香京julia种子在线播放

Subjects were eight captive-born adult male titi monkeys (Callicebus cupreus) housed at the California National Primate Research Center (CNPRC) in Davis, CA. Subjects were a mean age of 7.7 years old (median 7.0, range 4.0–12.8), and were living with their female pair mates for a mean of 2.5 years (median 1.7, range 0.7–9.9). All subjects were parents of offspring living in the cage. Animals were fed twice daily (0,830 and 1,330 h) a diet consisting of New World monkey chow, rice cereal, banana, apples, raisins, and baby carrots and water was available ad libitum. Further details of husbandry and training are available elsewhere (Tardif et al., 2006).

Functional imaging was used to examine how males viewing their pair mate in close proximity to a stranger male would differ in their regional cerebral glucose metabolism compared to viewing a stranger male next to a stranger female in adjacent cages. Subjects, pair mates and young offspring (less than 1 year old) were relocated to a metabolism room 48 h prior to their positron emission tomography (PET) scan. As in our previous PET studies (Bales et al., 2007; Hinde et al., 2016; Maninger et al., 2017), animals were relocated prior to the scan in order to reduce the possible effect of novel housing on brain metabolism. Animals were fasted 6–12 h prior to the scan, with water available throughout the pre-scan period. On the day of the scan, all of the animals were caught and removed from the cage. The subject was manually restrained while he received a bolus injection of [¹⁸F]-fluorodeoxyglucose (FDG, PETNET Solutions, Sacramento, CA, up to 2 mCi/kg IV, administered in a volume of <2 ml) into the saphenous vein.

Following the FDG injection, the male was put back in his cage alone (since his pair mate and offspring were removed) for the 30 min conscious uptake period, where he had visual access to another cage that housed two animals separated by a wire mesh. In the jealousy condition, the two animals in the viewing cage were the subject's female pair mate and a stranger male (Figure 1). The stranger was a male who was unfamiliar to the subjects. Viewing a stranger male adjacent to his female pair mate was designed to challenge the pair bond and induce “jealousy” in the male subjects. In the control condition, there was a stranger female and a stranger male monkey (note that this was a different animal from the jealousy condition) in the viewing cage. Because titi monkeys are territorial animals when paired and can show aggression to opposite-sex strangers (Fernandez-Duque et al., 2013), the male and female in the viewing cage were separated by a wire mesh in order to prevent any physical aggression (and potential wounding) between the unfamiliar animals. This was important because there were no humans in the room during the FDG uptake period to stop any fights. During the control condition, the female pair mate was moved out of the testing room. The offspring were moved out of the testing room for both the control and jealousy conditions. A camera was placed at the side of the subject's cage and the male was filmed during the uptake period for 30 min, while all of the humans left the room. Each of the eight males experienced both the jealousy and control conditions on separate days; there was a mean of 5.2 weeks (range 3–6.3) between testing days. The order of conditions was counter-balanced, such that four males experienced the jealousy condition before the control condition and four males experienced the control condition before the jealousy condition.

After the FDG uptake period, subjects were anesthetized with ketamine (25 mg/kg IM) and administered medetomidine (0.05 mg/kg IM). After the subject was sedated, a 1 ml blood sample was collected from the femoral vein and put into a heparin-containing tube, and a sample of cerebrospinal fluid (CSF) was collected and put on ice. In order to ensure that hormonal outcomes were not influenced by the considerable disturbances involved prior to collection of blood samples, care was taken to ensure that the time of day was comparable for a given subject tested in each condition. Testing started between 0,800 and 0,839 h for the first subject, and the second subject was tested approximately 1.5–2 h later. Cortisol concentrations for males tested in the first (earlier) group averaged 94.1 ± 5.8 μg/dl, while males tested in the later group averaged 55.7 ± 7.2 μg/dl. While time of day was correlated with cortisol concentrations (r = –0.575), this effect of circadian rhythm was accounted for in the design of the study by carrying out both of an individual's scans (control and jealousy) in the same time grouping (early or late). For example, both of 32878's scans were carried out in the early group, and both of 31716's scans were carried out in the late group. We also measured the duration of time between capture of the subject following FDG uptake until bloodand CSF sample collection (i.e., “disturbance time”). Disturbance time for blood samples was a mean of 5.22 min (median 4.80, range 1.72–16.15) after capture and sedation (raw data in Appendix 1). This disturbance time was not statistically correlated with plasma cortisol concentrations (r = 0.102). CSF samples were collected a mean of 9.50 min (median 9.88, range 5.75–13.13) after capture and sedation (raw data in Appendix 1; no statistical analysis on CSF data was carried out).

