Macaques Exhibit a Naturally-Occurring Depression Similar to Humans

Rodent models have dominated preclinical investigations into the mechanisms of depression. However, these models-which rely on subjecting individual rodents to physical stressors - do not realistically resemble the etiopathological development of depression, which occurs naturally in a social context. A non-human primate model that better reflects the social ethological aspects of depression would be more advantageous to investigating pathophysiological mechanisms and developing antidepressant therapeutics. Here, we describe and model a naturally-occurring depressive state in a non-human primate species, the cynomolgus monkey (Macaca fascicularis), in a realistic social ethological context and associate the depressed behavioral phenotype with significant serum metabolic perturbations. One to two subjects per stable social colony (17–22 subjects) manifested a depressive phenotype that may be attributed to psychosocial stress. In accordance with rodent and human studies, the serum metabolic phenotype of depressed and healthy subjects significantly differed, supporting the model's face validity. However, application of the fast-acting antidepressant ketamine failed to demonstrate predictive validity. This study proposes a non-human primate depression model in a realistic social ethological context that can better approximate the psychosocial stressors underlying depression.

D epression is a debilitating psychiatric illness with a lifetime prevalence of 21% and has been shown to significantly contribute to financial distress, workplace dysfunction, increased hospitalization, and poor health status [1][2][3] . Unfortunately, the relationship between the depressive phenotype and its underlying pathophysiology remains unclear 4 . Moreover, meta-analyses of clinical trials suggest that conventional antidepressants are only marginally efficacious compared to placebos and publication biases inflate their apparent efficacy 5 . From a clinical standpoint, the current antidepressant research and development (R&D) process has been questioned; specifically, non-detection of patient subpopulations, substandard clinical trial designs, nonideal clinical end points, lack of biomarkers, industry pressure on pharmaceutical scientists, marketing strategies, difficulty establishing public-private R&D partnerships, and publication bias in favor of positive effects have all been cited as causes for failures in the current antidepressant R&D process 6,7 . With these challenges in mind, developing a stronger understanding of the pathophysiology underlying depression is necessary to improve therapeutics for this debilitating mental disorder.
From a preclinical standpoint, rodent models have dominated preclinical investigations into the mechanisms of depression 8 . However, traditional rodent models, such as chronic unpredictable mild stress (CUMS), lipopolysaccharide (LPS) administration, foot-shock, the forced swimming test (FST), and tail suspension test (TST), all rely on subjecting individuals to physical stressors 9 . Thus, traditional rodent models do not realistically resemble the etiopathological development of depression, which occurs naturally in a social context 10 . Likewise, there is little or no discrimination between the state of stress and the state of depression in these models, although the two are quite easily identifiable as two different states in human beings; for example, most people do not become depressed after serious stressful experiences 11 .
Due to these limitations inherent in conventional rodent models of depression, Nestler et al. has commented that depression research now requires a focus on ethological models with animal subjects observed under more natural conditions 8 . Therefore, several depression research groups, including ours, have begun conducting work on more homologous non-human primate models (e.g., macaques) observed under more natural social etho-logical conditions 12 . Non-human primate studies have provided the best evidence for the occurrence of stress-induced social behaviors that parallel signs of depression, such as a stereotypical prostrated and socially unresponsive posture as well as exacerbation of socially submissive behaviors 13 . Interestingly, these signs have been partly reversed through chronic selective serotonin reuptake inhibitor (SSRI) administration and co-occur with cardiovascular, neuroendocrine, and neural abnormalities commonly observed in depressed patients 14 . Moreover, social ethological models of non-human primates display additional advantages in terms of understanding hierarchical dysfunction and individual psychopathology 15,16 .
