Treatment outcome results
Before treatment, there were no significant differences between the YMRS scores of responders and nonresponders, t(32)=.21; p=0.83. After treatment, there was a large, statistically significant difference between YMRS scores for the two groups, t(32)=6.04; p<0.001. However, both groups showed a statistically significant decrease in YMRS scores from pre- to post-treatment, nonresponders: t(11)=2.38, p=0.036; responders: t(21)=16.42, p<0.001, indicating that even the nonresponders' YMRS scores decreased somewhat. Finally, HC showed no change in YMRS scores between testing sessions, t(13)=0.0, p=1.0).
fMRI task behavioral results
At baseline, the PBD group responded significantly slower than HC on the pediatric affective color-matching task, F(1, 46)=5.34, p=0.025, but this difference was marginal after treatment, F(1, 46)=4.02, p=0.051) (see ). The PBD group responded slower than HC for the positive and neutral words in the first session [positive: t(46)=2.5, p=0.016; neutral: t(46)=2.6, p=0.013], but for the negative and neutral words in the second session [negative: t(46)=2.3, p=0.028; neutral: t(46)=2.2, p=0.035], as revealed by a significant three-way interaction, F(2, 92)=3.41, p=0.037. Thus, the patients' impairments may be specific to their emotional state, as they were manic pre-treatment and remained somewhat depressed after treatment. The PBD group was also less accurate overall, F(1, 46)=7.99, p=0.007, but this did not differ across word valences or sessions. Despite the differences between patients and HC, medication responders and nonresponders showed no significant differences in RTs or accuracy rates. Therefore, although the PBD group showed some performance impairments on this task, these differences did not predict whether a participant would respond to medication.
Behavioral reaction times for PBD versus HC across valence and testing session for the color-matching task.
Task engagement results
Responders, nonresponders, and HC engaged the Frontolimbic Affective Circuit differently depending on the valence of the stimuli both before and after treatment, as revealed by a significant three-way interaction, F(4,90)=2.91, p=0.026, (). Separate three-way ANOVAs separating the groups for each valence at each time point revealed that responders, nonresponders, and HC showed different network engagement in response to negative words before treatment, F(2,45)=3.57, p=0.037, and in response to positive words after treatment, F(2,47)=3.49, p=0.039. Post hoc t-tests showed that before treatment both responders and nonresponders showed decreased network engagement in response to negative words relative to HC, t(34)=2.56, p=0.015; and t(24)=2.27, p=0.032, respectively; but these two groups did not differ from one another, t(32)=0.45, p=0.657. After treatment, responders showed increased network engagement compared with HC in response to the positive word stimuli, t(34)=2.57, p=0.015, but nonresponders did not differ from either responders or HC.
Frontolimbic Affective Component task engagement (fit with model) for responders versus nonresponders versus HC for positive, negative, and neutral word blocks.
The Structure of the Frontolimbic Affective Regulation Circuit
A one sample t-test against a zero connectivity baseline revealed that the frontolimbic affect regulation circuit included increased functional connectivity in bilateral amygdala, parahippocampal gyrus, hippocampus, inferior VLPFC and orbitofrontal cortex, anterior insula, superior temporal pole, and the inferior cerebellar vermis (all regions significant at p<0.01 FWE; see ). Areas that showed decreases in functional connectivity across participants during the task included the superior vermis and the bilateral thalamus. These regions represent the areas that are a part of the frontolimbic circuit and, by extension, are linked to the amygdala and VLPFC ROIs.
Brain regions that are a part of the Frontolimbic Affective Component across all participants and testing sessions. One sample t-test of the Frontolimbic Affective Circuit at p=0.01 family-wise error rate (FWE).
Predictors of treatment outcome (responders vs. nonresponders)
The 2×2×3 mixed model ANOVA on the participants' amygdala connectivity values showed a significant main effect of group, F(2, 43)=3.48, p=0.040, (). In contrast, a 2×2×3 mixed model ANOVA on the VLPFC connectivity values showed no significant effects. Post-hoc t-tests on the amygdala connectivity values revealed that responders showed significantly greater connectivity values overall than nonresponders, t(31)=2.98, p=0.006. HC did not differ from responders, t(32)=.628, p=0.54, or from nonresponders, t(23)=1.71, p=0.10. The responders showed increased functional integration between the amygdala and the rest of the limbic circuit, whereas the nonresponders showed relatively decreased functional integration between the amygdala and the limbic circuit. This preliminary evidence suggests that the integration between the amygdala and the frontolimbic circuit is a potential trait-like marker of medication responsivity.
ROI-derived amygdala-frontolimbic connectivity values (beta-weights) from the Frontolimbic Affective Component across groups and sessions. resp, medication responder; non, medication nonresponder.
While not significantly different than either group, HC showed an intermediate level of connectivity in the amygdala, partway between the responders and nonresponders and not significantly different from either group. This finding suggests that there may be a combination of factors contributing to the changes between the patient groups (i.e., both that responders were compensating and that nonresponders were getting worse).
Further analyses indicated that responders showed significantly greater amygdala connectivity than nonresponders in the left amygdala at pre-treatment baseline, t(31)=2.05, p=0.049, but no significant difference in the right amygdala at pre-treatment, t(31)=.98, p=0.34. However, after treatment, only the right amygdala differed significantly between responders and nonresponders, t(31)=2.23, p=0.033 with a marginal effect in the left amygdala, t(31)=1.81, p=0.080. There were no significant differences between the HC group and the other two groups in the amygdala at either time point. Although the three-way interaction was not significant, F(2,43)=2.36, p=0.107, these results suggest that left amygdala connectivity better differentiated responders from nonresponders before they were treated, whereas right amygdala connectivity better distinguished responders from nonresponders after treatment.
Amygdala ROI values predict medication response and clinical outcome scores
The logistic regression model revealed that the left amygdala pre-treatment and right amygdala post-treatment ROI values successfully classified 76% of participants as responders or nonresponders (odds ratio [OR] for left amygdala at pre-treatment assessment=6.7, confidence interval [CI]=0.96–46.3, p=0.055; OR for right amygdala at post-treatment=11.0, CI=1.0–119.9, p=0.050). The model classified responders fairly well (18 as responders, 3 as nonresponders; 86% correct), but only classified 58% of the nonresponders correctly (7 as nonresponders and 5 as responders). Although the contribution of the left amygdala values before treatment to the model's predictive power were just below the level of statistical significance (p<0.055), if they were removed from the model, the overall fit of the model was significantly decreased (change in −2 Log Likelihood=5.65, p=0.017), and the model would classify only 61% of the participants correctly. Thus, left amygdala connectivity pre-treatment added some predictive power to the model beyond that provided solely by right amygdala connectivity after treatment.
The overall average amygdala ROI connectivity values for the patients also significantly predicted the improvement in YMRS scores, b=−11.1, t(32)=2.94, p=0.006, R2=0.22, most likely due to post-treatment data. Average amygdala ROI connectivity after treatment predicted the improvement in YMRS scores, b=−6.7, t(32)=2.51, p=0.018, R2=0.17, whereas amygdala connectivity pre-treatment did not, b=−5.2, t(32)=−1.56, p=0.129, R2=0.07. Finally, right amygdala connectivity post-treatment predicted improvement on mania symptoms, b=−7.5, t(32)=3.1, p=0.005, R2=0.23, whereas left amygdala connectivity post-treatment did not, b=−3.6, t(32)=1.54, p=0.133, R2=0.07.