Integrated single-cell atlases unveil the operation principles of whole-brain 5-HT neuronal subsystems Authors: Ziqi Song, Chun-Xiao Huang, Huixian Zhang, Chuanzhe Ye, Na Guan, Jianren Song Published: 2025 ABSTRACT Investigating 5-hydroxytryptamine (5-HT) neuronal subsystems and the operation principles is crucial for elucidating their complex roles in the regulation of various physiological and behavioral outcomes. Until now, these efforts faced methodological and analytical hurdles. Here, we developed resource datasets in zebrafish that include a single-cell projectomic dataset, two independent scRNA-seq datasets, and a whole-brain 5-HT neuron functional imaging database at cellular resolution. The integrated analysis uncovered seven 5-HT subpopulations, each with unique axon projection patterns, specific soma locations, and distinctive genetic profiles. Within each subsystem, 5-HT neurons were functionally homogeneous, whereas across different subsystems, they were functionally heterogeneous. We further demonstrate that each 5-HT subsystem functions as a distinct module playing a role in multiple tasks, rather than being exclusive to one function. Moreover, the involvement of multiple parallel 5-HT subsystems, rather than a single subsystem, is fundamental to the accomplishment of a specific task. These results provide insights into the operation of 5-HT neuronal system in mammals. INTRODUCTION The 5-hydroxytryptamine (5-HT) neuronal system is integral to a myriad of physiological and behavioral phenomena, including sensory perception, motor function, learning and memory, brain state transitions, and reward processing (1–6). The remarkable functional versatility of this system can largely be attributed to the multidimensional diversities of whole-brain 5-HT neurons, despite common serotonergic phenotypes (7–11). Miscellaneous 5-HT neuron subtypes have been delineated on the basis of increasing information on anatomical, morphological, electrophysiological, and developmental genetic properties (3, 4, 8, 9, 12–16). The categorization of 5-HT neurons in rodents was achieved on the basis of their anatomically and transcriptionally defined rhombomeric domains of origin, resulting in nine major subtypes originally designated as B1 through B9 (17–21). However, in most anatomical domains, rhombomere-derived 5-HT lineages converge or intermingle (9, 14, 18, 22), suggesting that the heterogeneity of the 5-HT neuron is driven by multifaceted factors operating across multiple dimensions. To avoid bias in the classification that arises from considering only one or two factors obtained with limited methodologies, it seems imperative to us to establish the atlases of the whole-brain 5-HT neurons at single-cell resolution, integrating soma localization, axon projections, genetic profiles, and functionality, to allow for a comprehensive analysis of subsystem characterization. The extensive projections of the 5-HT neuronal system enable serotonin release in most of the brain regions, which in turn notably affects physiological and behavioral outcomes (3, 4, 23–25). Emerging evidence suggests that 5-HT neurons innervating the same brain regions are organized as a module and, in general, share comparable functions (4, 14, 25–27). This raises the possibility that distinct 5-HT subpopulations or subsystems could be categorized by their axon projection patterns and that the operation of the 5-HT subsystems is organized in a modular manner (7, 22, 28). A critical question that remains unanswered is whether one single 5-HT subsystem is solely responsible for one specific task or extensively engages in several tasks. In other words, it needs to be determined whether accomplishing a specific task depends on the recruitment of a particular 5-HT subsystem or requires the involvement of multiple parallel 5-HT subsystems working in parallel. To answer these questions requires the concurrent analysis of the activity of individual 5-HT neurons and their belonging subsystems across the entire brain during various specific tasks, presenting a technical and analytical challenge. Here, we present a single-cell projectomic dataset of 5-HT neurons across the whole brain in zebrafish larvae using single-neuron genetic labeling. We have identified seven distinct 5-HT subpopulations, each with unique axonal projection patterns and somatic locations. Two independent single-cell RNA sequencing (scRNA-seq) datasets reveal that these seven subpopulations display distinct transcriptomic profiles and characteristic molecular markers which we suggest to be a putative homolog of anatomical subpopulations in the mouse 5-HT system. Multiplane two-photon calcium imaging of the whole-brain 5-HT neurons showed that multiple subpopulations are recruited in parallel and exhibit functional heterogeneity, with each displaying modular-specific activity and undertaking the dispatched subtask. Neurons within each subpopulation were found to be functionally homogeneous while displaying a high level of fidelity in their activity patterns specific to their respective subpopulation. Thus, our findings identify existence of seven 5-HT subsystems in the zebrafish brain, each constituting a distinct functional module differentially and coordinately engaged in diverse tasks. This approach may be useful in unraveling the complexity of 5-HT neuronal system in mammals. RESULTS The single-cell projectomic atlas delineates seven 5-HT neuronal subpopulations Both in fish and mouse, only tryptophan hydroxylase (Tph) can serve as a cell type–specific genetic marker for 5-HT producing neurons in the central nervous system compared with other markers for 5-HT metabolism (7, 29). In zebrafish, there are three tph genes (tph1a, tph1b, and tph2), among which tph1a and tph1b only specifically present in pineal gland, posterior tuberculum, and hypothalamus (29, 30). In zebrafish, 5-HT neurons were identified in both the brain and spinal cord (31). To acquire the spatial distribution of the somata of 5-HT neurons in the whole brain, we imaged the entire brain of larval zebrafish at 7 days postfertilization (dpf) using the tph2:GFP fish line, in which green fluorescent protein (GFP) was driven by tph2 (Fig. 1A). The specificity of tph2 promotor-driven labeling was confirmed by immunostaining with an antiserotonin antibody, showing that 96% of tph2+ neurons were 5-HT positive (fig. S1, A, B, and D). The somata of 5-HT neurons were confined to four brain regions/nuclei, consisting of pretectum (PT; ~40 neurons), superior raphe (SR; ~80 neurons), inferior raphe (IR; ~40 neurons), and medulla oblongata (MO; ~25 neurons) (Fig. 1B and fig. S1C). The previously reported hypothalamic population was not observed in this fish line due to the use of tph2 promoter (32). In mammals, the hypothalamic 5-HT–positive neurons do not express TPH, instead they appear to take up 5-HT through serotonin transporters (33). So, we further focus on the 5-HT neuron populations identified in these four brain regions above (PT, SR, IR, and MO). The single-cell projectome of the whole-brain 5-HT neurons reveals seven subpopulations. (A) 3D reconstruction of the somatic distribution of the whole-brain 5-HT neurons based on the confocal imaging of the tph2:GFP fish line. The soma surface was reconstructed using Imaris, and the brain volume was registered to the brain template generated from the web interface of mapzebrain (https://mapzebrain.org/downloads). Both dorsal and lateral views are shown. R, rostral; C, caudal; D, dorsal; V, ventral. (B) Anatomical annotation of the whole-brain 5-HT neurons in four major brain regions, including PT, SR, IR, and MO. (C) Experimental workflow of sparse labeling, single-neuron tracing, morphology reconstruction, and registration to the Z-brain reference atlas. (D) Seven neuronal clusters were revealed by unsupervised classification of the single-neuron projectome (n = 188 neurons, N = 175 fish) by K-means (left) and anatomical annotation of these seven clusters based on the somatic location of the neurons in the four brain regions (right). Results were visualized by the t-SNE method. (E and F) 3D reconstruction and visualization of the projectome of 188 individual 5-HT neurons after registering to the Z-brain reference atlas (outlined in gray). 3D surface models of four major brain regions (PT, SR, IR, and MO) and Mauthner cell somata (as a landmark) were created in 3D slicer and displayed in distinct colors. Neurons of the seven subpopulations were color-coded and displayed altogether (E) or separately (F). Both dorsal and lateral views were presented. (PT, n = 31; SR-R, n = 25; SR-M, n = 38; SR-C, n = 27; IR-R, n = 22; IR-C, n = 19; MO, n = 26). (G) Summary of the seven 5-HT neuronal subpopulations with the distinct somatic location segregated within four major brain regions. To determine whether whole-brain 5-HT neurons could be further classified into subpopulations based on axon projection patterns, we generated a resource dataset to illustrate the main morphology of the single 5-HT neurons in the zebrafish brain. A single neuron was labeled in the HuC:H2B-GCaMP6 fish line with pan-neuronal expression of GCaMP6f by injecting a tdTomato-expressing plasmid driven from the upstream activating sequence (UAS) together with a Gal4-VP16 under the tph2 promoter. Whole-brain z-stack images were obtained at 6 dpf and then registered to a standard digital brain (34, 35) for visualizing individual neurons and performing cellular reconstruction (Fig. 1C). A total of 188 tph2+ neurons were included in the single-cell projectomic dataset (fig. S2A). We next performed unsupervised K-means classification analysis based on neuronal projectome and somatic location to classify 5-HT neuronal subpopulations, which was displayed by t-distributed stochastic neighbor embedding (t-SNE) (Fig. 1D, left). As revealed, all labeled 5-HT neurons can be divided into seven subpopulations (PT, SR-R, SR-M, SR-C, IR-R, IR-C, and MO) each displaying distinct axon projection pattern and soma location (Fig. 1, D to G). 