Interaction of serotonin/GLP-1 circuitry in a dual preclinical model for psychiatric disorders and metabolic dysfunction

Abstract

Isolation of rodents throughout adolescence is known to induce many behavioral abnormalities which resemble neuropsychiatric disorders. Separately, this paradigm has also been shown to induce long-term metabolic changes consistent with a pre-diabetic state. Here, we investigate changes in central serotonin (5-HT) and glucagon-like peptide 1 (GLP-1) neurobiology that dually accompany behavioral and metabolic outcomes following social isolation stress throughout adolescence. We find that adolescent-isolation mice exhibit elevated blood glucose levels, impaired peripheral insulin signaling, altered pancreatic function, and fattier body composition without changes in bodyweight. These mice further exhibited disruptions in sleep and enhanced nociception. Using bulk and spatial transcriptomic techniques, we observe broad changes in neural 5-HT, GLP-1, and appetitive circuits. We find 5-HT neurons of adolescent-isolation mice to be more excitable, transcribe fewer copies of Glp1r (mRNA; GLP-1 receptor), and demonstrate resistance to the inhibitory effects of the GLP-1R agonist semaglutide on action potential thresholds. Surprisingly, we find that administration of semaglutide, commonly prescribed to treat metabolic syndrome, induced deficits in social interaction in group-housed mice and rescued social deficits in isolated mice. Overall, we find that central 5-HT circuitry may simultaneously influence mental well-being and metabolic health in this model, via interactions with GLP-1 and proopiomelanocortin circuitry.

Graphical abstract

Mice were either singly housed (top left) or group housed (top right) throughout adolescence. In adolescent-isolation mice, this induced changes in neural gene expression (center) associated with deficits in social interaction (bottom left). Acute injection of the GLP-1 receptor agonist semaglutide was sufficient to rescue social interaction behavior (bottom).

1. Introduction

Isolation of rodents throughout and beyond the age of adolescence is known to induce many behavioral abnormalities which resemble neuropsychiatric disorders (Matsumoto et al., 2019). These include depressive-like behaviors (Amiri et al., 2015; Du Preez et al., 2020; Koike et al., 2009), anxiety behaviors (Amiri et al., 2015; Koike et al., 2009; Okada et al., 2015), and unsurprisingly, social deficits (Kercmar et al., 2011; Lukkes et al., 2009; Okada et al., 2015). Therefore, many investigators find adolescent isolation to be a suitable model to characterize the co-presentation of, as well as to explore potential therapeutic interventions for these disorders (Grigoryan et al., 2022; Matsumoto et al., 2019; Powell and Swerdlow, 2023; Weiss and Feldon, 2001). Somewhat separately, isolation of rodents throughout adolescence has also been shown to induce long-term metabolic changes. These changes include impaired glucose homeostasis (Bove et al., 2022; Vargas et al., 2016), dysregulated lipid metabolism (Vargas et al., 2016), changes in appetitive behavior (Jahng, 2011; Ryu et al., 2008) and orexigenic balance (Yamada et al., 2015), and increased body fat without an overall change in body weight (Schiavone et al., 2017; Sun et al., 2014). Taken together, these changes suggest the induction of a metabolic syndrome consistent with a pre-diabetic state (Fonseca, 2009; O’Brien et al., 2014). Interestingly, central serotonin (5-HT) circuitry may simultaneously influence mental well-being and metabolic health in this adolescent isolation model.

Central 5-HT is implicated in anxiety (Akimova et al., 2009; Karayol et al., 2021; Lukkes et al., 2009), depression (Arango et al., 2001; Haleem, 2013; Moncrieff et al., 2022; Nemeroff and Owens, 2009), schizophrenia (Bleich et al., 1988; Eggers, 2013; Haleem, 2015; Sumiyoshi et al., 2014), and social interaction behaviors (Ago et al., 2013; Dekeyne et al., 2000; Kane et al., 2012; Kiser et al., 2012). 5-HT is also strongly implicated in metabolic dysfunction (Sakimura et al., 2018), particularly appetitive (D’Agostino et al., 2018; Heisler et al., 2006; Jean et al., 2007; Medeiros et al., 2005) and glycemic regulation (Cai et al., 2022; Jhanwar-Uniyal et al., 1994; Otlivanchik et al., 2015). Activation of the 5-HT_2C receptor (5-HT_2CR) in the nucleus tractus solitarius (NTS) reduces appetite (D’Agostino et al., 2018; Wagner et al., 2023), while genetic ablation of this receptor induces hyperglycemia (Berglund et al., 2013). Preproglucagon (PPG) neurons of the NTS (PPG^NTS) produce an insulinotropic hormone known as glucagon-like peptide-1 (GLP-1) (Holt et al., 2019), whose analogs are commonly prescribed to combat early diabetes. Both central 5-HT and its 5-HT_2AR are necessary for the therapeutic action of GLP-1 (Anderberg et al., 2017). Reciprocally, activation of the GLP-1 receptor (GLP-1R) modulates the activity of 5-HT neurons, and, concomitantly, appetite (Anderberg et al., 2017). This downstream appetitive regulation is thought to occur via interactions with proopiomelanocortin (POMC) neurons of the arcuate nucleus (POMC^ARC), where 5-HT and GLP-1 circuitry again intersect (Aklan et al., 2020; D’Agostino et al., 2018; Heisler et al., 2006; Sohn et al., 2011). Increased activation of POMC^ARC is further implicated in depressive-like behavioral deficits (Fang et al., 2022).

Here, we investigate changes in central 5-HT and GLP-1 neurobiology that may dually accompany behavioral and metabolic outcomes following social isolation stress during adolescence. We use an isolation paradigm that extends beyond adolescence, to encompass various approaches used across the existing literature (Medendorp et al., 2018). We further investigate any differential effect of the GLP-1R agonist semaglutide on 5-HT neurophysiology and social interaction behavior.

