Decoding the Tumor Ecosystem

For decades, studying the tumor microenvironment (TME) meant cataloguing its contents. But a new generation of technology is moving beyond characterization and toward predicting how tumors will behave and who will respond to treatment. The TME is more than a backdrop for cancer growth. It is a dynamic, densely populated ecosystem in which immune cells, stromal cells, blood vessels, and even microorganisms collectively determine whether a tumor thrives, spreads, or yields to therapy. Understanding this complexity has long been a priority; a convergence of single-cell genomics, high-parameter imaging, and multi-omics platforms is now making it possible.

Here we explore how these tools are being applied, what they are revealing about tumor biology, and what still stands in the way of translating these discoveries into clinical practice.

Single-cell and spatial transcriptomics

Traditional bulk RNA sequencing averages gene expression across millions of cells, a limitation that masks the cellular heterogeneity now recognized as a primary driver of treatment failure in cancer.¹ Single-cell RNA sequencing (scRNA-seq) addresses this by profiling individual cells, but sacrifices the critical dimension of location. Spatial transcriptomics restores that context, mapping gene expression directly onto intact tissue architecture.

By integrating scRNA-seq and spatial transcriptomics, researchers can jointly uncover cellular heterogeneity, stromal-immune interactions, and spatial niches underlying tumor progression and therapy resistance.

Pancreatic ductal adenocarcinoma (PDAC) carries a five-year survival rate of just 10–20%. Effective targeted therapies are scant, and checkpoint immunotherapy has largely failed due to the tumor's dense, immunosuppressive microenvironment. Contributing to that poor outlook is perineural invasion (PNI), the infiltration of tumor cells along peripheral nerves, which carries the same prognostic weight as lymph node metastasis.² Using this combined approach to investigate PNI, researchers found distinct immune and stromal cell populations co-localizing with sites of low and high invasion. These findings point toward specific, spatially defined targets and open new avenues in a disease where options have long been limited.³

Taking the spatial approach further still, researchers applied spatial transcriptomics, single-nucleus RNA sequencing (snRNA-seq), and CODEX multiplex protein imaging to tumor sections from 78 patients across six cancer types. snRNA-seq sequences RNA from isolated nuclei and is particularly suited to frozen tumor tissue, and CODEX uses DNA-barcoded antibodies to simultaneously image dozens of proteins in intact tissue. The analysis mapped genomically distinct spatial microregions within tumors and their surrounding microenvironments. Tumor centers showed increased metabolic activity, while leading edges were characterized by heightened immune cell activity. They also identified both immune hot and cold neighborhoods within the same tumor. The authors conclude that mapping these spatial patterns offers new ways of understanding how tumors evolve and resist treatment.⁴

Beyond individual cancers, a 2025 study constructed TabulaTIME, a pan-cancer single-cell compendium built from over 4.4 million cells across 36 malignancies, offering a comprehensive blueprint of the TME. This resource identified a profibrotic barrier present consistently across tumor types. A subset of cancer-associated fibroblasts, involved in extracellular matrix remodeling and immune suppression, were positioned at the leading edge between malignant and normal tissue. There, they co-localized with a profibrotic macrophage subset that promotes tissue fibrosis (the excessive buildup of connective tissue), stiffening the TME and further impeding immune infiltration. If this fibroblast-macrophage interaction is a shared feature, it becomes a therapeutic target not just for one disease, but potentially for many, a shift from targeting individual cancers to targeting the TME itself.⁵

High-parameter imaging

Single-cell compendiums like TabulaTIME reveal what cell types exist and where, but they do so through the lens of gene expression. Proteins tell a different part of the story. They are the functional output of gene activity, and mapping their distribution within intact tissue adds a layer of biological detail that transcriptomics alone cannot provide.