5-HT neurons in PT subpopulation mainly innervated optic tectum, dorsal thalamus, and tegmentum and ipsilaterally projected to the medial region of hindbrain (Fig. 1F and fig. S2, B and C). Neurons in SR and IR have regionally biased soma location along a rostro-caudal axis and distinct axon projection patterns and can be further subdivided into three (SR-R, SR-M, and SR-C) and two (IR-R and IR-C) subpopulations, respectively (Fig. 1, F and G. and fig. S2, B and C). Notably, the three SR subpopulations showed substantial innervation of the telencephalon (including pallium and subpallium). This is consistent with previous findings that 5-HT neurons in the raphe nucleus intensively project to forebrain regions including prefrontal cortex, amygdala, and dorsal hippocampus in mammals (26, 36, 37). The three SR subpopulations display distinct projection patterns. 5-HT neurons in the SR-M subpopulation predominantly innervated the hindbrain, with longer axonal projections. Conversely, SR-R and SR-C subpopulations exhibited stronger projections to the diencephalon and midbrain than to the hindbrain (Fig. 1, F and G, and fig. S2, B and C). Moreover, 5-HT neurons of the SR-C subpopulation displayed rare ascending projection to olfactory bulb and descending projections to MO (Fig. 1, F and G, and fig. S2, B and C). Both IR-R and IR-C subpopulations predominantly innervated hindbrain. 5-HT neurons of IR-C subpopulation showed longer descending projection that ultimately extended to the spinal cord. The IR-R neurons were distinctive as the exclusive origins of 5-HT innervation in cerebellum (Fig. 1, F and G, and fig. S2, B and C). Axon of the 5-HT neurons in MO subpopulations projected locally and contralaterally within the hindbrain (Fig. 1, F and G, and fig. S2, B and C). Immunostaining of 5-HT–positive fibers revealed an extensive serotonergic innervation in the optic tectum, including the dorsal neuropil layers (fig. S1E). Consistent with the immunostaining and the previous studies in zebrafish and mammals, we observed widespread and abundant innervation of the optic tectum (the homolog of mammal superior colliculus), a visual processing structure (6, 38), by 5-HT neurons in the projectomic dataset. These 5-HT neurons innervating the optic tectum were predominantly from the PT, SR, and IR subpopulations. These data demonstrate that 5-HT neurons in the zebrafish brain can be classified into seven distinct subpopulations based on the main morphology revealed by the single-cell projectomic dataset. Each subpopulation has the unique axon projection pattern and specific soma spatial location. The seven 5-HT neuronal subpopulations display distinct transcriptomic profiles To determine whether the seven 5-HT neuronal subpopulations revealed by projection pattern were genetically heterogeneous, we generated two independent single-cell transcriptome datasets. We took advantage of the separated soma locations of the seven subpopulations and manually collected the soma of neurons in each subpopulation via a glass pipette for scRNA-seq using the tph2:GFP fish line (Fig. 2A). A total of 71 sample pools from seven 5-HT neuronal subpopulations were generated as the first dataset and subjected to transcriptomic analysis. The intersample distances were calculated on the basis of their expression profiles to assess intrasubpopulation similarity and intersubpopulation divergence. Hierarchical clustering analysis on this dataset revealed that sample pools from each subpopulation reliably separated into distinct groups, although three SR subpopulations exhibiting some overlap (fig. S3A). The overall organizational structure was elucidated by conducting principal components analysis (PCA) on 3000 highly variable genes and subsequently projecting the data onto two scaled dimensions using t-SNE (Fig. 2B). Differential expression analysis revealed that each subpopulation expressed a characterized set of featured genes (Fig. 2C and fig. S3B), including specific markers for neurotransmitters, axon projection, neuron development, synapse, and ion channels (fig. S3C), indicating that the seven 5-HT neuronal subpopulations identified by the single-cell projectomic dataset display distinct genetic profiles. Transcriptomic profiles of seven 5-HT neuronal subpopulations. (A to C) Illustration of manual somata sample collection for scRNA-seq of seven subpopulations (n = 10 to 12 neurons, N = 71 sample pools) using the tph2:GFP fish line (A). PCA is performed on highly variable genes of 71 sample pools, followed by visualization using t-SNE (B). Dot plot shows key marker genes across neuronal subpopulations. For each cluster, three representative genes from the top 30 markers (identified by both manual and 10× scRNA-seq) are shown. Dot size indicates the percentage of expressing cells (0 to 100%) and color scale representing the average scaled normalized expression (C). (D to F) Processing of 10× scRNA-seq of tph2+ neurons by FACS from the whole brain (D). The t-SNE plot shows 1489 tph2+ neurons clustered into seven subpopulations by unsupervised clustering. Expression of the tph2 gene is overlaid, with the color bar representing scaled normalized expression values (E). Each cluster is annotated and color-coded on the basis of shared marker genes identified by both manual and 10× scRNA-seq, corresponding to the seven subpopulations: PT for C1 (n = 249); SR-R for C2 (n = 352); SR-M for C3 (n = 190); SR-C for C4 (n = 311); IR-R for C5 (n = 114); IR-C for C6 (n = 191); MO for C7 (n = 82) (F). (G) Feature plots show the expression pattern of marker genes tcf7l2, en2b, gstp1, cart4, calb2a, nkx6.1, and mab21l1 for the PT, SR-R, SR-M, SR-C, IR-R, IR-C, and MO subpopulations, respectively. The color bar indicates scaled normalized expression values. (H) HCR-based RNA-FISH for tcf7l2, en2b, cart4, msi2a, nkx6.1, and mab21l1 using the tph2:GFP fish line shows colocalization of GFP+ 5-HT neurons and probe signals in the corresponding brain regions (white dashed boxes are enlarged). The seven 5-HT neuronal subpopulations share common serotonergic ID genes for serotonin synthesis and transport apart from the PT subpopulation, which is pet1 negative, and the MO subpopulation, which lacks pet1 and sert (fig. S4A). Neurons in most of the subpopulations express the genes related to GABA or glutamate synthesis or transport, indicating the potential for neurotransmitter corelease (fig. S4A), which is consistent with the mammalian 5-HT system (7). Notably, there was a differential expression of vglut3 and gat1/viaat along the rostro-caudal axis across three SR subpopulations with the caudal subpopulations (SR-M and SR-C) highly expressing vglut3 and the rostral SR-R subpopulation enriched in gat1/viaat (fig. S4A). The different expression pattern of gad1b and vglut3 in the seven 5-HT neuronal subpopulations was further confirmed by hybridization chain reaction (HCR)–based RNA fluorescence in situ hybridization (RNA-FISH) using the tph2:GFP fish line at 7 dpf (fig. S4, B and C). To validate the distinct transcriptomic signatures of the seven neuronal subpopulations, we next expanded our dataset substantially via 10× scRNA-seq based on fluorescence-activated cell sorting (FACS)–isolated neurons (Fig. 2D). We identified a total of 1489 tph2+ neurons that grouped into seven distinct clusters (C1 to C7) depicted in the t-SNE plot (Fig. 2E). Using the feature genes discerned by manually collected scRNA-seq, these clusters were characterized as PT, SR-R, SR-M, SR-C, IR-R, IR-C, and MO (Fig. 2, F and G), corresponding to C1 to C7, respectively. To verify the transcriptomic profiles of these seven subpopulations, we selected 1 of the top 10 shared marker genes identified by both manually collected scRNA-seq and 10× scRNA-seq for each cluster. We then performed HCR-based RNA-FISH using the tph2:GFP fish line, which confirmed the specific expression of tcf7l2, en2b, cart4, msi2a, nkx6.1, and mab21l1 in PT, SR-R, SR-C, IR-R, IR-C, and MO subpopulations, respectively (Fig. 2H). In addition, the expression of Calb2a within the IR-R subpopulation was validated through immunostaining with an anti-calretinin antibody (fig. S4D). Last, by comparing the relative expression of a series of homologous genes in mouse 5-HT neurons (7–9, 12), we suggest putative relationship for 5-HT subpopulations in mice and zebrafish (fig. S5A). These results demonstrate that the seven 5-HT neuronal subpopulations identified by single-cell projectome analysis are genetically heterogeneous. Each 5-HT neuronal subpopulation having distinct soma locations, axon projection patterns, as well as molecular signature embraces a distinct 5-HT neuronal subsystem. A Gene Ontology (GO) analysis of the differentially expressed genes unique to each subpopulation underscored the functional heterogeneity of seven subsystems (fig. S6A). Neurons within 5-HT subpopulations exhibit a modularly organized activity pattern The 5-HT neurons are deeply involved in sensorimotor behaviors (5, 6, 22, 39–41). To investigate whether the identified 5-HT subpopulations are functionally heterogeneous in performing the same task, we