Neural networks of the mouse visceromotor cortex

Abstract

The medial prefrontal cortex (MPF) regulates autonomic and neuroendocrine responses to stress^1,2 and coordinates goal-directed behaviours such as attention, decision-making and social interactions^3,4,5,6,7,8. However, the underlying mechanisms remain unclear due to incomplete circuit-level MPF characterization⁷. Here, using integrated neuroanatomical, physiological and behavioural approaches, we construct a comprehensive wiring diagram of the MPF, focused on the dorsal peduncular area (DP)—a poorly understood prefrontal area. We identify its deep (DPd) and superficial (DPs) layers, along with the infralimbic area, as major components of the visceromotor cortex that directly project to hypothalamic and brainstem structures to govern neuroendocrine, sympathetic and parasympathetic output. Notably, the DP functions as a network hub integrating diverse cortical inputs and modulating goal-directed behaviour through a largely unidirectional cortical information flow. On the basis of the mesoscale MPF connectome, we propose a unified network model in which distinct MPF areas orchestrate physiological and behavioural responses to internal and external stimuli.

Main

Traditionally, identification of the MPF (or mPFC) relied on its connections to brain structures involved in autonomic and emotional responses, such as the hypothalamus and periaqueductal gray (PAG)^4,9. Accordingly, the MPF arguably includes the anterior cingulate (ACA), prelimbic (PL) and infralimbic (ILA) cortices^7,10. Yet, ambiguity persists in the precise definition of the rodent MPF, particularly concerning the DP, a unique medial prefrontal region that is situated between the six-layered neocortex and the three-layered olfactory cortex¹¹. Studies suggest that the DP, together with the ventrally adjacent dorsal tenia tecta (TTd), regulates autonomic and behavioural responses to psychological stress^12,13, anxiety and depression¹⁴, fear conditioning¹⁵, threat¹⁶ and opioid reward¹⁷, which aligns with MPF function; however, a thorough understanding of its neural circuits and structural–functional position within the MPF are obscure. Without the DP, proposed unified MPF network models would be incomplete and inaccurate.

DP is a unique MPF area

The DP, initially included as part of the ILA¹⁸, was defined as part of the olfactory cortex¹⁹. In the mouse brain, the anatomical borders of the DP are inconsistent across mouse brain atlases^7,11,20. To address this, we refined the borders of the DP (Allen Reference Atlas (ARA) levels 37–41; Supplementary Fig. 1) on the basis of its cytoarchitecture, neuronal connectivity and gene expression (Fig. 1a–d and Extended Data Figs. 1 and 2).

Fig. 1: Anatomic delineation and cell type organization of the DP.

a, DP delineation based on Nissl cytoarchitecture. Two DP sublayers were identified: the DPs and DPd. Connectivity confirms DPs and DPd distinction: the BLAal (PHAL, magenta) targets the DPs, while the BLAam (AAV-GFP) innervates the ILA and DPd. b, Inputs from the AUDv and RE terminate in DP layer 1, defining the DP boundaries. c, Connectivity-based characterization of DP sublayers. FG-labelled DP to ACAd neurons are confined to DPs, whereas MPO- and thalamus-projecting neurons localize in the DPd. LHA-projecting neurons span the DPd and deep DPs. ProSUB axons densely innervate the DPs. d, Cell-type-specific gene expression supports regional and laminar DP organization. The RNAscope analysis shows that Cux2 is absent in the DP, Etv1 is enriched in the DPs and Tle4 is expressed in the DPd. Cacna1h is expressed in the neocortex and DPs. DPd is distinguished with Vglut2 expression in VGLUT2 Cre × Ai14 mice and via in situ hybridization. Bright-field in situ hybridization images from the ABA (https://mouse.brain-map.org/). e, Schematic of neuron types in the DPs and DPd. BS, brainstem; d, deep; s, superficial. f, Dendritic morphology of reconstructed DP neurons (left). Morphological clustering (right, top) based on the size (x axis) and complexity (y axis) of dendritic arbours (n = 47) identifies two morphological groups: one with smaller, simpler dendrites (n = 32; dendritic length, 3,024 ± 833 µm; number of branches, 38 ± 12), and another with larger, more complex dendrites (n = 15; dendritic length, 5,821 ± 958 µm; number of branches, 76 ± 13). The 47 DP neurons were pooled from four MORF3 mice. On the basis of the somatic distance from the midline, neurons also were clustered as superficial-layer (n = 26; distance from the midline, 288 ± 81 µm) or deep-layer (n = 21; distance from the midline, 605 ± 106 µm) neurons and are plotted in the size and complexity dimensions (middle right). Examples of superficial and deep neurons are presented from a single brain (bottom left). The box plot (bottom right) demonstrates that superficial neurons have a greater dendritic length (superficial, 4,275 ± 1,717 µm; deep, 3,472 ± 1285 µm; one-tailed Wilcoxon signed-rank test, FDR-corrected P = 0.0465), a higher number of branches (superficial, 56 ± 24; deep, 42 ± 16; FDR-corrected P = 0.0317) and contain a greater proportion of dendrites between soma and midline (superficial, 65 ± 17%; deep, 53 ± 17%; FDR-corrected P = 0.026). The smaller–simpler neurons were more distant from the midline compared with the larger–complex neurons (small–simple distance, 472 ± 184 µm; large–complex distance, 340 ± 153 µm; FDR-corrected P = 0.026). The line inside the box plot shows the median, the top and bottom edges are the upper (0.75) and lower (0.25) quartiles, the top whisker connects the upper quartile to the non-outlier maximum, and the bottom whisker connects the lower quartile to the non-outlier minimum. Scale bars, 200 μm (a (middle), b, c and d (left)), 300 μm (d (right) and f) and 500 μm (a (left and right)). A full list of abbreviations is provided in Supplementary Table 1.

Source data

Cytoarchitecturally, the DP has a notably thicker layer 1 compared with the dorsally adjacent ILA and ventrally adjacent TTd (Fig. 1a, Supplementary Fig. 1a,b and Supplementary Methods). The DP also exhibits a unique laminar organization that distinguishes it from other six-layered neocortical areas. On the basis of Nissl staining (Fig. 1a), we identified two primary cellular DP layers: a superficial layer (DPs) characterized by loosely arranged, relatively larger cell bodied neurons, and a deep layer (DPd) in which neurons with smaller cell bodies are densely packed. This cytoarchitectonic feature is also distinct from the densely packed, darkly stained cellular layer 2 and cell-sparse layer 3 of the TTd.

Our connectivity-based anatomic parcellation validated the laminar-specific DP features. Axon projections from the ventral auditory cortex (AUDv), lateral entorhinal cortex (ENTl) and reuniens thalamic nucleus (RE) generate dense terminals specifically in DP layer 1 (Fig. 1b and Extended Data Fig. 1a–c). Projections originating from the piriform cortex (PIR), anterior basolateral amygdala (BLAa; lateral part, BLAal)²¹ and posterior BLA (BLAp) densely distribute in DPs, while the BLAa medial (BLAam) and caudal (BLAac) parts generate axons in the DPd and ILA (Fig. 1a and Extended Data Fig. 1d–f). The cortical amygdala posterior lateral part (COApl) generates dense axons in DP layer 1 and the DPd but avoids DPs, while the COA posterior medial part (COApm) generates dense projections in the DPd (Extended Data Fig. 1g), illustrating the clearly distinguishable DP laminar organization.

Connectionally defined IT, CPT and CT DP neurons

Using multifluorescence retrograde tracing, we found that intratelencephalic (IT) cortical projection neurons like those that target the ACAd and anterior COA (COAa) primarily distribute in the DPs (Fig. 1c and Extended Data Fig. 1h,i). By contrast, corticothalamic (CT) neurons projecting to the RE and parataenial thalamic (PT) nuclei are exclusively in the DPd (Fig. 1c and Extended Data Fig. 1j (injection sites))²². Hypothalamic and brainstem projecting corticofugal pyramidal tract (CPT) neurons display target-specific distribution patterns. Neurons that project to the medial preoptic area (MPO) are located primarily in the DPd, whereas those projecting to the lateral hypothalamic area (LHA) distribute in both the DPd and deeper portions of the DPs (Fig. 1c and Extended Data Fig. 1k–l).

Molecularly defined DP cell types

We investigated the expression patterns of over 100 genes from the Allen Brain Atlas (ABA) database, to examine their regional and laminar DP patterns. RNAscope analysis of the select marker genes Cux2, Etv1 and Tle4 confirmed specificities of DP neuron types (Fig. 1d and Extended Data Fig. 2a). Our analysis revealed several key findings (Fig. 1d and Extended Data Fig. 2b–g): (1) marker genes for cortical layer 5 (L5) IT cells, Cacna1h and Plxnd1 (ref. ²³), are expressed in the DPs but not in the DPd, suggesting that DPs neurons share similar cell-type-specific molecular features of L5 IT neurons; (2) the L6 CT or L6 CT/IT marker genes Tle4 and Pamr1 are expressed in the DPd; (3) the L5 CPT marker genes Etv1 (also known as Er81) and Fezf2 (also for L5 IT cells)²³ are expressed in both the DPd and the DPs. Cortical L2/3 marker genes such as Cux2 and TTd marker genes such as Lrmp2 do not extend into the DP. Additional genes such as Htr2c and Fxyd6, which are uniquely expressed in the DPs and DPd, respectively, lend further evidence for their distinction. Combining retrograde tracing with RNAscope demonstrated that a large population of CPT cells in the DPd expressing Fezf2 project to hypothalamic structures such as the MPN, LHA and dorsomedial hypothalamic nucleus (DMH), and hypothalamic paraventricular nucleus (PVH) (Extended Data Fig. 3). Finally, we identified a unique population of neurons that expresses Vglut2 in the DPd, but not in the DPs or other MPF areas (Fig. 1d).

In summary, the DPs is characterized by L5 IT-type neurons that project to cortical areas and to the amygdala, and CPT-like neurons that project to the LHA (Fig. 1e). However, the DPd predominantly contains CT- and CPT-type neurons, with the latter projecting to hypothalamic periventricular structures such as the MPO, as well as brainstem structures such as the parabrachial nucleus (PB)¹⁷.

Morphological DP cell types

Studies examining the morphological features of MPF neurons²⁴ have excluded the DP. We used the genetic sparse labelling MORF3 (ref. ²⁵) reporter line, brain clearing and 3D microscopy imaging to acquire and systematically categorize morphological features of sparsely labelled excitatory DP projection neurons (Fig. 1f, Extended Data Fig. 4 and Methods). In total, 47 reconstructed DP neurons were categorized into two morphological clusters on the basis of dendritic size (quantified by dendritic length) and complexity (quantified by number of branches). Cluster 1 included smaller, less-complex neurons (32 neurons, average total dendritic length, 3,024 ± 833 µm; average number of branches, 38 ± 12). Cluster 2 contained larger, more-complex dendritic arbours (15 neurons; average total dendritic length, 5,821 ± 958 µm; average number of branches, 76 ± 13). Next, DP neurons were classified into superficial- versus deep-layer clusters on the basis of the distances of their cell bodies from the midline. Comparing deep (21 neurons; distance from the midline, 605 ± 106 µm) and superficial (26 neurons; distance from the midline, 288 ± 81 µm) clusters revealed that deep neurons were smaller and less complex (average length, 3,472 ± 1,285 µm; average number of branches, 42 ± 16), whereas the arbours of superficial neurons were larger and more complex (average length, 4,275 ± 1,717 µm; average number of branches, 56 ± 24). In total, 86% of the deep neurons (18 out of 21) fell within the smaller and less complex morphological class. By contrast, superficial DP neurons fell across both (larger–complex and smaller–simpler) morphological clusters. Overall, superficial layer neurons were larger (one-tailed Wilcoxon signed-rank test, false-discovery rate (FDR)-corrected P = 0.0465) and more complex (one-tailed Wilcoxon signed-rank test, FDR-adjusted P = 0.0317) compared with deep cluster neurons. They also had a greater proportion of the dendritic arbour between the cell body and the midline (one-tailed Wilcoxon signed-rank test, FDR-adjusted P = 0.0261). These results align with the cytoarchitectonic data suggesting distinct cell-type-specific morphological features of DPs and DPd neurons.

