Age-Associated Deviations of Amygdala Functional Connectivity in Youths With Psychosis Spectrum Disorders: Relevance to Psychotic Symptoms

The final neuroimaging data set consisted of 1,062 participants 10–25 years old (typically developing control subjects, N=622; individuals with psychosis spectrum disorders, N=194; individuals with other psychopathology, N=246) from four different samples. Three data sets were acquired at the University of Pittsburgh and one at the University of Pennsylvania (28, 29). Information on the study participants is presented in Table 1. One data set was from a longitudinal sample, and the other three were cross-sectional. (Details on participant recruitment and inclusion and exclusion criteria are provided in the online supplement; see the Supplemental Methods section and Figure S1.)

Positive and negative symptoms were measured by summing the relevant Structured Interview for Prodromal Syndromes/PRIME Screen–Revised responses (0=definitely disagree, 1=somewhat disagree, 2=slightly disagree, 3=not sure, 4=slightly agree, 5=somewhat agree, 6=definitely agree). Table S1 in the online supplement lists the included questions.

For all samples, scanning data were acquired using Siemens 3-T Tim Trio scanners. Resting-state data were collected using an echo-planar sequence sensitive to blood-oxygen-level-dependent contrast (T₂*). A magnetization-prepared rapid gradient-echo sequence (MPRAGE) was acquired to measure brain structure and for alignment of the resting-state functional MR images. Table S2 in the online supplement includes scan instructions and parameters; details of resting-state fMRI data processing are provided in the Supplemental Methods section.

Developmental effects were observed for functional connectivity between the centromedial amygdala and 19 clusters (Table 2; see also Figure S2 in the online supplement). These clusters included the following bilateral brain regions: posterior cingulate, insula, parahippocampal cortex, and precentral gyrus/frontal eye fields. Significant clusters were also observed for the left ventrolateral prefrontal cortex, left caudate, left dorsolateral prefrontal cortex, right thalamus, and right postcentral gyrus. We observed age-related decreases in connectivity strength between the centromedial amygdala and all clusters, with children exhibiting positive centromedial amygdala connectivity (mean=0.19 at age 10) and adults exhibiting near-zero levels of connectivity (mean=0.04 at age 25).

Developmental effects were also observed for functional connectivity between the basolateral amygdala and one cluster, which encompassed the left uncus (Table 2). In this case, children exhibited positive centromedial amygdala connectivity (mean=0.27 at age 10), and adults exhibited positive connectivity as well, albeit to a lesser extent (mean=0.17 at age 25).

For all significant clusters, the inverse form of age was the best fit. All developmental effects remained significant when motion covariates (average framewise displacement) and MRI software version were included in the model and when high-motion subjects were excluded from the analysis (see Tables S3–S4 in the online supplement). Strikingly, age-associated changes were consistent across sites (see Figure S3 in the online supplement).

We also confirmed that site effects were appropriately accounted for by including the measure as a covariate. After regressing out the effects of site in each region, we then plotted the residuals (see Figure S4 in the online supplement). The residuals all clustered around zero, which provides evidence that we were able to effectively account for site in our model. Furthermore, when we conducted a mixed-model linear regression to compare residuals between sites, there were no significant differences between sites for any of the regions of interest (all χ²=0, p=1), further solidifying evidence that region-of-interest values were similar across sites once site was included in the model.

After correcting for multiple comparisons, significant age-associated deviations (inverse age-by-group interactions) were observed in the psychosis spectrum group for connectivity between the centromedial amygdala and six clusters in the following brain regions: left ventrolateral prefrontal cortex, right thalamus, left dorsolateral prefrontal cortex, left caudate, left putamen, and right middle occipital gyrus (Figure 1A). In five clusters (ventrolateral prefrontal cortex, thalamus, caudate, putamen, and middle occipital gyrus), slope comparison analyses revealed that the typically developing control group exhibited significant age-related decreases with increasing age, while the psychosis spectrum group failed to exhibit significant age-associated changes (see Table S5 in the online supplement). Contrasts revealed that during late childhood the psychosis spectrum group, in comparison to the typically developing control group, exhibited significantly lower connectivity in the following pairs: centromedial amygdala–ventrolateral prefrontal, centromedial amygdala–putamen, and centromedial amygdala–caudate. In adulthood, the psychosis spectrum group exhibited higher connectivity in comparison to the typically developing control group in centromedial amygdala–ventrolateral prefrontal cortex connectivity, centromedial amygdala–putamen connectivity, and centromedial amygdala–occipital cortex connectivity. These results, illustrated in Figure 1A, reflect a lack of developmental decreases in psychosis emerging from underconnectivity in childhood. For connectivity between the centromedial amygdala and dorsolateral prefrontal cortex, the psychosis spectrum group exhibited higher connectivity during late childhood and early adolescence in comparison to the control group. This significant difference was no longer observed in adulthood. Specific time periods in which group differences were observed are presented in Table S5. All significant age-by-group interactions remained when psychiatric medication status was included as a covariate. Amount of variance explained by the full model, the inverse age-by-group interactions, and the main effects of inverse age and group are reported in Table S7 in the online supplement.

