Projections from the dorsomedial division of the bed nucleus of the stria terminalis to hypothalamic nuclei in the mouse

Abstract As stressful environment is a potent modulator of feeding, we seek in the present work to decipher the neuroanatomical basis for an interplay between stress and feeding behaviors. For this, we combined anterograde and retrograde tracing with immunohistochemical approaches to investigate the patterns of projections between the dorsomedial division of the bed nucleus of the stria terminalis (BNST), well connected to the amygdala, and hypothalamic structures such as the paraventricular (PVH) and dorsomedial (DMH), the arcuate (ARH) nuclei and the lateral hypothalamic areas (LHA) known to control feeding and motivated behaviors. We particularly focused our study on afferences to proopiomelanocortin (POMC), agouti‐related peptide (AgRP), melanin‐concentrating‐hormone (MCH) and orexin (ORX) neurons characteristics of the ARH and the LHA, respectively. We found light to intense innervation of all these hypothalamic nuclei. We particularly showed an innervation of POMC, AgRP, MCH and ORX neurons by the dorsomedial and dorsolateral divisions of the BNST. Therefore, these results lay the foundation for a better understanding of the neuroanatomical basis of the stress‐related feeding behaviors.


| INTRODUCTION
The prevalence of obesity reached an alarming rate over the last decades. Obesity is a complex and multifactorial disease sensitive to environmental cues. Among those, daily stress plays an indisputable role in the alteration of feeding behavior and consequently in the control of body weight. In the brain, several hypothalamic nuclei are essential to control energy metabolism. The arcuate nucleus (ARH) and the lateral hypothalamic area (LHA) are ideal topics focusing most of the scientists attention as they are mainly composed of neurons involved in the tight control of energy balance and motivation (Diniz & Bittencourt, 2017;Ruud et al., 2017;Sohn, 2015;Stuber & Wise, 2016;Timper & Brüning, 2017;van der Klaauw & Farooqi, 2015). Indeed, the ARH contains the anorexigenic proopiomelanocortin (POMC)-expressing and the orexigenic Agouti Related Peptide (AgRP)/Neuropeptide Y (NPY)-co-expressing neurons. For its part, the LHA is made of neuronal cells containing the melanin-concentrating hormone (MCH) or orexin (ORX) thought to regulate feeding and reward-related behaviors (Diniz & Bittencourt, 2017;Stuber & Wise, 2016).
Stress, anxiety and mood disorders are often associated with deregulation of feeding behavior (O'Neil et al., 2014;Weinstein et al., 1997).
The interplay between psychological factors and changes in food intake during stress episode relies on telencephalic structures such as the amygdala (Ip et al., 2019). Interestingly, a few recent studies functionally linked the dorsal divisions of the anterior bed nucleus of the stria terminalis (BNST) with the LHA and the ARH neurons in controlling appetitive, aversive or exploratory behaviors (Giardino et al., 2018;González et al., 2016;Jennings et al., 2013). On the other hand, the dorsomedial division plays an indisputable role in integrating stress and anxiety behaviors (Ch'ng et al., 2018;Daniel & Rainnie, 2016;Duval et al., 2015).
Indeed, the dorsomedial division of the BNST display increased number of c-fos activated neurons upon stress exposure, and an increased dendritic arborization after chronic stress exposure (Daniel & Rainnie, 2016;Kovács et al., 2018;Vyas et al., 2003). Collectively, these data suggest an interplay between the BNST and hypothalamic neurons in the stressdriven feeding behavior. However, BNST projections to hypothalamic nuclei have only been succinctly described without precisely compiling both the source of the projections with the downstream targets. Here, we focused our study on the projections from the dorsomedial division of the BNST on neurons of hypothalamic nuclei involved in feeding and motivated behaviors: the PVH, the dorsomedial nucleus of the hypothalamus (DMH), the ARH and the LHA with a particular interest on POMC, AgRP/NPY, MCH, and ORX neurons.

