By Jason An, Ph.D., CEO, Medicus
Last updated: 2026-10

A spatial biomarker is a measurable feature of tissue that describes where cells are and how they are arranged relative to each other — for example, the density of T cells at the tumor's invasive margin or the distance between immune cells and tumor cells — rather than only how much of a marker is present. Because it preserves tissue context, a spatial biomarker can capture biology of the tumor microenvironment that bulk expression measurements average away.

Why "how much" is no longer enough

Most established tissue biomarkers answer a quantity question: what percentage of tumor cells stain for a protein, or how many mutations a tumor carries. These measurements are valuable, but they discard information about location. A tumor where cytotoxic T cells sit right next to cancer cells and a tumor where the same number of T cells are trapped in the surrounding stroma can produce a similar bulk count — yet they may behave very differently.

Single-cell sequencing has revealed enormous heterogeneity within tumors, but as Lewis and colleagues note in Nature Methods, those approaches lose information on cellular location within the tumor microenvironment; spatial technologies were developed to recover it [1].

The clinical case for spatial information is strongest in immuno-oncology. In a systematic review and meta-analysis covering more than 10 solid tumor types and 8,135 patients, multiplex immunohistochemistry/immunofluorescence (mIHC/IF) — a modality that keeps multiple cell types in their native position on the same section — had a pooled AUC of 0.79 for predicting response to anti-PD-1/PD-L1 therapy, compared with 0.65 for PD-L1 IHC, 0.69 for tumor mutational burden, and 0.65 for gene expression profiling [2]. The authors call for larger confirmatory studies, but the signal is consistent with the idea that cellular context carries predictive information.

What counts as a spatial biomarker?

"Spatial biomarker" is an umbrella term. In practice, teams measure one or more of the following.

1. Compartment-specific cell density

Counting a cell type separately in defined tissue regions — tumor core, invasive margin, stroma. The best-known example is the Immunoscore, which quantifies CD3+ and CD8+ T-cell densities in the tumor and invasive margin by digital pathology. In an international validation across 14 centres in 13 countries and 2,681 patients with stage I–III colon cancer, the assay was highly reproducible between observers and centres (r = 0.97), and a high Immunoscore was associated with the lowest 5-year recurrence risk [3].

2. Infiltration on routine H&E

Stromal tumor-infiltrating lymphocytes (TILs) can be assessed on standard H&E sections. The International TILs Working Group published a standardized methodology for breast cancer, noting TILs' prognostic and potentially predictive value particularly in triple-negative and HER2-positive disease [4].

3. Spatial patterns and architecture

Beyond counts, the shape of an infiltrate matters. Saltz et al. used a convolutional neural network to map TILs on H&E whole slide images from 13 TCGA tumor types and found that local spatial structure of TIL patterns correlated with overall survival and differed across tumor and immune subtypes [5].

4. Cell–cell proximity and neighborhoods

Distances between specific cell types (e.g., PD-L1+ cells and CD8+ T cells), or recurring "neighborhoods" of cell types, typically measured with multiplex IHC/IF or other multiplexed imaging [1][2].

5. Spatially resolved molecular profiles

Spatial transcriptomics and spatial proteomics read out RNA or protein in place, from tens to thousands of targets [1]. These are powerful discovery tools; most are currently research platforms rather than routine diagnostics.

Spatial biomarker types vs. measurement technology

How AI changes spatial biomarker measurement

Spatial metrics are, by nature, computational. A single whole slide image contains a very large number of cells; measuring every cell's type and position by eye is not feasible at scale. Reviews of digital pathology describe how whole slide imaging, faster networks and cheaper storage, combined with machine learning, enable image-based analysis that was previously impractical [7]. The Digital Pathology Association's white paper on tissue image analysis gives pathologists an overview of the general workflow, analysis strategies, technical considerations and typical algorithm readouts, along with their advantages and limitations [8].

AI also addresses a well-documented reproducibility problem. In a multi-institutional study of four PD-L1 assays in non-small cell lung cancer, 13 pathologists showed excellent concordance when scoring PD-L1 on tumor cells (ICC 0.832–0.882) but poor concordance when scoring immune cells (ICC 0.172–0.229) [6]. Quantitative, cell-level measurement is one route to more consistent readouts — provided the algorithm itself is validated.

A typical AI pipeline for a spatial biomarker looks like this:

1. Digitize slides as whole slide images (WSI) with consistent scanning and quality control.
2. Segment tissue compartments — tumor, stroma, invasive margin, necrosis.
3. Detect and classify cells — tumor cells, lymphocytes, macrophages, or marker-positive cells.
4. Compute spatial features — densities per compartment, nearest-neighbor distances, clustering, neighborhood composition.
5. Link features to outcomes — response, recurrence, survival — in a cohort with appropriate clinical annotation.
6. Lock and validate the feature definition on independent cohorts before any clinical use.

Clinical and research uses

• Prognosis. Compartment-specific T-cell density adds risk information in colon cancer [3]; TILs carry prognostic value in breast cancer subtypes [4].
• Prediction of therapy response. Spatial and multiplex readouts are being studied as predictors of immune checkpoint inhibitor response [2].
• Drug development. Pharma teams use spatial readouts to characterize mechanism of action, stratify trial populations and explore resistance — areas where tissue context is often the missing variable [1].
• Standardizing subjective scores. Quantification can reduce reader variability for difficult scores such as immune-cell PD-L1 [6].

