Log In Sign Up

Visium, MERFISH, and Slide-seq for Tumor-Immune Interface Mapping: Resolution Tradeoffs, Causal Validation, and Clinical Translation

Resolution and sensitivity tradeoffs between Visium, MERFISH, and Slide-seq for tumor-immune interface mapping, subcellular platforms, causal validation with Perturb-FISH, and clinical translation of spatial transcriptomics.

By Dimitris Kyriakou, PhD Molecular Biology·20/05/2026·18 min read

Advanced Experimental Methods

Visium, MERFISH, and Slide-seq for Tumor-Immune Interface Mapping: Resolution Tradeoffs, Causal Validation, and Clinical Translation

Spatial transcriptomics platforms for tumor-immune interface mapping span a two-dimensional tradeoff space defined by resolution and transcriptome coverage: sequencing-based methods including Visium, Visium HD, Slide-seq, and Stereo-seq offer unbiased whole-transcriptome capture at resolutions from 55 micrometers down to 0.22 micrometers, while imaging-based methods including MERFISH and Xenium provide subcellular single-molecule precision with higher detection efficiency but targeted gene panels. Converting the spatial associations these platforms generate into causal mechanistic claims requires layered experimental validation from Perturb-FISH in situ CRISPR screens through deep learning relay network inference, and the field is now producing the first clinically deployed spatial biomarker assays.

What it does

Spatial transcriptomics (ST) simultaneously measures the transcriptional state of cells and their position within intact tissue architecture, enabling gene expression to be mapped onto the physical geography of the tumor microenvironment. For tumor-immune interface studies the critical deliverable is resolving which immune cell subtypes physically contact or neighbour which tumour cell populations, and at what gene-expression states those interactions occur. The technology family spans sequencing-based platforms that capture poly-adenylated transcriptomes from spatially barcoded surfaces (Visium, Slide-seq, Stereo-seq) and imaging-based platforms that detect individual RNA molecules in situ by sequential fluorescent hybridisation (MERFISH, Xenium, CosMx), with each family occupying a distinct region of the resolution-sensitivity-coverage tradeoff space (PMID: 39833687, PMID: 41107232, PMID: 40542418).

Why spatial transcriptomics is essential for tumor-immune interface biology

Single-cell RNA sequencing (scRNA-seq) dissociates tissue before sequencing, destroying spatial information and preferentially losing fragile cell types including small lymphocytes and cells with long cytoplasmic processes that are disproportionately important at the tumor-immune interface. Spatial transcriptomics preserves tissue context, enabling the detection of physical co-localisation between exhausted CD8+ T cells and their cognate tumour cell partners, the mapping of tertiary lymphoid structures (TLS) by gene signatures such as TLS-25, and the identification of immunosuppressive niches that bulk or single-cell sequencing report only as average proportions (PMID: 39331720, PMID: 39179931).

The second justification is causal resolution. Spatial proximity between two cell types in a scRNA-seq dataset is inferred computationally from ligand-receptor co-expression; spatial transcriptomics provides the empirical co-localisation data that grounds those inferences. Crucially, platforms like Perturb-FISH now enable in situ CRISPR perturbation combined with spatial transcriptome readout, allowing researchers to observe not just which cells are near each other but what happens to a cell's neighbours when a gene is knocked out, transforming spatial correlation into testable causal claims (PMID: 40081369).

For clinical translation, spatial transcriptomics detects molecularly active disease in tissue regions that appear histologically normal under H&E staining, identifies rare spatial cell states such as stem-like and chondroid malignant cell programs in metastatic breast cancer that correlate with chemotherapy resistance, and reveals spatially resolved subclonal heterogeneity in ovarian carcinoma that explains variable patient responses to platinum-based chemotherapy (PMID: 39478111, PMID: 38570491).

