Subcellular spatially resolved gene neighborhood networks in single cells

Introduction:: Image-based spatial omics methods such as fluorescent in situ hybridization (FISH) generate molecular profiles of single cells at single molecule resolution. Current spatial transcriptomics analysis methods focus on the distribution of single genes. However, the spatial proximity of RNA transcripts can play an important role in cellular function.

In cells, complex regulatory relationships between genes maintain gene-expression phenotypes.¹ Gene regulatory networks (GRNs) have been widely used to provide intuitive visualization or gene regulation.

We demonstrate a spatially resolved gene neighborhood network (spaGNN) pipeline for the analysis of subcellular gene proximity relationships². Machine learning-based clustering of subcellular FISH data yields subcellular density classes. The nearest-neighbor analysis produces heterogeneous gene proximity maps in distinct subcellular regions. (Figure 1A) We then illustrate the cell type distinguishing capability of spaGNN using multiplexed error-robust FISH (MERFISH) data of fibroblast³ and U2-OS cells⁴ and sequential FISH (seqFISH) data of human bone marrow-derived (HBM) and human umbilical cord-derived (HUC) mesenchymal stem cells (MSCs).


Materials and Methods:: HBMs and HUCs were cultured on collagen-coated glass coverslips for 24 hours before fixing cells with 4% paraformaldehyde for 10 minutes, and permeabilized in 70% ethanol at -20°C overnight. The mRNA targets were then labeled with the hybridization chain reaction (HCR) protocol and imaged through a fluorescent microscope.⁵ Images were analyzed to extract the location of each mRNA molecule. The fluorescent labels were then removed by recombinant DNase I.⁶ The next set of mRNA targets were then labeled with the same HCR protocol. Segmentation immunofluorescent (IF) staining was conducted after imaging all mRNA targets.


Results, Conclusions, and Discussions:: By applying the clustering algorithm to the positions of all detected transcripts, the algorithm separates transcripts into local communities based on RNA density. Pearson’s correlation between markers among patches indicates the tendency for genes to colocalize in the same patch. The correlation value, denoted as R, is defined as the patch correlation of gene pairs. A higher R-value indicates genes localizing in similar subcellular compartments. (Figure 1B)

Localized spatial gene neighboring relationships were then analyzed through local neighborhood analysis. Inside each patch, the center transcript and its surrounding 9 neighbors form a 10-transcript neighborhood. Each dot is a detected transcript, and an edge indicates a neighboring event between two genes. 10-gene neighborhoods were computed for all transcripts in the same patch, and the number of neighboring events was counted. We used a permutation analysis with 1000 permutations to find the distribution of pairwise gene neighboring events given the gene count. We defined the z-score of the detected number of interaction events given the distribution as the proximity score between the pair of genes, denoted as Z. A higher Z-value indicates genes are more often immediate neighbors of each other. (Figure 1C)

To further utilize the spatially resolved gene proximity relationships, we performed clustering analysis based on three features, single-cell RAN count, patch correlations, and pair-wise gene proximity score variability. We first clustered fibroblast and U2-OS cells and found all three clustering analyses achieved high cell-type prediction accuracy. (Figure 1D) We then applied the SpaGNN pipeline to seqFISH data in mesenchymal stem cells (MSCs) and primary human chondrocytes (HCH). The gene neighborhood network variability achieved the highest cell type prediction accuracy. (Figure 1E)

In conclusion, The subcellular organization of mRNA and protein molecules can be studied using multiplexed RNA-FISH and IF staining. Machine learning methods reveal differences at both subcellular patch level and local RNA neighborhood level spatial heterogeneity. Spatially resolved features can improve cell classification performance. Such subcellular spatial profile can be applied to understand MSC heterogeneity further, and potentially benefit MSC-based therapy design.


Acknowledgements (Optional): :

References (Optional): : 1. Kitano H. Systems Biology: A Brief Overview. Science. 2002;295(5560):1662-1664. doi:10.1126/science.1069492

2. Fang Z, Ford AJ, Hu T, Zhang N, Mantalaris A, Coskun AF. Subcellular Spatially Resolved Gene Neighborhood Networks in Single Cells. in press at Cell Reports Methods. doi:10.1101/2022.08.03.502409

3. Chen KH, Boettiger AN, Moffitt JR, Wang S, Zhuang X. Spatially resolved, highly multiplexed RNA profiling in single cells. Science. 2015;348(6233):aaa6090. doi:10.1126/science.aaa6090

4. Mah CK, Ahmed N, Lam D, et al. Bento: A Toolkit for Subcellular Analysis of Spatial Transcriptomics Data. Bioinformatics; 2022. doi:10.1101/2022.06.10.495510

5. Choi HMT, Schwarzkopf M, Fornace ME, et al. Third-generation in situ hybridization chain reaction: multiplexed, quantitative, sensitive, versatile, robust. Development. 2018;145(12):dev165753. doi:10.1242/dev.165753

6. Coskun AF, Cai L. Dense transcript profiling in single cells by image correlation decoding. Nat Methods. 2016;13(8):657-660. doi:10.1038/nmeth.3895