Identification of a neural basis for anorexia nervosa

Abstract

Anorexia nervosa (AN) is a debilitating, often lethal, restrictive-type eating disorder without an effective cure. The underlying neural basis of AN has remained elusive without an animal model that has represented all typical AN symptoms. Here we show that aberrant activation of mediobasal hypothalamic (MBH) glutamatergic neurons led to lethal self-starvation, hyperactivity, anhedonia, social phobia, and increased anxiety, all of which represent typical symptoms of AN. These symptoms were selectively exhibited by targeted activation of MBH neurons expressing steroidogenic factor (SF1) and estrogen receptor alpha (ERa). Moreover, the elicited AN symptoms by activation of MBH glutamatergic or SF1/ERa neurons were rescued by removing release of glutamate or brain-derived neurotrophic factor (BDNF) from these neurons. Importantly, BDNF overexpression in SF1/ERa neurons promoted typical AN symptoms, which were suppressed by removing glutamate release. Thus, our findings identify aberrantly enhanced BDNF and consequent augmented glutamate release from SF1/ERa neurons as a neural basis underlying AN.

Introduction

Normal body weight homeostasis is maintained by balanced food intake and energy expenditure (Sternson 2013). Feeding is controlled by a coordinated neural network, disturbance of which can lead to overt feeding abnormalities. Anorexia nervosa (AN) is a devastating eating disorder with substantial morbidity and mortality, frequently occurring in adolescents, and presenting with a much higher incidence in females (Ulfvebrand et al 2015). AN patients exhibit key behavioral features that include voluntary restrictive feeding, increased anxiety, hyperactivity, social phobia, and anhedonia (Achamrah et al 2016, Esposito et al 2018, Kaye et al 2009, Murray et al 2021) (Kohl et al 2004, Levinson & Rodebaugh 2012, Lloyd et al 2018, Murray et al 2022, Treasure et al 2015). AN patients are typically preoccupied with regimens to reduce body weight, cited as owing to a self-perceived “fat body” perspective (Esposito et al 2018), and it is generally accepted that the onset of AN is often triggered by stressful and/or traumatic early life events (Francois et al 2021, Kaye et al 2009, Park et al 2014). Thus, AN is generally viewed as a psychiatric disorder with severe endocrine consequences (Bryant et al 2026). Despite the severe symptoms observed in human patients, the underlying neural basis of AN remains largely unknown.

Extensive research on human AN has provided a rich body of literature on the genetic, brain structural, behavioral and physiological alterations associated with AN development (Puckett et al 2021, Ulfvebrand et al 2015, Zipfel et al 2015). In particular, brain-derived neurotrophic factor (BNDF) has been identified to be involved in AN (Trinh et al 2023). BDNF, a growth factor belonging to the neurotrophic family that plays a pivotal role in neuron development (Park & Poo 2013), is known to be critically involved in the regulation of energy balance, and its expression in multiple brain sites has been shown to play roles in body weight control (Sequeira-Cordero et al 2025, Vanevski & Xu 2013). Notably, human genetic studies have revealed a significant association between BDNF polymorphisms, especially BDNF Val66Met, with AN (Madra & Zeltser 2016, Ramoz et al 2022, Ribasés et al 2005). Consistently, animal models that harbor this mutation exhibit susceptibility in developing AN symptoms under specific challenges (Madra & Zeltser 2016). However, the contribution of this variant to AN development has been questioned due to conflicting reports regarding levels of BNDF expression in human AN patient (Brandys et al 2013, Trinh et al 2023). Thus, whether and how BNDF is specifically involved in AN remains to be established.

Notably, the dearth of mechanistic understanding underlying the neural basis of AN has been hindered due to the lack of an effective animal model that represents and/or recaptures all typical AN symptoms (Francois et al 2021, Scharner & Stengel 2020). To date, most AN animal model studies have relied on inducing activity-based anorexia (ABA), while other recent studies have implemented BDNF mutant mouse models (Madra & Zeltser 2016, Schalla & Stengel 2019). However, all of these models in some way lack key symptoms observed in human AN patients, with the most notable being voluntary restrictive feeding and self-starvation despite a depleted nutritional state (Francois et al 2021). As such, it has been challenging to implement existing models to fully represent and most accurately investigate the underlying neural basis for AN.

Ventromedial hypothalamic neurons (VMH) that reside within the diencephalon have been directly implicated in feeding control. They are mainly glutamatergic, but also express a variety of neuropeptides including BDNF (Khodai & Luckman 2021, Tong et al 2007). Based on various anatomical and gene expression studies, the VMH has been found to harbor several groups of neurons, including overlapping SF1 and ERa expressing neurons. Notably, this SF1 and ERa cohort has been shown to be directly involved in mediating an array of physiological functions, including feeding, stress responses, and social behaviors (Choi et al 2013, Krause et al 2021, Lin et al 2011, Xu et al 2011), implicating VMH neurons and their respective circuits in the development of AN phenotypes.

Here, we show that aberrant activation of ventral hypothalamic (VH) glutamatergic neurons or VMH specific SF1/ERa neurons led to voluntary self-starvation, hyperactivity, anhedonia, social phobia, and increased anxiety, thereby recapitulating all typical symptoms of AN. Importantly, while we found that individual SF1 and ERa neuronal groups both contributing to voluntary starvation, anxiety, and anhedonia, but only ERa neurons contributing to the hyperactivity phenotype. Moreover, our data reveal that aberrantly enhanced BDNF action and consequent increased glutamate release from VMH SF1/ERa neurons drive AN phenotypes, and thus provides an effective animal model to represent and investigate key AN symptoms that manifest in humans.

Results

Chronic activation of VH glutamatergic neurons induced typical anorexia phenotypes

Broadly, brain disease states caused by abnormal neuronal function have been generally considered to be due to chronic alterations in neuron activity (Scharner & Stengel 2019). We have previously achieved chronic activation of various groups of brain neurons by the targeted genetic and cell type-specific expression of the mutated bacterial sodium channel NaChBac (Cai et al 2024, Li et al 2025, Li et al 2023, Zhu et al 2020, Patel et al 2019). To explore the impact of chronic activation of VMH glutamatergic neurons, we delivered Cre-dependent AAV-DIO-NaChBac-GFP or control viral vectors to the VMH of Vglut2-Cre mice. Towards this, we first injected the NaChBac vector to one side of the VMH, and a control AAV-DIO-mCherry vector contralaterally in Vglut2-Cre mice (Fig. 1A). Compared to the control side, the side with NaChBac exhibited a dramatic increase in c-Fos expression, suggesting increased neuron activity (Fig. 1B, arrows). We next delivered the NaChBac virus to both sides of the hypothalamus (Fig. 1C). Consistently, compared to AAV-DIO-eYFP injected controls, the NaChBac injected mice showed dramatic increases in c-Fos in both sides of the hypothalamus (Supplementary Fig. S1A), demonstrating chronic activation of VMNH glutamatergic neurons by targeted NaChBac expression.

Figure 1. Chronic activation of VMH glutamatergic neurons induces anorexia-like phenotypes.

(A) Schematic of unilateral viral delivery (AAV-DIO-EYFP, AAV-DIO-mCherry, or AAV-DIO-NaChBac) into the VMH of Vglut2-Cre mice. (B) Representative images from unilateral injections in the same brain section showing expression of mCherry, NaChBac-GFP, and robust c-Fos induction (arrows) on the NaChBac-injected side. (C) Schematic of bilateral viral delivery (AAV-DIO-EYFP, AAV-DIO-mCherry, or AAV-DIO-NaChBac) into the VH of Vglut2-Cre mice. (D, E) Body weight trajectories of female (D) and male (E) Vglut2-Cre mice after bilateral VH injection of AAV-DIO-NaChBac or AAV-DIO-GFP. Females, n = 10 per group; males, n = 8 per group; two-way ANOVA; *P < 0.05, **P < 0.01, ***P < 0.001. (F, G) Kaplan–Meier survival analysis of female (F) and male (G) mice expressing GFP or NaChBac. Mice were considered as non-survial when body weight loss reached 30%, log-rank test, P < 0.001 for both sexes. (H) Daily food intake in EYFP- and NaChBac-expressing female mice. n = 6 per group; two-way ANOVA; ***P < 0.001. (I) Cumulative food intake during fasting–refeeding in EYFP- and NaChBac-expressing female mice. n = 5 per group; two-way ANOVA; **P < 0.01, ***P < 0.001. (J) Real-time analysis of spontaneous locomotor activity of female mice measured in metabolic cages (TSE). (K) Quantification of locomotor activity during light and dark phases. n = 6 per group; unpaired two-tailed t test; P = 0.017 (light), P = 0.387 (dark). (L, M) Open-field test. (K) Representative movement trajectories. (L) Total distance traveled in 15 min. n = 5 per group; unpaired two-tailed t test, P=0.001. (N) Schematic of the sucrose preference test. (O) Sucrose preference in EYFP- and NaChBac-expressing female mice. n = 5 (EYFP), n = 6 (NaChBac); unpaired two-tailed t test, P=0.011. (P, Q) Light–dark box test. (O) Representative trajectories. (P) Time spent in the light room. n = 5 (EYFP), n = 6 (NaChBac); unpaired two-tailed t test, P=0.031. (R, S) Nesting behavior. (Q) Representative images. (R) Nesting scores over time. n = 5 per group; two-way ANOVA, P=0.048 at 2 h, P<0.001 at 4 h.(T, U) Three-chamber sociability test. (S) Representative trajectories. (T) Time spent interacting with a stranger mouse. n = 5 per group; unpaired two-tailed t test; P=0.019.