Although ethologically valid primate models of socioaffective deficits can demonstrate strong face validity with depression, our capability to investigate their underlying circuitry is limited by technical and ethical considerations 17 . Therefore, current non-human primate depression models still require further evidence to establish model validity (e.g., ecological, etiological, face, construct, and predictive validity) [18][19][20] . In order to construct an improved non-human primate model of depression, our research group has devised a novel animal model strategy. In our initially published study, we selected a wellstudied non-human primate species, the cynomolgus monkey (Macaca fascicularis) 21 . Next, we developed a repeatable social isolation procedure to construct a cynomolgus monkey model of depression 22 , which demonstrated that social isolation is effective in inducing depression-like behavior by significantly reducing socially dominant aggressive conflict behavior, communicative behavior, sexual behavior, and parental behavior. However, in contrast to our social isolation-induced model that assesses individual subjects isolated from their respective social groups, we have also consistently observed a subset of cynomolgus monkey subjects that display naturally-occurring depressive behavior in a social context.
To distinguish depressed macaques from healthy macaques in a social ethological context, here we assessed the behavioral changes that naturally occur in depressed cynomolgus monkeys in a colonybased social environment (i.e., the free enclosure environment). The behaviors of 20 adult female cynomolgus monkeys (M. fascicularis) displaying naturally-occurring depressive behavior, 10 cynomolgus monkeys displaying depressive behavior after a period of social isolation, and 20 age-and sex-matched healthy controls were recorded over an eight-day field observation period in a social ethological context. Then, the fast-acting antidepressant ketamine was applied to assess reversals in depressive behavior 23 . As depressed human patients have been shown to display significant differences from healthy individuals in blood-based metabolites 24 , serum metabolic analysis was performed to assess the metabolic differences (if any) between depressed subjects and healthy controls to further validate our model.

Methods
Study Site and Rearing Conditions. The cynomolgus monkey feeding base at the Zhongke Experimental Animal Co., Ltd. (hereinafter ''Zhongke'') is located in Suzhou, P.R.C. (E 31u079030 to 31u079060, N 120u199080 to 120u199150). Through importing cynomolgus monkeys from Guangdong and Vietnam in 1990, Zhongke established its feeding base in order to provide reliable subjects for basic experimental use and preclinical drug screening and then established a domestication and breeding base for these monkeys. Zhongke conducts regular veterinary care, daily cleanings, weekly sterilizations, and yearly inspections. Zhongke provides environmental conditions and surroundings approximating those found in the wild ( Fig. 1A-F). For more information on the study site, please see our previous study 25 .
All monkeys were reared in socially-stable colonies with negligible rates of conflict, were provided with water ad libitum, and fed daily with fresh fruit, vegetables, and compound high-nutrition monkey food ( Fig. 1D-F). To reflect wild populations they have a male: female ratio of 1: 7-11, each colony was housed in its own free enclosure composed of two males, 16-22 adult females, and their offspring of less than six months of age. There was no statistically significant difference in male: female ratios across all 52 free enclosures (x 2 5 6.2417, df 5 51, P 5 1.000). To better enable reproduction, the mean age in each colony was maintained near 10 6 5 years of age.
Subject Selection. From April to November 2010, by means of a scanning method on a total population of 1007 adult female cynomolgus monkeys across all 52 enclosures, we identified subjects displaying depressive behavior using the following operational definition according to Shively's criteria 26 : slumped or collapsed body posture ( Fig. 1C and Fig. 2B), diminished interest in feeding and sex, and diminished communication and reciprocal grooming with others (Health Subjects, Fig. 3A-D). In total, 50 subjects met these criteria, and the frequencies of these depressive behaviors in these 50 subjects were tabulated over a two-week observational period. The 20 subjects that displayed the highest frequencies of depressive behaviors over the observational period were selected (hereinafter, naturally-occurring depressed or NOD subjects, Fig. 4A-E), and a staff veterinarian ruled out disease in these 20 subjects. To match by age and sex, 20 healthy adult female M. fascicularis subjects (aged 9-13 years) were selected (hereinafter, healthy controls or controls) by means of simple random sampling (random seed: 20101207) from the original pool of 1007 adult female cynomolgus monkeys. In order to provide reliable behavioral reference points for depressive behaviors in cynomolgus monkeys, 10 subjects displaying depressive behavior after a period of social isolation (hereinafter, social isolationinduced depressed or SID subjects) were also selected. Please see our previous investigation on social isolation-induced depression in the cynomolgus monkey for details. www.nature.com/scientificreports SCIENTIFIC REPORTS | 5 : 9220 | DOI: 10.1038/srep09220 Experimental Procedures, Behavioral Observation, and Recording. From January 2011 to August 2013, all 40 subjects were observed for eight days by three trained observers (totaling .5000 hours of field observations). Behavioral recording methods and scored behavioral items have been described in our previous work 25 . Briefly, all behaviors of each subject were videotaped, and the three observers blindly scored all discreet behavioral items on the videotape footage using NOLDUS Observer XT software (version 10.0, Noldus Information Technology, Leesburg, PA). Data on twelve behavioral categories were gathered (Table 1) with an inter-observer reliability of greater than 85% for each discreet behavioral item.
