To ensure variability in the anxious temperament of the animals that were used, we screened 116 periadolescent monkeys selecting 36 animals with the most stable high (n
12, 9 female and 3 male), middle (n
12, 8 female and 4 male) and low (n
12, 9 female and 3 male) levels of threat-induced BI or freezing behavior (average age+SEM at screening
2.3+.09 yrs.; range
1.5–3.4 yrs; puberty onset in rhesus monkeys is approximately 3 yrs). Although more females than males were selected for further study, the overall stability in freezing behavior, as measured by the change in the amount of freezing between the two screening days, did not significantly differ between males and females (t(114)
.112). For initial selection and categorization, freezing was assessed by separating the monkey from its cage mate and relocating it to a test cage in which for 10 minutes it was exposed to the potential threat of a human intruder presenting her facial profile to the monkey [no eye contact condition (NEC) of the human intruder paradigm] (). Although freezing in response to NEC is considered adaptive, excessive freezing is analogous to extreme childhood BI 
. In addition to measuring freezing behavior, individual differences in spontaneous coo vocalizations were assessed for 10 minutes immediately prior to the NEC condition while separating the monkey from its cage mate and relocating it to the test cage [alone condition (ALN) of the human intruder paradigm] (). The ALN condition typically elicits separation induced coo vocalizations, which can be likened to calls for help. Less coo calling is associated with increased anxiety and increased amygdala activity 
. In addition to the behavioral measurements, each animal had blood drawn on two separate occasions following exposure to restraint stress, to assess stress-induced plasma cortisol levels.
On an average of 4.3 months later, brain activity was measured on 2 separate occasions with FDG-PET in 35 of the 36 monkeys during 30-minutes of exposure to the NEC and ALN conditions (one monkey that displayed high levels of freezing during screening was not scanned because she refused to leave her cage). The 30-minute period of stress exposure was selected because FDG enters active brain cells during a 30-minute uptake period, and remains stably detectable within these regions for prolonged periods (due to its 110-minute half-life) 
. This time course also allows for later anesthesia administration and imaging of the brain activity that occurred during the preceding period of stress exposure (). During the period of FDG uptake, NEC-induced freezing and ALN-induced cooing were assessed as increased BI, and decreased spontaneous vocalizations that are characteristic features of children with anxious temperament. Blood was collected at the end of these conditions to assess stress-induced cortisol levels. To create a composite measure of anxious temperament, we first z-scored each measure of stress responsivity (increased freezing during NEC, increased cortisol levels in response to NEC, decreased cooing during ALN, and increased cortisol levels in response to ALN; controlling for any age effects across all animals) across subjects and then computed the mean of the four z-scored measures for each subject. To understand the extent of the relation between the individual measures comprising the composite score, correlations between the individual variables (after being age-residualized and z-scored) were performed. Freezing during NEC and cooing during ALN were marginally negatively associated (r(33)
.052); all other pairs of variables were not significantly correlated (r's(33)<.253, p's>.143). Because the distribution of the composite measurement of anxious temperament did not differ from a normal distribution (Kolmogorov-Smirnov one-sample test against a normal distribution (KS)(n
.861) and revealed no discernable subgroups, we treated anxious temperament as a continuous measure in all subsequent analyses.
Since anxious temperament in children is relatively stable, we examined the extent to which this composite measure of stress responsivity was correlated with the 35 monkeys' composite measure of stress-related behavior and physiology that was collected 4.3 months earlier during their initial assessment 
. Consistent with the trait-like qualities of anxious temperament, results demonstrated that this composite measure was stable (r(33)
.533, p<.001) (). In a subsequent study involving 24 of these animals, the same parameters were assessed at 3.8 years of age or 1.5 years later. Results revealed that individual differences in stress responsivity remained stable as the animals matured (r(22)
.027). We observed no gender differences (t's(33)<1.045, p>.304) or evidence of non-normal distributions (KS's(n
35)<.923, p's>.361) within any of the composite measures of anxious temperament. These data demonstrate that a composite assessment of stress-related behavioral and physiological measurements reflecting anxious temperament is stable over development in non-human primates.
Figure 2 Anxious temperament, defined as increased freezing during NEC, decreased cooing during ALN and increased stress-induced cortisol, assessed during screening was significantly correlated with a similar measure taken 4-months later during FDG-PET testing. (more ...)
To assess the relation between individual differences in anxious temperament and brain metabolism, we performed, separately for the NEC and ALN conditions, voxelwise correlations between monkeys' anxious temperament scores and brain metabolism, while controlling for age. To account for anatomical differences and any small errors in inter-subject registration that could masquerade as stable functional effects, the voxelwise analyses were co-varied for the probability of gray-matter at each voxel in the brain 
. Results represent the relation between anxious temperament and brain metabolism that cannot be explained by gross anatomical differences or registration error. In the NEC condition, significant (p<.05 two-tailed, multi-FDR corrected) correlations between anxious temperament and brain activity were detected in the left and right amygdala, left hippocampus, and the left brain stem (pontine nuclei region) (, pink; ) 
. Analyses of the brain activity during the ALN condition revealed that anxious temperament was significantly (p<.05 two-tailed, multi-FDR corrected) related to activity in the right amygdala, and right hippocampus (, blue; ). Although some regions only reached statistical significance in one hemisphere, tests of hemispheric asymmetry revealed no significant differences (t's(32)<1.719, p's>.095), suggesting that hemispheric differences were an artifact of statistical thresholding.
