Journal Club: Endothelial Heterogeneity, Longitudinal Tissue Profiling, and Retinal Cell Atlas

Introduction

For decades, the cells lining our blood vessels were treated as a relatively uniform monolayer — convenient for textbook diagrams but biologically misleading. The same could be said of how we have analyzed tumors, brains, and other complex tissues: averaging over millions of cells obscures the rare subpopulations and spatial niches that actually drive disease. The three papers in this journal club sit at the sharp end of a methodological revolution that is changing all of that. They span conceptual synthesis (Zhong et al. on endothelial heterogeneity), measurement scale (Schroeder et al. on iSCALE for spatial transcriptomics), and clinical translation (Coullomb et al. on mosna for spatial-omics biomarker discovery). Read together, they trace a coherent pipeline from describing heterogeneous cell populations to mapping them in space and linking their interactions to patient outcomes.

The unifying premise is straightforward but powerful: tissue behavior is an emergent property of cell identity plus neighborhood context, not just average gene expression. Each paper addresses a different layer of that pipeline. Zhong et al. argue that single-cell RNA sequencing (scRNA-seq), paired with organoid models, can finally resolve intra-organ and inter-organ endothelial diversity. iSCALE tackles the spatial-coverage bottleneck that prevents conventional platforms from capturing full tissue architecture. mosna closes the loop by turning spatial neighborhoods and cell–cell interactions into quantitative predictors of immunotherapy response and survival. If you have ever wondered whether single-cell and spatial methods are overhyped, this triad makes a strong case that the field is moving from descriptive atlases toward mechanistic, outcome-linked maps.

In this post, I'll walk through each paper, then discuss what they say collectively about where tissue biology is heading.

Paper 1: Endothelial Heterogeneity Comes of Age

Key Findings

Zhong and colleagues (2025, Cell Regeneration) review how vascular endothelial cells — long thought to be a homogeneous lining — are now recognized as a strikingly diverse population. They organize endothelial heterogeneity into two complementary axes:

• Intra-organ heterogeneity: arterio-venous specification (cells lining arteries versus veins express distinct programs) and tip-stalk cell dynamics during angiogenesis (the leading "tip" cell versus the proliferating "stalk" cells that extend new vessels).
• Inter-organ heterogeneity: organ-specific transcriptomic profiles that allow, for example, brain microvascular endothelial cells to form a tight blood–brain barrier while liver sinusoidal endothelial cells are optimized for filtration and immune surveillance.

scRNA-seq has been the workhorse behind this reframing, enabling identification of rare subpopulations and continuous cell-state gradients that bulk RNA-seq averaged away. The review also highlights emerging evidence for disease-associated endothelial states that diverge from their homeostatic counterparts and may serve as biomarkers or therapeutic targets.

Methodology & Study Design

This is a synthesis paper rather than a primary research article. The authors integrate findings from:

• scRNA-seq studies across mouse and human tissues (brain, heart, liver, lung, kidney, retina, and others)
• Vascular organoids as 3D culture systems that recapitulate organ-specific endothelial behavior and allow gene-function studies in human cells
• Multi-omic integration of transcriptomics with ATAC-seq, proteomics, and emerging spatial approaches

A key conceptual contribution is the proposal that organ-specific vascular organoids — human-derived, 3D, and genetically tractable — offer a path to experimentally dissect inter-organ endothelial programs that are otherwise inaccessible.

Significance

This review crystallizes a shift from descriptive vascular biology to precision, cell-state-specific mechanistic understanding. By treating the endothelium as a collection of specialized subtypes rather than a uniform barrier, researchers can now ask which endothelial programs go awry in cardiovascular disease, cancer angiogenesis, and organ failure — and design interventions accordingly. It also provides a roadmap for combining transcriptomics with epigenetic and spatial readouts to capture endothelial diversity at unprecedented resolution.

Paper 2: iSCALE — Bringing Spatial Transcriptomics to Biobank-Scale Tissues

Key Findings

Schroeder et al. (2025, Nature Methods) introduce iSCALE, a computational framework that reconstructs super-resolution gene expression landscapes across tissue regions far larger than any single capture area of current spatial transcriptomics (ST) platforms. The headline applications include:

• Reconstruction of cellular-level tissue architecture beyond the field-of-view limits of platforms like 10x Visium.
• Automatic annotation of cell types across the reconstructed landscape.
• Detection of lesion-associated cellular features in human multiple sclerosis (MS) brain samples — structures that were invisible to both conventional ST and routine histopathology.

