FABBS Doctoral Dissertation Research Excellence Award

Nick Jacobson, The Pennsylvania State University, Department of Psychology and Psychiatry

“Differences in Neural Activation in Anxiety and Depressive Disorders: An fMRI Meta-Analysis”


Anxiety and depressive disorders affect approximately one third of the population at some point during their lifetime (Kessler et al., 2005). Of those with major depressive disorder (MDD), 73% have comorbid lifetime anxiety disorders, and 27-77% of those with an anxiety disorder have a lifetime diagnosis of depression (Brown et al., 2001). Although either of these disorders alone is associated with significant distress or impairment, their co-occurrence is associated with greater duration and severity of each disorder, greater occupational and psychosocial impairment, lowered quality of life, increased suicidal ideation, and greater health care utilization (Cairney et al., 2008; Shankman & Klein, 2002; Steffens & McQuoid, 2005). As such, it is essential to understand this comorbidity.

Contemporary work has also focused on regrouping anxiety and depression into a single transdiagnostic factor, under the terms of internalizing disorders (Caspi et al., 2014), neuroticism (Kotov et al., 2010), and negative affect (Clark & Watson, 1991). In support of these views, some prior reviews have concluded that neurobiological mechanisms underlying anxious and depressive disorders may be the same (Nan Lv, Jun Liu, Jing Zhang, & Tzeng, 2013; Nutt, 2004) However, there have been considerable differences in each review’s conclusions with respect to the active brain areas associated with each diagnosis.

Although several non-systematic reviews, systematic reviews, and meta-analyses have examined the primary activation and deactivation of one particular disorder, only three reviews have examined or discussed similarities and differences of brain activation across two or more disorders (Etkin & Wager, 2007; Nan Lv et al., 2013; Nutt, 2004). Two non-systematic review drew conclusions about overlapping activation based on a fewer than six studies or less (Nan Lv et al., 2013; Nutt, 2004). Lastly, only one meta-analysis looked at the overlapping and unique activation of social phobia, specific phobia, and posttraumatic stress disorder (PTSD) and consequently was unable to establish overlapping versus unique activation comparing other anxiety disorders or any depressive disorders (Etkin & Wager, 2007). Consequently, no prior reviews or meta-analyses have comprehensively examined overlapping and unique activation of anxiety and depressive disorders.

The current study addressed these gaps by doing a systematic meta-analysis, including all anxiety and depressive disorders, comparing overlapping and distinctive activation areas, focusing on specific spatial coordinates, and summarizing across many more studies. The goal of the present study was to investigate the unique neuroactivation patterns consistently associated with a particular disorder across emotional and cognitive tasks. We examined the following questions: (1) For each anxiety and depressive disorder, which areas of the brain were associated with significant hyperactivation or hypoactivation, relative to controls? (2) Which areas or the brain included overlapping activation versus activation areas specific to each anxiety and depressive disorder? and (3) What was the percentage of total shared activation (i.e. activation in the same areas and in the same direction), versus total activation that discriminated between each pair of disorders?


Obtaining Articles.  We performed 252 keyword searches in PsycINFO and MEDLINE and examined over 20,000 abstracts and obtained 7,232 articles for a larger meta-analysis on overlap between anxiety and depression, searching three groups of keywords related to specificity/overlap, anxiety, and depression. Each combination of groups was performed in both PsycINFO and Medline.

Article Selection. Inclusion criteria for this study were: (1) assessed anxiety and/or depressive disorders using a diagnostic interview; (2) had at least one group that met diagnostic criteria for an anxiety or depressive disorder and reported the rates of anxiety and depressive disorders in the group; (3) used fMRI methods; and (4) contrasted the results with a control group who did not meet diagnostic criteria for anxiety or depressive disorders. Using these criteria, 99 articles were included.

Article Coding. Each of the articles was coded on the following characteristics: sample size of control and clinical groups, percentage of the sample with each disorder (including GAD, panic disorder, OCD, PTSD, social phobia, and specific phobia) and depressive disorders (including MDD and dysthymia). Note that percentages of diagnoses were required to ensure that the analyses examined the functional activity non-comorbid cases, and cases with comorbidity. We coded the type of task as an emotional or cognitive. We utilized t-values as our primary fMRI effect sizes. For each fMRI coordinate listed, we recorded x, y, and z coordinates for the areas of activation, and the atlas type (Talaraich or MNI). A team of five coders, coded each article.

Planned Analyses. Inter-rater reliability was calculated for all items in the analyses using bivariate intraclass correlations and kappa coefficients. All primary analyses used signed differential mapping (SDM 4.13). SDM creates a statistical map of reported coordinates in each study. SDM is also the only fMRI meta-analyses software that incorporates effect size statistics into peak coordinates. The primary analyses proceeded in two phases: (1) effects of each disorder were considered alone using meta-regression (i.e. this examined significant primary activation), and (2) two disorders were included in the model and post-hoc comparisons were examined using activation associated with one disorder subtracted from activation associated with another disorder, yielding a brain map of statistically significant differences between disorders.

Similar to prior fMRI meta-analyses examining other questions (Cortese et al., 2012; Wager, Jonides, & Reading, 2004), tables summarize the number of voxels that were significantly and occurring jointly in the same direction different between each disorder. To account for spatial impression, voxels were considered to overlap in the same direction if they occurred within 20 millimeter diameters of the other disorder, which is a method that has been used to account for spatial impression in both prior meta-analyses and simulation studies (Stanley et al., 2013). To quantify proportion of similarity between each disorder, we calculated percentage of voxels that were associated with significant differences, by dividing the number of voxels that were significantly different between each pair of disorders by the total number of voxels activated in each pair of disorders similar to Cortese et al., (2012).


