GLP-1 Increases pre-ingestive satiation via hypothalamic circuits in mice and humans

GLP-1 receptor agonists (GLP-1RAs) are effective anti-obesity drugs. However, the precise central mechanisms of GLP-1RAs remain elusive. We administered GLP-1RAs to obese patients and observed heightened sense of pre-ingestive satiation. Analysis of human and mouse brain samples pinpointed GLP-1R neurons in the dorsomedial hypothalamus (DMH) as candidates for encoding pre-ingestive satiation. Optogenetic manipulation of DMHGLP-1R neurons caused satiation. Calcium imaging demonstrated that these neurons are actively involved in encoding pre-ingestive satiation. GLP-1RA administration increased the activity of DMHGLP-1R neurons selectively during eating behavior. We further identified an intricate interplay between DMHGLP-1R neurons and arcuate NPY/AgRP neurons (ARCNPY/AgRP), to regulate food intake. Our findings reveal a hypothalamic mechanism through which GLP-1RAs control pre-ingestive satiation, offering novel neural targets for obesity and metabolic diseases.

Methods
Analysis

Analysis was done using custom MATLAB code otherwise stated. 

Clinical trials

72 patients were assessed for eligibility, and 44 patients were screened, with 4 patients meeting the exclusion criteria. 40 obese individuals were enrolled, who were allocated randomly into two groups (Group A n=20, Group B n=20). The participants were prescribed liraglutide and received a weekly escalating dose of 0.6mg, 1.2mg, 1.8mg, 2.4mg daily over a four-weeks period. Patients were asked to write a diary to record the dose and date daily. Group A participants underwent control clinical tests prior to liraglutide injection, got 4 weeks of injection, and performed the same structured test scheme. Group B participants got 4 weeks of injection and underwent clinical tests, and after 2 weeks of washout period they performed the control clinical tests. A total of 28 patients were analyzed for study. (Withdrawn from trial: Group A n=0, Group B n=2. Excluded from analysis: Group A n=5, Group B n=5) Individual characteristics have been reported in Table S3. Clinical tests were performed as previously described as a structured test scheme, broken down into four distinct phases with a survey at the end of each phase. The survey was reconstructed based on the following questionnaires: VAS (34), RISE-Q (35), RED-13 (36), PFS-K (37), DEBQ (38). Missing values were omitted for analysis. This study was approved by the institutional review board (IRB) of Seoul National University College of Medicine/Seoul National University Hospital (IRB No. 2208-049-1349). Written informed consent was obtained from participants. Inclusion, exclusion, withdrawal, exclusion from analysis criteria have been reported in Table S2.

Optogenetics

Laser stimulation (473 nm for activation and 532 nm for inhibition, Shanghai DPSS Laser) was delivered through an FC-FC fiber patch cord (Doric Lenses) connected to the rotary joint, followed by the FC-ZF 1.25 fiber patch cord delivered stimulation to the cannula (200 µm core, NA 0.37, Doric Lenses or Inper). The laser intensity was approximately 10 mW at the tip. For open-loop stimulation, mice received 10-minute or 2-minute intervals of lasers at wavelengths of 473nm at 10hz, 50ms for neural activation or 532nm continuously for neural inhibition experiments. For closed-loop stimulation, mice received manual laser stimulation when they started to eat a high-fat diet (D12492, Research diets).

Fiber photometry 

Fiber photometry signal data were acquired using the Doric Studio software. 465 nm calcium and 405 nm isosbestic signals (for artifact correction), were obtained. 405 nm signals were linearly fitted to 465 signals. △F/F0 signals were corrected as follows to minimize artifact recordings △F/F0 = (465 nm signal − fitted 405 nm signal)/fitted 405 nm signal. Signals were decimated to obtain approximately 20, 25 or 30 data points in 1 s. The mean of the baseline (m) and standard deviation (σ) of the baseline were computed to normalize the corrected signals into Z-scores (Z = (corrected 465 nm − m)/σ). The behavior time points for each test were manually annotated. For the heatmap visualization, Z-Score was used or each trial was normalized as follows: normalized Z = (Z − minimum Z)/(maximum Z − minimum Z). In the GLP-1RAs injection test baseline was designated before injection of drugs to account for the stable state of mice (Fig. 4). To account for signal changes near initiation of the first behavior, when the dosage effect was strongest,  mean △F/F0 signals were quantified regarding the initial 5% of time between the initiation of behavior (Food Accessibility, Seeking Start, Consumption Start) and the next behavior (Seeking Start, Consumption Start, Consumption End, respectively) and was compared with baseline. Cumulative probability was quantified by first extracting a section of the Z-score from the behavior moment of interest to the next behavior moment. The section was divided into bins of 1 second and averaged. The averaged values were used for further quantification. Rate of change was quantified by computing the gradient at each behavior moment. Z-scored signals were smoothed using moving average function in MATLAB (movmean) by a sliding window of 1 second (Fig. 4K and R). Afterwards, the gradient of two moments near a behavior moment (before and after 1 second of a behavior moment) was computed. 

