Empirically supported psychological treatments and the Research Domain Criteria (RDoC)

Highlights

• The research domain criteria (RDoC) units of analysis have been examined in several clinical trials employing evidence-based treatments.

• Genes, cells, molecules, and circuits were examined as possible moderators and mediators of outcome for negative valence system psychopathology.

• The research reviewed suggests that there are several candidate units of analysis for moderation or mediation of treatment outcome.

• The findings are limited or investigations were not designed to fully identify these critical therapy-relevant factors.

Abstract

Background

The Research Domain Criteria (RDoC) has been developed as an alternative approach to studying psychiatric disorders. The RDoC constructs and units of analysis, from genes up through paradigms, are intended to describe a hierarchy of priority measurements. Several of these have been investigated in the context of empirically-supported treatments, as either moderators or mediators of outcome.

Method

This review considers the available research on the moderating and mediating role of genes, molecules, circuits and physiology in cognitive-behavior therapy (CBT) outcome studies for negative valence system conditions.

Findings

Based on the review, research has aspired to identify candidate genes, molecules, circuits and physiological moderators or mediators of treatment, but no definitive tests have been conducted. Instead, several candidate variables have been found that deserve further investigation.

Limitations

The available research is based on diagnoses from the DSM, whereas the RDoC initiative endeavors to determine empirically valid taxonomic signs.

Conclusions

The results of this review are discussed in the joint context of developments in empirically-supported psychological therapy and the specific aims of the RDoC initiative, and conclude with recommendations for future research.

Introduction

The Research Domain Criteria (RDoC) presents a compelling opportunity to develop, by empirical means, an alternative to the current descriptive taxonomy used in psychiatric research and treatment. As an empirically derived approach to classification, there is also an opportunity to develop a framework that could be meaningfully related to treatment recommendations, prognostic indicators, and expected treatment outcome. This would be a considerable advance as presently the Diagnostic and Statistical Manual (DSM-5; American Psychiatric Association, 2013) is a purely descriptive taxonomy that confers little specific information to guide treatment decisions and provides no information related to expected outcomes or course of illness (beyond that depicted in field trials).

The RDoC arranges broad areas of psychopathological signs and units of assessment into a matrix, with the expectation that the cells of the matrix can be the source of empirical findings. RDoC aims to improve understanding of psychopathology, with the aim of developing a more reliable and valid taxometric system (Insel et al., 2010). While in its current conceptualization the RDoC is not associated with specific DSM categories, one could roughly arrange the existing diagnoses into components of the aforementioned matrix. Accordingly, the following DSM diagnostic categories or diagnoses are can be characterized as ones with substantial elements of Negative Valence Systems in RDoC:

• Anxiety disorders, which are among the most highly prevalent psychiatric conditions and represent a significant source of psychiatric disability. Anxiety disorders are most closely related to the RDoC Negative Valence sub-domains of Acute Threat (“Fear”) and Potential Threat (“Anxiety”).

• Obsessive-compulsive disorder (OCD), which has significant anxiety features, and is associated with substantial work-related disability and impaired social and family relationships. The RDoC Negative Valence sub-domains of Acute Threat (“Fear”) and Potential Threat (“Anxiety”) seem to apply to many cases of OCD, although it is noted that some sufferers experience a sense of “incompleteness” rather than fear or anxiety (Summerfeldt, 2004).

• Posttraumatic stress disorder (PTSD), which also has significant associated anxiety, and is characterized by substantial reductions in quality of life. Many aspects of PTSD correspond to the RDoC Negative Valence sub-domains of Acute Threat (“Fear”) and Potential Threat (“Anxiety”), although the Sustained Threat, Loss, and Frustrative Nonreward subdomains may also apply.

• Depressive disorders, which are very common and can be highly debilitating. The RDoC sub-domain of Loss appears most relevant to depression, although Frustrative Nonreward may also be implicated.

Given that research has accumulated that evaluates the syndromal components of specific psychopathology in the context of levels of analysis as depicted in the RDoC matrix, this paper has two broad aims. First, we aim to examine some of the available research on the Negative Valence Systems constructs, with particular emphasis on studies that describe genes, molecules, circuits, and physiology for their potential moderating effects in empirically supported psychological treatment. The second aim is to examine how these matrix items can inform research for the Negative Valence Systems constructs, again in particular reference to genes, molecules, circuits, and physiology for their mediating effects in empirically supported psychological treatment. In this way, it is hoped that research may focus energy on identifying components of the RDoC matrix that may be under-represented in the research literature. The cells associated with behavior and self-report were excluded given that those domains have been heavily researched and the incorporation of biomedical indicators in psychosocial research is a comparably recent development.

We note that the present review is not intended to be exhaustive; a complete examination of all of the research on genes, molecules, circuits, and physiology and their relationship to psychological treatments would be well beyond the scope of this paper. Rather, we aim to highlight representative studies that illustrate the relationship between empirically supported psychological treatments and RDoC.

Standards for empirically supported psychosocial treatments (ESTs) have been in place for close to twenty years (Chambless et al., 1998). The original standards required that, in order for a treatment to be deemed “well established,” the highest designation of empirical support, there had to be at least two randomized controlled trials (RCTs) conducted by different research teams documenting efficacy, usually determined by symptom reduction (Chambless and Ollendick, 2001).

At the time the original standards were developed, RCTs were relatively rare in evaluations of psychosocial interventions. Since that time the number of RCTs conducted has grown considerably, and RCTs are fairly common in evaluating treatment outcome. Partly as a result of this, new standards for declaring empirical support have been developed and adopted by the Society of Clinical Psychology (SCP) (Tolin et al., 2015a, Tolin et al., 2015b). Under the new standards, the threshold for a “very strong recommendation,” the highest designation, requires high quality evidence of the treatment having a clinically meaningful effect on both symptoms of the disorder and functional outcomes, with sustained benefits at least three months post-treatment, and at least one high quality study in a non-research setting. The revised standards call for assessment of treatments by systematic reviews, such as meta-analyses, rather than a count of studies that demonstrated efficacious outcome. This would allow for a more comprehensive appraisal of research findings, including possibly contradictory results.

