1. Introduction
Depressive disorders impact about 16% of the population [
1]. Depression is associated with impairment, chronic suffering, and early death [
2]. Historical trends suggest that depression is on the rise and a leading cause of the global burden of disease worldwide [
3]. Depression is commonly first evident during adolescence, and earlier onset is associated with a poorer prognosis, including premature death [
4,
5,
6]. As adolescence is a sensitive period for both neurodevelopment and the onset of psychopathology [
7], early intervention during this window may be particularly fruitful in mitigating the significant impacts of depression. Fortunately, evidence-based treatments are available and include psychopharmacological agents and psychotherapy [
8,
9]. However, work in this area is still needed given that 30 to 50% of the adolescents who receive care using the most well-validated treatments do not achieve remission [
10,
11]. Precision medicine approaches hold great promise for identifying which treatment(s) will be most effective for each individual [
12], shortening the time it takes for an individual to receive the treatment that will ultimately bring them to remission.
Neuroimaging approaches have been shown to advance the understanding of treatment mechanisms and this line of work has the potential to guide individualized treatment. An accumulating body of imaging research implicates processing of negative affect for those with depression. The Salience and Emotion Network (SEN) is involved in detecting and regulating negative affect: the amygdala and other regions of the limbic system are involved in negative affect detection, while the anterior cingulate cortex (ACC) is an important structure that mediates the detection of negative affect and serves as a regulator of amygdala responses. There is a wealth of evidence showing that the SEN functions abnormally in depression with adults [
13,
14,
15] and adolescents [
16,
17,
18].
Relatively little research has been conducted on neural predictors of psychotherapy for those suffering from depression, particularly in adolescence, a period when connections within the SEN are changing rapidly [
19]. Because a primary goal of psychotherapy is to teach strategies to both prevent and regulate negative emotions, psychotherapy may involve changes in synaptic plasticity by retraining of the implicit memory systems involved in negative affect [
20,
21,
22]. Validated psychotherapy for adolescents with depression includes Interpersonal Psychotherapy (IPT) for Depressed Adolescents (IPT-A) [
10,
23,
24]. IPT/IPT-A focuses on emotions in the context of the patient’s relationships. Conflict and stress in interpersonal relations are a common source of negative affect in adolescents with depression and are associated with the development, maintenance, and recurrence of depression [
25,
26,
27]. IPT-A aims to decrease depressive symptoms by addressing interpersonal difficulties that arise from interpersonal role disputes, role transitions, interpersonal deficits, and loss. The interpersonal skills addressed in IPT-A include conflict negotiation, interpersonal problem solving, social approach/avoidance, response inhibition, and perspective-taking [
23]. Successful early intervention through IPT-A may, in some ways, be related to the functioning of the SEN, as this network is strongly implicated in interpersonal relationships [
28].
