Serotonin Neurons in the Dorsal and Medial Raphe Nuclei: from Single-Cell Transcriptomes to Whole-Brain Projections

Serotonin neurons of the dorsal and medial raphe nuclei (DR and MR) collectively innervate the entire forebrain and midbrain, modulating diverse physiology and behavior. To gain a fundamental understanding of their molecular heterogeneity, we used plate-based single-cell RNA-sequencing to generate a comprehensive dataset comprising eleven transcriptomically distinct serotonin neuron clusters. Systematic in situ hybridization mapped specific clusters to the principal DR, caudal DR, or MR. These transcriptomic clusters differentially express a rich repertoire of neuropeptides, receptors, ion channels, and transcription factors. We generated novel intersectional viral-genetic tools to access specific subpopulations. Whole-brain axonal projection mapping revealed that DR serotonin neurons co-expressing vesicular glutamate transporter-3 preferentially innervate the cortex, whereas those co-expressing thyrotropin-releasing hormone innervate subcortical regions in particular the hypothalamus. Reconstruction of 50 individual DR serotonin neurons revealed segregated axonal projection patterns at the single-cell level. Together, these results provide a molecular foundation of the heterogenous serotonin neuronal phenotypes.

2 groups B6 and B7) and median raphe (MR; groups B5 and B8) nucleus, and provide ascending 49 innervation to the forebrain and midbrain. The DR and MR serotonin systems have been linked with 50 the regulation of many mental states and processes, including anxiety, mood, impulsivity, aggression, 51 learning, reward, social interaction, and hence remain the focus of intense research. 52 Evidence has suggested that the DR and MR serotonin systems differ in developmental origin, 53 connectivity, physiology, and behavioral function (Calizo et al., 2011;Okaty et al., 2019). DR 54 serotonin neurons derive entirely from rhombomere 1 of the developing mouse brain, whereas MR 55 serotonin neurons derive predominantly from rhombomeres 1, 2, and 3 (Bang et al., 2012;Jensen et 56 al., 2008). Although the DR and MR receive similar inputs globally from specific brain regions 57 (Ogawa et al., 2014;Pollak Dorocic et al., 2014;Weissbourd et al., 2014), they project to largely 58 complementary forebrain targets. The MR serotonin neurons project to structures near the midline, 59 whereas the DR serotonin neurons target more lateral regions (Jacobs and Azmitia, 1992). Slice 60 physiology recording showed that the serotonin neurons in the MR and DR have 61 different electrophysiological characteristics, such as resting potential, resistance, and reaction to 62 serotonin receptor-1A agonist (Calizo et al., 2011). Finally, activation of these two raphe nuclei has 63 been suggested to mediate opposing roles in emotional regulation (Teissier et al., 2015). 64 Even within the MR or DR, there is considerable heterogeneity of serotonin neurons in multiple 65 aspects. Although MR serotonin neurons arising from different cell lineages are anatomically mixed 66 in the adult, they have distinct electrophysiological properties (Okaty et al., 2015) and potentially 67 distinct behavioral functions (Kim et al., 2009;Okaty et al., 2015). Diversity of serotonin neurons in 68 the DR has received particular attention in recent years. Accumulating evidence indicates that there 69 are subgroup-specific projection patterns within the DR serotonin system (Niederkofler et al., 2016;70 Ren et al., 2018). The electrophysiological properties of DR serotonin neurons vary according to the 71 projection patterns (Fernandez et al., 2016). Physiological recordings as well as optogenetic and 72 chemogenetic manipulations suggest heterogeneity of DR serotonin neurons in their behavioral 73 functions (Cohen et al., 2015;Marcinkiewcz et al., 2016;Niederkofler et al., 2016). As a specific 74 example, we recently found that DR serotonin neurons that project to frontal cortex and amygdala 75 constitute two sub-systems with distinct cell body locations, axonal collateralization patterns, biased 76 inputs, physiological response properties, and behavioral functions (Ren et al., 2018). Our 77 collateralization analyses also imply that there must be additional parallel sub-systems of DR serotonin 78 neurons that project to brain regions not visited by the frontal cortex-and amygdala-projecting sub-79 systems. 80 Ultimately, the heterogeneity of DR and MR serotonin neurons must be reflected at the 81 molecular level. Pioneering work has introduced the molecular diversity of serotonin neurons across 82 the midbrain and hindbrain (Okaty et al., 2015;Spaethling et al., 2014;Wylie et al., 2010), yet 83 systematic analysis and integration of multiple cellular characteristics at the single-cell resolution 84 within each raphe nucleus is still lacking. The rapid development of single-cell RNA sequencing 85 (scRNA-seq) technology in recent years has provided a powerful tool for unbiased identification of 86 transcriptomic cell types in the brain (Darmanis et al., 2015;Li et al., 2017a;Mickelsen et al., 2019;87 Rosenberg et al., 2018;Saunders et al., 2018;Tasic et al., 2016;Tasic et al., 2018;Welch et al., 2019;88 Zeisel et al., 2018;Zeisel et al., 2015). In neural circuits where cell types have been well studied by 89 anatomical and physiological methods, there is an excellent correspondence between cell types defined 90 by transcriptomes and the classical methods (Li et al., 2017a;Shekhar et al., 2016). Here, we combine 91 scRNA-seq, fluorescence in situ hybridization, intersectional labeling of genetic defined cell types, 92 whole-brain axonal projection mapping, and single neuron reconstruction to investigate the 93 relationship between molecular architecture of serotonin neurons, the spatial location of their cell 94 bodies in the DR and MR, and their axonal arborization patterns in the brain. 95 3 96