conducted multiplane two-photon imaging of whole-brain 5-HT neurons using the tph2:GCaMP6 fish line, in which six of the subpopulations (PT, SR-R, SR-M, SR-C, IR-C, and MO) are spatially distinguishable, except for IR-R which was not well labeled (fig. S7A), during a series of sensorimotor tasks consisting of visuomotor behavior, food exposure, and spontaneous locomotion (Fig. 3, A, E, and I). We analyzed the calcium responses of these six 5-HT neuronal subpopulations at both the subregion and single-neuron levels. Visual flow stimuli reliably induced optomotor behaviors measured by tail bending angle in larval zebrafish (Fig. 3, A and B), during which the six subregions exhibit distinct calcium dynamic patterns with respect to the onset and amplitude of the response. Application of visual flow (t0) immediately recruited PT and MO regions, while subsequently PT was inhibited and MO was steadily active preceding each induced motor behaviors (Fig. 3B). The three SR subregions were recruited after the occurrence of induced motor behaviors (t1) with variable response amplitude, suggesting that they likely receiving locomotion feedbacks (Fig. 3B and fig. S7B). IR-C remained relatively inactive across the entire process (Fig. 3B and fig. S7B). The PCA confirmed that six subregions exhibited distinct response patterns during the visuomotor behavior (Fig. 3A), suggesting that 5-HT neurons in each subregion serve as a distinct functional module to undertake their own subtasks. To examine the hypothesis that these six 5-HT neuronal subpopulations are functionally modular organized, we analyzed the calcium response of individual neurons and evaluated whether the activity of single neurons followed the response pattern of their belonging subpopulation during visuomotor behaviors. Then, unsupervised K-means classification was performed on the activity of a total of 413 single neurons in six fish to yield six functional clusters, within which single neurons were subsequently annotated by their belonging subpopulation (Fig. 3, C and D, and fig. S7C). We found that the activity of most neurons within each subpopulation (~70% or higher) aligns with the response pattern of the respective subpopulation to which they belong (Fig. 3D, bottom). The high fidelity of the activity of single neurons to their belonging subpopulation demonstrates that 5-HT neuronal subsystems operate in a modular-organized manner during visuomotor behaviors. The performance of visuomotor behavior requires the concurrent engagement of multiple parallel 5-HT subsystems. Distinct modular-organized activity of 5-HT neuronal subpopulations during sensorimotor behaviors. (A to D) Illustration of the two-photon calcium imaging during visuomotor behavior in the tph2:GCaMP6 line. PCA of calcium responses at the subregional level (N = 6 fish) (A). Typical calcium response traces of each 5-HT subpopulation at the subregional level. The onset of visual stimulation (t0) or motor behavior (t) are indicated (B). Heatmap of calcium responses of individual 5-HT neurons from one preparation (n = 72 neurons) (C). Unsupervised K-means assignment of single-neuron responses to a unique cluster (C1 to C6). All analysis neurons are anatomically annotated to the four major brain regions. Neurons whose response pattern aligned with their subpopulation are color-coded; the remaining neurons are shown as gray. Response loyalty quantifies the percentage of neurons in each anatomical area whose response matches their K-means assigned subpopulation’s pattern in (D). Each dot represents one neuron (n = 413 neurons from six fish). (E to H) Illustration and PCA of two-photon calcium imaging during paramecia supply induced prey-like behavior (N = 6 fish) (E). Typical calcium response traces of each 5-HT subpopulation. The time of paramecia supply stimulation (t0) and the start (t) of each prey-like behavior are indicated (F). Heatmap of the calcium responses of individual 5-HT neurons (n = 82 neurons) (G). Unsupervised K-means of single neurons (n = 422 neurons from six fish) (H). (I to L) Illustration and PCA of two-photon calcium imaging during spontaneous locomotion (N = 7 fish) (I). Typical calcium response traces of each 5-HT subpopulation. The start (t) of each major motor event is indicated (J). Heatmap shows calcium responses of individual 5-HT neurons (n = 80 neurons) (K). Unsupervised K-means of single neurons (n = 454 neurons from seven fish) (L). After fasting for 48 hours, a food exposure test was performed at 8 dpf (see Materials and Methods for details). Whole-brain calcium imaging was performed in conjunction with simultaneous tail tracking. Food exposure reliably triggered prey-like behaviors, during which the six 5-HT neuronal subregions also showed differential response patterns (Fig. 3E). PT and MO subpopulations were immediately activated after food supply (t0) and sustained this activity albeit with different amplitudes for each prey-like behavior (t1) (Fig. 3F). The three SR subregions displayed divergent response patterns with SR-R being inhibited, SR-M marginally responsive, and SR-C constantly recruited during food exposure–induced prey-like behaviors (Fig. 3F). Similar to visuomotor behavior, IR-C also exhibited subtle response to food exposure (Fig. 3F). Unsupervised K-means classification on the activity of 422 individual neurons from six fish showed that calcium activity of most neurons (~75%) within SR-R, SR-M, SR-C, or IR-C showed similar response patterns with their belonging subpopulation, while PT and MO neurons shared two similar activity patterns (Fig. 3, G and H, and fig. S7D). The high fidelity of the activity of individual neurons to their belonging subpopulation suggests that 5-HT subsystems operate in a modular-organized manner during food exposure–induced prey-like behaviors. The control group (administration of water alone) further confirmed that the prey-like behaviors were triggered by food exposure (fig. S7, F and G). Furthermore, for spontaneous locomotion, the IR-C was strongly activated after the start of locomotor events, indicating an involvement in locomotion feedback, while the other five subregions were differentially recruited with distinctive response pattern or amplitude before initiation of locomotion (Fig. 3, I and J). Unsupervised K-means classification on the activity of 454 neurons from seven fish revealed that most neurons (approximately or higher than 60%) shared the same response pattern of the subpopulations to which they belong (Fig. 3, K and L, and fig. S7E). The high fidelity of the activity of single neurons to their subpopulation indicates that 5-HT subsystems are recruited following a modular organization during spontaneous locomotion. These results demonstrate that the seven 5-HT subpopulations identified by projectomic and transcriptomic data are functionally heterogeneous. Neurons within each subpopulation form an independent 5-HT subsystem and display a modularly organized activity pattern. Multiple parallel 5-HT subsystems are concurrently recruited with distinct responsive patterns to regulate a specific sensorimotor behavior. Furthermore, the same 5-HT subpopulation exhibits differential response patterns, including both excitation and inhibition, for distinct sensorimotor behaviors, indicating that each subsystem is used to accomplish the corresponding subtasks under varying circumstances. Notably, these 5-HT subsystems can also be simultaneously recruited following a similar pattern when the animals are in danger or vigilance (42). An electrical shock induced the same respond pattern for all examined subpopulations (fig. S7, H to J). Three SR subsystems serve as distinct modules to regulate behavioral states There is an extensive literature showing that 5-HT neurons in the SR (also named the dorsal raphe nucleus) can modulate behavior states (3, 4, 41, 43–53). To evaluate the differential functions of the three SR subpopulations in regulating behavior states, we examined the calcium response evoked by a range of behavioral paradigms in the three 5-HT neuronal subpopulations in SR at the subregion and single-neuron levels (Fig. 4). Sleep deprivation (SD) was performed on 6 dpf zebrafish larvae for 24 hours using the optomotor response (OMR) (see Materials and Methods for details). Following SD, the fish fell into a stable rebound sleep state as confirmed by notably increased sleep time, as well as the absence of electromyogram (EMG) activity (Fig. 4, A and B, and fig. S7L). During the rebound sleep state, persistent low-frequency calcium oscillations were observed in SR with gradually decreased amplitude from SR-R (largest) to SR-M and SR-C (smallest) which separated clearly into three clusters using PCA (Fig. 4, A and B, and fig. S7K). We used the unsupervised K-means classification to analyze the activity of 269 neurons of the SR in six fish, which revealed that most neurons (~80%) shared the same response pattern with their belonging subpopulations (Fig. 4, C and D, and fig. S7M). These results can be interpreted as showing that the SR-R subsystem plays a more important role during the sleep state than the other two subpopulations. In another set of experiments designed to motivate prey-like behavior, the fish were deprived of food for 48 hours. When food was subsequently supplied, this induced prey-like J-turns behavior characterized by tail movement trajectory and tail angle analysis (fig. S7O) and resulted in differential responses in the three SR subregions with SR-R receiving inhibition, SR-C being excited, and SR-M showing a subtle response (Fig. 