2. Methods

2.1. Ethical approval

All animal procedures were reviewed and approved by the University of Iowa (UI) Office of Animal Resources (Protocol 1032080) that abided by the AVMA and NIH.

2.2. Animals

Male C57BL/6J mice (catalog #000664, Jackson Labs, Bar Harbor, ME, USA; RRID:IMSR_JAX:000664) were purchased for arrival at P20, shipped five mice per cage. At age P22, half of each cohort (i.e., 10 mice) was moved to individual housing. A total of 140 mice were used. Female mice were not included, because they have been reported to be resistant to the early-adult onset of metabolic syndrome (Casimiro et al., 2021; Daniels Gatward et al., 2021; Elzinga et al., 2021; Reynolds et al., 2019).

All mice were housed in a temperature- and humidity-controlled AALAC-approved vivarium at UI on a standard 12 h/12 h dark/light (reverse) cycle in accordance with institutional requirements. Mice were housed in conventional-style cob-bedding rodent cages with nestlets, containing separate food and water that could be obtained ad libitum. Mice were maintained on chow that was composed of 14% kcal fat, 60% kcal carbohydrate, and 26% kcal protein (Catalog #5P76; Land O’Lakes, Arden Hills, MN, USA).

Mice were experimentally assessed in the following manner following completion of the 8-week isolation or group-housed period. Cohort 1: von Frey, 7 rest days, Hargreaves, 7 rest days, body composition analysis, 3 rest days, open field assay, 3 rest days, brains used for RT-qPCR. Cohort 2: insulin tolerance test, 5 rest days, glucose tolerance test, 7 rest days, pAKT/AKT signaling assay. Cohort 3: plasma used for ELISA (post-mortem collection), pancreata used for IHC/RNAscope, brains used for RNAscope. Cohort 4: Promethion metabolic phenotyping chambers. Cohort 5: electrophysiology assessment of semaglutide modulation. Cohorts 6 & 7: social interaction testing, vehicle or semaglutide injection.

2.3. Metabolic assessments

2.3.1. Body composition analysis:

Body composition was assessed in vivo using an EchoMRI machine (EchoMRI LLC; RRID:SCR_017104) and manufacturer software. After calibration of the machine, mice were assessed for fat mass, lean mass, total water, and free water.

2.3.2. Promethion assessment:

Mice were metabolically profiled using indirect calorimetry by the UI Metabolic Phenotyping Core. Mice were housed in Promethion cages (Sable Systems International, North Las Vegas, NV, USA) to determine oxygen consumption (VO2), thermogenesis, respiratory exchange ratio (RER), locomotor activity, and ingestive behavior. All data were recorded in 5 min intervals, with each interval measurement representing the average value during a 30 s sampling period per cage. No data were acquired during the first 3 d in the chambers to permit acclimation to the environment, and then experimental data were computed for each light/dark phase over 24 h (Kolling et al., 2022). Time 0 reported for the analyses represents the start of the dark or light cycle, respectively.

2.3.3. Glucose and insulin tolerance tests:

Glucose tolerance tests were performed after an 18 h fast starting from the beginning of the dark phase (Starks et al., 2015). After initial determination of fasting blood glucose, 2 g/kg glucose was administered intraperitoneally in PBS. Glucose levels were measured from tail blood samples taken at 0 min (before injection), 15, 30, 60, and 120 min using a OneTouch Ultra 2 blood glucose meter (LifeScan, Malvern, PA, USA). Insulin tolerance tests were performed after a 5 h fast starting from the beginning of the light phase. 1 U/kg insulin (Humulin R; Eli Lilly, Indianapolis, IN, USA) was delivered intraperitoneally. Blood glucose levels were measured from tail blood samples taken at 0 min (before injection), 15, 30, 60 and 120 min using a OneTouch Ultra 2 blood glucose meter.

2.3.4. P-AKT/AKT insulin signaling assay:

Mice were fasted overnight, anesthetized with intraperitoneal injection of a ketamine-xylazine cocktail, then intravenously injected with either vehicle (PBS) or insulin (2.5 U/kg body weight). 15 min after injection, insulin-sensitive tissues (liver, skeletal muscle (gastrocnemius), brown adipose tissue, and white adipose tissue) were harvested. Proteins were extracted by homogenizing tissues in tissue lysate buffer (supplementary methods). Protein samples (20 μg) were subjected to SDS PAGE, electro-transferred on a polyvinylidene fluoride membrane, then probed with primary antibodies (1:1,000) targeting AKT (cat.#: 9272; Cell Signaling Technology, Danvers, MA, USA, RRID: AB 329827 ) and p-AKT(Ser473) (cat.#: 4060; Cell Signaling Technology, Danvers, MA, USA, RRID:AB_2315049). Membranes were subsequently probed with secondary antibody (1:10,000). As a loading control, β-actin was targeted with primary antibody (1:25,000, cat#: 60008-1-Ig; Proteintech, Rosemont, IL, USA) followed with secondary antibody (1:10,000). Protein expression was visualized with an Amersham ECL detection kit (Cytiva, Marlborough, MA, USA).

2.3.5. Plasma insulin ELISA:

Plasma preparations were performed using a heparin-free method. Plate-based sandwich ELISAs were performed for insulin (1:4 dilution, RayBiotech, cat# ELM-Insulin-1) in duplicate, and plates were analyzed using a BioTek Cytation 5 Imaging Reader (RRID:SCR_019732).