Results 97
Single-cell RNA-sequencing defines 11 transcriptomic clusters of serotonin neurons in the dorsal 98 and medial raphe 99 We performed a comprehensive survey of DR and MR serotonin neurons in the adult mouse brain by 100 scRNA-seq ( Figure 1A). To specifically label serotonin neurons, we crossed Sert-Cre mice (Gong et 101 al., 2007) with the tdTomato Cre reporter mouse, Ai14 (Madisen et al., 2010). (Serotonin transporter,102 or Sert, is a marker for serotonin neurons; see more details below.) We collected serotonin neurons 103 acutely dissociated from brain slices by fluorescence-activated cell sorting (FACS) and used Smart-104 seq2 (Picelli et al., 2013) to generate scRNA-seq libraries. We used both male and female adult mice 105 (postnatal day 40-45) and applied two dissection strategies to separate the serotonin neurons 106 originating from anatomically-distinct brain regions: 1) in the first set of experiments, we dissected 107 the brainstem region that contain the entire MR and DR; 2) in the second set of experiments, we 108 focused on the principal DR (pDR, corresponding to the traditional B7 group) region by dissecting 109 specifically the DR but excluding its caudal extension (cDR, corresponding to the traditional group 110 B6) (Figure 1-figure supplement 1A). After quality control (Materials and methods), we 111 determined the transcriptomes of 709 cells from eight samples that include MR,pDR,and cDR,and 112 290 cells from six pDR-only samples (999 cells in total). We sequenced to a depth of ∼1 million reads 113 per cell and detected ~10,000 genes per cell (Figure 1-figure supplement 1B, Supplemental Table  114 1). The data quality and serotonin identity were validated by the fact that all 999 neurons expressed: 115 1) tryptophan hydroxylase 2 (Tph2), a key enzyme for serotonin biosynthesis; 2) transcription factor 116 Pet1 (Fev) known to express in raphe serotonin neurons (Hendricks et al., 1999); 3) the plasma 117 membrane serotonin transporter (Sert), which recycles released serotonin back to presynaptic 118 terminals of serotonin neurons, and 4) vesicular monoamine transporter 2 (Vmat2), which transports 119 serotonin (and other monoamines) from presynaptic cytoplasm to synaptic vesicles (Figure 1-figure  120 supplement 1C). Each cluster contained cells from both sexes after removing genes located on the Y chromosome, 126 indicating that there were few sex-specific differences (Figure 1-figure supplement 1D). No 127 substantial batch effect was observed (Figure 1-figure supplement 1E). Each of the 11 clusters 128 expressed a set of cluster-discriminatory genes, including markers for specific neurotransmitter 129 systems, such as Vglut3, Gad1, and Gad2 (Figure 1C,D; Figure 1-figures supplement 3-6). 130

Anatomical organization of transcriptomically defined serotonin clusters 131
Of the 11 transcriptomic clusters, six (Cluster 1-6) consisted of only serotonin neurons dissected from 132 the pDR. We hypothesized that these six clusters represent serotonin neurons from the pDR, and the 133 remaining five clusters represent cells from MR and cDR. To test this hypothesis and to obtain 134 information about the anatomical organization of transcriptomically defined serotonin cell clusters 135 within the DR and MR, we chose 16 cluster marker genes and performed hybridization-chain reaction 136 (HCR)-based single-molecule fluorescence in situ hybridization (smFISH) (Choi et al., 2018). To 137 restrict our analysis within the serotonin neuron population, we simultaneously double-labeled Tph2, 138 a marker for serotonin neurons, for all the HCR-smFISH experiments. 139 Figure 2A summarizes the distribution of the Tph2 + serotonin neurons that express each of the 140 16 cluster markers in four coronal sections that cover the pDR, cDR, and MR. This summary was 141 4 derived from counting Tph2/cluster marker double-labeled cells from confocal sections of the  smFISH experiments ( Figure 2B and Figure 2-supplement figures 1-2). Specifically, we found that 143 the distribution of markers for Clusters 1-6, Ret, Trh, Gad1, Npas1, Syt2, and C1ql2 ( Figure 1D), 144 were mostly restricted to the pDR. The two common Cluster 7 markers Tacr3 and Met were both 145 highly concentrated in serotonin neurons under the aqueduct in the cDR. Dlk1 should be expressed in 146 the DR clusters as well as Cluster 8 (Figure 1D), and its expression was found in both the DR and 147 MR, suggesting Custer 8 serotonin neurons are located in the MR. Clusters 9-11 markers Irx2 and 148 Piezo2 were mostly found in the MR. Thus, these observations support the anatomical breakdown 149 suggested by the dissection of primary tissue, and additionally provide a more granular and detailed 150 description about finer boundaries. 151 Within the DR, Trh + , Gad1 + , and Gad2 + serotonin neurons were mainly located in the dorsal 152 DR, whereas Vglut3 + and Syt2 + serotonin neurons were mainly located in the ventral DR and cDR. 153 These data suggest that Clusters 1-3 correspond to the dorsal DR and Cluster 4-6 to the ventral DR. 154 Cluster 6 marker Npas1 + was largely excluded from the densest portion of Tph2 expression at the 155 midline and instead was found scattered in the more rostral and ventral portion of the lateral wings. 156 On the other hand, Crhr2, which should be expressed in all DR serotonin neuron clusters except 157 Cluster 1 and 6 ( Figure 1D), was localized preferentially near the midline and was absent from the 158 lateral wing. Thus, Cluster 6 likely corresponds to serotonin neurons located preferentially in the 159 lateral wings. In contrast to DR, the anatomical organization of the molecular features that define MR 160 clusters is less obvious and different clusters appear more intermingled. 161 In summary, our HCR-smFISH experiments support the notion that Clusters 1-6 correspond 162 to pDR serotonin neurons, Cluster 7 corresponds to cDR serotonin neurons, and Clusters 8-11 163 correspond to MR serotonin neurons. We thus rename hereafter Clusters 1-6 as DR-1-6, Cluster 7 as 164 cDR, and Clusters 8-11 as MR-1-4 (Figure 3). Detailed expression levels of marker genes across all 165 11 clusters can be found in Figure 3-figure supplement 1-3. 166

