Principal workflow of the spatial fly atlas.

a) overview of the workflow: adult flies were sectioned, sections were analyzed with repetitive FISH and data were annotated using cell segmentation, clusterization (i.e. grid) and neighborhood embedding (Methods). b) Example data for adult head sections showing various positions in the brain along the anterior-posterior axis. c) Examples of male whole-body sections. mRNAs from each gene are represented in a different color. The combination of colors reveals the different cell types. Scale bars represent 100 µm.

Adult body cell types.

a) Major cell types of adult males identified by marker genes. Scale bars represent 100 µm. Inset for gut shows zoom-ins of different regions. Inset for flight muscle, shows the percentage of marker gene molecules detected within the outlined area. b) Gene set scores for different main classes of cell types. Quantified using 5 µm x 5 µm grid. Class assignment shown on the right, based on maximum score. Genes used: Neurons: elav, Syt1, Sh, acj6, ey, VAChT, Gad1, VGlut, nAChRalpha7; male repr.: Awh, eyg, svp; epithelia: grh, alphaTry, betaTry, hth; heart: tin, Hand. c) UMAP showing clustering of 5 µm x 5 µm grid spots. d) Spatial location of grid clusters. e) tSNE from male accessory glands from FCA showing expression of marker genes for main gland cells. f) Spatial FISH of the main gland cells marker genes highlights a defined population. Detailed view shown in f’.

Molecular Cartography shows subcellular mRNA localization.

a) Molecular Cartography visualization of marker genes of muscle subtypes. White boxes mark zoom-in regions shown in (b, d, e, and Fig. S6). b) Zoom-in on flight muscle showing colocalization of sls mRNA and DAPI stained nuclei. c) Density plots showing the distance of each mRNA molecule to its nearest nucleus. Red dotted lines mark the peak density, black dotted lines the median distance. d) Zoom-in on flight muscle showing striped patterns of Act88F, Mhc and TpnC4 mRNA localization. e) Zoom-in on anterior flight muscle. Anterior nuclei show increased expression of sls, while patterns of Act88F, Mhc and TpnC4 are maintained. Scale bar in (a) represents 100 µm, scale bars in (b, d and e) represent 10 µm.

Adult head cell types.

a) tSNE showing expression of photoreceptor (ninaC), neuronal (para) and glial (repo) markers (left). Molecular Cartography of the same marker genes (right). b-c) Molecular Cartography of marker genes for olfactory projection neurons (OPN), in an anterior head slice (b) and of perineurial glia of the blood brain barrier (BBB) in a more central slice (c). d-e) Using Molecular Cartography to identify unknown clusters in scRNA-seq data. f) UMAP showing clustering of 5 µm x 5 µm grid spots (top). Spatial location of grid clusters (bottom). g) Differential expression of central brain, optic lobe and photoreceptor regions. h) Molecular Cartography of pros and scro. i) tSNE showing split in optic lobe clusters by expression of Wnt4 and Wnt10. Insert shows Molecular Cartography of Wnt4 and Wnt10, spatially localized in ventral and dorsal regions, respectively. Scale bars represent 100 µm.

Comparison of different techniques for annotating the adult head samples.

a) Overview of different spatial analysis methods which were used to annotate SRT with labels from single-cell RNA-seq: grid-based, neighborhood embedding and nuclei segmentation. Scale bar represents 100 µm. b) Zoom-in on a high-density region with corresponding segmented nuclei. Scale bar represents 10 µm. c-f) Comparison of annotation of spatial data with single-cell RNA-seq for three different quantification methods. Grid-based squares/neighborhoods/nuclei are colored based on matching single-cell clusters for (c) glia, (d) optic lobe, (e) central brain and (f) unknown clusters. In (e) two brain slices are shown at different depths: central (top) and posterior (bottom). NE: neighborhood embedding.

Expression of FISH marker genes in single-cell datasets.

(a) Heatmap showing expression in standardized log(CPM) of 100 marker genes for single-cell clusters in the head dataset. (b) Heatmap showing expression in standardized log(CPM) of 50 marker genes for single-cell clusters in the FCA body dataset.

Head samples.

(a) All detected mRNA molecules color coded for all head samples used in analysis. Scale bars represent 100 µm. (b) Detected expression for each gene in all samples. (c) Heatmap showing Pearson correlation between samples.

Body samples.

(a) All detected mRNA molecules color coded for all body samples used in analysis. Scale bars represent 100 µm. (b) Detected expression for each gene in all samples. (c) Heatmap showing Pearson correlation between samples.

Comparison of body spatial datasets with body single-cell datasets.

(a) Composite heatmap showing gene-gene co-expression based on Pearson correlation. Bottom triangle calculated on spatial datasets (using grid-based 5 µm squares). Top triangle calculated using single-cells. (b) Gene-gene correlation measured across grid-based 5 µm squares and cells.

Comparison of head spatial datasets with brain single-cell datasets.

(a) Overview of colocalization algorithm. For each mRNA spot a disk of 4 µm diameter was used as search space. Overlaps in the disk area were then used to calculate proximity between genes. (b) Heatmap showing co-occurrence of genes with each other. (c) Ward’s clustering of gene-gene proximities. (left) Dendrogram clustering of genes. (right) Spatial location of gene clusters.

Subcellular mRNA localization in flight, leg and head muscles.

(a) Molecular Cartography of sparse genes CG32121 and Salm in flight muscles. (b) Molecular Cartography of genes following nuclei-enriched and complementary striped bands. (c) Molecular Cartography of sls and Mhc in head muscles, showing nuclei enrichment for sls. Scale bars represent 10 µm.

Comparison of head spatial datasets with brain single-cell datasets.

(a) Gene-gene correlation measured across grid-based 5 µm squares and cells. Mismatches shown in red, matching co-expression in green. (b) Composite heatmap showing gene-gene co-expression based on Pearson correlation. Bottom triangle calculated on spatial datasets (using grid-based 5 µm squares). Top triangle calculated using single-cell data. (c) Molecular Cartography showing high co-expression of pros and dati. Scale bar represents 100 µm. (d) Molecular Cartography showing co-expression of Indy and disco. Scale bar represents 100 µm. (e) Molecular Cartography showing expression of Vmat and DAT. Zoom shows non overlapping expression and expression of Vmat outside of nuclei marked by white DAPI signal. Scale bar represents 100 µm in the main image, 50 µm in zoom. (f) Molecular Cartography showing expression of neuropeptides Ilp2 and Pdf. Zooms show expression outside of nuclei marked by white DAPI signal. Scale bar represents 100 µm, 50 µm in zoom. (g) Stacked barplot showing sample composition based on Leiden 1 clustering.

Results from Tangram on grid-based quantification.

a-d) Comparison of annotation of spatial data with single-cell RNA-seq. Grid-based squares are colored based on mapped single-cell cluster labels for (a) glia, (b) optic lobe, (c) central brain and (d) unknown clusters using Tangram. In (c) two brain slices are shown at different depths: central (top) and posterior (bottom). e) Effect of manually adjusted thresholds on Tangram score for nuclei segmentation (left) and grid-based quantification (right) for central brain clusters: 98%-quantile for assigning Pdf and 99%-quantile for IPC labels. f) Molecular Cartography showing expression of mirr. g) Molecular Cartography showing expression of caup. Scale bars represent 100 µm.