Decoding Tumor Architecture: A Comprehensive Guide to Spatial Transcriptomics in Cancer Research

Abstract

Spatial transcriptomics (ST) is revolutionizing cancer research by preserving the spatial context of gene expression, enabling an unprecedented view of the tumor microenvironment (TME). This article provides researchers and drug development professionals with a comprehensive resource on ST, from foundational concepts to cutting-edge applications. We explore how ST uncovers distinct spatial domains like the tumor-microenvironment interface and leading edge, detail the rapidly evolving landscape of sequencing-based and imaging-based technologies, and offer guidance for platform selection and data analysis. Furthermore, we discuss the critical integration of artificial intelligence and computational validation to translate spatial discoveries into clinically actionable insights, ultimately advancing our understanding of tumor heterogeneity, progression, and therapeutic response.

Unveiling the Spatial Landscape: How ST Reveals the Architecture of the Tumor Microenvironment

Spatial transcriptomics has emerged as a revolutionary technological paradigm that bridges the critical gap between cellular gene expression profiles and their native spatial context within tissues. This approach represents a fundamental advancement beyond single-cell RNA sequencing by preserving and quantifying the anatomical organization of transcriptomes, enabling researchers to decipher complex tissue architecture with unprecedented resolution. The core principle underpinning all spatial transcriptomics methodologies is the precise linking of quantitative gene expression data to specific physical locations within tissue sections, thereby creating comprehensive maps of transcriptional activity in situ. This technical guide examines the established and emerging technologies, computational frameworks, and experimental applications of spatial transcriptomics, with particular emphasis on its transformative potential for elucidating tumor organization architecture. As these methods continue to evolve toward higher resolution and greater multiplexing capacity, they are poised to redefine our understanding of cellular ecosystems in health and disease states.

Core Technological Principles and Methodologies

The fundamental objective of spatial transcriptomics is to measure genome-wide expression data while preserving spatial context, addressing a critical limitation of single-cell RNA sequencing technologies that require tissue dissociation [1]. The field has developed along two primary technological trajectories: sequencing-based approaches that capture positional information through spatial barcoding, and imaging-based approaches that directly visualize RNA molecules within intact tissue sections [2].

Sequencing-Based Approaches

Next-Generation Sequencing (NGS)-based methods represent one major category of spatial transcriptomics technologies. These approaches employ spatial barcoding strategies to encode positional information onto transcripts before sequencing [2]. The foundational innovation came from Ståhl et al. (2016), who developed a method to capture poly-adenylated RNA on spatially-barcoded microarray slides prior to reverse transcription, ensuring each transcript could be mapped back to its original spot using unique positional molecular barcodes [2]. This initial technology featured arrays with approximately a thousand spots, each 100μm in diameter with 200μm center-to-center spacing, enabling unbiased investigation of large tissue areas without pre-selecting gene targets [2].

Commercial implementations such as the 10x Genomics Visium platform have improved upon this foundation, offering enhanced resolution (55μm diameter spots with 100μm center-to-center spacing) and increased sensitivity (>10,000 transcripts per spot) [2]. Alternative NGS-based methods like Slide-Seq utilize randomly barcoded beads deposited onto slides for mRNA capture, achieving higher resolution (10μm) through in situ indexing of barcode positions [2]. Continued technological innovations have pushed resolutions further toward the single-cell and subcellular levels, with methods such as Seq-Scope achieving subcellular resolution spatial barcoding capable of visualizing nuclear and cytoplasmic transcripts separately [2].

The universal workflow for NGS-based approaches involves capturing RNA molecules on spatially barcoded oligos, converting them to cDNA with embedded positional information, performing high-throughput sequencing, and computationally reconstructing spatial expression patterns by mapping sequence reads back to their tissue origins using the barcode information [2].

Imaging-Based Approaches

Imaging-based spatial transcriptomics methodologies directly visualize and quantify RNA molecules within intact tissue sections through two primary strategies: in situ sequencing (ISS) and in situ hybridization (ISH) [2]. In situ sequencing-based methods involve reverse transcribing target RNAs directly in tissue, amplifying them via rolling circle amplification, and then performing sequencing-by-ligation or sequencing-by-synthesis in situ [2]. Techniques such as STARmap have incorporated advances in hydrogel chemistry with improved padlock and primer design to profile thousands of genes in complex tissues like mouse cortex [2].

In situ hybridization-based methods, including multiplexed error-robust fluorescence in situ hybridization (MERFISH) and sequential fluorescence in situ hybridization (seqFISH), use multiple rounds of hybridization with fluorescently labeled probes to detect hundreds to thousands of different RNA species [1]. These approaches can achieve subcellular resolution and high detection efficiency (recently reaching 80% relative to the gold standard smFISH) but typically require a priori selection of target genes [2].

More recently, commercial platforms such as the CosMx Human Whole Transcriptome (WTX) assay and Xenium platform have demonstrated the ability to generate spatially resolved, single-cell transcriptomic data across various tissues and experimental models, including FFPE tumors and CRISPR-edited spheroids [3]. These technologies increasingly integrate artificial intelligence-powered tools like InSituType and InSituCor to uncover spatially organized gene modules and pathway activity patterns that traditional approaches cannot resolve [3].

Table 1: Comparison of Major Spatial Transcriptomics Technology Categories

Feature	NGS-Based Approaches	Imaging-Based Approaches
Gene Throughput	Unbiased, whole transcriptome	Targeted (dozens to thousands of genes)
Resolution	Spot-based (10-100μm), recently reaching subcellular	Single-cell to subcellular (<1μm with expansion microscopy)
Sensitivity	~100 unique transcripts per square μm (rapidly improving)	High (up to 80% detection efficiency relative to smFISH)
Tissue Area	Standardized arrays (up to ~13.2cm for Stereo-seq)	Flexible, limited by imaging time
Sequence Information	Full cDNA sequence enables isoform detection	Limited to targeted sequences
Key Examples	10x Visium, Slide-Seq, Stereo-seq	MERFISH, STARmap, CosMx, Xenium

Diagram 1: Spatial transcriptomics technology classification showing two main approaches and their derivatives.

Experimental Design and Workflow Considerations

Implementing spatial transcriptomics requires careful consideration of multiple experimental parameters to ensure biologically meaningful results. Technology selection depends on the specific research question, with key factors including required resolution, gene throughput, tissue characteristics, and analytical objectives [2] [1].

Platform Selection Criteria

The choice between sequencing-based and imaging-based spatial transcriptomics methods involves balancing multiple technical and practical considerations [2]. For discovery-phase research where unbiased transcriptome coverage is prioritized, NGS-based approaches like Visium provide comprehensive gene expression profiling without requiring pre-specified targets [2]. When studying specific cellular mechanisms with known marker genes or when single-cell resolution is essential, imaging-based approaches such as MERFISH or CosMx offer superior spatial precision [2] [3].

Sensitivity requirements must also be evaluated, as imaging-based methods typically demonstrate higher detection efficiency (approximately 80% relative to smFISH) compared to NGS-based methods, though the sensitivity of the latter is rapidly improving [2]. Tissue size presents another consideration, with NGS-based methods typically utilizing standardized array sizes (approximately 6.5×6.5mm for Visium) while imaging-based methods can accommodate larger areas but require proportionally increased imaging time [2].

Recent benchmarking studies have systematically evaluated multiple sequencing-based spatial transcriptomics methods using reference tissues with well-defined histological architectures, including mouse embryonic eyes, hippocampal regions, and olfactory bulbs [4]. These comparisons revealed significant variability in performance metrics including molecular diffusion, capture efficiency, and effective resolution across different technological platforms [4].

Tissue Preparation and Workflow

Proper tissue handling and preparation are critical for successful spatial transcriptomics experiments. The optimal approach depends on whether fresh frozen or formalin-fixed paraffin-embedded (FFPE) tissue samples are available [1]. Fresh frozen tissues generally provide higher RNA quality and are compatible with both NGS-based and imaging-based methods, while FFPE tissues enable retrospective studies using clinical archives but may present challenges for RNA recovery due to cross-linking [1].

The core workflow for NGS-based methods like Visium involves cryosectioning tissue at appropriate thickness (typically 10-20μm), mounting sections on barcoded spatial capture slides, performing H&E staining and imaging for histological reference, permeabilizing tissue to release RNA for capture on spatially barcoded oligos, and then proceeding with library preparation and sequencing [2] [1]. For imaging-based methods, tissue sections undergo fixation and permeabilization followed by multiple rounds of probe hybridization and imaging for targeted approaches, or reverse transcription and amplification steps for in situ sequencing methods [2].

Quality control throughout the process is essential, including assessment of RNA integrity, optimization of permeabilization conditions for NGS-based methods, and verification of probe specificity for imaging-based approaches [1]. Integration with complementary data modalities such as histopathological imaging, protein detection, and single-cell RNA sequencing references further enhances the biological insights gained from spatial transcriptomics experiments [5] [6].

Diagram 2: Comprehensive workflow for spatial transcriptomics experiments showing parallel paths for different tissue types and technologies.

Computational Analysis Frameworks

The complex, high-dimensional data generated by spatial transcriptomics technologies demands sophisticated computational approaches for proper interpretation and biological insight extraction. The analysis workflow typically encompasses multiple stages from raw data processing to advanced spatial analytics.

Spatial Domain Identification and Clustering

A fundamental application of spatial transcriptomics data is the identification of spatial domains—groups of cells or spots exhibiting similar gene expression patterns that often correspond to functional tissue units [7]. Both non-spatial clustering methods that rely solely on gene expression (e.g., Seurat, Louvain algorithm) and spatial methods that integrate transcriptional profiles with spatial coordinates have been developed [7]. Spatial clustering methods like SpaGCN combine spatial locations and histology data to construct weighted graphs, while STAGATE employs graph attention auto-encoder networks to delineate spatial domains [7].

Recent methodological advances have introduced more sophisticated frameworks for handling complex spatial transcriptomics datasets. The spCLUE framework utilizes a graph-contrastive-learning paradigm to infer spatial domains and spot representations across both single-slice and multi-slice data [7]. Similarly, STAIG integrates gene expression, spatial coordinates, and histological images using graph-contrastive learning coupled with high-performance feature extraction, enabling integration of tissue slices without pre-alignment while effectively removing batch effects [6].

Benchmarking studies have demonstrated that these advanced methods significantly outperform traditional approaches in spatial domain identification. For human brain datasets, STAIG achieved the highest median Adjusted Rand Index (0.69 across all slices) and Normalized Mutual Information (0.71), precisely distinguishing cortical layers L1-L6 and white matter regions that correspond to known anatomical structures [6].

Cell-Type Deconvolution and Spatial Mapping

Most sequencing-based spatial transcriptomics technologies do not yet achieve true single-cell resolution, producing data where each spot contains transcripts from multiple cells [8]. Computational deconvolution methods address this limitation by estimating the cell-type composition within each spot using reference single-cell RNA sequencing data [8].

Multiple algorithmic strategies have been developed for this purpose, including probabilistic methods (RCTD, cell2location, stereoscope), non-negative matrix factorization approaches (SPOTlight), and other specialized frameworks (Tangram, DSTG) [8]. Comprehensive benchmarking of ten state-of-the-art deconvolution methods using diverse real datasets revealed that RCTD and stereoscope achieve the most robust and accurate inferences across different tissues and technological platforms [8].

These deconvolution methods enable researchers to map specific cell types within tissue architecture, revealing organizational principles such as immune cell exclusion zones in tumors or layered neuronal subtypes in brain regions [8]. When combined with spatial domain identification, deconvolution provides a comprehensive view of cellular ecosystems and their organizational logic within tissues.

Table 2: Key Computational Methods for Spatial Transcriptomics Analysis

Method	Primary Function	Algorithm Type	Key Features
spCLUE [7]	Spatial domain identification	Graph contrastive learning	Multi-slice integration, batch effect correction
STAIG [6]	Spatial domain identification	Image-aided graph learning	Histology integration, alignment-free integration
RCTD [8]	Cell-type deconvolution	Probabilistic (Poisson)	Robust reference-based decomposition
stereoscope [8]	Cell-type deconvolution	Probabilistic (negative binomial)	Accurate proportion estimation
cell2location [8]	Cell-type deconvolution	Probabilistic (Bayesian)	Comprehensive tissue architecture modeling
Tangram [8]	Cell-type mapping	Optimization	Single-cell resolution mapping
Spaco [9]	Spatial visualization	Color optimization	Spatially-aware color assignment

Advanced Analytical Capabilities

Beyond basic clustering and deconvolution, spatial transcriptomics data supports increasingly sophisticated analytical approaches. Cell-cell communication inference methods leverage spatial proximity information to identify potential ligand-receptor interactions between neighboring cell types [1]. Spatially variable gene detection algorithms identify transcripts with expression patterns that show significant spatial dependence, often revealing genes involved in local microenvironmental regulation [4].

Recent methodological innovations also enable the prediction of spatial transcriptomics patterns from standard H&E-stained histology images using deep learning approaches. The MISO framework demonstrates that spatial gene expression can be predicted from H&E morphology with near single-cell resolution, potentially expanding spatial transcriptomics insights to vast historical archives of histology samples [5].

Data integration frameworks have been developed to harmonize spatial transcriptomics datasets across different experiments, conditions, and technologies. Methods like SPIRAL enable joint analysis of disparate spatial datasets, facilitating meta-analyses and cross-study comparisons [5]. Visualization tools such as Spaco address the critical challenge of effectively visualizing complex spatial data by implementing spatially aware colorization that ensures biologically distinct adjacent cell types receive maximally distinguishable colors [9].

Applications in Tumor Organization Architecture

Spatial transcriptomics has proven particularly transformative in cancer research, where it has illuminated previously inaccessible dimensions of tumor architecture, heterogeneity, and microenvironment organization. The technology enables comprehensive mapping of cellular ecosystems within tumors, revealing how spatial relationships influence disease progression and therapeutic response.

Tumor Microenvironment Deconstruction

The tumor microenvironment represents a complex ecosystem comprising malignant cells, immune populations, stromal components, and vasculature organized in specific spatial patterns that dictate disease behavior [3]. Spatial transcriptomics has enabled systematic cataloging of these cellular neighborhoods and their association with clinical outcomes. In breast cancer studies, combined spatial transcriptomic and proteomic profiling has revealed distinct immune evasion signatures and microenvironmental cues across different molecular subtypes [3]. Similar approaches in triple-negative breast cancers from women of African ancestry have identified distinctive patterns of immune infiltration and checkpoint interactions that may underlie health disparities [3].

Analysis of colorectal cancer tissues using spatial whole transcriptome approaches has demonstrated superior detection of rare cell populations compared to single-cell RNA sequencing alone, while simultaneously preserving critical spatial context [3]. These analyses have revealed spatially organized gene modules and pathway activity patterns that traditional approaches cannot resolve, including epithelial-mesenchymal transition gradients and immune barrier formations [3].

Tumor-Host Interface and Metastasis

The interface between tumor tissue and adjacent normal stroma represents a critical battlefield where cancer progression is determined. Spatial transcriptomics has uncovered intricate signaling networks at these boundaries that facilitate invasion and immune evasion [1]. In cutaneous melanoma, high-plex spatial profiling has identified highly localized immunosuppressive niches containing PD-L1-expressing myeloid cells positioned at the invasive front [1].

Studies of tumor metastasis using spatial transcriptomics have revealed how cancer cells remodel distant tissue microenvironments to support secondary growth. In brain metastases, spatial profiling has demonstrated how metastatic cells co-opt local stromal signaling networks and create immune-privileged niches that protect them from elimination [1]. These insights are informing novel therapeutic strategies aimed at disrupting these supportive ecosystems.

Therapy Response and Resistance

Spatial transcriptomics provides unique insights into the mechanisms underlying variable responses to cancer therapies. By comparing pre- and post-treatment tumor samples, researchers can identify spatial patterns associated with treatment sensitivity or resistance [3]. In HER2-positive breast cancer, spatial analyses have revealed immunological correlates of complete response to targeted therapy, including specific spatial arrangements of immune cell subsets in relation to tumor cells [1].

The technology has also been deployed to study cellular dynamics in response to emerging therapeutic modalities. For example, CosMx spatial molecular imaging has been integrated with CRISPR screening to map gene edits across thousands of tumor spheroids at single-cell resolution, revealing how specific genetic perturbations alter spatial organization and cellular function [3]. Similarly, multiomic spatial profiling has enabled tracking of CAR-T cells in solid tumors, mapping their spatial distribution, persistence, and functional states within the challenging tumor microenvironment [3].

Diagram 3: Applications of spatial transcriptomics in analyzing tumor organization architecture across multiple biological scales.

Essential Research Reagents and Platforms

The successful implementation of spatial transcriptomics research requires specific reagent systems and platform technologies designed to preserve spatial information while capturing comprehensive molecular data.

Table 3: Essential Research Reagent Solutions for Spatial Transcriptomics

Reagent/Platform	Type	Primary Function	Key Applications
10x Visium [2] [1]	NGS-based spatial platform	Whole transcriptome spatial mapping	Tumor heterogeneity, developmental biology, neuroscience
CosMx Human WTX Assay [3]	Imaging-based spatial platform	Subcellular spatial transcriptomics	FFPE tumors, CRISPR-edited models, tissue microarrays
GeoMx Digital Spatial Profiler [3]	Multiomic spatial platform	Region-specific protein/RNA profiling	High-throughput discovery, tumor microenvironment
nCounter ADC Panel [3]	Targeted spatial profiling	ADC characterization in 3D models	Drug efflux, permeability studies in spheroids
CellScape Platform [3]	Spatial proteomics	High-plex single-cell proteomics	Immune dynamics, cell signaling, tumor-immune interactions
PaintScape Platform [3]	Spatial genomics	3D genome architecture visualization	Chromatin organization, structural variants in cancer
Spatial Barcoded Slides [2]	Consumable	Positional mRNA capture	Whole transcriptome spatial analysis on NGS platforms
Multiplex FISH Panels [1]	Probe library	Targeted RNA visualization	Validation studies, focused pathway analysis

Spatial transcriptomics has fundamentally expanded our ability to study biology in its native anatomical context, creating new opportunities to understand tissue organization in development, homeostasis, and disease. The core principle of linking gene expression to tissue location has proven exceptionally powerful across diverse research domains, particularly in cancer biology where cellular spatial relationships dictate disease behavior and therapeutic outcomes.

The field continues to evolve rapidly along several technological trajectories. Resolution improvements are progressing toward comprehensive single-cell and subcellular spatial transcriptomics, while multiplexing capacities are expanding to enable full transcriptome coverage with imaging-based methods [2]. Multiomic integration represents another frontier, with methods now simultaneously capturing spatial information for transcriptomes, proteomes, and epigenomes within the same tissue section [3]. These advances are coupled with computational innovations that extract increasingly sophisticated biological insights from complex spatial data.

For tumor organization architecture research specifically, spatial transcriptomics offers unprecedented opportunities to decode the functional ecology of cancer ecosystems. The technology enables researchers to move beyond compositional analysis to understand how cellular spatial organization influences clinical behavior, treatment response, and resistance mechanisms. As these methods become more accessible and scalable, they are poised to transform cancer diagnostics and therapeutic development by revealing spatially-defined biomarkers and targets.

In conclusion, spatial transcriptomics represents a paradigm shift in molecular biology that finally enables comprehensive mapping of gene expression within its native structural context. By linking transcriptional information to tissue location as its core principle, this approach has opened new dimensions for understanding cellular organization in health and disease, with particular significance for unraveling the complex architecture of tumors. As technologies mature and analytical frameworks become more sophisticated, spatial transcriptomics will increasingly become an indispensable tool for biomedical research and clinical translation.

The tumor-microenvironment interface and the leading edge (also known as the invasive tumor front) are critical spatial domains where dynamic interactions between cancer cells and non-malignant components directly influence tumor progression, therapeutic resistance, and patient outcomes. These regions serve as active frontiers where tumor cells interact with immune cells, fibroblasts, and other stromal components, creating specialized niches that drive key oncogenic processes. The architectural organization within these domains is not random; rather, it follows predictable patterns that can be quantified and linked to clinical phenotypes [10] [11]. Technological advances in spatial transcriptomics and multiplexed imaging have enabled researchers to move beyond merely cataloging cellular diversity to understanding how the precise spatial arrangement of cells within tumors creates functional biological systems.

This architectural perspective reveals that the leading edge represents a specialized compartment with unique molecular and cellular features distinct from the tumor core. Cells occupying this interface zone often exhibit enhanced proliferative capacity, stem-like properties, and specialized interaction patterns with adjacent non-malignant cells [12]. The clinical relevance of these spatial domains is increasingly recognized, with evidence showing that specific spatial patterns of immune cell localization relative to tumor interfaces can predict patient response to immunotherapy and overall survival outcomes across multiple cancer types [10] [13] [11]. Understanding the biological processes occurring at these spatial boundaries provides not only fundamental insights into cancer biology but also opportunities for developing spatially-informed diagnostic biomarkers and therapeutic strategies that target the tumor-stroma interaction network.

Molecular and Cellular Definitions of Key Spatial Domains

Defining the Tumor-Microenvironment Interface

The tumor-microenvironment interface is a transcriptionally distinct region where tumor cells directly contact adjacent non-malignant tissues. This domain is characterized by a specialized "interface cell state" where both tumor and microenvironment cells upregulate a common set of genes, creating a unique transitional zone between compartments. Research in zebrafish melanoma models has demonstrated that this interface is histologically invisible but transcriptionally distinct, with specialized tumor and microenvironment cells upregulating cilia genes specifically where the tumor contacts neighboring tissues [14]. This interface region displays a transcriptional profile more correlated with tumor (R = 0.33) than with adjacent muscle tissue (R = 0.06), despite histological resemblance to the latter [14]. The identification of this domain requires integrated spatial molecular profiling rather than histological examination alone.

The interface region exhibits distinct pathway activation patterns, with enrichment of biological processes related to extracellular structure organization and immune cell migration [14]. From a topological perspective, this domain typically manifests as a narrow band ranging from 50-500 micrometers in width, depending on cancer type and individual tumor characteristics [12] [14]. In intrahepatic cholangiocarcinoma (ICC), the stromal region within the interface acts as a barrier at the tumor-normal interface while also extending into the tumor region, dispersing or encircling tumor cells [12]. This compartmentalization creates physically distinct microniches that influence cellular behavior and therapeutic responses.

Characterizing the Leading Edge/Invasive Front

The leading edge or invasive tumor front represents the advancing boundary where tumor cells infiltrate adjacent normal tissues. This domain is characterized by tumor cells with enhanced proliferative activity, stemness properties, and epithelial-mesenchymal transition (EMT) features [12]. In intrahepatic cholangiocarcinoma, tumor cells at the leading edge demonstrate significantly higher proliferation rates compared to those in the tumor core, with enrichment of pathways including ribosome biogenesis, ECM receptor interaction, and cell adhesion molecules [12]. These cells exhibit elevated stemness and EMT behaviors alongside reduced hypoxic stress compared to their core counterparts [12].

The leading edge architecture typically includes a unique "triad structure" composed of POSTN+ FAP+ cancer-associated fibroblasts (CAFs), SPP1+ macrophages, and endothelial cells that collectively foster tumor growth and progression [12]. Immune cells within this region display distinct functional states, with CD8+ T cells showing a naïve phenotype with low cytotoxicity and signs of exhaustion, likely due to compromised antigen presentation by antigen-presenting cells [12]. The leading edge also serves as a compartment where mucosal-associated invariant T (MAIT) cells recruit SPP1+ macrophages within the stroma, establishing immunosuppressive networks that facilitate immune evasion [12].

Table 1: Key Characteristics of Spatial Domains in Solid Tumors

Characteristic	Tumor-Microenvironment Interface	Leading Edge/Invasive Front
Cellular Composition	Mixed tumor-stroma cell types; specialized "interface" cells	Predominantly tumor cells with infiltrating immune populations
Transcriptional Features	Upregulation of cilia genes; ETS-factor regulated	Enrichment of proliferation, stemness, and EMT pathways
Spatial Organization	Narrow band (50-500µm) at tumor-stroma boundary	Advancing margin with "triad structure" of CAFs, macrophages, endothelial cells
Immune Context	Macrophages predominantly residing at boundaries; variable T cell infiltration	CD8+ T cells with naïve phenotype, low cytotoxicity, exhaustion markers
Metabolic Features	Increased antigen presentation along edges	Increased metabolic activity at center of microregions
Clinical Significance	Conservation across human melanoma samples	Associated with enhanced proliferation and progression in ICC

Quantitative Methodologies for Spatial Domain Analysis

Computational Framework for Domain Identification

The accurate identification and quantification of spatial domains requires specialized computational approaches that integrate molecular, cellular, and topological features. The SpaLinker framework provides a comprehensive methodology for identifying tumor-normal interface (TNI) regions by detecting spatial distribution patterns of tumor cells rather than relying on pre-defined tumor areas [11]. This approach calculates TNI scores based on the abrupt decrease in tumor cell abundance from the tumor core side to the normal side, effectively addressing the challenge of diffusely distributed tumor cells [11]. The framework employs a stepwise identification procedure that integrates gene expression signatures with cellular co-distribution patterns to improve spatial domain recognition accuracy.

For the identification of tertiary lymphoid structures (TLS) and other specialized microdomains, SpaLinker utilizes a feature selection procedure to determine predictive features, integrating the LC.50sig gene set with the co-distribution of plasma/B cells and T cells to improve identification accuracy [11]. Validation against well-annotated renal cell carcinoma datasets demonstrated that this unit-integrated features approach consistently outperforms single-feature analysis across multiple samples, with predicted TLS scores showing high consistency with ground truth annotations [11]. The framework has been successfully validated across diverse cancer types including hepatocellular carcinoma, intrahepatic cholangiocarcinoma, breast cancer, and nasopharyngeal carcinoma, achieving precision-recall area under the curve (PR-AUC) values from 0.76 to 0.86 and precision for TLS spots from 0.56 to 0.85 [11].

Statistical Spatial Pattern Analysis

Advanced statistical frameworks are essential for distinguishing biologically significant spatial patterns from random distributions. Spatiopath provides a null-hypothesis framework that extends Ripley's K function to analyze both cell-cell and cell-tumor interactions [13]. This method uses embedding functions to map cell contours and tumor regions, enabling the quantification of spatial associations between immune cells and tumor epithelium beyond simple accumulation metrics [13]. The approach analytically computes hyperparameters rather than relying on computationally intensive Monte Carlo simulations, making it suitable for analyzing large, complex tissue regions.

Spatiopath has demonstrated utility in identifying significant spatial patterns such as mast cells accumulating near T cells and tumor epithelium in lung cancer sections, revealing distinct spatial organization patterns with mast cells clustering near the epithelium and T cells positioned farther away [13]. This statistical rigor is particularly important for interpreting immune cell localization patterns that have prognostic significance, such as the distribution of CD8+ T cells in triple-negative breast cancer or myeloid and T cell associations in colorectal cancer [13]. By providing a mathematical foundation for spatial analysis in histopathology, these tools enable robust quantification of spatial features that can serve as biomarkers for patient outcomes and immunotherapy responses.

Diagram Title: Spatial Domain Analysis Workflow

Experimental Workflows for Spatial Domain Characterization

Integrated Single-Cell and Spatial Transcriptomics Protocol

Comprehensive characterization of spatial domains requires the integration of single-cell RNA sequencing (scRNA-seq) with spatial transcriptomics (ST) technologies. A standardized protocol for leading edge analysis involves collecting matched tissue samples from three distinct regions: core tumor tissues (T), leading-edge areas (L), and corresponding non-neoplastic adjacent tissues (N) [12]. For intrahepatic cholangiocarcinoma, this approach has been applied using samples from nine patients, with seven core tumor samples, nine leading-edge samples, and nine adjacent normal samples processed for scRNA-seq, while well-preserved leading-edge samples (n=5) undergo spatial transcriptomics sequencing on the 10x Genomics Visium platform [12]. This integrated design enables the identification of approximately 230,000 high-quality single-cell transcriptomes after quality control, capturing six predominant cell types: myeloid cells, epithelial cells (including malignant tumor cells), fibroblasts, endothelial cells, T/NK cells, and B cells [12].

The analytical workflow for spatial domain characterization includes several critical steps: (1) identification of tumor cells using the inferCNV algorithm with immune cells as reference, combined with marker-based strategies; (2) extraction and re-clustering of tumor cells with identification of proliferating tumor cells based on MKI67, TOP2A, and UBE2C expression; (3) assessment of transcriptional factor regulation using SCENIC; (4) differential expression analysis between spatial domains followed by KEGG pathway enrichment; and (5) evaluation of hypoxia, stemness, and EMT behaviors using the "addmodulescore" algorithm [12]. This integrated approach has revealed that proliferating tumor cells are significantly enriched in the leading-edge area compared to the tumor-core area, with elevated expression of transcription factors E2F1 and CEBPB associated with proliferation and stemness [12].

Advanced Lineage Tracing with Spatial Mapping

The integration of lineage tracing with spatial positioning provides unprecedented insights into clonal dynamics within spatial domains. PEtracer represents an advanced lineage tracing tool that captures cellular family trees while maintaining spatial information through repeated addition of short, predetermined DNA codes to cellular genomes over time [15]. This system utilizes prime editing technology to directly rewrite stretches of DNA with minimal undesired byproducts, enabling each cell to acquire unique lineage tracing marks while maintaining ancestral marks [15]. When applied to metastatic tumors in mice, this approach enables the reconstruction of tumor growth histories by combining lineage relationships with spatial positioning and gene expression profiles [15].

The experimental workflow for PEtracer-based spatial analysis includes: (1) in vivo lineage tracing during tumor growth; (2) tissue collection and processing; (3) advanced imaging to capture lineage tracing marks, spatial positions, and RNA expression patterns; and (4) computational integration of lineage, spatial, and transcriptional data [15]. Application of this approach to lung metastases has revealed that tumors comprise four distinct cellular neighborhoods: nutrient-rich lung-adjacent regions with the highest fitness cells, diverse leading-edge regions with lower fitness, low-oxygen regions beneath the leading edge, and tumor core regions with mixed living and dead cells [15]. This methodology demonstrates that cancer cell traits are influenced by both environmental factors (evidenced by location-dependent expression of Fgf1/Fgfbp1) and inherited lineage factors (evidenced by ancestry-associated expression of Cldn4 in lung-adjacent cells) [15].

Table 2: Experimental Platforms for Spatial Domain Analysis

Technology Platform	Spatial Resolution	Molecular Coverage	Key Applications in Domain Analysis
10X Genomics Visium	55μm with 45μm gap	Whole transcriptome	Mapping microregional structures; identifying spatial subclones
CODEX Multiplex Imaging	Single-cell	100+ proteins	Characterizing cellular neighborhoods; immune cell localization
MERFISH/Vizgen MERSCOPE	Single-cell	Targeted transcript panels	High-resolution mapping of interface regions
PEtracer Lineage Tracing	Single-cell	Lineage barcodes + transcriptome	Reconstruction of clonal dynamics in spatial domains
DBiT-seq	10μm	Whole transcriptome + proteins	Integrated multi-omics for microenvironment analysis

The Scientist's Toolkit: Essential Research Reagents and Platforms

Spatial Transcriptomics and Multiplexed Imaging Platforms

The characterization of tumor-microenvironment interfaces and leading edges relies on specialized research platforms that enable molecular profiling while preserving spatial context. The 10X Genomics Visium platform provides spatial transcriptomics capabilities with a resolution of 55μm with 45μm gaps between spots, enabling whole transcriptome profiling of tissue sections while maintaining architectural information [16] [14]. This technology has been successfully applied to define tumor microregions and spatial subclones across breast cancer, colorectal carcinoma, pancreatic ductal adenocarcinoma, renal cell carcinoma, uterine corpus endometrial carcinoma, and cholangiocarcinoma [16]. For higher-resolution spatial mapping, MERFISH (Vizgen MERSCOPE) and NanoString CosMx platforms offer single-cell resolution through multiplexed error-robust fluorescence in situ hybridization, allowing targeted transcript profiling at subcellular levels [10].

