Disruption of Neural Homeostasis as a Model of Relapse and Recurrence in Late-Life Depression

Abstract

The significant public health burden associated with late-life depression (LLD) is magnified by the high rates of recurrence. In this manuscript, we review what is known about recurrence risk factors, conceptualize recurrence within a model of homeostatic disequilibrium, and discuss the potential significance and challenges of new research into LLD recurrence. The proposed model is anchored in the allostatic load theory of stress. We review the allostatic response characterized by neural changes in network function and connectivity and physiological changes in the HPA axis, autonomic nervous system, immune system and circadian rhythm. We discuss the role of neural networks’ instability following treatment response as a source of downstream disequilibrium, triggering and/or amplifying abnormal stress response, cognitive dysfunction and behavioral changes, ultimately precipitating a full-blown recurrent episode of depression. We propose strategies to identify and capture early change points that signal recurrence risk through mobile technology to collect ecologically-measured symptoms, accompanied by automated algorithms that monitor for state shifts (persistent worsening) and variance shifts (increased variability) relative to a patient’s baseline. Identifying such change points in relevant sensor data could potentially provide an automated tool that could alert clinicians to at-risk individuals or relevant symptom changes even in a large practice.

Introduction

Late-life depression (LLD) is associated with significant public health burden, high disability, increased risk for cognitive decline and dementia, elevated suicide risk, and greater all-cause mortality rates.¹ These negative outcomes are magnified by depression being a recurrent disorder, characterized by repeated episodes over a patient’s lifetime.

While antidepressant medications and psychotherapy can effectively treat depressive episodes in older adults,^2–4 individuals who respond remain at high risk for future episodes.⁵ This includes both relapse (an episode occurring within 6 months of response or remission) and recurrence (episodes occurring after 6 months); for simplicity, in this manuscript we refer to both as recurrence. High recurrence rates are observed in maintenance clinical trials, where over 2 years approximately 35% of elders receiving antidepressants experience recurrence, with higher rates for placebo.^6,7 Similarly, in a naturalistic cohort study, participants with LLD receiving antidepressant treatment exhibited recurrence rates of 43% in 2 years and 57% over 4 years.⁸

Despite some clinical characteristics that indicate high recurrence risk, we have only limited data in LLD on neurobiological predictors of recurrence. This critical omission limits our ability to identify individuals at high recurrence risk and to develop new therapeutic approaches designed to maintain remission. In this manuscript we review clinical, neuropsychological, and neurobiological factors influencing recurrence risk. We then discuss environmental influences and the interplay of stress reactivity and allostatic load, defined as the wear-and-tear resulting from stress and the dysregulation of mechanisms intended to promote homeostasis. We integrate this literature into a model of homeostatic disequilibrium, focusing on the instability of key neural networks. We posit that this core neural instability results in cognitive changes followed by the emergence or worsening of clinical symptoms and ultimately depression recurrence (Figure 1). Finally, we discuss the significance and challenges of new research into LLD recurrence, including ecological assessment methods that could be used for monitoring and signaling impending recurrence.

Figure 1. Model of homeostatic disequilibrium contributing to recurrence of late-life depression.

The homeostatic disequilibrium hypothesis of depression recurrence proposes that in remitted depression, neural networks are in fragile homeostasis. This homeostasis is threatened by stress exposure. Individuals at increased risk for recurrence exhibit altered neural responses to those stressors and develop further alterations in network connectivity and function. These network changes contribute to cognitive symptoms. Such cognitive symptoms may be initially characterized by a range of attentional changes, including concentration deficits, worsening attentional biases to negative stimuli, or difficulty diverting attention from emotional stimuli. Later it may manifest as poorer performance across multiple cognitive domains including executive control. In turn, the network dysfunction and changes in cognitive function contribute to the emergence or worsening of clinical symptoms such as increased anxiety and worry, increased rumination, or poor sleep. These early symptoms may potentially be detected through the use of technology, allowing for ecological monitoring and identification of elevated recurrence risk. If unchecked, cognitive and behavioral changes may contribute to further distress and homeostatic disequilibrium, resulting in further changes in functional network activity. Without intervention the individual is at increased risk for a new depressive episode. A number of individual factors may increase or decrease the likelihood of maintaining homeostasis and include pre-existing neural network alterations or other characteristics unique to the stressor or individual (Table 1).