MCH and ORX neurons
All animal procedures performed in this study were approved by the UK government (Home Office) and by Institutional Animal Welfare Ethical. Pmch-Cre (IMSR Cat# JAX:014099, RRID:IMSR_JAX:014099) (Kong et al., 2010) and orexin-Cre (kind gift from Prof Takeshi Sakurai) (Matsuki et al., 2009) male mice were single housed and kept on a standard 12-hr light-dark cycle (lights on at 0700 hr) and on standard mouse chow and water ad libitum.

| Anterograde tracer into the anterior subdivisions of the BNST in the mouse
Mice were anesthetized with isoflurane (0.5 L/min) before the stereotaxic surgery. Six mice received a unilateral iontophoretic injection of 2.5% PHAL diluted in sodium phosphate buffer saline (NaPBS) pH 7.2. Glass micropipettes (tip diameter: 10-20 μm) were used to inject the PHAL iontophoretically (intermittent current of 5 μA and 7 s on/off time for 20 min). The coordinates, based on standard atlas coordinates (Franklin & Paxinos, 2008) were: AP, +0.5 mm; ML, +0.9 mm; DV, −4.0 mm. To avoid PHAL diffusion along the micropipette track, the micropipette was left in place for another 5 min before being removed. At 15 days after the injection, the brains were processed for immunohistochemistry, as described below.

| MCH and orexin neurons
Data used for the present study have been generated in a previous study (González et al., 2016) where the monosynaptic retrograde tracing from MCH and ORX neurons has been extensively reported. Images have been taken from these experiments. In details, 50-60 nl of AAV2/1-EF1a-Flex-C-RVG (Addgene, 49101) and AAV2/1-EF1a-Flex-eGFP-TVA (Addgene, 26198) were stereotaxically injected into the LHA of Pmch-Cre or orexin-Cre mice. The injection coordinates were: 1.35 mm caudal from bregma, 0.75-0.9 mm lateral from midline, 5.3-5.4 mm ventral. Two days later we injected at the same site 70 nl of envelope A pseudotyped SADB19 rabies virus expressing tagRFP in place of rabies glycoprotein (DRV-RFP 39) and the animal was perfused 10-11 days later. Viruses were obtained from Dr Molly Strom and were made as described in this paper (Vélez-Fort et al., 2014).

Surgery
Monosynaptic retrograde tracing using rabies virus was performed as follow: adult 12-week-old Pomc-Cre (Balthasar et al., 2004) and AgRP-IRES-Cre (Tong et al., 2008) mice were anesthetized with a mix of ketamine/xylazine (100 mg/kg and 20 mg/kg, respectively). Virus were injected with a microsyringe (Hamilton, 35G) and microinjection pump (World Precisions, rate at 100 nl/min). Mice receive 300 nl of mixed AAV1-Syn-FLEX-splitTVA-eGFP-tTA and AAV1-TREtight-BFP2-B19G in one side of the ARH (AP: −1.4 mm; ML: −0.3 mm; DV: −5.8 mm). After 7 days, the same mice received a second injection of 300 nl of pseudotyped rabies virus EnvA-SADdG-mcherry (Salk Institute) using the same coordinates. Control mice were injected with helper virus or EnvA-SADdG-mcherry alone. One week later, mice were anesthetized and perfused with 4% ice-cold paraformaldehyde (PFA, Sigma) in PBS (pH 7.4) and frozen for brain cryosectioning. Table 1 lists the antigen, immunogen, manufacturer, catalog number, species in which the antibodies were raised, and working dilution for each of the primary antibodies. Information about the specificity of the antibodies in the following paragraphs are from the manufacturers and/or the cited references.