Multiplexed protein imaging platforms are essential for validating transcriptional findings and understanding protein-level interactions at spatial domains. CODEX (Co-Detection by Indexing) enables characterization of more than 100 antibodies in a single panel through cyclic fluorescence imaging with antibody-conjugated barcodes [16] [10]. This technology has been integrated with spatial transcriptomics to identify both immune hot and cold neighborhoods and enhanced immune exhaustion markers surrounding 3D subclones [16]. Alternative approaches include imaging mass cytometry (IMC) and multiplexed ion beam imaging (MIBI), which utilize antibody-metal conjugates detected by mass spectrometry, offering resolutions of 1μm and 300nm respectively with high signal-to-noise ratios for approximately 50 protein targets [10].

Computational Tools for Spatial Analysis

The interpretation of spatial domain biology requires specialized computational tools that can extract meaningful patterns from complex spatial data. SpaLinker represents an integrated framework specifically designed to decipher spatially resolved tumor microenvironment features at molecular, cellular, and tissue structure levels [11]. This tool enables the identification of specialized architectures including tertiary lymphoid structures and tumor-normal interface regions while linking these features to clinical phenotypes by integrating bulk RNA-seq data [11]. For deep learning-based integration of spatial omics with tumor morphology, MISO (deep learning-based multiscale integration of spTx with tumor morphology) predicts spatial transcriptomics from H&E-stained histological slides, significantly outperforming competing methods in extensive benchmarks [5].

Additional computational resources include Giotto, SPATA, and Squidpy, which facilitate basic processing and analysis of various spatial transcriptomics data types [11]. For statistical analysis of spatial patterns, Spatiopath provides a null-hypothesis framework that distinguishes significant immune cell associations from random distributions, extending Ripley's K function to analyze both cell-cell and cell-tumor interactions [13]. Specialized algorithms for spatial domain detection include Morph, used to refine tumor boundaries, determine distances of spots from boundaries, and construct layers of spots indexing their depths to tumor boundaries [16]. These computational tools collectively enable the quantitative analysis of spatial relationships that define functional domains within tumors.

Diagram Title: Research Toolkit for Spatial Domains

Clinical Translation and Therapeutic Implications

Prognostic and Predictive Biomarkers from Spatial Domains

The spatial organization of cells within tumor-microenvironment interfaces and leading edges provides clinically actionable information that can inform prognosis and treatment selection. Spatial patterns of immune cell infiltration within these domains have demonstrated significant prognostic value across multiple cancer types. For example, the spatial distribution of CD8+ T cells in triple-negative breast cancer and the association distances between myeloid cells and T cells in colorectal cancer have been correlated with patient outcomes [13]. Immunophenotypes defined by the degree and pattern of immune cell infiltration at tumor interfaces can serve as predictors of tumor recurrence and response to immunotherapy [10]. The identification of these spatial biomarkers moves beyond traditional quantitative assessments of cell densities to incorporate topological relationships that more accurately reflect functional immune responses.

Computational frameworks like SpaLinker enable the de novo linking of spatial TME features with clinical phenotypes by integrating rich clinical annotation information from bulk RNA-seq data with spatial transcriptomics [11]. This approach has identified clinically relevant spatial architectures across renal cell carcinoma, hepatocellular carcinoma, and melanoma, revealing features associated with distinct clinical outcomes without requiring direct clinical annotations of spatial omics samples [11]. The application of these methods has demonstrated that tumor cells and normal cells located at leading edges display elevated levels of unique molecules linked to immunotherapy response or patient prognosis [11]. These findings highlight the potential for spatial domain analysis to generate clinically validated biomarkers that can guide personalized treatment approaches.

Therapeutic Targeting of Domain-Specific Processes

The unique biological processes occurring at tumor-microenvironment interfaces and leading edges present opportunities for developing spatially-informed therapeutic strategies. The identification of a conserved cilia-enriched interface in human melanoma samples suggests that cilia-related pathways may represent therapeutic targets for impeding melanoma invasion and progression [14]. In intrahepatic cholangiocarcinoma, the "triad structure" composed of POSTN+ FAP+ fibroblasts, SPP1+ macrophages, and endothelial cells at the leading edge represents a multiparametric therapeutic target that could disrupt the synergistic interactions promoting tumor progression [12]. The specialized immune environment at leading edges, characterized by CD8+ T cells with naïve phenotypes and compromised cytotoxicity, suggests potential for immune-modulating approaches that reverse T cell exhaustion specifically within these domains.

Advanced lineage tracing approaches have revealed that targeting the most aggressive cellular populations within specific spatial domains may improve therapeutic efficacy [15]. The observation that cancer cells in nutrient-rich lung-adjacent regions exhibit the highest fitness highlights the potential for metabolic interventions that disrupt nutrient availability in these domains [15]. Similarly, the location-dependent expression of fitness-related genes such as Fgf1/Fgfbp1 suggests that microenvironmental factors shaping cellular behavior in specific domains could be therapeutically modulated [15]. The ability to characterize different populations of cells within tumors based on their spatial positioning enables the development of therapies that target the most aggressive populations more effectively, potentially overcoming resistance mechanisms rooted in spatial heterogeneity.

This case study examines a pivotal 2021 study that integrated spatially resolved transcriptomics (SRT), single-cell RNA-seq (scRNA-seq), and single-nucleus RNA-seq (snRNA-seq) to characterize the tumor-microenvironment (TME) interactions at the boundary of invasive melanoma [17] [18]. The research identified a previously unrecognized, histologically invisible "interface" cell state at the tumor-stroma junction, characterized by a conserved enrichment of cilia-related genes regulated by ETS-family transcription factors [17]. This discovery, conserved in human patient samples, underscores the critical power of SRT in uncovering spatial mechanisms of tumor adaptation and presents a potential new target for therapeutic intervention in melanoma progression [17] [18].

The architecture of the tumor microenvironment is a critical determinant of cancer progression, invasion, and therapeutic response. While traditional sequencing methods have revealed cellular heterogeneity, they necessitate tissue dissociation, thereby destroying the spatial context essential for understanding cell-cell interactions [17]. Spatially resolved transcriptomics (SRT) has emerged as a transformative technology, preserving tissue architecture while profiling gene expression [17]. This technical guide delves into a landmark study that leveraged SRT to deconstruct the spatial architecture of the melanoma-microenvironment interface, providing a framework for how spatial biology can elucidate fundamental mechanisms of tumor organization [17] [19].

Core Discovery: The Tumor-Microenvironment Interface

Identification of a Spatially Distinct Region

The study employed the 10× Genomics Visium platform to analyze frozen sections from adult zebrafish with BRAFV600E-driven melanomas [17]. This model allowed for the analysis of the entire tumor and all surrounding tissues in a single transverse section. Unsupervised clustering of the SRT data from 7,281 array spots revealed a transcriptionally distinct cluster of spots localized exclusively to the border between the tumor and adjacent muscle tissue [17]. Despite being histologically indistinguishable from the surrounding muscle, this "interface" region possessed a unique transcriptional profile.

The correlation analysis of averaged transcriptomes showed that the interface cluster was more similar to the tumor (R = 0.33) than to the muscle (R = 0.06), indicating its unique nature [17]. This demonstrated that transcriptional specialization at the boundary is not evident from histology alone and requires spatial transcriptomic profiling.

Transcriptional Hallmarks of the Interface State

Differential gene expression analysis identified key markers upregulated in the interface relative to both the tumor core and the muscle microenvironment. These included:

• Cilia-related genes: A common set of cilia genes was upregulated at the interface.
• Translational and stress response genes: atf3 and eif3ea.
• Ribosomal genes: Indicating increased translational activity.
• Microtubule cytoskeleton genes: tuba1a and tuba1c [17].

The upregulation of these genes, particularly the cilia ensemble, pointed to a specialized biological program active only at the tumor-microenvironment boundary.

Protein-Level Validation and Conservation in Human Melanoma

Immunofluorescence validation confirmed the enrichment of cilia proteins specifically where the tumor contacts the microenvironment, corroborating the transcriptional findings [17]. Crucially, the study demonstrated that this cilia-enriched interface is conserved in human melanoma patient samples, suggesting it represents a fundamental feature of melanoma biology with potential translational relevance [17] [18].

Experimental Protocols and Methodologies

Integrated Spatial Transcriptomics Workflow

The study employed a multi-modal approach to comprehensively characterize the interface. The following diagram illustrates the integrated experimental workflow:

Detailed Methodological Breakdown

1. Sample Preparation and SRT Processing:

• Tissue Source: Frozen sections from large, invasive melanomas in adult zebrafish BRAFV600E model [17].
• SRT Technology: 10× Genomics Visium platform (6.5 mm² capture area, 55 µm spot diameter with 45 µm gaps) [17].
• Data Output: 7,281 barcoded array spots across three samples, profiling 17,317 unique genes [17].

2. Bioinformatic Analysis:

• Data Integration: Combined SRT, scRNA-seq, and snRNA-seq datasets using an "anchoring framework" to identify common cell states [17].
• Cluster Identification: Community-detection based clustering performed on the integrated dataset, yielding 13 distinct spatial clusters [17].
• Spatial Gene Expression Analysis: Computed mean expression of Gene Ontology (GO) terms and measured spatial coherence of pathways by comparing distances between high-expression spots versus a null distribution [17].
• Differential Expression: Identified interface-specific genes by comparing the interface cluster to both tumor and muscle clusters [17].

3. Validation Methods:

• Protein Validation: Immunofluorescence staining performed to validate enrichment of cilia proteins at the tumor boundary [17].
• Human Conservation Analysis: Investigated human patient samples to confirm the presence of the cilia-enriched interface [17].

Quantitative Findings and Data Analysis

Spatial Transcriptomics Data Metrics

Table 1: Summary of Spatially Resolved Transcriptomics Data Metrics

Metric	Sample A/B	Sample C	Overall Dataset
Number of Array Spots	Information missing	Information missing	7,281 spots [17]
Transcripts (UMIs) per Spot	~1,000-15,000 [17]	Fewer than A/B [17]	Information missing
Unique Genes per Spot	~500-3,000 [17]	Fewer than A/B [17]	Information missing
Unique Genes Detected	Information missing	Information missing	17,317 genes [17]
UMIs in Tumor Regions	Higher than microenvironment [17]	Information missing	Information missing

Key Transcriptional Signatures

Table 2: Key Upregulated Genes and Pathways in the Interface Region

Gene/Pathway Category	Specific Examples	Function/Putative Role in Interface
Cilia-Related Genes	Multiple identified genes	Cell signaling, sensing microenvironmental cues [17]
Translational/Stress Response	atf3, eif3ea	Cellular stress response, increased protein synthesis [17]
Ribosomal Genes	Multiple ribosomal proteins	Increased translational capacity [17]
Microtubule Cytoskeleton	tuba1a, tuba1c	Structural support for cilia, cell shape [17]
Spatially Organized Pathways (GO Terms)	Extracellular structure organization, Lipid import, IMP biosynthetic process [17]	Tumor-stroma co-adaptation, metabolic reprogramming [17]

Biological Mechanism: ETS-Factor Regulation of Cilia Genes

The study identified ETS-family transcription factors as key regulators of the interface state. These factors normally act to suppress cilia genes outside of the interface. At the tumor-microenvironment boundary, this suppression is alleviated, leading to the specific upregulation of cilia genes [17]. This represents a clear example of how spatial context can dictate transcriptional regulation in cancer cells. The following diagram illustrates this regulatory mechanism:

The Scientist's Toolkit: Essential Research Reagents and Solutions

Table 3: Key Research Reagents and Computational Tools for Spatial Transcriptomics

Reagent/Tool Category	Specific Examples	Function/Application
SRT Platform	10× Genomics Visium [17]	Capture probe-based spatial transcriptomics; preserves tissue architecture.
Sequencing Methods	scRNA-seq, snRNA-seq [17]	Characterize cellular heterogeneity at single-cell resolution.
Bioinformatic Tools	SPOTlight, Stereoscope [17]	Deconvolute SRT data to infer single-cell resolution.
Bioinformatic Tools	Anchoring framework [17]	Integrate multiple datasets (SRT, scRNA-seq) to identify common cell states.
Image Analysis Pipeline	MARQO [20]	Streamlines whole-slide, single-cell resolution analysis of multiplexed tissue images.
Nuclear Segmentation Tool	StarDist [20]	Performs AI-based nuclear segmentation for cell identification.
Validation Technique	Multiplex Immunofluorescence / Immunofluorescence [17] [20]	Protein-level validation of transcriptional findings.

Discussion and Implications for Spatial Cancer Research

This case study exemplifies how SRT technologies can move beyond cataloging cell types to reveal spatially organized functional states that are invisible to histology. The discovery of the "interface" state, with its distinct cilia-based biology, challenges the traditional binary view of tumors and their microenvironment, revealing instead a specialized zone of co-adapted cells [17].

From a therapeutic perspective, this interface represents a novel target for disrupting the tumor-stroma crosstalk essential for invasion and progression. The conservation of this state in human melanoma underscores its potential clinical relevance [17] [18]. For the field of spatial biology, this study provides a methodological blueprint for integrating multi-omic spatial data to uncover the architectural principles of tumor organization, a approach that is being extended through newer technologies like hyperplex immunofluorescence and advanced computational analysis [20] [19]. Future research will likely focus on targeting this interface state and exploring its existence and role in other cancer types.

The spatial organization of the tumor microenvironment (TME) profoundly influences cancer biology and therapy response [21]. In oral squamous cell carcinoma (OSCC), a defining feature of this organization is the distinct architectural and functional relationship between the tumor core (TC) and the leading edge (LE), also known as the invasive front. Traditional sequencing methods, which require tissue dissociation, lose the critical spatial context necessary to understand the functional compartmentalization of tumors [22]. The emergence of spatial transcriptomics (ST) has overcome this limitation, enabling the precise mapping of gene expression within the intact tissue architecture [23] [22]. This case study leverages ST to perform an integrative analysis of OSCC, framing the investigation within broader research on tumor spatial architecture to comprehensively characterize the conserved and tissue-specific transcriptional programs that define the TC and LE [21]. The findings provide pan-cancer insights into mechanisms of tumor progression and invasion, with direct implications for prognosis prediction and the development of novel targeted therapies.

Biological Findings: Distinct and Conserved TC and LE Architectures

Integrative single-cell and spatial transcriptomic analysis of HPV-negative OSCC has revealed that the TC and LE are not merely morphological regions but represent functionally specialized units with unique transcriptional profiles, cellular compositions, and cell-cell communication networks [21].

Unique Transcriptional Profiles and Cellular Compositions

Unsupervised clustering of malignant spots from ST data partitions the OSCC TME into three major clusters: a definitive TC, a definitive LE, and a transitory region that shares attributes of both [21].

Table 1: Key Characteristics of OSCC Tumor Core and Leading Edge

Feature	Tumor Core (TC)	Leading Edge (LE)
Key Marker Genes	CLDN4, SPRR1B, SPRR2 family genes (SPRR2D, SPRR2E, SPRR2A) [21]	LAMC2, ITGA5, COL1A1, FN1, COL1A2, TIMP1, COL6A2 [21]
Major Biological Pathways & Hallmarks	Keratinization, epithelial cell differentiation, antimicrobial and immune-related pathways [21]	Epithelial-mesenchymal transition (EMT), extracellular matrix (ECM) organization, angiogenesis, cell cycle [21]
Activated Signaling Pathways	MSP-RON in macrophages, IL-33, p38 MAPK signaling [21]	GP6, EIF2, HOTAIR regulatory pathways [21]
Prognostic Association	Gene signature associated with improved prognosis across multiple cancer types [21] [24]	Gene signature associated with worse clinical outcomes across multiple cancer types [21] [24]
Pan-Cancer Conservation	Tissue-specific transcriptional program [21]	Conserved transcriptional program across different cancer types [21]

The TC gene signature is associated with epithelial differentiation, characterized by high expression of genes involved in keratinization (e.g., SPRR2D, SPRR2E, SPRR2A) and inhibition of epithelial-mesenchymal transition (e.g., DEFB4A, LCN2) [21]. In contrast, the LE is enriched for genes driving ECM remodeling and a partial EMT (p-EMT) program, including COL1A1, FN1, and TIMP1 [21]. Pathway analysis predicts the activation of distinct canonical pathways: the LE shows activation of GP6, EIF2, and HOTAIR regulatory pathways, which are implicated in invasion and metastasis, while the TC activates pathways like MSP-RON and IL-33 signaling, suggesting a role in immune modulation [21].

A critical finding is the conservation of the LE gene signature across various cancer types, indicating common mechanisms underlying tumor invasion. Conversely, the TC transcriptional program appears to be more tissue-specific [21]. This conservation has direct clinical relevance, as the LE gene signature is associated with worse clinical outcomes, while the TC signature correlates with improved prognosis across multiple cancers [21] [24].

Ligand-Receptor Interactions and Cellular Neighborhoods

The cellular composition and interaction networks differ significantly between the TC and LE. Spatial deconvolution analysis identifies distinct cellular neighborhoods [21] [25]. The LE demonstrates a high density of cancer-associated fibroblasts (CAFs), with specific enrichment of ecm-MYCAFs (marked by LRRC15 and GJB2) and detox-iCAFs (marked by ADH1B and GPX3) [21]. These fibroblasts create a pro-invasive microenvironment through the deposition of ECM and paracrine signaling. The unique cellular compositions facilitate spatially organized ligand-receptor interactions that drive tumor progression. For instance, information flow from the TC to the LE is a key feature of the OSCC spatial architecture, and disrupting this communication has been identified as a potential therapeutic strategy [21].

Experimental Protocols and Methodologies

The characterization of TC and LE architectures relies on a combination of sophisticated ST technologies and advanced computational analyses.

Spatial Transcriptomics Wet-Lab Protocol

The following workflow details the key experimental steps for generating ST data, as applied in the featured OSCC study [21]:

Key Steps Explained:

• Sample Preparation: The protocol begins with fresh-frozen surgically resected OSCC samples sectioned into 10 μm thick slices and mounted on a specialized glass slide from the 10x Genomics Visium platform [21] [23].
• Histological Staining and Annotation: The tissue sections are stained with Hematoxylin and Eosin (H&E) and imaged. A pathologist then meticulously annotates morphological regions, such as the tumor core, leading edge, and stroma, which is crucial for correlating molecular data with tissue histology [21].
• Spatial Barcoding and Sequencing: The tissue is permeabilized to release mRNA transcripts, which bind to spatially barcoded capture probes on the slide. Each spot on the array (~55 μm in diameter, capturing ~5-50 cells) has a unique spatial barcode [21] [23]. The bound RNA is then reverse-transcribed into cDNA, which is used to construct a sequencing library.
• Data Generation: The libraries are sequenced on a high-throughput platform. In the featured study, this yielded transcriptomes from 24,876 spots across 12 samples, with a post-normalization average of 43,648 reads per spot [21].

Computational and Bioinformatic Analysis Workflow

The raw sequencing data undergoes a multi-step computational process to identify and characterize the TC and LE regions.

Table 2: Key Computational Methods for Spatial Data Analysis

Analytical Step	Method/Tool	Purpose and Application in OSCC Study
Data Preprocessing	10x Genomics Space Ranger, SCANPY [26]	Alignment, demultiplexing, generation of count matrices, normalization, and batch effect correction.
Malignant Cell Identification	Copy Number Variation (CNV) inference, Deconvolution	Stringent classification of malignant spots (CNV prob. >0.99 or deconvolution score >0.99) to separate tumor from non-malignant cells [21].
Spatial Domain Identification	Unsupervised Louvain Clustering, Graph Neural Networks (e.g., SpaGCN) [21] [26]	To identify spatially coherent clusters like TC, LE, and transitory regions without prior biological knowledge.
Differential Expression & Pathway Analysis	Differential Gene Expression Analysis (DGEA), Ingenuity Pathway Analysis (IPA)	To find marker genes for TC and LE and identify activated upstream regulators and canonical pathways [21].
Cell-Cell Communication	Ligand-Receptor Analysis Tools	To infer spatially-regulated ligand-receptor interactions between TC, LE, and stromal cells [21].
Developmental Trajectory	RNA Velocity, Pseudotime Analysis	To infer patterns of tumor cell differentiation and state transitions from TC to LE [21].

A pivotal step is the use of unsupervised Louvain clustering on the expression profiles of pre-identified malignant spots. This analysis reproducibly generates clusters corresponding to the TC and LE, which are then validated through differential expression of known markers (e.g., CLDN4 for TC; LAMC2 for LE) [21]. Artificial intelligence, particularly graph neural networks (GNNs), can further enhance this process by integrating gene expression data with spatial coordinates to achieve superior clustering accuracy and identify these spatial domains [23] [26].

The Scientist's Toolkit: Key Research Reagents and Solutions

Successfully executing a spatial transcriptomics study requires a suite of specialized reagents and platforms.

Table 3: Essential Research Reagents and Platforms for Spatial Transcriptomics

Item	Function and Role in TC/LE Analysis
10x Genomics Visium Platform	A widely adopted spatial barcoding platform for unbiased, whole-transcriptome capture from intact tissue sections. It was used in the foundational OSCC study to profile 24,876 spots [21] [23].
Fresh-Frozen Tissue Sections	The preferred sample type for full whole-transcriptome assays with Visium. Preserves RNA integrity better than FFPE for this application, though FFPE-compatible targeted panels are available [23].
Spatially Barcoded Capture Probes	Oligonucleotide probes fixed on the Visium slide that capture mRNA from the overlying tissue. Each probe's unique barcode links gene expression data to a specific spatial coordinate [21] [22].
H&E Staining Reagents	Enable histological visualization of the tissue section. Pathologist annotation of H&E images is critical for correlating molecular clusters (TC, LE) with tissue morphology [21].
Single-Cell RNA-Seq Reference Dataset	A publicly available scRNA-seq dataset (e.g., from HNSCC) used for deconvolution. It helps infer the cellular composition of each ST spot and stringently identify malignant cells [21].
AI/ML Clustering Tools (e.g., SpaGCN)	Graph convolutional network tools designed specifically for ST data. They integrate gene expression and spatial location to more accurately identify spatial domains like the TC and LE [26].

Clinical Implications and Therapeutic Opportunities

The distinct biology of the TC and LE presents unique opportunities for clinical intervention and biomarker development.

• Prognostic Biomarkers: The conserved LE gene signature is a powerful biomarker for worse clinical outcomes across multiple cancer types, while the TC signature is associated with improved prognosis [21] [24]. This suggests that spatial gene signatures could enhance patient risk stratification beyond current clinical and pathological criteria.
• Novel Therapeutic Targets: The study identified spatially-regulated patterns of cell development and ligand-receptor interactions. Using in silico modeling, the authors proposed that disrupting the information flow from the TC to the LE could be a viable therapeutic strategy and identified potential drugs to achieve this [21]. The activated pathways in the LE, such as those involving GP6 and HOTAIR, represent novel targets for inhibiting invasion and metastasis [21].
• Predictive Modeling for Drug Response: The spatially-defined transcriptional architectures are predictably associated with drug response. This enables the use of computational models to simulate how drugs might affect the different tumor compartments, paving the way for more effective and targeted treatment strategies [21] [25].

This case study demonstrates that the Tumor Core and Leading Edge of OSCC are not arbitrary anatomical regions but are fundamentally distinct functional units with conserved molecular architectures. Spatial transcriptomics has been instrumental in uncovering the unique transcriptional profiles, cellular ecosystems, and communication networks that define these compartments. The conserved, pro-invasive nature of the LE across cancer types highlights it as a critical target for therapeutic intervention. The integration of these spatial insights with artificial intelligence and in silico drug modeling holds exceptional promise for developing the next generation of spatially-informed, personalized cancer therapies. The interactive spatial atlases generated from this work serve as a foundational resource for the scientific community to further explore OSCC biology and develop novel targeted therapies [21] [24].

The tumor microenvironment (TME) represents a highly complex and dynamic ecosystem where malignant cells coexist with diverse immune populations, stromal components, and the extracellular matrix (ECM) within a precise spatial architecture. The organization of these elements is not random; rather, it follows distinct patterns that dictate disease progression and therapeutic response [27] [28]. Spatial transcriptomics has emerged as a groundbreaking technological frontier that bridges the critical gap between single-cell resolution and tissue context preservation, enabling researchers to quantify gene expression patterns directly within intact tissue sections while maintaining their native spatial coordinates [22]. This advanced approach has revolutionized our understanding of how biological pathways are spatially organized, particularly the intricate interplay between ECM remodeling and immune cell migration.

The significance of this spatial relationship is profound. The ECM, once considered merely a structural scaffold, is now recognized as a dynamic signaling hub that actively regulates immune cell behavior, influencing their activation, migration, and functional phenotypes [28]. Malignant cells exploit ECM remodeling to create immunosuppressive niches that facilitate immune evasion and tumor progression. Understanding these spatially organized pathways is therefore critical for developing novel therapeutic strategies that can overcome the physical and biochemical barriers imposed by the tumor ECM [28] [29]. This technical guide explores how spatial transcriptomics technologies are illuminating these complex interactions, with practical methodological guidance for researchers investigating the spatial architecture of tumor organization.

Technological Foundations of Spatial Transcriptomics

Spatial transcriptomics encompasses a suite of technologies that can be broadly categorized into three main approaches: imaging-based methods, sequencing-based methods, and laser capture microdissection (LCM)-based techniques [22]. Each offers distinct advantages and limitations for investigating ECM-immune interactions in the TME.

Imaging-based approaches, including in situ hybridization (ISH) and in situ sequencing (ISS), utilize fluorescently labeled probes to directly detect RNA transcripts within tissues, achieving subcellular resolution. Key methodologies include multiplexed error-robust fluorescence in situ hybridization (MERFISH), sequential FISH (seqFISH), and fluorescence in situ sequencing (FISSEQ) [22]. These technologies enable highly multiplexed gene expression analysis while preserving spatial context, making them ideal for mapping intricate cellular relationships at nanoscale resolution. However, they typically require pre-defined gene panels, limiting discovery potential for novel targets.

Sequencing-based approaches employ spatially barcoded oligonucleotide arrays to capture transcriptome-wide information from tissue sections. The 10x Genomics Visium platform is a prominent example that utilizes glass slides patterned with millions of spatially barcoded spots, each capturing mRNA from adjacent tissue areas [30] [31]. While offering whole transcriptome coverage, traditional implementations have resolution limitations (55-100 μm spot size), potentially capturing multiple cells per spot. Recent advancements like Slide-seq and High-Definition Spatial Transcriptomics (HDST) have dramatically improved resolution to near-single-cell level (approximately 10 μm) [27].

LCM-based approaches combine laser capture microdissection with RNA sequencing, enabling transcriptomic analysis of specific tissue regions identified by morphological criteria [22]. While providing regional specificity, these methods are lower throughput and result in destruction of tissue architecture during microdissection.

Table 1: Comparison of Major Spatial Transcriptomics Technologies

Technology	Resolution	Throughput	Key Advantages	Limitations
MERFISH/seqFISH	Subcellular (single RNA molecules)	Hundreds to thousands of genes	High multiplexing capability, single-cell resolution	Requires pre-defined gene panels
Visium (10x Genomics)	55-100 μm (multi-cell spots)	Whole transcriptome	Unbiased discovery, compatible with FFPE	Lower spatial resolution
Slide-seq/HDST	~10 μm (near single-cell)	Whole transcriptome	High resolution, discovery-based	Complex data analysis, lower RNA capture efficiency
LCM-seq	Cellular to regional	Targeted or transcriptome	Precise region selection	Destructive to tissue, lower throughput

Spatially Organized ECM-Immune Axis in Cancer

ECM Remodeling Creates Spatial Niches for Immune Evasion

The ECM undergoes dynamic remodeling in the TME through processes mediated by cancer-associated fibroblasts (CAFs), tumor-associated macrophages (TAMs), and malignant cells themselves. These alterations include changes in composition, stiffness, and architecture that collectively establish spatially distinct immune regulatory niches [28]. Spatial transcriptomics has been instrumental in decoding these patterns across various cancer types.

In lung adenocarcinoma (LUAD), spatial analysis has revealed that CAFs represent the most abundant non-malignant cell type, playing crucial roles in TME remodeling and prognosis determination [30]. Distinct histological subtypes display unique cellular composition profiles, with the micropapillary pattern exhibiting higher macrophage proportions and distinct gene expression pathways related to extracellular matrix organization and receptor tyrosine kinase signaling [30]. These spatially restricted gene modules create microenvironments conducive to tumor progression and immune evasion.

Clear cell renal cell carcinoma (ccRCC) research using cyclic immunohistochemistry (cycIHC) has demonstrated that the tumor periphery, particularly the pseudocapsule, exhibits homogeneous organization across the 3D scale but distinct cellular distribution gradients of T and B cells [29]. These immune patterns correspond specifically to deposited collagen types I and VI, suggesting an instructive role for ECM proteins in defining immune spatial organization [29].

ECM-Mediated Immune Cell Exclusion and Dysfunction

The ECM creates physical barriers that limit immune cell infiltration into tumor cores while simultaneously transmitting biochemical signals that alter immune cell function. Spatial transcriptomic analysis of myocardial infarction models treated with ECM hydrogels has revealed that ECM composition directly influences macrophage polarization states, with specific ECM components promoting pro-reparative macrophage phenotypes (Lyve1, Lgals3, Mrc1) versus pro-inflammatory states in control conditions [31]. This demonstrates the direct instructional capacity of ECM environments on immune cell differentiation and function.

In cervical cancer, the integration of single-cell RNA sequencing and spatial transcriptomics has enabled construction of a comprehensive spatial molecular atlas, identifying 38 distinct cellular neighborhoods with unique molecular characteristics [32]. These neighborhoods exhibit specialized immune compositions, with immunoglobulin-related genes (IGLC2, IGHG1, IGHG2) showing unique spatial expression characteristics restricted to specific microenvironments [32]. This spatial compartmentalization of immune function directly impacts therapeutic response.

Table 2: Key ECM Components and Their Spatial Immune Functions in Solid Tumors

ECM Component	Spatial Distribution	Immune Regulatory Functions	Therapeutic Implications
Collagen I & VI	Tumor periphery, pseudocapsule in ccRCC [29]	Instructs T and B cell distribution gradients [29]	Potential target for normalizing immune infiltration
MMP2	Upregulated in ECM hydrogel zones in MI models [31]	ECM remodeling facilitating immune cell migration	Combination therapy with immunotherapies
SPP1	ECM-rich regions in subacute MI [31]	Immune response modulation	Biomarker for immune-active zones
Fibronectin	Stromal regions in multiple cancers [28]	T cell dysfunction through integrin signaling	Target for overcoming T cell exclusion
Hyaluronic Acid	Desmoplastic regions in pancreatic and breast cancers [28]	Physical barrier to immune cell infiltration	Enzymatic degradation to improve drug delivery

Experimental Framework for Spatial Analysis of ECM-Immune Interactions

Sample Preparation and Spatial Library Construction

Robust spatial transcriptomics analysis begins with optimal sample preparation. For FFPE tissues, assess RNA quality by calculating DV200 values following extraction using kits such as Qiagen RNeasy FFPE [30]. Section tissues at 5μm thickness and mount on appropriate slides (e.g., Sigma-Aldrich Poly Prep Slides for Visium CytAssist) [30]. After drying overnight, incubate slides at 60°C for 2 hours, then perform deparaffinization according to established protocols (e.g., Visium CytAssist Spatial Gene Expression for FFPE — Deparaffinization, Decrosslinking, Immunofluorescence Staining & Imaging Protocol) [30].

Following deparaffinization, stain sections with hematoxylin and eosin and image at 20x magnification using a high-resolution slide scanner (e.g., Leica Aperio Versa8) [30]. For sequencing-based approaches like Visium, decrosslinking of H&E-stained sections should be conducted immediately after imaging. Subsequently, apply whole transcriptome probe panels to the tissue, allowing probe pairs to hybridize to their target genes and ligate to one another [30]. Transfer the slides to the spatial transcriptomics instrument (e.g., Visium CytAssist) for RNase treatment and permeabilization, enabling the ligated probes to hybridize to spatially barcoded oligonucleotides in the capture area [30]. Finally, construct spatial transcriptomics libraries from the probes for sequencing on appropriate platforms (e.g., Illumina NovaSeq 6000 system) [30].

