The recent study by Wang and colleagues presents the first comprehensive head-to-head benchmark of three leading commercially available imaging spatial transcriptomics (iST) platforms, 10x Xenium, Vizgen MERSCOPE and NanoString CosMx, applied to a large panel of formalin-fixed paraffin-embedded (FFPE) tissues drawn from tissue microarrays (TMAs) spanning 17 tumour and 16 normal tissue types.
Spatial transcriptomics (ST) enables capturing gene expression within intact tissue architecture, preserving spatial relationships and enabling insight into cell-cell interactions, tissue heterogeneity, and microenvironmental context. Compared to dissociative single-cell RNA-seq methods, ST preserves both local and global spatial relationships.
However, most ST implementations have historically required fresh or specially preserved tissue, limiting use of archival clinical specimens. The ability to apply iST to FFPE is therefore a major step forward for translational research and retrospective studies.
The authors used three TMAs (tumour TMA-1, tumour TMA-2, and a normal-tissue TMA) covering 33 distinct FFPE tissue types. Sequential sections from each core were processed in parallel on Xenium, MERSCOPE, and CosMx, following manufacturers’ recommended protocols. In total, over 5 million cells and nearly 395 million RNA transcripts were profiled.
To enable fair cross-platform comparison, the authors restricted analyses to overlapping genes across panels (typically > 65 shared genes per panel-pair), while also capturing each platform’s characteristic sensitivity and transcript yield.
Downstream analyses included transcript quantification (per gene and per core), pseudo-bulk comparisons, specificity (on-target vs control), false discovery rate (FDR), cell segmentation & cell-type clustering, and concordance with orthogonal single-cell RNA-seq (scRNA-seq) data generated using 10x Chromium on sequential slices.
Key Findings
- Transcript detection sensitivity: Among the three, Xenium consistently produced the highest transcript counts per gene without compromising specificity. CosMx, with its larger 1 K-gene panel, captured the greatest absolute number of genes above background in many tissues; however, variability across tissue types was observed. MERSCOPE generally had lower transcript yield and appeared more sensitive to sample quality and processing variables.
- Specificity and FDR: Xenium showed the highest on-target fraction and the lowest false discovery rate across most tissue types. CosMx generally had higher FDRs, though updated base-calling algorithms in the 2024 datasets improved performance substantially.
- Cell typing and clustering: All three platforms were capable of spatially resolved cell typing. Xenium and CosMx identified slightly more sub-clusters than MERSCOPE, albeit with differing trade-offs in segmentation error frequencies and cluster reliability.
- Concordance with orthogonal scRNA-seq: Data from Xenium and CosMx correlated well with matched single-cell transcriptomic data, underscoring that, despite differences in chemistry and signal amplification, the major cell-type and expression signatures remain consistent.
This benchmarking study fills a gap in the spatial-omics toolkit, offering a practical, comparative evaluation of FFPE-compatible imaging ST methods under real-world conditions. As spatial transcriptomics moves from proof-of-concept to routine use the data and recommendations provided by Wang et al. will be essential in guiding method selection, panel design, and project planning.