From Cell Atlases to Spatial Mechanisms: Eight 2025 Omics Studies

For research use only. Not intended for clinical diagnosis, treatment, or individual health assessment.
Single-cell and spatial omics studies are no longer defined only by how many cells they profile or how detailed their tissue maps appear. The more useful question is what a study can infer because of its sampling design. A three-dimensional brain atlas can test evolutionary boundaries. A multi-organ atlas can separate shared aging programs from organ-specific responses. A spatial time series can connect a transient stem-like population to a developmental trajectory. A multimodal disease study can ask whether transcriptional and metabolic changes occupy the same tissue compartments.
Eight studies first released or made publicly available in 2025 illustrate this transition from cataloging cells to testing organized hypotheses. Their current citation records span 2025 and 2026 because some papers appeared online in 2025 and were assigned to 2026 print volumes; the amphioxus brain study also progressed from a 2025 preprint to a peer-reviewed 2026 article. References below use the most current verified publication records.
This is not a list of the year's largest atlases, a platform ranking, or eight equal-length paper summaries. Each study is used to answer one reusable design question: What spatial or biological scale was chosen, what integration problem had to be solved, and what additional evidence moved the result beyond description? Researchers looking for public datasets rather than design lessons can consult 26 single-cell and spatial omics atlases worth bookmarking.
Eight Studies at a Glance
| Study System | Main Modalities | Design Priority | Transferable Lesson |
|---|---|---|---|
| Adult amphioxus brain | snRNA-seq, targeted spatial transcriptomics, serial-section 3D reconstruction | Preserve three-dimensional anatomical axes | An atlas becomes comparative evidence when cell types are anchored to homologous boundaries |
| Adult Aedes aegypti | Multi-tissue snRNA-seq, sex and feeding-state comparisons | Cover organs, sexes, and relevant states | Breadth is useful only when metadata preserves the contrasts that motivated sampling |
| Mouse endocrine aging | scRNA-seq and snRNA-seq across endocrine organs, functional follow-up | Separate shared aging signals from organ-specific programs | Multi-organ harmonization should retain organ identity and biological replication |
| Moso bamboo meristem | snRNA-seq, spatial transcriptomics, ATAC-seq, imaging, functional validation | Link stage, tissue zone, and cell trajectory | Developmental time and anatomical position should be modeled together |
| Mouse diabetic kidney disease model | High-resolution spatial transcriptomics and full-length single-cell RNA sequencing | Localize a preclinical perturbation response | Treatment-associated expression should be tied to renal compartments and validated independently |
| Ovarian endometrioma | scRNA-seq, regional spatial transcriptomics, MALDI mass spectrometry imaging | Align cell states with metabolic regions | Modalities should meet at a biological coordinate, not merely a shared sample label |
| SSpMosaic | Cross-batch, cross-modal, and cross-species computational integration | Preserve interpretable programs during alignment | Integration should expose the features supporting a match and test what biology remains |
| FastCCC | Scalable ligand-receptor analysis with a reference panel | Increase computational scale without losing statistical context | Faster inference expands analysis, but does not turn predicted communication into observed signaling |
What These Eight Studies Share
The eight projects span an invertebrate chordate, an insect, mouse organs, bamboo, human tissue, and computational benchmarks. Their biological questions differ, yet four design moves recur.
First, each project defines a meaningful unit of organization. That unit may be a brain compartment, one of 19 mosquito tissues, an endocrine organ, a meristem zone, a kidney compartment, a lesion region, an interpretable gene program, or a cell-pair communication score. Without this unit, additional cells or spatial pixels increase volume rather than explanatory power.
Second, the studies treat metadata as analytical structure. Sex, feeding state, age, organ, developmental stage, perturbation, tissue region, species, and assay modality are not optional labels added after clustering. They define the comparisons. A batch-correction method that erases one of these variables can produce a visually integrated embedding while damaging the original question.