Many of the “well established” treatments, to use the designation of the original standards, dealt with diagnoses that would be considered part of the negative valence systems under RDoC. As noted above, while these treatment protocols have generally been yoked to DSM diagnoses, broad generalities may be drawn that correspond in many ways to cells of the RDoC matrix. The DSM-5 anxiety disorders, depressive disorders, obsessive-compulsive and related disorders, and trauma- and stressor-related disorders could all be reasonably classified as negative valence disorders. The empirical basis for this claim comes from research showing that these DSM entities are linked by broad underlying dimensions such as neuroticism, high negative affect, and low positive affect (Brown and Barlow, 2009, Brown et al., 1998; Watson, 2005).

For purposes of the present review, we will limit our discussion to a handful of ESTs with robust evidence of efficacy for Negative Valence disorders. These include:

• Exposure therapy, which has been extensively studied and is considered highly efficacious (Deacon and Abramowitz, 2004).

• Behavioral activation (Martell et al., 2010), which is based broadly on the idea that depression is associated with a loss of reinforcement from the environment (Ferster, 1973). As a result, depressed individuals lose interest in activities, leading to vegetative signs associated with the syndrome. The treatment entails developing schedules of activities that might not be regularly completed by the depressed individual in order to spur greater reward from the environment.

• Cognitive therapy and cognitive-behavioral therapy, which emphasize directly challenging dysfunctional cognitions, and may include a variety of behavioral exercises designed to activate these cognitive distortions in order to access them directly (e.g., Beck, 1976). However, most treatment programs incorporate varying levels of behavioral procedures as well. Indeed, some have also suggested that the behavioral exercises in cognitive therapy constitutes a type of exposure exercise that is in itself therapeutic (Abramowitz et al., 2005). Accordingly, we have reviewed and presented findings related to cognitive therapy and cognitive-behavioral therapy (CBT) together.

We have restricted the discussion to these therapeutic approaches for convenience and space as well as their empirical support. These therapeutic approaches are, collectively, part of a broader therapeutic approach with closely related theoretical bases for their development. A comparable review could be conducted for other ESTs, such as interpersonal psychotherapy for depression. Given the partial overlap in treatment methods among exposure therapy, behavioral activation, and cognitive therapy or CBT, the following review reasonably contains some potential redundancies in the findings across studies.

Traditionally, efficacy studies have relied on patient report, clinician- or other informant- rated symptom scales, or behavioral indices of improvement (see Hollon and Dimidjian, 2014; McKay, 2016). However, a growing body of research also addresses how the physiological RDoC domains affect, and are affected by, psychological treatments. Treatment researchers could employ the RDoC units of analysis (genes, molecules, circuits, and physiology) in two key ways. First, these components could be moderators of eventual clinical outcome: they could help determine which individuals will respond most favorably to a given intervention. The identification of treatment moderators benefits clinicians by identifying which patients are likely to respond to a given treatment, and may identify subpopulations with possibly different causal mechanisms (Kraemer et al., 2002). Treatment moderation is most easily conceptualized using the seminal model by Baron and Kenny (1986), shown in Fig. 1. Moderation is demonstrated via an interactive effect with treatment on clinical outcomes (e.g., symptoms of the illness), represented by path c in Fig. 1. Several studies have demonstrated that the RDoC components at pre-treatment correlate with eventual symptom improvement (path b in Fig. 1). However, few have demonstrated that these components interact with the treatment to result in different outcomes (path c in Fig. 1).

Imagine, as an example, an open trial in which patients receive a particular EST and experience symptom reduction (path a in Fig. 1). Imagine further that prior to treatment, patients are screened for a particular genetic polymorphism, and it is demonstrated that the presence of this polymorphism predicts less symptom reduction after treatment (path b in Fig. 1). Although we have demonstrated prediction of outcome, we cannot determine that the polymorphism moderates the effect of the treatment. The polymorphism could simply predict chronicity of symptoms over time, regardless of what treatment is administered. To determine a moderating effect, we would need to show an interactive effect of treatment X polymorphism—requiring a controlled trial in which some patients received a different intervention. The polymorphism would have to predict poorer outcome for our EST but not (or to a lesser extent) for an alternative treatment. Thus, there is currently a wealth of information about RDoC predictors of outcome, though much work needs to be done before these components can be considered treatment moderators.

The components could also be treatment mediators, or mechanisms by which a given treatment causes clinical change. Understanding the mechanisms by which treatments operate may facilitate the development of innovative and potentially more powerful treatments, emphasizing active therapeutic components and minimizing or eliminating inactive ingredients (Kazdin and Weisz, 1998). To the extent that various treatments operate via different mechanisms, understanding those mechanisms would not only lead to better, more synergistic treatment programs but would also prevent combining different treatments with potentially incompatible or mutually antagonistic mechanisms (Kraemer et al., 2002).

Baron and Kenny's (1986) model of treatment mediation is depicted in Fig. 2. Treatment mediation is demonstrated when all of the following conditions are met:

• The treatment is associated with changes in the proposed mediator (path a in Fig. 2).

• Changes in the proposed mediator are associated with changes in symptoms (path b in Fig. 2).

• The treatment is associated with changes in symptoms (path c in Fig. 2).

• The association between treatment and symptom change is eliminated or substantially reduced when accounting for paths a and b (path c’ in Fig. 2).

Although the analytic strategies of mediator analyses have been refined substantially over the years (e.g., MacKinnon et al., 2007), the Baron and Kenny model continues to have substantial heuristic value. Of note, very few studies have demonstrated true mediation using the RDoC components discussed here. Many studies have shown that treatment affects genetic expression, molecules, cells, circuits, and physiology (path a in Fig. 2). Fewer have shown that changes in those components are associated with symptom reduction (path b in Fig. 2), and even fewer have demonstrated that changes in the components account fully or partially for the relationship between treatment and symptom reduction (path c’ in Fig. 2).