Micro-endoscope

The raw signal output was preprocessed and computed into calcium dynamics (craw) using CNMF-E by using Inscopix Data Acquisition Software (IDAS ver.1.8). Afterwards, craw was computed into Z-scores (Z = (Craw − m)/σ), according to the mean (m) and standard deviation (σ) of the baseline for each trial (Start of mouse going into shelter until food accessibility). Tuning of each cell was computed using choice probability (CP), defined as how well a single cell’s neural activity could predictively discriminate between two behavioral phases as described from previous reports (44). All frames from the pre-consumption and consumption behavior bouts were used to compute the histogram for each cell. These distributions are computed into a cumulative distribution which are integrated to generate a ROC (receiver operating characteristics) curve. The area under the curve is then computed for each unit regarding the two behavior conditions. The significance of a cell’s CP was determined using shuffled bout timings. Shuffling was repeated 100 times in which the mean and standard deviation was acquired. CP that had a value of over 2 standard deviations above the mean was considered significant and used for analysis (Fig. 3 and fig. S7). To account for visualization, quantification and the responsive cells in different contexts, Z-score was acquired for each trial from each cell (baseline -10 ~ -5 from behavior of interest) (fig. S7 and S9). Representative traces were smoothed using movmean function in MATLAB with a sliding window of 1 second (Fig. 3S). For responsive cell analysis, cells were determined responsive if Z-score surpassed a value of 4 or -4 after behavior of interest (fig. S10). Heatmap visualization was done as stated above (Fig. 3O, T, fig. S10). For visualization of the whole population recordings from all mice, the whole trace of the individual cell itself was used as the baseline (fig. S6).

Statistical Analysis

Statistical data were analyzed using MATLAB, Graphpad Prism 8.0 software and figures were visualized using MATLAB or CorelDrawC8 (64bit). Paired t-tests or unpaired t-tests were used to compare data between two groups. One-way or two-way repeated-measures analyses of variance (ANOVA) were used for multiple comparisons. P-values for comparisons across multiple groups were corrected using the Greenhouse–Geisser, the Tukey, and the Sidak method. Cumulative probability distribution was analyzed with two sample Kolmogorov-Smirnov tests. Results are reported as mean ± SEM, including shades, unless indicated otherwise. Levels of significance were as follows: *p < 0.05. **p < 0.01, ***p < 0.001, ****p < 0.0001. Statistic methods are listed in Table S1.

Electrophysiology studies

Slice preparation

Brain slices were prepared from mice as previously described (13, 14). Briefly, male mice were deeply anesthetized with i.p. injection of 7% chloral hydrate and transcardially perfused with a modified ice-cold artificial CSF (ACSF) (described below). The mice were then decapitated, and the entire brain was removed and immediately submerged in ice-cold, carbogen-saturated (95% O2 and 5% CO2) ACSF (126 mM NaCl, 2.8 mM KCl, 1.2 mM MgCl2, 2.5 mM CaCl2, 1.25 mM NaH2PO4, 26 mM NaHCO3, and 5 mM glucose). Coronal sections (250 μm) were cut with a Leica VT1000S Vibratome and then incubated in oxygenated ACSF (32 °C–34 °C) for at least 1 h before recording. The slices were bathed in oxygenated ACSF (32 °C–34 °C) at a flow rate of ∼2 ml/min. All electrophysiology recordings were performed at room temperature.

Whole-cell recordings

The pipette solution for whole-cell recording was modified to include an intracellular dye (Alexa Fluor 350 hydrazide dye) for whole-cell recording: 120 mM K-gluconate, 10 mM KCl, 10 mM HEPES, 5 mM EGTA, 1 mM CaCl2, 1 mM MgCl2, and 2 mM MgATP, 0.03 mM Alexa Fluor 350 hydrazide dye (pH 7.3). K-gluconate was replaced with equimolar Cs-gluconate for recording of spontaneous inhibitory postsynaptic currents (IPSCs) in response to photostimulation (2 pulses 50ms interval) upon the ARC. Epifluorescence was used to target fluorescent cells, at which time the light source was switched to infrared differential interference contrast imaging to obtain the whole-cell recording (Zeiss Axioskop FS2 Plus equipped with a fixed stage and a QuantEM:512SC electron-multiplying charge-coupled device camera). Electrophysiological signals were recorded using an Axopatch 700B amplifier (Molecular Devices); low-pass filtered at 2–5 kHz and analyzed offline on a PC with patch-clamp (pCLAMP) electrophysiology data acquisition and analysis program (Molecular Devices). Membrane potentials and action potential were determined from GLP-1R, LepR and NPY expressing neurons in brain slices. Membrane potential values were not compensated to account for junction potential (-8 mV). Recording electrodes showed resistances of 2.5–5 MΩ when filled with the K-gluconate internal solution. Input resistance (IR) was assessed by measuring voltage deflection at the end of the response to a hyperpolarizing rectangular current pulse step (500 ms of −10 to −50 pA).