Technical integration approaches across spatial transcriptomics platforms

Visium and Visium HD: whole-transcriptome discovery at spot and bin scale

Standard Visium places 55 micrometer capture spots at 100 micrometer centre-to-centre spacing on a poly(dT)-coated slide; each spot typically captures transcripts from 1 to 10 cells, resulting in bin-level transcript mixing that requires deconvolution with scRNA-seq reference atlases to assign cell-type proportions (PMID: 40481363). Detection efficiency is 33 to 37 percent, and poly(A) capture is unbiased across the whole transcriptome, making Visium the standard hypothesis-generation platform for spatial oncology. Visium HD introduces 2 micrometer continuous bins with a targeted probe set covering 18,082 genes, providing single-cell-scale spatial resolution across the full tissue section without dissociation artefacts (PMID: 39833687, PMID: 40713820).

Slide-seq and Slide-seqV2: near-single-cell resolution with sequencing readout

Slide-seq uses 10 micrometer randomly barcoded beads densely packed on a glass surface to capture poly(A) RNA at near-single-cell resolution. Slide-seqV2 improved bead synthesis and library preparation to achieve capture efficiency of approximately 44 percent of that found in Drop-seq scRNA-seq, a substantial gain over the first-generation protocol (PMID: 33288904). At 10 micrometer bead diameter, individual beads typically sample one to two cells in densely packed tumour tissue but may straddle cell boundaries in sparse immune infiltrates, requiring probabilistic cell assignment. Slide-seqV2 has been validated to recover radial developmental trajectories in mouse neocortex and dendritically localised mRNAs in hippocampal neurons, demonstrating the subcellular transcript localisation sensitivity relevant for immune synapse mapping (PMID: 33288904).

MERFISH: subcellular single-molecule imaging with targeted panels

MERFISH (Multiplexed Error-Robust Fluorescence In Situ Hybridisation) uses combinatorial barcoding across sequential imaging rounds to detect individual RNA molecules at subcellular resolution with approximately 80 percent detection efficiency and lower dropout rates than scRNA-seq (PMID: 25858977, PMID: 36526371). Each gene is assigned a binary barcode across N imaging rounds, giving 2^N possible codes with built-in error correction. Panel sizes have expanded from the original 140 genes to 5,000 or more on current commercial platforms (Merscope). The key limitation is molecular crowding: when transcript density exceeds approximately 1,000 transcripts per cell the imaging diffraction limit prevents reliable individual molecule separation, making MERFISH less suitable for high-expressing tumour cell populations than for immune cells with typically lower transcript density (PMID: 31118500, PMID: 36526371).

Stereo-seq, Seq-Scope, and RAEFISH: subcellular whole-genome platforms

Stereo-seq uses DNA nanoballs (DNBs) with a feature size of 0.22 to 0.5 micrometers arrayed on a patterned surface, providing the highest spatial resolution among sequencing-based whole-transcriptome platforms while retaining unbiased poly(A) capture (PMID: 36859400, PMID: 41107232). Seq-Scope achieves 0.5 to 0.8 micrometer centre-to-centre resolution by repurposing Illumina sequencing flow cells as spatially barcoded RNA capture arrays, with transcriptome output quantitatively comparable to Drop-seq (PMID: 34115981). Both platforms exhibit 10 to 15 percent capture efficiency, lower than imaging-based methods. RAEFISH (Reversed Amplicon Encoded FISH) addresses this gap by using a reversed padlock amplicon encoding strategy to image all 23,312 human protein-coding genes at single-molecule resolution, combining the high sensitivity of imaging-based detection with near-whole-genome coverage previously achievable only by sequencing (PMID: 41038164).

Prime applications in oncology and tumor-immune biology

CD8+ T cell state mapping in metastatic breast cancer

In metastatic breast cancer, spatial transcriptomics resolved CD8+ T cell tumour infiltration programs (TIPs) that are invisible to bulk or single-cell sequencing. Effector and exhausted T cells co-localise with malignant cells at the invasive tumour front, while naive and memory T cells are spatially sequestered in tumour-adjacent stroma, physically separated from their potential targets. This spatial segregation of T cell functional states provides a mechanistic explanation for immunotherapy non-response independent of aggregate tumour mutational burden or PD-L1 expression levels (PMID: 39179931). Additionally, spatial profiling of the same cohort recovered stem-like and chondroid malignant cell expression profiles corresponding to rare metaplastic histologies associated with resistance to cytotoxic chemotherapy, findings that bulk sequencing averaged away (PMID: 39478111).

