The Inhibitory Circuit Architecture of the Lateral Hypothalamus Orchestrates Feeding

Abstract

The growing prevalence of overeating disorders is a key contributor to the worldwide obesity epidemic. Dysfunction of particular neural circuits may trigger deviations from adaptive feeding behaviors. The lateral hypothalamus (LH) is a crucial neural substrate for motivated behavior including feeding, but the precise functional neurocircuitry that controls LH neuronal activity to engage feeding has not been defined. We observed that inhibitory synaptic inputs from the extended amygdala preferentially innervate and suppress the activity of LH glutamatergic neurons to control food intake. These findings help explain how dysregulated activity at a number of unique nodes can result in a cascading failure within a defined brain network to produce maladaptive feeding.

Over half a century ago, experiments in rodents and other species have revealed that gross neuroanatomical manipulations of the LH alter diverse behaviors, including feeding (1–3). While direct anatomical and neuropharmacological manipulations (4, 5) within the LH produce profound alterations in a variety of motivated behaviors, they provide limited mechanistic insight into the discrete circuit connections within the LH that regulate precise behaviors such as feeding. Given the circuit complexity within the LH (6), we aimed to dissect the neurocircuitry between the LH and a principal afferent from the extended amygdala.

The bed nucleus of the stria terminalis (BNST), a neural component of the extended amygdala (7), is a key integrator of diverse motivational states through its interactions with various synaptic targets, including the ventral tegmental area (VTA) (8) and the LH (9). The BNST is comprised primarily of GABAergic cells (10) and consumption of food activates BNST neurons (11). Therefore, we considered the BNST and its inhibitory projections to the LH as an important candidate for regulating feeding. We targeted a Cre-inducible viral construct coding for Channelrhodopsin-2 fused to enhanced yellow fluorescent protein (ChR2-eYFP) into the BNST of Vgat-ires-Cre mice and positioned optical fibers above the LH for in vivo photostimulation of Vgat^BNST➝LH projection fibers (Fig. 1, A to E and fig. S1). Optogenetic activation of this inhibitory pathway rapidly produced voracious feeding behavior in well-fed mice (Fig. 1, F to I, and fig. S2 and movie S1). We explored the motivational valence of this pathway by testing mice in real-time place preference and self-stimulation assays (Supplementary Methods). Vgat^BNST➝LH∷ChR2 mice exhibited a significant place preference for a photostimulation-paired chamber (Fig. 1J and fig. S2) and actively nose poked for photoactivation of the circuit. Food deprivation significantly augmented, while satiety significantly attenuated Vgat^BNST➝LH∷ChR2 self-stimulation (Fig. 1K and fig. S2). The evoked feeding response was specific to the Vgat^BNST➝LH pathway. Photoactivation of Vgat^BNST➝VTA projections did not elicit feeding behavior (figs. S3 and S4).

Fig. 1. Vgat^BNST➝LH circuit activation induces feeding in well-fed mice.

(A) Vgat^BNST➝LH circuit targeting. (B) 10x (top) and 20x (bottom) images of the Vgat^BNST➝LH::ChR2-eYFP circuit (scale bars = 1 mm (top), 500 μm (bottom)). (C) Localized ChR2-eYFP expression in the BNST (top) and quantified eYFP fluorescence intensity (bottom) is significantly greater in the BNST compared to surrounding regions (F_5,29 = 11.22, P < 0.001, n = 5 sections from n = 5 mice; ac, anterior commissure; ldBNST, lateral-dorsal BNST; vBNST, ventral BNST; LS, lateral septum; LPO, lateral preoptic area; VP, ventral pallidum; DS, dorsal striatum; scale bar = 200 μm). (D and E) ChR2-eYFP expression in the BNST (D) and axonal projections in the LH (E) in Vgat-ires-Cre mice (LH, lateral hypothalamus; Fx, fornix; EP, entopeduncular nucleus; DMH, dorsomedial hypothalamus; 3V, third ventricle; D, dorsal; V, ventral; L, lateral; M, medial; green = ChR2-eYFP; red = Nissl stain; scale bars = 200 μm (top), 20 μm (bottom)). (F) Spatial location heat maps in 20 min epochs before, during, and after 20-Hz photostimulation-induced feeding. (G and H) Photostimulation of Vgat^BNST➝LH projections significantly increased grain-based (standard) food intake (F_2,24 = 201.6, P < 0.001) (G) and food zone time (F_2,24 = 18.61, P < 0.001, n = 5 mice per group) (H). (I) Higher photostimulation frequencies significantly decreased evoked feeding latencies (F_3,56 = 48.89, P < 0.001, n = 9 mice). (J) Vgat^BNST➝LH::ChR2 mice spent significantly more time in the photostimulation-paired side compared to controls (P < 0.001, n = 5 mice per group). (K) Food-deprived Vgat^BNST➝LH::ChR2 mice nose poked significantly more for 10- and 20-Hz photostimulation when compared to 2 days of standard fed ad libitum and to 2 days of standard fed ad libitum supplemented with 2 hr of high-fat food exposure before the self-stimulation session (F2,204 = 40.87, P < 0.001, n = 9 mice). All values for all figures represent mean ± s.e.m. * P < 0.05, ** P < 0.001 (Student's t-test or ANOVA followed by Bonferroni post-hoc comparisons, where applicable). Dagger symbol denotes significance compared to all manipulations.

