Journal Club: From Cell Atlases to Causal Maps — Integrating Identity, Space, and Regulation

Introduction

For decades, biologists chased genes one mutant at a time. Forward genetic screens in plants like Arabidopsis and maize gave us the low-hanging fruit — large-effect developmental regulators such as GLABRA1, SHOOTMERISTEMLESS, and KNOTTED1. That era is winding down. Most "obvious" loci have been mapped, and the remaining thousands of plant genes are functionally opaque, often with subtle, context-dependent, or cell-type-restricted roles. At the same time, single-cell RNA sequencing (scRNA-seq) has matured into a production-scale technology, but most published plant datasets remain fragmented across species, organs, and protocols — making cross-study comparisons painful and cross-species insights rare.

The three papers in this journal club tackle complementary facets of the same problem: how do we move from listing which genes are expressed in which cells to understanding which cells, in which places, are controlled by which regulatory variants? Xue et al. (2025, Cell) deliver a cross-species cell-type reference layer for vascular plants; Moskal et al. (2025, IJMS) provide a conceptual and methodological synthesis arguing that spatial context is non-negotiable in plant tissues; and Dorans et al. (2025, Nature Genetics) introduce pgBoost, a learned framework for linking noncoding variants to their target genes using single-cell multiome data. Read together, they trace the arc from descriptive atlases to mechanistic, spatially anchored, regulator-aware gene discovery.

Paper 1: A Cross-Species Plant Cell Atlas for Accelerated Gene Discovery

Key Findings

• A unified single-cell atlas of shoot apices across six vascular plant species spanning angiosperms, gymnosperms, and ferns was generated, enabling direct evolutionary comparison of cell-type identities.
• The authors identified a core set of "cell-type foundational genes" linked to conserved tissues such as epidermis, xylem, and phloem — genes whose expression is so tightly cell-type-restricted across hundreds of millions of years of evolution that they almost certainly play structural or regulatory roles.
• They discovered a previously unrecognized clade of X8-domain proteins and JULGI-LIKE genes as regulators of phloem development, opening a new functional gene family.
• Using the atlas, they identified companion-cell-like cells in ferns and gymnosperms that conventional histology cannot distinguish — cryptic cell types only visible at single-cell resolution.

Methodology & Study Design

The team performed scRNA-seq on shoot apices from six vascular plant species, computationally integrated the datasets to define cross-species cell-type clusters, and used the resulting unified embeddings to score genes by their cell-type specificity across the evolutionary range. They then released an automated cell-type annotation tool for vascular plant scRNA-seq data — a practical pipeline the community can apply directly to new datasets rather than redoing clustering from scratch.

Significance

This work is a major step toward realizing the long-envisioned Plant Cell Atlas initiative. By linking gene discovery to cell-type identity rather than to forward-genetic phenotypes, the authors effectively provide a shortcut around forward-genetic saturation. The identification of a clade of X8/JULGI-LIKE phloem regulators — genes that almost certainly would not have surfaced in classical screens — is a concrete proof-of-concept. For the broader community, the annotation tool democratizes cross-species interpretation: labs working on non-model crops can now drop their scRNA-seq data into a curated vascular-plant reference frame.

Paper 2: Why "Where" Matters as Much as "How Much" in Plant Transcriptomics

Key Findings

• A comprehensive review and roadmap for integrating single-cell and spatial transcriptomics in plants, emphasizing that spatial context is as decisive as expression magnitude for functional interpretation in layered plant tissues.
• Side-by-side comparison of scRNA-seq vs. snRNA-seq in plants, including how protoplast dissociation artifacts distort cell-type proportions and stress signatures — a notorious plant-specific problem given the cell wall.
• Practical workflows for integrating single-nucleus ATAC-seq (snATAC-seq) and Multiome assays with transcriptomics to add chromatin information.
• A discussion of spatial platforms (Visium HD, Stereo-seq, bead arrays) and targeted imaging methods (smFISH, HCR-FISH, MERFISH), including plant-specific adaptations such as cell-wall permeabilization.
• Recent case studies in Arabidopsis, soybean, and maize that show how spatial niches reveal developmental trajectories and stress responses invisible to dissociated single-cell data alone.