Imagine, as an example, an open trial in which patients receive a particular EST and experience symptom reduction (path c in Fig. 2). Imagine further that we examine a particular neural circuit before and after treatment, and we demonstrate that activity in that circuit decreases from pre- to post-treatment (path a in Fig. 2). Although we have demonstrated that the circuit is affected by treatment, we cannot determine that the circuit mediates the effect of the treatment. Changes in the circuit may simply be one of many effects of the treatment. To determine a mediating effect, we would need to show that changes in the circuit are associated with reduction in symptoms (path b in Fig. 2), and that the relationship between treatment and symptom reduction is accounted for, in whole or in part, by changes in the circuit (path c’ in Fig. 2). Thus, there is a great deal of research showing that RDoC components are affected by treatments, but there is less information about whether these components are treatment mediators.

Table 1 provides examples of genes, molecules, circuits, and physiology that show promise as possible moderators of EST outcome, as well as examples of these constructs that show promise as possible mediators of clinical change in ESTs.

Several studies have demonstrated that genetic factors predict the outcome of ESTs for negative valence disorders. In particular, researchers have focused on a polymorphism of the serotonin transporter gene (5-HTTLPR), which predicts response to antidepressant medications (Serretti et al., 2007). Findings are mixed regarding prediction of psychological treatments. In one study of CBT for PTSD (Bryant et al., 2010), patients with lower expression serotonin-transporter genotypes showed limited treatment gains at post-treatment and six-month follow-up. A different result was found, however, in a study of patients with panic disorder and agoraphobia receiving exposure therapy (Knuts et al., 2014), in which, after controlling for pre-treatment avoidance behavior, patients with the 5-HTTLPR low-expression genotype exhibited a stronger response to treatment. Among children diagnosed with anxiety disorders, those with the short-short 5HTTLPR genotype were significantly more likely to respond to CBT than were those with a long allele (Eley et al., 2012).

Other researchers have investigated a polymorphism of the catechol-o-methyl transferase gene (COMTVal¹⁵⁸Met) that results in higher extracellular dopamine levels in the prefrontal cortex and may inhibit fear extinction (Lonsdorf et al., 2009). In a study of patients with panic disorder (Lonsdorf et al., 2010), those with the COMTVal¹⁵⁸Met polymorphism showed a poorer response to exposure-based treatment. In a study of patients with depression, however, the COMTVal¹⁵⁸Met polymorphism did not predict response to CBT (Light et al., 2007).

Another example involves the BDNF Val66Met allele, which has been implicated in harm avoidance in animal models and in some human studies (e.g., Montag et al., 2010) (but see also Gatt et al., 2015 meta-analysis with non-significant association found). In one study of individuals with OCD who previously had an unsuccessful trial of pharmacology, CBT (primarily exposure) was administered (Fullana et al., 2012). They found that participants with the BDNF Val66Met allele had overall poorer outcomes in CBT (36% responders) compared to those without the gene (60% responders).

The investigators examined symptom subtypes and found that treatment response was poorest in individuals with contamination fears and washing compulsions who also lacked this allele. This study documents a potential mediator, since all the participants were previous medication nonresponders. A fuller investigation of the mediating effects of the BDNF Val66Met allele would require the addition of a different treatment group (such as, in this case, a medication condition).

Thus, the degree to which genetic factors predict the outcome of empirically supported psychological treatments is currently unclear. It may be that certain polymorphisms only have predictive value when combined with other biological, clinical, and demographic factors, as was found in a sample of anxious children (Hudson et al., 2013), or when multiple genetic factors are taken into account, as was found in a study of adults with depression (Keri et al., 2014). It is also possible that some genetic factors have high specificity in expression of psychopathology, such as shown with the BDNF Val66Met allele in the case of not just OCD, but in particular with contamination-based symptoms (Fullana et al., 2012).

The search for genetic moderators of treatment requires that a given polymorphism, be it 5-HTTLPR, COMTVal¹⁵⁸Met, or others, requires that the polymorphism not only predict symptom reduction over the course of treatment, but that it also shows an interactive effect with that treatment on symptom outcomes. Ideally, this would be done in a randomized trial in which some patients (with and without the polymorphism) received an EST and others (with and without the polymorphism) received some alternative treatment (e.g., a medication). If the polymorphism showed differential effects within different treatments (and vice versa), treatment moderation would be demonstrated, which would facilitate patient-treatment matching.

Although genes have been most commonly examined as pre-treatment predictors, some researchers have examined the extent to which psychological treatments might affect the epigenetic expression of certain genes. One example is the FKBP5 gene, which regulates glucocorticoid receptor sensitivity and influences the hypothalamic-pituitary-adrenal (HPA) stress response (Zannas and Binder, 2014). Research on the epigenetic effects of ESTs approaches a test of mediation by demonstrating not only that FKBP5 expression is affected by treatment (path a in Fig. 2), but also that changes in FKBP5 expression are associated with symptom reduction (path b in Fig. 2). In a study of PTSD patients, at baseline these patients showed lower FKBP5 gene expression compared to trauma-exposed subjects without PTSD. After receiving CBT, patients showed an increase in FKBP5 expression. Importantly, change in FKBP5 expression predicted improvement in PTSD symptoms (Levy-Gigi et al., 2013). In a study of children with anxiety disorders receiving CBT, patients with the greatest reduction in symptom severity showed a decrease in percentage DNA methylation of FKBP5 during treatment, and this effect was driven by those with one or more FKBP5 risk alleles (Roberts et al., 2015).

Thus, FKBP5 expression shows promise as a potential mediator of ESTs. A full test of mediation would be achieved by demonstrating that change in FKBP5 expression accounts, in full or in part, for the relationship between treatment and symptom reduction.

