Lymphoma encompasses over 100 subtypes recognized by the 2022 WHO classification, broadly divided into B-cell, T-cell, and NK-cell proliferations with further subcategorization based on clinicopathologic, molecular, and genetic data. Despite this complexity, what makes lymphoma especially difficult to treat is not just the malignant cells themselves but the intricate ecosystem surrounding them. The tumor microenvironment (TME) is composed of malignant cells, immune cells, stromal cells, blood vessels, and the extracellular matrix, and its composition varies substantially between lymphoma subtypes and even within the same tumor.
The TME as both adversary and opportunity: Immune cells infiltrating the TME are not passive bystanders. Rather than simply mounting an antitumor response, they can be co-opted by lymphoma cells to promote growth, suppress cytotoxic activity, and enable immune evasion. Tissue-resident immune cells can induce tertiary lymphoid structures (TLS) that may support antitumor immunity; in some solid cancers, TLS presence correlates with improved response to immune checkpoint blockade. Detailed characterization of these TME components in lymphoma and their influence on treatment response remains incomplete.
The spatial gap: Prior to spatial transcriptomics, the dominant approach for characterizing lymphoma biology at the transcriptome level was single cell RNA sequencing (scRNAseq). While scRNAseq enables high-resolution identification of cell populations and their gene expression profiles, it requires tissue dissociation, which destroys the architectural organization of the lymph node or tumor. The position of a cell relative to its neighbors, the proximity of a regulatory T cell to a malignant Hodgkin cell, or the location of an immunosuppressive macrophage niche are lost entirely. Spatial transcriptomics was named Nature Method of the Year for 2020, and this 2023 review from Newcastle University systematically examines how it is being deployed to restore that missing spatial context in lymphoma research.
The review covers both image-based and sequencing-based spatial transcriptomic technologies, summarizes all published lymphoma studies using these tools (seven studies, totaling 75 patient samples as of publication), and projects how the field will evolve toward personalized medicine.
Spatial transcriptomic technologies fall into two broad categories: image-based approaches and next-generation sequencing (NGS)-based approaches. The choice between them involves a fundamental trade-off between spatial resolution, transcript detection efficiency, transcriptome coverage, and tissue requirements. The review summarizes these trade-offs in a comparison table covering the major platforms.
Image-based technologies: These use fluorescent probes and microscopy to detect and quantify RNA molecules directly within intact tissue. Single molecule fluorescence in situ hybridization (smFISH) is the foundational method, labeling individual transcripts with complementary fluorescent probes. The commercial RNAScope platform builds on this with "double-Z" probe designs that dramatically improve sensitivity and specificity; it supports multiplexing for up to 12 targets from formalin-fixed paraffin-embedded (FFPE) tissue and up to 48 targets from fresh frozen tissue. More ambitious platforms include seqFISH+ and MERFISH, which use combinatorial labeling and sequential imaging to target thousands or tens of thousands of genes simultaneously, though both currently require fresh frozen tissue and carry high costs. In situ sequencing (ISS) employs padlock probes and rolling circle amplification; the commercialized version, 10x Genomics Xenium, automates this process and is compatible with FFPE tissue.
Sequencing-based technologies: Rather than imaging transcripts in place, these methods capture mRNA from tissue sections onto spatially barcoded surfaces, then sequence the captured RNA. The Visium platform from 10x Genomics places barcoded capture spots (55 micrometers in diameter) directly on slides; the tissue is permeabilized to release RNA, which binds to the capture probes and is reverse-transcribed on the slide before library preparation. This provides a transcriptome-wide, unbiased profile at each capture location. A newer instrument, CytAssist, automates slide transfer for FFPE tissue Visium workflows. The GeoMx Digital Spatial Profiler (DSP) from NanoString uses a different approach: barcoded probes bind RNA throughout the section, and the user then selects specific regions of interest (ROIs) using morphology or protein expression markers, after which the probes from those ROIs are released and sequenced. The number of sequencing libraries scales with the number of ROIs, making cost variable.
Key performance characteristics: Among the platforms compared in the review, image-based smFISH and ISS methods offer higher spatial resolution but are limited to targeted gene panels and provide lower throughput. Sequencing-based methods like Visium and GeoMx DSP are transcriptome-wide (hypothesis-generating) but suffer from lower transcript detection efficiency and a lack of single-cell resolution: each Visium spot corresponds to approximately 1-10 cells, and GeoMx DSP ROIs cover 700-800 micrometers. The two platforms used in all published lymphoma studies as of 2023 are Visium and GeoMx DSP, reflecting their commercial maturity and FFPE tissue compatibility.