2.3.6. Pancreatic immunohistochemistry:

Pancreata were embedded in paraffin, sectioned at 5 μm, stained with hematoxylin, and chromogenically stained for either insulin or glucagon. Images of pancreata were acquired using an Olympus VS200 slide scanner in the brightfield configuration at 20X. Images were analyzed using FIJI software, based on an approach adapted from Poudel et al., (2016). Islets were morphologically and histologically identified as encapsulated areas devoid of hematoxylin staining and blood cells. The total islet area per pancreas and the percentage of islet area that was immunopositive for insulin or glucagon stain was calculated.

2.4. Behavioral assessments

All behavior tests took place after roughly 8 weeks of isolation or group housing. Mice were given at least one “rest day” between behavior tests.

2.4.1. Open field test:

Mice were placed in the corner of a 50 x 50 x 25 cm plexiglass arena (20 lux) and allowed to freely explore the arena for 30 min. The assay was recorded using an overhead camera, and the total distance traveled (cm), time spent in the center of the arena, and time spent in the corners of the arena were quantified using Ethovision XT14 (Noldus, Leesburg, VA, USA). “Center” was defined as the central 15% of the arena.

2.4.2. Von Frey:

The von Frey test was performed as previously described (Khan et al., 2023). Briefly, mice were placed in a clear acrylic box (10 x 10 x 15 cm) and placed on the test apparatus. The test apparatus consisted of a raised wire mesh platform (holes were 5 x 5 mm). Mice were allowed to acclimate to the acrylic box and test apparatus for three h on two consecutive days. Mechanical sensitivity was evaluated by applying von Frey filaments (Stoelting, Wood Dale, IL, USA) of varying strengths to the plantar surface of each hind paw. The number of responses out of five applications, per filament per paw, was recorded and used to calculate the 50% withdrawal threshold.

2.4.3. Hargreaves:

Mice were allowed to acclimate in a clear acrylic box (10 x 10 x 15 cm) on the test apparatus for at least three h on two consecutive days prior to testing. The test apparatus consisted of a heated base (Model 400; IITC Life Science, Woodland Hills, CA, USA) with temperature maintained at 30° C. Thermal sensitivity was evaluated by focusing a heat-generating light beam onto the plantar surface of each hind paw. The time required to evoke a paw withdrawal was recorded three times per paw and averaged to calculate the paw withdrawal latency.

2.4.4. Social interaction test:

Mice were given a subcutaneous injection of 0.1mg/kg semaglutide (AstaTech, Bristol, PA, USA) or vehicle (sterile saline) 60 min before the beginning of the social interaction test, which was performed as previously described (Khan et al., 2023). Briefly, the animal was placed in the central chamber of a three-chamber arena (20 lux) and allowed to explore the environment for 10 min. Following this, a novel C57BL/6J mouse (“stranger mouse”) of the same sex and relative age as the test mouse was placed into one of the side chambers under a metal enclosure. An empty metal enclosure was placed in the alternate side chamber. The test mouse was allowed to explore the environment and interact with the stranger mouse for 10 min. The location of the stranger mouse was alternated between the right and left chambers to control for any potential side preferences. Behavior was recorded by an overhead camera and scored by a researcher blinded to experimental groups. The total time spent interacting with the stranger mouse or with the novel object (empty enclosure) was analyzed over a 10 min period for each trial, three trials per mouse in total. Ethovision software was used to calculate locomotor distance (with manual validation) and generate representative heat maps.

2.5. RT-qPCR

Brains were flash-frozen −80° C until dissection of brain regions was performed. Total RNA from each dissected region was extracted using a TRIzol reagent (Ambion, Life Technologies, USA) as described previously (Balasubramanian et al., 2022). DNAse digestion was performed using a DNA-free^™ DNA Removal Kit (Life Technologies, USA). The purity and quantity of RNA were determined using a NanoDrop 1000 spectrophotometer (Thermo Fisher Scientific, USA). cDNA preparation of total mRNA was performed using the iScript cDNA Synthesis Kit (Bio-Rad Laboratories, CA, USA) according to the manufacturer’s protocol. RT-qPCR for target genes was performed using SYBR green qPCR master mix (Bio-Rad Laboratories, USA) and primers (Supplementary Table 1) with a CFX96 Real-time-PCR system (Bio-Rad Laboratories, USA). Fold changes in mRNA levels were determined for each gene after normalizing to β-actin Ct values using the fold change 2^−ΔΔCT method (Schmittgen and Livak, 2008).

2.6. In-situ hybridization in mouse tissue (RNAscope)

Pancreata were embedded in paraffin and sectioned at 5 μm. Brain tissue was sectioned at 35 μm coronally. All sections were mounted on histobond glass slides. RNAscope was performed according to the ACDbio manufacturer’s protocol using commercially-available probes (Supplementary Table 2), modified as in Croze et al., (2021). Images were acquired on an Olympus VS200 slidescanner at 20X. RNAscope analysis was performed using Qu-Path software (0.4.3) in accordance with the protocol outlined in Secci et al., (2023). Validation of the subcellular detection algorithm was performed using manual puncta counts as suggested by the developer. Subcellular detection parameters were adjusted until the algorithm count was found to be within 5% of the manually-counted value, bounding a linear range from single cell to anatomical region. This validation was performed once for each experimental group within each anatomical region of interest.

2.7. Electrophysiology

2.7.1. Brain slice preparation:

The procedure has been described in detail previously (Khan et al., 2023). Briefly, deeply anesthetized mice were transcardially perfused with ice-cold oxygenated modified artificial cerebrospinal fluid. Brains were quickly dissected, and coronal slices containing the dorsal raphe nucleus (DRN, 300 μm) were obtained using a vibratome (VT1200S; Leica Biosystems, Wetzlar, Germany).