Molecular properties of MR and DR serotonin neurons 167
Having determined the spatial locations of transcriptomically defined serotonin cell types, we next 168 analyzed key groups of differentially expressed genes crucial for neuronal function, including markers 169 for neurotransmitter systems, neuropeptides, ionotropic and metabotropic (G-protein-coupled) 170 neurotransmitter receptors, wiring specificity molecules, and transcription factors (Figure 4; Figure  171 4-figure supplement 1). 172 Genes related to neurotransmitters other than serotonin. The majority of the clusters 173 express Slc17a8, which encodes vesicular glutamate transporter-3 (Vglut3). These include almost all 174 serotonin neurons from the cDR and the majority from MR clusters, and DR-3-6 clusters ( Figure 1D). 175 These observations suggest that glutamate is the most prevalent co-transmitter for serotonin neurons. 176 Glutamate co-release from Vglut3 + serotonin terminals has indeed been reported at the orbital 177 prefrontal cortex (Ren et al., 2018), nucleus accumbens (Liu et al., 2014), ventral tegmental area 178 (Wang et al., 2019), and basolateral amygdala (Sengupta et al., 2017). 179 Three DR clusters express Gad1 and Gad2, which encode biosynthetic enzymes for the 180 neurotransmitter GABA. Two MR clusters express Gad2 but not Gad1, and cDR expresses neither. 181 Interestingly, few of the Gad1 + or Gad2 + neurons express vesicular GABA transporter (Vgat, Figure  182 4A left), which is responsible for transporting GABA into synaptic vesicles for synaptic transmission. 183 However, it has been reported that vesicular monoamine transporters (Slc8a1 for Vmat1; Slc8a2 for 184 Vmat2) can transport GABA into synaptic vesicles (Stensrud et al., 2014), and virtually all serotonin 185 neurons expressed Vmat2 (Figure 1-figure supplement 1D, Figure 4A left). Nevertheless, it remains 186 to be determined if these serotonin neurons can actually release GABA. At the single cell level, 15% 187 5 of serotonin neurons do not express any of the gene markers for glutamate or GABA transmission. 50% 188 of serotonin neurons express Vglut3, 21% express either Gad1, Gad2, or both, and 13% express 189 markers for Vglut3 and Gad1 or Gad2 (Figure 4A left). 190 In addition to small-molecule transmitters, the majority of serotonin neurons also co-express 191 neuropeptides (Figure 4A right). Expression of several neuropeptides served as excellent cluster 192 markers. For example, thyrotropin-releasing hormone (Trh) is highly expressed in DR-1-3 (14% 193 serotonin neurons, Figure 1C). Corticotropin-releasing hormone (Crh) is highly expressed in the DR3 194 cluster but much less everywhere else (7% serotonin neurons). Neuropeptide B (Npb) is highly 195 expressed in cDR and MR4 but much less in pDR serotonin neurons. Many serotonin neurons express 196 multiple neuropeptides (Figure 4A,B). 197 Small-molecule neurotransmitter receptors. Each cluster has a distinct expression pattern 198 of neurotransmitter (including neuropeptide) receptors ( Figure 4B, Figure 4-figure supplement 1 Other notable genes: the DR and MR serotonin clusters are also distinguished by differential 224 expression of ion channels as well as axon guidance and cell adhesion molecules (Figure 4-figure  225 supplement 1), which can contribute to their differences in physiological properties and wiring 226 specificity. Notably, genes encoding a voltage-gated K + channel (Kchn8) and mechanosensitive 227 channel (Piezo2) are highly expressed in the MR-4 cluster but exhibit little expression in all other 228 clusters. The chemokine ligand-12 (Cxcl12) and cadherin-6 (Cdh6) are preferentially expressed in 229 MR-1 and DR-4 clusters, respectively. 230 Transcription factors: Transcription factors (TFs) have been shown to be the best molecular 231 feature for cell type definition (Tabula Muris et al., 2018). Within our data, we observed robust cluster-232 specific expression of multiple TF genes ( Figure 4B; Figure 4-figure supplement 1). Importantly, 233 6 among the genes we identified are those previously reported to be involved in neuronal function. For 234 example, Nfix and Nfib (Bedford et al., 1998) are preferentially expressed in DR-3. Irx2 (Wylie et al.,  235 2010) is specific to MR-2 and MR-3. Sox13 is highly enriched in cDR. Zeb2 (Okaty et al., 2015) is 236 uniquely expressed in cDR and MR-4. TFs associated with neurodevelopmental disorders, such as 237 Npas1 and Npas3 (Erbel-Sieler et al., 2004), are preferentially expressed in DR-6, and Aff2 (Mondal 238 et al., 2012) is enriched within DR-3. 239 Transcriptional networks: To understand the relationship between cluster-specific genes and 240 to gain new insights on transcriptional regulatory programs orchestrating the maintenance of serotonin 241 neuron subtype identity, we next performed a pairwise correlation analysis of gene expression across 242 all 999 neurons (Supplemental File 1). We used Pearson correlation coefficient (rp) as a measure of 243 gene co-expression and focused on the genes with average expression > 10 counts per million. We 244 found multiple genes to co-expressed (defined as rp> 0.5) in serotonin neurons, forming co-expression 245 modules among each other as well as around various TFs (Supplemental File 1). Focusing further on 246 cluster markers, we identified three transcriptional "hubs" that putatively govern molecular programs 247 within respective neurons. Interestingly, two of these hubs were centered around pDR-specific TFs, 248 and, independently, comprised of dorsal (DR-1, DR-2, DR-3) and ventral (DR-4, DR-5, DR-6) pDR 249 markers ( Figure 4C). Among dorsal pDR markers we identified Pax5 and Sox14 to strongly correlate 250 (rp>0.5) with Gad1, Gad2, and Trh, among other genes. Among ventral pDR markers, we found Crhr2 251 to be highly correlated with several TFs, receptors, and neurotransmitter-related genes, most notably 252 Vglut3 (Slc17a8) and Hcrtr2. These Crhr2 + serotonin neurons could use TFs Esr2 and Sox1 to 253 maintain their subtype identity. Npas3 is specifically expressed in the DR-6 cluster and is highly 254 correlated with Htr1d, suggesting a critical role of maintaining the characteristics of DR-6 serotonin 255 neurons. Similarly, we found several MR and cDR-specific TFs to be the hubs of co-expression 256 modules. Particularly, the expression of a large number of cell adhesion molecules, receptors, ion 257 channel proteins and neuropeptides strongly correlate with the expression of TFs Zeb2, Pou3f1,Irx2,258 and Zfp536. Based on the identified correlation among gene expression of multiple marker genes 259 across all the cells, we speculate that the identified genes are linked by one or multiple transcriptional 260 regulatory programs, which, in turn, drive cell type-specific functions of distinct serotonin neuron 261 populations. 262 Sexually dimorphic gene expression: Finally, even though there were no apparent sex-263 specific difference at the cluster level Figure 1-figure supplement 1D), we did detect several genes, 264 such as Sod1, Snx10, Inpp4a, Zscan26, Ncam1, showing sexual dimorphism across the majority of cell 265 subtypes (Figure 4-figure supplement 2). 266