Generating spatial transcriptomic data is only the first step; extracting biologically meaningful insights requires sophisticated bioinformatic analysis. The review outlines the key processing steps, which differ substantially depending on whether image-based or sequencing-based methods were used, but converge on a shared goal: a gene-by-cell count matrix combined with a spatial coordinate matrix.
Preprocessing and normalization: For image-based spatial transcriptomics, preprocessing includes image registration, high-throughput transcript signal detection, and cell segmentation. For sequencing-based methods such as Visium, steps include image tiling, alignment of sequencing reads to reference genomes, and integration of the spatial coordinates with gene count data. Across all technologies, normalization must account for variable sequencing depth across capture spots while avoiding removal of true biological variance due to genuine differences in cell density across tissue regions.
Dimensionality reduction, clustering, and spatial visualization: The resulting high-dimensional gene expression data requires dimensionality reduction via principal component analysis (PCA) or manifold learning to visualize cells in 2D space. Clustering methods such as agglomerative clustering group capture locations with similar transcriptional profiles. A key advantage of spatial transcriptomics over standard scRNAseq is that these cluster assignments can then be projected back onto the tissue image, allowing researchers to see exactly where gene expression programs are active within the tissue architecture.
Cell-type deconvolution: Because Visium and GeoMx spots contain multiple cells rather than single cells, researchers integrate reference scRNAseq datasets from matched tissue types to deconvolute the composition of each spot. Mapping approaches assign cell-type annotations established from scRNAseq data to spatial locations. Deconvolution strategies calculate the probability that specific cell-type transcriptomes contribute to each capture area, allowing visualization of spatial patterns of cellular heterogeneity. Downstream analysis can then explore cell-to-cell interactions, inferred ligand-receptor pairings, and cell trajectory analysis.
Standardization challenges: The review specifically flags the lack of analytical standardization as a major limitation. Over 1,500 software tools exist for scRNAseq analysis alone, and distinct research groups analyze and store data in different formats, limiting reproducibility and data reusability. Large datasets combining spatial transcriptomics with proteomics and chromatin accessibility data create storage and computational challenges even in well-resourced laboratories.
Classic Hodgkin lymphoma (cHL) accounts for around 15% of lymphoma diagnoses and presents with a bimodal age distribution. Despite cure rates of 60-90% in the frontline setting, patients with relapsed or refractory disease have historically had limited options. The addition of PD-1 inhibitors such as pembrolizumab and nivolumab has substantially improved outcomes in this setting, which reflects the particular importance of immune evasion in cHL biology.
A unique microenvironment: The cHL tumor microenvironment is unusual in that malignant Hodgkin and Reed-Sternberg cells (HRSCs) represent only a tiny fraction of total lymph node cellularity. The remainder consists predominantly of T cells, NK cells, mononuclear phagocytes (MNPs), monocytes, and dendritic cells. These non-malignant cells are reprogrammed by HRSCs to sustain a tumor-tolerant environment, with regulatory T cells and exhausted CD4+ T cells accumulating in close proximity to tumor cells and expressing checkpoint molecules including LAG3, PD-1, and CTLA-4.
The GeoMx DSP study (Stewart et al., 2023): This study analyzed ten FFPE cHL lymph nodes using the NanoString GeoMx platform, selecting ROIs in both PD-L1-high and PD-L1-low areas. Deconvolution against single-cell transcriptome profiles of normal and pathological lymph nodes revealed two neoplastic PD-L1-high clusters with divergent immune cell localization. One HRSC cluster was enriched for T helper cells, exhausted CD4+ T cells, and NK cells; the other was specifically enriched for MNPs including classical monocytes, macrophages, and conventional dendritic cells (cDC2). The cDC2 and monocyte populations in this second cluster expressed immunoregulatory checkpoint molecules PD-L1, TIM-3, and IDO1 (the tryptophan-catabolizing enzyme), thereby contributing to a tumor-tolerant TME. Classical monocytes appeared to retain immunosuppressive and exhausted T cells in this niche while excluding plasmacytoid dendritic cells.
Ligand-receptor interaction analysis confirmed expression of inhibitory molecules by MNPs in close proximity to HRSCs. High expression of genes associated with the MNP-rich module correlated with early treatment failure in gene expression data, demonstrating a clinically relevant prognostic signature. This spatial polarization of tumor-associated MNPs around HRSCs identifies the inflammatory cDC2-monocyte-macrophage niche as a potential therapeutic target, potentially deployable alongside PD-1 blockade to deplete immunosuppressive cells nearest the tumor.
Diffuse large B-cell lymphoma (DLBCL) is the most common high-grade non-Hodgkin lymphoma and is itself biologically heterogeneous, with multiple classification schemes based on cell-of-origin, gene expression profiling, and genetic data. The review covers two published spatial transcriptomic studies in DLBCL and two in primary central nervous system lymphoma (PCNSL), a rare subtype confined to the CNS that shares morphological features with DLBCL but arises in an immune-privileged site.