Clinical and Behavioral Predictors of Recurrence

Across adult and geriatric populations, recurrence is most strongly associated with a higher number of prior depressive episodes and greater severity of residual depressive symptoms.^8–11 In LLD, residual symptoms such as sleep disturbance, sadness and anxiety are associated with recurrence.^6,8,12 In contrast, depression severity during the index episode and age of initial depression onset are inconsistently associated with recurrence.^7,8,13,14

Maintenance of antidepressant treatment and treatment adherence are primary clinical features that reduce but do not eliminate risk of recurrence. Placebo-controlled maintenance trials in LLD clearly demonstrate that ongoing antidepressant medication treatment is associated with better long-term remission rates.^6,15 Unfortunately, non-adherence occurs in up to 60% of adults on maintenance antidepressant treatment and results in a shorter time to recurrence.^16,17 While adherence to antidepressant medications is better in older adults than in younger adults,^18,19 comorbid cognitive impairment may challenge adherence.²⁰

LLD is often characterized by cognitive impairment²¹ that, even with good adherence, predicts poor acute antidepressant response.^21–24 While cognitive performance can improve with successful treatment, specific deficits typically persist.^25–29 Even following remission, LLD is associated with accelerated cognitive decline.^27,30 Conclusions from studies examining the effect of cognitive impairment on recurrence risk are mixed.^7,8,31 However, many past studies tested individuals when acutely depressed rather than in remission and measured cognitive performance only cross-sectionally rather than longitudinally.

Personal and environmental factors also influence risk. Predisposing factors increasing recurrence risk include environmental stressors, stressful life events, and greater perceived stress.^8,32–34 Some report that medical morbidity, development of new medical problems, and greater disability also increase recurrence risk.^{8,14,32,34,35} In contrast, social support may be a protective factor that reduces recurrence risk or mitigates the effect of life stress on recurrence risk.^8,36 Although not well studied in LLD, greater physical activity may benefit depression^37,38 and decrease the risk of recurrence. These observations have clinical utility but they do not inform us about neurobiological processes or behavioral markers predictive of recurrence that could stratify individual risk or serve as targets for prevention studies.

Neurobiological Differences and Risk of Recurrence

Few studies prospectively examine neuroimaging measures as predictors of recurrence. Structural neuroimaging studies consistently associate LLD with markers of accelerated brain aging, specifically greater white matter hyperintensity (WMH) volumes and smaller hippocampal volumes.^39–42 However, in a recent study, cross-sectional measures of brain aging (e.g. total cerebral volume, total gray matter, ventricular volume, hippocampal volume, WMH) did not predict LLD recurrence.⁸ In contrast, longitudinal changes are associated with poor outcomes, as increases in WMH volume and decreases in hippocampal volume are observed in depressed elders who experience recurrence.^43,44

Differences in brain connectivity during depressive episodes may persist into remission and increase vulnerability to recurrence. Work in remitted LLD is limited; however, relative to never-depressed elders, on diffusion tensor imaging individuals with remitted LLD exhibit reduced structural connectivity between the left and right hemispheres.⁴⁵ Compared with non-depressed elderly, functional connectivity of the default mode network (DMN) is altered in LLD and remains altered even after remission.⁴⁶ Remitted LLD is also associated with resting-state functional connectivity alterations in network-based markers, including reduced global efficiency and network strength with greater longitudinal change and instability in these network properties.⁴⁵

Beyond LLD, data in adult MDD populations identify neural differences that persist into remission. Much of this work does not prospectively examine recurrence but focuses on cross-sectional, network-based differences between remitted MDD (rMDD) and never-depressed individuals. The DMN exhibits altered connectivity and function in rMDD, with fewer connections with other subnetworks,⁴⁷ lower fractional amplitude of low-frequency fluctuations (ALFF),⁴⁸ and lower deactivation of DMN regions during cognitive tasks.⁴⁹ Similarly, in rMDD the Executive-Control Network (ECN) exhibits lower within-network connectivity⁵⁰ and attenuated activity during “cold” cognitive tasks but greater neural response to emotional stimuli.^51,52 Finally, the Salience Network (SN) exhibits higher ALFF in the insula in rMDD that is correlated with the number of prior depressive episodes.⁴⁸

