Label-Free 4D Holotomography with Depth-Adaptive Segmentation for Quantitative Analysis of Lipid Droplet Dynamics in Hepatic Organoids

This study introduces a label-free, depth-adaptive 4D holotomography framework to longitudinally quantify single lipid droplet dynamics in living hepatic organoids, revealing that oleic acid and linoleic acid drive lipid accumulation through distinct mechanisms of droplet enlargement versus proliferation, respectively, while palmitic acid rapidly compromises organoid integrity.

The Gist

The Big Picture: Watching Fat Cells Grow Without Breaking Them

Imagine you have a tiny, living city made of liver cells (called an organoid). Inside this city, there are tiny storage bubbles called lipid droplets (LDs). These are like the city's warehouses where fat is kept.

Scientists have always wanted to watch how these warehouses grow, shrink, and multiply over time to understand diseases like fatty liver. But there's a problem: the old way of doing this is like trying to study a living city by taking a photo, then tearing the city apart to paint the buildings, taking another photo, and tearing it apart again. This kills the city and only gives you a snapshot, not a movie.

This paper introduces a new "super-vision" camera that lets scientists watch these living cities grow fat in real-time, without touching them or using any dyes.

---

The New Tool: The "X-Ray" Camera (Holotomography)

The researchers used a special microscope called Holotomography. Think of it like a 3D X-ray camera that sees density instead of bones.

• How it works: Fat is naturally denser (heavier) than the water-based stuff inside a cell. This camera detects that density difference.
• The Benefit: Because it doesn't need to inject any chemicals or dyes (it's "label-free"), the liver cells stay alive and happy. The scientists can film the same organoid for days, watching the fat storage bubbles change right before their eyes.

The Challenge: Seeing Through the Fog

Looking at a single cell is easy, but looking at a whole 3D organoid (which is thick and crowded) is like trying to find specific balloons in a foggy, crowded gym. The deeper you look, the harder it is to see clearly.

To solve this, the team wrote a smart computer program (an algorithm) that acts like a super-smart detective. Instead of using one rule for the whole image, it adjusts its "glasses" depending on how deep it is looking. It knows that a fat bubble deep inside the city looks different than one near the surface, so it adapts to find them all accurately.

The Experiment: Three Different Diets

The scientists fed these living liver cities three different types of fatty acids (fats) to see how they reacted:

1. Oleic Acid (OA): A common "healthy" unsaturated fat (found in olive oil).
2. Linoleic Acid (LA): Another unsaturated fat (found in vegetable oils).
3. Palmitic Acid (PA): A "saturated" fat (found in meat and butter), known to be toxic to cells.

The Results: Three Very Different Stories

1. The "Healthy" Fats (OA and LA)

Both unsaturated fats made the liver cells store more fat, but they did it in completely different ways.

• Oleic Acid (The "Giant Balloon" Strategy):
  Imagine the city already has a few small warehouses. When fed Oleic Acid, the city didn't build many new warehouses. Instead, it took the existing ones and pumped them up until they became giants.

  • Result: Fewer, but huge fat bubbles. The storage became very uneven (some tiny, some massive).

• Linoleic Acid (The "Swarm" Strategy):
  When fed Linoleic Acid, the city didn't make the existing warehouses bigger. Instead, it built thousands of tiny new warehouses everywhere.

  • Result: A massive explosion in the number of fat bubbles, but they stayed small and uniform. It was like a swarm of bees filling the city.

Why does this matter? Even though both fats made the liver "fatty," the way they did it was totally different. This helps scientists understand that not all fats affect the body in the same way, even if the end result looks similar.

2. The "Toxic" Fat (PA)

When the scientists fed the liver cells Palmitic Acid, the city didn't just get fat; it collapsed.

• The cells started to bubble and tear apart (membrane blebbing).
• The structure fell apart within hours.
• Lesson: This confirms that saturated fats can be toxic and destructive to liver cells, unlike the unsaturated fats which the cells could handle (even if they got fat).

The Takeaway

This paper is a breakthrough because it moves science from static snapshots to live movies.

• Before: We knew livers get fat, but we didn't know how the fat bubbles behaved inside living tissue.
• Now: We have a non-invasive way to watch the "dance" of fat storage. We learned that different fats use different "strategies" to store energy: some blow up existing bubbles, while others swarm with new tiny ones.

This new method is like giving doctors a time-traveling microscope that can watch the early stages of liver disease develop in real-time, potentially leading to better treatments for fatty liver disease, diabetes, and obesity.

Technical Summary

1. Problem Statement

Quantifying lipid droplet (LD) dynamics in 3D hepatic organoids is critical for understanding metabolic diseases like fatty liver disease (MAFLD). However, current methodologies face significant limitations:

• Fluorescence-based assays: Require fixation and staining (disrupting native metabolism) or live-cell imaging, which suffers from phototoxicity, photobleaching, and limited penetration depth in thick 3D tissues. These methods often restrict analysis to 2D projections or static endpoints, obscuring droplet-level heterogeneity.
• Label-free limitations: While refractive index (RI) imaging is non-invasive, standard global-threshold segmentation fails in thick organoids due to depth-dependent RI variations, optical aberrations, and heterogeneous background structures.
• Lack of granularity: Existing analyses often rely on bulk metrics (e.g., total lipid content), failing to distinguish whether lipid accumulation arises from an increase in droplet number, droplet size, or changes in spatial organization.

2. Methodology

The authors developed a comprehensive framework combining Holotomography (HT) with a custom depth-adaptive segmentation algorithm to enable longitudinal, label-free 4D (3D + time) analysis.