DLBCL stromal networks (Sangaletti et al., 2020): Using NanoString GeoMx DSP on eight human FFPE DLBCL samples, this study profiled four ROIs per sample defined by smooth muscle actin (SMA)-positive and nerve growth factor receptor (NGFR)-positive stromal networks, representing myofibroblastic/reticular and mesenchymal stromal/pericytic cell populations, respectively. Differential gene expression analysis revealed that SMA-rich networks were enriched in immunoregulatory and vascular stroma-associated transcripts compared to NGFR-rich networks. Critically, the spatial data revealed intra-lesional heterogeneity where NGFR-rich foci within a tumor corresponded to downregulated MYC expression in the same area, suggesting that stromal composition influences the transcriptional state of nearby lymphoma cells and could impact disease progression.
DLBCL macrophage subsets (Liu et al., 2023): A tissue microarray (TMA) study applying GeoMx DSP to 47 DLBCL samples characterized eight distinct tumor-associated macrophage (TAM) subsets with different biological characteristics and distinct spatial locations within tumors. Key macrophage signatures included upregulation of CD163, complement genes, and TNF-alpha/NF-kB signaling pathways conferring a pro-tumor immunoregulatory profile. The spatial localization of these macrophage subsets provides a basis for evaluating their interactions with lymphoma cells and other immune cells as potential therapeutic targets.
PCNSL spatial immune dynamics (Heming et al. and Xia et al.): Heming et al. performed Visium spatial transcriptomics on four FFPE PCNSL biopsies, integrating their scRNAseq data. They identified four malignant B-cell clusters (mBc1-4), with the mBc4 cluster showing focal spatial enrichment and increased expression of immune checkpoint exhaustion markers (LAG3, PDCD1, HAVCR2, TIGIT), suggesting this subpopulation induces a locally stronger immunosuppressive environment. Xia et al. used Visium to study four PCNSL samples representing distinct TME categories: "hot" (T cell-rich), "cold" (T cell-poor), "invasive margin excluded" (IME), and "invasive margin immunosuppressed" (IMS). Developmental trajectory analysis revealed a TME remodeling pathway progressing from hot toward cold or IME states, with IMS as a transitional state where tumor and immune cells are in competition. FKBP5 was identified as a key gene associated with the barrier-forming process and tumor progression.
Follicular lymphoma (FL) is classified as an indolent, low-grade lymphoma, but clinical behavior is heterogeneous. Patients with progression of disease within 24 months (POD24) have significantly inferior outcomes. Specific TME features, including immune response gene expression signatures, are associated with disease progression and survival. Spatial transcriptomics is beginning to reveal how the spatial organization of malignant B cells within the lymph node follicle drives this heterogeneity.
Attaf et al. (2022) in follicular lymphoma: After using scRNAseq to identify recurrent malignant B-cell states driven by functional plasticity in response to TME cues, particularly signaling from T follicular helper (Tfh) cells, the authors performed Visium spatial transcriptomics on a single fresh frozen FL lymph node section. Distinct malignant B-cell states mapped to different tissue zones: germinal center-like and memory B-cell-like states preferentially localized to centrofollicular and interfollicular zones, respectively, while perifollicular (PF) zones contained multiple malignant B-cell states alongside abundant Tfh-activated cell states and markers of follicular dendritic cells (FDCs). These localizations suggest that distinct stromal and immune cell populations promote survival of particular malignant B-cell states within specific tissue niches, providing a spatial framework for understanding variable therapy responses and eventual relapse in FL.
AITL and CCR4-axis immunosuppression (Du et al., 2022): Angioimmunoblastic T-cell lymphoma (AITL) is derived from T follicular helper cells and carries a poor prognosis. Du et al. applied Visium Spatial Gene Expression to a single fresh frozen excised lymph node. Differential gene expression analysis in the TFH-germinal center (TFH-GC) region showed significant upregulation of CCL17 and CCL22 with spatial colocalization of T-regulatory cells (Tregs). There was also increased proportions of cycling B cells and vascular smooth muscle cells within the core tumor, and decreased NK cells and CD8+ cytotoxic T cells, consistent with an immunosuppressed TME. Both CCL17 and CCL22 are ligands for CCR4 and are known to recruit Tregs to the TME in cutaneous T-cell lymphoma (CTCL). The anti-CCR4 agent mogamulizumab has shown efficacy in CTCL, and Du et al.'s spatial data suggests the CCR4 axis as a candidate therapeutic target in AITL.