Similar to the typically developing control group, the psychosis spectrum group exhibited a decline in centromedial amygdala connectivity with increases in age for clusters in the following regions: parahippocampal cortex, frontal eye fields, posterior cingulate cortex, precentral cortex, and postcentral cortex (Figure 1B).

When youths with other psychopathology were added to the model, all five models maintained the significant inverse age-by-group interactions (see Table S6 in the online supplement). Like the typically developing control group, the other psychopathology group showed typical significant age-related decreases with increasing age in connectivity between the centromedial amygdala and three regions: the putamen, caudate, and occipital cortex (see Figures S5 and S6 and Table S7 in the online supplement). However, like the psychosis spectrum group, the other psychopathology group failed to show age-associated changes in centromedial amygdala–ventrolateral prefrontal cortex connectivity and centromedial amygdala–thalamus connectivity, although there were no significant inverse age-by-group interactions between the typically developing control group and the other psychopathology group (p>0.07). Furthermore, the typically developing and other psychopathology groups did not differ in amygdala connectivity levels at any point in development (see Table S7). As seen in Figure 2A–C, the developmental trajectories of centromedial amygdala–ventrolateral prefrontal cortex connectivity, centromedial amygdala–dorsolateral prefrontal cortex connectivity, and centromedial amygdala–thalamus connectivity for the other psychopathology group fell between the trajectories of the typically developing and psychosis spectrum groups during late childhood and early adolescence. In comparison to the other psychopathology group, the psychosis spectrum group exhibited reduced connectivity during late childhood and early adolescence in the following connectivity pairs: centromedial amygdala–putamen (Figure 2D) and centromedial amygdala–occipital cortex (Figure 2F).

Confirmatory analyses revealed that there were no significant interactions of inverse age by group by sex, inverse age by group by site, sex by group, or group by site in the psychosis spectrum and typically developing groups in these regions. Exploratory voxelwise analyses of age-by-group interactions failed to find any other significant clusters.

After calculating brain maturation deviation scores, we found that greater age-related deviation in centromedial amygdala–thalamus connectivity was associated with greater severity of positive symptoms in the psychosis spectrum group (r=0.19, p=0.01, q=0.05) (Figure 3). Post hoc analyses revealed that increased severity of grandiose ideas (r=0.31, p<0.001) and hallucinations (r=0.23, p=0.003) were related to centromedial amygdala–thalamus age-associated deviation, but not unusual thought content (r=0.08, p=0.32). This relationship was not present in the other psychopathology group (centromedial amygdala–thalamus: r=−0.02, p=0.89). This relationship was not observed between the centromedial amygdala–thalamus brain maturation deviation score and negative symptoms (r=0.02, p=0.8), depressive (r=−0.06, p=0.42), or manic symptoms (r=−0.01, p=0.93).

We next characterized where in the developmental trajectory amygdala connectivity related to positive symptoms. During late childhood and early adolescence, lower amygdala-thalamus connectivity was associated with greater severity of positive symptoms (r=−0.27, p=0.01, q=0.05).

In line with our previous findings, significant typical developmental functional connectivity decreases occurred between the centromedial amygdala and multiple brain regions (13). These results are consistent with previous developmental neuroimaging resting-state fMRI studies reporting decreases in subcortical-cortical connectivity into adulthood (36–39). In comparison to typically developing youths, those with psychosis spectrum disorders exhibited reduced connectivity between the centromedial amygdala and the ventrolateral prefrontal cortex, striatum, thalamus, and occipital cortex during late childhood and early adolescence, with a lack of normative decreases from adolescence to adulthood. These findings suggest either that there is an accelerated developmental decrease in amygdala connectivity in psychosis preceding the normative timetable or that the earlier age at onset reflects deterioration of this circuitry, as is evident later in adulthood. Normatively, decreases in connectivity can be seen as a period of specialization that occurs at a critical time when higher-level systems are becoming established to form adult trajectories. The lack of a marker of specialization through adolescence could reflect impairments in optimal specialization that could contribute to abnormal processing of executive affective information processing in psychosis.