We will focus this discussion on primarily on one particular molecule, cortisol, which is one of the most widely-used biomarkers of hypothalamic-pituitary-adrenal (HPA) axis activity. Elevated cortisol levels are common in anxiety disorders (Vreeburg et al., 2010) and depression (Stetler and Miller, 2011), and some research suggests that the degree of cortisol release to initial exposures (potentially a marker of fear activation) has predictive value in exposure therapy. Specifically, in patients with panic disorder (Siegmund et al., 2011) and in patients with PTSD (S. A. Rauch et al., 2015), patients who exhibited greater cortisol release during the initial exposure showed a stronger response to exposure-based therapy than did those who released less cortisol.

Elevated cortisol may be associated with EST response. Among older adults with depression, higher salivary cortisol levels and a flatter diurnal slope of cortisol were associated with a poorer response to CBT (Holland et al., 2013). Similarly, among hospitalized patients with depression, elevated urinary free cortisol excretion was associated with a lower rate of CBT response; that relationship was not explained by clinical measures of symptom severity (Thase et al., 1996). In a study of patients with remitted depression who were given preventive cognitive therapy vs. treatment as usual, there was some evidence that higher cortisol levels predicted earlier time to relapse, whereas lower cortisol levels predicted earlier time to relapse in patients treated as usual (Bockting et al., 2006).

Studies of patients with GAD demonstrate that patients treated with CBT show a decrease in cortisol over time. In one study, plasma cortisol levels decreased in patients receiving cognitive therapy but not in patients receiving treatment as usual (Tafet et al., 2005); however, a lack of random assignment precludes interpretations of an interaction with treatment. In a study of older adults with GAD, patients receiving escitalopram plus CBT showed a greater decrease in peak (morning) cortisol levels over time than did patients receiving escitalopram alone; however, changes in cortisol levels were not significantly associated with changes in GAD symptoms (Rosnick et al., 2016). In an interesting twist on this research, Lass-Hennemann and Michael (2014) demonstrated that for patients with spider phobia, exposure therapy was more effective when conducted in the morning, when cortisol levels were highest, than in the evening, when cortisol levels were lower.

In contrast to anxiety and depression, some studies suggest lower baseline cortisol levels in PTSD (Meewisse et al., 2007). In a study of exposure therapy for survivors of the 2001 attack on the World Trade Center, conducted 3–12 months after the attack, urinary cortisol levels declined over time among patients nonresponsive to treatment, suggesting a progression to chronic PTSD, whereas they remained stable in treatment responders (Yehuda et al., 2009).

While cortisol has been most extensively examined in relation to negative valence systems, there are several other molecules that have been hypothesized to influence depression and anxiety, and may mediate treatment outcome. As the research on these molecules is fewer, we have reviewed them briefly here. One set of molecules, the catecholamines (e.g., epinephrine, norepinephrine, metanephrine) have been long implicated in depressed mood (Baldessarini, 1989), with lower levels associated with greater symptom severity. One study found changes in metanephrine over a twelve week course of cognitive therapy for depression, but that these changes were not associated with degree of symptom improvement (Free et al., 1998). Another molecule that is present in lower levels in depression, γ-aminobutyric acid (GABA; Brambilla et al., 2003), has been examined for its possible mediating relationship in treatment. In one small study, individuals were treated with cognitive therapy (Sanacora et al., 2006). The results suggested that CT had limited impact on cortical GABA. However, these findings should be interpreted with caution, given that nearly half of the individuals enrolled in the study dropped out before completing therapy. In another study (Abdallah et al., 2014), change in GABA was unrelated to outcome in a group of depressed patients receiving CBT (n=30) or SSRI medication (n=9). However, glutamate levels were lower in CBT responders, and unchanged in CBT nonresponders. Finally, in a small study, it was found that glutamate metabolites were significantly lower in individuals with OCD at baseline (n=8) than controls, and increased significantly following intensive (90-min/day, 5 days/week, over 4 weeks) CBT. From this limited pool of studies, it appears that while there are molecules associated with negative mood states, it is unclear whether these mediate treatment outcome. These studies all identify potential mediators of outcome in the molecules investigated, but only one fully evaluated mediation (Abdallah et al., 2014) through the differential change in GABA and its associated outcome with CBT or SSRI pharmacotherapy. Thus, there are promising potential mediating molecules in empirically supported therapies, but none that have yet been identified.

Research on the predictive value of pre-treatment neural circuitry in anxiety disorders has largely focused on limbic structures (e.g., amygdala), which are associated with fear responses (Dilger et al., 2003), as well as prefrontal cortical regions that may serve to inhibit amygdala responsiveness (Delgado et al., 2008). In one study, patients with social phobia underwent functional magnetic resonance imaging (fMRI) scanning prior to treatment while viewing pictures of fearful and angry (vs. happy) faces. After receiving CBT, those with higher pre-treatment dorsal anterior cingulate cortex (ACC) and dorsomedial prefrontal cortex (dmPFC) activity showed the strongest response to treatment (Klumpp et al., 2013). In a study of children with mixed anxiety disorders (primarily generalized anxiety disorder), patients underwent fMRI scanning prior to treatment while attending to pictures of fearful vs. happy faces. Those with greater pretreatment activation in the left amygdala during the task showed a stronger response to CBT (McClure et al., 2007). In another study, patients with panic disorder received fMRI scans while performing an emotion regulation task in response to negatively-valenced pictures. After a brief course of CBT, those with greater activation in insula and left dorsolateral prefrontal cortex (dlPFC) showed the greatest improvement in symptoms (Reinecke et al., 2014). In a larger panic disorder study, treatment response was associated with an inhibitory functional coupling between the ACC and amygdala (Lueken et al., 2013).