4, E and F, and fig. S7N). Unsupervised K-means classification analysis on 255 neurons of the SR in eight fish demonstrated that most neurons followed the response pattern of their respective subpopulation (Fig. 4, G and H, and fig. S7P). These data indicate that the coordinated but antagonistic actions of SR-R and SR-C subsystems collectively modulate the prey motivation following food deprivation. We next examined the response of three SR subpopulations to acoustic startle response habituation, an evolutionarily conserved nonassociative form of learning (54, 55). Repetitive sound stimulation applied 40 times at intervals of 10 s induced startle response habituation characterized by decreased P startle response (fig. S7R), during which SR-M was strongly and persistently activated, while both SR-R and SR-C showed far smaller responses that decayed quickly (Fig. 4, I and J, and fig. S7Q). Single neuronal activity analyzed by unsupervised K-means classification confirmed that the vast majority of neurons adhered to the activity pattern with their respective subpopulation (Fig. 4, K and L, and fig. S7S). This suggests that the SR-M subsystem plays a major role in acoustic startle response habituation. Functional heterogeneity of three SR subpopulations for specific behavioral states. (A to D) Illustration of two-photon calcium imaging of the SR 5-HT neurons after SD for 24 hours using the tph2:GCaMP6 line. PCA of calcium responses in the three SR subregions (N = 6 fish) (A). Typical calcium response traces of each 5-HT subpopulation at the subregion level and EMG signal (B). Heatmap of the calcium responses of individual 5-HT neurons from one preparation (n = 29 neurons) (C). Unsupervised K-means classification of single-neuron responses into unique clusters (C1 to C3). All analysis neurons were anatomically annotated to the SR based on their somatic location. Only neurons whose response pattern aligned with their subpopulation companions were color-coded; the remainder are shown in gray. The response loyalty, represented by the percentage of neurons whose response aligned with that of their assigned subpopulation, is shown in the bottom in (D). Each dot represents one neuron (n = 269 neurons from six fish) (D). (E to H) Illustration of two-photon calcium imaging during food deprivation–induced behavioral motivation. PCA of calcium responses (N = 8 fish) (E). Typical calcium response traces of each 5-HT subpopulation (F). Heatmap of the calcium responses of individual 5-HT neurons from one preparation (n = 41 neurons) (G). Unsupervised K-means classification of single-neuron responses into unique clusters (C1 to C3) (n = 255 neurons from eight fish) (H). (I to L) Illustration of two-photon calcium imaging during acoustic startle response habituation. PCA of calcium responses (N = 7 fish) (I). Typical calcium response traces of each 5-HT subpopulation (J). Heatmap of the calcium responses of individual 5-HT neurons from one preparation (n = 37 neurons) (K). Unsupervised K-means classification of single-neuron responses in unique clusters (C1 to C2) (n = 269 neurons from seven fish) (L). Collectively, these results suggest that neurons within each SR subsystem form a functional module and that the three SR 5-HT neuronal systems, categorized by the projectome, transcriptome, and functionality play differential roles in regulating wake-sleep homeostasis, behavior motivation, and nonassociative learning. The 5-HT neuronal subsystems cooperatively accomplish a given task with each responsible for a specific subtask We demonstrated above that multiple parallel 5-HT neuronal subsystems are concurrently recruited, each with a distinct responsive pattern during a given sensorimotor task (Fig. 3), suggesting that a recruited subsystem might manage specific subtasks. To testify this hypothesis, we designed loss-of-function experiments to examine how single 5-HT neuronal subsystems contribute to the visuomotor behavior (Fig. 5A and fig. S8, A to C). As expected, laser ablation of the 5-HT neurons in the PT, SR-R, and MO subpopulations influenced different parameters of the induced visuomotor behavior. The absence of the PT 5-HT neuronal subpopulation specifically reduced the occurrence of locomotor events without affecting any other parameters (Fig. 5B), indicating that the PT subpopulation gates the transformation from the visual stimuli to locomotor events. Ablation of SR-R subpopulation decreased slightly the occurrence of locomotor events and tail beat amplitude while increasing tail beat frequency within each locomotor event (Fig. 5C), indicating that it might play a role in proprioception to balance between the intended and actual locomotor outcomes. MO ablation markedly changed all locomotor parameters (Fig. 5D), suggesting that it regulates the initiation and execution of locomotion, which is consistent with the finding in cats and rodents (56, 57). Sham ablation of non–5-HT cells in these four brain regions showed no effect on visuomotor behavior (fig. S8, D to G). These results demonstrate that while a single 5-HT neuronal subsystem is primarily responsible for a subtask, multiple subsystems coordinate and cooperate their activity to accomplish a given task. 5-HT neuronal subsystems cooperatively accomplish a specific task. (A) Experimental setup used to perform the visuomotor behavior test and the laser ablation of 5-HT neurons in PT, SR-R, or MO subpopulations using the tph2:GFP fish. (B to D) Typical recordings of tail movements during the visuomotor behavior test before and after laser ablation of 5-HT neurons in PT, SR-R, or MO subpopulation. Single locomotor events as indicated by the arrowheads were enlarged and shown in the dashed boxes to the right. Quantification of the important features of the induced locomotion: number of locomotor events in 3 min (each dot represents one fish), number of locomotor cycles per event, tail beat frequency, and amplitude within each locomotor event [n = 55 and 34 events from seven fish for (B), n = 71 and 49 events from eight fish for (C), n = 63 and 45 events from seven fish for (D)] [(B) to (D)]. (E to G) Schematic model of the SD device used together the laser ablation technique of 5-HT neurons in SR-R of the tph2:GFP fish (E). The sleep cycle duration was clearly affected after laser ablation (F). Quantification of the total sleep time, number of sleep bouts, and sleep bout length during night (G). Each dot represents one fish. (H to J) Illustration of optogenetic activation of 5-HT neurons in the SR-R subpopulation after paramecia supply in the food-deprived tph2:ChR2-mCherry fish (H). Typical traces show tail movement after paramecia supply with (top) or without (bottom) optogenetic blue-light stimulation for 60 s. One J-turn event (prey-like behavior) is enlarged (black dashed boxes) (I). Quantification of the number of J-turns in 60 s (n = 8 fish) (J). All data are presented as mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001, significant difference. ns, not significant. Each 5-HT neuronal subsystem exhibits a distinctive response pattern during different tasks (Figs. 3 and 4), implying that the same subsystem adjusts to play different roles under different tasks. We performed experiments to evaluate whether a 5-HT neuronal subsystem takes care of the same subtask or different subtasks under various circumstances. After SD, ablation of the SR-R subpopulation resulted in a condition where the fish had notably reduced rebound sleep in terms of decreased total sleep duration, reduced sleep bouts, and shorter periods of individual sleep bouts (Fig. 5, E to G, and fig. S8H), suggesting that this SR-R 5-HT subsystem is responsible for wake-sleep homeostasis. We found that the activity of neurons in SR-R subpopulation is inhibited by food supply after food deprivation (Fig. 4). To counteract the inhibition of the SR-R subpopulation underlying prey-like behaviors, we used the tph2:ChR2-mCherry fish line where blue light stimuli can be applied to precisely excite 5-HT neurons in the SR-R subpopulation with SR-M and SR-C unaffected (fig. S8, I to K). Switching the inhibition to excitation in 5-HT neurons of SR-R subsystem by using blue light stimuli reduced the prey-like behaviors (Fig. 5, H to J, and fig. S8, L and M), indicating a critical role of the SR-R 5-HT subsystem in controlling motivation for food intake. Solely optogenetic stimulation of SR-R 5-HT neurons did not elicit prey-like behaviors in zebrafish (fig. S8, N and O). Together with the revealed function of the SR-R subsystem in visuomotor behavior (Fig. 5C), these results suggest that a single 5-HT neuronal subsystem can play diverse roles across distinct circumstances. In summary, it is not just one 5-HT subsystem that is exclusively devoted to accomplish a specific physiological or behavioral task. Instead, a coordinated recruitment of multiple parallel 5-HT subsystems is required with each assigned a specific subtask. Each 5-HT subsystem functions as a separate module that adjusts its output in response to diverse physiological or behavioral processes. DISCUSSION The multidimensional diversity of whole-brain 5-HT neurons has hindered the characterization of 5-HT subsystems. There are two envisions as to how the 5-HT neuronal subsystems operate. The first one is that a single 5-HT subsystem has sole charge of one function. This can be interpreted to mean that one major 5-HT subsystem is active during a particular behavioral or physiological task. The second considers that the coordinated involvement of multiple 5-HT subsystems operating in