Comparison of imaging based single-cell resolution spatial transcriptomics profiling platforms using formalin-fixed paraffin-embedded tumor samples

Abstract

Imaging-based spatial transcriptomics (ST) is evolving as a pivotal technology in studying tumor biology and associated microenvironments. However, the strengths of the commercially available ST platforms in studying spatial biology have not been systematically evaluated using rigorously controlled experiments. We use serial 5 μm sections of formalin-fixed, paraffin-embedded surgically resected lung adenocarcinoma and pleural mesothelioma samples in tissue microarrays to compare the performance of the ST platforms (CosMx, MERFISH, and Xenium (uni/multi-modal)) in reference to bulk RNA sequencing, multiplex immunofluorescence, GeoMx, and hematoxylin and eosin staining data. In addition to an objective assessment of automatic cell segmentation and phenotyping, we perform a manual phenotyping evaluation to assess pathologically meaningful comparisons between ST platforms. Here, we show the intricate differences between the ST platforms, reveal the importance of parameters such as probe design in determining the data quality, and suggest reliable workflows for accurate spatial profiling and molecular discovery.

Fig. 1. Experimental design, panel comparison, and nuclear staining with the ST platforms.

a TMAs containing tissue samples from patients with pleural mesothelioma (MESO1, MESO2; n = 22) or non-small cell lung cancer (NSCLC; ICON1, ICON2; n = 22). Tumor samples were sectioned at a thickness of 5 μm and submitted to CosMx, MERFISH, Xenium-UM, and Xenium-MM assays based on tissue availability. Data analyses were performed using Seurat pipeline and evaluation of pathologists with guidance of mIF and H&E staining. b Venn diagram displays shared genes between panels of the ST platforms. c DAPI staining images of ICON2 and MESO2 TMAs and CosMx whole images of the ICON2 and MESO2 TMAs with staining with morphology markers before Field of View (FOV) selection. Red/white squares display selected FOVs of CosMx. The CosMx FOV, MERFISH, Xenium-UM, and Xenium-MM images display nuclear staining of selected FOV regions and whole tissue cores. Only TMAs with data available from all ST platforms were selected for comparisons. Created in BioRender. Ozirmak lermi, N. (2025) https://BioRender.com/y68w242.

Fig. 2. Technical comparison of the ST platforms.

a, b Box plots of the transcript count and uniquely expressed genes per cell that are captured on each ST platform for CosMx (blue) ICON1 (n = 36512), CosMx ICON2 (n = 38395), CosMx MESO1 (n = 58996), CosMx MESO2 (n = 45898), MERFISH (purple) ICON1 (n = 31431), MERFISH ICON2 (n = 94539), MERFISH MESO2 (n = 98149), Xenium-UM (yellow) ICON2 (n = 133019), Xenium-UM MESO1 (n = 142721), Xenium-UM MESO2 (n = 85196), Xenium-MM (orange) ICON2 (n = 113387), Xenium-MM MESO1 (n = 128042), Xenium-MM MESO2 (n = 70472). Data were normalized by panel size. Each dot represents a cell. The center of the box plot is denoted by the median, a horizontal line dividing the box into two equal halves. The bounds of the box are defined by the lower quartile (25th percentile) and the upper quartile (75th percentile). The whiskers extend from the box and represent the data points that fall within 1.5 times the interquartile range (IQR) from the lower and upper quartiles. Any data point outside this range is considered an outlier and plotted individually. The red diamonds correspond to the average transcript counts and uniquely expressed genes in the blocks. Significance assessed using Mann–Whitney–Wilcoxon two-sided test (p < 2.2e−16, p = 0.00039). c Dot plot of the total transcript counts per probe across the ST platforms. Each dot represents a gene probe (red) or a negative control probe (green). d Bar graph of False Discovery Rates (FDR) calculated using negative control or blank probes and total read counts. The y-axis represents FDR as a percentage. The gray bars represent unavailable data. The blue and green bars represent FDR calculated using blank and negative control probes, respectively. Source data are provided as a Source Data file.

Fig. 3. Comparative analysis of cell segmentation across the ST platforms.

a Zoomed images of nuclear staining (green) overlaid with cell boundaries (white and red) obtained on all platforms using a tissue core of ICON2 TMA (CosMx (n = 14), MERFISH (n = 21), Xenium-UM (n = 26), Xenium-MM (n = 22). Cells with red boundaries represent filtered cells after quality controls. b Bubble plot of the mean cell area sizes (μm²) across the tissue blocks in CosMx (blue), MERFISH (purple), Xenium-UM (yellow) and Xenium-MM (orange) assays. The bubble size corresponds to the difference between the smallest and largest cell areas within each block. c Percentages of the remaining cells after filtering. The dashed lines indicate transcript count thresholds for: Xenium (yellow and orange) (10), MERFISH (purple) (10), and CosMx (blue) (30). Source data are provided as a Source Data file.

Fig. 4. Concordance of RNA levels between ST platforms and bulk RNA-seq data.

Scatter plots of the average expression of overlapping genes in ST platform and bulk RNA-seq data of patients of the ICON2 (a) and MESO2 (b) TMAs. The red lines represent linear regression of mean log-normalized expression per cell type ± 95% confidence interval. Pearson correlation coefficients (R) are provided in the top left corner of each plot. Outlier genes are annotated. Source data are provided as a Source Data file.

Fig. 5. Concordance of ST platforms with DSP WTA.

Scatter plots display correlation of the ST platforms and DSP WTA based on shared genes in ICON2 (a) and MESO2 (b) TMAs. The blue lines represent linear regression of mean log-normalized expression per cell type ± 95% confidence interval. The Pearson correlation coefficient (R) is provided in the top left corner of each plot. Source data are provided as a Source Data file.

Fig. 6. Comparison of gene count detection among the different ST platforms.

Scatter plots display correlation of different ST platforms based on shared genes in ICON2 (a) and MESO2 (b) TMAs. The green lines represent linear regression of mean log-normalized expression per cell type ± 95% confidence interval. The Pearson correlation coefficient (R) is provided in the top left corner of each plot. Source data are provided as a Source Data file.

Fig. 7. Cell type annotation performance of the ST platforms using the ICON2 TMA.

a UMAP and spatial locations of manually annotated cell type clusters across the platforms. Clusters are labeled with their corresponding cell types based on top-expressed genes. Each color represents a cell type. b Box plots of the F1-scores display performance of cell segmentation and cell type annotations in ST platforms. Each dot represents number of tissue cores in CosMx (blue) ICON2 (n = 14), MERFISH (purple) ICON2 (n = 21), Xenium-MM (orange) ICON2 (n = 21). The center of the box plot is denoted by the median, a horizontal line dividing the box into two equal halves. The bounds of the box are defined by the lower quartile (25th percentile) and the upper quartile (75th percentile). The whiskers extend from the box and represent the data points that fall within 1.5 times the interquartile range (IQR) from the lower and upper quartiles. Any data point outside this range is considered an outlier and plotted individually. Source data are provided as a Source Data file.

Fig. 8. Cell type annotation performance of the ST platforms in the MESO2 TMA.

UMAP and spatial locations of annotated cell types across the platforms. Each color represents a cell type. The cell type labels were transferred from reference scRNA-seq data of pleural mesothelioma cohorts using Seurat in R.