Neurobiology of Antidepressant Withdrawal: Implications for the Longitudinal Outcome of Depression Brian H. Harvey, Bruce S. McEwen, and Dan J. Stein Inappropriate discontinuation of drug treatment and non- compliance are a leading cause of long-term morbidity during treatment of depression. Increasing evidence sup- ports an association between depressive illness and dis- turbances in brain glutamate activity, nitric oxide synthe- sis, and -amino butyric acid. Animal models also confirm that suppression of glutamate N-methyl-D-aspartate re- ceptor activity or inhibition of the nitric oxide-cyclic guanosine monophosphate pathway, as well as increasing brain levels of -amino butyric acid, may be key elements in antidepressant action. Imaging studies demonstrate, for the most part, decreased hippocampal volume in patients with depression, which may worsen with recurrent depres- sive episodes. Preclinical models link this potentially neurodegenerative pathology to continued stress-evoked synaptic remodeling, driven primarily by the release of glucocorticoids, glutamate, and nitric oxide. These stress- induced structural changes can be reversed by antidepres- sant treatment. In patients with depression, antidepressant withdrawal after chronic administration is associated with a stress response as well as functional and neurochemical changes. Preclinical data also show that antidepressant withdrawal evokes a behavioral stress response that is associated with increased hippocampal N-methyl-D-as- partate receptor density, with both responses dependent on N-methyl-D-aspartate receptor activation. Drawing from both clinical and preclinical studies, this article proposes a preliminary molecular perspective and hypoth- esis on the neuronal implications of adherence to and discontinuation of antidepressant medication. Biol Psy- chiatry 2003 54:1105���1117 �� 2003 Society of Biological Psychiatry Key Words: Antidepressant, discontinuation, adherence, GABA, glutamate, relapse Introduction Clong-term linicians have long been aware of the importance of adherence to antidepressants (Keller and Boland 1998). In placebo-substitution studies of depressed patients who have responded to antidepressants, those switched to placebo relapse 2 to 4 times more often than those maintained on medication (Nierenberg 2001). Nev- ertheless, patients with depression and anxiety disorders all too often discontinue antidepressants prematurely (Basco and Rush 1995). Among the understandable reasons for premature anti- depressant discontinuation is the poor tolerability of cer- tain agents, although the introduction of novel antidepres- sants has diminished the relevance of this issue (Anderson 2000). More important, however, is that premature discon- tinuation may be a product of a lack of knowledge of the neurobiological mechanisms mediating untreated depres- sion and antidepressant withdrawal and of their implica- tions for the course and prognosis of depression. Although current antidepressants exert initial effects within hours of administration by increasing intrasynaptic levels of serotonin (5HT) and/or noradrenaline (NA), clinical response is delayed for up to 4 to 6 weeks, indicating that a cascade of subcellular events is ultimately responsible for their therapeutic effects (Harvey 1997 Popoli et al 2002) however, many patients report signif- icant drug-related side effects within the first days of treatment, including anxiety, irritability, nausea, headache, and other symptoms that are unrelated to clinical response (Harvey 1997). Similarly, patients that have been on long-term antidepressants, such as the selective serotonin reuptake inhibitors (SSRIs), and who are abruptly with- drawn report an array of ���withdrawal effects��� that are distinct from relapse of clinical depression (Schatzberg et al 1997 Zajecka et al 1997). It is thus pertinent to ask whether, just as treatment initiation-related side effects may be forerunners of subcellular neuronal changes in- volved in clinical response some weeks later, are there salient longer term effects of antidepressant withdrawal on From the Division of Pharmacology (BHH), School of Pharmacy, Potchefstroom University for Christian Higher Education, Potchefstroom, South Africa Laboratory of Neuroendocrinology (BSM), Rockefeller University, New York, New York and Medical Research Council Anxiety and Stress Disorders Research Unit (DJS), Department of Psychiatry, University of Stellenbosch, Tygerberg, South Africa. Address reprint requests to Brian H. Harvey, Division of Pharmacology, School of Pharmacy, Potchefstroom University for Christian Higher Education, Potchef- stroom, 2520, South Africa. Received October 31, 2002 revised March 27, 2003 accepted May 5, 2003. �� 2003 Society of Biological Psychiatry 0006-3223/03/$30.00 doi:10.1016/S0006-3223(03)00528-6