Viral-genetic tools to access different serotonin neuron subtypes 267
The gene expression patterns of specific serotonin neuron clusters can in principle allow genetic access 268 to these specific subpopulations for anatomical tracing, physiological recording, and functional 269 perturbation (Luo et al., 2018). However, DR and MR contain not only serotonin neurons but also 270 GABAergic and glutamatergic neurons that do not express Tph2 (and hence do not release serotonin), 271 some of which may project to the same target regions (McDevitt et al., 2014). To precisely investigate 272 the function of transcriptome-based serotonin neuronal types, we need to use an intersectional strategy 273 to target serotonin neurons that express specific additional markers (Jensen et al., 2008). To this end, 274 we generated Sert-Flp mice through homologous recombination-based knock-in in embryonic stem 275 cells (Materials and methods), and used Sert-Flp mice to intersect with transgenic Cre mice that 276 allow expression of a fluorescent reporter only in Flp + /Cre + (AND gate), so as to genetically label only 277 specific Cre-positive clusters ( Figure 5A). 278 To characterize the Sert-Flp mouse line, we crossed it with H11-CAG-FRT-stop-FRT-EGFP 279 mice we newly generated (Materials and methods). Anti-Tph2 staining on the brain slices containing 280 7 pDR showed that 98.5% GFP + neurons are Tph2 + and 100% Tph2 + neurons are GFP + (Figure 5B). To  281  further verify the intersectional strategy and to label the serotonin neurons co-expressing markers for  282 glutamate or GABA transmission, we crossed Sert-Flp with the IS reporter mice (Rosa-CAG-loxP-283 stop-loxP-FRT-tdTomato-FRT-EGFP) (He et al., 2016) and either Vglut3-Cre or Gad2-Cre.  Tph2 staining showed that all GFP-labeled neurons are Tph2 + (Figure 5C,D). In the pDR, Vglut3 + 285 serotonin neurons were mainly located ventrally, whereas Gad2 + serotonin neurons were mainly 286 located dorsally, consistent with our previous study (Ren et al., 2018) and the HCR-smFISH results 287 (Figure 2). 288 To map the axonal projection pattern of serotonin subtypes defined by intersection of Flp and 289 Cre expression, we developed a new AAV vector (AAV-CreON/FlpON-mGFP) that expressed 290 membrane-targeted GFP under the dual gates of Flp and Cre ( Figure 6A). Based on our scRNA-seq 291 and HCR-smFISH results, the Vglut3 + and Trh + pDR serotonin neurons consist of largely 292 complementary cell types at the transcriptomic level and have a distinct distribution along the dorsal-293 ventral axis in the pDR. To visualize these two subpopulations of serotonin neurons, we injected AAV-294 CreON/FlpON-mGFP into pDR of either Vglut3-Cre;Sert-flp (n = 3, Figure 6B) or Trh-Cre;Sert-flp 295 mice (n = 3, Figure 6C). Anti-Tph2 staining showed that 98.2% GFP + neurons were Tph2 + . As 296 predicted, Vglut3 + Sert + GFP cells were mostly located in the ventral pDR ( Figure 6B), whereas 297 Trh + Sert + cells were located in the dorsal pDR ( Figure 6C). As negative controls, we injected the 298 same virus into mice carrying only the Sert-flp transgene or only the Vglut3-Cre transgene and did not 299 find any mGFP + cell bodies or fibers (n = 3 for each; data not shown). 300 The intersectional strategy allowed us to trace the projection of GFP + axons from these two 301 groups of serotonin neurons across the brain. We next examined their projections by staining every 302 four coronal sections across the brain with anti-GFP antibody. We found that serotonin axons from 303 Vglut3 + population preferentially targeted cortical regions ( Figure 6D), consistent with our previous 304 results ( Ren et al., 2018). By contrast, no GFP-labeled axons were observed in the cortical regions 305 from Trh-Cre:Sert-flp mice. Instead, Trh + serotonin axons project to the anterior and medial 306 hypothalamus, posterior amygdala, and the lateral geniculate nucleus in the thalamus, none of which 307 were targeted by Vglut3 + axons ( Figure 6D). 308

Whole-brain axonal projections of selected serotonin neuron subpopulations 309
While suggestive of an anatomical division in targets, assessing the full extent to which projections of 310 Vglut3 + or Trh + pDR serotonin neuron populations segregate requires quantifying axonal innervation 311 at the whole-brain level. We used the iDISCO-based brain clearing technique AdipoClear (Chi et al.,312 2018) to visualize, align, and summarize whole-brain projectomes ( Figure 7A). Individual 313 hemispheres of either Vglut3-Cre;Sert-flp (n = 3) or Trh-Cre;Sert-flp mice (n = 3) injected with AAV-314 CreON/FlpON-mGFP at pDR were imaged by light-sheet microscopy. We developed deep learning 315 models to automatically trace whole-brain axonal projections by segmenting volumes with a 3D UNet-316 based convolutional neural network we developed (Materials and methods). The resulting 317 volumetric probability maps were thinned and thresholded before aligning to the Allen Institute's 2017 318 common coordinate framework as previously described (Ren et al., 2018; Figure 7A). 319 We visualized the axon terminals in brain regions targeted by either the Trh + or the Vglut3 + 320 population of serotonin neurons. Initial assessment of selected target regions suggested a strong 321 segregation of axonal projection patterns between these two populations ( Figure  there were extensive differences in innervation patterns of terminal axon fields between the Vglut3 + 326 and Trh + populations. Whole-brain quantitative and statistical analyses showed Vglut3 + axons 327 8 preferentially in anterolateral cortical regions and adjacent structures such as olfactory bulb, agranular 328 insular cortex, endopiriform, piriform, and claustrum as well as other cortical regions such as 329 entorhinal, primary motor, and barrel cortices. By contrast, Trh + axons were largely absent from these 330 structures. Conversely, subcortical regions primarily in thalamus (zona incerta and medial geniculate) 331 and hypothalamus (anterior and dorsomedial nuclei) were preferentially targeted by Trh + axons and 332 largely avoided by the Vglut3 + population ( Figure 7E; Video 1). 333 Given the variability of locations and amount of viral transduction, individual brains from the 334 same genotype exhibit considerable variation in total axons labeled ( Figure 7E top) and in detailed 335 projection patterns (Figure 7-figure supplement 1-2). These variabilities further highlighted regions 336 that were targeted densely but exclusively in one or two individual brains Notable examples include 337 anterior bed nucleus of stria terminalis (BNST), posterior amygdala, and globus pallidus external 338 segment (GPe) in Trh + projections and the lateral central amygdala (CeA) and dentate nucleus of the 339 cerebellum for Vglut3 + projections. While we did identify large-scale patterns of collateralization for 340 these two subtypes of serotonin neurons, one possible contribution to this inter-individual variability 341 in projection patterns is heterogeneity within molecularly defined subpopulations of serotonin neurons. 342

