Comprehensive Identification and Spatial Mapping of Habenular Neuronal Types Using Single-Cell RNA-Seq

Abstract

The identification of cell types and marker genes is critical for dissecting neural development and function, but the size and complexity of the brain has hindered the comprehensive discovery of cell types. We combined single-cell RNA-seq (scRNA-seq) with anatomical brain registration to create a comprehensive map of the zebrafish habenula, a conserved forebrain hub involved in pain processing and learning. Single-cell transcriptomes of ∼13,000 habenular cells with 4× cellular coverage identified 18 neuronal types and dozens of marker genes. Registration of marker genes onto a reference atlas created a resource for anatomical and functional studies and enabled the mapping of active neurons onto neuronal types following aversive stimuli. Strikingly, despite brain growth and functional maturation, cell types were retained between the larval and adult habenula. This study provides a gene expression atlas to dissect habenular development and function and offers a general framework for the comprehensive characterization of other brain regions.

Keywords: adult; habenula; in situ atlas; larva; neuronal types; single-cell RNA-seq; transcriptomes; zebrafish.

Figure 1. Unbiased Clustering of scRNA-seq Data Identifies 15 Molecular Distinct Neuronal Clusters in the Larval Habenula

A. Schematic of the zebrafish habenula showing the anatomical subdivisions corresponding to the dorso-medial (orange), dorso-lateral (red) and ventral (blue) regions. These subdivisions are known to have distinct gene expression patterns and functionality. B. Overview of the experimental strategy. Transgenic heads with gng8-GFP positive cells were dissected, pooled and dissociated, followed by enrichment of GFP⁺ habenular cells using fluorescent activated cell sorting (FACS). Single cell libraries were prepared using droplet-based droplet and plate-based Smart-seq2. Raw reads were processed to obtain a gene expression matrix (genes x cells). PCA and graph clustering was used to divide cells into clusters and identify cluster specific markers. Validation and spatial localization was performed using fluorescent RNA in situ hybridization (FISH) of statistically significant cluster-specific markers (see STAR Methods). C. 2D visualization of single cell clusters using t-distributed Stochastic Neighbor Embedding (tSNE). Individual points correspond to single cells and are color-coded according to their cluster membership determined by graph-based clustering. The tSNE mapping was only used for post hoc visualization of the clustering but not to define the clusters themselves. D. Gene Expression profiles (columns) of select cluster-specific markers identified through differential expression analysis (DEA) of previously known (labeled with an asterisk (*)) and new habenular types (rows). Bar on the right displays the percent of total dataset represented in every cluster, showing the abundance of each cell type found by clustering analysis. E. A dendrogram representing global inter-cluster transcriptional relationships. The dendrogram was built by performing hierarchical clustering (correlation distance, average linkage) on the average gene-expression profiles for each cluster restricting to the highly variable genes in the dataset. See also Figure S1, Table S1

Figure 2. Validation and Spatial Distribution of Previously Described Neuronal Types along with Identified Novel Markers

A. Expression profiles of known and novel habenular marker genes that are specific or enriched in the five clusters displaying previously described gene expression signatures. Green bar on top represents new markers and orange bar represents known markers. B–G. In vivo expression patterns of known and novel marker genes that are enriched in clusters harboring previously characterized habenular genes (Hb01, Hb02, Hb06, Hb09, and Hb15). Each type was characterized by both previously described markers, and new markers found from single cell analysis. RNA-FISH (green) was performed with a total-Erk (pale gray) co-stain for registration (see Figure 3). In some cases, a non-linear filter (gamma = 0.3) was applied to the total-Erk (gray) channel to aid visualization of the in situ signal (green). B–D. FISH labeling of B) Previously known marker (tac3a) and new markers for C) Hb01 (murcb, tacr3l) and D) Hb02 (adrb2a) found by single cell analysis. Insets show regionalized expression of the gene without total-Erk. murcb/tacr3l⁺(Hb01) and adrb2a⁺(Hb02) domains form subdivisions within the tac3a⁺ domains. D. FISH labeling of new markers pou3f1 and pnoca enriched in the lrrtm1⁺ and foxa1⁺ cluster Hb06. E. FISH labeling of new marker igf2a enriched in adcyap1a⁺ left-only cluster Hb09. F. FISH labeling of new marker wnt11r specific to the aoc1⁺ ventral habenular cluster Hb15. Scale bars indicate 50 μm. See also Figures S2 and S3, Table S1, Movie S1

Figure 3. Validation and Spatial Distribution of 10 Novel Habenular Neuronal Types