Methodology & Study Design

This is a methodological synthesis rather than a primary research paper. The authors review platform selection, sample preparation (with particular attention to plant recalcitrance to protoplasting), computational workflows for label transfer, deconvolution, spatial embedding, and neighborhood-aware cell–cell communication, and integration with epigenomic assays. They also lay out a roadmap from descriptive atlases to gene-regulatory network inference and low-pleiotropy trait engineering in crops.

Significance

Paper 1 builds a cell-type atlas; Paper 2 argues why that atlas is incomplete without space. In plant tissues, where epidermis, cortex, vasculature, and meristem occupy precise anatomical positions, two cells with identical transcriptomes can have completely different neighbors and developmental fates. The review thus functions as a bridge paper, situating plant scRNA-seq atlases like Xue et al.'s within the broader shift toward spatially resolved, multi-modal plant biology.

Paper 3: pgBoost — Linking Variants to Genes Using Single-Cell Multiome Data

Key Findings

• Existing methods for linking noncoding regulatory elements to target genes from single-cell multiome data (paired RNA-seq + ATAC-seq) show low concordance with each other and rarely incorporate genomic distance as an explicit feature.
• pgBoost, a gradient boosting framework that non-linearly combines multiple existing linking strategies (including genomic distance) using eQTL data as training labels, outperforms individual methods on held-out eQTL, activity-by-contact (ABC), CRISPR, and GWAS-derived evaluation sets.
• Restricting pgBoost to features from a focal cell type boosts power to find links relevant to that specific lineage.
• The authors highlight GWAS examples where pgBoost linked fine-mapped variants to experimentally validated or biologically plausible target genes that competing methods missed.

Methodology & Study Design

The team trained a gradient boosting model (pgBoost) on eQTL-derived SNP–gene links, using as input features the scores from existing linking methods (e.g., correlation-based, enhancer-promoter proximity, ABC-like metrics) plus raw genomic distance. The model learns weights that reflect each method's reliability, allowing it to outperform any single strategy. A key methodological choice is the ability to focalize on one cell type at a time, exploiting the cell-type resolution that only single-cell multiome data provides.

Significance

This paper tackles one of the central problems in modern functional genomics: which gene does my GWAS hit actually regulate? Most disease- and trait-associated variants lie in noncoding regions and regulate genes far away, often in a cell-type-specific manner. Bulk approaches average over cell types and miss these contexts. By learning a distance-aware, cell-type-aware, multi-method integrator, pgBoost turns single-cell multiome data into a more actionable map of regulatory wiring — exactly the kind of tool needed to move from descriptive chromatin atlases to causal variant interpretation.

Synthesis & Discussion

A Three-Layer Framework for Modern Functional Genomics

Read together, the three papers articulate a coherent three-layer model:

Layer	Paper	Question Answered
Cell-type identity	Xue et al. (2025)	Which genes define which cell types across species?
Spatial context	Moskal et al. (2025)	Where are those cell types located within the tissue, and does location matter?
Regulatory wiring	Dorans et al. (2025)	Which noncoding variants control which genes, in which cell types?

Agreements Across the Three Papers

All three treat cell-type specificity as the central organizing principle for gene interpretation. All three implicitly or explicitly argue that single-modality data are insufficient — whether the missing modality is space (Paper 2), chromatin state (Paper 3), or evolutionary breadth (Paper 1). And all three reflect a field-wide pivot from cataloging toward predicting: predicting gene function, predicting spatial niche, predicting variant target.

Tensions and Open Questions

• Conservation vs. context: Paper 1 emphasizes evolutionary conservation of cell-type genes, while Paper 2 emphasizes that spatial context can radically change what a gene does. How do foundational genes identified across six species behave under stress, in different organs, or at different developmental stages?
• Spatial resolution trade-offs: Paper 2 leaves open what spatial resolution is enough. Subcellular resolution (MERFISH, smFISH) reveals gradients and polarity; tissue-level (Visium HD) covers more cells but blurs boundaries. Which plant biology questions demand which?
• Validation of regulatory links: Paper 3's pgBoost predictions need experimental validation, especially for long-range or cell-state-specific links. The plant field in particular lacks scalable perturbation platforms (CRISPRi/CRISPRa in specific cell types) compared to human cell-line systems.
• Generalization across species: Paper 1's foundational genes work across six vascular plants, but Paper 3's pgBoost is trained on human multiome data. Can plant multiome atlases enable a plant-equivalent of pgBoost? The data infrastructure is not yet there.