Following collection of blood and CSF samples, an endotracheal tube was placed and a catheter was placed in the saphenous vein in order to administer IV fluids (lactated ringers solution, 10 ml/kg/h). Atipamazole was used to reverse medetomidine, and anesthesia was maintained with isoflurane (1–2%), while the male was positioned on the scanner bed feet first and the brain of the animal was positioned in the center of the scanner. PET imaging was performed on a microPET P4 scanner (Siemens Preclinical Solutions, Knoxville, TN). Image acquisition began a mean of 69.49 (SD ± 7.52) minutes post-FDG administration, and static PET scans were acquired for 60 min. Anesthesia was maintained throughout the scan. Animals were housed in metabolism cages for 24 h after scanning, at which time radiation was decayed to background levels and animals were returned to their home cages.

Structural magnetic resonance imaging (MRI) scans were conducted in a GE Signa LX 9.1 scanner (General Electric Corporation, Milwaukee, WI) with a 1.5 T field strength and a 3″ surface coil. Each male was fasted 8–12 h before the procedure. At the start of the procedure, the male was sedated with ketamine (10 mg/kg IM) and medazolam (0.1 mg/kg IM), and an endotracheal tube was placed. A catheter was also placed in the saphenous vein in order to administer fluids as necessary. Anesthesia was maintained with isoflurane (1–2%) while the male was positioned in the MRI scanner. Each scan lasted approximately 20 min and consisted of a 3D SPGR pulse sequence in a coronal plane. Images of the entire brain were collected using the following parameters: echo time TE = 7.9 ms, repetition time TR = 22.0 ms, flip angle = 30.0°, field of view = 8 cm, number of excitations = 3, matrix = 256 × 256, and slice thickness = 1 mm. As a precautionary measure, the male's EtCO2, oxygen saturation, heart rate and blood pressure were monitored throughout.

We determined which regions of interest (ROIs) to quantify based on three groups of studies: rodent studies of aggression (lateral septum, medial amygdala, posterior cingulate cortex, and anterior hypothalamus), the Rilling rhesus monkey study of consortship (superior temporal cortex, insular cortex), and human studies of social pain and jealousy (anterior cingulate cortex, nucleus accumbens, ventral pallidum, caudate, and putamen) (Figure 2).

ROI structures were individually drawn on each subject's MRI image, for both left and right hemispheres, using landmarks as a guide, in Siemen's Inveon Research Workplace software (IRW, Siemens Healthcare, USA). ROIs were drawn prior to co-registrations with the PET image, so they were drawn blind with regard to PET image/FDG uptake and to experimental condition. The same ROIs were used for both the jealousy and control conditions. Static PET images were reconstructed with a 3DRP reconstruction protocol. MRI images were co-registered with PET scan images using the automatic rigid registration algorithm in IRW and checked visually for registration accuracy. Mean activity for the PET images were determined in IRW by applying ROIs defined on the MRI images to the PET images. Data are presented in proportions of whole brain activity, which was calculated by dividing the mean activity in the ROI (in units of microcuries per cubic centimeter) by mean activity of whole brain ROI.

Males were filmed during the 30 min FDG uptake period. After all of the PET scans were completed, the videos were scored by a trained coder (T.S.) who was blind to experimental condition and validated against previous scoring done in the laboratory. Videos were scored on Behavior Tracker 1.5 (behaviortracker.com) for duration of the behaviors in the ethogram (see Table 1). Behaviors included lip smacking (an affiliative behavior), tail lashing (an arousal behavior), arching (an arousal behavior), as well as looking across at the stimulus cage, locomotion, chewing, drinking, and “off camera.” Data analyses were performed on the total duration of each behavior (i.e., the absolute length of time the behavior was performed).