Ketamine Testing. The fast-acting antidepressant ketamine, a noncompetitive NMDA receptor (NMDAR) antagonist, was used to assess reversals in depressive behavior. To compare ketamine's effects in the three groups, two subjects were randomly selected from each group by simple random sampling (random seed: 20120705). Ketamine, administered at a dose of 2.5 ml/kg I.M., was injected into the distal portion of the hind limb at 9:00 A.M. We observed the six drugged subjects daily from 9:30-11:30 A.M. and 2:30-4:30 P.M. over a period of seven days.

GC-MS Sample Preparation and Derivation.
For gas chromatography-mass spectrometry (GC-MS) analysis, 15 ml of sampled serum was added to 10 ml Lleucine-13C6 (0.5 mg/ml) as an internal standard. After vortexing for 30 s, 75 ml of a methanol and chloroform mixture (methanol/chloroform: 351) was added. After centrifuging at 14000 rpm for 15 min, 80 ml of supernatant was evaporated to dryness under a stream of nitrogen gas. The dried residue was added to 30 ml of methoxamine hydrochloride (20 mg/ml pyridine) and incubated at 37uC for 90 min with continuous shaking. Subsequently, the solution was treated with 30 ml of BSTFA (1% TMCS) at 70uC for 60 min and placed at room temperature for 30 min.
GC-MS Analysis. A 1 ml volume from each derived sample was injected into an Agilent 7890A/5975C GC-MS system (Agilent Technologies Inc., USA). An HP-5 MS fused silica capillary column (30 m 3 0.25 mm 3 0.25 mm, Agilent, USA) with a helium carrier gas flow rate of 1 ml/min was applied for metabolite separation. The injector temperature was set at 280uC. The column temperature was initially kept at 80uC for 2 min and then increased to 320uC at 10uC/min, where it was maintained for 6 min. The column effluent was introduced into the ion source of an Agilent 5975 mass selective detector (Agilent Technologies). The MS quadrupole temperature was set at 150uC, and the ion source temperature was set at 230uC. MS detection was conducted with electron impact ionization in the full scan mode (m/z, 50-600).
Metabonomic Data Analysis. GC-MS metabolite profiles of all 40 subjects were processed after conversion into a NetCdf file format using TagFinder 27 . This processing enabled deconvolution, alignment, and data reduction to produce a list of mass and retention time pairs with corresponding intensities for all detected peaks from each data file in the data set. The resulting three-dimensional data set - Figure 2 | (A-G) Lower-ranking subjects facing aggression from higher-ranking subjects. Red arrow indicates the higher-status monkey, aggressing against the lower-status monkeys (indicated by blue arrow). All photographs were taken by Dr. Fan Xu. www.nature.com/scientificreports including peak index (RT-m/z pair), sample names (observations), and normalized peak area percentages -were imported into SIMCA-P 11.0 (Umetrics, Umeå, Sweden) for statistical analysis.
Partial least-squares discriminant analysis (PLS-DA), a supervised multivariate approach, was performed to visually discriminate between the two experimental groups. The quality of the PLS-DA models was described by three parameters (R 2 X, R 2 Y, and Q 2 Y), which were calculated by the default leave-one-out procedure. R 2 X and R 2 Y were used to quantify the goodness-of-fit; Q 2 Y was applied to assess the predictability of the model. If the Q 2 and R 2 values resulting from the original model were higher than the corresponding values from the permutation test, the model was considered valid. The discriminating metabolites were obtained using a statistically significant threshold of variable influence on projection (VIP) values obtained from the OPLS-DA model (VIP . 1) and two-tailed Student's t-test (P-value , 0.05).