Distinct neural network features of DPs and DPd

Extensive research has been conducted on MPF connectivity in rats and primates^4,26,27. In mice, investigations have focused on cortical and thalamic MPF connectivity at regional, cell-type-specific and single-neuron resolutions^{24,28,29,30,31}, although the global neural network of individual MPF regions, including the DP, has not been systematically generated.

Our data production and annotation pipelines were implemented to investigate the brain-wide input/output connectivity of the DPd, DPs, ILA, PL, ACAv and ACAd. Anterograde (AAV, PHAL, BDA) and retrograde (CTB, Fluorogold (FG)) tracer injections were placed into each MPF area (Methods, Extended Data Fig. 5 and Supplementary Methods). Image data were registered, and tracer labels were thresholded and annotated using our custom software, Outspector^21,32,33,34 (Extended Data Fig. 6a–e, Supplementary Figs. 2–4 and Supplementary Methods; the specificity and accuracy of the Outspector pipeline is shown in Supplementary Fig. 5).

To ensure consistent labelling, 2D hierarchical clustering was performed on data from multiple injections into individual regions of interest (ROIs; Extended Data Fig. 6f). Annotated data were normalized and analysed, with visualizations of anterograde, retrograde and reciprocal connections presented in connectivity matrices. Aggregated anterograde and retrograde connectivity data from select cases with injections in each MPF area—DPs (n = 4 (anterograde) and n = 3 (retrograde)), DPd (n = 2 and n = 1), ILA (n = 3 and n = 2), PL (n = 3 and n = 3), ACAd (n = 3 and n = 3) and ACAv (n = 3 and n = 3)—were quantitatively analysed to construct 2D hierarchical clustering for the projection fraction matrix, input fraction matrix and reciprocity fraction matrix for each MPF area (Supplementary Methods). Accordingly, we constructed comprehensive global input/output MPF connectivity matrices (Extended Data Fig. 7) and MPF connectivity flatmaps (Extended Data Fig. 8) that illustrate the distinct global input/output connectivity patterns of each MPF area. An online tool for viewing all reconstructed connectivity maps is also available (https://brain.neurobio.ucla.edu/mpf/). An additional 96 tracer-injection experiments targeting the cortex, hippocampus, amygdala, thalamus, hypothalamus and brainstem were performed to ensure comprehensiveness, cross-validate anterograde and retrograde datasets, and to address potential tracer tropism issues (the specificity and reliability of the tracing data is shown in Extended Data Fig. 9 and Supplementary Fig. 6). Moreover, AAV1-Cre-based trans-synaptic tracing and TVA-mediated rabies tracing were conducted to delineate multisynaptic and cell-type-specific MPF networks.

DPs and DPd receive distinct cortical inputs

Retrograde tracer injections made into either the DPs or DPd revealed their distinct cortical inputs (Fig. 2a–c, Extended Data Figs. 9 and 10, Supplementary Fig. 7a–f and Supplementary Results), which were all validated with anterograde tracer injections made into their upstream source structures (Extended Data Fig. 11a). Cortical inputs to the DPs and DPd can be categorized as follows: (1) both the DPs and DPd receive dense inputs from olfactory cortical areas and the COA. The DPs receives much denser inputs from the PIR and dorsal endopiriform nucleus (EPd), which processes main olfactory information, while the DPd, but not the DPs, receives dense input from the COApm and moderate input from the posterior amygdalar nucleus (PA). The COApm and PA receive pheromonal information directly from the accessory olfactory bulb (AOB) and are involved in sexual and social behaviours^35,36. (2) Inputs from the ENTl: the DPs receives inputs primarily from ENTl L2/3, while ENTl L5 neurons (not L2/3) robustly project to the DPd and ILA²⁸ (Fig. 2a–c, Extended Data Fig. 10a,b and Supplementary Fig. 7d,e). (3) Inputs from the temporal association (TEa), ectorhinal (ECT) and perirhinal (PERI) cortical areas, together termed the rhinal cortex³⁷: these areas receive multimodal information from somatosensory and motor cortical areas²⁸ and generate direct projections to both the DPs and DPd (Fig. 2a,b, Extended Data Fig. 10a,b and Supplementary Fig. 7d). (4) Input from the AUDv (Figs. 1b and 2a, Extended Data Fig. 10a,b and Supplementary Fig. 7c), which is responsible for processing auditory information associated with fear conditioning³⁸, is primarily to the DPs. (5) The claustrum (CLA), which projects to the DPd, but not the DPs, shares massive bidirectional connections with the ILA and PL²⁸ (Extended Data Figs. 9g,h and 10a,b and Supplementary Fig. 6g,h and 7b). (6) Inputs from the subiculum and CA1: consistent with our previous report³³, anterograde tracer injections into the prosubiculum (ProSUB) resulted in dense axonal terminals in the DPs (Fig. 1c), while anterograde tracer injections into the ventral subiculum (SUBv) and CA1 resulted in axonal terminals in the DPd and ILA (Fig. 2a and Extended Data Fig. 10a,b). (7) Inputs from the BLAa, which is integrally involved in fear conditioning³⁹: the BLAal and BLAp, which integrates olfactory and visceral information, project to the DPs, while the BLAac and BLAam project to the DPd and ILA²¹ (Figs. 1a and 2a,b and Extended Data Figs. 10a,b and 11b). (8) The dorsal, ventral and posterior agranular insular areas (AId, AIv and AIp), which process gustatory and visceral sensory information⁴⁰, project to the DPd, DPs and ILA (Fig. 2a,b, Extended Data Fig. 10a,b and Supplementary Fig. 7b). (9) In contrast to the DPs, the DPd also receives input from subcortical structures including, but not limited to, the lateral septum (LS) (Extended Data Fig. 10c).

Fig. 2: Cortical input/output organization of the DPs, DPd and other MPF areas.

a, Distribution map of cortical neurons projecting to the DPs, DPd and other MPF areas from retrograde tracing experiments, revealing diverse cortical inputs (left). Right, quantified cortical inputs to the DPs and DPd (representative CTB-labelled neurons are shown; the full datasets and connectivity map are shown in Extended Data Figs. 9 and 10 and Supplementary Fig. 7; all MPF cases are available at https://brain.neurobio.ucla.edu/mpf/; validation of the connections is provided in Extended Data Fig. 11a,b). b, Summary of inputs to the DPs and DPd from cortical, amygdalar and hippocampal regions. c, Anterograde tracing confirmed distinct ENTl inputs: Cux2-positive L2/3 neurons project to the DPs, while L5/6 neurons target the DPd. d, Quantitative comparison of DPs outputs to, and inputs from, other MPF (ILA, PL, ACAv, ACAd) and cortical (medial orbitofrontal area (ORBm), MOp, MOs) areas (top). Middle, PHAL-labelled axons from the DPs travel through layer 1 and deep layer 5, terminating densely in the MPF, MOp, MOs and association cortices (for example, PTLp, RSPd, visual areas; Extended Data Fig. 11e). Bottom, by contrast, retrograde CTB labelling reveals sparse DPs input from these regions (see Extended Data Figs. 5a,b and 10a for PHAL and CTB injections). These data highlight the primarily unidirectional DPs to cortex connections. e, Schematic of the DP as a network hub relaying inputs from cortical areas within the lateral cortical network²⁸, olfactory cortex (OLF), amygdala (AMY) and hippocampus (HPF) to other MPF and cortical areas within the medial cortical network²⁸ and somatic sensorimotor cortices. f, 2D hierarchical clustering of the MPF to cortex projection fraction matrix reveals distinct clusters of cortical targets for the DPs, DPd and other MPF areas (based on anterograde tracer injections made into each MPF area). Prefrontal (PFC) regions are grouped to the left of the grey line; additional clusters are boxed in colour. Further details are provided in the Supplementary Results. Scale bars, 200 μm (d (right)), 300 μm (c (top)) and 500 μm (a, c (bottom) and d (left and middle)). A full list of abbreviations is provided in Supplementary Table 1.

Source data

DP projections to the MPF and cortical areas

The DPs and DPd generate minor projections back to regions that heavily innervate them, such as olfactory cortex (Supplementary Fig. 6a–f). Instead, the DPs projects densely to other MPF areas, including the ILA, PL, ACAv and ACAd, which, except for the ILA, send only light projections back to the DP (Fig. 2d–f and Supplementary Fig. 7a). This is atypical for MPF structures, which are predominantly reciprocally interconnected (Extended Data Fig. 11c,d). Moreover, the DPs projects substantially to the secondary motor (MOs) and primary motor (MOp) domains involved in controlling whisker, upper limb and trunk movements, as well as to the dorsal retrosplenial (RSPd), posterior parietal (PTLp) and anteromedial visual (VISam) cortical areas (Fig. 2e,f, Extended Data Fig. 11e and Supplementary Fig. 7f). Although these cortical structures share reciprocal connections with the ACAd, ACAv, PL and ILA²⁸, they send little to no projections to the DP (Supplementary Fig. 8). Further details are provided in the Supplementary Results.

Together, the data show that the DP mediates a predominantly unidirectional flow of cortical information—a distinguishing DP feature. It receives convergent inputs from olfactory areas, amygdala, hippocampus, ENTl and other areas along the lateral cerebral mantle such as the AUDv and AIp and, in turn, densely innervates other MPF and cortical areas along the medial neocortex (Fig. 2e and Supplementary Fig. 8). Supporting this notion, TVA-receptor-mediated cell-type-specific rabies viral tracing demonstrated that DPs to ACA-projecting neurons receive monosynaptic inputs from the AON, PIR, SUB, CA1 and ENTl (Extended Data Fig. 11f). Further details are provided in the Supplementary Results.

DP connections with midline thalamic nuclei

The MPF is known for its reciprocal connections with the mediodorsal thalamic nucleus (MD) as well as other medial and midline thalamic nuclei⁴, although the thalamic connectivity of the DP has not been investigated. Comparative distributions of thalamocortical projection neurons and corticothalamic projection terminals for MPF regions are displayed in connectivity maps (Extended Data Fig. 12a; https://brain.neurobio.ucla.edu/mpf/). Quantitative analyses of all thalamic connectivity are visualized in matrices (Extended Data Fig. 12e and Supplementary Fig. 9; additional details are provided in the Supplementary Results).

The DPs (layer 1) receives extensive input from the RE (Extended Data Fig. 12b; validation of RE to DP projections is shown in Fig. 1b and Extended Data Fig. 1c), which accounts for 76.79% of its total thalamic input. The DPs does not provide significant thalamic projections given its lack of CT neurons. By contrast, the DPd bidirectionally connects with several midline and medial thalamic nuclei, including the RE, paraventricular (PVT), parataenial, medial part of the mediodorsal (MDm), rhomboid (RH) and intermediodorsal (IMD) thalamic nuclei (Extended Data Fig. 12c and Supplementary Fig. 9b). Quantitative analysis revealed that the DPd and ILA share similar thalamic connections specifically with the RE, PT, PVT and MDm, while the PL, ACAd and ACAv demonstrate distinct connections with different clusters of thalamic nuclei (Extended Data Fig. 12e, Supplementary Fig. 9a,b and Supplementary Results).

Anterograde/retrograde tracer co-injections were made into the PVT, PT or RE (Extended Data Figs. 12f and 13a,b), which validated thalamocortical and corticothalamic connections, but also identified information relayed to the DP and ILA through these thalamic nuclei. The PVT bidirectionally connects with structures in a network that regulates visceral sensorimotor activities (Extended Data Fig. 12f,g). These include (1) the PB and nucleus of the solitary tract (NTS), the primary structures that process visceral sensory inputs. Notably, the PB also generates direct projections to the ILA⁴¹ (Supplementary Fig. 10); (2) the median preoptic nucleus (MEPO), which processes information related to fluid balance, thirst and cardiovascular function⁴²; (3) the agranular insular areas (AId, AIv, AIp), central nucleus of the amygdala (CEA), bed nuclei of the stria terminalis (BST) and PAG, all of which are involved in autonomic function control⁴³.

By contrast, the RE and PT bridge the DP and ILA with hypothalamic networks that govern two basic classes of social behaviour (namely, reproductive and defensive) as well as with structures implicated in spatial navigation and conditioned memory (Extended Data Figs. 12f,g and 13a,b). Given that the RE is a primary source of thalamic input to the DPs, we used a TVA-receptor-mediated rabies tracing technique to further examine which structures provide monosynaptic inputs to the RE to DP neurons (Extended Data Fig. 13c,d). Additional details are provided in the Supplementary Results.

