📄 cell biology

HiExM Enables Scalable Mapping of Organelle Morphology and Spatial Heterogeneity

This paper introduces HiExM, a scalable high-throughput expansion microscopy pipeline that overcomes the trade-off between spatial resolution and sample size to enable robust, quantitative profiling of individual organelle morphology and spatial heterogeneity across large cell populations.

Original authors: Day, J. H., Farrell, J. D., Yang, D., Neira, F. N., Allen, E. A., Byrne, A. M., Leksa, N. C., Klinger, K. W., de Nola, G., Al-Jazrawe, M., Boyer, L. A.

Published 2026-07-14

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Day, J. H., Farrell, J. D., Yang, D., Neira, F. N., Allen, E. A., Byrne, A. M., Leksa, N. C., Klinger, K. W., de Nola, G., Al-Jazrawe, M., Boyer, L. A.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine trying to take a group photo of a bustling city, but your camera is so blurry that every person looks like a fuzzy blob, and all the buildings merge into one giant, indistinct smudge. That's essentially what scientists face when they try to study the tiny, busy "organelles" inside our cells using standard microscopes. These organelles, specifically the ones called late endosomes and lysosomes (LELs), are like the cell's recycling centers and trash trucks. They are small (some as tiny as 100 nanometers), packed tightly together, and constantly moving. Because they are so close to each other, regular microscopes can't tell them apart; they just see a glowing, messy blob.

For a long time, scientists had to choose between two bad options: look at a few cells in extreme detail (but miss the big picture) or look at thousands of cells quickly (but see nothing but fuzzy blobs). This paper introduces a new "super-power" called HiExM (High-Throughput Expansion Microscopy) that solves this puzzle.

Think of HiExM as a magical, microscopic "swelling gel." Instead of trying to zoom in closer with a better lens, the scientists actually make the cells bigger. They treat the cells with a special gel that expands them about 4 times their original size. It's like blowing up a tiny, crumpled map until it's the size of a billboard; suddenly, the tiny streets and buildings are easy to see and measure. Because the gel stretches everything evenly, the relationships between the parts stay the same, but the details become crystal clear.

Using this "swelling" trick, the team could finally see individual LELs as distinct, separate objects rather than a merged mess. They even trained a computer program (a machine learning model called StarDist) to act like a super-accurate detective, counting and measuring these expanded organelles one by one across huge groups of cells.

Here is what they discovered using this new, high-resolution view:

1. The "Fuzzy" vs. "Sharp" Difference
When the scientists looked at how different antibody drugs (tiny delivery vehicles) entered the cells, standard microscopes said, "Hey, both drugs look like they're in the same place." But HiExM showed a different story. One drug (a bivalent antibody) seemed to rush into the recycling centers faster and stick to them more strongly than the other (a monovalent antibody). The standard view was too blurry to catch this race; the expanded view showed the finish line clearly.

2. The Shape of Things
The team didn't just count the organelles; they studied their shapes and where they lived in the cell. They found that LELs aren't all the same. Some are big and clustered together near the center of the cell, while others are small and lonely near the edges. By using a special type of artificial intelligence (a "variational autoencoder"), they turned the images of these organelles into a unique "fingerprint" code. This allowed them to see subtle differences in the organelles' shapes that human eyes or simple measurements would miss.

3. The Drug Test
To see if their new method could spot real changes, they treated cells with three different chemicals that mess with the recycling centers: apilimod, bafilomycin, and chloroquine.

What the paper suggests: When they looked at the basic measurements (like size and distance from the nucleus), the three chemicals looked very similar.
What the paper found: However, when they used the AI "fingerprint" to look at the detailed shapes, the chemicals looked very different from each other. The AI could tell that chloroquine changed the organelles in a unique way compared to the other two, even though the basic measurements couldn't. This suggests that the AI is catching subtle, hidden changes that standard tools miss.

What the paper rules out
The authors are careful to say that just because the organelles look different in the AI's "fingerprint," it doesn't mean we know exactly why they changed or what specific protein is doing what. They also note that while they could see the differences, they didn't prove that one drug is "better" than the other for curing diseases yet; they just proved that the new method can see the differences that were previously invisible.

The Bottom Line
This paper doesn't claim to have cured a disease or solved a medical mystery. Instead, it suggests that by "swelling" cells and using smart computer vision, scientists can finally stop guessing about what's happening inside the cell's recycling centers. They can now map out the exact shape, size, and location of thousands of these tiny organelles at once. It's like going from looking at a foggy city through a keyhole to having a drone fly over a perfectly clear, expanded map, revealing the hidden patterns of how our cells really work.

Technical Summary: HiExM Enables Scalable Mapping of Organelle Morphology and Spatial Heterogeneity

Problem Statement
Quantitative analysis of subcellular organization faces a fundamental tradeoff between spatial resolution and throughput. Many organelles, such as late endosomes and lysosomes (LELs), exist at spatial scales near or below the diffraction limit of light microscopy (~200 nm) and are often densely packed. Conventional fluorescence microscopy frequently produces merged or ambiguous signals, preventing the accurate segmentation of individual organelles required for object-based feature extraction. While electron microscopy offers nanometer-scale resolution, it lacks compatibility with scalable, multiplexed, protein-specific measurements. Conversely, super-resolution techniques often suffer from limited field of view or acquisition speed, restricting their use to small-scale studies. Consequently, there is a lack of high-throughput methods capable of resolving discrete organelles across large cell populations to capture both intra-cellular and inter-cellular heterogeneity.

