Spatial Proteomics vs Spatial Transcriptomics for IO Biomarker Discovery

Spatial Proteomics vs Spatial Transcriptomics for IO Biomarker Discovery

Immuno-oncology biomarkers have moved beyond simply asking which cells are present. More and more, the question is where those cells sit and which cells they sit next to. A CD8+ T cell touching a tumour cell does not tell you the same thing as one held back in the stroma, and a tertiary lymphoid structure at the invasive margin is a different signal from scattered B cells. Spatial biology measures this structure directly, from a tissue section.

Spatial biology has two main families. Spatial proteomics images proteins in situ, using multiplex immunofluorescence, cyclic staining or mass-tag imaging. Spatial transcriptomics maps RNA, either by capturing it on barcoded arrays for sequencing or by imaging it molecule by molecule. Both now work on FFPE tissue, and both have been used in IO studies. They answer different questions, though, and they sit at very different distances from a validated clinical assay.

This guide compares the two for translational scientists and biomarker leads. It covers what each technology measures, how the main platform classes compare, the published IO evidence, the hurdles on the way to the clinic, and how to choose for your programme.

Key takeaways

• Spatial proteomics measures a curated panel of proteins at single-cell resolution. It reads out cell phenotype and functional state directly, and multiplex immunofluorescence has the most clinical-grade IO evidence of any spatial method.
• Spatial transcriptomics measures hundreds of genes up to the whole transcriptome. It is the stronger discovery tool, but data are sparser per cell and the analysis burden is higher.
• In a 2019 JAMA Oncology meta-analysis, multiplex IHC/IF had a higher pooled AUC (0.79) for predicting anti-PD-1/PD-L1 response than PD-L1 IHC, TMB or gene expression profiling.
• A practical pattern: discover broadly with transcriptomics or high-plex proteomics, then lock a low-plex protein panel for trial use.
• The vendor landscape has consolidated. Bruker bought NanoString’s business in 2024, Quanterix acquired Akoya in 2025 and Bio-Techne bought Lunaphore in 2023.

What spatial proteomics measures

Spatial proteomics detects proteins in intact tissue with labelled antibodies and records where each signal sits, usually at single-cell or subcellular resolution. Proteins are the functional layer: PD-L1 on a macrophage, Ki-67 in a T cell, or granzyme B in a cytotoxic cell are all read directly, rather than inferred from mRNA. The main technology classes are:

• Multiplex IHC and multispectral immunofluorescence (mIHC/mIF). Tyramide signal amplification with spectral unmixing on a standard autostainer and whole-slide scanner. Panels are usually small: the AstroPath and MITRE studies discussed below both used six markers. The approach is closest to routine pathology and has the most multi-site reproducibility data.
• Cyclic immunofluorescence. Repeated rounds of staining, imaging and signal removal build up much larger panels on one section. Examples include t-CyCIF (Lin et al., eLife 2018), CODEX, now sold as PhenoCycler (Goltsev et al., Cell 2018), and COMET, an automated sequential immunofluorescence instrument developed by Lunaphore.
• Mass-tag imaging. Antibodies carry metal isotopes instead of fluorophores and are read by mass spectrometry, which removes spectral overlap. Imaging mass cytometry (IMC) uses laser ablation (Giesen et al., Nature Methods 2014), and multiplexed ion beam imaging (MIBI) uses a primary ion beam (Angelo et al., Nature Medicine 2014). Both reach tens of markers, but acquisition is slower, so studies usually image selected regions rather than whole slides.

Every one of these depends on antibody performance. Validation of each clone, in each panel position, on each tissue type, is where most of the effort goes.

What spatial transcriptomics measures

Spatial transcriptomics locates RNA in tissue. It can cover far more genes than any antibody panel, which makes it powerful for discovering unexpected cell states, signalling pathways and ligand-receptor interactions. It does not measure protein directly, and mRNA levels do not always track protein levels.

Sequencing-based methods

Transcripts (or probes that bind them) are captured on a surface carrying spatial barcodes, then read by next-generation sequencing. The field grew out of the array method described by Ståhl et al. in Science (2016).