A. Imaging Platform

• Technique: Holotomography (HT-X1, Tomocube Inc.), a quantitative phase imaging technique that reconstructs 3D RI distributions from multiple angle-resolved intensity measurements.
• Sample: Living mouse hepatic organoids (mHOs) cultured in 96-well plates under physiological conditions (37°C, 5% CO₂).
• Protocol: Time-lapse imaging at 6-hour intervals for up to 42 hours. Organoids were treated with free fatty acids (FFAs): Oleic Acid (OA), Linoleic Acid (LA), and Palmitic Acid (PA).
• Validation: Correlative imaging with LipiDye II fluorescence was used to verify that high-RI structures correspond to LDs.

B. Algorithmic Pipeline

1. Organoid Segmentation: Adapted from a previous MP-HT pipeline using adaptive Otsu thresholding and morphological closing to define the organoid boundary and lumen.
2. Depth-Adaptive LD Segmentation:
   • Challenge: Global thresholds fail due to depth-dependent contrast loss.
   • Solution: A rule-based algorithm in MATLAB detects LD candidates across nine distinct RI contrast thresholds ($\Delta n = 0.005$ to $0.030$).
   • Filtering: Candidates are filtered using 3D morphological criteria (compactness, solidity, aspect ratio) and a 2D Feret diameter ratio to ensure near-sphericity.
   • Refinement: A size-adaptive refinement step uses Maximum Intensity Projection (MIP) to suppress false positives from small noise while preserving large droplets.
3. Quantitative Feature Extraction:
   • Metrics: LD count, volume (fL), surface area, sphericity ($\Psi$), effective lipid concentration ($c_{LD}$), and dry mass ($M_{LD}$).
   • Calculation: Lipid concentration is derived from RI using the formula $c_{LD} = (n_{LD} - n_m) / \alpha$ (where $\alpha = 0.135$ mL/g). Dry mass is calculated as $M_{LD} = c_{LD} \times V_{LD}$.

C. Validation

• Performance: The segmentation achieved a Precision of 0.676, Recall of 0.900, and an F1-score of 0.771 when compared against fluorescence ground truth (using a volume-weighted overlap criterion to account for axial resolution limits).

3. Key Contributions

1. Label-Free 4D Framework: Established a non-invasive, longitudinal method to track LD dynamics in intact, living 3D organoids without phototoxicity or staining artifacts.
2. Depth-Adaptive Segmentation: Developed a novel multi-threshold algorithm that overcomes depth-dependent RI variations, enabling robust single-droplet resolution throughout the entire organoid volume.
3. Single-Droplet Resolution: Moved beyond population averages to quantify individual droplet kinetics (number, size, mass, concentration) over time.
4. Discovery of Divergent Accumulation Strategies: Revealed that different unsaturated fatty acids induce lipid storage through fundamentally distinct cellular mechanisms (enlargement vs. proliferation).

4. Key Results

A. Differential Responses to Fatty Acids

• Unsaturated FAs (OA & LA): Both induced robust LD accumulation while largely preserving organoid structural integrity.
• Saturated FA (PA): Triggered rapid structural collapse, membrane blebbing, and organoid disintegration within hours, confirming its lipotoxic nature. PA-treated organoids were excluded from quantitative kinetic analysis due to this instability.

B. Distinct Accumulation Modes (OA vs. LA)

Despite both increasing total lipid burden, OA and LA drove accumulation via different temporal strategies:

• Oleic Acid (OA):
  • Mechanism: Volume-dominated accumulation.
  • Dynamics: Characterized by the preferential enlargement of a smaller number of existing droplets.
  • Outcome: Increased size heterogeneity (higher coefficient of variation in volume) and a plateau in total droplet count after initial growth.
• Linoleic Acid (LA):
  • Mechanism: Number-dominated accumulation.
  • Dynamics: Characterized by a sustained, continuous increase in the number of LDs.
  • Outcome: A more uniform population of smaller droplets with stable size distribution (low heterogeneity).
• Biophysical Insight: LA-treated LDs showed slightly higher lipid concentrations, potentially reflecting the sequestration of polyunsaturated fatty acids (susceptible to peroxidation) into neutral lipid stores as a protective mechanism.

C. Organoid-Level Metrics

• Total LD volume and dry mass increased significantly in OA and LA groups compared to controls.
• Organoid volume expanded more significantly in LA-treated groups, suggesting an injury-like swelling response or lumen collapse in some cases.

5. Significance and Impact

• Methodological Advance: Solves the "missing cone" and depth-contrast problems in 3D RI imaging, providing a robust tool for quantitative analysis in thick biological tissues.
• Biological Insight: Demonstrates that the mode of lipid storage (hypertrophy vs. hyperplasia of droplets) is dictated by the specific fatty acid class, a nuance invisible to bulk measurements.
• Clinical Relevance: The ability to distinguish between microvesicular (many small droplets) and macrovesicular (few large droplets) steatosis dynamically is crucial for studying metabolic disorders (NAFLD/MAFLD, diabetes) and drug screening.
• Future Applications: This non-invasive framework is ideal for longitudinal drug screening and studying metabolic remodeling in complex, multi-lineage organoids without the perturbation caused by fluorescent labels.

Limitations Noted:

• Axial resolution is lower than lateral resolution due to the "missing cone" problem in HT, potentially biasing shape/sphericity measurements.
• RI-based dry mass estimates can be affected by missing-cone artifacts.
• Segmentation accuracy in deep regions can be confounded by high-RI non-lipid structures.