Methodology
The authors developed an integrated framework combining High-Throughput Expansion Microscopy (HiExM) with robust computational image analysis and machine learning.

HiExM Pipeline: The study utilized a scalable HiExM protocol on human cerebral microvascular endothelial cells (hCMECs). Samples were fixed using a tandem MeOH/PFA method to preserve LAMP1 staining while enabling expansion. The protocol involved labeling with Acryloyl X-SE, embedding in a hydrogel, digestion with Proteinase K, and physical expansion in water, achieving a ~4-fold linear expansion. This process was performed in multi-well formats compatible with automated high-content imaging (Opera Phenix).
Segmentation and Ground Truth: To resolve individual LELs, the authors manually generated ground-truth segmentations of expanded LAMP1+ organelles. They trained a StarDist model, chosen for its ability to segment star-convex shapes, to automate the segmentation of LELs in 3D image volumes.
Feature Extraction:
- Measured Features: Five quantitative descriptors were extracted for each LEL: volume, distance to the nucleus, distance to the cell membrane, number of proximal LELs, and the number of branches in a minimal graph connecting LEL centroids.
- Learned Features: A 3D $\beta$ -variational autoencoder ( $\beta$ -VAE) was trained on 100,000 segmented LEL image volumes (stained for LAMP1 and LAMP3). The model learned a 256-dimensional latent space encoding morphological features, providing an orthogonal data modality to the measured features.
Analytical Framework: The pipeline integrated these features to analyze:
- Antibody Uptake: Comparing bivalent (AbB) and monovalent (AbM) anti-transferrin receptor antibodies, as well as a fusion protein (AbM-GAA), over a 5-hour time course.
- Perturbation Response: Treating cells with apilimod, bafilomycin, and chloroquine to induce distinct lysosomal perturbations.
- Clustering: Using Leiden clustering on both measured and learned feature spaces to identify distinct organelle states and map their spatial distributions.

Key Results

Resolution and Segmentation: HiExM provided a 4-fold increase in effective resolution, allowing for the disambiguation of vertically stacked organelles and the visualization of sub-organelle structures, such as the distinct lumen of LELs and the membrane localization of LAMP3 versus the lumenal concentration of LAMP3.
Antibody Trafficking Analysis:
- Standard Pearson correlation coefficient (PCC) analysis on whole fields of view failed to distinguish between AbB and AbM localization using conventional microscopy.
- HiExM revealed that AbB colocalized with LAMP1 significantly earlier (90 minutes) than AbM.
- Object-based analysis showed that while AbM and AbB occupied similar LEL volumes, AbB exhibited higher signal intensity within those volumes.
- The AbM-GAA fusion protein showed a distinct distribution pattern (residing in a smaller population of LELs with a "lengthened tail" of intensity) compared to the global shift seen with AbB.
Morphology-Spatial Relationships:
- Clustering of measured features revealed 14 distinct LEL clusters, showing that heterogeneity is an intrinsic feature of the LEL population within single cells.
- The study demonstrated a non-random relationship between spatial positioning and morphology: large-volume LELs were enriched in specific learned-feature clusters and tended to be spatially clustered, while peripheral LELs were smaller and less connected.
Perturbation Sensitivity:
- When comparing lysosomal perturbations (apilimod, bafilomycin, chloroquine), the measured-feature space showed minimal differentiation (<2-fold difference between compounds).
- In contrast, the learned-feature space (VAE latent space) revealed distinct, compound-specific morphological signatures, with several clusters showing >2-fold differential enrichment/depletion.
- The pipeline successfully identified specific cluster combinations that could distinguish the effects of chloroquine from apilimod/bafilomycin, demonstrating that machine learning can detect subtle morphological shifts masked by traditional metrics.

Significance and Claims
The paper establishes HiExM as a generalizable platform for scalable, single-organelle profiling. The authors claim that this integrated approach bridges the gap between high-throughput screening and high-resolution imaging, enabling the quantification of discrete organelles across large cell populations.

Key claims regarding the significance of this work include:

Overcoming Resolution Limits: The ability to resolve individual organelles in densely packed environments allows for accurate segmentation, which is a prerequisite for advanced computational analyses like object-based feature extraction and machine learning.
Enhanced Sensitivity: The pipeline detects biologically meaningful differences in intracellular trafficking and drug response that are invisible to conventional colocalization analysis or bulk measurements.
Unsupervised Discovery: By combining HiExM with unsupervised representation learning, the framework can reveal hidden relationships between organelle morphology and spatial context, as well as distinct phenotypic states induced by different perturbations.
Drug Screening Application: The authors posit that this approach offers a deployable strategy for drug design and screening, particularly for optimizing antibody-based drugs and identifying organelle-level phenotypes that may be missed by conventional viability or narrow-readout screens.

The work concludes that this framework provides a quantitative, organelle-resolved imaging strategy that advances the understanding of cell biology by capturing the coordinated behaviors of organelles within the cellular context.

More like this