Researchers Map Question-Driven Framework for Designing Spatial Transcriptomics Studies

A conceptual roadmap published in Frontiers in Genetics establishes an intuitive, question-driven guide for life science researchers navigating the rapidly expanding domain of spatial transcriptomics. Authored by a team from the University of Mississippi led by Saroj Nepal, Anita Shrestha, Dikshya Niraula, and Bivek Pokharel, the review addresses a critical gap facing biologists who transition from conventional genomic tools to spatially resolved omics. While dissociated single-cell sequencing and bulk transcriptomic profiling obscure cellular relationships by averaging molecular signals across tissue homogenates, spatial transcriptomics measures RNA profiles while preserving their native physical coordinates within intact tissue architectures. This enables researchers to observe developmental patterning, uncover specialized microenvironmental niches, map pathological tissue remodeling, and characterize region-specific gene expression. Rather than focusing strictly on technical instrument parameters, the authors emphasize that experimental design must begin with the underlying biological inquiry, determining the exact spatial information required before selecting laboratory platforms and analytical pipelines.

The authors emphasize that spatial transcriptomics is not a monolithic method but a diverse family of complementary technologies designed to address distinct biological objectives. Choosing the appropriate system requires balancing key trade-offs between physical spatial resolution, gene coverage, transcript detection sensitivity, sample throughput, tissue area, and computational overhead. Imaging-based methods such as single-molecule fluorescence in situ hybridization, RNAscope, seqFISH+, MERFISH, Xenium, and CosMx Spatial Molecular Imager provide subcellular to cellular resolution and high detection sensitivity by directly visualizing RNA transcripts within intact specimens. These imaging systems frequently require cell segmentation to assign detected molecules to individual cells and typically utilize targeted gene panels ranging from single targets to several thousand genes. In contrast, capture-based sequencing platforms including 10x Genomics Visium, Visium HD, Slide-seq, and Stereo-seq offer unbiased, transcriptome-wide coverage across moderate to very large tissue sections, making them particularly well suited for discovering novel spatially regulated pathways despite differing spatial unit definitions and resolution limits.

Selecting an appropriate tissue preparation method represents another pivotal decision point that determines whether molecular integrity or anatomical structure is prioritized in a study. Fresh-frozen tissues remain the preferred standard when maximizing RNA capture and sequence preservation is paramount, especially for unbiased gene discovery assays where degraded RNA directly compromises downstream measurements. Conversely, formalin-fixed paraffin-embedded tissue blocks provide exceptional histological preservation over many years, allowing researchers to study archived clinical specimens and interpret structural morphology accurately, even though chemical crosslinking fragments RNA and can reduce detection efficiency. The authors advise researchers to evaluate whether a selected assay possesses validated compatibility with the target preservation format, noting that commercial options like Stereo-seq and Visium have progressively introduced specific workflows adapted to fixed clinical specimens.

To conceptualize data generation and primary quality assessment, the authors use capture-based Visium sequencing as an illustrative model compared to constructing an environmental weather map. Just as meteorologists record landscape features before gathering local atmospheric metrics, Visium couples high-resolution histological images with position-specific RNA capture across hundreds to thousands of spatial array spots. Because each Visium spot encompasses several adjacent cells, the resulting transcriptional profile reflects a localized cellular neighborhood rather than an isolated single-cell profile. Processing software like Space Ranger subsequently aligns raw sequencing reads with histological imagery to reconstruct the spatial expression atlas. Before proceeding to downstream biological interpretation, rigorous quality control must evaluate individual spatial spots to identify technical failures and tissue damage.

Normalization represents an essential computational stage that allows investigators to conduct valid quantitative comparisons across distinct tissue locations without confounding technical artifacts. Drawing an analogy to measuring rainfall with different-sized collection buckets, the authors explain that a bucket collecting more water simply possesses larger capacity rather than capturing a heavier rainstorm. Normalization adjusts for these technical discrepancies, placing measurements onto comparable scales across the spatial grid. Consequently, emerging computational methods such as SpaNorm are being engineered to decouple technical noise from meaningful spatial biological differences.

Deriving biological insights begins by grouping spatially referenced measurements into functional tissue domains through sequential feature selection, variance scaling, dimensionality reduction, and unsupervised clustering. Selecting highly variable genes isolates informative spatial markers while eliminating ubiquitously expressed housekeeping signals, after which scaling ensures that highly expressed genes do not artificially dominate computational distance metrics. Dimensionality reduction methods such as principal component analysis, UMAP, and t-SNE condense multi-gene coordinated networks into lower-dimensional representations that group transcriptionally similar spots together into discrete clusters. To illustrate domain identification, the authors highlight an Arabidopsis thaliana seedling case study profiled with Visium HD, which revealed seven distinct spatial clusters that mapped closely onto recognized anatomical structures when projected back onto the histology image. Cross-referencing cluster-specific marker genes with the TAIR database and conducting functional pathway enrichment identified specialized developmental domains, confirming that clusters enriched for photosynthesis and plastid organization corresponded directly to light-capturing cotyledon tissues rather than germinating seed regions.