| Characterization of primary antisera
The rabbit and goat polyclonal PHAL antibodies (Vector Laboratories Cat# AS-2300, RRID: AB_2315142 and Cat# AS-2224, RRID: AB_10000080) target both PHAL and the related Phaseolous vulgaris erythroagglutinin. None of them is found in the mammalian brain. The specificity was established by the absence of immunoreactivity in brain sections from naïve animals, from cases in which the uptake and transport of PHAL failed, and from regions that do not receive innervation from the area of tracer injection (Balcita-Pedicino et al., 2011).
The rabbit polyclonal TH antibody (Merck Millipore, AB152, RRID: AB_390204) has been produced against denatured tyrosine hydroxylase from rat pheochromocytoma and targets catecholamine neurons. To test this polyclonal antibody, sections of liver has been used as negative control and brain sections (corpus striatum) and adrenal glands as positive control and produced a pattern of staining similar to that reported elsewhere in the literature (Goff et al., 2015).
The rabbit polyclonal salmon MCH (sMCH) antibody (Risold laboratory, RRID: AB_2616562) recognized the synthetic sMCH (full17-aminoacid, sequence: DTMRKMVGRVYRPCWEV). The specificity of the sMCH antisera was tested by blotting (Risold et al., 1992). The sMCH antibody has been tested on hypothalamic sections from many species (Chometton et al., 2014;Croizier et al., 2013). Its specificity has been verified by liquidphase inhibition, dot blot, and affinity column analyses (Fellmann et al., 1987;Risold et al., 1992). For mice, it was also shown that the labeling was observed exclusively in MCH-GFP cells in the lateral hypothalamus (Croizier et al., 2011). Moreover, double labeling experiments detecting the prepro-MCH mRNA by in situ hybridization and MCH peptides by indirect immunofluorescence were performed in pigs. Both signals were observed in the same cell bodies in the posterior LHA (Chometton et al., 2014).
The mouse monoclonal ORX antibody (Angio-Proteomie, Cat# hAP-0500, ABIN1983384) was produced from a hybridoma (mouse myeloma fused with spleen cells from a mouse immunized with a peptide, aa 35-65, O43612) from human Orexin-A protein. The immunohistochemistry produced a pattern of staining similar to that described elsewhere in the literature (Barbier et al., 2018).

| Enzymatic immunohistochemistry
After rinsing in PBS with 0.3% Triton X100 (PBS-T), free-floating sections were incubated with the anti-PHAL antibodies (
Tissues were washed three times with PBS-T (5 min each) and incubated for 2 hr with appropriate secondary antibodies ( Neither additional treatment was made, except to enhance fluorescent intensity. Nomenclature and nuclear parcellation are based on Mouse Brain Atlas from Franklin and Paxinos (Franklin & Paxinos, 2008), on Rat Brain Atlas from Swanson (Swanson, 2004) and on other studies summarized in Giardino et al., 2018 (Figure 1a).

| Analysis of PHAL fibers density
For the quantitative analysis of PHAL fibers density, each image plane was binarized using Image J to isolate labeled fibers from the background and to compensate for differences in fluorescence intensity. A 220 × 220 μm region of interest (ROI) was placed over the structures of interest. The integrated intensity, which reflects the total number of pixels in the binarized image, was then calculated for each image.

| General distribution patterns of PHAL axons
In all experiments, the PVH in the anterior hypothalamus, and most nuclei of the tuberal hypothalamus including the DMH, the LHA and the ARH received light to intense inputs (Figures 2 and 3). By contrast,

| Pattern of innervation of the periventricular nuclei
In mice, the PVH is poorly compartmentalized when compared to that described in rats (Biag et al., 2012;Simmons & Swanson, 2009). In the neuroendocrine division, it is nearly impossible to distinguish mag- Tyrosine hydroxylase, TH) neurons are not as segregated as in rats (Biag et al., 2012). In this study, we did not intend to precisely delineate PVH divisions in mice, but we based our descriptive and structural analyses on published data, on immunostainings and in situ hybridization for known PVH markers such as AVP, TH and Crh.   (Bittencourt, 2011;Croizier et al., 2010;Cvetkovic et al., 2004;Lein et al., 2007). When used with Pmch-Cre mice, TVA and RG allow the rabies infection in MCH neurons of the capsule of the VMH, the DMH and the LHA as shown by GFP fluorescence (Figures 11b, 12a,b, and 13a,b).
After 9  Monosynaptic inputs to ORX neurons of the LHA were observed in the same BNST divisions as described for MCH (Figure 13c-e).