A–J. RNA-FISH (green) was performed for specific markers for novel clusters A) Left-enriched clusters: Hb07 (pcdh7b), Hb08 (wnt7aa), Hb10 (ppp1r1c); B) Posterior habenular clusters: Hb04 (cbln2b), Hb11 (cpne4a), Hb12/11 (pyya), C) Non-regionalized or rare neuronal types: Hb03 (spx), Hb05 (c1ql4b), Hb14 (slc32a1), and Hb13 (tubb5), each overlaid with a total-Erk co-stain (pale gray) for registration. In each case, representative habenular slices with expression are shown. Full stacks are available through a linked website [See Data Availability Section]. D. Slices through the registered reference habenula simultaneously showing six marker genes that are expressed in a regionalized pattern: wnt7aa (La_Hb08), adcyap1a (La_Hb07), cbln2b (La_Hb04), murcb (La_Hb01), lrrtm1 (La_Hb06), gpr139 (La_Hb15). E. Schematic of representative transverse slices through the habenula displaying rough spatial co-ordinates of previously described as well as new neuronal types found by single-cell analysis. Cells are color-coded based on their identity in the t-SNE plot (see Figure 1C). Depth is indicated by the z slice in microns. The sectioning extends from z = μm (Dorsal) and z = 75μm (Ventral). Only regionalized markers are represented. Schematic is a simplified representation of an accompanying stack of registered habenular markers overlaid onto one another [see Movie S1]. Scale bars indicate 50 μm. See also Figure S3, Table S1, Movie S1

Figure 4. Correspondence of Larval Habenular Neuronal Types and their Molecular Identities Between the Droplet and SMART-seq2 Datasets

A. t-SNE visualization of single cell clusters obtained by clustering of the SMART-seq2 (SS2) data. B. Dot plot (confusion matrix) showing the proportion of cells in each SS2 cluster (rows) that were classified to droplet clusters (columns) using a multiclass random forest classifier(RF). A cell was assigned to a droplet cluster label if > 15 % of the decision trees in the RF classifier contributed to the majority vote (given that there are 16 classes, 6.25% vote would constitute a majority). * represents SS2 clusters in which greater than 70% of the cells of the cluster maps to single droplet clusters. C. Same as A, but where each cell is annotated according to its RF assigned droplet cluster label. Rough demarcations of the SS2 clusters as in A are sketched. D. Top 10 differentially expressed genes in each habenular type computed using a post hoc test on the SS2 data based on the RF-assigned cluster label as in C. Highlighted on the right are anecdotal examples of genes that were not detected among the top 15 differentially expressed genes in the corresponding droplet clusters. See also Figure S4

Figure 5. Comparative Analysis of Habenular Neuronal Types between Larval and Adult Stages

A. t-SNE visualization of adult single cell clusters obtained by clustering of the adult dataset. Clusters have been labeled post hoc after comparison to the larval dataset (See Figures 5C and 5D). B. Gene Expression profiles (columns) of select cluster-specific markers identified through differential expression analysis (DEA) across all adult clusters. Bar on the right displays percent of total dataset represented in every adult cluster, showing the abundance of each cell type found by clustering analysis. C. Dot plot (confusion matrix) showing the proportion of gng8⁺ cells in the adult dataset (rows) that were classified to larval cluster labels (columns). Each adult habenular type was assigned to a larval cluster label if >15% of the trees in the RF model contributed to the majority vote. Proportion of cells in each row should add to a 100%. D. Dot plot (confusion matrix) showing the proportion of larval cells (rows) that were classified to cluster labels of the gng8⁺ cells in the adult dataset (columns). Each adult habenular type was assigned to a larval cluster label if >15% of the trees in the RF model contributed to the majority vote. Proportion of cells in each column should add to a 100%. This training on the adult dataset was performed to validate the robustness of the RF analysis. E. FISH validation and localization of select dorsal habenular cluster markers. F. FISH validation of the genes that are expressed in all ventral clusters (aoc1) and across three other ventral sub-clusters (cd82a, mprip and zgc:173443). See also Figure S5 and Table S2.

Figure 6. Divergent Expression Patterns of Functionally Relevant Genes among the Larval Habenular Neuronal Types

(A–F) Gene expression profiles of select functionally relevant genes among larval habenular types visualized in the form of a dot plot. Representation as in Figure 1D. A) Neuropeptides and Neuropeptide Receptors B) Transporters C) Neurotransmitter Receptors D) Calcium Channels E) Potassium Channels. Only genes expressed in >20% of cells in at least each habenular type are shown. See also Figure S6.

Figure 7. Noxious Electric Shocks Activate a Sub-Population of Neurons in the Ventro-lateral Habenula Labeled by mprip

A. ISH analysis of cfos expression in the habenula 30 minutes after exposure to electric shocks. [Scale bars represent 50 μm] B. Registration of cfos signals to habenular molecular atlas reveals co-regionalization with the mprip⁺ ventrolateral population [Scale bars represent 50 μm]. C. Double in situ hybridization for c-fos and mprip, (marker for ventrolateral neuronal type) showing a co-localization of cfos⁺ and mprip⁺ domains in the larval habenula in response to electric shocks [Scale bars represent 10 μm]. D. Double in situ hybridization of cfos and mprip showing the conservation of electric shocks-induced cfos responses in mprip⁺ ventro-lateral neuronal type in the adult habenula. Nuclei borders are demarcated in the zoomed in panels on the right using dotted circles [Scale Bar represents 10μm unless otherwise stated]. See also Figure S7 and Movie S2