Data Processing and Analytical Workflow

Process raw sequencing data using dedicated spatial analysis pipelines (e.g., Space Ranger pipelines version 2.0.0) [30], which performs tissue detection, fiducial detection, read alignment, and barcode/UMI counting against an appropriate reference genome (GRCh38 for human samples). Generate feature-spot matrices based on spatial barcodes for subsequent analysis with specialized R packages (e.g., Seurat V3.1.2) [30].

To normalize sequencing depth variance across spatial spots, particularly for technical artifacts and tissue anatomy, use the SCTransform function based on regularized negative binomial regression [30]. For multi-sample studies, integrate data from multiple spatial slides using reciprocal principal component analysis (RPCA) integration workflow to correct for potential batch effects [30]. Effectiveness of batch correction can be confirmed by ensuring spots do not primarily cluster by sample origin in UMAP projections.

Perform dimensionality reduction with principal component analysis (PCA), followed by shared nearest neighbor (SNN) construction based on Jaccard index between spots using the first 50 dimensions [30]. Cluster determination can be performed using the FindClusters function at resolution 0.6 by SNN modularity optimization [30]. The top 20 PCA dimensions are typically used for UMAP dimensional reduction, with clusters visualized in UMAP space using DimPlot and SpatialDimPlot functions [30].

Identify spatially variable features using the FindSpatiallyVariables function with the markvariogram method [30]. For cell type annotation, employ multi-step approaches including cell type deconvolution using specialized packages (e.g., SpaCET R package) [30], which utilizes reference single-cell RNA sequencing datasets to estimate the proportion of various cell types within each spatial spot. Each spot can then be assigned a dominant cell type based on the highest estimated proportion.

For further characterization of functional states and pathway enrichments within annotated spots, apply gene set variation analysis (GSVA) [30]. Calculate GSVA scores for each gene set per spot, allowing assessment of relative pathway activity within spatially defined regions and cell populations.

Figure 1: Experimental workflow for spatial analysis of ECM-immune interactions

Specialized Analytical Techniques for ECM-Immune Interactions

For investigating spatial ligand-receptor interactions, use computational tools like CellPhoneDB (version 3.1.0) with built-in databases for humans [30]. Input metadata and count matrix files, with p-values calculated using the proportion of means that exceeded the actual mean, ranked based on significance [30].

To analyze cellular differentiation states and plasticity, employ trajectory inference tools such as CytoTRACE (v.0.3.3), which uses transcriptional diversity as a proxy for developmental potential and assigns CytoTRACE scores to each cell [30]. Calculate these scores for each cluster independently using default parameters, identifying cell clusters with the lowest median CytoTRACE scores as potentially representing dedifferentiated states [30].

For deeper investigation of gene co-expression relationships, apply Weighted Gene Co-expression Network Analysis (WGCNA) to identify functional gene modules [32]. Construct gene adjacency matrices and topological overlap matrices, followed by hierarchical clustering approaches for co-expressed gene module identification. For specific cell type analysis, utilize hdWGCNA methodology with dynamic tree cutting techniques to identify functional gene modules, establishing minimum module sizes of 30 genes [32].

Visualization and Interpretation of Spatial Data

Mapping Multi-gene Spatial Expression Patterns

Construct continuous spatial expression maps using spatial interpolation algorithms to predict gene expression levels in unsampled regions [32]. Identify expression boundaries and transition zones by calculating spatial gradients of gene expression for key ECM and immune markers. Systematic analysis should include spatial expression distribution patterns for epithelial markers (MUC1, CDH1, KRT16), stromal markers (COL1A1, COMP, DCN), and immune markers (CD3G, FCGR1A) [32].

Spatial autocorrelation assessment should utilize methods that evaluate spatial clustering patterns of gene expression, with Moran's I index providing quantitative measures of spatial autocorrelation degrees [32]. This approach helps identify whether specific gene expression patterns are randomly distributed, clustered, or dispersed.

Advanced Computational Integration Methods

The integration of single-cell RNA sequencing with spatial transcriptomics has emerged as a powerful strategy for resolving the spatial and functional complexity of the TME [33]. Multimodal intersection analysis (MIA) can integrate scRNA-seq and ST data to map spatial cell-type relationships, as demonstrated in pancreatic ductal adenocarcinoma where stress-associated cancer cells were found to colocalize with inflammatory fibroblasts [33].

Emerging methods like SPIRAL enable integration and alignment of spatially resolved transcriptomics data across different experiments, conditions, and technologies [5]. Deep learning approaches such as MISO (multiscale integration of spatial omics with tumor morphology) can predict spatial transcriptomics from H&E-stained histological images, significantly outperforming competing methods in extensive benchmarks and enabling near single-cell-resolution, spatially-resolved gene expression prediction [5].

Figure 2: ECM-immune signaling pathway in tumor microenvironment

Research Reagent Solutions for Spatial Transcriptomics

Table 3: Essential Research Reagents for Spatial ECM-Immune Studies

Reagent/Technology	Function/Application	Key Features	Reference
CosMx Human Whole Transcriptome (WTX) Assay	Spatially resolved, single-cell transcriptomic and proteomic data	Subcellular resolution, wide tissue compatibility, AI-powered analysis tools	[3]
CellScape Precise Spatial Proteomics	High-plex spatial proteomics with multiomic integration	EpicIF technology for iterative staining cycles, customizable workflows	[3]
GeoMx Discovery Proteome Atlas (DPA)	1,100+ plex protein spatial profiling	Pairs with GeoMx Whole Transcriptome Atlas for same-section multiomics	[3]
nCounter ADC Development Panel	High-throughput molecular characterization	Robust performance with fragmented RNA, ideal for 3D tumor models	[3]
PaintScape Platform	In situ visualization of 3D genome architecture	Powered by jebFISH technology, maps chromatin folding in cancer	[3]
Visium CytAssist Spatial Gene Expression	Spatial transcriptomics from FFPE tissues	Compatible with archived samples, whole transcriptome coverage	[30]
10x Genomics Visium Platform	Capture-based spatial transcriptomics	Genome-wide expression profiling with spatial context	[32]

Spatial transcriptomics has fundamentally transformed our understanding of the spatially organized biological pathways connecting ECM remodeling to immune cell migration in the tumor microenvironment. The experimental frameworks and analytical workflows detailed in this technical guide provide researchers with comprehensive methodologies for investigating these critical interactions. As spatial technologies continue to evolve toward higher resolution and increased multiplexing capacity, and as computational methods for data integration become more sophisticated, we anticipate accelerated discovery of novel spatially-organized biomarkers and therapeutic targets. The convergence of spatial multi-omics with artificial intelligence approaches promises to unlock unprecedented insights into the spatial architecture of tumor organization, ultimately advancing precision oncology through spatially-informed diagnostic and therapeutic strategies.

A Researcher's Toolkit: Spatial Transcriptomics Technologies and Their Applications in Oncology

Spatial transcriptomics (ST) has emerged as a pivotal technology for studying tumor biology and its microenvironment by mapping gene expression data directly within the architectural context of intact tissue sections [34]. The loss of spatial information in conventional bulk and single-cell RNA sequencing represents a critical weakness in cancer research, where the functional organization of cells defines therapeutic responses and disease progression [35]. For researchers investigating tumor organization architecture, selecting between sequencing-based and imaging-based spatial methodologies represents a fundamental strategic decision with profound implications for data quality, biological insights, and resource allocation [36]. This technical guide provides a comprehensive comparison of these core methodologies, framing their capabilities within the specific context of tumor microenvironment research.

Core Technological Principles

Imaging-Based Spatial Transcriptomics

Imaging-based technologies utilize single-molecule fluorescence in situ hybridization (smFISH) as their foundational principle, enabling highly multiplexed detection of RNA transcripts through cyclic imaging processes [35]. These platforms differ primarily in their probe design, hybridization strategies, and signal amplification approaches, but share the common advantage of providing subcellular resolution, making them exceptionally valuable for dissecting cellular heterogeneity within complex tumor ecosystems [34] [36].

• Xenium: This hybrid technology combines in situ sequencing (ISS) and in situ hybridization (ISH) through a padlock probe system. An average of 8 gene-specific padlock probes hybridize to target RNA, undergo ligation to form circular DNA constructs, and are enzymatically amplified via rolling circle amplification (RCA). Fluorescently labeled oligonucleotides then bind to barcodes within these probes across multiple imaging rounds, generating unique optical signatures for each target gene [35].

• MERFISH: This platform employs a binary barcoding strategy where each gene is assigned a unique barcode of "0"s and "1"s. Thirty to fifty primary probes with "hangout tails" hybridize to target genes. Fluorescent secondary probes bind these tails across multiple imaging cycles, with fluorescence detection representing "1" and its absence representing "0" in the barcode sequence. This approach reduces optical crowding and incorporates error correction [35].

• CosMx SMI: This method incorporates both hybridization and optical signature approaches with an additional positional dimension. It uses pools of five gene-specific probes containing a target-binding domain and a readout domain with 16 sub-domains. Branched, fluorescently labeled secondary probes provide signal amplification, with 16 cycles of hybridization and imaging generating unique color-position combinations for each gene [35].

Sequencing-Based Spatial Transcriptomics

Sequencing-based technologies integrate spatially barcoded arrays with next-generation sequencing to determine transcript locations and abundance. These methods typically capture mRNA using polyT tails incorporated into spatially barcoded probes on arrays, with these spatial barcodes becoming incorporated into cDNA during reverse transcription [35]. The fundamental difference among platforms primarily lies in feature size, which determines spatial resolution.

• Visium and Visium HD: These platforms rely on spatially barcoded RNA-binding probes attached to slides, containing spatial barcodes, unique molecular identifiers (UMIs), and oligo-dT sequences for mRNA capture. The V2 workflow, suitable for FFPE tissues, uses adjacent probe pairs that hybridize to target mRNA and ligate before capture. Visium HD maintains the same core technology but reduces spot size to 2μm, significantly enhancing resolution compared to the standard 55μm spots [35].

• Stereo-seq: This technology utilizes DNA nanoball (DNB) patterning for RNA capture. Oligo probes containing barcoded sequences, coordinate identities (CIDs), molecular identifiers (MIDs), and poly(dT) are circularized and amplified via rolling circle amplification to form DNBs. These are loaded onto grid-patterned arrays, with DNBs of approximately 0.2μm diameter and 0.5μm center-to-center spacing, providing exceptionally high spatial density [35].

• GeoMx DSP: This platform employs a combination of barcoded probes and region-of-interest (ROI) selection rather than comprehensive spatial mapping. UV-cleavable oligonucleotide tags bound to RNA or protein targets are released from user-selected tissue regions, collected, and sequenced to quantify expression within morphologically defined areas [35].

Figure 1: Core workflow differences between imaging-based and sequencing-based spatial transcriptomics technologies. Imaging methods detect transcripts directly in tissue through cyclic fluorescence, while sequencing methods capture RNA onto barcoded arrays for subsequent sequencing and computational mapping.

Technical Comparison of Platform Performance

Resolution, Sensitivity, and Coverage Characteristics

The choice between sequencing-based and imaging-based technologies involves fundamental trade-offs between resolution, gene coverage, and practical considerations like cost and throughput [36]. These parameters directly influence the biological questions that can be effectively addressed in tumor research.

Table 1: Technical Parameter Comparison Between Major Spatial Transcriptomics Platforms

Platform	Technology Type	Spatial Resolution	Gene Coverage	Tissue Type Compatibility	Key Strengths
10X Visium	Sequencing-based	55μm spots (multi-cell)	Whole transcriptome	FFPE, Fresh Frozen	Unbiased discovery, standard workflows
10X Visium HD	Sequencing-based	2μm bins (single-cell)	Whole transcriptome	FFPE, Fresh Frozen	Single-cell resolution with full transcriptome
Stereo-seq	Sequencing-based	0.5μm center-to-center (subcellular)	Whole transcriptome	FFPE, Fresh Frozen	Ultra-high resolution, large tissue areas
Xenium	Imaging-based	Single-cell to subcellular	Targeted panels (300-500 genes)	FFPE, Fresh Frozen	High sensitivity, precise localization
MERFISH	Imaging-based	Single-cell to subcellular	Targeted panels (500-1,000 genes)	FFPE, Fresh Frozen	Low error rate, quantitative accuracy
CosMx SMI	Imaging-based	Single-cell to subcellular	Targeted panels (1,000-6,000 genes)	FFPE, Fresh Frozen	Large panel size, high plex capability
GeoMx DSP	Sequencing-based	ROI-based (cellular to regional)	Whole transcriptome or targeted	FFPE, Fresh Frozen	Morphology-guided selection, high plex RNA/protein

Performance Metrics in Tumor Samples

Recent benchmarking studies using formalin-fixed paraffin-embedded (FFPE) tumor samples provide critical performance comparisons directly relevant to cancer research. These evaluations reveal platform-specific characteristics in sensitivity, accuracy, and practical implementation.

Table 2: Experimental Performance Metrics from FFPE Tumor Tissue Evaluation [34]

Performance Metric	CosMx	MERFISH	Xenium (Unimodal)	Xenium (Multimodal)
Transcripts per Cell	Highest detection	Moderate to high (tissue age dependent)	Lower than imaging	Lowest detection
Unique Genes per Cell	Highest detection	Moderate (improved in newer tissues)	Lower than imaging	Lowest detection
Negative Control Performance	Some target genes expressed at control levels	Limited data (lacks negative controls)	Minimal target genes at control levels	Few target genes at control levels
Cell Segmentation Basis	Morphology-based	Morphology-based	Transcript-based	Multi-modal (transcript + morphology)
Tissue Coverage	Limited (545μm × 545μm FOVs)	Whole tissue	Whole tissue	Whole tissue

A 2025 systematic comparison using lung adenocarcinoma and pleural mesothelioma samples highlighted crucial performance differences. CosMx demonstrated the highest transcript and unique gene counts per cell, though it showed variability in target gene probe performance relative to negative controls, with some key markers for cell type annotation (e.g., CD3D, CD40LG, FOXP3) expressing similarly to negative controls in older tissue samples [34]. MERFISH performance was notably dependent on tissue age, with significantly better detection in newer FFPE samples, while Xenium showed more consistent performance across sample types but with lower overall sensitivity [34].

Methodological Implementation for Tumor Research

Experimental Design Considerations

Choosing between sequencing-based and imaging-based approaches depends primarily on the research objective: discovery versus validation [36]. Sequencing-based methods are ideal for unbiased exploration of tumor heterogeneity and microenvironment composition, while imaging-based approaches excel at validating spatial patterns of known markers at high resolution.

For sequencing-based approaches, Visium HD now enables true single-cell resolution across the whole transcriptome, making it suitable for comprehensive tumor atlas construction [35]. Stereo-seq offers even higher spatial density for capturing rare cell populations and subtle tumor microenvironments [35]. For imaging-based platforms, CosMx provides the largest targeted panels (up to 6,000 genes), enabling detailed characterization of specific cellular programs within tumor ecosystems [35].

Sample Preparation and Protocol Selection

FFPE tissues represent the standard for clinical cancer samples, and all major platforms now support FFPE compatibility [34]. However, tissue age and preservation quality significantly impact data quality, particularly for imaging-based methods [34]. For sequencing-based approaches, the Visium V2 workflow with CytAssist instrument simplifies the process by transferring probes from standard slides to Visium slides, optimizing handling of precious clinical samples [35].

Protocol duration and complexity differ substantially between approaches. Sequencing-based methods typically follow standardized library preparation pipelines that are more easily scalable for multiple samples [36]. Imaging-based experiments require specialized equipment, custom probe panels, and extended imaging times, increasing overall time and cost per sample [36].

Essential Research Reagents and Solutions

Table 3: Key Research Reagent Solutions for Spatial Transcriptomics

Reagent Category	Specific Examples	Function	Platform Applications
Gene Expression Panels	CosMx Human Universal Cell Characterization Panel (1,000-plex), MERFISH Immuno-Oncology Panel (500-plex), Xenium Human Lung Panel (289-plex + custom)	Targeted gene detection for cellular phenotyping	Imaging-based platforms (Xenium, MERFISH, CosMx)
Whole Transcriptome Kits	Visium HD Gene Expression Kit, Stereo-seq WTA Kit	Comprehensive transcriptome coverage	Sequencing-based platforms (Visium HD, Stereo-seq)
Multimodal Integration Reagents	CellScape EpicIF reagents, GeoMx Protein Panels	Combined RNA and protein detection	Multiomic platforms (CellScape, GeoMx)
Sample Preparation Kits	Visium FFPE Tissue Optimization Kit, Xenium FFPE Protocol reagents	Tissue preparation, permeabilization, and RNA accessibility	All platforms with FFPE support
Signal Amplification Systems	CosMx branched readout domains, Xenium RCA reagents	Enhanced detection sensitivity	Imaging-based platforms

Data Analysis and Visualization Approaches

Analytical Pipelines for Spatial Tumor Data

Spatial transcriptomics data analysis requires specialized computational approaches that integrate gene expression with spatial coordinates [37]. Common analytical tasks include dimensionality reduction, clustering, cell-type identification, and spatial pattern detection [37]. Popular frameworks like Seurat, Giotto, Scanpy, and Squidpy provide standardized workflows for these analyses [37] [9].

For sequencing-based data, analysis typically begins with Space Ranger for alignment, tissue detection, barcode counting, and feature-spot matrix generation [37]. Normalization methods like Scran or SCNorm address technical variability, followed by clustering using Louvain, Leiden, or other community detection algorithms [37]. Cell-type identification can be performed through projection to reference datasets (scmap, SingleR) or signature-based methods (Cell-ID) [37].

For imaging-based data, analytical workflows must account for cell segmentation challenges. As demonstrated in comparative studies, segmentation approach (transcript-based vs. morphology-based) significantly impacts cell calling and downstream analysis [34]. Methods like JSTA use deep learning for joint cell segmentation and type annotation in imaging data [37].

Visualization Strategies for Tumor Architecture

Effective visualization is crucial for interpreting spatial relationships within tumor ecosystems. Spaco (Spatial Palette Optimization) addresses the critical challenge of colorizing categorical spatial data by introducing a Degree of Interlacement (DOI) metric that models spatial relationships between cell types [9]. This ensures adjacent cell types receive maximally distinguishable colors, significantly enhancing visual interpretation in complex tumor microenvironments.

Figure 2: Spatial transcriptomics data analysis workflow with emphasis on visualization strategies. The Spaco method optimizes color assignment based on spatial relationships between cell types, enhancing interpretation of complex tumor microenvironments.

Standard visualization approaches include plotting cell positions as centroids or polygons, colored by metadata such as cell type or gene expression [38]. For exploring tumor microenvironments, highlighting specific cell types of interest while muting background cells can reveal spatial patterns of immune infiltration or stromal organization [38]. Neighborhood analysis techniques identify recurrent cellular communities within tumors, providing insights into microenvironmental organization [37].

Platform Selection Framework for Tumor Studies

Decision Framework Based on Research Objectives

The choice between sequencing-based and imaging-based spatial technologies should be guided by specific research questions, sample characteristics, and analytical requirements [36]. This decision framework provides guidance for selecting optimal approaches based on common scenarios in tumor biology research.

Choose sequencing-based technologies when:

• Conducting discovery-phase research to identify novel biomarkers or cell states
• Requiring whole transcriptome coverage without prior knowledge of key genes
• Studying heterogeneous tumors with undefined cellular composition
• Processing multiple samples in parallel for cohort studies
• Integrating with existing single-cell RNA-seq datasets

Choose imaging-based technologies when:

• Validating spatial patterns of previously identified gene signatures
• Requiring single-cell or subcellular resolution for precise localization
• Studying spatial organization of known cell types in tumor microenvironments
• Working with limited tissue samples where maximizing information from small areas is critical
• Combining RNA detection with protein markers in multiomic assays

Integrated Approaches for Comprehensive Tumor Characterization

For the most complete understanding of tumor architecture, combined approaches leveraging both sequencing-based and imaging-based methods often provide superior insights [36]. Sequencing-based spatial transcriptomics can identify novel gene signatures and cellular heterogeneity across entire tissue sections, while follow-up imaging-based validation confirms spatial localization at high resolution [36].

Additionally, integrating spatial data with single-cell RNA sequencing helps resolve mixed cellular signals in sequencing-based spatial data and informs panel design for imaging-based approaches [36]. This integrated framework enables both discovery and validation within the same research program, leveraging the complementary strengths of both technological approaches.

The complex spatial organization of cells within a tumor is a critical determinant of cancer progression, therapeutic response, and patient outcome. Traditional sequencing methods, which require tissue dissociation, irrevocably lose this architectural context. Spatial transcriptomics (ST) has emerged as a transformative technology that enables the mapping of gene expression data within the intact tissue landscape, preserving the precise spatial relationships between malignant, immune, and stromal cells [39]. This capability is particularly vital for immuno-oncology research, where the cellular composition and organization of the tumor microenvironment (TME) directly influence immune evasion and therapy efficacy [34]. By integrating deep transcriptome profiling with histological imaging, ST technologies allow researchers to visualize the functional interactions and heterogeneity that define cancer ecosystems. This technical guide provides an in-depth comparison of four major spatial platforms—10x Visium, Slide-seq, Stereo-seq, and GeoMx DSP—framed within the context of investigating tumor organization and architecture. We detail their core methodologies, present comparative performance data, and outline experimental protocols to inform platform selection for cancer research.

Spatial transcriptomics technologies can be broadly categorized into two groups: sequencing-based and imaging-based methods [35]. Sequencing-based technologies (like Visium, Slide-seq, and Stereo-seq) capture mRNA using spatially barcoded probes on a surface, followed by library preparation and next-generation sequencing (NGS) to decode the spatial origin and identity of each transcript. In contrast, imaging-based technologies (a category that includes GeoMx DSP's readout, though it differs in profiling approach) utilize in situ hybridization or sequencing to directly visualize RNA molecules within tissue sections through iterative cycles of fluorescent probing and imaging [35] [39].

The table below provides a quantitative comparison of the core technical parameters for the platforms covered in this guide.

Table 1: Core Technical Specifications of Spatial Transcriptomics Platforms

Platform	Core Technology	Spatial Resolution	Key Strength	Tissue Compatibility	Species Compatibility	Multimodal Capability
10x Visium HD	Sequencing-based (spatially barcoded 2µm spots) [40]	2 µm spot size (single-cell scale) [40]	Balanced resolution & whole transcriptome coverage [40]	FFPE, Fresh Frozen, Fixed Frozen [40]	Human, Mouse (HD WT Panel); Agnostic (HD 3' Gene Expression) [40]	Gene Expression, Protein (IF), Morphology (H&E) [40]
Stereo-seq	Sequencing-based (DNA Nanoball array) [35]	500 nm center-to-center (subcellular) [41] [35]	Extremely high resolution & massive FOV (up to 13cm x 13cm) [41]	Fresh Frozen [35]	Agnostic [35]	Information not available
GeoMx DSP	Sequencing-based (UV-cleavable barcoded probes from ROI) [42]	10 µm (region-of-interest guided) [42]	Flexible, biology-driven profiling of predefined regions; high-plex RNA + protein from same section [42] [43]	FFPE, Fresh Frozen [42]	Customizable via spike-in probes [42]	Same-section RNA + Protein (1,100-plex protein, 18,000-plex RNA) [43]

Platform-Specific Workflows and Tumor Biology Applications

10x Visium HD

Workflow: The Visium HD assay for FFPE tissue begins with tissue sectioning onto a specialized glass slide containing millions of spatially barcoded 2 µm x 2 µm spots [40]. The workflow requires the CytAssist instrument to transfer gene-specific probes from a standard glass slide onto the Visium slide, optimizing mRNA capture from potentially degraded FFPE RNA [40] [35]. After probe hybridization and ligation, the probe complexes are released, and the library is constructed for sequencing. Bioinformatic analysis then maps the sequenced reads back to their spatial coordinates using the barcodes [35].

Application in Tumor Research: Visium HD's single-cell-scale resolution is ideal for mapping intratumoral heterogeneity and delineating distinct cellular neighborhoods within the TME. Its whole transcriptome coverage supports unsupervised discovery of novel gene expression signatures directly from the spatial context of the tumor [40].

Stereo-seq

Workflow: Stereo-seq utilizes DNA Nanoball (DNB) technology. Synthesized oligo probes containing spatial coordinate barcodes are circularized and amplified via rolling circle amplification (RCA) to form DNBs [35]. These DNBs are then patterned onto a chip to create the capture array. With a DNB diameter of 220 nm and a center-to-center distance of 500 nm, this array offers nanoscale resolution [41] [35]. mRNA from fresh frozen tissue sections is captured by the poly(dT) sequences on the DNBs, followed by on-slide cDNA synthesis, library preparation, and sequencing [35].

Application in Tumor Research: Stereo-seq's combination of subcellular resolution and a massive field of view (up to 1 cm x 1 cm standard, customizable up to 13 cm x 13 cm) is uniquely powerful for pan-cancer atlas projects and studying rare tumor populations or metastatic niches across large tissue areas without the need for tiling [41].

GeoMx Digital Spatial Profiler (DSP)

Workflow: GeoMx DSP employs a fundamentally different, region-of-interest (ROI) driven approach. Tissue sections are stained with fluorescent morphology markers (e.g., for tumor, immune cell compartments) and oligonucleotide-tagged probes for RNA (and/or protein) [42]. After imaging, the user selects ROIs based on the tissue morphology. The instrument then uses a digital micromirror device to project UV light onto the selected ROIs, photocleaving and releasing the oligonucleotide barcodes for collection [42]. These barcodes are quantified via NGS or the nCounter system to determine analyte abundance in each specific ROI.

Application in Tumor Research: GeoMx DSP is exceptionally well-suited for hypothesis-driven spatial biology. It allows researchers to quantitatively compare gene expression profiles between specific, clinically relevant compartments—for instance, comparing the immune infiltrate in the tumor core versus the invasive margin, or profiling regions with high versus low PD-L1 protein expression [42] [34]. Its high-plex, same-section multiomic capability is a key asset for comprehensive biomarker discovery.

Diagram 1: GeoMx DSP workflow for spatially resolved omics.

Experimental Protocol for Tumor Analysis

This section outlines a generalized protocol for a spatial transcriptomic study of FFPE tumor tissue, integrating steps common to platforms like Visium HD and GeoMx DSP.

A. Sample Preparation and Sectioning

• Tissue Source: Use human or mouse FFPE tumor tissue blocks. Tissue Microarrays (TMAs) can be used for high-throughput cohort studies on GeoMx DSP and Visium [3] [34].
• Sectioning: Cut serial sections of 5-10 µm thickness using a microtome.
• Mounting: For Visium HD, mount sections onto the specific Visium HD slide using the CytAssist instrument for probe transfer [40]. For GeoMx DSP, mount on standard glass slides.

B. On-Slide Assay

• Deparaffinization and Rehydration: Standard xylene and ethanol series.
• Antigen Retrieval and Permeabilization: Use target retrieval solutions and proteases to expose RNA and/or epitopes.
• Probe Hybridization:
  • Visium HD: Gene-specific probes are hybridized to the tissue and then transferred to the slide's capture oligos via CytAssist-mediated ligation [40].
  • GeoMx DSP: Incubate tissue with the chosen RNA (Whole Transcriptome Atlas) and/or protein (Discovery Proteome Atlas) probe sets, along with fluorescent morphology markers [42] [43].
• Washing: Stringent washes to remove unbound probes.

C. Imaging and Profile Generation

• Imaging: Acquire high-resolution images of the entire tissue section using the platform's imaging system. For GeoMx DSP, this image is used to select ROIs.
• Spatial Barcode Collection:
  • Visium HD: Synthesize cDNA, release the library from the slide, and prepare sequencing libraries.
  • GeoMx DSP: Select ROIs and segments based on fluorescent morphology. UV light is used to cleave and collect oligonucleotide tags from the selected areas [42].
• Sequencing and Data Generation: Prepare sequencing libraries from the collected material and run on an NGS platform. The final output is a digital count matrix of gene expression (and protein abundance for GeoMx multiomics) linked to spatial coordinates or ROIs.

The Scientist's Toolkit: Essential Research Reagents and Materials

Successful spatial transcriptomics experiments rely on a suite of specialized reagents and instruments. The table below lists key components for setting up a spatial biology workflow in a cancer research lab.

Table 2: Key Research Reagent Solutions for Spatial Transcriptomics

Item	Function	Example Kits/Assays
Spatial Expression Slide	Solid support with spatially barcoded oligos for mRNA capture.	10x Visium HD Slide (2 Capture Areas) [40]
Gene Expression Panel	Probe sets designed to profile transcriptome or targeted gene panels.	Visium HD HD WT Panel (Human/Mouse), CosMx Human Whole Transcriptome Panel, GeoMx Whole Transcriptome Atlas [40] [43] [34]
Multiomics Panel	Antibody-based panels for simultaneous protein detection.	GeoMx Discovery Proteome Atlas (1,200+ proteins), CosMx Multiomics (76 proteins with RNA) [43]
Morphology Markers	Fluorescent antibodies or dyes to visualize tissue and cell structures for ROI selection.	Pan-cytokeratin (tumor), CD45 (immune), SYTO13 (nuclei) [42]
Library Prep Kit	Reagents for constructing sequencing libraries from spatially barcoded cDNA or oligos.	Illumina-Compatible Library Kit (platform-specific)
Data Analysis Suite	Software for processing, visualizing, and analyzing spatial data.	10x Loupe Browser, Bruker DSPDA, STUtility [42]

The choice of an optimal spatial transcriptomics platform is dictated by the specific research question in tumor biology. 10x Visium HD offers a robust, discovery-oriented solution with whole transcriptome coverage at a resolution suitable for analyzing cellular neighborhoods. Stereo-seq pushes the boundaries of resolution and scale, making it ideal for constructing detailed atlases and studying rare cellular events across vast tissue landscapes. GeoMx DSP provides unparalleled flexibility for targeted, hypothesis-driven research, enabling direct, quantitative comparison of predefined tissue compartments and integrated multiomic profiling from the same section. As these technologies continue to mature, their integration with advanced computational methods, such as deep learning models that predict gene expression from routine histology slides [5], promises to further democratize and enhance our ability to decode the complex architecture of cancer.

Imaging-based spatial transcriptomics (iST) has emerged as a pivotal technology for studying tumor biology and associated microenvironments by characterizing gene expression profiles within their native histological context [34]. These platforms preserve the spatial architecture of tissues while enabling single-cell or subcellular resolution mapping of RNA molecules, providing unprecedented insights into cellular states and interactions within complex tissues [44]. The ability to study the "whole panorama of cellular and molecular interactions in tissues accurately and within their functional context is vital for understanding health and disease" [34], particularly in cancer research where tumor development and accompanying immune responses depend on the location of different cell-type populations and tissue organization [45].

Among commercially available iST platforms, CosMx (NanoString), Xenium (10x Genomics), and MERFISH (Vizgen) have gained significant traction, each employing variations of fluorescence in situ hybridization (FISH) with distinct chemical approaches, probe designs, and signal amplification strategies [44]. These technologies differ fundamentally in their sample preparation protocols, amplification methods, gene selection for panel design, and cell-segmentation processes [34]. Understanding their comparative strengths and limitations is essential for researchers designing studies involving precious tumor samples, especially in translational oncology research utilizing formalin-fixed paraffin-embedded (FFPE) tissues, which represent the current standard for sample processing and archiving in pathology [34] [44].

Core Technological Specifications

Table 1: Core Technical Specifications of Major iST Platforms

Feature	CosMx	Xenium	MERFISH
Primary Technology	Branch chain hybridization amplification [44]	Padlock probes with rolling circle amplification [44] [46]	Direct probe hybridization with transcript tiling [47] [44]
Spatial Resolution	Subcellular [34]	Subcellular [48] [46]	Subcellular, nanometer precision [47]
Gene Panel Size	1,000-plex (standard panel) [34], 6K panel available [49]	289-392 genes (customizable) [34], up to 5,000 genes [49]	500-plex (standard panel) [34], customizable [47]
Sample Compatibility	FFPE, fresh frozen [34]	FFPE, fresh frozen [48] [46]	FFPE, fresh frozen [47]
Cell Segmentation	Manufacturer's algorithm + CellPose [50]	Uni/multi-modal segmentation [34], DAPI-based with expansion [46]	Cell boundary staining with manufacturer's algorithm [51]
Key Differentiator	Largest standard panel size [34]	High sensitivity and specificity [49] [46]	Single-molecule resolution with error-robust barcoding [47] [51]

Technology Workflow Comparison

Performance Benchmarking in Tumor Research

Detection Sensitivity and Specificity

Systematic benchmarking studies using controlled experimental conditions with serial sections of tumor tissues provide critical insights into platform performance characteristics. A comprehensive evaluation using colon adenocarcinoma, hepatocellular carcinoma, and ovarian cancer samples revealed distinct detection patterns across platforms [49]. Xenium 5K demonstrated superior sensitivity for multiple marker genes including the epithelial cell marker EPCAM, which showed well-defined spatial patterns consistent with H&E staining and Pan-Cytokeratin immunostaining on adjacent sections [49]. In comparative analyses, Stereo-seq v1.3, Visium HD FFPE, and Xenium 5K showed high correlations with single-cell RNA sequencing (scRNA-seq) data, while CosMx 6K detected a higher total number of transcripts than Xenium 5K but showed substantial deviation from matched scRNA-seq references [49].