| The arcuate nucleus
From the dorsomedial division of the BNST, fibers traveled caudally and ventrally near the third ventricle following a periventricular pathway to reach the ARH (Figures 2d,e and 14a,b). These parallel running axons displayed abundant varicosities (Figure 14a,b). Varicosities, collaterals and putative terminal boutons were also generated at several levels of the ARH (Figure 14c-g) such as the retrochiasmatic area  Figure 17c,d). We also noticed a few mcherry-positive cells in the adjacent septohypothalamic nucleus (Figure 17d).

| DISCUSSION
In the present study, we combined anterograde and retrograde tract tracing, histological and immunohistochemical approaches to describe the hypothalamic projections from the dorsomedial division of the BNST. The overall pattern of projections was similar to that described in rats by Dong and Swanson (2006), and we identified several hypothalamic nuclei including the PVH, the DMH, the tuberal nucleus, the LHA and the ARH as main targets of these projections. We

| The paraventricular nucleus
Our study showed a light to moderate innervation of distinct parts of the PVH from the dorsomedial BNST, in line with that described in rats (Dong & Swanson, 2006). Past retrograde studies confirmed the innervation of PVH neurons by neurons of the dorsomedial BNST and most of them have been performed in rats after fluorogold, true blue or cholera toxin B injection centered in the PVH (Cullinan et al., 1993;Moga & Saper, 1994;Prewitt & Herman, 1998;Sawchenko & Swanson, 1983;Spencer et al., 2005;Ulrich-Lai et al., 2011). In mice, rabies-based retrograde tracing showed a specific innervation of Corticotropin-releasing factor receptor 1 (CRFR1)-expressing PVH cells by dorsomedial BNST neurons (Jiang et al., 2018).
In the mouse and contrary to the rat, cytoarchitectonic limits between magnocellular and parvicellular subdivisions of the PVH are mostly undistinguishable (Biag et al., 2012;Simmons & Swanson, 2009). Nevertheless, the distributions of AVP-and CRH-expressing neurons are segregated, somehow defining parvicellular and magnocellular compartments within the mouse PVH. We observed a denser innervation of the CRH-containing area while the AVP-expressing subdivisions contained few PHAL axons.
Therefore, despite a poor compartmentalization of the PVH, our data are in agreement with the rat data, where parvicellular part is more densely innervated than magnocellular subdivision (Dong & Swanson, 2006;Simmons & Swanson, 2009). Although we were unable to verify that CRH neurons are innervated, we identified parvicellular TH-positive cells as being targeted by these projections.

| The dorsomedial nucleus
In mice, the DMH received a moderate innervation from the dorsomedial BNST also in line with that described in rats (Dong & Swanson, 2006). Retrograde tracing based on fluorogold injections in  (Picard et al., 2016), NPY neurons in the lateral parts (Bi et al., 2012;Lein et al., 2007). Moreover, the brain-derived neurotrophic factor (BDNF) receptor, TrkB is expressed in all the divisions of the DMH (Liao et al., 2019). To our knowledge, no published data on the projections from the BNST to the above-mentioned neurons are available in mice neither data on retrograde tracing from the aforementioned neurons. Our results although preliminary, seem to indicate that the dorsomedial BNST may innervate specifically some of those populations and not others (i.e. MCH).