Using these models, we next evaluated how chronic activation of VMH glutamatergic neurons impacted physiology and behavior. Strikingly, mice with chronic activation of VMH glutamatergic neurons exhibited steep body weight reduction starting 5 days post viral delivery in both females (Fig. 1D) and males (Fig. 1E), which eventually resulted in death (Figs. 1F and 1G). Notably, these mice exhibited voluntary reduction in food intake, measured in females by either daily ad libitum food consumption (Fig. 1H) or fast-refeeding (Fig. 1I). We determined that the voluntary starvation was not due to a secondary effect of reduced locomotion since these mice exhibited overall increased locomotion as detected by more beam breaks during arena monitoring (Figs. 1J and 1K) and in open field tests (OPT, Figs. 1L and 1M). These behavioral changes resemble those seen in AN patients (Kaye et al 2009), prompting us to further explore other behavioral changes known to be associated with AN, including anhedonia, anxiety, and sociability.

To test anhedonia, we performed a two-bottle sucrose preference tests (SPT), in which mice were allowed to choose between water and sucrose (Fig. 1N). Shown here in female mice, the NaChBac expressing models showed a dramatic reduction in sucrose consumption compared to controls (Fig. 1O), suggesting a state of anhedonia. Also, when subjected to light-dark tests (LDT), the NaChBac models spent significantly less time in the light room (Figs. 1P and 1Q). In addition, NaChBac mice exhibited less nest building behavior compared to controls (Figs. 1R and 1S), which is another indicator for anxiety in mice. Together these data suggest increased behavioral signs of anxiety as a result of increased VMH glutamatergic neuron activity. To next evaluate sociability, we used a 3-chamber test, in which mice were placed in the center chamber with the two side chambers, with one side chamber containing an empty wire cage, and the other a stranger conspecific (Fig. 1T). Compared to controls, the NaChBac females spent significantly less time with conspecifics (Fig. 1U), suggesting a sign of social phobia. Collectively, these behavioral phenotypes demonstrate that chronic activation of VMH glutamatergic neurons causes lethal voluntary starvation, hyperactivity, anxiety, social phobia and anhedonia, all typical symptoms of AN. Importantly, we observed that, while the NaChBac mice exhibited no obvious jumping behaviors in home cages, these mice showed frequent jumping behaviors in unfamiliar test areas, suggesting an elevated level of anxiety triggered by new environmental cues, reminiscent of the observations in humans, where it has been found that AN can be triggered by stressful or traumatic challenges (Madra & Zeltser 2016, Steiger & Thaler 2016). Of note, all of these behavioral alterations were similarly observed in male mice (Supplementary Figs. S1B–S1N), suggesting no sexual dimorphism in anorexia-like phenotypes by chronic activation of VH glutamatergic neurons.

Chronic activation of VMH SF1 and ERa neurons elicits typical anorexia phenotypes

Stereotaxic delivery of Cre-dependent AAV vectors targeting the VH area using Vglut2-Cre may hit regions beyond the VMH area as the adjacent regions are known to contain glutamatergic and express Vglut2. However, the VMH area can be more precisely defined by the expression of SF1. In addition, within the VMH, SF1 expression partially overlaps with ERα (Cheung et al 2013), which is primarily localized to the ventrolateral region of the VMH. Also, within this area female mice harbor more cells that express ERα than males (Supplementary Fig. S2A). Thus, to achieve more specific targeting VMH neurons, we first generated compound SF1-Cre::ERα-Cre mice and delivered the AAV-DIO-NaChBac-GFP virus to the VMH, with the individual Cre lines serving as control groups (Fig. 2A). We first confirmed targeted expression of NaChBac and consequent c-Fos expression in both the compound the individual Cre lines (Fig. 2B). Consistent with the partial overlap between SF1 and ERa neurons, the compound mice exhibited more NaChBac and c-Fos expression throughout the VMH compared to individual Cre drivers (Fig. 2B, arrows), suggesting successful chronic activation of the overlapping SF1/ERα neuronal population. Interestingly, chronic activation of SF1/ERα neurons led to gradual reduction in body weight and eventual death in both males (Fig. 2C) and females (Fig. 2D–E), whereas mice with either chronic SF1 or ERa neuron activation alone showed intermediate body weight reductions without lethality (Figs. 2C and 2D). Notably, females routinely showed earlier times of death compared to males (Fig. 2E), suggesting a sexual dimorphic effect. We reasoned this likely reflected the more numerous ERα-positive neurons present in the VMH of female versus males. Importantly, we found that the dramatic lethal body weight reduction was associated with reduced food intake in both females (Fig. 2F) and males (Fig. 2G), and that reduced fasting induced refeeding in females (Fig. 2H). Consistent with the body weight phenotype, the reduction of food intake in the double Cre mice was much more significant compared to either single Cre mice in both sexes (Figs. 2F and 2G).

Figure. 2. Chronic activation of VMH SF1 and ERa neurons leads to typical anorexia like behaviors.

(A) Schematic of viral delivery (AAV-DIO-EYFP or AAV-DIO-NaChBac) into the VMH of SF1-Cre, ERa-Cre, and SF1-Cre::ERa-Cre mice. (B) Representative images showing EYFP or NaChBac expression patterns in VMH, and c-Fos induction (arrows) in NaChBac-expressing mice. (C, D) Body weight trajectories of female (C) and male (D) SF1-Cre, ERa-Cre, and SF1-Cre::ERa-Cre mice after VMH injection of AAV-DIO-NaChBac or AAV-DIO-EYFP. Females, n=8 (EYFP), n=6 (SF1-Cre/NaChBac), n=6 (ERa-Cre/NaChBac), n=8 (SF1-Cre::ERa-Cre/NaChBac); males, n=6 (EYFP), n=6 (SF1-Cre/NaChBac), n=8 (ERa-Cre/NaChBac), n=8 (SF1-Cre::ERa-Cre/NaChBac); two-way ANOVA, ***P<0.001. (E) Kaplan–Meier survival analysis of female and male SF1-Cre::ERa-Cre mice expressing NaChBac. Mice were considered as non-survival when body weight loss reached 30%, log-rank test, P=0.021 between males and females. (F, G) Daily food intake in females (F) and males (F) EYFP- and NaChBac-expressing mice. Females, n=7 (EYFP), n=6 (SF1-Cre/NaChBac), n=6 (ERa-Cre/NaChBac), n=6 (SF1-Cre::ERa-Cre/NaChBac); males, n=6 (EYFP), n=5 (SF1-Cre/NaChBac), n=6 (ERa-Cre/NaChBac), n=6 (SF1-Cre::ERa-Cre/NaChBac); two-way ANOVA; *P<0.05, **P<0.01, ***P<0.001 versus EYFP group; &p=0.023 versus ERa-Cre/NaChBac group.. (H) Cumulative food intake during fasting–refeeding in EYFP- and NaChBac-expressing female mice. n=5 (EYFP), n=5 (SF1-Cre/NaChBac), n=7 (ERa-Cre/NaChBac), n=6 (SF1-Cre::ERa-Cre/NaChBac); two-way ANOVA, **P=0.007, ***P<0.001. (I) Sucrose preference in EYFP- and NaChBac-expressing female SF1-Cre::ERa-Cre mice. n = 5 per group; unpaired t test, P=0.003. (J-M) total distance traveled in open the field test (J), Time spent in the light room in light-dark test (K), nesting score (L), and social time with stranger in three-chamber test (M) in female mice. n=7 (EYFP), n=6 (SF1-Cre/NaChBac), n=6 (ERa-Cre/NaChBac), n=6 (SF1-Cre::ERa-Cre/NaChBac); two-way ANOVA; *P<0.05, **P<0.01, ***P<0.001 versus EYFP; ###P<0.001 versus SF1-Cre/NaChBac.