DP projects to structures regulating behaviour

DP projections to cerebral nuclei that control visceral motor activities

All neocortical areas, including MPF areas, except for the DP, project heavily to the caudoputamen (CP)³². Instead, the DPs exhibits highly dense projections to the CEA and substantia innominata (SI), and moderate projections to the anterolateral division of the BST (BSTal) (Extended Data Fig. 14a–e), all of which are extensively interconnected in a core network that regulates autonomic function⁴⁴. By contrast, the DPd densely projects to the anteromedial BST (BSTam) (Extended Data Fig. 14a,d,e), which regulates neuroendocrine activities of the hypothalamic–pituitary–adrenal axis in response to stress^1,45,46. The DPd and ILA provide significantly weaker projections to the CEA and, instead, topographically project to the medial amygdalar nucleus (MEA), posterior division of the BST (BSTp) and hypothalamic medial nuclei (Extended Data Fig. 14a,c,e), which form two parallel subnetworks that govern reproductive and defensive behaviours^36,43. Like the ILA, the DPs and DPd also topographically project to the olfactory tubercle (OT) and nucleus accumbens (ACB) (Extended Data Fig. 14b–d), known for their roles in reward and social interactions⁴⁷.

DPs projections to preautonomic hypothalamic structures

The MPF coordinates autonomic, neuroendocrine and behavioural responses that are important for maintaining emotional equilibrium^1,2,3. The DPs exhibits extensive projections that innervate the descending parts of the PVH (PVHd), namely the lateral parvicellular (PVHlp) and forniceal (PVHf) (Fig. 3a (i and iii) and Extended Data Fig. 15a,b). These regions contain VGLUT2-positive preautonomic neurons that send descending projections to the spinal cord’s intermediolateral (IML) column (Fig. 3a (ii and iv)) for sympathetic output control^43,48. Moreover, the DPs generates dense projections to the LHA (Fig. 3a (v) and Extended Data Figs. 15a,c and 16a), which also contains preautonomic neurons that project to the spinal cord (Fig. 3a (v and vi)), suggesting that DPs neurons probably innervate spinal-projecting LHA neurons.

Fig. 3: DP projections to brain structures regulating autonomic outputs.

a, The DPs innervates two parts of the PVHd (1–3), marked by VGLUT2 expression (2; Bright-field in situ hybridization images from the ABA (https://Brain-map.org)): the PVHlp and PVHf, which contain spinal-projecting neurons labelled by CTB spinal cord injection (4; see Extended Data Fig. 15a for PHAL injection). DPs axons also densely innervate the LHA (5), which contains CTB-labelled spinal-projecting neurons (6). fx, fornix; mtt, mammilothalamic tract; v3, third ventricle. b, In MORF3 mice, AAV-RFP was injected into the DP (1 and 2) and AAVretro-Cre was injected into the spinal cord to identify potential synaptic contacts between DP axons and hypothalamic spinal-projecting neurons. Cre-induced MORF3 expression revealed the morphology of spinal-projecting neurons (3). High-resolution light-sheet (1–4 and 7) and confocal (5, 6 and 8–12) images show close appositions between DP terminals and MORF3-labelled somas and dendrites in PVHd (6) and LHA (7–12). The blue arrowheads mark corresponding neurons across magnifications; the white arrowheads indicate putative axodendritic contacts (Supplementary Video 1). c, AAV1-Cre trans-synaptic tagging and Cre-dependent anterograde tracing confirmed the DP to LHA to IML column (sympathetic preganglionic neurons) pathway. AAV1-Cre DP injection (top left) transported Cre to postsynaptic LHA neurons, which were labelled through Cre-dependent AAV-GFP (bottom left). Middle, GFP-labelled axons in the IML (thoracic levels T1–T2). Right, magnified images showing terminal boutons apposed to neurons in the IML (1–3) and near the central canal (4). d, Pathways through which the DPs regulates vagal parasympathetic output. AAV1-Cre DPs injection (1) anterogradely transported Cre to the CEA, where Cre-dependent AAV-GFP was injected (2) to label axons from Cre-positive postsynaptic neurons (3–5). CEA starter cells co-expressed Cre and GFP (2). GFP-labelled axons in the DMX and NTS (5), confirming the DPs to CEA to DMX pathway. Projections targeted other structures that regulate autonomic function, for example, BST and SI (3), PB (4), PARN and IRN (5). Image 6 shows DMX/NTS labelling from non-selective CEA neurons. V, trigeminal motor nucleus. e, Schematic of the DPs, DPd and ILA in a cerebral network regulating autonomic outputs. f, Direct projections to Barrington’s nucleus (B) from the ILA (left) and DPd (right; green), indicating disynaptic MPF connections controlling lumbosacral parasympathetic outputs (DPd/ILA to Barrington’s nucleus to spinal cord). Note that PL also projects to Barrington’s nucleus (right; red). Scale bars, 5 μm (b (10 and 11)), 7 μm (b (12)), 10 μm (b (9)), 20 μm (c (right)), 30 μm (b (6)), 100 μm (c (top left, inset)), 200 μm (a and c (bottom left and middle)), 300 μm (b (2–5, 7 and 8)), 500 µm (d and f) and 1 mm (c (top left, main image)). A full list of abbreviations is provided in Supplementary Table 1.

To test this, in MORF3 mice²⁵, AAV-RFP was injected into the DP to label its axonal projections, while AAVretro-Cre was injected into the spinal cord (Fig. 3b (i and ii)). Cre was retrogradely transported to spinal projecting neurons and stochastically unlocked MORF3 expression. This revealed the dendritic morphology of MORF3-labelled PVHd and LHA neurons, which heavily intermingled with axonal terminals arising from the DP (Fig. 3b (iii–v, vii and viii)). Despite MORF3-labelled neurons accounting for only a small fraction (3–5%) of total spinal projecting neurons owing to the sparse labelling MORF3 properties²⁵, significant numbers of close appositions were observed between DP axon terminals and MORF3-labelled somas and dendrites (Fig. 3b (vi and ix–xii) and Supplementary Video 1), indicating putative synaptic connectivity through which the DP innervates LHA spinal projecting neurons.

Next, using AAV1-Cre anterograde trans-synaptic tagging alongside Cre-dependent AAV anterograde tract tracing, we confirmed that the postsynaptic DP-input-recipient neurons in the LHA project directly to IML preautonomic neurons in the spinal cord (Fig. 3c, Extended Data Fig. 16b1–b3; see Supplementary Fig. 11a for the unidirectional DP to LHA connection). Meanwhile, those DP-input-recipient LHA neurons also generate dense projections to the SI, lateral habenular nucleus and brainstem structures such as PAG, PB and nucleus raphe magnus (RM) (Extended Data Fig. 16b4–b19), all of which directly regulate autonomic function. Additional details are provided in the Supplementary Results.

DP projections to autonomic brainstem structures

In rats^26,27, the ILA sends direct excitatory projections to the dorsal motor nucleus of the vagus nerve (DMX), which controls vagal parasympathetic output. Our findings in mice confirmed that the ILA, but not the DP, projects directly to the DMX (Extended Data Fig. 16c). Instead, the DPs projects to the CEA (Extended Data Fig. 14a,c,d), which in turn projects to the DMX. Using AAV1-Cre trans-synaptic tagging combined with Cre-dependent AAV-GFP anterograde tracing, we verified that CEA neurons receiving direct inputs from the DP display dense axonal terminals in the DMX (Fig. 3d; DP to CEA projections are unidirectional; Supplementary Fig. 11b). Considering the CEA’s GABAergic nature, these data establish a disynaptic pathway, DPs to CEA to DMX, through which the DPs can inhibit vagal parasympathetic outputs. Notably, DPs-input-recipient CEA neurons generate extensive projections to the BSTal, PB and NTS (Fig. 3d). The PB and NTS are primary structures that process visceral sensory information. While the ILA generates dense projections to both structures, the DPd also projects to the PB⁴¹ (Supplementary Fig. 10). Together, the ILA, DPs, CEA, BSTal, PB and NTS form a network that regulates vagal parasympathetic output (Fig. 3e and Extended Data Fig. 16d).

Furthermore, the ILA and DPd project to Barrington’s nucleus (Fig. 3f), which directly innervates parasympathetic motor neurons in the lumbosacral spinal cord that generate the pelvic nerves to regulate micturition, defecation and penile erection⁴⁹. Finally, like other MPF components⁴, the ILA, DPs and DPd generate topographic projections to the PAG (Supplementary Fig. 12) through which they can regulate autonomic and behavioural outputs. Further details are provided in the Supplementary Results.

DPd projections to the hypothalamic neuroendocrine zone

The prevailing understanding is that the MPF (ILA and PL) does not directly innervate hypothalamic neuroendocrine cells, but instead regulates neuroendocrine activities indirectly through the BST and other structures^46,50. Our data confirm the lack of direct projections from ILA and PL to the hypothalamic neuroendocrine zone. However, we show that the DPd sends direct and extensive projections to the entire hypothalamic neuroendocrine zone along the third ventricle (Fig. 4a,b and Extended Data Fig. 17a–d). Hypothalamic nuclei within this zone, such as the PVH, supraoptic nucleus (SO) and arcuate nucleus (ARH), contain various parvo- and magnocellular neurosecretory neurons^43,48 (Extended Data Fig. 17d).

Fig. 4: DPd projections to the hypothalamic neuroendocrine zone.

a, The DPd generates dense terminals throughout the hypothalamic periventricular zone, which houses neuroendocrine neurons. BSTif, bed nuclei of the stria terminalis, interfascicular nucleus. See Supplementary Table 1 for a full list of abbreviations. b, MPF projections to the PVH (neuroendocrine, preautonomic) and PV (neuroendocrine). The Pareto chart shows the DPd as main input source to both nuclei. The curved line represents cumulative projections. DPs projections to the PVH target the PVHd. c, DPd labelled neurons after PVH CTB injection confirms the direct DPd to PVH connection. d, AAV-FLEX-RFP DP injection in VGLUT2 Cre mice shows projections to hypothalamic neuroendocrine cells (labelled through tail-vein (intravenously, iv) FG injection; middle and right). DPd axon terminals formed close appositions with neuroendocrine somatostatin (SS)-positive intermediate periventricular nucleus (PVi) and PVHpv neurons and CRH PVHmpd neurons. e, AAV-synaptophysin-GFP DP injection in CRH-Cre Ai14 mice. High-resolution ×60 images show that GFP labelled DP terminals (green) form close appositions with CRH PVH neurons (red) across planes. f, Schematic showing how the DP, MEA and BST regulate hypothalamic neuroendocrine outputs directly through the periventricular zone or indirectly through the AVPV, MEPO, MPNm and DMH. g, In CRH-Cre mice (n = 5), ChR2 was injected into the DP, while Cre-dependent AAV-RFP was injected into PVH (left). Middle, DP ChR2 injection and CRH-positive PVH neurons. The boxed region shows a representative biocytin-labelled recorded CRH neuron. The magnified images show co-localized biocytin (left) and CRH (right). Right, the distribution of peak amplitude responses evoked by optogenetic stimulation of DP inputs (12 out of 14 neurons responded). The box boundaries mark the 25th and 75th percentiles, the centre line shows the median and the whiskers (error bars) indicate the 90th and 10th percentiles. The outlying points (dots) are shown. The trace (right) shows a representative CRH neuron response. oEPSC, optically evoked excitatory postsynaptic current; vh, holding potential. h, AAV1-Cre was injected into the SUBv and BLAa to trans-synaptically deliver Cre to DPd neurons, subsequently targeted with Cre-dependent AAV-FLEX-ChR2 (AAV-mCherry for the control). Optogenetic DPd neuron stimulation induced FOS in the PVHmpd. The top micrographs show the fibre tip and Cre, ChR2 and FOS expression. Bottom, ChR2⁺ axons and FOS induction in the PVHmpd across ARA levels. i, Plasma CORT levels were significantly elevated in the ChR2 group (n = 6) compared with the controls (n = 9). Statistical analysis was performed using a two-sided unpaired t-test (t₁₃ = 3.379, P = 0.0049) (left) and a Mann–Whitney U-test (U = 4, P = 0.0048) (right). Data are mean ± s.e.m. Scale bars, 10 μm (d (right), e (top right) and g (bottom)), 20 μm (e (middle right and bottom) and g (top right)), 30 μm (e (middle left)), 50 μm (e (top left)), 100 μm (h (bottom)), 150 μm (g (top middle)), 200 μm (c, d (left and middle) and h (top right)), 500 μm (a and h (top left)) and 700 μm (g (top left)).