2.7.2. Ex vivo electrophysiological recordings:

Neurons were visualized using an upright microscope (BX51W1; Olympus, Tokyo, Japan) accompanied by a differential interference contrast imaging system. Membrane currents were amplified with a Multiclamp 700B amplifier (Molecular Devices, San Jose, CA, USA), filtered at 3 kHz, and sampled at 20 kHz with a Digidata 1550B digitizer (Molecular Devices). Data were acquired via pClamp 11 software (Molecular Devices). Access resistance was monitored and changes greater than 20% would lead to discontinuation of the recordings.

After establishing a whole-cell configuration, cells were first sampled for proper access resistance (less than 40 MΩ) prior to initiating a series of current-clamp recordings. Cells were held at −55 mV between pulses to compare data across cells. Peri-threshold current levels were determined by incrementally injecting a family of current pulses of 1000 ms duration. Current was injected every 10 s and was stepped in 10 pA increments from −100 pA to 200 pA to obtain action potential characteristics and current-dependence data. Following the determination of spike threshold, cells were stimulated with 10 super-threshold sweeps using a P_d of 5000 ms every 10 s to acquire latency data. After acquiring baseline recordings, 1 μM semaglutide was added to the bath solution and allowed to wash over the slice for a period of 10 min. Recordings were then repeated for within-cell comparisons. No washout recordings were performed due to the slow k_off of semaglutide (> 35 min).

2.10. Experimental design and statistical analyses

Data analyses and statistical tests were performed using Prism version 10 (GraphPad Software, Inc., La Jolla, CA, USA; RRID:SCR_002798). Outliers were first determined using a ROUT test (Q = 1%). Data were then checked for normal distribution and homogeneity of variance using the Fmax test. Where data violated homogeneity of variance, a non-parametric analysis was used.

1-dimensional plots (violin plots) report individual data points with median (garnet) and quartile (black) bars. Data shown in 2-dimensional plots are reported as the mean ± SEM. Statistical difference was defined at α < 0.05, and all statistical tests used a two-tailed hypothesis. Exact p values, statistical tests, and specific variables required for independent replication are found in the Statistical Summary Table.

3. Results

3.1. Adolescent isolation alters body composition, ingestive behavior, and glucose homeostasis

After weaning to group-housed or isolated caging (age P22), mice were maintained for a period of 8 weeks (to P78) before performing behavioral or physiological assessments. While adolescent isolation does not alter body weight (Figure 1A), we observed changes in lean tissue mass (*p = 0.0463) and free body fluid (*p = 0.0174) with a trend toward increased body fat (Figure 1B–D). Adolescent-isolation mice consume more chow during the dark phase only (*p = 0.0232, Figure 1E), with no change in water consumption during either dark or light phase (Figure 1G–H). Using Promethion metabolic phenotyping chambers, we observed no overall changed in the whole-animal metabolic measures of oxygen consumption, thermogenesis, or respiratory exchange ratio (Figure 1I–N). This indicated that the change in body composition was likely due to hyperphagia.

Figure 1.

Body composition, ingestive behaviors, and metabolism. A: Violin plot comparison of body weights between group-housed mice (Group; black circles) and adolescent-isolation mice (Isolation; blue squares) after 8-week challenge. B-D: Comparison of fat tissue (B), lean tissue (C), and free body fluid (D) as assessed by EchoMRI. E-N assessed using Promethion metabolic phenotyping cages, mice singly-housed for assessment. E-F: Total food intake per mouse over a 12-hour dark phase (E) or light phase (F). G-H: Same as E-F, but water intake. I: 24-hour plot of relative oxygen consumption (VO₂), with (J) integrated area under the curve (iAUC) separated by dark (left) and light (right) phase. K-L: Same as I-J, but for thermogenesis as determined using indirect calorimetry. M-N: Same as I-J, but for respiratory exchange ratio (RER; VO₂/VCO₂). *p < 0.05

We next performed an intraperitoneal glucose tolerance test (GTT). A 2-way mixed model with repeated measures (2w-ANOVA) revealed a main effect of housing (*p = 0.0113, F (1, 18) = 7.970), and a housing x time interaction (**p = 0.0055, F (4, 72) = 3.999, Figure 2A). iAUC was larger in group-housed mice when corrected for baseline levels (*p = 0.0227, Figure 2B). Interestingly, adolescent-isolation mice demonstrated higher fasted blood glucose levels versus group-housed mice (****p = < 0.0001, Figure 2C). During performance of an intraperitoneal insulin tolerance test (ITT), 2w-ANOVA revealed a non-significant trend toward a main effect of housing (p = 0.0594, Figure 2D). iAUC was not different between groups, however, we found glucose clearance within the first 15 min of the ITT to be larger in magnitude in adolescent-isolation mice (**p = 0.0055, Figure 2F). To investigate peripheral insulin signaling, we performed western blot to determine the ratio of phosphorylated AKT (phosphorylated protein kinase B; P-AKT) to unmodified AKT in various tissues. In skeletal muscle, 1-way ANOVA (1w-ANOVA) revealed that adolescent-isolation mice had a lower P-AKT/AKT ratio in response to insulin injection than group-housed mice (****p = 0.0001, F (3, 15) = 46.29, Figure 2I). Quantitative densitometry determined that the amount of total AKT could not account for this difference, as it was found to be higher in group-housed insulin-injected mice versus adolescent-isolation insulin-injected mice when normalized to β-actin loading controls (**p = 0.0029, Figure 2G, purple annotation). We further used IHC and RNAscope to characterize relevant pancreatic function. In adolescent-isolation mice, we found less immune-positive area for insulin as a percentage of total islet area (*p = 0.0211, Figure 2M). Using RNAscope, we profiled transcription levels for Ins2 (insulin), Gcg (glucagon), and Sst (somatostatin). In pancreata, somatostatin inhibits the secretion of insulin and glucagon. We found the Sst/Ins2 ratio to be diminished in adolescent-isolation mice (*p = 0.0434, Figure 2Q) without an apparent change in the Sst/Gcg ratio, suggesting an increase in relative Ins2 transcription levels. Interestingly, an ELISA assessment of plasma insulin found no difference between group-housed and adolescent-isolation mice (Figure 2S).