Many of the regions that exhibited disrupted age-associated amygdala connectivity in the psychosis spectrum group are related to perception and salience (e.g., thalamus, striatum, and occipital cortex [40–47]). A primary function of the amygdala is to determine what is salient in one’s environment and to facilitate learning for these items (48–50). Projections of the amygdala to the thalamus are thought to modulate attentional orientation and arousal, broadly speaking (51). Projections of the amygdala to the visual cortex are known to enhance sensitivity, discrimination, and subjective vividness of perceived stimuli (52). Finally, projections of the amygdala to the striatum are thought to provide an interface between the amygdala and dopamine systems, which are known to regulate motivation, learning, and behavioral activation (53). Thus, abnormal connectivity of the amygdala with the thalamus, visual cortex, and striatum could reflect processes that result in the misattribution of salience to stimuli in psychosis, as well as a facilitation of learning associations about threat-related items. Critically, this type of aberrant salience processing may provide the foundation for higher-order features of positive symptoms, such as delusions (54).

Connections between the centromedial amygdala and lateral prefrontal regions also exhibited a disruption in age-associated changes in the psychosis spectrum group. During late childhood and early adolescence, the psychosis spectrum group exhibited reduced amygdala–ventrolateral prefrontal cortex connectivity; however, during adulthood the psychosis spectrum group exhibited increased connectivity between these two regions. Amygdala–ventrolateral prefrontal cortex connectivity is necessary during the reappraisal phase of regulating one’s emotions (55–59), and how these two structures interact during emotion regulation changes during adolescent development (60). Indeed, impairments in lateral prefrontal–mediated circuitry are related to emotional deficits typically observed in psychosis (61–63) and age-associated amygdala–ventrolateral prefrontal cortex functional connectivity alterations during an affective labeling task have been observed in youths at clinical high risk for developing psychosis (64). Thus, centromedial amygdala–ventrolateral prefrontal cortex age-associated disruptions may underlie the emotional dysregulation that often precedes and predicts increased psychotic symptoms (65–67). Studies using experience sampling have shown that adults with schizophrenia report more intense negative emotions and greater social stress than healthy control subjects (68–70). The heightened amygdala–ventrolateral prefrontal cortex connectivity we observed in adults experiencing psychosis spectrum symptoms may reflect underlying biological vulnerability to psychosis onset, which interacts with these environmental stressors. Studies integrating experience sampling methods (71) with neuroimaging in youths at high risk for developing psychosis are necessary to test this hypothesis.

We also observed a unique, altered age-associated pattern of centromedial amygdala–dorsolateral prefrontal cortex connectivity in the psychosis spectrum group compared with the typically developing control group. During late childhood and early adolescence, the psychosis spectrum group exhibited increased centromedial amygdala–dorsolateral prefrontal cortex coupling compared with the typically developing group. This group difference was no longer present in adulthood, with both the typically developing and the psychosis spectrum groups exhibiting similar levels of connectivity. Previously, in adults with schizophrenia, absent or reduced amygdala–dorsolateral prefrontal cortex functional connectivity has been observed during emotional distraction during a working memory task (72) and at rest (7, 21). Our developmentally sensitive results of increased centromedial amygdala–dorsolateral prefrontal cortex connectivity in the psychosis spectrum group during late childhood and early adolescence contrast with these findings and highlight the importance of examining how developmental stage may affect the directionality and interpretation of brain connectivity (73). Reduced GABA levels are consistently observed in the prefrontal cortex in schizophrenia. While GABA-ergic deficits have not been identified in the amygdala in schizophrenia, the majority of neurons projecting from the centromedial amygdala are GABA-ergic (74). Altered connections may generate down-regulation of GABA interneuron activity in the prefrontal cortex, resulting in a lack of inhibition, which may be responsible for the increased connectivity observed between the amygdala and dorsolateral prefrontal cortex during late childhood and early adolescence in youths with psychosis spectrum disorders.