Source data

Retrograde tracer injections in the PVH (Fig. 4c) and ARH (Extended Data Fig. 17e) confirmed these connections and showed retrogradely labelled neurons only in DPd and none in the ILA or other MPF areas. Using VGLUT2-Cre mice, we showed that DPd VGLUT2-positive neurons directly innervate hypothalamic parvocellular and magnocellular neuroendocrine cells, which were labelled through an intravenous FG injection⁴⁸. Axon terminals originating from VGLUT2 DPd neurons intermingled with FG-labelled cells in the PVH, SO and periventricular nucleus (PV) (Fig. 4d and Extended Data Fig. 18a–c), suggesting putative synaptic connections between DPd axonal terminals and neuroendocrine cells. Using RNAscope in the same animals, we confirmed that the preoptic (PVpo) and anterior (PVa) PV and the periventricular part of the PVH (PVHpv), of which the neurons synthesize SS, receive dense input from the DPd (Fig. 4d). Importantly, the dorsal medial parvicellular part of the PVH (PVHmpd), which contains CRH- and TRH-synthesizing neurons, receives substantial input from DPd VGLUT2-expressing neurons (Fig. 4d).

To confirm direct innervation of CRH neurons by DP axons, we injected an AAV1-synaptophysin-GFP anterograde tracer into the DPd, labelling its axonal terminals in the PVH of CRH-Cre × Ai14 mice in which CRH neurons express tdTomato (Fig. 4e and Methods). High-resolution ×60 confocal imaging of the PVH revealed numerous GFP-labelled axonal boutons in close apposition to CRH neurons (Fig. 4e). Furthermore, cell-type-specific monosynaptic rabies tracing in CRH-Cre mice identified back-labelled neurons in the DPd to confirm direct monosynaptic input from the DPd to PVH CRH neurons (Extended Data Fig. 18e).

Within the ARH, a highly dense DPd axonal terminal plexus was concentrated dorsomedially (Fig. 4a), where dopamine neuroendocrine motoneurons are located. Comparatively, lighter inputs from the DPd were observed in ventrolateral ARH regions, where neurons synthesizing growth-hormone-releasing hormone predominate. Moreover, immunostaining confirmed that DPd axons form putative synaptic connections with magnocellular oxytocin-positive neurons in the PVH and SO (Extended Data Fig. 18d).

The DPd densely innervates the BSTam (Fig. 4a and Extended Data Fig. 14a,e). In turn, the BSTam provides extensive projections to the entire hypothalamic neuroendocrine zone⁴⁵ (Extended Data Fig. 18f). Both the DPd and BSTam send dense projections to other hypothalamic structures, including the anteroventral periventricular (AVPV), MEPO, medial preoptic (MPN) and dorsomedial (DMH) hypothalamic nuclei and MPO (Fig. 4a and Extended Data Fig. 17a,f,g). These structures, which are components of the hypothalamic visceral motor pattern generator⁴³, densely project to the hypothalamic neuroendocrine zone (Extended Data Fig. 19a) and other structures involved in autonomic function (for example, the PVHlp, LHA, PAG, PB and Barrington’s nucleus) and social behavioural activities (for example, hypothalamic ventromedial, ventral and dorsal premammillary nuclei)⁴³. Together, the DPd, BSTam and these hypothalamic structures form a network that regulates neuroendocrine and motor responses underlying social behaviour (Fig. 4f and Extended Data Fig. 19b). Additional details are provided in the Supplementary Results.

DPd neurons regulate neuroendocrine responses

We validated functional monosynaptic connections from the DPd onto CRH neuroendocrine cells by performing single-cell electrophysiological recordings using channelrhodopsin-assisted circuit mapping (CRACM) combined with anterograde trans-synaptic tracing (Fig. 4g and Methods). Cre-dependent AAV1-expressing tdTomato was injected into CRH-Cre mice to label PVH CRH neurons, while AAV-ChR2-YFP was injected into the DPd to label its axon projections. After 4 weeks, 350-µm brain slices containing the PVH were prepared to patch and record from tdTomato-labelled CRH neurons, with optical stimulation of ChR2-labelled axonal terminals originating from the DPd. The data showed that 12 out of 14 tdTomato-labelled PVH neurons responded to channelrhodopsin stimulation, with an average response amplitude of 83 ± 21 pA (Fig. 4g), confirming monosynaptic inputs from the DPd onto PVH CRH neurons.

We next examined which neural inputs might drive the DPd to regulate motor outputs of CRH neuroendocrine cells as indicated by changes in corticosterone (CORT) levels. We focused on two upstream structures: the SUBv, which is associated with psychological stress and emotional regulation¹, and the BLAa, which is implicated in fear conditioning. Using a combination of AAV1-Cre anterograde trans-synaptic tagging and multifluorescence anterograde tract tracing, we found that postsynaptic DPd neurons receiving inputs from the SUBv and BLAa project extensively to the PVH (Extended Data Fig. 20a–h; unidirectional BLA/SUBv to DP connections are shown in Supplementary Fig. 11d,e). Our findings also suggest that inputs from the SUBv and BLAa potentially converge onto the same DPd projection neurons (Extended Data Fig. 20i–n). Together, these data validate a disynaptic SUBv/BLAa to DPd to PVH circuit.

To functionally confirm this hypothesis, we administered AAV1-Cre into the SUBv and the BLAa, and Cre-dependent AAV-ChR2 into the DP. As a result, postsynaptic DP neurons that receive inputs from the SUBv and BLAa would express ChR2 and send downstream projections to the PVH (Fig. 4h). Optogenetic activation of these specific DPd postsynaptic neurons led to a significant increase in plasma CORT levels compared with the control group (P = 0.0049, unpaired t-test; P = 0.0048, Mann–Whitney U-test) (Fig. 4i). This evidence of a functional circuit suggests that DPd neurons can directly modulate plasma CORT levels (and potentially other neuroendocrine activities) in response to signals from the SUBv and BLAa. Consistent with our results, a recent study also showed that chemogenetic stimulation of DP/TTd results in increased plasma CORT¹⁴.

Defining the functional relevance of DP neurons

We showed that the DP receives an array of cortical and thalamic inputs and, in turn, regulates neuroendocrine, autonomic and behavioural responses to environmental stimuli. To further validate this hypothesis and explore its functional relevance, we conducted several different experiments.

DP neurons receive convergent sensory and contextual memory synaptic inputs

The DP is the only prefrontal cortical area that receives converging input from olfactory (PIR and EPd), auditory (AUDv) and lateral entorhinal (ENTl) cortices (Fig. 1b and Extended Data Fig. 1a,b,d). The integration of inputs from these cortical areas is important because olfactory and auditory signals are important for social cues and for responding to predators, while the ENTl provides spatial and contextual memory information^36,38.

To functionally validate the synaptic convergence of these neural inputs onto DP neurons, we conducted single-cell electrophysiological recording experiments using CRACM combined with anterograde trans-synaptic tracing (Methods). We injected AAV.ChR2 into the ENTl and AAV1-Cre into the AUDv (Fig. 5a; see Supplementary Fig. 11f for the unidirectional AUDv to DP connection). The AAV1-Cre anterogradely transported from the AUDv to DP to trans-synaptically infect postsynaptic DP neurons, which expressed tdTomato. Meanwhile, ChR2-labelled axons arising from the ENTl distributed in DP layer 1. Patch-clamp recordings of tdTomato-labelled DP neurons showed that nearly all of the neurons (n = 10 out of 11) responded to stimulation of ChR2 axons (Fig. 5b), suggesting that ENTl and AUDv inputs converge onto the same DP neurons. Using the same strategy, we demonstrated that DP neurons (n = 12 out of 17) receive convergent auditory (from the AUDv) and olfactory (from the PIR) inputs (Fig. 5b,c).

Fig. 5: The functional role of DP neurons in multimodal sensory integration and social behaviour regulation.

a,c, AAV1-Cre was injected into the AUDv, AAV-ChR2 into the ENTl (a; n = 6) or PIR (c; n = 5), and Cre-dependent AAV-RFP into the DP to reveal postsynaptic neurons receiving direct input from the AUDv. The magnified images show biocytin-labelled neurons recorded by patch clamp. b, The distribution of peak amplitude (amp) responses evoked by optogenetic stimulation of ENTl (top; 12 out of 17 neurons responded) or PIR (bottom; 10 out of 11 neurons responded) inputs (left). Right, representative responses from RFP-tagged DP neurons (see Extended Data Fig. 20o for isolated monosynaptic responses). The median peak current amplitudes evoked by optogenetic stimulation were not significantly different (ENTl, 218.3 pA, n = 12; PIR, 137.4 pA, n = 11; Mann–Whitney U-test, U = 61, P = 0.782). Whisker plot definitions are as described in Fig. 4g. d, Miniscope calcium imaging was performed during non-social (open-field, object exploration) and social (same-sex and opposite-sex) interactions synchronized with behavioural camera recording. e, Calcium imaging field of view and representative extracted traces. f,g, The percentage of time that mice (under miniscope analysis) spent interacting with non-social versus social targets. DP activity (f); ILA activity (g). The interaction time significantly differed across interaction type. Statistical analysis was performed using one-way ANOVA: DP: F(1.473, 5.154) = 13.47, *P = 0.0108; ILA: F(1.646, 8.230) = 6.883, *P = 0.0205; Tukey’s multiple-comparison test (family-wise error rate controlled) for DP: **P = 0.0051. Dots represent the n per group. Data are mean ± s.e.m. h,i, The percentage of behaviourally responsive cells during interactions with non-social (object) or social targets. DP activity (h); ILA activity (i). No significant differences for the DP or ILA were detected. j, The percentage of all behaviourally responsive cells (excited (E) + inhibited (I)). DP contained significantly more opposite-sex interaction responsive cells than object exploration responsive cells (Tukey’s multiple-comparison test: *P = 0.0168). Significantly more opposite-sex interaction-responsive cells were identified in the DP than in the ILA (Šidák’s multiple-comparison test with adjusted P values: **P = 0.0027). k, The frequency of calcium events during non-social (open-field test) and social (same-sex, opposite-sex) trials. In the DP, the calcium event frequency is significantly higher during same-sex interaction compared with during the open-field test (Tukey’s multiple-comparison test, *P = 0.0303). In the ILA, the frequencies of calcium events during both same-sex (Tukey’s multiple-comparison test, **P = 0.0086) and opposite-sex (Tukey’s multiple-comparison test, **P = 0.0094) interactions are significantly higher than that during the open-field test. Between the DP and ILA, the frequencies of calcium events during the open-field test (Šidák’s multiple-comparison test, *P = 0.0416) and same-sex interaction (Šidák’s multiple-comparison test, *P = 0.0199) are higher in the DP than in the ILA. Scale bars, 250 μm (a (top right) and e), 500 μm (a (bottom right)) and 1 mm (a (left)). A full list of abbreviations is provided in Supplementary Table 1.

Source data

DP neurons’ role in perceiving social and environmental stimuli

We explored the potential functional relevance of DP neurons in social cognition and interaction—an extensively studied MPF function^3,5,6,51. Social interactions necessitate the perception and integration of diverse neural inputs, including olfactory, pheromonal, auditory, visual and somatosensory cues. These inputs undergo instantaneous modifications through dynamic, mutual feedback between participants, as well as past social experiences³⁶. The DP receives and integrates information relayed from cortical and thalamic inputs and, in turn, projects to the MEA, BST, hypothalamic and other structures implicated in social behaviour. These connectivity data lead us to hypothesize that DP neurons have a role in perceiving and regulating social behaviour.

We performed calcium imaging using miniaturized microscopy in DP during both social and non-social behavioural contexts (Fig. 5d and Methods). We expressed GCAMP6f in VGLUT2-positive DP neurons in VGLUT2 Cre mice (Fig. 5e). In social trials, we monitored DP activity in mice engaged in interactions including chasing, grooming or mounting either with a same-sex or opposite-sex partner (Fig. 5f). In non-social trials, we recorded DP activity in mice within an open-field arena or during exploration of an object (Fig. 5f). We used receiver operating characteristic (ROC) analysis to identify DP neurons that were either significantly inhibited or excited during interactions with same-sex partners, opposite-sex partners or objects. Among the recorded DP neurons (6 male mice, 63.3 ± 9.9 per animal), a substantial proportion responded during either same-sex (6 male mice, 30.3 ± 4.0%) or opposite-sex (3 male mice, 42.2 ± 1.5%) social interactions (Fig. 5h–j). Notably, socially excited neurons outnumbered socially inhibited neurons (Fig. 5h and Extended Data Fig. 20p).