• Visium and Visium HD (10x Genomics). Standard Visium uses 55 µm spots spaced 100 µm apart, so each spot averages several cells. Visium HD replaces the spots with a continuous grid of 2 µm barcoded squares, which are usually binned (for example to 8 µm) for analysis. Both support whole-transcriptome, probe-based workflows on FFPE.
• Stereo-seq (STOmics). Uses DNA nanoball-patterned arrays to combine high resolution with a large field of view (Chen et al., Cell 2022).
• Slide-seq. Captures RNA on a monolayer of 10 µm DNA-barcoded beads (Rodriques et al., Science 2019).

Imaging-based methods

Imaging-based methods detect individual RNA molecules in situ through cycles of hybridisation and imaging, giving subcellular resolution with targeted panels. The main commercial platforms are Xenium (10x Genomics), CosMx (Bruker, formerly NanoString) and MERSCOPE (Vizgen). All three support FFPE. Panels now run from a few hundred genes to several thousand, for example the Xenium Prime 5K panels. Cell segmentation, which assigns transcripts to cells, is a key analytical step and a known source of error.

Region-based profiling, such as GeoMx Digital Spatial Profiler, falls between these groups. It measures RNA or protein from user-selected regions rather than single cells, and it supplied the digital spatial profiling data in the melanoma TLS study discussed below.

Spatial proteomics vs spatial transcriptomics: comparison table

The table compares platform classes rather than products, because specifications change with each release. Cost tiers are relative and depend heavily on panel size, tissue area and whether the work is done in-house or through a service provider.

In short, protein methods trade breadth for interpretability and assay maturity, and RNA methods trade those for breadth. For a broader view of protein platforms outside the tissue context, see our multiplex immunoassay platform comparison guide.

IO use cases and the published evidence

Tumour microenvironment phenotyping

High-plex spatial proteomics has shown that tumours sharing a histological subtype can differ greatly in architecture. Using MIBI on triple-negative breast cancer, Keren et al. (Cell, 2018) quantified 36 proteins in 41 patients and described “mixed” and “compartmentalised” tumour-immune architectures, with ordered immune structures at the tumour-immune border linked to survival. Jackson et al. (Nature, 2020) used imaging mass cytometry with 35 markers on tumours from 352 patients to define breast cancer subgroups by cellular composition and tissue organisation, with distinct clinical outcomes.

CD8 location and proximity as response biomarkers

Spatial context was central to one of the first mechanistic studies of PD-1 blockade. In metastatic melanoma treated with pembrolizumab, Tumeh et al. (Nature, 2014) found that responders had more CD8+, PD-1+ and PD-L1+ cells at the invasive margin and inside the tumour, with close proximity between PD-1 and PD-L1. A model based on CD8 density at the invasive margin was validated in a small independent cohort.

Lu et al. (JAMA Oncology, 2019) pooled data from more than 8,000 patients across over ten solid tumour types. Multiplex IHC/IF had a pooled AUC of 0.79 for predicting anti-PD-1/PD-L1 response, compared with 0.65 for PD-L1 IHC, 0.65 for gene expression profiling and 0.69 for TMB. The authors called for larger confirmatory studies. The AstroPath study (Berry et al., Science, 2021) then used whole-slide six-marker multispectral imaging in melanoma. It identified pretreatment features, including CD163+PD-L1− myeloid cells and CD8+FoxP3+PD-1low/mid T cells, that stratified survival after anti-PD-1 therapy, and validated the signature in a cohort from another institution.

For context on how these compare with the approved single-analyte markers, see our articles on PD-L1 as a predictive biomarker and TMB.

Tertiary lymphoid structures

Three papers in Nature in January 2020 put B cells and tertiary lymphoid structures (TLS) at the centre of IO biomarker research. Cabrita et al. found that co-occurring CD8+ T cells and CD20+ B cells were linked to better survival in metastatic melanoma. They confirmed TLS by immunofluorescence and derived a TLS gene signature that predicted outcomes in cohorts treated with checkpoint inhibitors. Helmink et al. found B cell markers enriched in responders and located those B cells within TLS. Later, Meylan et al. (Immunity, 2022) used spatial transcriptomics in renal cell carcinoma to show that TLS support B cell maturation into antibody-producing plasma cells. In that study, IgG-stained tumour cells correlated with response and progression-free survival in patients treated with checkpoint inhibitors. TLS research is a good example of the two modalities working together: protein imaging defines the structure, and transcriptomics explains what happens inside it.