| The arcuate nucleus
As observed in rats (Dong & Swanson, 2006), very intense terminals are centered in the dorsomedial ARH in mice, where most of the POMC and dopamine neuroendocrine neurons are concentrated (Lein et al., 2007;Markakis & Swanson, 1997). We observed a much lighter input ventromedially and ventrolaterally where AgRP/NPY and growth hormone-releasing hormone are, respectively, expressed (Lein et al., 2007;Sawchenko et al., 1985). In our study, the contact between terminal boutons with POMC neurons suggested the presence of synapses. In line with this observation, a previous study used retrograde tracing approach to show a specific innervation of POMC and AgRP neurons by the BNST (Wang et al., 2015), but the precise divisions of the BNST that were concerned were not clearly detailed.
Here, we clarified that mostly dorsomedial divisions of the BNST projected onto POMC and AgRP neurons as well as the adjacent F I G U R E 1 1 (a) Experimental approach. A mix of AAV-EF1a-Flex-C-RVG and AAV-EF1a-Flex-eGFP-TVA was injected at day 0 in the DMH and the dorsal part of the capsule of the VMH of 9-to 10-week-old Pmch-Cre male mice. Two days later, mice received injection of EnvA-Gdeleted-tagRFP pseudotyped rabies. Animals were perfused nine days later for further analyses.

| The dorsal part of the capsule of the VMH and the tuberal nucleus
Similarly to that described in rats (Dong & Swanson, 2006), the core of the VMH is devoid of PHAL-positive fibers arising from the dorsomedial subdivision of the BNST in mice. However, neurons of the capsule of the VMH are innervated and in particular MCH neurons.
Other neuronal populations are also found in this restricted area such as the RF (Arg-Phe) amide-related peptides (RFRP) neurons (Legagneux et al., 2009).
The tuberal nucleus is immediately adjacent to the medial border of the VMH and interrupt the capsule of the VMH (Canteras et al., 1994). The tuberal nucleus provides intense innervation to nuclei containing neurosecretory motoneurons such as the parvicellular part of the PVH, the ARH and to the periaqueductal gray (Canteras et al., 1994). Taken with anatomical data, these observations suggest a role of the tuberal nucleus in the reproductive, defensive and more interestingly for this study in ingestive behaviors (Canteras et al., 1994).

| The lateral hypothalamic area
The LHA is a complex and vast area receiving and sending projections from plethora of brain regions (Bittencourt et al., 1992;Hahn & Swanson, 2010, 2012, 2015. We decided to focus our study only in F I G U R E 1 2 (a) Experimental approach. A mix of AAV-EF1a-Flex-C-RVG and AAV-EF1a-Flex-eGFP-TVA was injected at day 0 in the perifornical area and the LHA of 9-to 10-week-old Pmch-Cre male mice. Two days later, mice received injection of EnvA-G-deleted-tagRFP pseudotyped rabies. Animals were perfused 9 days later for further analyses. tum (Risold & Swanson, 1997). In our study, we cannot exclude that terminal boutons observed in contact with ARH cells come from the F I G U R E 1 3 (a) Experimental approach. A mix of AAV-EF1a-Flex-C-RVG and AAV-EF1a-Flex-eGFP-TVA was injected at day 0 in the perifornical area and the LHA of 12 weeks old orexin-Cre male mice. Two days later, mice received injection of EnvA-G-deleted-tagRFP pseudotyped rabies. Animals were perfused 9 days later for further analyses. nucleus of the BNST . However, in these two cases, we were able to visualize PHAL-labeled fibers in the ARH, suggesting that these projections probably only arose from the dorsomedial division of the BNST. Nonetheless, the projections from the oval nucleus strongly innervate the LHA. In this study we cannot exclude that a significant proportion of PHAL-labeled fibers observed in the LHA also come from the dorsolateral division of the BNST. rabies strain has been shown not to efficiently spread when compared to N2c strain (Reardon et al., 2016). In this study, we used B19 strain, putatively restricting its spread and potentially influencing the number of pre-synaptic neurons. However, it should not influence the distribution of retrogradely-labeled neurons.