Practical pitfalls to plan for

• Pre-analytics. Fixation, section thickness and staining variation change what an algorithm sees. Standardize upstream before blaming the model.
• Scanner and format diversity. Algorithms trained on one scanner's images may degrade on another's. Data infrastructure that normalizes and tracks image provenance reduces this risk.
• Definition drift. "Invasive margin" or "intratumoral" must be defined precisely and identically across sites — the Immunoscore consortium's reproducibility depended on standardized protocols [3].
• Cohort bias. A feature discovered in one hospital's population needs external validation.
• Regulatory status. Most spatial biomarker assays today are research-use tools. Check the intended use and regulatory status of any assay before considering clinical decisions.

How Medicus approaches this

Medicus, Inc. is a healthcare AI company focused on spatial biomarkers and cell-based artificial intelligence. Our Spatial Biomarkers work converts the location, type and tissue-structure context of cells into quantitative indicators for diagnostic, prognostic and drug-development research — a biomarker approach that goes beyond expression levels to interpret spatial context. Our models are built on gastric and breast cancer clinical datasets developed and validated together with tertiary hospitals in Korea.

For PD-L1, where interpretation can be subjective and vary by tissue region, MeDICus® is designed to support quantitative, consistent assessment through AI quantification (research use; it is not presented here as an approved diagnostic). Because spatial analysis depends on well-organized image data, many teams start by consolidating whole slide images with MeDIAuto®, our pathology data infrastructure, and analyze sensitive data in a controlled Cloud Safe Zone.

If you are planning a spatial biomarker study or pilot, talk to our team.

FAQ

What is the difference between a spatial biomarker and a conventional biomarker?

A conventional tissue biomarker usually reports an amount — such as the percentage of cells expressing a protein. A spatial biomarker reports location and relationships: where cells sit, how densely they populate specific regions, and how close different cell types are to each other.

Do spatial biomarkers require special staining?

Not always. Some spatial features, such as stromal TILs or TIL spatial patterns, can be derived from routine H&E slides [4][5]. Others, such as proximity between specific immune phenotypes, typically require multiplex IHC/IF or other multiplexed imaging [1][2].

Are spatial biomarkers better than PD-L1 IHC for predicting immunotherapy response?

In one meta-analysis, multiplex IHC/IF had a higher pooled AUC (0.79) than PD-L1 IHC (0.65) for anti-PD-1/PD-L1 response [2]. The authors state that larger confirmatory studies are needed, so spatial approaches should be viewed as promising rather than as replacements for approved companion diagnostics.

What is tumor microenvironment AI?

Tumor microenvironment AI refers to algorithms that analyze tissue images to identify tumor, immune and stromal cells and quantify how they are organized. It is the computational engine that makes spatial biomarkers measurable at the scale of whole slide images.

Can spatial biomarkers be used in clinical diagnosis today?

Some, such as the Immunoscore in colon cancer, have been extensively validated [3]. Many others remain research tools. Clinical use depends on the specific assay's validation and regulatory status in each country.

What data do I need to start a spatial biomarker project?

Digitized whole slide images with consistent quality, matched clinical annotations (treatment, response, outcome), and a clear question. Organizing images from multiple scanners into a single, well-governed repository is often the first practical step.

This guide is educational and is not clinical advice.

References

1. Lewis SM, Asselin-Labat ML, Nguyen Q, et al. Spatial omics and multiplexed imaging to explore cancer biology. Nat Methods. 2021;18(9):997-1012. PubMed 34341583
2. Lu S, Stein JE, Rimm DL, et al. Comparison of biomarker modalities for predicting response to PD-1/PD-L1 checkpoint blockade: a systematic review and meta-analysis. JAMA Oncol. 2019;5(8):1195-1204. PubMed 31318407
3. Pagès F, Mlecnik B, Marliot F, et al. International validation of the consensus Immunoscore for the classification of colon cancer: a prognostic and accuracy study. Lancet. 2018;391(10135):2128-2139. PubMed 29754777
4. Salgado R, Denkert C, Demaria S, et al. The evaluation of tumor-infiltrating lymphocytes (TILs) in breast cancer: recommendations by an International TILs Working Group 2014. Ann Oncol. 2015;26(2):259-271. PubMed 25214542
5. Saltz J, Gupta R, Hou L, et al. Spatial organization and molecular correlation of tumor-infiltrating lymphocytes using deep learning on pathology images. Cell Rep. 2018;23(1):181-193.e7. PubMed 29617659
6. Rimm DL, Han G, Taube JM, et al. A prospective, multi-institutional, pathologist-based assessment of 4 immunohistochemistry assays for PD-L1 expression in non-small cell lung cancer. JAMA Oncol. 2017;3(8):1051-1058. PubMed 28278348
7. Niazi MKK, Parwani AV, Gurcan MN. Digital pathology and artificial intelligence. Lancet Oncol. 2019;20(5):e253-e261. PubMed 31044723
8. Aeffner F, Zarella MD, Buchbinder N, et al. Introduction to digital image analysis in whole-slide imaging: a white paper from the Digital Pathology Association. J Pathol Inform. 2019;10:9. PubMed 30984469