Cellular neighbourhoods

Spatial data can also be analysed as recurring neighbourhoods of cell types, as well as by distances between pairs of cells. Using CODEX re-engineered for FFPE tissue microarrays with 56 markers, Schürch et al. (Cell, 2020) identified nine conserved cellular neighbourhoods at the colorectal cancer invasive front and linked how they were coupled and fragmented to outcome. In cutaneous T cell lymphoma treated with pembrolizumab, Phillips et al. (Nature Communications, 2021) found no difference in cell frequencies between responders and non-responders. They did find differences in the spatial arrangement of effector PD-1+CD4+ T cells, tumour cells and Tregs. The resulting “SpatialScore” was then checked on a simpler, clinically accessible imaging platform. That step, from high-plex discovery to a low-plex readout, is the translational pattern to plan for.

Clinical translation hurdles

Standardisation

Pre-analytical variables (fixation, section age, antigen retrieval), antibody clone choice, panel design, scanner settings and image analysis parameters all change the output. The Society for Immunotherapy of Cancer published best practices for mIHC/mIF staining and validation in 2020, covering optimisation against chromogenic IHC and cross-platform comparability. Spatial transcriptomics does not yet have an equivalent consensus for clinical use, and RNA quality in archival FFPE blocks is a further variable.

Analytical validation

Reproducibility across sites is the precondition for any multi-centre trial. The MITRE study (Taube et al., 2021) tested a six-plex PD-1/PD-L1 axis panel across six institutions with a locked image analysis algorithm. After optimisation it reported an average 90% equivalence with single-plex IHC and strong inter-site concordance for cell densities and PD-1/PD-L1 proximity. Few spatial transcriptomics assays have comparable multi-site data so far. For the wider framework of fit-for-purpose, analytical and clinical validation, see our guide on how to validate an IO biomarker in a clinical trial.

Regulatory path

In FDA’s list of authorised companion diagnostics (content current as of 28 September 2026), the tissue-based tests are single-marker IHC or in situ hybridisation assays, including the PD-L1 IHC assays used for checkpoint inhibitors. We found no multiplex immunofluorescence or spatial imaging assay on the list. A spatial biomarker that selects patients would therefore be a first of its kind, needing a locked assay, a locked algorithm and a co-development plan with the drug. In the US, the FDA’s 2024 rule on laboratory-developed tests was vacated by a federal court in March 2025 and formally rescinded in September 2025. Central CLIA laboratories remain a common route for exploratory and trial-stage spatial assays. Our FDA companion diagnostic approval guide and CDx co-development article cover these pathways in detail.

Digital pathology integration

Every spatial readout is an image analysis output, so it depends on a digital pathology stack. The precedents exist: FDA authorised the first whole slide imaging system for primary diagnosis in April 2017, and in 2021 Paige Prostate became, according to the company, the first AI-based pathology product to receive De Novo authorisation. Multiplex fluorescence images, segmentation models and spatial metrics add further complexity: file formats, storage, algorithm locking, version control, and pathologist review of fluorescence images that look nothing like H&E.

How to choose for your programme

Start with the question, then the stage of the programme. The table below is a starting point, not a rule.

Some practical rules:

• Budget tissue first. Biopsy cores are finite. Decide early which sections go to H&E, single-plex IHC, spatial and sequencing, and whether you can run protein and RNA on the same or adjacent sections.
• Plan the down-selection. If the endpoint is a trial assay, design discovery so that its output is a short, antibody-addressable marker list.
• Pre-specify the analysis. Spatial datasets offer thousands of possible features. Lock the feature definitions and the cut-off before unblinding.
• Check the data plan. Imaging-based transcriptomics and whole-slide high-plex imaging create very large files. Agree storage, compute and the analysis owner before the first run.

For more on where IO biomarker strategy is heading, see our article on immuno-oncology biomarker trends.