| Functional considerations of the dorsomedial BNST projections to hypothalamic areas
Our study reveals that projections from the dorsomedial division of the BNST are connected to hypothalamic nuclei somehow linked to motivation and feeding behaviors such as the PVH (Atasoy et al., 2012;Betley et al., 2013;Hill, 2012) regions forming the visceromotor pattern generator network in rats (Thompson & Swanson, 2003). The DMH is composed of various neurons that are somehow involved in central feeding control and emotional component of eating behavior (Bello & Hajnal, 2010;Bi et al., 2012;de La Serre et al., 2016;Liao et al., 2019;Narayanan et al., 2010;Ryan et al., 2013;Volkow et al., 2003;Yang et al., 2009).
In agreement with published data, our study revealed in the tuberal hypothalamus, a strong innervation of MCH and ORX neurons of the LHA and POMC and AgRP neurons of the ARH, all well-known effectors in the control of appetitive, aversive and goal-oriented behaviors (Diniz & Bittencourt, 2017;Giardino et al., 2018;González et al., 2016;Jennings et al., 2013;Smith et al., 2019;Sohn, 2015;Stuber & Wise, 2016). Interestingly, in addition of being directly innervated by neurons of the dorsomedial divisions of the BNST, the aforementioned hypothalamic nuclei are also strongly interconnected suggesting a reinforcement of their role in feeding and motivated behaviors ( Figure 18). In particular, PVH and DMH neurons are interconnected (Thompson et al., 1996;Thompson & Swanson, 1998) and both innervate arcuate neurons such as POMC and AgRP to control for instance feeding (Garfield et al., 2016;Jeong et al., 2017;Krashes et al., 2014;Wang et al., 2015). In return, ARH neurons innervate the PVH and the DMH (Baquero et al., 2015;Bouret et al., 2004;van der Klaauw et al., 2019). The LHA also receives and send projections to the DMH (Hahn & Swanson, 2010;Thompson & Swanson, 1998) and receive innervation from the arcuate neurons (Bouret et al., 2004;Vogt et al., 2014).

| CONCLUSION
Our study suggests a topographic organization of the projections of the dorsolateral and dorsomedial divisions of the BNST and the adjacent septohypothalamic nucleus onto hypothalamic areas. In agreement with rat data Dong & Swanson, 2006) most lateral divisions of the BNST project to lateral areas of the tuberal hypothalamus including the perifornical area and the LHA, while those arising from medial structures such as the dorsomedial BNST and the septohypothalamic nucleus mostly innervate the periventricular including the PVH and the ARH. In particular, the septohypothalamic nucleus projections more intensely innervate the lateral ARH where most of the POMC neurons are observed, than projections arising from the dorsomedial BNST as confirmed by our retrograde study analyses. Collectively, these data suggest a convergent role in feeding and motivated behaviors of these telencephalic structures through projections onto hypothalamic nuclei with potential functional subtleties involving specific neuronal populations.
The anterior BNST is considered as a well-known stress integrator and is composed of neurons involved in stress and anxiety-like behaviors (Bowers et al., 2012;Füzesi et al., 2016;Giardino et al., 2018;Khan et al., 2018;Knoll & Carlezon, 2010;Vialou et al., 2014). It receives strong afferences from the amygdala, a well-described telencephalic structure detecting emotional and biological stressors Ip et al., 2019).
Altogether, our data argue in favor of a neuroanatomical basis for the interplay between stress and feeding behavior. However, the molecular and functional characterization of these specific neurocircuits is poorly described and would require further analyses.

PEER REVIEW
The peer review history for this article is available at https://publons. com/publon/10.1002/cne.24988.
F I G U R E 1 8 (a) Sagittal view of a brain showing general organization of the projections from the dorsomedial (BNSTdm) and dorsolateral (BNSTdl) divisions of the bed nucleus of the stria terminalis (BNST) to the studied hypothalamic nuclei. The relative strength of each pathway is proportional to the thickness of the black lines. Projections from the BNSTdm mostly innervate periventricular nuclei such as the paraventricular (PVH), the dorsomedial (DMH), the arcuate (ARH) nuclei of the hypothalamus, while those arising from BNSTdm innervate more lateral areas such as the lateral hypothalamic area (LHA). The flatmap is based on Franklin and Paxinos (2008). (b) Schematic illustrating the connections (dark grey) between the hypothalamic nuclei. The PVH, DMH, ARH and LHA are strongly interconnected. V3, third ventricle