In summary, we identified developmentally sensitive alterations in cortico-limbic and intralimbic resting-state fMRI connectivity in youths with psychosis spectrum disorders. These neurodevelopmental alterations provide support for multiple theories associated with schizophrenia and complement the growing body of literature that shows progressive maturational disturbances in those who go on to develop psychosis (75–78).

Two age-associated amygdala connectivity alterations were distinct to youths with psychosis spectrum disorders. In comparison to both the typically developing and other psychopathology groups, the psychosis spectrum group exhibited reduced connectivity during late childhood and early adolescence in two connectivity pairs: centromedial amygdala–putamen and centromedial amygdala–occipital cortex. Although psychotic symptoms rarely separate clinical samples into discrete groups, our results suggest that these brain abnormalities are unique to youths with psychosis spectrum disorders and, in the future, could potentially differentiate psychotic disorders from other psychiatric disorders. Alternatively, in other amygdala connectivity measures, the developmental trajectories of the other psychopathology group fell in between those of the typically developing and psychosis spectrum groups. These findings suggest that there is a less severe neurobiological impact on the other psychopathology group in these connectivity metrics.

The growth charting methods employed in this study establish a novel connection between deviations from amygdala-thalamus connectivity development and increased positive symptom severity. In our exploratory analyses, we found that deviation from the normative trajectory of neurodevelopment is relevant to positive symptoms. Although this relationship was statistically significant after multiple comparisons, it is a small effect and must be replicated in future studies. These findings, along with others (17, 18, 73), highlight the importance of examining the role that (dys)maturation patterns play in the development of psychiatric disorders. While small effect sizes may not indicate a direct intervention, it is important to identify these deviations to fully characterize the pathophysiology of psychosis risk.

This study also represents a proof-of-principle approach for merging multiple resting-state fMRI data sets to inform normative developmental trajectories and identify aberrant trajectories in youths with psychosis spectrum disorders. Despite the samples having multiple sites, protocols, and recruitment methods, the age-associated changes were remarkably consistent across the different samples of typically developing youths (see Figure S3), and including sample as a covariate effectively removed any site differences (see Figure S4). In conjunction with recent work (10, 17, 79–81), these results support using publicly available data to assess developmental changes in brain function and relevance to psychiatric disorders. Given that age-associated changes can be small but significant, an approach that takes advantage of large sample sizes will be necessary for identifying distinct periods of development in which there are disruptions related to psychiatric disorders or symptoms.

Our study was limited by the fact that cross-sectional data were available only for the psychosis spectrum group. Thus, the neurodevelopmental trajectories in psychosis spectrum disorders do not reflect within-person change. Our cross-sectional sample cannot definitively show whether our results are due to altered development or abnormalities due to time of onset of psychosis. Thus, longitudinal studies of youths with psychosis spectrum disorders, with multiple visits per individual, can extend our understanding of psychosis by identifying how the shape and rate of maturation of subject-specific developmental trajectories in youths with psychosis spectrum disorders diverge, converge, or remain stable in comparison to typical development (82–84). Furthermore, many developmental changes that occur during adolescence are nonlinear, and these patterns are most accurately captured with longitudinal analyses (85). Additionally, psychotic symptoms are dynamic and change over time (86, 87), and these changes need to be taken into account when characterizing neurodevelopmental change. Recently, using novel time-varying analytic approaches (88, 89), we found that connectivity measures were differentially related to individual differences in anxiety and depression at different points in adolescent development (13). A similar approach could be applied to longitudinal neuroimaging and psychotic symptom data, to identify particular periods of development in which psychic symptoms are linked to resting-state fMRI connectivity metrics.

Finally, while we are fairly confident that we were able to appropriately account for site in our analyses (see Figure S4), we observed a statistically significant effect of site in many of the regions of interest (see Table S4). Despite all scans being performed on the same scanner model, there were still differences in task instructions, MRI resolution, and duration of scan. Site effects may have obscured our ability to identify smaller developmental changes in normative amygdala connectivity developmental trajectories. It is also possible that we failed to identify more subtle age-related deviations in amygdala connectivity between the psychosis spectrum and typically developing groups because of site effects. Recently, methodology from genetics has been used to harmonize structural MRI data across sites (90, 91); modifying and applying this method to resting-state connectivity data is a logical next step. Despite these site differences, we still see a significant interaction between group and age; these findings suggest that multisite neuroimaging data sets will be important for understanding how biomarkers may be sensitive or specific to developmental stage.