These canonical networks are also implicated by studies examining functional MRI (fMRI) predictors of recurrence in rMDD. Recurrence is predicted by high ventromedial PFC DMN activity in response to sad stimuli⁵³ and by low dorsomedial PFC ECN activity during a go-no go task.⁵⁴ Recurrence is also predicted by altered subgenual anterior cingulate cortex (sgACC) SN resting connectivity⁵⁵ and high connectivity between the sgACC and anterior temporal lobe during a self-blame task.⁵⁶ Recent data in midlife depression shows that the DMN alterations go beyond changes in connectivity strength and involve lower connectivity stability within key DMN regions such as mPFC and PCC⁵⁷, a finding relevant to our proposed homeostasis model.

Dysfunction in these intrinsic networks is not limited to adult MDD but is also observed in LLD.^58–64 Further, successful antidepressant treatment in LLD results in changes in network connectivity^64,65 and function.^62,66,67 These data support the hypothesis that residual alterations in network activity or connectivity observed in remission may be associated with increased risk of recurrence.

Environmental Influences and Stress Reactivity

Residual neural network abnormalities likely do not directly cause recurrence in most individuals. Instead, they may contribute to recurrence by adversely affecting response to environmental experiences, particularly to stressful events. “Stress” commonly refers to experiences that cause feelings of anxiety, dysphoria, or frustration by threatening one’s security or challenging one’s ability to cope with the experience.⁶⁸ Although this definition includes severe trauma, more often it refers to interpersonal conflict, economic insecurity, poor health, and even time pressure or daily hassles. Severe or persistent stressors can result in a number of behavioral and mood changes, including anxiety, dysphoric mood, sleep disruption, altered appetite, and withdrawal from social and pleasurable activities.

Stressful life events are a recognized risk factor for depression recurrence,^69,70 a relationship that persists in LLD^71,72 where a higher incidence of stressful events and greater perceived stress results in poorer antidepressant response.⁷³ Different classifications of stress have been proposed,⁶⁸ including “good stress” that is characterized by pressure related to positive events, or overcoming a challenge and feeling rewarded by a positive outcome. “Tolerable stress” refers to negative situations where the individual is able to cope with the problem, often with social supports. “Toxic stress” refers to situations that may be more severe where the individual is less able to cope with the problem and may have limited or no supports. Many individuals susceptible to depression have greater difficulty coping with stressors or have less social support, resulting in a higher level of perceived stress and potentially more severe behavioral symptoms.

A key element linking stress exposure with new depressive episodes may include variability in the physiological and behavioral responses to stressors. Alterations in physiological stress responses are apparent in depression across the adult age range. In non-geriatric populations, rMDD is associated with altered physiological stress reactivity to both mild and more potent stressors, including activity in the hypothalamic-pituitary-adrenal (HPA) axis and sympathetic nervous system.^74–77 Such altered physiologic responses may predict subsequent recurrent episodes.⁷⁸

Physiologic stress responses parallel neural responses to a range of stress-inducing tasks. Most work examining adult rMDD reports that stress induction and exposure to negative emotional stimuli results in increased activity in the amygdala, hippocampus, and striatum, with less activation in dorsal prefrontal and cingulate regions.^{76,77,79–82} These data parallel functional connectivity studies associating higher stress levels with higher within-network SN connectivity, lower within-network ECN connectivity and lower connectivity between the ECN and both the SN and DMN.⁸³

Behavioral manifestations of stress reactivity may include affective instability, or the frequent fluctuation of emotions.⁸⁴ Remitted MDD is characterized by greater behavioral responses to stress, specifically greater declines in positive affect and greater increases in negative affect.^85,86 Ecological assessment studies report that rMDD is characterized by greater affective instability over short periods in emotions of feeling down, worried, irritated, and restless.⁴⁷ This behavioral instability is associated with reduced connectivity between the SN and other subnetworks.⁴⁷

Allostasis and Contributors to Allostatic Load

These changes may be framed in the context of allostasis and stress-related increase in allostatic load. Allostasis refers to the body’s process of responding to environmental changes and events in order to maintain homeostasis. Allostatic load refers to the wear-and-tear on the body and brain resulting from excessive stress or from dysregulation of mechanisms intended to promote homeostasis.⁸⁷ Core brain regions involved in emotion reactivity and regulation also mediate the response to stress.⁸⁸ Thus, the networks involved in emotional reactivity and regulation are the first neural sites to show evidence of long-term wear and tear⁸⁸. These networks affect multiple regulatory systems that play key roles in allostasis, including the HPA axis, the autonomic nervous system, the immune system and the circadian system. When disrupted, these systems contribute to allostatic overload.⁸⁹