parallel is required to accomplish a specific task and that any single 5-HT subsystem can be activated in diverse activity patterns depending on the task. Our study identified seven 5-HT neuronal subsystems across the whole brain by generating single-cell projectomic and transcriptomic atlases of the brain 5-HT neurons that characterize the spatial, morphological, and molecular signatures of these seven subsystems. Functional imaging of the whole-brain 5-HT neurons at single-cell resolution further confirmed these subsystems in terms of functionality and revealed the operating rules of the 5-HT subsystems during various behavioral and physiological processes. We found that the seven 5-HT subpopulations could display functional heterogeneity and neurons within each subsystem were activated in a modular-organized manner depending on the task at hand. Our results support the hypothesis that a single 5-HT subsystem directs a specific subtask and that coordinated recruitment of multiple parallel 5-HT subsystems is necessary to complete a given behavioral or physiological task. We generated and subsequently integrated single-cell projectomic and transcriptomic atlases of the brain and analyzed the functional subpopulations of brain 5-HT neurons during various tasks. Using an unsupervised clustering strategy based on the single-neuron projectomic dataset, which provides soma location and the brain-wide distribution of the main neurites (axons, axon collaterals, and dendrites), characterized whole-brain 5-HT neurons into seven distinct subpopulations. The manually collected scRNA-seq transcriptome and 10× scRNA-seq dataset describes not only the transcriptome but also the spatial location of the seven identified 5-HT subpopulations. Single-neuron transcriptome analysis verified that the projectome-defined subpopulations were also described by the expression of distinct molecular markers, especially neuron projection–related genes. The integration of the soma location, axon projection, gene expression, and neurotransmitter cotransmission of 5-HT subpopulations promoted the targeting analysis of functional organization of 5-HT neuronal subsystems. In mice, the diversity of 5-HT neuronal system has been investigated previously by considering anatomical domains of somata, developmental cell lineages, and single-cell transcriptome (8, 9, 12, 26, 40, 58–60). On the basis of the exclusive expression pattern of a series genes homologous to 5-HT subpopulation markers identified in mice (7–9, 12), we suggested putatively related anatomical populations in mammalian system for these seven zebrafish subpopulations (fig. S5 and table S1). For example, the SR-R higher expression of gad1b, gad2, and pax5 was found to be similar to that expressed by DR 5-HT neurons in the En1-Pet1 and Drd2-Pet1 mice. Both SR-M and SR-C are enriched for drd2a/b, en1b, and slc17a8 (vglut3). While the different expression of piezo2 and oxtr in SR-M and SR-C indicates that they are homologous to mouse DR (vm) Drd1-Pet1 and DR Drd2-Pet1 5-HT neurons, respectively. The expression of meis1b, meis2a, hoxd3a, and nkx6.1 in the zebrafish IR-C are remarkably similar to the genetic profile found in the mouse ROb-RPa (nucleus raphe obscurus-nucleus raphe pallidus). The IR-R in zebrafish is similar to the RMg (nucleus raphe magnus) in mice that both express hoxa2b and nkx2.2a, and the MO is homologous to ROb-RPa in that both highly express hoxc4a, hoxd3a, meis1b, and meis2a. 5-HT neurons were not identified in the pretectal region in mammal (29, 61–65), but the PT subpopulation was revealed in our results to be potentially homologous to the MR with high expression of penkb, trcr3a, and nr2f2. It revealed that some 5-HT neuronal modules in zebrafish correspond to multiple populations in mice, and multiple modules in zebrafish may map to the same population in mammal. This lack of one-to-one correspondence likely reflects both phylogenetic divergence and developmental stage differences (larval zebrafish versus adult mouse). Last, we also reported distinct coreleasing neurotransmitters (including glutamate, GABA, and other neuropeptides) for different 5-HT subpopulations consistent with findings in mammals (table S1) (7, 49, 66–70). The multiplane two-photon calcium imaging at a single-neuron level enables us to characterize response properties of nearly 3000 5-HT neurons in zebrafish during a wide range of behavioral tasks, including visuomotor behavior, food exposure, spontaneous locomotion, sleep-wake homeostasis, behavior motivation, and acoustic startle habituation. The large-scale analysis with combination of single-neuron activity, the subpopulation response pattern, as well as various behavioral or physiological paradigms enabled us to uncover the principles governing the operation of 5-HT neuronal system. Neurons within a defined subpopulation display a homogeneous activity pattern, while neurons across subpopulations exhibit diverse response features. This intersubpopulation heterogeneity was also reported in previous studies under similar tasks (41). Neurons from the same subpopulation show a homogeneous axon projection pattern and innervate the same brain regions. The functional diversity among the neurons belonging to distinct subpopulations can be inferred from their different projectomic features. Although 5-HT subsystems may innervate a few brain regions in common, their functionality is determined largely by their projections to other brain regions. For example, while the three SR subpopulations largely innervate the pallium (homologous to the mammalian cortex), it is the distinct projections to many other brain regions that are crucial for their functional diversity. The SR-R 5-HT subsystem sends an intense axon projection to the hypothalamus and preoptic area, which matches with its role in wake-sleep homeostasis. The SR-M 5-HT subsystem innervates the locus coeruleus and interpeduncular nucleus and exerts an effect in nonassociative learning, and lastly, the SR-C subsystem projects their axons to the tegmentum and thalamus that are responsible for food rewards. It should be noted that there are brain regions receiving overlapping projections from almost all the 5-HT subsystems. This does not necessarily mean that these subsystems share a common function related to this brain region but rather that they modulate the output of the common innervated brain regions. As shown in the present study, all neurons from six subpopulations (except for the SR-C) project to the hindbrain, a locomotion-related brain region. However, the activity patterns of the six subpopulations are notably diverse with differences in amplitude and timing during different sensorimotor behaviors, indicating that they contribute in different ways to sensory perception, sensorimotor integration, locomotion initiation, or sensory feedback. This scenario would explain the previous seemingly conflicting results showing the different function of 5-HT neuronal system during a unique behavior paradigm. 5-HT immunostaining revealed extensive thin axonal arborizations throughout target brain regions. Because of inherent technique constrains, the sparse labeling–based morphological reconstruction only allowed us to manually trace the primary structure of each neurite in the single 5-HT neurons. The finest fibers exhibiting extremely delicate structures and lacking direct connection to the primary branches remained largely untraceable. In this study, we modestly enhanced the fluorescent signal of labeled fine axonal arbors and manually traced those that were substantially connected to the main neurite branches. Nonetheless, the very fine axonal arbors that were disconnected from the main branches were still excluded from the reconstruction dataset to ensure methodological rigor. This conservative approach may have resulted in insufficient defined boundary of certain axonal innervations. Consequently, a discrepancy persists between 5-HT immunostaining and single-cell projectomic dataset with respect to arborization density in target regions. The approach modestly enhancing the fluorescence enabled us to achieve higher resolution of 5-HT axonal distribution, particularly in the optic tectum and hindbrain. But the signal enhancement and manual tracing processes may introduce observer bias, potentially limiting the absolute accuracy of fine axon reconstruction. In addition, the transient expression methods used may have biased our sampling toward certain subpopulations, potentially limiting our ability to capture the complete projectome. Given these technical limitations, our dataset primarily characterizes the major projection patterns of 5-HT neurons. A more comprehensive mapping of serotonergic innervation will require future studies using more advanced labeling techniques, precise tracing tools, and fully automated reconstruction pipelines. In summary, we report a resource that identifies 5-HT subsystems across the whole brain in zebrafish. This classification scheme shows interspecies conservation of the 5-HT neuronal system in both zebrafish and mice and broadens our strategic approach to the development of treatments for serotonin-related diseases. MATERIALS AND METHODS Animals Zebrafish (Danio rerio) were raised and maintained in an automatic fish-housing system (with water temperature of 25° to 28°C) under a 14-hour light and 10-hour dark cycle at the Tongji University according to established procedures. Wild-type and transgenic lines including Tg(tph2:GFP) (30), Tg(HuC:H2B-GCaMP6f) (71) (abbreviated as HuC:H2B-GCaMP6f), Tg(vmat2:GFP) (72) (abbreviated