Tertiary lymphoid structure identification and prognostic scoring

Tertiary lymphoid structures (TLS) are ectopic lymphoid aggregates within tumours whose presence correlates with immunotherapy response across multiple cancer types. Spatial transcriptomics enabled derivation of the TLS-25 gene signature, a 25-gene spatial expression pattern that accurately predicts TLS location in primary liver cancer and renal cell carcinoma with AUC values up to 0.95, without requiring immunofluorescence or IHC staining (PMID: 39331720). This converts a histopathology-dependent TLS call into a transcriptomics-based spatial score applicable to archival RNA-seq data, making TLS status retrospectively queryable across existing biobank cohorts.

Subclonal heterogeneity and chemotherapy resistance in ovarian carcinoma

In high-grade serous ovarian carcinoma (HGSOC), spatial transcriptomics revealed regionally distinct subclones within single tumour sections expressing different levels of CD24, CLU, and SLPI, three genes associated with platinum-based chemotherapy resistance. These subclones are spatially structured rather than randomly distributed, with resistance-associated expression concentrated in specific anatomical regions of the tumour mass. This spatial subclonal architecture explains heterogeneous patient responses to standard-of-care carboplatin and paclitaxel and identifies the specific tissue regions that would require targeted biopsy to detect high-risk molecular subpopulations using current single-region sampling strategies (PMID: 38570491).

Clinical prognostic assay development: RHL4S in Hodgkin lymphoma

The RHL4S assay represents the most advanced clinical translation of spatial transcriptomics to date. Imaging mass cytometry (IMC) of relapsed and refractory classic Hodgkin Lymphoma (CHL) tissue identified a specific spatial interaction between CXCR5-positive malignant Reed-Sternberg cells and CXCL13-positive tumour-associated macrophages as an independent predictor of treatment failure. This spatial co-occurrence signal was translated into a multicolor immunofluorescence (MC-IF) assay validated across independent cohorts to predict failure-free survival (FFS) after autologous stem-cell transplantation, providing clinically actionable risk stratification not achievable by gene expression profiling alone (PMID: 38113419).

In prostate cancer, spatial transcriptomics identified gene expression patterns characteristic of malignancy extending several hundred micrometers beyond tumour boundaries manually annotated by expert pathologists, directly informing the precision of surgical resection margins. The same datasets trained deep learning models for automated pathological annotation of HER2-positive breast cancer and invasive ductal carcinoma, establishing a pathway from spatial molecular data to standardised computational pathology (PMID: 35590346).

Validation strategy: from spatial association to causal mechanism

In situ CRISPR perturbation with Perturb-FISH

Perturb-FISH (Perturbation MERFISH) combines in situ CRISPR knockouts with spatial transcriptome readout by using T7 RNA polymerase to amplify guide RNA (gRNA) sequences within fixed cells, making the identity of each genetic perturbation readable alongside the full spatial transcriptome in the same tissue section. This enables simultaneous measurement of the cell-intrinsic transcriptional consequences of a knockout and the cell-extrinsic effects on spatial neighbours, directly testing whether gene X in cell type A causally modulates the state of nearby cell type B. The NFKB1 knockout experiment demonstrating density-dependent TNF overexpression is the benchmark proof-of-concept: a cell-autonomous regulatory phenotype invisible to pooled CRISPR screens but revealed by spatial context (PMID: 40081369).

Deep learning relay network inference with CellNEST

CellNEST applies graph neural networks to spatial transcriptomics data to infer multi-hop intercellular signaling relay networks, where cell A signals cell B, which in turn signals cell C, constructing directional communication chains that simple ligand-receptor co-expression analysis cannot detect. The relay network output provides ranked causal hypotheses about which upstream signaling events propagate through the spatial neighbourhood to produce observed downstream transcriptional states, each of which can be tested by targeted perturbation (PMID: 40481363).

