Limbic–frontal circuitry in major depression: a path modeling metanalysis
Introduction
To test a previously proposed limbic–cortical network model developed with positron emission tomography (PET) measures of brain glucose metabolism (see Fig. 1), we present findings of an across-lab metanalysis examining effective connectivity in major depression (MDD) using Structural Equation Modeling.
Depression is a common affective disorder characterized by persistent negative mood and selective deficits in cognitive, circadian, and motor functioning. Neuroimaging studies of both cerebral blood flow (CBF) and glucose metabolism (FDG) have repeatedly identified regional abnormalities in the untreated depressed state Baxter et al., 1989, Bench et al., 1992, Drevets et al., 1992, Mayberg et al., 1994. Consistently reported are frontal and cingulate changes. While less common, other limbic and subcortical regions including hippocampus, amygdala, posterior cingulate, striatum, and thalamus are also implicated. Frontal and cingulate changes involve multiple but distinct sites, with anatomical convergence across functional, structural, and post-mortem pathological studies: dorsolateral (BA9/46) and ventrolateral prefrontal cortex (BA10/47), dorsomedial and ventromedial frontal cortex (BA9/10/11/32), and dorsal, rostral, and subgenual cingulate (BA24b, 24a, 25). In addition to abnormalities identified in the pretreatment depressed state, changes in many of these same regions are seen with various types of pharmacological, cognitive, and somatic antidepressant treatments Brody et al., 1999, Brody et al., 2001, Goldapple et al., 2004, Kennedy et al., 2001, Mayberg et al., 2000, Pizzagalli et al., 2001a.
While there is growing consensus that an array of brain areas are involved in depression, not all regions are reported in all studies. Furthermore, there is variability in the direction of CBF and FDG changes, particularly in those areas of frontal and anterior cingulate cortex considered most critical. Our own studies, as well as others Brannan et al., 2000, Pizzagalli et al., 2001a, demonstrate distinct differences in anterior cingulate activity that distinguish eventual responders and nonresponders scanned before pharmacotherapy Mayberg et al., 1997, Wu et al., 1999. As well, change patterns in responders and nonresponders to identical treatment show mirror metabolic change patterns in certain specific regions, including regions unaffected in the baseline state (i.e., posterior cingulate, and hippocampus). In addition, there are clear regional change pattern differences across unique treatments (CBT, medication, ECT) affecting similar regions in different ways (i.e., frontal decreases and hippocampal increases with CBT; frontal increases and hippocampus decreases with pharmacotherapy; decreases in both with ECT) Buchsbaum et al., 1997, Goldapple et al., 2004, Henry et al., 2001, Kennedy et al., 2001, Martinot et al., 1990, Mayberg, 2003, Nobler et al., 2001. Unique treatment-specific effects are also seen; notably brainstem and thalamic changes with medication and medial and orbital frontal changes with CBT. In general, subdivisions of cingulate, frontal cortex, hippocampus, and thalamus figure predominantly across all studies.
The nature of these reported changes suggests a complex interaction between the pre-treatment brain state, brain responsivity, and different treatment interventions that is not intuitive. Nonetheless, the findings suggest a testable hypothesis that variations in the state of connections across a set of critical regions (effective connectivity), as measured using a multivariate technique, might better explain the reported variability across independent depression patient treatment samples than the more typical approaches examining relative differences in discrete regions among groups (change distribution analysis, SPM) Friston, 1994, Horwitz et al., 1999.
To provide further theoretical context for such an approach, a major depressive episode is considered, at the brain level, the net result of maladaptive functional interactions among a highly integrated network of limbic–cortical regions normally responsible for maintaining homeostatic emotional control in response to cognitive and somatic stress Mayberg, 2003, McEwen, 2003. Network dysfunction combined with variations in active intrinsic compensatory processes might therefore account for heterogeneity of depressive symptoms observed clinically, as well as variations in pretreatment scan patterns described experimentally. Non-imaging studies implicate various contributors to these adaptive differences including genetic vulnerability, affective temperament, and developmental insults and environmental stressors Bagby et al., 2003, Caspi et al., 2003, Heim and Nemeroff, 2001. Treatments for depression can be similarly viewed within such a systems framework, whereby different modes of treatment facilitate recovery via initiation of additional adaptive chemical and molecular changes Hyman and Nestler, 1996, Vaidya and Duman, 2001. Progressively, more aggressive treatments needed to ameliorate symptoms in some patients may reflect poor adaptive capacity of this network in these patient subgroups.
