Metabolomic profiles associated with a mouse model of antipsychotic-induced food intake and weight gain

Antipsychotic drugs (AP) are used to treat a multitude of psychiatric conditions including schizophrenia and bipolar disorder. However, APs also have metabolic side effects including increased food intake and body weight, but the underlying mechanisms remain unknown. We previously reported that minocycline (MINO) co-treatment abrogates olanzapine (OLZ)-induced hyperphagia and weight gain in mice. Using this model, we investigated the changes in the pharmacometabolome in the plasma and hypothalamus associated with OLZ-induced hyperphagia and weight gain. Female C57BL/6 mice were divided into groups and fed either i) control, CON (45% fat diet) ii) CON + MINO, iii) OLZ (45% fat diet with OLZ), iv) OLZ + MINO. We identified one hypothalamic metabolite indoxylsulfuric acid and 389 plasma metabolites (including 19 known metabolites) that were specifically associated with AP-induced hyperphagia and weight gain in mice. We found that plasma citrulline, tricosenoic acid, docosadienoic acid and palmitoleic acid were increased while serine, asparagine and arachidonic acid and its derivatives were decreased in response to OLZ. These changes were specifically blocked by co-treatment with MINO. These pharmacometabolomic profiles associated with AP-induced hyperphagia and weight gain provide candidate biomarkers and mechanistic insights related to the metabolic side effects of these widely used drugs.

Metabolomics is a rapidly expanding technology based on either NMR or LC/GC-based mass spectrometry which allows for the identification of complex metabolite profiles and establish their correlation with disease. Data derived from metabolomics can provide a deeper mechanistic understanding of drug-induced changes induced by APs at the metabolite level and also identify biomarkers of efficacy and side-effects to further facilitate clinical trials during the evaluation of future AP therapies. Here, using metabolomics, we determined the specific AP-induced biomarkers of weight gain and hyperphagia in the plasma and hypothalamus of OLZ-treated mice.

Methods
Mouse studies. All procedures were approved by UCSD IACUC and all experiments were performed in accordance with relevant guidelines and regulations. Female C57BL/6 (stock #000,664) were purchased from Jackson Labs at 9 weeks of age and were acclimatized to the experimental environment (12:12 light/dark, 20-22 °C, 60% humidity) with normal chow until 10 weeks of age. Mice were then randomized to four treatment groups: control (CON), OLZ, MINO, and OLZ + MINO. CON animals were fed a high fat diet (HFD, 45 kcal% from fat, Research Diets, Inc., D09092903, New Brunswick, NJ). OLZ was compounded into the HFD at a concentration of 54 mg/kg of diet 28,29 . MINO was administered in drinking water at a dose of 0.6 mg/ ml 28,30 . After 2 weeks of treatment, mice were sacrificed and blood collected in heparinized capillaries followed by centrifuge at 12,000 rpm for 10 min at 4 °C to collect plasma. Hypothalami were dissected and snap-frozen in liquid nitrogen. Samples were stored at -80 °C until analysis. Four to five samples per group were processed for metabolomics analysis.
Untargeted metabolomics. Metabolomics analysis was performed on collected mouse plasma and hypothalamus samples as previously described [31][32][33] . Briefly, small bioactive lipid metabolites and polar, hydrophilic metabolites were extracted from plasma samples using addition of organic solvent followed by offline solid phase extraction (SPE), as described [31][32][33] . Metabolites were chromatographically separated using a Thermo Vanquish UPLC system with a Phenomenex Kinetex C18 (1.7um particle size, 100 × 2.1 mm) or Merck-SeQuant ZIC-pHILIC (5 μm particle size, 100 × 2.1 mm) column for measure of bioactive lipids and polar metabolites, respectively. Mass spectra were acquired on a Thermo QExactive orbitrap mass spectrometer with electrospray ionization in negative mode and positive mode, for measure of bioactive lipids and polar metabolites, respectively, as previously described [31][32][33] . Metabolites were identified by matching accurate mass, retention time, and MS/MS fragmentation patterns to an in-house library of commercially available standards, as described 31-33 . Bioinformatic analysis. Differential metabolomic analysis was conducted in R statistical software, using MetaboDiff version 0.9.3 34 (https ://githu b.com/andre asmoc k/Metab oDiff /), starting with the table of metabolite measurements. Following the recommended MetaboDiff workflow, K-nearest neighbor imputation was used to impute values for metabolites which were missing in less than 60% of samples. Metabolites that were missing in more than 60% of samples were excluded. Variance stabilizing normalization was then applied. Abundance changes were assessed on the normalized data, with t-tests computed in R, and the Benjamini-Hochberg multiple test correction was applied for each comparison. To identify the metabolites specifically associated with APinduced hyperphagia, and unchanged by MINO treatment alone, we used the following filtering criteria: Firstly, we identified metabolites that were significantly changed in abundance by OLZ treatment compared with CON. This list was then refined to only include metabolites that were specifically blocked by co-treatment with MINO (OLZ + MINO). Finally, this list was refined to include only metabolites that were not differentially changed in abundance between CON versus MINO-treated samples. Therefore, metabolites were filtered by the following criteria: significant in OLZ vs CON (adj pval < 0.3), AND significant in OLZ + MINO vs OLZ (adj pval < 0.3) and direction of effect opposite in OLZ vs CON compared to OLZ + MINO vs OLZ, and not significant in MINO vs CON (adj pval > 0.3). Heatmaps were generated using ClustVis 2.0 software 35 (https ://biit.cs.ut.ee/clust vis). Two-tailed Pearson correlation analysis between the AP-induced metabolites and food intake, weight gain and gonadal adipose tissue (gWAT) weight were performed using GraphPad Prism with significance set at p < 0.05.
Prediction of pathway activity from untargeted mass spectral data was determined using the mummichog algorithm 36 in Metaboanalyst 4.0, https ://www.metab oanal yst.ca 37 . The mummichog algorithm enables pathway analysis from m/z values, bypassing the need for metabolite identification. Metabolites which were significantly changed in OLZ vs CON (adj. pval < 0.3), as well as MINO vs OLZ (adj pval < 0.3), but which were not significantly altered in MINO vs CON (adj pval > 0.3) were included in the analysis. These metabolites were further filtered by those which changed in different directions in OLZ vs CON compared to MINO vs OLZ. This resulted in 389 metabolites for input to Metaboanalyst. The Metaboanalyst MSPeaksToPathways parameters were: Mass Accuracy (ppm) = 0.01, Analytical mode = Negative, P-value Cutoff = 0.05.