Specificity assessments using metrics such as Negative Co-expression Purity (NCP) reveal important distinctions between platforms. NCP quantifies the percentage of non-co-expressed genes in reference single-cell datasets that do not appear to be co-expressed in each spatial transcriptomics dataset, with values closer to 1 indicating higher specificity [46]. In such analyses, Xenium demonstrates consistently higher specificity than CosMx, which presents the lowest values among commercial platforms [46]. MERFISH quantitatively reproduces bulk RNA-seq and scRNA-seq results with improvements in overall dropout rates and sensitivity compared to sequencing-based methods [51].

Technical Performance Metrics

Table 2: Performance Benchmarking Across iST Platforms Using Tumor Samples

Performance Metric	CosMx	Xenium	MERFISH
Transcripts/Cell	Highest in TMAs (p < 2.2e−16) [34]	186.6 reads/cell (average) [46]	Lower in older tissues, improves in newer samples [34]
Unique Genes/Cell	Highest among platforms (p < 2.2e−16) [34]	Varies by segmentation mode [34]	Dependent on tissue quality and age [34]
Sensitivity	High total transcripts but lower correlation with scRNA-seq [49]	Superior sensitivity for marker genes [49], 1.2-1.5× higher than scRNA-seq [46]	Improved dropout rates vs. sequencing methods [51]
Specificity (NCP)	Lowest among commercial platforms [46]	High (>0.8), slightly lower than other platforms [46]	High specificity in gene detection [51]
Tissue Age Compatibility	Detected target genes expressed same as negative controls in older tissues [34]	Consistent performance across tissue ages [34]	Performance decreases in older tissues [34]
Cell Segmentation Accuracy	Requires filtering (30 transcripts/cell) [34]	76.8% reads assigned to cells [46], multimodal segmentation available [34]	Relies on cell boundary staining [51]

Experimental Design Considerations for Tumor Studies

The selection of appropriate iST platforms for tumor microenvironment studies requires careful consideration of several experimental parameters. Tissue quality and age significantly impact data quality, particularly for MERFISH and CosMx platforms. In comparative studies using lung adenocarcinoma and pleural mesothelioma samples, CosMx displayed multiple target gene probes that expressed at the same level as negative control probes across all tissue microarrays (TMAs), with this effect more pronounced in older tissue samples (19.6% in MESO1 and 31.9% in MESO2) [34]. These affected genes included important cell type annotation markers such as CD3D, CD40LG, FOXP3, MS4A1, and MYH11 [34].

Panel design represents another critical consideration, as platforms offer different degrees of customizability. CosMx provides a standard 1,000-plex panel with optional add-on genes, Xenium offers either fully customizable panels or standard panels with optional add-ons, and MERFISH provides similar customizability options [44]. For tumor immunology applications, researchers must carefully select panels that encompass relevant immune, stromal, and malignant cell markers appropriate for their cancer type.

Cell segmentation approaches vary significantly between platforms and impact downstream analysis. Xenium utilizes both unimodal (Xenium-UM) and multimodal (Xenium-MM) segmentation, with unimodal assays demonstrating higher transcript and gene counts per cell than multimodal assays (p < 2.2e−16) [34]. CosMx requires filtering of cells with fewer than 30 transcript counts and those five times larger than the geometric mean of cell area sizes [34], while MERFISH relies on cell boundary staining in conjunction with nuclear markers for segmentation [51].

Tumor Architecture Applications and Biological Insights

Mapping Tumor Microenvironment Organization

Spatial transcriptomics platforms have enabled unprecedented insights into tumor organization architecture. In high-grade serous ovarian cancer (HGSC), comprehensive mapping of over 2.5 million cells from 130 tumors revealed a fundamental macro-organization principle where "malignant cells and fibroblasts form spatially distinct compartments (which we refer to as the malignant and stromal compartments), such that T/NK cells preferentially localized in the stromal rather than the malignant compartment (P < 1 × 10−4)" [45]. This organization pattern was consistently observed across patients and validated in multiple datasets, demonstrating how spatial biology influences immune cell infiltration patterns in tumor ecosystems.

In vulvar high-grade squamous intraepithelial lesions (vHSIL) studied in relation to immunotherapy response, CosMx analysis of 20 pre-treatment lesions identified 18 cell clusters and 99 distinct non-epithelial cell states from over 274,000 single cells mapped in situ [50]. This deep profiling revealed that complete responders to immunotherapy exhibited "a higher ratio of immune-supportive to immune-suppressive cells—a pattern mirrored in other solid tumors following neoadjuvant checkpoint blockade" [50]. Key immune populations enriched in complete responders included CD4+CD161+ effector T cells and chemotactic CD4+ and CD8+ T cells, while partial responders showed increased proportions of T helper 2 cells and CCL18-expressing macrophages [50].

Analytical Workflows for Tumor Architecture

Essential Research Reagents and Experimental Protocols

Research Reagent Solutions for iST Experiments

Table 3: Essential Research Reagents for Spatial Transcriptomics Workflows

Reagent Category	Specific Examples	Function in Workflow
Gene Expression Panels	CosMx Human Universal Cell Characterization Panel (1,000-plex) [34], Xenium human lung panel (289-plex + custom genes) [34], MERFISH Immuno-Oncology Panel (500-plex) [34]	Targeted transcript detection with cell type resolution
Sample Preparation Kits	FFPE tissue preparation kits [34] [44], Fresh frozen tissue preservation solutions [48] [47]	Tissue preservation and processing for optimal RNA integrity
Cell Segmentation Reagents	DAPI nuclear stain [46], Cell boundary markers [51], Antibodies for multimodal segmentation [34]	Cellular compartment identification and boundary definition
Signal Amplification Systems	Branch chain amplification reagents (CosMx) [44], Rolling circle amplification kit (Xenium) [44] [46], Readout probe amplifiers (MERFISH) [47]	Signal enhancement for transcript detection
Validation Tools	Multiplex immunofluorescence panels [34], RNAscope assays [49], CODEX protein profiling [49]	Orthogonal validation of spatial findings

Detailed Methodological Protocols

For researchers implementing iST technologies in tumor architecture studies, several key methodological considerations emerge from benchmarking studies:

Sample Preparation Protocol: For FFPE tissues, which represent the standard in clinical pathology, sectioning at 5μm thickness provides optimal results across platforms [34]. Tissue quality assessment should include H&E staining evaluation, and when possible, RNA integrity measurement (DV200 > 60% is recommended for MERFISH) [44]. For studies involving archival tissues, note that "the more recently constructed MESO TMAs had higher numbers of transcripts and uniquely expressed genes per cell with CosMx and MERFISH than Xenium" [34], indicating that tissue age impacts performance differently across platforms.

Quality Control and Data Processing: Implement platform-specific quality thresholds, such as filtering cells with fewer than 30 transcript counts for CosMx and fewer than 10 transcripts for MERFISH and Xenium [34]. Carefully evaluate negative control probes, as some platforms exhibit target gene probes expressing at similar levels to negative controls, particularly in older tissues [34]. For cell segmentation, consider using improved algorithms like CellPose, which has been shown to be one of the most reliable methods across platforms [50].

Multi-platform Integration: When integrating iST data with complementary modalities, leverage established workflows such as the "contamination ratio metric" for pre-emptively excluding genes likely to return spurious results due to imperfect cell segmentation [50]. For cell type annotation, semi-supervised methods like InSituType can effectively classify cells using immuno-oncology-based reference profiles while allowing for unsupervised clustering to characterize novel cell states [50].

The rapid evolution of high-throughput subcellular resolution spatial transcriptomics platforms has fundamentally transformed our ability to decipher tumor architecture. CosMx, Xenium, and MERFISH each offer distinct advantages—CosMx with its large standard panel size, Xenium with its high sensitivity and robust performance across tissue ages, and MERFISH with its single-molecule resolution and error-robust barcoding [34] [49] [46]. Systematic benchmarking reveals that platform selection must be guided by specific research questions, tissue characteristics, and analytical requirements rather than assuming universal superiority of any single technology [34] [49] [44].

For tumor biology applications, these technologies have enabled the discovery of fundamental organization principles of tumor microenvironments, including spatially distinct compartments that orchestrate immune cell infiltration [45] and cellular ecosystems that determine immunotherapy responses [50]. As the field advances, increasing gene panel sizes, improving segmentation algorithms, and enhancing multi-omic integration will further empower researchers to unravel the spatial complexities of cancer. The continued benchmarking and methodological refinement of these platforms will ensure that spatial transcriptomics realizes its potential to revolutionize both basic cancer biology and translational drug development.

Spatial transcriptomics (ST) has emerged as a transformative technology in cancer research, enabling the precise quantification and visualization of gene expression within the intact spatial context of tumor tissues. Unlike conventional bulk or single-cell RNA sequencing that lose spatial organization, ST technologies preserve the architectural relationships between cells, providing critical insights into the tumor microenvironment (TME), cellular heterogeneity, and molecular interactions that drive cancer progression [52]. The spatial context of cellular interactions is particularly crucial in oncology, where tumor heterogeneity and immune microenvironment composition serve as critical components of oncologic disease progression and treatment response [52].

The evolution of ST technologies from early in situ hybridization methods to current high-plex spatial barcoding and imaging platforms has fundamentally expanded our investigative capabilities in tumor biology. These advances allow researchers to move beyond mere cataloging of cellular components toward understanding functional organization within tumors—how cellular positioning influences signaling networks, metabolic cooperation, and therapeutic vulnerability [52] [53]. This technical guide examines the key applications of ST in mapping tumor architecture, profiling immune responses, and identifying novel therapeutic targets, providing both methodological frameworks and practical considerations for implementation in cancer research.

Creating Comprehensive Spatial Maps of Tumor Ecosystems

Technological Platforms for Spatial Mapping

Spatial mapping of tumors requires platforms that balance resolution, multiplexing capability, and tissue compatibility. The selection of an appropriate technology depends on specific research objectives, whether focused on transcriptome-wide discovery or targeted high-plex validation.

Table 1: Comparison of Spatial Transcriptomics Platforms for Tumor Mapping

Platform	Methodology	Resolution	Maximum Targets	Sample Types	Best Applications in Cancer Research
10x Genomics Visium	Spatial barcoding with sequencing	55 μm (single-cell with HD)	All 3' mRNA	FFPE, Fresh frozen	Tumor heterogeneity, spatial domains [52]
NanoString CosMx	In situ hybridization	Subcellular	18,000+ RNAs	FFPE, Fresh frozen	Single-cell spatial phenotyping, rare cell detection [3] [52]
10x Genomics Xenium	Padlock probe with rolling circle amplification	Subcellular	5,000 RNAs	FFPE, Fresh frozen	High-plex targeted imaging, tumor microenvironments [52]
GeoMx Digital Spatial Profiler	UV-cleavable oligo tags	Single-cell to multicellular regions	18,000+ RNAs (Whole Transcriptome Atlas)	FFPE, Fresh frozen	Region-specific profiling, immune oncology [3] [52]
CellScape	Iterative staining/bleaching cycles	Single-cell	30+ proteins	FFPE on coverslips	Spatial proteomics, immune cell tracking [3]
Akoya PhenoCycler	Cyclic immunofluorescence	Single-cell	~100 proteins	FFPE, Fresh frozen	Multiplexed tissue imaging, immune contexture [52]

Experimental Workflow for Spatial Tumor Mapping

The standard workflow for creating spatial maps of tumor architecture involves coordinated wet-lab and computational steps:

Tissue Preparation and Processing:

• Collect fresh frozen or FFPE tumor tissues sectioned at 4-10 μm thickness [52]
• For FFPE samples, perform deparaffinization, hematoxylin and eosin (H&E) staining, and imaging prior to ST processing
• Optimize permeabilization conditions to ensure optimal mRNA capture efficiency while preserving tissue morphology

Spatial Library Preparation and Sequencing:

• For sequencing-based platforms (Visium, Xenium): Implement spatial barcoding, cDNA synthesis, and library construction followed by next-generation sequencing [52]
• For imaging-based platforms (CosMx, MERFISH): Perform sequential hybridization and imaging cycles to decode spatial RNA positions [54] [52]
• Incorporate quality control measures including RNA quality assessment (RIN >7 for fresh frozen) and control gene validation

Data Processing and Integration:

• Align sequencing reads to reference genomes and assign spatial barcodes to generate gene-spot matrices
• Implement computational alignment tools (STalign, PASTE) for integrating multiple tissue sections and constructing 3D tumor architectures [55]
• Apply spatial clustering algorithms (BayesSpace, GraphST) to identify histopathological domains and spatial expression patterns [56] [55]

Figure 1: Experimental workflow for creating spatial maps of tumor architecture, integrating both sequencing-based and imaging-based spatial transcriptomics platforms.

Computational Analysis and 3D Reconstruction

Advanced computational methods are essential for transforming raw spatial data into biologically meaningful tumor maps:

Spatial Deconvolution: Apply algorithms (Cell2location, STRIDE, SPOTlight) to infer cell-type compositions within capture spots, leveraging single-cell RNA-seq references to resolve cellular heterogeneity beyond platform resolution limits [56]. These methods use probabilistic modeling, non-negative matrix factorization, or deep learning to estimate the proportion of different cell types in each spatial location.

Multi-Slice Alignment and 3D Reconstruction: Implement tools (PASTE, STalign, SPIRAL) to align consecutive tissue sections and reconstruct three-dimensional tumor architecture [55]. These methods employ optimal transport theory, image registration, or graph-based matching to create cohesive spatial models across multiple tissue layers, preserving spatial relationships across the z-axis.

Spatial Domain Identification: Utilize clustering algorithms (BayesSpace, GraphST) that incorporate spatial neighborhood information to identify histologically and molecularly distinct tumor regions, immune niches, and stromal compartments [56] [55]. These domains often correlate with functional specializations, such as proliferative centers, invasive margins, and immunosuppressive niches.

Immune Profiling within the Spatial Context

Mapping the Tumor Immune Microenvironment

Spatial transcriptomics enables comprehensive profiling of immune cell distribution, functional states, and interactions within the tumor ecosystem. This spatial context is critical for understanding immune evasion mechanisms and predicting immunotherapy responses.

Table 2: Spatial Immune Profiling Applications in Cancer Research

Application	Methodology	Key Readouts	Clinical Relevance
Immune Cell Typing and Localization	Integration with scRNA-seq references + deconvolution algorithms	Immune cell densities, spatial distribution, neighborhood patterns	Identification of immune-excluded vs. immune-inflamed phenotypes [57] [53]
Tertiary Lymphoid Structure (TLS) Characterization	High-plex protein and RNA detection (CODEX, CellScape)	Immune cell organization, germinal center formation, lymphocyte maturation	Positive prognostic indicator across multiple cancer types [3] [53]
Immune Checkpoint Spatial Mapping	Multiplexed protein imaging (PhenoCycler, IMC)	PD-1/PD-L1, LAG-3, TIM-3 distribution relative to tumor cells	Predictors of response to checkpoint inhibitor therapy [57] [53]
Tumor-Immune Interface Analysis	Spatial boundary identification + differential expression	Cytolytic activity, immunosuppressive signals, metabolic competition	Mechanisms of immune resistance and sensitivity [3] [57]
CAR-T Cell Tracking	Multiomic spatial profiling (CellScape)	CAR-T persistence, activation state, tumor engagement	Optimization of cell therapy protocols [3]

Experimental Design for Spatial Immune Profiling

Effective spatial immune profiling requires strategic panel design and multimodal integration:

Targeted Panel Design:

• Select marker genes that distinguish immune cell subtypes (T cells: CD3D, CD8A; B cells: CD79A, MS4A1; Macrophages: CD68, CD163)
• Include immune activation markers (IFNG, GZMB, PRF1), checkpoint molecules (PDCD1, CTLA4, LAG3), and functional state indicators
• For protein co-detection, incorporate validated antibodies with minimal cross-reactivity in multiplexed panels

Multimodal Integration:

• Combine spatial transcriptomics with multiplexed protein detection (CODEX, PhenoCycler) to correlate transcriptional states with protein expression and post-translational modifications [57] [52]
• Integrate with spatial metabolomics (MALDI-MSI) to map nutrient availability, metabolic waste products, and oncometabolites that influence immune function [58] [59]
• Register spatial data with H&E and IHC stains to connect molecular profiles with standard pathological assessment

Spatial Analysis Framework:

• Apply neighborhood analysis to identify recurrent cellular communities and interaction patterns
• Calculate cell-cell proximity metrics to quantify immune-tumor interactions and spatial exclusion
• Perform gradient analysis to identify spatial patterns of immune activation and suppression

Figure 2: Integrated workflow for spatial immune profiling in the tumor microenvironment, combining transcriptomic, proteomic, and metabolomic data.

Analytical Approaches for Spatial Immunology

Computational methods for spatial immune profiling have evolved to capture the complexity of tumor-immune interactions:

Cell-Cell Communication Inference: Tools like CellChat and NicheNet adapted for spatial data predict ligand-receptor interactions between neighboring cells, revealing autocrine and paracrine signaling networks that shape the immune microenvironment [57].

Spatial Trajectory Analysis: Methods such as SpatiAlign and STAligner reconstruct the migration and differentiation paths of immune cells across tissue space, tracking T cell exhaustion gradients or macrophage polarization states from blood vessels into tumor cores [55].

Multiscale Integration: Frameworks like MISO employ deep learning to predict spatial gene expression patterns from standard H&E histology, potentially enabling retrospective analysis of clinical archives and connecting spatial immune features with morphological patterns recognized by pathologists [5].

Identifying Novel Therapeutic Targets

Spatial Discovery of Vulnerabilities

Spatial transcriptomics reveals therapeutic targets through identification of spatially restricted disease mechanisms, compartment-specific dependencies, and resistance pathways that are invisible to bulk analyses.

Target Identification Strategies:

• Region-Specific Differential Expression: Compare gene expression between spatial domains (e.g., invasive margin vs. tumor core, treatment-resistant niches vs. sensitive regions) to identify territory-specific vulnerabilities [3] [53]

• Cell Neighborhood Analysis: Identify expression programs associated with specific cellular microenvironments, such as immune-suppressive niches or stromal interaction zones that promote tumor survival [3] [53]

• Spatial Synthetic Lethality: Discover gene pairs where spatial co-localization creates unique dependencies, particularly targeting interactions between tumor and stromal compartments [53]

• Resistance Niche Mapping: Analyze pre- and post-treatment samples to identify spatial patterns associated with therapeutic resistance, including protected niches that serve as reservoirs for persistent cells [53] [59]

Experimental Protocols for Target Validation

Longitudinal Spatial Monitoring:

• Collect paired tumor biopsies before and during treatment from clinical trials
• Apply spatial barcoding platforms to track spatial evolution under therapeutic pressure
• Identify early spatial biomarkers of response and resistance

Functional Validation Workflow:

• Prioritize candidate targets based on spatial specificity, druggability, and clinical association
• Implement CRISPR-based perturbation (CRISPRi, CRISPRa) in 3D tumor models followed by spatial readouts (CosMx CRISPR workflow) to validate target necessity [3]
• Develop targeted therapeutic agents and assess spatial distribution using mass spectrometry imaging
• Evaluate efficacy in patient-derived organoids and xenografts with spatial endpoint analysis

Integration with Drug Development Pipelines:

• Utilize spatial biomarker signatures for patient stratification in clinical trials
• Employ spatial pharmacodynamics to assess target engagement and mechanism of action in tissue context
• Apply spatial data to optimize drug combinations that address heterogeneous tumor ecosystems

Table 3: Spatial Transcriptomics in Therapeutic Development

Development Stage	Spatial Application	Technology Platform	Output
Target Discovery	Regional vulnerability identification	CosMx WTX, GeoMx DPA	Spatially restricted targets, microenvironmental dependencies [3]
Lead Optimization	Tissue distribution and penetration assessment	MALDI-MSI, DESI-MSI	Drug and metabolite spatial localization [59]
Preclinical Efficacy	Tumor-immune modulation tracking	CellScape, PhenoCycler	Spatial mechanisms of action, immune activation [3]
Biomarker Development	Response signature discovery	Visium, Xenium	Predictive spatial signatures, patient stratification [52] [53]
Clinical Trial Analysis	Resistance mechanism elucidation	Multi-platform integration	Spatial evolution under treatment, resistance niches [53]

Integrated Research Workflows and Reagent Solutions

Comprehensive Spatial Multi-omics Workflow

Advanced spatial biology now integrates multiple molecular modalities to create comprehensive maps of tumor biology:

Figure 3: Integrated spatial multi-omics workflow for comprehensive tumor profiling, simultaneously capturing multiple molecular layers from a single tissue section.

Essential Research Reagent Solutions

Table 4: Key Research Reagents and Platforms for Spatial Cancer Research

Reagent/Platform	Type	Function in Spatial Analysis	Example Applications
CosMx Human Whole Transcriptome (WTX) Assay	Panel-based assay	Subcellular spatial transcriptomics with 18,000+ RNA targets	Tumor heterogeneity, rare cell detection, CRISPR validation [3]
GeoMx Discovery Proteome Atlas	Protein assay	1,100+ plex spatial proteomics paired with whole transcriptome	Immune profiling, signaling pathway activation, cell typing [3]
CellScape Precise Spatial Proteomics	Platform	High-plex iterative staining for protein and RNA detection	CAR-T tracking, tumor-immune interactions, checkpoint mapping [3]
nCounter ADC Development Panel	Targeted panel	High-throughput characterization of antibody-drug conjugates	ADC mechanism of action, resistance studies [3]
PaintScape Platform	Genomic architecture tool	In situ visualization of 3D genome organization	Chromatin folding, ecDNA detection, structural variation [3]
CellSP Computational Framework	Software tool	Identification of gene-cell modules with coordinated subcellular patterns	RNA localization patterns, functional module discovery [54]

Spatial transcriptomics and related spatial technologies have fundamentally transformed our approach to cancer research by preserving the architectural context of molecular measurements. The applications in spatial mapping, immune profiling, and target identification provide unprecedented insights into tumor organization and therapeutic opportunities. As these technologies continue to evolve toward higher resolution, greater multiplexing capacity, and improved integration across molecular modalities, they promise to accelerate the development of precisely targeted therapies that account for the spatial complexity of human tumors. The implementation of robust experimental and computational frameworks outlined in this guide will enable researchers to fully leverage spatial approaches in advancing cancer understanding and treatment.

Integrating ST with Single-Cell RNA-seq for Enhanced Cellular Deconvolution

The tumor microenvironment (TME) is a complex ecosystem comprising malignant cells and diverse non-malignant components, including immune cells, cancer-associated fibroblasts, vascular endothelial cells, and tissue-resident stromal cells, all embedded within the extracellular matrix [33]. Traditional bulk RNA sequencing obscures cellular heterogeneity by averaging gene expression across mixed cell populations, while single-cell RNA sequencing (scRNA-seq), though providing high-resolution transcriptomic profiles, requires tissue dissociation that eliminates critical spatial context [33] [27]. Spatial transcriptomics (ST) has emerged as a revolutionary complementary technology that maps gene expression within intact tissue sections, preserving the native spatial architecture and enabling researchers to investigate cellular organization and communication within the TME [33] [27].

The integration of scRNA-seq and ST technologies provides a powerful synergistic approach for deciphering the complexity and spatial organization of the TME with unprecedented resolution [33]. This technical guide explores the computational frameworks, experimental protocols, and analytical tools that enable effective data integration, focusing specifically on their application to cellular deconvolution – the process of inferring cellular composition and organization from spatially barcoded gene expression data. By bridging single-cell resolution with spatial localization, researchers can now uncover cellular heterogeneity, stromal-immune interactions, and spatial niches that drive tumor progression and therapy resistance, ultimately advancing precision oncology through spatially-informed biomarkers and diagnostic tools [33].

Core Computational Strategies for Data Integration

Deconvolution and Mapping Approaches

Table 1: Computational Methods for scRNA-seq and ST Data Integration

Method	Underlying Algorithm	Primary Function	Key Advantages	References
TACIT	Unsupervised thresholding with graph-based clustering	Cell type annotation in spatial multiomics	No training data required; handles sparse marker panels; identifies rare cell types	[60]
iSORT	Transfer learning via neural networks	Maps gene expression to spatial locations; identifies spatial-organizing genes	Infers pseudo-growth trajectories using SpaRNA velocity concept	[61]
SPOTlight	Non-negative matrix factorization	Spot deconvolution	Efficient for decomposing mixed spot expressions into constituent cell types	[61]
Cell2location	Hierarchical Bayesian framework	Spot deconvolution	Accounts for tissue heterogeneity and technical variations	[61]
Tangram	Deep neural networks	Maps single-cell data to spatial coordinates on discrete spots	High accuracy in spatial alignment of cell types	[61]
novoSpaRc	Optimal transport method	Predicts spatial probability distribution for individual cells	Reconstructs spatial organization without prior spatial information	[61]

Deconvolution approaches primarily aim to resolve the cellular composition of ST spots, each of which typically captures transcriptomes from multiple cells. Sequencing-based ST platforms such as 10X Visium provide whole transcriptome coverage but at a resolution that encompasses multiple cells per spot, necessitating computational methods to infer the specific cell types contributing to each spot's expression profile [61] [27]. The integration of scRNA-seq data as a reference enables this deconvolution by providing cell type-specific gene expression signatures.

Mapping approaches focus on projecting single-cell transcriptomes onto spatial coordinates to reconstruct tissue architecture at cellular resolution. These methods use various computational frameworks to position individual cells within the spatial context of tissues, effectively "imputing" spatial information for scRNA-seq data [61].

Advanced Integration Frameworks

Recent advancements in integration methodologies have addressed specific challenges in spatial transcriptomics. The TACIT (Threshold-based Assignment of Cell Types from Multiplexed Imaging Data) algorithm employs an unsupervised approach for cell annotation using predefined signatures without requiring training data [60]. TACIT uses unbiased thresholding to distinguish positive cells from background, focusing on relevant markers to identify ambiguous cells in multiomic assays. Validation across five datasets encompassing 5,000,000 cells and 51 cell types from three biological niches (brain, intestine, gland) demonstrated that TACIT outperforms existing unsupervised methods in both accuracy and scalability [60].

The iSORT (integrative Spatial Organization of cells using density Ratio Transfer) framework utilizes transfer learning to decipher spatial organization of cells by integrating scRNA-seq and ST data [61]. iSORT trains a neural network that maps gene expressions to spatial locations, enabling the identification of spatial-organizing genes (SOGs) that drive tissue patterning, and infers pseudo-growth trajectories using a novel concept called SpaRNA velocity, which projects RNA velocity onto the physical space of ST slices [61].

For large-sized tissues that exceed the capture area of conventional ST platforms, iSCALE (inferring Spatially resolved Cellular Architectures in Large-sized tissue Environments) provides a machine learning framework that reconstructs large-scale, super-resolution gene expression landscapes by leveraging the relationship between gene expression profiles and histological image characteristics [62]. This approach enables comprehensive gene expression prediction and tissue annotation across entire large tissue sections, including regions without direct gene expression measurements, making it particularly valuable for studying sizable human tissue samples common in clinical research [62].

Experimental Protocols for Integrated Analysis

Workflow for Combined scRNA-seq and ST Profiling

Diagram: Integrated scRNA-seq and ST Analysis Workflow

A robust experimental workflow for integrating scRNA-seq and ST data begins with careful sample preparation. For colorectal cancer studies, researchers have successfully profiled 41,700 cells from three CRC tumor-normal-blood pairs using this integrated approach [63]. The protocol involves:

• Sample Collection and Processing: Collect matched tumor tissues, adjacent normal tissues, and peripheral blood mononuclear cells (PBMCs) from patients. For solid tumors, process tissues immediately for either single-cell dissociation or optimal cutting temperature (OCT) compound embedding for cryosectioning [63].

• Single-Cell RNA Sequencing: Generate single-cell suspensions using appropriate enzymatic digestion protocols. Perform scRNA-seq library preparation using platforms such as 10X Genomics. Sequence to a depth of approximately 150 G reads per sample, achieving median sequencing saturation of 90% or higher [63].

• Spatial Transcriptomics Profiling: Prepare tissue sections of appropriate thickness (typically 10-16 μm) for ST platforms. For sequencing-based approaches like 10X Visium, follow standard protocols for tissue permeabilization and library preparation. For imaging-based platforms like CosMx or MERFISH, optimize hybridization and imaging conditions [3].

• Quality Control Metrics: Apply stringent quality control filters for both scRNA-seq and ST data. For scRNA-seq, retain cells with at least 1,000 genes and 2,500 unique molecular identifiers (UMIs). Remove cells with high mitochondrial gene percentage indicative of stress or apoptosis [63].

Cell Type Annotation and Malignant Cell Identification

Diagram: Cell Type Deconvolution and Annotation Pipeline

Cell type annotation in spatial transcriptomics data leverages scRNA-seq reference data to identify both major cell populations and rare cell subtypes. A typical workflow includes:

• Reference-Based Annotation: Transfer cell type labels from scRNA-seq to ST data using canonical marker genes. For colorectal cancer, major populations include epithelial cells, fibroblasts, endothelial cells, monocytes, T cells, NK cells, B cells, and mast cells, identified through expression of established markers such as EPCAM (epithelial cells), PTPRC (T cells), CD19 (B cells), and LUM (fibroblasts) [63].

• Malignant Cell Identification: Distinguish malignant epithelial cells from normal epithelial cells through copy number variation (CNV) analysis. Calculate large-scale gene expression patterns across genomic regions to infer CNV alterations characteristic of cancer cells [63].

• Subpopulation Analysis: Further subcluster epithelial cells to identify malignant subpopulations. In CRC, researchers have identified seven subtypes of malignant cells reflecting heterogeneous states in tumors, including tumorCAV1, tumorATF3JUN|FOS, tumorZEB2, tumorVIM, tumorWSB1, tumorLXN, and tumorPGM1, each with distinct transcriptional programs [63].

• Spatial Regional Annotation: Transfer cellular annotations from scRNA-seq to ST spots to define tissue regions such as tumor core, stroma, immune infiltration zones, and normal epithelium. Validate regional annotations through histopathological examination and marker gene expression patterns [63].