neuronal function? More importantly, can these events represent neuronal events that are in some ways opposite to those seen in association with clinical response and so negatively influence long-term treatment outcome? Although significant progress has been made in the molecular neurobiology of depression and of antidepres- sant action, such questions about the neurobiology of antidepressant withdrawal have not been comprehensively addressed. Nevertheless, current knowledge makes it pos- sible to propose a preliminary molecular perspective on the importance of adherence to antidepressant medication and this will be the aim of the current paper. Such a model could have important implications for understanding and improving clinical outcome, especially since repeat epi- sodes of depression are often of longer duration, more severe and less treatment-responsive. Similarly, greater number and longer duration of episodes of depression are associated with a higher risk of relapse and recurrence (Judd et al 1998 Keller and Boland 1998). To propose such a model, clear understanding of the neurobiology of depression is essential, as is a deeper understanding of the molecular action of antidepressants, and we begin with a brief review of these areas. Depression: A Model for a Chronic, Recurring Illness The association between stressful life events and the onset and chronicity of major depression has been the subject of intense debate over the past years. Kraepelin (1921) first proposed that psychosocial stressors play a greater role in the initial than in subsequent episodes of depressive disorder. In other words, over the course of the illness, the onset of depressive episodes may become increasingly autonomous and less reliant on environmental adversity. Post (1992) proposed that this pattern results from a sensitization process to the state of depression. The kin- dling theory of recurrent affective illness (see Post and Weiss 1998 for review) has been viewed as having particular relevance for how synapses react to environ- mental and neurochemical stressors and how these can influence clinical response and outcome. Recent clinical studies suggest that the association between previous number of depressive episodes and the pathogenic impact of stressful life events on major depres- sion is likely causal and biphasic (Kendler et al 2001a). Such data are congruent with the kindling theory but suggest the presence of a threshold at which further sensitization does not occur. Furthermore, increased sus- ceptibility to depression with subsequent episodes, or a ���kindled state,��� is distinctly linked to the number of prior episodes, the presence of stressful life events, and high genetic risk (Kendler et al 2001b). Of particular interest are recent studies that highlight the critical interaction between the course of illness and brain morphologic changes occurring in depression. Thus, MacQueen et al (2003) found that hippocampal volume reduction is not present in first episode patients but becomes increasingly evident in those with multiple episodes. Clearly, depression is a chronic and recurrent illness that is associated with considerable disability, morbidity, and mortality (Keller et al 2002 Kuhn et al 2002). Although the kindling hypothesis has been questioned (Mueller et al 1999) and requires further validation, it represents one way to integrate a range of data describing the neurophysiology and neurochemistry of the onset, long-term development, and treatment responsiveness of the illness. The primary neural substrate of kindling involves glutamate and N-methyl-D-aspartate (NMDA) receptor activation, with -amino butyric acid (GABA) pathways exerting essentially a permissive role on the kindling action of glutamate (Post and Weiss 1998). This involvement of GABA and glutamate may have particular relevance in the neurobiology underlying the treatment and prognosis of depressive illness. Functional and Molecular Correlates of Antidepressant Action The monoamine models have played an important heuris- tic role in our understanding of antidepressant action however, it remains unclear, for example, how some patients with chronic depression are able to remit without treatment or why depression relapse rates are so high even in patients maintained on antidepressants (Keller and Boland 1998). While these differences may have their origin in the complex phenomenological and biological heterogeneity of depression itself, as well as the heteroge- neity of antidepressants, troubling questions such as these can only be addressed by fully understanding how antide- pressants work. The original monoamine deficiency hypothesis assumed that depression is associated with a deficiency of principle monoamines, serotonin, noradrenaline, and dopamine (DA). Later, the monoamine receptor hypothesis empha- sized the sensitization state of receptors during depression and first introduced the concept that both the effects and delayed response to antidepressants may rest on time- dependent changes in receptor sensitivity. Positron emis- sion tomography studies have found significantly altered 5HT transporter (5HTT) densities in depression (Ichimiya et al 2002 Kasper et al 2002), and these may be altered by antidepressant treatments (Kasper et al 2002). Further, antidepressant efficacy is related to changes in 5HT2 1106 B.H. Harvey et al BIOL PSYCHIATRY 2003 54:1105���1117