While we also observed that DP neurons respond to interactions with a non-social object (23.9 ± 2.8%), the frequency of calcium events during non-social behaviour trials was significantly lower compared with during social behaviour trials (same-sex interaction) (Fig. 5k; DP, object versus same-sex, *P = 0.0303; Tukey’s multiple-comparison test).

As a control, we examined ILA neuronal activity in response to the same social and non-social cues (6 male mice; average number of ILA neurons recorded, 133.5 ± 14.4). As the ILA lacks VGLUT2-positive neurons, we expressed GCaMP6f in VGLUT1-positive glutamatergic ILA neurons (Fig. 5e). The ILA displays distinct input/output patterns compared with the DP, notably lacking significant inputs from the COApm or PA, and its projections to the hypothalamus primarily target the anterior hypothalamic nucleus (AHN), which is involved in defensive behaviour⁴³. Notably, compared with the DP, the ILA contains significantly fewer opposite-sex interaction-responsive cells (Šidák’s multiple-comparison test, **P = 0.0027; Fig. 5g,i,j) and a lower frequency of calcium events during non-social open-field test (Šidák’s multiple-comparison test, *P = 0.0416) and same-sex social interaction (Šidák’s multiple-comparison test, *P = 0.0199) (Fig. 5g,k and Extended Data Fig. 20q).

Consistent with these findings, using FOS as a marker of neuronal activity in male mice, we show that DP neurons respond to the presence of male (intruders) and female mice, as well as pups (Supplementary Fig. 13a). Furthermore, DP neurons are activated by a variety of stressors and perceived threats, including acute-restraint stress, sudden looming sounds and foot-shock-based fear conditioning (Supplementary Fig. 13b,c). These findings, coupled with the broad range of neural inputs received by the DP, support its role in processing environmental cues that are important for social interactions and evaluation of threat. This ability is essential for assessing risks, making decisions and preparing for appropriate behavioural responses to safeguard well-being. Consistent with this hypothesis, recent research suggests that neurons in the DP/TTd are implicated in psychological stress resulting from social defeat^12,13, top-down control of defensive behaviour¹⁶, depression, anxiety and encoding of fear memories related to auditory cues^14,15, as well as opioid reward¹⁷.

Discussion

The concept of the primary visceromotor cortex

The distinct input/output connectivity pattern of the DP compared with other MPF areas and the TTd is striking. Furthermore, the DPs and DPd also exhibit distinct connectivity patterns. On the basis of these findings, we propose the concept of the primary visceromotor cortex (Fig. 6a). Although the notion of the visceromotor cortex has been speculated^4,9, our comprehensive MPF mapping determined its main components and their network organization. We propose that the DPd, DPs and ILA compose the core components of the primary visceral motor cortex, respectively regulating neuroendocrine, sympathetic and parasympathetic functions, thereby governing different aspects of visceral motor activities. This network model follows a similar organizational logic to that of the primary motor cortex (MOp) in its control of somatic movements of the head, neck and limbs through its direct innervations of motor neurons in the cranial motor nuclei and ventral horn of the spinal cord⁴³.

Fig. 6: Proposed working model of the MPF based on network analysis.

a, Schematic showing that the DPs, DPd and ILA form the primary visceral motor cortex, regulating neuroendocrine, sympathetic and parasympathetic outputs. Right, their network pathways across the whole mouse brain. The 3D brain outline is from the Allen Institute Common Coordinate Framework (CCF: https://alleninstitute.github.io/abc_atlas_access/descriptions/Allen-CCF-2020.html). b, 2D anatomic location of the DPs and DPd (left). The ARA digital atlas is available online (https://atlas.brain-map.org/). Middle, schematic and wiring diagram illustrating that the DP is a key node mediating a predominantly unidirectional cortical information flow from caudal to rostral and lateral to medial. The DP integrates convergent inputs from lateral cortical subnetworks, the olfactory cortex, amygdala and hippocampus, and relays this information to the MPF and other cortical areas within the medial cortical network and somatic sensorimotor areas. Right, 3D anatomical views of the DP, ILA and other MPF regions. The 3D brain outline is from the Allen Institute CCF as described in a. c, A proposed unified MPF model based on network analysis. The DP serves as an integrative centre, receiving comprehensive information from both external and internal environments. It transmits this information to other cortical areas including the ILA, PL, ACAv, ACAd, MOs and MOp. Each of these MPF areas, along with the MOp/MOs, carries out specific physiological and motor responses to various stress stimuli and has a role in regulating different goal-directed behaviours to ensure long-term homeostasis through their projections to different brain structures (see the main text for details). For example, the ACAd (and its adjacent MOs frontal eye field (MOs-fef)) and ACAv send dense projections to the dorsomedial striatum or the CP and SC, which are involved in coordinating eye and head movement during navigation^32,34 and attention³. The ACA and its subcortical targets, such as the SC, are also important for escape or prey-capture behaviours³⁴. These downstream effectors send projections to the thalamus, creating a feedback loop that allows for the regulation of MPF activities. This network model aligns well with the classic perception–evaluation–action model^3,5. Additional details are provided in the Supplementary Discussion. A full list of abbreviations is provided in Supplementary Table 1.

The DPd, DPs and ILA send direct projections to motor or premotor neurons that regulate neuroendocrine and autonomic functions. Notably, we identify monosynaptic projections from the DPd to neuroendocrine cells in the PVH and to the broader hypothalamic neuroendocrine zone (Fig. 6a), enabling direct cortical control of CORT release. The DPs densely innervates preautonomic neurons in PVHd and LHA, presumably regulating sympathetic output through a feedforward DPs to PVHd/LHA to IML pathway. The DPs also inhibits parasympathetic output through a disynaptic DPs to CEA to DMX/NTS pathway. By contrast, the ILA directly excites DMX neurons (ILA to DMX), thereby promoting vagal parasympathetic outputs. Together, the DPs and ILA generate direct cortical projections to brainstem and spinal preautonomic structures that regulate digestive, respiratory and cardiovascular functions. Notably, the ILA and DPd, but not the DPs, also directly innervate Barrington’s nucleus, which governs lumbosacral parasympathetic activities⁴⁹.

Similar to the way MOp regulates motor functions through its projections to the basal ganglia, the DPd, DPs and ILA generate dense projections in a topographic manner to several cerebral nuclei (striatal or pallidal-like structures with GABAergic projection neurons⁴³), such as the CEA, MEA, LS, ACB and BST, through which they regulate visceral motor activities, innate behaviour and reward. Among them, the BSTam, which receives dense inputs from the DPd, is known for its role in regulating neuroendocrine activities^43,45,46,50. The CEA and BSTal, which receive dense inputs from the DPs, have essential roles in regulating autonomic activities through their projections to the DMX and other brainstem structures^43,44. Moreover, the DPd and ILA generate dense projections to the MEA and BST posterior division, which in turn project to the hypothalamic medial behavioural column to regulate social behaviours^35,36,43. Like direct projections of the MOp to the subthalamic nucleus (STN) and superior colliculus (SC), the DPd, DPs and ILA also project densely to the LHA (in the vicinity of the STN) and PAG (in the vicinity of the SC), which contain preautonomic neurons that regulate cardiovascular, respiratory and other autonomic functions.

The DPd, DPs and ILA maintain bidirectional connections with several thalamic nuclei, including the RE, PVT and PT, which relay hypothalamic and brainstem information related to homeostasis (for example, visceral sensory input, fluid balance) and social behaviours (for example, defensive, reproductive)^35,36,43. The PVT and PT receive dense inputs from the PB and NTS—key regions that process visceral sensory information. Moreover, the PB exhibits connectivity with both the DP and ILA, as shown here and elsewhere⁴¹. Further details on DPs, DPd and ILA involvement in neural circuits regulating social behaviour are provided in the Supplementary Discussion.

A proposed unified MPF model for motivated behaviour

Our understanding of the prefrontal cortex has been hindered by the lack of a unified working model^3,5,7. On the basis of our newly constructed, comprehensive whole-brain MPF wiring diagram, we propose a testable network model to understand how different MPF components coordinate and synchronize into a unified system to regulate physiological and motor actions in response to environmental and social challenges.

The DP mediates a predominantly unidirectional cortical information flow (Fig. 6b). It receives and integrates extensive inputs from cortical areas along the lateral cortical mantle, both dorsal and ventral to the rhinal fissure. These areas include the olfactory cortex, ENTl, amygdala, subiculum and neocortical areas within the lateral cortical subnetworks²⁸, such as the PERI, ECT, TEa, AUDv, AIp, AId and AIv. Consequently, the DP processes a vast range of information from both external and internal environments, including olfactory, pheromonal, visceral sensory, auditory, fear conditioning and spatial orientation signals. Integrating these diverse modalities is crucial for perception, risk evaluation, decision-making and memory formation^3,5.

The DP then projects densely to other MPF and cortical areas involved in motor control (Fig. 6c): (1) the DPd, DPs and ILA comprise the primary visceral motor cortex, regulating neuroendocrine and autonomic functions. (2) The ILA and PL target the ACB, which is linked to reward processing through the VTA. (3) The ACAv, ACAd and MOs-fef, along with other medial network areas (for example, PTLp, RSPv), project to the dorsomedial striatum³² and SC³⁴, coordinating eye, head and neck movements for spatial orientation and navigation^3,43. (4) The DPs and other MPF areas also send direct inputs to the MOs and MOp, supporting control of trunk and limb movements. These circuits form the structural basis for integrating visceral and somatic motor actions in goal-directed behaviours^3,5. See the Supplementary Discussion for further discussion topics.

Related to the primates and human

On the basis of connectivity, the DPd, DPs, ILA and PL probably correspond to components of the ventromedial prefrontal cortex (vmPFC) in primates and humans^4,7,9,10,51. Subgenual vmPFC regions, including Brodmann area 25, are part of the central autonomic network^7,10, but their role in neuroendocrine regulation is unclear. Studies showing vmPFC involvement in mood disorders have yielded mixed results^51,52. Our findings suggest that functional divergence between the anatomically adjacent two areas may underlie these inconsistencies: the DPd enhances plasma CORT levels¹⁴, whereas the ILA suppress neuroendocrine and cardiovascular stress responses¹.

Recent studies show that the DP and TTd are activated by social interactions and stressors like restraint, looming sounds, fear memory, predatory threats and opioid exposure^{12,13,14,16,17}. We found that DP neurons respond to both social (sex dependent) and non-social stimuli, whereas the ILA does not exhibit sex-specific social responses. Defining the homologous regions of the DPs, DPd and ILA in primates and humans remains essential, alongside validating their roles in physiological and social responses critical for homeostasis and social behaviour^3,5,6,36. The DP functions as a hub for integrating global cortical information flow (Fig. 6b). Lesions affecting the DP or ILA (or the vmPFC in humans) presumably impair integration of sensory and interoceptive signals, disrupt MPF connectivity, and lead to disorganized emotional or behavioural responses (Fig. 6c and Supplementary Discussion). This is often observed in various psychiatric disorders, including classic Phineas Gage-like personality disorders, major depressive disorder as well as post-traumatic stress disorder^3,6,39.

Methods

Animals

For neuroanatomical tracing experiments, most of the animals used were C57BL/6J male mice aged 2 months (n = 95; Jackson Laboratories). Mice were housed in a temperature-controlled (21–22 °C), humidity-controlled (51%) and light-controlled (12 h–12 h light–dark cycle; lights on, 06:00; lights off, at 18:00) room. Food and water were given ad libitum. After arrival at the facility, the mice were allowed a minimum of 1 week to adapt to the housing environment before surgeries were performed. All of the experiments were conducted according to the standards set by the National Institutes of Health Guide for the Care and Use of Laboratory Animals and the institutional guidelines of the University of Southern California (USC) and University of California Los Angeles (UCLA).