Blood and CSF samples were collected after animals were sedated for the PET scan following the FDG uptake period, and placed on ice. Blood samples in heparin-containing tubes were centrifuged at 3,000 RPM for 15 min at 4°C. Plasma was aliquoted, and plasma and CSF samples were stored at −70°C until assay. CSF samples were assayed for oxytocin (OT) and vasopressin (AVP). Plasma samples were assayed for testosterone, cortisol, OT, and AVP. While the veterinarians collected as many CSF samples as possible, often they were unable to get a sample due to the small size of the animals (male subjects weighed a mean of 1.3 kg, median 1.2, range 1.1–1.6). Therefore, we present CSF values in Appendix 1, but did not have an adequate sample size to analyze them statistically.

AVP and OT concentrations were estimated in duplicate using commercial enzyme immunoassay kits (Enzo Life Sciences, Farmingdale, NY) previously validated for titi monkeys. Assay sensitivity was 2.34 pg/ml for AVP and 15.55 pg/ml for OT. Intra- and inter-assay coefficients of variation (CV) were 3.36 and 14.34% respectively for AVP, and 10.62 and 12.78%, respectively for OT. Plasma cortisol and testosterone concentrations were estimated in duplicate using commercial radioimmunoassay kits (Siemens Healthcare, Malvern, PA). Prior to cortisol assay, plasma samples were diluted 1:4 in PBS gel buffer. Cortisol assay procedures were modified with the addition of 0.5 and 2.35 μg/dl concentrations of standards along with the provided range of 1.0–49 μg/dl. Assay sensitivity was 0.261 μg/dl. Intra- and inter-assay CV were 3.20 and 6.26%, respectively. Prior to testosterone assay, plasma samples were diluted 1:2 in PBS gel buffer. Testosterone assay procedures were modified with the addition of 57 and 197.5 ng/dl concentrations of standards along with the provided range of 24–1,667 ng/dl. Testosterone assay sensitivity was 4.58 ng/dl. All samples were run in the same assay and intra-assay CV was 1.02%.

All models were fitted in a fully Bayesian multivariate multilevel framework for several reasons. First, due to our small sample size and large number of outcomes, multivariate models could not be estimated with least squares or maximum likelihood methods. Bayesian multivariate methods allowed for estimation of hypothesized regions of interest that included numerous correlated outcomes in one model. In addition, Bayesian multilevel methods fully account for uncertainty across levels of hierarchically structured data (McElreath, 2015), which was important due to our within-subjects design. Third, Bayesian methods allow for incorporating prior information into the model which improves precision of the parameter estimates (Gelman et al., 2008; Kruschke and Vanpaemel, 2015; see Supplementary Material for model details). Finally, parameter estimates have probabilistic interpretations, which allows for estimating the probability of a positive or negative experimental effect (Zucker et al., 1997; Lee, 2011).

In total, we fit five multivariate multilevel models. The first three models were based on hypothesized brain regions previously implicated as modulating jealousy-like behavior in different species: (1) mate-guarding in rodents; (2) jealousy or social pain in humans; and (3) bilateral regions from the rhesus monkey study. The next two multivariate multilevel models assessed hormonal and behavioral differences. The final models were exploratory, in that outcomes were determined from our results. These final models were not multilevel, but multivariate examining correlations between look duration and hormones as well as FDG uptake in ROIs (for the jealousy condition only). For these models, we standardized the predictors and response variables so that the estimates were on r scale (correlation coefficient). All model based estimates are provided in Tables 2–6, whereas the raw means and standard deviations are provided with the model checks (Appendix 2).