Third, the strongest findings use orthogonal evidence. Spatial localization supports an anatomical claim. Imaging supports cell identity or morphology. Chromatin accessibility can reinforce a regulatory interpretation. Functional experiments can test a candidate regulator. Cross-dataset validation can assess whether a computational program transfers. Each layer should contribute evidence that is not simply a reformatted version of the same RNA counts.
Fourth, the studies expose the importance of claim boundaries. An atlas describes what was sampled, not every cell in a species. A trajectory is an inferred ordering, not direct lineage observation. Spatial proximity makes an interaction plausible, not causal. A computational benchmark shows performance in selected scenarios, not universal superiority. These distinctions are central to credible single-cell and spatial omics writing.
Figure 2. The eight studies form a research path from atlas construction at several biological scales to spatial mechanism testing and interpretable computational analysis.
Four Atlas Studies, Four Scales
The first four studies all build atlases, but "atlas" means something different in each. Their value comes from matching the sampling frame to the scientific comparison rather than maximizing one headline metric.
A Three-Dimensional Evolutionary Atlas
Lin and colleagues combined single-nucleus RNA sequencing with spatial transcriptomics and serial-section reconstruction to map the adult amphioxus central nervous system [1]. The peer-reviewed study described a tripartite organization along the anterior-posterior axis, examined ZLI-like and midbrain-hindbrain-boundary-like regions, and compared amphioxus organization with vertebrate brain features.
The reusable lesson is that a comparative atlas needs a coordinate system shared across levels of evidence. Cell clusters alone cannot establish whether a population belongs to a proposed brain compartment. Spatial gene domains, anatomical position, serial-section registration, and cross-species comparisons must converge. A three-dimensional reconstruction is particularly useful when a boundary changes across adjacent sections or when a small structure would be ambiguous in one plane.
Cross-species claims also require careful homology language. Similar marker combinations or positions can support correspondence, while absence of expected markers can argue against a proposed region. Neither result should be reduced to one-to-one cell labels without considering gene orthology, developmental timing, and lineage-specific innovation. For project planning, define which structures are to be compared before selecting sections and probe panels; otherwise, the final dataset may be rich but anatomically incomplete.
A Multi-Organ and Sex-Aware Atlas
Goldman and colleagues profiled more than 367,000 nuclei from 19 dissected tissues of adult female and male Aedes aegypti mosquitoes [2]. The study examined broad cell diversity, sensory receptor expression, sex-specific features, and brain responses following blood feeding. One notable design feature was the deliberate inclusion of both sexes and state-specific time points rather than treating a whole-animal atlas as a single baseline.
This illustrates the difference between coverage and comparison. Sampling many organs creates a reference, but biological insight depends on preserving tissue, sex, state, and time in the analysis. A receptor detected in several sensory tissues and a sex-associated program restricted to one cell population answer different questions. Pooling too early could obscure both.
For new atlas projects, construct a specimen-to-question matrix before library preparation. List each tissue or compartment, relevant biological states, replicate structure, expected rare populations, and required validation. This reveals whether the design can distinguish tissue specificity from sex, treatment, or processing effects. It also prevents one deeply sampled organ from dominating a cross-organ embedding. Research teams planning cell inventories across complex samples can use a single-cell sequencing service to align isolation strategy, nuclei or whole-cell input, multiplexing, sequencing depth, and analysis outputs with the intended comparisons.
A Multi-Organ Aging Atlas
Wei and colleagues generated single-cell or single-nucleus profiles across eight mouse endocrine organs in young and aged animals [3]. The study reported organ-specific aging programs alongside shared features such as increased immune infiltration and stress-related changes in functional endocrine cells. It also investigated a GZMK-associated MHC-I and unfolded-protein-response axis with additional experiments.
The design lesson is that a system-level aging atlas needs two analytical views. The first is within-organ: which cell types, proportions, and pathways change with age in each tissue? The second is across organs: which responses recur after accounting for baseline differences among pituitary, thyroid, adrenal, pancreatic, gonadal, and other endocrine compartments? A global old-versus-young comparison without organ-aware modeling can mistake normal tissue identity for an aging effect.