Prior research has yet to examine neural correlates or predictors of IPT-A with depressed adolescents or adults. Almost two decades ago, positron emission tomography (PET) scans were used to measure the change in regional cerebral glucose metabolism in adults with depression in the context of IPT (although the studies focused on brain mechanism but not predictors of IPT response) [
29,
30]. While we are not aware of subsequent research published on neural mechanisms or predictors of IPT/IPT-A, other forms of evidence-based psychotherapy that target negative emotions may yield some clues about possible predictors of treatment response. For example, research on neural predictors of Cognitive Behavior Therapy (CBT) may be relevant to consider, particularly research that examines volume, task activation, and resting-state functional connectivity (RSFC) of key nodes of the SEN. Regarding brain structure, there is accumulating evidence showing that larger ACC volume is related to a more favorable treatment response to CBT [
31,
32,
33]. One of the only studies that focused on adolescents with depression considered RSFC of the amygdala and ACC as a whole-brain seed within the context of a trial of CBT group psychotherapy. It was found that higher baseline connectivity of both the amygdala and right subgenual ACC with the dorsolateral prefrontal cortex (DLPFC) was associated with larger improvements in depression scores [
34], centrally implicating these key regions of the SEN. Research is not as consistent with regard to neural activation within the context of emotionally salient tasks, perhaps in part because a range of paradigms has been used. Ritchey and colleagues [
35] found that greater activity of an area near the ACC in the context of viewing positive, negative, and neutral images was related to more improvement of depressive symptoms within the context of a trial for CBT in adults with depression. Rubin-Falcone and colleagues [
36] failed to find that ACC or amygdala activation predicted CBT response using similar stimuli. Further, in a series of studies by Fu and colleagues [
37,
38] in which a viewing of sad faces task was used at baseline, greater ACC activation was a predictor of poorer CBT treatment response. Others have used paradigms in which negative words were viewed prior to the CBT trial. Interestingly, Siegle and colleagues [
39] found that greater amygdala activation and less ACC activation predicted favorable treatment response, while Doerig and colleagues [
40] found that greater amygdala activation was associated with a less favorable response to treatment. More work is needed, but these identified patterns provide some key evidence that the SEN may be a useful predictor of psychotherapy treatment response and may have implications for IPT-A.
While randomized trials are crucial to confirm neurobiological predictors of different treatment responses for personalization, an early step is to examine neural correlates of treatment response in the context of a validated treatment. In this current pilot project, we aimed to identify key neurobiological constructs that are associated with abatement of depressive symptoms within the context of a trial of IPT-A. We predicted that a greater ACC volume and greater amygdala-ACC RSFC at baseline would be associated with a better response to treatment. We also hypothesized that ACC and amygdala activation during a baseline administered task that probed negative emotion would predict greater improvement in depressive symptoms, although directional predictions were not made. The findings of this pilot study are preliminary and intended to be hypothesis-generating rather than confirmatory.
4. Discussion
There is an urgent need for new knowledge to guide the personalization and tailoring of treatment approaches to optimize clinical response outcomes. This process may be especially critical in the earliest stages of illness when there is an opportunity to divert significant worsening in clinical trajectories and restore health. Further, most of the existing research has been conducted with adults. Given the differences in brain development, the extent to which the work with adults identifying neural predictors of treatment response is relevant to adolescents when brain plasticity is heightened is unclear. In a recent review of the literature of predictors of treatment response for adolescents with depression, Ang and Pizzagalli [
54] concluded that “Studies on biomarkers that truly reflect pathophysiology are scarce and difficult to draw conclusions from” (p. 18). Indeed, this line of work is just beginning to be undertaken. While the current work has the strength of a prospective design to evaluate neurobiological predictors of treatment response, given its small sample and single-arm design, the present study is intended to only serve as hypothesis generation for future studies evaluating neural predictors or moderators of psychotherapy response for adolescents with depression.
One of the consistently implicated regions in studies examining predictors of treatment response has been the ACC (e.g., [
55]). Similarly, the most consistent results of the current study implicate the ACC as a predictor of change in depressive symptoms in the context of IPT-A. While our results failed to show consistent evidence of ACC structure as a predictor of treatment response, greater baseline activation of the right ACC in response to negative emotion stimuli was shown to predict a more favorable treatment response. This result was found across W8 and W16 outcomes for the BDI, an index based on child self-report. Of interest, this finding shows some similarities to work with adults [
35], even if the broader literature shows a wide variation in the direction of results and the exact brain regions implicated (e.g., different hemispheres, subgenual ACC). It is possible that individuals with greater ACC activation at baseline have the capacity for greater change post-treatment, thereby driving this effect. A different perspective for how ACC is functioning may support regulatory capacity. In the current work, greater left amygdala-ACC RSFC also predicted greater improvement in depressive symptoms. Together these results suggest that ACC and the broader SEN may be important to examine when attempting to identify biomarkers that may predict treatment response.