Analytical Tools and Research Reagent Solutions

Computational Tools for Spatial Deconvolution

Table 2: Key Computational Tools for Spatial Deconvolution and Analysis

Tool	Primary Function	Input Data	Output	Access
TACIT	Cell type annotation from multiplexed imaging data	Spatial transcriptomics/proteomics data, cell type signatures	Annotated cell types with confidence scores	Available upon request [60]
iSORT	Transfer learning for spatial organization prediction	scRNA-seq data, ST reference	Spatial-organizing genes, SpaRNA velocity	GitHub: xiaojierzi/iSORT [61]
ReDeconv	Bulk RNA-seq deconvolution accounting for transcriptome size	Bulk RNA-seq data, reference signatures	Cell type proportions with size correction	https://redeconv.stjude.org [64]
iSCALE	Large-scale spatial gene expression prediction	H&E images, ST training captures	Predicted gene expression for large tissues	Available upon request [62]
Cell2location	Bayesian deconvolution of spatial transcriptomics	scRNA-seq reference, ST data	Cell type abundance maps	Standard Python package [61]
Tangram	Deep learning-based spatial mapping	scRNA-seq data, ST data	Aligned single-cell spatial coordinates	Standard Python package [61]

Research Reagent Solutions

Table 3: Essential Research Reagents and Platforms for Spatial Transcriptomics

Platform/Reagent	Type	Key Features	Applications in Tumor Research	References
10X Visium	Sequencing-based ST	Whole transcriptome, 6.5×6.5mm capture area	Spatial mapping of tumor heterogeneity and TME	[62] [27]
CosMx Human WTX	Imaging-based ST	Subcellular resolution, 1,000+ RNA targets	Single-cell spatial analysis in FFPE tumors	[3]
CellScape Platform	Spatial proteomics	High-plex protein detection (65+ markers)	Immune cell phenotyping in tumor microenvironments	[3]
GeoMx DPA	Spatial multiomics	1,100+ plex protein assay with WTA	Comprehensive tumor microenvironment characterization	[3]
MERFISH	Imaging-based ST	Single-molecule resolution, high-plex RNA detection	Cellular neighborhoods and rare cell populations in tumors	[33] [61]
Akoya Phenocycler-Fusion	Spatial proteomics	50+ protein markers, single-cell resolution	Immune contexture analysis in colorectal cancer	[60]

Applications in Tumor Biology and Clinical Translation

Insights into Tumor Architecture and Metastasis

Integrated scRNA-seq and ST analyses have revealed fundamental principles of tumor organization and progression. Studies comparing primary hepatocellular carcinoma (HCC) and liver metastases have uncovered distinct spatial architectures: HCC displays an ordered lineage architecture with transformed hepatocyte-like tumor cells broadly dispersed across the tissue, while liver metastases show sharply compartmentalized domains including an invasion zone where proliferative stem-like tumor cells occupy TAM-rich boundaries adjacent to hypoxia-adapted tumor-core cells [65].

Notably, despite these organizational differences, both tumor types converge on shared metabolic programs, such as "porphyrin overdrive" characterized by reduced cytochrome P450 expression, enhanced oxidative phosphorylation gene expression, and upregulation of FLVCR1 and ALOX5, reflecting coordinated rewiring of heme and lipid metabolism that may represent a therapeutic vulnerability [65].

Revealing Cell-Cell Communication Networks

The integration of scRNA-seq and ST enables the inference of spatially organized cell-cell communication networks within the TME. In colorectal cancer, analyses have revealed intensive intercellular interactions between stroma and tumor regions that are extremely proximal in tissue sections. Specifically, the ligand-receptor pair C5AR1-RPS19 has been identified as playing key roles in the crosstalk between stroma and tumor regions [63].

Spatial characterization of tumor regions has identified TMSB4X as a highly expressed feature in CRC tumor regions, suggesting its potential as a diagnostic marker, while stroma regions are characterized by VIM-high expression, indicating a stromal niche fostering tumor progression [63]. These spatially resolved interactions provide potential targets for disrupting pro-tumorigenic signaling within the TME.

Biomarker Discovery and Therapeutic Implications

The integration of scRNA-seq and ST technologies is advancing precision oncology by enabling the discovery of spatially informed biomarkers. For instance, spatial analysis of lung cancer tissues has identified distinct gene expression signatures at the invasive front that correlate with metastatic potential and patient prognosis [27]. Similarly, studies in triple-negative breast cancer have revealed spatial patterns of immune cell exclusion that may predict response to immunotherapy [3].

Three-dimensional spatial profiling and multimodal integration with proteomic and epigenomic data are further enhancing our understanding of tumor biology, revealing complex relationships between genetic alterations, gene expression patterns, protein activity, and metabolic pathways within the spatial context of tumors [27]. These advances are paving the way for more precise diagnostic approaches and therapeutic strategies that target specific spatial compartments or cellular interactions within the TME.

The integration of single-cell RNA sequencing with spatial transcriptomics represents a transformative approach for deconvoluting cellular complexity within tissue architecture. Through sophisticated computational methods such as TACIT, iSORT, and deconvolution algorithms, researchers can now reconstruct cellular landscapes with unprecedented resolution, revealing spatial niches, cellular communication networks, and tissue organizational principles that drive tumor progression and therapeutic response.

As these technologies continue to evolve, with improvements in resolution, throughput, and multimodal integration, they hold immense promise for advancing precision oncology. The identification of spatially informed biomarkers and therapeutic targets will enable more effective diagnostic and treatment strategies tailored to the unique spatial architecture of individual tumors. Future developments in artificial intelligence, deep learning, and three-dimensional spatial profiling will further enhance our ability to decipher the complex spatial biology of cancer and other diseases.

Navigating Technical Challenges: A Framework for Benchmarking and Optimizing ST Data

In the field of cancer research, spatial transcriptomics (ST) has emerged as a pivotal technology for elucidating the intricate spatial organization of tumors. It bridges a critical gap left by single-cell RNA sequencing (scRNA-seq) by linking molecular profiles to their spatial context within the tissue architecture [66] [1]. The ability to study the tumor microenvironment (TME), cellular heterogeneity, and cell-cell interactions in situ has profound implications for understanding tumor initiation, progression, and therapeutic response [22]. However, the rapid evolution of commercial ST platforms necessitates a rigorous and standardized approach to evaluate their performance. For researchers investigating tumor organization, three metrics are paramount: sensitivity (the ability to detect true transcript signals), specificity (the ability to avoid false-positive signals), and spatial resolution (the minimal distance at which two distinct transcript signals can be discerned). This guide provides a technical framework for the critical assessment of these metrics, underpinned by recent benchmarking studies using human tumor samples.

Core Performance Metrics and Platform Comparison

The performance of an ST platform is not a singular characteristic but a combination of interdependent metrics that directly impact data quality and biological interpretation. Systematic benchmarking studies, which utilize serial sections from the same tumor samples and orthogonal validation datasets, provide the most objective performance assessments [66] [34].

Defining the Metrics

• Sensitivity refers to the platform's transcript capture efficiency. It is commonly measured by the number of unique molecules or transcripts detected per cell. High sensitivity is crucial for identifying rare but biologically significant transcripts and for robust cell type annotation, especially in heterogeneous tumor tissues.
• Specificity indicates the signal-to-noise ratio. It is evaluated using negative control probes and blank code words included in the assay panels. A high-specificity platform minimizes off-target hybridization, ensuring that the detected spatial patterns are biologically real and not technical artifacts [34].
• Spatial Resolution defines the smallest discernible detail in a spatial measurement. For ST, this ranges from the subcellular level (∼0.5 µm) to the multi-cellular level (∼55 µm). Higher resolution is essential for accurate cell segmentation, precise localization of transcripts within cellular compartments, and studying subcellular phenomena in tumor biology [66] [54].

Comparative Performance of Major Platforms

Recent independent benchmarking efforts have profiled the leading high-throughput, subcellular-resolution ST platforms using formalin-fixed paraffin-embedded (FFPE) human tumor samples [66] [34]. The table below synthesizes key quantitative findings from these studies.

Table 1: Performance Comparison of High-Throughput Spatial Transcriptomics Platforms

Platform	Technology Type	Reported Sensitivity (Transcripts/Cell)	Specificity Assessment	Spatial Resolution	Key Strengths in Tumor Analysis
Xenium (10x Genomics)	Imaging-based (iST)	Consistently high; superior sensitivity for marker genes like EPCAM [66].	High; minimal target gene probes expressed similarly to negative controls [34].	Single-molecule precision [66].	Excellent concordance with scRNA-seq and protein data (CODEX) [66].
CosMx (NanoString)	Imaging-based (iST)	Highest raw transcript counts per cell [34].	Some target gene probes (e.g., CD3D, FOXP3) expressed at levels similar to negative controls, potentially impacting immune cell annotation [34].	Single-molecule precision [66].	High-plex gene panels; capable of detecting extensive transcriptomes [34].
Visium HD (10x Genomics)	Sequencing-based (sST)	High correlation with scRNA-seq gene counts; outperformed Stereo-seq in cancer cell marker detection in ROIs [66].	N/A (Relies on poly(dT) capture; specificity managed bioinformatically).	2 µm resolution [66].	Unbiased whole-transcriptome analysis [66].
Stereo-seq (BGI)	Sequencing-based (sST)	High correlation with scRNA-seq gene counts [66].	N/A (Relies on poly(dT) capture; specificity managed bioinformatically).	0.5 µm resolution [66].	Extremely high spatial resolution for an sST platform [66].
MERFISH (Vizgen)	Imaging-based (iST)	Lower transcript counts per cell in older archival samples; performance is tissue-age dependent [34].	Lacks negative control probes for direct assessment [34].	Single-molecule precision [22].	High detection efficiency for targeted panels [22].

Experimental Protocols for Benchmarking Metrics

Robust benchmarking requires carefully controlled experiments that use serial sections from the same tumor block to eliminate biological variability and incorporate multi-omics ground truth data for validation [66] [34].

Establishing Ground Truth with Multi-Omics Profiling

To objectively evaluate ST platform performance, a foundational step is the creation of orthogonal validation datasets from the same sample.

• Sample Preparation: Collect treatment-naïve tumor samples (e.g., colon adenocarcinoma, hepatocellular carcinoma, lung adenocarcinoma) and process them into FFPE blocks. Generate serial tissue sections of 5 µm thickness for parallel profiling on different ST platforms and validation assays [66] [34].
• Orthogonal Validation Data:
  • Single-cell RNA Sequencing (scRNA-seq): Perform on dissociated cells from the same tumor sample. This data serves as a spatial-agnostic reference for evaluating the sensitivity and gene detection accuracy of ST platforms [66] [34].
  • Multiplexed Protein Imaging (CODEX/Immunofluorescence): Profile proteins on tissue sections adjacent to those used for ST using technologies like CODEX (co-detection by indexing) or multiplex immunofluorescence (mIF). This provides an independent spatial context to validate transcriptomic findings, especially for key tumor and immune markers [66] [34].
  • Histopathological Annotation: Have trained pathologists review H&E-stained sections and mIF data to manually annotate cell types and tissue regions. This serves as the morphological ground truth for assessing cell segmentation and phenotyping accuracy [34].

Methodologies for Metric-Specific Evaluation

The following experimental and analytical procedures are used to quantify each core metric.

• Evaluating Sensitivity:

  • Data Processing: For each platform, generate cell-by-gene count matrices following the manufacturer's recommended bioinformatics pipeline.
  • Calculation: Calculate the mean and median number of transcripts per cell and the number of unique genes detected per cell across the entire dataset and within specific, pathologist-annotated cell populations.
  • Comparison: Correlate gene-wise transcript counts from the ST data with the matched scRNA-seq reference profile. A higher correlation coefficient indicates better sensitivity and accuracy in capturing the true transcriptional landscape [66].

• Evaluating Specificity:

  • Control Probe Analysis: For platforms that include them (e.g., CosMx, Xenium), extract the expression counts of negative control probes and blank code words. These probes are designed to not bind any biological target.
  • Threshold Determination: Establish an expression threshold for true signal detection based on the distribution of counts from these negative controls (e.g., mean + 2 standard deviations).
  • Signal-to-Noise Assessment: Identify and report any target gene probes whose expression falls below this established threshold, as these may not be reliably detectable above background noise [34].

• Evaluating Spatial Resolution and Cell Segmentation:

  • Manual Annotation: Manually segment nuclei from high-resolution DAPI or H&E images to create a "gold standard" set of cell boundaries.
  • Algorithmic Segmentation: Apply each platform's native cell segmentation algorithm (e.g., unimodal based on DAPI, or multimodal incorporating protein markers) to the same tissue region.
  • Performance Quantification: Calculate metrics like the F1-score for cell detection by comparing algorithmic segmentation to manual annotation. Further, assess segmentation accuracy by evaluating the co-detection of genes known to be mutually exclusive in different cell types, which can indicate erroneous segmentation or transcript diffusion [34].

Diagram 1: Experimental benchmarking workflow for ST platforms.

The Scientist's Toolkit: Essential Reagents and Materials

The following reagents and tools are critical for executing the benchmarking protocols described above and for conducting rigorous spatial transcriptomics studies of tumor tissues.

Table 2: Essential Research Reagent Solutions for Spatial Transcriptomics

Item	Function / Explanation
Formalin-Fixed Paraffin-Embedded (FFPE) Tissue Blocks	The standard for sample processing and archiving in pathology. Essential for benchmarking with clinically relevant samples and ensures compatibility with most commercial ST platforms [34].
Visium CytAssist Tissue Slide Alignment Quick Reference Card	A guide tool for 10x Genomics Visium workflows to demarcate the viable region on a microscope slide for probe transfer, ensuring the sample is positioned correctly for analysis [67].
Custom-Targeted Gene Panels	Pre-designed probe sets for imaging-based platforms (e.g., CosMx, Xenium, MERFISH). Panels are often tailored to immuno-oncology, containing genes relevant to tumor, stromal, and immune cell populations [34].
Negative Control & Blank Probes	Probes included in commercial panels that do not target any biological sequence. They are fundamental for quantifying background noise and establishing thresholds for assessing assay specificity [34].
CODEX Antibody Panels	Multiplexed antibody panels for protein co-detection. Used on adjacent serial sections to validate protein-level expression of targets identified by ST, providing a multi-omic ground truth [66].
Collagen-Coated Microscope Slides	Used in specialized protocols for profiling 2D cell cultures or engineered tissues. The coating facilitates cell adhesion when traditional tissue sectioning is not feasible [67].

Advanced Analysis: From Single Cells to Subcellular Modules

Beyond core metrics, advanced computational tools are now enabling the discovery of biologically meaningful patterns in the rich data generated by high-resolution ST.

Functional Interpretation with CellSP

For subcellular resolution data, tools like CellSP (Cell Subcellular Patterns) can identify "gene-cell modules"—sets of genes that exhibit coordinated spatial distribution patterns (e.g., peripheral, radial, punctate) within a common set of cells [54]. This analysis moves beyond single-gene localization to uncover systems-level organization. For example, in mouse brain and human kidney cancer data, CellSP has been used to discover modules related to myelination, axonogenesis, and immune responses that change between healthy and diseased states [54]. The process involves:

• Pattern Discovery: Using tools like SPRAWL and InSTAnT to identify significant subcellular distribution patterns for individual genes or gene pairs in each cell.
• Module Discovery: Applying a biclustering algorithm to find genes that co-exhibit the same spatial pattern in a significant subset of cells.
• Module Characterization: Interpreting modules via Gene Ontology enrichment and training machine learning classifiers to identify genes and pathways predictive of module membership [54].

Diagram 2: CellSP workflow for subcellular module discovery.

The systematic evaluation of sensitivity, specificity, and spatial resolution is a critical prerequisite for generating biologically and clinically impactful spatial transcriptomics data in cancer research. As benchmarking studies demonstrate, platform performance varies significantly, influencing the detection of key immune markers, the accuracy of cell typing, and the ability to resolve subtle spatial features of the tumor microenvironment. There is no single "best" platform; the choice depends on the specific research question, requiring a balance between whole-transcriptome discovery and targeted, high-sensitivity hypothesis testing. By adopting the standardized evaluation frameworks and protocols outlined in this guide, researchers can make informed decisions, ensure the rigor of their data, and fully leverage spatial transcriptomics to unravel the complex architecture of human tumors.

Systematic Benchmarking of ST Platforms Using Clinical Cancer Samples

Spatial transcriptomics (ST) has emerged as a revolutionary technology that bridges the critical gap between single-cell molecular profiling and tissue architecture context. In cancer research, where cellular spatial relationships and tumor microenvironment interactions dictate disease progression and therapeutic response, ST provides unprecedented insights into tumor organization. However, the rapid proliferation of commercial ST platforms with distinct technological approaches, resolutions, and sensitivities has created an urgent need for systematic benchmarking, particularly using clinically relevant formalin-fixed, paraffin-embedded (FFPE) samples. This technical review synthesizes comprehensive benchmarking data from recent studies to guide researchers in selecting appropriate platforms, designing robust experiments, and accurately interpreting spatial data within tumor architecture research.

Spatial transcriptomics technologies can be broadly categorized into sequencing-based (sST) and imaging-based (iST) platforms, each with distinct methodological foundations and performance characteristics. sST platforms utilize spatially barcoded poly(dT) oligos on arrays to capture poly(A)-tailed RNA for subsequent sequencing, enabling unbiased whole-transcriptome analysis. In contrast, iST platforms employ multiple rounds of fluorescently labeled probe hybridization, imaging, and destaining to localize transcripts through combinatorial barcoding at single-molecule resolution [49].

Recent technological advancements have produced platforms with substantially enhanced spatial resolution and gene detection capacity. Key commercial platforms now offer subcellular resolution (≤2 μm) and high-throughput gene detection (>5,000 genes), including Stereo-seq v1.3, Visium HD FFPE, CosMx 6K, and Xenium 5K [49]. These platforms represent the current state-of-the-art for clinical cancer samples, particularly FFPE tissues, which constitute over 90% of clinical pathology specimens [68].

Table 1: Technical Specifications of Major Spatial Transcriptomics Platforms

Platform	Technology Type	Spatial Resolution	Gene Detection Capacity	FFPE Compatibility	Key Strengths
Stereo-seq v1.3	Sequencing-based	0.5 μm	Whole transcriptome	Yes (Fresh Frozen)	Highest resolution, unbiased detection
Visium HD FFPE	Sequencing-based	2 μm	18,085 genes	Yes	Whole transcriptome, standardized workflow
Xenium 5K	Imaging-based	Single molecule	5,001 genes	Yes	High sensitivity, optimized panels
CosMx 6K	Imaging-based	Single molecule	6,175 genes	Yes	Large gene panels, subcellular localization
MERSCOPE	Imaging-based	Single molecule	500-1,000 genes	Yes	High specificity, custom panels

Experimental Design for Systematic Benchmarking

Sample Selection and Preparation

Robust benchmarking requires carefully controlled experimental designs using matched clinical samples processed under uniform conditions. Recent comprehensive studies have utilized:

• Multiple cancer types: Colon adenocarcinoma (COAD), hepatocellular carcinoma (HCC), ovarian cancer (OV), and breast cancer specimens from treatment-naïve patients to represent diverse tumor biology [49].
• Tissue Microarrays (TMAs): Containing 17 tumor and 16 normal tissue types from clinical FFPE archives, enabling high-throughput platform assessment across diverse tissue contexts [68].
• Serial sectioning: Generation of consecutive tissue sections (4-5 μm thickness) from the same FFPE blocks for parallel processing across different platforms, ensuring maximal comparability.
• Multi-omics ground truth: Integration of CODEX spatial proteomics on adjacent sections and single-cell RNA sequencing (scRNA-seq) from the same samples to establish comprehensive reference datasets [49].

Platform Evaluation Framework

Systematic benchmarking should assess multiple performance dimensions critical for cancer research applications:

• Sensitivity and Specificity: Transcript detection efficiency and false positive rates using known marker genes and orthogonal validation.
• Spatial Resolution and Diffusion Control: Ability to resolve single cells and subcellular features while minimizing transcript diffusion.
• Cell Segmentation Accuracy: Performance in delineating cell boundaries using DAPI, membrane stains, or computational methods.
• Concordance with Reference Data: Correlation with matched scRNA-seq and spatial proteomics data.
• Cell Type Annotation and Spatial Clustering: Capacity to identify biologically relevant cellular neighborhoods and tissue domains.

Performance Benchmarking Results

Molecular Capture Efficiency

Marker Gene Detection: Evaluation of established cell marker genes reveals platform-specific sensitivity patterns. The epithelial cell marker EPCAM shows well-defined spatial patterns across all platforms, consistent with H&E staining and Pan-Cytokeratin immunostaining on adjacent sections [49]. Quantitative assessments within shared tissue regions demonstrate that Xenium 5K consistently achieves superior sensitivity for multiple marker genes compared to other platforms [49].

Gene Panel-Wide Performance: When assessing entire gene panels, Stereo-seq v1.3, Visium HD FFPE, and Xenium 5K show high correlations with matched scRNA-seq references (Figure 1D) [49]. CosMx 6K detects a higher total number of transcripts than Xenium 5K but shows substantial deviation from scRNA-seq reference data, indicating potential technical biases in transcript recovery [49].

Table 2: Quantitative Performance Metrics Across Platforms

Platform	Transcripts per Cell	Genes per Cell	Correlation with scRNA-seq	Cell Segmentation Accuracy
Stereo-seq v1.3	Medium	High	0.89	Variable (depends on segmentation method)
Visium HD FFPE	Medium	High	0.85	High (with nuclear staining)
Xenium 5K	High	Medium-High	0.91	High (with membrane staining)
CosMx 6K	High	Medium-High	0.78	Medium-High
MERSCOPE	Medium	Medium	0.82	Medium

Spatial Resolution and Cell Segmentation

Spatial Specificity: Imaging-based platforms (Xenium, CosMx, MERSCOPE) inherently provide single-cell resolution due to their imaging-based detection system. Sequencing-based platforms achieve subcellular resolution through small capture feature sizes (0.5-2 μm) but require computational integration for single-cell analysis [49].

Cell Segmentation Performance: Assessment of nuclear and membrane segmentation reveals platform-specific strengths. Xenium demonstrates improved segmentation capabilities with additional membrane staining, while CosMx and MERSCOPE show varying degrees of segmentation accuracy depending on tissue type and autofluorescence [68]. All platforms achieve spatially resolved cell typing with varying sub-clustering capabilities, with Xenium and CosMx identifying slightly more clusters than MERSCOPE, albeit with different false discovery rates [68].

Concordance with Orthogonal Data

Spatial Proteomics Alignment: Integration with CODEX spatial proteomics data from adjacent sections reveals strong concordance for key protein-RNA pairs across platforms, validating biological findings [49]. However, instances of RNA-protein decoupling highlight the importance of multi-omics validation for comprehensive tumor characterization [69].

scRNA-seq Integration: Stereo-seq v1.3, Visium HD FFPE, and Xenium 5K demonstrate high concordance with matched scRNA-seq data, supporting their application for cell atlas construction [49]. Cross-platform comparisons reveal strong concordance among these three platforms, highlighting their consistent ability to capture biologically relevant gene expression variation [49].

Experimental Protocols

Sample Processing Workflow

Diagram Title: Sample Processing Workflow for ST Benchmarking

Platform-Specific Methodologies

Sequencing-based Platforms (Stereo-seq, Visium HD):

• Tissue Permeabilization: Optimization of permeabilization time to balance transcript capture efficiency and spatial resolution.
• cDNA Synthesis: On-slide reverse transcription with spatial barcodes.
• Library Preparation: Platform-specific library construction with unique dual indices.
• Sequencing: High-throughput sequencing on Illumina platforms with recommended read depths of 50-100K reads per spot.

Imaging-based Platforms (Xenium, CosMx, MERSCOPE):

• Probe Hybridization: Incubation with gene-specific probe panels (500-6,000 genes).
• Signal Amplification: Platform-specific amplification (rolling circle amplification for Xenium, branch chain hybridization for CosMx, probe tiling for MERSCOPE).
• Cyclic Imaging: Multiple rounds of fluorescent staining, imaging, and destaining.
• Image Processing: Computational reconstruction of transcript locations with subcellular resolution.

Platform Selection Guide

Diagram Title: Platform Selection Decision Tree

The Scientist's Toolkit

Table 3: Essential Research Reagents and Computational Tools

Resource	Type	Function	Application in ST
SCNT R Package	Computational Tool	Data analysis and visualization of single-cell and spatial data	Streamlines quality control, dimensionality reduction, and visualization for ST data [70]
SPATCH Web Server	Data Resource	User-friendly web server for data visualization and download	Enables exploration of benchmarking datasets without computational expertise [49]
CODEX Multiplexed Imaging	Experimental Reagent	High-plex spatial protein profiling	Provides ground truth protein data for ST validation [49]
10X Visium HD Gene Expression	Commercial Kit	Whole transcriptome spatial analysis	Standardized workflow for sequencing-based spatial transcriptomics [49]
Xenium Gene Panels	Commercial Reagent	Targeted gene panels for in situ analysis	Optimized probe sets for specific tissue types and research questions [68]
SurvBoard	Computational Framework	Standardized benchmarking for multi-omics survival models	Enables evaluation of ST clinical prediction performance [71]

Systematic benchmarking of spatial transcriptomics platforms reveals a maturing technological landscape with multiple robust options for clinical cancer samples. Each platform presents distinct strengths: sequencing-based approaches (Stereo-seq, Visium HD) offer unbiased whole-transcriptome coverage ideal for discovery research, while imaging-based platforms (Xenium, CosMx) provide superior single-cell resolution and sensitivity for targeted panels. The consistent high performance of Xenium 5K across multiple metrics, coupled with the expanding gene panels of CosMx 6K and Stereo-seq v1.3, provides researchers with powerful options for diverse research applications.

Future developments in spatial transcriptomics will likely focus on further increasing multiplexing capacity, improving accessibility through streamlined workflows, and enhancing computational integration with histopathology and clinical outcomes. As these technologies become increasingly integral to cancer research, continued systematic benchmarking using standardized frameworks will be essential to guide platform selection and methodological advancement in spatial tumor profiling.

The choice between fresh-frozen (FF) and formalin-fixed paraffin-embedded (FFPE) tissue preservation represents one of the most fundamental methodological decisions in spatial transcriptomics research on tumor architecture. This decision profoundly impacts every subsequent analytical step, from data quality to biological interpretation. In the context of investigating tumor organization architecture, where preserving both spatial context and molecular integrity is paramount, understanding these impacts is not merely technical but foundational to research validity.

FFPE preservation has served as the gold standard in pathology for over a century, with billions of specimens archived worldwide [72]. These archives represent an invaluable resource for retrospective studies linking long-term clinical outcomes with spatial molecular patterns. Conversely, FF preservation is often considered the benchmark for molecular integrity, particularly for sensitive techniques like spatial transcriptomics that aim to capture the intricate cellular relationships within the tumor microenvironment. This technical guide examines the comparative advantages, limitations, and appropriate applications of each method within modern spatial oncology research.

Molecular Integrity: A Comparative Analysis

The fixation and preservation methods employed directly influence the quantity, quality, and analytical potential of nucleic acids and proteins recovered from tissue specimens. These differences stem from the fundamental mechanisms of each process.

Nucleic Acid Quality and Analytical Performance

Table 1: Comparative Nucleic Acid Quality and Sequencing Performance

Parameter	Fresh-Frozen (FF) Tissue	FFPE Tissue	Research Implications
DNA Integrity	High molecular weight DNA [73]	Fragmented DNA; cross-linked [72] [74]	FFPE requires specialized extraction protocols
RNA Integrity	Preserved RNA integrity (RIN >8) [75]	Degraded RNA; reduced RIN (mean 2.2) [76]	FF preferred for RNA-Seq applications
RNA Yield	Higher yields [76]	2-fold less RNA yield [76]	FFPE may require input normalization
Major Artefacts	Minimal sequence artefacts [72]	C>T/G>A transitions from deamination; oxidation artefacts [72]	FFPE artefacts problematic for low-frequency variants
Gene Detection	Higher gene detection in RNA-Seq [77]	~90% overlap with FF in optimized protocols [77]	FF provides more comprehensive transcriptome
Methylation Analysis	Accurate β-values [73]	Overestimated β-values in 21.4% of CpG sites [73]	Caution required for FFPE epigenomic studies

Proteomic and Epigenomic Comparisons

Proteomic studies reveal substantial methodological biases. FFPE samples typically yield approximately 40% fewer protein identifications compared to OCT-embedded frozen samples (approximately 700 vs. 1200 proteins) [78]. Mitochondrial proteins involved in TCA cycle and electron transport are particularly underrepresented in FFPE proteomes, indicating specific vulnerability to formalin fixation [78]. However, when protocols are optimized, shotgun proteomics can identify thousands of proteins with 92% overlap between FFPE and frozen specimens [79].

For chromatin accessibility profiling, recent advances in spatial FFPE-ATAC-seq enable mapping of open chromatin regions in archived tissues, though with notable technical distinctions. While this method maintains expected enrichment at transcription start sites, it produces smaller fragment sizes without clear nucleosome periodicity compared to fresh-frozen spatial ATAC-seq [80].

Spatial Transcriptomics Workflows: Technical Protocols

The integration of FF and FFPE samples into spatial transcriptomics requires distinct preparatory workflows, each with critical steps that influence experimental outcomes.

Fresh-Frozen Tissue Protocol for Spatial Transcriptomics

The optimal FF protocol for skull base tumors emphasizes rapid processing (<15 minutes from resection to freezing) and omission of isopentane snap-freezing to significantly improve RNA quality (p=0.004) and histomorphological integrity (p=0.02) [75]. Fresh tissue washed with cold PBS before OCT embedding and snap-freezing currently represents the best method for preparing spatial sections, with RNA Integrity Number (RIN) ≥6 serving as a sufficient quality threshold for spatial transcriptomics [75].

FFPE Tissue Protocol for Spatial Transcriptomics

For spatial FFPE-ATAC-seq, target retrieval optimization is critical. The highest transcription start site (TSS) enrichment scores are achieved using Tris-EDTA buffer (pH 9.0) at 65°C combined with proteinase K digestion (10 ng/μl for 45 minutes) [80]. This specialized processing helps overcome formalin-induced crosslinking that would otherwise obstruct Tn5 transposase access to genomic DNA [80].

The Scientist's Toolkit: Essential Research Reagents

Table 2: Key Reagents for Tissue Processing and Spatial Analysis

Reagent/Category	Function	Application Notes
OCT Compound	Tissue embedding medium for cryosectioning	Optimal for FF spatial transcriptomics [78] [75]
RNA/DNA Defender	Nucleic acid stabilizer	For fresh reference tissue stabilization [77]
Tris-EDTA Buffer (pH 9.0)	Target retrieval solution	Optimal for FFPE chromatin accessibility (spatial FFPE-ATAC-seq) [80]
Proteinase K	Enzyme for breaking protein-nucleic acid crosslinks	Critical for FFPE epitope retrieval (10 ng/μl for 45 min) [80]
Magnetic Bead-Based Kits	Nucleic acid purification	Gentle deparaffinization and crosslink reversal for FFPE [74]
WT-Ovation FFPE System	Whole transcriptome amplification	Optimized for degraded FFPE RNA [76]
CORALL FFPE Kit	Library preparation	Specialized for FFPE whole transcriptome sequencing [77]
Chromium Single Cell Gene Expression Flex	Single-cell RNA sequencing	Enables scRNA-seq on fixed tissues including FFPE [74]

Impact on Spatial Data Quality and Analytical Outcomes

The preservation method directly influences spatial data quality through multiple technical dimensions. In mass spectrometry-based proteomics, the preservation method introduces greater variation than biological differences between tumor stages, complicating direct comparisons [78]. Multivariate analyses demonstrate that samples cluster primarily by preservation method rather than biological characteristics, necessitating careful normalization when integrating datasets from different sources [78].

For nucleic acid-based spatial analyses, FFPE specimens consistently show reduced library complexity, higher duplication rates, and less uniform coverage [74]. Despite these challenges, gene expression profiles from FFPE tissues can achieve high correlation with matched FF samples (r > 0.89-0.95) when optimized protocols are employed [74]. Single-cell RNA sequencing of FFPE tissues now enables robust preservation of clinically relevant cell type information, with high correlations in signaling pathways between matched fresh and FFPE specimens [74].

The choice between FF and FFPE tissue preservation for spatial transcriptomics of tumor organization involves balancing molecular integrity against architectural preservation, clinical relevance, and resource availability. FF tissues provide superior biomolecule quality and are preferred for discovery-phase research where comprehensive molecular capture is prioritized. FFPE specimens offer unparalleled access to clinically annotated archives and excellent tissue morphology, enabling retrospective longitudinal studies linking spatial organization to clinical outcomes.

Future methodological developments will continue to bridge the gap between these platforms. Techniques like spatial FFPE-ATAC-seq [80] and single-cell sequencing of FFPE tissues [74] are rapidly advancing, unlocking the potential of vast archival collections for spatial tumor research. By understanding the specific impacts of each preservation method and implementing appropriate protocols, researchers can maximize the scientific return from both fresh and archived specimens in spatial studies of tumor architecture.