Results
Principal components analysis (PCA) of the plasma metabolomics (Fig. 1A) revealed that CON and OLZ show clear separation (red vs green). In addition, MINO and OLZ + MINO (yellow and blue) can be clearly differentiated from CON (green) or OLZ (red). MINO and OLZ + MINO groups had clear similarities, as shown by overlapping space in the PCA plot (blue/yellow). PCA analysis of the hypothalamic metabolites ( Fig. 1B) revealed that CON (green) and OLZ (red) groups were clearly separated. OLZ, MINO and OLZ + MINO share overlapping space suggesting that differences between these groups were less pronounced.
We detected 7140 metabolites in the plasma, comprising 396 identified metabolites and 6744 unknown metabolites (Supplemental Table 1). Differential analysis revealed that 1188 metabolites were significantly changed between OLZ vs CON adjusted p value < 0.3 (Supplemental table 2). To further identify metabolites that were specifically associated with AP-induced hyperphagia and weight gain, we applied another set of filters: metabolites  (Fig. 2).
In the hypothalamus, we detected a total of 414 metabolites of which 92 were identified and 322 unclassified metabolites (Supplemental table 5). Of these, 151 metabolites were significantly changed in OLZ compared with CON treated groups (Supplemental table 6). To identify hypothalamic metabolites that were specifically associated with AP-induced hyperphagia and weight gain, we applied the same filter that we used with plasma metabolites (metabolites that were significantly changed with OLZ treatment compared with CON and these metabolites should not be differentially changed between CON vs MINO). This revealed just one metabolite, indoxylsulfuric acid, that was increased by ~ twofold by OLZ treatment compared with CON and OLZ + MINO normalized levels to CON range ( Fig. 3 and Supplemental table 7).
As we previously reported, OLZ increased food intake (Fig. 4A), weight gain (Fig. 4B) and gWAT weight (Fig. 4C) compared to CON while co-treatment MINO reversed these OLZ-induced effects 28 . We then determined whether these metabolites correlate with food intake, weight gain and adipose tissues mass (Fig. 4). Plasma metabolites, citrulline, tricosenoic acid and palmitoleic acid were strongly positively correlated with weight gain, food intake and/or gWAT weight while HEPE, asparagine, PGA3 and serine were strongly negatively correlated with weight gain, food intake and/or gWAT weight (Fig. 4D). In addition, hypothalamic indoxylsufuric acid levels were significantly positively correlated with weight gain and tended to be positively correlated with food intake and gWAT weight (Fig. 4D).
Pathway analysis. We then conducted pathway analysis using Metaboanalyst software to determine which pathways the plasma metabolites associated with antipsychotic hyperphagia belong to (Supplemental Fig. 1). There was a significant number of hits related to arachidonic acid metabolism, steroid hormone biosynthesis and retinol metabolism ( Table 1). The arachidonic acid metabolism pathway was particularly enriched with significant changes in metabolite abundance after OLZ treatment compared with CON (Table 1, Fig. 5).

