A cell’s function is influenced by the environment, or niche, in which it resides. Studies of niches usually require assumptions about the cell types present, which impedes the discovery of new cell types or interactions. Here we describe ProximID, an approach for building a cellular network based on physical cell interaction and single-cell mRNA sequencing, and show that it can be used to discover new preferential cellular interactions without prior knowledge of component cell types. ProximID found specific interactions between megakaryocytes and mature neutrophils and between plasma cells and myeloblasts and/or promyelocytes (precursors of neutrophils) in mouse bone marrow, and it identified a Tac1+ enteroendocrine cell–Lgr5+ stem cell interaction in small intestine crypts. This strategy can be used to discover new niches or preferential interactions in a variety of organs.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank J. Korving for help with the microdissection microscope; the animal facility; A. de Graaf and the microscope facility; the sequencing facility; K. Wiebrands for suggesting a name for the method and assistance in intestinal crypt dissociation; N. Battich and B. De Barbanson for help with machine learning; L. Kester for help with the Monte Carlo simulations; and R. van der Linden for help with the sorting experiments. This work was supported by the European Research Council (advanced grant ERC-AdG 742225-IntScOmics to A.v.O.) and a Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO) OPEN award (NWO-ALWOP189 to A.v.O.). This work is part of the Oncode Institute, which is partly financed by the Dutch Cancer Society. In addition, we thank the Hubrecht Sorting Facility and the Utrecht Sequencing Facility, subsidized by the University Medical Center Utrecht, Hubrecht Institute, and Utrecht University.
Integrated supplementary information
(a,b) t-SNE map of transcriptome similarities with experimental replicates (a), and sub-dissected units (b) represented as color code. (c) Cumulative percentage of the fraction of different clusters present in the different units.
(a-i) t-SNE map of transcriptome similarities with color-coded representation of transcript count for the sum of ribosomal genes (a), sum of S100a8 and S100a9 (b), Beta-s (c), sum of Ear1 and Ear2 (d), Ppbp (e), Igj (f), Hmox1 (g), sum of C1qa and C1qb (h), and Mcpt8 (i).
(a-f) t-SNE map of transcriptome similarities with color-coded representation of transcript count for Elane (a), Ltf (b), Ngp (c), Fpr1 (d), Slc11a1 (e), and Retnlg (f).
(a-c) t-SNE map of transcriptome similarities. (a) Colors represent the different RaceID2 clusters. The putative cell types for the different clusters are indicated. (b) Color representation of the origins of the cells. (c) Color indication of a significantly (p < 0.00001) higher frequency of picked cells (red) or sorted cells (blue). Non-significant differences are represented in grey.
(a) t-SNE map of transcriptome similarities with color-coded representation of raw transcript counts (UMIs) per cells. (b) Bar plot representing the average total number of transcripts per cell according to their RaceID2 clusters. The blue line represents the mean transcript count across all filtered cells, and the red line represents the filter threshold value. (c) Line chart showing the frequency of detected interactions: macrophages – erythroblasts (green), megakaryocytes – neutrophils (cyan), plasma cells – myeloblasts (magenta), and all 3 previously mentioned interactions in a single analysis (red), for 100 different downsampling, in function of the filtering threshold based on total transcript count per cells. (d) Line chart showing the frequency of detection of the megakaryocytes – neutrophils, plasma cells – promyelocytes/myeloblasts and macrophages – erythroblasts interaction as function of the number of sampled structures. The line and the shaded area are the running average and standard deviation respectively of 10 consecutive data points. Each data point is the frequency based on 20 different samplings.
(a-f) t-SNE map of transcriptome similarities with color-coded representation of RaceID2 clusters (a), embryonic day (b), and expression of Beta-s (c) and Hmox1 (d). (e) Representation of enriched (red) and depleted (blue) interactions per clusters. Nodes represent cluster centers. Edges are interactions inter-clusters, and filled circles are interactions intra-clusters. (f) t-SNE map of transcriptome similarities with magenta edges representing all cell-cell interactions stemming from cluster 6 (macrophages, in cyan).
(a,b) t-SNE map of transcriptome similarities, with node colors representing the RaceID2 clusters (a) or strain of origin (b) (n=2 independent experiments). The edges connecting cells represent the physical interaction in between cells (a), with color-coding of inter- versus intra-strain interaction (b). 62% of the interactions are intra-strain. The putative cell types for the different clusters are indicated in (a). (c) Probability distribution of the frequency of inter-strain interactions after label permutation simulations (grey area, and blue bar representing the mean of the distribution, 10,000 simulations), compared to the actual number (red bar).
(a,b) Raw single-plane images of BM cryosections corresponding to Figure 3a and 4a, stained with smFISH probes against Elane (green) and Igj (red) mRNAs, conjugated with Cy5 and TAMRA respectively (a) (n=1), or biotin anti-CD41 antibodies and alexa 488-streptavidin conjugates (grey), and smFISH probes against Retnlg (green) and Ngp (red) mRNAs, conjugated with Cy5 and TAMRA respectively (b) (n=1). Nucleuses were stained with DAPI (blue). The asterisk (*) denotes autofluorescent cells not used in the analysis, and the pound sign (#) denotes red blood cells. Scale bar corresponds to 10 μm.
Supplementary Figures 1–8
Code and demo data for ProximID, ProximID with different cell-type-recognition methods, and image analysis of interacting cells
Doublet microdissection. The video shows a representative example of the separation of two interacting cells from the bone marrow using two glass needles, and their subsequent mouth-pipetting