In females, activation of SF1/ERα neurons reduced sucrose consumption in SPT (Fig. 2I), suggesting a state of anhedonia. Interestingly, compared to controls, while activation of SF1 neurons caused a reduction in locomotion, mice with elevated activity in ERα neurons-only or SF1/ERa neurons led to increased locomotion in the OFT (Fig. 2J), suggesting a direct role of ERα expressing neurons in promoting locomotion. Notably, given the increased locomotion, activation of SF1/ERα neurons led to a more dramatic reduction in time spent in the light during the LDT (Fig. 2K), suggesting increased anxiety. Additionally, activation of SF1/ERα neurons consistently caused a more dramatic reduction in both nesting (Fig. 2L) and sociability behaviors (Fig. 2M), whereas activation of individual SF1 or ERa neuron groups caused a modest reduction (Figs. 2L and 2M). Worth noting, the behavioral differences we described in females were also observed in males (Supplementary Fig. S2B–S2E). Collectively, these results suggest that chronic activation of SF1/ERα neurons induces distinct AN phenotypes, with individual SF1 and ERa neuronal groups both contributing to voluntary starvation, anxiety, and anhedonia, but only ERa neurons contributing to the hyperactivity phenotype.

Glutamate and BDNF are both required for anorexia phenotypes

VMH neurons are known to release multiple neurotransmitters, including fast-acting glutamate and BDNF (Harvey & Rios 2024). To first examine the role of glutamate release from these neurons, we employed a CRISPR-Cas9 approach by delivering either AAV-DIO-sgVglut2, which specifically lesions the VGLUT2 locus (Hunker et al 2020), or control AAV-DIO-sgRNA vectors to the VMH area of SF1-Cre::ERα -Cre mice (Fig. 3A). Compared to mice with control AAV-DIO-EYFP vector injections, those that received a mixture of AAV-DIO-NaChBac-GFP and control AAV-DIO-sgRNA vectors gradually reduced body weight within 5 days after receiving viral injections (Fig. 3B). Strikingly, this led to a lethal phenotype with 21 days (Fig. 3C), but could be completely rescued in mice that received injections of a mixture of AAV-DIO-NaChBac-GFP and the selective AAV-DIO-sgVglut2 construct (Figs. 3B and 3C). Moreover, we noted that Vglut2 deletion rescued the NaChBac-mediated voluntary starvation (Fig. 3D), hyperactivity (Fig. 3E), reduced time spent in the light during the LDT (Fig. 3F), and reduced sociability (Fig. 3G). These results collectively suggest that augmented glutamate release mediates the anorexia phenotype induced by chronic activation of SF1/ERα neurons.

Figure. 3. BDNF and Glutamate release are both required towards anorexia phenotypes.

(A) Diagram of viral delivery (AAV-DIO-EYFP, AAV-DIO-NaChBac + AAV-DIO-sgRNA, AAV-DIO-NaChBac + AAV-DIO-sgVglut2) into the VMH of SF1-Cre::ERa-Cre male mice. (B) Body weight trajectories of male SF1-Cre::ERa-Cre mice. n=7 per group, two-way ANOVA, ***P<0.001. Numbers shown in the figure represent the number of remaining surviving mice included in the analysis at each time point. (C) Kaplan–Meier survival analysis of mice in EYFP, NaChBac+sgRNA, and NaChBac+sgVglut2 groups. Mice were considered non-survival when body weight loss reached 30%, log-rank test, ***P<0.001 versus EYFP, ###P<0.001 versus NaChBac+sgVglut2. (D) Daily food intake. n=7 per group, one-way ANOVA, **P=0.003 versus EYFP, #P=0.031 versus NahBac+sgRNA. (E-G) Total distance traveled in open field test(E), time spent in the light room in light-dark test (F), social time with stranger in three-chamber test (G) n=7 per group, one-way ANOVA, **P<0.01, ***P<0.001 versus EYFP; #P<0.05, ##P<0.01 versus NaChBac+sgRNA. (H) Diagram of viral delivery (AAV-DIO-EYFP, AAV-DIO-NaChBac) into VH of SF1-Cre::ERa-Cre::BDNFflox/+and SF1-Cre::ERa-Cre::BDNFflox/flox female mice. (I) Body weight trajectories of female mice. n=8 (SF1-Cre::ERa-Cre::BDNFflox/++EYFP), n=6 (SF1-Cre::ERa-Cre::BDNFflox/++NaChBac), n=6 ( SF1-Cre::ERa-Cre::BDNFflox/flox+NaChBac ); two-way ANOVA, ***P<0.001. Numbers shown in the figure represent the number of mice included in the analysis at each time point. (J) Kaplan–Meier survival analysis of mice in SF1-Cre::ERa-Cre::BDNFflox/++EYFP, SF1-Cre::ERa-Cre::BDNFflox/++NaChBac, SF1-Cre::ERa-Cre::BDNFflox/flox+NaChBac groups. Mice were considered as non-survival when body weight loss reached 30%, log-rank test, ***P<0.001 versus EYFP, ###P<0.001 versus SF1-Cre::ERa-Cre::BDNFflox/flox+NaChBac. (K) Daily food intake. n=5–6 per group, one-way ANOVA, ***P<0.001 versus EYF; ##P=0.009 versus SF1-Cre::ERa-Cre::BDNFflox/++NaChBac (L-N) Total distance traveled in open field test(L), time spent in the light room in light-dark test (M), social time with stranger in three-chamber test (N). n=6–8 per group, one-way ANOVA, *P<0.05, **P<0.01, ***P<0.001 versus EYFP.

To next examine the role for BDNF in SF1/ERa neurons, we generated SF1-Cre::Esr1-Cre::BDNF^flox/flox mice and delivered AAV-DIO-NaChBac-GFP vectors to the VMH area (Fig. 3H). We first confirmed effective deletion of BDNF in the VMH of SF1-Cre::Esr1-Cre::BDNF^flox/flox mice, compared to control SF1-Cre::Esr1-Cre::BDNF^flox/+ mice (Supplementary Fig. 3). Notably, we did not observe any differences in body weight between SF1-Cre::ERa-Cre::BDNF^flox/+ and SF1-Cre::ERa-Cre::BNDF^flox/flox mice during the age period this study was conducted (Fig. 3I), suggesting that BDNF deletion has no impact on baseline body weight in an early age perod. As expected from our prior experiments, NaChBac injected SF1-Cre::ERa-Cre::BDNF^flox/+ mice exhibited a gradual body weight loss after day 5 of viral delivery (Fig. 3I), which eventually caused a lethal phenotype (Fig. 3J). Interestingly however, SF1-Cre::ERa-Cre::BDNF^flox/flox mice showed an initial reduction in body weight, with a lesser degree compared to BNDF^flox/+ mice, but then gradually recovered to a nearly complete rescue 28 days after viral delivery. In line with this, contrasting to the lethality we observed in SF1-Cre::ERa-Cre::BDNF^flox/+mice, no lethality was observed in SF1-Cre::ERa-Cre::BDNF^flox/flox mice (Fig. 3J), suggesting a delayed rescue in body weight with selective loss of BDNF.

Consistent with the observed body weight effects, BDNF deletion rescued the effect of NaChBac-induced voluntary starvation (Fig. 3K). Notably, BDNF deletion also rescued hyperactivity (Fig. 3L), and reduced time spent in the light during the LDT (Fig. 3M) at 4 wks, but not earlier (2 wks after viral delivery), suggesting again a delayed rescue of AN-like phenotypes with BDNF deletion. Finally, reduced sociability was not ever fully rescued at any time tested (Fig. 3N). Together, these results collectively suggest that both glutamate and BDNF release mediate the anorexia phenotype induced by chronic activation of SF1/ERα neurons with a delayed effect with loss of BDNF.