The spatial biology vendor landscape

The spatial biology market has consolidated sharply since 2023. The verified changes, listed without any ranking of the platforms:

• Bio-Techne and Lunaphore. Bio-Techne completed its acquisition of Lunaphore, developer of the COMET platform, in July 2023. It has positioned COMET alongside its RNAscope in situ hybridisation assays for combined protein and RNA detection.
• Bruker and NanoString. After NanoString’s Chapter 11 filing, Bruker acquired substantially all of NanoString’s business, including the GeoMx, CosMx, nCounter and AtoMx product lines, in a court-supervised sale that closed on 6 May 2024. The business now sits within Bruker Spatial Biology.
• Vizgen and Ultivue. The two companies announced a merger in October 2024, with the combined company operating as Vizgen.
• Quanterix and Akoya Biosciences. Quanterix, known for ultrasensitive blood-based protein detection, completed its acquisition of Akoya (PhenoCycler and PhenoImager platforms) in July 2025. It describes the combination as spanning blood and tissue biomarkers. Akoya was not acquired by 10x Genomics.
• Standard BioTools. In 2026, Standard BioTools agreed to divest its mass cytometry business, which includes the Hyperion imaging systems used for IMC, as part of its proposed merger with Treeline Biosciences. At the time of writing the sale had not closed.

10x Genomics (Visium, Visium HD, Xenium) remains independent, as do companies such as Ionpath (MIBI) and STOmics (Stereo-seq). The direction is clear: protein and RNA, and tissue and blood, are being combined under fewer corporate roofs. For biomarker teams, that makes it more important to ask about product roadmaps, long-term reagent supply and service continuity before building a trial assay on any platform.

FAQs

What is the difference between spatial proteomics and spatial transcriptomics?

Spatial proteomics maps proteins in tissue with labelled antibodies, from a handful to tens of markers at single-cell resolution. Spatial transcriptomics maps RNA, from targeted panels of hundreds to thousands of genes up to the whole transcriptome. Proteomics reads functional state directly; transcriptomics offers far greater breadth for discovery.

Which spatial method best predicts response to checkpoint inhibitors?

Multiplex IHC/IF has the strongest evidence so far. A 2019 JAMA Oncology meta-analysis found a pooled AUC of 0.79 for mIHC/IF, compared with 0.65 to 0.69 for PD-L1 IHC, gene expression profiling and TMB. The authors noted that larger confirmatory studies are needed, and no spatial assay is yet an FDA-authorised companion diagnostic.

Do spatial biology platforms work on FFPE tissue?

Yes, most major platforms now support FFPE, including multiplex immunofluorescence, cyclic IF, imaging mass cytometry, probe-based Visium and Visium HD, Xenium, CosMx and MERSCOPE. RNA quality in archival blocks varies, so check sample quality requirements before committing a cohort.

Should a clinical trial use spatial proteomics or spatial transcriptomics?

For exploratory and patient-selection endpoints, a locked, low-plex protein panel with a pre-specified image analysis algorithm is usually the more practical choice, because it is closer to standard pathology workflows and has published multi-site reproducibility data. Spatial transcriptomics is usually best kept for discovery and mechanism work on a subset of samples.

What happened to NanoString and Akoya Biosciences?

Bruker acquired NanoString’s business, including GeoMx and CosMx, in a sale that closed in May 2024. Quanterix completed its acquisition of Akoya Biosciences, including PhenoCycler and PhenoImager, in July 2025.

Discuss spatial biomarkers at ImmunoMark San Diego, 10–11 February 2027

Choosing between spatial proteomics and spatial transcriptomics is now a strategic decision that shapes trial design and diagnostic partnering, as well as the science. ImmunoMark Summit San Diego, our immuno-oncology conference in San Diego, brings together 300+ senior pharma, biotech and diagnostics leaders across two parallel streams: Next-Gen Immuno-Oncology and Biomarkers & Companion Diagnostics. Tissue-based and spatial biomarkers fit squarely within the Biomarkers & CDx stream, and pre-arranged 1:1 meetings make it easy to compare notes with the teams solving the same translation problems.

Register for ImmunoMark San Diego 2027

Spatial biology platform companies: meet the translational and biomarker teams choosing platforms for their next programmes. If your company wants to meet this audience, see the San Diego 2027 sponsorship opportunities.