Behavioral Data Analysis. Behavioral data was coded as duration (in seconds) for each discreet behavioral item per each 30-minute observational phase and presented as means 6 S.D.'s. To assess the behavioral differences between NOD subjects and healthy controls and those between NOD subjects and SID subjects, Student's t-test was performed if the data was normally distributed; otherwise, the Mann-Whitney U test was applied. One-way analysis of variance (ANOVA) was performed to compare the behavioral characteristics of all three groups. In order to reduce type I errors during multiple comparisons, Bonferroni correction was used to adjust P-values. P-values of less than 0.05 were deemed significant for all analyses. All data management and statistical analysis were performed using Stata 12.0 (StataCrop LP, College Station, Texas 77845, USA).
Ethics Statement. All animal work was conducted according to relevant national and international guidelines. In accordance with the recommendations of the Weatherall report, ''The use of non-human primates in research,'' the following statement has been included to document the details of animal welfare and the steps taken to ameliorate suffering in all work involving non-human primates. This study was performed in strict accordance with the recommendations in the ''Guide for the Care and Use of Laboratory Animals'' of the Institute of Neuroscience at Chongqing Medical University (approval no.: 20100031). State regulators and the Committee on Ethics of Animal Experimentation at Chongqing Medical University approved the protocols prior to implementation. For more detail, please refer to our previous publication 25 .

Behaviors of Greater Duration in Healthy Controls
Relative to NOD Subjects. The behaviors of greater duration in healthy controls relative to NOD subjects fell into the following behavioral  categories: ingestion, rutting and mating, resting, locomotive, amicable, and miscellaneous (Tables 1, 2). First, as to ingestion behaviors, healthy controls spent longer periods 'drinking' during the feeding phase than NOD subjects ( Fig. 1D-E). Second, with respect to resting behaviors, healthy controls spent more time resting on elevated perches (i.e., 'perching on shelf' and 'hanging on iron chain') than NOD subjects. Third, respecting locomotive behavior, healthy controls presented with significantly longer durations moving along elevated perches (i.e., 'walking on the shelf' and 'walking on the iron chain') and the floor (i.e., 'quadrupedal walking on the floor' and 'standing') than NOD subjects. Fourth, as to amicable behavior, healthy controls received more amicable grooming (i.e., 'being groomed'), groomed others more (i.e., 'grooming'), and nursed their own infants more (i.e., 'nursing infant') relative to NOD subjects. Finally, as to miscellaneous behavior, healthy controls spent more time on one self-directed behavior (i.e., 'scratching by foreleg') than NOD subjects.
Behaviors of Greater Duration in NOD Subjects Relative to Healthy Controls. The behaviors of greater duration in NOD subjects relative to healthy controls fell into the following behavioral categories: ingestion, resting, locomotive, thermoregulatory, conflict, and miscellaneous (Tables 1, 3 and Fig. 2, 3). First, as to ingestion behaviors, NOD subjects more often selected safer and more remote places to feed (i.e., 'feeding while hanging') to avoid confrontation. Second, in terms of resting behaviors, NOD subjects selected safer and more remote areas to rest such as 'sitting on floor.' Third, with regard to locomotive behaviors, NOD subjects were more predominantly found walking on the skylight (i.e., 'walking on skylight') to avoid potential attack relative to healthy controls. Fourth, with respect to thermo-regulatory behavior, NOD subjects spent more time on solitary huddling or embracing a conspecific (i.e., 'huddling'/'embracing'). Finally, as to miscellaneous behaviors, NOD subjects spent more time on two particular self-directed behaviors (i.e., 'scratching by hind leg' and 'licking tail') than healthy controls.