The HPA axis has long been associated with stress response. A rich literature describes the effect of cortisol on neurogenesis, neuronal architecture and structural changes in key regions such as hippocampus and amygdala^90–93. Although HPA axis alterations in remitted LLD are not well characterized, older depressed adults exhibit a high degree of HPA axis dysregulation,⁹⁴ perhaps because with aging the HPA axis is less adaptive in its responses and is characterized by a flattened diurnal rhythm.⁹⁵ HPA axis dysregulation in LLD may have significant negative long-term consequences as higher cortisol levels are associated with smaller regional brain volumes, poorer cognitive function, and may be a pathway linking depression with increased risk for dementia.^40,96–99

Stress-related changes in autonomic nervous system function contribute to the increase in peripheral and central vascular burden through multiple mechanisms including vasoconstriction, high blood pressure and microinfarcts.^100,101 These effects are amplified by chronic inflammation leading to endothelial disfunction and arterial damage.¹⁰² Brain regions involved in autonomic regulation overlap with those involved in emotion regulation – the central nucleus of the amygdala, the dorsal anterior cingulate, the insula and medial PFC.^103–105 As proposed by the vascular depression hypothesis, cerebrovascular disease has been recognized for over two decades as a key pathway predisposing and perpetuating late-life depression^106–108, also contributing to comorbidity between LLD and dementia.¹⁰⁹

The immune system, through pro-inflammatory markers such as cytokines and interferons, also responds to stress exposure and is associated with midlife and late-life depression^106,110. Proinflammatory markers are associated with volume reductions in the same areas involved in emotion regulation (hippocampus, PFC, amygdala) and with higher white matter hyperintensities burden and executive dysfunction in LLD.^111,112 Greater amygdala activity and greater PFC-amygdala functional connectivity is also associated with enhanced inflammatory responses to social stress.¹¹³

Stress exposure also disrupts sleep, increasing allostatic load by disturbing circadian homeostatic systems.⁸⁹ The circadian system regulates the timing of sleep and activity cycles and controls biological rhythms in the brain and body through direct neural signals and through hormonal messengers including cortisol. The relationship between circadian rhythms, sleep, and allostatic load may be cyclical, where stress may disrupt circadian rhythms while sleep disruption may contribute to difficulty regulating responses to stress. Notably, sleep disruption is a cardinal symptom of depression and is increasingly recognized as a risk factor for Alzheimer’s disease, potentially by several mechanisms including systemic inflammation that in turn increases beta-amyloid burden¹¹⁴ and decreased beta-amyloid clearance due to reduced slow-wave sleep.^115,116

These canonical stress-response systems interact⁸⁸ but – importantly for our model - they also have significant bottom-up effects on the neural centers of allostasis. Moreover, activity in these systems fluctuate over short periods of time, requiring novel environmental assessment approaches to capture their contributions to depression recurrence.

Stress, Allostatic Load, and Cognitive Decline

Beyond developing depressive episodes, stress exposure also affects cognitive function. Acutely, stress exposure requires the allocation of cognitive resources to cope with environmental demands. This results in a corresponding decrease in the attentional resources available for information processing and relatively poorer cognitive performance compared with periods of low stress, a finding that may be magnified in older adults.^117,118 These effects may be mediated by greater negative affect reactivity to stressors.¹¹⁹ Chronic stress exposure and greater perceived stress are associated with poorer cognitive performance,^120,121 more rapid cognitive decline,¹²² and increased risk for mild cognitive impairment and dementia.^123,124

These relationships may be interpreted in the context of increased allostatic load related to chronic stress exposure. Greater allostatic load and physiological markers of stress such as higher cortisol levels, high sympathetic tonus, inflammation, and sleep disruption are associated with poorer cognitive performance^97,98,125 and altered structure of brain regions underlying cognition.^93,126 Various markers of high allostatic load are further associated with neurodegeneration, increased cerebrovascular burden, microglial activation, tau protein phosphorylation and other pathological markers of accelerated aging.¹²⁷