Whole-brain axonal arborization patterns of individual serotonin neurons 343
Our whole-brain projection analyses indicate that axonal arborization patterns of molecularly defined 344 serotonin neuron subpopulations are still very complex (Figure 7). To examine the extent to which 345 this reflects projection patterns of individual serotonin neurons, we combined the cell-type-specific 346 sparse labeling strategy we recently developed ( (containing >50% of the axon length within in olfactory cortex, n = 7), prefrontal cortex-projecting 364 (>30% of the axon length, n = 4), entorhinal cortex-projecting (>30% of the axon length, n = 3), and 365 dorsal cortex-projecting (including motor, somatosensory, retrosplenial, and visual cortex; n = 3). In 366 the olfactory area-projecting subgroup, three brains had branches in the olfactory bulb (OB). Axons 367 of the three dorsal cortex-projecting neurons traveled the largest distance, entering the cortex 368 anteriorly and extending to the posterior end. Interestingly, seven cortex-projecting serotonin neurons 369 also sent substantial branches to non-cortical regions (>10% of axon length), including three that 370 innervated the cerebellum. 371 Hypothalamus-projecting. Each cell of this group (n = 11) dedicated more than 33% axon 372 length to innervate the hypothalamus ( Figure 8B; Figure 8-figure supplement 2B). None of them 9 had collateralization in the cortical regions. Four cells had substantial branches to the pons, one to the 374 amygdala and one to the thalamus (> 10% axon length). 375 Amygdala-projecting. The axons of all amygdala-projecting DR serotonin neurons (n = 7; 376 Figure 8C; Figure 8-figure supplement 3A) followed either or both of the two distinct routes to 377 reach the amygdala (Figure 8-figure supplement 4A1, A2). Six cells had collateralized axonal 378 terminals in the hypothalamus (>10% axon length). One cell sent a branch to the contralateral cortex. 379 Two cells had branches that were almost equally distributed in the bed nucleus of stria terminalis 380 (BNST), central amygdala (CeA), and medial amygdala (MeA). Interestingly, the axon arbors at the 381 CeA or MeA were highly restricted and dense (Figure 8-figure supplement 4A3). 382 Other groups. The rest of the DR serotonin neurons were divided into thalamus-projecting, 383 striatum-projecting, and caudal brainstem-projecting groups based on the highest density of axons 384 (Figure 8-figure supplement 3B-D), even though most also innervated other regions including the 385 medulla and spinal cord (Figure 8-figure supplement 1C). While most projections of the DR 386 serotonin neurons were unilateral (37/50), one of the thalamus-projecting neurons had symmetrical 387 bilateral projections in the thalamic target (Figure 8-figure supplement 4B). For the two DR 388 serotonin neurons that innervated ventral striatum (also called nucleus accumbens, or NAc), 389 arborization appeared to be restricted to either the core or the shell (Figure 8-figure supplement 4C). 390 In summary, these results revealed remarkable complexity and heterogeneity of serotonin 391 neuron projections at the single-cell level. They nevertheless followed certain patterns. For example, 392 the cell body locations of these traced neurons (Figure 8- tracing, we could now directly quantify the total length of axons of the 50 DR serotonin neurons. We 400 found that the total axon length of these DR serotonin neurons exhibited considerable heterogeneity, 401 from 1.2 cm to 22.7 cm. When examined across the six groups, cortex-projecting serotonin neurons 402 indeed had the longest axons ( Figure 8H) depth single-cell RNAseq in combination with systematic fluorescence in situ hybridization, whole 414 brain projection mapping via intersectional methods and single-axon tracing, we begin to shed light 415 on the relationship between transcriptomic clusters, the spatial location of their cell bodies, and brain-416 wide projection patterns of serotonin neurons. 417

Relationship between molecular identity and cell body distribution 418
Our single-cell transcriptome analysis identified 11 molecularly distinct types of serotonin neurons in 419 the DR and MR (Figure 1). Based on tissue source from which scRNA-seq data was collected and 420 fluorescent in situ hybridization using transcriptomic cluster markers, we were able to assign six types 421 to principal DR (pDR), one type to caudal DR (cDR), and four types to MR (Figure 3). The fact that 422 we can assign specific transcriptomic clusters to specific groups of raphe nuclei indicate that 423 molecularly defined serotonin populations are spatially segregated at least at this coarse level. Our 424 results are broadly consistent with previous findings that utilized developmental origin to differentiate 425 raphe serotonin neurons (Okaty et al., 2015). 426 The six types of serotonin neurons within principal DR exhibit further specificity in spatial 427 distributions. Specifically, serotonin neurons from DR-1-3 clusters are preferentially localized in 428 dorsal pDR, whereas those from DR-4-6 in ventral pDR, with DR-6 neurons preferentially localized 429 to the ventral lateral wings. These data support and extend our previous finding (Ren et al., 2018) for 430 the preferential ventral pDR location of Vglut3 + serotonin neurons, which is highly expressed in DR-431 4 and DR-5 clusters. Together, these findings revealed the molecular basis for the differentiation of 432 dorsal/ventral DR sub-systems (Figure 2 and 3). 433 cDR has been suggested to be more similar to the MR than to pDR in their connectivity 434 (Commons, 2015;Kast et al., 2017). Our single-cell transcriptomic analysis indicated that serotonin 435 neurons in the cDR are strikingly homogenous and profoundly different from both pDR and MR at 436 the molecular level ( serotonin neurons. This illustrates the value of using genetically targeted strategies to characterize 448 important but rare types of cells in the brain. 449