Behavioral Similarities between NOD and SID Subjects. We compared the behaviors of the subjects from the two models of depression, NOD and SID, to determine their similarities, which fell into the following behavioral categories: ingestion, rutting and mating, resting, amicable, conflict, vigilance, and locomotive (Tables 1, 4 and Fig. 4). As to ingestion behavior, both NOD and SID subjects shared similar levels of feeding after others had finished feeding (i.e., 'licking residue from floor'), chewing food (i.e., 'chewing'), and feeding from a perched position (i.e., 'feeding while perched'). As to rutting and mating behavior, both NOD and SID subjects shared similar levels of sexual intercourse (i.e., 'copulating').
As to resting behaviors, both NOD and SID subjects shared similar levels of resting in safer and more remote places (i.e., 'hanging on iron chain,' 'hanging on skylight,' and 'hanging on ventilator') in addition to other resting behaviors (i.e., 'sitting on floor,' 'perching on shelf,' 'lying on floor,' 'lying on the board,' and 'sitting and sleeping'). As to amicable behavior, both NOD and SID subjects shared similar levels of grooming others (i.e., 'grooming'). As to conflict behavior, both NOD and SID subjects shared similar levels of conflict with others (i.e., 'driving,' 'attacking,' 'fleeing,' 'pulling foreleg,' 'biting,' and 'being attacked'), which may be related to their shared levels of vigilance behavior (i.e., 'watching company' and 'miscellaneous calling'). In regard to locomotive behavior, both NOD and SID subjects shared similar levels of moving in safer and more remote areas (i.e., 'climbing') and walking on the floor (i.e., 'quadrupedal walking on floor').
Behavioral Differences between NOD and SID Subjects. We also compared the behaviors of the subjects from the two models of depression, NOD and SID, to determine their differences (Table 5). SID subjects spent almost six times more time than NOD subjects 'drinking'. SID subjects also spent more time than NOD subjects on certain self-directed behaviors (i.e., 'rubbing paw on floor'). SID subjects also spent more time than NOD subjects 'mounting,' 'being groomed,' and 'standing.' In contrast, NOD subjects spent more time than SID subjects resting (i.e., 'hanging on window or door'). NOD subjects also spent more time than SID subjects on maternal-infant care (i.e., 'nursing infant' and 'holding infant'). NOD subjects also spent more time than SID subjects receiving aggression from others (i.e., 'being threatened'), which may have contributed to increased submissive behavior (i.e., 'presenting buttocks').
Behavioral Differences among Control, NOD, and SID Subjects. In order to better define the reliable behavioral aspects of NOD subjects, the behavioral presentations of all three groups were compared (Table 6). Notably, according to the post hoc test, both NOD and SID subjects presented similar behavioral characteristics that were significantly different from the control group, including 'sitting on floor,' 'hanging on iron chain,' 'nursing infant,' 'walking on shelf,' and 'walking on iron chain.' First, with regard to first two actions in the resting category, both NOD and SID subjects spent more time 'sitting on floor' and 'hanging on iron chain' relative to controls. Second, both NOD and SID subjects spent less time 'nursing infant' than controls. Third, both NOD and SID subjects spent less time 'walking on shelf' and 'walking on iron chain' relative to controls.
Behavioral Effects of Ketamine Administration. In order to assess prediction validity, the fast-acting anti-depressant ketamine was used to comparatively assess reversals in depressive behavior in NOD and SID subjects relative to healthy controls. With regard to the duration of behaviors, both NOD and SID subjects displayed decreased durations of 'chewing' and increased durations of 'hanging on window or door' post-ketamine administration with no such changes observed in healthy controls. With respect to the frequency of behaviors, both NOD and SID subjects displayed decreased frequencies in 'feeding while sitting' and 'chewing' post-ketamine administration with no such changes observed in healthy controls (Table 7).
Three behaviors were reduced across all three groups post-ketamine administration. The duration of 'feeding while perched' was completely eliminated in healthy controls, was reduced by a factor of 0.64 in NOD subjects, and was reduced by a factor of 0.018 in SID subjects. The frequencies of 'watching company' were reduced by factors of 0.77, 0.81, and 0.64 in healthy controls, NOD subjects, and SID subjects, respectively. The frequencies of 'scratching by foreleg' were reduced by factors of 0.61, 0.73, and 0.44, respectively.