Multiplex immunohistochemistry (mIHC), multiplex immunofluorescence (mIF), and imaging mass cytometry (IMC) bring protein-level spatial resolution to the TME, and the technology is rapidly advancing. IMC uses metal isotope-labeled antibodies and laser ablation to detect up to 40 or more protein targets simultaneously in intact tissues, avoiding the spectral overlap limitations that constrain conventional fluorescence-based imaging. Standard IMC operates at a spatial resolution of approximately one micron, sufficient to resolve individual cells and large subcellular compartments. More recently, scientists were able to achieve a resolution under 350 nanometers in IMC, approaching that of light microscopy. This advance moves spatial protein profiling from the cell level to the organelle level, enabling visualization of subcellular structures without disrupting native tissue context.⁶

By simultaneously profiling dozens of proteins across hundreds of thousands of cells in their tissue landscape, these platforms are already generating findings with direct therapeutic relevance that bulk approaches cannot provide. Applying a 35-marker IMC panel to tumor tissues from 26 melanoma patients receiving anti-PD-1(programmed death-1 receptor) therapy, researchers profiled over 662,000 single cells, clustering cells by shared protein expression patterns and characterizing distinct tumor-immune microenvironments. The analysis identified six distinct TME subtypes associated with different clinical responses to treatment, illustrating how spatially resolved protein data can help identify the patients most likely to benefit from immunotherapy.⁷

While IMC offers unmatched multiplexing depth, mIHC and mIF bring a distinct and practical advantage: they work on standard fluorescence microscopes already found in most research and clinical pathology laboratories. Rather than staining for one or two markers per slide, mIHC and mIF use sequential cycles of staining and imaging on the same section, each round adding new protein targets. The resulting images are computationally aligned to produce a single composite view of up to 40 or more proteins simultaneously, all within preserved tissue organization.

For these tools to reach the clinic, findings need to be reproducible across platforms. A 2025 study found consistent prognostic biomarkers across three different multiplex imaging platforms in 102 breast cancer patients, suggesting that spatially defined protein biomarkers can hold up across technologies, an important step toward standardization.⁸

Several commercial platforms now support this workflow, including Akoya's PhenoCycler-Fusion and the ZEISS Axioscan 7, an open platform compatible with multiple mIF assay formats. Their compatibility with FFPE tissue, the standard format for archived clinical specimens, removes one significant barrier to clinical adoption, though challenges around protocol standardization and image analysis workflows remain.

Multi-omics integration

No single technology can capture the TME in its entirety. That recognition has driven a shift toward multiomics frameworks that layer genomic, transcriptomic, epigenomic, proteomic, and metabolomic data onto the same tumor samples, with artificial intelligence providing the computational power to interpret the result.

A recent Stanford study illustrates the diagnostic and therapeutic power of this integration. Gliomas are among the most lethal brain cancers, with a five-year survival rate below 10% for the most aggressive form and limited therapeutic options. Applying MIBI-TOF for spatial protein mapping, digital spatial profiling for transcriptomics, and MALDI mass spectrometry imaging to map N-glycans (cell-surface sugar modifications that affect immune recognition) across 670 lesions from 310 adult and pediatric patients, researchers quantified the expression of key tumor antigens currently targeted in clinical trials, including epidermal growth factor receptor (EGFR) and programmed death-ligand 1 (PD-L1). Most gliomas expressed these targetable antigens in less than 50% of tumor cells, a finding that may explain why patients who initially respond to targeted therapy often relapse. Bulk approaches, which average signal across the entire tumor, would have likely obscured these antigen-negative pockets, which can drive recurrence.⁹

Two recently described deep learning models illustrate how artificial intelligence is beginning to lower the cost and accessibility barriers that have historically limited spatial omics. MISO is trained to predict spatial gene expression patterns from standard hematoxylin and eosin (H&E) stained slides, the same slides generated routinely in every pathology laboratory. Validated across hundreds of tissue samples spanning five cancer types, MISO produces near single-cell-resolution spatial gene expression data from slides that would otherwise yield no molecular information.¹⁰

HEX takes a complementary approach, predicting spatial protein expression from the same standard H&E slides. Trained on over 819,000 image tiles from 382 tumor samples, HEX accurately predicts the expression of 40 protein biomarkers spanning immune, structural, and functional programs, and has been shown to improve survival and treatment response prediction in lung cancer patients.¹¹ Together, these models suggest that the molecular detail embedded inside routine pathology slides may be far greater than previously appreciated, and extracting it may not require specialized equipment.

Looking forward

The challenge now is translating these discoveries into routine practice, and the evidence that these tools are ready is growing. In 2025, the Society for Immunotherapy of Cancer convened a task force to establish best practices for mIF and mIHC, identifying these approaches as among the most promising emerging biomarkers for predicting immunotherapy response and noting their superior predictive accuracy over established modalities currently used in the clinic.¹² Cost and standardization remain real barriers, but the tools are maturing, the evidence is accumulating, and the distance between discovery and practice is closing.