To fully test such hypotheses, baseline patterns in patients with known clinical response to various treatments are first required. In addition, a more deliberate assessment of these state–region–treatment interactions is needed. The multivariate technique partial least squares (PLS) combined with structural equation modeling provides one such approach Horwitz et al., 1999, McIntosh, 2000, McIntosh and Gonzalez-Lima, 1994, whereby relationship between regions in a defined theoretical model can be tested across different patient cohorts where variations in treatment response are known.
This study had three main goals. The first and most critical was to create a formal depression model structure that would both represent our previous theoretical construct (Fig. 1) and show specificity and good reliability across multiple depressed patient samples. The second was to characterize path differences within such a model that would distinguish responders from nonresponders. Third, differences associated with response to different treatments were considered. We hypothesized that the combined use of PLS and SEM, constrained by an a priori focus on those brain regions consistently identified in past studies, and knowledge of known anatomical connections among these specified regions would identify such a model. It was further hypothesized that paths involving anterior cingulate would be critical to responder–nonresponder differences. While the design of this study is largely exploratory, it is seen as the first step to establishing a simplified model system for future prospective studies.
Section snippets
Subjects
This metanalysis combined resting state FDG-PET data from independent studies performed at two institutions (University of Texas Health Sciences Center at San Antonio and the University of Toronto). All data were acquired using similar recruitment inclusion and exclusion criteria and a comparable imaging protocol on a similar generation PET scanner and were thus, considered comparable for the proposed cross-group analyses. Baseline, pre-treatment scans from 119 patients with major depression
Results
A single model was created for all three depressed cohorts based on the criteria for model construction given above. Within the context of this model, paths differentiating groups as a function of treatment–response interactions were also identified, but not differences due to either cohort site (Toronto vs. San Antonio) or the specific treatment alone (Drug vs. CBT).
Discussion
These findings define, using structural equation modeling of resting state FDG-PET scan data, patterns of effective connectivity that differentiate distinct groups of unmedicated unipolar depressed patients. Furthermore, specific cortical–cortical, cortical–limbic, and limbic–subcortical path interactions within this specified depression model distinguished patient subgroups. Most interestingly, these differences were not simply gross cohort differences, but rather more complex interactions
Acknowledgements
The authors thank Steven Brannan MD, J. Arturo Silva MD, and Janet Tekell MD for assessing treatment response in the San Antonio patient group and Craig Easdon, PhD for his assistance with SEM.
References (102)
- et al.
Reduced glutamate in the anterior cingulate cortex in depression: an in vivo proton magnetic resonance spectroscopy study
Biol. Psychiatry
(2000) - et al.
Frontal and parietal metabolic disturbances in unipolar depression
Biol. Psychiatry
(1994) - et al.
Serotoninergic but not noradrenergic neurons in rat central nervous system adapt to long-term treatment with monoamine oxidase inhibitors
Neuroscience
(1985) - et al.
Brain metabolic changes in major depressive disorder from pre- to post-treatment with paroxetine
Psychiatry Res.
(1999) - et al.
Effect of sertraline on regional metabolic rate in patients with affective disorder
Biol. Psychiatry
(1997) - et al.
Psychosurgery
Neurosurg. Clin. N. Am.
(1995) - et al.
Stereotactic cingulotomy
Neurosurg. Clin. N. Am.
(2003) - et al.
Neural plasticity to stress and antidepressant treatment
Biol. Psychiatry
(1999) The neuropsychological profile in unipolar depression
Trends Cogn. Sci.
(1998)- et al.
The role of childhood trauma in the neurobiology of mood and anxiety disorders: preclinical and clinical studies
Biol. Psychiatry
(2001)