Serum Metabonomic Analysis. PLS-DA analysis was carried out to explore the metabolic differences between NOD subjects and healthy controls. The resulting score plots of the PLS-DA model revealed that NOD subjects were significantly distinguishable from healthy controls (R 2 X 5 0.802, R 2 Y 5 0.936, Q 2 5 0.444). The values of all three parameters (R 2 X, R 2 Y, and Q 2 Y) were positive, demonstrating a robust metabolic difference between NOD subjects and healthy controls.
To identify the differential metabolites responsible for the discrimination between NOD subjects and healthy controls, the corresponding loading plots of the PLS-DA model were analyzed. This analysis resulted in the identification of 27 differential metabolites with a value of VIP . 1 and p , 0.05 (Table 8). As compared to healthy controls, NOD subjects were characterized by higher levels of silanamine, benzamide, sedoheptulose, and glycerol, in addition to lower levels of urea, leucine, serine, threonine, butanoic acid, threitol, methionine, proline, phenylalanine, asparagine, monoamidomalonic acid, 2-keto-gluconic acid, arabinofuranose, phosphoric acid, ornithine, citric acid, 1,4-butanediamine, fructose, glucose, inositol, myo-inositol, turanose, and a-D-glucopyranoside. Functional analysis of these differential metabolites demonstrated that changes in NOD subjects primarily involved disturbances in amino acid metabolism (i.e., leucine, serine, threonine, methionine, proline, phenylalanine, asparagine, and ornithine).

Discussion
The cynomolgus monkey (Macaca fascicularis) model described here better resembles the etiological development of depression and contains three key elements characteristic of a naturally-occurring model of depression: (i) observation is conducted in a natural colony-based social environment; (ii) the colony's social hierarchy produces stress-inducing social competition; and (iii) the depressed disease state naturally arises with no experimental manipulation to disturb the etiological process. This study is the first to conclusively associate the depressed behavioral phenotype in the cynomolgus monkey with significant serum metabolic changes, indicating that the depressed behavioral phenotype relates to specific underlying pathophysiological changes.
Studies on other animal species have also taken a social ethological approach to investigating depression with similar findings. Carole et al. observed the behavior of 59 working horses in their domestic environment and found 24% of the horses presented with ''withdrawn'' behavior suggestive of a depressive syndrome (e.g., atypical posture (stretched neck), unusual gaze, fixity of head and ears) 28 . As compared to the ''non-withdrawn'' horses (healthy controls) from the same stable, ''withdrawn'' horses appeared more indifferent to environmental stimuli in their home environment and reacted more emotionally to challenging situations. Carole et al. also noted that similar apathetic behavior has been observed in other domesticated animal species living under unfavorable conditions (e.g., sows, sheep, chicks, pigs). Moreover, numerous rodent studies have investigated dominance-subordination relationships to examine the effects of Bonferroni correction used to reduce type I error for multiple comparisons. *P , 0.05. social stress 29 . For example, Golden reported that C57BL/6J mice repeatedly subjected to bouts of social defeat by larger and more aggressive CD-1 mice resulted in the development of a depressivelike syndrome characterized by enduring deficits in social interactions 30 . Although these previous animal models of depression have been valuable, an ideal depression model should increase the ease with which an analogue of depression or dysthymia may be evoked and should demonstrate five types of validity: ecological, construct, etiological, face, and predictive [18][19][20] As discussed below, our social ethological model of depression more adequately addresses four of these five types of validity relative to traditional animal models of depression.
With respect to ecological validity (i.e., how closely our model resembles the natural environmental conditions in which the human disease state occurs) 31 , our naturally-occurring depression model is superior to traditional rodent models as it occurs in a realistic social ethological context. Similar to many modern human societies, our model possesses a number of interacting subjects that constitute a relatively stable socially-hierarchical society 32 . Notably, adult female cynomolgus monkeys were selected for this study, as they organize themselves into a stable social hierarchy in conjunction with one or two males 25 . Also similar to many modern human societies, scarce resources that form the basis of stress-inducing social competition (e.g., food, sex) are unevenly distributed among different social classes 33 . Thus, compared to traditional rodent models of depression, our macaque model more closely approximates the natural environmental conditions under which depression arises in humans.
