A number of treatments are available for depression including medications, psychotherapy and various somatic treatments such as electroconvulsive therapy [ECT], vagus nerve stimulation [VNS] and transcranial magnetic stimulation [TMS]. Unfortunately, up to two-thirds of patients remain symptomatic following first-line treatment [5] and a third fail to achieve remission (defined as full resolution of depressive symptoms) after four established treatments [6]; approximately 10%–20% of depressed patients may show virtually no improvement despite multiple, often aggressive treatments. Thus, a conservative estimate places the U.S. prevalence of TRD at 1%–3%. Importantly, this estimate does not include the prevalence of dysthymia or treatment-resistant bipolar depression.

Side effects can include cognitive detriments, personality changes, and seizures [14]. A number of non-surgical but still relatively aggressive treatments were also employed in the early 20^th century for severe psychiatric illness, including malarial pyrotherapy [15], camphor-induced seizures [16], and hypoglycemic coma [17]. In general, these treatments were developed to treat “schizophrenia” though formal, reliable diagnostic criteria were not employed. Also, while these treatments were often associated with convulsions, this was not necessarily the presumed mechanism of action at the time (except for camphor-induced seizures). Developed within this context, ECT was introduced in 1938. Over time, ECT has become established as the most effective acute treatment for a depressive episode, with remission rates over 50% [18–20]. However, cognitive side effects can be significant and occasionally persistent [21], and relapse remains a significant problem [18].

For many years, these remained the only treatments available for patients with severe psychiatric illness. However, in the mid-20^th century, an explosion of psychopharmacologic research occurred. Contemporary with the discovery of the antimanic properties of lithium [22, 23] (now a standard first-line treatment for bipolar disorder) and the antipsychotic/antimanic properties of chlorpromazine [24, 25] (a treatment for schizophrenia), the first antidepressant medications were identified in the mid-1950’s. These included the tricyclic compound imipramine [26] and the monoamine oxidase inhibitor (MAOI) iproniazid [27, 28], leading to the “monoamine hypothesis” of depression [29–31]. Research in depression then largely shifted to delineating the neurochemical bases of the disorder and helped establish a role for serotonin, norepinephrine and dopamine dysfunction in the pathophysiology and treatment of depression. Medications modulating monoaminergic neurotransmission have consistently shown response rates (defined as a ≥50% decrease in depression severity from baseline) of approximately 60%.

Other neuromodulatory systems have also been explored as potential drug targets. Abnormalities of the hypothalamic-pituitary-adrenal (HPA) axis have been clearly identified in depressed patients, consistent with findings that trauma and psychosocial stress predisposes an individual to depression, suggesting the stress hormone axis as a possible target for antidepressant medications [32, 33]. Abnormalities within the hypothalamic-pituitary-thyroid axis have also been linked with depression [34], and thyroid hormone augmentation is an accepted strategy for TRD, though with limited efficacy [35]. Dysfunction of glutamate and γ-aminobutyric acid (GABA) systems have been found in depressed patients and certain animal models of depression, and these systems have been suggested as targets for further treatment development [36]. The NMDA glutamate receptor antagonist ketamine, for instance, has been shown to have acute antidepressant effects in severely depressed patients, though relapse occurs relatively quickly [37–39]; and riluzole (an agent that appears to inhibit glutamate release, increase glutamate reuptake and block glutamate receptors) has been shown to have antidepressant efficacy [40]. More recently, the antimuscarinic agent scopolamine has been shown to have acute antidepressant effects [41]. The opioid system has also been proposed as a target of antidepressant treatment [42]. Inflammatory processes have been implicated in depression, and anti-inflammatory therapies may have antidepressant efficacy [43]. Nociceptive neurokinins may also play a role in the neurobiology of depression [44]. Finally, an increasing number of studies are focusing on dysregulation of second messenger systems, gene transcription, various neurotrophic factors, cell turnover and epigenetic mechanisms in mood disorders, including depression [45–47]. It is encouraging to see this expansion of psychopharmacological research beyond a fairly limited focus on the monoaminergic systems. Unfortunately, this work has yet to yield treatments that substantially change the overall response rates and course of illness for depressed patients [48].

In conjunction with the development of somatic treatments, formal psychotherapeutic approaches to depression have evolved (Figure 1). Psychoanalysis was developed in the late 19^th century [49], and the 20^th century saw a dramatic increase in types of psychotherapy, many influenced by advances in behavioral psychology. The most studied psychotherapies for depression include cognitive behavioral therapy (CBT)[50] and interpersonal psychotherapy [51], with reported efficacy virtually identical to that of antidepressant medications used as monotherapy (e.g., response rates of approximately 60%).

