Figure 3. The transcription factor (TF) regulatory landscape in whole-brain neuronal clusters.

(A) Schematic showing the strategies to assess the cluster similarity based on effector gene and TF profiles. We focused on clusters of whole-brain glutamatergic/GABAergic neurons and neuromodulator neurons. Similar pair clusters were identified by two strategies: the first strategy was based on hierarchical sister clusters. The second strategy was based on population-level statistical analysis, in which we calculated and ranked the distances of every two clusters from 39 neurotransmitter-type clusters ($C_{39}^2$) and chose the ones with the lowest 10% distance as similar pair clusters. (B) Left: schematic showing the criteria of three patterns: pair clusters that were similar in both TF and effector gene profiles as ‘matched pattern’, those pair clusters that were similar in effector gene profiles but not in TF profiles as ‘convergent pattern’, and those pair clusters that were similar in TF profiles but not in effector gene profiles as ‘divergent pattern’. Right: the plot showing the number of each pattern using two strategies in A. The red dashed circle showing the number of cluster pairs with given pattern based on hierarchical sister cluster analysis; the black solid circle showing the number of cluster pairs with given pattern based on population-level statistical analysis. (C–D) Violin plots showing the expression of TFs (yellow) or effector genes (black) in glutamatergic and GABAergic similar pair clusters of convergent pattern. (E) The bar plot showing the proportions of different patterns for neuronal clusters with neurotransmitter or neuromodulator types. The numbers of each pattern were indicated. Fisher′s exact test was used to test the significant association of different patterns, p = 0.02564, *p < 0.05. (F–G) Violin plots showing the expression of TF profiles (yellow) or effector gene profiles (black) in neuronal clusters of divergent pattern.

Figure 3—figure supplement 1. Hierarchical clustering analysis of whole-brain glutamatergic/GABAegric neurotransmitter-type neurons based on effector gene and transcription factor (TF) profiles.

(A) Hierarchical clustering of 39 glutamatergic/GABAegric neurotransmitter-type neurons based on 1099 effector gene profiles in highly variable genes. Red boxes were highlighted to indicate the terminus sister clusters strongly supported by pvclust (AU > 0.95, p < 0.05). (B) Visualizing the matching nodes between TF-based tree and effector gene-based tree with R package ‘TreeDist’. Red line indicated the matching node (node 13, Clusters 9 and 61). Each split was labeled with a measure of similarity score. Bottom tree distance (0.7127991) indicated the different organization of TF-based and effector-based tree. (C) Similarity score of each split in (B).

Figure 3—figure supplement 2. The transcription factor (TF) regulatory landscape in whole-brain glutamatergic/GABAergic neuronal clusters.

(A) Schematic showing the sister clusters identification based on hierarchical clustering of 39 whole-brain glutamatergic/GABAergic neurotransmitter neuronal clusters (IIa: glutamatergic neurons and IIb: GABAergic neurons, in blue and red region of t-distributed stochastic neighbor embedding [t-SNE] plot, respectively). Upper hierarchical clustering was based on 1099 effector genes, red lines indicated 11 terminus sister pairs. The bottom hierarchical clustering was based on 283 TFs, red lines indicated 14 terminus sister pairs. (B) A schematic for matched pattern (left) and a summary table (right) showing that neuronal clusters at similar termini of TF-based hierarchy were matched as sister cluster pairs within effector gene-based hierarchy (n = 1). (C) A schematic for convergent pattern (left) and a summary table (right) showing that glutamatergic/GABAergic neuromodulator-type neuronal clusters at different termini of TF-based hierarchy were found as sister cluster pairs within effector gene-based hierarchy (n = 10). (D) Scatter plot showing the distance distribution of glutamatergic/GABAergic neurotransmitter-type neuronal clusters with matched pattern (shown below) according to the effector-based (white) and TFs-based (gray) distance. Distance defined as (1 − (Pearson correlation coefficient between clusters))/2. Red and black dots showed the distance of each pair clusters. In matched patter, the red and black dots were the lowest distance in both effector gene-based (white) and TF-based (gray) distances. (E) The table showing the brain regions and neurotransmitter type of each pair clusters with matched pattern (n = 5). Regional origins and neurotransmitter type are color-coded. (F) Scatter plot showing the distance distribution of glutamatergic/GABAergic neurotransmitter-type neuronal clusters with convergent pattern (shown below) with other clusters according to the effector-based (white) and TFs-based (gray) distance. Distance defined as (1 − (Pearson correlation coefficient between clusters))/2. Red and black dots showed the distance of each pair clusters. In convergent pattern, the red and black dots were the lowest distance in effector gene-based (white) distances, but not the lowest distance in TF-based (gray) distances. (G) The table showing the brain regions and neurotransmitter type of each pair clusters with convergent pattern (n = 14). Regional origins and neurotransmitter type are color-coded. (H) Heatmap showing the distances distribution of 39 glutamatergic/GABAergic neurons based on effector gene profiles. The hierarchical clustering in the heatmap is similar as Figure 3—figure supplement 2A (upper). (I) Heatmap showing the distances distribution of 39 glutamatergic/GABAergic neurons based on subsampling 80% of effector genes. The hierarchical clustering in the heatmap is similar as Figure 3—figure supplement 2A (upper). The tree distance before and after subsampling was 0.1407203 (calculated by ‘TreeDist’). (J) Heatmap showing the distances distribution of 39 glutamatergic/GABAergic neurons based on TF profiles. The hierarchical clustering in the heatmap is similar as Figure 3—figure supplement 2A (bottom). (K) Heatmap showing the distances distribution of 39 glutamatergic/GABAergic neurons based on subsampling 80% of TF profiles. The hierarchical clustering in the heatmap is similar as Figure 3—figure supplement 2A (bottom). The tree distance before and after subsampling was 0.2032071 (calculated by ‘TreeDist’). (L) Summary table showing the pairs of neuronal clusters with three patterns before and after subsampling. Re-identified patterns using subsampling completely recapitulated those pairs identified using the population-level statistical analysis based on total genes.