Because high-caloric diets can facilitate overeating (12), we determined whether consumption induced by Vgat^BNST➝LH circuit activation was directed towards palatable energy-dense foods (Supplementary Methods). Well-fed Vgat^BNST➝LH∷ChR2 mice showed a strong preference for high-fat food during photostimulation exposure (table S1 and movie S1), suggesting that activation of the Vgat^BNST➝LH circuit is sufficient for eliciting feeding that is preferential for calorie-dense substances even when energy requirements are satisfied.

To examine whether endogenous activity of the Vgat^BNST➝LH pathway is important for feeding, we transduced BNST-GABAergic neurons (Fig. 2A) and their axons that innervate the LH (Fig. 2B and fig. S5) with the inhibitory opsin, archaerhodopsin (eArch3.0-eYFP) (13, 14). Suppression of presynaptic BNST-GABAergic signaling via eArch3.0 activation resulted in enhanced LH-postsynaptic neuronal activity during anesthetized extracellular recordings (Fig. 2, C to E). Despite the influence of hunger, photoinhibition of the Vgat^BNST➝LH circuit reduced feeding in food-deprived mice (Fig. 2, F to J, and fig. S6). Furthermore, photoinhibition of Vgat^BNST➝LH projections led to a significant avoidance of a photoinhibition-paired chamber (Fig. 2K and fig. S6).

Fig. 2. Vgat^BNST➝LH circuit inhibition diminishes feeding in food-deprived mice and is aversive.

(A and B) eArch3.0-eYFP expression in the BNST (A) and axonal projections in the LH (B) in Vgat-ires-Cre mice (scale bars = 200 μm (top), 20 μm (bottom)). (C) Schematic for anesthetized in vivo extracellular recordings in the LH. (D) Example trace from a single LH unit (top) and its representative peri-event histogram and raster (bottom). (E) The average firing rate of light-responsive LH units significantly increased during the 5-s photoinhibition trials (F_2,12 = 19.52, P < 0.001, n = 5 units from n = 3 mice). (F) Spatial location heat maps in 10 min epochs before, during, and after photoinhibition. (G and H) Photoinhibition of Vgat^BNST➝LH projections significantly decreased standard food intake (F_1,44 = 16.30, P < 0.001) and time spent in the food zone (I and J) (F_1,44 = 2.43, P = 0.028, n = 6 mice per group). (K) Vgat^BNST➝LH::eArch3.0 mice spent significantly less time in the photoinhibition-paired side when compared to controls (P = 0.004, n = 6 mice per group).

The hypothalamus contains numerous genetically distinct neuronal populations (15–21). We thus characterized the molecular phenotype of the postsynaptic LH neuronal targets that receive functional Vgat^BNST➝LH innervation. We paired photostimulation of Vgat^BNST➝LH inputs with whole-cell recordings in conjunction with multiplexed gene expression profiling of individual LH neurons in brain slices (Fig. 3A, Supplementary Methods). We focused on a set of genes known to be heterogeneously expressed in the LH and whose products have been implicated in feeding (22). BNST-GABAergic inputs formed strong functional connections with postsynaptic LH neurons that expressed significantly higher levels of Vglut2. In contrast, weakly innervated LH neurons displayed significantly lower levels of Vglut2 and higher levels of Vgat expression (Fig. 3, B and C, and fig. S7, and table S2).