Male MORF3 mice (n = 20; aged 2 months), were also used for neuroanatomical tract tracing. The MORF3 mouse line (C57BL/6-Gt(ROSA)26Sortm3(CAG-sfGFP*)Xwy/J), generated by X. W. Yang’s laboratory at UCLA²⁵, is a Cre reporter mouse line that uses a mononucleotide repeat frameshift (MORF) as a translation switch for cell labelling in vivo. MORF3 mice express a Cre-dependent tandem ‘spaghetti monster’ fluorescent protein with 20 V5 epitopes (smFP-V5) preceded by a polycytosine repeat (C22) MORF switch under the control of a CAG promoter. Cre recombination combined with a spontaneous frameshift results in sparse and stochastic labelling of neural cells.

Two other transgenic mouse lines, including VGLUT2 Cre (B6J.129S6(FVB)-Slc17a6tm2(cre)Lowl/MwarJ) and CRH-Cre (B6(Cg)-Crhtm1(cre)Zjh/J) were purchased from JAX and breeding colonies for these mouse lines were established at UCLA. CRH-Cre mice were crossed with Ai14 mice to generate CRH-Cre Ai14 mice for some of the experiments. A total of 5 male VGLUT2 Cre, 5 CRH-Cre and 4 CRH-Cre Ai14 mice aged 2 months were used for the neuroanatomic tracing experiments. Animal information for non-tracing studies is included in their corresponding subsections below.

Sample sizes for each of the tracing and morphology experiments were chosen based on previous experience. Randomization was not performed for the neural tracing experiments as animals were not assigned to different groups. Randomization is necessary for reducing bias and controlling variability. Instead, the data were validated in different ways (see the ‘Data reproducibility’ section). Randomization was also not necessary for morphology experiments. Owing to the characteristics of MORF3, we achieve random labelling of neurons and then reconstruct the traceable neurons. For neuroanatomic tracing and morphology experiments, blinding also was not necessary because the animals were not assigned to different groups.

Ethics statement

We are committed to promoting ethical research practices and ensuring the welfare of all animals involved in our studies. All procedures adhered to regulatory standards as delineated in the National Institutes of Health Guide for the Care and Use of Laboratory Animals, as well as institutional guidelines set forth by the Institutional Animal Care and Use Committees at the University of Southern California (USC) and the University of California, Los Angeles (UCLA).

Tracers

Phaseolus vulgaris leucoagglutinin (PHAL; 2.5%; Vector Laboratories) was used as the main chemical anterograde tracer. Chemical retrograde tracers included cholera toxin subunit b (CTB Alexa Fluor 647 conjugate, 0.25%; Invitrogen) and FG (1%; Fluorochrome). FG was also injected through the tail vein (2%, 30 μl) to label neuroendocrine cells. Each anterograde and retrograde tracer has distinct characteristics and exhibits varied neurotropism that can meaningfully affect connectivity results. For example, anterograde AAVs label fibres of passage, while PHAL does not. These differences underscore the importance of data validation. In this work, the connectivity data were validated in multiple ways to ensure the reliability of the results. Retrograde tracers were placed in regions of anterograde terminations, while anterograde tracers were placed in regions of retrogradely labelled cells. For example, Extended Data Fig. 10 shows retrograde tracer injection in the DPs and DPd that reveal their distinct brain-wide inputs (brain-wide ROIs to the DPs or DPd). To validate these results, Extended Data Fig. 11 shows anterograde tracer injections delivered to those brain-wide ROIs to validate their projections to either the DPs or DPd. Furthermore, we have previously demonstrated that PHAL and AAV produce similar brain-wide anterograde labelling patterns²¹ (Supplementary Fig. 6).

AAVretro-hSyn-Cre-WPRE (AAV retro Cre; 1.6 × 10¹³ genome copies (GC) per ml; Addgene, 105553), AAV1-Syn-Flex-GCaMP6f-WPRE-SV40, AAV1-hSyn-SIO-stGtACR2-FusionRed and AAV-Ef1a-mCherry-IRES-Cre (retrograde)⁵³ were produced by Addgene.

Viral anterograde tracers AAV encoding enhanced green fluorescent protein (AAV-GFP; AAV2/1.hSynapsin.EGFP.WPRE.bGH) and tdTomato (AAV1.CAG.tdtomato.WPRE.SV40) and AAV1-hSyn-Cre-WPRE-hGh (anterograde trans-synaptic) were packaged by UPenn Vector Core. AAV5-Ef1a-DIO-hChR2(H134R)-eYFP, AAV9-CAG-FLEX-GFP (Cre-dependent channelrhodopsin expressing AAV) was produced by the University of North Carolina vector core facility.

For experiments investigating the synaptic contacts of DP axon terminals onto PVH CRH neurons, an AAV-synaptophysin-GFP injection was made into the DP of CRH-Cre Ai14 male mice (n = 4, aged 2 months). A 1:1 mixture of AAV Cre and AAV synaptophysin-GFP was generated and a 100 nl pressure injection was made into the desired ROI. The viruses used for the mixture were AAV1-hSyn-FLEX-tdTomato-T2A-SypGFP-WPRE (Addgene, 51509, 1.3 × 10¹³ GC per ml) and AAV1-hSyn-Cre-WPRE (Addgene, 105553, 1.8 × 10¹³ GC per ml).

The viral vectors used for the genetically engineered rabies tracing system were produced by Wickersham Lab at MIT and included AAV2/9-CAG-FlexmKate-T2A-TVA, AAV2/9-CAG-Flex-mKate-T2A-N2c-G and Rbv-CVS-N2c-ΔG-GFP (the modified rabies virus). All viral vectors were aliquoted and stored at 80 °C until use. Importantly, rabies virus has been widely used as a tool for reliably mapping presynaptic inputs to a specific brain region or starter cell population^54,55,56. While its tropism appears broad, it remains possible that specific cell types in different pathways may be over- or under-represented relative to the performance of other retrograde tracers and viruses (for example, AAVretro, CAV2, retrobeads and so on)^57,58. Moreover, the potential for cytotoxicity within the starter cell population may provide an opportunity for rabies to non-specifically spread to nearby synapses rather than through those specifically contacting starter cells. To control for these possibilities, retrograde tracing results must be validated using complementary anterograde approaches, such as by tracing axonal connections from the upstream region, or by optogenetically stimulating upstream axons and recording synaptic responses in the target cell population. As such, the data that we present were validated in several different ways.

Stereotaxic surgeries for neuroanatomical tract tracing

Surgical and microscopy imaging procedures for neuroanatomical tract tracing experiments have been described previously^28,32,59. Stereotaxic coordinates of targeted injection centres were determined through the ARA¹¹ and empirically adjusted when needed. On the day of the experiment, mice were deeply anaesthetized and mounted into a Kopf stereotaxic apparatus where they were maintained under isoflurane gas anaesthesia (Datex-Ohmeda vaporizer). Before the surgery, mice were given one subcutaneous injection of Ketoprofen (4 mg per kg) and a protective ophthalmic ointment was applied to their eyes. For single anterograde tracer injection experiments (PHAL or AAV), tracers were iontophoretically delivered through glass micropipettes (inner tip diameter, 24–32 μm) using alternating 7 s on–7 s off pulsed positive electrical current (Stoelting, current source) for 10 min, and AAVs were delivered through the same method for 2 min (inner tip diameter, 8–12 μm). For anterograde/retrograde co-injection experiments (PHAL/CTB-647 and AAV/FG), tracer cocktail iontophoretic injections were made through glass micropipettes (inner tip diameter, 28–32 μm) for 10 min (PHAL/CTB-647) or 5 min (AAV/FG).

For multiple retrograde tracing experiments, at each injection site, 50 nl of the retrograde tracer was individually pressure-injected through glass micropipettes at a rate of 10 nl min⁻¹ (Drummond Nanoject III). All injections were placed into the right hemisphere. After injections, incisions were sutured, and mice were returned to their home cages for recovery.

TRIO tracing (Cre-dependent TVA receptor mediated rabies tracing)

To reveal monosynaptic inputs to projection-defined neuronal populations in the DP (for example, DP neurons projecting to the ACA), we used a modified TRIO (tracing the relationship between input and output) strategy⁶⁰. In brief, AAVretro-Cre was injected into a downstream projection target of DP (for example, ACA), and Cre-dependent TVA- and RG-expressing helper virus (AAV8-hSyn-FLEX-TVA-P2A-GFP-2A-oG) and mCherry-expressing G-deleted rabies virus (produced by the laboratory of I. Wickersham) were injected into the DP to label the DP projection neurons population (first order) and their brain-wide monosynaptic inputs (second order). The same strategy also was applied to trace brain structures that generate monosynaptic inputs to the DP-projecting neurons in the RE.

We used a modified rabies tracing strategy to identify monosynaptic inputs from the DPd to CRH-expressing neuroendocrine neurons in the PVH. In CRH-Cre transgenic mice, a Cre-dependent helper virus (AAV8-hSyn-FLEX-TVA-P2A-GFP-2A-oG) was injected into the PVH to drive expression of TVA and rabies glycoprotein (RG), followed by injection of mCherry-expressing G-deleted rabies virus (RbV-ΔG-mCherry). This approach selectively labels neurons providing direct monosynaptic input to CRH neurons. All viral vectors were produced by the Wickersham laboratory at MIT.

AAV1-Cre-based anterograde trans-synaptic tracing

This technique leverages the fact that when AAV1 is injected at sufficiently high concentration into a neuronal population, viral particles will travel down the axons and be released from the synaptic terminals where they can infect postsynaptic neurons. Detailed methodology was described previously⁶¹. In brief, anaesthetized mice were iontophoretically injected with Cre-dependent AAV-FLEX-RFP or GFP in the target structures (for example, the LHA), and pressure injected (20–80 nl) with AAV1-Cre in an upstream structure (for example, the DP). The AAV1-Cre is transported anterogradely down the axons and is released from the terminals, where it transfects postsynaptic cells that have been infected with high concentrations of Cre-dependent AAV-FLEX-RFP. The scant Cre expression is sufficient to unlock strong fluorophore expression in the downstream neurons, therefore revealing their axonal projections and terminals. After a 3-week post-operative recovery, the mice were anaesthetized with pentobarbital and perfused. The Cre injection site was verified by staining with mouse anti-Cre recombinase monoclonal primary antibody (see the ‘Tissue preparation and immunohistochemistry’ section for details). One caveat to using AAV1-Cre is that the virus can also travel retrogradely and should be used in situations in which the connection is known to be primarily unidirectional. In all of the experiments in which AAV1-Cre was used in the current paper, the connections were shown to be predominantly unidirectional (BLA/SUBv/AUDv to DP and DP to CEA/LHA/PAG; Supplementary Fig. 11). As an example, we injected AAV1-Cre into the DP and a Cre-dependent AAV into the LHA to show that LHA neurons that receive input from the DP project to the spinal cord (DP to LHA to spinal cord). If LHA projected back to the DP, we would possibly be showing that LHA neurons that project to DP also project to the spinal cord. However, we know that the DP to LHA connection is unidirectional (Supplementary Fig. 11b), which lends reasonable confidence to our conclusion of DP to LHA to spinal cord.

Details regarding how injection site locations were accurately determined are provided in the ‘Injection Site Analysis’ section of the Supplementary Information.

Data reproducibility

Reported connections underwent validation through at least one of the following methods. Injections targeting different regions were repeated to evaluate consistency of labels. The Cre-dependent anterograde and TVA receptor mediated rabies tracing also validated the connections. Furthermore, retrograde tracers were introduced into regions displaying anterograde terminal labelling to confirm anterograde connections, whereas anterograde tracers were administered into the sites of retrogradely-labelled projection cells to validate retrograde injection data. In some instances, channelrhodopsin assisted circuit mapping also validated connections.

Tissue preparation and immunohistochemistry

Animals were euthanized with an overdose injection of sodium pentobarbital (6 mg per kg) 7 days (chemical tracers) or 14 days (viral tracers) after surgeries. Each animal was transcardially perfused with approximately 50 ml of 0.9% NaCl followed by 50 ml of 4% paraformaldehyde solution (PFA; pH 9.5). The brains were post-fixed in 4% PFA for 24–48 h at 4 °C, after which they were embedded in 3% type I-B agarose (Sigma-Aldrich) before sectioning. Four series of coronal sections were sliced at a thickness of 50 µm using a Compresstome (VF-700, Precisionary Instruments) and prepared for processing. One of the four-section series was immunostained for imaging. The series that contained coronal level 53 of the Allen Reference Atlas was selected to maintain similar section-level distributions across experiments.