Discussion
Studies in psychiatric patients have highlighted changes in metabolite abundance associated with APs treatment [38][39][40][41][42][43][44] . However, these studies are based on relatively small, diverse human populations suffering from psychiatric disease and exposed to various APs with differing lengths of treatment. Therefore, it has been a challenge to unravel the complexity of these datasets and to specifically focus on the food-intake related side effects of AP-treatment. In the present study, we used MINO co-treatment to specifically block OLZ-induced hyperphagia in mice to identify metabolites that were specifically associated with AP-induced hyperphagia and weight gain.
In the hypothalamus, we identified a single metabolite, indoxylsulfuric acid, that was increased by ~ twofold by OLZ treatment compared with CON and this increase was specifically blocked by co-treatment MINO. Indoxylsulfuric acid (also known as Indoxyl sulfate) is a metabolite of the common amino acid tryptophan 45 . Indoxylsulfuric acid has been shown to induce oxidative stress and inflammation in central nervous system cells 46 and high levels in the brain are associated with neuroinflammation and cerebral dysfunction 47,48 . Our studies identify a novel potential role of indoxylsulfuric acid in the hypothalamus in food intake and body weight regulation.
In the plasma, we identified 4 known metabolites that increased in concentration in response to OLZ and 19 that decreased in abundance in response to OLZ (Fig. 2). The four metabolites that were increased included the amino acid citrulline and fatty acids tricosenoic acid (C23:0), docosadienoic acid (C22:2) and palmitoleic acid (C16.1). Plasma citrulline is highly induced after AP treatment and specifically blocked by the co-treatment with MINO. Citrulline is a nonessential amino acid synthesized almost exclusively by enterocytes of the small intestine. Serum citrulline levels have previously been linked to changes in food intake in mice 49 and human www.nature.com/scientificreports/ studies 50,51 . Notably, in fasted mice, significantly higher levels of serum citrulline were observed in the morning compared to the evening 49 . In addition, increased plasma citrulline in mice is associated with diet-induced obesity and may predict the development of the metabolic syndrome 52 . Accordingly, in human studies, circulating citrulline levels have been found to be reduced by approximately 10-20% in the post-prandial period 50, 51 and lowered by 30% after prolonged starvation 53 . These studies provide evidence that serum citrulline has previously been associated with changes in food intake in mouse and human studies, but further studies are required to address the role of serum citrulline in AP-induced food intake. Docosadienoic acid is a natural ω-6 polyunsaturated fatty acid (PUFA) and acts as an agonist of free fatty acid receptor 4 (FFAR4, also known as GPR120). Docosadienoic acid treatment in vitro in mouse gastric cells inhibits the secretion of the pro-feeding hormone ghrelin 54 .
Other published studies have suggested ghrelin levels are increased in response to AP-treatment 55 . Therefore, it is possible the AP-induced increase in docosadienoic acid is a response to attempt to block the AP-induced increase in ghrelin. Interestingly, there is no significant change in docosadienoic acid levels when comparing plasma from lean and obese people suggesting this increase may be specific to AP treatment 56 . Tricosenoic acid is an odd chain fatty acid (23:0) that is highly prevalent in human plasma compared with other odd chain fatty acids. However, tricosenoic acid has no known link to food intake and is unchanged in abundance in studies comparing lean and obese human plasma 57 . Finally, palmitoleic acid (palmitoleate 16:1n-7) is one of the most abundant fatty acids in serum and tissues, particularly adipose tissue and liver 58 . A recent study grouped schizophrenia patients treated with APs by quartiles of increasing metabolic impairments (indicated by fasting insulin and BMI) and found palmitoleic acid was significantly elevated with increasing metabolic impairments (p = < 0.0001, fdr 0.02%) 42 . Furthermore, higher palmitoleic acid has also been observed in plasma of the obese children 59 and adults 60 compared with control lean groups. However, administration of palmitoleic acid in rats resulted in a lower food intake 61 and interestingly, this study proposed that the increased circulating concentrations observed in obesity may result in a loss of the metabolic beneficial effects