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

Our plain-English summary of Comparison of imaging-based single-cell resolution spatial transcriptomics profiling platforms using formalin-fixed, paraffin-embedded tumor samples, by N.O. Lermi, M.M. Ayala, S.H. Ruiz, et al (Research Square, 2025).

The study

• Authors: N.O. Lermi, M.M. Ayala, S.H. Ruiz, W. Lu, K. Khan, A. Serrano, I. Lubo, L. Hamana, K. Tomczak, S. Barnes, J. Dou, Q. Liang, T.E.A.M. Rti, M.G. Raso, X. Tang, M. Jiang, B. Sanchez-Espiridion, A. Weissferdt, J. Heymach, J. Zhang, B. Sepesi, T. Cascone, A. Tsao, M. Altan, R. Mehran, D. Gibbons, I. Wistuba, C. Haymaker, K. Chen, L.M. Solis Soto
• Journal: Research Square
• Published: 2025
• Read it: PubMed · Publisher · Free full text

Abstract

Imaging-based spatial transcriptomics (ST) is evolving rapidly as a pivotal technology in studying the biology of tumors and their associated microenvironments. However, the strengths of the commercially available ST platforms in studying spatial biology have not been systematically evaluated using rigorously controlled experiments. In this study, we used serial 5-μm sections of formalin-fixed, paraffin-embedded surgically resected lung adenocarcinoma and pleural mesothelioma tumor samples in tissue microarrays to compare the performance of the single cell ST platforms CosMx, MERFISH, and Xenium (uni/multi-modal) platforms in reference to bulk RNA sequencing, multiplex immunofluorescence, GeoMx Digital Spatial Profiler, and hematoxylin and eosin staining data for the same samples. In addition to objective assessment of automatic cell segmentation and phenotyping, we performed pixel-resolution manual evaluation of phenotyping to carry out pathologically meaningful comparison between ST platforms. Our study detailed the intricate differences between the ST platforms, revealed the importance of parameters such as tissue age and probe design in determining the data quality, and suggested reliable workflows for accurate spatial profiling and molecular discovery.