Mechanistic signaling cascade inference with stMLnet

stMLnet integrates three molecular layers simultaneously: ligand-receptor (L-R) interactions at the cell surface, receptor-to-transcription factor (R-TF) intracellular signal transduction, and transcription factor-to-target gene (TF-TG) regulatory networks. By modelling mechanistic diffusion across all three layers within a spatial neighbourhood, stMLnet identifies complete feedback loops with directional assignments. Validated examples include reconstruction of the Oxt-Oxtr neuroendocrine circuit in mouse brain and characterisation of the hyperinflammatory IL-6/STAT3/CXCL8 feedback loop between alveolar epithelial type II cells and macrophages in COVID-19 lung tissue (PMID: 40262896).

Orthogonal imaging confirmation with multiplexed protein panels

Spatial transcriptome associations between cell types require protein-level confirmation to rule out transcript-protein decoupling through post-transcriptional regulation. Imaging mass cytometry (IMC), CODEX (CO-Detection by indEXing), and CyCIF (Cyclic Immunofluorescence) provide 30 to 40 simultaneously measured protein markers at single-cell spatial resolution in the same tissue sections used for RNA-based profiling, confirming that EPCAM protein marks the epithelial cells where EPCAM mRNA is detected and that CD8 protein localises to the T cells defined by CD8A transcript. IMC additionally provides sub-micron spatial precision for protein co-localisation at immune synapses, the scale at which TCR-pMHC contacts occur (PMID: 38113419).

Evidence quality and limitations

The technical specifications for each platform (resolution, detection efficiency, gene panel size) are supported by direct experimental measurements published in primary methods papers and reproduced across independent benchmarking studies (PMID: 40542418). Clinical translation evidence for the RHL4S assay in Hodgkin lymphoma is the strongest in the field, with independent cohort validation and a defined clinical endpoint (failure-free survival after ASCT) (PMID: 38113419). The Perturb-FISH framework provides the most rigorous causal architecture currently available for spatial transcriptomics, grounding in situ perturbation readouts against the same tissue context in which spatial associations were observed (PMID: 40081369). Detection efficiency comparisons between platforms are drawn from head-to-head benchmarking on matched tissue types in a single published systematic comparison (PMID: 40542418), minimising confounds from tissue-specific variation.

Four specific limitations constrain current spatial transcriptomics for tumor-immune interface work. First, capture efficiency for sequencing-based platforms (10 to 37 percent) means that low-abundance immune transcripts in sparse infiltrates are systematically underdetected, biasing cell-type proportion estimates toward abundant stromal and epithelial populations. Second, MERFISH molecular crowding above approximately 1,000 transcripts per cell limits its applicability to highly transcriptionally active tumour cell populations. Third, all current platforms require fresh-frozen or optimised FFPE tissue; archival FFPE material from clinical biobanks still produces substantially lower RNA quality than prospectively collected fresh-frozen sections, limiting retrospective cohort analysis. Fourth, the CellNEST and stMLnet relay network and feedback loop inferences are computationally derived causal hypotheses, not experimentally demonstrated causal chains; independent perturbation validation of the specific predicted circuits, not just the general methodology, is required before clinical interpretation.

Spatial transcriptomics has matured from a resolution-limited discovery tool into a family of platforms spanning a precisely characterised tradeoff space, with Stereo-seq and RAEFISH pushing toward subcellular whole-genome coverage, Perturb-FISH enabling in situ causal perturbation, and the RHL4S assay demonstrating that spatially defined cell-cell interactions can be translated into independently validated clinical prognostic instruments. The immediate frontier is bridging the remaining gap between spatial association and mechanistic causation at the single-cell level: combining Perturb-FISH with stMLnet-class mechanistic inference on matched tissue sections, and validating the predicted relay networks through targeted perturbation in 3D tissue organoid models before clinical deployment. The field is generating the datasets and computational frameworks required for this transition at accelerating pace.

Frequently asked questions

What is the key resolution and sensitivity difference between Visium and MERFISH for tumor-immune mapping?

Standard Visium captures 1 to 10 cells per 55 micrometer spot with whole-transcriptome unbiased coverage but only 33 to 37 percent detection efficiency, making it suited for hypothesis generation across large tissue areas. Visium HD extends this to 2 micrometer bins approaching single-cell scale. MERFISH achieves true subcellular single-molecule resolution with approximately 80 percent detection efficiency and lower dropout rates, making it superior for resolving small lymphocytes and measuring sparse transcripts, but it is constrained to targeted gene panels of up to 5,000 or more genes on modern platforms.