Shared programs also require donor- or animal-level evidence. Thousands of cells from one organ do not replace independent animals. Pseudobulk summaries, mixed models, or other replicate-aware approaches help separate consistent age effects from cell-level sampling noise. Functional follow-up is especially important when a shared axis is proposed as a driver: expression association can nominate the axis, while perturbation tests whether it can influence the observed phenotype in the model.
A Spatiotemporal Plant Atlas
Qin and colleagues integrated a chromosome-scale bamboo genome, anatomical imaging, snRNA-seq, spatial transcriptomics, and other assays across three stages of intercalary meristem activity [4]. They identified meristem-associated populations, localized short-columnar cells adjacent to elongating ground-tissue parenchyma, reconstructed transitions from proliferative to elongating states, and tested a WOX2 homolog in functional experiments.
This project demonstrates why developmental atlases should not force time and space into separate stories. Pseudotime can order transcriptional states, but it does not reveal where those states occur. Spatial measurements can localize populations, but a static map cannot by itself determine developmental progression. Sampling initial division, rapid division, and rapid elongation stages creates an external temporal axis against which the inferred trajectory can be checked.
The same principle applies beyond plants. Regeneration, embryogenesis, wound repair, and tissue maturation often involve transient states that occupy specific zones. A defensible design samples both the stages and the anatomical structures needed to test progression. Marker localization and functional assays then provide evidence that a candidate population is not merely a computational midpoint. Readers interested in other spatially resolved applications can compare these design choices with the cases in Stereo-seq in action, while keeping platform capability separate from biological inference.
Mechanistic Studies in Tissue
The next two studies begin with disease-associated tissue and use spatial measurements to narrow mechanistic hypotheses. Their value lies in connecting perturbation or disease context to specific cell types and anatomical regions, not in treating a spatial map as proof of efficacy or causality.
Localizing a Preclinical Response
Wu and colleagues combined high-resolution spatial transcriptomics with full-length single-cell RNA sequencing in a mouse model of diabetic kidney disease to study total flavones from Abelmoschus manihot and an irbesartan comparison [5]. The paper reported kidney cell and spatial maps, treatment-associated changes in regulators and pathways, and candidate receptor or regulatory mechanisms.
The reusable question is: where does a perturbation-associated molecular change occur? Kidney tissue contains glomerular, tubular, vascular, interstitial, and immune compartments with different baseline programs. A bulk expression change may reflect altered cell composition, a state change within one compartment, or both. Spatial localization can separate these explanations when region annotation and cell identity are sufficiently supported.
This remains a preclinical mouse study. Similarity between treatment groups in selected model outcomes does not establish clinical equivalence, and molecular docking or co-localization does not establish target engagement. A robust perturbation design should distinguish at least four evidence levels:
- Physiological or histological change in the experimental model
- Reproducible cell-type or region-specific expression change
- Spatial alignment between the candidate pathway and affected tissue compartment
- Independent perturbation or molecular evidence for the proposed target relationship
Spatial omics is most informative at the middle two levels. It can identify where to focus validation and which competing explanations are less consistent with the tissue map. It should not be asked to carry the entire causal argument.
Linking Transcription and Metabolism
Qi and colleagues integrated scRNA-seq, regional spatial transcriptomics using a whole-transcriptome digital spatial profiling workflow, and MALDI mass spectrometry imaging in ovarian endometrioma and adjacent ovarian cortex samples [6]. The study described epithelial and perivascular transcriptional programs alongside spatial metabolic heterogeneity.
This case illustrates a frequent multi-omics challenge: the modalities do not observe the same unit. Single-cell RNA data provide cell-level expression without native coordinates. Regional spatial transcriptomics measures selected tissue areas that may contain several cell types. MALDI imaging measures mass features on another spatial grid, and many features may not have immediate molecular identification. Integration therefore requires a shared biological coordinate, such as an epithelial region, stromal compartment, vascular neighborhood, or lesion boundary.