Figure 2.

Glucose homeostasis, insulin signaling, and pancreatic function. A: Blood glucose levels in response to intraperitoneal injection of glucose (glucose tolerance test, GTT), with (B) iAUC. C: Comparison of fasted blood glucose levels. D-E: Same as A-B, but following intraperitoneal injection of insulin (insulin tolerance test, ITT). F: Comparison of blood glucose slopes from baseline (t₀) to 15 min post-injection (t₁). G-K: Insulin signaling in various tissue types as determined by AKT signaling. Two animals per treatment (insulin (Ins) or vehicle (Veh) injection) from each housing condition were blotted for phosphorylated AKT (P-AKT) or base state AKT (blots). Corresponding graphs plotting the ratio of P-AKT/AKT as determined using quantitative densitometry for liver (H), skeletal muscle (I), brown adipose tissue (BAT, J) and white adipose tissue (WAT, K). L-O: Immunohistochemical assessment of insulin and glucagon production in pancreata. L: Representative brightfield images of pancreata immunolabeled (brown) for insulin in group-housed (left) and adolescent-isolation (right). M: Pixels immunopositive for insulin as a percentage of total islet area. N: Same as M, but for glucagon. O: Total islet area per pancreas section. P-S: RNAscope assessment of relevant genes in pancreata. P: Representative images for pancreata labeled with in situ probes for Ins2 (insulin, cyan), Gcg (glucagon, yellow), and Sst (somatostatin, magenta) for group-housed (left) and adolescent-isolation (right). Q: Sst/Ins2 ratio across averaged across entire pancreatic sections. R: Same, but Sst/Gcg ratio. S: Comparison of fast plasma insulin levels as assessed by ELISA. *p < 0.05, **p < 0.01, ****p < 0.0001

3.2. Adolescent isolation affects sleep, locomotion, and pain sensitivity

Using Promethion metabolic phenotyping chambers, we examined locomotion and sleep behaviors in a home cage environment. We observed no difference in locomotion (Figure 3A–B), with a reduction in time spent asleep during the light phase only (*p = 0.0288, Figure 3D). We then performed an open field assay, and found that adolescent-isolation mice traveled less distance throughout the test (Figure 3E,H,J) without an apparent change in time spent in the center (Figure 3F,I,L) or corners of the arena (Figure 3G,J,M). We further found no change in time spent in the center or corners of the arena when data were analyzed in a time-dependent manner (comparisons not shown).

Figure 3.

Locomotion, sleep, and anxiety-like behavior. A-D assessed using Promethion metabolic phenotyping cages, mice singly-housed for assessment. A-B: Total locomotor distance throughout dark phase (A) and light phase (B). C-D: Total time spent asleep during dark phase (C) and light phase (D). E-M: Open field assay. E-G: Total distance traveled (E), time spent in center of the arena (F), and time spent in corners of the arena (G) as determined by Ethovision software over the first 10 min of the test. H-J: Same as E-G, but for the first 20 min of the test. K-M: Same as E-G, but for all 30 min of the test. *p < 0.05, **p < 0.01, ***p < 0.001

Adolescent isolation has been shown to increase mechanical pain thresholds, without affecting thermal sensitivity in ddY and CFW mouse strains (Horiguchi et al., 2013; Konecka and Sroczynska, 2008). Using a Von Frey assessment, we observed a decrease in mechanical pain thresholds in adolescent-isolation C57 mice as compared to group-housed controls. 2w-ANOVA revealed a housing x filament interaction (*p = 0.0193, F (8, 144) = 2.384, Figure 4A). Adolescent-isolation further showed a shallower average slope across filament gauges (*p = 0.0235, Figure 4B), and may have a reduction in 50% withdrawal threshold (p = 0.0503, Figure 4C) We observed no difference in thermal sensitivity by measuring paw withdrawal time during a Hargreaves test (Figure 4D).

Figure 4.

Mechanical and thermal pain sensitivity. A-C: Von Frey test. A: Percent response throughout a von Frey test, plungers represent SEM. B: Comparison of average slope between filaments. C: 50% withdrawal threshold as determined using the Chaplain up-down method. D: Hargreaves test comparing thermal paw withdrawal latency for both paws averaged. *p < 0.05

3.3. Adolescent isolation induces transcriptional changes in neural 5-HT, GLP-1, and appetitive circuits

To probe changes in neural gene expression that could accompany the phenotypic changes observed in adolescent-isolation mice, we used RT-qPCR to examine the olfactory bulb (OB), arcuate nucleus and ventromedial hypothalamus (ARC/VMH), suprachiasmatic nucleus (SCN), lateral hypothalamic area (LHA), periaqueductal grey (PAG), laterodorsal tegmental nucleus (LDT), and rostral ventromedial medulla (RVM). These regions were selected for their known roles in eating behavior, regulation of circadian behaviors, and nociception. We observed broad transcriptional changes in 5-HT-related genes (Slc6a4, Htr1a, Htr2c), insulinotropic genes (Gcg, Insr, Glp1r), and appetitive genes (Faah, Cnr1, Cart1, Hcrt) (Figure 5).

Figure 5.