Interrater Reliability. Bivariate intraclass correlations were excellent for the coding of anxiety disorders, (0.87-1.00, Median = 1.00), clinical group sample size (0.66-1.00, Median = 1.00), control sample size (0.87-1.00, Median = 1.00), percentage of depressive disorders (0.87-1.00, Median = 1.00), percentage of dysthymia (0.86-1.00, Median = 1.00), percentage of GAD (0.90-1.00, Median = 0.99), percentage of OCD (0.90-1.00, Median = 0.99), percentage of panic disorder (0.93-1.00, Median = 0.99), percentage of social phobia (0.91-1.00, Median = 0.98), percentage of specific phobia (0.91-1.00, Median = 0.98), and t-statistics for ROIs (0.76-1.00, Median = 0.94). Kappa coefficients for agreement of number of tesla (0.81-1.00, Median = 1.00), block/event design (1.00-1.00, Median = 1.00), whether the task was emotional or cognitive (0.67-1.00, Median = 1.00), whether the task was auditory (0.66-1.00, Median = 1.00), and whether the task was visual (0.70-0.89, Median = 0.78) all were good.

Similarities and Differences between Disorders. In the analyses, there were 187 persons with GAD, 143 persons with panic disorder, 207 persons with social phobia, 47 persons with specific phobia, 417 persons with OCD, 482 persons with PTSD, 805 persons with MDD, and 61 persons with dysthymia. The median of all the pairwise comparisons was of 16% similarities relative to differences in cognitive tasks. In the cognitive tasks, the results showed that across nearly every disorder, there were significantly more differences than there were more differences than there were similarities (which are indicated by similarities percentages below 50%). An exception occurred between dysthymia and GAD, which were more similar than they were different cognitively, and also between specific phobia and dysthymia which was also more similar than they were different cognitively.

In contrast, in emotional tasks, most disorders showed had an approximately equal degree of differences and overlap. The median of all the pairwise comparisons was of 47% similarities relative to differences in emotional tasks. This was particularly salient with panic disorder, social phobia, and dysthymia, which tended a greater degree of similarity than difference with other disorders. In contrast, MDD, OCD, GAD, PTSD, and specific phobia tended to be more dissimilar than they were compared to most other disorders.


Overall, these findings suggest that within cognitive tasks, disorders tended to show far more differences from one another than similarities. In contrast, within emotional tasks, disorders tended to show approximately equal levels of similarities to differences. These findings both complement and extend findings from prior qualitative reviews and meta-analyses (Etkin & Wager, 2007; Nan Lv, Jun Liu, Jing Zhang, & Tzeng, 2013; Nutt, 2004). These findings highlight that there is overlap between each pair of anxiety or depressive disorders, but disorders tend to have more in common during emotionally evocative times, compared to cognitively engaging times. In addition, in contrast some prior theories (Nan Lv et al., 2013; Nutt, 2004), the results also showed that each disorder exhibited significant differences from one another both cognitively and emotionally, suggesting that there are substantial differences between each anxiety and depressive disorder in both cognitive and emotionally evocative times.

In addition to the general pattern of moderate overlap between disorders emotionally, and distinct activation cognitively, it is important to examine which disorders were most different from one another, and which disorders showed the greatest similarity. Within emotional tasks, the disorders with the greatest similarities with other disorders were dysthymia (averaging 51% similarities relative to differences), panic disorder (averaging 50% similarities relative to differences), and social phobia (averaging 48% similarities relative to differences).

Instead of favoring generality, the current study provides evidence for criterion validity of the Diagnostic and Statistical Manual’s separation of disorders (American Psychiatric Association, 2013). This corroborates research that supports heterogeneity of the genetic structure of the anxiety disorders (Kendler et al., 1995). This is also consistent with clinical research (Goldberg, 2015; Huberty, 2012; Lang & McTeague, 2009) showing that these disorders reflect different pathological entities with different corresponding biological processes. Further our findings suggest that the 25 pages devoted to differential diagnoses in the DSM-5 among anxiety and depressive diagnoses may hold clinical utility (American Psychiatric Association, 2013).

Recent movements in psychopathology research have emphasized the importance of examining “cross-cutting” and “transdiagnostic” aspects of psychopathology. This movement has been largely propelled by the National Institute of Mental Health’s Research Domain Criteria (RDoc). The RDoc movement emphasizes the study of transdiagnostic processes due to recent failures to find specificity in neural activation across disorders (Etkin & Cuthbert, 2014). Nevertheless, the current findings suggest that there are important differences across disorders and ignoring these substantial differences may impair our account for the substantial heterogeneity across diagnosis. Although conflicting with RDoc’s origins of abandoning diagnoses (Cuthbert & Insel, 2013), the current results nevertheless support RDoc’s initiative of tying neurological activation areas to specific types of psychopathology (Weinberger & Goldberg, 2014).

Current findings may also suggest that specific treatments for disorders should not be abandoned in favor of transdiagnostic treatments (Ellard, Fairholme, Boisseau, Farchione, & Barlow, 2010). Nevertheless, these findings suggest that it would likely be more efficacious for any transdiagnostic therapies to focus on emotional, rather than cognitive components, given the greater degree of overlap in emotional, compared to cognitive domains.