There is overlap in the neurobiological relationships observed between LLD, cognitive impairment and dementia, and allostatic load effects. Altered ECN connectivity and activation is observed in LLD and is also associated with poorer cross-sectional cognitive performance.^59,62,128 Pathological brain aging that is also observed in LLD, including progression of WMH and hippocampal atrophy, is associated with greater allostatic load^43,44 and with accelerated cognitive decline and increased risk of dementia.^129–131 While depression and cognitive impairment are highly comorbid, recent work supports that depression has an effect on cognition that is independent of the underlying neuropathology, meaning that cognitive impairment in context of depression is greater than would be expected based on the severity of underlying neuropathology alone.¹³² Depressive symptoms thus moderate or magnify the relationship between brain pathology and cognitive performance.^132,133 We propose that neurobiological factors contributing to recurrence vulnerability, including altered network connectivity and altered neural and physiological responses to stress, also contribute to the risk for cognitive decline in LLD. We also posit that neurobiological and cognitive factors act in a cascade-type reaction to trigger recurrence (Figure 1).

Recurrence as a Consequence of Allostatic Load and Homeostatic Disequilibrium

We propose that depressive episodes be considered within the concept of neural homeostasis, or the tendency of neural circuits and networks to develop a stable equilibrium. In individuals at low risk for depression, exposure to stressful events and high levels of perceived stress will result in an allostatic response. This allostatic response is characterized by neural changes in network function and connectivity and physiological changes in the HPA axis, autonomic nervous system, immune system and circadian rhythm. In turn, these allostatic physiological responses contribute to short- or long-term behavioral and mood changes that include fluctuations in affect, anxiety symptoms, and sleep disruption. A successful allostatic response involves maintenance of homeostasis and a resolution of the behavioral symptoms, although the likelihood and speed of such a resolution depends on many factors specific to the individual and the stressor (Table 1).

Table 1.

Factors influencing allostatic responses to stress

Individual Factors		Stressor Factors
Vulnerability factors	Protective factors (resilience)
Personality traits such as neuroticism Poor coping skills Limited social support Lifestyle (exercise/sleep/diet) Substance abuse Cumulative burden of other stressors	Adequate/extended coping skills Extended social support Lifestyle (exercise/sleep/diet) Spirituality/mindfulness training Intact microbiome	Intensity of stressor and degree of perceived threat Level of control over stressor Duration or chronicity of stressor Involvement of social conflict Contribution to disability

Numerous factors specific to the individual and the stressor influence the intensity of the allostatic response and how successfully and quickly homeostasis can be restored. This is not meant to be a definitive list, but rather examples of characteristics that may facilitate or inhibit successful allostatic responses and the likelihood of a return to homeostasis.

This process may be disrupted in individuals with depression. During active depressive episodes, allostatic load increases as emotional and cognitive circuits are in disequilibrium. This disequilibrium contributes to emotional dysregulation, clinical symptoms, and deterioration in cognitive performance. Successful treatment may facilitate the stabilization of neural network connectivity and function, a return to homeostasis, and subsequent symptomatic remission. However, in some individuals this new homeostatic state is fragile with persistent differences in neural network function, cognitive performance, and residual behavioral and mood symptoms that continue into remission and indicate ongoing vulnerability to future episodes (Table 2). Neural and physiological responses to stressors are increased in depressed and remitted individuals compared with never-depressed individuals.¹³⁴ These functional neural network changes and accompanying physiological changes contribute to impairment in cognitive performance and behavioral signs and symptoms (Figure 1). Subsequent behavioral symptoms may further increase allostatic load. For example, higher perceived stress may negatively affect task performance, increasing subjective and physiological stress. Similarly, perceived stress can disrupt sleep, that then negatively affects circadian rhythms. Without intervention, these cognitive and behavioral changes could progressively worsen, further disrupt homeostasis, and result in a new depressive episode.

Table 2.