RT-qPCR of target genes. Relative changes in transcription are in adolescent-isolation mice (Isolation) compared to group-housed mice (Group) separated by olfactory bulb (OB, A), arcuate nucleus with ventromedial hypothalamus (ARC/VMH, B), lateral hypothalamic area (LHA, C), periaqueductal grey (PAG, D), laterodorsal tegmental nucleus (LDT, E), and rostral ventromedial medulla (RVM, F). G: Color-coded heat map denoting fold change for all assessed genes. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001

We next used RNAscope to spatially determine differences in the transcription of genes associated with the 5-HT, GLP-1, and POMC circuitry. In 5-HT neurons of the raphe nuclei (5-HT^RN), these included Tph2 (which regulates 5-HT production) and Glp1r (GLP-1R) (Figure 6A). Activation of GLP-1R can increase the activity of 5-HT^RN, modulating appetite (Anderberg et al., 2017). We first profiled several raphe nuclei separately for the number of in situ Glp1r mRNA spots per Tph2-positive cell, then pooled relevant raphe nuclei together. We found no change in the relative percentage of Tph2-positive cells that expressed Glp1r, within each or across all raphe nuclei, respectively (Figure 6D–E). We found that 5-HT^B9 of adolescent-isolation mice transcribed fewer copies of Glp1r compared to group-housed mice (*p = 0.0227), and this held true across all raphe nuclei pooled together (*p = 0.0447) (Figure 6F–G).

Figure 6.

Spatial transcriptomic assessment of raphe nuclei, nucleus tractus solitarius (NTS), and arcuate nucleus (ARC) using RNAscope. A: Coronal sections containing B9 5-HT neurons (5-HTNs) and median raphe nucleus (MRN) (left) or dorsal raphe nucleus (DRN) and raphe magnus (RM) (right) labeled with in situ probes for Tph2 (green) and Glp1r (magenta), (inset, left). B: Coronal section containing PPG neurons (PPGNs) of the NTS labeled with in situ probes for Gcg (green), Htr1a (cyan) and Htr2c (magenta). C: Coronal section of the ARC containing POMC neurons (POMCNs) labeled with Pomc (green), Htr2c (cyan) and Fos (magenta). D: Percentage of Tph2-positive cells within each serotonergic nucleus without or with co-labeling of Glp1r. E: Same as D, but combined across all sampled nuclei. F: Quantitation of in situ probe spots per Tph2-positive cell, separated by serotonergic nucleus. Each data point represents the mean across an entire tissue section. G: Same as F, but combined across all sampled nuclei. H: Percentage of Gcg-positive cells without co-labeling (Gcg only) or with co-labeling of Htr1a, Htr2c, or both. I: Quantitation of in situ probe spots per Gcg-positive cell. J: Percentage of Pomc-positive cells without co-labeling (Pomc only) or with co-labeling of Fos, Htr2c, or both. K: Quantitation of in situ Glp1r probe spots per Pomc-positive cell. L: Same, but for Htr2c. *p < 0.05, **p < 0.01

In PPG^NTS, we spatially profiled transcription of Gcg (GLP-1), Htr1a (5-HT_1AR), and Htr2c (5-HT_2CR) (Figure 6B). Activation of 5-HT_1AR is known to inhibit PPG^NTS, and activation of 5-HT_2CR is known to excite them (Holt et al., 2017). We found no change in the relative percentage of PPG^NTS that expressed either Htr1a, Htr2c, or both (Figure 6H). PPG^NTS of adolescent-isolation mice trended toward increased Gcg transcription (p = 0.0624), while exhibiting a decrease in Htr1a (*p = 0.0227), and no change in Htr2c (Figure 6I). In the hypothalamus, POMC^ARC are a potent inhibitor of appetitive drive and represent an intersection of 5-HT and GLP-1 circuitry. Herein, we first spatially profiled the transcription of Pomc (POMC), Htr2c, and Fos (c-Fos) to infer basal activation of POMC^ARC (Figure 6C). We found that more Fos-negative POMC^ARC expressed Htr2c in adolescent-isolation mice, with no apparent change in the overall percentage of Fos-positive POMC^ARC (Figure 6J). To complement this, we further profiled the transcription levels of Glp1r and Htr2c. POMC^ARC of adolescent-isolation mice exhibited an increase in Glp1r transcription (**p = 0.0045, Figure 6K) with no change in Htr2c puncta per neuron (Figure 6L). These findings align with those found using RT-qPCR of brain punches (Figure 5).

3.4. 5-HT neurons of isolated mice are resistant to the GLP-1R agonist semaglutide, which rescues deficits in social interaction

Since we discovered reduced levels of Glp1r in the 5-HT^RN of adolescent-isolation mice (Figure 6), we sought to determine the effects of the GLP-1R agonist semaglutide on neuronal physiology. We first used ex vivo electrophysiology in the whole-cell configuration to profile evoked and innate properties of 5-HT neurons in the dorsal raphe nucleus (5-HT^DRN). To profile the electrical activity of 5-HT circuitry, we blocked GABAergic and glutamatergic transmission while leaving 5-HT transmission intact. A 2w-ANOVA revealed that the 5-HT^DRN of adolescent-isolation mice exhibited higher action potential frequency (*p = 0.0170, F (1, 462) = 5.736, Figure 7A), and shorter interspike interval across current stimuli (****p < 0.0001, F (1, 440) = 18.01, Figure 7B). Though baseline rheobase (a measure of excitability) did not differ between groups, 2w-ANOVA revealed that bath application of 1 μM semaglutide affected this property in only 5-HT^DRN of group-housed mice (*p = 0.0456, F (1, 21) = 4.518, Figure 7D). A housing x drug interaction was also observed (*p = 0.0269, F (1, 21) = 5.665). This suggests that the loss of Glp1r in adolescent-isolation 5-HT^DRN (Figure 7B) may render them resistant to some effects of semaglutide. Semaglutide reduced the overall action potential frequency of 5-HT^DRN in both conditions (Group: ****p < 0.0001; Isolation: **p = 0.0066) without an interaction effect of housing (3-way ANOVA; current x housing x drug interaction p = 0.9846, Figure 7E). Interestingly, the effect size of semaglutide on frequency was notably smaller in 5-HT^DRN of adolescent-isolation mice (Group: 30.41, Isolation: 7.459). Lastly, we categorized the individual within-cell effects of semaglutide on normalized metrics of frequency (Figure 7F), interspike interval (Figure 7G), and latency to first spike (Figure 7H). We find the 5-HT^DRN to be highly heterogenous in their responses to semaglutide, with no apparent up- or down-regulation bias aligning with housing condition (2w-ANOVA; Frequency: p = 0.2189; Interspike Interval: p = 0.5985; Latency to First Spike: p = 0.9603).