as vmat2:GFP), Tg(tph2:Gal4;UAS:mCherry) (42) (abbreviated as tph2:mCherry), Tg(tph2:Gal4;UAS:ChR2(H134R)-mCherry) (42) (abbreviated as tph2:ChR2-mCherry), Tg(tph2:Gal4;UAS:GCaMP6) (31, 42) (abbreviated as tph2:GCaMP6), and Tg(olig2:DsRed2) (73) (abbreviated as olig2:DsRed2) at 6 to 7 dpf were used in this study. All experimental protocols were approved by the Animal Use Committee of Tongji University (TJAA00621B04). Care was taken throughout the experiments to reduce the number of animals used and to minimize distress. Embryonic microinjections To perform genetic sparse labeling of the single 5-HT neurons, a pzTol2 tph2-Gal4/VP16 construct containing a 3408-bp fragment from the tryptophan hydroxylase 2 (tph2) promoter (6) (Vector Builder) and membrane targeted miniTol2-14×UAS:tdTomato-CAAX construct were coinjected at a concentration of 10 to 20 ng/μl with Tol2 transposase mRNA (10 ng/μl) into the HuC:H2B-GCaMP6f transgenic embryos at the two to four-cell stage. Confocal live imaging For visualization of 5-HT neurons in the whole brain, the tph2:GFP and the tph2:GCaMP6 fish line at 7 dpf were paralyzed using pancuronium dibromide (1 mM, Sigma-Aldrich) and embedded in 2% low-melting point agarose (Sigma-Aldrich) in a custom-made chamber. For characterization of 5-HT neural projectome, the HuC:H2B-GCaMP6f larvae with sparse labeling of a single neuron were preselected at 5 dpf and prepared for imaging at 6 to 7 dpf as described above. Confocal imaging was carried out using an Olympus Fluoview 3000 confocal microscope (FV3000, Olympus) under a 40× [numerical aperture (NA), 0.8] or 20× (NA, 1.0) water-immersion objective with z-stack steps at 1 μm using a laser scanning confocal microscope (FV3000, Olympus). 3D reconstruction and quantification of whole-brain 5-HT neurons The three-dimensional (3D) reconstruction of single 5-HT neuron across the whole brain was performed on the basis of confocal z-stack images of the tph2:GFP fish line at 7 dpf, which were rendered using Imaris v9.8 (OXFORD). First, the soma surface of 5-HT neurons was reconstructed on the basis of the GFP signals using the following custom settings: surface detail of 2.00 μm (smooth) and thresholding background subtraction (Local Contract). Then, tiff files of all brain regions from the web interface of mapzebrain (https://mapzebrain.org/downloads) were downloaded and merged in one channel using Fiji v1.53c (National Institutes of Health) (74) to generate a whole-brain region, which was used as template for 3D reconstruction, after the surface reconstruction performed by Imaris. Then, the confocal brain volume was registered to this brain template, and four brain regions (35) including PT, SR, IR, and MO were surface-rendered and displayed in different colors with transparency. The numbers of 5-HT neurons in the four major brain regions were manually quantified on confocal images using Fiji (N = 6 fish). All the images were subjected to threshold adjustment such that the nonneuron background was reduced to zero leaving only fluorescent signals originating from cells. Manual neuron tracing and image registration to standard brain template The HuC:H2B-GCaMP6f, a marker present in the common coordinate framework, was used as a reference channel of original confocal stacks and aligned to the Z-brain reference atlas, a standard brain template (https://mapzebrain.org/downloads), using the Computational Morphometry Toolkit plugin in Fiji in an iOS environment. Registration parameters were set to Cachero, Ostrovsky2010, and a nonrigid alignment was performed using the command string (-X 26 -C 8 -G 80 -R 4 -A “--accuracy 0.4” -W “--accuracy 0.4”) (34). A reformatted anatomy was resized to 646 (pixels) in width, 1054 (pixels) in height, and 359 (planes) in depth, and a registration text file containing affine transformation commands was computed after alignment preprocessing. To perform the projectome characterization of 5-HT neurons, the raw image stack containing the single neuron with nonoverlapping projections was subjected to manual tracing based on tdTomato signals. The soma location was marked, while axon projections, collaterals, and dendrite morphologies were semiautomatically reconstructed along each branch point using the Simple Neurite Tracer plugin in Fiji and saved as individual SWC files. For the fine axon arborizations of 5-HT neurons, additional manual tracing was performed by enhancing the fluorescent signal intensity of labeled images by 1.5- to 2.5-fold to capture as many fine axonal arbors as possible. To ensure data rigor, projections that were still disconnected from the main branches were excluded from the reconstruction. In total, 175 fish were scanned for a total of 188 tracings. Transformation commands were applied to the SWC files containing single-neuron tracing based on the registration text file before being coordinated to the standard brain template. The 3D reconstruction and visualization of 5-HT neuron morphologies were performed using 3D Slicer v5.6.1 (https://slicer.org/). One hundred eighty-eight neurons from different subpopulations were color-coded and displayed altogether, separately as groups or a single neuron. Both dorsal and lateral views were presented. The 3D surface models of four major brain regions (PT, SR, IR, and MO) and Mauthner cell soma (as a landmark) were created from segmented image data from the mapzebrain in 3D slicer and displayed in different colors. Neuronal clustering based on the single-cell projectomic dataset For neuronal clustering, we measured a series of parameters of the individual 5-HT neuron morphology. First, the soma location of each neuron was determined by measuring x, y, and z coordinates within a 3D space. To standardize these coordinates, we adjusted them to a unified hemisphere, ensuring a consistent reference frame across all measurements. Then, the distribution of axon, including their collateral branches, and dendrite within 38 different brain regions/nuclei was counted on the basis of reference brain segmentations and browsed in 3D Slicer (https://slicer.org/). In addition, the distance from projection terminal to the central part of Mauthner cell soma was measured. Last, the unsupervised clustering of all 5-HT neurons was performed on the basis of all these 42 parameters above with K-means and visualized with t-SNE using a Python script. Anatomical annotation of the seven 5-HT neuron subpopulations was performed according to the corresponding somatic location. To differentiate the projection patterns between subpopulations, the percentage of neurite distribution in different brain regions was calculated as the number of neurons with neurite distribution in a certain brain region divided by the total neuron number for each subpopulation. Heatmap of the projection distribution percentage covering 38 brain regions was plotted using Prism 9.0 (GraphPad). Single-cell collection and RNA-seq For transcriptome analysis of different 5-HT neuronal subpopulations, the single-cell collection was conducted using the tph2:GFP line at 7 dpf. Fish were fixed in a recording chamber containing extracellular solution containing 134 mM NaCl, 2.9 mM KCl, 2.1 mM CaCl2, 1.2 mM MgCl2, 10 mM Hepes, and 10 mM glucose (pH 7.8, 290 mOsm). The somata of 5-HT neurons from seven subpopulations (PT, SR-R, SR-M, SR-C, IR-R, IR-C, and MO) were visualized under a fluorescence microscopy, then targeted, and collected separately using glass electrodes. In particular, the boundaries of three SR subpopulations were determined and labeled using the Qimage software, and then the most rostral and caudal neurons were collected for SR-R and SR-C samples, respectively, to avoid the neurons around the boundary. To minimize the unspecific contamination from other cell types or surrounding tissue, a single cell was collected at a time. Electrodes with a small tip opening, approximately two-thirds the diameter of a single cell, were used. First, the electrode was filled with extracellular solution and moved to the target neurons using a motorized micromanipulator (Scientifica). Then, a negative pressure was applied to draw a single cell into the pipette, and the cell integrity and specificity were assessed under both bright field and fluorescence microscopy. Any samples that did not exhibit clear cell boundaries (intact cell membranes) or GFP signals were discarded to ensure the quality of the sample collection. For each sampling, a total of 10 to 12 neurons were collected one by one and transferred to lysis buffer solution. Total RNA was extracted for construction of cDNA libraries, involving fragmentation, polymerase chain reaction enrichment, N7 index linking, and purification. A total of 71 QuantSeq libraries (N = 9 to 12 samples for each subpopulation) were prepared using the TruePrep DNA Library Prep Kit V2 for Illumina (Vazyme) according to the manufacturer’s instructions. Sequencing was carried out using Illumina NovaSeq 6000 with 150-bp paired-end reads. Gene expression level analysis was performed using the R toolkit Seurat (version 5.0.2), and the mean gene expression matrix of all samples was obtained after processing by the NormalizeData function. Distances between samples were calculated on the basis of the correlation of all genes using the dist function, and hierarchical clustering was performed on the distance matrix using the hclust function. Subsequently, the dendrogram of the clustering results was plotted using the R package ggtree. Three thousand highly variable genes were selected for PCA dimensionality reduction, and then the result was visualized with t-SNE. For marker gene analysis of each subpopulation, the FindAllMarkers function in Seurat was used, with parameters set as min.pct = 0.1 and logfc.threshold = 0.25. The expression of genes was sorted by avg_log2FC, and a heatmap of the top 20 genes list was plotted. Dot plot was performed for 3 of top 30 marker genes of each cluster shared by manually collected scRNA-seq and 10× scRNA-seq data. GO analysis was performed on 3000 highly variable genes, and dot plot was performed for top five marker genes related to neuron development, axon projection, neurotransmitter and receptor, synapse formation, and ion channels, separately. To compare the seven 5-HT neuron subpopulations identified from our manually collected scRNA-seq data with subtype classifications in mice, we used the reported marker genes from three scRNA-seq studies on mouse 5-HT neuron subtypes as a reference (8, 9, 12). We analyzed the relative expression pattern of all gene homologs in our data compared to the 5-HT subtype markers identified in mice. Some of these homologs showed subpopulation-specific expression patterns. Then, a heatmap of genes homologous to 5-HT subtype markers identified in mice was generated using R, suggesting cross-species homology accordingly. 