Co-registration should be explicit. Researchers need to document which sections are adjacent or identical, how regions are aligned, how segmentation differs among modalities, and what spatial uncertainty remains. A metabolite enriched in an epithelial region is not automatically produced by epithelial cells, and a transcript-metabolite correlation does not establish a biochemical reaction. The most useful output is a prioritized, anatomically anchored hypothesis for targeted molecular identification or validation.
This study also shows why discovery features should not be presented as validated diagnostic markers. Candidate genes, pathways, or metabolites require independent cohorts, predefined assays, and performance evaluation before any clinical interpretation. In research-use projects, integration should focus on biological compartments, mechanism hypotheses, and validation plans. Our guide to integrating scRNA-seq with spatial transcriptomics explains how reference mapping and deconvolution fit into that process.
Figure 3. Cell states become more informative when they are localized to tissue compartments and supported by independent molecular or functional evidence; association remains distinct from causality.
Interpretable Integration and Communication
The final two studies address computational bottlenecks created by larger and more diverse datasets. One asks how to align batches, modalities, species, and spatial resolutions through interpretable gene programs. The other asks how to scale cell-cell communication analysis without costly permutations. Both expand analytical reach, but both still depend on input quality and biological validation.
Interpretable Programs Across Modalities
Zhang and colleagues introduced SSpMosaic, a framework that uses interpretable gene programs and higher-order metaprograms as anchors across single-cell and spatial omics datasets [7]. The study evaluated integration and annotation across batches, modalities, species, and spatial resolutions, and applied the framework to several tissue contexts.
Gene-program anchors can be easier to interpret than a latent dimension with no direct feature definition. They may also tolerate missing individual markers when a coordinated program is preserved. However, interpretability is not guaranteed merely because a factor has a gene list. Each metaprogram still needs biological annotation, stability testing, and examination for confounding by cell cycle, stress, mitochondrial content, or technical effects.
Before selecting an integration framework, specify what must remain after alignment:
- Cell-type distinctions that are central to the study
- Continuous states or transitions that should not be over-corrected
- Condition-associated programs that represent biology rather than batch
- Species-specific features that should not be forced into false equivalence
- Spatial domains and boundaries that must remain anatomically coherent
Then test whether those features survive plausible parameter choices and alternative integrations. Benchmark scores summarize selected aspects of mixing and conservation; they do not prove that the aligned representation answers the biological question. Our spatial multi-omics algorithm guide provides additional context for matching algorithm families to analysis goals.
Scaling Communication Analysis
Hou, Ma, and Zhou introduced FastCCC, a permutation-free framework for scalable, robust, and reference-based cell-cell communication analysis [8]. The method computes analytical significance rather than relying on extensive permutations, offers modular scoring choices, and includes a large human communication reference panel. The paper applied the framework to contexts including immune responses and T-cell development.
Computational scale matters because modern atlases contain many cell types, conditions, and specimens. An analysis that is practical for one dataset may become prohibitive when repeated across donors, contrasts, and sensitivity settings. Faster inference can make replicate-aware and parameter-sensitivity analyses more feasible rather than limiting a project to one pooled run.
Speed, however, does not solve the biological identification problem. Ligand and receptor transcripts may be expressed without productive signaling. Multi-subunit complexes can be incompletely measured. Cell abundance and library size can affect scores. Dissociated single-cell data do not retain the original distance between sender and receiver cells. A reference panel can stabilize analysis in some settings, but mismatch in tissue, condition, cell labels, or assay can transfer inappropriate expectations.
A communication result should therefore be described as a predicted or candidate interaction unless additional evidence supports signaling. Useful follow-up includes spatial co-localization, receptor and ligand protein detection, downstream response programs in the proposed receiver, reproducibility across specimens, and perturbation. CD Genomics' research-use single-cell omics bioinformatics and data mining service can incorporate communication analysis within a broader workflow that documents cell annotation, contrasts, statistical units, and validation boundaries.