With respect to construct validity (i.e., how closely our model resembles the proposed theoretical construct of the human disease Bonferroni correction used to reduce type I error for multiple comparisons. *P , 0.05. Post hoc testing found that controls presented with significant differences from NOD and SID subjects, but there were no differences found between NOD and SID subjects.

'
Bonferroni correction used to reduce type I error for multiple comparisons. *P , 0.05. www.nature.com/scientificreports SCIENTIFIC REPORTS | 5 : 9220 | DOI: 10.1038/srep09220 state), several biological constructs for the development of depression have been proposed with considerable overlap between them (e.g., HPA-axis hyperactivity, the monoamine theory, the cytokine hypothesis/macrophage theory, and neurostructural change); however, in the case of all these constructs, psychosocial stress has been proposed as a bridge between the pathophysiology of depression and its predominantly psychosocial risk factors 34 . Our naturallyoccurring depression model is superior to traditional rodent models in terms of construct validity, as our model includes psychosocial stressors in a realistic social ethological context (e.g., conflicts over feeding, sex, and resting areas) that are akin to the psychosocial stressors that underlie depression development in humans. Moreover, these psychosocial stressors have been physiologically validated, as previous studies in adult female cynomolgus monkeys meeting the identical selection criteria applied here have been shown to display elevated cortisol levels and HPA-axis dysfunction 26,35,36 .
With respect to etiological validity (i.e., how closely our model resembles the causative conditions of the human disease state) 37 , our naturally-occurring model of depression is superior to traditional rodent models, as our model does not rely upon direct experimental manipulation but naturally occurs in a social environment designed to approximate wild conditions in terms of colony size and male: female ratio. Accordingly, our group and Shively et al. have previously demonstrated that depressive behavior in captive cynomolgus monkeys is spontaneous and not induced by specific experimental manipulation 12 . In addition, similar to modern  humans, our model involves chronic mild psychosocial stressors that are randomly distributed throughout daily social life that accumulate over time. As a result, psychosocial stressors naturally arise from resource-based competition in the social context of the macaque colony. Thus, our model demonstrates superior etiological validity relative to conventional rodent models by better approximating the psychosocial stressors encountered by adult humans living in a resource-based competitive society 8 . That being said, the etiology of depression is a complex, multifactorial process that involves more than just psychosocial stressors but has also been shown to involve lifestyle (e.g., diet, sleep, and exercise) and genetic factors as well 38,39 . Our model does not account for these other key etiological factors, a fact which reduces the etiological validity of our model. With respect to face validity (i.e., how closely the model resembles the human disease state) 40 , NOD subjects presented with a slumped or collapsed body posture and maintained a low dominance rank in the social hierarchy as demonstrated by feeding last and selecting relatively remote and safe places to rest (Tables 2, 3). (Macaque colonies are socially hierarchical with higher-ranked subjects, such as the alpha-male and alpha-female, characterized by eating first and resting on elevated perches for observing their colony; NOD subjects spent significantly less time feeding first and resting on these elevated perches.) Moreover, NOD subjects received more aggression from other subjects and were less involved in reciprocal grooming behavior with other subjects. These findings are similar to those previously found in Shively's work on cynomolgus monkeys 26 and resemble the abnormal posture and low-status, relational victimization associated with depressed human patients [41][42][43] . It should be noted here that juvenile monkeys were not selected for this study, as juveniles may not possess the psychological maturity to fully experience the depressive mood state and/or physically manifest the depressive phenotype and would, therefore, not qualify to support the face validity of our naturally-occurring depression model.