Figure 7.

Electrophysiological assessment of putative 5-HT neurons, as assessed via whole-cell access. A: Comparison of action potential frequency across a range of electrical stimuli. B: Interspike interval across the same range of electrical stimuli as in A. C: Representative recordings for a pair of cells across electrical stimuli. D: Comparison of rheobase before and after bath application of 1 μM semaglutide. E: Two-way plots demonstrating the effect of 1 μM semaglutide on action potential frequency of 5-HT neurons from group-housed mice (left) and adolescent-isolation mice (right). Remaining 3-way ANOVA comparison is shown above. F-H: Normalized plots showing relative changes in action potential frequency (F), interspike interval (G), and latency to first spike (H) at suprathreshold current before and after semaglutide application. *p < 0.05, **p < 0.01, ****p < 0.0001

Though our findings suggest that adolescent-isolation 5-HT^DRN are resistant to semaglutide (Figure 6D–G, Figure 7D–E), other brain areas relevant to social interaction behaviors (Fang et al., 2022; Krzywkowski et al., 2020; Lo et al., 2019) exhibited an increase in Glp1r, specifically the arcuate nucleus of the hypothalamus (Figure 5B, Figure 6K). We therefore injected mice with 0.1 mg/kg semaglutide or vehicle (sterile saline) 60 min prior to the commencement of a 3-chamber social interaction test to explore any differential effects on sociability. Since adolescent-isolation mice exhibited hypolocomotive activity during the open field assay (Figure 3K), we first assessed the total distance traveled during the task. A 2w-ANOVA revealed a main effect of drug (***p = 0.0003, F= (1, 36) = 16.08, Figure 8B), with semaglutide reducing locomotor activity in only adolescent-isolation mice (**p = 0.0016). No difference was observed in vehicle-injected mice across housing conditions. We next assessed the number of entrances per chamber, with a 3w-ANOVA revealing a housing x drug x chamber interaction (*p = 0.0267, F (0.6676, 5.340) = 10.29, Figure 8C). Interestingly, Tukey’s multiple comparison test found that semaglutide reduced the number of entrances to the social chamber in group-housed mice (*p = 0.0229), while also reducing the number of entrances to the empty chamber in adolescent-isolation mice (*p = 0.0454). A 3w-ANOVA further revealed a housing x drug x chamber interaction for total time spent in either the social or empty chambers (**p = 0.0032, F (1, 9) = 15.84, Figure 8D). Compared to vehicle-injected group-housed mice, Tukey’s multiple comparisons test demonstrated social deficits in both semaglutide-injected group-housed mice (*p = 0.270) and vehicle-injected adolescent-isolation mice (*p = 0.0237), but not in semaglutide-injected adolescent-isolation mice (p > 0.9999), suggesting a housing-dependent effect of semaglutide injection- suppressing sociability in group-housed mice, and enhancing sociability in adolescent-isolation mice. Given that semaglutide did not affect the locomotion of group-housed mice (Figure 8B), it is unlikely that differences in locomotor activity could account for the observed differential housing-dependent effects of semaglutide on social interaction.

Figure 8.

Social interaction behaviors, and the effects of semaglutide administration. Group-housed (Group, Gr) and adolescent-isolation (Isolation, Iso) mice were subcutaneously injected with either sterile saline (vehicle, Veh) or 0.1 mg/kg semaglutide (Sema) 60 mins prior to testing. A: Representative heat maps generated using Ethovision software for a single mouse per condition, representing mouse position across the entirety of a 10 min experimental trial. B: Total distance traveled through the experimental trial. Main and interactive effects determined by 2-way mixed model with Tukey’s multiple comparisons test. C: Number of entries into either a chamber containing an enclosure populated by a stranger mouse (Social Chamber) or a chamber containing an empty enclosure (Empty Chamber). Full main and interactive effects of a 3-way ANOVA performed for housing (H), drug (D) and chamber (C) shown at top, post-hoc testing performed using Tukey’s multiple comparisons test. D: Total time spent within either the social or empty chamber, statistical testing and experimental conditions performed as in C. *p<0.05, **p < 0.01

4. Discussion

4.1. Summary of key findings

Isolation of rodents throughout adolescence has long been used as an experimental paradigm to study neuropsychiatric disorders. Herein, we report that much of the affected central circuitry may dually influence mental well-being and metabolic health. This circuitry includes the neural 5-HT, GLP-1, POMC and endocannabinoid systems. We further report reciprocal changes in receptor expression within the DRN-NTS circuit, and changes in receptor expression at their point of convergence in the hypothalamus. Most interestingly, we report that the GLP-1R agonist semaglutide, commonly prescribed for the treatment of type 2 diabetes, was sufficient to differentially affect social interaction. This is especially interesting since semaglutide is not known to cross the blood brain barrier (Salameh et al., 2020), but rather directly accesses GLP-1Rs on the brainstem and hypothalamus (Gabery et al., 2020). The neural circuitry discussed in this report may underlie the central action of GLP-1 analogues, separate from effects mediated by the gut-brain axis. These findings elevate the role of neural GLP-1, often overshadowed by its peripherally-produced counterpart, and suggest a role for central serotonin circuitry in its action.