Augmented BDNF action from VH glutamatergic neurons promotes anorexia phenotypes

Given that we observed clear rescue effects on the anorexia phenotypes induced by VMH neuron activation with BDNF deletion, we next examined whether augmented BDNF action from these neurons was sufficient to cause AN phenotypes. Towards, we generated AAV-DIO-BDNF-p2A-mCherry viruses, which conditionally overexpress BDNF in a Cre-dependent manner. To first examine the efficiency of viral mediated BNDF expression, we delivered the AAV-DIO-BDNF-p2A-mCherry virus to one side of the hypothalamus and control GFP vectors to the contralateral side (Fig. 4A). After 2 wks of expression, BDNF immunostaining showed that the side with BNDF viral injection harbored numerous neurons with BDNF and mCherry colocalization, while the control injection side had sparse BDNF positive neurons (Fig. 4B), confirming successful BDNF overexpression. We next bilaterally delivered the BDNF virus to the VMH area of Vglut2-Cre mice (Fig. 4C), and subsequently confirmed targeted bilateral delivery to the hypothalamus (Supplementary Fig. S4A). Using this model, we observed that BDNF-injected mice showed a gradual reduction in body weight, and by week 8 post-injection, body weights were reduced by 20% in both females (Fig. 4D) and males (Fig. 4E). Associated with this rapid weight loss, daily food intake was dramatically reduced in both females (Fig. 4F) and males (Fig. 4G). Further, fasting-refeeding responses were also dramatically reduced (Figs. 4H and 4I), suggesting voluntary restrictive feeding.

Figure. 4. Augmented BDNF action from VH glutamatergic neurons promoted anorexia like behaviors.

(A) Diagram of unilateral viral delivery (AAV-DIO-EYFP, AAV-DIO-BDNF-mCherry) into the VMH of Vglut2-Cre mice. (B) Representative images from unilateral injections in the same brain section showing EYFP (green), BDNF-mCherry (red) expression, and BDNF immunostaining (cyan). (B1) a higher-magnification view of the left boxed area on the AAV-DIO-EYFP injection side. (B2) a higher-magnification view of the right boxed area on the AAV-DIO-BDNF-mCherry injection side. Arrows indicate BDNF immunostaining positive cells. (C) Diagram of bilateral viral delivery (AAV-DIO-EYFP, AAV-DIO-BDNF-mCherry) into the VMH of Vglut2-Cre mice. (D, E) Body weight trajectories of female (D) and male (E) Vglut2-Cre mice after VMH injection of AAV-DIO-EYFP or AAV-DIO-BDNF. Females, n=6 per group; males, n=10 (EYFP), n=9 (BDNF); two-way ANOVA, *P<0.05, **P<0.01, ***P<0.001. (F, G) Daily food intake in EYFP- and BDNF-expressing Vglut2-Cre female (F) or male (G) mice. n=6 per group for both sexes; unpaired t test; P=0.001 (female), P=0.006 (male). (H, I) Cumulative food intake during fasting–refeeding in EYFP- and BDNF-expressing female (H) or male (I) Vglut2-Cre mice. Females, n = 5 per group; males, n=6–7 per group; two-way ANOVA; **P < 0.01, ***P < 0.001. (J) Sucrose preference in EYFP- and BDNF-expressing female Vglut2-Cre mice. n=6 per group; unpaired two-tailed t test, P=0.001. (K-N) Total distance traveled in the open field test (K), time spent in the light room in light-dark test (L), nesting behavior (M), social time with stranger in three-chamber test (N) in female mice. n=6 per group; unpaired two-tailed t test (J,K,M), two-way ANOVA (L); *P<0.05, **P<0.01 versus EYFP.

Alongside changes in food consumption, compared to controls the BDNF injected female mice showed reduced sucrose consumption (Fig. 4J) in the SPT, and increased locomotion in the OFT (Fig. 4K). Notably, even with increased locomotion, these mice spent less time in the light during the LDT (Fig. 4L). Similarly, BDNF-injected mice exhibited significantly lower scores in nest building (Fig. 4M). In addition, these mice exhibited spontaneous jumping behaviors upon being placed in a new cage, which was not observed in control mice. Thus, augmented BDNF action from VMH glutamatergic neurons causes anxiety and induces an increased stress response in a new environment. The BDNF injected female mice also showed significantly less time interacting with their conspecifics than controls (Fig. 4N), suggesting a behavioral sign of social phobia. These data collectively suggest that BNDF overexpression in VMH glutamatergic neurons led to typical AN phenotypes. Importantly, we observed the same phenotypes in males (Supplementary Fig. S4B–S4J), suggesting a shared neural pathway in promoting BNDF-induced anorexia phenotypes between males and females.

Augmented BDNF action from SF1/ERa neurons promotes anorexia phenotypes

To examine the impact of selective expression of BDNF in SF1/ERa neurons, we delivered AAV-DIO-BDNF-p2A-mCherry viral vectors to the VMH of SF1-Cre::ERa-Cre mice, in addition to the individual Cre mice, using AAV-DIO-EYFP vectors as controls (Fig. 5A). Selective overexpression of BDNF was first confirmed in both SF1/ERa neurons and the individual Cre driver neurons (Fig. 5B). When monitoring physiological and behavioral outputs, we noted that targeted overexpression of BDNF in VMH SF1, ERa, and SF1/ERa neurons all caused a gradual reduction in body weight in both females (Fig. 5C) and males (Fig. 5D). Notably, the degree of body weight reduction was greater in SF1/ERa targeted models, compared to individual SF1 or ERa targeted groups (Figs. 5C and 5D). As expected, body weight reduction was directly associated with reduced daily food intake and decreased fasting-induced refeeding in both females and males (Fig. 5E–H), suggesting a BDNF- induced voluntary restrictive feeding effect.

Figure. 5. Augmented BDNF action from SF1/ERa neurons leads to anorexia like phenotypes.

(A) Diagram of viral delivery (AAV-DIO-EYFP, AAV-DIO-BDNF-mCherry) into the VMH of SF1-Cre, ERa-Cre, and SF1-Cre::ERa-Cre mice. (B) ) Representative images showing patterns of EYFP or NaChBac expression in the VMH of SF1-Cre, ERa-Cre, and SF1-Cre::ERa-Cre mice. (C, D) Body weight trajectories of female (C) and male (D) SF1-Cre, ERa-Cre, and SF1-Cre::ERa-Cre mice after VH injection of AAV-DIO-EYFP or AAV-DIO-BDNF. Females, n=6 (EYFP, SF1-Cre+BDNF, ERa-Cre+BDNF), n=12 (SF1-Cre::ERa-Cre+BDNF); males, n=7 (EYFP), n=4 (SF1-Cre+BDNF), n=6 (ERa-Cre+BDNF, SF1-Cre::ERa-Cre+BDNF); two-way ANOVA; *P<0.05, **P<0.01, ***P<0.001. (E, F) Daily food intake in EYFP- and BDNF-expressing female (E) or male (F) mice. females, n=6 per group; males, n=4–7 per group; one-way ANOVA; *P<0.05, **P<0.01, ***P<0.001 versus EYFP. (G, H) Cumulative food intake during fasting–refeeding in EYFP- and BDNF-expressing female (G) or male (H) mice. females, n=6 per group; males, n=4–7 per group; two-way ANOVA; *P<0.05, ***P<0.001. (I) Sucrose preference in EYFP- and BDNF-expressing female SF1-Cre::Esr-Cre+BDNF mice. n=6 per group; unpaired two-tailed t test, P=0.002. (J) Total number of jumps during the 10-min test in female mice. n=6 (EYFP, SF1-Cre+BDNF, ERa-Cre+BDNF), n=12 (SF1-Cre::ERa-Cre+BDNF); one-way ANOVA; **P<0.01 versus EYFP. (K) Quantification of locomotor activity during dark phase (TSE system). n = 5 per group; unpaired two-tailed t test; P=0.042. (L-N) time spent in the light room in light-dark test (L), nesting behavior (M), social time with stranger in three-chamber test (N) in female mice. N=6–8 per group; one-way ANOVA (L, N), two-way ANOVA (M); *P<0.05, ***P<0.001 versus EYFP.