of palmitoleic acid. This phenomenon of obesity-induced elevation of circulating factors and subsequent resistance to these signals is frequently observed in obesity where OLZ OLZ+MINO CON MINO  Fig. 2). Arachidonic acid (20:4 n-6) is a polyunsaturated fatty acid (PUFA) and serves as a precursor to eicosanoids (prostaglandins, thromboxanes, leukotrienes) which mediate inflammatory responses 63 . Plasma levels of metabolites synthesized from n-6 PUFAs have been implicated in previous metabolomic and lipidomic analyses of AP treated psychiatric patients. For example, AP treatment significantly lowered plasma levels of metabolites synthesized from n-6 PUFAs including reduced prostaglandins PGE1, PGF2a and PGG2 in bipolar individuals compared with controls 64 . The arachidonic acid signaling pathway has been implicated in food intake in a variety of studies [65][66][67] . Inhibition of food intake has been reported after treatment of rodents with arachidonic acid or PGF2 alpha 65 or PGE2 65,66 through the EP4 receptor 66 . Therefore, it is possible that reduced prostaglandin signaling may contribute to the increase in food intake observed during AP treatment. However, other reports suggest that other members of the prostanoid family, specifically PGD2, stimulate food intake 67,68 , and thus, further investigation is needed to specifically understand the role of the arachidonic acid pathway in AP-induced hyperphagia. OLZ treatment also resulted in significantly lower levels of amino acids serine and asparagine compared with CON mice. In previous studies, AP treatment was associated with decreased plasma serine in AP treated patients with high BMI patients compared with lower BMI groups 42 . Furthermore, administration of serine significantly reduced food intake 69,70 in mouse studies. Therefore, we speculate that lower levels of serine observed in our studies may play a role in potentiating food intake as seen in AP-treatment. In other studies, plasma asparagine levels do not change between lean and obese people but are significantly lower in obese patients with visceral obesity (which has a stronger association with metabolic disturbances and cardiovascular risks) compared with than obesity resulting from increased subcutaneous adipose tissue 71 . Follow-up studies are warranted to investigate the role of the amino acids serine and asparagine in AP-induced food intake and weight gain.
A potential caveat of our study is that we used our mouse model of OLZ-induced hyperphagia and weight gain. Future studies will be necessary to determine if these metabolite changes are associated with hyperphagia in general, or specific to AP-induced hyperphagia. We have related our mouse-based studies to similar human studies measuring changes in the metabolome after AP treatment in various psychiatric disease populations. However, these human studies are highly complex datasets with many confounding elements including length of treatment, medication interactions, dietary variation and complex psychiatric phenotypes that all contribute to changes in the metabolic profiles. To overcome this in the future, it would be ideal to treat a control population without underlying psychiatric disease to specifically investigate the changes in the metabolome associated with AP-induced hyperphagia and weight gain.

Conclusions
In conclusion, our mouse-based study has identified metabolite biomarkers in the plasma that were associated with AP-induced hyperphagia and weight gain. Some of these AP-induced metabolites have been previously implicated in food intake and body weight regulation based on published literature. In addition, we have related these findings  www.nature.com/scientificreports/ to other human studies investigating the pharmacometabolome during AP treatment. Further studies are warranted to determine whether these AP-induced metabolites play a mechanistic role in energy balance regulation or to investigate their utility as candidate biomarkers during drug evaluation for AP-induced side-effects.

Data availability
All data generated or analyzed during this study are included in this published article and its supplementary information files.
Received: 22 May 2020; Accepted: 9 October 2020 Figure 5. Arachidonic acid metabolism pathway. Metabolites colored green were decreased by OLZ treatment compared with CON, direction of change reversed after MINO co-treatment. Prediction of pathway activity from untargeted mass spectral data was determined using the mummichog algorithm 36 in Metaboanalyst 4.0, https ://www.metab oanal yst.ca 37 .