In plain English

This preprint compared three commercial imaging-based spatial transcriptomics platforms (CosMx, MERFISH, Xenium), which measure gene activity in single cells on tissue slides. Xenium was tested with two cell-outlining methods, unimodal and multimodal. The authors used archived tumor tissue from MD Anderson Cancer Center from 22 patients with lung adenocarcinoma (2016 to 2018) and 22 with pleural mesothelioma (2020 to 2022). The study did not examine asbestos or any other exposure. CosMx and MERFISH did better on newer samples, while Xenium was uniform across tissue ages. CosMx detected the most transcripts per cell but had high background: in one mesothelioma set, 31.9% of target gene probes overlapped negative controls. CosMx and MERFISH had higher estimated false discovery rates than Xenium. Three pathologists scored lung adenocarcinoma results against stained sections. Segmentation scores were similar across platforms, but for cell-type annotation Xenium multimodal scored best and MERFISH lowest. Xenium unimodal was excluded, and only Xenium multimodal had H&E staining of its exact section. The authors concluded that tissue age and probe design affected data quality. They noted the samples came from one institution and few tumor types.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: Laboratory methods comparison of three imaging-based single-cell spatial transcriptomics platforms (CosMx, MERFISH, Xenium unimodal and multimodal) on serial sections of archived tumor tissue, checked against bulk RNA sequencing, GeoMx DSP whole-transcriptome data, multiplex immunofluorescence, H&E staining and pathologists' scoring
• Who was studied: Formalin-fixed, paraffin-embedded surgically resected tumor samples from MD Anderson Cancer Center, arranged in four tissue microarrays: 22 patients with lung adenocarcinoma (2 arrays, 53 cores, ICON1 and ICON2, collected 2016 to 2018) and 22 patients with pleural mesothelioma (2 arrays, 49 cores, MESO1 and MESO2, collected 2020 to 2022)
• Exposure: No asbestos or other environmental or occupational exposure was studied. What was compared was the laboratory platform and gene panel: CosMx 1,000-plex, MERFISH 500-plex immuno-oncology panel, and Xenium 289-plex human lung panel plus 50 custom genes, with 93 genes shared by all three
• Main results:
  • CosMx detected the highest transcript counts and uniquely expressed gene counts per cell among all tissue microarrays
  • CosMx and MERFISH performed better with the newer tissue microarrays, whereas Xenium performed uniformly across different tissue ages
  • With CosMx, target gene probes overlapped the negative control probes: ICON1 n = 8 (0.8%), ICON2 n = 27 (2.7%), MESO1 n = 196 (19.6%)
  • In the MESO2 microarray, CosMx had 319 (31.9%) target gene probes overlapping negative controls, while Xenium-MM had 2 (0.6%); Xenium-UM had none
  • Estimated false discovery rates were above 10% (ICON) and above 5% (MESO) for CosMx, and above 4% (ICON and MESO) for MERFISH
  • CosMx and MERFISH had higher false discovery rates than Xenium (both unimodal and multimodal)
  • After quality filtering, MERFISH retained a smaller fraction of cells (ICON1 22.6%, ICON2 40.0%, MESO2 64.4%) than the other platforms
  • Correlation with bulk RNA sequencing in the ICON2 lung adenocarcinoma microarray was R = 0.43 for CosMx and R = 0.39 for MERFISH, lower than for Xenium-MM
  • Correlation with bulk RNA sequencing in the newest microarray (MESO2, pleural mesothelioma) was R = 0.58 for CosMx and R = 0.64 for MERFISH
  • CosMx correlated best with GeoMx DSP whole-transcriptome data (ICON2 R = 0.84; MESO2 R = 0.85)
  • In CosMx lung adenocarcinoma data, signals for the T-cell marker CD3E and the B-cell marker CD19 came from every cell type, so these cell types could not be told apart
  • In the pathologists' evaluation of the ICON2 microarray, F1-scores for cell segmentation were similar among platforms
  • Xenium-UM was excluded from the pathologists' evaluation because it lacked cell membrane detection for cell segmentation
  • In the pathologists' evaluation of the ICON2 microarray, Xenium-MM had better F1-scores (median above 75%) than the other platforms for all cell types
  • MERFISH had the lowest F1-scores for every phenotype interrogated in the pathologists' evaluation
  • For the pleural mesothelioma samples, labeling cell types using matched single-cell RNA sequencing data increased the accuracy of cell-type annotations
• The authors' conclusion: The authors concluded that the study detailed differences between the platforms and showed that parameters such as tissue age and probe design affect data quality. They recommended visual inspection of simplified cell-type annotations against histology.
• Limitations the authors state:
  • Samples came from a single institution. The study lacked a broader range of tumor types, whole tissue sections, and samples with special processing such as decalcification.
  • CosMx required selecting fields of view, so whole tissue cores could not be analyzed with CosMx
  • Not every tissue microarray could be run on every platform (ICON1 was unavailable in Xenium and MESO1 in MERFISH)
  • MERFISH could not be included in the negative-control comparison because it lacked negative control probes
  • Only Xenium-MM had H&E staining of its exact section available for the pathologists' evaluation

Topics

Medical terms in this study, explained in our Lexicon: Lung Adenocarcinoma, Pleural Mesothelioma, Discovery.

Related studies

• Comparison of imaging based single-cell resolution spatial transcriptomics profiling platforms using formalin-fixed paraffin-embedded tumor samples — N. Ozirmak Lermi, M. Molina Ayala, et al. (2025)
• Numerical and structural genomic aberrations are reliably detectable in tissue microarrays of formalin-fixed paraffin-embedded tumor samples by fluorescence in-situ hybridization — H. Horn, J. Bausinger, et al. (2014)
• Differential expression of N-cadherin in pleural mesotheliomas and E-cadherin in lung adenocarcinomas in formalin-fixed, paraffin-embedded tissues — A.C. Han, A. Peralta-Soler, et al. (1997)
• Validation of a Gene Expression Test for Mesothelioma Prognosis in Formalin-Fixed Paraffin-Embedded Tissues — A. De Rienzo, R.W. Cook, et al. (2017)
• Discrepancies of asbestos body and fiber content between formalin-fixed and corresponding paraffin embedded lung tissue — B.K. Kuhn, F. Barresi, et al. (2021)
• Comparative analysis of asbestos body and fiber content in formalin-fixed vs. paraffin-embedded lung tissue — B.K. Kuhn, B. Vrugt, et al. (2025)

Browse return to medical research.