Of note, BDNF overexpression in SF1/ERa neurons also exhibited reduced sucrose consumption in SPT (Fig. 5I), suggesting increased anhedonia. However, distinct from BDNF expression in Vglut2-Cre mice which caused hyperactivity, its expression in ERa and SF1/ERa neurons exhibited no changes in OPT activity levels in either females (Supplementary Figs. S5A and S5B) or males (Supplementary Fig. S5C). Notably though, BDNF expression in ERa and SF1/ERa neurons led to strong spontaneous jumping behaviors in both females (Fig. 5J) and males (Supplementary Fig. S5D), which may have affected their locomotor activity measures in the OPT testing chambers. Indeed, when measured using TSE metabolic home cage chambers, mice with BDNF overexpression in SF1/ERa neurons showed increased locomotion (Fig. 5K). Notably, BDNF expression in SF1 neurons alone caused reduced locomotion (Supplementary Figs. S5A, S5B and S5C) but not jumping behavior (Fig. 5J and Supplementary Fig. S5D), which is consistent with the results from NaChBac-mediated activation of SF1 neurons (Fig. 2I), again suggesting specific contribution of ERa neurons to the AN associated hyperactivity phenotype.

Finally, female mice with BDNF expression in SF1, ERa, or SF1/ERa neurons showed reduced times spent in the light (Fig. 5L), as well as lower nesting scores (Fig. 5M) compared to controls, collectively suggesting increased levels of anxiety. We also subjected these mice to a 3-chamber sociability test, and female mice with BDNF overexpression in all 3 models showed reduced time interacting with conspecifics, with SF1/ERa BDNF expressing mice showing a much greater reduction compared to controls (Fig. 5N). We also observed a similar reduction in males, with the exception that BDNF overexpression in only ERa neurons caused no differences in the LDT or social phobia assay (Supplementary Figs. S5E–S5J). Together these results demonstrate that targeted BDNF overexpression in SF1/ERa neurons produced a collection of behavioral phenotypes that closely resemble AN phenotypes, with differential contributions from Sf1 and ERa neurons towards hyperactivity and anxiety between males and females.

BDNF-induced anorexia phenotypes are rescued by deletion of glutamate release

Our prior experiments reveal that loss of glutamate release fully rescued the anorexia-like phenotypes we observed with chronic activation of SF1/ERa neurons, while BDNF deletion only resulted in a delayed rescue, suggesting a possibility that BDNF acts upstream of glutamate release. To test this, we first examined the impact on downstream target neuron activity with deletion of either glutamate release or BDNF. For this we first examined c-Fos expression in known downstream target area of VMH SF1/ERa neurons. Notably, we found that NaChBac expression in SF1/ERa neurons caused a significant increase in c-Fos expression within the paraventricular thalamus (PVT), which was completely abrogated by deletion of glutamate release (Fig. 6A and 6B). We also noted similar results in 4 other known downstream targets of VMH SF1/ERa neurons, including the lateral septum (Supplementary Fig. S6A and S6B), preoptic area (POA, Supplementary Fig. S6C and S6D), periaqueductal grey (PAG, Supplementary Fig. S6E and S6F) and the peri-locus coeruleus regions (Peri-LC, Supplementary Fig. S6G and S6H). Together these changes in VMH outputs suggest that ablation of glutamate release effectively blocked excitatory inputs to these downstream neurons. Interestingly, deletion of BDNF also caused a similar reduction of c-Fos in the PVT (Fig. 6C and 6D) LS, POA, PAG, and Peri-LC (Supplementary Fig. 7), suggesting that deletion of BDNF mimics the effect of abating glutamate release in blocking excitatory inputs to these downstream targets.

Figure. 6. Rescuing effects on BNDF-induced anorexia phenotypes by deletion of glutamate release.

(A) Representative images showing GFP and c-Fos (arrows) expression patterns in the PVT, a downstream target of the VMH, in AAV-DIO-EYFP, AAV-DIO-NaChBac+sgRNA, AAV-DIO-NaChBac+sgVglut2 injection SF1-Cre::ERa-Cre mice. scale bar=250 um. (B) Representative images showing GFP (green) and c-Fos (red, arrows) expression patterns in the PVT, a downstream target of the VMH, in SF1-Cre::ERa-Cre::BDNFflox/++EYFP, SF1-Cre::ERa-Cre::BDNFflox/++NaChBac, SF1-Cre::ERa-Cre::BDNFflox/flox+NaChBac mice. scale bar=250 um. (C) Number of c-Fos positive cells in the PVT region related to panel A. n=4–5 per group; one-way ANOVA; ***P<0.001 versus EYFP; ###P<0.001 versus NaChBac+sgRNA. (D) number of c-Fos positive cells in the PVT region related to panel A. n=4–5 per group; one-way ANOVA; ***P<0.001 versus EYFP; ###P<0.001 versus SF1-Cre::ERa-Cre:: BDNFflox/flox+NaChBac. (E) Diagram of viral delivery (AAV-DIO-mCherry, AAV-DIO-BDNF-mCherry+sgRNA, AAV-DIO-BDNF-mCherry+sgVglut2) into the VMH of SF1-Cre::ERa-Cre mice. (F) Body weight trajectories of SF1-Cre::ERa-Cre mice after VH injection of virus. n=4 (EYFP, BDNF+sgVglut2), n=5 (BDNF+sgRNA); two-way ANOVA; *P<0.05, **P<0.01, ***P<0.001 versus mCherry; #P<0.05 versus BDNF+sgRNA. (G) Daily food intake. n=4–5 per group, one-way ANOVA, **P=0.006 versus mCherry; #P=0.043 versus BDNF+sgRNA. (H) Total number of jumps during a 10-min test. n=4–5 per group; one-way ANOVA; ***P<0.01 versus mCherry; ###P<0.001 versus BDNF+sgRNA. (I, J) time spent in the light room in light-dark test (I), social time with stranger in three-chamber test (J). n=4–5 per group; one-way ANOVA; *P<0.05, ***P<0.01 versus mCherry; #P<0.05 versus BDNF+sgRNA. (K) Representative images showing mCherry (green) and c-Fos (magenta, arrows) expression patterns in the PVT, a downstream of VMH, in mCherry, BDNF+sgRNA, BDNF+sgVglut2 injection SF1-Cre::ERa-Cre mice. scale bar=250 um. (L) number of c-Fos positive cells in PVT region related to panel K. n=4–5 per group; one-way ANOVA; ***P<0.001 versus mCherry; ###P<0.001 versus BDNF+sgRNA.

We then sought to determine whether the observed anorexia phenotypes induced by BDNF overexpression depends on glutamate release. Toward this, we injected a mixture of AAV-DIO-BDNF-p2A-mCherry and AAV-DIO-Cas9-sgVglut2 or control sgRNA to the VMH of SF1-Cre::ERa-Cre mice (Fig. 6E). As expected, the BDNF and control sgRNA injected mice exhibited gradual reduction in body weight, compared to controls (Fig. 6F), whereas the BDNF and sgVglut2 injected mice exhibited a complete rescue in body weight (Fig. 6F). Consistently, our data reveal that CRISPR-mediated deletion of Vglut2 rescued voluntary restrictive feeding (Fig. 6G), spontaneous jumping (Fig. 6H), reduced time in light (Fig. 6I), and reduced social interaction time (Fig. 6J). These results indicate that the effect of BDNF overexpression is mediated via increased glutamate release. To further confirm this, we also examined c-Fos expression in downstream neurons in these mice and found that Vglut2 deletion reversed the increase in c-Fos expression in PVT (Figs. 6K and 6L), LS (Supplementary Figs. S8A and S8B), POA (Supplementary Fig. S8C and S8D), PAG (Supplementary Fig. S8E and S8F), and Peri-LC (Supplementary Figs. S8G and S8H), confirming that glutamate release mediates the increased excitatory effect of BDNF onto downstream targets.

BDNF has been suggested to increase glutamate release via both postsynaptic augmentation of glutamate receptor action and presynaptic tonic glutamate release (Carvalho et al 2008). We reasoned that if BDNF acts by postsynaptic augmentation of glutamate receptor action, then overexpression of BDNF in SF1/ERa neurons should increase c-Fos expression in these neurons. However, we found that c-Fos expression in SF1/ERa neurons was not increased compared to control mCherry injected mice, or those receiving BDNF injections with concomitant deletion of Vglut2 (Supplementary Fig. 9). This evidence supports the idea that BDNF expression facilitates tonic glutamate release from SF1/ERa glutamatergic neurons.