Variance was partitioned into two components (Gelman and Hill, 2007): (1) the variance (σu2) between subjects (i.e., varying intercepts); and (2) the residual variance (σe2). As a measure of residual variance explained by subject, we computed intra-class correlation coefficients (ICC) (Quene and Van Den Bergh, 2004). Each multivariate model included one fixed effect, which provided a contrast from the control group. The parameter estimates were summarized with 95% credibility intervals (CrI) that, by definition, have a 95% probability of containing the true parameter (Morey et al., 2015). Bayesian methods provide probabilistic estimates that allow for explicit statements about likely values for the true treatment effect. We thus computed posterior probabilities of a positive or negative effect (Pr(b > 0) and Pr(b < 0)) (Gelman, 2013; Greenland and Poole, 2013). This is not possible with classical methods, since parameters are assumed to be fixed point estimates (probability distributions cannot exist). Posterior probabilities >80% are reported in the results section, but all estimates are provided in tables. The 80% figure was chosen merely as a convenient figure in order to simplify the reporting of the results. Finally, an effect size parameter (δ_T) was obtained from dividing the estimate (b) by the square root of the variance components summed (σu 2+σe2) (Hedges, 2007). Interpretation of (δ_T) follows Cohen's d (Cohen, 2009; small = 0.2, medium = 0.5, large = 0.8).

The jealousy condition for one male was not usable for technical reasons, thus the final sample size was 8 control scans and 7 jealousy scans, for 8 total subjects. Due to the complexity of the multivariate multilevel models, we examined fit with posterior predictive checks and posterior predictive p-values (Gelman et al., 1996; de la Horra and Rodriguea-Bernal, 1999). Here, the fitted model was used to simulate data, from which a properly specified model will provide replications that look like the observed data and non-extreme p-values (0.95 > p-value > 0.05). For most models, the posterior predictive p-values indicated that model fit was adequate (Appendix 2).

All computation was done in R (Team, 2016). The package brms (Buerkner, 2015), a front end to the probabilistic programming language Stan (Stan_Development_Team, 2015), was used to fit all regression models (all R script and data are available upon request).

The use of Bayesian statistics remains relatively uncommon. This may be seen as a limitation when comparing our results to the extant literature. Indeed, our use of Bayesian methods has different goals than typically pursued: we did not focus on rejecting a null hypothesis. Instead, our analysis sought to quantify the most probable values for the “true” effect of jealousy on regional cerebral glucose metabolism, hormones and behavior in titi monkeys. This is not possible with classical methods (e.g., ANOVA) in which evidential quantities (e.g., p-values) are in reference to counterfactual sampling procedures. When inferring from our results, the posterior probabilities can be directly interpreted as probabilities (how probability is used in everyday language). The present approach does not include thresholds (i.e., cut-offs, but of course a probability of 99% provides stronger evidence than 85%, assuming equal prior odds). A meaningful probability can be determined in light of theory, past research, the quality of this study (including limitations), and the reported results (Harrell and Shih, 2001; Gelman, 2013; Greenland and Poole, 2013; Gelman et al., 2014; Kruschke, 2014).

Our first multivariate multilevel model simultaneously estimated areas implicated by rodent studies as modulating mate-guarding behavior: the lateral septum (LS), anterior hypothalamus (AH), posterior cingulate cortex (PCC), and medial amygdala (MeA). The probability of a positive effect of jealousy condition on FDG uptake in the right LS was 93% (b = 0.04, CrI = [−0.02–0.10], δ_T = 0.55; Figure 3A) and, similarly, 99% in the left PCC (b = 0.04, CrI = [0.01–0.07], δ_T = 1.02; Figure 3B), while there was some evidence for reduced uptake in the right MeA (Pr(b < 0) = 85%, b = −0.03, CrI = [−0.10–0.03], δ_T = −0.40; Figure 3A). The posterior probabilities for the other comparisons were below 80% and are reported in Table 2. Notably, the amount of variation explained by subject across outcomes ranged from 9% (left AH: ICC = 0.09, CrI = [0.00–0.44]) to 42% (right LS: ICC = 0.42, CrI = [0.004–0.87]).

Our second multivariate multilevel model examined several cortical areas that were shown to be associated with jealousy or social pain in human studies [anterior cingulate cortex (AC), caudate (Ca), putamen (P), nucleus accumbens (NAcc), and ventral pallidum (VP)]. The probability of a positive effect of jealousy condition on FDG uptake in the left AC was 96% (b = 0.05, CrI = [−0.01–0.10], δ_T = 0.79)], while there was some evidence for reduced uptake in the right Ca (Pr(b > 0) = 90%, b = 0.04, CrI = [−0.03–0.10], δ_T = 0.41), right VP (Pr(b > 0) = 86%, b = 0.05, CrI = [–0.03–0.13], δ_T = 0.38), and the left VP (Pr(b> 0) = 88%, b = 0.04, CrI = [−0.03–0.13], δ_T = 0.38). According to the posterior predictive checks, model fit was adequate. The other comparisons are provided in Table 3. The amount of residual variation explained by subject across outcomes ranged from 13% (left VP: ICC = 0.13, CrI = [0.00–0.56]) to 53% (right Ca: ICC = 0.53, CrI = [0.02–0.89]).