Spatial transcriptomics has emerged as a pivotal technology for investigating tumor organization architecture, enabling the precise mapping of gene expression within intact tissue sections. This capability is particularly crucial for deciphering tumor heterogeneity, immune microenvironment composition, and cellular communication networks that drive cancer progression and therapeutic resistance [52]. Unlike traditional bulk or single-cell RNA sequencing that requires tissue dissociation and loses spatial context, spatial transcriptomics preserves the architectural relationships between malignant, stromal, and immune cells within the tumor ecosystem [39] [52]. However, the immense value of spatially resolved data comes with significant computational challenges that researchers must overcome to extract biologically meaningful insights.

The computational pipeline for spatial transcriptomics begins with raw data generation from either sequencing-based platforms (e.g., 10X Visium, Stereo-seq) or imaging-based platforms (e.g., Xenium, MERSCOPE, CosMx), each producing distinct data types and analytical challenges [35]. The subsequent steps include cell segmentation to assign transcripts to individual cells, spatial gene expression analysis to identify patterns and gradients, cell-type deconvolution for multi-cell resolutions, and cellular communication inference to map interaction networks [81]. Each stage demands specialized computational methods and robust data management strategies, particularly as datasets grow in size and complexity. For tumor biology research, accurately resolving these computational challenges is essential for identifying novel therapeutic targets, understanding mechanisms of resistance, and developing predictive biomarkers for personalized cancer treatment [52].

Core Computational Challenges and Solutions

Cell Segmentation: From Pixels to Cells

Cell segmentation represents the foundational computational step in imaging-based spatial transcriptomics, where individual RNA molecules must be accurately assigned to their cell of origin. Inaccurate segmentation leads to misassignment of mRNAs, introducing errors in downstream analyses such as cellular phenotyping, differential expression, and cell-cell communication inference [82]. This challenge is particularly acute in tumor tissues characterized by high cellular density, complex morphology, and heterogeneous cell types.

Table 1: Comparison of Cell Segmentation Methods for Spatial Transcriptomics

Method	Algorithm Type	Required Inputs	Key Advantages	Limitations
Proseg [83]	Probabilistic model	Nuclei staining, RNA locations	Reduces suspicious gene co-expression; improves T-cell detection in tumors	Platform-specific adaptation needed
BOMS [82]	Mean shift algorithm	RNA spatial locations and gene labels	No auxiliary image needed; fast execution; simple implementation	May struggle with highly transcriptionally similar adjacent cells
Baysor [82]	Bayesian mixture modeling	RNA locations (optional: auxiliary image)	Elegant mathematical foundation; flexible confidence in auxiliary data	Long runtimes on large datasets; difficult parameter tuning
BIDCell [82]	Deep learning	RNA data, scRNA-seq reference, marker genes	Incorporates biological prior knowledge	Requires single-cell reference and marker gene knowledge
Cellpose [82]	Deep learning	Nuclei or membrane staining	Excellent nuclei segmentation performance	Does not capture full cell body; unassigned mRNAs

The segmentation challenge has spurred development of innovative computational approaches. Proseg, a recently developed tool, utilizes a probabilistic model that defines cell boundaries based on RNA transcript distribution patterns. It leverages the principle that RNA transcripts are typically randomly distributed throughout the cell, simulating cells that best explain the observed transcript distribution using a Cellular Potts Model approach. Validation studies demonstrated that Proseg significantly reduces the frequency of biologically implausible gene co-expression pairs compared to existing methods and has revealed previously undetected T-cell populations in renal cell carcinoma samples due to improved segmentation accuracy [83].

Alternative approaches like BOMS (Based On Mean Shift) offer segmentation without requiring auxiliary images by leveraging the spatial locations and gene labels of mRNA spots. The algorithm operates on the principle that molecules belonging to the same cell form local neighborhoods that are transcriptionally similar. It computes Neighborhood Gene Expression (NGE) vectors for each molecule based on its k nearest neighbors, then uses a mean shift procedure to identify modes in the joint spatial-NGE domain, effectively grouping molecules that converge to the same mode into individual cells [82]. This method demonstrates particularly value for complex tissues where high-quality staining is difficult to obtain.

Analytical Frameworks for Spatial Data

Beyond segmentation, spatial transcriptomics data demands specialized analytical approaches that leverage spatial information to derive biological insights. Key computational challenges include identifying spatially variable genes, delineating tumor microenvironment domains, inferring cell-cell communication, and integrating multi-omic data.

Spatially variable gene (SVG) detection methods identify genes whose expression exhibits significant spatial patterns, which may correspond to functional niches within the tumor microenvironment. Methods employing Gaussian processes, generalized linear models, and spatial autocorrelation analysis can classify different patterns of spatial variation such as linear gradients or periodic expression, potentially revealing mechanisms of tumor-immune interaction and microenvironment-driven gene regulation [81].

For sequencing-based technologies with multi-cellular resolution, computational deconvolution approaches are essential to infer cell-type proportions within each spatial spot. These methods typically integrate cell-type-specific transcriptomic profiles from single-cell RNA sequencing references, enabling resolution of cellular heterogeneity within the constraints of the spatial technology's resolution [81]. More advanced methods now incorporate spatial information directly into the deconvolution process, improving accuracy by leveraging the similarity between neighboring spots.

Table 2: Computational Methods for Spatial Transcriptomics Data Analysis

Analytical Task	Computational Approach	Key Applications in Tumor Research
Spatially Variable Gene Identification	Gaussian processes, spatial autocorrelation analysis [81]	Identifying tumor niche-specific expression patterns, microenvironment gradients
Cell-Cell Communication Inference	Graph convolutional networks, optimal transport, spatial cross-correlation [81]	Mapping tumor-immune interactions, paracrine signaling networks
Spatial Domain Detection	Hidden Markov random fields, graph-based clustering	Delineating tumor regions, immune niches, stromal compartments
Multi-omics Integration	Multi-view learning, manifold alignment [3]	Linking spatial gene expression with protein activity, genetic alterations
Trajectory Inference	RNA velocity in situ, spatial pseudotime	Modeling tumor evolution, cell state transitions across spatial contexts

The integration of spatial transcriptomics with histology images represents another promising analytical frontier. Deep learning approaches like MISO (Multiscale Integration of Spatial Omics) can predict spatial gene expression directly from H&E-stained histological slides, potentially enabling spatial transcriptomic analysis from vast archives of existing clinical specimens [5]. Such methods significantly expand the potential for retrospective studies linking long-term clinical outcomes with spatial tumor organization.

Managing Large-Scale Spatial Data

The data management challenges in spatial transcriptomics are substantial, with imaging-based technologies generating terabytes of raw image data per experiment [81]. Effective data handling requires specialized computational infrastructure and optimized processing pipelines.

Cloud computing platforms have emerged as essential resources for managing spatial transcriptomics data, providing scalable storage and computational resources that can accommodate the massive datasets [84]. The democratization of data access through cloud platforms enables researchers worldwide to analyze large spatial datasets without requiring extensive local computational infrastructure. Containerization technologies like Docker and Singularity further enhance reproducibility by encapsulating complete analytical environments.

Data compression strategies are particularly important for spatial transcriptomics, given the size of raw imaging files. Efficient file formats optimized for sparse spatial data can significantly reduce storage requirements while maintaining fast access for analysis. Establishing centralized data repositories with standardized organization principles will be critical for sharing spatial transcriptomics data across the research community [81].

Experimental Protocols for Tumor-Focused Spatial Transcriptomics

Integrated Workflow for Tumor Microenvironment Analysis

A comprehensive spatial analysis of tumor architecture requires careful experimental design and computational execution. The following protocol outlines an integrated approach for characterizing cellular organization and interactions in the tumor microenvironment.

Figure 1: Integrated computational workflow for spatial analysis of tumor architecture

Sample Preparation and Technology Selection

• Tissue Processing: Optimal spatial transcriptomics requires careful tissue preservation. For frozen tissues, optimal cutting temperature (OCT) compound embedding with rapid freezing preserves RNA integrity. For FFPE tissues, standard clinical pathology protocols are compatible with newer spatial technologies like 10X Visium HD and Xenium [35] [52].
• Platform Selection: Choose spatial technology based on resolution requirements and sample type. For discovery-phase studies requiring whole transcriptome coverage, sequencing-based approaches like Visium HD provide genome-wide profiling. For high-resolution validation studies, imaging-based platforms like Xenium or CosMx offer subcellular resolution [35].

Computational Processing Pipeline

• Data Preprocessing: For sequencing-based data, process raw FASTQ files using space-aware alignment tools (e.g., SpaceRanger) that assign reads to spatial barcodes. For imaging-based data, perform image segmentation and spot calling using platform-specific pipelines [81].
• Cell Segmentation: Apply segmentation algorithms appropriate for your data type and quality. For tissues with clear nuclear staining, Proseg or Cellpose provide excellent performance. For tissues without high-quality staining, staining-free methods like BOMS offer a viable alternative [83] [82].
• Quality Control: Implement rigorous QC metrics including genes per cell, counts per cell, mitochondrial percentage, and segmentation confidence scores. Identify and exclude poor-quality cells or regions [81].

Spatial Analysis and Interpretation

• Cell Type Annotation: Integrate single-cell RNA sequencing references to annotate cell types using transfer learning approaches. Incorporate prior knowledge of cell-type-specific marker genes for validation.
• Spatial Pattern Analysis: Identify spatially variable genes using methods like spatial autocorrelation analysis. Detect spatial domains with similar cellular composition or gene expression patterns [81].
• Cell-Cell Communication: Infer ligand-receptor interactions between spatially proximal cells using tools that incorporate spatial proximity rather than just transcriptional similarity [81].

Deep Learning-Based Prediction from Histology Images

The integration of spatial transcriptomics with digital pathology represents a powerful approach for leveraging extensive histology archives. The MISO framework demonstrates how deep learning can predict spatial gene expression patterns directly from H&E-stained images [5].

Figure 2: Deep learning workflow for predicting spatial gene expression from H&E images

Implementation Protocol

• Training Data Preparation: Curate paired H&E images and spatial transcriptomics data from the same tissue section. The MISO model was trained on 72 10X Genomics Visium samples and validated on 348 samples from five cancer indications [5].
• Multiscale Feature Extraction: Process H&E images at multiple resolutions to capture both cellular and tissue-level features. Convolutional neural networks extract morphological features correlated with gene expression patterns.
• Model Training and Validation: Train deep learning models to predict spatial gene expression from image features. Validate predictions using held-out spatial transcriptomics measurements and through biological validation such as comparison to known spatial patterns of key oncogenes and tumor suppressor genes.

Table 3: Research Reagent Solutions for Computational Spatial Transcriptomics

Resource Category	Specific Tools	Function and Application
Cell Segmentation Tools	Proseg [83], BOMS [82], Baysor [82], Cellpose [82]	Assigning RNA molecules to individual cells based on spatial distributions and transcriptional profiles
Spatial Analysis Platforms	MISO [5], SPIRAL [5], CytoSPACE [5]	Predicting gene expression from histology, data integration across technologies, spatial alignment
Cloud Computing Resources	AWS, Google Cloud, Azure [84]	Providing scalable computational infrastructure for large dataset storage and analysis
Data Visualization Tools	Giotto, Squidpy, Vitessce	Visualizing spatial gene expression patterns, cellular neighborhoods, and tissue domains
Reference Datasets	HEST-1k [5], TCGA [5], MOSAIC Consortium [5]	Providing benchmark data for method development and validation across diverse tumor types

The computational challenges in spatial transcriptomics represent significant but surmountable hurdles in the quest to comprehensively characterize tumor architecture. Advances in cell segmentation algorithms like Proseg and BOMS are improving the accuracy of cellular profiling, while innovative analytical frameworks are unlocking the potential of spatial data to reveal new biology. As these computational methods mature and become more accessible, they promise to transform our understanding of tumor organization, progression, and therapeutic response. The integration of artificial intelligence with spatially resolved data particularly powerful for extracting maximum information from precious clinical samples, potentially accelerating the development of novel cancer diagnostics and therapeutics. For the research and drug development community, embracing these computational approaches will be essential for fully leveraging the power of spatial biology in oncology.

Leveraging AI and Machine Learning for Automated Feature Extraction and Spatial Pattern Recognition

In oncology, tumors are not merely aggregates of malignant cells but complex, organized tissues with intricate spatial architectures. The spatial relationships between cancer cells, immune cells, stromal components, and vasculature create specialized microenvironments that critically influence disease progression, therapeutic response, and resistance mechanisms [52]. Spatial transcriptomics (ST) has emerged as a revolutionary technology that enables the precise quantification and visualization of gene expression within the intact spatial context of tissues, unlike conventional transcriptomics which loses this crucial architectural information [52]. This capability is particularly vital for genitourinary cancers (e.g., prostate, bladder, kidney), which demonstrate significant spatial heterogeneity affecting treatment resistance and immune evasion [52]. The integration of Artificial Intelligence (AI) and Machine Learning (ML) with ST data is now pushing the boundaries of our understanding, allowing for the automated extraction of biologically meaningful features and the recognition of complex spatial patterns that were previously inaccessible. These advanced computational methods are transforming raw, high-dimensional spatial omics data into actionable biological insights, thereby accelerating discovery in tumor biology and drug development.

AI and Machine Learning Foundations for Spatial Data

Core Concepts in AI Feature Extraction

AI feature extraction is a fundamental process in machine learning that converts raw data into a set of meaningful, non-redundant features that effectively represent the underlying information for algorithmic processing [85]. In the context of spatial biology, this involves identifying and isolating characteristic spatial patterns or structures within data, such as tissue images or spatial gene expression matrices [86]. The primary goals are to reduce data dimensionality, eliminate redundancy, enhance model performance, and improve interpretability [85]. This process is crucial for managing the enormous scale and complexity of spatial transcriptomics datasets, where manual analysis is infeasible.

Various feature types require distinct processing approaches. Numerical features (e.g., gene expression counts) enable precise mathematical computations, while categorical features (e.g., cell type classifications) provide essential distinctions between biological classes [85]. The features most relevant to spatial transcriptomics include spatial point patterns (cell locations), textural features (tissue morphology), and interaction features (cell-cell communication metrics) that collectively describe the tumor ecosystem.

Machine Learning Techniques for Spatial Pattern Recognition

Multiple machine learning techniques have been adapted and developed specifically for spatial pattern recognition in biological contexts:

• Convolutional Neural Networks (CNNs) automatically extract hierarchical features from images, identifying patterns from simple edges to complex shapes through layered filters [85] [86]. In spatial transcriptomics, CNNs analyze histology images from H&E-stained slides to predict spatial gene expression patterns [5].

• Graph Neural Networks (GNNs) process data structured as graphs, making them ideal for modeling cellular neighborhoods and interaction networks where cells represent nodes and spatial proximities represent edges [5].

• Transformer architectures with attention mechanisms capture long-range dependencies within tissue sections, effectively modeling interactions between distant but biologically connected tissue regions [5].

• Autoencoders serve as powerful tools for dimensionality reduction, learning compressed representations of high-dimensional spatial data while preserving biologically relevant information [85]. These are particularly valuable for identifying latent patterns in spatial omics datasets.

• Hybrid approaches that combine multiple architectures, such as transformers with graph neural networks, have demonstrated superior performance in predicting spatial gene expressions from histology images by jointly modeling local and global tissue contexts [5].

AI-Driven Methodologies in Spatial Transcriptomics

Deep Learning for Multiscale Data Integration

A significant challenge in spatial biology is integrating information across multiple scales, from subcellular features to tissue-level organization. The MISO (Multiscale Integration of Spatial Omics) framework represents a cutting-edge deep learning approach that addresses this challenge by predicting spatial transcriptomics data from routinely available H&E-stained histology slides [5]. This methodology demonstrates how AI can leverage existing pathological resources to generate spatially resolved molecular information.

The MISO framework employs a sophisticated pipeline that processes whole slide images (WSIs) through a deep learning network trained on matched H&E and spatial transcriptomics data from 72 10X Genomics Visium samples [5]. The model learns the complex relationships between tissue morphology and gene expression patterns, enabling it to predict spatial gene expression from H&E morphology alone. When validated on 348 samples across five cancer indications from the MOSAIC consortium, MISO significantly outperformed competing methods in extensive benchmarks [5]. This approach demonstrates particular strength in predicting spatially variable genes and capturing biological processes with clear morphological correlates, such as immune infiltration and stromal reactions.

Table 1: Commercial Spatial Transcriptomics Platforms Enabled by AI Analysis

Platform	Company	Methodology	Resolution	Maximum Targets	Best for AI Applications
Xenium	10x Genomics	Padlock probe with rolling circle amplification	Subcellular	5000 RNAs	High-plex subcellular mapping
Visium	10x Genomics	Spatially barcoded spots for mRNA capture	55 μm (single-cell with HD)	All 3' mRNA	Whole transcriptome analysis
CosMx	NanoString	Branched DNA probes with multiple readout sequences	Subcellular	18,000+ RNAs	Ultra-high-plex single-cell analysis
MERSCOPE	Vizgen	Multiple probes per RNA with unique readout sequences	Subcellular	1000 RNAs	Single-molecule imaging
GeoMx	NanoString	UV-cleavable oligo tags on probes	Region of Interest	18,000+ RNAs	High-throughput discovery

AI-Powered Feature Extraction Techniques

AI enables several sophisticated feature extraction paradigms specifically designed for spatial transcriptomics data:

Spatial Gene Expression Prediction: Deep learning models like MISO [5] and SEPAL [5] can predict spatial gene expression patterns directly from histological images. These models typically use a CNN backbone (e.g., ResNet) to extract visual features from tissue tiles, which are then processed through transformer or graph neural network layers to model spatial dependencies and predict gene expression values for each spatial location.

Cellular Neighborhood Identification: Unsupervised and self-supervised learning methods cluster cells or tissue regions based on their spatial transcriptomic profiles to identify recurrent cellular neighborhoods – spatially coherent units with distinct biological functions. AI methods enhance this by simultaneously considering gene expression, spatial proximity, and morphological context.

Cell-Cell Interaction Inference: Graph neural networks model tissue sections as spatial graphs where cells represent nodes and physical proximities define edges. These models can then infer communication patterns based on ligand-receptor co-expression in spatially proximal cells, revealing tumor-immune interactions and stromal signaling networks.

Domain Adaptation from Histology: As demonstrated by MISO [5], domain adaptation techniques enable knowledge transfer from widely available H&E-stained histological slides to spatial transcriptomics domains. This is particularly valuable given that H&E slides are routinely generated for most cancer patients, while spatial transcriptomics remains a specialized, costly technology.

Experimental Protocols for AI-Enhanced Spatial Analysis

Protocol 1: Predicting Spatial Transcriptomics from H&E Morphology

This protocol is based on the MISO methodology [5] and enables researchers to infer spatial gene expression from standard histology slides.

• Sample Preparation: Collect paired H&E-stained whole slide images (WSIs) and spatial transcriptomics data from the same tissue section. For validation studies, 10X Genomics Visium provided ground truth data [5].

• Data Preprocessing:

  • Tile H&E images into smaller patches (e.g., 256×256 pixels) at multiple magnification levels (e.g., 5X, 10X, 20X).
  • Align spatial transcriptomics spots with corresponding H&E image regions.
  • Normalize gene expression counts using standard methods (e.g., logCPM, SCTransform).

• Model Architecture:

  • Implement a multi-scale CNN (e.g., ResNet50) to extract features from H&E tiles at different resolutions.
  • Incorporate transformer layers with attention mechanisms to model spatial relationships between tissue regions.
  • Include graph neural network components to capture neighborhood interactions.

• Training Procedure:

  • Train the model using paired H&E and spatial transcriptomics data.
  • Employ a mean squared error loss between predicted and actual gene expression.
  • Use cross-validation across tissue sections to assess generalizability.

• Validation:

  • Benchmark against competing methods using metrics like root mean square error (RMSE) and correlation coefficients.
  • Perform biological validation by confirming that predictions recapitulate known spatially variable genes.

Protocol 2: Spatially Resolved Cell-Type Deconvolution

This protocol enables the identification of cell types within spatial transcriptomics spots that typically contain multiple cells.

• Reference Generation:

  • Generate a single-cell RNA sequencing reference atlas from similar tissue types.
  • Annotate cell types using established marker genes.

• Integration Framework:

  • Implement a conditional variational autoencoder (cVAE) to integrate single-cell and spatial transcriptomics data.
  • Use spatial coordinates as conditional variables to maintain spatial context.

• Deconvolution:

  • Model each spatial spot as a mixture of cell types from the reference atlas.
  • Estimate cell-type proportions using non-negative matrix factorization or Bayesian approaches.

• Spatial Pattern Analysis:

  • Apply spatial autocorrelation statistics (e.g., Moran's I) to identify non-random distributions of cell types.
  • Construct spatial graphs to identify recurrent cellular neighborhoods.

Table 2: AI Model Architectures for Spatial Transcriptomics Analysis

Model Type	Primary Application	Key Advantages	Implementation Considerations
Convolutional Neural Networks (CNNs)	Image-based feature extraction from histology	Hierarchical feature learning, translation invariance	Requires large datasets, GPU acceleration
Graph Neural Networks (GNNs)	Modeling cell-cell interactions	Naturally models spatial relationships, flexible topology	Graph construction critical for performance
Transformers	Long-range spatial dependencies	Attention mechanisms, excellent scalability	Computationally intensive for large tissues
Autoencoders	Dimensionality reduction, denoising	Learns compressed representations, removes noise	Risk of losing biologically relevant information
Hybrid Models (CNN+GNN)	Multimodal data integration	Combines visual and spatial information	Complex training, potential overfitting

The Scientist's Toolkit: Essential Research Reagents and Platforms

Successful implementation of AI-driven spatial transcriptomics requires both wet-lab reagents and computational tools. The following table details essential components of the spatial biology workflow.

Table 3: Research Reagent Solutions for Spatial Transcriptomics

Item	Function	Example Products/Platforms
Spatial Barcoding Beads	Capture location-tagged mRNA from tissue sections	10x Genomics Visium Gene Expression Slide
Morphology Preservation Buffers	Maintain tissue architecture during processing	Visium Tissue Preservation Solution
Permeabilization Reagents	Enable mRNA release from fixed tissues	Visium Permeabilization Enzyme
Probe Sets	Target-specific oligonucleotides for transcript detection	CosMx Human Whole Transcriptome Panel
Fluorescent Reporters	Visualize spatial gene expression patterns	Readout Fluorescent Tags (MERSCOPE)
Multiomic Integration Panels	Simultaneous detection of RNA and protein	GeoMx Discovery Proteome Atlas (1,100+ plex protein assay)
Nucleic Acid Amplification Kits	Signal amplification for low-abundance transcripts	Hybridization Chain Reaction (HCR) Amplification
Library Preparation Kits	Prepare sequencing libraries from barcoded cDNA	Visium Spatial Gene Expression Library Kit
Image Analysis Software	Process and visualize spatial omics data	MISO Pipeline, Giotto, Seurat, SpaceRanger
AI Modeling Frameworks	Implement machine learning for pattern recognition	PyTorch, TensorFlow, Scanpy, Squidpy

Visualization and Computational Workflows

Effective visualization is crucial for interpreting the complex spatial relationships uncovered by AI methodologies. The following workflow represents an integrated pipeline for AI-powered spatial transcriptomics analysis.

Future Frontiers and Implementation Challenges

The integration of AI with spatial transcriptomics presents several implementation challenges that researchers must address. Computational resource requirements are substantial, as processing whole slide images and spatial transcriptomics datasets demands significant GPU memory and storage capacity [5]. Data integration complexity arises when combining multimodal data sources (histology, transcriptomics, proteomics) with different resolutions and noise profiles [5] [52]. Interpretability and explainability remain crucial for biological validation, as complex deep learning models can function as "black boxes" without clear mechanistic insights [85]. Additionally, technical variability between platforms, batches, and experimental conditions requires careful normalization and domain adaptation approaches [52].

Future developments are likely to focus on several key frontiers. Multimodal foundation models pre-trained on large-scale histology and omics data will enable transfer learning for specific cancer types with limited data [5]. Spatial dynamical modeling will extend beyond static snapshots to model temporal changes in tumor architecture during treatment and progression. Clinical translation will see increased development of AI-driven spatial biomarkers for diagnosis, prognosis, and treatment selection, particularly in immuno-oncology [3] [52]. Finally, real-time analysis platforms will emerge, integrating spatial omics with AI for intraoperative decision support and rapid diagnostic pathology.

As these technologies mature, the combination of AI and spatial transcriptomics will fundamentally enhance our understanding of tumor organization, enabling more precise targeting of cancer's spatial vulnerabilities and advancing the development of next-generation therapeutics that account for the complex architectural principles of human tumors.

From Data to Discovery: Validating Spatial Findings and Cross-Platform Comparisons

The intricate spatial organization of solid tumors is a critical regulator of cancer progression, therapeutic response, and patient prognosis. While spatial transcriptomics (ST) has revolutionized our ability to map gene expression within intact tissue architecture, transcriptomic data alone provides an incomplete picture of the tumor microenvironment (TME). Establishing reliable ground truth datasets through the integration of ST with protein-level data from technologies like CODEX (Co-Detection by indEXing) and histological validation is paramount for advancing spatially resolved cancer research. This multi-omic approach bridges the gap between molecular expression, protein function, and tissue morphology, enabling researchers to decipher the complex cellular communication networks and functional niches that define tumor biology. The correlation of these complementary data types ensures that transcriptional signatures are contextualized within their protein and morphological frameworks, significantly enhancing the biological relevance and translational potential of discoveries in precision oncology.

Core Methodologies for Multi-Omic Spatial Profiling

Spatial Transcriptomics Platforms and Principles

Spatial transcriptomics technologies can be broadly categorized into two classes: sequencing-based (sST) and imaging-based (iST) platforms [27]. Sequencing-based methods, such as Visium HD (10x Genomics) and Stereo-seq (BGI), capture polyadenylated RNA using spatially barcoded poly-dT oligonucleotides arrayed on a surface, enabling unbiased whole-transcriptome analysis [49]. In contrast, imaging-based platforms, including CODEX, CosMx (NanoString), and Xenium (10x Genomics), utilize iterative hybridization of fluorescently labeled probes with sequential imaging to profile hundreds to thousands of genes at single-molecule resolution within intact tissue sections [27] [33]. A key advantage of iST platforms is their inherent compatibility with protein co-detection, either through antibody-based methods or integrated multimodal assays.

CODEX Multiplexed Protein Profiling

CODEX (Co-Detection by indEXing) is a highly multiplexed protein imaging technology that enables simultaneous detection of dozens of protein markers in formalin-fixed paraffin-embedded (FFPE) or frozen tissue sections while preserving spatial context [49]. The methodology utilizes a library of DNA-barcoded antibodies that are hybridized simultaneously and detected through successive rounds of fluorescent imaging with complementary fluorescently labeled oligonucleotides. This iterative staining and imaging process allows for the precise spatial localization of numerous protein epitopes within complex tissues. The resulting high-dimensional protein data serves as an ideal ground truth for validating protein-level expression patterns inferred from ST data, particularly for cell-type identification, cellular state characterization, and cell-cell interaction analysis.

Integrated Experimental Workflow for Ground Truth Establishment

Sample Preparation and Tissue Sectioning

Robust multi-omic integration begins with meticulous sample preparation. For comprehensive studies, tumor samples should be divided and processed into both FFPE blocks and fresh-frozen optimal cutting temperature (OCT) compound-embedded blocks to accommodate the specific requirements of different ST platforms and CODEX [49]. Serial tissue sections (typically 4-10 μm thick) are then cut from adjacent regions of the same tissue block and allocated to different technologies—one section for ST, the immediately adjacent section for CODEX, and subsequent sections for H&E staining and other histological analyses. This serial sectioning approach is critical for ensuring that similar cellular regions are profiled across modalities, enabling direct cross-platform comparison.

Technical Optimization for Multi-Omic Assays

Successful integration requires careful optimization of experimental conditions to balance mRNA preservation with protein epitope integrity. Key considerations include:

• Fixation Conditions: While standard ST protocols often use methanol fixation for optimal mRNA preservation, this is suboptimal for antibody-based protein detection. Paraformaldehyde (PFA) fixation better preserves protein epitopes but can reduce mRNA accessibility. Optimization of PFA concentration and fixation time is essential [87].
• Permeabilization Enhancement: Standard tissue permeabilization enzymes may be insufficient for PFA-fixed tissues. Combining the permeabilization enzyme with 1% sodium dodecyl sulfate (SDS) can significantly increase yields for both mRNA and antibody-derived tags (ADTs) while maintaining tissue architecture [87].
• Antibody Validation: For protein detection assays, antibodies must be rigorously validated for specificity in the tissue type of interest. Control experiments without primary antibodies and isotype-matched controls are essential to establish signal specificity.

The following workflow diagram illustrates the integrated experimental design for correlating ST with CODEX and histology:

Data Generation and Alignment Pipeline

Following sample processing, the next critical phase involves generating and aligning multi-omic data to establish spatial ground truth. The sequential steps in this pipeline ensure precise registration of transcriptional, protein, and histological information from adjacent tissue sections.

Benchmarking Spatial Transcriptomics Platforms Against CODEX-Derived Ground Truth

Systematic benchmarking studies have evaluated the performance of various high-throughput ST platforms against CODEX-derived protein ground truth. The following table summarizes key performance metrics across four advanced platforms, based on a comprehensive analysis of colon adenocarcinoma, hepatocellular carcinoma, and ovarian cancer samples [49]:

Table 1: Performance Benchmarking of ST Platforms Against CODEX Protein Ground Truth

Platform	Technology Type	Resolution	Gene Panel Size	Sensitivity for Marker Genes	Correlation with scRNA-seq	Concordance with CODEX
Visium HD FFPE	Sequencing-based (sST)	2 μm	18,085 genes	Moderate to High	High	High
Stereo-seq v1.3	Sequencing-based (sST)	0.5 μm	Whole transcriptome	Moderate	High	High
Xenium 5K	Imaging-based (iST)	Single molecule	5,001 genes	High	High	High
CosMx 6K	Imaging-based (iST)	Single molecule	6,175 genes	Moderate	Moderate	Moderate

The evaluation of molecular capture efficiency reveals important distinctions between platforms. When assessing shared regions across FFPE serial sections, Xenium 5K consistently demonstrated superior sensitivity for multiple marker genes compared to other platforms [49]. Gene-wise correlation analysis with matched single-cell RNA sequencing (scRNA-seq) data showed that Stereo-seq v1.3, Visium HD FFPE, and Xenium 5K maintained high correlations with scRNA-seq profiles, while CosMx 6K showed substantial deviation despite detecting a higher total number of transcripts [49].

Successful integration of ST and CODEX data requires carefully selected reagents, platforms, and computational tools. The following table details essential components of the multi-omic spatial profiling toolkit:

Table 2: Research Reagent Solutions for Multi-Omic Spatial Profiling

Category	Specific Product/Platform	Function/Application	Key Considerations
Spatial Transcriptomics Platforms	10x Genomics Visium HD	Whole transcriptome mapping at 2μm resolution	Compatible with FFPE and fresh frozen tissues
	NanoString CosMx 6K	Targeted transcriptomics with single-cell resolution	6,175-plex RNA panel with protein co-detection capability
	10x Genomics Xenium 5K	In-situ analysis of 5,001 genes	Optimized for FFPE tissues with integrated morphology analysis
Multiplex Protein Imaging	CODEX/IBEX systems	Highly multiplexed protein detection (50+ markers)	Enables immune cell phenotyping and spatial neighborhood analysis
	Akoya Phenocycler	Whole-slide multiplexed protein imaging	Suitable for discovery and validation phases
Antibody Resources	DNA-barcoded antibodies (CODEX)	Multiplexed protein detection via DNA barcoding	Require validation for specific tissue types and fixation conditions
	CITE-seq/SPOTS antibodies	Simultaneous protein and transcript detection	Polyadenylated antibody-derived tags for sequencing-based detection
Computational Tools	SpaLinker	Links spatial TME features to clinical phenotypes	Identifies tertiary lymphoid structures and tumor-normal interfaces
	Giotto Suite	Comprehensive ST data analysis	Spatial domain detection, cell-cell communication analysis
	SPATA	Spatial transcriptomics analysis framework	Integrates with single-cell references for cell type decomposition

Analytical Framework for Cross-Modal Data Integration

Spatial Registration and Coordinate Alignment

The first critical step in multi-omic integration is the precise spatial alignment of datasets generated from adjacent tissue sections. This process involves:

• Landmark Identification: Using histological features such as blood vessels, tissue boundaries, or prominent morphological structures as reference points across serial sections.
• Non-linear Transformation: Applying advanced image registration algorithms (e.g., elastic, B-spline) to align the coordinate systems of ST, CODEX, and histological images, accounting for tissue distortion during sectioning.
• Cellular-Level Alignment: For high-resolution data, implementing cell segmentation based on DAPI or H&E staining to enable single-cell cross-modal correlation when combined with nuclear staining references.