Beyond segmenting tissues into predefined functional clusters, spatial transcriptomics provides powerful statistical methods to identify unannotated spatial patterns and continuous biological gradients. Spatially variable gene analysis tests whether a gene’s expression exhibits non-random spatial structure across a specimen, employing spatial autocorrelation metrics such as Moran’s I to determine if nearby measurement locations display correlated expression levels. In the Arabidopsis case study, Moran’s I identified XYP2 as a significant spatially variable gene concentrated within vascular tissues, aligning with its known biological role in vascular differentiation. While autocorrelation confirms the presence of spatial organization, generalized additive models allow researchers to trace continuous expression changes across tissue geometry. In a case study examining inflamed murine lacrimal glands modeling Sjögren-like disease, generalized additive models demonstrated that the invariant-chain component CD74 and an antigen-presentation gene module shared coordinated, non-uniform spatial gradients, revealing that MHC class II immune activity concentrates within discrete inflammatory microenvironments rather than dispersing uniformly throughout the diseased gland.

A critical conceptual warning emphasized in the roadmap concerns the proper design and statistical execution of comparative studies conducted across multiple biological specimens. The authors contrast within-specimen spatial characterization, where individual spots or cells serve as valid observational units, with between-specimen comparisons, where spots from the same tissue section must never be treated as independent biological replicates. To illustrate this fundamental distinction, the authors compare specimens to separate residential houses and measurement spots to thermometers distributed across different rooms. Measuring temperatures in ten rooms inside a single house provides rich spatial detail regarding that dwelling’s climate, but it cannot transform that single structure into ten independent buildings. Researchers should instead summarize measurements within specimens using pseudobulk aggregation or apply hierarchical mixed-effects models that explicitly account for non-independent, nested correlation structures.

The framework underscores that computational findings require systematic validation across analytical strategies, physical replicates, and independent measurement modalities before establishing biological conclusions. Because results can fluctuate based on clustering algorithms or normalization parameters, cross-validating findings using multiple computational angles—such as confirming that differential expression, spatially variable gene analysis, and gradient modeling indicate identical pathways—substantially strengthens confidence. Furthermore, evaluating local spatial coherence by assessing whether neighboring spots share concordant cluster assignments confirms domain stability, as demonstrated in the lacrimal gland Visium analysis. Investigators must also evaluate whether observed patterns replicate across independent biological subjects, employing the analogy of tasting sweet apples from one tree, which demonstrates within-tree consistency but cannot establish that apples from different orchards will also be sweet. Finally, orthogonal techniques like RNAscope or immunohistochemistry should verify that identified RNA signatures correspond to localized protein expression, while carefully controlling for potential confounders such as animal age, sex, tissue handling delay, and batch integration effects.

Looking toward future horizons, the authors highlight significant technical and computational frontiers that promise to broaden the biological utility of spatial transcriptomics studies. Existing sequencing platforms predominantly capture polyadenylated messenger RNA transcripts, leaving non-coding RNAs, transfer RNAs, and microRNAs largely unmapped due to secondary structures and chemical base modifications. Emerging chemistries utilizing random priming, in situ polyadenylation, padlock probes, rolling circle amplification, and enzymatic demethylation are being adapted to profile small regulatory RNAs in their native tissue microenvironments. Ultimately, the authors remind investigators that computational spatial patterns serve as hypothesis-generating frameworks rather than definitive causal mechanisms, urging life scientists to ground their spatial transcriptomics experimental designs in clear biological questions backed by rigorous orthogonal validation.

Subject of Research: Conceptual design and analytical roadmap for spatial transcriptomics studies

Article Title: Thinking spatially: a conceptual roadmap for designing spatial transcriptomics studies

Article References: Nepal, S., Shrestha, A., Wasti, A., Dawadi, P., Roy, M., Koirala, P., Islam, M. T., Emu, N. J., Bristy, S. A., Bynum, K. G., Nighswander, N., Singh, A. P., Miura, S., Niraula, D., & Pokharel, B. (2026). Thinking spatially: a conceptual roadmap for designing spatial transcriptomics studies. Frontiers in Genetics, 17, Article 1958622. https://doi.org/10.3389/fgene.2026.1958622

Image Credits: AI Generated

DOI: 10.3389/fgene.2026.1958622

Keywords: spatial transcriptomics, computational biology, gene expression mapping, tissue architecture, single-cell analysis, quality control, bioinformatics, spatial domain detection, Thinking, spatially, conceptual, roadmap