the scOrganellomics pipeline. This involves:

  • 3D Confocal Microscopy: High-resolution imaging of fixed mouse liver tissue.
  • Organelle Labeling: Simultaneous visualization of mitochondria, peroxisomes, and lipid droplets using fluorescent markers.
  • Cellpose-based Segmentation: A deep learning model (Cellpose v2) is used for automated segmentation of cells and their organelles. This is a key advancement for robust and scalable organelle identification.
  • Liv-Zones Pipeline: A custom Python package for processing the segmented images, extracting approximately 100 organelle metrics per cell.
  • Computational Analysis: Principal Component Analysis (PCA) and Gaussian Mixture Models (GMM) are used to cluster hepatocytes based on their organelle "signatures."
  • Machine Learning: Multilayer Perceptron (MLP) models are trained using organelle signatures to predict nutritional status, hepatocyte categories, and disease progression.

Datasets

The study utilized datasets generated from:

  • Mouse Liver Samples: From mice subjected to three different conditions:
  • Control diet
    • Overnight fasting
    • High-fat/high-carbohydrate "Western Diet" (WD)
  • MASLD Progression Model: Liver samples from mice fed a WD for varying durations (0 to 50 days) to model early stages of Metabolic dysfunction-associated steatotic liver disease (MASLD).

The authors generated a dataset of over 4000 hepatocytes and millions of organelles for their analysis.

Key Insights and Ideas

  • Organelles as "Structural Fingerprints": The paper posits that the specific arrangement and characteristics of organelles within a cell act as a unique "signature" that reflects its functional state and adaptation.
  • Beyond Transcriptomics: This work demonstrates that organelle morphology and organization provide a powerful layer of information for understanding cellular heterogeneity, complementing or even surpassing transcriptional data in certain contexts, especially when spatial information is critical.
  • Nutritional Perturbations Constrain Heterogeneity: Surprisingly, changes in diet (fasting or WD) didn't just shift organelle signatures; they actively reduced the diversity of these signatures within the liver, suggesting a mechanism for adapting to new demands by simplifying cellular states.
  • Spatial Organization Matters: Hepatocyte categories defined by organelle signatures are not randomly distributed but show distinct patterns along the portal vein (PV) to central vein (CV) axis, suggesting a functional zonation based on organelle architecture.
  • Predictive Power of Organelle Signatures: The ability of organelle signatures alone to predict nutritional status and disease progression with high accuracy is a significant finding, opening doors for non-invasive diagnostics or prognostics.

Key Results

  • scOrganellomics Pipeline: Successfully segmented hepatocytes and ~100 organelle metrics from millions of organelles in intact liver tissue.
  • Hepatocyte Categorization: Identified 5 distinct hepatocyte categories in control mice based on organelle signatures, with differential distribution across liver zones.
  • Nutritional Impact: Nutritional perturbations (fasting, WD) reduced the number of distinct hepatocyte categories from 5 to 3 or 4, indicating a decrease in overall hepatocyte heterogeneity.
  • Machine Learning Accuracy:
    • Predicted nutritional status with 99.4% accuracy.
  • Distinguished between all 12 hepatocyte categories (across all groups) with 94.8% accuracy.
  • Predicted MASLD progression stages with ~80% accuracy.

Limitations and Future Work

  • Organelle Contact Analysis: The current resolution limits the analysis of inter-organelle contacts, which could provide further insights into differential functions. Future work could involve higher-resolution imaging techniques like volumetric EM or Expansion Microscopy.
  • Transgenic Lines: While they used fluorescent markers, developing transgenic lines that combine multiple organelle markers would allow for more comprehensive analysis in living organisms.
  • Broader Tissue Application: The authors envision extending this strategy to other tissues and species, suggesting its broad applicability beyond the liver.
  • Human Liver Assessment: The pipeline could be adapted to assess the status of human livers under various disease conditions, potentially transforming disease modeling and personalized therapy.

Summary updated

Abstract

Abstract Cellular heterogeneity within complex tissues and organs is essential to coordinate biological processes across biological scales. The effect of local cues and tissue microenvironments on cell heterogeneity has been mainly studied at the transcriptional level. However, it is within the subcellular scale - the organelles - that lays the machinery to conduct most metabolic reactions and maintain cells alive, ensuring proper tissue function. How changes in subcellular organization under different microenvironments define the functional diversity of cells within organs remains largely unexplored. Here we determine how organelles adapt to different microenvironments using the mouse liver as model system, in combination with computational approaches and machine-learning. To understand organelle adaptation in response to changing nutritional conditions, we analyzed 3D fluorescent microscopy volumes of liver samples labeled to simultaneously visualize mitochondria, peroxisomes, and lipid droplets from mice subjected to different diets: a control diet, a high-fat diet, and a control diet plus fasting. A Cellpose based pipeline was implemented for cell and organelle segmentation, which allowed us to measure ~100 different organelle metrics and helped us define subcellular architectures in liver samples at the single cell level. Our results showed that hepatocytes display distinct subcellular architectures within different regions of the liver-close to the central vein, in the middle region, and near the portal vein- and across the various diet groups, thus reflecting their adaptation to specific nutritional inputs. Principal component analysis and clustering of hepatocytes based on organelle signatures revealed 12 different hepatocyte categories within the different experimental groups, highlighting a reduction in hepatocyte heterogeneity under nutritional perturbations. Finally, using single cell organelle signatures exclusively, we generated machine learning models that were able to predict with high accuracy different hepatocyte categories, diet groups, and the stages of MASLD. Our results demonstrate how organelle signatures can be used as hallmarks to define hepatocyte heterogeneity and their adaptation to different nutritional conditions. In the future, our strategy, which combines subcellular resolution imaging of liver volumes and machine learning, could help establish protocols to better define and predict liver disease progression.
BibTeX
@misc{hillsley2024spatial,
  author = {Alexander Hillsley and Raghabendra Adhikari and Alexander Johnson and Isabel Espinosa-Medina and Jan Funke and Daniel Feliciano},
  title = {{Spatial single-cell Organellomics reveals nutrient dependent hepatocyte heterogeneity and predicts pathophysiological status <i>in vivo</i>}},
  year = {2024},
  doi = {10.1101/2024.12.06.627285},
  howpublished = {bioRxiv},
}
Figure