FAQ: Decoding the Tumor Ecosystem

Q1: What is shifting TME research beyond just cataloging tumor contents?

New tech like single-cell/spatial transcriptomics, high-parameter imaging, and multiomics predicts tumor behavior and treatment responses by mapping dynamic cellular interactions in the tumor microenvironment.

Q2: How do spatial tools reveal cancer resistance mechanisms?

They expose heterogeneity, like profibrotic barriers in pan-cancer studies or perineural invasion in PDAC, identifying immune-cold niches and stromal interactions that block therapy.

Q3: What role does high-parameter imaging play?

Techniques like IMC and mIHC profile dozens of proteins at subcellular resolution, linking TME subtypes to outcomes—e.g., six subtypes predicting anti-PD-1 response in melanoma patients.

Q4: Are these tools ready for clinical use?

Yes, despite standardization challenges; SITC 2025 guidelines endorse mIF/mIHC as top biomarkers, with AI models like MISO/HEX unlocking insights from routine pathology slides.

References

1. Fu YC, Liang SB, Luo M, Wang XP. Intratumoral heterogeneity and drug resistance in cancer. Cancer Cell Int. 2025;25(1):103. Published 2025 Mar 18. doi:10.1186/s12935-025-03734-w

2. Selvaggi F, Bannone E, Melchiorre E, Diana M, Cotellese R, Aceto GM. Perineural invasion in pancreatic cancer: Current biological function in R status, prognosis, and pain. Surg Open Sci. 2025;24:58-60. Published 2025 Feb 26. doi:10.1016/j.sopen.2025.02.007

3. Chen MM, Gao Q, Ning H, et al. Integrated single-cell and spatial transcriptomics uncover distinct cellular subtypes involved in neural invasion in pancreatic cancer. Cancer Cell. 2025;43(9):1656-1676.e10. doi:10.1016/j.ccell.2025.06.020

4. Mo CK, Liu J, Chen S, et al. Tumour evolution and microenvironment interactions in 2D and 3D space. Nature. 2024;634(8036):1178-1186. doi:10.1038/s41586-024-08087-4

5. Han, Y., Zhang, L., Sun, D. et al. Spatiotemporal analyses of the pan-cancer single-cell landscape reveal widespread profibrotic ecotypes associated with tumor immunity. Nat Cancer 6, 1880–1898 (2025). doi.org/10.1038/s43018-025-01039-5

6 Bollhagen A, Whipman J, Coelho R, Heinzelmann-Schwarz V, Jacob F, Bodenmiller B. High-resolution imaging mass cytometry to map subcellular structures. Nat Methods. 2025;22(12):2601-2608. doi:10.1038/s41592-025-02889-8

7. Xiao X, Guo Q, Cui C, et al. Multiplexed imaging mass cytometry reveals distinct tumor-immune microenvironments linked to immunotherapy responses in melanoma. Commun Med (Lond). 2022;2:131. Published 2022 Oct 21. doi:10.1038/s43856-022-00197-2

8. Eng JR, Bucher E, Hu Z, et al. Highly multiplexed imaging reveals prognostic immune and stromal spatial biomarkers in breast cancer. JCI Insight. 2025;10(3):e176749. Published 2025 Jan 14. doi:10.1172/jci.insight.176749

9. Piyadasa H, Oberlton B, Ribi M, et al. Multi-omic landscape of human gliomas from diagnosis to treatment and recurrence. Cancer Cell. 2026;44(1):112-128.e6. doi:10.1016/j.ccell.2025.11.006

10. Schmauch B, Herpin L, Olivier A, et al. A deep learning-based multiscale integration of spatial omics with tumor morphology. Nat Commun. 2025;16(1):11674. Published 2025 Nov 27. doi:10.1038/s41467-025-66691-y

11. Li Z, Li Y, Xiang J, et al. AI-enabled virtual spatial proteomics from histopathology for interpretable biomarker discovery in lung cancer. Nat Med. 2026;32(1):231-244. doi:10.1038/s41591-025-04060-4

12. Taube JM, Sunshine JC, Angelo M, et al. Society for Immunotherapy of Cancer: updates and best practices for multiplex immunohistochemistry (IHC) and immunofluorescence (IF) image analysis and data sharing. J Immunother Cancer. 2025;13(1):e008875. Published 2025 Jan 8. doi:10.1136/jitc-2024-008875