More recently, treatments have been developed to specifically target medication-resistant depression. VNS was approved by the U.S. Food and Drug Administration (FDA) in 2005 for the treatment of depression not responding to at least four antidepressant treatments. VNS does not have acute antidepressant efficacy, but has a 30% response rate (15% remission rate) with 12 months of open-label treatment [52, 53]. Although relapse rates over 1–2 years may be as high as 56% [54, 55], VNS patients appear to relapse less often than patients receiving only “treatment-as-usual” [54, 56, 57]. Repetitive TMS was FDA approved in 2008 for depression not responding to one antidepressant medication. TMS has consistently demonstrated statistically significant antidepressant effects [58–60]. However, response rates are generally less than 30%, efficacy appears limited in patients failing more than one adequate antidepressant treatment [61], and long-term relapse rates following the acute course are unknown but likely to be similar to those seen with ECT [62].

More invasive antidepressant treatments have also been revisited in the past few years. Subcallosal cingulate (SCC) DBS was associated with a 60% response rate in TRD patients following 6 months of chronic, high frequency stimulation in an uncontrolled, open-label study [63]; these effects were largely maintained over an additional 6 months [63]. DBS of other targets may also have antidepressant efficacy. These include the ventral anterior internal capsule/ventral striatum (40% response rate following 6 months of chronic stimulation, 53% response rate at last follow-up)[64], nucleus accumbens (50% response rate following 6 months of chronic stimulation)[65], inferior thalamic peduncle [66], and habenula [67]. Aside from DBS, epidural stimulation of bilateral medial and dorsolateral prefrontal cortices was associated with remission in 3 of 5 TRD patients following 7 months of treatment [68]. It should be emphasized that, despite encouraging early results with these various surgical interventions, these data are extremely preliminary and based on small, open-label, uncontrolled studies.

It is commonly presumed that the clinical heterogeneity of depression reflects biological heterogeneity, such that specific interventions may be needed for specific (clinical) subtypes of depressed patients. This view has led to attempts to define clinically significant depressive subtypes based on phenomenological differences. Early apparent success was seen with the recognition that patients with atypical depression may preferentially respond to MAOIs vs. tricyclic antidepressants (TCAs) [69]; however, a substantial number of patients with atypical depression show a very good response to a TCA, thus limiting the utility of this subtype for clinical and research purposes [70]. Similarly, it is generally accepted that patients with bipolar depression should be initially treated with a “mood stabilizer”, such as lithium or an anticonvulsant [71, 72]. However, compared with typical mood stabilizers, monotherapy with a typical antidepressant medication may be as or more effective in achieving and maintaining euthymia in bipolar disorder, at least in patients with bipolar II disorder (consisting of depressive and milder hypomanic episodes, but not more severe manic episodes) [73, 74]. Furthermore, it is not clear that the neurobiology of unipolar vs. bipolar depression is significantly different [75]. Therefore, the definition of depressive subtypes based on clinical criteria has not led to improved overall response rates, nor has this helped identify biologically distinct subtypes to guide further treatment development.

Variability also exists in antidepressant treatment response across depressed patients, and even within clinically defined subtypes. Some patients fail to respond to one medication, but will show an excellent response to a second (sometimes very similar) medication [76]. Others may not respond well to medications, but achieve clinical remission with psychotherapy [77], and vice versa; many patients may need both to achieve remission (ie. a full resolution of all symptoms) [78]. Finally, it is widely acknowledged, although less well studied, that some patients may do very well with a particular medication early in their illness, but either relapse during maintenance treatment or show limited response to the same medication in a later episode. To this point, it is noted that the vast majority of patients enrolled in studies of TRD have recurrent depression – implying they achieved remission during one or more prior depressive episodes [53, 64, 79].

Proof-of-principle testing helped confirm the hypothesis that directly modulating BA25 activity and white matter tracts connecting this brain region to other nodes in a presumed depression network would have antidepressant effects in TRD patients [63, 79]. However, only 60% of patients responded. With DBS for patients with severe, treatment-resistant movement disorders [eg. Parkinson’s disease (PD)], nearly all patients show clinically significant improvement (of motor symptoms) with chronic stimulation [82–84].

The critical question then becomes: if BA25 is a critical node in a depression network, whose activity and connectivity with other brain regions is specifically dysfunctional in TRD, why were response and remission rates not higher? One simple answer is, as above: depression is a complex disorder with multiple neurobiological etiologies – thus, only patients with “BA25-dependent” TRD may respond to this treatment. We reject this possibility for several reasons: (1) Several decades of neurobiological research have failed to identify clinically meaningful biological subtypes of “depression” as currently defined. (2) The vast majority of TRD patients have recurrent illness [i.e., they previously remitted with adequate (standard) treatment]. Thus, at an earlier stage of illness, TRD patients show response to the same biological interventions as non-TRD patients. TRD then cannot be simply defined as a “non-monoaminergic” depression. Further, this highlights the recurrence and potential malignant transformation of the disorder over time – and the limits of current treatments to address this. (3) Nearly all TRD patients enrolled in the SCC DBS study showed at least some (>30%) improvement, and these effects persisted over time [63], suggesting that modulation of a BA25-associated network leads to sustained positive effects on mood regulation in nearly all patients. Thus, we conclude: that BA25 and connected brain regions are involved in the regulation of negative mood, and that modulation of this network via focal stimulation is likely to impact mood regulation in a potentially beneficial way in most patients. However, while this beneficial mood effect is necessary for the full syndrome of depressive symptoms to resolve, it may not in itself be sufficient to achieve complete remission in all patients.