Given the importance of the VMH in feeding behavior, we also considered it’s potential role in controlling candidate downstream targets our activity mapping data revealed. A previous study suggested a role for VMH projections to hindbrain me5 trigeminal neurons in controlling jaw movements, and thereby affecting food ingestion (Kosse et al 2024). To examine if this pathway contributes to the anorexia related phenotypes we observed, we further mapped the downstream pathways of SF1/ERa neurons within the Peri-LC, in which me5 neurons are located. We found that the region of the peri-LC that showed the highest levels of c-Fos expression was the lateral dorsal tegmental region (LDTg). Notably, however, minimal numbers of c-Fos expressing me5 neurons were present in either the NaChBac-injected (Supplementary Figs. S6G and Fig. S6H) or BDNF-injected mice (Supplementary Figs. S7G and S7H). To specifically examine the contribution of Peri-LC neurons in the observed AN phenotypes, we delivered AAV-DIO-NaChBac-GFP viruses to the VMH of SF1-Cre::ERa-Cre mice to activate these neurons, and at the same time, co-injected AAV-DIO-WGA-GFP viruses to synaptically trace connected neurons. We then delivered the AAV-Flp-DOG to the Peri-LC region to render all synaptically connected neurons in this region Flp active, and at the same time co-injected AAV-fDIO-Cre and AAV-DIO-Kir2.1-dTomato (or AAV-DIO-mCherry controls) to selectively target expression of Kir2.1 or mCherry within the Peri-LC that are downstream from SF1/ERa neurons (Supplementary Fig. S10A). We first conformed targeted expression of Kir2.1 and mCherry in the Peri-LC, and that Kir2.1 abrogated elevated c-Fos expression seen with NaChBac manipulations (Supplementary Fig. S10B). These data suggest effective inhibition of Peri-LC neurons that were activated by targeted NaChBac expression in SF1/ERa neurons. However, we found that Kir2.1 inhibition of the traced neurons in Peri-LC failed to rescue both the reduction in body weight (Supplementary Fig. S10C) and lethality (Supplementary Fig. S10D). Collectively, these results suggest that activation of downstream neurons in Peri-LC does not contribute significantly to the observed AN phenotypes by activation of SF1/ERa neurons.

Refractory anorexic effects by activation of ventral hypothalamic glutamatergic neurons

It’s well known that human anorexia is very difficult to treat (Foldi et al 2020), and thus new therapeutic approaches are needed. To determine whether anorexia induced by activation of hypothalamic glutamatergic neurons in our model is reversible, we tested whether this phenotype could be ameliorated using an established animal model of positive energy balance. In our previous work, NaChBac-mediated activation of AgRP neurons induced robust hyperphagia and severe obesity (Zhu et al 2020). We therefore tested whether activation of AgRP neurons could reverse the voluntary starvation induced by activation of VMH glutamatergic neurons. Toward this, we crossed AgRP-Cre and Vglut2-Cre to generate AgRP-Cre::Vglut2-Cre mice, and delivered AAV-DIO-NaChBac-GFP viral vectors to both the arcuate nucleus and the ventromedial area of the hypothalamus of these mice (Fig. 7A). We first validated accurate NaChBac targeting and evaluated c-Fos expression in both VMH and Arc AgRP neurons (Fig. 7B). As expected, NaChBac-mediated activation of AgRP neurons alone caused rapid weight gain, whereas activation of Vglut2 neurons alone caused reduced body weight (Fig. 7C). However, mice with NaChBac expression in both AgRP and Vglut2 neurons phenocopied the reduced body weight and lethal phenotype typical of what we observed with activation of Vglut2 neurons alone (Figs. 7C and 7D). Moreover, mice with NaChBac expression in both AgRP and Vglut2 neurons exhibited an indistinguishable voluntary starvation phenotype to that Vglut2 neurons along (Fig. 7E). These results suggest that hyperphagia produced by activation of AgRP neurons failed to reverse voluntary starvation induced by NaChBac-mediated activation of glutamatergic neurons.

Fig. 7. Refractory anorexic effects by activation of ventral hypothalamic glutamatergic neurons.

(A) Diagram of viral delivery (AAV-DIO-EYFP, AAV-DIO-NaChBac) into VH of AgRP-Cre, Vglut2-Cre, and AgRP-Cre::Vglut2-Cre. (B) Representative images showing EYFP or NaChBac expression patterns in the VMH, and robust c-Fos induction (arrows) in NaChBac-expressing mice. (C) Body weight trajectories of AgRP-Cre, Vglut2-Cre, AgRP-Cre::Vglut2-Cre mice after EYFP or NaChBac injection. n=8 (EYFP), n=6 (AgRP-Cre), n=8 (Vglut2-Cre), n=7 (AgRP-Cre::Vglut2-Cre); two-way ANOVA; ***P<0.001. (D) Kaplan–Meier survival analysis of AgRP-Cre, Vglut2-Cre, AgRP-Cre::Vglut2-Cre mice after EYFP or NaChBac injection. Mice were considered as non-survival when body weight loss reached 30%, log-rank test, ***P<0.001 versus EYFP. (E) Daily food intake. n=5 per group; one-way ANOVA, ***P<0.001 versus EYFP, ###P<0.001 versus AgRP-Cre/NaChBac. (F) Diagram of viral delivery (AAV-DIO-EYFP, AAV-DIO-NaChBac) into VH of ob/ob::Vglut2-Cre mice. (H) Body weight trajectories of ob/ob::Vglut2-Cre mice after EYFP or NaChBac injection. n=6 (EYFP), n=7 (NaChBac); two-way ANOVA; *P<0.05, **P<0.01. (I) Kaplan–Meier survival analysis of ob/ob::Vglut2-Cre mice after EYFP or NaChBac injection. Mice were considered non-survival when body weight loss reached 30%, log-rank test, ***P<0.001. (J) Daily food intake. n=6–7 per group; two-way ANOVA, ***P<0.001.

Finally, we then bred Vglut2-Cre mice with leptin deficient ob/ob mice, which are well known to develop hyperphagia and severe obesity. Towards this, we delivered AAV-DIO-NaChBac-GFP or control AAV-DIO-GFP viruses to the VMH area of Vglut2-Cre::ob/ob mice (Fig. 7F). As expected, control GFP virus-injected mice showed increased body weight and developed obesity (Fig. 7G). In contrast, the NaChBac-injected mice exhibited gradual reduction in body weight, and eventually, all of these mice died within 8 days after viral delivery (Figs. 7G and 7H), recapitulating the lethality phenotype observed in NaChBac-injected Vglut2-Cre mice. Consistently, the NaChBac-injected mice showed a dramatic reduction in food intake, compared to control mice (Fig. 7I), suggesting voluntary starvation is the driving force of the lethality. Thus, hyperphagia from leptin deficiency failed to reverse voluntary starvation induced by NaChBac-mediated activation of glutamatergic neurons.

Discussion

The understanding of the neural basis underlying AN has been greatly hindered by the void of animal models that replicate clinical symptoms observed in AN patients, calling for new animal models with better replication of human AN symptoms (Francois et al 2021, Stern & Bulik 2020). The most widely used ABA model involves imposed food restriction and involuntary starvation to induce a negative energy state, which is in opposite to voluntary food restriction, the major driving force of AN pathogenesis (Francois et al 2021, Spadini et al 2021). As such, these existing models do not capture some of the most critical aspects of the human AN phenotypes, and therefore may be considered limited in their ability to provide needed insights on the neural basis underlying AN. Other BDNF-based animal models, although based on human alleles, also rely on forced fasting and exhibit variations in AN symptom (Madra & Zeltser 2016). Other genetic models, including the anx/anx mouse, show primarily self-starvation, but not other typical symptoms of AN (Scharner & Stengel 2021). Our current animal models with VMH SF1/ERa neuron chronic activation or overexpression of BDNF exhibited voluntary fasting, hyperactivity, anxiety, anhedonia, and a heightened response to environmental stressors, all typical symptoms in AN. Thus, our study establishes an AN animal model that faithfully recaptures many of the critical AN symptoms observed in humans, and further identifies VMH SF1/ERa neurons as one brain site capable of eliciting AN pathogenesis.