Neurobehavioral markers of challenged homeostasis

Neural circuits
	At rest	Increased network instability
		DMN- reduced disengagement
		SEN - increased intra-network connectivity
		ECN - decreased intra-network and inter-network connectivity
	During stress reactivity task	Increased stress-related limbic hyperactivity
Behavioral processes
	Stress Reactivity
	Affective Instability
Cognitive processes
	Decreased attentional function
	Cognitive performance fluctuation
Clinical Signs/Symptoms
	Increased Perceived Stress
	Sleep disruption
	Mood/anxiety fluctuations

Such clinical manifestations could be monitored using ecological momentary assessment (EMA), defined as brief symptom, state, or physiological measures assessed in the individual’s natural environment. EMA approaches may provide for early detection of homeostatic disequilibrium, characterized as symptom development, allowing for intervention before a new episode develops. For individuals at high recurrence risk, EMA approaches could bridge the gap between provider visits and detect impending recurrence.

When homeostasis is threatened in individuals with remitted depression, there are two potential paths. Optimally, the system and individual can adapt to the challenge, maintain homeostasis, and avoid depression recurrence. This would be facilitated by resilience factors, such as positive coping strategies, ongoing high treatment adherence, social support and resolution of the stress. Alternatively, adaptation could fail due to poor coping strategies, poor treatment adherence, biological vulnerabilities that accumulate following repeated depressive episodes, poor social support, or due to the severity or persistence of the stressor. This results in worsening disequilibrium and the development of a recurrent episode. The time in recovery since the last episode may also be important, where greater periods of good mental health may promote good coping strategies, improved social support, and healthy lifestyle activities.

Using a musical analogy, the concept of harmonics can illustrate homeostasis and disequilibrium (Figure 2). Overall allostasis represents the fundamental tone (or first harmonic) that dictates the homeostatic response. This is however modified by multiple second (neural networks fragility) and third harmonics (cognitive performance changes) that together set the final harmony (or disharmony) in the system.

Figure 2. Homeostatic Disequilibrium Illustrated using Harmonics.

Harmonics are higher frequencies superimposed on the fundamental waveform that create a more complex wave pattern. Harmonics are a shared domain in music (allowing for complex sound patterns) and electronics (where their effect is less beneficial, usually causing voltage waveform distortion and overloading or overheating). In our analogy, allostasis represents the fundamental tone (or first harmonic) that dictates the general pattern of the homeostatic response. Neural network synchrony would represent the second harmonic that changes the output of the homeostatic response. Cognitive performance would represent the third harmonic. Together, the harmonics set the final output of the system. As illustrated on the right, with disruption of homeostasis by increasing fragility of neural networks and worsened cognitive performance, we observe greater disharmony and variability in the system.

Caveats to the Hypothesis of Homeostatic Disequilibrium

Our concept of homeostatic disequilibrium as a precursor to and characteristic of depressive episodes is not limited to older adults and could be applied to depression across the lifespan. However, maintenance of homeostasis and avoidance of recurrence may be particularly challenging in older adults. Biological aging is characterized by homeostenosis, or a reduction in the reparative potential in organs and a decreased ability to maintain homeostasis in response to stress.¹³⁵ It is possible that repeated depressive episodes with their associated increases in allostatic load may accelerate homeostenosis, increase the challenge of maintaining homeostasis in the context of stress, and increase the risk for future recurrent episodes. LLD’s high rates of medical comorbidity¹³⁶ and the associated effects of medical disease on endocrine function, autonomic activity, and inflammation may contribute to pathological brain aging and increase homeostenosis.

This homeostatic disequilibrium hypothesis has similarities with other theories of depression recurrence. The “kindling” theory proposes that the vulnerability to future mood episodes grows with the number of past episodes.¹³⁷ In the allostasis framework, load increases iteratively with each depressive episodes. Prior load carried forward may decrease the ability to adapt to a current stressor.⁸⁸ This perspective is reminiscent of the “scar” hypothesis that emphasizes the persistence of residual psychological deficits after a depressive episode.¹³⁸

While aging and repeated episodes may contribute to homeostenosis, the effect of time on one’s ability to maintain homeostasis and continue in remission is not always necessarily negative. With prolonged periods of sustained homeostasis and clinical remission, homeostasis could become more “entrenched” making it easier to adapt to new challenges and reducing risk of recurrence. This could be observed by resolution of residual depressive symptoms and stability or even normalization of neural and cognitive markers over time. Moreover, these processes may be enhanced by preferential processing of positive information in aging as evidenced by the “positivity bias” described in older adults.¹³⁹ However, many individuals may require ongoing treatment as antidepressant cessation may allow the re-emergence of network alterations that would predispose those individuals to new depressive episodes.