Figure 3 a) Significant (p<.05, two-tailed multi-FDR corrected) relationships between regions of the amygdala (outlined in blue) and anxious temperament (Time 2) during the ALN (cyan) and NEC (pink) conditions are overlaid on coronal (green-border) and (more ...)
Brain Areas During Stressful Contexts Predict Anxious Temperament
We next used a logical AND conjunction analysis to identify the brain regions that across the different stressful contexts predicted individual differences in anxious temperament. Here we assessed the overlap between the brain regions that were correlated with anxious temperament in NEC with those that were correlated with anxious temperament in ALN. Analysis of the correlations in the NEC and ALN conditions revealed that the same regions of the right amygdala were significantly correlated with anxious temperament (p<.05, two-tailed multi-FDR corrected in both NEC and ALN) (, purple; ) 
. Thus, activity in a region of the right amygdala assessed in 2 different stressful contexts is consistently related to behavioral and physiological measures of stress responsivity assessed in those contexts. Notably, hierarchical linear regressions revealed the composite measure of anxious temperament to explain significant variance in amygdala activation beyond any one individual measure of stress (R2
-Change's>.191, F-Change's(1,32)>10.742, p's<.003 Furthermore, there was only one voxel within the amygdala (as defined by our whole-amygdala ROI) that showed a significantly greater correlation with an individual measure of stress (cooing) than the composite measure of anxious temperament (R2-Chage
.031; other p's >.05). We also found that individual differences in metabolic activity in this overlapping amygdala region assessed during NEC or ALN were positively correlated with individual differences in the screening measures of anxious temperament that were determined 4.3 months prior to the FDG studies (early temperament vs. later NEC amygdala activity, r(33)
.049; early temperament vs. later ALN amygdala activity, r(33)
.028). This result suggests that the relation between anxious temperament and amygdala activation reflects the stable components of anxious temperament.
Although we predicted that amygdala activation is fundamental to anxious temperament, we were also interested in identifying the brain systems that interact with the amygdala to produce the behavioral and physiological changes. We further investigated additional regions that consistently demonstrated significant (p<.005, two-tailed, uncorrected) correlations between brain activity and anxious temperament in both the NEC and ALN conditions using a logical AND conjunction analysis 
. This analysis revealed overlapping regions in bilateral amygdala, bilateral Bed Nucleus of Stria Terminalis (BNST), bilateral hippocampus, and periaqueductal gray (PAG) to be significantly related to anxious temperament (, purple; Table S1
). Follow-up paired t-tests comparing the stressful (NEC and ALN) conditions to the home-cage conditions (H-ALN and H-CM) revealed that each of these regions, including the amygdala, had significantly greater activity during the stressful conditions (t's(34)>2.223, p's<.033). These results suggest that anxious temperament involves a neural circuit that is consistent with that previously characterized in preclinical mechanistic studies of stress and anxiety 
Figure 4 Regions where anxious temperament (Time 2) was significantly (p<.005, two-tailed uncorrected, two-condition logical AND conjunction) related to metabolism in both of the two stressful conditions [Alone (ALN) and No Eye Contact (NEC)] (more ...)
Because we were interested in the stable neural circuitry of anxious temperament that we hypothesized would persist in non stressful environments, we examined whether individuals with anxious temperament would still display increased amygdala activation when assessed in the security of their home cage. Thus, the relation between anxious temperament and brain activation was assessed while the monkeys were in their home cages with their cage mates (H-CM) and when in their home cages without their cage mates (H-ALN). The H-ALN condition was studied to assess brain activity in a familiar environment in the absence of social interaction. Overall, activity from the overlapping amygdala region defined from the NEC and ALN conditions (, purple) was significantly less in the home cage conditions compared to the stressful conditions (t(34)
2.626, p's <.013). To investigate the stability of the relationship between brain activation and anxious temperament across both stressful and secure contexts, we searched for regions where anxious temperament was significantly correlated with brain metabolism in each of the four conditions (NEC, ALN, H-CM, H-ALN). This logical AND conjunction analysis revealed that anxious temperament was significantly (p<.005 two-tailed, uncorrected) associated with activity in bilateral amygdala, bilateral hippocampus, and PAG in all four conditions (, orange; ). Because cortisol samples were not always taken at the same time of day, we verified these results while statistically controlling for the time of the cortisol sampling. Controlling for time of day did not affect the highly significant (r's(31)>.493, p's<.004) relationships between anxious temperament and brain activity within the amygdala and the identified stress network. Importantly, the entire overlapping amygdala region identified from the NEC and ALN conditions (, purple) was significantly (p<.05 two-tailed, small-volume FDR-corrected within the amygdala) related to anxious temperament in the H-CM and H-ALN conditions. Moreover, the correlation between anxious temperament and this overlapping amygdala region was not significantly higher in the stressful conditions when compared to the secure conditions (t's(32)<1.313, p's>.19). We used an additional approach to further examine the relationship between anxious temperament and context dependent amygdala metabolism. Specifically, we correlated our measure of anxious temperament with the change in brain metabolism between each pair of conditions, as well as with the variability across all conditions within the amygdala and the identified stress network. These results revealed no significant correlations between anxious temperament and either the change between conditions (|r|'s(33)<.175, p's>.316) or the variability across conditions (|r|'s(33)<.179, p's>.304) within any of the regions tested. These results suggest our measure of anxious temperament in relation to individual differences in amygdala activity is not state related, but is stable across contexts. Overall, these results demonstrate that the relationship between anxious temperament and the amygdala, along with an extended stress network, is not restricted to the NEC and ALN conditions but extends to secure, non threatening, settings.
Brain Areas Across Stressful and Secure Contexts Predict Anxious Temperament