What This Tells Us About the Field

The era of single-species, single-modality, descriptive cell catalogs is closing. The new frontier is integrative, mechanistic, and spatially aware — anchored in cross-species references where possible, validated by perturbation where possible, and increasingly powered by machine learning that can combine disparate evidence streams. The Plant Cell Atlas initiative, the rapid spread of spatial platforms in plant labs, and the maturation of multiome-based regulatory inference in human genomics are all manifestations of the same underlying transition.

For Your Lab Meeting

1. Cross-species generalization: Xue et al. identified "foundational genes" using six vascular plant species. How many species do we need before a gene's cell-type specificity can be called "foundational"? Is six enough, or is this an arbitrary threshold? What would you add (moss? Marchantia? algae?) to test the limits?
2. Spatial resolution vs. throughput: Paper 2 surveys Visium HD, Stereo-seq, and targeted FISH methods. If your lab had budget for one platform to study, say, Arabidopsis root development, which would you choose and why? What would you lose either way?
3. Distance as a feature: Dorans et al. find that explicitly including genomic distance in pgBoost improves variant–gene linking. In plants, enhancers can act over very long ranges (sometimes >50 kb). Does this mean pgBoost-style frameworks will need plant-specific retraining, or is distance a sufficiently universal prior?
4. Protoplast bias: Paper 2 highlights that scRNA-seq in plants systematically under-represents certain cell types due to cell-wall digestion resistance. For your favorite plant tissue, which cell types do you suspect are missing from published scRNA-seq datasets, and how would you test that?
5. From atlases to interventions: All three papers gesture toward translational impact — crop engineering, GWAS interpretation, low-pleiotropy trait design. What is the single biggest bottleneck preventing plant biology from making the same leap that human single-cell genomics has made toward clinical application?

Key Terms

• scRNA-seq / snRNA-seq: Single-cell and single-nucleus RNA sequencing. snRNA-seq isolates nuclei rather than whole cells, avoiding the need to digest cell walls — particularly useful in plants where protoplasting introduces stress and cell-type bias.
• Multiome / paired RNA + ATAC: A single-cell assay that simultaneously profiles gene expression (RNA-seq) and chromatin accessibility (ATAC-seq) from the same cell, allowing direct linkage of regulatory element activity to transcription in defined cell types.
• eQTL (expression quantitative trait locus): A genomic variant associated with variation in mRNA expression levels of a particular gene across individuals. Used as training labels for variant–gene linking methods.
• Activity-by-Contact (ABC) model: A computational framework that predicts enhancer–promoter links by combining enhancer activity (e.g., from ATAC-seq or H3K27ac), promoter activity, and the 3D contact frequency between them.
• Foundational gene: A term introduced (or popularized) by Xue et al. for a gene whose expression is so tightly associated with a specific cell type across multiple species that it likely plays a core, evolutionarily conserved role in that cell type's identity or function.

---

References

1. Hao-Chen Xue, Zhou-Geng Xu, Yu-jie Liu, Long Wang, Xin Ming, Zechi Wu, et al. (2025). A unified cell atlas of vascular plants reveals cell-type foundational genes and accelerates gene discovery.. Cell. https://doi.org/10.1016/j.cell.2025.07.036

2. Kinga Moskal, Marta Puchta-Jasińska, Paulina Bolc, Adrian Motor, Rafał Frankowski, Aleksandra Pietrusińska-Radzio, et al. (2025). Why “Where” Matters as Much as “How Much”: Single-Cell and Spatial Transcriptomics in Plants. International Journal of Molecular Sciences. https://doi.org/10.3390/ijms262411819

3. E. Dorans, K. Jagadeesh, K. Dey, & A. Price (2025). Linking regulatory variants to target genes by integrating single-cell multiome methods and genomic distance. Nature Genetics. https://doi.org/10.1038/s41588-025-02220-3