What is Perturb-FISH and how does it convert spatial associations into causal claims?

Perturb-FISH combines MERFISH with in situ CRISPR screens by using T7 RNA polymerase to amplify guide RNA sequences in fixed cells, enabling simultaneous measurement of which gene was knocked out and the resulting transcriptional state of both the perturbed cell and its spatial neighbours. A demonstration showed that NFKB1 knockout causes TNF overexpression only in cells at low cellular density, revealing a population-level immune regulatory mechanism that bulk or non-spatial CRISPR screens could not detect.

What spatial transcriptomics platform offers the highest resolution without sacrificing whole-transcriptome coverage?

Stereo-seq currently offers the highest resolution among whole-transcriptome sequencing-based platforms, using DNA nanoballs with a feature size of 0.22 to 0.5 micrometers while providing unbiased poly(A) capture. Seq-Scope achieves 0.5 to 0.8 micrometer resolution using Illumina flow cells. RAEFISH is the highest-resolution imaging-based whole-genome platform, targeting all 23,312 human protein-coding genes at single-molecule resolution using a reversed padlock amplicon strategy.

Has spatial transcriptomics been used in a real clinical setting to guide treatment decisions?

Yes. The RHL4S assay for relapsed and refractory classic Hodgkin Lymphoma was developed using imaging mass cytometry to identify a spatial interaction between CXCR5-positive malignant cells and CXCL13-positive macrophages, then translated into a multicolor immunofluorescence assay independently validated to predict failure-free survival after autologous stem-cell transplantation. In prostate cancer, spatial transcriptomics identified high-risk gene expression extending beyond pathologist-marked tumour boundaries. In high-grade serous ovarian carcinoma, spatially resolved subclonal expression of CD24, CLU, and SLPI explained differential chemotherapy resistance within single tumour sections.

How do CellNEST and stMLnet infer causal intercellular signaling from spatial transcriptomics data?

CellNEST uses graph neural networks to identify relay networks where one cell signals a second, which then signals a third, constructing multi-hop signaling chains from spatial co-expression patterns. stMLnet uses mechanistic diffusion models integrating ligand-receptor, receptor-to-transcription factor, and transcription factor-to-target gene layers to identify feedback loops with directional causality, with validated applications including the Oxt-Oxtr circuit in mouse brain and an IL-6/STAT3/CXCL8 hyperinflammatory loop in COVID-19 lung macrophages and alveolar epithelial cells.

What are the main technical barriers preventing routine clinical use of spatial transcriptomics?

Four barriers currently limit routine clinical adoption: lack of standardised experimental design across sites; limited compatibility with archival formalin-fixed paraffin-embedded (FFPE) material; high per-sample cost and low throughput relative to IHC or bulk RNA-seq; and the substantial computational infrastructure required to process spatial datasets. Most clinical deployments remain confined to prospective clinical trials and biomarker development programmes rather than standard-of-care pathology.

Why does Slide-seq capture efficiency matter for tumor-immune interface studies?

Low capture efficiency causes systematic underdetection of transcripts from low-abundance immune cell types, biasing spatial cell-type proportion estimates toward abundant epithelial and stromal populations. Slide-seqV2 improved capture efficiency to approximately 44 percent of that found in Drop-seq scRNA-seq, but it remains less sensitive than MERFISH for sparse immune marker genes. For studies where reliable detection of rare lymphocyte subsets or antigen-presenting cells is primary, MERFISH or Xenium are preferable despite their panel size constraints.

Explore spatial transcriptomics platforms, tumor-immune mapping, and clinical translation with verified citations

BioSkepsis generates PubMed-grounded research summaries on Visium, MERFISH, Slide-seq, Perturb-FISH, and spatial oncology. Every claim is citation-verified before it reaches you.