Fig. 3. Vgat^BNST➝LH projections preferentially target LH glutamatergic neurons.

(A) Schematic for ChR2-assisted circuit mapping with single-cell gene expression profiling. (B) Color-coded fold expression of all target genes from all recorded LH neurons (Vglut2, vesicular glutamate transporter-2; Vgat, vesicular GABA transporter; DYN, dynorphin; MCH, melanin-concentrating hormone; NTS, neurotensin; OX, orexin/hypocretin; TH, tyrosine hydroxylase). The average fold expression for Vglut2 was significantly higher in postsynaptic LH neurons that display large optically-evoked inhibitory postsynaptic current amplitudes (strongly innervated) compared to weakly innervated LH neurons (U = 169.0, P = 0.016, n = 6 mice, n = 48 cells). (D) Schematic for modified rabies virus tracing. (E and F) Images from a Vglut2-ires-cre mouse showing FLEX-TVA-mCherry expression in LH glutamatergic neurons (E) and appreciable SADΔG-GFP labeling of BNST neurons (F). (G and H) FLEX-TVA-mCherry expression in LH GABAergic neurons (G) and minimal SADΔG-GFP labeling of BNST neurons (H) (green = SADΔG-GFP; red = FLEX-TVA-mCherry; blue = Nissl stain; scale bars = 200 μm). (I) Significantly more BNST neurons innervate LH glutamatergic neurons compared to LH GABAergic neurons (F_1,20 = 38.50, P < 0.001, n = 3 mice per group).

We confirmed these findings by utilizing modified rabies virus tracing techniques to identify the monosynaptic inputs to glutamatergic and GABAergic neurons in the LH (Fig. 3D). In Vglut2-ires-Cre and Vgat-ires-Cre mice, we targeted Cre-inducible TVA (AAV5-FLEX-TVA-mCherry) and RG (AAV8-FLEX-RG) proteins that allow for rabies virus infection and subsequent transsynaptic viral propagation, respectively (23, 24), to LH-glutamatergic or -GABAergic neurons. Two weeks after AAV transduction, the modified rabies virus, SADΔG-GFP(EnvA), was injected into the LH, and BNST-containing slices were obtained 7 days later for confocal imaging. Vglut2^LH∷Rabies tracing revealed dense populations of transsynaptically labeled BNST neurons (Fig. 3, E and F), while Vgat^LH∷Rabies tracing resulted in minimal BNST labeling (Fig. 3, G to I, and fig. S8).

Because BNST-GABAergic projection neurons promote feeding and selectively target LH glutamatergic neurons, we considered Vglut2^LH neurons as a critical downstream circuit node for regulating food intake. Photoactivation of Vglut2^LH neurons (Fig. 4A and fig. S9) suppressed feeding in food-deprived mice (Fig. 4, B to F, and fig. S10), while photoinhibition of Vglut2^LH neurons induced feeding in well-fed mice (figs. S11 to S13). Photostimulation of Vglut2^LH neurons produced aversion (Fig. 4G and fig. S10) and Vglut2^LH inhibition produced a preference for palatable foods (table S1).

Fig. 4. Photoactivation of Vglut2^LH neurons suppresses feeding in food-deprived mice and is aversive.

(A) ChR2-eYFP expression in the LH of a Vglut2-ires-Cre mouse (scale bars = 200 μm (top), 20 μm (bottom)). (B) Spatial location heat maps in 10 min epochs before, during, and after 5-Hz photostimulation. (C and D) Photostimulation of Vglut2^LH neurons significantly decreased food intake (F_1,36 = 13.31, P < 0.001) and food zone time (E and F) (F_1,36 = 13.12, P < 0.001, n = 5 mice per group). (G) Vglut2^LH::ChR2 mice spent significantly less time in the photostimulation-paired side when compared to controls (P < 0.001, n = 5 mice per group).

Until now, the precise neurocircuit elements responsible for the feeding and reinforcement phenomena observed five decades ago by electrical stimulation of the LH (1, 2, 25) have remained a mystery. Inhibitory inputs from the BNST specifically innervate and suppress LH glutamatergic neurons to promote feeding. Further unraveling of the specific patterns of gene expression and projection targets of LH glutamatergic neurons could identify novel points for therapeutic intervention within these circuits for the treatment of eating disorders and obesity.