The third multivariate multilevel model estimated bilateral ROIs from the rhesus monkey study [insular cortex (IC) and superior temporal sulcus (ST)]. This model produced negligible probabilities for an effect of jealousy, and residual variation attributed to subjects was minimal (all instances <5%). Importantly, posterior predictive check indicated a misfit between the observed and the model implied standard deviations (Appendix 2; Table 4). Assuming equal variances was problematic, but unfortunately a heteroskedastic model could not be fit (due to an already complex model).

Our multivariate multilevel model estimated plasma hormone concentrations of oxytocin (OT), vasopressin (AVP), cortisol, and testosterone in the jealousy condition compared to the control condition. There was a positive effect (Pr(b > 0) = 93%) of jealousy condition on plasma testosterone (b = 190.17, CrI = [−85.38–440.42], δ_T = 0.48) as well as similar evidence (Pr(b > 0) = 92%) for a positive effect on plasma cortisol (b = 10.13, CrI = [−4.83–24.48], δ_T = 0.40). Posterior predictive checks indicated that the model adequately described the observed data. The amount of residual variation explained by subjects ranged from 17% (AVP: ICC = 0.17, CrI = [0.00–0.67]) to 63% (cortisol: ICC = 0.62, CrI = [0.03–0.92]). Plasma OT and AVP concentrations had lower probabilities (65 and 74%, respectively) for differences between conditions (Table 5; also presented are model estimates and confidence intervals).

The multivariate multilevel model estimated differences in total durations of behavior between the control and jealousy conditions (for ethogram see Table 1). Due to excessive zeroes, drinking and time off camera were not analyzed. There was some evidence for a positive effect (Pr(b > 0) = 90%) of jealousy condition on lip smacking duration (b = 7.60, CrI = [−4.19–19.75], δ_T = 0.55). The residual variance explained by subject ranged from 14% (chewing: ICC = 0.14, CrI = [0–0.59]) to 99% (tail lashing: ICC = 0.99, CrI = [0.97–1.0]). All estimates are reported in Table 6.

Durations of behaviors were standardized prior to analysis, resulting in correlations (r). Look duration via a single-level multivariate model was positively correlated with cortisol (b = 0.63, CrI = [−0.07–0.98], Pr(b > 0) = 97%). The correlation with testosterone was also positive, but the interval was very wide (b = 0.31, CrI = [−0.67–0.94], Pr(b > 0) = 78%). Using a single-level multivariate model to investigate associations between behavior and brain region of interest, we found that look across duration had a probability of a positive correlation on FDG uptake in the right LS of 92% (b = 0.53, CrI = [−0.32–0.97]; Figure 4A). There were substantial negative correlations with look duration in the left PCC (b = −0.46, CrI = [−0.97–0.35], Pr(b < 0) = 90%; Figure 4B) and the right MeA (b = −0.74, CrI = [−0.99 to −0.22], Pr(b < 0) = 99%; Figure 4C).

The probability of a negative correlation between plasma testosterone and FDG uptake in the right MeA was 96% (b = −0.57, CrI = [−0.97–0.07]), 48% for the right LS (b = −0.03, CrI = [−0.84–0.80]), and 79% for the left PCC (b = −0.57, CrI = [−0.93, 0.60]). There were negative correlations between cortisol concentrations and FDG uptake in the left PCC (b = −0.19, CrI = [−0.91, 0.70], Pr(b < 0) = 72%), and right MeA (b = −0.54, CrI = [−0.97, 0.22], Pr(b < 0) = 93%). There was a positive correlation between cortisol concentrations and FDG uptake in the right LS (b = 0.31, CrI = [−0.66, 0.94], Pr(b > 0) = 81%), but the interval was very wide.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.