Abstract Multiplexed spatial profiling can enable biological insights by characterizing gene expression within discrete physical locations of a tissue. However, these advanced techniques can be confounded by variability of sample collection, storage, or profiling protocols, making it difficult to accurately compare data generated by different laboratories. Further, as more spatial profiling datasets become publicly available, having methods to enable meta-analysis of samples collected on different studies will maximize the learnings to support the advancement of treatments or identifying patient segments. To quantify the variability of spatial profiling data at different laboratories and to advance data normalization methodologies, 4 independent laboratories used the NanoString® GeoMx® Digital Spatial Profiling (DSP) to profile serial 5 µm sections of tissue and cell pellet arrays (CPAs). GeoMx DSP enables high throughput, spatially resolved analysis of gene or protein expression from fresh or archival human tissues. In this study, the GeoMx Cancer Transcriptome Atlas was used to profile >1800 genes simultaneously. We examined the concordance of GeoMx data generated in the different laboratories when controlling for methodical variation (e.g., reagents, tissue source) and experimentally varying region of interest (ROI) size, collection site, and sample preservation methods. Sections of tonsil, colon, and 2 CPAs were profiled separately at the 4 laboratories. Each analyzed fresh cut (FC) tissues and two sites examined sample stability by analyzing the impact of storing slides at -80°C for 1 month prior to spatial profiling. Concordance analysis was performed using the Horn-Morisita Index on raw data comparing paired and unpaired ROIs across each set of slides. In CPA samples where each pellet was a different tumor type (e.g., NSCLC, melanoma), we observed strong clustering by cell line. While data initially showed varying degrees of clustering by slide, factoring out this variable removed the association of slide, allowing integration of the data across profiling locations without affecting concordance within slides. In tonsil, ROIs with increasing area were profiled. Comparing expression between pairs of samples for a given area, concordance increased with ROI size (R = -0.40, p<6e-06). Finally, we observed little impact of preservation method (FC vs -80°C) in these data. In this study, we quantify slide-specific variation observed in high-plex RNA profiling by the DSP platform and detail methods for accounting for this variation. We note that many downstream analyses (e.g., differential expression) already model slide effects during the analysis, but modeling it explicitly allows for direct comparison of concordance with other approaches (e.g., clustering, PCA). These methods support the use of multi-institution studies leveraging the GeoMx platform. Citation Format: Tyler Hether, Tim Howes, David Scoville, Charlie Glaser, Yanyun Li, Rami Vanguri, Neeman Mohibullah, Wan-Jung Chang, Todd Yoder, Minnal Gupta, Kathy Ton, Yan Liang, Ying Huang, Zach Herbert, Jason Reeves, Elizabeth Mittendorf, Simon Lacey, Travis Hollmann, Sarah Warren, Theresa LaVallee. A multi-institution examination of concordance in spatial transcriptomics using the GeoMx Cancer Transcriptome Atlas [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 66.
Read full abstract