A significant stress exposure may not be necessary to disrupt homeostasis and to experience recurrence. Although stressful life events are associated with onset of new depressive episodes, with repeated episodes less stress may be required to develop a new depressive episode in midlife.¹⁴⁰ This may reflect that the accumulated allostatic load developing over repeated depressive episodes could worsen homeostenosis so that even mild or trivial stressors may have a disproportionate effect and disrupt homeostasis in some individuals.

While this model proposes that individual differences in neural network connectivity and function that persist into remission contribute to recurrence, there are other hypotheses. It is possible that pharmacological or physiological changes related to long-term use and repeated exposure to antidepressant medications could contribute to a loss of efficacy and emergence of new depressive episodes. This concept of tachyphylaxis¹⁴¹ has been explained through several mechanisms such as pharmacokinetic or pharmacodynamic tolerance, long-term changes in serotonin signaling pathways, and effects of the HPA axis on serotonin receptor function or modulation of gene expression.¹⁴² These mechanisms underlying tachyphylaxis may be complementary to our hypotheses. Long-term pharmacological effects that contribute to a decline in clinical response may have the greatest effect in individuals with persistent neural network differences that contribute to residual depressive symptoms or altered responses to environmental stressors.

Scientific and clinical significance of studying depression recurrence

While acute depressive episodes may be conceptualized within a model of disrupted homeostasis, the principle value of the homeostatic disequilibrium hypothesis may be to aid our understanding of the development and recurrent nature of depressive episodes. Unfortunately, we currently have a limited understanding of neurobiological factors that influence recurrence. Our best clinical markers of recurrence are primarily limited to an individual’s past psychiatric history and residual symptoms. These residual symptoms can be subtle or mild⁸ and do not provide clinical support to identify the optimal timing for when to intervene to avoid an impending depressive episode.

A better understanding of neurobiological mechanisms influencing recurrence risk provides an opportunity to identify therapeutic targets for future maintenance studies. Such work is critical for the development of mechanism-based treatments aimed at prevention of future episodes, whether such future interventions be pharmacological, psychotherapeutic, or neuromodulatory. Our current interventions for treating acute depression all have limitations for maintenance treatment. Specifically, psychotherapy may be less effective for maintenance than for acute treatment of LLD,^6,143 antidepressant medications may carry the risk of tachyphylaxis, and the benefits of neuromodulation strategies diminish once treatment is complete.^144,145 Once remission is achieved, we need better strategies to maintain remission and prevent new episodes.

We need better methods to identify individuals prone to recurrence and stratify their individual risk. We also need a process to monitor those individuals for early warning signs of recurrence, allowing clinicians to intervene early in the course of a new, emerging depressive episode. EMA may provide such a process, allowing ongoing assessment in one’s normal environment of clinical symptoms, cognition, and physiological measures such as sleep. Recent advances in smartphone technology and mobile health provide new opportunities for monitoring¹⁴⁶ that could allow for early detection of new episodes.

If these lines of research can reduce the risk for recurrence or increase the time between depressive episodes, this may improve other adverse outcomes associated with LLD. This may also benefit negative outcomes of depression, such as disability and suicide, or improve outcomes of comorbid conditions such as vascular disease and dementia.

Challenges and Next Steps

Although crucial to improve long-term health and prognosis of individuals with recurrent depression, studying the neurobiology of recurrence carries significant challenges. This requires a longitudinal study design, following individuals with remitted depression while monitoring for recurrence. Longitudinal cohort studies can be challenging and require robust plans for subject engagement and cohort maintenance. Moreover, they require sufficient power to achieve the study aims, a value directly related to the number of anticipated events. Past work in LLD suggests that even with ongoing treatment, we should expect recurrence rates of 43% in 2 years and 57% over 4 years.⁸ Although recurrence is not rare, power determinations need to account for the natural attrition seen in any longitudinal study of older adults. Moreover, if the research strategy is to recruit and treat currently depressed individuals, some will not remit or respond to treatment, thus requiring an even larger initial sample. These questions cannot be definitively answered in a small sample at a single site, but instead require a larger collaborative team across several institutions.

Clinical changes need to be examined in context of the underlying neurobiology in order to test and improve the scientific model and identify mechanistic targets for intervention. This requires monitoring physiological systems involved in the stress response and repeat neuroimaging. This could elucidate how homeostatic disequilibrium contributes to recurrence and how allostatic load contributes to pathological brain aging.