Figure 4. SSpMosaic and FastCCC address different scale problems: aligning heterogeneous modalities through interpretable programs and expanding communication analysis across large cell atlases.
From Maps to Testable Hypotheses
Across the eight studies, the transition from descriptive map to testable hypothesis follows a recurring evidence chain. Each link should be visible in the analysis plan and final report.
1. Define the biological contrast. State whether the comparison concerns anatomy, sex, feeding state, age, developmental stage, perturbation, lesion compartment, modality, or species. If several contrasts are present, identify the primary one and prevent it from being confounded with batch.
2. Build a reproducible cell or state definition. Use multiple markers, reference information, and sample-aware checks. A cluster name should not rest on one high-expression gene. When a state is continuous, report scores or trajectories rather than forcing a categorical label.
3. Anchor the state to tissue or system context. This may mean a three-dimensional brain boundary, an organ, a meristem zone, a kidney compartment, a lesion region, or a spatial neighborhood. The coordinate must be independently interpretable and reproducible across specimens.
4. Test a directional prediction. A useful mechanism hypothesis predicts what should change when a regulator, cell population, pathway, developmental stage, or local environment is altered. A spatial association that cannot generate a distinguishing prediction remains descriptive.
5. Add independent validation. Imaging, protein assays, chromatin measurements, targeted spatial panels, perturbation, longitudinal sampling, or another appropriate experiment can test different parts of the hypothesis. Reusing the same RNA markers in another visualization is a consistency check, not orthogonal validation.
This chain also protects against two common overstatements. Pseudotime is not lineage tracing, and ligand-receptor inference is not observed communication. Both can organize evidence and nominate experiments. Neither establishes directionality on its own.
Reusable Study-Design Lessons
The studies suggest a compact set of decisions that can be reused across organisms and research areas.
| Design Decision | Question to Ask Before Profiling | What Can Go Wrong | Evidence to Reserve |
|---|---|---|---|
| Biological scale | Is the target a cell, state, organ, boundary, three-dimensional structure, or whole system? | The assay has high resolution but misses the relevant structure | Histology, anatomy annotation, or serial sections |
| Sampling matrix | Which organs, stages, sexes, conditions, regions, and replicates are required? | Biology is confounded with batch or one group is underrepresented | Independent specimens and balanced processing |
| Modality role | What unique question does each assay answer? | Multiple assays duplicate the same signal without resolving a new uncertainty | Orthogonal protein, chromatin, metabolite, or imaging evidence |
| Integration target | Which identities, states, and condition effects must be preserved? | Over-correction removes real biology or under-correction retains technical separation | Alternative integration and held-out labels |
| Spatial coordinate | How will cells or regions be aligned to anatomy? | A cluster map is interpreted as a tissue mechanism without a stable coordinate | Morphology, compartment markers, or image registration |
| Statistical unit | Are conclusions supported across specimens rather than cells alone? | Pseudoreplication produces narrow confidence intervals | Specimen-level summaries or hierarchical models |
| Mechanism boundary | Which claim is descriptive, associative, directional, or causal? | Proximity or correlation is written as signaling or regulation | Functional perturbation and independent response readouts |
Discovery and validation should also be separated at the specimen and feature levels. If one cohort is used to discover a cell state, choose its markers, optimize a spatial score, and report the same score as confirmation, every step remains dependent on the original data. A stronger plan fixes the cell-state definition or spatial signature after discovery, then tests it in additional specimens with the same inclusion rules. When an independent cohort is not feasible, leave-one-specimen-out analysis, held-out tissue regions, or a targeted assay performed on reserved sections can still reveal how much the result depends on one sample.