Relationship between molecular identity and projection-defined serotonin sub-systems 450
In our previous study, we characterized two projection-defined parallel DR serotonin sub-systems. We 451 found that serotonin neurons that project to the orbitofrontal cortex (OFC) and central amygdala (CeA) 452 differ in input and output patterns, physiological response properties, and behavioral functions (Ren 453 et al., 2018). Whole-brain collateralization patterns of these two sub-systems indicate that there must 454 be additional sub-systems projecting to regions not visited by either of these two sub-systems projects 455 to, including much of the thalamus and hypothalamus. What is the relationship between molecularly 456 defined serotonin neurons and projection-defined sub-systems? 457 Using viral-genetic intersectional approaches to access specifically Vglut3 + pDR serotonin 458 neurons in combination with staining in histological sections ( Figure 6) and iDISCO-based whole 459 brain imaging (Figure 7), we found that these neurons project profusely to cover much of the 460 neocortex, as well as the olfactory bulb, cortical amygdala, and lateral hypothalamus. Comparisons of 461 the projection patterns of Vglut3 + (this study) with OFC-projecting DR serotonin neurons (Ren et al.,462 2018) suggest that the latter is a large subset of the former. Brain regions that are innervated by Vglut3 + 463 but not by OFC-projecting serotonin neurons include the somatosensory barrel cortex, ventral striatum, 464 and a specific subset of CeA. Interestingly, our previous study indicated that ~23% of CeA-projecting 465 DR serotonin neurons are Vglut3 + (Ren et al., 2018), so it is possible that even within a refined nucleus 466 like CeA, molecularly distinct serotonin projections are confined to sub-regions with a finer resolution. 467 We also assessed the whole-brain projection patterns of a largely complementary population 468 of DR serotonin neurons, namely those that express Trh and thus belong to DR-1-3 clusters. We found 469 that Trh + serotonin neurons predominantly project to subcortical regions, most notably anterior and 470 medial nuclei of the hypothalamus and several thalamic nuclei, a pattern mostly complementary to the 471 Vglut3 + population (Figures 6-7, Video S1). Given that our previous CeA-projecting DR serotonin 472 neurons do not innervate most of the hypothalamus, and Trh + serotonin neurons only partially 473 innervate CeA, these two populations are at most partially overlapping. 474 These comparisons support a broad correspondence between molecular identity and axonal 475 projection patterns at the level of DR serotonin neuronal populations that include multiple 476 transcriptomic clusters. These results will enable future testing of whether a more precise 477 correspondence exists at the level of single transcriptomic clusters that we have defined here. Our 478 transcriptome data suggest that each DR/MR serotonin neuron type can be distinguished from others 479 by the expression of two marker genes ( Figure 1D; Figure3-figure supplement 1-3), supporting 480 the view that neuronal subtypes are generally specified by unique combination of genes rather than 481 single genes (Li et al., 2017b). With the generation of drivers based on these marker genes, 482 intersectional methods in combination with location-specific viral targeting can be used in the future 483 to dissect the projection patterns of the individual transcriptomic clusters. 484

Insights from single-cell reconstruction 485
The complexity of whole-brain projection patterns of Vglut3 + and Trh + populations discussed above 486 can be driven by: 1) heterogeneity of projection patterns of different transcriptomic clusters within the 487 Vglut3 + or Trh + population; 2) heterogeneity of projection patterns of serotonin neurons within the 488 same transcriptomic cluster; 3) complex collateralization patterns of individual serotonin neurons; and 489 4) a combination of some or all of the above. If scenario 1 were the only contributing factor, then there 490 would be only six different projection patterns for the pDR serotonin neurons. However, our single-491 cell reconstruction of DR serotonin neurons revealed many more than 6 branching patterns (e.g., 492 Figure 8G), indicating that there must be diverse collateralization patterns even within the same 493 transcriptomic cluster, and highlighting the complexity of individual serotonin neuron projections. 494 Despite the complexity, these single-cell tracing data nevertheless suggest certain rules obeyed by 495 serotonin neurons. 496 First, there is a general segregation of cortical-and subcortical-projecting serotonin neurons. 497 Of the 46 forebrain-projecting DR serotonin neurons, 31 have a strong preference (>90% total axon 498 length) for innervating either cortical or subcortical regions. This is likely an underestimate of the 499 preference, especially for cortical-projecting ones, as their axons necessarily need to travel through 500 the subcortical regions to reach the cortex. (As a specific example, most forebrain-projecting DR 501 serotonin neurons pass through the lateral hypothalamus to reach their targets; it is thus difficult to 502 determine whether axons in the lateral hypothalamus play a local function.) Second, most of the 503 serotonin neurons tend to focus a majority of their arborization within one brain region (e.g., prefrontal 504 vs entorhinal cortex, Figure 8B; CeA vs MeA, Figure 8-figure supplement 4A3). The subcortical-505 projecting serotonin neurons appear to have more specificity, with most neurons exhibiting dense 506 arborization in one or two nuclei. The cortical-projecting serotonin neurons can have elaborate 507 arborization patterns across multiple cortical areas (e.g., Figure 8-figure supplement 1C) and the 508 longest axon lengths per cell ( Figure 8H). 509 12 Together with our study on the projection-defined serotonin sub-systems (Ren et al., 2018), we 510 believe it is unlikely that the major function of the forebrain-projecting serotonin system is to broadcast 511 information non-discriminately. Our study is limited by the scope (50 reconstructed cells out of 9000 512 DR serotonin neurons). To fully reveal the organizational logic of the serotonin system, efforts should 513 be put into larger scale single-cell reconstruction integrated with molecular identity and functional 514 studies of individual transcriptomic clusters of serotonin neurons. 515

Integrating multiple features within individual serotonin sub-systems 516
The molecular features of different serotonin cell types suggest their functional properties. For 517 example, several studies have reported that subgroups of serotonin neurons in the MR and DR 518 express Vglut3 and indeed, subsequent slice recording confirmed that serotonin terminals can co-519 release glutamate and serotonin. In addition to neurochemical properties, each serotonin neuron 520 population reveals a specific combination of distinct genes responsible for electrophysiological (ion 521 channels), connectivity (wiring molecules), and functional (neurotransmitter/peptide receptors) 522 properties (Figure 4; Figure 4-figure supplement 1A). For example, most Crhr2 + neurons co-523 express Vglut3 and Npy2r, which suggests that these serotonin neurons use glutamate as co-524 transmitter in their cortical targets, and are in turn modulated by corticotropin-releasing hormone and 525 neuropeptide Y. Meanwhile, most Trh + serotonin neurons co-expression Gad1, Kcns1, and α1A 526 adrenergic receptors (Adra1a) specifically. We can speculate that these serotonin neurons use Trh 527 (and perhaps GABA) as co-transmitters to regulate the physiology of their subcortical targets, and 528 are in turn modulated by locus coeruleus norepinephrine neurons. 529 Our previous study suggests that DR serotonin sub-systems have biased input but segregated 530 output (Ren et al., 2018). Here we found that each of the transcriptomic clusters of serotonin neurons 531 have distinct combination of axon guidance and cell adhesion molecules (Figure 4-figure  532 supplement 1A). These differentially expressed wiring molecules might be used during development 533 to set up distinct projection patterns of different serotonin neuron types (Deneris and Gaspar, 2018; 534 Kiyasova and Gaspar, 2011;Maddaloni et al., 2017), and/or used in adults to maintain wiring 535 connectivity or contribute to the remarkable ability of serotonergic axons to regenerate after injury 536 (Jin et al., 2016). 537 In conclusion, our comprehensive transcriptomic dataset and its 11 distinct groups of the DR 538 and MR serotonin neurons reveals the molecular heterogeneity of the forebrain-projecting serotonin 539 system. We have shown that the molecular features of these distinct serotonin groups reflect their 540 anatomical organization and provide tools for future exploration of the full projection map of 541 molecularly defined serotonin groups. The molecular architecture of serotonin system lays the 542 foundation to integrate connectivity, neurochemical, physiological, and behavioral functions. This 543 integrated understanding of serotonin can in turn provide novel therapeutic strategies to treat brain 544 disorders involving this important neuromodulator. 545 546