10× single-cell RNA-seq Whole-brain tissues of the vmat2:GFP fish line at 7 dpf were dissociated using papain solution [papain (28 U/ml), 1% deoxyribonuclease, and l-cysteine in Dulbecco’s Modified Eagle’s Medium/F12 (12 mg/ml)] and incubated at 37°C for 15 min with mixing. Dissociated cells were washed twice with washing buffer [100 ml: 650 μl of 45% glucose, 500 μl of 1 M Hepes, and 5 ml of fetal bovine serum into 93.85 ml of 1× Dulbecco’s phosphate-buffered saline (DPBS), sterilized by passage through a 0.22-μm pore size filter] and filtered with 40-μm cell strainers (BD Falcon) into washing buffer (72). Subsequently, FACS was performed using a Beckman Coulter Moflo Astrios Cell Sorter (Beckman Coulter). All sampling was carried out using a 10× Genomics Chromium Single Cell Kit (version 2). Downstream cDNA synthesis, library preparation, and sequencing were carried out according to the manufacturer’s instructions (https://10xgenomics.com/solutions/single-cell/). Sequencing data were converted to a matrix of expression counts using Cell Ranger (version 2.0.1). Transcriptome libraries were mapped to zebrafish genome assembly (GRCz11). After removing cells with low unique molecular identifiers (UMIs) (<600) and molecule-to-gene ratio (<1.2), as well as genes detected in fewer than 20 cells or more than 60% of all cells, the remaining cells with 5-HT marker gene expression (tph2+) were selected for clustering analysis. In total, 1489 tph2-expressed cells were obtained, the top 10 principal components were used for clustering, and marker genes for each cluster were calculated using the FindAllMarkers function in Seurat (parameters set as min.pct = 0.1 and min.diff.pct = 0.25). Cluster annotation was performed using the top 20 marker genes shared by manually collected scRNA-seq and 10× scRNA-seq data. In addition, FeaturePlot was performed for one of the shared markers of each subpopulation. In addition, GO analysis was performed on the 10× scRNA-seq marker genes using R package clusterProfiler, and the top 10 enriched GO terms for each category (Biological Process, Cellular Component, and Molecular Function) were presented in dot plot. HCR-based RNA fluorescence For HCR RNA FISH staining, the tph2:GFP fish larvae at 6 dpf were anaesthetized with 0.03% tricaine methane sulfonate (MS-222, Sigma-Aldrich) solution, euthanized, and fixed with 4% paraformaldehyde (PFA) in DPBS overnight at 4°C with gentle shaking. All reagents, including probe sets (tcf7l2, en2b, cart4, msi2a, nkx6.1, mab21l1, and gad1b), amplifier, and hairpin were purchased from Molecular Instruments (Los Angeles, CA, USA), and the staining was performed as previously described (75). Briefly, after wash with DPBST (1× DPBS + 0.1% Tween 20), samples were treated with ice-cold 100% methanol for dehydration and permeabilization. Then, samples were rehydrated before conducting hybridization with probe solution overnight at 37°C. The preamplification was performed using amplification buffer for 30 min at room temperature after washing with wash buffer and Saline-Sodium Citrate with Tween-20 (SSCT). Samples were then incubated in the hairpin solution containing the h1 and h2 hairpins for 12 to 16 hours in the dark environment at room temperature. Excess hairpins were washed using 5× SSCT at room temperature the following day. The zebrafish larvae were then mounted in agarose for imaging. Immunohistochemistry The tph2:GFP, tph2:ChR2-mCherry, or HuC:H2B-GCaMP6f fish at 7 dpf were anesthetized, euthanized, and fixed for 24 hours with 4% PFA, and then the brain was dissected out and washed three times each for 5 min with 1× PBS and permeabilized in 1× PBS containing 1% Triton X-100 (PBST) for 2 hours. After blocking with 5% bovine serum albumin in PBST at RT for 1 hour, the brain was then incubated with anti-GFP (1:2000; Abcam, ab1218), antiserotonin (1:2000; Sigma-Aldrich, S5545), or anti-calretinin (1:1000; Swant, CR7697) for 48 hours at 4°C. After rinsing with 1× PBS, the brain was incubated with secondary antibodies at 4°C overnight. The secondary antibodies were as follows: Alexa Fluor 647 donkey anti-mouse (1:500; Invitrogen A-31571), Alexa Fluor 488 donkey anti-rabbit (1:200; the Jackson Laboratory 711-545-152), and Cy5 donkey anti-rabbit (1:200; the Jackson Laboratory, 711-175-152). After washing with PBS, the tissue was placed dorsal side up on a slide and mounted with the antifading fluorescent mounting medium (Vectorlabs) for imaging using a laser scanning confocal microscope (FV3000, Olympus). Behavioral setup and two-photon calcium imaging To perform two-photon calcium imaging, the tph2:GCaMP6 fish line at 7 dpf was used. Zebrafish larvae were embedded using 2% low-melting point agarose in a custom-made chamber in a dorsal side up position, with head fixed and tail freed. Calcium imaging of 5-HT neuron responses was performed using a two-photon (900 nm) laser scanning microscope (Scientifica). Five planes of whole-brain images were captured using raster scanning pattern with a 16× water-immersion objective (NA, 0.80; Olympus) at a scan rate of 3.84 frames/s (1024 pixels by 1024 pixels). For examining the SR 5-HT neurons, single-plane raster scanning mode was used under a 40× water-immersion objective (NA, 0.8; Olympus) at a scan rate of 15.48 frames/s (1024 pixels by 1024 pixels). Frames were acquired using SciScan 1.3 software (Scientifica). Tail movements were simultaneously monitored under infrared (850 nm) illumination at 200 frames per second (fps) with a high-speed camera (MV-CH040-A0CM camera, HIK ROBOT), and tail coordinates were extracted using the free Python-based software ZebraZoom (76). Tail movement trajectories were analyzed and plotted using ZebraZoom tracking software. For simultaneous EMG recording (motor nerve recording), the tail skin of fish was carefully removed, and extracellular solution was continuously perfused. The electrode filled with extracellular solution was gently put on the top of the skin-free region. EMG signals were acquired with a MultiClamp 700B amplifier (Molecular Devices), low pass filtered at 10 kHz, digitized with a DigiData 1550B, and analyzed using Clampfit (Molecular Devices) software. The calcium responses of 5-HT neuron subpopulations under a wide range of behavioral tasks, including visual stimulation, food exposure, spontaneous movement, electric shock, SD, and nonassociative learning, were evaluated with the general protocol described below. Visual stimulation Head-fixed and tail-free fish at 7 dpf was placed in a virtual reality environment in which a diffusive screen showing white and black moving gratings (each 4 mm thick) was placed in front of the fish, with the distance between the fish and the display being 1 cm. Fictive forward swimming can be induced by optic flow aligned from side to center at a speed of 4 mm/s. The whole-brain calcium imaging and tail tracking were performed simultaneously. Tail movement was recorded with a high-speed camera (MV-CH040-A0CM camera, HIK ROBOT). The tail movement trajectory of the fictive locomotion was analyzed using ZebraZoom (76). Tail angle was defined as the angle between the body axis of the fish and the tip of the tail. Food exposure Before testing, zebrafish larvae were allowed to feed on paramecia from 5 to 6 dpf and then fasted for 48 hours from 6 to 8 dpf. Food exposure test was performed at 8 dpf as previously described (32). The fish was embedded in agarose with the eyes and nostrils carefully exposed. After a 10-min acclimatization, paramecia suspension (100 to 200 μl) or water alone (control group) was introduced beneath the medium via careful pipetting to avoid behavioral disturbance. Whole-brain calcium imaging and detailed examination of SR activities were performed in conjunction with simultaneous tail tracking and tail angle analysis. To determine the prey capture behavior, the tail movement trajectory was analyzed using ZebraZoom (76), and the maximum tail angle was measured for each identified swim bouts. Then, the prey capture–associated J-turns were defined on the basis of previous studies (77, 78). Electric shock The zebrafish larvae at 7 dpf were placed in the chamber and embedded in agarose with tail free. After 10 min of acclimatization, a single pulse of electrical shock for 50 ms with an initial intensity of 