Notably, no published spatial transcriptomic studies had yet been conducted in CTCL at the time of this review, though imaging-based scRNAseq work by Phillips et al. had demonstrated that topographical differences in PD-1-positive T cells correlate with pembrolizumab response, highlighting what spatial transcriptomics could add in that subtype.
Small sample sizes and lack of generalizability: The seven studies summarized in the review include a combined total of only 75 patient samples, a number that severely limits translational impact given the significant heterogeneity both between and within lymphoma subtypes. Some studies were performed on a single patient sample. This is partially a consequence of the maturity of the technology and the cost involved, but it also reflects the difficulty of obtaining sufficient numbers of high-quality tissue samples from rare lymphoma subtypes. Small datasets restrict statistical power and increase the risk that observed spatial patterns reflect individual variation rather than generalizable biology.
Resolution limitations: Neither Visium nor GeoMx DSP currently offers true single-cell resolution. Visium's 55-micrometer capture spots correspond to approximately 1-10 cells, and GeoMx DSP ROIs cover 700-800 micrometers. Deconvolution methods mitigate this but introduce uncertainty: the accuracy of cell-type assignment depends on the quality and comprehensiveness of the reference scRNAseq dataset and the degree to which the reference tissue matches the pathological sample being deconvolved. While higher-resolution technologies such as seqFISH+ and MERFISH can resolve individual transcripts at near-cellular resolution, they remain limited to fresh frozen tissue and are substantially more expensive.
Bias from region-of-interest selection: The GeoMx DSP studies selected ROIs based on morphology and protein expression markers such as CD20, CD3, and PD-L1. While this targeted approach improves efficiency of transcript detection within cells of interest, it introduces selection bias: the resulting data does not represent the full complexity and heterogeneity of the tumor. Important but unanticipated spatial patterns outside the selected ROIs may be missed entirely.
Technical complexity, cost, and infrastructure: Spatial transcriptomic techniques remain technically demanding in both sample preparation and downstream bioinformatic analysis. The expense associated with high-throughput gene capture, sequencing depth, and dedicated computational infrastructure is substantially higher than conventional IHC and immunofluorescence methods. The lack of standardized reporting guidelines for spatial transcriptomic studies (analogous to MIAME for microarray data) makes cross-study comparison difficult. The rapidly expanding tooling landscape, with over 1,500 scRNAseq analysis tools, creates further fragmentation in analytical approaches.
Deep learning integration: The combination of spatial transcriptomics and deep learning models offers a direct path toward improved pathological classification of lymphoma and the identification of diagnostic, prognostic, and predictive biomarkers. Deep learning models trained on spatial transcriptomic data could identify individual gene markers, cell subpopulation abundances, and gene expression signatures associated with treatment response, potentially enabling tissue-based precision medicine in routine clinical samples. This integration is particularly promising for improving subtype classification in histologically ambiguous cases and for identifying resistance mechanisms before they become clinically apparent.
Spatial epigenomics and multi-omics: Epigenetic alterations, including DNA methylation (DNMT1), histone acetylation (CREBBP), histone methylation (EZH2), and non-coding RNA (miR-155), play key roles in lymphoma pathogenesis by regulating gene expression, tumor cell biology, and immune cell activation within the TME. The technology to study epigenomics at single-cell resolution through chromatin accessibility profiling (ATAC-seq) can now be combined with spatial barcoding to produce spatial-ATAC-seq, enabling spatial epigenetic mapping alongside transcriptomics. This is clinically relevant given that personalized therapies targeting epigenetic alterations, such as tazemetostat (an oral EZH2 inhibitor) for relapsed/refractory follicular lymphoma, are already in use. Spatial multi-omics will identify the genomic and epigenomic targets most likely to benefit individual patients within specific TME contexts.
Spatial organization of tumor cell heterogeneity: Beyond TME-immune interactions, spatial transcriptomics also reveals the spatial organization of lymphoma cell subpopulations themselves. Identifying lymphoma cell clusters that induce variable local immunosuppression at the tumor invasive margin versus the tumor core will clarify how spatial architecture drives therapy resistance and relapse. This information supports the development of therapeutic strategies targeting the spatial organization of tumor subclones rather than treating all tumor cells uniformly.
Retrospective and prospective applications: The compatibility of Visium and GeoMx DSP with FFPE tissue enables retrospective analysis of archived biopsy material from clinical trials, allowing spatial transcriptomic data to be correlated with long-term outcome data already available in completed trial cohorts. As datasets accumulate, they will provide spatially resolved single-cell transcriptional profiles across lymphoma subtypes, generating fundamental insights into immune evasion mechanisms and defining spatial biomarkers that predict response to immunotherapy. The authors conclude that spatial transcriptomics will be an essential tool in the march toward precision medicine for lymphoma.