Materials and methods 547
Animals 548 All procedures followed animal care and biosafety guidelines approved by Stanford University's 549 Administrative Panel on Laboratory Animal Care and Administrative Panel of Biosafety in accordance 550 with NIH guidelines. For scRNA-seq (Figure 1 and 4

Transcriptome analysis 574
Single-cell isolation and sequencing. Lysis plates were prepared by dispensing 4 μl lysis buffer 575 as described in (Tabula Muris et al., 2018). After dissociation, single tdTomato + cells were sorted 576 in 96-well plates using SH800S (Sony). Immediately after sorting, plates were sealed with a pre-577 labelled aluminum seal, centrifuged, and flash frozen on dry ice. cDNA synthesis and library 578 preparation were performed using the Smart-seq2 protocol (Picelli et al., 2014). Wells of each 579 library plate were pooled using a Mosquito liquid handler (TTP Labtech). Pooling was followed by 580 two purifications using 0.7x AMPure beads (Fisher, A63881 counted as expressed if it has at least one read mapping to it and is detected in at least 3 cells. Cells 594 with fewer than 50,000 reads were excluded. Counts were log-normalized for each cell using the 595 natural logarithm of 1 + counts per million [ln(CPM+1)]. All genes were projected onto a low-596 dimensional subspace using principal component analysis. Cells were clustered using a variant of 597 the Louvain method that includes a resolution parameter in the modularity function ( Tabula Muris  598  Gene co-expression networks. The relationship between gene expression was measured using rank 603 correlation statistics. Pearson correlations were computed across all cells. We first removed low 604 expressed genes by selecting genes with mean expression CPM>2, leaving ~8000 genes in the dataset. 605 Pearson correlation coefficients (rp) were computed for each gene and significance was tested by 606 bootstrapping (1,000 iterations). A correlation table containing rp above 0.3 and below -0.3 can be 607 found in Supplemental Table 3. Reported values are mean from the bootstrapped values. Gene 608 functional categories were retrieved from HGNC resource (https://www.genenames.org). Genes 609 assigned to more than one functional category were re-assigned a single category in the following 610 priority order : SNAREs, Secreted Ligands, ACh and Monoamine Receptors, Glutamate Receptors, 611 Axon Guidance and Cell Adhesion Molecules (CAMs), GABA Receptors, GPCRs, Ion Channel 612 Proteins, Transcription Factors, a full list of genes assigned to each category can be found in 613 Supplemental Table 2. Gene pairs for which rp<0.4 were removed and remaining pairs were 614 visualized as a network using igraph and visNetwork R packages. To further refine the final list of co-615 expressed genes and generate Figure 4D we focused on gene pairs for which:1) rp>0.5; 2) at least one 616 gene of the pair is found among cluster markers; 3) both genes of the correlating pair belong to one of 617 the above listed functional categories. 618 Data availability. The datasets generated and analyzed in the study are available in the NCBI Gene 619 Expression Omnibus (GEO) (currently waiting for the access number is relative to the brain surface when AP is -1.0). After surgery, mice recovered on a heated pad until 628 ambulatory and then returned to their homecage. 629