Figure 3—figure supplement 3. Similar pair clusters of neuromodulator-type neurons.

(A-B) Hierarchical clustering of neuromodulator-type neurons based on 1783 effector genes or 319 transcription factors (TFs) in highly variable genes. The tree plots visualizing the matching nodes between TF-based tree and effector gene-based tree with R package ‘TreeDist’. Red line indicated the matching node. Bottom tree distance (0.3766735) indicated the similarity organization of TF-based and effector-based tree. (C) Each split node in (A-B) were labeled with a measure of similarity score. Matching nodes were highlighted. (D-E) The table showing the terminus sister pairs of vmat2⁺ neuromodulator neurons from effector gene-based (D) or TF-based (E) hierarchical sister cluster analysis. The neuromodulator identities, brain region origins, and co-expression neurotransmitters were color-coded. (F-G) A schematic for matched pattern (F), convergent pattern (G) showing five and three similar neuromodulator pairs with matched and convergent pattern, respectively, based on hierarchical sister cluster analysis. The neuromodulator identities were color-coded. (H) Scatter plots showing the distribution of the distances of clusters of each vmat2⁺ neuromodulator-type clusters exhibiting matched (light red) and convergent (light orange) patterns to other clusters using the effector-based (white) or TF-based (gray) population-level statistical analysis. Red and black dots represented the distances between two similar clusters. In matched pattern, the red and black dots had the lowest distances in both effector gene-based (white) distance cohort, and TF-based (gray) distance cohort. In convergent pattern, the red and black dots had the lowest distance in effector gene-based (white) cohort, but had higher distances in TF-based (gray) distance cohort. (I) Summary table showing the pairs of similar neuromodulator clusters with two patterns in (H) or after subsampling (I). (J) Venn plots showing the number of each pattern using two strategies (hierarchical sister cluster and population-level statistical analysis) in Figure 3A. The red dashed circle showing the number of cluster pairs with given pattern based on hierarchical sister clusters; the black solid circle showing the number of cluster pairs with given pattern based on population-level statistical analysis. (K) Dot plot showing the TFs expression in matched terminus pairs from vmat2⁺ neuromodulator neurons. Sister clusters in a pair exhibited similar TFs expression patterns (red box). (L) Pheatmap showing the similar effector gene expression between each sister clusters of vmat2⁺ neuromodulator neurons with matched pattern.

Figure 3—figure supplement 4. Divergent pattern of glutamatergic/GABAergic neurotransmitter-type neuronal clusters.

(A) A schematic for divergent pattern (left) and a summary table (right) showing that neuronal clusters as sister cluster pairs within transcription factor (TF)-based hierarchy were divergent at different termini within effector gene-based hierarchy (n = 13). Colored dots represent the regional origins of each cluster. (B) Scatter plot showing the distance distribution of glutamatergic/GABAergic neurotransmitter-type neuronal clusters with divergent pattern (shown with fork) with other clusters according to the effector-based (white) and TF-based (gray) distance measures. Distance defined as (1 − (Pearson correlation coefficient between clusters))/2. Red and black dots showing the distance of each pair clusters with divergent pattern. In divergent pattern, the red and black dots were the lowest distances in TF-based (gray) distance, but not the lowest distance in effector gene-based (white) distances.