0.1 mA was delivered (42). The current was delivered by placing an anode and a cathode near the left and right sides of the fish, respectively. The whole-brain calcium imaging and tail tracking were performed simultaneously as described above. Spontaneous locomotion The zebrafish larvae were randomly selected and mounted in the chamber. After 10 min of acclimatization, whole-brain calcium imaging was performed while spontaneous tail movement (free swimming activity independent of any stimuli) was monitored by a high-speed camera. Sleep deprivation SD was performed on 6 dpf zebrafish larvae for 24 hours using the OMR treadmill as previously described (79). Fish were placed in a rectangular tank filled with system water. A monitor displays either a moving (1.1 cm/s) or nonmoving grating pattern (control), consisting of black-white stripes (1.1 cm in width). To induce persistent swimming, the moving direction was changed every 35 s. The fish were kept in a constant ambient condition with maintained daytime level of illumination. Also, the water temperature was monitored, and the fish were normally fed to keep healthy. Exposure of fish to 24-hour OMR SD was from 22:00. For sleep/wake behavior assay, fish of each condition were placed into individual wells of a custom-made multiwell plate, and their locomotor activity was traced for the next 24 hours from 22:00. Locomotor activity was monitored using a monochrome camera (10 FPS, Daheng Imaging, MER-231-413U3M) fitted with a variable-focus lens and infrared filter that was housed inside a custom-made sound-proof box. The fish were housed continuously, illuminated with infrared lights and illuminated with white light from 8:00 a.m. to 10:00 p.m. Video tracking and data analysis were performed using ZebraZoom (76). The parameters used for movement detection were a detection threshold of 0.15 cm/s and a bin size of 60 s. For calcium imaging, the fish were subjected to 30-min sleep recovery in the dark, and then two-photon imaging of 5-HT neurons in SR and EMG recording was performed. Acoustic startle response habituation For nonassociative learning analysis, head-fixed and tail-free fish at 7 dpf was subjected to acoustic startle response habituation assay by performing repetitive sound stimulation as previously described (55). After 10 min of acclimatization, an acoustic stimulus (400 Hz) was triggered 40 times with interstimulus interval of 10 s controlled by a Direct Digital Synthesis (DDS)-based arbitrary waveform generator (FY6800, FeelTech) connected to the amplifier and delivered by an 8-W loudspeaker. The experimental setup was synchronized and controlled with a DigiData 1550B. The whole-brain calcium imaging and tail tracking were performed simultaneously as described above. The escape-associated C-bend behavior was detected on the basis of tail angle using ZebraZoom (76). Also, P startle response for every 10 stimuli was calculated as the probability of startle response. Two-photon calcium imaging analysis Acquired confocal images of calcium responses were analyzed by Fiji and a MATLAB (MathWorks) script (80). Neural dynamics of the 5-HT neuronal subpopulations was analyzed in volumetric images of five planes encompassing the whole brain. The image correction for motion artifacts was performed using the Template Matching plugin in Fiji. The individual region of interests (ROIs) was drawn manually on the average image calculated from the entire series. Each ROI represented one specific brain region (the PT, SR-R, SR-M, SR-C, IR-C, and MO) or a single 5-HT neuron. ΔF/F values of each ROI were calculated from raw traces after normalization to the baseline levels. For calcium responses in different brain subregions, PCA was performed using a Python script (N = 6 to 8 fish). The sample traces of calcium activity dynamics shown for each brain region were subjected to Gaussian smoothing using a MATLAB script (80). For the single-neuron analysis, a heatmap was plotted using raw traces of calcium responses from one fish. In addition, unsupervised clustering analysis was conducted for single-neuron responses during visual stimulation, food exposure, spontaneous movement, and SD using raw traces lasting 30 to 90 s, covering the first response after 5-s baseline, by K-means. To analyze nonassociative learning, traces of the whole recording period (300 to 400 s) were used. Subsequently, t-SNE dimensionality reduction was conducted for visualization. After unsupervised clustering of calcium signal traces of more than 400 single neurons, each trace was annotated to a specific anatomical subpopulation (PT, SR-R, SR-M, SR-C, IR-C, and MO) according to the soma location information. Then, the response loyalty was calculated as the percentage of neurons within each subpopulation whose response aligned with the response of the respective subpopulation to which they belonged. The 3D reconstruction of the soma locations in the four major brain regions was performed after normalization as described above, and the extent of response loyalty is indicated by color coding. Optogenetic stimulation For optogenetic activation of the SR-R neurons, the tph2:ChR2-mCherry fish at 7 dpf were used, and the tph2:mCherry fish were used as negative control (42). The zebrafish larvae were mounted in a custom-made chamber, with head fixed and tail freed and subjected to food exposure task as described above. For the SR-R stimulation, Polygon 1000 photostimulation device (Mightex) was used to control illuminated region being specifically limited to the SR-R. The 470-nm blue light with a power of 15 mW was triggered by a DigiData 1550B for 1 min, and tail tracking and tail angle analysis were simultaneously performed. As described above, the tail movement trajectory was analyzed using ZebraZoom (76), and the prey capture–associated J-turns were defined on the basis of previous studies (77, 78). The number of J-turn events within 1 min was counted manually. Extracellular recordings were performed on three SR subpopulations of 5-HT neurons to evaluate the efficiency and precision of activation. In vitro preparation of the tph2:mCherry fish were obtained as mentioned in single-cell collection. The recording electrode was filled with extracellular solution and advanced into the brain through a motorized micromanipulator (Scientifica). Signals were recorded in current-clamp gap-free mode, detected by a MultiClamp 700B amplifier (Molecular Devices) and a 1550B Digidata data acquisition system (Molecular Devices) with the low-pass filtered at 10 kHz. Laser ablation of neurons and behavioral analysis For laser ablation of specific 5-HT neuronal subpopulations, the tph2:GFP fish at 7 dpf were anesthetized with 0.03% MS-222 and then embedded in 2% low-melting point agarose with dorsal side up in a custom-made chamber. The olig2:DsRed2 fish were used for sham ablation. A two-photon microscope (Scientifica) equipped with IR-differential interference contrast optics and a charge-coupled device camera with image grabber (QIMAGING) was used for targeted ablation of GFP+ 5-HT neurons. Ablation was performed using a single laser pulse (wavelength: 900 nm, 340 mW, 1 to 5 ms) operating at an 80-MHz repetition rate with a <140-fs pulse duration focused through a 16× objective (NA, 0.8; Olympus). Targeted ablation of 5-HT neurons in PT, SR-R, or MO subpopulations or non–5-HT cells in these four brain regions was performed. Successful ablation was confirmed by two-photon images showing a development of tissue scarring (cavitation and obvious tissue distortion). Any ablation that resulted in severe damage was excluded. Fish were then released and allowed to recover in an aquarium at room temperature for 2 hours after which behavioral tests were performed. A successful ablation was confirmed after the behavioral tests were completed. Visuomotor behavior analysis was performed as described above. The tail movement trajectory of the fictive locomotion was analyzed using ZebraZoom (76). The number of locomotion occurrence within 3 min and the event number within each locomotion event were manually counted. Analysis of tail beat frequency and amplitude were performed using ZebraZoom (76). For sleep/wake analysis, SD was performed for 24 hours as described above before neuronal ablation. Then, the fish of each condition were individually tracked for 24 hours on a 14:10 light:dark cycle to assess sleep/wake behavior. Statistical analysis The calcium response curves and heatmaps were generated by MATLAB R2020a (MathWorks). Photoshop (Adobe) and Fiji were used to process images. Schematic diagrams were generated and modified in Adobe Illustrator. Unsupervised clustering analysis was generated using K-means and visualized by t-SNE in Python (v.3.7.9). The PCA was constructed in Python (v.3.7.9). Clampex v10.7.0.3 and Clampfit v10.6 (Molecular Devices) software was used for collection and analysis of electrophysiological signals. Behavior tracking of zebrafish was carried out using the Python-based free software ZebraZoom v1.1783 (76). All significant statistical results are indicated in the figures following the conventions: *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. The results were presented as means ± SEM, and statistical analysis was performed using Prism 9.0 (GraphPad software). In particular, for the analysis of locomotor events in the subpopulation ablation experiments and for the analysis of J-turn events during optogenetic activation of SR-R, we used paired t tests. For other two-group comparisons, statistical significance was determined by unpaired t tests.