In the area of OCD, predictor research has largely focused on the cortico-striato-thalamo-cortical (CSTC) circuit (S. L. Rauch and Carlezon, 2013; Saxena and Rauch, 2000), in which the orbitofrontal cortex (OFC), striatum (including caudate nucleus), and thalamus are thought to be hyperconnected, activating one another in a continuous loop and resulting in intrusive mental activity (Beucke et al., 2013, Jung et al., 2013). In one study, OCD patients with contamination fears underwent fMRI with symptom provocation prior to CBT. Greater pre-treatment activity in the anterior temporal pole and amygdala, and lower pre-treatment activity in dlPFC, was associated with stronger treatment response (Olatunji et al., 2014). This demonstrates prediction, but not moderation. However, two studies come closer to demonstrating moderation of EST effects. In one, OCD patients received behavior therapy after receiving [18 F]fluorodeoxyglucose positron emission tomography (FDG-PET) scans. Those who exhibited higher metabolism in OFC at pre-treatment showed a stronger response to behavior therapy. Interestingly, for a second group of OCD patients treated with fluoxetine, the opposite pattern was obtained, with higher OFC metabolism associated with poorer outcome (Brody et al., 1998). This study demonstrates that pre-treatment OFC metabolism interacted with treatment in its impact on symptoms. An absence of random assignment to behavior therapy vs. fluoxetine, however, limits the conclusions that can be drawn. That problem was remedied in a structural MRI study, in which symptom improvement following CBT was significantly correlated with larger pretreatment gray matter volume within the right medial PFC, and a different pattern was demonstrated for patients randomized to fluoxetine, for whom smaller pretreatment gray matter volume in right middle lateral OFC was associated with a stronger response (Hoexter et al., 2013).

PTSD research has emphasized cortico-limbic abnormalities thought to underlie fear conditioning. In particular, it is thought that the medial prefrontal cortex is under-engaged, resulting in decreased inhibition of fear signals from the amygdala and related structures (Bremner et al., 1999, Charney et al., 1993). In one study, patients underwent fMRI while viewing fearful (vs. neutral) facial expressions before receiving CBT. Those with greater amygdala and ventral ACC activation to fearful faces showed a poorer response to treatment (Bryant et al., 2008).

Depression imaging research has emphasized cortical and limbic regions associated with depressed mood. Specifically, depressed mood is characterized by decreased dorsolateral and ventral prefrontal cortical activity, as well as increased activity in the amygdala, orbitofrontal cortex, and thalamus (Drevets, 2000, Sheline, 2003). In a study of patients with MDD, patients underwent fMRI while viewing negatively-valenced (vs. positively- and neutrally-valenced) words prior to receiving CBT. The strongest treatment response was seen in patients who showed low sustained reactivity to negative words in the subgenual cingulate cortex, and high reactivity in the amygdala (Siegle et al., 2006). Conversely, Fu et al. (2008) found that greater pre-treatment ACC response was associated with a stronger response to CBT, and no relationship was found between amygdala activity and treatment response. In another study, patients with MDD received pre-treatment fMRI scanning while viewing negative (vs. positive and neutral) pictures. After receiving CBT, it was found that patients with greater baseline activity in ventromedial PFC (vmPFC), anterior temporal lobe, and dlPFC showed the strongest response to treatment (Ritchey et al., 2011). In a sample of adolescents with MDD, those with greater striatal activity and lower medial PFC activity during a monetary reward task showed the strongest response to CBT (Forbes et al., 2010).

It therefore appears that there is some evidence for neural circuits that serve as treatment moderators. The available evidence is suggestive of a treatment-moderating effect of prefrontal neural activity in OCD, with greater OFC activity and larger medial PFC volume at pre-treatment showing an interactive effect of EST vs. fluoxetine. Within the areas of anxiety disorders, PTSD, and depression, there is abundant evidence of treatment prediction, with the general trend suggesting that greater baseline activity in fear structures such as ACC, insula, and amygdala, and greater activity in inhibitory frontal regions, predicts better outcome of ESTs for anxiety and depression, although an opposite pattern was found for PTSD. These promising results would translate into a determination of treatment moderation if these circuits and regions of interest showed an interactive effect with treatment, meaning they were predictive of outcomes for some treatments but not others.

A growing body of research demonstrates that ESTs affect neural circuitry. Within the anxiety disorders, exposure-based therapy is based on fear extinction, in which limbic responses habituate (Gottfried and Dolan, 2004, Knight et al., 2004, LaBar et al., 1998, Phelps et al., 2004). Fear extinction also engages frontal regions, particularly lateral (LaBar et al., 1998, Molchan et al., 1994, Yaguez et al., 2005), dorsomedial and ventromedial (Phelps et al., 2004, Yaguez et al., 2005), and orbitofrontal (Finger et al., 2008, Gottfried and Dolan, 2004, Hugdahl et al., 1995) structures, which likely play an inhibitory role (Delgado et al., 2008). Extensions of this basic research to clinical trials of ESTs largely show the same pattern. In studies of patients with spider phobia, those receiving exposure therapy showed greater reductions of hyperactivity in the insula and ACC (Straube et al., 2006) and increased medial OFC activity (Schienle et al., 2007) while viewing videos or pictures of spiders, compared to patients on a wait list. The Schienle et al. (2007) study goes a step further by demonstrating that reductions in anxiety symptoms were positively correlated with activation decreases in amygdala and insula.

In an FDG-PET study, patients with panic disorder were randomized to CBT or antidepressant medications. In the CBT group, metabolism decreased in right inferior temporal gyrus and superior and inferior frontal gyri, and increased in left inferior frontal gyrus, middle temporal gyrus, and insula (Prasko et al., 2004). In the medication group, metabolism decreased in right superior, middle, medial and inferior frontal gyri and superior and middle temporal gyri, and increased in left medial and middle frontal gyri and superior, middle and transverse temporal gyri. Although two different treatments were used, interaction effects were not examined. Furthermore, the relationship between changes in these neural regions of interest and symptom reduction was not assessed. Therefore, it cannot be determined whether changes in these regions of interest mediated CBT response. In a subsequent open trial of CBT for panic, 11 of the 12 patients who showed improvement after CBT exhibited decreased metabolism in right hippocampus, left ACC, left cerebellum, and pons, and increased metabolism in bilateral medial PFC (Sakai et al., 2006).