With this supposition in mind, we propose another answer: the syndrome of depression, as currently defined, is too broad to serve as a useful construct for treatment development, especially for more focal treatments. By analogy, PD is recognized as a complex neurological syndrome (with motor, sleep, cognitive, mood and autonomic abnormalities), and it is widely recognized that certain treatments may improve a subset of symptoms but not affect, or even worsen, others. For example, subthalamic DBS can improve the core motor symptoms of PD [82], but may not significantly affect postural instability, mood or cognition [85, 86] – and negative effects on mood and cognition have been reported [87, 88]. A similar approach to depression would identify specific symptoms or subsets of symptoms (e.g. depressed mood/anhedonia vs. sleep vs. anxiety) as targets for treatment; the efficacy of a particular treatment could then be gauged by its effect on these symptoms, rather than using improvement in the entire syndrome as the primary efficacy measure. A version of this approach has been recently suggested (using a dimensional construct for the depressive syndrome) as a way to develop novel antidepressant treatment strategies [89]. Such approaches should not diminish the goal of achieving full symptomatic and functional recovery. But, they may be more useful in developing specific treatment targets for clinical and basic research studies.

However, it is our opinion that a re-definition of depression should go further. We recognize that the observed symptoms of “depression” clearly track together in presentation and improvement with treatment, including domains of mood (e.g., sadness, irritability, anxiety), cognition (eg. guilt, low self-esteem, poor concentration/attention, decreased/increased psychomotor speed, hopelessness, suicidal ideation), and somatic state (eg. sleep/appetite/libido disturbances, somatic anxiety, motoric retardation/agitation). But, we conclude that this depressive state, as currently defined, is not abnormal per se, but reflects an etiologically non-specific response to stress/distress (e.g., physical or emotional stress, infection, inflammation, etc.) that includes a stereotypical set of emotional, cognitive-behavioral and somatic responses. This is supported by the currently accepted clinical criteria, where the diagnosis of a MDE can only be made if the presentation is not clearly due to a state caused by withdrawal from an addictive substance, effects of a medical illness, effects of a medication or normal bereavement highlighting the non-specific nature of the phenomenology.

We instead propose that the primary abnormality of depression is not the depressive state itself, but rather the inability to appropriately regulate that state (Figure 2): depressed patients become “depressed” without a trigger, become more severely “depressed” following a trigger (e.g, death of a loved one) than is considered normal, and/or remain “depressed” for longer than would be expected. By analogy, with cardiac arrhythmias (e.g., Wolff-Parkinson-White syndrome, sick sinus syndrome, etc.), the actual arrhythmia (e.g., tachycardia) is the symptomatic presentation that may need acute treatment, but treatment of the underlying disease requires addressing the tendency of the system to enter an aberrant rhythm it cannot exit. Similarly, we conceptualize the depressive state as an aberrant neural rhythm, but the depressive disorder as the brain’s tendency to go into and stay in that rhythm inappropriately.

Such a view has important implications for treatment development. With cardiac arrhythmias, a number of diverse interventions with different mechanisms of action (e.g., most medications, cardioversion, time [i.e., spontaneous recovery]) may return the system to the physiological state, but others (eg. ablation, pacing, certain medications) may be needed to help maintain the “normal” state over time. Likewise, diverse “antidepressant” treatments (e.g., various medications, psychotherapy, ECT, or TMS) can shift a patient out of the depressive state but may not necessarily prevent re-entry into that state. This is especially true of treatments with purported acute “antidepressant” effects (such as sleep deprivation [90], ketamine [38], and scopolamine [41]), where a positive mood/behavioral change is seen within minutes to hours, but effects are short-lived (with return of depressive symptoms within hours or days). Other treatments, with mechanisms of action distinct from those of more acute treatments, may be needed to prevent the abnormal recurrence of the depressive state. For example, certain psychotherapies may provide positive steps in this direction [9, 10, 91–93]. It has been suggested that such treatments may work by strengthening the cognitive cortical-subcortical systems responsible for regulating mood [94–96]).

In light of our proposed re-definition of depression, we suggest a number of directions that may prove useful to take with future research on MDD (Box 2). One important focus of such research should be on identifying interventions that not only alter the underlying state, but actually prevent the brain from entering or remaining in that state at inappropriate times. Although highly preliminary, emerging data for DBS suggest that such a treatment intervention may be possible. With SCC DBS, there are immediate changes in mood with initial acute stimulation that may or may not persist, with resolution of the full syndrome of the depressive episode requiring weeks to months [63, 79]. But critically, once patients are well, they appear to stay well over time: although the published follow-up data are currently limited to one year of treatment, it is notable that >70% of patients responding to SCC DBS at 6 months were still responders at 12 months [63]. Though not described in detail, similar results may be seen with ventral capsule/ventral striatal DBS for TRD [64]. Further, if stimulation ceases (e.g., the device is turned off or the battery depletes), patients will almost invariably return to the depressive state [63, 64, 97].