The following primary antibodies were used across the studies: rabbit anti-PHAL (1:1,000; Vector Laboratories, AS-2300), rabbit anti-fluorogold antibody (1:5,000; Millipore-Sigma, AB153-I), mouse anti-Cre recombinase (1:4,000; EMD Millipore, MAB3120), rat monoclonal FOS antibody (1:1,000; Synaptic Systems, 226017), mouse anti-oxytocin antibody (4G11, 1:5,000; Millipore-Sigma, MAB5296), rabbit anti-vasopressin antibody (1:5,000, Millipore-Sigma, AB1565), rabbit-anti somatostatin antibody (1:500, Millipore-Sigma, SAB4502861) and streptavidin for biocytin-labelled recorded neurons (1:1.000; Thermo Fisher Scientific, S21374).

In brief, sections were transferred to a blocking solution containing normal donkey serum (Vector Laboratories) and Triton X-100 (VWR) for 1 h. After three 5-min rinses, the sections were incubated in a KPBS solution comprising donkey serum, Triton X-100 and a primary antibody at the appropriate dilution (the concentration for each primary antibody is described above) for 48–72 h at 4 °C (Vector Laboratories, AS-2300). The sections were rinsed three times in KPBS and then soaked for 3 h in the secondary antibody solution, which contained donkey serum, Triton X-100 and the corresponding secondary antibody solution at the appropriate dilution (1:500 for anti-rabbit IgG conjugated with Alexa Fluor 488, 555 or 647; Invitrogen, A-21206 (488), A-31573 (647); anti-mouse IgG conjugated with Alexa Fluor 488 or 647: Life Technology: A-21202 (488) and A-31571 (647); anti-mouse IgG conjugated with Alexa Fluor 647: Jackson ImmunoResearch, 715-605-150). The sections were again rinsed with KPBS three times. All selected section series were counterstained with a fluorescent Nissl stain, NeuroTrace 435/455 (NT, 1:500, Invitrogen, N21479). The sections were then mounted and coverslipped using 65% glycerol.

RNAscope experiments

Deeply anaesthetized animals were perfused and brains were post-fixed in 4% PFA overnight followed by sequential immersion in 15% and 30% sucrose for cryoprotection. The brains were then sectioned (20 µm) and the RNAscope Multiplex Fluorescent v2 assay was performed according to the manual provided by Advanced Cell Diagnostics. In brief, slides containing tissue sections were post-fixed in the prechilled 4% PFA for 15 min at 4 °C, and then dehydrated in a series of ethanol solutions 50%, 70% and 100% for 5 min at room temperature. Hydrogen peroxide was applied to each section for 10 min at room temperature, and target retrieval was then performed at 99 °C for 5 min. Protease III was then added to the samples and incubated in a HybEZ oven for 30 min at 40 °C. The subsequent RNAscope assay was composed by multiple steps of hybridization (probe and AMP binding) and signal development (HRP channel binding), each followed by washing the slides in 1× wash buffer as described in the manufacturer’s protocol (Multiplex fluorescent v2 kit). Opal fluorophores (Akoya Biosciences) were diluted at 1:1,500 to visualize RNA signals. After the RNAscope assay, the sections were counterstained with DAPI (1:500, Thermo Fisher Scientific, D1306) and mounted. Images were acquired using a Dragonfly High Speed Confocal Microscope System (Andor). The following probes were used in this study: Cux2, Etv1, Tle4, SS, CRH, and TRH.

Image acquisition and 2D processing

All 50-µm sections were scanned as high-resolution virtual slide image (VSI) files using the Olympus VS120 high-throughput microscope. Post-acquisition processing methods, including image registration, tracer segmentation and quantification, are described in detail below. In brief, all images were flipped to ensure that the injection site and its corresponding ipsilateral labelling were in the correct right hemisphere. This was performed using the VS-Desktop software that accompanies the microscope. Images were then imported into Outspector—our custom software for image processing (see the ‘Outspector: our proprietary 2D image processing pipeline’ section of the Supplementary Methods) and processed on a high-performance computing cluster. A first-pass fully automatic registration was executed and subsequently its results were inspected and fine-tuned when needed. Tracer segmentation was performed on registered images with the default parameters. Segmentation results were also inspected, followed by interactive parameter adjustments if needed. Segmented pixel counts and cell counts in each brain region at all section levels were generated and used for analysis.

Outspector

Outspector is our informatics workflow specifically designed to reliably warp, reconstruct, annotate and analyse the labelled pathways in a high-throughput manner. Tissue sections from each analysed case were assigned and registered to standard templates of the corresponding ARA levels. All images shown in this Article are from the raw data of unwarped, unregistered VSI images. Threshold parameters were individually adjusted for each case and tracer. Adobe Photoshop was used to correct conspicuous artifacts in the threshold output files. Each colour channel was brightness/contrast adjusted to maximize labelling visibility (Nissl Neurotrace 435/455 is converted to bright field), and TIFF images are then converted to JPEG file format. The detailed calculations used for all Outspector processing steps and analysis are provided in the ‘Outspector: our proprietary 2D image processing pipeline’ section of the Supplementary Methods.

Quantifying colocalization

Using spots function on Imaris, thresholds were set to select most positive cells for each channel (405, 488, 561 and 642). After automatic detection, manual quality control was necessary. The spatial relationship between DAPI in 405 and RNA puncta for Fezf2 was used as a visual aid to correct false negatives and remove false-positive spots from the automatic detection. Each channel was quality controlled multiple times to ensure accuracy.

Tissue processing and imaging for 3D single-neuron labelling

To elucidate the finely detailed morphology of DP neurons, we injected with either AAV-retro Cre or AAV1-Cre into MORF3 mice. Then, 3–4 weeks after the injections, the mice were perfused. Immunostaining with the V5 antibody allows the visualization of the finely detailed neuronal morphology of these MORF3 neurons. This approach enabled us to image, reconstruct and analyse the cell-type-specific morphological features of DP cortical projection neurons. Following the same strategy, we also revealed sparsely labelled neurons with finely detailed dendritic morphology in the LHA.

The SHIELD clearing protocol was used for the 3D tissue processing⁶². Mice were transcardially perfused with cold saline and SHIELD perfusion solution. Brains were dissected, subjected to a SHIELD-based brain clearing method and immunostained for V5 (to reveal MORF3 with secondary antibody conjugated with Alexa Fluor 647) using eFLASH^25,62,63. Brains were processed using the SmartClear and SmartLabel machines (LifeCavas SmartBatch). Whole brains were imaged using the Life Canvas SmartSPIM lightsheet microscope at ×15 (0.42 μm × 0.42μm × 0.42 μm). Three lasers (488, 555, 647 nm) were used, respectively, to image Syp–eGFP axonal terminals, tdTomato-positive cortical neurons and their axons and MORF3-positive postsynaptic neurons.

For acquiring images on thick sections, the whole brain was cut into 500 µm sections and were cleared in the SDS buffer at 37 °C for 72 h. The sections were then washed three times with KPBS and incubated in KPBS at 4 °C for 24 h. The sections were mounted and cover-slipped onto 25 × 75 × 1 mm glass slides with an index matching solution (100%, EasyIndex, LifeCanvas Technologies, EI-Z1001). The sections were imaged with a high-speed spinning-disk confocal microscope (Andor Dragonfly 202 Imaging System, Andor, Oxford Instruments Company, CR-DFLY-202-2540). ×10 magnification (NA 0.40, Olympus, UPLXAPO10X) was used to acquire an overview, after which ×30 magnification (Olympus, UPLSAPO30xSIR, NA 1.05, W.D.: 0.80 mm) at 1 µm z steps or ×60 magnification (Olympus, UPLASAPO60XS2, NA 1.30, W.D.: 0.30 mm) were used to acquire high-resolution images for fine-detailed dendritic morphology and putative synaptic connections. The sections were imaged with four excitation wavelengths: 405 nm (blue Nissl background), 488 nm (for AAV-GFP) and 647 nm (far red for MORF3) with respective emission detection wavelengths of 450 nm, 525 nm and 600 nm.

3D reconstructions, visualizations and analysis of neuronal morphology

Manual reconstruction of neurons was performed using Vaa3D^64,65 or Fast Neurite Tracer⁶⁶, followed by morphological repairing, sorting and typing of the dendritic arbours. Reconstructions were finalized by loading them in Imaris and neuTube⁶⁷ and finalized into the widely used swc file format⁶⁸. In total, 47 reconstructed DP neurons were morphologically analysed in MATLAB, leveraging functions from the Trees Toolbox environment⁶⁹. The 47 DP neurons were pooled from four MORF3 mice. Specifically, analysis scripts measured all neurons across a series of parameters including total dendritic wiring length, number of branches, average branch angle, average root angle, branch asymmetry, average branch order, average Strahler order and the distance from the midline. The coefficients of variances across the parameters were then measured, followed by classification using k-means clustering. The size and complexity of dendritic arbours, defined by total dendritic length and the total number of branches, emerged as the most varied. We then classified all of the neurons, first based on dendritic arbour size and complexity, followed by superficial versus deep-layer classification based on the distance of all cell bodies from the midline. Both classifications created two neuronal clusters each. The superficial versus deep-layer neurons were then compared for size, complexity and dendritic orientation. Similarly, the two morphological clusters (small-simple and large-complex) were compared for their distance from midline.

Statistical analyses

These comparisons were made using one-tailed nonparametric Wilcoxon signed rank tests. The P values from the tests were corrected using FDR correction to account for multiple comparisons.

In vitro patch-clamp slice recordings

In wild-type mice, AAV1-Cre (anterograde trans-synaptic Cre) was injected into the AUDv, while AAV-hSyn-ChR2-YFP (UPenn Vector Core) was injected into either the ENTI (n = 5; Fig. 5a) or PIR (n = 6 Fig. 5c) of different animals. In separate experiments, the AAV-hSyn-ChR2-YFP was injected into the DP of CRH-Cre Ai14 mice (n = 7; Fig. 4g). Sample sizes were chosen on the basis of previous experience. Randomization and blinding were not performed as the mice were not assigned to different groups.

Two weeks after channelrhodopsin and tracer injections, acute brain slices were prepared for electrophysiological recordings. After anaesthesia with isoflurane, the mouse was decapitated and the brain was rapidly removed and immersed in ice-cold high sucrose slice solution: 208 mM sucrose, 2.5 mM KCl, 1.25 mM NaH₂PO₄, 26 mM NaHCO₃, 1.3 mM MgCl₂, 8 mM MgSO₄ and 10 mM glucose; saturated with 95% O₂ and 5% CO₂; pH 7.4. Brain slices (thickness, 350 μm) were cut in the coronal plane using a vibrating microtome (Leica, VT1000S). Slices were allowed to recover for at least 30 min in a submersion chamber filled with warmed (32–34 °C) ACSF containing 130 mM NaCl, 3 mM KCl, 1.25 mM NaH₂PO₄, 26 mM NaHCO₃, 2 mM CaCl₂, 2 mM MgCl₂ and 10 mM glucose, oxygenated with 95% O₂ and 5% CO₂, pH 7.2–7.4, 290–310 mOsm, and then cooled gradually to room temperature until recording. The presence of RFP and ChR2 (YFP) labelling was determined with green and blue fluorescence excitation, respectively, in the slices before recording. Patch pipettes with about 3–4 MΩ impedance were used for whole-cell patch clamp recordings. Glass pipettes contained a Cs-methanesulfonate-based internal solution with the following salt concentrations: 130 mM Cs-methanesulfonate, 10 mM CsCl, 4 mM NaCl, 1 mM MgCl₂, 5 mM mM 150 mM MgATP, 5 mM EGTA, 10 mM HEPES, 5 mM GTP, 10 mM phosphocreatine and 0.1 mM leupeptin (pH 7.2 with 151 CsOH, 270 mOsm) and biocytin (0.2%). For isolation and recording of monosynaptic responses to blue-light stimulation, tetrodotoxin (Na⁺ channel blocker, 1 μM) and 4-aminopyridine (K⁺ channel blocker, 100 μM) were added to the external solution. Signals were recorded from red-labelled neurons with a MultiClamp 700B amplifier (Molecular Devices) running pClamp software (v.10.5) under voltage clamp mode at a holding voltage of −70 mV for excitatory currents, filtered at 1 kHz and sampled at 10 kHz. Blue light (470 nm) stimulation was delivered using a 0.5 ms pulse at ~1 mW power for 5 trials, delivered through a LED (CoolLED). Signals were analysed using Clampfit (v.10.5). For responding neurons, peak responses were averaged across trials (technical replicate). The averaged value of each cell was then combined with the other cells for each experiment to calculate the mean, median, s.d. and so on (biological replicate). To demonstrate that light-evoked synaptic responses were indeed excitatory, the selective glutamate receptor antagonists CNQX (20 µM) for AMPA receptors and APV (50 µM) for NMDA receptors were added to the bathing solution.