Among patients with social phobia randomized to CBT or citalopram, oxygen 15-labeled water PET scans were used to detect regional cerebral blood flow (rCBF) during a symptom provocation task. Response to CBT was associated with decreased rCBF in bilateral amygdala and hippocampus, and in the periamygdaloid, rhinal, and parahippocampal cortices. These changes were comparable to those seen in patients treated with citalopram (Furmark et al., 2002). In the absence of an interaction effect, although it might be concluded that these changes in rCBF are markers of improvement, they cannot be interpreted as CBT mediators. An fMRI study of socially phobic patients receiving CBT showed reductions in insula, dmPFC, and OFC; however, these changes did not correlate significantly with changes in symptom severity (Klumpp et al., 2013).

In OCD research, exposure-based therapy has been demonstrated to alter activity in the CSTC circuit described above. In fMRI research, it was shown that exposure therapy resulted in decreased frontostriatal activation during a probabilistic learning task (Freyer et al., 2011). In children with OCD who underwent fMRI during a test of executive function, activity in dorsolateral prefrontal and parietal regions was low compared to a group of healthy controls. Following CBT, activity in these regions increased (Huyser et al., 2010). Some recent studies come closer to establishing treatment mediation. In one, OCD patients receiving intensive exposure therapy showed significant FDG-PET decreases in bilateral thalamic activity, and significant increases in ACC activity, and these changes were significantly associated with symptom improvement (Saxena et al., 2009). In a longitudinal investigation in which fMRI scans were conducted at baseline, midway through treatment, and at posttreatment, symptom change was incrementally associated with decreased activity during symptom provocation in ACC and left OFC (Morgieve et al., 2014).

PTSD research has also suggested a possible mediating effect of neural circuitry. In a small study of patients with PTSD, fMRI scans were performed while patients viewed pictures of fearful (vs. neutral) faces before and 6 months after CBT. At post-treatment, compared to pre-treatment, patients showed increased activation of ventral ACC, left middle temporal gyrus, right inferior frontal gyrus, left parietotemporal gyrus, and right hippocampus; and decreased activation in right postcentral gyrus, right middle temporal gyrus, and left superior temporal gyrus. Symptom reduction was associated with increases in rostral ACC activity, and decreases in amygdala activity (Felmingham et al., 2007).

Research evaluating circuitry in the treatment of depression has examined changes in brain activation during exposure to sad theme pictorial images (such as faces). In one study employing a brief behavioral activation intervention, results showed significant changes in several regions associated with cognitive control and emotion regulation, notably the right OFC and paracingulate gyrus (Dichter et al., 2010). Another study found that CBT resulted in increases in vmPFC activation (Ritchey et al., 2011); however, these studies did not assess the relationship between changes in these regions of interest and changes in depressive symptoms. A closer step to identifying a treatment mediator is found in an earlier study of FDG-PET before and after treatment with CBT vs. venlafaxine (Kennedy et al., 2007). In that study, response to either treatment was associated with decreased metabolism in OFC and left medial prefrontal cortex. However, response to CBT was also associated with decreased metabolism in posterior cingulate and increased metabolism in left inferior temporal cortex, whereas response to venlafaxine was associated with changes in the opposite directions in those regions. Because treatment responders and nonresponders were analyzed separately, it is difficult to ascertain the relationship between metabolic change and degree of symptom improvement. A relationship between neural change and symptom change was found, however, in a study using an in-scanner emotionally relevant task (Yoshimura et al., 2014). Results showed that symptom reduction following CBT was associated with reduced activation in medial PFC and ventral ACC.

These results point to several promising candidates for EST mediators. In the anxiety disorders, studies tend to show that various regions of interest are affected by ESTs; specifically, CBT tends to result in decreased activity in fear circuitry (e.g., insula, ACC) and increased activity in frontal regions (e.g., OFC, PFC), although some exceptions are noted. Importantly, the changes in fear-related regions are correlated with changes in anxiety symptoms. A different profile is seen in OCD and depression, in which treatment generally seems to decrease activity in certain frontal regions, and those decreases are correlated with symptom reduction. The next step in this line of research would be to demonstrate that changes in these circuits account, in full or in part, for the relationship between treatment and symptom reduction.

Peripheral physiology research in anxiety disorders has focused on signs of sympathetic arousal, including heart rate and heart rate variability, skin conductance, electromyographic measures of startle, and hyperventilation. Interest in cardiac factors as predictors of EST outcome peaked with the emergence of emotional processing theory (Foa and Kozak, 1986, Lang, 1979), which posited that the degree of evoked arousal during exposure therapy would predict clinical outcomes. Some research has demonstrated that peak heart rate response during the initial exposure predicted the eventual symptom reduction in patients with specific phobia (Alpers and Sell, 2008, Lang et al., 1970; J. P. Watson and Marks, 1971), OCD (Kozak et al., 1988), and PTSD (Pitman et al., 1996). However, subsequent research raised questions about the predictive value of initial physiologic activation in exposure (see Craske et al., 2008, for review). Studies in specific phobia (Telch et al., 2004), panic disorder (Meuret et al., 2012), and agoraphobia (Michelson et al., 1990) have failed to find that early peak heart rate predicts the eventual response to treatment. Other cardiac variables, such as heart rate variability, have yielded mixed results as predictors of outcome (Bornas et al., 2007, Davies et al., 2015).

Respiratory variables, particularly those associated with hyperventilation, may predict EST outcome. In a sample of patients with mixed anxiety disorders, low baseline levels of end-tidal partial pressure of carbon dioxide (pCO₂), a sign of hyperventilation, predicted poorer response to both CBT and Acceptance and Commitment Therapy (Davies and Craske, 2014). Curiously, however, low baseline pCO₂ did not predict outcome of a breathing retraining intervention for patients with panic disorder (Meuret et al., 2008).