4.2. Preclinical models

Though reduction of Glp1r rendered 5-HT^DRN resistant to semaglutide, the drug was still sufficient to rescue social interaction behavior (Figure 8), a deficit commonly attributed to 5-HT function. 5-HT and GLP-1 circuitries separately converge onto POMC^ARC, which express 5-HT_2CR and GLP-1R. Our finding of increased Glp1r in POMC^ARC (Figure 6K) may underlie chronic housing-dependent activation of these neurons (Fang et al., 2022), accounting for this rescuable deficit in social interaction (Figure 8C,F). However, this circuitry does not explain the hyperphagic/obesogenic phenotype commonly observed in this paradigm (Figures 1,2). Interestingly, we observed increased Hcrt in the ARC/VMH of adolescent-isolation mice (Figure 5B). Orexin is known to directly activate 5-HT^DRN, which then reciprocally projects to orexin neurons to suppress their activity (Liu et al., 2002). Our findings suggest a loss of serotonin-mediated inhibition of orexin neurons, complementing previous reports of hyperactive AGRP neurons in hyperphagic adolescent-isolation mice (Yamada et al., 2015). Despite this, basal Fos levels within POMC neurons did not differ between groups (Figure 6J). This finding suggests the involvement of other neural circuitry, possibly including joint modulation of olfactory neurons by serotonin and locally-produced GLP-1 (Huang et al., 2021, 2017; Thiebaud et al., 2019). Though we observed reduced Gcg in the OB of adolescent-isolation mice, the involvement of olfactory circuitry is beyond the scope of this report.

Limited data demonstrate that GLP-1R agonists are sufficient to increase 5-HT activity at the single-cell level, without broadly profiling these effects on full data sets (Anderberg et al., 2017). 5-HT nuclei are subject to de facto lateral inhibition, wherein activation of the 5-HT_1AR on 5-HT neurons is sufficient to reduce their activity (Quentin et al., 2018). Accordingly, we find that the GLP-1 analog semaglutide suppressed the activity of most 5-HT^DRN when 5-HT transmission was left intact, despite pharmacological block of glutamatergic and GABAergic transmission. This suggests a heterogenous population of 5-HT neurons within the DRN that do not uniformly express 5-HT_1AR and GLP-1R. In seeming opposition to our findings, other groups have reported decreased 5-HT^DRN activity following isolation throughout adolescence (Sargin et al., 2016). However, this investigation was performed using transgenic mice without blockade of glutamatergic transmission, and the electrical stimuli used were unconventionally large in magnitude. Within the stimulus range that we now report, the data contained in Sargin et al., (2016) display no apparent difference in 5-HT neuron excitability for adolescent-isolation mice. Despite this effect of adolescent isolation on 5-HT^DRN function, we did not observe typical anxiety-like behaviors during the open field assay. However, given that we observed no housing-dependent differences in locomotor activity while housed in a home cage environment (Figure 3A–B) or during the social interaction test (Figure 8B), the difference in locomotion during the open field assay could possibly represent a milder anxiety-like behavior (Figure 3E,H,K).

4.3. Metabolic dysregulation, mental health disorders, and social isolation in human populations

In 2020, social distancing and stay-at-home orders were implemented as a non-pharmacological measure to stem the spread of SARS-CoV-2 worldwide. This period ranged in duration from 3-30 months (CDC, 2022; Moreland et al., 2020), far greater than the 10-day duration that is known to cause lasting psychiatric effects (Brooks et al., 2020; Leigh-Hunt et al., 2017; Pietrabissa and Simpson, 2020; Rogers et al., 2020). Social isolation can be particularly harmful to children and adolescents (Saurabh and Ranjan, 2020; Schinka et al., 2013; Sprang and Silman, 2013), greatly impacting development and increasing the risk of disease in adulthood (Andersen et al., 1980; Caspi et al., 2006; Danese et al., 2009; Koss et al., 2014; Lacey et al., 2014). During the pandemic, the rate of body weight gain nearly doubled in persons aged 2-19. From August 2019 to August 2020, obesity prevalence increased from 19.3% to 22.4% in this age group (Lange et al., 2021). Separately, the perception of social isolation greatly increased the occurrence of depression during covid (Pietrabissa and Simpson, 2020). The co-presentation of metabolic dysfunction and depression was not assessed in these groups, but first-episode psychosis patients often present with metabolic alterations. This is thought to be due to their voluntary social isolation (Bocchio-Chiavetto et al., 2018), which is known to be a risk factor for type 2 diabetes (Henriksen et al., 2023). In human populations, it is unclear whether psychiatric-related isolation drives metabolic alterations, if impaired insulinotropic signaling promotes the development of psychiatric disorders, or if dysfunction of central 5-HT could be simultaneously driving both conditions. Further investigation is needed to determine if central 5-HT/GLP-1 circuitry may represent a pharmaceutical intersection for persons co-presenting with metabolic dysregulation and psychiatric symptoms.

Highlights.

• Adolescent isolation of rodents dually affects metabolism and mental well-being

• Adolescent isolation changes gene transcription in neural 5-HT & GLP-1 circuits

• 5-HT neurons of isolated mice are resistant to the GLP-1R agonist semaglutide

• Semaglutide rescues deficits in social interaction induced by this paradigm