In further support of our model's face validity, GC-MS-based metabolomic profiling displayed significant disturbances amino acid metabolites (i.e, leucine, serine, threonine, methionine, proline, phenylalanine, asparagine, and ornithine) as well as carbohydrate and fatty acid metabolites (e.g., glucose, fructose, citric acid, glycerol) in NOD subjects, findings which are consistent with previous serum/ plasma metabolomic analyses that have also observed significant abnormalities in these metabolic pathways in rodent models of depression and depressed human patients 24,44,45 . More specifically, significant decreases in three differential metabolites identified in NOD subjects here --leucine, glucose, and myo-inositol --were also observed in our previous NMR-based plasma metabolomic study in depressed humans 24 . However, several other amino acid, carbohydrate, and fatty acid metabolites that were found to be dysregulated in depressed human patients (e.g., valine, lactate, alanine, acetate, pyruvate, glutamine, taurine, glycine, unsaturated lipids) were not significantly affected in our NOD subjects here. Moreover, in a previous GC-MS metabolomic study in older adult patients, significant decreases in three other differential metabolites --inositol, citrate, and glycerol -were found in the plasma of depressed older adults 46 . Specifically, although our results in NOD subjects for inositol and citrate were consistent with the previous study, glycerol was upregulated in NOD subjects here but was found to be downregulated in depressed older adults. According to Paige et al.'s commentary, these changes in citrate and glycerol in depressed older patients are suggestive of a reduction in fatty acid oxidation; consequently, these alterations were accompanied by changes in glycerate and medium-chain fatty acids (e.g., myristate, palmitate, stearate, oleate). However, we did not observe changes in glycerate or fatty acid levels in the NOD subjects here. Moreover, several differential metabolites observed in NOD subjects here (e.g., silanamine, benzamide, sedoheptulose, urea, butanoic acid, threitol, monoamidomalonic acid, 2-keto-gluconic acid, arabinofuranose, phosphoric acid, 1,4-butanediamine, turanose, and a-D-glucopyranoside) have not been previously associated with plasma/serum metabolite profiles in human depression, which may be partly attributable to inter-species differences in dietary intake, commensal microbiota composition, and metabolic processing. These combined findings suggest some significant similarities and differences between the NOD metabolic phenotype and the human depressive metabolic phenotype.
With regard to predictive validity (i.e., the model's selectivity in response to specific compounds, such as antipsychotics or antidepressants) 23,47,48 , the fast-acting anti-depressant ketamine was used to comparatively assess reversals in depressive behavior in NOD subjects relative to healthy controls. NOD subjects displayed decreased durations of 'chewing' and increased durations of 'hanging on window or door' and decreased frequencies in 'feeding while sitting' and 'chewing' post-ketamine administration with no such changes observed in healthy controls. Therefore, unfortunately, ketamine administration did not produce reversals in depressive behavior in NOD subjects. As field observations were limited to only two subjects over a course of seven days, future studies should include a larger sample set of NOD subjects treated with conventional antidepressant therapy observed over a longer duration.
There are several notable limitations to this study. First, a selfcontrolled study design that allowed for the direct comparisons before and after social isolation as well as before and after ketamine administration would have better controlled for individual differences. Future studies involving interventions in this species should seek to self-control to account for individual variations. Second, this study failed to provide insight as to why ketamine administration did not produce expected reversals in depressive behavior in NOD subjects; thus, future studies should apply conventional antidepressant therapy observed over a longer duration in order to gather more evidence to support the predictive validity of this model. Third, our metabolic data was captured at one specific time point, while behavioral observation was conducted over a longer period of time, making the performance of correlation/regression analyses to examine potential relationships between behavioral and metabolic measurements impossible. A future study will have to monitor both behavioral and serum metabolite changes in tandem over a period of time in order to robustly perform these analyses.

Conclusion
Here, we describe and model a naturally-occurring depressive state in a non-human primate species, the cynomolgus monkey (Macaca fascicularis), in a realistic social ethological context and associated the depressed behavioral phenotype with significant serum metabolic perturbations. In accordance with rodent and human studies, the metabolic phenotype of the NOD subjects and healthy controls significantly differed. This model's stable depression phenotype demonstrates superior validity to traditional rodent models of depression by better approximating the psychosocial stressors encountered by adult humans living in a resource-based competitive society and can aid further investigation into the mechanisms underlying depression.