The MS application is particularly striking: iSCALE uncovered cellular neighborhoods associated with MS lesions that expert pathologists had not flagged, suggesting that computational reconstruction can extend the reach of existing data and reveal biologically meaningful microenvironments.

Methodology & Study Design

iSCALE is a machine-learning reconstruction method that:

1. Combines histology images with sparse spatial transcriptomics measurements as inputs.
2. Uses a learning-based model to infer dense, super-resolution gene expression maps across large tissue sections.
3. Performs downstream automatic cell-type annotation and microenvironment mapping.

The authors validate the approach through:

• Benchmarking experiments against gold-standard datasets.
• Immunohistochemistry (IHC) staining of matching tissue sections.
• Manual annotation by pathologists to confirm that reconstructed structures are biologically plausible.

The framework is technology-agnostic at the input level — it can leverage any spatial transcriptomics platform that provides gene expression plus matched histology.

Significance

Most ST platforms are constrained by high costs, long turnaround times, low resolution, limited gene coverage, and small capture areas (often only a few square millimeters). iSCALE addresses the scale and cost barriers by computationally extending sparse measurements, making spatial biology more accessible for whole-section, biobank-scale, and clinical workflows. For diseases like MS, where lesions are spatially heterogeneous and span large brain regions, this is a critical step toward mapping pathology at the scale it actually occurs.

Paper 3: mosna — Turning Spatial Organization into Clinical Predictors

Key Findings

Coullomb et al. (2026, Molecular & Cellular Proteomics) present mosna, a Python package for analyzing spatial omics data in integration with clinical or biological metadata. Key features and results include:

• Compatibility with diverse platforms: sub-cellular resolved transcriptomics (MERFISH, seqFISH, Xenium), spatial proteomics (CODEX, MIBI-TOF, low-plex immunofluorescence), and spot-based ST (10x Visium, Slide-seq, Stereo-seq).
• Multi-step analysis pipeline: from network construction (via the companion tool tysserand) through increasingly complex spatial features, with integration into Squidpy for downstream analysis.
• Clinical integration: handles binary conditions (responder vs. non-responder) and survival data.
• Demonstration results: on real datasets, mosna identified interaction patterns and cellular neighborhoods predictive of immunotherapy response and patient survival in cancer — features that simple cell-type abundance analyses missed.

Methodology & Study Design

mosna is a modular analysis workflow with the following structure:

1. Network construction: builds accurate spatial graphs of cell–cell proximity using tysserand.
2. Feature extraction: computes compositional features, spatial patterns, and cellular niches, leveraging mosna-algorithms or external tools.
3. Clinical integration: tests each feature against binary outcomes or survival using descriptive statistics, visualization, and machine-learning models.
4. Predictive modeling: trains classifiers, identifies the most predictive variables, and produces interpretable outputs.

The authors validate the approach on:

• Two spatially resolved proteomic datasets.
• One spatial transcriptomics dataset with immunotherapy response labels.
• A manually annotated ST dataset for benchmarking the niche-discovery method.

The code is open source (https://github.com/AlexCoul/mosna), with Jupyter notebook tutorials and full documentation.

Significance

Immunotherapy response is notoriously variable, and bulk or even single-cell profiling often fails to explain why some patients respond and others do not. mosna operationalizes the idea that tissue organization itself contains prognostic information — that how cells are arranged matters as much as which cells are present. By providing a unified, platform-agnostic workflow, mosna lowers the barrier for translational researchers to move from descriptive spatial maps to quantitative, outcome-linked biomarkers.

Synthesis & Discussion

A Three-Layer Pipeline

Read together, these papers describe three sequential layers of a modern tissue-biology pipeline:

1. Biology (Paper 1): Defining what cell states exist — here, the diverse endothelial subtypes across organs and vessel types.
2. Measurement (Paper 2): Capturing those states in their spatial context at tissue scale — iSCALE reconstructs whole-tissue architecture from sparse measurements.
3. Interpretation (Paper 3): Linking spatial organization to clinical outcomes — mosna extracts interaction features that predict immunotherapy response and survival.

The pipeline is enabled by a common technological foundation: high-throughput single-cell and spatial omics, supported by machine-learning-based reconstruction and statistical inference.