Although there is little consistent evidence to suggest that baseline peripheral physiology predicts EST outcome (at least for the anxiety disorders in which they have been studied), there is a wealth of evidence suggesting that these variables are affected by the ESTs (see Goncalves et al., 2015, for review). In early research on specific phobias, Lang et al. (1970) found that heart rate and skin conductance decreased over the course of exposure therapy; furthermore, the degree of heart rate decrease (though not the degree of skin conductance decrease) was related to the degree of self-reported fear reduction. Studies of patients with panic disorder have yielded mixed results in terms of heart rate decrease; some have found a decrease over the course of treatment (Garakani et al., 2009, Mavissakalian and Michelson, 1982, Roth et al., 1987), while others have not (Meuret et al., 2012). Heart rate variability was not affected by exposure therapy for claustrophobia (Kamphuis and Telch, 2000), but increased following CBT for panic disorder (Garakani et al., 2009). In CBT for depression, heart rate variability improved over the course of treatment for severely depressed patients, but not for mildly depressed patients (Carney et al., 2000).

Examinations of within-session habituation of physiological response have yielded mixed results. In one study of in vivo exposure for fear of flying, those patients who eventually flew after the treatment had exhibited a greater reduction in heart rate from the beginning to the end of the treatment flight (Beckham et al., 1990), although other factors such as socioeconomic status could also have affected flight likelihood. In another study, within-flight habituation of heart rate was not associated with clinical measures at follow-up, or with flying behavior after the treatment (Busscher et al., 2015).

In studies of PTSD, heart rate reactivity to trauma cues (usually comparing pre-cue heart rate to post-cue heart rate) has been examined as a potential mechanism of ESTs. Heart rate reactivity to a trauma imagery script decreases over the course of trauma-focused CBT (Dunne et al., 2012). Two studies also demonstrated that the degree of decrease in heart rate reactivity was greater for patients receiving CBT than for patients receiving supportive therapy (Blanchard et al., 2002) or wait list (Blanchard et al., 2002, Rabe et al., 2006), and that changes in heart rate reactivity were significantly associated with decreases in PTSD symptoms.

Fear-potentiated startle, in which facial EMG response to a sudden (usually acoustic) stimulus in the presence of fear cues, has been demonstrated to decrease following exposure therapy for spider phobia (de Jong et al., 1993, Kashdan et al., 2012). Unpotentiated startle, in which EMG startle response is measured in the absence of fear cues, has shown an association with CBT response. In children with mixed anxiety disorders, CBT responders showed a decrease in startle response over time, whereas nonresponders showed no decrease or even an increase in startle (Bakker et al., 2011). Among women with PTSD, startle response was reduced to a greater degree in CBT responders compared to nonresponders (Griffin et al., 2012).

Respiratory variables, particularly pCO₂, have been investigated as possible mediators of treatments for panic disorder. In a study of exposure therapy, pCO₂ did not improve over the course of treatment (Meuret et al., 2012). However, when a different treatment, capnometry-assisted breathing retraining, was used, changes in pCO₂ partially mediated changes in anxiety sensitivity, a cognitive vulnerability factor for panic (Meuret et al., 2009). In another study, pCO₂ was shown to improve for patients receiving the breathing retraining intervention, but not for patients receiving cognitive therapy. Furthermore, among patients receiving the breathing retraining intervention, changes in pCO₂ preceded and mediated changes in symptom appraisal and perceived control of symptoms (cognitive vulnerability factors), and were associated with subsequent changes in panic disorder severity (Meuret et al., 2010).

Cardiac variables, startle response, and respiratory abnormalities are all potential EST mediators. Most of our understanding of physiological changes with ESTs comes from the anxiety disorders, which are usually characterized by baseline sympathetic activation. We have substantial evidence (with some exceptions) that these physiological factors are affected by ESTs; exposure and CBT treatments generally decrease heart rate and heart rate reactivity to fear cues, may increase heart rate variability, decrease baseline and fear-potentiated startle response, and increase pCO₂. We further have evidence that some of these changes are linked to clinical outcomes; in particular, changes in heart rate reactivity, startle response, and pCO₂ correlate with decreases in anxiety-related symptoms. Formal tests of mediation are sparse, although Meuret et al. (2010) finding that pCO₂ mediates changes in certain panic-related symptoms is promising.

The research reviewed here is revealing in several important ways. First, while RDoC assigns prominence to genetic research, it appears that there is very little empirical support for genes in moderating or mediating outcome for cognitive-behavioral interventions. Promising potential genes or polymorphisms have been suggested in prior research, such as 5-HTTLPR, COMTVal¹⁵⁸Met, or BDNF Val66Met (moderators), or FKBP5 (mediator). This may not be surprising given the comparably limited support found for genetic contributions generally in larger genome wide association studies evaluating genes or polymorphisms in depression (Hek et al., 2013), anxiety (Otowa et al., 2016), or OCD (Stewart et al., 2013). It appears that no gene or genotype is solely associated with treatment outcome with CBT or exposure, but instead complex interactions between genes may contribute to outcome (Eley et al., 2012). Accordingly, conceptualization of psychopathology and putative moderators or mediators may require combinations of genes or polymorphisms. This may be the more sensible next step in understanding how genetic makeup might predict outcome with cognitive-behavioral therapy. This would also be in line with the call for more complex modeling in understanding the role of genes and heritability in psychiatric conditions generally (Kendler, 2002). Research into the role of specific molecules as moderators or mediators of outcome is comparable to that observed for genes. Namely, there are molecules that may be promising moderators (cortisol) or mediators (cortisol, catecholamines, glutamate metabolites, GABA), but most of the research has either been purely predictive in nature, or only suggest mediation.

