The most common pitfalls in spatial biology study design occur when the biological question, tissue, sampling strategy and analysis plan are considered separately. A study can generate technically impressive spatial transcriptomics data and still fail to support a clear development decision.
For biotech and pharmaceutical teams working with limited clinical tissue, that can mean losing irreplaceable material, delaying a programme or producing exploratory findings that cannot be validated. These six pitfalls should be addressed before the first section is placed on a slide.
1. Choosing a platform before defining the biological question
Starting with “We want to use spatial transcriptomics” is not a study objective. The study needs to specify what must be learned, such as which cells express a candidate biomarker, where treatment response occurs or whether immune exclusion differs between responders and non-responders.
Without that definition, teams may prioritise resolution, gene coverage or novelty that does not improve the required conclusion. The better approach is to write the intended biological comparison and endpoint first. Platform selection should then follow the question, tissue type, required resolution, capture area and analytical needs.
2. Assuming available tissue is suitable tissue
The presence of a formalin-fixed paraffin-embedded (FFPE) block does not confirm that it can support the intended spatial workflow. Fixation, storage, sectioning, tissue folds, necrosis, RNA quality and tissue morphology can all affect assay performance.
Poor preparation can compromise an expensive experiment before data generation begins. Official tissue preparation guidance also makes clear that preserving morphology and messenger RNA integrity is critical to data quality.
Assess block orientation, tissue area, morphology, RNA quality and assay compatibility before committing the study cohort. A pilot using representative material can expose risks while preserving the most valuable samples.
3. Confusing more cells with more biological evidence
A single tissue section may generate measurements from thousands of spots or cells. Those observations are not thousands of independent biological replicates.
When the study compares treatment groups, disease states or patient outcomes, the independent unit is usually the patient, donor or animal. Serial sections and multiple regions can improve measurement within a specimen, but they do not replace biological replication. Treating cells or spots as independent can produce overconfident results.
Define the biological, experimental and observational units in advance. Randomise samples across slides and processing batches where possible, and ensure the statistical model reflects the study hierarchy. The Bioconductor spatial transcriptomics design guidance provides a useful explanation of replication, pseudo replication and batch confounding.
4. Sampling the easiest region rather than the relevant biology
Spatial studies rarely profile every part of every specimen. Region-of-interest or field-of-view selection can therefore determine which biology is visible.
Selecting only large, visually obvious or technically convenient regions may exclude invasive margins, rare immune structures or heterogeneous stromal compartments. Manual selection may also vary between reviewers. Research on systematic region-of-interest selection shows why representativeness and reproducibility matter when capture area is limited.
Define selection rules before reviewing outcomes. Use pathology input, sample comparable compartments across groups and document why each region was included. If heterogeneity is central to the hypothesis, include enough regions and specimens to represent it.
5. Designing a targeted panel without the downstream analysis in mind
A targeted panel can contain biologically interesting genes yet still fail to distinguish the cell types, states or pathways required for the study.
For cell-resolved spatial transcriptomics, the panel may need markers for cell identification, state definition, the primary hypothesis, technical controls and possible validation targets. Adding genes without considering detectability, redundancy or analytical use can consume panel capacity without improving interpretation.
Build the panel with the pathologist, spatial scientist and bioinformatician involved. Test whether the proposed genes can support the intended annotations and comparisons before finalising the assay.
6. Treating bioinformatics as a post-study service
Analysis choices influence the experimental design. They determine which metadata is required, how samples should be balanced, how regions will be compared and whether the study can test its stated hypothesis.
Cell segmentation is one example. A recent study found that segmentation errors can confound differential expression, neighbourhood and ligand-receptor analyses by assigning molecules to the wrong cells. The authors concluded that these errors can sometimes dominate downstream results (Nature Genetics).
Align the analysis and quality-control plan before data generation. Define how segmentation will be reviewed, how cell types will be annotated, which comparisons are confirmatory or exploratory and which findings require orthogonal protein-level validation.