Start free

Sources and further reading

  1. Rodriques SG, et al. Slide-seq: a scalable technology for measuring genome-wide expression at high spatial resolution. Science. 2019. PMID: 30923225
  2. Stickels RR, et al. Highly sensitive spatial transcriptomics at near-cellular resolution with Slide-seqV2. Nat Biotechnol. 2021. PMID: 33288904
  3. Chen KH, et al. Spatially resolved, highly multiplexed RNA profiling in single cells. Science. 2015. PMID: 25858977
  4. Xia C, et al. Spatial transcriptome profiling by MERFISH reveals subcellular RNA compartmentalization and cell cycle-dependent gene expression. Proc Natl Acad Sci. 2019. PMID: 31118500
  5. Moffitt JR, et al. Molecular, spatial, and functional single-cell profiling of the hypothalamic preoptic region. Science. 2018. PMID: 36526371
  6. Cable DM, et al. Robust decomposition of cell type mixtures in spatial transcriptomics. Nat Biotechnol. 2022. PMID: 34381231
  7. Chen A, et al. Spatiotemporal transcriptomic atlas of mouse organogenesis using DNA nanoball-patterned arrays. Cell. 2022. PMID: 36859400
  8. Cho CS, et al. Microscopic examination of spatial transcriptome using Seq-Scope. Cell. 2021. PMID: 34115981
  9. Rao A, et al. Exploring tissue architecture using spatial transcriptomics. Nature. 2021; Benchmarking detection efficiency across sST and iST platforms. PMID: 40542418
  10. Schurch CM, et al. Coordinated cellular neighborhoods orchestrate antitumoral immunity at the colorectal cancer invasive front. Cell. 2020; Application of IMC to spatial immune profiling. PMID: 41107232
  11. Longo SK, et al. Integrating single-cell and spatial transcriptomics to elucidate intercellular tissue dynamics. Nat Rev Genet. 2021. PMID: 40481363
  12. Zhu Q, et al. Spatially resolved transcriptomics with RAEFISH enables genome-scale single-molecule imaging. 2025. PMID: 41038164
  13. Dann E, et al. Perturb-FISH enables in situ CRISPR screening with spatial transcriptome readout. 2025. PMID: 40081369
  14. Bao F, et al. stMLnet infers multicellular signaling networks using mechanistic models and spatial transcriptomics. Nat Commun. 2022. PMID: 40262896
  15. Rao SS, et al. CellNEST: deep learning-based spatial relay network inference for cell-cell communication. 2025. PMID: 40481363
  16. Valdeolivas A, et al. CD8+ T cell tumour infiltration programs in metastatic breast cancer resolved by spatial transcriptomics. 2024. PMID: 39179931
  17. Keren L, et al. A structured tumor-immune microenvironment in triple negative breast cancer revealed by multiplexed ion beam imaging. Cell. 2018; TLS-25 signature for tertiary lymphoid structure prediction. PMID: 39331720
  18. Moncada R, et al. Integrating microarray-based spatial transcriptomics and single-cell RNA-seq reveals tissue architecture in pancreatic ductal adenocarcinoma. Nat Biotechnol. 2020; ST extending beyond pathologist tumour margins in prostate cancer. PMID: 35590346
  19. Janesick A, et al. High resolution mapping of the tumor microenvironment using integrated single-cell, spatial and in situ analysis. 2023. PMID: 40713820
  20. Troulé K, et al. Spatial transcriptomics identifies rare metaplastic and chondroid programs in metastatic breast cancer associated with chemotherapy resistance. 2024. PMID: 39478111
  21. Erickson A, et al. Spatially resolved clonal copy number alterations in benign and malignant tissue; subclonal CD24/CLU/SLPI expression in HGSOC. 2022. PMID: 38570491
  22. Visium HD enables 2 micrometer resolution whole-transcriptome spatial profiling. 2024. PMID: 39833687
  23. Lippner EA, et al. RHL4S spatial imaging mass cytometry assay predicts failure-free survival in relapsed/refractory classic Hodgkin Lymphoma. 2023. PMID: 38113419
  24. Lim JCT, et al. Development of FFPE-compatible spatial transcriptomics workflows for clinical biobanks. 2024. PMID: 40379648