Section relationships deserve equal attention. Measurements on the same section provide the closest spatial correspondence but may be technically incompatible or consume the tissue. Adjacent sections support multimodal comparison, yet small structures and sharp boundaries may shift between planes. Distant serial sections improve three-dimensional coverage but require registration and cannot be treated as identical microenvironments. Reports should state whether modalities were measured on the same, adjacent, or non-adjacent sections; how landmarks were aligned; and what spatial tolerance was used when calling features co-localized. This distinction is essential when connecting regional transcriptomics to imaging or metabolite maps.
Finally, atlas reuse should be planned as a versioned data product. Record the reference genome, feature annotation, cell ontology, tissue hierarchy, preprocessing software, and release-specific identifiers. Preserve donor- or specimen-level metadata even when a public browser presents aggregated views. These details determine whether a later project can compare states, transfer labels, or reproduce a reported spatial domain without silently changing the biological definition.
Three practical habits follow. First, design the metadata table before the sample sheet. Second, reserve tissue or material for validation rather than consuming every section in discovery. Third, define analysis deliverables in terms of claims: cell inventory, spatial localization, multimodal alignment, candidate interactions, or functional hypotheses. This makes it easier to choose suitable assays and prevents an attractive method from becoming the project objective.
For spatial projects, a spatial transcriptomics data analysis service should document image registration, cell or spot annotation, spatial domains, neighborhood definitions, replicate-aware statistics, and uncertainty. Experimental and computational plans should be developed together; otherwise, the analysis team may discover that key contrasts, sections, or validation samples were never collected.
Plan an Integrated Omics Study
The eight studies show that successful omics projects are built around matched scales. Cell atlases need a sampling frame. Spatial studies need an anatomical coordinate. Multi-omics studies need a shared biological unit. Computational frameworks need preservation tests and evidence limits. The right design is the smallest combination of samples, modalities, and validation experiments that can answer the defined question without relying on unsupported inference.
CD Genomics supports research-use planning and execution for single-cell, spatial transcriptomics, and integrated omics studies. Explore our single-cell sequencing service, spatial transcriptomics services, single-cell omics bioinformatics and data mining service, and spatial transcriptomics data analysis service to align sample design, assay selection, integration, spatial analysis, and validation-ready reporting. Our services are intended for research use only and are not intended for clinical diagnosis or treatment.
FAQs
References
- Lin CY, Hsu WH, Lu MYJ, et al. A 3D amphioxus brain atlas illuminates the blueprint of the ancestral chordate brain and highlights unique vertebrate innovations. Science Advances. 2026;12(27):eaee5076.
- Goldman OV, DeFoe AE, Qi Y, et al. A single-nucleus transcriptomic atlas of the adult Aedes aegypti mosquito. Cell. 2025;188(25):7267-7290.e26.
- Wei R, Du Z, Wang J, et al. A single-cell transcriptomic landscape characterizes the endocrine system aging in the mouse. Protein & Cell. 2026;17(1):27-45.
- Qin N, Liu X, Li S, et al. Single-nucleus and spatial transcriptomics reveal the cell populations of intercalary meristems in bamboo. Proceedings of the National Academy of Sciences of the United States of America. 2025;122(51):e2511701122.
- Wu C, Tang H, Yu Y, et al. Single-cell spatial transcriptomics reveals potential molecular mechanisms of Abelmoschus manihot (L.) medic in treating diabetic kidney disease. iMeta. 2025;4(6):e70099.
- Qi Y, Chen X, Zheng S, et al. Single-cell and spatially resolved omics reveal transcriptional and metabolic signatures of ovarian endometriomas. Nature Communications. 2025;16(1):11539.
- Zhang Y, Ming W, Yu B, et al. Robust integration and annotation of single-cell and spatial omics data using interpretable gene programs. Cell Genomics. 2026;6(4):101105. First published online in 2025.
- Hou S, Ma W, Zhou X. FastCCC: a permutation-free framework for scalable, robust, and reference-based cell-cell communication analysis in single cell transcriptomics studies. Nature Communications. 2025;16(1):11428.