Such a study should apply a longitudinal “deep phenotyping” approach, wherein repeated data measures are gathered across multiple domains using a variety of methods. This is critical to cast a wide net to identify biological or behavioral markers that may change over time and signify the development of or resilience to recurrence risk. Data can include assessment of standard neurocognitive domains along with measures of emotional cognitive performance, medical morbidity, environmental measures of stress exposure, social support, and assessment of a range of relevant clinical symptoms using both standardized assessments and self-report tools.

This can be facilitated by EMA. EMA methods include active monitoring of depressive symptoms, stress levels, and cognitive performance, an approach that could detect new episodes but may be limited by patient adherence. It also includes passive collection of smartphone data, or the opportunity to gather 24-hour data without any active demands on the user. Such passive approaches incorporate data from ambient light sensors, actigraphy, global positioning system (GPS) location, and text or call logs.¹⁴⁶ Newer techniques collect typing kinematics data as individuals use their smartphone in daily life, examining typing speed as a marker of processing speed.¹⁴⁷ Many of these measures correlate with self-report and structured interview measures of depression severity.^146,147 Ecological measures using either active or passive approaches will be important to assess symptoms or function in daily life, which may provide more robust markers of vulnerability than clinically-administered retrospective instruments. EMA could also assess and modify lifestyle factors such as sleep, diet, exercise, mindfulness practices and treatment adherence.

There are challenges to EMA, as not all patients have smartphones and questions remain about assessment validity and individual privacy. However, these techniques are feasible and acceptable to older adults,¹⁴⁸ although it raises the challenge of accounting for potential cognitive difficulties (e.g. speed, attention and visuospatial abilities). A benefit to EMA is the ability to affordably collect of large amounts of data. The repeated measures approach allowed by EMA may mitigate the limited time and resources inherent in any study design.

EMA may allow for prospective monitoring and early identification of new episodes. However, there remains questions of what symptoms to monitor and how to define a “clinically relevant” change. Although change in many symptoms could serve as markers of an impending depressive episode, we propose that a simple assessment of mood and sadness is a reasonable starting point. Sadness is a key characteristic of depressive episodes and we previously found that greater levels of sadness are associated with increased recurrence risk.⁸ Such evaluations would be accompanied by other symptom assessments and passive sensor measures, allowing us to test alternative hypotheses about what measures may be the best markers of impending recurrence. Passive and active smartphone-based EMA collection can be accompanied by automated algorithms monitoring for state shifts (persistent worsening) and variance shifts (increased variability) relative to a patient’s baseline. Identifying change points in sensor data either as mean shifts or variance shifts is a well-known problem in statistical control research¹⁴⁹ and could provide an automated tool that may alert clinicians to at-risk individuals or relevant symptom changes even in a large practice. The same platform may also deliver tailored therapies such as e-CBT targeting specific symptoms (e.g. sleep/pain/anxiety) when detecting signs of recurrence.

Conclusions

The proposed framework could inform LLD treatment in several ways. First, by identifying high-risk individuals who could be better monitored and treated more aggressively even during remission. Second, through the delivery of timely interventions such as augmentation with either pharmacotherapy or psychotherapy when validated tipping-points are identified by neurobiological markers, cognitive performance, or clinical symptoms. Third, this could inform future prevention studies by identifying key mechanistic targets.

Such research is critically important to better understand the neurobiological factors underlying depression’s recurrent pattern. Our knowledge of the neurobiology of depression is strongest in relationship to acute episodes, and to a lesser extent of changes that occur during recovery from acute episodes. In contrast, we have less data on neurobiological differences that persist into remission and how they influence recurrence risk. Simply put, if we consider the cumulative burden of depressive episodes over the lifespan, we need better therapeutic targets and preventive strategies. We can only achieve these goals with a better understanding of underlying neurobiological changes in context of real-life symptoms and function.

Highlights.

• What is the primary question addressed by this study? What factors are predictive of and contribute to recurrence of depressive episodes?

• What is the main finding of this study? We propose that depression recurrence is related to stressor-induced disruption of neural homeostasis, resulting in changes in behavior, mood, and cognition.

• What is the meaning of the finding? This hypothesis provides a framework for future studies examining depression recurrence that include monitoring approaches to detect newly developing episodes.