Research on circuitry may hold greater immediate promise in understanding the relationship between brain activation and treatment outcome. Assuming that progress will be made in the ability to conduct fMRI or other brain scans, it is conceivable that models of brain circuitry associated broadly with fear or anxious-misery conditions that undergird neurotic conditions could be conducted to determine likely success in treatment or risk for relapse. This would be a condition where circuitry fulfills its role as a moderator of treatment. On the other hand, models of psychopathology have promoted the potential to rely on circuitry as a mediator of treatment, whereby recruitment of specific brain areas as part of cognitive-behavioral interventions could be tracked in order to determine likelihood of outcome and long-term maintenance of gains. For both anxiety and depressive disorders, there appear to be several broad areas, such as regions of the prefrontal cortex, anterior cingulate cortex, and amygdala that are associated with treatment outcome. Collectively, moderator and mediator research on genes, molecules, and circuitry in evidence-based therapy would fulfill an assumption that undergirds that RDoC model, as articulated by Insel et al. (2010): “…the RDoC framework assumes that data from genetics and clinical neuroscience will yield biosignatures that will augment clinical symptoms and signs for clinical management” (p. 749). This assumption underlying the RDoC initiative may be the most explicit part that calls for identification of moderators and mediators of treatment outcome through biomedical and neuroscience methods.

Physiology has the promise for the most immediate treatment relevant value, and yet is the least researched of the domains covered in this essay. At present there is readily available equipment that could be used to assess treatment relevant parameters of symptom response. This line of work may also be most fiscally possible in future research given the ease with which these areas can be assessed, with comparably little specialized training, low participant risk and discomfort, and widely available software for data reduction and interpretation. High priority should be given to research into putative physiological moderators of treatment outcome. This domain warrants specific attention as a mediator of outcome, given that models of treatment involving exposure are predicated on activation of arousal in order to facilitate change (Foa and Kozak, 1986, Lang, 1979).

It is important to note that the research reviewed here is limited in its generality to RDoC constructs. This review conceptualized a set of existing diagnoses in the DSM system as putative Negative Valence System conditions. This proxy assignment was for convenience and due to the general defining features of each condition that included negative valence components (i.e., acute, potential, and sustained threat, loss, frustrative non-reward). The impetus of RDoC, however, is to develop a more reliable and valid taxonomy, and as a result future research will be warranted that tests moderation and mediation where participant recruitment involves screening specifically for these constructs rather than according to the existing DSM criteria. Accordingly, the lack of moderation or mediation found for the majority of the domains evaluated could be a result of error variance inherent in the diagnoses. In addition, the areas reviewed here could also be conceptualized in relation to developmental trajectories. For example, Brent et al. (2010) found that FKBP5 polymorphisms were not associated with treatment outcome in adolescents treated for depression. This candidate gene was identified in this review as a potential mediator of outcome for PTSD. Accordingly, in this example, the influence of developmental context as well as specific genes, or more generally units of analysis in the RDoC matrix, should be evaluated.

Further, in light of space constraints, the disorders surveyed here were conceptualized as specifically members of the negative valence system. However, it is reasonable to consider that many of these same disorders include processes under other elements of the RDoC matrix. Notably, most of the disorders here have symptoms demarcated under Arousal/Regulatory system (sleep disturbance; Smith et al., 2005). Additional symptoms within other aspects of the RDoC matrix are relevant to the disorders reviewed here, such as: Positive Valence System (e.g., lack of sensitivity to reward in OCD; Palminteri et al., 2012); Cognitive System (e.g., working memory in depression and posttraumatic stress; Schweizer and Dalgleish, 2016); and Social Processes (e.g., understanding of mental states in social anxiety disorder; Washburn et al., 2016).

Until recently, ESTs have been developed to target specific DSM diagnoses. This led to challenges in treatment selection and application given the proliferation of protocols (Tolinet al., 2015b). Transdiagnostic treatment models, such as the Unified Protocol (Payne et al., 2014) attempt to address this problem by targeting the central mechanism(s) of psychopathology (e.g., neuroticism) rather than putative diagnostic criteria. Further research into moderators and mediators has potential to refine treatment delivery through these transdiagnostic protocols and identify the active ingredients of therapy.

As noted earlier, the recently revised standards for empirically supported treatments emphasize clinically-relevant outcomes, demonstrated maintenance of treatment gains, and functional outcomes in determining a recommendation for wider application of any protocol (Tolin et al., 2015b). In its current form, the RDoC delineates units of analysis that can be evaluated in the context of different functioning areas. These units of analysis can be construed as possible moderators or mediators of outcome. It is also possible that these could be outcomes in their own right. At present, the majority of empirically supported treatments have accumulated efficacy evidence based on self-report instruments. However, in the event that other units of analysis (i.e., genes, molecules, circuits) are found to be robust indicators of outcome, these could be the assessments that serve as the basis for declaring interventions empirically supported.

Treatment research provides important additional information regarding basic psychopathological processes. To cite one example, differential treatment outcome for symptom manifestations of OCD (i.e., better outcome for washing/contamination fears versus hoarding; Abramowitz et al., 2003) under prior editions of the DSM led to research that showed hoarding is syndromally separate from other manifestations of the disorder (Pertusa et al., 2010). This ‘discovery’ of mechanisms through the conduct of controlled research trials has been suggested elsewhere as part of a research agenda aimed at developing ESTs (Weisz et al., 2000). This necessarily requires that evidence accumulate through a discovery phase followed by a priori verification tests.

In light of the current review, as it relates to negative valence systems, it is clear that a great deal of additional research is warranted before genes, molecules, circuits, or physiology may be robust moderators or mediators of outcomes, which has been noted as an important potential goal of the RDoC initiative (e.g., Craske, 2012). Finally, given that translational research has the potential to evaluate multiple units in the RDoC matrix concurrently, it is possible to evaluate multiple moderators and mediators to empirically test the assumed hierarchical nature of the units of measurement. That is, it is possible that circuits are robust moderators of treatment outcome, molecules are significant mediators, and genes are not contributors to either. This would then address criticisms leveled at the RDoC initiative (e.g., Weinberger et al., 2015) and potentially allow for identification of the most efficient methods of assessing moderators or evaluating mediators to facilitate treatment development and planning.