Last week I discussed two reviews of neuroimaging studies attempting to untangle the biology of depression by studying activity of specific brain regions and neurotransmitter-associated proteins as well as at the prospect of identifying biomarkers to predict treatment response.
In addition to identifying patients who would respond to specific treatments, biomarkers have been sought to diagnose depression, to identify those at risk for whom prophylactic treatment might be appropriate, and to identify biological components or dimensions of depression.
Savitz and colleagues note that brain imaging has an established but small role in detecting conditions such as neoplasm, hematoma, hydrocephalus, cerebrovascular disease and gross atrophy in patients with psychiatric symptoms. It has not, however, been demonstrated to be effective in diagnosing mood disorders, predicting who in at-risk populations will become ill, or selecting treatment. Several studies have shown associations of significant utility (greater than 80% sensitivity and specificity) in single samples, but none has been replicated by another group of investigators using a different sample. They illustrate these issues of establishing diagnostic efficacy with the successful example of F-18 florbetapir (Amyvid) scans for β-amyloid protein, which has been approved for use in the evaluation of patients for Alzheimer’s disease.
A fundamental problem in using any biomarker for a psychiatric disorder is that, unlike Alzheimer’s disease, where the diagnosis can be confirmed at autopsy, the “gold standard” for psychiatric disorders is the DSM, which is based on signs, symptoms, and course of illness. Since clinically-defined disorders may comprise multiple distinct pathologies, judging the accuracy of a biomarker against clinical diagnoses amounts to trying to “forcibly align neurobiology with DSM diagnoses,” which Savitz and colleagues call “regressive.” The Research Domain Criteria, which I wrote about several months ago, takes a more biologically grounded approach, looking at putative neural dysfunctions such as HPA axis dysregulation, sustained amygdala activity, or reward response involving the nucleus accumbens and orbitofrontal cortex.
Savitz and colleagues identify an "emerging consensus” of depression biology involving greater sensitivity to punishment and impaired hedonic capacity. Compared with non-depressed people, some depressed patients have greater amygdala reactivity to negative stimuli such as sad faces and reduced response to happy ones. Some have less activation of the ventral striatum and orbitofrontal cortex with reward stimuli, which may be correlated with anhedonia. Importantly, there is evidence for these abnormalities in non-depressed adolescent children of depressed parents, suggesting that both limbic reactivity and reduced reward responses may be stable traits rather than aspects of the depressed state or artifacts of treatment. Another candidate for a stable, likely genetically determined, marker for vulnerability to depression is reduced gray matter volume in the hypothalamus, hippocampus, and palladium, which may be the consequence of glutamate-induced excitotoxicity.
Disrupted connections among the medial prefrontal cortex, the dorsolateral prefrontal cortex, and the amygdala show up as white-matter pathology in patients with late-life, so-called vascular depression, and also in fMRI studies of patients exposed to negatively-valenced stimuli.
Savitz and colleagues also describe well-replicated findings of reduced postsynaptic 5HT1A receptors in the mesiotemporal cortex, and increased serotonin transporter in the anterior cingulate cortex, thalamus, and insula in currently depressed patients, though not all studies agree.
Unreplicated studies have found that a whole-brain morphometry algorithm predicts response to fluoxetine but is less precise at diagnosing depression; structural MRI findings prospectively predict treatment resistance in medication-naïve patients; and pretreatment response of limbic cortical areas to sad faces predicts subsequent response to cognitive-behavioral therapy.
Dunlop and Mayberg's review discusses much of the same research with a somewhat different emphasis. They see using biomarkers to personalize treatment as different from selecting treatment based on clinical trial results, but I would argue that clinical trials are still required to establish the sensitivity and specificity of a biomarker, and the prediction would still be statistical.
They note that depressed patients consistently have “hypofrontality,” i.e. reduced metabolism of the dorsolateral prefrontal cortex and increased activity in limbic regions such as the amygdala and insula. They also typically have hyperactivity of the subgenual cingulate cortex. Some patients have reduced hippocampal volume. Heterogeneity is an issue; different groups of patients may "neurologically adapt" in different ways. They also note the possible confounding effects of pathological inflammation in these studies.
Dunlop and Mayberg describe efforts to associate specific symptom clusters with neuroimaging signatures, and to focus on core symptom constructs, such as anhedonia, reward processing, and emotion regulation. They also refer to three networks which seem to mediate specific categories of mental activity. The default mode network is most active during self-referential processing and includes the medial prefrontal cortex, the posterior cingulate cortex, the inferior parietal cortex, and the medial temporal lobe. The central executive network is involved in externally-oriented, goal-directed tasks requiring working memory and planning, and corresponds anatomically to the dorso-lateral prefrontal and posterior parietal cortices. The salience network monitors for and orients to potentially relevant internal and external stimuli. It involves the ventrolateral prefrontal cortex, anterior insula, and dorsal anterior cingulate cortex. Researchers have looked at differences in activation of these networks associated with depression.
Dunlop and Mayberg identify anterior insula activity as a possible treatment selection biomarker. The dorso-anterior insula, in addition to its role in the salience network, is involved in processing risk, reward, consciousness, and monitoring of performance. Other parts of the insula have different functions--posterior regions process pain and sensory input from the viscera, while the ventro-anterior part is involved in olfactory and gustatory processing. A number of studies support the conclusion that greater anterior insula activity, particularly in response to emotional stimuli, suggests less likelihood of response a variety of treatments, including pharmacotherapy, cognitive-behavioral therapy (CBT), and vagus nerve stimulation. Such patients may be candidates for earlier trials of other therapy, such as electro-convulsive therapy (ECT), transcranial magnetic stimulation (TMS), or experimental treatments like ketamine.
The authors also suggest that lower resting state activity in this region represents downregulation by intact emotional regulation circuitry and might identify good candidates for CBT or other therapies which are thought to work with emotional regulation.
Reduced hippocampal volume is also associated with poor treatment response but has not been found useful in treatment selection. Another predictor of poor response to multiple treatment modalities is hyperactivity of the subcallosal cingulate cortex.
What conclusions can we draw from last week’s and this week’s reviews? First, a picture is coalescing of depression as a state of overactivity of limbic regions, such as the amygdala, subcallosal cingulate cortex, and anterior insula, unmodified by top-down regulation by the dorso-lateral prefrontal cortex. Depression is likely to have biological components, which can be considered either dimensions or subtypes; these include high limbic reactivity to negative emotional stimuli and impaired reward processing. While Dunlop and Mayberg refer to anhedonia as another dimension, I am not clear about its biology and to what extent it is an aspect of reward processing. Meyer, in the review I discussed last week, attempted to link a similar list of clinical dimensions with specific genes.