After recording, the slices were fixed with PFA and then transferred to PBS for biocytin staining to reveal and validate the location of the recorded neurons. The biocytin staining also revealed the morphology of the recorded DP or PVH neurons (see the ‘Tissue preparation and immunohistochemistry’ section for details).

Statistical analyses

To compare the median peak amplitude of currents evoked by optogenetic stimulation of AUDv to DP neurons receiving either ENTl or PIR input, nonparametric Mann–Whitney U-tests were used and P value adjustments were not necessary.

Optogenetic stimulation and measurements of plasma CORT hormone

A total of 15 mice (male C57BL/6, aged 2 months) were randomly assigned to either the experimental or control group. Sample sizes for each of the experiments were chosen based on our previous experience and, based on the results, sufficient power was achieved to detect significant differences. Each of the animals received bilateral injections of AAV1-Cre into the SUBv and the BLAa, allowing Cre to anterogradely transport to and trans-synaptically label DP neurons. The animals also received a bilateral injection of Cre-dependent AAV-FLEX-ChR2 (30 nl) into the DP. Consequently, only DP neurons postsynaptic to the SUBv and BLAa express ChR2 and generate projections to the hypothalamic neuroendocrine zone, including the PVH. Optic fibres (200 μm core diameter, 0.37 numerical aperture; Inper) were implanted above the injection site at a depth of −1.6 mm ventral to the bregma skull surface and secured to the skull using super glue and dental cement.

The optogenetic stimulation was performed 7–8 weeks after the surgery. Before testing, the mice were individually habituated to the testing cage (new mouse cage) and optic fibres over 4 consecutive days, for 30 min daily. On the test day, the mice were habituated to the testing cage alone for 2 h and attached to optic fibres for 15 min before optogenetic stimulation. A 473 nm laser, pulsing at 20 Hz with 15 ms pulses, was applied for 30 s on and 30 s off cycles, for a total of 20 min. The light irradiance was around 4–5 mW mm⁻² in the target region. After laser stimulation, the optic fibres were removed, and the mice remained undisturbed in the testing cage for 25 min before euthanasia. To minimize circadian rhythm effects on CORT levels, animals were tested between 1.5 to 5 h after lights turned off in the dark phase, with control and ChR2 groups counterbalanced during the test.

Serum collection and CORT assaying

After rapid cervical dislocation, terminal trunk blood was collected with heparinized microcapillary tubes (Thermo Fisher Scientific) and transferred into microfuge tubes, which were left on ice for at least 15–20 min. The blood samples were centrifuged at 2,500 rpm for 15 min at 4 °C. The supernatant (serum) was collected and transferred into new tubes and the serum samples were frozen at −80 °C. The samples were packed and shipped on dry ice to the Center for Research in Reproduction Ligand Assay and Analysis Core at the University of Virginia School of Medicine, where they were processed (in duplicate) using a radioimmunoassay kit for CORT measurements. The brains were sectioned at 50 μm thickness and Nissl stained to assess the location of the implanted optic fibres. The researcher assessing the accuracy of the bilateral implant location was blinded to the treatment condition of the animal. Only animals with accurate bilateral implants in the DP were included in the analysis.

Statistical analyses

To compare the CORT levels between the experimental and control groups, Student’s t-test and nonparametric Mann–Whitney U-tests were used. Owing to the two-group comparisons, P value adjustment was not necessary.

Social interaction experiments

Surgery

Six VGLUT2-Cre (for DP) and six VGLUT1-Cre (for ILA) male mice (aged 6–8 weeks) were anaesthetized using a 1–2% isoflurane–oxygen mixture and injected unilaterally with 350 nl (1 nl s⁻¹) pAAV.Syn.Flex.GCaMP6f.WPRE.SV40 (Addgene, 100833-AAV1) into the DP. Then, 30 min after virus injection, a 6.1-mm-long (0.5 mm diameter) relay lens (Inscopix) was implanted over the DP: anteroposterior, +2.1 mm, mediolateral, 0.6 mm; dorsoventral, −3.2 mm. The lens was securely affixed to the skull using cyanoacrylate glue and dental cement and then covered with Kwik-Sil (WPI). Sample sizes for each of the experiments were chosen based on previous experience. Randomization was not necessary as VGLUT2-Cre animals received injection into the DP and the VGLUT1-Cre mice received injection in the ILA. Also, each of the animals underwent both social and non-social interaction behaviour tests (see below).

Three weeks later, the mice were anaesthetized as described previously and a UCLA V4 miniscope with an attached baseplate was positioned over the lens to visualize GCaMP6f expression. The focal plane was adjusted to optimize cell visibility, and the baseplate was secured in place with dental cement. A plastic cap was used to cover and protect the lens over the baseplate.

Behavioural tests

Mice with lens implantation in the DP or ILA were habituated to handling and the miniscope for 4–5 days. On the day of the open-field test, the miniscope was attached to the baseplate and the mice freely ran in an open arena (45 cm × 45 cm × 30 cm) for 20 min. On the day of same-sex social interaction, mice performed social interaction counterbalanced with object exploration in the same arena as the open-field test for 8 min per session. The next day, mice performed social interaction with opposite sex social targets for 8 min. Each social trial included an 8-min recording session, beginning with a 1-min period of the subject mouse alone, followed by a 7-min interaction with the social partner. On the day of same-sex social interaction counterbalanced with object exploration, after the first trial, the subject mouse rested with the LED off for approximately 5 min before starting the second trial. Calcium imaging and behaviour (miniCam) were recorded simultaneously at 30 Hz.

Pre-processing and extraction of calcium signals

Calcium imaging videos were motion corrected (NoRMCorre: https://github.com/flatironinstitute/NoRMCorre; MiniAn: https://github.com/denisecailab/minian) and analysed with extended constrained non-negative matrix factorization (CNMF-E)-algorithm-based packages (MiniscopeAnalysis: https://github.com/etterguillaume/MiniscopeAnalysis; MiniAn: https://github.com/denisecailab/minian) to identify cells and extract cell activity. Denoised and demixed trace data were used for further analysis.

Identification of behaviour responsive cell

To determine whether an individual neuron’s activity was significantly modulated by social interaction, we used ROC analysis to calculate an area under the curve (AUC) value for denoised calcium trace correspondence to binarized manually scored interaction bouts. The interaction bouts were defined when the imaged mouse actively investigated the social target/object within a distance of one-quarter to one-third of mouse body length. To determine whether the AUC values indicated that the neuron was significantly excited or inhibited by the interaction, we compared observed AUC values to a null distribution derived from circularly shuffled calcium signals (1,000 shuffles). Significance was determined using percentile thresholds, classifying neurons as significantly responsive if their auROC values surpassed the 97.5th percentile for excited neurons or the 2.5th percentile for inhibited neurons. Neurons that were neither excited nor inhibited were deemed non-responsive.

Calcium event calculation

Calcium transients were identified using automatic detection algorithms in MATLAB (peakfinder) (https://github.com/GradinaruLab/striatum2P).

Statistical analyses

To examine whether there were differences in DP-miniscoped or ILA-miniscoped animals in the time spent interacting during non-social (object) versus social (same- or opposite-sex partner) testing, an omnibus one-way ANOVA was performed followed by post hoc Tukey’s multiple-comparison test, which controls for the family-wise error rate. To assess the percentage of behaviourally responsive cells and frequency of calcium events during non-social (open-field) and social (same- or opposite-sex partner) testing, an omnibus two-way ANOVA was performed followed by post hoc analyses using either Tukey’s tests (for within DP or ILA comparisons) or Šidák’s tests with adjusted P values (for across DP and ILA comparisons) for multiple comparisons.

DP neuronal activities in response to social and environmental stimuli

Male mice in response to social stimuli

Twenty-four male mice, aged between 12 and 14 weeks, were used for the study. To reduce potential stress by the testing environment, each mouse was habituated to a clean cage for four consecutive days, spending 2 to 4 h in the cage and with approximately 30 min in the recording position daily. On the testing day, the mice were habituated in the clean cage for about 2 h before the experiment. Unfamiliar male mice, female mice and postnatal day 7 pups were used as the social stimuli. Each mouse was exposed to three instances of the same type of social stimulus, lasting about 6 min each, for a total duration of 20 min, with six mice in each group. In the control group, the cages were opened three times as well, but without introducing any intruders. The mice remained undisturbed in their individual cages for 80 min before perfusion.

Acute-restraint stress

Animals (n = 4, male) subjected to acute-restraint stress were removed from the colony room, transferred to a behavioural testing room for 1 h and then placed into a Tailveiner restrainer for mice (Braintree Scientific). This device consists of a clear tube, within which a plug was adjusted to limit voluntary movement. Acute restraint stress was applied for 30 min and terminated 30 min before the animals were euthanized. Throughout all studies, control mice (n = 4, male) underwent numerous sensory manipulations that were carried out in parallel with those for the mice in the restraint stress group.

Acute looming sound stimuli in comparison with white noise and foot shock

Subjects

TRAP2-CreER⁷⁰ and Ai14 mice were acquired from the Jackson Laboratory (www.jax.org/strain/030323; www.jax.org/strain/007908). Homozygous TRAP2-CreER and Ai14 mice were bred to generate TRAP2-CreER/Ai14 offspring.

Drug preparation

4-Hydroxytamoxifen (4-OHT; Sigma-Aldrich, H6278) was dissolved at 20 mg ml⁻¹ in ethanol by shaking at 37 °C for 15 min, then corn oil was added to give a final concentration of 10 mg ml⁻¹, and the ethanol was evaporated by vacuum under centrifugation. 4-OHT (intraperitoneally (i.p.), 50 mg per kg) was injected in all TRAP2-CreER experiments^70,71.

Stimulation for TRAP

Four TRAP-Cre mice were randomly assigned into two groups: control (n = 2) and foot-shock-based fear conditioning (n = 2). For auditory stimulation (noise group), on day 1 and 2, the TRAP2/Ai14 mice were adapted to custom-made boxes in a sound-proofed room for 1 h per day. On day 3, the mice were placed into their home cage for 12 h in a sound-proofed room before white-noise exposure. The mice were then placed in the custom-made boxes and exposed to 1 h broadband white-noise stimulation (70 dB SPL).

For the foot-shock group, on day 1, the mice were acclimatized in custom-made boxes in a sound-proofed room for 1 h. On day 2 (conditioning), the mice were put into custom-built boxes with a metal grid floor. After 10 min habituation to the conditioning chamber, the animals were exposed to the 20 s noise (70 dB SPL) co-terminated with a 0.75 mA foot shock (5 Hz for 1 s with the duration of each pulse = 100 ms) five times⁷². On day 3, the mice were subjected only to 1 h broadband white-noise exposure. Thereafter, mice received an injection of 4-OHT (i.p., 50 mg per kg) and were returned to their home cage. The cage was moved back to our animal facility 12 h after injection of 4-OHT. Sound stimulation and injection of 4-OHT were conducted in a sound-proofed room. On day 12, the mice were further presented with looming sound stimuli. A train of ten consecutive crescendo sound stimuli (with intensity increased linearly from 20 to 70 dB SPL in 0.4 s and followed by 0.6 s interval with intensity at 20 dB SPL) was repeated for 10 min with an interval of 10-s between the train stimuli. The mouse was euthanized after a 1.5 h rest.

Control mice were treated the same procedures as the auditory stimulation group but without any sound stimuli.

Images used from open-access resources

Images from the mouse ARA used in this paper are available online (https://mouse.brain-map.org/)¹¹. Images of the flatmaps are freely downloadable from the cited article⁷³. Bright-field in situ hybridization images from the ABA are available online (https://mouse.brain-map.org/)⁷⁴. The Common Coordinate Framework (CCF) was used to generate the background 3D images of the brain, and are available online (https://alleninstitute.github.io/abc_atlas_access/descriptions/Allen-CCF-2020.html)⁷⁵.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.