Studies with adults that have incorporated neuroimaging into other types of psychotherapy have also implicated cingulate and limbic regions. Given that the ACC is a general predictor for a range of pharmacological and psychotherapeutic interventions, the extent to which the ACC will be a useful biomarker for differential prediction of IPT-A treatment response for those suffering from depression is unclear. An example of this type of work includes evidence in an adjacent brain region, also part of the SEN: Mayberg and colleagues have demonstrated that functioning of the insula can differentially predict response to psychopharmacological versus psychotherapeutic interventions [
56], with lower activation of the anterior insula predicting a better response to CBT and higher activation of the anterior insula predicting a better response to SSRI. However, it is premature to discount the activation of the ACC to assign youth to treatment, for fully powered randomized control trials will be needed to make this determination.
In this study, results suggesting the predictive value of the amygdala were very scarce. The results failed to show evidence of amygdala volume as a predictor. In addition to the findings discussed above regarding connectivity, left amygdala activation during the emotion faces viewing task was associated with CDRS raw score improvement at W8 (for both percentage improvement and raw change scores). A similar pattern was only occasionally noted in psychotherapy studies with adults [
39] but has also been found in adolescents treated with SSRIs [
57]. While not a treatment trial, Canli and colleagues [
58] found that greater amygdala activation to emotional faces predicts a greater decline in depressive symptoms over an eight-month period, controlling for medication status. Adjacent brain structures have also been shown to predict treatment response. Although not the focus of this study, based on other work [
59], we explored hippocampal structure and activation during the emotion-matching task (
Supplementary Materials), showing some tentative evidence that hippocampal volume may be a candidate for future consideration, as the emotional memories laid down when addressing interpersonal threats are critically important to the adaptive capacities of the individual. Future work should consider other brain networks likely to be implicated in IPT that were not examined here, including the default mode network and more distributed networks involving social processes, cognitive control network (e.g., [
34]), and reward network which has been found to predict CBT treatment responses in adolescents [
60] and Behavioral Activation Therapy in adults [
61].
The Research Domain Criteria (RDoC) initiative [
62] suggests that multiple-level approaches obtain the most informative outcomes. Including multiple units/levels of analysis allows examining relevant systems across the brain. This approach is increasingly used with adults [
63]. This multilevel approach has recently been used with adolescents. For example, in an open-label study on youth with depression who were undergoing psychopharmacological treatment with SSRIs [
57], in addition to neural activation and connectivity predictors, elevations of the hypothalamic pituitary axis were identified in those more likely to benefit from pharmacotherapy. While here we addressed multiple-levels of neural structure and function, this multilevel approach has yet to be routinely applied to psychotherapeutic interventions with depressed adolescents. Together, these findings point to the utility of taking a multiple index approach to assess treatment response.
There are significant study limitations. This work is exceptionally challenging to undertake given that in addition to the ongoing psychotherapy sessions, there is extensive coordination between parents and children’s schedules for the lengthy assessments conducted before, during and after treatment. Indeed, only about a third of the participants who initially consented were included in this analysis. This also resulted in a sample of convenience that was generally well resourced and represented limited racial or ethnic diversity. Additionally, due to issues related to subject motion and technical problems, the number of scans with usable data was smaller than expected. In the end, the exceptionally small sample size precluded any findings from being confirmatory; instead, they should be considered as possible avenues to explore in future work with considerably larger samples. Overall, models were simplistic and did not account for multiple comparisons. If avenues relevant to personalization are to be realized, future work will need to go beyond considering prediction of IPT-A treatment response and consider randomized control trials with multiple arms of assignments to different treatments. Future work may also apply machine learning approaches for algorithm development [
64] and integration of clinical, cognitive, physiological, neural, and genetic data in prediction algorithms [
65]. When more fully realized, this translational line of work holds enormous potential for advancing understanding of the pathophysiology of depression, promoting the development of more effective neurobiologically-informed interventions, and eventually aiding in the efforts to identify biological markers that can advance personalized treatment [
12].