Multi-Modal Cell Type Annotation and Validation

CODEX protein data provides an essential ground truth for validating and refining cell type annotations derived from ST data. The integrated analytical approach includes:

• Protein-Guided Clustering: Using protein expression patterns from CODEX to inform clustering parameters for ST data, particularly for immune cell subsets that may have distinct protein signatures but overlapping transcriptional profiles.
• Cross-Modal Marker Validation: Identifying concordant and discordant patterns between mRNA and protein expression for key cell type markers (e.g., CD3ε, CD4, CD8 for T cells; CD19, CD20 for B cells; EpCAM for epithelial cells) [87].
• Spatial Distribution Analysis: Comparing the spatial distributions of specific cell populations identified through protein markers (CODEX) with transcriptional signatures (ST) to identify regions of agreement and biological divergence.

Signaling Pathway and Cellular Interaction Analysis

The integration of transcriptional and protein data enables more robust analysis of active signaling pathways and cell-cell communication:

• Ligand-Receptor Interaction Mapping: Combining expression of ligand and receptor pairs at both mRNA and protein levels to identify functionally active communication axes within the TME.
• Pathway Activity Inference: Using protein expression and phosphorylation status (when available) to validate inferred pathway activity from transcriptional signatures.
• Spatial Neighborhood Analysis: Defining cellular neighborhoods based on both transcriptional and protein signatures, then examining how these neighborhoods correlate with histological features and clinical outcomes.

Applications in Cancer Research and Clinical Translation

Identification of Therapeutically Relevant Spatial Features

The correlation of ST with CODEX has enabled the discovery and validation of spatial features with clinical significance:

• Tertiary Lymphoid Structures (TLS): Integrated analysis has revealed TLS as organized immune aggregates containing B cells, T cells, and dendritic cells in specific spatial arrangements, which correlate with improved response to immunotherapy across multiple cancer types [11]. Tools like SpaLinker leverage both gene expression and spatial information to accurately identify TLS regions and link them to patient outcomes [11].
• Tumor-Normal Interface (TNI) Regions: The spatial interface between tumor and normal tissue harbors unique cellular communities and molecular gradients that influence invasion and metastasis. SPOTS analysis has revealed specialized macrophage subsets (CD169+) and fibroblast populations (CD29+) occupying distinct spatial niches at these interfaces [87].
• Immunosuppressive Niches: Combined protein and transcript profiling has identified spatial compartments enriched for immunosuppressive cell types (Tregs, M2 macrophages) and checkpoint expression (PD-1, PD-L1, CTLA-4) that may represent resistance mechanisms to immunotherapy [33].

Biomarker Discovery and Validation

The multi-omic ground truth approach enhances biomarker discovery by:

• Differentiating Functional States: Identifying markers that distinguish cell states (e.g., exhausted vs. activated T cells) through correlated protein and RNA expression patterns.
• Spatial Contextualization: Determining whether biomarker expression is diffuse or restricted to specific spatial contexts with functional implications.
• Prognostic Stratification: Developing spatial signatures that integrate both transcriptional and protein information to improve patient stratification beyond conventional histopathological grading.

The establishment of ground truth through correlation of spatial transcriptomics with CODEX protein profiling and histology represents a paradigm shift in cancer research. This multi-omic framework moves beyond singular molecular perspectives to provide a comprehensive, spatially resolved understanding of tumor ecosystems. As these technologies continue to evolve, several exciting directions emerge: the development of fully integrated assays that simultaneously capture RNA and protein from the same tissue section, the implementation of artificial intelligence for automated pattern recognition across modalities, and the creation of standardized reference maps for normal and diseased tissues. Ultimately, the rigorous correlation of transcriptional, proteomic, and morphological information will accelerate the translation of spatial oncology discoveries into clinically actionable insights, paving the way for more precise diagnostic and therapeutic strategies in cancer care.

Spatial transcriptomics (ST) has emerged as a revolutionary technological paradigm, integrating spatial data with transcriptomic information to generate high-resolution maps of gene expression within the intact architectural context of tissues [88]. This capability is fundamentally transforming cancer research by preserving the spatial relationships that are lost in single-cell RNA sequencing (scRNA-seq), thereby enabling unprecedented insights into cellular heterogeneity, intercellular interactions, and the functional organization of the tumor microenvironment (TME) [17] [23]. The TME comprises a complex ecosystem of malignant cells, immune cells, stromal cells, blood vessels, and extracellular matrix, all interacting in spatially coordinated ways that influence tumor progression, invasion, metastasis, and therapy response [13] [17]. Understanding this spatial architecture is critical, as the distribution of immune cells within the TME has demonstrated significant prognostic value and potential for predicting immunotherapy outcomes [13] [89].

The rapid development of ST platforms, however, presents both opportunities and challenges. These technologies can be broadly classified into imaging-based (iST) and sequencing-based (sST) approaches, each with distinct strengths and limitations concerning spatial resolution, transcriptome coverage, and sample compatibility [90] [49] [23]. Imaging-based methods, such as Xenium, Merscope, and Molecular Cartography, utilize multiplexed single-molecule RNA fluorescence in situ hybridization (smRNA-FISH) for targeted analysis with single-cell or subcellular resolution [90]. In contrast, sequencing-based methods like Visium capture transcripts using spatially barcoded arrays for unbiased whole-transcriptome analysis, though often at a coarser resolution that encompasses multiple cells per spot [90] [17]. This technological diversity makes selecting the appropriate platform for specific research objectives a complex decision, requiring careful consideration of parameters such as sensitivity, specificity, gene coverage, and the accuracy of transcript assignment to individual cells [90].

As ST technologies advance, a parallel boom has occurred in the development of statistical and computational frameworks designed to extract biologically meaningful patterns from the complex, high-dimensional data they generate [89] [88]. These frameworks are essential for moving beyond descriptive accounts of spatial organization to quantitative, validated models of tumor architecture and function. This guide focuses on introducing and detailing key validation frameworks, with a particular emphasis on SpaTopic and other complementary tools, providing researchers with the methodologies needed to advance spatial transcriptomics in tumor organization research.

The computational analysis of spatial transcriptomics data presents unique challenges, including managing multimodality (integrating gene expression with spatial coordinates and histology), high dimensionality, and spatial noise [23]. Several sophisticated frameworks have been developed to address these challenges. The table below summarizes the core tools discussed in this guide.

Table 1: Key Statistical and Computational Frameworks for Spatial Transcriptomics

Framework Name	Core Methodology	Primary Application	Key Advantages
SpaTopic [91] [92]	Bayesian topic modeling (Latent Dirichlet Allocation)	Identifying recurrent spatial patterns ("topics") in cell types across tissue images.	High interpretability, scalability to millions of cells, identifies biologically meaningful spatial domains.
Spatiopath [13]	Null-hypothesis framework extending Ripley's K function	Distinguishing statistically significant spatial associations from random cell distributions.	Robustly quantifies cell-cell and cell-tumor epithelium interactions; distinguishes real associations from fortuitous accumulations.
Cell2Spatial [93]	Information-theoretic gene selection & spatial likelihood modeling	Mapping single cells to spatial transcriptomics spots to reconstruct tissue architecture at single-cell resolution.	Effectively handles unmatched single-cell and ST datasets; improves signal fidelity and spatial coherence.
SpatialTopic [91]	Bayesian topic model with spatial priors	Decoding spatial tissue architecture from multiplexed images by integrating cell type and spatial information.	High computational efficiency (minutes for 100,000 cells); identifies recurrent spatial patterns like Tertiary Lymphoid Structures (TLS).

These frameworks represent a paradigm shift from simple descriptive analyses to robust, statistically grounded inference of spatial patterns. SpaTopic and SpatialTopic leverage topic modeling to reduce complexity and identify latent structures, while Spatiopath provides a rigorous statistical foundation for testing hypotheses about cellular interactions. Cell2Spatial addresses the critical need for enhanced resolution in sST data, enabling detailed reconstructions of tissue architecture [93]. Together, they form a powerful toolkit for validating and interpreting the spatial architecture of tumors.

Detailed Framework Analysis: SpaTopic Methodology and Workflow

SpaTopic is a statistical learning framework designed specifically to identify and annotate pathology-relevant spatial domains by harmonizing spot clustering and cell-type deconvolution [92]. Its power lies in integrating single-cell transcriptomics (scRNA-seq) with spatially resolved transcriptomics (SRT) data through a topic modeling approach, treating spatial domains as documents composed of different cell types (words) [92]. This allows it to stratify the TME into spatial domains with coherent cellular organization, moving beyond gene expression-based clustering alone.

The SpaTopic workflow consists of four methodical steps:

• Input and Deconvolution: The process begins with SRT data and matched scRNA-seq data with pre-existing cell-type annotations. SpaTopic first uses a deconvolution method (e.g., CARD) to infer the cell-type composition of each spot in the SRT data. Simultaneously, an unsupervised clustering method (e.g., STAGATE) is applied to aggregate spots into initial clusters based on their spatial gene expression profiles [92].

• Cell Type-Specific Scoring: Next, SpaTopic applies the Kolmogorov-Smirnov (KS) test to determine a cell type–specific enrichment score for each initial cluster. This generates a matrix (S matrix) that quantifies how specific each cell type is to each spatial cluster, leveraging the results from the deconvolution and clustering steps [92].

• Topic Modeling via LDA: The core of SpaTopic involves applying the Latent Dirichlet Allocation (LDA) model to decompose the S matrix into two probability distributions:

  • Topic-Cell Type Distribution (C1): This matrix defines the "cell-type topics," representing the probability of each cell type within a given topic. It reveals the predominant cell-type compositions that characterize functional units in the TME.
  • Cluster-Topic Distribution (C2): This matrix defines the probability that a spatial cluster belongs to a specific topic [92].

• Spatial Domain Annotation: Finally, the cluster-topic matrix is binarized, assigning each initial cluster to one or more specific topics, now termed CellTopics. This step refines the initial spot clusters into final spatial domains based on the learned cell-type topics, enabling the characterization and quantitative comparison of these domains across different SRT datasets [92].

Table 2: Experimental Protocol for Applying SpaTopic to Tumor Data

Step	Protocol Detail	Purpose & Rationale
1. Sample Preparation	Generate serial sections from tumor samples (FFPE or fresh-frozen).	To ensure compatibility with SRT platforms and matched scRNA-seq.
2. Data Generation	- Perform SRT (e.g., using 10x Visium, Xenium).- Perform scRNA-seq on the same or matched sample.	To acquire spatial gene expression data and a reference for cell type annotation.
3. Preprocessing	- Annotate cell types from scRNA-seq using standard clustering/markers.- Quality control of SRT data (filtering spots/genes).	To create a clean, annotated reference for deconvolution and topic modeling.
4. SpaTopic Execution	- Run deconvolution (CARD) and spatial clustering (STAGATE).- Execute the SpaTopic workflow to infer CellTopics.	To identify spatial domains based on coherent cell-type composition.
5. Validation	- Compare SpaTopic domains with manual histopathological annotations.- Validate using Adjusted Rand Index (ARI).	To quantitatively assess the accuracy and biological relevance of the identified domains.

SpaTopic has been rigorously validated, outperforming methods like STAGATE, SpaGCN, and BayesSpace in accurately capturing the underlying spatial organization of tissues, as measured by the Adjusted Rand Index (ARI) against manual annotations [92]. For example, in a human pancreatic ductal adenocarcinoma (PDAC) dataset, SpaTopic identified distinct CellTopics corresponding to cancer, stromal, ductal, and normal pancreatic regions. The cancer region (CellTopic3) was characterized by enrichment of neoplastic cells and fibroblasts, and its highly expressed genes showed significant enrichment in stromal and immune-related processes, providing insights into tumorigenesis and potential chemoresistance [92].

Experimental Protocols for Key Analyses

Implementing the computational frameworks described requires robust experimental design and data generation protocols. The following section details the methodologies for benchmarking ST technologies and for conducting spatial pattern analysis, which are foundational to any subsequent computational validation.

Protocol for Benchmarking Spatial Transcriptomics Platforms

Systematic benchmarking is crucial for selecting the appropriate ST technology and interpreting results accurately. A robust benchmarking protocol involves:

• Sample Preparation: Collect treatment-naïve tumor samples (e.g., colon adenocarcinoma, hepatocellular carcinoma). Divide each sample and process it into matched FFPE and fresh-frozen (OCT-embedded) blocks. Generate serial tissue sections (4-10 μm thickness) from these blocks for parallel profiling across multiple ST platforms and complementary assays [49].
• Multi-Omics Profiling:
  • ST Platforms: Process adjacent serial sections on the high-throughput platforms to be benchmarked (e.g., Stereo-seq v1.3, Visium HD FFPE, CosMx 6K, Xenium 5K). This controls for biological variability.
  • Ground Truth Data:
    • scRNA-seq: Perform single-cell RNA sequencing on dissociated cell suspensions from the same tumor sample to provide a cell-type annotated reference transcriptome [49].
    • Protein Profiling: Use multiplexed protein imaging (e.g., CODEX) on tissue sections adjacent to those used for ST to validate spatial patterns at the protein level [49].
    • Histology: Perform H&E staining and high-resolution imaging of consecutive sections. Manually annotate nuclear boundaries and tissue regions for segmentation validation [49].
• Performance Metrics: Systematically evaluate each platform across several critical metrics:
  • Sensitivity & Specificity: Assess the detection sensitivity for marker genes (e.g., EPCAM) and calculate the false discovery rate (FDR) using negative control probes [90] [49].
  • Transcript Diffusion: Quantify the degree of transcript diffusion away from the nucleus, which affects localization accuracy [49].
  • Cell Segmentation Accuracy: Compare automated segmentation results (using tools like Cellpose, Baysor, or Mesmer) against manually segmented nuclei from DAPI/H&E images [90] [49].
  • Concordance with scRNA-seq and CODEX: Evaluate the correlation of gene expression profiles with scRNA-seq and the spatial alignment of marker expression with CODEX protein data [49].

Protocol for Spatial Pattern Analysis with Spatiopath

Spatiopath provides a statistical framework for distinguishing significant spatial associations from random distributions. The experimental and analytical protocol is as follows:

• Data Input Preparation:
  • Cell Type Identification: From ST data (either from iST or deconvoluted sST data), identify and label all immune cells and tumor cells using known marker genes.
  • Tumor Region Segmentation: Manually or computationally segment the tumor epithelium boundary from the tissue image [13].
• Define Spatial Objects:
  • Let set A represent the spatial objects of interest (e.g., the closed 2D contour of the tumor epithelium boundary, or the coordinates of a specific immune cell type).
  • Let set B represent the coordinates of the immune cell population whose spatial association with A is being tested [13].
• Generalized Accumulation Function:
  • Spatiopath generalizes Ripley's K function to handle interactions between points (cells) and arbitrarily shaped objects (tumor boundaries). The generalized accumulation function counts the accumulation of points in B to the shapes in A, corrected for boundary effects [13].
• Null Hypothesis Testing:
  • The core of Spatiopath is a null hypothesis model where immune cells are randomly distributed. The framework computes whether the observed accumulation of B cells near A is statistically significant compared to this random null distribution, thereby distinguishing fortuitous accumulation from true spatial association [13].
• Distance Quantification:
  • The analysis outputs the physical distance at which significant spatial apposition occurs, for example, revealing that mast cells accumulate significantly within a specific micrometer range from the tumor epithelium, while T cells may be positioned farther away [13].

Signaling Pathways and Tumor-Microenvironment Interactions

A primary application of these frameworks is the dissection of signaling pathways and cellular communication within the TME. SpaTopic, for instance, not only identifies spatial domains but also enables the inference of communication patterns between them [92]. In a PDAC analysis, CellTopic1 (stromal region) showed blocked integrin signaling pathways and minimal interaction with other regions, consistent with the known role of PDAC stroma as a physical barrier [92]. Similarly, the identification of a conserved tumor-microenvironment interface enriched in cilia genes, as revealed by ST in zebrafish melanoma models and validated in human samples, suggests a specialized zone of tumor-stroma crosstalk potentially regulated by ETS-family transcription factors [17].

The integration of ST data with cell-cell communication tools (e.g., CellChat, NicheNet) allows for the mapping of ligand-receptor interactions across spatially defined domains. Spatiopath enhances this by quantitatively determining if interacting cell pairs are significantly co-localized or spatially segregated, adding a layer of statistical robustness to inferred communication networks [13] [92].

The Scientist's Toolkit: Essential Research Reagents and Materials

Successfully executing a spatial transcriptomics study with the described validation frameworks requires careful selection of reagents and platforms. The following table catalogs essential components for building a robust experimental pipeline.

Table 3: Research Reagent Solutions for Spatial Transcriptomics

Category	Item / Platform	Specification / Function	Considerations for Selection
Spatial Platforms	10x Visium / Visium HD	Sequencing-based; whole transcriptome; 55μm (HD: 2μm) resolution.	Ideal for unbiased discovery; resolution limits single-cell analysis in standard Visium [90] [23].
	Xenium, Merscope, CosMx	Imaging-based (smRNA-FISH); targeted panels; single-cell/subcellular resolution.	Best for high-resolution targeted studies; gene number limited by panel [90] [49] [23].
Sample Types	Fresh-Frozen (FF) Tissue	Snap-frozen tissue sections.	Often superior RNA integrity; compatible with most platforms [90].
	Formalin-Fixed Paraffin-Embedded (FFPE)	Archival clinical tissue samples.	Essential for translational studies; compatibility varies by platform (e.g., Xenium, CosMx support FFPE) [49] [23].
Probes & Panels	Targeted Gene Panels	Pre-defined sets of genes for iST.	Crucial for iST; must be carefully designed to cover cell types and pathways of interest [90] [23].
	Whole Transcriptome Probes	Poly(dT) capture oligos for sST.	Used in Visium, Stereo-seq for unbiased profiling [49].
Stains & Reagents	DAPI / H&E Stains	Nuclear counterstain and histological reference.	Enables cell segmentation and correlation with tissue pathology [90] [49].
	Fluorescent Antibodies (CODEX)	For multiplexed protein profiling.	Provides ground truth validation for protein expression and cell typing [49].
Computational Tools	SpaTopic / SpatialTopic R Packages	Software for spatial domain identification via topic modeling.	Requires input of cell types and locations [91] [92].
	Spatiopath Algorithm	Software for statistical spatial pattern analysis.	Used to quantify significant cell-cell and cell-region interactions [13].
	Cell2Spatial R Package	Software for mapping single cells to spatial spots.	Useful for enhancing the resolution of sequencing-based ST data [93].
Reference Data	scRNA-seq Dataset	Annotated single-cell transcriptomes from the same sample.	Mandatory for deconvolution and for SpaTopic analysis [92] [93].

This toolkit, combining wet-lab reagents with dry-lab computational packages, provides the foundation for a rigorous and reproducible spatial transcriptomics research program aimed at decoding tumor architecture.

Spatial transcriptomics (ST) has revolutionized the study of tumor organization by enabling the precise quantification of gene expression within the native tissue architecture. Unlike bulk or single-cell RNA sequencing that requires tissue dissociation and loses spatial context, ST technologies preserve the spatial relationships between cells, offering unprecedented insights into the tumor microenvironment (TME), cellular neighborhoods, and spatially variable gene patterns [94]. This capability is particularly valuable for understanding cancer biology, tumor heterogeneity, and the mechanisms underlying therapy resistance.

However, the rapid proliferation of commercial ST platforms has raised critical questions about the reproducibility and concordance of findings across different technologies. As researchers increasingly employ these methods to answer fundamental biological questions and develop clinical diagnostics, understanding cross-platform reproducibility becomes essential for interpreting results and validating discoveries [95]. This technical guide examines the reproducibility of spatial findings across leading ST platforms, with a specific focus on applications in tumor architecture research, providing researchers with methodologies for assessing concordance and frameworks for experimental design.

Imaging-based spatial transcriptomics (iST) platforms represent the cutting edge for single-cell resolution spatial analysis, particularly for Formalin-Fixed Paraffin-Embedded (FFPE) tissues—the standard in clinical pathology. Three leading commercial platforms have emerged: 10X Genomics Xenium, Vizgen MERSCOPE, and NanoString CosMx. While they share the common goal of mapping gene expression in situ, they employ distinct chemical approaches and signal amplification strategies that significantly impact their performance characteristics [68].

Xenium uses a small number of padlock probes with rolling circle amplification (RCA). CosMx employs a low number of probes amplified via branch chain hybridization. MERSCOPE utilizes direct probe hybridization but amplifies signal by tiling transcripts with many probes [68]. These fundamental differences in chemistry create platform-specific strengths and limitations that researchers must consider when designing experiments, especially those involving precious biobanked FFPE samples.

Beyond these iST platforms, other technologies play important roles in the spatial biology ecosystem. Digital Spatial Profiling (DSP) platforms like NanoString's GeoMx allow researchers to select regions of interest (ROIs) based on histology for expression analysis, bridging traditional pathology with high-plex molecular analysis [95]. The emerging CellScape platform enables high-plex spatial proteomics through iterative staining and imaging cycles [3], while the PaintScape platform visualizes 3D genome architecture in situ [3].

Experimental Workflow for Cross-Platform Assessment

The general workflow for conducting a rigorous cross-platform assessment of ST technologies involves careful experimental design, sample preparation, data generation, and computational analysis. The following diagram illustrates the key stages in this process:

Figure 1: Cross-platform assessment workflow for spatial transcriptomics technologies. The process begins with FFPE tissue blocks, proceeds through tissue microarray construction and serial sectioning, then processes sections across multiple platforms for comparative analysis. RCA: Rolling Circle Amplification.

Benchmarking Performance Across Platforms

Systematic Benchmarking in FFPE Tissues

A comprehensive 2025 benchmarking study systematically evaluated three commercial iST platforms—10X Xenium, Vizgen MERSCOPE, and Nanostring CosMx—using serial sections from tissue microarrays (TMAs) containing 17 tumor and 16 normal tissue types [68]. This study provides the most direct head-to-head comparison available to date, with critical implications for tumor organization research.

The experimental design involved creating three TMAs: two tumor TMAs (tTMA1 with 170 cores from 7 cancer types; tTMA2 with 48 cores from 19 cancer types) and one normal tissue TMA (nTMA with 45 cores from 16 normal tissue types) [68]. To emulate real-world conditions using standard biobanked FFPE tissues, samples were not pre-screened based on RNA integrity, though they were screened by H&E during TMA assembly—reflecting typical workflows for clinical pathology specimens.

For platform matching, the researchers utilized the CosMx 1K panel, Xenium human breast, lung, and multi-tissue panels, and designed custom MERSCOPE panels to match the Xenium breast and lung panels, ensuring adequate gene overlap (>65 genes) across platforms [68]. Data collection occurred in multiple rounds (2023 and 2024), with intentional standardization of tissue preparation conditions in the 2024 round to enable fair head-to-head comparisons.

Quantitative Performance Metrics

The benchmarking study generated massive datasets encompassing over 394 million transcripts and 5 million cells, enabling robust statistical comparisons across platforms [68]. The table below summarizes key performance metrics with implications for tumor architecture studies:

Table 1: Performance metrics of imaging-based spatial transcriptomics platforms from systematic benchmarking in FFPE tissues

Performance Metric	Xenium	CosMx	MERSCOPE
Transcript Counts per Gene	Highest	High	Lower
Concordance with scRNA-seq	High	High	Not Reported
Cell Sub-clustering Capability	High	High	Moderate
False Discovery Rates	Variable	Variable	Variable
Cell Segmentation Error Frequency	Variable	Variable	Variable
Total Transcripts Recovered (2024 data)	High	Highest	Lower

The study found that Xenium consistently generated higher transcript counts per gene without sacrificing specificity, while both Xenium and CosMx demonstrated strong concordance with orthogonal single-cell transcriptomics data [68]. All three platforms successfully performed spatially resolved cell typing, though with varying sub-clustering capabilities—Xenium and CosMx identified slightly more clusters than MERSCOPE, albeit with different false discovery rates and cell segmentation error frequencies [68].

Platform-Specific Technical Considerations

Each platform demonstrated distinct technical characteristics that influence their application in tumor research:

Sensitivity and Specificity: On matched genes, Xenium showed superior sensitivity with higher transcript counts per gene, while maintaining specificity. CosMx also demonstrated high sensitivity, particularly with its whole transcriptome approach [68] [3].

Single-Cell Resolution: All three iST platforms provide single-cell resolution, but with different segmentation approaches. Xenium improved its segmentation capabilities between 2023 and 2024 by adding additional membrane staining, highlighting how rapidly these platforms are evolving [68].

Multimodal Integration: CosMx's Whole Transcriptome (WTX) assay has demonstrated strong performance in detecting rare cells and representing tissue composition more accurately than scRNA-seq alone, while preserving spatial context—particularly valuable for identifying rare tumor subpopulations or immune cells [3].

Assessing Concordance and Reproducibility

Analytical Frameworks for Cross-Platform Validation

Establishing concordance across platforms requires multiple analytical approaches that assess different aspects of data quality and biological validity. The benchmarking study employed several rigorous methods:

Orthogonal Validation with scRNA-seq: The comparison of iST data with matched single-cell transcriptomics data from 10x Chromium Single Cell Gene Expression FLEX provides a critical ground truth for assessing the accuracy of gene expression measurements independent of spatial information [68].

Spatial Clustering Consistency: Evaluating the consistency of spatially resolved cell typing across platforms tests whether each platform identifies similar cellular neighborhoods and tissue domains—essential for tumor microenvironment studies where cellular organization impacts function [68].

Cell Segmentation Accuracy: Assessing co-expression patterns of known disjoint markers helps validate cell segmentation and boundary detection, which is fundamental for accurate single-cell analysis within tissues [68].

Reproducibility in Clinical Samples

A separate study focusing on rigor and reproducibility examined spatial transcriptomics performance in clinically sourced human kidney tissues, including both nephrectomy specimens and biopsies [95]. This research provides critical insights for tumor architecture studies, particularly regarding:

Technical Reproducibility: The study demonstrated high technical reproducibility for digital spatial profiling when applied to FFPE tissues, with consistent results across replicate sections and regions of interest [95].

Normalization Impact: The research highlighted how normalization approaches can significantly impact biological interpretation of spatial transcriptomics data, emphasizing the need for careful computational processing in cross-platform studies [95].

Sensitivity Tradeoffs: The comparison between multicellular (GeoMx DSP) and single-cell resolution (CosMx SMI) platforms revealed tradeoffs in cost, execution time, and detection sensitivity that must be balanced based on research objectives [95].

Table 2: Key research reagents and solutions for spatial transcriptomics studies

Reagent/Solution	Function	Platform Examples
Custom Gene Panels	Targeted gene expression profiling	Xenium, MERSCOPE
Whole Transcriptome Panels	Comprehensive transcriptome coverage	CosMx WTX
Membrane Stains	Cell segmentation and boundary identification	Xenium
Immunostaining Panels	Protein expression alongside transcriptomics	CellScape
CRISPR Screening Panels	Spatial mapping of gene edits	CosMx CRISPR
Multi-omics Integration	Combined RNA and protein profiling	GeoMx DPA

Methodologies for Cross-Platform Concordance Experiments

Experimental Design Considerations

Implementing robust cross-platform concordance studies requires meticulous experimental design:

Tissue Selection and Preparation: The benchmarking study used tissue microarrays containing multiple tumor and normal types, enabling assessment of platform performance across diverse tissue architectures [68]. For tumor-specific studies, including various cancer subtypes and grading patterns is essential.

Sectioning Protocol: Consecutive serial sectioning (typically 5-10μm thickness) ensures that nearly identical cellular regions are profiled across platforms. The 2024 benchmarking data specifically controlled for baking times after slicing to standardize tissue condition across platforms [68].

RNA Quality Assessment: While the benchmarking study intentionally used typical biobanked tissues without RNA quality pre-screening to reflect real-world conditions, the MERSCOPE platform recommends DV200 > 60% for optimal performance [68]. Researchers should consider RNA quality metrics when interpreting results, especially for archival samples.

Data Processing and Analysis Framework

The computational workflow for cross-platform concordance involves several critical steps:

Spatial Data Alignment: Multiple computational tools exist for aligning and integrating spatial transcriptomics slices, with at least 24 methodologies recently reviewed [55]. These can be categorized as:

• Statistical mapping approaches (e.g., Splotch, GPSA, PASTE)
• Image processing and registration tools (e.g., STIM, STalign)
• Graph-based methods (e.g., SpatiAlign, STAligner) [55]

Cell Segmentation Standardization: The benchmarking study used each manufacturer's standard base-calling and segmentation pipeline, reflecting typical user experience [68]. For more controlled comparisons, consistent segmentation algorithms could be applied across platforms.

Cross-Platform Integration: Emerging tools like SPIRAL enable integration and alignment of spatially resolved transcriptomics data across different experiments, conditions, and technologies [55], facilitating direct comparative analyses.

The following diagram illustrates the computational workflow for assessing cross-platform concordance:

Figure 2: Computational workflow for assessing cross-platform concordance in spatial transcriptomics data. The process begins with raw data from multiple platforms, proceeds through spatial alignment and feature extraction, and culminates in multiple concordance metrics calculation.

Implications for Tumor Architecture Research

Applications in Tumor Biology

The reproducibility of spatial findings across platforms has profound implications for advancing our understanding of tumor architecture:

Tumor Microenvironment Deconstruction: Consistent identification of cellular neighborhoods across platforms validates the biological reality of these structures rather than being technical artifacts. The benchmarking study demonstrated that all three iST platforms could perform spatially resolved cell typing with varying degrees of sub-clustering capabilities [68].

Therapeutic Target Discovery: Cross-platform concordance increases confidence in potentially targetable spatial patterns, such as immune exclusion zones or stromal barrier formations. CosMx WTX has been used to project over 2,000 measured pathways directly onto tumor tissues, enabling visualization of epithelial-mesenchymal transition, immune barriers, and tissue-specific pathway activation [3].

Clinical Translation: Reproducibility across platforms is fundamental for developing spatial biomarkers for diagnostic use. The high rigor and reproducibility demonstrated for DSP in clinically sourced tissues supports the potential for clinical translation [95].

Emerging Applications and Multimodal Integration

Spatial transcriptomics is increasingly being integrated with other data modalities to provide deeper insights into tumor biology:

Spatial Multi-omics: Platforms like the GeoMx Discovery Proteome Atlas (1,100+ plex protein assay) paired with the GeoMx Whole Transcriptome Atlas (18,000+ plex) enable same-section spatial profiling of RNA and protein targets [3]. This multiomic approach demonstrates high sensitivity, reproducibility, and improved biological resolution for cancer research.

3D Genome Architecture: The PaintScape platform enables in situ, single-cell visualization of 3D genome architecture in cancer, revealing structural genome differences across localized, metastatic, and triple-negative breast cancer models [3].

AI-Enhanced Spatial Analysis: Deep learning approaches like MISO (Multiscale Integration of Spatial Omics with tumor morphology) can predict spatial transcriptomics from H&E stained histological slides, potentially increasing the accessibility of spatial biology to larger patient cohorts [5].

The assessment of cross-platform concordance in spatial transcriptomics reveals both substantial agreement and important technical differences across leading technologies. The systematic benchmarking of Xenium, CosMx, and MERSCOPE demonstrates that all three platforms can generate biologically meaningful spatial data from FFPE tissues, with varying strengths in sensitivity, resolution, and analytical capabilities.

For researchers studying tumor organization architecture, this concordance framework provides methodological guidance for platform selection, experimental design, and analytical validation. As spatial technologies continue to evolve toward higher plex, improved resolution, and multimodal integration, establishing reproducibility across platforms remains fundamental to advancing our understanding of cancer biology and translating spatial discoveries into clinical applications.