The VMH is traditionally viewed as a “satiety center” given that VMH lesions cause obesity and hyperphagia (King 2006, Krause & Ingraham 2017) (Zhao et al 2008). Importantly, acute activation or inhibition of VMH SF1 neurons results in hypophagia or hyperphagia, respectively (Viskaitis et al 2017), further suggesting that VMH SF1 neurons modulate feeding behavior in a bi-directional way. Importantly, these neurons have also been identified to regulate defensive and fear responses, as well as social behaviors (Chang & Gean 2019, Lin et al 2011, Silva et al 2013, Todd et al 2018). Within the VMH, ERa-expressing neurons, partially overlapping with SF1 neurons, mediate sexual dimorphic effects in feeding and anxiety-related behaviors (Cheung et al 2015),(Xu et al 2011), and also promote locomotor activity (Fujikawa et al 2016, Krause et al 2021). These studies collectively demonstrate that VMH neurons are implicated in the regulation of feeding, locomotion, and psychiatric behaviors that include anxiety and social interactions, all of which are known to be severely dysregulated in AN. Of note, these studies are largely based on chronic loss of function (i.e. lesion or gene deletion), or acute manipulations of neuron activity, and thus are unable to reveal the impact from chronic activation of VMH neurons. Given that many brain diseases, including AN, are likely caused by a chronic change in brain state (Scharner & Stengel 2019), it was our goal to generate a new model that chronically alters neuronal activity in a genetically targeted and cell type-specific manner. Towards this, our current study utilizes a targeted NaChBac approach to achieve chronic activation of SF1/ERa neurons, which revealed striking behavioral manifestations that closely resemble AN, and thus further extend previous findings by linking them to a disease state.

Here we show that chronic activation of VMH SF1/ERa neurons led to an AN phenotype comparable to that of VH glutamatergic neuron activation, suggesting a major contribution of SF1/ERa neurons to the development of AN associated behaviors. Interestingly, although both SF1 and ERa neurons each individually contributed to voluntary food restriction, anxiety, and reduced sociability, chronic activation of SF1 neurons surprisingly led to reduced locomotion while activation of ERa neurons led to heighted locomotion. The effect of SF1 neuron activation on locomotion is consistent with previous observations showing that SF1 neurons influence muscle metabolism (Fujikawa et al 2016). Whereas ERa neuron activation is also known to drive locomotion, especially in females (Krause et al 2021), our results suggest that chronic activation of ERa neurons caused a pervasive hyperactivity effect that overrides the locomotion-reducing effect from SF1 neurons, giving rise to a heightened locomotion in mice with chronic activation of SF1/ERa neurons. Thus, overactivation of ERa neurons is required to produce hyperactivity, a core symptom in AN. Importantly, while activation of ERa neurons in females produced signs of increased anxiety and reduced sociability in females, we did not observe these effects in male mice with both NaChBac and BDNF expression in ERa neurons, demonstrating another layer of sexual dimorphism. Notably, this differential involvement of ERa-expressing neurons may provide a mechanistic link to the higher penetrance of AN occurrence in adolescent girls, who are known to have a developmental estrogen surge during adolescence.

Consistent with the known association between BDNF polymorphisms and AN, our results suggest that BDNF deletion is capable of rescuing the AN phenotypes we observed in the mouse, and conversely overexpression of BDNF in SF1/ERa neurons produces AN phenotypes that mimic those seen with chronic activation of these neurons, pointing to a role for enhanced BDNF action in SF1/ERa neurons in promoting AN associated behaviors. These observations are further supported by obesity development with disrupted BNDF signaling in the hypothalamus (An et al 2015, Kosse et al 2024, Li et al 2024, Unger et al 2007) (Spiteri et al 2010, Unger et al 2007). On this note, reports of changes in BDNF levels in AN patients have been inconsistent, some reporting increases while others describing reduced or no changes (Borsdorf et al 2021, Mercader et al 2007, Shobeiri et al 2022). Given the complex etiology of AN pathogenesis, differential reporting of changes in BDNF levels in AN patients may reflect the complex nature of disease-causing defects that occur either upstream or downstream, for example, in the case of downstream, it may lead to reduced BDNF levels owing to presumed feedback. Intriguingly, the AN-associated BDNF Vav66Met allele appears to show defective function of BDNF (Bachmann et al 2012), which is not fully consistent with our current observations that enhanced BNDF action promotes AN phenotypes, or existing literature on the impact of disrupted BDNF function on obesity development (Harvey & Rios 2024). Taken together, further studies may be warranted to more directly examine the effects of this specific allele in SF1/ERa neurons on AN development.

Regarding insight into mechanisms of VMH action, we found that targeted disruption of glutamate release completely rescued the AN phenotypes associated with chronic action of glutamatergic SF1/ERa neurons, suggesting glutamate is the major mediator from these neurons that elicits AN associated phenotypes. The observed heightened glutamate release with chronic activation is supported by increased c-Fos in all downstream target areas examined, which was effectively reversed by loss of glutamate release. Loss of BNDF caused a delayed rescue, suggesting that the BDNF effect is likely mediated by reducing glutamate release. Supporting this, c-Fos in neurons downstream of SF1/ERa neurons with chronic activation was reversed with BDNF deletion, and BDNF overexpression in these neurons increased c-Fos expression in downstream neurons. Importantly, BDNF is known to promote glutamate release from presynaptic terminals (Carvalho et al 2008). Our data also suggest that BDNF induced c-Fos expression in downstream neurons was reversed by loss of glutamate release. In addition, we did not observe any increase in c-Fos levels in the SF1/ERa neurons themselves with BDNF expression. These results collectively suggest that BDNF does not affect postsynaptic glutamate receptor function, but instead enhances presynaptic glutamate release onto downstream targets involved in AN-associated behaviors.

It is striking that, while the dramatic voluntary starvation phenotypes could be rescued by selective loss of glutamate release or BDNF deletion from SF1/ERa neurons, it was not at all impacted by two genetic manipulations well known to cause hyperphagia and severe obesity (i.e. leptin-deficiency or chronic activation of AgRP neurons), suggesting that the observed voluntary starvation is persistent, forceful, and overrides homeostatic mechanisms despite a sever hunger state. This may underscore the known difficulty in curing AN through persuading more feeding (Jennifer et al 2009). Also worth noting, a previous study suggests that VMH BDNF action on jaw movements through trigeminal me5 neurons may directly gate feeding behavior (Kosse et al 2024). However, our current data show that the restrictive feeding we observe with chronic activation of SF1/ERa neurons was not reversed by chronic inhibition of the broad Peri-LC areas that harbor me5 neurons, arguing against a potential implication of defective jaw moments. In addition to brain stem projections, VMH SF1/ERa neurons are known to send multiple output projections to other brain regions, including PAG), PVT, bed nucleus of stria terminalis (BNST), amygdala, and striatum among others (Meek et al 2016, Wang et al 2015, Zhang et al 2020) (Kunwar et al 2015, Lin et al 2011, Timper & Bruning 2017, Wang et al 2015, Zhang et al 2020). Many of these targets are known to regulate feeding, social interactions, and anxiety-related behaviors. Importantly, human imaging studies that show heightened neuronal activities in some of these brain regions in AN patients (Fladung et al 2010, Miyake et al 2010, Vocks et al 2010), suggesting direct relevance of the identified pathway to human AN patients. Since the effect of chronic activation of SF1/ERa neurons dominates over that of well-known feeding circuit, direct targeting the disease-causing neurons or their signaling pathways may be useful towards developing effective treatments for AN. In summary, this study generated and validated a new AN animal model that faithfully captures all key AN symptoms, including voluntary starvation, increased locomotion, signs of anxiety, social phobia, and sexual dimorphic manifestations via the putative involvement of ERa neurons. It remains to be seen if this new model will help bridge animal research to human genetics through further investigating VMH BDNF and SF1/ERa-expressing neurons towards better understanding the causes and treatments of AN.

Limitations of the study

We used viral approaches to achieve chronic activation and BDNF overexpression in this study and therefore, our animals are unable to fully capture distinct fo AN occurring in adolescent females. An inducible model will be required to test the temporal effect of estrogens on SF1/ERa neurons during the adolescent age. In addition, behavioral and physiological data from our genetic animals support a role for BDNF in enhancing synaptic glutamate release, no specific data were presented to support this is the case in SF1/ERa neurons, although there exists rich supporting literature (Carvalho et al 2008). Future studies are required to identify how BDNF enhances glutamatergic action.