Histology and Imaging 646
Animals were perfused transcardially with phosphate buffered saline (PBS) followed by 4% 647 paraformaldehyde (PFA). Brains were dissected, post-fixed in 4% PFA for 12-24 hours in 4 ºC, then 648 placed in 30% sucrose for 24-48 hours. They were then embedded in Optimum Cutting Temperature 649 (OCT, Tissue Tek) and stored at-80ºC until sectioning. For the antibody staining in Figure 1, 50-µm 650 sections containing DR were collected onto Superfrost Plus slides to maintain the anterior to posterior 651 sequence. All working solutions listed below included 0.2% NaN3 to prevent microbial growth. Slides 652 were then washed 3x10 min in PBS and pretreated overnight with 0.5 mM SDS at 37ºC. Slides were 653 then blocked for 4 hours at room temperature in 10% normal donkey serum (NDS) in PBS with 0.3% 654 Triton-X100 (PBST), followed by incubation in primary antibody (Novus, rabbit anti-Tph2) diluted 655 1:1000 in 5% NDS in PBST for 24 hours at RT. After 3x10 min washes in PBS, secondary antibody 656 was applied for 6 hours at room temperature (donkey anti-rabbit, Alexa-647 or Alexa-488, Jackson 657 ImmunoResearch), followed by 3x10 min washes in PBST. Slides were then stained for NeuroTrace 658 Blue (NTB, Invitrogen). For NTB staining, slides were washed 1x5 min in PBS, 2x10 min in PBST, 659 incubated for 2-3 hours at room temperature in (1:500) NTB in PBST, washed 1x20 min with PBST, 660 and 1x5 min with PBS. Sections were additionally stained with DAPI (1:10,000 of 5 mg/mL, Sigma-661 Aldrich) in PBS for 10-15 min and washed once more with PBS. Slides were mounted and coversliped 662 with Fluorogel (Electron Microscopy Sciences). After that, the slides were then imaged either using a 663 Zeiss 780 confocal microscope or a 3i spinning disk confocal microscope (CSU-W1 SoRa), and 664 images were processed using NIH ImageJ software. After that, whole slides were then imaged with a 665 5x objective using a Leica Ariol slide scanner with the SL200 slide loader. 666 For DR-containing slices in Figure 5 and 6, staining was applied to floating sections. Primary 667 antibodies (Novus, rabbit anti-Tph2, 1:1000; Rockland, rabbit anti-RFP, 1:1000; Abcam, goat anti-668 Tph2, 1:500; Aves Labs Inc., chicken anti-GFP, 1:2000) were applied for 48 hours and secondary 669 antibodies for 12 hours at 4ºC. For serotonin terminal staining in Figure 6, floating sections were 670 stained with Primary antibodies (Aves Labs Inc., chicken anti-GFP, 1:2000) for 60 hours and 671 secondary antibodies for 18 hours at 4ºC. 672 Whole Brain Imaging of Vglut3 + and Trh + Serotonin Projections 673 After 8-10 weeks of virus expression, mice were transcardially perfused with 20 ml 1x PBS containing 674 10 µg/ml heparin followed by 20 ml of 4% PFA before removing each brain and allowing them to 675 postfix overnight at 4 C. The clearing protocol largely follows the steps outlined in (Chi et al., 2018). 676 Brains were washed at room temperature with motion at least 30 minutes each step: 3x in 1x PBS 677 before switching to 0.1% Triton X-100 with 0.3 M glycine (B1N buffer) and being stepwise 678 (20/40/60/80%) dehydrated into 100% methanol. Delipidation was carried out by an overnight 679 incubation in 2:1 mixture of DCM:methanol and a 1 hour incubation in 100% DCM the following day. 680 After 3x washes in 100% methanol, brains were bleached for 4 hours in a 5:1 mix of methanol: 30% 681 H2O2 and then stepwise (80/60/40/20%) restored into B1N buffer. Samples were permeabilized with 682 2 washes of PTxwH buffer containing 5% DMSO and 0.3M glycine for 3 hours before being washed 683 in PTxwH overnight. Antibody labeling was carried out in PTxwH buffer at 37 C with motion. 684 Primary antibody (Aves 1020; chicken anti-GFP, 1:1000) was added and samples were incubated for 685 11 days. After 3 days of washes, secondary antibody (Thermo A-31573, AlexaFluor 647 donkey anti-686 chicken, 1:1000) was added for 8 days, followed by another 3 days of washes. One final day of 687 washing in 1x PBS preceded clearing. Samples were again dehydrated stepwise into methanol, using 688 water as the alternative fraction. Delipidation proceeded as before with a DCM/methanol mixture 689 overnight and 2x 1 hour DCM-only incubations the next day. Brains were finally cleared for 4 hours 690 in dibenzyl ether and then stored in a fresh tube of dibenzyl ether at least 24 hours before imaging. 691 Samples were imaged with a LaVision Ultramicroscope II lightsheet using a 2x objective and 692 3 µm z-step size. Antibody fluorescense was collected from a 640 nm laser and autofluorescense 693 captured from 488 nm illumination. Image volumes were processed and analyzed with custom Python 694 and MATLAB scripts. In short, we trained a 3D U-Net convolutional neural network (Çiçek, et al., 695 2016) to identify axons in volumes and post-processed the resulting probability-based volumes as 696 previously described (Ren et al., 2018). Using the autofluorescent channel, we aligned samples to the 697 Allen Institute's Common Coordinate Framework (Renier et al., 2016), applied the same 698 transformation vectors to the volumetric projection of axons, and quantified total axon content in each 699 brain region listed in Supplemental Table 4. 700

Hybridization Chain Reaction in situs 701
Probes were generated for use with HCR v3.0 (Molecular Technologies) (Choi et al., 2018). Wildtype 702 8 week old mice were perfused and brains were removed and fixed as described above. Following an 703 overnight postfix, brains were cryoprotected in 30% sucrose until they sank and subsequently frozen 704 at -80 C. The midbrain raphe nuclei were sectioned coronally at either 16 or 20 µm directly onto a 705 glass slide, and dried at room temperature for 4-6 hours before storing at -20 C overnight. Aldrich) three times (for 2 h each) and embedded in Lowicryl HM20 resin (Electron Microscopy Sciences, 716 14340). We use a fluorescence micro-optical sectioning tomography (fMOST) system to acquire the 717 brain-wide image dataset at high resolutions (0.23 × 0.23×1 μm for 10 brains and 0.35 × 0.35×1 μm for 718 the other 9 brains). Embedded brain samples were mounted on a 3D translation stage in a water bath with 719 propidium iodide (PI). The fMOST system automatically performs the coronal sectioning with 1um steps 720 and imaging with 2 color channels in 16-bit depth. The green channel of GFP is for visualization of 721 neurons and the red channel of PI counterstaining is for visualization the whole brain cytoarchitecture. 722 Image annotation and skeletonization. Amira software (v 5.4.1, Visage imaging, Inc) were used for 723 semi-automatically reconstruction of single neurons. First, we use Amira to load a relatively large volume 724 but low-resolution data and find the position of soma or major axon as a start position. Then, one by one, 725 we load each small volume (800× 800×400 voxels) of highest resolution data along the axons and 726 dendrites to label the full structure of each neuron. We have totally reconstructed 50 high quality neurons 727 with complete and clear axon terminals. The reconstructed neurons were checked by one another person. 728 Image registration and visualization. Reconstructed neurons were aligned to the Allen Common 729 Coordinate Framework (CCF   1901-1912. 930 Stuart, T., Butler, A., Hoffman, P., Hafemeister, C., Papalexi,E.,Mauck,W.M.,3rd,Hao,Y.,931 Stoeckius, M., Smibert, P., and Satija, R. (2019). Comprehensive Integration of Single-Cell Data. Cell 932 177, 1888-1902 Computational data, a., Cell type, a., Writing, g., et al. (2018). Single-cell transcriptomics of 20 mouse 935 organs creates a Tabula Muris. Nature 562, 367-372. 936 Tasic, B., Hippenmeyer, S., Wang, C., Gamboa, M., Zong, H., Chen-Tsai, Y., and Luo, L. (2011). 937 Site-specific integrase-mediated transgenesis in mice via pronuclear injection. Proc Natl Acad Sci U 938 S A 108, 7902-7907. 939 Tasic   analysis such that areas defined by individual layers (e.g., cortical layers I-VI), cell identity, and 1176 anatomical cardinal directions are collapsed into their parent region. Individual normalized regional 1177 densities for each brain are aligned to the heat maps from Figure 7E. Number Identity Expression Level Identity Expression Level Identity Expression Level  T o x P u 3 f1 n f1 R x r g fp 5 0 3 r 2 s 1 N r 2 f1 R u n 1 N fi l3 A r n tl N r 2 f2 R o r b E g r 1 P u r g A r i 3 N p