The integration of spatial transcriptomics with other data modalities—including proteomics, chromatin organization, and histopathological imaging—promises to unlock deeper insights into tumor architecture and function. Through rigorous cross-platform validation and standardized analytical approaches, the spatial biology community can ensure that findings reflect biological reality rather than technical artifacts, ultimately accelerating discoveries in tumor biology and therapeutic development.

The tumor microenvironment (TME) represents a highly complex and organized ecosystem where the spatial coordination of malignant, immune, and stromal cells fundamentally influences disease progression and therapeutic response [33]. Traditional bulk and single-cell RNA sequencing technologies, while powerful for cataloging cellular heterogeneity, inherently destroy the critical spatial context that governs cell-cell communication and functional tissue organization [22]. Spatial transcriptomics has emerged as a transformative technology that bridges this gap by quantifying gene expression patterns within the intact architectural framework of tissues [22]. This technical guide examines the clinical validation of spatially resolved gene signatures and their established utility in predicting patient prognosis across multiple cancer types, providing researchers and drug development professionals with methodologies, analytical frameworks, and clinical evidence supporting their implementation.

Clinical Evidence: Prognostic Spatial Signatures Across Cancers

Robust clinical studies have successfully linked specific spatial gene expression patterns to patient outcomes, demonstrating superior prognostic capability compared to non-spatial approaches. The following table synthesizes key validated spatial signatures from recent literature:

Table 1: Clinically Validated Spatial Gene Signatures for Cancer Prognosis

Cancer Type	Spatial Signature	Prognostic Value	Clinical Validation	Reference
Non-Small Cell Lung Cancer (NSCLC)	Resistance Signature: Proliferating tumor cells, granulocytes, vessels	HR = 3.8 for shorter PFS	Validated in 3 independent cohorts (n=234)	[96]
	Response Signature: M1/M2 macrophages, CD4+ T cells (stroma)	HR = 0.4 for longer PFS	Validated in external cohorts	[96]
Melanoma	S100B+ Tumor Compartment (8-gene signature)	Predicts response to immune checkpoint inhibitors	Validated in independent cohort (n=45); outperformed bulk signatures	[97]
Gastric Cancer (GC)	Intratumoral TLS (iTLS) Signature: CXCL13+ T cells, CXCR5+ B cells, LAMP3+CD80+ DCs	Improved OS and PFS	Associated with better immunotherapy response	[98]

Methodological Framework: Generating Spatially-Resolved Signatures

Core Spatial Transcriptomics Technologies

The development of prognostic spatial signatures relies on multiple technological platforms, each with distinct advantages and resolutions:

• Digital Spatial Profiling (DSP): This platform, exemplified by the GeoMx system, enables compartment-specific transcriptomic profiling within user-defined tissue regions of interest (ROIs) [3] [97]. Using UV-photocleavable oligonucleotide tags, it allows for high-plex spatial whole transcriptome analysis (18,000+ genes) while preserving tissue architecture [3] [97]. Its key advantage for clinical validation is compatibility with formalin-fixed, paraffin-embedded (FFPE) tissues, the standard in pathology [97].

• In Situ Sequencing and Imaging-Based Platforms: Technologies like CosMx and CellScape provide single-cell or subcellular resolution by imaging barcoded probes hybridized to RNA targets directly in tissue sections [3]. The CosMx Human Whole Transcriptome (WTX) assay, for instance, can simultaneously profile transcriptomic and proteomic data from FFPE tissues, enabling AI-powered analysis of spatially organized gene modules [3].

• High-Plex Spatial Multiomics: Integrated approaches now enable same-section spatial profiling of RNA and protein. For example, the GeoMx Discovery Proteome Atlas (1,100-plex protein assay) pairs with its Whole Transcriptome Atlas for comprehensive multiomic analysis [3].

Experimental Workflow for Signature Development

The standard pipeline for developing and validating prognostic spatial signatures involves multiple critical stages, visualized in the following workflow:

Computational Analysis and Signature Training

The analytical framework for transforming spatial data into prognostic signatures employs sophisticated statistical and machine learning approaches:

• Spatial Data Preprocessing: Raw spatial transcriptomics data undergoes normalization, batch effect correction, and quality control. For barcode-based technologies like 10X Visium, spots are typically clustered based on gene expression similarity [99].

• Cell Type Deconvolution: Computational methods like non-negative matrix factorization or reference-based mapping with single-cell RNA-seq data are used to infer cell-type proportions within each spatial spot [99] [100] [33]. This enables the creation of spatial maps of immune, stromal, and malignant cell distributions.

• Spatial Analytics and Neighborhood Analysis: Advanced algorithms identify spatially variable genes and characterize cellular neighborhoods—recurrent multicellular communities within the TME [96]. In NSCLC, for example, Voronoi diagrams and cellular neighborhood analysis have revealed distinct spatial architectures enriched with either response-associated (M2 macrophages) or resistance-associated cell types (vessels, PD-L1+ tumor cells) [96].

• Signature Training Using Machine Learning: Prognostic signatures are typically trained using regularized Cox proportional hazards models. The LASSO (Least Absolute Shrinkage and Selection Operator) penalty is particularly valuable for selecting the most predictive features from high-dimensional spatial data while preventing overfitting [96] [101]. For NSCLC, this approach identified a resistance signature comprising proliferating tumor cells, granulocytes, and vessels, and a response signature comprising M1/M2 macrophages and CD4+ T cells [96]. Models are typically trained on a discovery cohort with internal cross-validation before external validation.

Key Biological Pathways and Cellular Ecosystems

Spatial transcriptomics has revealed that prognosis is intimately linked with specific cellular ecosystems organized within the TME. The following diagram illustrates two key prognostic pathways and ecosystems:

Tertiary Lymphoid Structures as Prognostic Hubs

As illustrated above, tertiary lymphoid structures (TLS) represent organized immune aggregates that form within the TME. In gastric cancer, integrated single-cell and spatial transcriptomics has revealed that intratumoral TLS (iTLS) are enriched with specific cellular populations including CXCL13+ T lymphocytes, CXCR5+ germinal center B lymphocytes, and activated dendritic cells [98]. The development of these structures depends on a coordinated molecular cascade initiated by high endothelial venule (HEV) cells expressing VCAM1 and ICAM1, which recruit and activate CXCL13+ T cells through the CXCL13-ACKR1 pathway [98]. This subsequently promotes B lymphocyte recruitment via CXCL13-CXCR5 crosstalk, culminating in TLS formation. From a clinical perspective, the presence of iTLS is associated with significantly improved overall survival and progression-free survival in gastric cancer patients, highlighting its role as a favorable prognostic ecosystem [98].

Immunosuppressive Spatial Niches

Conversely, spatial transcriptomics has identified distinct immunosuppressive cellular neighborhoods associated with poor prognosis. In NSCLC, resistance to immunotherapy is characterized by spatial co-localization of proliferating tumor cells, granulocytes, and vascular structures [96]. These resistance niches likely create a physical and biochemical barrier to effective immune cell infiltration and function. The prognostic significance of these niches is demonstrated by the resistance signature (proliferating tumor cells, granulocytes, vessels) that predicted significantly worse outcomes with a hazard ratio of 3.8 for progression-free survival [96].

The Scientist's Toolkit: Essential Research Reagents and Platforms

Table 2: Essential Research Tools for Spatial Prognostic Signature Development

Tool Category	Specific Technologies/Platforms	Key Function	Application in Prognosis
Spatial Profiling Platforms	GeoMx Digital Spatial Profiler, CosMx SMI, 10X Visium, CellScape	High-plex RNA/protein mapping in FFPE tissues	Compartment-specific signature discovery [3] [97]
In Situ Imaging Panels	CODEX, MERFISH, seqFISH, RNAscope	Single-cell resolution spatial imaging	Validation of signature localization [96] [22]
Analysis Suites	Visium CytAssist, Xenium Analyzer, DSP DA	Spatial data processing and visualization	Cellular neighborhood identification [3]
Validation Assays	nCounter Analysis System, VistaPlex Assay Kits	Targeted spatial signature quantification	Clinical assay translation [3]

The clinical validation of spatial gene signatures represents a paradigm shift in cancer prognosis, moving beyond mere compositional analysis to incorporate the critical dimension of spatial organization within the TME. The methodologies and evidence presented in this technical guide demonstrate that spatial context provides biologically meaningful and clinically actionable insights that outperform traditional bulk tissue biomarkers. As spatial technologies continue to evolve toward higher plex and resolution, and as computational methods for spatial data integration become more sophisticated, the translation of spatial signatures into clinical practice will accelerate. Future developments will likely focus on standardizing analytical pipelines, validating signatures in prospective clinical trials, and integrating artificial intelligence for automated spatial pattern recognition. For researchers and drug development professionals, mastering these spatial technologies and analytical approaches is now essential for advancing precision oncology and developing the next generation of prognostic tools.

Spatial transcriptomics has revolutionized our understanding of solid tumor organization by preserving the architectural context of gene expression. This technical review synthesizes current research demonstrating that despite the histological and molecular diversity across cancer types, a fundamental organizational principle exists: the leading edge (LE) of tumors exhibits conserved transcriptional programs linked to invasion and poor prognosis, while the tumor core (TC) displays more tissue-specific signatures associated with varied clinical outcomes. This pan-cancer architectural framework, elucidated through advanced computational integration of spatial datasets, reveals conserved mechanisms of progression and unveils novel therapeutic targets for drug development.

The tumor microenvironment (TME) is not a chaotic collection of cells but a highly organized ecosystem with distinct spatial domains that play specialized roles in cancer progression. The emergence of high-resolution spatial transcriptomics technologies has enabled the systematic mapping of these domains across cancer types, revealing consistent architectural patterns that transcend tissue of origin. This technical guide examines the evidence for conserved versus tissue-specific spatial architectures in solid tumors, focusing on the robust dichotomy between the invasive leading edge and the tumor core.

Understanding these pan-cancer principles provides a framework for developing novel therapeutic strategies that target conserved invasive mechanisms while accounting for tissue-specific contextual factors. For drug development professionals, these insights offer opportunities to design treatments that disrupt critical spatial communication networks and metabolic dependencies within the TME.

Core Findings: Conserved Leading Edge and Tissue-Specific Tumor Core

Transcriptional and Functional Dichotomy

Integrative single-cell and spatial transcriptomic analysis of HPV-negative oral squamous cell carcinoma (OSCC) has established that the TC and LE represent functionally distinct compartments with unique transcriptional profiles, cellular compositions, and ligand-receptor interactions [21].

Table 1: Core Transcriptional and Functional Features of Tumor Spatial Domains

Feature	Tumor Core (TC)	Leading Edge (LE)
Key Marker Genes	CLDN4, SPRR1B, SPRR2 family, DEFB4A, LCN2 [21]	LAMC2, ITGA5, COL1A1, FN1, TIMP1 [21]
Hallmark Pathways	Keratinization, cell differentiation, antimicrobial immunity [21]	Epithelial-mesenchymal transition (EMT), cell cycle, angiogenesis [21]
Activated Signaling	MSP-RON (macrophages), IL-33, p38 MAPK [21]	GP6, EIF2, HOTAIR regulatory pathways [21]
Cellular Processes	Immune modulation, differentiation [21]	ECM remodeling, invasion, proliferation [21]
Pan-Cancer Conservation	Tissue-specific [21]	Conserved across cancer types [21]
Clinical Prognosis	Associated with improved outcomes [21]	Predicts worse survival across multiple cancers [21]

The LE gene signature is characterized by extracellular matrix (ECM) remodeling genes (COL1A1, FN1, COL1A2, TIMP1, COL6A2) and demonstrates elevated activity in cell cycle, EMT, and angiogenesis pathways [21]. In contrast, the TC expresses genes involved in keratinization (SPRR2D, SPRR2E, SPRR2A) and inhibition of EMT (DEFB4A, LCN2) [21]. This fundamental dichotomy is conserved across patients, with high correlation within TC and LE compartments across different individuals, but low correlation between TC and LE within the same patient [21].

Pan-Cancer Validation in Primary and Metastatic Liver Tumors

The conserved nature of invasive programs is further evidenced in liver tumors. A direct high-resolution spatial comparison of primary hepatocellular carcinoma (HCC) and liver metastases revealed fundamentally different spatial architectures, yet shared metabolic vulnerabilities [102].

HCC displays an ordered lineage architecture with transformed hepatocyte-like tumor cells broadly dispersed across the tissue, while liver metastases show sharply compartmentalized domains: an invasion zone containing proliferative stem-like tumor cells adjacent to TAM-rich boundaries, and a plasticity zone forming a heterogeneous niche of cancer-testis antigen-positive germline-like cells [102]. Despite these organizational differences, both tumor types converged on a shared program of "porphyrin overdrive" metabolism, characterized by reduced cytochrome P450 expression, enhanced oxidative phosphorylation, and upregulation of FLVCR1 and ALOX5, reflecting coordinated rewiring of heme and lipid metabolism [102].

Methodological Framework: Spatial Transcriptomics and Computational Integration

Experimental Workflow for Spatial Domain Mapping

The identification of conserved spatial architectures requires standardized experimental and computational workflows. The following diagram illustrates the integrated process for spatial transcriptomics analysis and domain identification:

Multi-Slice Integration with Community-Enhanced Graph Contrastive Learning

The integration of multiple spatial transcriptomics datasets across different platforms and biological conditions presents significant computational challenges due to batch effects and different spatial resolutions. Tacos (mulTiple spAtial transcriptomiCs data integratiOn using community-enhanced graph contraStive learning) represents a state-of-the-art approach that addresses these limitations [103].

Tacos constructs spatial graphs for each slice based on spatial coordinates, then employs a graph contrastive learning-based encoder to extract spatially aware embeddings. The method incorporates two key innovations for handling heterogeneous spatial structures:

• Communal attribute voting: Detects node features more likely to be masked
• Communal edge dropping: Computes edge mask probabilities based on community structure [103]

The model detects mutual nearest neighbor (MNN) pairs between spots from different slices and uses triplet loss to pull positive pairs close while pushing negative pairs apart, effectively aligning slices while preserving biological structures [103]. This approach has demonstrated superior performance in integrating slices from different platforms (e.g., 10x Visium, Slide-seqV2, Stereo-seq) while maintaining specific structural features unique to each dataset [103].

Table 2: Computational Methods for Spatial Transcriptomics Integration

Method	Core Algorithm	Strengths	Limitations
Tacos [103]	Community-enhanced graph contrastive learning	Handles different resolutions; preserves specific structures	Computational complexity with large datasets
STAligner [103]	Graph neural networks	Effective for similar resolutions	Limited with heterogeneous structures
SPIRAL [103]	Graph neural network + optimal transport	Good alignment performance	Less effective at preserving annotated layers
SpaOTsc [104]	Structured optimal transport	Infers spatial relationships from scRNA-seq	Requires spatial measurements of some genes
Harmony [103]	Linear integration	Fast batch correction	Loses spatial relationships

Spatial Cell-Cell Communication Inference

Understanding signaling relationships between spatial domains is crucial for decoding TME organization. SpaOTsc (Spatial Optimal Transport for single cells) infers spatial and signaling relationships between cells from single-cell transcriptomic data by utilizing spatial measurements of a relatively small number of genes [104].

The method establishes a spatial metric for individual cells in scRNA-seq data based on a map connecting it with spatial measurements, then obtains cell-cell communications by "optimally transporting" signal senders to target signal receivers in space [104]. This approach has been validated for reconstructing spatial cellular dynamics in tissues and predicting spatial gene expression patterns [104].

The Scientist's Toolkit: Essential Research Reagents and Platforms

Table 3: Essential Research Reagents and Platforms for Spatial Architecture Studies

Category	Specific Tools/Reagents	Function	Considerations
Spatial Technologies	10x Visium (55μm) [103], Slide-seqV2 (10μm) [103], Stereo-seq (subcellular) [103], seqFISH [9]	Spatial gene expression profiling	Resolution, gene coverage, tissue compatibility
Computational Tools	Tacos [103], Spaco [9], SpaOTsc [104], STAligner [103], SPIRAL [103]	Data integration, visualization, analysis	Scalability, resolution handling, batch correction
Analysis Frameworks	Seurat [9], Giotto [9], Scanpy [103], Squidpy [9]	General data analysis and visualization	Integration with spatial methods, customization
Visualization Tools	Spaco [9] with DOI metric	Spatially-aware colorization	Color contrast, CVD support, perceptual clarity
Reference Datasets	Human DLPFC [103], Mouse Olfactory Bulb [103], HCC & Metastases [102]	Benchmarking, validation	Annotation quality, technical variability

Advanced Visualization with Spaco

Effective visualization of spatial transcriptomics data requires specialized tools that account for spatial relationships between cell types. Spaco (Spatial colorization) introduces the Degree of Interlacement (DOI) metric to construct a weighted graph evaluating spatial relationships among different cell types, refining color assignments to enhance visual clarity [9].

The DOI is computed via a modified spatial k-nearest neighbor network incorporating a dual-outlier-free strategy that excludes both spatially sparse cells and cell types to enhance stability [9]. This approach generates a cluster interlacement graph (CI-graph) that ensures cluster pairs with larger DOIs (more spatial interlacement) are visualized with more distinct colors, significantly improving interpretation of complex tissue architectures, particularly in brain and tumor microenvironments [9].

Therapeutic Implications and Drug Discovery Applications

Targeting Conserved Invasive Programs

The conserved nature of LE transcriptional programs across cancer types suggests they represent fundamental mechanisms of tumor invasion and metastasis that could be targeted therapeutically. In silico modeling of OSCC has identified spatially-regulated patterns of cell development that are predictably associated with drug response [21]. This approach can prioritize compounds that disrupt information flow from TC to LE regions, potentially inhibiting metastatic progression.

The workflow for translating spatial architectural insights into therapeutic discovery is illustrated below:

Metabolic Vulnerabilities Across Spatial Architectures

The discovery of "porphyrin overdrive" as a conserved metabolic program in both HCC and liver metastases highlights how spatial transcriptomics can reveal convergent vulnerabilities despite divergent cellular origins and organizational structures [102]. This shared program of reduced cytochrome P450 expression, enhanced oxidative phosphorylation gene expression, and upregulation of FLVCR1 and ALOX5 reflects coordinated rewiring of heme and lipid metabolism that may be therapeutically exploitable [102].

Targeting this metabolic convergence could yield broad efficacy across different liver tumor types, illustrating how pan-cancer spatial analysis can identify unexpected therapeutic opportunities that transcend classical histopathological classifications.

Spatial transcriptomics has established that solid tumors across different tissues of origin share fundamental organizational principles, particularly the conservation of invasive programs at the leading edge alongside more tissue-specific signatures in the tumor core. This architectural framework provides a new dimension for understanding cancer biology and developing therapeutic strategies.

Future research directions should focus on:

• Multi-omic spatial integration combining transcriptomic, proteomic, and epigenomic data within architectural contexts
• Dynamic spatial modeling to track architectural evolution during progression and treatment
• High-throughput drug screening integrated with spatial readouts to identify compounds that disrupt conserved invasive programs
• Clinical translation of spatial signatures as biomarkers for prognosis and treatment selection

For drug development professionals, these pan-cancer spatial insights offer a roadmap for targeting conserved mechanisms of invasion and metastasis while accounting for tissue-specific contextual factors that may modulate therapeutic response.

Conclusion

Spatial transcriptomics has fundamentally shifted our approach to studying cancer, moving beyond single-cell suspensions to a holistic view of the tumor ecosystem. The key takeaway is that tumor architecture is not random; it is organized into functional spatial domains—such as the conserved leading edge and the tissue-specific tumor core—that dictate disease progression and therapy response. The integration of high-resolution ST platforms with advanced computational methods, particularly AI, is essential to decode this complexity. Future efforts must focus on standardizing analytical pipelines, increasing accessibility, and translating these rich spatial maps into novel therapeutic strategies and biomarkers. The ultimate goal is to usher in an era of spatial pathology, where a deep understanding of cellular neighborhoods directly informs precision oncology and improves patient outcomes.

References

• 1. An introduction to spatial transcriptomics for biomedical research [https://genomemedicine.biomedcentral.com/articles/10.1186/s13073-022-01075-1]
• 2. Exploring tissue architecture using spatial transcriptomics - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC8475179/]
• 3. Spatial Biology Takes Center Stage at AACR 2025 [https://brukerspatialbiology.com/spatial-biology-takes-center-stage-at-aacr-2025/]
• 4. Systematic comparison of sequencing-based spatial ... [https://www.nature.com/articles/s41592-024-02325-3]
• 5. A deep learning-based multiscale integration of spatial ... [https://www.nature.com/articles/s41467-025-66691-y]
• 6. STAIG: Spatial transcriptomics analysis via image-aided ... [https://www.nature.com/articles/s41467-025-56276-0]
• 7. spCLUE: a contrastive learning approach to unified spatial ... [https://genomebiology.biomedcentral.com/articles/10.1186/s13059-025-03636-0]
• 8. A comprehensive comparison on cell-type composition ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9294426/]
• 9. Spaco: A comprehensive tool for coloring spatial data at single ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10935509/]
• 10. Quantifying and interpreting biologically meaningful spatial ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11897387/]
• 11. SpaLinker identifies phenotype-associated spatial tumor ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12278630/]
• 12. Single‐Cell and Spatial Transcriptomics Delineate the Microstructure... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11831447/]
• 13. Statistical analysis of spatial patterns in tumor ... [https://www.nature.com/articles/s41467-025-57943-y]
• 14. Spatially resolved transcriptomics reveals the architecture ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8560802/]
• 15. Mapping cells in time and space: New tool reveals a detailed history of... [https://news.mit.edu/2025/new-tool-reveals-detailed-history-of-tumor-growth-0730]
• 16. Tumour evolution and microenvironment interactions in 2D ... [https://www.nature.com/articles/s41586-024-08087-4]
• 17. Spatially resolved transcriptomics reveals the architecture ... [https://www.nature.com/articles/s41467-021-26614-z]
• 18. Spatially resolved transcriptomics reveals the architecture of ... [https://pubmed.ncbi.nlm.nih.gov/34725363/]
• 19. From Spatial Patterns to Prognosis: Decoding Single-Cell ... [https://pubmed.ncbi.nlm.nih.gov/40959106/]
• 20. Multiparametric cellular and spatial organization in cancer ... [https://www.nature.com/articles/s41551-025-01475-9]
• 21. Spatial transcriptomics reveals distinct and conserved ... [https://www.nature.com/articles/s41467-023-40271-4]
• 22. Advances in spatial and its applications in cancer... transcriptomics [https://molecular-cancer.biomedcentral.com/articles/10.1186/s12943-024-02040-9]
• 23. Artificial Intelligence–driven spatial transcriptomics in OSCC [https://pmc.ncbi.nlm.nih.gov/articles/PMC12639440/]
• 24. reveals distinct and conserved Spatial core... transcriptomics tumor [https://pubmed.ncbi.nlm.nih.gov/37596273/]
• 25. Spatial profiling technologies for research and clinical ... [https://www.sciencedirect.com/science/article/pii/S2590262825000528]
• 26. Artificial intelligence-enabled spatially resolved transcriptomics ... [https://accscience.com/journal/TD/articles/online_first/1282]
• 27. Spatial transcriptomics in solid tumor studies [https://www.sciencedirect.com/science/article/abs/pii/S0093775425000818]
• 28. Extracellular matrix dynamics in tumor immunoregulation [https://jhoonline.biomedcentral.com/articles/10.1186/s13045-025-01717-y]
• 29. Spatial Correlation of the Extracellular Matrix to Immune ... [https://www.sciencedirect.com/science/article/pii/S0023683725000406]
• 30. Spatial transcriptomics uncovers immune-cell plasticity and ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12408323/]
• 31. Regional and cell specific bioactivity of injectable ... [https://www.nature.com/articles/s41467-025-65351-5]
• 32. integrative single-cell and spatial RNA sequencing analysis [https://pmc.ncbi.nlm.nih.gov/articles/PMC12589693/]
• 33. Single-cell and spatial transcriptomics integration [https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2025.1649468/full]
• 34. Comparison of imaging based single-cell resolution spatial ... [https://www.nature.com/articles/s41467-025-63414-1]
• 35. A practical guide for choosing an optimal spatial ... [https://bmcgenomics.biomedcentral.com/articles/10.1186/s12864-025-11235-3]
• 36. Sequencing vs. Imaging in Spatial Transcriptomics [https://www.scdiscoveries.com/blog/sequencing-vs-imaging-in-spatial-transcriptomics/]
• 37. Analysis and Visualization of Spatial Transcriptomic Data - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC8829434/]
• 38. Visualizing Spatial Transcriptomics: A Guide to Effective Plotting [https://nanostring-biostats.github.io/CosMx-Analysis-Scratch-Space/posts/spatial-plotting/]
• 39. Opportunities and challenges in the application of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12568615/]
• 40. Visium Spatial Assays | 10x Genomics [https://www.10xgenomics.com/platforms/visium/product-family]
• 41. STOmics Stereo-seq | Advanced Spatial Multi-omics Technology [https://en.stomics.tech/]
• 42. GeoMx DSP Spatial Genomics Overview [https://nanostring.com/products/geomx-digital-spatial-profiler/geomx-dsp-overview/]
• 43. October 2025 Newsletter [https://brukerspatialbiology.com/october-2025-newsletter/]
• 44. Systematic benchmarking of imaging spatial ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10723440/]
• 45. Mapping spatial organization and genetic cell-state ... [https://www.nature.com/articles/s41590-024-01943-5]
• 46. Optimizing Xenium In Situ data utility by quality assessment ... [https://www.nature.com/articles/s41592-025-02617-2]
• 47. MERFISH Spatial Transcriptomics [https://vizgen.com/technology/]
• 48. Xenium In Situ Platform [https://www.10xgenomics.com/platforms/xenium]
• 49. Systematic benchmarking of high-throughput subcellular ... [https://www.nature.com/articles/s41467-025-64292-3]
• 50. Single-cell spatial transcriptomics unravels cell states and ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11907085/]
• 51. Concordance of MERFISH spatial transcriptomics with bulk ... [https://www.life-science-alliance.org/content/6/1/e202201701]
• 52. Current Role and Future Frontiers of Spatial ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12427188/]
• 53. Spatial Data Analysis of Patient Tumor Samples [https://blog.crownbio.com/spatial-data-analysis-patient-tumor-samples]
• 54. CellSP enables module discovery and visualization for ... [https://www.nature.com/articles/s42003-025-08891-2]
• 55. A comprehensive review of spatial transcriptomics data ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12199153/]
• 56. From pixels to cell types: a comprehensive review of ... [https://genomicsinform.biomedcentral.com/articles/10.1186/s44342-025-00055-2]
• 57. Immune profiling in oncology: bridging the gap between ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12380968/]
• 58. Driving innovations in cancer research through spatial ... [https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2025.1589943/full]
• 59. Spatial Metabolomics in Cancer Research: Unlocking the ... [https://www.metwarebio.com/spatial-metabolomics-cancer-research-precision-oncology/]
• 60. Deconvolution of cell types and states in spatial multiomics ... [https://www.nature.com/articles/s41467-025-58874-4]
• 61. Transfer learning of multicellular organization via single - cell and... [https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1012991]
• 62. Scaling up spatial transcriptomics for large-sized tissues [https://www.nature.com/articles/s41592-025-02770-8]
• 63. and Integrating - spatial reveals single ... cell transcriptomics tumor [https://www.nature.com/articles/s41419-024-06598-6?error=cookies_not_supported&code=2c4550b6-1239-45e5-abc4-e5309d158146]
• 64. St. Jude researchers create tool incorporating ... [https://www.stjude.org/media-resources/news-releases/2025-medicine-science-news/st-jude-researchers-create-tool-incorporating-transcriptome-size-to-improve-rna-seq-analysis.html]
• 65. Spatial Transcriptomics Reveals Distinct Architectures but ... [https://pubmed.ncbi.nlm.nih.gov/41097737/]
• 66. Systematic benchmarking of high-throughput subcellular ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12534522/]
• 67. Protocol for single-cell spatial transcriptomic profiling of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12491524/]
• 68. Systematic benchmarking of imaging spatial ... [https://www.nature.com/articles/s41467-025-64990-y]
• 69. Benchmarking of spatial transcriptomics platforms across ... [https://sciforum.net/paper/view/26187]
• 70. SCNT: an R package for data analysis and visualization of ... [https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-025-06209-x]
• 71. standardized benchmarking for multi-omics cancer survival ... [https://pubmed.ncbi.nlm.nih.gov/41031875/]
• 72. A critical spotlight on the paradigms of FFPE-DNA ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10415133/]
• 73. Fresh and frozen cardiac tissue are comparable in DNA ... [https://www.nature.com/articles/s41598-023-43788-2]
• 74. With Great FFPE Challenges, Comes Novel Research ... [https://www.admerahealth.com/blog/optimizations-for-ffpe-samples]
• 75. Improvements and challenges of tissue preparation for ... [https://www.sciencedirect.com/science/article/pii/S2405844023013403]
• 76. Quantitative Expression Profiling in Formalin-Fixed Paraffin ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC2893624/]
• 77. Fresh frozen vs FFPE samples for next-generation ... [https://www.lexogen.com/blog/fresh-frozen-vs-ffpe-samples-for-next-generation-sequencing-studies/]
• 78. Proteomic comparison between different tissue ... [https://www.nature.com/articles/s41598-021-87003-6]
• 79. Equivalence of protein inventories obtained from ... - PubMed [https://pubmed.ncbi.nlm.nih.gov/19467989/]
• 80. Spatial profiling of chromatin accessibility in formalin-fixed ... [https://www.nature.com/articles/s41467-025-60882-3]
• 81. Computational challenges and opportunities in spatially ... [https://www.nature.com/articles/s41467-021-25557-9]
• 82. Cell segmentation in spatial transcriptomics with BOMS [https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0311458]
• 83. New cell segmentation methods for better spatial ... [https://www.fredhutch.org/en/news/spotlight/2025/10/vidd-jones-natmethods.html]
• 84. Bioinformatics in 2025: Key Innovations and Trends ... [https://bioinformy.com/blog/bioinformatics-in-2025-key-innovations-and-trends]
• 85. AI Feature Extraction: Techniques, Benefits, and Applications [https://flypix.ai/ai-feature-extraction/]
• 86. How is a spatial features extraction done? [https://milvus.io/ai-quick-reference/how-is-a-spatial-features-extraction-done]
• 87. Integrated protein and transcriptome spatial profiling - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC10272089/]
• 88. Spatial transcriptomics: a bibliometric analysis with large ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12554094/]
• 89. How to get the most out of your cancer spatial ... [https://www.sciencedirect.com/science/article/abs/pii/S1084952125000679]
• 90. Comparison of spatial transcriptomics technologies using ... [https://genomebiology.biomedcentral.com/articles/10.1186/s13059-025-03624-4]
• 91. Scalable topic modelling decodes spatial tissue ... [https://www.nature.com/articles/s41467-025-61821-y]
• 92. SpaTopic: A statistical learning framework for exploring tumor ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11430467/]
• 93. Cell2Spatial is a computational framework that maps single ... [https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3003477]
• 94. A guidebook of spatial transcriptomic technologies, data ... [https://www.sciencedirect.com/science/article/pii/S2001037023000156]
• 95. Rigor and Reproducibility of Spatial Transcriptomics ... [https://www.sciencedirect.com/science/article/abs/pii/S002368372500100X]
• 96. Spatial signatures for predicting immunotherapy outcomes ... [https://www.nature.com/articles/s41588-025-02351-7]
• 97. Spatially informed gene signatures for response to ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11326985/]
• 98. Single-cell and spatial transcriptomics implicate a ... [https://www.nature.com/articles/s41467-025-65421-8]
• 99. Spatial single cell analysis of tumor microenvironment ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10317840/]
• 100. Single-cell and spatial transcriptomics analysis of non- ... [https://www.nature.com/articles/s41467-024-48700-8]
• 101. Molecular dynamics simulation and single-cell and spatial ... [https://pubmed.ncbi.nlm.nih.gov/41234880/]
• 102. Spatial Transcriptomics Reveals Distinct Architectures but ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12523610/]
• 103. Integrating multiple spatial transcriptomics data using ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11990772/]
• 104. Inferring spatial and signaling relationships between cells ... [https://www.nature.com/articles/s41467-020-15968-5]