Points of Convergence

• All three treat cell identity + neighborhood context as the unit of biological meaning, rather than bulk expression averages.
• All three push for higher resolution than bulk methods can offer — rare subpopulations, spatial niches, and interaction patterns carry information that is invisible at the tissue level.
• All three are explicitly translational in orientation: endothelial subtypes as therapeutic targets (Paper 1), lesion detection in MS (Paper 2), and immunotherapy biomarkers (Paper 3).

Points of Divergence

• Cell type of focus: Paper 1 is endothelial-centric; Paper 3 is tumor-microenvironment-centric; Paper 2 is tissue-agnostic.
• Methodology: Paper 1 leans on wet-lab models (organoids); Paper 2 is purely computational reconstruction; Paper 3 is a software/workflow contribution.
• Outcome variable: Paper 1 emphasizes mechanistic understanding; Paper 2 emphasizes technical scalability; Paper 3 emphasizes clinical prediction.

Open Questions

• Generalizability: How well do interaction signatures identified by mosna transfer across tumor types, platforms, and patient cohorts?
• Causality: Are the spatial interaction patterns in immunotherapy responders drivers of response, or downstream markers? Causal experiments — perhaps in vascular or tumor organoids à la Paper 1 — are needed.
• Integration: Can iSCALE-style reconstruction be combined with organoid systems to validate that spatial niches observed in vivo are functionally meaningful?
• Endothelial-immune crosstalk: Given that endothelial cells shape immune infiltration in tumors, can the endothelial subtypes from Paper 1 be linked to the immunotherapy-prediction features from Paper 3?

For Your Lab Meeting

1. Methodology trade-offs: Zhong et al. emphasize scRNA-seq plus organoids; iSCALE emphasizes histology-driven reconstruction. If you had to choose one platform to profile the tumor vasculature for an immunotherapy-resistance study, which would you pick and why? What would you lose by not having the other?
2. Validation strategies: iSCALE was validated against IHC and pathologist annotations. Is this enough for clinical adoption, or do we need orthogonal single-molecule FISH validation? Where should the field draw the line?
3. Interaction feature interpretation: mosna identifies predictive cellular neighborhoods, but are these biologically interpretable as mechanisms, or are they mostly black-box features for ML models? How would you design a follow-up study to distinguish the two?
4. Cross-tissue generalization: The MS application of iSCALE and the cancer application of mosna both show that spatial features carry disease-relevant information. Would you expect the same feature-engineering approaches to work across both tissues, or will each disease require bespoke pipelines?
5. Future directions: If you could design the "iSCALE + mosna + organoid" experiment of 2030, what question would you ask? Which of the three papers would you extend, and how?

Key Terms

• scRNA-seq (single-cell RNA sequencing): A technique that profiles gene expression in thousands of individual cells, enabling discovery of rare subpopulations and continuous cell states that bulk RNA-seq averages together.
• Spatial transcriptomics (ST): Methods that measure gene expression while preserving the spatial location of each measurement within a tissue section — e.g., 10x Visium, MERFISH, Stereo-seq.
• Vascular organoid: A 3D, self-organizing tissue culture derived from stem cells that recapitulates aspects of vascular development and can be used to study endothelial cell behavior in a human, experimentally tractable system.
• Tip-stalk cell dynamics: During angiogenesis, the leading "tip" cell guides new vessel sprouting while proliferating "stalk" cells elongate the sprout — a paradigmatic example of intra-organ endothelial heterogeneity.
• Cellular niche / neighborhood: A spatially defined group of interacting cell types, often inferred from spatial omics data, whose composition and arrangement carry functional or clinical meaning beyond what any single cell type provides.

---

References

1. Jing Zhong, Rong-rong Gao, Xin Zhang, Jia-xin Yang, Yang Liu, Jinjin Ma, et al. (2025). Dissecting endothelial cell heterogeneity with new tools. Cell Regeneration. https://doi.org/10.1186/s13619-025-00223-3

2. Amelia Schroeder, Melanie L. Loth, Chunyu Luo, Sicong Yao, Hanying Yan, Daiwei Zhang, et al. (2025). Scaling up spatial transcriptomics for large-sized tissues: uncovering cellular-level tissue architecture beyond conventional platforms with iSCALE. Nature Methods. https://doi.org/10.1038/s41592-025-02770-8

3. Coullomb A, Foulon C, van Haastrecht B, Monsarrat P, & Pancaldi V (2026). mosna Reveals Different Types of Cellular Interactions Predictive of Response to Immunotherapies and Survival in Cancer.. Molecular & cellular proteomics : MCP. https://doi.org/10.1016/j.mcpro.2026.101536