Animals

Animal care and experimental procedures were approved by the Institutional Animal Care and Use Committee of the University of Texas Health Science Center at Houston. Mice were maintained at 21–22 °C under a 12-h light/12-h dark cycle with standard pellet chow (LabDiet 5053; 20% protein diet and 4.5%; 3.02 kcal/g) and water provided ad libitum, unless otherwise specified for fasting experiments. Animals were group-housed under standard conditions, and singly housed when measuring daily food intake or when placed in metabolic cages. Vglut2-Cre, SF1-Cre, AgRP-Cre, ob, Esr1-Cre (ERa-Cre) mice were purchased from The Jackson Laboratory. BDNF^flox/flox mice were kindly provided by Dr. Baoji Xu from U. Florida. For all experiments, mice from the same litters were randomly assigned to different treatment groups and were 6–8 weeks of age at the time of surgical procedures.

Stereotaxic Surgery

Stereotaxic surgeries for viral delivery were performed as previously described. Briefly, mice were anesthetized with a ketamine/xylazine cocktail (100 and 10 mg/kg, respectively), and their heads were secured in a stereotaxic frame. Viral vectors were delivered through a 0.5 mL syringe (Neuros Model 7000.5 KH, point style 3; Hamilton, Reno, NV, USA) mounted on a motorized stereotaxic injector (Quintessential Stereotaxic Injector; Stoelting, Wood Dale, IL, USA) at a rate of 30 nL/min. Viral preparations were titered to ~10¹² particles/ml. Injection coordinates were as follows: for VMH region, anteroposterior (AP) −1.50 mm, mediolateral (ML) ±0.35 mm and ±0.70 mm, dorsoventral (DV) −5.60 mm; 60 nl per site. For me5 region, coordinates were AP −5.35 mm, ML ±0.90 mm, DV −3.45 mm; 100 nl per site. For three days following neurosurgery, mice received carprofen (Rimadyl; Zoetis Inc., MI, USA; 5 mg/kg, i.p.) for analgesia.

Body Weight Measurement

Body weight was measured daily in mice injected with AAV-DIO-NachBac and weekly in mice injected with AAV-DIO-BDNF following surgery under a standard chow diet.

Food Intake Measurement

Daily food intake was measured beginning 3 days after surgery in NachBac-injected Vglut2-Cre mice, 1 week after surgery in NachBac-injected SF1-Cre, Esr1-Cre, SF1-Cre::Esr1-Cre mice, and 3 weeks after surgery in BDNF-injected mice. Mice were singly housed 3 days before the start of food intake measurements. Food intake was recorded for at least 3 consecutive days, and the average value was used as the daily food intake.

Fasting Refeeding Assay

Fasting refeeding experiments were performed after completion of food intake measurements. Mice were fasted from 9:00 to 18:00, after which standard chow pellets were provided. Food intake was recorded at 30-, 60-, 120-, and 240-min post refeeding.

Open Field Test (OFT)

In this study, the open-field test is used to evaluate locomotor activity. Mice were placed individually into an open-field chamber, and their movement was recorded for 15 min using overhead cameras. Locomotion parameters were quantified with EthoVision XT software (Noldus Information Technology).

Light-Dark box Test (LDT)

Anxiety-like behavior was assessed using the light–dark box test. Mice were initially placed in the dark compartment, and their movement between the light and dark chambers was recorded using overhead cameras and analyzed with EthoVision XT software (Noldus Information Technology). The time spent in the light compartment was used as an index of anxiety-like behavior, with increased light-chamber exploration reflecting reduced anxiety.

Nest-building Behavior

Nest-building behavior was assessed after mice had been housed individually for one week. Approximately 2.5 g of compressed white cotton nesting material was placed in each cage at ~09:00, and nests were evaluated in 4 h. Nest quality was scored on a 5-point scale: 1, no nest attempted; 2, poor nest, with incomplete use of nesting material and minimal structure; 3, fair nest, with full use of nesting material but lacking organized walls; 4, good nest, with full use of nesting material and low but discernible walls; 5, excellent nest, with full use of nesting material and well-defined high walls.

Three-chamber Sociability Test

The three-chamber test is used to evaluate sociability. The three-chamber apparatus (60 cm × 42.5 cm × 22 cm) was constructed of clear polycarbonate wall and grey Komatex floor, with three equally divided chambers (42.5 cm × 20 cm × 22 cm). Chambers were connected by sliding doors (5.5 cm × 18 cm) in two dividing walls, which control access to left and right chambers. Two stainless-steel barred cups were put in the left and right chambers, besides the walls. During the initial habituation phase, the mice were placed in the middle chamber, and freely explore each of the three chambers with empty stainless-steel barred cups. Immediately after this, an unfamiliar same sex same age mouse which had no previous interaction with the subject mouse, was enclosed in a wire cup. The opposite chamber contained a similar empty wire cup, allowing the subject mouse to explore for 10 min freely. The interaction duration with the stranger mice and empty wire cage was recorded.

Sucrose Preference Test (SPT)

In the present study, the sucrose preference test was used to assess anhedonic behavior. Mice were singly housed for one week with ad libitum access to food and water. Over the subsequent 3-day adaptation period, each mouse was given two bottles: one containing tap water and the other containing 1% sucrose solution. On the day before testing, mice were fasted from water but not food. At 8:00 a.m. on the test day, both bottles were provided, and their positions were switched at 1:00 p.m. Bottle weights were recorded at 8:00 a.m. and 6:00 p.m. Sucrose preference was calculated as: $\mathrm{S}\mathrm{u}\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{s}\mathrm{e}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{f}\mathrm{e}\mathrm{r}\mathrm{e}\mathrm{n}\mathrm{c}\mathrm{e}\left(\mathrm{\%}\right)=[\mathrm{s}\mathrm{u}\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{s}\mathrm{e}\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{k}\mathrm{e}/(\mathrm{s}\mathrm{u}\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{s}\mathrm{e}\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{k}\mathrm{e}+\mathrm{w}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{k}\mathrm{e}\left)\right]\times 100\mathrm{\%}$.

Metabolic Cages Test

Mice were individually housed in chambers of the PhenoMaster cages (TSE systems, Chesterfield, Missouri,USA). Mice were given ad libitum access to a normal chow diet and water. Locomotion activity levels were measured using indirect calorimetry continuously at different time points. Data was averaged through different time points in light and dark cycles, respectively, for comparison. The data from the first day and the last day was removed.

Perfusion and Tissue Dissection

Mice were anaesthetized with ketamine/xylazine (100 and 10 mg kg⁻¹, respectively). After loss of the pedal reflex, animals were transcardially perfused with 15 ml of 0.9% saline followed by 15 ml of 10% formalin. Brains were collected, post-fixed overnight in 10% formalin at room temperature and cryoprotected in 30% sucrose in PBS. Coronal sections (30 μm) were cut on a freezing microtome and stored in PBS containing 0.1% sodium azide at 4 °C.

Immunostaining and Imaging

Brain sections were washed three times for 5 min each in PBS containing 0.1% Triton X-100, followed by blocking in PBS with 0.1% Triton X-100 and 5% normal donkey serum for 1 h at room temperature. Sections were then incubated overnight at 4 °C in primary antibody solution (primary antibody, 5% normal donkey serum, 0.1% Triton X-100 in PBS). The following primary antibodies were used: rabbit anti-cFos (1:1000, #2250, Cell Signaling Technology), rabbit anti-BDNF(1:200, ANT-010, Alomone Labs), goat anti-GFP (1:1000, 51370, ROCKLAND), rabbit anti-ER alpha (1:500, NBP1–84827), chicken anti-mCherry (1:500, MCHERRY-0020, aves labs).The next day, sections were rinsed three times in PBS with 0.1% Triton X-100 and incubated with secondary antibodies prepared in the same diluent (secondary antibody, 5% normal donkey serum, 0.1% Triton X-100 in PBS). Floating sections were mounted onto glass slides and coverslipped with Fluoromount (Diagnostic BioSystems Inc., Sigma-Aldrich). Images were acquired using a Leica TCS SP5 confocal microscope (Leica Microsystems, Wetzlar, Germany). The fluorescence-positive neurons were recognized and quantified in the software QuPath 0.4.2 (https://qupath.github.io).

Quantification and Statistical Analysis

GraphPad Prism 9.5.1 (GraphPad Software, Inc., La Jolla, CA, USA) was used for all statistical analyses and construction parts of the figures. Comparisons were made by unpaired two-tailed Student’s t-test, or one-way or two-way ANOVA followed by Tukey’s multiple-comparison post hoc test. Error bars in graphs represent ± s.e.m. All statistical analyses and sample sizes are indicated in the corresponding figure legends.